Citation
A real-time expert system for citrus microirrigation management

Material Information

Title:
A real-time expert system for citrus microirrigation management
Creator:
Xin, Jiannong, 1961-
Publication Date:
Language:
English
Physical Description:
xvi, 204 leaves : ill. ; 29 cm.

Subjects

Subjects / Keywords:
Crops ( jstor )
Expert systems ( jstor )
Fertigation ( jstor )
Irrigation ( jstor )
Irrigation management ( jstor )
Irrigation scheduling ( jstor )
Irrigation systems ( jstor )
Rain ( jstor )
Sensors ( jstor )
Tensiometers ( jstor )
Agricultural and Biological Engineering thesis, Ph. D
Citrus -- Irrigation ( lcsh )
Dissertations, Academic -- Agricultural and Biological Engineering -- UF
Irrigation engineering ( lcsh )
Microirrigation -- Computer programs ( lcsh )
City of Gainesville ( local )
Genre:
bibliography ( marcgt )
non-fiction ( marcgt )

Notes

Thesis:
Thesis (Ph. D.)--University of Florida, 1995.
Bibliography:
Includes bibliographical references (leaves 189-203).
General Note:
Typescript.
General Note:
Vita.
Statement of Responsibility:
by Jiannong Xin.

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
The University of Florida George A. Smathers Libraries respect the intellectual property rights of others and do not claim any copyright interest in this item. This item may be protected by copyright but is made available here under a claim of fair use (17 U.S.C. §107) for non-profit research and educational purposes. Users of this work have responsibility for determining copyright status prior to reusing, publishing or reproducing this item for purposes other than what is allowed by fair use or other copyright exemptions. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder. The Smathers Libraries would like to learn more about this item and invite individuals or organizations to contact the RDS coordinator (ufdissertations@uflib.ufl.edu) with any additional information they can provide.
Resource Identifier:
021985520 ( ALEPH )
33838959 ( OCLC )

Downloads

This item has the following downloads:


Full Text










A REAL-TIME EXPERT SYSTEM
FOR CITRUS MICROIRRIGATION MANAGEMENT












By

JIANNONG XIN

















A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 1995




























Copyright 1995

by

Jiannong Xin














ACKNOWLEDGEMENTS


I would like to express my sincere gratitude to the supervisory committee. Special thanks go to Dr. Fedro S. Zazueta, supervisory committee chairman, for his advice and encouragement during the study and for the opportunity to study as a graduate research assistant. Special thanks go to Dr. Allen G. Smajstrla, supervisory committee member, for making himself available on many occasions and for contributing his expertise toward this study. I would also like to thank the supervisory committee members, Dr. James W. Jones, Dr. Pierce H. Jones, Dr. Douglas D. Dankel, II, and Dr. Louis H. Motz, for their advice and support during this study.

Special thanks also go to Dr. Thomas A. Wheaton and Dr. Lawrence R. Parsons of the Lake Alfred Citrus Research Center for their expertise, advice, and time toward this study.

I would like to extend my appreciation to the following people for their evaluation and comments on the software: Dr. Robert M. Peart and Dr. Dorota Z. Haman of the Agricultural and Biological Engineering Department, University of Florida; Dr. James J. Ferguson and Dr. J. David Martsolf Jr. of the Horticultural Department, University of Florida; and Ms. Cynthia Moore, St. Johns River Water Management District, Florida.

Finally, I would like to thank my wife, Qiuping Jian, for her support during this study.

111















TABLE OF CONTENTS


ACKNOWLEDGMENTS ......... ............................ 111

LIST OF TABLES ........................................ Viii

LIST OF FIGURES ...................................... ......... x

ABSTRACT ......... ......................................... xvii

CHAPTER 1 INTRODUCTION ......................................... I1

1.1 Statement of the Problem .................................... 1
1.2 Objectives of the Dissertation ............................... 5

CHAPTER 2 REVIEW OF THE LITERATURE ............................ 7

2.1 Citrus Irrigation in Florida ................................... 7
2.2 Soil Moisture Sensors ...................................... 8
2.3 Irrigation Scheduling ....................................... 9
2.3.1 Monitoring Method ................................. 10
2.3.2 Computer Simulation ................................ 11
2.4 Irrigation Control ....................................... 13
2.5 Expert Systems in Agriculture .............................. 15
2.6 Summary ........................................ 17

CHAPTER 3 GENERAL EXPERT SYSTEM CONCEPTS ................... 19

3.1 Expert Systems ........................................ 19
3.1.1 Inference Engine ................................... 20
3.1.2 Knowledge Base .................................. 21
3.1.3 User Interface ..................................... 22
3.2 Real-Time Expert Systems .................................. 22
3.2.1 Why Use anRTES? ............................... 24
3.2.2 Characteristics of RTESs ............................ 24




iv









3.3 Knowledge Acquisition .................................... 25
3.3.1 Basic Approaches ................................... 26
3.3.2 Potential Problems .................................. 27
3.3.3 Practical Issues ..................................... 28
3.4 Knowledge Representation .................................. 29
3.4.1 Semantic Network ............................... .. 30
3.4.2 Frame ............................................ 30
3.4.3 Objects ........................................... 31
3.4.4 Rules ........................................... 32
3.5 Rule-Based Expert Systems ............................... 33
3.5.1 Rule-Based Architectures ............................ 34
3.5.2 Uncertainty Management ........................... 35

CHAPTER 4 SYSTEM SPECIFICATION AND DESIGN .................... 37

4.1 Domain of the Problem .................................... 37
4.2 Requirements Specification ................................. 38
4.2.1 Goalof the System .................................. 38
4.2.2 System Inputs ..................................... 38
4.2.3 System Outputs .................................... 39
4.3 Knowledge Specification ................................... 40
4.4 Knowledge Representation Paradigm .......................... 41
4.4.1 Reasoning Method ................................ 41
4.4.2 System Performance Requirements ...................... 42
4.5 Development Tools ....................................... 42
4.5.1 Expert System Shells ................................ 43
4.5.2 CLIPS ......................................... 44
4.6 Hardware Specification ............... ............... 44
4.6.1 Soil Moisture Sensor ............................. 45
4.6.2 Personal Computer .................................. 49
4.6.3 Automated Weather Station ........................... 49
4.6.4 Data Logger ..................................... 49
4.6.5 PC Digital Input/Output Board ......................... 50
4.6.6 Irrigation Control Board .............................. 50
4.6.7 Overview of the Hardware ............................ 52
4.7 Paradigm of the Real-Time Expert System ...................... 53

CHAPTER 5 PROBABILITY OF RAINFALL ............................. 55

5.1 Introduction. .......................................... 55
5.2 Markov Chain ......................................... 56
5.3 Rainfall Data ........ .................................. 58
5.4 Frequency of Rainfall ...................................... 59

V









5.5 Statistical Test ........................................... 60
5.6 Irrigation Decision with Rainfall Probability .................... 62

CHAPTER 6 CITRUS IRRIGATION SCHEDULING ....................... 64

6.1 Introduction. ......................................... 64
6.2 Citrus Water Requirements ................................. 64
6.3 Evapotranspiration and Management Allowed Depletion ........... 67
6.4 Irrigation Depth and Duration ............................... 69
6.5 Soil-Water Budget ...................................... ..71
6.6 Irrigation Scheduling Using Tensiometers ...................... 72
6.6.1 Tensiometer Installation Depth ......................... 73
6.6.2 Soil-Water Potential and Allowable Water Depletion ........ 74

CHAPTER 7 CITRUS COLD PROTECTION AND FERTIGATION ........... 77

7.1 Introduction. ....................................... 77
7.2 Cold Protection Application ................................. 78
7.2.1 Principle of Cold Protection ........................... 78
7.2.2 Critical Application Temperature ....................... 79
7.2.3 Water Application Rate .............................. 80
7.3 Fertigation ............... ............................ 80
7.3.1 Application of Fertigation ............................. 81
7.3.2 System Components of Fertigation ...................... 84
7.3.3 Fertilizer Materials ................................ 85

CHAPTER 8 CONSTRUCTION OF THE KNOWLEDGE BASE .............. 88

8.1 Introduction. ........................................ 88
8.2 The Process of Control and Reasoning ......................... 89
8.3 The Sensor Data ................................... ....93
8.3.1 Download the Sensor Data ............................ 93
8.3.2 Uncertainty Management of the Sensor Data .............. 94
8.4 Irrigation Management .................................... 100
8.4.1 Irrigation Strategies ................................ 100
8.4.2 Criteria for Starting an Irrigation ...................... 102
8.4.3 Criteria for Stopping an Irrigation ..................... 105
8.5 Cold Protection ....................................... 108
8.5.1 When to Turn On .............................. 109
8.5.2 When to Turn Off ................................. 109
8.6 Fertigation ........................... ............. 110




vi









CHAPTER 9 SYSTEM IMPLEMENTATION AND TESTS ................. 112

9.1 Function Requirements of CIMS ............................ 112
9.2 Module Design of CIMS .................................. 112
9.2.1 Expert System Module .............................. 114
9.2.2 Control Panel ................................... 115
9.2.3 Scheduling ....................................... 116
9.2.4 Database ......... ..................... 117
9.2.5 Simulation ..................................... 117
9.2.6 Tools ........................................ 118
9.2.7 Help ........................... ............. 118
9.2.8 User Interface .................................... 119
9.3 Data and Message Passing of CIMS .......................... 119
9.3.1 Data Flow of the RTES Module ...................... 119
9.3.2 Data Requirements of the Simulation Module ............ 120
9.3.3 Data Requirements of the Scheduling Module ............ 121
9.4 Maintenance of CIMS .................................. 121
9.5 System Tests of CIMS .................................... 123
9.5.1 Predictive Tests ................................... 124
9.5.2 Field Tests ..................................... 124
9.5.3 Simulated Crop Water Use ........................... 125

CHAPTER 10 SUMMARY AND CONCLUSION .................... 130

APPENDIX A SAMPLE SENSOR DATA ..................... 135

APPENDIX B SAMPLE TEST CASES OF KNOWLEDGE BASE ....... 142

APPENDIX C USER'S GUIDE OF CIMS ........................... 151

LIST OF REFERENCES ........................................ 189

BIOGRAPHICAL SKETCH ........................................ 204













vii
















LIST OF TABLES


Table ae

4.1 System input and output requirements ......................... 39

4.2 Characteristics of pressure transducer model 141PC ............... 47

5.1 Markov chain wet-day frequency ............................. 59

5.2 Results of paired t-test for rainfall probabilities within each season ... 61 6.1 Citrus irrigation water requirements in central Florida ............. 65

6.2 Citrus crop coefficients and recommended MAD in Florida ......... 68 6.3 Average soil-water content for Candler fine sand by volume ........ 74

6.4 Estimated soil-water tension in corresponding to soil-water
depletion for Candler fine sand .............................. 75

7.1 Pounds of nitrogen fertilizer to be applied to furnish nitrogen
requirement of orange and grapefruit trees under normal conditions .. 83 7.2 Solubility of common fertilizers in water ....................... 87

8.1 Criteria for checking possible sensor Failure B .................. 97

8.2 Criteria for checking possible sensor Failure C for sensors at the
same depth from different locations ........................... 97

8.3 A sample propagation of CF ................................. 98

8.4 Sensor readings and constraints to start an irrigation ............. 103




viii








8.5 Critical sensor readings (Criteria II) to start an irrigation .......... 105

9.1 DatainputofCIMS .............................. 121

9.2 Accumulated citrus net irrigation requirements and number of
irrigations for 22 years in central Florida ...................... 127










































ix















LIST OF FIGURES


Figure aPse 3.1 Major components of an expert system ........................ 20

3.2 Major components of an RTES .............................. 23

3.3 Knowledge acquisition cycle ............................... 25

3.4 The architecture and execution cycle of rule-based systems ......... 34

4.1 A regular tensiometer and a tensiometer with micro-pressure
transducer ................... ........................ 46

4.2 Pressure transducers (Model 141PC) from Micro Switch ........... 46

4.3 Tensiometer calibration equipment .................. ......... 48

4.4 Tensiometer calibration curve ............................. 48

4.5 Solenoid control relays of the irrigation system .................. 51

4.6 Hardware layout of the control system ......................... 52

4.7 Paradigm of the real-time expert system ........................ 53

5.1 Annual rainfall distribution in Orlando from 1952 to 1992 .......... 60

6.1 Irrigation by threshold of soil-water content ..................... 72

6.2 Soil-water content versus soil-water tension for Candler fine sand .... 76 7.1 Major components of a fertigation system ...................... 85

8.1 Major inputs and outcomes of the expert system .................. 88

x









8.2 Decision flow of the expert system ............................ 90

8.3 Paradigm of the knowledge base ............................. 91

8.4 Process of downloading weather and soil moisture sensor data ....... 93 8.5 CF propagation by checking range of sensor data ................. 95

8.6 CF propagation for sensors S1 and S3 ......................... 96

8.7 CF propagation for sensor S2 ................................ 96

8.8 Process of selecting valid sensor readings from different locations .... 99 8.9 Decision process to use a full or deficit irrigation strategy ......... 101

8.10 Decision process to start an irrigation (criteria I) and sensor readings
to trigger an irrigation for trees during different growth stages. ..... 103

8.11 Decision process (criteria II) to start an irrigation and critical sensor
readings for trees during different growth stages. ................ 104

8.12 Decision process to stop an irrigation ......................... 106

8.13 Cold protection decision processes based on the critical air
temperature. ..................................... 108

9.1 Program modules of CIMS ............................... 113

9.2 Control panel of CIMS ................... ........... 116

9.3 Irrigation system database ................... ....... 117

9.4 Data flow of CIMS .................. ............ 120

9.5 Structure of the knowledge base and initial facts ............ 123

12.1 Tensiometer readings on Juilan day 213, 1994 ............... 139

12.2 Tensiometer readings on Juilan day 277, 1994. ............. 140

12.3 Tensiometer readings on Juilan day 278, 1994. ............. 141


xi









13.1.1 Main menu of CIMS ............................. 152

13.2.1 Submenu of the Facts ............................. 154

13.2.2 Irrigation block and valve definition .................... 155

13.2.3 Fertigation block and valve definition ................... 156

13.2.4 Initial facts of the expert system ...................... 157

13.2.5 Fertigation schedule ............. .............. 157

13.2.6 Submenu of the Expert main menu ..................... 159

13.2.7 Execution screen of the RTES . . . . . . . . . . . 159

13.2.8 Screen of sensor readings and application status ............. 160

13.2.9 Submenu to view application history .................... 160

13.3.1 Control panel of the system ......................... 161

13.4.1 Submenu of user defined control schedules ............... 162

13.4.2 User defined irrigation schedule screen .................. 163

13.4.3 User defined fertigation schedule screen .................. 163

13.4.4 Irrigation application screen .......................... 164

13.4.5 Irrigation valve on or off display ...................... 165

13.4.6 Fertigation dialog screen ........................... 166

13.4.7 Fertigation valve on or off display ..................... 167

13.4.8 Irrigation and fertigation dialog screen .................... 168

13.5.1 Database control button ............................. 169

13.5.2 Delete dialog screen ................................ 179

13.5.3 Search dialog screen .............................. 170

xii










13.5.4 Browse dialog window ............................ 171

13.5.5 Submenu of the Database main menu ................... 172

13.5.6 Farm database screen ............................. 172

13.5.7 Weather database screen ........................... 173

13.5.8 Irrigation database screen ........................... 173

13.5.9 Crop database screen .............................. 173

13.5.10 Crop coefficient database screen ...................... 174

13.5.11 Soil database screen .............................. 174

13.6.1 Simulation submenu .............................. 175

13.6.2 Screen to set initial condition of the simulation ............. 176

13.6.3 Simulation dialog window .......................... 176

13.6.4 Irrigation prognosis from the simulation ................. 177

13.6.5 Simulated soil-water content ......................... 178

13.7.1 Submenu of Tools ............................... 179

13.7.2 Read weather data from weather station ................. 180

13.7.3 ET method dialog window ............................ 180

13.7.4 Penman ET screen ............................... 181

13.7.5 Blaney-Criddle ET screen .......................... 181

13.7.6 Modified Blaney-Criddle ET screen ...................... 182

13.7.7 Stephens-Stewart ET screen ......................... 182

13.7.8 Estimate irrigation duration screen ..................... 183








13.7.9 Help screen of irrigation duration ..................... 183

13.7.10 A dummy field layout map ......................... 184

13.8.1 Submenu of the Help ............................. 185

13.8.2 CIMS help screen ............................... 185

13.8.3 Calculator .................................... 186

13.8.4 Calendar and diary screen .......................... 186

13.8.5 Clock ....................................... 187

13.8.6 Text editor .................................... 187

13.8.7 Puzzle ...................................... 187

13.8.8 About the CIMS ................................ 187

13.8.9 More about the CIMS ............................. 188

13.8.8 Screen to quit from the CIMS ........................ 188
























xiv













Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy


A REAL-TIME EXPERT SYSTEM FOR CITRUS MICROIRRIGATION MANAGEMENT By

Jiannong Xin

August, 1995

Chairman: Dr. Fedro S. Zazueta
Major Department: Agricultural and Biological Engineering

Elaborate techniques are commonplace in modem farm management and microirrigation scheduling for citrus. Water management practices involve complex decisions and daily operations that are affected by water and nutrient requirements of the trees, temporal distribution of rainfall, and extreme weather conditions. A computer-based system (CIMS) was developed using a real-time expert system (RTES) and conventional control techniques to assist citrus microirrigation, cold protection, and fertigation management. The system integrates water management technologies into an effective control technique that can be used as a tool by farm managers. CIMS combines the RTES, conventional control, and irrigation management tools into a single system to help the decision-making of irrigators. On-site soil moisture sensors and an automated weather station provide data to the system. CIMS activates or deactivates the control devices of an xv








irrigation system based upon its knowledge base. The expert system operates continuously to select an irrigation strategy and to schedule the application of irrigation, fertigation, and cold protection. Data uncertainty management approaches were used to validate the sensor readings. Conventional control routines were also developed so that irrigation and fertigation could be applied according to user defined schedules with control flexibility and few hardware requirements. A simulation model of the crop root zone was developed to estimate crop water requirements to help the user to define irrigation schedules. A shortterm prognosis of an irrigation requirement can be generated from the simulation. Database and farm management utilities were also included in the system to assist the decision-making of farm managers. Both laboratory and field tests showed that the integrated system worked as expected as a management tool for irrigation, fertigation, and cold protection. The system is highly automated and has the potential to improve microirrigation management, to achieve water and energy savings, and to prevent water pollution due to improper fertigation management.


















xvi













CHAPTER 1
INTRODUCTION


1.1 Statement of the Problem


Modern farm management involves complex decisions and daily operations that are affected by water and nutrient requirements of crops, temporal distribution of rainfall, environmental protection, and extreme weather conditions. In recent years, increasing costs of energy, increasing water demands from non-agricultural users, and adverse weather cycles are forcing the agricultural industry to use new technologies to improve water management capabilities and to increase the efficiency of resources used in production.

Irrigation is the largest consumer of fresh water in the world. In Florida, agriculture accounts for over 40 percent of total fresh water use (Fernald and Patton, 1984) -- about 3,000 million gallons per day. Citrus, one of the major crops in Florida, is a billion dollar industry and consumes millions of gallons of water for irrigation each year. Thus, even modest increases in water use efficiency will result in substantial water savings and reduce energy cost.

Operational costs for irrigation are increasingly due to increasing energy costs. Agricultural energy consumption varies greatly among the different commodities and agricultural practices in Florida. For citrus, an energy survey (Stanley et al., 1980) showed that 11,588.90 billion BTUs were used in 530,000 acres of production with 32.2 percent of

1








2

the energy used in irrigation. Currently, close to 16,000 billion BTUs (the energy equivalent of approximately 128 million gallons of gasoline) (Zazueta et al., 1994) are needed by about 750,000 acres of citrus production. If 32.2 percent of this energy is used in irrigation, about 5,152 billion BTUs are consumed only for irrigation. A conservative 10 percent reduction in energy use by using better control systems would result in a savings of 515 billion BTUs.

Fertilizer and chemical applications are common in agricultural practice. Because water is used to convey many of the chemicals through microirrigation systems, the efficiencies of fertigation and chemigation are directly related to the water application efficiency. Furthermore, improper chemical applications may result in adverse environmental impacts. Ground and surface water can be contaminated if applied chemicals are transported out of the crop root zone due to inadequate management.

It is generally accepted that improved management techniques are necessary to increase water use efficiency, particularly in the scheduling and application of irrigation. Current irrigation management practice attempts to satisfy crop water demands and relies, for the most part, on manual operation or timers for control of the irrigation system. Many irrigation schedules rely on calendars, evapotranspiration (ET) estimation, and the grower's experience. Water can be wasted due to poor irrigation management, and irrigation may even be applied during rainfall when a preset timer is used to control the system (Xin et al., 1993).

Computer-based irrigation scheduling has received much attention. Simulation models have been developed for various irrigation strategies and to assist irrigation scheduling using historical weather records (Lembke and Jones, 1972; Swaney et al., 1983;








3

Villalobos and Fereres, 1989; Rogers and Elliott, 1989). Irrigation scheduling approaches based on crop growth models and soil water budget components have been developed by researchers (Jensen et al., 1971; Chesness et al., 1986; Smajstrla and Zazueta, 1987; Jones and Ritchie, 1990). Models have also been developed to determine optimal irrigation strategies using stochastic and probabilitic models of weather variables (Khanjani and Busch, 1982). Long-term historical weather data are commonly used by such simulation models. Many agricultural simulation models, which use historical weather data, have succeeded in planning and long-term prediction of irrigation management. However, factors such as the difficulty of model development, uncertainty of future conditions, and limitations of available data combine to make the use of simulation models difficult for real-time application.

With increasing competition for the use of water and high energy costs associated with irrigation, microirrigation systems have become common in Florida, particularly for high cash value crops like citrus. Because a microirrigation system wets only the soil volume around the emitters, microirrigation systems must apply water at a high frequency. Water should be applied at a rate equal to plant uptake (Phene et al., 1992). The soil water potential can be maintained reasonably constant under high frequency irrigation scheduling. Irrigation can be applied at least daily at a rate equal to the ET requirement; consequently, there is a need for "real-time" irrigation scheduling and control systems (Phene et al., 1989b).

Cold protection is an important issue for citrus growers. Cold weather has caused severe economic damage to the Florida citrus industry in the past, particularly in 1989. The








4

primary cold protection method for Florida citrus is irrigation (Parsons et al., 1989; Parsons and Wheaton, 1990). Effective irrigation management for cold protection can reduce tree loss and increase profitability. However, cold protection management requires timely and accurate climatic data so that adequate protection measures can be taken. On-site real-time monitoring of weather data and expert knowledge on cold protection are necessary for farm management.

As personal computers have become increasingly common, the potential for computer-based decision support systems for farm water management has also increased. Computerized irrigation scheduling systems have been developed by Cahoon et al. (1990), Phene et al. (1992), and Zazueta et al. (1984a, 1994). Expert systems techniques can be used to represent the heuristic knowledge required for decision making. Unlike simulation systems, which are based on mechanistic biological or mathematical models, expert systems use expert knowledge in the decision process like that used by human decision-makers. Real-time expert systems (RTES) operate in a real-time domain and deal with dynamic data and time critical responses, applying expert systems technology to control engineering.

In an RTES, most of the inputs come from sensors, while many of the outputs go to effectors. Soil moisture sensors and weather stations can monitor soil water content and climatic conditions, respectively. Expert knowledge can be acquired to develop several alternative strategies and apply the one most suited to a specific problem. With the real-time soil and weather data monitoring integrated with expert knowledge on farm management, the system can be operated in real-time.








5

1.2 Objective of the Dissertation


Agricultural production is related to many factors including crop, soil, and biological conditions, and management decisions. Many complex decisions must be made daily. New techniques are needed to assist farm managers. With the complexity of modern farm management and available computer technology, it appears that an RTES is a means to assist farm management. The primary goal of this research was to develop a methodology using an RTES to improve the management of citrus microirrigation systems. The specific objectives of this dissertation were

1) To acquire expert knowledge on citrus microirrigation management.

2) To develop control routines and a control panel to turn on or off user

specified valves from a local or remote computer.

3) To develop a user-friendly RTES for citrus irrigation, fertigation, and cold

protection management.

4) To provide alternative control functions so that irrigation and fertigation can

be applied according to user defined schedules.

5) To use farm databases, crop water requirement simulation, and other

computer tools to assist the decision-making of farm managers.

6) To demonstrate the use of an RTES as an operational tool to improve

management of an irrigation system.

Acquiring expert knowledge is crucial in the development of an expert system. Experts need to be identified to acquire their knowledge in the problem domain: citrus








6

microirrigation management. Since the expert system must be operated in the real-time domain, control hardware is required. The control process can be accomplished by using conventional programs. As an RTES applied to microirrigation management, its reasoning process is not as time critical as military applications. In other words, the system is not a hard RTES. The use of the term RTES is to distinguish the system from expert systems in which time is not a factor at all or which acquire data only from static databases.

The components of farm databases, crop water requirement simulation, and computer-controlled irrigation systems have been successfully applied. The integration of an expert system with simulation models, databases, and user defined control needs to be resolved in this study. This integration can rely heavily on the structure of the system design, functionality of the expert system shell, and design of the user interface. With modem software development tools, operating systems with multitasking capabilities, and object-oriented software design, this integration can be achieved. After the system is developed, system validation can be conducted by running generated test cases, expert evaluation, field tests, or a simulation approach.













CHAPTER 2
REVIEW OF THE LITERATURE


2.1 Citrus Irrigation in Florida


Citrus is one of the major crops in Florida. The total acreage of citrus was 853,742 acres in 1994 (Florida Agricultural Statistics, 1994). The citrus industry is a significant contributor to the economy of Florida. Its annual economic impact on the state's economy has been estimated at billions of dollars.

Although the average yearly rainfall in Florida varies from 50 to 62 inches, irrigation is required to achieve maximum production and improve the quality of citrus fruit (Koo, 1963; Tucker, 1983). Citrus irrigation systems are also used for cold protection purposes. Microirrigation systems are common today in Florida for citrus irrigation. Of Florida's 1,855,390 irrigated acres, 19 percent is citrus (Smajstrla et al., 1995). Billions of gallons of water are required for the industry each year. Thus, even modest increases in water use efficiency will result in substantial water savings.

In Florida, microirrigation is the preferred method for citrus irrigation. Microirrigation is an efficient and convenient means of supplying water directly to a crop root zone. It provides an effective means for utilizing small continuous streams of water for irrigation. Furthermore, microirrigation systems more easily realize computerized control than do other types of irrigation systems.

7








8


2.2 Soil Moisture Sensors


The ability to measure soil moisture in-situ is important for irrigation management. Irrigation water can be saved by using soil moisture sensors (Zazueta et al., 1993). However, the choice of soil moisture sensor is crucial to the success of irrigation control and management. Usually, "the most intractable barrier to the full implementation of automatic process control is the lack of adequate on-line sensors (p. v)" (Carr-Brion, 1986). A poor choice of a sensor at the design stage is commonly caused by lack of adequate appreciation of the limitations of the type of sensor used or by lack of knowledge of what is available.

Many literature reviews can be found on soil-moisture measurement by a variety of techniques (Taylor, 1955; Schmugge et al., 1980; McKim et al., 1980; Erbach, 1983; Wheeler and Duncan, 1984; Gardner, 1986; Stafford, 1988; Zazueta and Xin, 1992). The techniques commonly used in soil moisture sensors include (1) electromagnetic, (2) nuclear,

(3) remote sensing, (4) hygrometric, (5) tensiometric, (6) optical, and (7) time domain reflectometry (TDR). Not all these soil moisture sensing techniques are suitable for automation. The sensor must have the capability of interfacing with a computer or other electronic devices. Sensor cost is another major concern for agricultural applications. Some soil moisture sensors, such as TDR and neutron probes, can achieve high accuracy (Topp and Davis, 1985; Simpson and Meyer, 1987), but costs of the devices are also high. Tensiometers are relatively inexpensive and are easy to use.

Tensiometers measure the matric potential (capillary tension) directly, which is related to the energy required for plants to extract water from the soil. Tensiometers are the








9

primary method for measuring matric potential in soil. They have a fairly fast response time when used for irrigation (Towner, 1980; Stone et al., 1986). A pressure transducer can be installed on a tensiometer and interfaced to a data acquisition or readout system to realize automation. The use of tensiometers with pressure transducers for soil-water potential measurement has been successful in many applications (Fitzsimmons and Young, 1972).

The advantages of tensiometers are (1) low cost and easy construction, (2) easy installation and maintenance, (3) long periods of operation if properly maintained, and (4) adaptable to automatic measurement with pressure transducers. The disadvantages are (1) a limited range of 0 to -0.8 bar that is not adequate for some soils, (2) hysteresis, and (3) potential breakage during installation and cultural practices.


2.3 Irrigation Scheduling


Irrigation scheduling requires making decisions on when to irrigate and how much water to apply. The main techniques used for scheduling include (1) monitoring of soil moisture, (2) physiological indicators, and (3) soil water balance models. Proper irrigation scheduling should result in savings of water and energy without yield reduction. Irrigation scheduling decisions may relate to crop response to water stress, management objectives, water quality control, system constraints, and public policies. Maintaining adequate soil moisture levels in the crop root zone is critical for crop growth. Inadequate soil moisture not only limits water supply to the roots, but also reduces root conductivity directly (Wiersum and Harmanny, 1983).








10

Although irrigation scheduling has been studied in various ways for a long time, research is needed to reduce the consumption of water and energy and to increase profitability through better scheduling. Numerous studies have been conducted on irrigation scheduling (Pleban et al., 1983, 1984; Zoldoske, 1988; Rogers and Elliott, 1989; Shayya et al., 1990; Protopapas and Georgakakos, 1990). Monitoring of soil water content, crop growth, and weather conditions is important for irrigation scheduling.

2.3.1 Monitoring Method

Monitoring methods are primarily based on either crop or soil measurement. Monitoring can rely on instruments or one's intuition. Soil moisture is usually monitored by using a sensor to measure soil water potential (Campbell and Campbell, 1982). Irrigation is applied when the monitored crop or soil data reach some critical value. Irrigation scheduling can also rely on the monitoring of weather data (Howell et al., 1984).

Soil moisture sensors are one of the major tools used to assist decision-making on irrigation water applications. Tensiometers and gypsum blocks are widely used in the field (Cary and Fisher, 1983). Augustine and Snyder (1984) and Snyder et al. (1984) used tensiometers to schedule irrigation for bermudagrass turf. Their results showed that irrigation water savings of 42 to 95 percent were obtained in sensor controlled plots over conventionally irrigated plots. A study using tensiometers to schedule cotton drip irrigation was conducted by Wierenga et al. (1987). In all of these studies tensiometers were successfully used for irrigation scheduling.








11

Cassell and Klute (1986) studied soil effects on tensiometers. They found that the use of tensiometers for irrigation scheduling was more successful in coarse textured soils than it was in fine textured soils. This is because a greater percentage of the water available to a plant is retained by coarse textured soils at suctions less than 0.8 bar than is the case for fine-textured soils. Tensiometers only operate from zero to about 0.8 bar. Tensiometers are an effective tool to assist irrigation decision-making, but soil water potential must be maintained within their operational range.

2.3.2 Computer Simulation

As computer systems have become widespread, simulation-based approaches have been developed (Lembke and Jones, 1972; Swaney et al., 1983; Villalobos and Fereres, 1989; Rogers and Elliott, 1989). Soil water balance and crop growth simulation models are two common approaches.

Soil water balance

This method applies the principle of continuity to the root zone. It describes soil moisture change in the root zone over time. Using this approach to manage irrigation involves estimating the amount of water in the crop root zone. To maintain the soil water content in the crop root zone within a desired range, irrigation should be applied to satisfy evapotranspiration (ET) demands.

Jensen et al. (1970) reported a scheduling method based upon soil-crop-climate data. Soil water balance simulation models were developed by researchers (Jensen et al., 1971; Zazueta et al., 1986; Smajstrla and Zazueta, 1987; Anderson et al., 1978; Cahoon et al.,








12

1990) for different locations and crop types. Maintaining a soil water balance is a widely used and effective approach for irrigation scheduling. However, this approach can require substantial weather data, and these data are not available in many cases. Models to generate weather data have been developed for these purposes (Richardson, 1981, 1985; Richardson and Wright, 1984; Villalobos and Fereres, 1989; Jones, 1993). Simulation of crop growth

Crop growth models can be developed to simulate crop growth. A crop growth model can be a physically based representation of the dynamics of the soil-crop-atmosphere system. The crop yield can be predicted by explicit models of the plant growth process, such as assimilation, respiration, and transpiration (Protopapas and Georgakakos, 1990).

Crop growth models have been developed as aids for irrigation water management (Swaney et al., 1983; Rogers and Elliott, 1989; Jones and Ritchie, 1990, Jones, 1993). Although crop growth models have been successfully used in irrigation management and decision-making, the technique has some difficulties in practice. First, models are currently not available for all crops. This is because it is difficult to develop an accurate crop growth model. A crop system can be complex and affected by many factors such as weather, insects, weeds, diseases, soil physical and chemical factors, and the interactions of these factors. Second, factors such as uncertainty of future conditions and limitations of available data make the use of simulation models difficult for real-time applications.








13


2.4 Irrigation Control


Irrigation system control includes a variety of topics, ranging from on-stream storage and water diversions to agronomic practices (Duke et al., 1990). The control topics may relate to (1) hydraulic, (2) mechanical, (3) electro-mechanical, (4) electronic, and (5) computerized control (Duke et al., 1990). The major irrigation control modes are (1) on-off control, (2) stepwise control, and (3) continuous control (Phene, 1986). Computerized systems have shown great potential in irrigation control and farm management. This is because a single hardware configuration can serve a wide range of control functions, and control strategies can be easily modified by software modifications.

Studies (Phene et al., 1973; Phene and Howell, 1984; Phene, 1989) have been conducted of irrigation control using soil moisture sensors. Sensors were used in a feedback mode to maintain a nearly constant soil moisture content in the root zone. They concluded that the performance of the irrigation controller depended on four basic factors: (1) adequate operation of the system's control hardware, (2) the proper algorithm for the system's software, (3) a reliable soil moisture sensor installed in the field, and (4) adequate operation of the system's output, the solenoid valves, the pressure regulators, flow meter, and filter. Further studies (Phene et al., 1989b) indicated that an irrigation controller should have the following characteristics to monitor soil matric potential in real-time and control irrigation systems: (1) ability to sample the sensor data automatically, (2) means of comparing the sensor output to a threshold value, and (3) ability to control and monitor irrigation devices.








14

Field tests demonstrated that this scheduling method should be easily adaptable to irrigation control, particularly in a sandy soil with low soil-water holding capacity.

Although most current irrigation management practices rely on manual operation or timers, many researchers have focused on computer controlled irrigation systems (Duke et al., 1984; Zazueta et al., 1984a, 1989; Phene et al., 1989a; Burns et al., 1990; Shayya et al., 1990; Zazueta and Smajstrla, 1992). Vellidis et al. (1990) developed a microcomputerbased data acquisition system for soil water potential measurements. The system consisted of commercially available components: tensiometers, pressure transducers, a data acquisition system, control devices, and a portable computer. In tests, the system was found to be effective to monitor temporal variation of soil moisture potential.

Computerized irrigation control systems have the potential for water and energy savings. Stombaugh et al. (1992) studied frost protection of strawberries using an automated pulsed irrigation system. Their studies showed that the automation of irrigation could potentially reduce water use by 89 percent under mild frost conditions. For nursery irrigation, Zazueta et al. (1984b) reported that when closed loop computer control was added to the system, water savings of about 20 percent (for a well-managed system) to 60 percent or more (for a poorly managed system) were achieved. These water savings were achieved by control of water deliveries using preset irrigation schedules. Even better results might be achieved if the system integrated expert knowledge on irrigation management.








15


2.5 Expert Systems in Agriculture


In the past few years there has been significant interest among researchers in the concept of expert systems. Expert systems have been widely applied in medicine, military, industry, and agriculture (Laffey et al., 1988; Jones, 1989a; Feigenbaum et al., 1994). As the development of expert system technology continues, expert systems are increasingly being used in applications that sense the environment and directly influence it through action. Techniques of real-time problem solving have been studied (Strosnider and Paul, 1994). In practice, real-time expert systems (RTESs) have been successfully developed for control, monitoring, and diagnosis applications (Padalkar et al., 1991; Schnelle and Mah, 1992; Harrison and Harrison, 1994). Advanced personal computers and commercially available easy-to-use expert system shells allow many people from different disciplines to develop expert systems (Durkin, 1994). The use of expert system design methodology in building agricultural decision support systems has shown great potential in recent years.

In agriculture, expert systems have been developed to assist the transfer of technology from agricultural researchers and extension services to producers. Many expert systems have been developed for management of nutrients, irrigation machinery, insect and weed control, disease diagnosis, harvesting, and marketing (Jones and Haldeman, 1986; Peart et al., 1986; Kalkar and Goodrich, 1986; Lemmon, 1986; Kline et al., 1987; Morey et al., 1988; Muttiah et al., 1988; McClendon et al., 1989; Batchelor and McClendon, 1992; Merlo, 1992; Kumar et al., 1992). Expert systems for crop management integrated with simulation models have been developed (Plant, 1989; Palmer, 1986). Expert systems can








16

help facilitate the use of simulation models in several ways (Jones, 1985): (1) estimate model parameters, (2) provide input for models, and (3) restrict scenarios for model analyses.

As expert system technology has evolved, applications to irrigation scheduling and operation have been developed. Wright et al. (1986) developed a real-time expert control system (Hexscon). They concluded that the important issue in developing an RTES is combining the best of conventional and expert-system controllers. Their results suggested that real-time expert control can be built on a microcomputer and has enough sophistication and capacity to be effective for real-world problems. Jacobson et al. (1987) developed an RTES that supervised a tomato greenhouse environment controller. Jacobson et al. (1989) implemented real-time greenhouse monitoring and control, linking a conventional expert system with a set of utilities for data acquisition and control. Conventional expert systems have been linked with models and data acquisition to make management recommendations. Thomson et al. (1989) reported on an expert system that was coupled with simulation of a peanut growth model and databases. This system evaluated moisture sensor readings combined with a crop growth model to make estimates of irrigation timing. An RTES has been applied to turf irrigation by Zazueta et al. (1989). In tests, it was found that the RTES can apply irrigation in response to crop water demand. A major defect of the system was the lack of heuristic knowledge available for irrigation control. An expert system for irrigation management was developed in Thailand (Srinivasan et al., 1991). The system demonstrated its effectiveness in improving water management decisions.








17

Although many expert systems have been developed for agricultural applications, these systems were successful to some extent from a purely pedagogical viewpoint; but very few of the systems are considered to be successful from a commercial viewpoint (Jones, 1989a). RTESs appear to have the potential to be successful because of their well-defined domains. However, developing such systems can be extremely difficult due to factors such as critical timing, knowledge acquisition, temporal reasoning, and uncertainty management.


2.6 Summary


Modem farm management and irrigation scheduling are complex tasks. There are many factors that need to be considered to achieve successful farm management. There is a need for more research in the field of farm management, which includes (1) irrigation management, (2) chemigation, (3) maintaining water quality, (4) pest control, (5) environment impact, (6) labor and energy conservation, and (7) cold protection. The "optimal" management of an agricultural farm involves complex daily operation and management decisions because of the temporal distribution of rainfall and extreme microclimate.

One way of dealing with the problems is by the introduction of expert systems integrated with control engineering techniques into irrigation management. With an RTES, a computer can be used to implement these tasks: (1) decision-making on irrigation, fertigation, and cold protection, (2) monitoring the performance of the irrigation system, (3) adjusting irrigation applications as climatic or other conditions change during the irrigation,








18

(4) processing feedback data to evaluate the irrigation process, and (5) maintaining a complete record of all applications.

Development of this RTES requires extended knowledge including (1) artificial intelligence, (2) software engineering, (3) citrus irrigation and fertigation management, and

(4) control engineering. Studies are needed to integrate the technology of water management into effective control engineering. An automatic weather station and soil moisture sensors are essential for the system. An RTES could offer a better tool for the management of irrigation, fertigation, and cold protection. RTESs are a feasible and necessary approach for citrus microirrigation management.













CHAPTER 3
GENERAL EXPERT SYSTEM CONCEPTS


3.1 Expert Systems


One of the most significant results of artificial intelligence (AI) research to date is the expert system (ES). An ES is a computer system that emulates the decision-making capability of a human expert. Feigenbaum (1982), an early pioneer of ES technology, defined an ES as

an intelligent computer program that uses knowledge and inference procedures to solve problems that are difficult enough to require significant
human expertise for their solution. (p. 15)

An ES attempts to perform like a human expert to solve problems. Instead of relying on statistical or algorithmic methods, ESs solve problems by applying a symbolic knowledge representation of human expertise. Consequently, ESs try to encode domain-specific knowledge rather than comparatively domain-free methods derived from computer science or a mathematical approach. Its application is normally restricted to a specific problem domain or well-defined domain. Applications of expert systems include medical, industrial, agricultural, and space technologies. ESs provide the advantages of increased availability, reliability, fast response, and multiple expertise. The structure of an ES can be a rule-base, frame, model or other approaches. The major components of an ES are the inference engine, knowledge base, user interface, and knowledge acquisition facility (Figure 3.1).

19








20




elopmnt TOW for ExpWt Sys

Ex rt Syst
Explanellon




En3.1.1 Inference Enine











The inference engine is the heart of an ES used for drawing conclusions based upon a knowledge base. Schalkoff (1990) stated that the inference engine can be considered to be a finite state machine with states representing typical actions such as (1) match rules, (2) select rules, (3) execute rules, and (4) check stopping conditions (i.e., goal satisfaction). Following are some commonly used reasoning approaches. Forward3.1.1 Inference ngine

Forward chaining is also called data driven. It is a reasoning method from facts (data) to conclusions. For instance, if a traffic light is green (fact), then one can drive through the traffic light (conclusion). Because of the nature of the forward reasoning process, this reasoning method is suitable for problem domains such as monitoring and realtime control systems, where data or facts are continuously acquired or updated.








21

Backward chaining

Backward chaining is a goal driven method. In contrast to forward chaining, backward chaining reverses the process. It reasons from a hypothesis, a potential goal to be proved, to the facts which support the hypothesis. This reasoning approach is more applicable to problems having many more inputs than possible conclusions, such as diagnosis and classification problems. The approach was used in Prolog and the medical expert system MYCIN.

Opportunistic chaining

Opportunistic chaining combines the forward and backward reasoning methods. For applications with many inputs and many possible conclusions, neither forward nor backward reasoning is an efficient approach. Thus, the two reasoning methods are applied together in one system to achieve efficiency. However, such a method may expand the difficulty of development.

Advanced reasoning methods

Advanced reasoning approaches (Gonzalez and Dankel, 1993) are model-based reasoning, qualitative reasoning, case-based reasoning, temporal reasoning, and artificial neural networks.

3.1.2 Knowledge Base

A knowledge base contains expert-level information required to solve problems in a specific domain. A knowledge base consist of a human expert's knowledge acquired by a knowledge engineer and encoded into the system. In other words, a knowledge base








22

contains the representation of domain specific knowledge. The essence of the knowledge base must fit the structure of the knowledge representation scheme.

The strength of an ES lies in the knowledge base because the knowledge base contains a representation of the human expertise of the problem-solving decision. Therefore, the whole process of knowledge acquisition is crucial in the system's development. The transferring of domain expertise to an ES's knowledge base has proved difficult and time consuming, in part because the process requires the interposition of a knowledge engineer between the human expert and a computer.

3.1.3 User Interface

Because ESs are generally interactive and involve users with little or no computer experience, the user interface should be designed to be friendly, explainable, and easy to use. A clear definition of the user interface requirements for an ES is essential to the success of the system. In particular, for users to accept the interface, it must accomplish the task in a straightforward way and still meet the entire range of problem solving requirements. To develop a friendly and explainable user interface, Ege and Stary (1992) suggested that the designers need to provide a global system perspective to create task-oriented, or usercentered user interfaces.


3.2 Real-Time Expert Systems


Historically, AI researchers have focused on problems in which the time response is not a concern, such as the medical diagnostic ES (i.e., MYCIN, Shortliffe and Buchanan, 1975). This kind of system is asking humans to supply necessary inputs, and the response








23

time can be slow or is not considered a major factor. This scenario is different from the environment where real-time expert systems (RTES) are used. In RTESs, data change rapidly and the input data are often collected automatically. RTESs typically need to respond to changing task environments, timely handling of data, and execution of diverse functions. This may involve an asynchronous flow of events and dynamically changing requirements with limitations on time, hardware, and other factors. Figure 3.2 shows additional components that may be required for an RTES. Sensors may be used to provide facts to the knowledge base, and external hardware can be controlled through an ES. Thus, the control process can be accomplished through conventional logic and procedures. Applications of RTESs in different areas have been reported by many researchers (Wright et al., 1986; Laffey et al., 1988; Nann et al., 1991; Ingrand et al., 1992).



Dsvelpment Tod for Expr Sypa n

Ept Syalem

- co i
Ene




UON Irdefte
Add New





User

Figure 3.2. Major components of an RTES.








24

3.2.1 Why Use an RTES?

ESs have great potential value as control devices for many applications. In this role, it is important to make provisions for an easy-to-use interface of sensor inputs, system outputs, and control actuators. In the real world, systems that are designed to control complex and dynamic processes (such as on-line monitoring) require fast handling of data and execute diverse functions. The control decisions, which may require deep knowledge or expertise, should be made based upon timely data. RTESs might be useful for domains where conventional ES approaches have failed or are impractical. These may include situations in which humans fail to effectively monitor data, make costly mistakes, miss optimizing opportunities, are unable to solve conflicting constraints, or suffer from cognitive over load. Turner (1986) pointed out that the main reason for using an RTES is to reduce the cognitive load on users to enable them to increase their productivity without the cognitive load on them increasing.

3.2.2 Characteristics of RTESs

Three factors are of main concern for RTESs. First, conclusions must be reached and actions must be taken in real-time to respond to the sensor's perceptions and environmental change. Second, the system must be able to provide tentative conclusions based on initial evidence if not all of the data are available at once. Third, the system must operate safely and reasonably on inaccurate and uncertainty of data input.

Laffey et al. (1988) perceived a series of characteristics of RTESs that differ from conventional ESs. Those major characteristics are as follows.








25

Non-monotonicity of data and uncertainty of missing data,

Continuous operation with high performance and guaranteed response times,

Ability to deal with asynchronous events,

Ability to communicate with the external environment (sensors and

effectors), and

Integration with procedural components.


3.3 Knowledge Acquisition


Knowledge acquisition is the transfer and transformation of problem-solving expertise from some knowledge source to a computer program (Buchanan and Shortliffe, 1984). This is a process of eliciting, structuring, and organizing knowledge from human experts or other sources so that the expertise can be encoded into an ES (Figure 3.3). The process E" consists of elicitation of the knowledge from the sources and representation of this knowledge. The goal of knowledge Encod
acquisition is to produce and verify the knowledge required by the system.
es
This process is the most important step, and normally, the most time-consuming Figure 3.3. Knowledge acquisition cycle.








26

phase in the development of a knowledge-based system. Researchers refer to this as the knowledge acquisition bottleneck (Feigenbaum, 1979) in the development of knowledgebased systems.

The success of current ES technology is highly related to the strict separation between a domain-dependent knowledge base and an inference engine. The power of a knowledge-based system relies more upon the quality of the knowledge base rather than the characteristics of the inference engine.

3.3.1 Basic Approaches

As Figure 3.3 shows, knowledge acquisition is mainly eliciting and organizing knowledge from human experts. The role of a knowledge engineer is to communicate the basis of the performance with the experts and to specify it in a form suitable for a computer. The basic approaches to knowledge acquisition can be summarized as

Interviews,

Questionnaires and observation of the expert at work,

Intuition, and

Using knowledge engineering facilitators and inductive tools.

Interviewing is one of the major approaches of knowledge acquisition. The interview can be structured or unstructured, and the communication can be one-to-one or many-to-one. Giarratano and Riley (1989) described the basic procedure of the knowledge acquisition and extraction task. First, an acquisition strategy should be decided. This includes specifying how knowledge will be acquired and the methods to be used in the interview. Second, the knowledge elements or specific knowledge that could be used by the








27

system must be identified. Third, the information to aid the knowledge understanding and verification by the developer needs to be classified and organized. Fourth, the detailed functional capabilities of the system have to be laid out. Then, a description of the general functional capability must be given in detail. This includes studying the flow of knowledge from both the developer's and user's viewpoint. Fifth, the system task needs to be defined.

Questionnaires mean that the knowledge engineer prepares some question sheets and lets the expert answer them. This approach can be used in combination with the interview approach. Observation allows the knowledge engineer to learn how the expert solves a real problem.

Intuition refers to how a knowledge engineer attempts to be a pseudo-expert and applies his knowledge to the domain. This process can only serve as an aid to knowledge acquisition because the knowledge engineer is not a true expert and lacks expertise in the domain.

Tools for the processes of automated knowledge acquisition have been developed by researchers (Quinlan, 1986; Michalski et al., 1986). These tools attempt to help bridge the gap between the expert/knowledge engineer and computer implementations using learning algorithms. One of the earliest and best-known algorithms is ID3 (Quillan, 1986). However, these tools provide very limited capabilities in solving real-world problems.

3.3.2 Potential Problems

Interviews are the most common approach for knowledge acquisition, but this process is not simple because, in general, experts do not structure their decision-making in








28

any formal way; and they may have difficulty in explicitly describing their reasoning. These problems were summarized by Harandi and Lange (1990):

Vocabulary. Knowledge engineering is virtually impossible unless the

knowledge engineer has a basic understanding of the problem domain. An essential part of that understanding is familiarity with domain terminology.

Completeness. A knowledge engineer must be able to identify pieces of

information or knowledge that are missing from the knowledge base.

Integration. A knowledge engineer should find out how new information fits

into the current knowledge base because the new information could interact

with already available information in an undesirable way.

Analysis. Usually, experts have difficulty in explaining exactly how and

why they reach certain conclusions. Therefore, knowledge engineers may have to conduct an interview that may require substantial communication

skills.

3.3.3 Practical Issues

In the real world, the process of knowledge acquisition should consider many practical issues; and there is no standard approach to follow. The practical considerations of knowledge acquisition were discussed by Jones (1989b), and Gonzalez and Dankel (1993). For instance, how to find a "real" expert who is articulate and very knowledgeable in the problem domain, how to plan and to conduct an interview, and how to capture the detailed knowledge are problems that must be addressed.








29

As a result, knowledge acquisition is an art. Each problem may require specific acquisition strategies and could sometimes involve psychological issues. To reduce the errors caused by human intervention, more efficient and reliable approaches for acquiring knowledge are required. These should automate the elicitation process based on a representation scheme that will completely and efficiently denote all the domain traits and encompass all the essential knowledge.


3.4 Knowledge Representation


Knowledge representation consists of encoding real-world expert knowledge into a format both readable and understandable by a computer. Some way to represent knowledge is needed that allows the computer to derive new conclusions about its environment by manipulating the representation.

In the process of knowledge representation, the primary problem is to find a kind of format or knowledge representation language. Usually, knowledge representation is not straightforward. First, the knowledge engineer should understand the concepts required to solve the problem. Second, these concepts should be represented precisely and unambiguously at all granularity levels. Third, these concepts should be easy to understand and applicable to many systems. Ringland and Duce (1988) stated some issues that should be raised in knowledge representation:

Is the approach expressively adequate to the domain?

Is reasoning efficient enough to allow the inference to perform in an

acceptable time?








30

How do we construct meta-knowledge representation?

What are the primitives and how does one manage incomplete knowledge? The commonly applied knowledge representation methods in production systems are (1) semantic network, (2) frame, (3) objects, and (4) rules.

3.4.1 Semantic Network

The semantic network was initially applied by Quillian (1968) to analyze words and sentences. Since then, the approach has become widely used. A semantic net is a formal graphic language for representing facts about entities. Its structure is shown graphically in terms of nodes (objects) and the arcs (links) connecting them. The directed arcs that connect the nodes represent relationships between objects. A semantic net can virtually represent any relationship that holds among the objects or concepts in some domain of interest.

The graphical relational representation of a semantic net is explicit and succinct to the state of knowledge. In addition, because nodes are directly connected with related nodes, the search can be efficient. However, the nets offer no standard definition of link names among nodes; and there are some practical difficulties in performing computer reasoning with completely general semantic nets.

3.4.2 Frame

Frame-based representation was developed to manage information overload inherent in large semantic nets without sacrificing their expressive power (Minsky, 1975; Bobrow and Winograd, 1977; Fikes and Kehler, 1985; Brachman and Levesque, 1985). The basic characteristics of a frame are that it maintains the fundamental notions of abstraction hierarchies and inheritance of properties from superclasses, but it packages the descriptive








31

attributes associated with each class or instance into more compact local data structures. A frame is mainly a group of slots (attribute) andfillers (values) that define a stereotype of knowledge. The fillers can also be subdivided into facets. Each slot, filler, and facet has its own associated values.

Frame systems are very suitable for those well-defined features (stereotype knowledge), so that many of its slots have default values. Therefore, the main advantage of frame systems is that their knowledge representation is significantly better structured and organized than knowledge in semantic net systems. Furthermore, the system can only trigger specific actions through demons during the processing of information, instead of repeatedly testing a rule in a rule-based system. This will significantly increase the efficiency of knowledge processing. The frame system, however, may have difficulty dealing with heuristic knowledge and coping with a new situation beyond the default values.

3.4.3 Obiects

An object is a representation corresponding to a conceptual entity in the real world and how the information related to the entity is manipulated. The basic idea behind an object-oriented representation is that information should be clustered around the "object." The difference between frame and object representation is that an object-oriented approach creates a tight bond between the code and data instead of separating them into two complex, separate structures (Gonzalez and Dankel, 1993).

The characteristics of object-orientation are the levels of abstraction that can be achieved and its ability of encapsulation, inheritance, and polymorphism. The approach, however, possesses drawbacks similar to those of the frame method.








32

3.4.4 Rules

The classic and most common way of knowledge representation is the use of a generic form of IF-THEN rules. The rules may have the simple form: IF antecedent(s), THEN consequence(s).

Rules provide a readily understandable form. The antecedents define a pattern to be matched against the content of the working memory, which is a global database of facts used by the rules. The rule is fired if such a pattern is matched. Thus, the consequences change the working memory and play the inferences in an ES. Each rule is an independent unit in the entire knowledge base.

Rule-based systems have been widely applied in many areas and are often misunderstood because their IF-THEN structure is similar to the condition structure in conventional programming language. Two factors distinguish rules from a conditional statement in conventional languages (Ringland and Duce, 1988):

The antecedents are expressed as a pattern rather than a boolean expression;

and such antecedents can be in a simple or very complex form.

A rule-based system allows separation of knowledge from control of how the

knowledge is applied. The condition of conventional language is a flow of control. The flow in rules does not pass from one rule to the next in lexical sequence, but is determined completely separately through the inference

engine.

Rule-based systems are the most widely used production systems. All knowledge in the knowledge base is represented in a single uniform format, and each rule is a distinct








33

individual element of knowledge that can be updated independently of the other rules. The major drawback of the system is inefficiency due to the use of infinite chaining. Moreover, there may exist contradictory and inconsistent knowledge among the rules when new rules are added.

Knowledge representation is the foundation of AI. Many researchers (Bobrow and Winograd, 1977; Brachman and Levesque, 1985; Fox, 1990) have attempted to improve the ways of representing knowledge. Advanced knowledge representation approaches include the techniques of spatial, causal or temporal models, and neural nets. A general knowledge representation system cannot be constructed easily. The selection of knowledge representation methods will mainly depend on the inherent structure of the knowledge and what knowledge representation the expert system tool will support.


3.5 Rule-Based Expert Systems


Many ESs have been developed using rule-based structure or so called production systems. Rule-based ESs have been successfully applied in many domains, including medical diagnosis, mathematic discovery, and hardware configuration. Rule-based systems have been widely applied because the systems have advantages of modularity, uniformity, and naturalness (Gonzalez and Dankel, 1993). In addition, many development tools (MetaMYCIN, CLIPS, and LEVEL 5) with relatively low costs are available for development of rule-based systems.








34

3.5.1 Rule-Based Architectures

Rule-based systems or production systems have three main components: (1) working memory, (2) rule memory, and (3) inference engine. The architecture and execution cycle of rule-based systems is illustrated in Figure 3.4.

The working memory functions as a storage facility of these objects representing facts about the world. The rule memory contains rules or the knowledge base of the system. The inference engine is the active element in the system. It selects rules from the rule memory that matches the contents of the working memory and executes the associated actions. If a rule is matched with the content of the working memory, the rule is said to be fired. The conflict resolution strategy will affect system behavior (Gonzalez and Dankel, 1993). It should be chosen with care.




Sh a Infexrci Wancof Base Engine Mmory (Rul-) (AOen) (Fa )




Fa Fally


uer


Figure 3.4. The architecture and execution cycle of rule-based systems.








35

3.5.2 Uncertainty Management

The quality of an ES is related to the knowledge base. Each piece of knowledge acquired from experts may involve some kind of uncertainty. Bronowski (1965) stated that

in trying to formalize a rule, we look for truth, but what we find is
knowledge, and what we fail to find is certainty. (p. 32)

Uncertainty arises from a variety of sources: (1) unreliable information, (2) imprecise descriptive languages, (3) inferences with incomplete information, and (4) a poor combination of knowledge from different experts (Bonissone and Tong, 1985). Uncertainty may prevent a system from making the best decision and may even cause a bad decision to be made. The basic numeric approaches to deal with uncertainty are Bayesian probability, Dempster-Shafer theory (Dempster, 1967; Shafer, 1976), and certainty factors. Bayesian and Dempster-Shafer approaches were developed before expert systems became popular. Because their practical implementation is complex and the approach requires a prodigious amount of data that are usually not available, these approaches are not widely applied in the development of expert systems.

Shortliffe and Buchanan (1975) developed an uncertainty approach to represent uncertain information in MYCIN. A certainty factor (CF) is calculated from a measure of belief (MB) and a measure of disbelief (MD): CF MD 3-1
1 minfMB, MD]








36

The combined CF for the hypothesis is calculated:



CF1 + CF2(1 -CF,) bodi >0 CF, CF,
CF CF 2 ae<0 3-2
1 min(ICF1I, ICF2)

CF1 + CF2(1 + CFI) bod <0





One of the advantages the CF has in comparison with Bayesian theory is that the CF avoids the need to establish prior probability. Moreover, the CF represents and combines the effects of multiple sources of evidence in terms of joint beliefs or disbeliefs in each hypothesis. Consequently, the CF is an easy to apply and widely used approach for uncertainty management.













CHAPTER 4
SYSTEM SPECIFICATION AND DESIGN


4.1 Domain of the Problem


Citrus farm management practices include decisions regarding irrigation, fertigation, cold protection, and plant disease control. Management decisions are affected by many factors including soil, meteorological, and biological conditions. In Florida, microirrigation systems are common today in citrus irrigation. Because of the reduced area of coverage of microirrigation and the low water-holding-capacities of Florida's sandy soils, irrigation requires high frequency applications. Computer controlled irrigation systems offer considerable labor savings.

In Florida citrus groves, chemicals and cold protection are commonly applied through microirrigation systems. Fertigation application decisions require knowledge about plant nutrient demands and environmental impacts. For cold protection water must be applied in a timely and accurate manner to avoid severe economic damage. Farmers need to make daily management decisions to maximize their net income. An expert system that contains expert knowledge on irrigation, fertigation, and cold protection is desirable to improve citrus microirrigation management. Thus, the goal of this research was to develop a real-time expert system (RTES) to assist in the management of citrus irrigation, fertigation, and cold protection.

37








38

4.2 Requirements Specification


4.2.1 Goal of the System

The overall goal is to develop a computer-based tool integrated with expert knowledge as a decision aid for irrigation, fertigation, and cold protection of citrus. The system should have the ability in response to current weather and soil-water moisture changes to realize real-time performance. The system must also meet the following requirements:

Be easy to use,

Achieve irrigation system automation with minimum maintenance,

Handle missing data or unreliable sensor data,

Record each event conducted by the system,

Develop several control schemes to minimize hardware requirements, and

Be accessible through telecommunication systems.

4.2.2 System Inputs

Table 4.1 shows the basic input and output requirements of the system. Because soil water status is a crucial factor for crop growth, sensors are needed to monitor soil-water content in the crop root zone. Rainfall and evapotranspiration (ET) are two important parameters that determine crop water requirements. An automated weather station is essential to obtain weather data for real-time control. The weather data are also used for crop ET estimation. Crop, soil, and irrigation system data are required in the decision process. Crop data include effective root depth, percentage of crop land coverage, and age








39

of trees. Soil data include soil depth and soil-water-holding capacity. Microirrigation (micro-spray) is assumed to be used in this application. The following microirrigation system data are required:

Number of emitters used per tree,

Emitter flow rate,

Wetted diameter, and

Irrigation application efficiency.



Table 4.1 System input and output requirements. Inputs Outputs
Soil-water tension Turn on or off specified irrigation valves
Weather data Turn on or off fertigation pump
ET coefficients Apply pre-injection and flush for fertigation
Tree status (age of tree) Apply cold protection
Soil characteristics Simulate crop water requirement
Irrigation application rate Display sensor maintenance messages


4.2.3 System Outputs

Outputs of the system need to be specified to apply irrigation, fertigation, and cold protection. These outputs can also be a message displayed on the screen or a control signal sent to an external device. After the decision-making process is accomplished, the system control procedures activate or deactivate irrigation control valves and pumps. Thus, irrigation, fertigation, and cold protection are applied according to the decision of the expert








40

system. Because the real-time system operates continuously, the system should display all sensor readings and warning messages to request maintenance when a sensor failure is identified. In addition, each control action should be saved to a file for future reference.


4.3 Knowledge Specification


Expert knowledge is the heart and power of an expert system. Knowledge required for the system development, in general, involves citrus irrigation management, fertigation application, cold protection, and sensor behavior.

For irrigation management, irrigation strategies and criteria for turning on or off the irrigation system must be resolved. These criteria relate to the soil-water content in the crop root zone and weather conditions. Irrigation management decisions should maintain the soilwater content between certain levels in the crop root zone and should avoid applying too little or too much water.

Fertigation should be scheduled at the right time and be applied for the proper duration. Application of fertigation should be in a sequence of pre-injection, nutrient application, and flush period. Expert decisions on fertigation and irrigation are needed to achieve water savings and to avoid environmental pollution. For cold protection, factors that affect cold damage of citrus trees should be understood. These factors include the principle of cold damage, air temperature, and wind speed effect. In particular, critical air temperatures for application of cold protection need to be specified. Knowledge of when to start and when to stop an application need to be acquired for the reasoning process.








41

Sensor failure can be a main cause of making an unreliable or poor decision. Thus, safety measures or data validation should be used to evaluate the sensor data. One approach is to install a redundant sensor; then sensor readings from the redundant sensors can be evaluated to increase the data reliability. However, this approach increases the cost of system hardware. An alternative approach is to use data uncertainty analysis, such as the certainty factor (CF) discussed in Chapter 3. Sensor readings with low CFs should be discarded from the input of the decision process. In addition, a simulation approach can be used to verify the actual irrigation duration to avoid excess water application.


4.4 Knowledge Representation Paradigm


Commonly used knowledge representation paradigms are rules, logic, frame, objects, or semantic networks. The choice of the paradigm should be suitable to represent the domain knowledge and be in the consideration of the selection of development tools. Decision processes in citrus irrigation management are heuristic in nature. The knowledge required in this application can be considered shallow knowledge. Rule-based systems are the best currently available means for codifying the problem-solving knowledge of human experts (Hayes-Roth, 1985). Because of the heuristic nature of the decision-making process and available development tools, a rule-based knowledge representation paradigm is used in the system.

4.4.1 Reasoning Method

Since rules are selected as the knowledge representation paradigm, reasoning methods can be either forward, backward, or opportunistic (bi-directional). For this








42

application, many inputs are acquired automatically from sensors. The decision-making progress is from initial facts (data from the sensors), to intermediate facts, and finally to a conclusion. Thus, a development shell with forward reasoning should be selected.

4.4.2 System Performance Requirements

As an expert system which operates in the real-time domain, the system imports initial data (facts) from sensors and the sensor data are varied with time. The system must respond to this variation to realize real-time performance. In particular, the system should achieve the following performance:

Must be operated within a fixed time constraint,

Should have the ability to react to the changing external environment and

must be operated continuously and as information is updated,

Must deal with incomplete and faulty input data from external devices, and Must allow procedure calls to other systems to bring back the necessary

information for reasoning.


4.5 Development Tools


The software tools available for development of expert systems can be categorized into four classes: General-Purpose Programming Language (GPPL), General-Purpose Representation Language (GPRL), expert-system building frameworks (shells), and expert system development environments (Collins et al., 1990).

The GPPL includes computer languages such as C, Lisp, and Prolog. The GPPL languages have high programming flexibility, but the development time and the cost may








43

be higher than with other developing tools because the development needs to be started from scratch in many cases.

The GPRL are the languages which are especially written for expert system development, such as OPS5 and UNITS. These languages have relatively higher flexibility in comparison with shells and development environments. Expert system development environments (such as KEE, LEVEL5, and ART) are complete development environments which provide sophisticated features such as knowledge representation, debugging routines, and development facilities. Although these development environments provide many features, the use of these software packages may require a substantial amount of time to learn the package to take the advantage of their features. Also, the cost of the packages is high.

4.5.1 Expert System Shells

The introduction of expert system shells has made expert system development much faster and attracted more new developers. A range of expert system shells have been developed to match the cost and application requirements. Some of these shells are CLIPS, EXSYS, and VP-EXPERT. The shells have the advantages that they are low cost, easy to learn, and save development time, but their flexibility and capability are the major constraints for development of expert systems.

Choosing the right tool for implementing a particular application is difficult because there is almost no absolutely one 'right' choice. One needs to ponder the advantages of any selection against its limitations to determine the most suitable tool for a particular problem. If a shell exists and satisfies the application requirements, the shell may be the better choice








44

than an AI language. Barrett and Beerel (1988) stated "use a shell if you can, an environment where you should, and an AI language when you must (p. 69)." This study selected an expert system shell, C Language Integrated Production System (CLIPS), as the development tool.

4.5.2 CLIPS

CLIPS, developed by NASA's Johnson Space Center, is a forwarding chaining rulebased language which uses the Rete Algorithm for pattern matching. CLIPS can be used as an embedded application or child process. The tool can easily call an external executable routine. Because a real-time system requires access to external devices, these features are important for this application. CLIPS is delivered with a complete source code so that the user can modify and re-compile the program for special purposes.


4.6 Hardware Specification


Agricultural decisions are highly related to climate data such as rainfall and air temperature. Because the system is designed for real-time performance, hardware is required to acquire external data and to realize control actions. Sensors are needed for measuring climate parameters and soil-water content. The following hardware is necessary:

Soil moisture sensor,

Personal computer,

Automated weather station,

Data logger,








45

PC digital input/output board, and

Irrigation control system.

4.6.1 Soil Moisture Sensor

Numerous soil moisture sensors are available commercially. The types include electrical, magnetic, nuclear, optical, and tensiometric sensors. A literature review of soil moisture sensors was conducted by researchers (Schmugge et al., 1980; Mckim et al., 1980; Zazueta and Xin, 1992). Among the many available soil moisture sensors, the tensiometer is one of the most widely used sensors mainly because of its performance and its low cost. In this study, tensiometers are used to measure soil-water potential at the crop root zone. Tensiometers

Figure 4.1 shows a vacuum gauge tensiometer and a tensiometer with a micropressure transducer. A tensiometer consists of a ceramic porous cup, plastic body tube, a gauge, and a service cap. The tube is filled with water. In the field the ceramic cup is installed in the active root zone of the soil. As the soil dries, water in the tube is pulled through the ceramic cup and the tension is displayed on the vacuum gauge or the pressure transducer. This tension is equivalent to the soil-water tension when equilibrium is reached. The tensiometer, in this way, measures the force exerted by the soil to extract water from the ceramic cup. Thus, the tensiometer measures the soil-water potential or the energy status of water for a plant rather than the quantity of water in the soil. The range of soil-water potential that tensiometers measure is from 0 to about -0.8 bar.








46












Naft bod MW




Cwmh pma up


Reulr Tweimr Twsmwr wl Prmu Ser Figure 4. 1. A regular tensiometer and a tensiometer with micro-pressure transducer. Pressure transducer

Tensiometers continuously measure the soil-water tension. The To tenmtmr
tension must be converted into an electrical signal to output to a computer. Tensiometers with micropressure transducers have been constructed at the Soil and Water sup'py Gmund
Hydraulics Laboratory, Agricultural and Biological Engineering Figure 4.2. Pressure transducer (Model Department, University of Florida 141PC) from Micro Switch.








47

(Smajstrla, Personal communication). A Micro Switch' model 141PC pressure transducer was used in this research. Figure 4.2 illustrates the pressure sensor. Junction P1 is the reference pressure input, which is air pressure for this application. Junction P2 is connected to the tensiometer to measure soil-water tension or gauge pressure. The pressure transducer outputs an electric voltage signal in response to its resistance change caused by variation of the input pressure. Output voltage of the sensor is proportional to the input pressure signal. Table 4.2 lists the major characteristics of the sensor.



Table 4.2 Characteristics of pressure transducer Model 141PC.


Parameter Minimum Typical Maximum
Output 4.85 V 5.0 V 5.15 V + 2.5 V
Excitation 7 VDC 8 VDC 16 VDC
Input Pressure 15 psi 15 psi
Operation temperature 40"C to + 85"C
Size 2.35 x 1.18 x 0.75 inches


Calibration of tensiometers

Before tensiometers are installed in the field, a calibration test is needed to determine the relationship of soil-water tension and output voltage from the pressure sensor. Figure 4.3 shows the calibration equipment designed at the Soil and Water Hydraulics Laboratory,



'The manufacturers listed in this dissertation are for illustration only. No endorsement of these companies or their products is implied by the authors or the University of Florida.











48







BBBS88888 oI88.8888











Pump


Watr L"l Cago Chmber Figure 4.3. Tensiometer calibration equipment.























10
50



4030



20-



10




-2500 -2000 -1-1 1000 -500 0 50 1000 Prmsu Seom Oulpe (nm)



Figure 4.4. Tensiometer calibration curve.








49

Agricultural and Biological Engineering Department, University of Florida (Smajstrla, Personal communication). Tensiometers are placed in a chamber which is partially filled with water and sealed. A vacuum pump is used to alter the pressure inside the chamber. Outputs (voltages) from the pressure sensors vary with the chamber pressure (tension), as indicated by a pressure gauge. The sensor outputs can be recorded through a data logger to a computer or one can simply use a multimeter. The relationship between the tension and the sensor output voltage is linear. Figure 4.4 is one of the typical calibration curves. Soilwater tension can be calculated from the pressure sensor output based upon this calibration curve.

4.6.2 Personal Computer

An IBM or any compatible PC with 386 or higher CPU is recommended to run the expert system. Four MB of RAM and five megabytes free space on a hard disk are required. The PC must have at least one communication port.

4.6.3 Automated Weather Station

The weather station used in this project was a Weather 2000 system from Campbell Scientific, Inc. This station is an automated system designed for commercial, agricultural, and irrigation scheduling applications. This weather station measures meteorological conditions that affect crop water consumption. The station provides the following data: rainfall, air temperature, solar radiation, wind speed, relative humidity, and soil temperature.

4.6.4 Data Loger

The Campbell Scientific data logger (CR10) is a fully programmable module, which provides sensor measurement, time keeping, communications, data reduction, data/program








50

storage, and control functions. A multitasking operation system allows simultaneous communication and measurement functions.

The device is protected in a sealed, rugged, stainless steel canister in order to be installed for out-door conditions. The input signal can be either analog or digital. The maximum analog input ranges from -2.5 V to +2.5 V. The interface to the CR10 can be a portable CR10OKD Keyboard Display or a computer. The PC communication software to access data from the CRIO is supplied by the vendor.

4.6.5 PC Digital Input/Output Board

A digital input/output (I/O) board was used as the interface between the computer and the irrigation control board. The PC-DIO072, a general digital I/O card from the Industrial Computer Source, was used for this application. The card can be applied to relay monitoring, control, sensing switches, security systems, and energy management. This board provides user selectable buffered inputs and outputs based on the 8255 chips by Intel. Major features of the PC-DIO are (1) 72 channels of digital I/O, (2) interrupt and interrupt disable capability, and (3) four or eight bit groups independently selectable for I/O. The output source current (output high) is 15 mA. The base address used in this application is H310, H314, and H318 to set the ports as output.

4.6.6 Irrigation Control Board

An irrigation control board was used to activate or deactivate the automated irrigation valves. The control board consists of 70 solenoid relays OACQ5 (Figure 4.5). The relays are installed on a relay rack (PB24Q). Both the relay and the relay rack are manufactured by Opto22. Each relay controls a specific electronic valve of the irrigation








51








Comp PC-DI2 Nolr

o cabl CAB 50-10 One trnOWrfeed ioehivu oj -a~dl.owwuv 24VNAc Iay Ru c P24Q (18CQIneleled
SOL 8 Rol. OCQ5 SOL Sotd ROL





ist is s eit i I a Ionce



10ASw BloFuse

115V







Figure 4.5. Solenoid control relays of the irrigation system.



system. In addition to the automated relay control, the control board provides manual switches and timers to turn on or off irrigation valves in case the relay system fails. In this design, valve 64 is the default for the fertigation pump, and valve 40 is the master valve.








52

4.6.7 Overview of the Hardware

The selected hardware performs data collection and controls the irrigation system. The system is an on-line irrigation controller (Figure 4.6). Soil moisture sensors (tensiometers) measure the soil-water potential. An automated weather station measures meteorological data, which provides current rainfall and sufficient data for ET estimation. Sensor readings and weather data are stored in the data logger (CRIO). The computer retrieves the data from the data logger at each given time interval. Connection between the computer and the data logger can be either wire or radio links depending on the distance and hardware cost. After the sensor data are input to the computer, the expert system uses the







Daogge
station---= n DO





/




Irgation\


Tenslonter



Figure 4.6. Hardware layout of the control system.








53

data as initial facts to conduct its reasoning process. Then, the irrigation control valves can be turned on or off through the irrigation control board according to the results of the reasoning process.

4.7 Paradigm for the Real-Time Expert System

Since CLIPS provides a built-in inference engine for forward reasoning, the development is focused mainly on the formation of the knowledge base (KB), external control procedures, and the user interface. Figure 4.7 shows the structure of the system. Input data from the sensors are collected and stored by external devices. Users can select the frequency of downloading data from the data logger to the CLIPS fact base (FB). Before the data are transferred to FB, a data pre-process procedure is needed to reorganize the data format into the CLIPS data format.




Daa CmWmn n









nContr o Evet
-- v






Figure 4.7. Paradigm of the real-time expert system.








54

The KB was developed using the CLIPS language which contains expert knowledge represented in rules, logical analysis, uncertainty management, and calls to external procedures. After the knowledge is implemented into the rule base, it needs to be verified by an domain expert. The KB development is one of the main tasks for the development.

External control procedures were designed to activate or deactivate the irrigation and fertigation devices. Irrigation control procedure can turn on or off user specified irrigation valves for a desired water application. Thus, irrigation can be applied block by block for different irrigation management strategies.

Fertigation control procedure was implemented with the capability of turning on or off both the injection pump and the irrigation valves. Cold protection control procedure is similar to the irrigation control procedure in that they turn the irrigation valves on or off. However, all the electronic valves should be turned on or off at once to cover the entire field for a cold protection application.

These control actions are accomplished by the results of the reasoning process of the KB. Each control action is displayed on the screen so that the user can view the current event. In addition, all control events, including application types and duration, are saved in a control event data log file.













CHAPTER 5
PROBABILITY OF RAINFALL

5.1 Introduction


Many agricultural operations and activities are affected by rainfall frequency and amount. To improve water use, an irrigator must consider probable occurrence of rainfall. Ideally, irrigation should be managed to maximize effective rainfall while satisfying crop water demands. Although rainfall cannot be predicted with certainty, estimated probability of rainfall is useful for irrigation management.

Rainfall probability is commonly predicated through weather forecasting according to meteorological observations. Stochastic modeling is another available approach to generate daily weather data from the use of observed weather data. Estimation of daily or seasonal rainfall sequences can be obtained by examining past precipitation records. The rainfall sequence can be estimated by using rainfall occurrence models (Schmidt, 1992): (1) alternating wet and dry interval models, (2) wet and dry day models (Markov-chain models), and (3) point process models. Such prediction is based upon the assumptions that sequences will tend to be the same in the future as they were during the period of record.

Markov chain models are widely used because of their simplicity, flexibility, seasonality, and number of states. Markov chain probability models for daily precipitation occurrences have been studied extensively. This approach has been implemented with


55








56

success for various locations (Gabriel and Neumann, 1962; Jones et al, 1972; Todorovic and Woolhiser, 1975). The rainfall probabilities according to wet-dry day sequences have also been applied for irrigation management (Safley et al., 1974). A Markov chain rainfall probability model was used to estimate the rainfall occurrence to assist irrigation scheduling in this study. This model is not a physical explanation of rainfall occurrence, amount of precipitation, or other meteorological observations, but merely a statistical description of the past observed behavior. The purpose of using the Markov chain probability of rainfall was to couple the rainfall probability with the irrigation decision-making process.


5.2 Markov Chain


A Markov chain is one particular type of stochastic process. Feller (1969) defined a Markov chain as

a stochastic process in which the future development
depends only on the present state, but not the past history
of the process or the manner in which the present state
was reached. (p.444)

A stochastic model, in general, provides only the probability associated with a set of possible future outcomes. Thus, a state X is followed by state Y with probability p, and by state Z with probability q = 1 p, where X and Y are the only possible occurrences. The Markov approach can be applied to wet-dry day sequences. Let Ci denote the occurrence of a wet or dry day.


1, if day I is wet 5-1 .Ifday Iis dry








57

Let N, denote the number of wet days in the n day period.




N. CI 5-2



The possible values of the random value Nw are 0, 1, ..., n.

Gabriel and Neumann (1962) studied a first-order Markov chain model for daily rainfall occurrence in Tel Aviv, Israel. Their assumption was that the probability of rainfall on any day depends only on whether the previous day was wet or dry. Such a probability model is a Markov chain with two conditional probabilities:



p, = P, (wet day I previous day wet)

po = P, {wet day I previous day dry)



Although this model obtained satisfactory results in Tel Aviv (Gabriel and Neumann, 1962) and other regions, previous studies (Schmidt et al., 1987; Jones and Thornton, 1993) showed that using a first-order Markov chain to estimate rainfall probability may not be adequate for a subtropical or tropical region such as Florida. Their studies suggested that a higher order Markov chain should be used for tropical and subtropical weather conditions. Jones and Thornton (1993) have applied a third-order Markov chain for tropical and subtropical regions. For a third-order Markov chain model, the probability of a rainfall event on any given day is assumed only depending on the states of the three previous days.








58

Thus, a Markov chain of wet day probability with order 3 can be formed with the conditional probability:



P({WID1D2D3} = P{W= XtIXt-I= xt-., Xt-2= Xt-2, X.3= X.3} 5-3



where W = wet day,

D, X = daily sequence of rainfall event, and

x, = random variable of a wet and dry day at day t.

The occurrence of the previous three consecutive days (DID2D3) could be {000 001 010 011 100 101 110 111}, where 0 represents a dry day and 1 represents a wet day. This third-order Markov chain was used to describe rainfall occurrence only. The amount and intensity of rainfall is not described by this equation. The probability of a wet day in the immediate future is based on past long-term weather data rather than the changing meteorological conditions.

5.3 Rainfall Data


Forty years of daily rainfall data from 1952 to 1992 were used to study the sequence of wet and dry days. This weather station is located in Orlando, Florida (latitude 28:27:00, longitude 81:19:00). These data were obtained from the NOAA weather station and the EarthInfo CD-ROM disk (NOAA, 1952-1992).








59

5.4 Frequency of Rainfall


A third-order Markov chain analysis was conducted using the forty-year rainfall data in Orlando, Florida. The prediction derived here is a long term estimation of rainfall occurrence. The probability of rainfall occurrence of each day is considered only dependent upon the wet-dry sequence of the three previous days. Each day can be either wet or dry. Days with a trace of rainfall (less than 0.01 inch) are considered to be a dry day. Table 5.1 shows the wet-day probabilities.

Table 5.1 Markov chain wet-day frequency. Previous Month of year Case Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec 000 0.26 0.21 0.23 0.15 0.20 0.38 0.45 0.39 0.33 0.19 0.18 0.22
001 0.20 0.30 0.29 0.19 0.22 0.33 0.40 0.53 0.37 0.27 0.20 0.20
010 0.22 0.29 0.22 0.21 0.35 0.45 0.53 0.47 0.46 0.22 0.18 0.20
011 0.23 0.20 0.15 0.22 0.35 0.37 0.51 0.50 0.45 0.24 0.10 0.17
100 0.40 0.54 0.43 0.39 0.50 0.65 0.59 0.70 0.54 0.56 0.44 0.35
101 0.44 0.37 0.49 0.47 0.52 0.62 0.65 0.63 0.69 0.59 0.44 0.53
110 0.36 0.41 0.47 0.45 0.56 0.70 0.70 0.70 0.70 0.59 0.51 0.37
111 0.35 0.46 0.45 0.47 0.57 0.70 0.75 0.69 0.65 0.61 0.39 0.43

Rainfall data is for Orlando, Florida from 1952 to 1990. 0-Dry day
1-Wet day

This probability indicates the likelihood or frequency of rainfall during a particular month and previous wet-dry day sequence. As the results show, the rainfall frequency ranges from 10 percent to 70 percent. Summer and fall have a higher wet day frequency than winter and spring. Higher probability of rainfall during June through September is due to the higher rainfall occurrence during these months. Figure 5.1 shows the annual rainfall distribution in Orlando over forty-year average data. Approximately 70 percent of the annual rainfall occurs during the summer and fall seasons. Thus, higher wet-day frequencies (> 0.6 from








60

Table 5.1) occurred in the summer and early fall. In contrast, lower wet-day frequencies occurred during the dry season, as expected.




16

14ure 5.1. Annual rainfall distribution in Orlando from 1952 to 1992.

12


















Figure 5.1. Annual rainfall distribution in Orlando from 1952 to 1992.



5.5 Statistical Test


The rainfall probability has shown variation among months and seasons. A statistical analysis can be used to evaluate data significance among them. A paired t-test can be used to test whether there are significant differences among the monthly or seasonal rainfall








61

probabilities. The one sample t statistic (Moore and McCabe, 1989) has the t distribution with n- degrees of freedom:


t = 5-4




Within each season, a paired t-test between monthly rainfall probability was conducted. Subjects were matched in pairs, and the outcomes were compared within each matched pair. The results of the paired t-test are shown in Table 5.2. In winter and spring, there are no significant differences in rainfall probability among the months at the 95 percent confidence level. The same test showed a significant difference in rainfall probabilities for summer and fall at the 95 percent confidence level. Because wet-day frequencies during winter and spring showed no significant differences, the average value of the seasonal rainfall probability can be used to represent the seasonal rainfall probability. However, probabilities of rainfall during summer and fall must be treated as monthly because they are significantly different among months within the seasons.

Table 5.2 Results of paired t-test for rainfall probabilities within each season. Winter Spring Summer Fall Jan vs Nov NS Feb vs Mar NS May vs Jun S Aug vs Sep S* Jan vs Dec NS Feb vs Apr NS May vs Jul S Aug vs Oct S Nov vs Dec NS Mar vs Apr NS Jun vs Jul S Sep vs Oct S NS = not significant different at 95% confidence level. S = significant different at 95% confidence level. S* = significant different at 90% confidence level.








62


5.6 Irrigation Decision with Rainfall Probability


One of the irrigation management goals is to maximize the use of effective rainfall. For an automated irrigation management system, obtaining real-time rainfall data is important to achieve this goal. The results of the Markov chain wet-dry day probability of rainfall were integrated into the knowledge base to aid irrigation decision-making. Because an automated weather station was installed in the field, daily rainfall for the past three consecutive days can be used to estimate today's rainfall probability based on the Markov chain probability of rainfall. The irrigation decision, then, is coupled with probability levels of rainfall. Thus, irrigation may be delayed or less water may be applied if a high probability of rainfall occurs and an irrigation is required. A practical issue related to irrigation scheduling using the wet day frequencies is what value of wet-day frequency should be considered a threshold for high probability of rainfall occurrence. This threshold is a critical value that can affect an irrigation decision. When 60 percent is considered the high rainfall probability, only summer and fall could have the possible values that are larger than 0.6. For mature citrus trees, summer and fall are not critical growth stages. This implies that it may be feasible to maintain the soil moisture at a lower level during these seasons without causing yield loss. To maximize utilization of effective rainfall and to reduce cost, irrigation may be delayed or less water may be applied when rainfall probability is greater than the threshold value.

In addition to the stochastic model, another way to obtain rainfall probability is to directly access a weather forecasting database. If the user can obtain a short-term weather








63

forecasting data, such as probability of rainfall for the next few days, those data can be input to the system to assist the irrigation decision-making process. Because computer networks are widely used nowadays, the system can link to a network to retrieve weather forecasts and historical weather data. The computer network at the Institute of Agricultural and Food Science (IFAS), University of Florida, supports a weather forecasting database.













CHAPTER 6
CITRUS IRRIGATION SCHEDULING


6.1 Introduction


Irrigation scheduling is important to maintain adequate soil-water content for high productivity and the resulting economic benefits. Studies have shown that citrus irrigation can increase fruit production (Myers and Harrison, 1978; Koo and Smajstrla, 1984; Smajstrla and Koo, 1984; Adams, 1992). Irrigation scheduling involves decisions on when to irrigate and how much water to apply. Irrigation scheduling methods can be based on (1) soil properties, (2) plant properties, or (3) a soil-water balance modeling approaches. Each method has advantages and disadvantages. In this study, soil properties and a soil-water balance model were used.


6.2 Citrus Water Requirements


Citrus water use involves a process of soil-water extraction by the roots and transpiration from leaves. Irrigation should provide water to crops to meet the evapotranspiration (ET) demand imposed by climate. Citrus water requirements have been studied by researchers (Rogers and Tucker, 1978; SCS, 1982; Smajstrla et al., 1986). Table 6.1 shows the citrus water requirements for central Florida (SCS, 1982). The estimated citrus annual ET is 39.74 inches in central Florida. Similar results have been obtained by 64








65



Table 6.1 Citrus irrigation water requirements in central Florida.

ET Normal year Dry year

Month (in/month) ER NIR ER NIR Jan 1.68 1.03 0.65 0.88 0.80 Feb 1.75 1.26 0.49 1.08 0.67 Mar 2.54 1.72 0.82 1.48 1.06 Apr 3.33 1.42 1.91 1.21 2.12 May 4.29 1.68 2.61 1.44 2.85 Jun 4.84 3.42 1.42 2.93 1.91 Jul 5.11 4.01 1.10 3.45 1.66 Aug 4.88 3.66 1.22 3.14 1.74 Sep 4.16 3.16 1.00 2.71 1.45 Oct 3.24 2.03 1.21 1.74 1.50 Nov 2.19 0.89 1.30 0.77 1.42 Dec 1.73 0.99 0.74 0.85 0.88 Total 39.74 25.27 14.27 21.68 18.06 Note: NIR = net irrigation requirement (inches),
ER = effective rainfall (inches), and
ET = evapotranspiration.
Source: SCS (1982) (p.4-30)








66

several researchers (Gerber et al., 1973; Reitz et al., 1977). They estimated that annual ET for citrus is about 48 inches in Florida. Monthly mean ET rates vary from a low of 0.08 inches per day in the winter to a peak of 0.17 to 0.2 inches per day in the summer (Tucker, 1983). In Florida, it was reported that annual water use is about 47.6 inches for ridge citrus (Koo, 1963), and 44.6 inches for flatwood citrus (Rogers et al., 1987). Citrus irrigation requirements have been recommended by researchers based upon crop ET requirement and effective rainfall.

Citrus irrigation requirements are different for young trees and mature trees. Young trees are usually managed to grow as quickly as possible for early production. Moreover, young trees are less able to resist water stress than mature trees. Therefore, adequate irrigation is especially important for young trees.

For mature trees, irrigation management should be different for critical and noncritical growth stages. The critical growth stage for mature trees refers to the months of leaf expansion, bloom, fruit set, and fruit enlargement. This occurs mostly during the spring months. Irrigation during this stage is very important to both fruit quality and yield. The spring in Florida usually has the lowest rainfall, thus the greatest moisture stress occurs. A sufficient amount of water is essential for mature trees during the critical growth stage. Sound irrigation practices should be emphasized during this critical state (Tucker, 1983). The remaining months are considered to be a non-critical growth stage. Irrigation application during the non-critical stage should be considered only when tree stress is imminent.








67

Estimations of citrus water requirements provide only a general guideline for irrigation. Actual irrigation applications vary due to several factors including (1) irrigation management strategies, (2) irrigation system, (3) variability of rainfall and other climatic factors, (4) soil characteristics, (5) planting density, and (6) crop growth characteristics. Because of the variability inherent in these factors, it is difficult to create a general irrigation schedule. Therefore, field measurement of soil-water content or maintaining a soil-water budget is useful to determine crop water requirements.


6.3 Evapotranspiration and Management Allowed Depletion


Knowledge of ET is important to the management and design of irrigation systems. Actual crop ET is determined from reference ET and experimentally obtained crop coefficients:


ETa Kc ETo 6-1


where K = crop coefficient,

ET. = actual ET, in/day, and

ETo = reference ET, in/day.

Microirrigation systems supply water only to the immediate vicinity of each plant being irrigated. Tree canopies shade only a portion of the soil surface area and intercept only a portion of the incoming radiation. Conventional estimation of water requirements assumes that part of the applied water will be lost to non-beneficial consumptive use, which is the loss from evaporation of wetted soil surfaces and plant transpiration from undesirable








68

vegetation. Consequently, conventional estimation of consumptive use, which assumes wetting the entire field surface, needs to be modified for microirrigation.

The transpiration rate under microirrigation is a function of the conventionally computed consumptive use rate and the extent of the plant canopy (Sharples et al., 1985). Keller and Biliesner (1990) used a simple equation for estimating the average daily transpiration rate:


ETm ET [0.1 (P)o0s] 6-2 where ETm = average daily transpiration rate for a crop under microirrigation, in/day,

ET. = conventionally estimated average daily consumptive use, in/day, and

Pd = percentage of soil surface area shaded by crop canopies at midday (solar noon), %.

In Florida, citrus crop coefficients (Rogers et al., 1983), Kc, with grass coverage and citrus irrigation Management Allowed Depletion (Koo, 1963), MAD, are given in Table 6.2. MAD is used to express the amount of water that can be depleted in the crop root zone without adversely affecting the plant.

Table 6.2 Citrus crop coefficients and recommended MAD in Florida.

Month Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Kc 0.90 0.90 0.90 0.90 0.95 1.0 1.0 1.0 1.0 1.0 1.0 1.0
MAD (%) 67 67 33 33 33 33 67 67 67 67 67 67








69

Irrigation decision-making relies highly on the skill of the irrigation manager. Irrigation should be applied when the soil-water content reaches the MAD. SCS (1982) recommended that available soil-water depletion should not exceed 30 percent between fruit set (February to March) and the period when young fruit has reached more than I-inch in diameter (June to July). During remaining months of the year, soil-water depletion should not exceed 50 percent. Studies in Florida (Koo, 1963; Gerber et al., 1973; Smajstrla et al., 1987) suggested that mature trees should be irrigated when one-third (33 percent) of the soil moisture in the root zone is depleted in spring or two-thirds (66 percent) is depleted during the rest of year.

As a rule of thumb, MAD within the root zone should not fall below 50 percent of the total available water-holding capacity (Keller and Bliesner, 1990). For young trees and mature trees during critical growth stages, irrigation should be applied when 25 to 35 percent of depletion has occurred.


6.4 Irrigation Depth and Duration


Irrigation scheduling requires making decisions on irrigation duration and frequency to meet crop ET demands. The amount of water and application frequency are related to water management, soil property, and economic considerations. Irrigation duration is associated with the application rate. For young trees, irrigation duration should average between one and three hours for microsprinklers with flow rates of 10 to 15 gallons per hour, and three to six hours for most drip systems depending on soil type, frequency of irrigation, and number of emitters (Davies et al., 1989).








70

The maximum net depth per irrigation (Dx) should replace the soil moisture deficit. The net irrigation depth for microirrigation can be described as the following equation (Keller and Bliesner, 1990): MAAD
D W, Z 6-3 100


where MAD = management allowed depletion, %

W, = available water-holding capacity of the soil, in/ft, and

Z = plant root zone, ft.

For microsprinkler irrigation, irrigation time to bring root zone to field capacity can be expressed as (Parsons et al., 1993) AW Df Z
Ig 6-4 Pr

where Id = irrigation duration (hour),

AW = available soil-water content (in/ft),

Df = depletion of AW prior to irrigation (%),

Z = root depth (ft),

P, = precipitation rate (in/hr), and


2.04 Fr N Ef
P, 6-5 d2

where Fr = flow rate of emitter (gal/hr),

N = number of emitters per tree,








71

E, = overall efficiency of the irrigation system (%), and

d = diameter of spray pattern (ft).

Because of the low water-holding capacity of Florida sandy soils, irrigation applications need to be more frequent to maintain the proper range of soil-water contents. With a computer controlled microirrigation system, water can be applied frequently in a timely manner to maintain less variation of soil-water content and without increasing application costs.


6.5 Soil-Water Budget


A soil-water budget is commonly used to describe the amount of available water in a crop root zone. For irrigation scheduling, it is convenient to calculate the amount of water used (depletion) in the root zone instead of estimating the remaining water. The water balance equation can be expressed as E T, *R-ER -NI
Ot 0a 6-6
D

where Oi+1 0i = soil profile depletions at the end and beginning of a period,

ER = effective rainfall (in),

ET, = actual daily evapotranspiration (in/day),

D = crop effective root depth (in),

NI = net irrigation (in), and

R = runoff from surface and deep percolation (in/day).








72

Soil-water depletion at any time is the amount of water needed to irrigate the current \ crop root zone to field capacity s

(FC). Normally, irrigation is ITtrpin Evwts
applied when the depletion exceeds MAD or when the Tkne
managed soil-water content is Figure 6.1. Irrigation by threshold of soil-water content.
less than a threshold value (SWCo). Thus, soil-water content is maintained at a certain level (Figure 6.1). Because each irrigation raises soil-water content from SWCo to FC, the amount of applied water is the same for all irrigations. However, the application frequency varies between irrigation events. On the other hand, irrigation can also be applied at a fixed interval (every day or two days), but irrigating with different amounts of water each time.


6.6 Irrigation Scheduling Using Tensiometers


Researchers reported that tensiometers can be used effectively to schedule irrigation by measuring soil-water potential, but proper installation and maintenance are required for the application (Smajstrla et al., 1985a; Smajstrla and Koo, 1986; Fitzsimmons and Young, 1972; Creighton et al., 1989). Two practical issues should be resolved in using tensiometers for citrus irrigation scheduling:







73

Depth and number of tensiometers that should be installed, and

Soil water tension at which irrigation should be initiated.

6.6.1 Tensiometer Installation Depth

Because tensiometers measure soil-water potential in only a small volume of soil immediately surrounding the ceramic cup, installation sites should be representative of the surrounding field conditions and water content in the effective root zone. Citrus root zone moisture extraction depths in unsaturated soils range from 3 to 5 feet and the minimum root zone moisture extraction depth required is 1.5 to 2 feet (SCS, 1982). The maximum effective rooting depths for citrus are 3.0 to 5.4 feet (Martin et al., 1990). In Florida, Tucker (1983) reported that citrus rooting depths extend to 5 feet for well-drained sandy soils; groves on flatwood soils rarely exceeded 2 feet in rooting depth. For citrus under microirrigation, the effective root zone should be defined as the upper 1.5 to 3 feet of the root zone for ridge citrus; and 1 to 1.5 feet for flatwoods citrus (Smajstrla et al., 1987).

How many tensiometers should be installed in the field is a compromise between cost and accuracy. One set of tensiometers, in general, is desirable for every five acres (Smajstrla et al., 1985b). At least two tensiometers should be used per location in order to check soilwater depletion in the effective root zone (Smajstrla et al., 1986). One tensiometer should be located near the soil surface (6 to 12 inches from soil surface) where most of the roots are located. The second one should be located near the bottom of the effective zone (24 to 36 inches from soil surface) that will be irrigated. Because the upper portion of the effective root zone contains the most roots actively involved in water uptake, it is important to








74

concentrate on both water applications and observations in this zone. In practice, tensiometers at three depths are desirable for deep rooting crops.

To account for sensor failures and different soil characteristics, tensiometers should be installed at several different sites adequately to represent the water status in large areas. How many and at what depth tensiometers should be installed at each site needs to be justified for each specific application.

6.6.2 Soil-Water Potential and Allowable Water Depletion

Irrigation scheduling is usually determined by allowable soil-water depletion. However, tensiometers measure soil-water potential. Hydraulic characteristics for the irrigated soil are needed to establish the relationship between the soil-water potential and the amount of soil-water depletion. Table 6.3 shows the soil-water tension versus the soil-water content for Candler fine sand at different soil depths (Carlisle et al., 1978).

Table 6.3 Average soil-water content for Candler fine sand by volume.

Depth Soil Water Temion, Cntibar
(ft) 0 2 3 4.5 6 8 10 15 20 33 1500 0 1.7 36.8 34.8 31.4 1.7 11.9 8.0 6.8 5.7 5.1 4.5 1.5
0 4 36.3 34.7 32.0 17.6 11.3 7.5 6.3 5.3 4.7 4.2 1.4
Source: Carlisle et al. (1978)
For Candler fine sand, assuming the permanent wilting point (PWP) is at -15 bars, the available water-holding capacity ranges from 7.5 to 9.5 percent by volume (Martin et al., 1990). These values approximately correspond to -8 and -7 cb according to Table 6.3. Thus, the relationship between soil-water potential and soil-water depletion can be approximately established. Table 6.4 shows the relationship of soil-water potential and soil-









75

water depletion for Candler fine sand. One-third (33 percent) of the depletion approximately corresponds to 11 cb, and 50 percent depletion approximately corresponds to 20 cb of soilwater tension (Figure 6.2). Then, irrigation can be applied when soil-water potential or soilwater depletion reaches a threshold value.




Table 6.4 Estimated soil-water tension in corresponding to soil-water depletion for
Candler fine sand.


Depletion, % AWC % SWC-7.5, % SWC-9.5, % ASWT, % ESWT, CB

100 0 1.40 1.40 1.40 -1500

90 10 2.21 2.01 2.11

80 20 3.02 2.62 2.82

70 30 3.83 3.23 3.53 67 33 4.07 3.41 3.74 -38 60 40 4.64 3.84 4.24 -33 50 50 5.45 4.45 4.95 -20 40 60 6.26 5.06 5.66 -13 33 67 6.83 5.49 6.16 -11 20 80 7.88 6.28 7.08 -9 10 90 8.69 6.89 7.79 -8 0 100 9.50 7.50 8.50 -7.5 Note: AWC = available water content,
SWC-7.5 = soil-water content for field capacity at 7.5 percent by volume, SWC-9.5 = soil-water content for field capacity at 9.5 percent by volume,
ASWC = average soil-water content of above two, and
ESWT = estimated soil-water tension.










76


















35I 30





j20


15 10 5


0
0 5 10 15 20 25 30 35 So Wter Tensa (cMner)


Figure 6.2. Soil-water content versus soil-water tension for Candler fine sand.













CHAPTER 7
CITRUS COLD PROTECTION AND FERTIGATION


7.21 Introduction


Cold protection refers to methods used to prevent cold damage to the crop. This term is typically used for (1) frost protection, (2) freeze protection, (3) frost/freeze protection, and (4) chilling protection (Barfield et al., 1990). Cold protection is always important to citrus production. Cold weather has caused severe economic damage in Florida's citrus industry in January, 1985 and February, 1989. Although several cold protection approaches are available, such as tree wraps, heaters, and wind machines, irrigation is the primary means of cold protection in Florida citrus. Microirrigation is a valuable tool for cold protection. Major cold protection in Florida (estimated over 100,000 acres of citrus) is accomplished with microsprinkler irrigation (Parsons et al., 1989; Parsons and Wheaton, 1990). Experience has indicated that micro spray jet systems are effective for cold protection (Harrison et al., 1987; Hardy, 1989; Parsons and Wheaton, 1990), particularly for young trees. Studies of cold protection methods with computer aided decision systems have been conducted by researchers (Holland, 1990; Heinemann et al., 1991, 1994; Martsolf et al., 1991). Their results showed that computerized systems could improve decision-making on cold protection. Irrigation equipment must be specifically designed for cold protection purposes. The irrigation 77








78

system must have a sufficient capacity so that the entire crop area being protected can be simultaneously watered to achieve adequate cold protection. With aid of an on-site weather station and a real-time expert control system, cold protection management can be implemented automatically. The computer can be used to turn on the irrigation system when critical environmental conditions occur.


7.2 Cold Protection Application


7.2.1 Principle of Cold Protection

Cold protection is based on thermodynamic principles, which have been discussed by Harrison et al. (1987), Martsolf (1990, 1992), and Barfield et al. (1990). A plant gains or looses heat from its surroundings through a heat transfer process. Heat transfer can occur as conduction and convention, evaporation and transpiration, and radiant energy exchange (Harrison et al., 1987). Irrigation water provides cold protection because the heat loss from the plant to its surroundings is replaced by the sensible heat and the heat of fusion of water. The latent heat of fusion is released when water changes from liquid to ice. The total latent heat input to the air, ignoring heat from the soil or atmosphere, is equal to the heat lost in cooling to wet bulb temperature, plus that lost as a portion of the drops freeze. The latent heat flux released from the water can be expressed as equation 7-1 (Barfield et al., 1990):

QL = 0.27 x 106 p, I [C (T, Twb)+ Yf Ff] 7-1 where QL = total latent heat flux in W m2,

Pw = density of water in kg m-3,

I = irrigation application rate in mm h-1,








79

C = specific heat of water in J kg1' "C-',

T,, = water temperature in *C,

Twb = the wet bulb temperature in *C,

yf = the latent heat of fusion in J kg-', and

Ff = the fraction of water that has become ice (fused) when it strikes the ground. Ff depends on drop size and environmental conditions. As equation 7-1 shows, cold protection is mainly related to (1) air temperature, (2) wind speed, which affects evaporation, and (3) irrigation application rate. In this application, it was assumed that there were adequate water and energy supplies and the irrigation system was properly designed for cold protection. This implies that the system can irrigate an entire citrus grove simultaneously with an adequate application rate.

7.2.2 Critical Application Temperature

Knowledge of weather data, particularly air temperature, is crucial for cold protection. Most citrus growers in Florida receive weather data from sources such as the National Weather Service, commercial radio and TV, and county extension offices. During the winter season, growers carefully track weather changes to make decisions related to cold protection.

Since air temperature is a crucial factor for cold protection, it is important to determine the critical air temperature to start cold protection. Applying cold protection too early or too late may result in water waste or crop damage. Harrison et al. (1987) reported that the freezing temperature for Tangelo is about 30.1 *F in Orlando, Florida. However, irrigation must be initiated before the temperature reaches freeze point because irrigation








80

pipe lines can be frozen at this temperature. A survey report (Ferguson et al., 1989) indicated that 45 percent of the growers initiated irrigation for cold protection at 33-35 "F and turned it off at 35-39 "F (43 percent). The Institute of Food and Agricultural Science (IFAS) at the University of Florida recommends initiating cold protection when air temperature reaches 360F and finishing when air temperature reaches 36 to 40 "F. Thus, air temperature at 360F was used as the critical application temperature in this system.

7.2.3 Water Application Rate

The water that is most effective for cold protection is that which covers the foliage, and not the ice formed on the ground (Harrison et al., 1987). Uniformity of application is very important to cold protection. Thus, uniformity design criteria of the irrigation system must be strictly met. The application rate must be high enough to provide sufficient energy to the system and guarantee plant coverage. A previous study (Parsons et al., 1989) showed 33.3 gal/acre/min (12 gal/hr/tree) was recommended for frost protection. Young trees may potentially use less water. Furthermore, in Florida, emitters should be installed properly. Parsons and Wheaton (1990) reported that emitters placed north or northwest (upwind) of the tree provided better cold protection.


7.3 Fertigation


Microirrigation offers the opportunity for precise application of fertilizer to the soil. Fertigation is the addition of soluble nutrients or agricultural chemicals through irrigation systems to crops. Fertilizer application through irrigation systems is desirable because of labor and energy savings, flexibility in timing of application, and easy and precise control








81

of application rate. Because of the high efficiency and centralized control of microirrigation systems, fertilizer placement through microirrigation systems can improve its efficiency of application (Keller and Bliesner, 1990). For this application it was assumed that the irrigation system was properly designed and it was adequate for cold protection and fertigation.

Fertigation management decisions are affected by available fertilizer concentrations, desired application rates, types of fertilizer, and the crop. Application of too little fertilizer may not obtain the desired results, and excessive applications of chemicals may result in unnecessary expenses and potential crop or environmental damage.

7.3.1 Application of Fertigation

Fifteen chemical elements have been found to be essential nutrients to satisfactory growth and functioning of citrus trees (Jackson, 1991). Among the fifteen chemical elements, three elements (carbon, hydrogen, and oxygen) are adequately provided in the environment suited to tree growth and are largely beyond the control of the grower. The other twelve are fertilizer elements or "plant food." The major chemical elements are nitrogen, phosphorus, and potassium. Numerous researches have been conducted to study fertigation application and fertigation effects on fruit quality and growth (Koo et al., 1984; Rolston, et al., 1986; Robinson,1990; Willis and Davies, 1991; Hearn, 1993; Boman, 1993). However, because of the complexity of the crop nutrient requirement, it is difficult to obtain a generally accepted fertigation schedule. A typical orange grove might require fertilizer in the following amounts (Jackson, 1991). This assumes a yield of 500 boxes per acre.








82

Nitrogen needed = 0.4 lbs. N/box x 500 boxes = 200 lbs.

Potash needed = 0.4 lbs. K20/box X 500 boxes = 200 lbs.

Other nutrients applied on basis of need.

Table 7.1 shows the nitrogen requirement for orange and grapefruit under normal conditions (Koo et al., 1984). Nitrogen requirement for young citrus trees is approximately 0.16 pound per tree per year in Florida (Fisher, 1990). Although some general recommendations have been given for citrus fertigation, with current knowledge it is difficult to develop a general and precise fertigation strategy. There are many different opinions about rates, concentrations, and times at which fertigation should be applied. Research is needed to determine the proper amount of fertilizer and application frequency when microirrigation is used.

For this application, a set of fertigation schedules is created in the knowledge base. These schedules were determined by interviewing citrus fertigation experts at the University of Florida. The knowledge base continuously checks the current time and the fertigation schedules. Fertigation is applied when the computer time matches the predefined fertigation schedules. The user should modify the schedules based upon expert recommendation only. Chemical injection rate can be computed by the following equation (Keller and Bliesner, 1990):



Fr A
c 7-2 c' t, To









Table 7.1 Pounds of nitrogen fertilizer to be applied to furnish the nitrogen
requirement of orange and grapefruit trees under normal conditions.

Fruit Pounds of nitrogen (N)' Pounds of nitrogen fertilizer needed per acre per year" production needed per acre per year
15.5 % N 33.5 % N 45.0% N (boxes/acre)
Orange Grapefruit Orange Grapefruit Orange Grapefruit Orange Grapefruit

< 200 100 60 645 485 300 225 220 165

300 120 90 775 580 360 270 265 200 400 160 120 1030 775 480 360 355 265 500 200 150 1290 965 600 450 445 335 600 240 180 1580 1160 715 538 535 400 700 280 210 1805 1355 835 620 620 465

> 800 300 240 1935 1450 895 670 665 500

a. Nitrogen needed is based on 0.4 pound per box of fruit for oranges and 0.3 pound per box for grapefruits. In most
cases, one should not use less than 100 pounds or more than 300 pounds of nitrogen for orange trees and not less
than 60 pounds or more than 240 pounds for grapefruit trees per acre per year.
b. The figures given should be divided by the number of applications per year to obtain pounds per application.








84

where q, = rate of injection of liquid fertilizer solution into the system, gph,

F, = fertilizer application rate (quantity of nutrients to be applied) per irrigation cycle, lb/A,

A = area irrigated in T,, ha ,

T. = irrigation application or set time, hr,

c' = concentration of actual nutrients in the liquid fertilizer, lb/gal, and

tr = ratio between fertilizer time and irrigation application time.

Because of the uncertainty of rainfall, a risk is always present that fertigation may be applied immediately before a rain. Thus, plant nutrients can be leached out of the crop root zone. To reduce this risk, the Markov chain probability of rainfall can be applied to fertigation management. Fertigation may be delayed when a high probability of rainfall occurs. When irrigation and fertigation schedules conflict, an algorithm is needed to merge the two schedules and to satisfy both the application requirements. Fertigation should serve as a part of irrigation whenever possible.

7.3.2 System Components of Fertigation

Fertigation systems consist of several components including an irrigation pumping station, a fertilizer injection device, an injection port, a solute fertilizer reservoir, a backflow prevention system, and calibration devices. Figure 7.1 is a common arrangement for fertilizer injection equipment. The electrical control board controls the operation of the fertigation pump (valve 64) and water supply valves (valve 1 to 63). Fertigation can be applied based upon the expert knowledge or user defined schedules.




Full Text
CHAPTER 4
SYSTEM SPECIFICATION AND DESIGN
4.1 Domain of the Problem
Citrus farm management practices include decisions regarding irrigation, fertigation,
cold protection, and plant disease control. Management decisions are affected by many
factors including soil, meteorological, and biological conditions. In Florida, microirrigation
systems are common today in citrus irrigation. Because of the reduced area of coverage of
microirrigation and the low water-holding-capacities of Florida's sandy soils, irrigation
requires high frequency applications. Computer controlled irrigation systems offer
considerable labor savings.
In Florida citrus groves, chemicals and cold protection are commonly applied
through microirrigation systems. Fertigation application decisions require knowledge about
plant nutrient demands and environmental impacts. For cold protection water must be
applied in a timely and accurate manner to avoid severe economic damage. Farmers need
to make daily management decisions to maximize their net income. An expert system that
contains expert knowledge on irrigation, fertigation, and cold protection is desirable to
improve citrus microirrigation management. Thus, the goal of this research was to develop
a real-time expert system (RTES) to assist in the management of citrus irrigation, fertigation,
and cold protection.
37


47
(Smajstrla, Personal communication). A Micro Switch1 model 141PC pressure transducer
was used in this research. Figure 4.2 illustrates the pressure sensor. Junction PI is the
reference pressure input, which is air pressure for this application. Junction P2 is connected
to the tensiometer to measure soil-water tension or gauge pressure. The pressure transducer
outputs an electric voltage signal in response to its resistance change caused by variation of
the input pressure. Output voltage of the sensor is proportional to the input pressure signal.
Table 4.2 lists the major characteristics of the sensor.
Table 4.2 Characteristics of pressure transducer Model 141PC.
Parameter
Minimum
Typical
Maximum
Output
4.85 V
5.0 V
2.5 V
5.15 V
Excitation
7 VDC
8 VDC
16 VDC
Input Pressure
- 15 psi
15 psi
Operation temperature
- 40C to + 85 C
Size
2.35 x 1.18 x 0.75 inches
Calibration of tensiometers
Before tensiometers are installed in the field, a calibration test is needed to determine
the relationship of soil-water tension and output voltage from the pressure sensor. Figure
4.3 shows the calibration equipment designed at the Soil and Water Hydraulics Laboratory,
1The manufacturers listed in this dissertation are for illustration only. No endorsement of
these companies or their products is implied by the authors or the University of Florida.


164
Valve number: irrigation control valve number,
Start time: irrigation start time (hour and minute) in 24-hour format,
Duration: irrigation duration (hour and minute) in 24-hour format, and
Skip day: irrigation skip days.
Fertigation schedules use the same data structure that irrigation uses. To define
fertigation schedule, click Fertigation Schedule on the submenu of Scheduling. Then, enter
the similar data that irrigation uses (Figure 13.4.3).
13.4.2 Apply Irrigation
Irrigation can be applied based upon the user defined irrigation schedule. The system
continuously checks the user defined irrigation schedule against the computer clock. If the
computer date and time satisfy the irrigation schedule, the irrigation valves, then, are
activated. The steps to run the user defined irrigation are as follows.
Figure 13.4.4. Irrigation application screen.


APPENDIX A
SAMPLE SENSOR DATA
Weather and tensiometer data at Conserv II, Orlando in 1994
J. Day
Time
Mini
C
MaxT
C
RH
%
Solar
kW/m2
WS
mi/h
Rain
mm
SI-6"
cb
Sl-12"
cb
S2-6"
cb
S2-12"
cb
60
100
16.9
17.1
79.5
0
4.1
0
10.3
9.96
18.5
9.41
60
200
17
17.1
81.4
0
3.6
0
10.3
9.92
18.7
9.39
60
300
16.6
17
83.8
0
1.9
0
10.5
10.1
18.7
9.43
60
400
16.5
16.8
86.2
0
2.2
0
1.25
9.94
18.5
9.16
60
500
16.3
16.5
88.1
0
2.2
0
3.36
10.1
19.1
9.58
60
600
16.3
16.4
88.8
0
2.6
0
4
9.96
19
9.42
60
700
16.1
16.4
88.8
0
3.4
0
4.36
10
19.6
9.8
60
800
16.4
17.5
87.6
0.14
4.1
0
4.55
10
21
10.21
60
900
17.6
20.1
82.6
0.42
7.6
0
4.8
10.1
23.1
10.7
60
1000
20.1
22.1
74
0.87
11
0
4.91
9.06
20.2
8.58
60
1100
21.8
23.2
68.5
0.91
13
0
5.05
8.81
19.5
8.08
60
1200
21.8
22.4
67.49
0.5
13
0
5.1
8.56
19
7.58
60
1300
21.8
23.7
66.97
1.08
12
0
5.11
8.89
21
7.96
60
1400
22.7
23.8
63.48
0.7
13
0
5.09
8.32
19.4
7.03
60
1500
22.5
23.4
66.33
0.61
13
0
5.29
8.03
19.7
7.03
60
1600
22.5
23.2
69.25
0.51
14
0
5.39
8.1
19.6
7.12
60
1700
22
22.7
72.3
0.26
12
0
5.51
8.17
19.8
7.12
60
1800
21.5
22
74.7
0.08
11
0
5.63
8.15
19.9
7.1
60
1900
21.2
21.5
75.3
0
11
0
5.64
8.07
19.9
7.06
60
2000
20.8
21.2
78.7
0
13
0
5.7
8.23
20.3
7.11
60
2100
20.6
20.9
80.4
0
14
0
5.91
8.33
20.5
7.33
60
2200
20.6
20.9
78.2
0
14
0
5.79
8.23
20.6
7.22
60
2300
20.2
20.8
79.5
0
13
0
5.89
8.27
20.3
7.27
60
2400
19
20.5
85.4
0
12
0
5.92
8.31
19.9
7.31
61
100
18.7
19
92.5
0
10
0
5.95
8.36
20.6
7.36
61
200
18.6
18.7
95.2
0
10
1
5.99
8.43
21
7.43
61
300
18.6
18.9
96.8
0
12
9
2.24
8.41
5.22
7.29
61
400
18.9
19.4
96.8
0
14
0
2.96
8.03
5.76
4.939
61
500
19.3
19.8
97.6
0
9.8
9
1.35
4.52
4.69
3.436
61
600
19.7
20.5
98.4
0
9.8
7
2.96
5.04
6.01
4.563
61
700
20.5
20.9
98.3
0
13
3
3.3
5.27
6.19
4.961
61
800
20.9
21.6
98.2
0.02
16
2
2.7
5.17
5.41
4.86
61
900
21.5
21.7
95.8
0.1
18
2
3.23
5.3
6.13
4.905
61
1000
21.6
23.4
90.7
0.73
18
1
3.43
5.6
6.55
5.174
61
1100
22.4
23.9
85.9
0.88
18
1
3.88
5.95
6.73
5.471
61
1200
22.5
24.6
85.7
0.87
17
0
3.98
6.05
7.07
5.74
61
1300
22.2
24.5
77.5
1.52
17
0
4.67
6.57
7.94
6.037
61
1400
22
23.2
70.2
1.42
15
0
3.91
6.25
7.28
6.155
135


125
The Conserv II research site is equipped with an automated weather station, data
logger, tensiometers, solenoid valves, computer, and communication system. A fertigation
system is also installed at the site. A variety of chemicals are injected through the
irrigation system.
Tests ofCIMS
The main objective of the field tests was to test the reasoning process and hardware
interactions. The field tests were separated into three parts: (1) control modules, (2)
reasoning process, and (3) complete system. The control modules were tested to turn on or
off user-defined solenoid valves to ensure that the control actions were reliable. Then, the
reasoning process was tested by running a combination of sensor inputs. The control
actions, which are conclusions of the reasoning process, were evaluated against known
results. These tests showed that the reasoning process made decisions as expected to turn
on or off the external devices. Finally, the complete system was tested in the field.
Although long-term field tests are recommended to study the water savings and reliability
of the system, short-term field tests were conducted to avoid interrupting on-going
researches.
9,5,3 Simulated Crop Water Use
To study the possible performance of the system within the limits of this research,
a modeling approach was used. A soil-water budget model combined with Markov chain
probability of rainfall was developed using the concept on which the RTES decision process
is based. The soil-water budget model, as described in Chapter 6, was used to simulate the
soil-water content in the crop root zone with and without the effects of Markov chain


137
63
1900
15.4
16.9
52.34
0.02
7.2
0
6.6
8.5
9.6
8.1
63
2000
13.8
15.5
65.48
0
5.7
0
6.6
8.48
9.54
8.14
63
2100
12.1
13.9
76.2
0
1.9
0
6.67
8.59
9.42
8.26
63
2200
11.6
12.4
83.2
0
0
0
6.71
8.62
9.55
8.36
63
2300
11.2
12.2
86.8
0
0
0
6.6
8.47
9.42
8.27
63
2400
10.6
11.3
90.6
0
0
0
6.59
8.52
9.28
8.28
64
100
9.69
10.7
93.3
0
0
0
6.51
8.44
9.14
8.22
64
200
9.49
10.6
95.7
0
0.8
0
6.48
8.45
9.21
8.21
64
300
9.49
10
96.6
0
0.4
0
6.56
8.57
9.36
8.33
64
400
9.48
9.82
97.5
0
0.5
0
6.6
8.64
9.47
8.4
64
500
9.57
9.95
97.7
0
0
0
6.5
8.49
9.47
8.37
64
600
9.29
10.4
98
0
0
0
6.52
8.46
9.32
8.15
64
700
8.76
9.81
98.2
0.02
0
0
6.86
8.88
9.94
8.69
64
800
9.81
12.5
98.8
0.35
0.6
0
6.97
9.03
10.9
8.94
64
900
12.5
15.7
93.7
0.73
6.6
0
6.77
9.08
10.7
8.8
64
1000
15.7
18.7
75.9
1.04
9.7
0
6.52
8.67
10.5
8.67
64
1100
18.7
20.9
60.12
1.91
9.8
0
6.75
8.67
9.63
8.16
64
1200
20.8
22.7
51.09
2.13
8.8
0
6.84
8.9
9.89
8.31
64
1300
22.4
24
45.6
2.33
7
0
6.84
9.13
9.84
8.18
64
1400
23.8
24.9
42.08
2.31
7.1
0
6.63
8.59
9.69
8.35
64
1500
24.8
25.8
39.98
2.01
7.7
0
7.22
9.04
10.5
8.52
64
1600
25.4
25.8
39.15
1.52
7.7
0
6.44
8.66
9.26
7.48
64
1700
24.1
25.8
41.15
0.98
8.2
0
7.08
9
10.2
8.38
64
1800
21.7
24.1
47.13
0.42
8.1
0
6.82
8.7
9.73
7.8
64
1900
18
21.6
58.47
0.04
5.3
0
6.89
8.62
9.92
8.05
64
2000
14.9
18
71
0
1.2
0
6.92
8.64
9.89
8.14
64
2100
13.8
14.9
82.8
0
0
0
7.06
8.57
9.67
8.16
64
2200
13
13.8
87.9
0
0
0
7.17
8.85
9.78
8.44
64
2300
12.3
13
91.4
0
0
0
6.96
8.64
9.3
8.18
64
2400
12
12.3
94.5
0
0
0
7.17
8.92
9.8
8.71
65
100
11.7
12.2
96.2
0
0
0
7.27
8.89
9.77
8.66
65
200
11.3
11.7
97.3
0.01
0
0
7.19
8.84
9.74
8.63
65
300
11
11.9
98.2
0
0
0
7.33
9.03
10.2
8.89
65
400
11.9
12.3
98.8
0
0
0
7.23
8.88
10.2
8.6
65
500
11.8
12.2
99.1
0
0
0
7.09
8.76
10
8.47
65
600
11.7
12
99.3
0
0
0
7.26
8.89
10.4
8.75
65
700
12
12.5
99.5
0.01
0
0
7.24
8.91
10.6
8.8
65
800
12.5
13
99.5
0.14
0.1
0
7.26
8.99
10.8
8.89
65
900
12.9
14.2
99.4
0.62
0.3
0
7.27
8.84
11.2
8.86
65
1000
13.8
16
99.1
1.12
1.5
0
6.69
8.57
11.3
8.72
65
1100
16
20.8
83.4
1.94
2.4
0
6.61
8.83
10.5
9.29
65
1200
20.9
24.4
59.75
2.25
1.5
0
6.92
8.86
10.3
8.59
65
1300
24.3
26.9
45.16
2.4
4.3
0
6.75
8.67
9.95
8.04
65
1400
26.8
27.5
35.47
2.34
6.4
0
5.78
8.15
9.31
7.65
65
1500
27.3
28.6
31.27
2.02
3.5
0
7.71
9.19
11.4
9.27
65
1600
27.7
29.1
32.95
1.49
3.4
0
7.7
9.25
11.1
9
65
1700
28.1
29.8
32.16
1.05
3.7
0
7.1
8.78
10.9
8.7
65
1800
24.1
28.4
41.71
0.32
3.3
0
7.4
9.03
10.5
8.25
65
1900
22
24
56.56
0.03
3.1
0
7.44
9.02
10.6
8.22
65
2000
20.4
21.9
71.9
0
5.4
0
7.56
9.04
10.8
8.46
65
2100
19.4
20.4
79.7
0
1.6
0
7.6
9.1
11
8.61
65
2200
18.3
19.4
85
0
0.8
0
7.51
8.96
11
8.49


13.5.4 Browse dialog window 171
13.5.5 Submenu of the Database main menu 172
13.5.6 Farm database screen 172
13.5.7 Weather database screen 173
13.5.8 Irrigation database screen 173
13.5.9 Crop database screen 173
13.5.10 Crop coefficient database screen 174
13.5.11 Soil database screen 174
13.6.1 Simulation submenu 175
13.6.2 Screen to set initial condition of the simulation 176
13.6.3 Simulation dialog window 176
13.6.4 Irrigation prognosis from the simulation 177
13.6.5 Simulated soil-water content 178
13.7.1 Submenu of Tools 179
13.7.2 Read weather data from weather station 180
13.7.3 ET method dialog window 180
13.7.4 Penman ET screen 181
13.7.5 Blaney-Criddle ET screen 181
13.7.6 Modified Blaney-Criddle ET screen 182
13.7.7 Stephens-Stewart ET screen 182
13.7.8 Estimate irrigation duration screen 183
xiii


51
Computer PC-DI072
Notas:
- One transformer feeds 10 solenoid valvas
(maxfrnum surge currant Is 1 A for aach valva)
- Each of the other 8transformers feed 0 sol. valvas
- All solenoid valvas ara 24 VAC
nrtrn
L0J 115V
Figure 4.5. Solenoid control relays of the irrigation system.
system. In addition to the automated relay control, the control board provides manual
switches and timers to turn on or off irrigation valves in case the relay system fails. In this
design, valve 64 is the default for the fertigation pump, and valve 40 is the master valve.


148
Test case 27
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 18 12 18 24 24)
(sensor-reading id2 6 18 12 18 24 18)
Test case 28
(TREE-STATUS MATURE)
(sensor-reading idl 6 18 12 18 24 32)
(sensor-reading id2 6 24 12 18 24 18)
(NON-CRITICAL GROWTH)
Test case 29
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 18 12 18 24 18)
(sensor-reading id2 6 18 12 24 24 18)
Test case 30
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 18 12 18 24 18)
(sensor-reading id2 6 18 12 18 24 24)
Test case 31
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 36 12 28 24 28)
(sensor-reading id2 6 28 12 28 24 28)
Test case 32
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 28 12 36 24 28)
(sensor-reading id2 6 28 12 28 24 28)
Data OK
Not turn on
Reject data from ID 1-3
(CF= -0.28)
Not turn on
Data OK
Not turn on
Data OK
Not turn on
Reject data from ED 1-1
Turn on the system
Reject data from ID1-1
Turn on the system


170
Find
Figure 13.5.2. Delete dialog screen.
a search function to find a particular record from a database. A
search dialog window (Figure 13.5.3) will display after clicking the
Find button. The search procedure is described as follows.
Click down arrow symbol to select a search field.
Enter the actual value of string one wants to search at the
Find field.
Click the Find button to locate the record.
Click the Find Again button to repeat the search.
Click the Cancel button to exit the screen.
Search for:
Find:
¡ |
In Field:
DATE
Mi
Figure 13.5.3. Search dialog screen.


144
(time-constrain no)
(weather-constrain no)
Test case 4
(TREE-STATUS YOUNG)
(sensor-reading idl 6 9 12 9 24 9)
(sensor-reading id2 6 13 12 9 24 9)
(time-constrain no)
(weather-constrain no)
Test case 5
(TREE-STATUS YOUNG)
(sensor-reading idl 6 9 12 9 24 9)
(sensor-reading id2 6 9 12 13 24 9)
(time-constrain no)
(weather-constrain no)
Test case 6
(TREE-STATUS YOUNG)
(sensor-reading idl 6 9 12 9 24 9)
(sensor-reading id2 6 9 12 9 24 13)
(time-constrain no)
(weather-constrain no)
Test case 7
(TREE-STATUS YOUNG)
- 0.28),
(sensor-reading idl 6 16 12 9 24 9)
(sensor-reading id2 6 16 12 9 24 9)
Test case 8
(TREE-STATUS YOUNG)
ID2-2
(sensor-reading idl 6 11 12 16 24 11)
(sensor-reading id2 6 11 12 16 24 11)
Data validation OK
Not turn on
Data validation OK
Not turn on
Data validation OK
Turn on by sensor at depth 3
Reject data from ED1-1(CF=
ID2-2 (CF = 0.28)
Not turn on
Reject data from ID 1-2 and
(CF = 0.7)
Not turn on


195
Koo, R. C. J. and A. G. Smajstrla. 1984. Effects of trickle irrigation and fertigation on
fruit production and juice quality of 'Valencia' orange. Proc. Fla. State Hort. Soc.
97:8-10.
Koo, R. C. J., C. A. Anderson, I. Stewart, D. P. H. Tucker, D. V. Calvert, and H. K.
Wutscher. 1984. Recommended fertilizers and nutritional sprays for citrus. Bui.
536D, IFAS, Univ. of Florida, Gainesville, FL, 30p.
Kumar, D., C. D. Heatwole, B. B. Ross, T. A. Dillaha. 1992. A knowledge-based
system for preliminary selection and economic evaluation of sprinkler irrigation
systems. Applied Eng. in Agriculture 8(4):441 -447.
Laffey, T. J., P. A. Cox, J. L. Schmidt, S. M. Kao, and J. Y. Read. 1988. Real-time
knowledge based systems. AI Magazine, Spring, pp. 27-45.
Lembke, W. D. and B. A. Jones, Jr. 1972. Selecting a method for irrigation scheduling
using a simulation model. Trans.of the ASAE 15(2):284-286.
Lemmon, H. 1986. COMAX: An expert system for cotton crop management. Science
233(7):29-33.
Martin, D. L., E. C. Stegman, and E. Fereres. 1990. Irrigation scheduling principles.
In Management of Farm Irrigation Systems, ed. G. J. Hoffman, T. A. Howell, and
K. H. Solomon, ASAE Monograph, St. Joseph, MI, pp. 155-203.
Martsolf, J. D. 1990. Cold protection strategies. Proc. Fla. State Hort. Soc. 103:72-78.
Martsolf, J. D., P. H. Heinemann, and C. T. Morrow. 1991. Protection from frost or
freeze with sprinklers. HortScience 26(6):729.
Martsolf, J. D. 1992. Cold protection mechanism. Proc. Fla. State Hort. Soc. 105:91-94.
McClendon, R. W. W. D. Batchelor, and J. E. Hook. 1989. An expert simulation
system for irrigation management. ASAE Paper No. 89-2460, St. Joseph, MI.
Mckim, H. L., J. E. Walsh, and D. N. Arion. 1980. Review of techniques for measuring
soil moisture in situ. United States Army Corps of Engineers, Special Report 80-
31, Cold Regions Research and Engineering Lab, Hanover, NH.
Merlo, G. 1992. Possible applications of experts systems in the management of water
distribution systems. Water Supply 10:87-95.


198
Prerau, D. S., A. S. Gunderson, R. E. Reinke, and M. R. Adler. 1990. Maintainability
techniques in developing large expert systems. IEEE Expert, June, pp.71-79.
Protopapas, A. L., and A. P. Georgakakos. 1990. An optimal control method for real
time irrigation scheduling, Water Resources Research 26:647-669.
Quillian, M. R. 1968. Semantic memory. In Semantic Information Process, ed. by M.
Minsky. The MIT Press, Cambridge, pp. 227-270.
Quinlan, J. R. 1983. Learning efficient classification procedures and their application to
chess endgames. In Machine Learning: An Artificial Intelligence Approach, ed.
R. S. Michalski, J. G. Carbonell, and T. M. Mitchell. Tioga Publishing, Palo
Alto, CA, pp. 463-482.
Reitz, H. J., L. G. Albrigo, L. H. Allen, Jr., J. F. Bartholic, D. V.Calvert, H. W. Ford,
J. F. Gerber, L. C. Hammond, D. S. Harrison, R. C. J. Koo, R. S. Mansell, J.
M. Myers, J. S. Rogers, and D. P. H. Tucker. Water requirements for citrus.
WRC-4. Water Resources Council, IFAS, Univ. of Florida, Gainesville, FL.
Richardson, C. W. 1981. Stochastic simulation of daily precipitation, temperature, and
solar radiation. Water Resources Research 17(1)182-190.
Richardson, C. W. 1985. Weather simulation for crop management models. Trans, of the
ASAE 28(5): 1602-1606.
Richardson, C. W. and Wright D. A. 1984. WGEN: A model for generating daily
weather variables. ARS-8, U. S. Department of Agricultural Res. Service,
Temple, TX, 80p.
Ringland, G. A. and D. A. Duce. 1988. Approaches to Knowledge Representation: An
Introduction. John Wiley & Sons Inc., New York.
Robinson, M. R. 1990. Report on fertilizers and irrigation. Proc. Fla. State Hort. Soc.
13:140-145.
Rogers, J. S., L. H. Allen Jr., and D. V. Calvert. 1983. Evapotranspiration from a
humid-region developing citrus grove with grass cover. Trans, of ASAE
26:1778-1783, 1792.
Rogers, D. H. and R. L. Elliott. 1989. Irrigation scheduling using crop growth
simulation, risk analysis, and weather forecasts. Trans, of ASAE 32:1669-1667.


96
Checking the variation among sensor readings
Sensor readings may also be
validated by comparing readings
from different sensors or the rate
change of sensor readings. Three
assumptions were made about the
soil characteristics in this study.
First, the soil profile is
homogeneous. Second, the sensors
are properly installed in the crop
root zone. Third, irrigation water is
Figure 8.6. CF propagation for sensors SI and
uniformly provided to the crop root
zone. Under these assumptions,
variation of the sensor readings at
different depths should be within a
certain range under static conditions.
When the output of a sensor differs
from other sensor's readings more
than the expert given value (EGV), a
lower CF value is assigned to the
sensor. This possible failure is
denoted as Failure B. Since three
Figure 8.7. CF propagation for sensor S2.


60
Table 5.1) occurred in the summer and early fall. In contrast, lower wet-day frequencies
occurred during the dry season, as expected.
Figure 5.1. Annual rainfall distribution in Orlando from 1952 to 1992.
5,5 Statistical Test
The rainfall probability has shown variation among months and seasons. A statistical
analysis can be used to evaluate data significance among them. A paired t-test can be used
to test whether there are significant differences among the monthly or seasonal rainfall


129
water. Although high Markov chain probability of rainfall should be used in the irrigation
decision process, there might be very few rainfall sequences which can match the high
Markov chain probability of rainfall if such threshold value is too high. The results showed
that 50 percent Markov chain probability of rainfall may be an acceptable threshold value
of rainfall probability.
Although Markov chain probability of rainfall, theoretically, provides a better chance
to increase effective rainfall, this result showed that water savings are not significant if the
system only relies on the results of Markov chain probability of rainfall. Finally, this
simulation is only a preliminary study of water savings affected by the decision process used
in the RTES. Because the simulation assumes a well-managed irrigation system, higher
water savings are expected when using an RTES for a conventionally managed system.


186
Figure 13.8.3. Calculator
Figure 13.8.4. Calendar and diary screen.
13,8,3 Irrigation Calendar and Diary
Figure 13.8.4 shows the calendar and diary. The left side of the window displays a
calendar and the right side of the window shows the diary box. This calendar and diary can
be used as a tool for calendar based irrigation management.
Click the < Month button to display previous month.
Click the Month > button to display next month.
Click the < Year button to display previous year.
Click the Year > button to display next year.
Click the Today button to display highlight today's date.
Click any day in the month to view that day's diary or message.
13.8 4 Clock
Click the Clock on the submenu of Help to display the clock in Figure 13.8.5.


Table 9.2 Accumulated citrus net irrigation requirements and number of irrigations for 22 years in central Florida.
Model
SCS
AFSIRS
Soil Water Balance
Case 1 Case 2 Case 3 Case 4
60% rainfall nrobabilitv 50% rainfall probability 40% rainfall probability Deficit irrieation
NIR NIR-80 NIR-70 NIR-80 NIR-70 NIR-80 NIR-70 NIR-80 NIR-70
Irrigation
depth
(inches)
318.34
297.00
310.67
301.52
298.61
285.19
279.82
267.07
251.61
247.39
225.89
Difference
from NIR
(inches)
-7.67
13.67
0.0
9.15
12.08
25.48
30.85
43.60
59.06
63.28
84.78
Number of
deficit
irrigations
46
51
159
181
322
381
441
547
Number of
full
irrigations
596
568
569
518
524
445
447
353
355
Number of
total
irrigations
596
614
620
677
705
767
828
794
902
Note: NIR
NIR-80
NIR-70
scs
AFSIRS
Net irrigation requirements.
Irrigation applied to 80 percent of field capacity.
Irrigation applied to 70 percent of field capacity.
This value is the normal year citrus water requirement estimated by Soil Conservation
Service (SCS, 1982) and multiplied by 22.
Agricultural Field Scale Irrigation Requirements Simulation model (Smajstrla, 1990).


12
1990) for different locations and crop types. Maintaining a soil water balance is a widely
used and effective approach for irrigation scheduling. However, this approach can require
substantial weather data, and these data are not available in many cases. Models to generate
weather data have been developed for these purposes (Richardson, 1981, 1985; Richardson
and Wright, 1984; Villalobos and Fereres, 1989; Jones, 1993).
Simulation of crop growth
Crop growth models can be developed to simulate crop growth. A crop growth
model can be a physically based representation of the dynamics of the soil-crop-atmosphere
system. The crop yield can be predicted by explicit models of the plant growth process, such
as assimilation, respiration, and transpiration (Protopapas and Georgakakos, 1990).
Crop growth models have been developed as aids for irrigation water management
(Swaney et al., 1983; Rogers and Elliott, 1989; Jones and Ritchie, 1990, Jones, 1993).
Although crop growth models have been successfully used in irrigation management and
decision-making, the technique has some difficulties in practice. First, models are currently
not available for all crops. This is because it is difficult to develop an accurate crop growth
model. A crop system can be complex and affected by many factors such as weather,
insects, weeds, diseases, soil physical and chemical factors, and the interactions of these
factors. Second, factors such as uncertainty of future conditions and limitations of available
data make the use of simulation models difficult for real-time applications.


33
individual element of knowledge that can be updated independently of the other rules. The
major drawback of the system is inefficiency due to the use of infinite chaining. Moreover,
there may exist contradictory and inconsistent knowledge among the rules when new rules
are added.
Knowledge representation is the foundation of AI. Many researchers (Bobrow and
Winograd, 1977; Brachman and Levesque, 1985; Fox, 1990) have attempted to improve the
ways of representing knowledge. Advanced knowledge representation approaches include
the techniques of spatial, causal or temporal models, and neural nets. A general knowledge
representation system cannot be constructed easily. The selection of knowledge
representation methods will mainly depend on the inherent structure of the knowledge and
what knowledge representation the expert system tool will support.
3.5 Rule-Based Expert Systems
Many ESs have been developed using rule-based structure or so called production
systems. Rule-based ESs have been successfully applied in many domains, including
medical diagnosis, mathematic discovery, and hardware configuration. Rule-based systems
have been widely applied because the systems have advantages of modularity, uniformity,
and naturalness (Gonzalez and Dankel, 1993). In addition, many development tools (Meta-
MYCEN, CLIPS, and LEVEL 5) with relatively low costs are available for development of
rule-based systems.


CHAPTER 9
SYSTEM IMPLEMENTATION AND TESTS
9.1 Function Requirements of CIMS
The overall objective of the citrus irrigation management system (CIMS) is to
provide a tool to improve citrus microirrigation management. To achieve this objective, the
software development of CIMS should meet the following requirements:
The system must be easy to use.
The system program must be structured and modularized.
The system should provide control routines to turn on or off the solenoid
values and pumps of an irrigation system.
The knowledge base must realize real-time performance to achieve the
management goals.
The system should be able to deal with sensor data uncertainty.
The system must also provide conventional control approaches and utilities
to satisfy different system hardware requirements.
9,2 Module Design of CIMS
System structure and module design are crucial to achieve the functional
requirements of CIMS. The structural design of CIMS is not only related to these functional
112


26
phase in the development of a knowledge-based system. Researchers refer to this as the
knowledge acquisition bottleneck (Feigenbaum, 1979) in the development of knowledge-
based systems.
The success of current ES technology is highly related to the strict separation
between a domain-dependent knowledge base and an inference engine. The power of a
knowledge-based system relies more upon the quality of the knowledge base rather than the
characteristics of the inference engine.
3,3,1 Basic Approaches
As Figure 3.3 shows, knowledge acquisition is mainly eliciting and organizing
knowledge from human experts. The role of a knowledge engineer is to communicate the
basis of the performance with the experts and to specify it in a form suitable for a computer.
The basic approaches to knowledge acquisition can be summarized as
Interviews,
Questionnaires and observation of the expert at work,
Intuition, and
. Using knowledge engineering facilitators and inductive tools.
Interviewing is one of the major approaches of knowledge acquisition. The
interview can be structured or unstructured, and the communication can be one-to-one or
many-to-one. Giarratano and Riley (1989) described the basic procedure of the knowledge
acquisition and extraction task. First, an acquisition strategy should be decided. This
includes specifying how knowledge will be acquired and the methods to be used in the
interview. Second, the knowledge elements or specific knowledge that could be used by the


69
Irrigation decision-making relies highly on the skill of the irrigation manager.
Irrigation should be applied when the soil-water content reaches the MAD. SCS (1982)
recommended that available soil-water depletion should not exceed 30 percent between fruit
set (February to March) and the period when young fruit has reached more than 1-inch in
diameter (June to July). During remaining months of the year, soil-water depletion should
not exceed 50 percent. Studies in Florida (Koo, 1963; Gerber et al., 1973; Smajstrla et al.,
1987) suggested that mature trees should be irrigated when one-third (33 percent) of the soil
moisture in the root zone is depleted in spring or two-thirds (66 percent) is depleted during
the rest of year.
As a rule of thumb, MAD within the root zone should not fall below 50 percent of
the total available water-holding capacity (Keller and Bliesner, 1990). For young trees and
mature trees during critical growth stages, irrigation should be applied when 25 to 35 percent
of depletion has occurred.
6,4 Irrigation Depth and Duration
Irrigation scheduling requires making decisions on irrigation duration and frequency
to meet crop ET demands. The amount of water and application frequency are related to
water management, soil property, and economic considerations. Irrigation duration is
associated with the application rate. For young trees, irrigation duration should average
between one and three hours for microsprinklers with flow rates of 10 to 15 gallons per
hour, and three to six hours for most drip systems depending on soil type, frequency of
irrigation, and number of emitters (Davies et al., 1989).


154
13.2. Real-Time Expert System
The RTES uses expert knowledge and data from on-site sensors to make decisions
on citrus irrigation, fertigation, and cold protection. The RTES has two modules: Facts and
\
Facts
Expert .Control Scheduling database Simulation Tools Help .Quit
Irriqat
on
fertigation
Freeze Protection
Initial Facts
Fertigation Schedule
Figure 13.2.1. Submenu of the Facts.
Expert as Figure 13.2.1. The Facts module defines the initial facts for the expert system.
The facts can be (1) valve on or off definition for the applications of irrigation, fertigation,
or cold protection, (2) threshold values and the crop growth stage for the decision-making,
(3) fertigation schedules. The Expert module executes the expert system to perform the
following tasks: (1) read data from the soil moisture sensors and the weather station, (2) read
initial facts defined in the Facts module, (3) conduct reasoning process based on the
knowledge base it contains, and (4) take control actions to activate or deactivate irrigation
control valves and pumps.
13.2,1 Define Initial Facts
Click Facts on the main menu and submenu of the Facts showed in Figure 13.2.1.
The steps to define the initial facts are described as follows.


14
Field tests demonstrated that this scheduling method should be easily adaptable to irrigation
control, particularly in a sandy soil with low soil-water holding capacity.
Although most current irrigation management practices rely on manual operation or
timers, many researchers have focused on computer controlled irrigation systems (Duke et
al., 1984; Zazueta et al., 1984a, 1989; Phene et al., 1989a; Bums et al., 1990; Shayya et al.,
1990; Zazueta and Smajstrla, 1992). Vellidis et al. (1990) developed a microcomputer-
based data acquisition system for soil water potential measurements. The system consisted
of commercially available components: tensiometers, pressure transducers, a data acquisition
system, control devices, and a portable computer. In tests, the system was found to be
effective to monitor temporal variation of soil moisture potential.
Computerized irrigation control systems have the potential for water and energy
savings. Stombaugh et al. (1992) studied frost protection of strawberries using an automated
pulsed irrigation system. Their studies showed that the automation of irrigation could
potentially reduce water use by 89 percent under mild frost conditions. For nursery
irrigation, Zazueta et al. (1984b) reported that when closed loop computer control was added
to the system, water savings of about 20 percent (for a well-managed system) to 60 percent
or more (for a poorly managed system) were achieved. These water savings were achieved
by control of water deliveries using preset irrigation schedules. Even better results might
be achieved if the system integrated expert knowledge on irrigation management.


76


LIST OF FIGURES
Figure Page
3.1 Major components of an expert system 20
3.2 Major components of an RTES 23
3.3 Knowledge acquisition cycle 25
3.4 The architecture and execution cycle of rule-based systems 34
4.1 A regular tensiometer and a tensiometer with micro-pressure
transducer 46
4.2 Pressure transducers (Model 141PC) from Micro Switch 46
4.3 Tensiometer calibration equipment 48
4.4 Tensiometer calibration curve 48
4.5 Solenoid control relays of the irrigation system 51
4.6 Hardware layout of the control system 52
4.7 Paradigm of the real-time expert system 53
5.1 Annual rainfall distribution in Orlando from 1952 to 1992 60
6.1 Irrigation by threshold of soil-water content 72
6.2 Soil-water content versus soil-water tension for Candler fine sand .... 76
7.1 Major components of a fertigation system 85
8.1 Major inputs and outcomes of the expert system 88
x


70
The maximum net depth per irrigation (Dx) should replace the soil moisture deficit.
The net irrigation depth for microirrigation can be described as the following equation
(Keller and Bliesner, 1990):
D
X
MAD
100
wa z
6-3
where MAD = management allowed depletion, %
W, = available water-holding capacity of the soil, in/ft, and
Z = plant root zone, ft.
For microsprinkler irrigation, irrigation time to bring root zone to field capacity can
be expressed as (Parsons et al., 1993)
AW DfZ
6-4
where Id =
irrigation duration (hour),
AW =
available soil-water content (in/ft),
Df =
depletion of AW prior to irrigation (%),
Z
root depth (ft),
Pr
precipitation rate (in/hr), and
2.04 F, N Ef
p 1
d2
where Fr =
flow rate of emitter (gal/hr),
N
number of emitters per tree,
6-5


71
Ef = overall efficiency of the irrigation system (%), and
d = diameter of spray pattern (ft).
Because of the low water-holding capacity of Florida sandy soils, irrigation
applications need to be more frequent to maintain the proper range of soil-water contents.
With a computer controlled microirrigation system, water can be applied frequently in a
timely manner to maintain less variation of soil-water content and without increasing
application costs.
6 5 Soil-Water Budget
A soil-water budget is commonly used to describe the amount of available water in
a crop root zone. For irrigation scheduling, it is convenient to calculate the amount of water
used (depletion) in the root zone instead of estimating the remaining water. The water
balance equation can be expressed as
where 0i+1 0¡
ET.R-ER-NI
- 8, 5 6-6
soil profile depletions at the end and beginning of a period,
ER
= effective rainfall (in),
ETa
= actual daily evapotranspiration (in/day),
D
= crop effective root depth (in),
NI
= net irrigation (in), and
R
= runoff from surface and deep percolation (in/day).


156
Define fertieation control valve (on or off)
Fertigation application is different from irrigation and cold protection. Fertigation
is applied in a sequence of preinjection, fertigation, and flush. After the irrigation pipe line
is pressurized during preinjection, fertigation is then applied. Because the potential back
flow and the chemicals may remain inside the pipe line, a flush period is needed to flush the
chemicals out of the pipe line after chemical injection is stopped. The steps to define
fertigation facts are as follows.
Click Fertigation on the submenu of Facts. An editing window is displayed
as Figure 13.2.3.
Enter the number of
preinjection time and
flush time (minutes) in
lines two and four.
Define the valve on or
off status using the same
procedures of irrigation
valve definition.
Figure 13.2.3. Fertigation block and valve
definition.
Define or modify initial facts
Click Initial Facts to open the fact editing window (Figure 13.2.4).
Select the facts to edit. Use semi-colon at the beginning of each line to
omit the line from the initial facts.


61
61
61
61
61
61
61
61
61
61
62
62
62
62
62
62
62
62
62
62
62
62
62
62
62
62
62
62
62
62
62
62
62
62
63
63
63
63
63
63
63
63
63
63
63
63
63
63
63
63
63
63
136
1500
21.6
23.3
66.17
1.08
15
0
4.38
6.57
7.21
6.087
1600
20.4
22
68.61
0.61
14
0
4.08
6.45
6.73
6.329
1700
19.4
21.5
68.57
0.59
15
0
4.66
7.05
7.83
6.755
1800
18.7
19.8
68.11
0.25
15
0
4.59
6.98
7.52
6.599
1900
17.3
18.7
70.7
0.02
14
0
4.74
7.1
7.73
6.723
2000
16.3
17.3
73
0
13
0
4.76
7.12
7.89
6.83
2100
15.6
16.3
76.4
0
13
0
4.73
7.12
7.86
6.859
2200
15.6
16.4
72
0
12
0
4.91
7.2
8.03
6.96
2300
15.8
16.4
70.7
0
15
0
4.95
7.25
8.05
6.993
2400
15
15.7
72.1
0
16
0
5
7.28
8.05
6.976
100
13.9
14.9
71.2
0
16
0
4.99
7.24
8.07
7.03
200
12.7
13.8
69.43
0
19
0
5.08
7.37
8.06
7.11
300
12.1
12.7
71.7
0
17
0
5.18
7.46
8.22
7.23
400
11.7
12.1
76
0
14
0
5.24
7.51
8.27
7.3
500
11.9
12.1
72.9
0
15
0
5.27
7.55
8.31
7.3
600
11.8
11.9
71.8
0
15
0
5.34
7.56
8.37
7.34
700
11.2
11.8
74.8
0
15
0
5.39
7.6
8.44
7.39
800
11.2
11.3
77.6
0.02
13
0
5.4
7.6
8.46
7.39
900
11.2
11.4
76.1
0.12
16
0
5.48
7.69
8.6
7.5
1000
11.4
11.6
75.1
0.22
14
0
5.51
7.73
8.75
7.58
1100
11.3
11.7
73.9
0.26
16
0
5.51
7.71
8.79
7.59
1200
13
14
63.01
2.14
3
0
5.97
8.09
9.15
7.8
1300
14
15.6
57.81
1.55
18
0
5.59
7.92
9.4
7.89
1400
15.4
17
52.89
2.34
17
0
5.45
7.72
8.84
7.77
1500
16.1
16.5
49.63
2.04
17
0
5.04
7.7
8.34
7.16
1600
15.8
16.7
48.39
1.57
16
0
6
8.34
9.44
8.11
1700
14.8
16
51.64
1
15
0
5.65
8.24
8.48
7.31
1800
13.5
15
55.23
0.39
15
0
5.67
8.04
8.74
7.58
1900
12.3
13.5
57.34
0.03
12
0
5.78
8.04
8.65
7.66
2000
11.6
12.2
60.5
0
9.8
0
5.94
8.15
8.72
7.79
2100
10.8
11.7
63.13
0
8.8
0
5.86
8.05
8.76
7.88
2200
9.94
10.8
64.48
0
6.5
0
6.05
8.18
8.85
8
2300
10.1
10.5
62.99
0
7.3
0
5.99
8.22
8.92
7.99
2400
9.11
10.4
61.22
0
4.6
0
5.87
8.03
8.62
7.79
100
7.42
9.06
69.02
0
1.1
0
6
8.12
8.71
7.98
200
5.74
7.54
77.2
0
0.3
0
6.04
8.16
8.75
8.09
300
5.5
7.39
76.1
0
1.1
0
6.1
8.3
9.06
8.22
400
6.95
7.83
66.09
0
3.8
0
6.12
8.27
8.86
8
500
6.55
7.88
63.78
0
2.2
0
5.86
7.91
8.73
7.88
600
4.78
6.48
72
0
0
0
6.06
8.21
8.81
8.08
700
4.71
6.33
75.1
0.01
1.1
0
6.29
8.38
9.42
8.38
800
6.26
9.65
68.19
0.27
2.6
0
6.59
8.76
10.4
8.69
900
9.73
13.6
50.61
0.76
8.4
0
6.1
8.41
9.32
8.05
1000
13.6
15.9
40.65
1.14
9.1
0
6.59
8.54
9.1
7.79
1100
15.8
17.6
36.53
2.02
8.2
0
7.12
9.16
10
8.67
1200
17.3
18.6
34.66
2.32
8.6
0
6.46
8.58
9.72
8.33
1300
18.4
20.6
30.85
2.39
9
0
5.96
8.38
9.02
7.72
1400
20.4
21.4
28.13
2.15
9.7
0
6.3
8.28
9.37
7.96
1500
21.3
22.3
27.34
2.05
11
0
6.16
8.29
9.49
8
1600
20
22.4
28.39
1.24
10
0
6.44
8.54
9.62
8.16
1700
18.8
20.3
30.69
0.76
11
0
6.66
8.87
10.3
8.52
1800
17
19.8
36.85
0.35
9.9
0
6.49
8.43
9.43
7.97


192
Fisher, J. 1990. Fertigation: microsprinklers provide new options in fertilization
programs. Citrus Industry, January, pp. 12-18.
Fitzsimmons, D. W. and N. C. Young. 1972. Tensiometer-pressure transduce system for
studying unsteady flow through soils. Trans, of the ASAE 15:272-275.
Florida Agricultural Statistics. 1994. Commercial Citrus Inventory, Orlando, FL.
Fox, M. S. 1990. AI and expert system myths, legends and facts. IEEE Expert, February,
pp. 8-20.
Gabriel, K. R. and J. Newmann. 1962. A Markov chain model for daily rainfall
occurrence at Tel Aviv. Jour, of the Royal Meteorological Society 88(375): 90-95.
Gardner, W. H. 1986. Water content. In Methods of Soil Analysis. Part 1. Physical and
Mineralogical Methods, ed. A. Klute. Agronomy Series No. 9. Am. Soc.
Agronomy, 2nd ed., pp. 493-544.
Gerber, J. F., A. H. Krezdorn, J. F. Bartholic, J. R. Conner, H. J. Reitz, R. C. J. Koo,
D.S. Harrison, E. T. Smerdon, J. G. Georg, L. H. Harris, and J. T. Bradley.
1973. Water needs for Florida Citrus. Proc. Fla. State Hort. Soc. 86:61-64.
Giarratano, J. and G. Riley. 1989. Expert Systems Principles and Programming. PWS-
Kent Publishing Company, Boston.
Gonzalez, A. and D. Dankel II. 1993. Engineering Knowledge-based Systems: Theory
and Practice. Prentice Hall, Inc., Englewood Clitts, NJ.
Harandi, M. T. and R. Lange. 1990. Model-based knowledge acquisition. In Knowledge
Engineering, ed. H. Adeli. McGraw-Hill Publishing Company, New York, vol.
I pp. 103-129.
Hardy, N. G. 1989. Using irrigation for frost protection. Citrus Industry, October, pp.
39-41.
Harrison, D. S., J. F. Gerber, and R. E. Choate. 1987. Sprinkler irrigation for cold
protection, Florida Cooperative Extension Service, Circular 348. IFAS, Univ. of
Florida, Gainesville, FL.
Harrison, P. R. and P. A. Harrison. 1994. Validating an embedded intelligent sensor
control system. IEEE Expert, June, pp. 49-53.


140
-+- Rain (cm) S1-6" ^_S1-12" -*-82-6" __S2-12"
Figure 12.2. Tensiometer readings on Juilan day 278, 1994.
Note:
SI-6"

Tensiometer at site 1 with 6" depth.
Sl-12"

Tensiometer at site 1 with 12" depth.
S2-6"

Tensiometer at site 2 with 6" depth.
S2-12"

Tensiometer at site 2 with 12" depth.
Tensiometer Sl-6"
had bad data at 1400 minutes on day 213 and recovered later.


176
Click Water Budget on
the submenu of
Simulation to open the
simulation dialog window
(Figure 13.6.3). Enter
starting and ending date of
Figure 13.6.2. Screen to set initial condition of the
simulation.
the simulation from the
screen.
Click the Cancel button
to exit the screen or
click the Execute
button to run the
simulation. The
simulation model
obtains its required
data from the databases. The simulation results are stored in three files: (1) simulated
soil-water content from the simulation period, (2) the most recent simulated soil-
water content, and (3) one week prognosis of irrigation requirement, which assumes
no rainfall to occur in that week and same ET rate as the previous week.
Figure 13.6.3. Simulation dialog window.


TABLE OF CONTENTS
ACKNOWLEDGMENTS iii
LIST OF TABLES viii
LIST OF FIGURES x
ABSTRACT xvii
CHAPTER 1 INTRODUCTION 1
1.1 Statement of the Problem 1
1.2 Objectives of the Dissertation 5
CHAPTER 2 REVIEW OF THE LITERATURE 7
2.1 Citrus Irrigation in Florida 7
2.2 Soil Moisture Sensors 8
2.3 Irrigation Scheduling 9
2.3.1 Monitoring Method 10
2.3.2 Computer Simulation 11
2.4 Irrigation Control 13
2.5 Expert Systems in Agriculture 15
2.6 Summary 17
CHAPTER 3 GENERAL EXPERT SYSTEM CONCEPTS 19
3.1 Expert Systems 19
3.1.1 Inference Engine 20
3.1.2 Knowledge Base 21
3.1.3 User Interface 22
3.2 Real-Time Expert Systems 22
3.2.1 Why Use an RTES? 24
3.2.2 Characteristics of RTESs 24
IV


31
attributes associated with each class or instance into more compact local data structures. A
frame is mainly a group of slots (attribute) and fillers (values) that define a stereotype of
knowledge. The fillers can also be subdivided into facets. Each slot, filler, and facet has
its own associated values.
Frame systems are very suitable for those well-defined features (stereotype
knowledge), so that many of its slots have default values. Therefore, the main advantage
of frame systems is that their knowledge representation is significantly better structured and
organized than knowledge in semantic net systems. Furthermore, the system can only
trigger specific actions through demons during the processing of information, instead of
repeatedly testing a rule in a rule-based system. This will significantly increase the
efficiency of knowledge processing. The frame system, however, may have difficulty
dealing with heuristic knowledge and coping with a new situation beyond the default values.
3,4,3 Objects
An object is a representation corresponding to a conceptual entity in the real world
and how the information related to the entity is manipulated. The basic idea behind an
object-oriented representation is that information should be clustered around the "object."
The difference between frame and object representation is that an object-oriented approach
creates a tight bond between the code and data instead of separating them into two complex,
separate structures (Gonzalez and Dankel, 1993).
The characteristics of object-orientation are the levels of abstraction that can be
achieved and its ability of encapsulation, inheritance, and polymorphism. The approach,
however, possesses drawbacks similar to those of the frame method.


108
8.5 Cold Protection
Making a decision on cold protection requires balance cold damage risk and resource
conservation. The decision to apply cold protection is associated with (1) when to start cold
protection, (2) when to stop the irrigation, and (3) insuring that there are sufficient resources
(water and energy) available. The system assumes that sufficient resources are available.
Figure 8.13 illustrates the cold protection decision process. An alert message is displayed
when air temperature reaches the warning temperature, which indicates potential cold
damage may occur. Cold protection starts at the critical temperature. After a certain period
of application, cold protection stops when air temperature reaches the stop temperature.
Stop temperature
Critical temperature
On
Off
Time
Figure 8.13. Cold protection decision processes based on the critical air temperature.


56
success for various locations (Gabriel and Neumann, 1962; Jones et al, 1972; Todorovic and
Woolhiser, 1975). The rainfall probabilities according to wet-dry day sequences have also
been applied for irrigation management (Safley et al., 1974). A Markov chain rainfall
probability model was used to estimate the rainfall occurrence to assist irrigation scheduling
in this study. This model is not a physical explanation of rainfall occurrence, amount of
precipitation, or other meteorological observations, but merely a statistical description of the
past observed behavior. The purpose of using the Markov chain probability of rainfall was
to couple the rainfall probability with the irrigation decision-making process.
5.2 Markov Chain
A Markov chain is one particular type of stochastic process. Feller (1969) defined
a Markov chain as
a stochastic process in which the future development
depends only on the present state, but not the past history
of the process or the manner in which the present state
was reached, (p.444)
A stochastic model, in general, provides only the probability associated with a set of possible
future outcomes. Thus, a state X is followed by state Y with probability p, and by state Z
with probability q = 1 p, where X and Y are the only possible occurrences. The Markov
approach can be applied to wet-dry day sequences. Let (, denote the occurrence of a wet or
dry day.

1, if day i is wet
5, 5-1
0, if day i is dry


BIOGRAPHICAL SKETCH
The author was bom on February 16, 1961, in Yulin County, Shaanxi, China. After
receiving his high school diploma from Middle School of Yulin in 1978, he studied at the
Beijing University of Aeronautics and Astronautics (BUAA), China. In 1982, he received
a Bachelor of Science degree from BUAA. After graduating from BIAA, he was an
assistant engineer employed at the Research Institute of CFTRC, Xi'an, China. In 1987, he
was sent to the United States through the visiting scholar program at the University of
Florida. In 1988, he studied in the Agricultural and Biological Engineering Department at
the University of Florida and received a Master of Engineering degree in May, 1990.
204


184
13 7 4 Map of Irrigation System Layout
An irrigation system layout map can be created in the system. The soil moisture
sensors and weather station locations can also be displayed on the map.
To view the map, click the
Field Layout Map on the
submenu of Tools. Figure
13.7.10 shows a dummy
field layout map.
Click the Close button to Figure. 13.7.10. A dummy field layout
Pipe layout map
II
exit the screen.
map.


APPENDIX B
TEST CASES AND RESULTS
B. 1 Test Data Description
The following test data are in CLIPS fact format.
Test data Data description
(TREE-STATUS YOUNG) Young trees are planted in the site
(TREE-STATUS MATURE)
(sensor-reading idl 6 13 12 12 24 16)
(sensor-reading id2 6 9 12 12 24 14)
(time-constrain no)
(weather-constrain no)
(weather-data ws 30 rh 70
rain 0.0 airtemp 38)
(FTSCH1 MONDAY START-TIME
3 10 END-TIME 4 10)
(FTSCH2 1994 11 23 START-TIME
3 20 END-TIME 3 30)
EDl-i
ID2-
Mature trees are planted in the site
Tensiometer readings at site 1 (idl)
at depth 1, (6"~12 cb), 2 (12' 12
cb), and 3 (24 "--16 cb)
Tensiometer readings at site 2 (id2)
at depth 1, (6"~9 cb), 2 (12' 12
cb), and 3 (24"14 cb)
Irrigation without time constraint
Irrigation without weather constraint
Weather data with wind speed 30
mph, relative humidity 70 %, rainfall
0.0 inches, and air temperature 38 C
Fertigation schedule by day of the
week. Fertigation is scheduled on
Monday. It starts at 3:10 a.m. and
ends at 4:10 a.m.
Fertigation schedule by date of the
year. Fertigation is scheduled on
November 23, 1994. It starts at 3:20
a.m. and ends at 3:30 a.m.
Sensor at site 1 and depth i (i = 1, 2,
and 3)
Sensor at site 2 and depth i (i = 1, 2,
and 3)
142


161
events on the subject. The application history files display the starting and ending time of
each event which has occurred.
13.3 Control Panel
CIMS provides an irrigation control screen so that the user can turn on or off the
control valves by simply clicking a button from the control screen. The steps to run the
control panel are described as follows.
Click the Control on the main menu. A control panel screen appears as Figure
13.3.1.
Figure 13.3.1. Control panel of the system.
Select radio button Irrigation, Fertigation, or Freeze Protection for a particular
application.
Click the Cancel button to exit the screen and click the Help button to view help
messages on the screen.
Click the Turn On or Turn Off button to turn on or off the control valves. This
procedure turns on or off the valves according to the user predefined valve on or off
status from Facts menu.


irrigation system based upon its knowledge base. The expert system operates continuously
to select an irrigation strategy and to schedule the application of irrigation, fertigation, and
cold protection. Data uncertainty management approaches were used to validate the sensor
readings. Conventional control routines were also developed so that irrigation and
fertigation could be applied according to user defined schedules with control flexibility and
few hardware requirements. A simulation model of the crop root zone was developed to
estimate crop water requirements to help the user to define irrigation schedules. A short
term prognosis of an irrigation requirement can be generated from the simulation. Database
and farm management utilities were also included in the system to assist the decision-making
of farm managers. Both laboratory and field tests showed that the integrated system worked
as expected as a management tool for irrigation, fertigation, and cold protection. The system
is highly automated and has the potential to improve microirrigation management, to achieve
water and energy savings, and to prevent water pollution due to improper fertigation
management.
xvi


191
Dempster, A. P. 1967. Upper and lower probabilities induced by a multivalued mapping.
Annals Math. Statistics 28(2):325-339.
Duke, H. R., L. E. Stetson, and N. C. Ciancaglini. 1990. Irrigation system control. In
Management of Farm Irrigation Systems, ed. G. J. Hoffman, T. A. Howell and
K. H. Solomon. ASAE Monograph, St. Joseph, MI, pp. 265-312.
Duke, H. R., D. F. Heermann, and M. C. Blue. 1984. Computer control of irrigation for
electrical load management. Trans, of the ASAE 27(2):597-602,608.
Durkin, J. 1994. Expert systems: an overview of the field. PC AI, January/February, pp
37-39.
Ege, R. K. and C. Stary. 1992. Designing maintainable, reusable interfaces. IEEE
Software 11:24-32.
Erbach, D. C. 1983. Measurement of soil moisture and bulk density, ASAE Paper No.
83-1553, St. Joseph, MI.
Feigenbaum, E. A. 1979. Themes and case studies of knowledge engineering. In Expert
System in the Micro-Electronic Age, ed. D. Michie. Edinburgh Univ. Press,
Edinburgh, Scotland, pp. 3-25.
Feigenbaum, E. A. 1982. Knowledge Engineering in the 1980's. Dept, of Computer
Science, Stanford University, Stanford, CA.
Feigenbaum, E. A., P. E. Friedland, B. B. Johnson, H. P. Nil, H. Schorr, H. Shrobe,
and R. S. Engelmore. 1994. Knowledge-based systems research and applications
in Japan, 1992. AI Magazine, Summer, pp. 29-43.
Feller, W. 1969. An Introduction to Probability Theory and Its Applications. Vol. 1. 3rd
ed. New York: John Wiley & Sons.
Ferguson, J., C. Taylor, G. Israel, and B. Summerhill. 1989. Citrus production survey:
young tree care. Citrus Industry, September, pp. 38-41.
Fernald, E. A. and D. J. Patton. 1984. Water Resources Atlas of Florida, Florida State
University, Tallahassee, FL.
Fikes, R. and T. Kehler. 1985. The role of frame-based representation in reasoning.
Communications of the ACM 28(9): 904-920.


67
Estimations of citrus water requirements provide only a general guideline for
irrigation. Actual irrigation applications vary due to several factors including (1) irrigation
management strategies, (2) irrigation system, (3) variability of rainfall and other climatic
factors, (4) soil characteristics, (5) planting density, and (6) crop growth characteristics.
Because of the variability inherent in these factors, it is difficult to create a general irrigation
schedule. Therefore, field measurement of soil-water content or maintaining a soil-water
budget is useful to determine crop water requirements.
6,3 Evapotranspiration and Management Allowed Depletion
Knowledge of ET is important to the management and design of irrigation systems.
Actual crop ET is determined from reference ET and experimentally obtained crop
coefficients:
ETa Kc ETQ 6-1
where Kc = crop coefficient,
ETa = actual ET, in/day, and
ET0 = reference ET, in/day.
Microirrigation systems supply water only to the immediate vicinity of each plant
being irrigated. Tree canopies shade only a portion of the soil surface area and intercept
only a portion of the incoming radiation. Conventional estimation of water requirements
assumes that part of the applied water will be lost to non-beneficial consumptive use, which
is the loss from evaporation of wetted soil surfaces and plant transpiration from undesirable


13.1.1 Main menu of CIMS 152
13.2.1 Submenu of the Facts 154
13.2.2 Irrigation block and valve definition 155
13.2.3 Fertigation block and valve definition 156
13.2.4 Initial facts of the expert system 157
13.2.5 Fertigation schedule 157
13.2.6 Submenu of the Expert main menu 159
13.2.7 Execution screen of the RTES 159
13.2.8 Screen of sensor readings and application status 160
13.2.9 Submenu to view application history 160
13.3.1 Control panel of the system 161
13.4.1 Submenu of user defined control schedules 162
13.4.2 User defined irrigation schedule screen 163
13.4.3 User defined fertigation schedule screen 163
13.4.4 Irrigation application screen 164
13.4.5 Irrigation valve on or off display 165
13.4.6 Fertigation dialog screen 166
13.4.7 Fertigation valve on or off display 167
13.4.8 Irrigation and fertigation dialog screen 168
13.5.1 Database control button 169
13.5.2 Delete dialog screen 179
13.5.3 Search dialog screen 170
xii


25
Non-monotonicity of data and uncertainty of missing data,
Continuous operation with high performance and guaranteed response times,
Ability to deal with asynchronous events,
Ability to communicate with the external environment (sensors and
effectors), and
Integration with procedural components.
3,3 Knowledge Acquisition
Knowledge acquisition is the transfer and transformation of problem-solving
expertise from some knowledge source to a computer program (Buchanan and Shortliffe,
1984). This is a process of eliciting,
structuring, and organizing knowledge
from human experts or other sources so
This process is the most important step,
this knowledge. The goal of knowledge
knowledge required by the system.
acquisition is to produce and verify the
from the sources and representation of
consists of elicitation of the knowledge
that the expertise can be encoded into
an ES (Figure 3.3). The process
Manual
Acquisition
and normally, the most time-consuming Figure 3.3. Knowledge acquisition cycle


52
4.6.7 Overview of the Hardware
The selected hardware performs data collection and controls the irrigation system.
The system is an on-line irrigation controller (Figure 4.6). Soil moisture sensors
(tensiometers) measure the soil-water potential. An automated weather station measures
meteorological data, which provides current rainfall and sufficient data for ET estimation.
Sensor readings and weather data are stored in the data logger (CRIO). The computer
retrieves the data from the data logger at each given time interval. Connection between the
computer and the data logger can be either wire or radio links depending on the distance and
hardware cost. After the sensor data are input to the computer, the expert system uses the


124
systems can be validated by running test cases and comparing results against known results
or expert opinion. Field tests place expert systems in the field, and then seek to perceive
performance errors as they occur.
9.5.1 Predictive Tests
Predictive tests must be conducted at both the system and component level. Many
test cases for a variety of control scenarios were generated to test the reasoning process and
data uncertainty management. The system was validated by running the test cases against
known results and expert opinions. The test results showed that CEMS performed the
reasoning process as was expected. Some sample test cases and reasoning results of the
system are listed in Appendix B. These test cases were also used to debug and verify the
system.
9.5.2 Field Tests
Test site description
Field tests were conducted in the Kresdorn research grove, which is funded by the
Mid-Florida Citrus Foundation and is located at the Conserv II project at Orlando, Florida.
The Kresdorn Grove is a 20-acre grove designed to conduct experiments related to irrigation
demand, fertigation, herbigation, pest management, in-row spacing, and cold protection.
Three-year-old Ambersweet orange trees are planted at this site. Irrigation is applied using
a microirrigation system (micro spray). One emitter was installed per tree. It provides a
360-degree fan pattern of approximately 5 feet diameter. The flow rate of each emitter is
about 16 gallons per hour. Soil type at this site is classified as Candler fine sand.


94
downloading action is accomplished through a conventional program to access the data
logger. After the sensor data are downloaded, the data must be rearranged to the CLIPS
(expert system shell) data format to accomplish the reasoning process.
8.3.2 Uncertainty Management of the Sensor Data
One of the important characteristics of RTES is that the system reasons with missing
data. In practice, sensors may provide bad or unreliable data when they have failed or as
they begin to fail. The sensor outputs can also vary with soil characteristics and nonuniform
root zones. Thus, a lack of adequate information may jeopardize the decision-making
process. This may prevent the outcome from being the best decision and may also result in
a bad decision. Because the sensor data are crucial to the reasoning process, uncertainty
management is essential so that the system operates reliably with missing data.
A physical approach to increase reliability of the sensor reading is using redundant
sensors, but this adds more hardware costs. The approach used here is data uncertainty
management. A certainty factor (CF) analysis is applied to the sensor data. The values of
CF range from -1 to 1 indicating two extremes: false (invalid) and true (valid). An initial
assumption was made that the sensor reading was valid. Thus, an initial high CF value
(0.98) was assigned to each sensor reading. Rules were developed to validate the sensor
readings. These rules interpret the data to detect deviations from normal or desired
behavior. The CF values are decreased when the sensor data are identified as abnormal.
Thus, a sensor reading may be rejected from the reasoning process when its CF value is at
a less than acceptable confidence level. The sensor readings were checked in two ways: (1)
checking the range of sensor readings, and (2) checking the variation among sensor readings.


Table 7.1 Pounds of nitrogen fertilizer to be applied to furnish the nitrogen
requirement of orange and grapefruit trees under normal conditions.
Fruit
production
(boxes/acre)
Pounds of nitrogen (N)a
needed per acre per year
Pounds of nitrogen fertilizer needed per acre per yearb
15.5 %N
33.5 %N
45.0% N
Orange
Grapefruit
Orange
Grapefruit
Orange
Grapefruit
Orange
Grapefruit
<200
100
60
645
485
300
225
220
165
300
120
90
775
580
360
270
265
200
400
160
120
1030
775
480
360
355
265
500
200
150
1290
965
600
450
445
335
600
240
180
1580
1160
715
538
535
400
700
280
210
1805
1355
835
620
620
465
>800
300
240
1935
1450
895
670
665
500
a. Nitrogen needed is based on 0.4 pound per box of fruit for oranges and 0.3 pound per box for grapefruits. In most
cases, one should not use less than 100 pounds or more than 300 pounds of nitrogen for orange trees and not less
than 60 pounds or more than 240 pounds for grapefruit trees per acre per year.
b. The figures given should be divided by the number of applications per year to obtain pounds per application.


132
operate. In addition, CIMS also provides farm databases and tools to estimate crop
evapotranspiration and irrigation duration that can be used by the system manager.
CIMS was validated by (1) face validation, (2) predictive validation, and (3) field
tests. Face validation is only a preliminary approach. Experts and potential users were
requested to evaluate the system against their opinions. For predictive validation, many test
data for variety of control scenarios were created to test the system's functionality and
reasoning behavior. Field tests were conducted mainly to test the control components and
reasoning process. The system was tested at the Kresdorn Grove which is located at the
Conserv II water reuse project in Orlando, Florida. The control routines of CIMS have been
successfully implemented and tested at the site for two years. A short-term field test of the
reasoning process of the knowledge base has been conducted at the site. Both the predictive
and field tests showed that the expert system makes decisions as expected to turn on or off
the control system in response to variations of the input data. Irrigation and fertigation
control by user defined schedules worked reliably. More field experiments are needed to
study the long-term performance of the system, especially to assess the crop responses of
different irrigation strategies.
In conclusion, CIMS integrates expert system with control techniques for citrus
microirrigation management. The system provides several options of irrigation management
and it has potential to improve citrus microirrigation management. Therefore, CIMS is new
in its design and application. This work mainly accomplished the following:


CHAPTER 10
SUMMARY AND CONCLUSIONS
As personal computers (PC) become increasingly common, the demand for
computerized farm management tools is increasing. A citrus microirrigation management
system (CIMS) was developed using real-time expert system (RTES) and conventional
control techniques. The CIMS is a PC-based system integrated with RTES, conventional
control, simulation, database, and irrigation management utilities. The system can respond
to external environmental variations and operate continuously in order to make rapidly
decisions without human intervention on matters of irrigation, fertigation, and cold
protection. An automated weather station and soil moisture sensors were used to collect
real-time field data. A knowledge base, which represents the heuristic knowledge of experts
required for the decision-making of citrus irrigation management, was developed for the
reasoning process. Control programs were developed to turn on or off the solenoid valves
and pumps based upon the results of the reasoning process.
With the real-time weather data and soil-water potential, the RTES provides a high
level of automation for microirrigation management. The system has the potential to reduce
labor costs and to improve water, chemicals, and energy conservation. CIMS can be
operated without human presence or may be operated remotely via telephone, cellular or
130


160
making process. The reasoning process will continue until interrupted by a user. To
interrupt the continuous process, double click left button of the mouse. Data from
the soil moisture sensors, the weather station, and system on or off status of an
application are displayed as Figure 13.2.8 during the real-time reasoning process.
Figure 13.2.8. Screen of sensor readings and application status.
View application history
Since the RTES runs continuously, control actions can occur any time due to the
variation of input data and the reasoning process. The user may not know what has
happened in the past. Therefore, the systems automatically store control actions to a history
log file. To view the application history, click View Application on the submenu of Expert
(Figure 13.2.9). Click either Irrigation, Fertigation, or Freeze Protection to view the past
Figure 13.2.9. Submenu to view application history.


166
Figure 13.4.6. Fertigation dialog screen.
Procedures to apply the user defined fertigation schedule are
Click Apply Fertigation on the submenu of Scheduling. A dialog screen appears to
define preinjection time and flush time (Figure 13.4.6).
Click the Cancel button to exit the dialog screen.
Enter preinjection and flush time in minutes showed in Figure 13.4.6.
Click the Proceed button to apply fertigation according to the user defined schedule.
A screen display on or off status of the control valves and pump (Figure 13.4.7).
Double click left button of the mouse to interrupt the system.


145
Test case 9
(TREE-STATUS YOUNG)
ID2-3
(sensor-reading idl 6 9 12 9 24 16)
(sensor-reading id2 6 9 12 9 24 16)
Test case 10
(TREE-STATUS YOUNG)
(sensor-reading idl 6 10 12 10 24 10)
(sensor-reading id2 6 19 12 10 24 10)
Test case 11
(TREE-STATUS YOUNG)
(sensor-reading idl 6 12 12 10 24 10)
(sensor-reading id2 6 12 12 10 24 19)
Test case 12
(TREE-STATUS YOUNG)
(sensor-reading idl 6 10 12 10 24 10)
(sensor-reading id2 6 10 12 10 24 19)
Test case 13
(TREE-STATUS MATURE)
(CRITICAL GROWTH)
(sensor-reading idl 6 17 12 9 24 9)
(sensor-reading id2 6 16 12 9 24 9)
(time-constrain no)
(weather-constrain no)
Test case 14
(TREE-STATUS MATURE)
(CRITICAL GROWTH)
(sensor-reading idl 6 9 12 9 24 17)
(sensor-reading id2 6 9 12 9 24 16)
(time-constrain no)
(weather-constrain no)
Reject data from ID 1-3 and
Not turn on
Reject data from ED 1-1
Not turn on
Reject data from ID2-3
(CF=0.7)
Turn on the system by ID1-1
or ID2-1
Reject data from ID2-3
Not turn on
Data OK
Turn on by sensor ID1-1 or
ED2-1
Data OK
Turn on by sensor ID 1-3 or
ID2-3


6
microirrigation management. Since the expert system must be operated in the real-time
domain, control hardware is required. The control process can be accomplished by using
conventional programs. As an RTES applied to microirrigation management, its reasoning
process is not as time critical as military applications. In other words, the system is not a
hard RTES. The use of the term RTES is to distinguish the system from expert systems in
which time is not a factor at all or which acquire data only from static databases.
The components of farm databases, crop water requirement simulation, and
computer-controlled irrigation systems have been successfully applied. The integration of
an expert system with simulation models, databases, and user defined control needs to be
resolved in this study. This integration can rely heavily on the structure of the system
design, functionality of the expert system shell, and design of the user interface. With
modem software development tools, operating systems with multitasking capabilities, and
object-oriented software design, this integration can be achieved. After the system is
developed, system validation can be conducted by running generated test cases, expert
evaluation, field tests, or a simulation approach.


50
storage, and control functions. A multitasking operation system allows simultaneous
communication and measurement functions.
The device is protected in a sealed, rugged, stainless steel canister in order to be
installed for out-door conditions. The input signal can be either analog or digital. The
maximum analog input ranges from -2.5 V to +2.5 V. The interface to the CRIO can be a
portable CR10KD Keyboard Display or a computer. The PC communication software to
access data from the CRIO is supplied by the vendor.
4.6.5 PC Digital Input/Output Board
A digital input/output (I/O) board was used as the interface between the computer
and the irrigation control board. The PC-DI072, a general digital I/O card from the
Industrial Computer Source, was used for this application. The card can be applied to relay
monitoring, control, sensing switches, security systems, and energy management. This
board provides user selectable buffered inputs and outputs based on the 8255 chips by Intel.
Major features of the PC-DIO are (1) 72 channels of digital I/O, (2) interrupt and interrupt
disable capability, and (3) four or eight bit groups independently selectable for I/O. The
output source current (output high) is 15 mA. The base address used in this application is
H310, H314, and H318 to set the ports as output.
4.6.6 Irrigation Control Board
An irrigation control board was used to activate or deactivate the automated
irrigation valves. The control board consists of 70 solenoid relays OACQ5 (Figure 4.5).
The relays are installed on a relay rack (PB24Q). Both the relay and the relay rack are
manufactured by Opto22. Each relay controls a specific electronic valve of the irrigation


CHAPTER 3
GENERAL EXPERT SYSTEM CONCEPTS
3.1 Expert Systems
One of the most significant results of artificial intelligence (AI) research to date is
the expert system (ES). An ES is a computer system that emulates the decision-making
capability of a human expert. Feigenbaum (1982), an early pioneer of ES technology,
defined an ES as
an intelligent computer program that uses knowledge and inference
procedures to solve problems that are difficult enough to require significant
human expertise for their solution, (p. 15)
An ES attempts to perform like a human expert to solve problems. Instead of relying on
statistical or algorithmic methods, ESs solve problems by applying a symbolic knowledge
representation of human expertise. Consequently, ESs try to encode domain-specific
knowledge rather than comparatively domain-free methods derived from computer science
or a mathematical approach. Its application is normally restricted to a specific problem
domain or well-defined domain. Applications of expert systems include medical, industrial,
agricultural, and space technologies. ESs provide the advantages of increased availability,
reliability, fast response, and multiple expertise. The structure of an ES can be a rule-base,
frame, model or other approaches. The major components of an ES are the inference engine,
knowledge base, user interface, and knowledge acquisition facility (Figure 3.1).
19


27
system must be identified. Third, the information to aid the knowledge understanding and
verification by the developer needs to be classified and organized. Fourth, the detailed
functional capabilities of the system have to be laid out. Then, a description of the general
functional capability must be given in detail. This includes studying the flow of knowledge
from both the developer's and user's viewpoint. Fifth, the system task needs to be defined.
Questionnaires mean that the knowledge engineer prepares some question sheets and
lets the expert answer them. This approach can be used in combination with the interview
approach. Observation allows the knowledge engineer to learn how the expert solves a real
problem.
Intuition refers to how a knowledge engineer attempts to be a pseudo-expert and
applies his knowledge to the domain. This process can only serve as an aid to knowledge
acquisition because the knowledge engineer is not a true expert and lacks expertise in the
domain.
Tools for the processes of automated knowledge acquisition have been developed
by researchers (Quinlan, 1986; Michalski et al., 1986). These tools attempt to help bridge
the gap between the expert/knowledge engineer and computer implementations using
learning algorithms. One of the earliest and best-known algorithms is ID3 (Quillan, 1986).
However, these tools provide very limited capabilities in solving real-world problems.
3.3.2 Potential Problems
Interviews are the most common approach for knowledge acquisition, but this
process is not simple because, in general, experts do not structure their decision-making in


158
Define fertigation schedule
Fertigation schedules can be defined as two formats: by date or by day of week.
Click Fertigation Schedule to open the editing window (Figure 13.2.5).
Enter the starting and ending time. To omit a schedule, use the semi-colon in
front of each schedule.
A sample fertigation schedule by day of week is shown below.
(FTSCH1 MONDAY START-TIME 3 10 END-TIME 4 10)
FTSCH1
schedule ID,
MONDAY
day of the week,
START-TIME 3 10 -
fertigation starting hour and minute, and
END-TIME 4 10
fertigation ending hour and minute.
A sample of the fertigation schedule by date showed below.
(FTSCH2 1994 11 23 START-TIME 3 20 END-TIME 3 30)
FTSCH2
fertigation schedule ID,
1994 11 23
year, month, and day,
START-TIME 3 10 -
fertigation starting hour and minute, and
END-TIME 4 10
fertigation ending hour and minute.
Note that all the text in the facts must be upper case and each attribute must be
separated by at least one space.
13 2,2 Execution of the RTES
Before executing the RTES, initial facts must be defined. In particular, the on-off
status of control valves must be defined for each irrigated block. To run the RTES, click


13
2.4 Irrigation Control
Irrigation system control includes a variety of topics, ranging from on-stream storage
and water diversions to agronomic practices (Duke et al., 1990). The control topics may
relate to (1) hydraulic, (2) mechanical, (3) electro-mechanical, (4) electronic, and (5)
computerized control (Duke et al., 1990). The major irrigation control modes are (1) on-off
control, (2) stepwise control, and (3) continuous control (Phene, 1986). Computerized
systems have shown great potential in irrigation control and farm management. This is
because a single hardware configuration can serve a wide range of control functions, and
control strategies can be easily modified by software modifications.
Studies (Phene et al., 1973; Phene and Howell, 1984; Phene, 1989) have been
conducted of irrigation control using soil moisture sensors. Sensors were used in a feedback
mode to maintain a nearly constant soil moisture content in the root zone. They concluded
that the performance of the irrigation controller depended on four basic factors: (1) adequate
operation of the system's control hardware, (2) the proper algorithm for the system's
software, (3) a reliable soil moisture sensor installed in the field, and (4) adequate operation
of the system's output, the solenoid valves, the pressure regulators, flow meter, and filter.
Further studies (Phene et al., 1989b) indicated that an irrigation controller should have the
following characteristics to monitor soil matric potential in real-time and control irrigation
systems: (1) ability to sample the sensor data automatically, (2) means of comparing the
sensor output to a threshold value, and (3) ability to control and monitor irrigation devices.


139
Rain (cm) -o- S1-6" _^S1-12" S2-6" _H_S2-12"
Figure 12.1. Tensiometer readings on Juilan day 213, 1994.
Note: SI-6"
51-12"
52-6"
S2-12"
Tensiometer SI-6"
Tensiometer at site 1 with 6" depth.
Tensiometer at site 1 with 12" depth.
Tensiometer at site 2 with 6" depth.
Tensiometer at site 2 with 12" depth,
was failed at 1400 minutes on day 213.


32
3.4.4 Rules
The classic and most common way of knowledge representation is the use of a
generic form of IF-THEN rules. The rules may have the simple form:
IF antecedent(s), THEN consequence(s).
Rules provide a readily understandable form. The antecedents define a pattern to be
matched against the content of the working memory, which is a global database of facts used
by the rules. The rule is fired if such a pattern is matched. Thus, the consequences change
the working memory and play the inferences in an ES. Each rule is an independent unit in
the entire knowledge base.
Rule-based systems have been widely applied in many areas and are often
misunderstood because their IF-THEN structure is similar to the condition structure in
conventional programming language. Two factors distinguish rules from a conditional
statement in conventional languages (Ringland and Duce, 1988):
The antecedents are expressed as a pattern rather than a boolean expression;
and such antecedents can be in a simple or very complex form.
A rule-based system allows separation of knowledge from control of how the
knowledge is applied. The condition of conventional language is a flow of
control. The flow in rules does not pass from one rule to the next in lexical
sequence, but is determined completely separately through the inference
engine.
Rule-based systems are the most widely used production systems. All knowledge
in the knowledge base is represented in a single uniform format, and each rule is a distinct


10
Although irrigation scheduling has been studied in various ways for a long time,
research is needed to reduce the consumption of water and energy and to increase
profitability through better scheduling. Numerous studies have been conducted on irrigation
scheduling (Pleban et al., 1983, 1984; Zoldoske, 1988; Rogers and Elliott, 1989; Shayya et
al., 1990; Protopapas and Georgakakos, 1990). Monitoring of soil water content, crop
growth, and weather conditions is important for irrigation scheduling.
2.3.1 Monitoring Method
Monitoring methods are primarily based on either crop or soil measurement.
Monitoring can rely on instruments or one's intuition. Soil moisture is usually monitored
by using a sensor to measure soil water potential (Campbell and Campbell, 1982). Irrigation
is applied when the monitored crop or soil data reach some critical value. Irrigation
scheduling can also rely on the monitoring of weather data (Howell et al., 1984).
Soil moisture sensors are one of the major tools used to assist decision-making on
irrigation water applications. Tensiometers and gypsum blocks are widely used in the field
(Cary and Fisher, 1983). Augustine and Snyder (1984) and Snyder et al. (1984) used
tensiometers to schedule irrigation for bermudagrass turf. Their results showed that
irrigation water savings of 42 to 95 percent were obtained in sensor controlled plots over
conventionally irrigated plots. A study using tensiometers to schedule cotton drip irrigation
was conducted by Wierenga et al. (1987). In all of these studies tensiometers were
successfully used for irrigation scheduling.


163
Figure 13.4.2. User defined irrigation schedule screen.
Figure 13.4.3. User defined fertigation schedule screen


133
Integrated a variety of water management techniques into a single system
including an expert system, user defined irrigation control, crop water
requirement simulation, farm databases, and irrigation management utilities.
Developed a new technique for citrus microirrigation management in the real
time domain. This system provides a highly automated tool for irrigation
management and has potential to improve irrigation management.
Potentially, this technique can be applied to other sites or crops.
Applied Markov chain probability of rainfall in the irrigation decision
process. Although the Markov chain probability of rainfall is a well-known
approach, the system applied the approach to an actual expert system in the
process of irrigation decision-making. The simulated results showed that
irrigation decision based on the Markov chain probability of rainfall can
achieve water savings, but the amount of savings is limited. Further studies
are needed to investigate what amount of water can be saved by using the
approach.
Applied the expert system shell CLIPS to a real-time problem domain. Since
the expert system shell CLIPS cannot be directly applied to deal with a real
time problem, efforts were made to achieve reasoning with time and
continuous operation of the system. In addition, uncertainty management of
the sensor data was conducted to increase the system reliability.
In terms of irrigation management, CIMS is in an early stage of development.
Problems faced by decision makers can be much more complex than CIMS addresses.


103
Figure 8.10. Decision process to start an irrigation (criteria I) and sensor readings to
trigger an irrigation for trees during different growth stages.
Table 8.4 Sensor readings and constraints to start an irrigation.
Young
Mature (critical)
Mature (non critical)
Sensor
-8 < S¡< -10 cb
-12 < S¡< -15 cb
-22 < Sj < -25 cb
Time
00:00 am 10:00 am
00:00 am -10:00 am
00:00 am 10:00 am
Wind Speed
< 5 mph
< 5 mph
< 5 mph
RH
> 70%
> 70%
> 70%
Where Si denotes sensor output at depth i (i = 6", 12", and 24").


A REAL-TIME EXPERT SYSTEM
FOR CITRUS MICROIRRIGATION MANAGEMENT
By
JIANNONG XIN
A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
UNIVERSITY OF FLORIDA
1995

Copyright 1995
by
Jiannong Xin

ACKNOWLEDGEMENTS
I would like to express my sincere gratitude to the supervisory committee. Special
thanks go to Dr. Fedro S. Zazueta, supervisory committee chairman, for his advice and
encouragement during the study and for the opportunity to study as a graduate research
assistant. Special thanks go to Dr. Allen G. Smajstrla, supervisory committee member, for
making himself available on many occasions and for contributing his expertise toward this
study. I would also like to thank the supervisory committee members, Dr. James W. Jones,
Dr. Pierce H. Jones, Dr. Douglas D. Dankel, II, and Dr. Louis H. Motz, for their advice and
support during this study.
Special thanks also go to Dr. Thomas A. Wheaton and Dr. Lawrence R. Parsons of
the Lake Alfred Citrus Research Center for their expertise, advice, and time toward this
study.
I would like to extend my appreciation to the following people for their evaluation
and comments on the software: Dr. Robert M. Peart and Dr. Dorota Z. Haman of the
Agricultural and Biological Engineering Department, University of Florida; Dr. James J.
Ferguson and Dr. J. David Martsolf Jr. of the Horticultural Department, University of
Florida; and Ms. Cynthia Moore, St. Johns River Water Management District, Florida.
Finally, I would like to thank my wife, Qiuping Jian, for her support during this
study.
iii

TABLE OF CONTENTS
ACKNOWLEDGMENTS iii
LIST OF TABLES viii
LIST OF FIGURES x
ABSTRACT xvii
CHAPTER 1 INTRODUCTION 1
1.1 Statement of the Problem 1
1.2 Objectives of the Dissertation 5
CHAPTER 2 REVIEW OF THE LITERATURE 7
2.1 Citrus Irrigation in Florida 7
2.2 Soil Moisture Sensors 8
2.3 Irrigation Scheduling 9
2.3.1 Monitoring Method 10
2.3.2 Computer Simulation 11
2.4 Irrigation Control 13
2.5 Expert Systems in Agriculture 15
2.6 Summary 17
CHAPTER 3 GENERAL EXPERT SYSTEM CONCEPTS 19
3.1 Expert Systems 19
3.1.1 Inference Engine 20
3.1.2 Knowledge Base 21
3.1.3 User Interface 22
3.2 Real-Time Expert Systems 22
3.2.1 Why Use an RTES? 24
3.2.2 Characteristics of RTESs 24
IV

3.3 Knowledge Acquisition 25
3.3.1 Basic Approaches 26
3.3.2 Potential Problems 27
3.3.3 Practical Issues 28
3.4 Knowledge Representation 29
3.4.1 Semantic Network 30
3.4.2 Frame 30
3.4.3 Objects 31
3.4.4 Rules 32
3.5 Rule-Based Expert Systems 33
3.5.1 Rule-Based Architectures 34
3.5.2 Uncertainty Management 35
CHAPTER 4 SYSTEM SPECIFICATION AND DESIGN 37
4.1 Domain of the Problem 37
4.2 Requirements Specification 38
4.2.1 Goal of the System 38
4.2.2 System Inputs 38
4.2.3 System Outputs 39
4.3 Knowledge Specification 40
4.4 Knowledge Representation Paradigm 41
4.4.1 Reasoning Method 41
4.4.2 System Performance Requirements 42
4.5 Development Tools 42
4.5.1 Expert System Shells 43
4.5.2 CLIPS 44
4.6 Hardware Specification 44
4.6.1 Soil Moisture Sensor 45
4.6.2 Personal Computer 49
4.6.3 Automated Weather Station 49
4.6.4 Data Logger 49
4.6.5 PC Digital Input/Output Board 50
4.6.6 Irrigation Control Board 50
4.6.7 Overview of the Hardware 52
4.7 Paradigm of the Real-Time Expert System 53
CHAPTER 5 PROBABILITY OF RAINFALL 55
5.1 Introduction 55
5.2 Markov Chain 56
5.3 Rainfall Data 58
5.4 Frequency of Rainfall 59
v

5.5 Statistical Test 60
5.6 Irrigation Decision with Rainfall Probability 62
CHAPTER 6 CITRUS IRRIGATION SCHEDULING 64
6.1 Introduction 64
6.2 Citrus Water Requirements 64
6.3 Evapotranspiration and Management Allowed Depletion 67
6.4 Irrigation Depth and Duration 69
6.5 Soil-Water Budget 71
6.6 Irrigation Scheduling Using Tensiometers 72
6.6.1 Tensiometer Installation Depth 73
6.6.2 Soil-Water Potential and Allowable Water Depletion 74
CHAPTER 7 CITRUS COLD PROTECTION AND FERTIGATION 77
7.1 Introduction 77
7.2 Cold Protection Application 78
7.2.1 Principle of Cold Protection 78
7.2.2 Critical Application Temperature 79
7.2.3 Water Application Rate 80
7.3 Fertigation 80
7.3.1 Application of Fertigation 81
7.3.2 System Components of Fertigation 84
7.3.3 Fertilizer Materials 85
CHAPTER 8 CONSTRUCTION OF THE KNOWLEDGE BASE 88
8.1 Introduction 88
8.2 The Process of Control and Reasoning 89
8.3 The Sensor Data 93
8.3.1 Download the Sensor Data 93
8.3.2 Uncertainty Management of the Sensor Data 94
8.4 Irrigation Management 100
8.4.1 Irrigation Strategies 100
8.4.2 Criteria for Starting an Irrigation 102
8.4.3 Criteria for Stopping an Irrigation 105
8.5 Cold Protection 108
8.5.1 When to Turn On 109
8.5.2 When to Turn Off 109
8.6 Fertigation 110
vi

CHAPTER 9 SYSTEM IMPLEMENTATION AND TESTS 112
9.1 Function Requirements of CEMS 112
9.2 Module Design of CEMS 112
9.2.1 Expert System Module 114
9.2.2 Control Panel 115
9.2.3 Scheduling 116
9.2.4 Database 117
9.2.5 Simulation 117
9.2.6 Tools 118
9.2.7 Help 118
9.2.8 User Interface 119
9.3 Data and Message Passing of QMS 119
9.3.1 Data Flow of the RTES Module 119
9.3.2 Data Requirements of the Simulation Module 120
9.3.3 Data Requirements of the Scheduling Module 121
9.4 Maintenance of CEMS 121
9.5 System Tests of CEMS 123
9.5.1 Predictive Tests 124
9.5.2 Field Tests 124
9.5.3 Simulated Crop Water Use 125
CHAPTER 10 SUMMARY AND CONCLUSION 130
APPENDIX A SAMPLE SENSOR DATA 135
APPENDIX B SAMPLE TEST CASES OF KNOWLEDGE BASE 142
APPENDIX C USER'S GUIDE OF CIMS 151
LIST OF REFERENCES 189
BIOGRAPHICAL SKETCH 204
vii

LIST OF TABLES
Table Page
4.1 System input and output requirements 39
4.2 Characteristics of pressure transducer model 141PC 47
5.1 Markov chain wet-day frequency 59
5.2 Results of paired t-test for rainfall probabilities within each season ... 61
6.1 Citrus irrigation water requirements in central Florida 65
6.2 Citrus crop coefficients and recommended MAD in Florida 68
6.3 Average soil-water content for Candler fine sand by volume 74
6.4 Estimated soil-water tension in corresponding to soil-water
depletion for Candler fine sand 75
7.1 Pounds of nitrogen fertilizer to be applied to furnish nitrogen
requirement of orange and grapefruit trees under normal conditions 83
7.2 Solubility of common fertilizers in water 87
8.1 Criteria for checking possible sensor Failure B 97
8.2 Criteria for checking possible sensor Failure C for sensors at the
same depth from different locations 97
8.3 A sample propagation of CF 98
8.4 Sensor readings and constraints to start an irrigation 103
Vlll

8.5 Critical sensor readings (Criteria II) to start an irrigation 105
9.1 Data input of QMS 121
9.2 Accumulated citrus net irrigation requirements and number of
irrigations for 22 years in central Florida 127
IX

LIST OF FIGURES
Figure Page
3.1 Major components of an expert system 20
3.2 Major components of an RTES 23
3.3 Knowledge acquisition cycle 25
3.4 The architecture and execution cycle of rule-based systems 34
4.1 A regular tensiometer and a tensiometer with micro-pressure
transducer 46
4.2 Pressure transducers (Model 141PC) from Micro Switch 46
4.3 Tensiometer calibration equipment 48
4.4 Tensiometer calibration curve 48
4.5 Solenoid control relays of the irrigation system 51
4.6 Hardware layout of the control system 52
4.7 Paradigm of the real-time expert system 53
5.1 Annual rainfall distribution in Orlando from 1952 to 1992 60
6.1 Irrigation by threshold of soil-water content 72
6.2 Soil-water content versus soil-water tension for Candler fine sand .... 76
7.1 Major components of a fertigation system 85
8.1 Major inputs and outcomes of the expert system 88
x

8.2 Decision flow of the expert system 90
8.3 Paradigm of the knowledge base 91
8.4 Process of downloading weather and soil moisture sensor data 93
8.5 CF propagation by checking range of sensor data 95
8.6 CF propagation for sensors SI and S3 96
8.7 CF propagation for sensor S2 96
8.8 Process of selecting valid sensor readings from different locations .... 99
8.9 Decision process to use a full or deficit irrigation strategy 101
8.10 Decision process to start an irrigation (criteria I) and sensor readings
to trigger an irrigation for trees during different growth stages 103
8.11 Decision process (criteria II) to start an irrigation and critical sensor
readings for trees during different growth stages 104
8.12 Decision process to stop an irrigation 106
8.13 Cold protection decision processes based on the critical air
temperature 108
9.1 Program modules of CIMS 113
9.2 Control panel of CIMS 116
9.3 Irrigation system database 117
9.4 Data flow of CIMS 120
9.5 Structure of the knowledge base and initial facts 123
12.1 Tensiometer readings on Juilan day 213, 1994 139
12.2 Tensiometer readings on Juilan day 277, 1994 140
12.3 Tensiometer readings on Juilan day 278, 1994 141
xi

13.1.1 Main menu of CIMS 152
13.2.1 Submenu of the Facts 154
13.2.2 Irrigation block and valve definition 155
13.2.3 Fertigation block and valve definition 156
13.2.4 Initial facts of the expert system 157
13.2.5 Fertigation schedule 157
13.2.6 Submenu of the Expert main menu 159
13.2.7 Execution screen of the RTES 159
13.2.8 Screen of sensor readings and application status 160
13.2.9 Submenu to view application history 160
13.3.1 Control panel of the system 161
13.4.1 Submenu of user defined control schedules 162
13.4.2 User defined irrigation schedule screen 163
13.4.3 User defined fertigation schedule screen 163
13.4.4 Irrigation application screen 164
13.4.5 Irrigation valve on or off display 165
13.4.6 Fertigation dialog screen 166
13.4.7 Fertigation valve on or off display 167
13.4.8 Irrigation and fertigation dialog screen 168
13.5.1 Database control button 169
13.5.2 Delete dialog screen 179
13.5.3 Search dialog screen 170
xii

13.5.4 Browse dialog window 171
13.5.5 Submenu of the Database main menu 172
13.5.6 Farm database screen 172
13.5.7 Weather database screen 173
13.5.8 Irrigation database screen 173
13.5.9 Crop database screen 173
13.5.10 Crop coefficient database screen 174
13.5.11 Soil database screen 174
13.6.1 Simulation submenu 175
13.6.2 Screen to set initial condition of the simulation 176
13.6.3 Simulation dialog window 176
13.6.4 Irrigation prognosis from the simulation 177
13.6.5 Simulated soil-water content 178
13.7.1 Submenu of Tools 179
13.7.2 Read weather data from weather station 180
13.7.3 ET method dialog window 180
13.7.4 Penman ET screen 181
13.7.5 Blaney-Criddle ET screen 181
13.7.6 Modified Blaney-Criddle ET screen 182
13.7.7 Stephens-Stewart ET screen 182
13.7.8 Estimate irrigation duration screen 183
xiii

13.7.9 Help screen of irrigation duration 183
13.7.10 A dummy field layout map 184
13.8.1 Submenu of the Help 185
13.8.2 CIMS help screen 185
13.8.3 Calculator 186
13.8.4 Calendar and diary screen 186
13.8.5 Clock 187
13.8.6 Text editor 187
13.8.7 Puzzle 187
13.8.8 About the CIMS 187
13.8.9 More about the CIMS 188
13.8.8 Screen to quit from the CIMS 188
xiv

Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy
A REAL-TIME EXPERT SYSTEM
FOR CITRUS MICROIRRIGATION MANAGEMENT
By
Jiannong Xin
August, 1995
Chairman: Dr. Fedro S. Zazueta
Major Department: Agricultural and Biological Engineering
Elaborate techniques are commonplace in modern farm management and
microirrigation scheduling for citrus. Water management practices involve complex
decisions and daily operations that are affected by water and nutrient requirements of the
trees, temporal distribution of rainfall, and extreme weather conditions. A computer-based
system (CIMS) was developed using a real-time expert system (RTES) and conventional
control techniques to assist citrus microirrigation, cold protection, and fertigation
management. The system integrates water management technologies into an effective
control technique that can be used as a tool by farm managers. CIMS combines the RTES,
conventional control, and irrigation management tools into a single system to help the
decision-making of irrigators. On-site soil moisture sensors and an automated weather
station provide data to the system. CIMS activates or deactivates the control devices of an
xv

irrigation system based upon its knowledge base. The expert system operates continuously
to select an irrigation strategy and to schedule the application of irrigation, fertigation, and
cold protection. Data uncertainty management approaches were used to validate the sensor
readings. Conventional control routines were also developed so that irrigation and
fertigation could be applied according to user defined schedules with control flexibility and
few hardware requirements. A simulation model of the crop root zone was developed to
estimate crop water requirements to help the user to define irrigation schedules. A short
term prognosis of an irrigation requirement can be generated from the simulation. Database
and farm management utilities were also included in the system to assist the decision-making
of farm managers. Both laboratory and field tests showed that the integrated system worked
as expected as a management tool for irrigation, fertigation, and cold protection. The system
is highly automated and has the potential to improve microirrigation management, to achieve
water and energy savings, and to prevent water pollution due to improper fertigation
management.
xvi

CHAPTER 1
INTRODUCTION
1.1 Statement of the Problem
Modem farm management involves complex decisions and daily operations that are
affected by water and nutrient requirements of crops, temporal distribution of rainfall,
environmental protection, and extreme weather conditions. In recent years, increasing costs
of energy, increasing water demands from non-agricultural users, and adverse weather cycles
are forcing the agricultural industry to use new technologies to improve water management
capabilities and to increase the efficiency of resources used in production.
Irrigation is the largest consumer of fresh water in the world. In Florida, agriculture
accounts for over 40 percent of total fresh water use (Fernald and Patton, 1984) -- about
3,000 million gallons per day. Citrus, one of the major crops in Florida, is a billion dollar
industry and consumes millions of gallons of water for irrigation each year. Thus, even
modest increases in water use efficiency will result in substantial water savings and reduce
energy cost.
Operational costs for irrigation are increasingly due to increasing energy costs.
Agricultural energy consumption varies greatly among the different commodities and
agricultural practices in Florida. For citrus, an energy survey (Stanley et al., 1980) showed
that 11,588.90 billion BTUs were used in 530,000 acres of production with 32.2 percent of
1

2
the energy used in irrigation. Currently, close to 16,000 billion BTUs (the energy equivalent
of approximately 128 million gallons of gasoline) (Zazueta et al., 1994) are needed by about
750,000 acres of citrus production. If 32.2 percent of this energy is used in irrigation, about
5,152 billion BTUs are consumed only for irrigation. A conservative 10 percent reduction
in energy use by using better control systems would result in a savings of 515 billion BTUs.
Fertilizer and chemical applications are common in agricultural practice. Because
water is used to convey many of the chemicals through microirrigation systems, the
efficiencies of fertigation and chemigation are directly related to the water application
efficiency. Furthermore, improper chemical applications may result in adverse
environmental impacts. Ground and surface water can be contaminated if applied chemicals
are transported out of the crop root zone due to inadequate management.
It is generally accepted that improved management techniques are necessary to
increase water use efficiency, particularly in the scheduling and application of irrigation.
Current irrigation management practice attempts to satisfy crop water demands and relies,
for the most part, on manual operation or timers for control of the irrigation system. Many
irrigation schedules rely on calendars, evapotranspiration (ET) estimation, and the grower's
experience. Water can be wasted due to poor irrigation management, and irrigation may
even be applied during rainfall when a preset timer is used to control the system (Xin et al.,
1993).
Computer-based irrigation scheduling has received much attention. Simulation
models have been developed for various irrigation strategies and to assist irrigation
scheduling using historical weather records (Lembke and Jones, 1972; Swaney et al., 1983;

3
Villalobos and Fereres, 1989; Rogers and Elliott, 1989). Irrigation scheduling approaches
based on crop growth models and soil water budget components have been developed by
researchers (Jensen et al., 1971; Chesness et al., 1986; Smajstrla and Zazueta, 1987; Jones
and Ritchie, 1990). Models have also been developed to determine optimal irrigation
strategies using stochastic and probabilitic models of weather variables (Khanjani and
Busch, 1982). Long-term historical weather data are commonly used by such simulation
models. Many agricultural simulation models, which use historical weather data, have
succeeded in planning and long-term prediction of irrigation management. However, factors
such as the difficulty of model development, uncertainty of future conditions, and limitations
of available data combine to make the use of simulation models difficult for real-time
application.
With increasing competition for the use of water and high energy costs associated
with irrigation, microirrigation systems have become common in Florida, particularly for
high cash value crops like citrus. Because a microirrigation system wets only the soil
volume around the emitters, microirrigation systems must apply water at a high frequency.
Water should be applied at a rate equal to plant uptake (Phene et al., 1992). The soil water
potential can be maintained reasonably constant under high frequency irrigation scheduling.
Irrigation can be applied at least daily at a rate equal to the ET requirement; consequently,
there is a need for "real-time" irrigation scheduling and control systems (Phene et al.,
1989b).
Cold protection is an important issue for citrus growers. Cold weather has caused
severe economic damage to the Florida citrus industry in the past, particularly in 1989. The

4
primary cold protection method for Florida citrus is irrigation (Parsons et al., 1989; Parsons
and Wheaton, 1990). Effective irrigation management for cold protection can reduce tree
loss and increase profitability. However, cold protection management requires timely and
accurate climatic data so that adequate protection measures can be taken. On-site real-time
monitoring of weather data and expert knowledge on cold protection are necessary for farm
management.
As personal computers have become increasingly common, the potential for
computer-based decision support systems for farm water management has also increased.
Computerized irrigation scheduling systems have been developed by Cahoon et al. (1990),
Phene et al. (1992), and Zazueta et al. (1984a, 1994). Expert systems techniques can be used
to represent the heuristic knowledge required for decision making. Unlike simulation
systems, which are based on mechanistic biological or mathematical models, expert systems
use expert knowledge in the decision process like that used by human decision-makers.
Real-time expert systems (RTES) operate in a real-time domain and deal with dynamic data
and time critical responses, applying expert systems technology to control engineering.
In an RTES, most of the inputs come from sensors, while many of the outputs go to
effectors. Soil moisture sensors and weather stations can monitor soil water content and
climatic conditions, respectively. Expert knowledge can be acquired to develop several
alternative strategies and apply the one most suited to a specific problem. With the real-time
soil and weather data monitoring integrated with expert knowledge on farm management,
the system can be operated in real-time.

5
1.2 Objective of the Dissertation
Agricultural production is related to many factors including crop, soil, and biological
conditions, and management decisions. Many complex decisions must be made daily. New
techniques are needed to assist farm managers. With the complexity of modem farm
management and available computer technology, it appears that an RTES is a means to assist
farm management. The primary goal of this research was to develop a methodology using
an RTES to improve the management of citrus microirrigation systems. The specific
objectives of this dissertation were
1) To acquire expert knowledge on citrus microirrigation management.
2) To develop control routines and a control panel to turn on or off user
specified valves from a local or remote computer.
3) To develop a user-friendly RTES for citrus irrigation, fertigation, and cold
protection management.
4) To provide alternative control functions so that irrigation and fertigation can
be applied according to user defined schedules.
5) To use farm databases, crop water requirement simulation, and other
computer tools to assist the decision-making of farm managers.
6) To demonstrate the use of an RTES as an operational tool to improve
management of an irrigation system.
Acquiring expert knowledge is crucial in the development of an expert system.
Experts need to be identified to acquire their knowledge in the problem domain: citrus

6
microirrigation management. Since the expert system must be operated in the real-time
domain, control hardware is required. The control process can be accomplished by using
conventional programs. As an RTES applied to microirrigation management, its reasoning
process is not as time critical as military applications. In other words, the system is not a
hard RTES. The use of the term RTES is to distinguish the system from expert systems in
which time is not a factor at all or which acquire data only from static databases.
The components of farm databases, crop water requirement simulation, and
computer-controlled irrigation systems have been successfully applied. The integration of
an expert system with simulation models, databases, and user defined control needs to be
resolved in this study. This integration can rely heavily on the structure of the system
design, functionality of the expert system shell, and design of the user interface. With
modem software development tools, operating systems with multitasking capabilities, and
object-oriented software design, this integration can be achieved. After the system is
developed, system validation can be conducted by running generated test cases, expert
evaluation, field tests, or a simulation approach.

CHAPTER 2
REVIEW OF THE LITERATURE
2.1 Citrus Irrigation in Florida
Citrus is one of the major crops in Florida. The total acreage of citrus was 853,742
acres in 1994 (Florida Agricultural Statistics, 1994). The citrus industry is a significant
contributor to the economy of Florida. Its annual economic impact on the state's economy
has been estimated at billions of dollars.
Although the average yearly rainfall in Florida varies from 50 to 62 inches, irrigation
is required to achieve maximum production and improve the quality of citrus fruit (Koo,
1963; Tucker, 1983). Citrus irrigation systems are also used for cold protection purposes.
Microirrigation systems are common today in Florida for citrus irrigation. Of Florida's
1,855,390 irrigated acres, 19 percent is citrus (Smajstrla et al., 1995). Billions of gallons of
water are required for the industry each year. Thus, even modest increases in water use
efficiency will result in substantial water savings.
In Florida, microirrigation is the preferred method for citrus irrigation.
Microirrigation is an efficient and convenient means of supplying water directly to a crop
root zone. It provides an effective means for utilizing small continuous streams of water for
irrigation. Furthermore, microirrigation systems more easily realize computerized control
than do other types of irrigation systems.
7

8
2.2 Soil Moisture Sensors
The ability to measure soil moisture in-situ is important for irrigation management.
Irrigation water can be saved by using soil moisture sensors (Zazueta et al., 1993).
However, the choice of soil moisture sensor is crucial to the success of irrigation control and
management. Usually, "the most intractable barrier to the full implementation of automatic
process control is the lack of adequate on-line sensors (p. v)" (Carr-Brion, 1986). A poor
choice of a sensor at the design stage is commonly caused by lack of adequate appreciation
of the limitations of the type of sensor used or by lack of knowledge of what is available.
Many literature reviews can be found on soil-moisture measurement by a variety of
techniques (Taylor, 1955; Schmugge et al., 1980; McKim et al., 1980; Erbach, 1983;
Wheeler and Duncan, 1984; Gardner, 1986; Stafford, 1988; Zazueta and Xin, 1992). The
techniques commonly used in soil moisture sensors include (1) electromagnetic, (2) nuclear,
(3) remote sensing, (4) hygrometric, (5) tensiometric, (6) optical, and (7) time domain
reflectometry (TDR). Not all these soil moisture sensing techniques are suitable for
automation. The sensor must have the capability of interfacing with a computer or other
electronic devices. Sensor cost is another major concern for agricultural applications. Some
soil moisture sensors, such as TDR and neutron probes, can achieve high accuracy (Topp
and Davis, 1985; Simpson and Meyer, 1987), but costs of the devices are also high.
Tensiometers are relatively inexpensive and are easy to use.
Tensiometers measure the matric potential (capillary tension) directly, which is
related to the energy required for plants to extract water from the soil. Tensiometers are the

9
primary method for measuring matric potential in soil. They have a fairly fast response time
when used for irrigation (Towner, 1980; Stone et al., 1986). A pressure transducer can be
installed on a tensiometer and interfaced to a data acquisition or readout system to realize
automation. The use of tensiometers with pressure transducers for soil-water potential
measurement has been successful in many applications (Fitzsimmons and Young, 1972).
The advantages of tensiometers are (1) low cost and easy construction, (2) easy
installation and maintenance, (3) long periods of operation if properly maintained, and (4)
adaptable to automatic measurement with pressure transducers. The disadvantages are (1)
a limited range of 0 to -0.8 bar that is not adequate for some soils, (2) hysteresis, and (3)
potential breakage during installation and cultural practices.
2,3 Irrigation Scheduling
Irrigation scheduling requires making decisions on when to irrigate and how much
water to apply. The main techniques used for scheduling include (1) monitoring of soil
moisture, (2) physiological indicators, and (3) soil water balance models. Proper irrigation
scheduling should result in savings of water and energy without yield reduction. Irrigation
scheduling decisions may relate to crop response to water stress, management objectives,
water quality control, system constraints, and public policies. Maintaining adequate soil
moisture levels in the crop root zone is critical for crop growth. Inadequate soil moisture
not only limits water supply to the roots, but also reduces root conductivity directly
(Wiersum and Harmanny, 1983).

10
Although irrigation scheduling has been studied in various ways for a long time,
research is needed to reduce the consumption of water and energy and to increase
profitability through better scheduling. Numerous studies have been conducted on irrigation
scheduling (Pleban et al., 1983, 1984; Zoldoske, 1988; Rogers and Elliott, 1989; Shayya et
al., 1990; Protopapas and Georgakakos, 1990). Monitoring of soil water content, crop
growth, and weather conditions is important for irrigation scheduling.
2.3.1 Monitoring Method
Monitoring methods are primarily based on either crop or soil measurement.
Monitoring can rely on instruments or one's intuition. Soil moisture is usually monitored
by using a sensor to measure soil water potential (Campbell and Campbell, 1982). Irrigation
is applied when the monitored crop or soil data reach some critical value. Irrigation
scheduling can also rely on the monitoring of weather data (Howell et al., 1984).
Soil moisture sensors are one of the major tools used to assist decision-making on
irrigation water applications. Tensiometers and gypsum blocks are widely used in the field
(Cary and Fisher, 1983). Augustine and Snyder (1984) and Snyder et al. (1984) used
tensiometers to schedule irrigation for bermudagrass turf. Their results showed that
irrigation water savings of 42 to 95 percent were obtained in sensor controlled plots over
conventionally irrigated plots. A study using tensiometers to schedule cotton drip irrigation
was conducted by Wierenga et al. (1987). In all of these studies tensiometers were
successfully used for irrigation scheduling.

11
Cassell and Klute (1986) studied soil effects on tensiometers. They found that the
use of tensiometers for irrigation scheduling was more successful in coarse textured soils
than it was in fine textured soils. This is because a greater percentage of the water available
to a plant is retained by coarse textured soils at suctions less than 0.8 bar than is the case for
fine-textured soils. Tensiometers only operate from zero to about 0.8 bar. Tensiometers are
an effective tool to assist irrigation decision-making, but soil water potential must be
maintained within their operational range.
2.3.2 Computer Simulation
As computer systems have become widespread, simulation-based approaches have
been developed (Lembke and Jones, 1972; Swaney et al., 1983; Villalobos and Fereres,
1989; Rogers and Elliott, 1989). Soil water balance and crop growth simulation models are
two common approaches.
Soil water balance
This method applies the principle of continuity to the root zone. It describes soil
moisture change in the root zone over time. Using this approach to manage irrigation
involves estimating the amount of water in the crop root zone. To maintain the soil water
content in the crop root zone within a desired range, irrigation should be applied to satisfy
evapotranspiration (ET) demands.
Jensen et al. (1970) reported a scheduling method based upon soil-crop-climate data.
Soil water balance simulation models were developed by researchers (Jensen et al., 1971;
Zazueta et al., 1986; Smajstrla and Zazueta, 1987; Anderson et al., 1978; Cahoon et al.,

12
1990) for different locations and crop types. Maintaining a soil water balance is a widely
used and effective approach for irrigation scheduling. However, this approach can require
substantial weather data, and these data are not available in many cases. Models to generate
weather data have been developed for these purposes (Richardson, 1981, 1985; Richardson
and Wright, 1984; Villalobos and Fereres, 1989; Jones, 1993).
Simulation of crop growth
Crop growth models can be developed to simulate crop growth. A crop growth
model can be a physically based representation of the dynamics of the soil-crop-atmosphere
system. The crop yield can be predicted by explicit models of the plant growth process, such
as assimilation, respiration, and transpiration (Protopapas and Georgakakos, 1990).
Crop growth models have been developed as aids for irrigation water management
(Swaney et al., 1983; Rogers and Elliott, 1989; Jones and Ritchie, 1990, Jones, 1993).
Although crop growth models have been successfully used in irrigation management and
decision-making, the technique has some difficulties in practice. First, models are currently
not available for all crops. This is because it is difficult to develop an accurate crop growth
model. A crop system can be complex and affected by many factors such as weather,
insects, weeds, diseases, soil physical and chemical factors, and the interactions of these
factors. Second, factors such as uncertainty of future conditions and limitations of available
data make the use of simulation models difficult for real-time applications.

13
2.4 Irrigation Control
Irrigation system control includes a variety of topics, ranging from on-stream storage
and water diversions to agronomic practices (Duke et al., 1990). The control topics may
relate to (1) hydraulic, (2) mechanical, (3) electro-mechanical, (4) electronic, and (5)
computerized control (Duke et al., 1990). The major irrigation control modes are (1) on-off
control, (2) stepwise control, and (3) continuous control (Phene, 1986). Computerized
systems have shown great potential in irrigation control and farm management. This is
because a single hardware configuration can serve a wide range of control functions, and
control strategies can be easily modified by software modifications.
Studies (Phene et al., 1973; Phene and Howell, 1984; Phene, 1989) have been
conducted of irrigation control using soil moisture sensors. Sensors were used in a feedback
mode to maintain a nearly constant soil moisture content in the root zone. They concluded
that the performance of the irrigation controller depended on four basic factors: (1) adequate
operation of the system's control hardware, (2) the proper algorithm for the system's
software, (3) a reliable soil moisture sensor installed in the field, and (4) adequate operation
of the system's output, the solenoid valves, the pressure regulators, flow meter, and filter.
Further studies (Phene et al., 1989b) indicated that an irrigation controller should have the
following characteristics to monitor soil matric potential in real-time and control irrigation
systems: (1) ability to sample the sensor data automatically, (2) means of comparing the
sensor output to a threshold value, and (3) ability to control and monitor irrigation devices.

14
Field tests demonstrated that this scheduling method should be easily adaptable to irrigation
control, particularly in a sandy soil with low soil-water holding capacity.
Although most current irrigation management practices rely on manual operation or
timers, many researchers have focused on computer controlled irrigation systems (Duke et
al., 1984; Zazueta et al., 1984a, 1989; Phene et al., 1989a; Bums et al., 1990; Shayya et al.,
1990; Zazueta and Smajstrla, 1992). Vellidis et al. (1990) developed a microcomputer-
based data acquisition system for soil water potential measurements. The system consisted
of commercially available components: tensiometers, pressure transducers, a data acquisition
system, control devices, and a portable computer. In tests, the system was found to be
effective to monitor temporal variation of soil moisture potential.
Computerized irrigation control systems have the potential for water and energy
savings. Stombaugh et al. (1992) studied frost protection of strawberries using an automated
pulsed irrigation system. Their studies showed that the automation of irrigation could
potentially reduce water use by 89 percent under mild frost conditions. For nursery
irrigation, Zazueta et al. (1984b) reported that when closed loop computer control was added
to the system, water savings of about 20 percent (for a well-managed system) to 60 percent
or more (for a poorly managed system) were achieved. These water savings were achieved
by control of water deliveries using preset irrigation schedules. Even better results might
be achieved if the system integrated expert knowledge on irrigation management.

15
2.5 Expert Systems in Agriculture
In the past few years there has been significant interest among researchers in the
concept of expert systems. Expert systems have been widely applied in medicine, military,
industry, and agriculture (Laffey et al., 1988; Jones, 1989a; Feigenbaum et al., 1994). As
the development of expert system technology continues, expert systems are increasingly
being used in applications that sense the environment and directly influence it through
action. Techniques of real-time problem solving have been studied (Strosnider and Paul,
1994). In practice, real-time expert systems (RTESs) have been successfully developed for
control, monitoring, and diagnosis applications (Padalkar et al., 1991; Schnelle and Mah,
1992; Harrison and Harrison, 1994). Advanced personal computers and commercially
available easy-to-use expert system shells allow many people from different disciplines to
develop expert systems (Durkin, 1994). The use of expert system design methodology in
building agricultural decision support systems has shown great potential in recent years.
In agriculture, expert systems have been developed to assist the transfer of
technology from agricultural researchers and extension services to producers. Many expert
systems have been developed for management of nutrients, irrigation machinery, insect and
weed control, disease diagnosis, harvesting, and marketing (Jones and Haldeman, 1986;
Peart et al., 1986; Kalkar and Goodrich, 1986; Lemmon, 1986; Kline et al., 1987; Morey et
al., 1988; Muttiah et al., 1988; McClendon et al., 1989; Batchelor and McClendon, 1992;
Merlo, 1992; Kumar et al., 1992). Expert systems for crop management integrated with
simulation models have been developed (Plant, 1989; Palmer, 1986). Expert systems can

16
help facilitate the use of simulation models in several ways (Jones, 1985): (1) estimate
model parameters, (2) provide input for models, and (3) restrict scenarios for model
analyses.
As expert system technology has evolved, applications to irrigation scheduling and
operation have been developed. Wright et al. (1986) developed a real-time expert control
system (Hexscon). They concluded that the important issue in developing an RTES is
combining the best of conventional and expert-system controllers. Their results suggested
that real-time expert control can be built on a microcomputer and has enough sophistication
and capacity to be effective for real-world problems. Jacobson et al. (1987) developed an
RTES that supervised a tomato greenhouse environment controller. Jacobson et al. (1989)
implemented real-time greenhouse monitoring and control, linking a conventional expert
system with a set of utilities for data acquisition and control. Conventional expert systems
have been linked with models and data acquisition to make management recommendations.
Thomson et al. (1989) reported on an expert system that was coupled with simulation of a
peanut growth model and databases. This system evaluated moisture sensor readings
combined with a crop growth model to make estimates of irrigation timing. An RTES has
been applied to turf irrigation by Zazueta et al. (1989). In tests, it was found that the RTES
can apply irrigation in response to crop water demand. A major defect of the system was
the lack of heuristic knowledge available for irrigation control. An expert system for
irrigation management was developed in Thailand (Srinivasan et al., 1991). The system
demonstrated its effectiveness in improving water management decisions.

17
Although many expert systems have been developed for agricultural applications,
these systems were successful to some extent from a purely pedagogical viewpoint; but very
few of the systems are considered to be successful from a commercial viewpoint (Jones,
1989a). RTESs appear to have the potential to be successful because of their well-defined
domains. However, developing such systems can be extremely difficult due to factors such
as critical timing, knowledge acquisition, temporal reasoning, and uncertainty management.
2,6 Summary
Modem farm management and irrigation scheduling are complex tasks. There are
many factors that need to be considered to achieve successful farm management. There is
a need for more research in the field of farm management, which includes (1) irrigation
management, (2) chemigation, (3) maintaining water quality, (4) pest control, (5)
environment impact, (6) labor and energy conservation, and (7) cold protection. The
"optimal" management of an agricultural farm involves complex daily operation and
management decisions because of the temporal distribution of rainfall and extreme
microclimate.
One way of dealing with the problems is by the introduction of expert systems
integrated with control engineering techniques into irrigation management. With an RTES,
a computer can be used to implement these tasks: (1) decision-making on irrigation,
fertigation, and cold protection, (2) monitoring the performance of the irrigation system, (3)
adjusting irrigation applications as climatic or other conditions change during the irrigation,

18
(4) processing feedback data to evaluate the irrigation process, and (5) maintaining a
complete record of all applications.
Development of this RTES requires extended knowledge including (1) artificial
intelligence, (2) software engineering, (3) citrus irrigation and fertigation management, and
(4) control engineering. Studies are needed to integrate the technology of water
management into effective control engineering. An automatic weather station and soil
moisture sensors are essential for the system. An RTES could offer a better tool for the
management of irrigation, fertigation, and cold protection. RTESs are a feasible and
necessary approach for citrus microirrigation management.

CHAPTER 3
GENERAL EXPERT SYSTEM CONCEPTS
3.1 Expert Systems
One of the most significant results of artificial intelligence (AI) research to date is
the expert system (ES). An ES is a computer system that emulates the decision-making
capability of a human expert. Feigenbaum (1982), an early pioneer of ES technology,
defined an ES as
an intelligent computer program that uses knowledge and inference
procedures to solve problems that are difficult enough to require significant
human expertise for their solution, (p. 15)
An ES attempts to perform like a human expert to solve problems. Instead of relying on
statistical or algorithmic methods, ESs solve problems by applying a symbolic knowledge
representation of human expertise. Consequently, ESs try to encode domain-specific
knowledge rather than comparatively domain-free methods derived from computer science
or a mathematical approach. Its application is normally restricted to a specific problem
domain or well-defined domain. Applications of expert systems include medical, industrial,
agricultural, and space technologies. ESs provide the advantages of increased availability,
reliability, fast response, and multiple expertise. The structure of an ES can be a rule-base,
frame, model or other approaches. The major components of an ES are the inference engine,
knowledge base, user interface, and knowledge acquisition facility (Figure 3.1).
19

20
Figure 3.1. Major components of an expert system.
3,1,1 Inference Engine
The inference engine is the heart of an ES used for drawing conclusions based upon
a knowledge base. Schalkoff (1990) stated that the inference engine can be considered to
be a finite state machine with states representing typical actions such as (1) match rules, (2)
select rules, (3) execute rules, and (4) check stopping conditions (i.e., goal satisfaction).
Following are some commonly used reasoning approaches.
Forward chaining
Forward chaining is also called data driven. It is a reasoning method from facts
(data) to conclusions. For instance, if a traffic light is green (fact), then one can drive
through the traffic light (conclusion). Because of the nature of the forward reasoning
process, this reasoning method is suitable for problem domains such as monitoring and real
time control systems, where data or facts are continuously acquired or updated.

21
Backward chaining
Backward chaining is a goal driven method. In contrast to forward chaining,
backward chaining reverses the process. It reasons from a hypothesis, a potential goal to be
proved, to the facts which support the hypothesis. This reasoning approach is more
applicable to problems having many more inputs than possible conclusions, such as
diagnosis and classification problems. The approach was used in Prolog and the medical
expert system MYCIN.
Opportunistic chaining
Opportunistic chaining combines the forward and backward reasoning methods. For
applications with many inputs and many possible conclusions, neither forward nor backward
reasoning is an efficient approach. Thus, the two reasoning methods are applied together
in one system to achieve efficiency. However, such a method may expand the difficulty of
development.
Advanced reasoning methods
Advanced reasoning approaches (Gonzalez and Dankel, 1993) are model-based
reasoning, qualitative reasoning, case-based reasoning, temporal reasoning, and artificial
neural networks.
3,1,2 Knowledge Base
A knowledge base contains expert-level information required to solve problems in
a specific domain. A knowledge base consist of a human expert's knowledge acquired by
a knowledge engineer and encoded into the system. In other words, a knowledge base

22
contains the representation of domain specific knowledge. The essence of the knowledge
base must fit the structure of the knowledge representation scheme.
The strength of an ES lies in the knowledge base because the knowledge base
contains a representation of the human expertise of the problem-solving decision. Therefore,
the whole process of knowledge acquisition is crucial in the system's development. The
transferring of domain expertise to an ES's knowledge base has proved difficult and time
consuming, in part because the process requires the interposition of a knowledge engineer
between the human expert and a computer.
3.1.3 User Interface
Because ESs are generally interactive and involve users with little or no computer
experience, the user interface should be designed to be friendly, explainable, and easy to use.
A clear definition of the user interface requirements for an ES is essential to the success of
the system. In particular, for users to accept the interface, it must accomplish the task in a
straightforward way and still meet the entire range of problem solving requirements. To
develop a friendly and explainable user interface, Ege and Stary (1992) suggested that the
designers need to provide a global system perspective to create task-oriented, or user-
centered user interfaces.
3 2 Real-Time Expert Systems
Historically, AI researchers have focused on problems in which the time response
is not a concern, such as the medical diagnostic ES (i.e., MYCIN, Shortliffe and Buchanan,
1975). This kind of system is asking humans to supply necessary inputs, and the response

23
time can be slow or is not considered a major factor. This scenario is different from the
environment where real-time expert systems (RTES) are used. In RTESs, data change
rapidly and the input data are often collected automatically. RTESs typically need to
respond to changing task environments, timely handling of data, and execution of diverse
functions. This may involve an asynchronous flow of events and dynamically changing
requirements with limitations on time, hardware, and other factors. Figure 3.2 shows
additional components that may be required for an RTES. Sensors may be used to provide
facts to the knowledge base, and external hardware can be controlled through an ES. Thus,
the control process can be accomplished through conventional logic and procedures.
Applications of RTESs in different areas have been reported by many researchers (Wright
et al., 1986; Laffey et al., 1988; Nann et al., 1991; Ingrand et al., 1992).
Figure 3.2. Major components of an RTES.

24
3.2.1 Whv Use an RTES?
ESs have great potential value as control devices for many applications. In this role,
it is important to make provisions for an easy-to-use interface of sensor inputs, system
outputs, and control actuators. In the real world, systems that are designed to control
complex and dynamic processes (such as on-line monitoring) require fast handling of data
and execute diverse functions. The control decisions, which may require deep knowledge
or expertise, should be made based upon timely data. RTESs might be useful for domains
where conventional ES approaches have failed or are impractical. These may include
situations in which humans fail to effectively monitor data, make costly mistakes, miss
optimizing opportunities, are unable to solve conflicting constraints, or suffer from cognitive
over load. Turner (1986) pointed out that the main reason for using an RTES is to reduce
the cognitive load on users to enable them to increase their productivity without the
cognitive load on them increasing.
3.2.2 Characteristics of RTESs
Three factors are of main concern for RTESs. First, conclusions must be reached
and actions must be taken in real-time to respond to the sensor's perceptions and
environmental change. Second, the system must be able to provide tentative conclusions
based on initial evidence if not all of the data are available at once. Third, the system must
operate safely and reasonably on inaccurate and uncertainty of data input.
Laffey et al. (1988) perceived a series of characteristics of RTESs that differ from
conventional ESs. Those major characteristics are as follows.

25
Non-monotonicity of data and uncertainty of missing data,
Continuous operation with high performance and guaranteed response times,
Ability to deal with asynchronous events,
Ability to communicate with the external environment (sensors and
effectors), and
Integration with procedural components.
3,3 Knowledge Acquisition
Knowledge acquisition is the transfer and transformation of problem-solving
expertise from some knowledge source to a computer program (Buchanan and Shortliffe,
1984). This is a process of eliciting,
structuring, and organizing knowledge
from human experts or other sources so
This process is the most important step,
this knowledge. The goal of knowledge
knowledge required by the system.
acquisition is to produce and verify the
from the sources and representation of
consists of elicitation of the knowledge
that the expertise can be encoded into
an ES (Figure 3.3). The process
Manual
Acquisition
and normally, the most time-consuming Figure 3.3. Knowledge acquisition cycle

26
phase in the development of a knowledge-based system. Researchers refer to this as the
knowledge acquisition bottleneck (Feigenbaum, 1979) in the development of knowledge-
based systems.
The success of current ES technology is highly related to the strict separation
between a domain-dependent knowledge base and an inference engine. The power of a
knowledge-based system relies more upon the quality of the knowledge base rather than the
characteristics of the inference engine.
3,3,1 Basic Approaches
As Figure 3.3 shows, knowledge acquisition is mainly eliciting and organizing
knowledge from human experts. The role of a knowledge engineer is to communicate the
basis of the performance with the experts and to specify it in a form suitable for a computer.
The basic approaches to knowledge acquisition can be summarized as
Interviews,
Questionnaires and observation of the expert at work,
Intuition, and
. Using knowledge engineering facilitators and inductive tools.
Interviewing is one of the major approaches of knowledge acquisition. The
interview can be structured or unstructured, and the communication can be one-to-one or
many-to-one. Giarratano and Riley (1989) described the basic procedure of the knowledge
acquisition and extraction task. First, an acquisition strategy should be decided. This
includes specifying how knowledge will be acquired and the methods to be used in the
interview. Second, the knowledge elements or specific knowledge that could be used by the

27
system must be identified. Third, the information to aid the knowledge understanding and
verification by the developer needs to be classified and organized. Fourth, the detailed
functional capabilities of the system have to be laid out. Then, a description of the general
functional capability must be given in detail. This includes studying the flow of knowledge
from both the developer's and user's viewpoint. Fifth, the system task needs to be defined.
Questionnaires mean that the knowledge engineer prepares some question sheets and
lets the expert answer them. This approach can be used in combination with the interview
approach. Observation allows the knowledge engineer to learn how the expert solves a real
problem.
Intuition refers to how a knowledge engineer attempts to be a pseudo-expert and
applies his knowledge to the domain. This process can only serve as an aid to knowledge
acquisition because the knowledge engineer is not a true expert and lacks expertise in the
domain.
Tools for the processes of automated knowledge acquisition have been developed
by researchers (Quinlan, 1986; Michalski et al., 1986). These tools attempt to help bridge
the gap between the expert/knowledge engineer and computer implementations using
learning algorithms. One of the earliest and best-known algorithms is ID3 (Quillan, 1986).
However, these tools provide very limited capabilities in solving real-world problems.
3.3.2 Potential Problems
Interviews are the most common approach for knowledge acquisition, but this
process is not simple because, in general, experts do not structure their decision-making in

28
any formal way; and they may have difficulty in explicitly describing their reasoning. These
problems were summarized by Harandi and Lange (1990):
Vocabulary. Knowledge engineering is virtually impossible unless the
knowledge engineer has a basic understanding of the problem domain. An
essential part of that understanding is familiarity with domain terminology.
Completeness. A knowledge engineer must be able to identify pieces of
information or knowledge that are missing from the knowledge base.
Integration. A knowledge engineer should find out how new information fits
into the current knowledge base because the new information could interact
with already available information in an undesirable way.
Analysis. Usually, experts have difficulty in explaining exactly how and
why they reach certain conclusions. Therefore, knowledge engineers may
have to conduct an interview that may require substantial communication
skills.
3,3,3 Practical Issues
In the real world, the process of knowledge acquisition should consider many
practical issues; and there is no standard approach to follow. The practical considerations
of knowledge acquisition were discussed by Jones (1989b), and Gonzalez and Dankel
(1993). For instance, how to find a "real" expert who is articulate and very knowledgeable
in the problem domain, how to plan and to conduct an interview, and how to capture the
detailed knowledge are problems that must be addressed.

29
As a result, knowledge acquisition is an art. Each problem may require specific
acquisition strategies and could sometimes involve psychological issues. To reduce the
errors caused by human intervention, more efficient and reliable approaches for acquiring
knowledge are required. These should automate the elicitation process based on a
representation scheme that will completely and efficiently denote all the domain traits and
encompass all the essential knowledge.
3.4 Knowledge Representation
Knowledge representation consists of encoding real-world expert knowledge into a
format both readable and understandable by a computer. Some way to represent knowledge
is needed that allows the computer to derive new conclusions about its environment by
manipulating the representation.
In the process of knowledge representation, the primary problem is to find a kind of
format or knowledge representation language. Usually, knowledge representation is not
straightforward. First, the knowledge engineer should understand the concepts required to
solve the problem. Second, these concepts should be represented precisely and
unambiguously at all granularity levels. Third, these concepts should be easy to understand
and applicable to many systems. Ringland and Duce (1988) stated some issues that should
be raised in knowledge representation:
Is the approach expressively adequate to the domain?
Is reasoning efficient enough to allow the inference to perform in an
acceptable time?

30
How do we construct meta-knowledge representation?
What are the primitives and how does one manage incomplete knowledge?
The commonly applied knowledge representation methods in production systems are (1)
semantic network, (2) frame, (3) objects, and (4) rules.
3.4.1 Semantic Network
The semantic network was initially applied by Quillian (1968) to analyze words and
sentences. Since then, the approach has become widely used. A semantic net is a formal
graphic language for representing facts about entities. Its structure is shown graphically in
terms of nodes (objects) and the arcs (links) connecting them. The directed arcs that connect
the nodes represent relationships between objects. A semantic net can virtually represent
any relationship that holds among the objects or concepts in some domain of interest.
The graphical relational representation of a semantic net is explicit and succinct to
the state of knowledge. In addition, because nodes are directly connected with related nodes,
the search can be efficient. However, the nets offer no standard definition of link names
among nodes; and there are some practical difficulties in performing computer reasoning
with completely general semantic nets.
3.4.2 Frame
Frame-based representation was developed to manage information overload inherent
in large semantic nets without sacrificing their expressive power (Minsky, 1975; Bobrow
and Winograd, 1977; Fikes and Kehler, 1985; Brachman and Levesque, 1985). The basic
characteristics of a frame are that it maintains the fundamental notions of abstraction
hierarchies and inheritance of properties from superclasses, but it packages the descriptive

31
attributes associated with each class or instance into more compact local data structures. A
frame is mainly a group of slots (attribute) and fillers (values) that define a stereotype of
knowledge. The fillers can also be subdivided into facets. Each slot, filler, and facet has
its own associated values.
Frame systems are very suitable for those well-defined features (stereotype
knowledge), so that many of its slots have default values. Therefore, the main advantage
of frame systems is that their knowledge representation is significantly better structured and
organized than knowledge in semantic net systems. Furthermore, the system can only
trigger specific actions through demons during the processing of information, instead of
repeatedly testing a rule in a rule-based system. This will significantly increase the
efficiency of knowledge processing. The frame system, however, may have difficulty
dealing with heuristic knowledge and coping with a new situation beyond the default values.
3,4,3 Objects
An object is a representation corresponding to a conceptual entity in the real world
and how the information related to the entity is manipulated. The basic idea behind an
object-oriented representation is that information should be clustered around the "object."
The difference between frame and object representation is that an object-oriented approach
creates a tight bond between the code and data instead of separating them into two complex,
separate structures (Gonzalez and Dankel, 1993).
The characteristics of object-orientation are the levels of abstraction that can be
achieved and its ability of encapsulation, inheritance, and polymorphism. The approach,
however, possesses drawbacks similar to those of the frame method.

32
3.4.4 Rules
The classic and most common way of knowledge representation is the use of a
generic form of IF-THEN rules. The rules may have the simple form:
IF antecedent(s), THEN consequence(s).
Rules provide a readily understandable form. The antecedents define a pattern to be
matched against the content of the working memory, which is a global database of facts used
by the rules. The rule is fired if such a pattern is matched. Thus, the consequences change
the working memory and play the inferences in an ES. Each rule is an independent unit in
the entire knowledge base.
Rule-based systems have been widely applied in many areas and are often
misunderstood because their IF-THEN structure is similar to the condition structure in
conventional programming language. Two factors distinguish rules from a conditional
statement in conventional languages (Ringland and Duce, 1988):
The antecedents are expressed as a pattern rather than a boolean expression;
and such antecedents can be in a simple or very complex form.
A rule-based system allows separation of knowledge from control of how the
knowledge is applied. The condition of conventional language is a flow of
control. The flow in rules does not pass from one rule to the next in lexical
sequence, but is determined completely separately through the inference
engine.
Rule-based systems are the most widely used production systems. All knowledge
in the knowledge base is represented in a single uniform format, and each rule is a distinct

33
individual element of knowledge that can be updated independently of the other rules. The
major drawback of the system is inefficiency due to the use of infinite chaining. Moreover,
there may exist contradictory and inconsistent knowledge among the rules when new rules
are added.
Knowledge representation is the foundation of AI. Many researchers (Bobrow and
Winograd, 1977; Brachman and Levesque, 1985; Fox, 1990) have attempted to improve the
ways of representing knowledge. Advanced knowledge representation approaches include
the techniques of spatial, causal or temporal models, and neural nets. A general knowledge
representation system cannot be constructed easily. The selection of knowledge
representation methods will mainly depend on the inherent structure of the knowledge and
what knowledge representation the expert system tool will support.
3.5 Rule-Based Expert Systems
Many ESs have been developed using rule-based structure or so called production
systems. Rule-based ESs have been successfully applied in many domains, including
medical diagnosis, mathematic discovery, and hardware configuration. Rule-based systems
have been widely applied because the systems have advantages of modularity, uniformity,
and naturalness (Gonzalez and Dankel, 1993). In addition, many development tools (Meta-
MYCEN, CLIPS, and LEVEL 5) with relatively low costs are available for development of
rule-based systems.

34
3 5.1 Rule-Based Architectures
Rule-based systems or production systems have three main components: (1) working
memory, (2) rule memory, and (3) inference engine. The architecture and execution cycle of
rule-based systems is illustrated in Figure 3.4.
The working memory functions as a storage facility of these objects representing facts
about the world. The rule memory contains rules or the knowledge base of the system. The
inference engine is the active element in the system. It selects rules from the rule memory that
matches the contents of the working memory and executes the associated actions. If a rule
is matched with the content of the working memory, the rule is said to be fired. The conflict
resolution strategy will affect system behavior (Gonzalez and Dankel, 1993). It should be
chosen with care.
Figure 3.4. The architecture and execution cycle of rule-based systems.

35
3.5.2 Uncertainty Management
The quality of an ES is related to the knowledge base. Each piece of knowledge
acquired from experts may involve some kind of uncertainty. Bronowski (1965) stated that
in trying to formalize a rule, we look for truth, but what we find is
knowledge, and what we fail to find is certainty, (p. 32)
Uncertainty arises from a variety of sources: (1) unreliable information, (2) imprecise
descriptive languages, (3) inferences with incomplete information, and (4) a poor
combination of knowledge from different experts (Bonissone and Tong, 1985). Uncertainty
may prevent a system from making the best decision and may even cause a bad decision to
be made. The basic numeric approaches to deal with uncertainty are Bayesian probability,
Dempster-Shafer theory (Dempster, 1967; Shafer, 1976), and certainty factors. Bayesian
and Dempster-Shafer approaches were developed before expert systems became popular.
Because their practical implementation is complex and the approach requires a prodigious
amount of data that are usually not available, these approaches are not widely applied in the
development of expert systems.
Shortliffe and Buchanan (1975) developed an uncertainty approach to represent
uncertain information in MYCIN. A certainty factor (CF) is calculated from a measure of
belief (MB) and a measure of disbelief (MD):
MB MD
1 min [MB, MD]
3-1

36
The combined CF for the hypothesis is calculated:
CF
combintd
CF, CF2(1 -CF,)
CF, CF2
1 minflCF,|, |CF2|)
CF, CF2(1 CF,)
both >0
one <0
both <0
3-2
One of the advantages the CF has in comparison with Bayesian theory is that the CF
avoids the need to establish prior probability. Moreover, the CF represents and combines
the effects of multiple sources of evidence in terms of joint beliefs or disbeliefs in each
hypothesis. Consequently, the CF is an easy to apply and widely used approach for
uncertainty management.

CHAPTER 4
SYSTEM SPECIFICATION AND DESIGN
4.1 Domain of the Problem
Citrus farm management practices include decisions regarding irrigation, fertigation,
cold protection, and plant disease control. Management decisions are affected by many
factors including soil, meteorological, and biological conditions. In Florida, microirrigation
systems are common today in citrus irrigation. Because of the reduced area of coverage of
microirrigation and the low water-holding-capacities of Florida's sandy soils, irrigation
requires high frequency applications. Computer controlled irrigation systems offer
considerable labor savings.
In Florida citrus groves, chemicals and cold protection are commonly applied
through microirrigation systems. Fertigation application decisions require knowledge about
plant nutrient demands and environmental impacts. For cold protection water must be
applied in a timely and accurate manner to avoid severe economic damage. Farmers need
to make daily management decisions to maximize their net income. An expert system that
contains expert knowledge on irrigation, fertigation, and cold protection is desirable to
improve citrus microirrigation management. Thus, the goal of this research was to develop
a real-time expert system (RTES) to assist in the management of citrus irrigation, fertigation,
and cold protection.
37

38
4.2 Requirements Specification
4.2.1 Goal of the System
The overall goal is to develop a computer-based tool integrated with expert
knowledge as a decision aid for irrigation, fertigation, and cold protection of citrus. The
system should have the ability in response to current weather and soil-water moisture
changes to realize real-time performance. The system must also meet the following
requirements:
Be easy to use,
Achieve irrigation system automation with minimum maintenance,
Handle missing data or unreliable sensor data,
Record each event conducted by the system,
Develop several control schemes to minimize hardware requirements, and
Be accessible through telecommunication systems.
4.2.2 System Inputs
Table 4.1 shows the basic input and output requirements of the system. Because soil
water status is a crucial factor for crop growth, sensors are needed to monitor soil-water
content in the crop root zone. Rainfall and evapotranspiration (ET) are two important
parameters that determine crop water requirements. An automated weather station is
essential to obtain weather data for real-time control. The weather data are also used for
crop ET estimation. Crop, soil, and irrigation system data are required in the decision
process. Crop data include effective root depth, percentage of crop land coverage, and age

39
of trees. Soil data include soil depth and soil-water-holding capacity. Microirrigation
(micro-spray) is assumed to be used in this application. The following microirrigation
system data are required:
Number of emitters used per tree,
Emitter flow rate,
Wetted diameter, and
Irrigation application efficiency.
Table 4.1 System input and output requirements.
Inputs
Outputs
Soil-water tension
Turn on or off specified irrigation valves
Weather data
Turn on or off fertigation pump
ET coefficients
Apply pre-injection and flush for fertigation
Tree status (age of tree)
Apply cold protection
Soil characteristics
Simulate crop water requirement
Irrigation application rate
Display sensor maintenance messages
4,2,3 System Outputs
Outputs of the system need to be specified to apply irrigation, fertigation, and cold
protection. These outputs can also be a message displayed on the screen or a control signal
sent to an external device. After the decision-making process is accomplished, the system
control procedures activate or deactivate irrigation control valves and pumps. Thus,
irrigation, fertigation, and cold protection are applied according to the decision of the expert

40
system. Because the real-time system operates continuously, the system should display all
sensor readings and warning messages to request maintenance when a sensor failure is
identified. In addition, each control action should be saved to a file for future reference.
4.3 Knowledge Specification
Expert knowledge is the heart and power of an expert system. Knowledge required
for the system development, in general, involves citrus irrigation management, fertigation
application, cold protection, and sensor behavior.
For irrigation management, irrigation strategies and criteria for turning on or off the
irrigation system must be resolved. These criteria relate to the soil-water content in the crop
root zone and weather conditions. Irrigation management decisions should maintain the soil-
water content between certain levels in the crop root zone and should avoid applying too
little or too much water.
Fertigation should be scheduled at the right time and be applied for the proper
duration. Application of fertigation should be in a sequence of pre-injection, nutrient
application, and flush period. Expert decisions on fertigation and irrigation are needed to
achieve water savings and to avoid environmental pollution. For cold protection, factors that
affect cold damage of citrus trees should be understood. These factors include the principle
of cold damage, air temperature, and wind speed effect. In particular, critical air
temperatures for application of cold protection need to be specified. Knowledge of when
to start and when to stop an application need to be acquired for the reasoning process.

41
Sensor failure can be a main cause of making an unreliable or poor decision. Thus,
safety measures or data validation should be used to evaluate the sensor data. One approach
is to install a redundant sensor; then sensor readings from the redundant sensors can be
evaluated to increase the data reliability. However, this approach increases the cost of system
hardware. An alternative approach is to use data uncertainty analysis, such as the certainty
factor (CF) discussed in Chapter 3. Sensor readings with low CFs should be discarded from
the input of the decision process. In addition, a simulation approach can be used to verify
the actual irrigation duration to avoid excess water application.
4 4 Knowledge Representation Paradigm
Commonly used knowledge representation paradigms are rules, logic, frame, objects,
or semantic networks. The choice of the paradigm should be suitable to represent the
domain knowledge and be in the consideration of the selection of development tools.
Decision processes in citrus irrigation management are heuristic in nature. The knowledge
required in this application can be considered shallow knowledge. Rule-based systems are
the best currently available means for codifying the problem-solving knowledge of human
experts (Hayes-Roth, 1985). Because of the heuristic nature of the decision-making process
and available development tools, a rule-based knowledge representation paradigm is used
in the system.
4,4,1 Reasoning Method
Since rules are selected as the knowledge representation paradigm, reasoning
methods can be either forward, backward, or opportunistic (bi-directional). For this

42
application, many inputs are acquired automatically from sensors. The decision-making
progress is from initial facts (data from the sensors), to intermediate facts, and finally to a
conclusion. Thus, a development shell with forward reasoning should be selected.
4.4.2 System Performance Requirements
As an expert system which operates in the real-time domain, the system imports
initial data (facts) from sensors and the sensor data are varied with time. The system must
respond to this variation to realize real-time performance. In particular, the system should
achieve the following performance:
Must be operated within a fixed time constraint,
Should have the ability to react to the changing external environment and
must be operated continuously and as information is updated,
Must deal with incomplete and faulty input data from external devices, and
Must allow procedure calls to other systems to bring back the necessary
information for reasoning.
4.5 Development Tools
The software tools available for development of expert systems can be categorized
into four classes: General-Purpose Programming Language (GPPL), General-Purpose
Representation Language (GPRL), expert-system building frameworks (shells), and expert
system development environments (Collins et al., 1990).
The GPPL includes computer languages such as C, Lisp, and Prolog. The GPPL
languages have high programming flexibility, but the development time and the cost may

43
be higher than with other developing tools because the development needs to be started from
scratch in many cases.
The GPRL are the languages which are especially written for expert system
development, such as OPS5 and UNITS. These languages have relatively higher flexibility
in comparison with shells and development environments. Expert system development
environments (such as KEE, LEVEL5, and ART) are complete development environments
which provide sophisticated features such as knowledge representation, debugging routines,
and development facilities. Although these development environments provide many
features, the use of these software packages may require a substantial amount of time to
learn the package to take the advantage of their features. Also, the cost of the packages is
high.
4.5.1 Expert System Shells
The introduction of expert system shells has made expert system development much
faster and attracted more new developers. A range of expert system shells have been
developed to match the cost and application requirements. Some of these shells are CLIPS,
EXSYS, and VP-EXPERT. The shells have the advantages that they are low cost, easy
to learn, and save development time, but their flexibility and capability are the major
constraints for development of expert systems.
Choosing the right tool for implementing a particular application is difficult because
there is almost no absolutely one 'right' choice. One needs to ponder the advantages of any
selection against its limitations to determine the most suitable tool for a particular problem.
If a shell exists and satisfies the application requirements, the shell may be the better choice

44
than an AI language. Barrett and Beerel (1988) stated "use a shell if you can, an
environment where you should, and an AI language when you must (p. 69)." This study
selected an expert system shell, C Language Integrated Production System (CLIPS), as the
development tool.
4.5.2 CLIPS
CLIPS, developed by NASA's Johnson Space Center, is a forwarding chaining rule-
based language which uses the Rete Algorithm for pattern matching. CLIPS can be used as
an embedded application or child process. The tool can easily call an external executable
routine. Because a real-time system requires access to external devices, these features are
important for this application. CLIPS is delivered with a complete source code so that the
user can modify and re-compile the program for special purposes.
4,6 Hardware Specification
Agricultural decisions are highly related to climate data such as rainfall and air
temperature. Because the system is designed for real-time performance, hardware is
required to acquire external data and to realize control actions. Sensors are needed for
measuring climate parameters and soil-water content. The following hardware is necessary:
Soil moisture sensor,
Personal computer,
Automated weather station,
Data logger,

45
PC digital input/output board, and
Irrigation control system.
4.6.1 Soil Moisture Sensor
Numerous soil moisture sensors are available commercially. The types include
electrical, magnetic, nuclear, optical, and tensiometric sensors. A literature review of soil
moisture sensors was conducted by researchers (Schmugge et al., 1980; Mckim et al., 1980;
Zazueta and Xin, 1992). Among the many available soil moisture sensors, the tensiometer
is one of the most widely used sensors mainly because of its performance and its low cost.
In this study, tensiometers are used to measure soil-water potential at the crop root zone.
Tensiometers
Figure 4.1 shows a vacuum gauge tensiometer and a tensiometer with a micro
pressure transducer. A tensiometer consists of a ceramic porous cup, plastic body tube, a
gauge, and a service cap. The tube is filled with water. In the field the ceramic cup is
installed in the active root zone of the soil. As the soil dries, water in the tube is pulled
through the ceramic cup and the tension is displayed on the vacuum gauge or the pressure
transducer. This tension is equivalent to the soil-water tension when equilibrium is reached.
The tensiometer, in this way, measures the force exerted by the soil to extract water from
the ceramic cup. Thus, the tensiometer measures the soil-water potential or the energy status
of water for a plant rather than the quantity of water in the soil. The range of soil-water
potential that tensiometers measure is from 0 to about -0.8 bar.

46
Pressure transducer
Tensiometers continuously
measure the soil-water tension. The
tension must be converted into an
electrical signal to output to a
computer. Tensiometers with micro
pressure transducers have been
constructed at the Soil and Water
Hydraulics Laboratory, Agricultural
and Biological Engineering
Department, University of Florida
Figure 4.2. Pressure transducer (Model
141PC) from Micro Switch.

47
(Smajstrla, Personal communication). A Micro Switch1 model 141PC pressure transducer
was used in this research. Figure 4.2 illustrates the pressure sensor. Junction PI is the
reference pressure input, which is air pressure for this application. Junction P2 is connected
to the tensiometer to measure soil-water tension or gauge pressure. The pressure transducer
outputs an electric voltage signal in response to its resistance change caused by variation of
the input pressure. Output voltage of the sensor is proportional to the input pressure signal.
Table 4.2 lists the major characteristics of the sensor.
Table 4.2 Characteristics of pressure transducer Model 141PC.
Parameter
Minimum
Typical
Maximum
Output
4.85 V
5.0 V
2.5 V
5.15 V
Excitation
7 VDC
8 VDC
16 VDC
Input Pressure
- 15 psi
15 psi
Operation temperature
- 40C to + 85 C
Size
2.35 x 1.18 x 0.75 inches
Calibration of tensiometers
Before tensiometers are installed in the field, a calibration test is needed to determine
the relationship of soil-water tension and output voltage from the pressure sensor. Figure
4.3 shows the calibration equipment designed at the Soil and Water Hydraulics Laboratory,
1The manufacturers listed in this dissertation are for illustration only. No endorsement of
these companies or their products is implied by the authors or the University of Florida.

48
Figure 4.3. Tensiometer calibration equipment.

49
Agricultural and Biological Engineering Department, University of Florida (Smajstrla,
Personal communication). Tensiometers are placed in a chamber which is partially filled
with water and sealed. A vacuum pump is used to alter the pressure inside the chamber.
Outputs (voltages) from the pressure sensors vary with the chamber pressure (tension), as
indicated by a pressure gauge. The sensor outputs can be recorded through a data logger to
a computer or one can simply use a multimeter. The relationship between the tension and
the sensor output voltage is linear. Figure 4.4 is one of the typical calibration curves. Soil-
water tension can be calculated from the pressure sensor output based upon this calibration
curve.
4.6.2 Personal Computer
An IBM or any compatible PC with 386 or higher CPU is recommended to run the
expert system. Four MB of RAM and five megabytes free space on a hard disk are required.
The PC must have at least one communication port.
4.6.3 Automated Weather Station
The weather station used in this project was a Weather 2000 system from Campbell
Scientific, Inc. This station is an automated system designed for commercial, agricultural,
and irrigation scheduling applications. This weather station measures meteorological
conditions that affect crop water consumption. The station provides the following data:
rainfall, air temperature, solar radiation, wind speed, relative humidity, and soil temperature.
4.6.4 Data Logger
The Campbell Scientific data logger (CRIO) is a fully programmable module, which
provides sensor measurement, time keeping, communications, data reduction, data/program

50
storage, and control functions. A multitasking operation system allows simultaneous
communication and measurement functions.
The device is protected in a sealed, rugged, stainless steel canister in order to be
installed for out-door conditions. The input signal can be either analog or digital. The
maximum analog input ranges from -2.5 V to +2.5 V. The interface to the CRIO can be a
portable CR10KD Keyboard Display or a computer. The PC communication software to
access data from the CRIO is supplied by the vendor.
4.6.5 PC Digital Input/Output Board
A digital input/output (I/O) board was used as the interface between the computer
and the irrigation control board. The PC-DI072, a general digital I/O card from the
Industrial Computer Source, was used for this application. The card can be applied to relay
monitoring, control, sensing switches, security systems, and energy management. This
board provides user selectable buffered inputs and outputs based on the 8255 chips by Intel.
Major features of the PC-DIO are (1) 72 channels of digital I/O, (2) interrupt and interrupt
disable capability, and (3) four or eight bit groups independently selectable for I/O. The
output source current (output high) is 15 mA. The base address used in this application is
H310, H314, and H318 to set the ports as output.
4.6.6 Irrigation Control Board
An irrigation control board was used to activate or deactivate the automated
irrigation valves. The control board consists of 70 solenoid relays OACQ5 (Figure 4.5).
The relays are installed on a relay rack (PB24Q). Both the relay and the relay rack are
manufactured by Opto22. Each relay controls a specific electronic valve of the irrigation

51
Computer PC-DI072
Notas:
- One transformer feeds 10 solenoid valvas
(maxfrnum surge currant Is 1 A for aach valva)
- Each of the other 8transformers feed 0 sol. valvas
- All solenoid valvas ara 24 VAC
nrtrn
L0J 115V
Figure 4.5. Solenoid control relays of the irrigation system.
system. In addition to the automated relay control, the control board provides manual
switches and timers to turn on or off irrigation valves in case the relay system fails. In this
design, valve 64 is the default for the fertigation pump, and valve 40 is the master valve.

52
4.6.7 Overview of the Hardware
The selected hardware performs data collection and controls the irrigation system.
The system is an on-line irrigation controller (Figure 4.6). Soil moisture sensors
(tensiometers) measure the soil-water potential. An automated weather station measures
meteorological data, which provides current rainfall and sufficient data for ET estimation.
Sensor readings and weather data are stored in the data logger (CRIO). The computer
retrieves the data from the data logger at each given time interval. Connection between the
computer and the data logger can be either wire or radio links depending on the distance and
hardware cost. After the sensor data are input to the computer, the expert system uses the

53
data as initial facts to conduct its reasoning process. Then, the irrigation control valves can
be turned on or off through the irrigation control board according to the results of the
reasoning process.
4.7 Paradigm for the Real-Time Expert System
Since CLIPS provides a built-in inference engine for forward reasoning, the
development is focused mainly on the formation of the knowledge base (KB), external
control procedures, and the user interface. Figure 4.7 shows the structure of the system.
Input data from the sensors are collected and stored by external devices. Users can select
the frequency of downloading data from the data logger to the CLIPS fact base (FB). Before
the data are transferred to FB, a data pre-process procedure is needed to reorganize the data
format into the CLIPS data format.
Figure 4.7. Paradigm of the real-time expert system.

54
The KB was developed using the CLIPS language which contains expert knowledge
represented in rules, logical analysis, uncertainty management, and calls to external
procedures. After the knowledge is implemented into the rule base, it needs to be verified
by an domain expert. The KB development is one of the main tasks for the development.
External control procedures were designed to activate or deactivate the irrigation and
fertigation devices. Irrigation control procedure can turn on or off user specified irrigation
valves for a desired water application. Thus, irrigation can be applied block by block for
different irrigation management strategies.
Fertigation control procedure was implemented with the capability of turning on or
off both the injection pump and the irrigation valves. Cold protection control procedure is
similar to the irrigation control procedure in that they turn the irrigation valves on or off.
However, all the electronic valves should be turned on or off at once to cover the entire field
for a cold protection application.
These control actions are accomplished by the results of the reasoning process of the
KB. Each control action is displayed on the screen so that the user can view the current
event. In addition, all control events, including application types and duration, are saved in
a control event data log file.

CHAPTER 5
PROBABILITY OF RAINFALL
5 1 Introduction
Many agricultural operations and activities are affected by rainfall frequency and
amount. To improve water use, an irrigator must consider probable occurrence of rainfall.
Ideally, irrigation should be managed to maximize effective rainfall while satisfying crop
water demands. Although rainfall cannot be predicted with certainty, estimated probability
of rainfall is useful for irrigation management.
Rainfall probability is commonly predicated through weather forecasting according
to meteorological observations. Stochastic modeling is another available approach to
generate daily weather data from the use of observed weather data. Estimation of daily or
seasonal rainfall sequences can be obtained by examining past precipitation records. The
rainfall sequence can be estimated by using rainfall occurrence models (Schmidt, 1992): (1)
alternating wet and dry interval models, (2) wet and dry day models (Markov-chain models),
and (3) point process models. Such prediction is based upon the assumptions that sequences
will tend to be the same in the future as they were during the period of record.
Markov chain models are widely used because of their simplicity, flexibility,
seasonality, and number of states. Markov chain probability models for daily precipitation
occurrences have been studied extensively. This approach has been implemented with
55

56
success for various locations (Gabriel and Neumann, 1962; Jones et al, 1972; Todorovic and
Woolhiser, 1975). The rainfall probabilities according to wet-dry day sequences have also
been applied for irrigation management (Safley et al., 1974). A Markov chain rainfall
probability model was used to estimate the rainfall occurrence to assist irrigation scheduling
in this study. This model is not a physical explanation of rainfall occurrence, amount of
precipitation, or other meteorological observations, but merely a statistical description of the
past observed behavior. The purpose of using the Markov chain probability of rainfall was
to couple the rainfall probability with the irrigation decision-making process.
5.2 Markov Chain
A Markov chain is one particular type of stochastic process. Feller (1969) defined
a Markov chain as
a stochastic process in which the future development
depends only on the present state, but not the past history
of the process or the manner in which the present state
was reached, (p.444)
A stochastic model, in general, provides only the probability associated with a set of possible
future outcomes. Thus, a state X is followed by state Y with probability p, and by state Z
with probability q = 1 p, where X and Y are the only possible occurrences. The Markov
approach can be applied to wet-dry day sequences. Let (, denote the occurrence of a wet or
dry day.

1, if day i is wet
5, 5-1
0, if day i is dry

57
Let Nw denote the number of wet days in the n day period.
- E c, 5-2
ui
The possible values of the random value Nw are 0, 1, n.
Gabriel and Neumann (1962) studied a first-order Markov chain model for daily
rainfall occurrence in Tel Aviv, Israel. Their assumption was that the probability of rainfall
on any day depends only on whether the previous day was wet or dry. Such a probability
model is a Markov chain with two conditional probabilities:
P! = Pr {wet day | previous day wet}
p0 = Pr (wet day | previous day dry}
Although this model obtained satisfactory results in Tel Aviv (Gabriel and Neumann,
1962) and other regions, previous studies (Schmidt et al., 1987; Jones and Thornton, 1993)
showed that using a first-order Markov chain to estimate rainfall probability may not be
adequate for a subtropical or tropical region such as Florida. Their studies suggested that
a higher order Markov chain should be used for tropical and subtropical weather conditions.
Jones and Thornton (1993) have applied a third-order Markov chain for tropical and
subtropical regions. For a third-order Markov chain model, the probability of a rainfall
event on any given day is assumed only depending on the states of the three previous days.

58
Thus, a Markov chain of wet day probability with order 3 can be formed with the
conditional probability:
P{W|D1D2D3} = P{Wt= xJXw- x, X= x,2) Xt_3~ x,_3} 5-3
where W = wet day,
D, X = daily sequence of rainfall event, and
xt = random variable of a wet and dry day at day t.
The occurrence of the previous three consecutive days (D,D2D3) could be {000 001
010011 100 101 110 111}, where 0 represents a dry day and 1 represents a wet day. This
third-order Markov chain was used to describe rainfall occurrence only. The amount and
intensity of rainfall is not described by this equation. The probability of a wet day in the
immediate future is based on past long-term weather data rather than the changing
meteorological conditions.
5.3 Rainfall Data
Forty years of daily rainfall data from 1952 to 1992 were used to study the sequence
of wet and dry days. This weather station is located in Orlando, Florida (latitude 28:27:00,
longitude 81:19:00). These data were obtained from the NOAA weather station and the
Earthlnfo CD-ROM disk (NOAA, 1952-1992).

59
5.4 Frequency of Rainfall
A third-order Markov chain analysis was conducted using the forty-year rainfall data
in Orlando, Florida. The prediction derived here is a long term estimation of rainfall
occurrence. The probability of rainfall occurrence of each day is considered only dependent
upon the wet-dry sequence of the three previous days. Each day can be either wet or dry.
Days with a trace of rainfall (less than 0.01 inch) are considered to be a dry day. Table 5.1
shows the wet-day probabilities.
Table 5.1 Markov chain wet-day frequency.
Previous
Case Jan
Feb
Mar
Apr
Month of year
May Jun Jul
Aug
Sep
Oct
Nov
Dec
000
0.26
0.21
0.23
0.15
0.20
0.38
0.45
0.39
0.33
0.19
0.18
0.22
001
0.20
0.30
0.29
0.19
0.22
0.33
0.40
0.53
0.37
0.27
0.20
0.20
010
0.22
0.29
0.22
0.21
0.35
0.45
0.53
0.47
0.46
0.22
0.18
0.20
Oil
0.23
0.20
0.15
0.22
0.35
0.37
0.51
0.50
0.45
0.24
0.10
0.17
100
0.40
0.54
0.43
0.39
0.50
0.65
0.59
0.70
0.54
0.56
0.44
0.35
101
0.44
0.37
0.49
0.47
0.52
0.62
0.65
0.63
0.69
0.59
0.44
0.53
110
0.36
0.41
0.47
0.45
0.56
0.70
0.70
0.70
0.70
0.59
0.51
0.37
111
0.35
0.46
0.45
0.47
0.57
0.70
0.75
0.69
0.65
0.61
0.39
0.43
Rainfall data is for Orlando, Florida from 1952 to 1990.
0-Dry day
1-Wet day
This probability indicates the likelihood or frequency of rainfall during a particular month
and previous wet-dry day sequence. As the results show, the rainfall frequency ranges from
10 percent to 70 percent. Summer and fall have a higher wet day frequency than winter and
spring. Higher probability of rainfall during June through September is due to the higher
rainfall occurrence during these months. Figure 5.1 shows the annual rainfall distribution
in Orlando over forty-year average data. Approximately 70 percent of the annual rainfall
occurs during the summer and fall seasons. Thus, higher wet-day frequencies (> 0.6 from

60
Table 5.1) occurred in the summer and early fall. In contrast, lower wet-day frequencies
occurred during the dry season, as expected.
Figure 5.1. Annual rainfall distribution in Orlando from 1952 to 1992.
5,5 Statistical Test
The rainfall probability has shown variation among months and seasons. A statistical
analysis can be used to evaluate data significance among them. A paired t-test can be used
to test whether there are significant differences among the monthly or seasonal rainfall

61
probabilities. The one sample t statistic (Moore and McCabe, 1989) has the t distribution
with n-1 degrees of freedom:
t
5-4
Within each season, a paired t-test between monthly rainfall probability was conducted.
Subjects were matched in pairs, and the outcomes were compared within each matched pair.
The results of the paired t-test are shown in Table 5.2. In winter and spring, there are no
significant differences in rainfall probability among the months at the 95 percent confidence
level. The same test showed a significant difference in rainfall probabilities for summer and
fall at the 95 percent confidence level. Because wet-day frequencies during winter and
spring showed no significant differences, the average value of the seasonal rainfall
probability can be used to represent the seasonal rainfall probability. However, probabilities
of rainfall during summer and fall must be treated as monthly because they are significantly
different among months within the seasons.
Table 5.2 Results of paired t-test for rainfall probabilities within each season.
Winter
Spring
Summer
Fall
Jan vs Nov
NS
Feb vs Mar
NS
May vs Jun
S
Aug vs Sep
S*
Jan vs Dec
NS
Feb vs Apr
NS
May vs Jul
S
Aug vs Oct
S
Nov vs Dec
NS
Mar vs Apr
NS
Jun vs Jul
S
Sep vs Oct
S
NS = not significant different at 95% confidence level.
S = significant different at 95% confidence level.
S* = significant different at 90% confidence level.

62
5.6 Irrigation Decision with Rainfall Probability
One of the irrigation management goals is to maximize the use of effective rainfall.
For an automated irrigation management system, obtaining real-time rainfall data is
important to achieve this goal. The results of the Markov chain wet-dry day probability of
rainfall were integrated into the knowledge base to aid irrigation decision-making. Because
an automated weather station was installed in the field, daily rainfall for the past three
consecutive days can be used to estimate today's rainfall probability based on the Markov
chain probability of rainfall. The irrigation decision, then, is coupled with probability levels
of rainfall. Thus, irrigation may be delayed or less water may be applied if a high
probability of rainfall occurs and an irrigation is required. A practical issue related to
irrigation scheduling using the wet day frequencies is what value of wet-day frequency
should be considered a threshold for high probability of rainfall occurrence. This threshold
is a critical value that can affect an irrigation decision. When 60 percent is considered the
high rainfall probability, only summer and fall could have the possible values that are larger
than 0.6. For mature citrus trees, summer and fall are not critical growth stages. This
implies that it may be feasible to maintain the soil moisture at a lower level during these
seasons without causing yield loss. To maximize utilization of effective rainfall and to
reduce cost, irrigation may be delayed or less water may be applied when rainfall probability
is greater than the threshold value.
In addition to the stochastic model, another way to obtain rainfall probability is to
directly access a weather forecasting database. If the user can obtain a short-term weather

63
forecasting data, such as probability of rainfall for the next few days, those data can be input
to the system to assist the irrigation decision-making process. Because computer networks
are widely used nowadays, the system can link to a network to retrieve weather forecasts and
historical weather data. The computer network at the Institute of Agricultural and Food
Science (IF AS), University of Florida, supports a weather forecasting database.

CHAPTER 6
CITRUS IRRIGATION SCHEDULING
6.1 Introduction
Irrigation scheduling is important to maintain adequate soil-water content for high
productivity and the resulting economic benefits. Studies have shown that citrus irrigation
can increase fruit production (Myers and Harrison, 1978; Koo and Smajstrla, 1984;
Smajstrla and Koo, 1984; Adams, 1992). Irrigation scheduling involves decisions on when
to irrigate and how much water to apply. Irrigation scheduling methods can be based on (1)
soil properties, (2) plant properties, or (3) a soil-water balance modeling approaches. Each
method has advantages and disadvantages. In this study, soil properties and a soil-water
balance model were used.
6,2 Citrus Water Requirements
Citrus water use involves a process of soil-water extraction by the roots and
transpiration from leaves. Irrigation should provide water to crops to meet the
evapotranspiration (ET) demand imposed by climate. Citrus water requirements have been
studied by researchers (Rogers and Tucker, 1978; SCS, 1982; Smajstrla et al., 1986). Table
6.1 shows the citrus water requirements for central Florida (SCS, 1982). The estimated
citrus annual ET is 39.74 inches in central Florida. Similar results have been obtained by
64

65
Table 6.1 Citrus irrigation water requirements in central Florida.
ET
Normal year
Dry year
Month
(in/month)
ER
NIR
ER
NIR
Jan
1.68
1.03
0.65
0.88
0.80
Feb
1.75
1.26
0.49
1.08
0.67
Mar
2.54
1.72
0.82
1.48
1.06
Apr
3.33
1.42
1.91
1.21
2.12
May
4.29
1.68
2.61
1.44
2.85
Jun
4.84
3.42
1.42
2.93
1.91
Jul
5.11
4.01
1.10
3.45
1.66
Aug
4.88
3.66
1.22
3.14
1.74
Sep
4.16
3.16
1.00
2.71
1.45
Oct
3.24
2.03
1.21
1.74
1.50
Nov
2.19
0.89
1.30
0.77
1.42
Dec
1.73
0.99
0.74
0.85
0.88
Total
39.74
25.27
14.27
21.68
18.06
Note: Nil
R. = net irrigation requirement (inches),
ER = effective rainfall (inches), and
ET = evapotranspiration.
Source: SCS (1982) (p.4-30)

66
several researchers (Gerber et al., 1973; Reitz et al., 1977). They estimated that annual ET
for citrus is about 48 inches in Florida. Monthly mean ET rates vary from a low of 0.08
inches per day in the winter to a peak of 0.17 to 0.2 inches per day in the summer (Tucker,
1983). In Florida, it was reported that annual water use is about 47.6 inches for ridge citrus
(Koo, 1963), and 44.6 inches for flatwood citrus (Rogers et al., 1987). Citrus irrigation
requirements have been recommended by researchers based upon crop ET requirement and
effective rainfall.
Citrus irrigation requirements are different for young trees and mature trees. Young
trees are usually managed to grow as quickly as possible for early production. Moreover,
young trees are less able to resist water stress than mature trees. Therefore, adequate
irrigation is especially important for young trees.
For mature trees, irrigation management should be different for critical and non-
critical growth stages. The critical growth stage for mature trees refers to the months of leaf
expansion, bloom, fruit set, and fruit enlargement. This occurs mostly during the spring
months. Irrigation during this stage is very important to both fruit quality and yield. The
spring in Florida usually has the lowest rainfall, thus the greatest moisture stress occurs. A
sufficient amount of water is essential for mature trees during the critical growth stage.
Sound irrigation practices should be emphasized during this critical state (Tucker, 1983).
The remaining months are considered to be a non-critical growth stage. Irrigation
application during the non-critical stage should be considered only when tree stress is
imminent.

67
Estimations of citrus water requirements provide only a general guideline for
irrigation. Actual irrigation applications vary due to several factors including (1) irrigation
management strategies, (2) irrigation system, (3) variability of rainfall and other climatic
factors, (4) soil characteristics, (5) planting density, and (6) crop growth characteristics.
Because of the variability inherent in these factors, it is difficult to create a general irrigation
schedule. Therefore, field measurement of soil-water content or maintaining a soil-water
budget is useful to determine crop water requirements.
6,3 Evapotranspiration and Management Allowed Depletion
Knowledge of ET is important to the management and design of irrigation systems.
Actual crop ET is determined from reference ET and experimentally obtained crop
coefficients:
ETa Kc ETQ 6-1
where Kc = crop coefficient,
ETa = actual ET, in/day, and
ET0 = reference ET, in/day.
Microirrigation systems supply water only to the immediate vicinity of each plant
being irrigated. Tree canopies shade only a portion of the soil surface area and intercept
only a portion of the incoming radiation. Conventional estimation of water requirements
assumes that part of the applied water will be lost to non-beneficial consumptive use, which
is the loss from evaporation of wetted soil surfaces and plant transpiration from undesirable

68
vegetation. Consequently, conventional estimation of consumptive use, which assumes
wetting the entire field surface, needs to be modified for microirrigation.
The transpiration rate under microirrigation is a function of the conventionally
computed consumptive use rate and the extent of the plant canopy (Sharpies et al., 1985).
Keller and Biliesner (1990) used a simple equation for estimating the average daily
transpiration rate:
ETm ETa[0.\ (.PJ)05] 6-2
where
ETm =
average daily transpiration rate for a crop under
microirrigation, in/day,
ETa =
conventionally estimated average daily consumptive use,
in/day, and
Pa
percentage of soil surface area shaded by crop canopies at
midday (solar noon), %.
In Florida, citrus crop coefficients (Rogers et al., 1983), Kc, with grass coverage and citrus
irrigation Management Allowed Depletion (Koo, 1963), MAD, are given in Table 6.2.
MAD is used to express the amount of water that can be depleted in the crop root zone
without adversely affecting the plant.
Table 6.2 Citrus crop coefficients and recommended MAD in Florida.
Month
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
Kc
0.90
0.90
0.90
0.90
0.95
1.0
1.0
1.0
1.0
1.0
1.0
1.0
MAD (%)
67
67
33
33
33
33
67
67
67
67
67
67

69
Irrigation decision-making relies highly on the skill of the irrigation manager.
Irrigation should be applied when the soil-water content reaches the MAD. SCS (1982)
recommended that available soil-water depletion should not exceed 30 percent between fruit
set (February to March) and the period when young fruit has reached more than 1-inch in
diameter (June to July). During remaining months of the year, soil-water depletion should
not exceed 50 percent. Studies in Florida (Koo, 1963; Gerber et al., 1973; Smajstrla et al.,
1987) suggested that mature trees should be irrigated when one-third (33 percent) of the soil
moisture in the root zone is depleted in spring or two-thirds (66 percent) is depleted during
the rest of year.
As a rule of thumb, MAD within the root zone should not fall below 50 percent of
the total available water-holding capacity (Keller and Bliesner, 1990). For young trees and
mature trees during critical growth stages, irrigation should be applied when 25 to 35 percent
of depletion has occurred.
6,4 Irrigation Depth and Duration
Irrigation scheduling requires making decisions on irrigation duration and frequency
to meet crop ET demands. The amount of water and application frequency are related to
water management, soil property, and economic considerations. Irrigation duration is
associated with the application rate. For young trees, irrigation duration should average
between one and three hours for microsprinklers with flow rates of 10 to 15 gallons per
hour, and three to six hours for most drip systems depending on soil type, frequency of
irrigation, and number of emitters (Davies et al., 1989).

70
The maximum net depth per irrigation (Dx) should replace the soil moisture deficit.
The net irrigation depth for microirrigation can be described as the following equation
(Keller and Bliesner, 1990):
D
X
MAD
100
wa z
6-3
where MAD = management allowed depletion, %
W, = available water-holding capacity of the soil, in/ft, and
Z = plant root zone, ft.
For microsprinkler irrigation, irrigation time to bring root zone to field capacity can
be expressed as (Parsons et al., 1993)
AW DfZ
6-4
where Id =
irrigation duration (hour),
AW =
available soil-water content (in/ft),
Df =
depletion of AW prior to irrigation (%),
Z
root depth (ft),
Pr
precipitation rate (in/hr), and
2.04 F, N Ef
p 1
d2
where Fr =
flow rate of emitter (gal/hr),
N
number of emitters per tree,
6-5

71
Ef = overall efficiency of the irrigation system (%), and
d = diameter of spray pattern (ft).
Because of the low water-holding capacity of Florida sandy soils, irrigation
applications need to be more frequent to maintain the proper range of soil-water contents.
With a computer controlled microirrigation system, water can be applied frequently in a
timely manner to maintain less variation of soil-water content and without increasing
application costs.
6 5 Soil-Water Budget
A soil-water budget is commonly used to describe the amount of available water in
a crop root zone. For irrigation scheduling, it is convenient to calculate the amount of water
used (depletion) in the root zone instead of estimating the remaining water. The water
balance equation can be expressed as
where 0i+1 0¡
ET.R-ER-NI
- 8, 5 6-6
soil profile depletions at the end and beginning of a period,
ER
= effective rainfall (in),
ETa
= actual daily evapotranspiration (in/day),
D
= crop effective root depth (in),
NI
= net irrigation (in), and
R
= runoff from surface and deep percolation (in/day).

72
Soil-water depletion at
any time is the amount of water
needed to irrigate the current
crop root zone to field capacity
(FC). Normally, irrigation is
applied when the depletion
exceeds MAD or when the
managed soil-water content is
less than a threshold value
(SWCo). Thus, soil-water content is maintained at a certain level (Figure 6.1). Because
each irrigation raises soil-water content from SWCo to FC, the amount of applied water is
the same for all irrigations. However, the application frequency varies between irrigation
events. On the other hand, irrigation can also be applied at a fixed interval (every day or two
days), but irrigating with different amounts of water each time.
6.6 Irrigation Scheduling Using Tensiometers
Researchers reported that tensiometers can be used effectively to schedule irrigation
by measuring soil-water potential, but proper installation and maintenance are required for
the application (Smajstrla et al., 1985a; Smajstrla and Koo, 1986; Fitzsimmons and Young,
1972; Creighton et al., 1989). Two practical issues should be resolved in using tensiometers
for citrus irrigation scheduling:
S FC
-2
3
\i\A/\/\A
m SkVCO
i
3
Irrigation Events
Time
Figure 6.1. Irrigation by threshold of soil-water
content.

73
Depth and number of tensiometers that should be installed, and
Soil water tension at which irrigation should be initiated.
6.6.1 Tensiometer Installation Depth
Because tensiometers measure soil-water potential in only a small volume of soil
immediately surrounding the ceramic cup, installation sites should be representative of the
surrounding field conditions and water content in the effective root zone. Citrus root zone
moisture extraction depths in unsaturated soils range from 3 to 5 feet and the minimum root
zone moisture extraction depth required is 1.5 to 2 feet (SCS, 1982). The maximum
effective rooting depths for citrus are 3.0 to 5.4 feet (Martin et al., 1990). In Florida, Tucker
(1983) reported that citrus rooting depths extend to 5 feet for well-drained sandy soils;
groves on flatwood soils rarely exceeded 2 feet in rooting depth. For citrus under
microirrigation, the effective root zone should be defined as the upper 1.5 to 3 feet of the
root zone for ridge citrus; and 1 to 1.5 feet for flatwoods citrus (Smajstrla et al., 1987).
How many tensiometers should be installed in the field is a compromise between cost
and accuracy. One set of tensiometers, in general, is desirable for every five acres (Smajstrla
et al., 1985b). At least two tensiometers should be used per location in order to check soil-
water depletion in the effective root zone (Smajstrla et al., 1986). One tensiometer should
be located near the soil surface (6 to 12 inches from soil surface) where most of the roots are
located. The second one should be located near the bottom of the effective zone (24 to 36
inches from soil surface) that will be irrigated. Because the upper portion of the effective
root zone contains the most roots actively involved in water uptake, it is important to

74
concentrate on both water applications and observations in this zone. In practice,
tensiometers at three depths are desirable for deep rooting crops.
To account for sensor failures and different soil characteristics, tensiometers should
be installed at several different sites adequately to represent the water status in large areas.
How many and at what depth tensiometers should be installed at each site needs to be
justified for each specific application.
6,6,2 Soil-Water Potential and Allowable Water Depletion
Irrigation scheduling is usually determined by allowable soil-water depletion.
However, tensiometers measure soil-water potential. Hydraulic characteristics for the
irrigated soil are needed to establish the relationship between the soil-water potential and the
amount of soil-water depletion. Table 6.3 shows the soil-water tension versus the soil-water
content for Candler fine sand at different soil depths (Carlisle et al., 1978).
Table 6.3 Average soil-water content for Candler fine sand by volume.
Depth
Soil Water Tension, Centibar
()
0
2
3
4.5
6
8
10
IS
20
33
1500
0-1.7
36.8
34.8
31.4
17.7
11.9
8.0
6.8
5.7
5.1
4.5
1.5
0-4
36.3
34.7
32.0
17.6
11.3
7.5
6.3
5.3
4.7
4.2
1.4
Source: Carlisle et al. (1978)
For Candler fine sand, assuming the permanent wilting point (PWP) is at -15 bars,
the available water-holding capacity ranges from 7.5 to 9.5 percent by volume (Martin et al.,
1990). These values approximately correspond to -8 and -7 cb according to Table 6.3.
Thus, the relationship between soil-water potential and soil-water depletion can be
approximately established. Table 6.4 shows the relationship of soil-water potential and soil-

75
water depletion for Candler fine sand. One-third (33 percent) of the depletion approximately
corresponds to 11 cb, and 50 percent depletion approximately corresponds to 20 cb of soil-
water tension (Figure 6.2). Then, irrigation can be applied when soil-water potential or soil-
water depletion reaches a threshold value.
Table 6.4 Estimated soil-water tension in corresponding to soil-water depletion for
Candler fine sand.
Depletion, %
AWC %
SWC-7.5, %
SWC-9.5, %
ASWT, %
ESWT, CB
100
0
1.40
1.40
1.40
-1500
90
10
2.21
2.01
2.11
-
80
20
3.02
2.62
2.82
-
70
30
3.83
3.23
3.53
-
67
33
4.07
3.41
3.74
-38
60
40
4.64
3.84
4.24
-33
50
50
5.45
4.45
4.95
-20
40
60
6.26
5.06
5.66
-13
33
67
6.83
5.49
6.16
-11
20
80
7.88
6.28
7.08
-9
10
90
8.69
6.89
7.79
-8
0
100
9.50
7.50
8.50
-7.5
Note: AWC = available water content,
SWC-7.5 = soil-water content for field capacity at 7.5 percent by volume,
SWC-9.5 = soil-water content for field capacity at 9.5 percent by volume,
ASWC = average soil-water content of above two, and
ESWT = estimated soil-water tension.

76

CHAPTER 7
CITRUS COLD PROTECTION AND FERTIGATION
7.1 Introduction
Cold protection refers to methods used to prevent cold damage to the crop. This
term is typically used for (1) frost protection, (2) freeze protection, (3) frost/freeze
protection, and (4) chilling protection (Barfield et al., 1990). Cold protection is always
important to citrus production. Cold weather has caused severe economic damage in
Florida's citrus industry in January, 1985 and February, 1989. Although several cold
protection approaches are available, such as tree wraps, heaters, and wind machines,
irrigation is the primary means of cold protection in Florida citrus. Microirrigation is a
valuable tool for cold protection. Major cold protection in Florida (estimated over
100,000 acres of citrus) is accomplished with microsprinkler irrigation (Parsons et al.,
1989; Parsons and Wheaton, 1990). Experience has indicated that micro spray jet systems
are effective for cold protection (Harrison et al., 1987; Hardy, 1989; Parsons and
Wheaton, 1990), particularly for young trees. Studies of cold protection methods with
computer aided decision systems have been conducted by researchers (Holland, 1990;
Heinemann et al., 1991, 1994; Martsolf et al., 1991). Their results showed that
computerized systems could improve decision-making on cold protection. Irrigation
equipment must be specifically designed for cold protection purposes. The irrigation
77

78
system must have a sufficient capacity so that the entire crop area being protected can be
simultaneously watered to achieve adequate cold protection. With aid of an on-site
weather station and a real-time expert control system, cold protection management can be
implemented automatically. The computer can be used to turn on the irrigation system when
critical environmental conditions occur.
1.2 Cold Protection Application
7,2,1 Principle of Cold Protection
Cold protection is based on thermodynamic principles, which have been discussed
by Harrison et al. (1987), Martsolf (1990, 1992), and Barfield et al. (1990). A plant gains
or looses heat from its surroundings through a heat transfer process. Heat transfer can occur
as conduction and convention, evaporation and transpiration, and radiant energy exchange
(Harrison et al., 1987). Irrigation water provides cold protection because the heat loss from
the plant to its surroundings is replaced by the sensible heat and the heat of fusion of water.
The latent heat of fusion is released when water changes from liquid to ice. The total latent
heat input to the air, ignoring heat from the soil or atmosphere, is equal to the heat lost in
cooling to wet bulb temperature, plus that lost as a portion of the drops freeze. The latent
heat flux released from the water can be expressed as equation 7-1 (Barfield et al., 1990):
Ql = 0.27 x 10-6 Pw I [C (Tws Twb) + Yf Ff] 7-1
where QL = total latent heat flux in W m'2,
pw = density of water in kg m'3,
irrigation application rate in mm h'1,
I

79
C = specific heat of water in J kg'1 "C1,
Tws = water temperature in C,
Twb = the wet bulb temperature in C,
Yf = the latent heat of fusion in J kg'1, and
Ff = the fraction of water that has become ice (fused) when it strikes the
ground. Ff depends on drop size and environmental conditions.
As equation 7-1 shows, cold protection is mainly related to (1) air temperature, (2)
wind speed, which affects evaporation, and (3) irrigation application rate. In this
application, it was assumed that there were adequate water and energy supplies and the
irrigation system was properly designed for cold protection. This implies that the system
can irrigate an entire citrus grove simultaneously with an adequate application rate.
7,2,2 Critical Application Temperature
Knowledge of weather data, particularly air temperature, is crucial for cold
protection. Most citrus growers in Florida receive weather data from sources such as the
National Weather Service, commercial radio and TV, and county extension offices. During
the winter season, growers carefully track weather changes to make decisions related to cold
protection.
Since air temperature is a crucial factor for cold protection, it is important to
determine the critical air temperature to start cold protection. Applying cold protection too
early or too late may result in water waste or crop damage. Harrison et al. (1987) reported
that the freezing temperature for Tngelo is about 30.1 F in Orlando, Florida. However,
irrigation must be initiated before the temperature reaches freeze point because irrigation

80
pipe lines can be frozen at this temperature. A survey report (Ferguson et al., 1989)
indicated that 45 percent of the growers initiated irrigation for cold protection at 33-35 F
and turned it off at 35-39 F (43 percent). The Institute of Food and Agricultural Science
(EFAS) at the University of Florida recommends initiating cold protection when air
temperature reaches 36F and finishing when air temperature reaches 36 to 40 F. Thus,
air temperature at 36F was used as the critical application temperature in this system.
7.2.3 Water Application Rate
The water that is most effective for cold protection is that which covers the foliage,
and not the ice formed on the ground (Harrison et al., 1987). Uniformity of application is
very important to cold protection. Thus, uniformity design criteria of the irrigation system
must be strictly met. The application rate must be high enough to provide sufficient energy
to the system and guarantee plant coverage. A previous study (Parsons et al., 1989) showed
33.3 gal/acre/min (12 gal/hr/tree) was recommended for frost protection. Young trees may
potentially use less water. Furthermore, in Florida, emitters should be installed properly.
Parsons and Wheaton (1990) reported that emitters placed north or northwest (upwind) of
the tree provided better cold protection.
7,3Fertigation
Microirrigation offers the opportunity for precise application of fertilizer to the soil.
Fertigation is the addition of soluble nutrients or agricultural chemicals through irrigation
systems to crops. Fertilizer application through irrigation systems is desirable because of
labor and energy savings, flexibility in timing of application, and easy and precise control

81
of application rate. Because of the high efficiency and centralized control of microirrigation
systems, fertilizer placement through microirrigation systems can improve its efficiency of
application (Keller and Bliesner, 1990). For this application it was assumed that the
irrigation system was properly designed and it was adequate for cold protection and
fertigation.
Fertigation management decisions are affected by available fertilizer concentrations,
desired application rates, types of fertilizer, and the crop. Application of too little fertilizer
may not obtain the desired results, and excessive applications of chemicals may result in
unnecessary expenses and potential crop or environmental damage.
7,3,1 Application of Fertigation
Fifteen chemical elements have been found to be essential nutrients to satisfactory
growth and functioning of citrus trees (Jackson, 1991). Among the fifteen chemical
elements, three elements (carbon, hydrogen, and oxygen) are adequately provided in the
environment suited to tree growth and are largely beyond the control of the grower. The
other twelve are fertilizer elements or "plant food." The major chemical elements are
nitrogen, phosphorus, and potassium. Numerous researches have been conducted to study
fertigation application and fertigation effects on fruit quality and growth (Koo et al., 1984;
Rolston, et al., 1986; Robinson, 1990; Willis and Davies, 1991; Fleam, 1993; Boman,
1993). However, because of the complexity of the crop nutrient requirement, it is difficult
to obtain a generally accepted fertigation schedule. A typical orange grove might require
fertilizer in the following amounts (Jackson, 1991), This assumes a yield of 500 boxes per
acre.

82
Nitrogen needed = 0.4 lbs. N/box x 500 boxes = 200 lbs.
Potash needed = 0.4 lbs. K20/box X 500 boxes = 200 lbs.
Other nutrients applied on basis of need.
Table 7.1 shows the nitrogen requirement for orange and grapefruit under normal
conditions (Koo et al., 1984). Nitrogen requirement for young citrus trees is approximately
0.16 pound per tree per year in Florida (Fisher, 1990). Although some general
recommendations have been given for citrus fertigation, with current knowledge it is
difficult to develop a general and precise fertigation strategy. There are many different
opinions about rates, concentrations, and times at which fertigation should be applied.
Research is needed to determine the proper amount of fertilizer and application frequency
when microirrigation is used.
For this application, a set of fertigation schedules is created in the knowledge base.
These schedules were determined by interviewing citrus fertigation experts at the University
of Florida. The knowledge base continuously checks the current time and the fertigation
schedules. Fertigation is applied when the computer time matches the predefined fertigation
schedules. The user should modify the schedules based upon expert recommendation only.
Chemical injection rate can be computed by the following equation (Keller and Bliesner,
1990):
Fr
C'fr Ta
7-2

Table 7.1 Pounds of nitrogen fertilizer to be applied to furnish the nitrogen
requirement of orange and grapefruit trees under normal conditions.
Fruit
production
(boxes/acre)
Pounds of nitrogen (N)a
needed per acre per year
Pounds of nitrogen fertilizer needed per acre per yearb
15.5 %N
33.5 %N
45.0% N
Orange
Grapefruit
Orange
Grapefruit
Orange
Grapefruit
Orange
Grapefruit
<200
100
60
645
485
300
225
220
165
300
120
90
775
580
360
270
265
200
400
160
120
1030
775
480
360
355
265
500
200
150
1290
965
600
450
445
335
600
240
180
1580
1160
715
538
535
400
700
280
210
1805
1355
835
620
620
465
>800
300
240
1935
1450
895
670
665
500
a. Nitrogen needed is based on 0.4 pound per box of fruit for oranges and 0.3 pound per box for grapefruits. In most
cases, one should not use less than 100 pounds or more than 300 pounds of nitrogen for orange trees and not less
than 60 pounds or more than 240 pounds for grapefruit trees per acre per year.
b. The figures given should be divided by the number of applications per year to obtain pounds per application.

84
where qc =
rate of injection of liquid fertilizer solution into the system,
gph,
Fr
fertilizer application rate (quantity of nutrients to be applied) per
irrigation cycle, lb/A,
A
area irrigated in T ha,
T. =
irrigation application or set time, hr,
c'
concentration of actual nutrients in the liquid fertilizer, lb/gal, and
tr
ratio between fertilizer time and irrigation application time.
Because of the uncertainty of rainfall, a risk is always present that fertigation may
be applied immediately before a rain. Thus, plant nutrients can be leached out of the crop
root zone. To reduce this risk, the Markov chain probability of rainfall can be applied to
fertigation management. Fertigation may be delayed when a high probability of rainfall
occurs. When irrigation and fertigation schedules conflict, an algorithm is needed to merge
the two schedules and to satisfy both the application requirements. Fertigation should serve
as a part of irrigation whenever possible.
7.3.2 System Components of Fertigation
Fertigation systems consist of several components including an irrigation pumping
station, a fertilizer injection device, an injection port, a solute fertilizer reservoir, a backflow
prevention system, and calibration devices. Figure 7.1 is a common arrangement for
fertilizer injection equipment. The electrical control board controls the operation of the
fertigation pump (valve 64) and water supply valves (valve 1 to 63). Fertigation can be
applied based upon the expert knowledge or user defined schedules.

85
Figure 7.1. Major components of a fertigation system.
7.3.3 Fertilizer Materials
Many soluble materials are suitable for application through microirrigation systems.
Selection of a fertilizer material should consider the solubility, convenience, and cost of the
desired nutrients. Table 7.2 shows some of the soluble fertilizer formulas. Liquid fertilizers
can contain a single nutrient or combinations of nitrogen (N), phosphate (P), and potash (K).
Nitrogen injection is relatively easy because anhydrous ammonia (82-0-0) and aqua
ammonia (24-0-0) are completely soluble and can be injected directly into irrigation water.
However, some nutrients of the fertilizer may be lost because gaseous ammonia is likely to
volatilize (Keller and Bliesner, 1990). In addition, nitrate-nitrogen tends to persist in the soil
in solution and to move with water. Thus, this material is highly susceptible to loss due to
leaching if excessive water is applied. Materials like nitrogen and potassium are easily

86
applied by fertigation. This is because potassium oxide is so soluble that the fertilizer moves
freely into the soil. However, potassium molecules are adsorbed on the soil complex and
are not readily leached away (Keller and Bliesner, 1990).
Phosphorus fertigation is difficult because the treble-superphosphate (0-45-0) is only
moderately water-soluble (Keller and Bliesner, 1990). The quality of irrigation water must
be considered for phosphorous fertigation. If the water contains considerable amount of
calcium, it can cause clogging because any form of phosphorus will precipitate as
dicalcium phosphate in the pipeline and emitters.
Although there are general recommendations on fertigation, such as how many
pounds of nitrogen should be applied each year for citrus, precise knowledge on how long,
how much, and when to fertigate is not currently available. This is mainly due to the
complexity of crop development. However, management decisions on fertigation are
available. Fertigation may be applied as a part of irrigation, and irrigation application
should always avoid leaching the nutrients out of the crop root zone.

87
Table 7.2 Solubility of common fertilizers in water.
Fertilizer Formula
Temperature range
(kg fertilizer per m3 water)
Cold
Lukewarm
Hot
Ammonium chloride
NH4CL
297 (0)
758 (100)
Ammonium nitrate
nh4no3
1183 (0)
1950 (20)
3440 (50)
Mono ammonium
phosphate
nh4h2po4
227 (0)
282 (20)
417 (50)
Diammonium phosphate
(NH4)2HP04
429 (0)
575 (10)
1060 (70)
Ammonium sulfate
(NH4)2so4
706 (0)
760 (20)
850 (50)
Potassium chloride
KC
280 (0)
347 (20)
430 (50)
Potassium nitrate
KNOj
133 (0)
316 (20)
860 (50)
Potassium sulfate
k2so4
69 (0)
110 (20)
170 (50)
Monopotassium
phosphate
kh2po4
- -
330 (25)
835 (90)
Dipotassium
phosphate
k2hpo4
- -
1670 (20)
- -
Calcium nitrate
Ca(N03)2
1020 (0)
3410 (25)
3760 (100)
Magnesium nitrate
Mg(N03)2
423 (18)
578 (90)
Monocalcium phosphate
Ca(H2P04)2
18 (30)
Phosphoric acid
h3po4
5480 (25)
Urea
(NH2)2CO
780 (5)
1193 (25)
Source: Hodgman, C. (ed) 1949. Handbook of Chemistry and Physics. Chemical Rubber, Cleveland, Ohio.
* Numbers in parentheses are solution temperature, *C.

CHAPTER 8
CONSTRUCTION OF THE KNOWLEDGE BASE
8.1 Introduction
The crucial step in the development of an expert system is the construction of the
knowledge. The knowledge base deals mainly with the issues of uncertainty management
of the sensor data, irrigation, fertigation, and cold protection. In particular, decisions on
citrus irrigation management may involve (1) when and how long an irrigation should be
applied, (2) when cold protection should be applied, and (3) when and how long fertigation
should be applied. To make management decisions in real time, the system needs on-site
Figure 8 .1. Inputs and outcomes of the expert system.
88

89
data and expert knowledge on the subject. Figure 8.1 shows the major inputs and outcomes
of the expert system. Inputs to the system consist of soil-water content, time of day, weather
data, crop information, fertigation requirements, and Markov chain rainfall probabilities.
Soil moisture sensors are used to monitor the soil-water potential in the crop root zone. An
automated weather station provides wind speed, air temperature, solar radiation, relative
humidity, and rainfall data. With the continuous monitoring of soil-water potential and
weather data, the reasoning process can be conducted to make decisions on citrus irrigation,
fertigation, and cold protection. The decision-making process may involve (1) irrigation
strategy, (2) uncertainty management of the sensor data, and (3) application of irrigation,
fertigation, and cold protection. Figure 8.2 shows the decision flow diagram of the system.
8.2 The Process of Control and Reasoning
In development of the real-time expert system (RTES), two essential tasks are (1)
to control external devices, and (2) to represent the expert knowledge required to realize
such control. The system provides the following control functions:
Monitoring sensor and weather data at a given time interval,
Analyzing sensor data to identify possible sensor failure, and
Controlling electronic valves and pumps for irrigation, fertigation, and cold
protection applications.
These control actions are conventional control problems. Thus, conventional
software programs were developed for each of the control tasks. The control tasks can

90
Figure 8.2. Decision flow of the expert system.

91
be an initialization of the hardware, and turning on or off the control valves according to
predefined values. These control programs were integrated into the knowledge base so that
the control actions take place when the proper rule is fired. In other words, when a rule
"condition" is satisfied, the "action" executes external control programs. To realize these
control capabilities and real-time performance, two basic technical problems were overcome
in developing the knowledge base:
How to deal with constantly changing data in real time, and
How to reason about the behavior of the decision process.
Figure 8.3. Paradigm of the knowledge base.

92
The solution of these problems involves the development of a special architecture
for real time and the use of reasoning with time. Figure 8.3 shows the paradigm used for
the knowledge base that performs the following tasks:
Recognize events that may indicate a problem or need to take an action,
Determine priority of these events or actions,
Execute the action based upon the priority, and
Explain the reasoning to the user.
A scheduler is used to determine when to execute a specific group of rules. Rules
can be scheduled to fire at given time intervals or may be event driven. Data from soil
moisture sensors and the weather station should be accessed at given time intervals. The
validity of the data decreases with time after a reading. Rules to perform a task are valid or
expire in a certain period of time depending on the validity of the sensor data.
Event detection recognizes the current state of the system. Event detection involves
determining the current status of sensor readings, whether a valve (irrigation, fertigation, or
cold protection) is on or off, and rainfall occurrence. The event priority is used to determine
priority of rules. Rules having higher priority are executed first in the reasoning process.
Then, execution of the control actions is based upon the conclusion of the reasoning process.
The explanation facility displays control actions that have occurred. In addition, all
previous control actions are stored in an application log file. This decision-making process
must operate continuously to deal with constantly changing data and maintain real-time
performance.

93
8.3 The Sensor Data
8.3.1 Download the Sensor Data
To increase reliability of the sensor data, tensiometers should be properly calibrated
and installed. In this design, three tensiometers (depths 6", 12", and 24") were used at each
of two locations. EDI and ID2 are used to denote sensors at the two locations. SI, S2 and
S3 denote sensor readings at the three depths (6", 12", and 24", respectively). The sensor
installation depth can be adjusted based on field conditions without affecting the knowledge
base.
Figure 8.4 illustrates
the process of downloading
sensor data. A data logger
(CRIO) was used to acquire
sensor data. Rules were
developed to access the data
logger at a specified time
interval. The system keeps
track of the elapsed time since
the last data downloading
time. The sensor and weather
data are downloaded at the
specified time intervals. This
Figure 8.4. Process of downloading weather and soil
moisture sensor data.

94
downloading action is accomplished through a conventional program to access the data
logger. After the sensor data are downloaded, the data must be rearranged to the CLIPS
(expert system shell) data format to accomplish the reasoning process.
8.3.2 Uncertainty Management of the Sensor Data
One of the important characteristics of RTES is that the system reasons with missing
data. In practice, sensors may provide bad or unreliable data when they have failed or as
they begin to fail. The sensor outputs can also vary with soil characteristics and nonuniform
root zones. Thus, a lack of adequate information may jeopardize the decision-making
process. This may prevent the outcome from being the best decision and may also result in
a bad decision. Because the sensor data are crucial to the reasoning process, uncertainty
management is essential so that the system operates reliably with missing data.
A physical approach to increase reliability of the sensor reading is using redundant
sensors, but this adds more hardware costs. The approach used here is data uncertainty
management. A certainty factor (CF) analysis is applied to the sensor data. The values of
CF range from -1 to 1 indicating two extremes: false (invalid) and true (valid). An initial
assumption was made that the sensor reading was valid. Thus, an initial high CF value
(0.98) was assigned to each sensor reading. Rules were developed to validate the sensor
readings. These rules interpret the data to detect deviations from normal or desired
behavior. The CF values are decreased when the sensor data are identified as abnormal.
Thus, a sensor reading may be rejected from the reasoning process when its CF value is at
a less than acceptable confidence level. The sensor readings were checked in two ways: (1)
checking the range of sensor readings, and (2) checking the variation among sensor readings.

95
Checking the ranee of sensor readings
A common irrigation
management strategy is to
maintain soil-water content in a
certain range. Irrigation is
applied when sensor outputs
indicate that the maximum
allowable depletion has been
exceeded. A possible sensor
failure may have occurred when
its output is out of the normal
range. This possible failure is
denoted as Failure A. Figure
8.5 shows the propagation of
CF values. When the sensor output is out of the predefined range, a lower value of CF' is
assigned to the sensor. The combined new confidence value (CF") is calculated through CF
propagation from equation 3-1. Thus, confidence in the sensor reading is reduced because
of the lower CF value. However, this reduction of the CF value does not definitely indicate
sensor failure. Only the final propagation of CF determines whether a sensor has failed at
Figure 8.5. CF propagation by checking range of
sensor data.
a certain confidence level.

96
Checking the variation among sensor readings
Sensor readings may also be
validated by comparing readings
from different sensors or the rate
change of sensor readings. Three
assumptions were made about the
soil characteristics in this study.
First, the soil profile is
homogeneous. Second, the sensors
are properly installed in the crop
root zone. Third, irrigation water is
Figure 8.6. CF propagation for sensors SI and
uniformly provided to the crop root
zone. Under these assumptions,
variation of the sensor readings at
different depths should be within a
certain range under static conditions.
When the output of a sensor differs
from other sensor's readings more
than the expert given value (EGV), a
lower CF value is assigned to the
sensor. This possible failure is
denoted as Failure B. Since three
Figure 8.7. CF propagation for sensor S2.

97
sensors are installed at different soil depths, several combinations of sensor output
differences can be checked. For sensors SI and S3, the combination is showed in Figure 8.6
and Table 8.1. Sensor S2 has two adjacent sensors. The combinations checked are shown
in Figure 8.7 and Table 8.1. The "R" notation in Figures 8.6 and 8.7 represents rules used
for data uncertainty management. Sensor readings are also compared among sensors at
different locations with the same installation depth. Table 8.2 shows the comparison pairs
to check the possible failure. This possible failure is denoted as Failure C.
Table 8.1 Criteria for checking possible sensor Failure B.
Sensor S1
Sensor S2
Sensor S3
|S1 -S2| > EGV
|S2 SI | > EGV
| S2 S31 > EGV
| S2 S31 < EGV
| S3 S21 > EGV
|S1 -S2| < EGV
Table 8.2 Criteria for checking possible sensor Failure C for sensors
at the same depth from different locations.
Sensor S1
Sensor S2
Sensor S3
iSi-Sil > EGV
| S¡ Si | > EGV
|Si- Sil > EGV
'lote: i and j represent sensors at same depth but in difieren
Propagation of CF
locations.
The three possible failures (Failures A, B and C) reduce confidence in sensor
readings. Overall confidence in sensor readings relies on the final propagation of CF. The
decision that a sensor failure has occurred depends on what CF value is assigned to the

98
Table 8.3 A sample propagation of CF.
Sensors 1 or 3
Instance
Initial CF
F ailure A
Failure B
Failure C
Combined CF
1
0.95
-0.9
i
o
-0.8
-0.75
2
0.95
-0.9
-0.4
N/A
0.17
3
0.95
N/A

1
l
o
bo
0.55
4
0.95
N/A
N/A
-0.8
0.75
5
0.95
N/A

1
N/A
0.91
6
0.95
-0.9
N/A
N/A
0.50
Sensor 2
Instance
Initial CF
Failure A
Failure B
Failure C
Combined CF
7
0.95
-0.9
i
o
4^
-0.4
-0.28
8
0.95
l

VO
-0.4
N/A
0.17
9
0.95
N/A

1
i
o
0.86
10
0.95
N/A
N/A
-0.4
0.92
11
0.95
N/A
i
o
N/A
0.92
12
0.95
-0.9
N/A
N/A
0.50
possible failure and at what confidence level the sensor reading is considered valid. In other
words, the value of the final propagation of CF and the threshold of valid CF values
determine whether a sensor reading is valid or not. For example, if an initial CF value is
assigned as 0.95 and the CF values of Failures A, B, and C are -0.8, -0.7, and -0.4,
respectively, Table 8.3 shows the likely combined CF results. If the value 0.88 was used as
the valid threshold, instances 5, 10, and 11 are valid readings. Instances 4 and 9 would also
be valid if the threshold was reduced to 0.75.

99
Since soil moisture sensors are
installed in at least two locations to
reduce the effects of non-uniform soil
characteristics and crop root
distributions, only the valid sensor
readings from the two sites are
averaged and then used in the
reasoning process. As Figure 8.8
shows, sensor readings (Sx, x=1, 2,
and 3) at the same depth from
different locations (ID1 and ID2) are
averaged if the readings are valid. If
both sensor readings at the same depth
are valid, the average value is used
(Fig. 8.8 (a)). When one of the sensors is considered to have failed, only the valid sensor
reading is used (Fig. 8.8 (b) and (c)). The invalid reading is discarded and a warning
message is displayed for maintenance purposes. When both sensors at the same depth
from different locations are considered to have failed (Fig. 8.8 (d)), both of the readings
at this depth are discarded from the reasoning process. Thus, a dependable decision can
be made when one sensor has failed or when sensor data are missing. Some sample sensor
data are included in Appendix A.
(d)
Note: ID1-Sx = Sensor readngs at different sol depths at location A
ID2-Sx = Sensor readings at afferent sol depths at location B
Figure 8.8. Process of selecting valid sensor
readings from different locations.

100
8.4 Irrigation Management
Irrigation management involves daily decision-making on (1) irrigation strategies,
(2) when to start an irrigation, and (3) when to stop an irrigation.
8.4.1 Irrigation Strategies
Irrigation strategies can be classified as full or deficit irrigation. Full irrigation
provides 100 percent of irrigation demands. In practice, most irrigations are managed as
full irrigation. Deficit irrigation refers to a strategy under which crops are deliberately
allowed to sustain some degree of water deficit. The fundamental goal of deficit irrigation
is to increase water use efficiency and reduce energy costs, while controlling water stress
and yield losses. Irrigation can be managed by user selected full or deficit irrigation, or by
an irrigation strategy obtained from the expert reasoning process. Figure 8.9 illustrates the
decision-making process required to determine an irrigation strategy.
Full irrigation
Young trees are normally managed to grow rapidly for early production. For mature
trees, irrigation during the critical growth stage is crucial to yield. Thus, full irrigation is
applied for young trees and mature trees during the critical growth stages.
Deficit irrigation
Although deficit irrigation potentially increases water and energy saving, this
strategy increases difficulty of irrigation management. In particular, uncertainty of rainfall
is one of the factors.

101
Figure 8.9. Decision process to use a full or deficit irrigation strategy.
Expert reasoning
Expert reasoning determines a full or deficit irrigation strategy according to age of
tree, crop growth stages, and probability of rainfall (Figure 8.9). Since the system can
obtain real-time climate data from the automated weather station, the wet-dry day sequence
for the most recent three days can be used to estimate today's rainfall probability based on
the Markov chain results as discussed in Chapter 6. A high probability of rainfall was
assumed when the rainfall probability was greater than a specified value (for instance, > 60

102
percent). If a weather station is not available, the user can enter probability of rainfall to the
system according to a weather forecasting network.
During the non-critical growth stages, mature trees can sustain some water stress
without causing significant yield losses. A deficit (partial) irrigation strategy may be
suitable for this situation. For mature trees during the non-critical growth stage, deficit
irrigation may be applied when the probability of rainfall is high. Irrigation can be delayed
or less water may be applied if a high probability of rainfall occurs at a given time. Full
irrigation is always suggested for the rest of the growing season.
8,4,2 Criteria for Starting an Irrigation
Irrigation is usually scheduled during a period of low evaporative demand. Thus,
both time and weather constraints are attached to the criteria. Time constraints can be
imposed by regulatory requirements, and weather constraints may result from favorable
weather conditions such as low wind speed and high relative humidity. Overall, time and
weather constraints ensure that irrigation is applied during a low evaporative demand period.
Soil-water potential plays an important role in irrigation management. The critical
value of soil-water potential, related to the maximum allowed depletion, varies with the crop
growth stage. Young trees and mature trees during the critical growth stage require a higher
soil-water content than mature trees during the non-critical growth stage. Irrigation may be
applied when the soil-water potential is in the trigger irrigation range (Figure 8.10).
Irrigation starting criteria based on soil-water potential and constraints of time and weather
are denoted as Criteria I. A set of sample values of Criteria I is shown in Table 8.4.

103
Figure 8.10. Decision process to start an irrigation (criteria I) and sensor readings to
trigger an irrigation for trees during different growth stages.
Table 8.4 Sensor readings and constraints to start an irrigation.
Young
Mature (critical)
Mature (non critical)
Sensor
-8 < S¡< -10 cb
-12 < S¡< -15 cb
-22 < Sj < -25 cb
Time
00:00 am 10:00 am
00:00 am -10:00 am
00:00 am 10:00 am
Wind Speed
< 5 mph
< 5 mph
< 5 mph
RH
> 70%
> 70%
> 70%
Where Si denotes sensor output at depth i (i = 6", 12", and 24").

104
A range of soil-water potentials is specified to increase the possibility of satisfying these
constraints simultaneously.
As Table 8.4 shows, the soil-water potentials of -10 cb, -15 cb, and -25 cb
approximately correspond to 25 percent, 45 percent, and 55 percent soil-water depletion for
Candler fine sand, respectively. These values can be easily modified by the user to satisfy
a unique soil type or management requirements.
Starting irrigation by Criteria I considers both crop water requirements and reduction
of water loss due to evaporation. However, these conditions may not be satisfied
simultaneously even when the sensor readings are over the maximum allowed values.
Irrigation starting Criteria II (Figure 8.11) is used when Criteria I is not satisfied and the
Figure 8.11. Decision process (criteria II) to start an irrigation and critical sensor
readings for trees during different growth stages.

105
soil-water potential is over the maximum allowed value. Only two conditions, soil-water
potentials and growth stages of trees, are required to start an irrigation. An irrigation starts
when the soil-water potentials are over the maximum allowed values. Table 8.5 shows a set
of sample data for Criteria D.
Table 8.5. Critical sensor readings (Criteria II) to start an irrigation.
Young
Mature (critical)
Mature (non critical)
Sensor;
S¡ < R1 (-10 cb)
Si S; Where S¡ denotes sensor outputs at depth i (i = 6", 12", and 24" .
Irrigation Criteria I has higher priority than irrigation Criteria II. Criteria II plays
a role of preventing crop water stress due to extremely low soil-water potential.
Furthermore, sensor readings at different soil depths have the same priority to start an
irrigation. This implies that any one of the sensors can trigger an irrigation if the starting
criteria are satisfied.
8,4,3 Criteria for Stopping an Irrigation
After an irrigation system is turned on, a decision needs to be made on when to turn
it off. Figure 8.12 exhibits alternative approaches to stopping irrigation. Irrigation is
stopped by the following criteria: (1) rainfall, (2) predefined duration, and (3) readings of
soil moisture sensors.

106
Rainfall
base checks rainfall and
irrigation events to prevent these two events from occurring at the same time.
Predefined duration
In practice, irrigations are stopped after certain predefined durations. The duration
of an irrigation could be suggested by irrigation experts. When deficit irrigation is applied
at the user's request or due to a high probability of rainfall, the duration is reduced to a
predefined percentage of full irrigation. In addition, the system also provides a simulation
model to estimate soil-water content at the crop root zone. The simulation model is a simple
water-budget as discussed in Chapter 6. By running the simulation model, a prognosis of
crop water requirements and irrigation duration is given. Thus, the simulation results can
be used as a reference to determine irrigation schedules.

107
Soil moisture sensor reading
Theoretically, irrigation should be turned off when soil-water content reaches field
capacity. Using tensiometers to turn off an irrigation is more difficult than using it to turn
on an irrigation system. This is because of variation of soil characteristics, root distribution,
wetted front of irrigation water, and lag time of tensiometer response. Two practical
questions need to be answered to use tensiometers for this purpose. First, at what value of
soil-water potential should irrigation be turned off? Ideally the value of soil-water potential
should be field capacity. Second, which sensor at the different soil depths should be used
to stop irrigation?
Because tensiometers have a certain response time to irrigation water, irrigation
should be turned off before they indicate that field capacity has been reached. Unlike the
"turn on" process, tensiometers at different root depths should have different priorities when
turning off an irrigation. Deeper sensors should have higher priorities to turn off irrigations.
Since there are some practical problems with using tensiometers to turn off an irrigation, this
is an optional approach in the system.
A new device called wetting depth probe (Zur et al., 1994) may be an alternative
means that can be used to stop irrigations. The probe measures the movement of the wetting
front, and irrigation can be stopped when the wetting front reaches a certain depth.
However, more field tests are needed for practical application of the device.

108
8.5 Cold Protection
Making a decision on cold protection requires balance cold damage risk and resource
conservation. The decision to apply cold protection is associated with (1) when to start cold
protection, (2) when to stop the irrigation, and (3) insuring that there are sufficient resources
(water and energy) available. The system assumes that sufficient resources are available.
Figure 8.13 illustrates the cold protection decision process. An alert message is displayed
when air temperature reaches the warning temperature, which indicates potential cold
damage may occur. Cold protection starts at the critical temperature. After a certain period
of application, cold protection stops when air temperature reaches the stop temperature.
Stop temperature
Critical temperature
On
Off
Time
Figure 8.13. Cold protection decision processes based on the critical air temperature.

109
8.5.1 When to Turn On
The dry bulb air temperature is measured during the decision-making process.
Critical temperature indicates that the crop is in risk of cold damage. This critical
temperature has been determined through cold protection trials. The Institute of Food and
Agricultural Science (IFAS) at the University of Florida recommends initiating cold
protection at a temperature 36F and stopping at 36 to 40 F. Thus, the critical air
temperature used here was assigned the value of 36F.
8.5.2 When to TurnOff
The decision of when to stop cold protection is more difficult to determine and less
certain than that of when to start the cold protection. The rate of air temperature increase
is generally much greater than the rate of fall (Barfield et al., 1990). If the plant part
remains at or below the plant critical temperature for a sufficient length of time, the plant
can be damaged. The decision to stop cold protection must balance the risk and resource
conservation. Cold protection should last long enough to achieve effective protection. The
stop temperature (36F to 40F) recommended by IFAS is used for this system. The critical
temperature to start or stop a cold protection can easily be modified by the user according
to the crop's cold resistance and the environment. Three dummy rules have been developed
based on the decision process.
Rule 1: If air temperature < critical temperature
and cold protection is not on
Then apply cold protection
Rule 2: If air temperature > critical temperature

110
and cold protection is on
Then stop cold protection
Rule 3: If air temperature > critical temperature
and air temperature < critical temperature + 2F
Then display warning message
8 6 Fertigation
Fertigation is used to provide the necessary nutrients to plants. Fertigation requires
proper irrigation to maintain an adequate soil-water content and to enable the plant to utilize
the nutrients. In practice, factors that affect the fertigation application include fertilizer
concentration, application rate, type of fertilizer, and crops. Each irrigation system may
have different application rates and different management strategies. As discussed in
Chapter 7, expert decisions on fertigation are difficult because of the complexity of plant
development. Specific knowledge on crop nutrient requirements is not currently available,
but knowledge on fertigation management is available. For this application, a set of
fertigation schedules was created in the knowledge base. Although the scheduling approach
is not a perfect solution, this approach allows the user to modify the schedule according to
expert recommendation for a specific crop and soil type. By using the schedule, the rule
base continuously checks current time and the fertigation schedule. Fertigation is applied
at the appropriate time in the scheduled period. When irrigation and fertigation schedules

Ill
conflict, fertigation is always arranged at the end of the irrigation. The fertigation should
serve as a part of irrigation whenever possible.

CHAPTER 9
SYSTEM IMPLEMENTATION AND TESTS
9.1 Function Requirements of CIMS
The overall objective of the citrus irrigation management system (CIMS) is to
provide a tool to improve citrus microirrigation management. To achieve this objective, the
software development of CIMS should meet the following requirements:
The system must be easy to use.
The system program must be structured and modularized.
The system should provide control routines to turn on or off the solenoid
values and pumps of an irrigation system.
The knowledge base must realize real-time performance to achieve the
management goals.
The system should be able to deal with sensor data uncertainty.
The system must also provide conventional control approaches and utilities
to satisfy different system hardware requirements.
9,2 Module Design of CIMS
System structure and module design are crucial to achieve the functional
requirements of CIMS. The structural design of CIMS is not only related to these functional
112

113
requirements, but also associated with factors such as ease of use and maintenance. Figure
9.1 shows the main modules of CIMS, which can be categorized as (1) expert system, (2)
control panel, (3) scheduling, (4) database, (5) simulation, (6) tools, (7) help, and (8)
graphical user interface (GUI). Specification of each module is described as following.
Figure 9.1. Program modules of CIMS.

114
9.2.1 Expert System Module
The RTES consists of (1) data input, (2) knowledge base, (3) valve control, and (4)
application log modules. The RTES reads the sensor data and conducts a reasoning process
according to the rule base to make application decisions on irrigation, fertigation, and cold
protection.
Data input modules
The data input modules (read sensor data and reformat data structure) deal with
reading sensor data from external devices, and reorganizing it into the CLIPS data format.
The read sensor data module reads real-time data from the soil moisture sensors and the
weather station at user-specified time intervals. Sensor data are stored in a data logger and
retrieved through a communication link between the computer and data logger. Because
CLIPS requires its own data format, the downloaded sensor data are processed before they
are used by the knowledge base (KB).
Knowledge base IKB1
The KB is the brain of the system. Expert knowledge on citrus irrigation, fertigation,
and cold protection is acquired to build the KB. After the knowledge is acquired, production
rules are used to represent the expert knowledge. These rules are created using the CLIPS
IF-THEN rule-based language and stored in ASCII text format.
The RTES also maintains fact base s (FBs) (Figure 9.1). FBs are used to define facts
such as valve on or off position specifications for each application. The user can modify
these facts to achieve a specific control platform. Control actions are executed based on the
reasoning process. The irrigation system's pump and electronic valves are activated when

115
a decision to irrigate is made after the reasoning process. Data uncertainty management as
discussed in Chapters 3 and 8 is conducted during the reasoning process to handle the
missing and unreliable sensor data. A warning message is displayed when a sensor is
identified to have failed.
Valve control
After a decision is made to start the irrigation system, the external hardware (pumps
and solenoid valves) must be activated. The valve control module consists of three control
procedures: (1) irrigation, (2) fertigation, and (3) cold protection (Figure 9.1). These three
procedures are conventional programs used to turn on or off pumps and electronic valves.
The valve control module is shared by the: (1) RTES, (2) control panel, and (3) scheduling
modules.
Application log
Because the RTES can be operated continuously without a human presence, the user
may not know what control events have been implemented. Application log files are created
to store the detailed information of each control event that was conducted by the system.
From the user's point of view, this function allows past events to be viewed and necessary
reports to be created. For the developer, this log file can be used as a debugging tool.
9.2.2 Control Panel
Besides the RTES, the control panel provides an additional means to control the
irrigation system. This module uses the control screen (Figure 9.2 ) so that the user can turn
control valves on or off by simply clicking a button. The control panel also displays the on
or off status of irrigation, fertigation, and cold protection. Each of these control functions

116
can activate a pump or electronic valves according to the system's pre-defined on or off
status. The user can activate the pump and control valves from a local or remote computer.
ON/OFF Control Screen
ON/OFF
() Irrigation
OFF
O Fertigation
OFF
O Freeze Protection
OFF
Figure 9.2. Control panel of CIMS.
9,2,3 Scheduling
The cost of an RTES can be high because of the hardware and complexity of the
software development, particularly knowledge engineering. Consequently, CIMS provides
an alternative approach to assist irrigation managers. The scheduling module is a
conventional program that enables users to define their irrigation and fertigation schedules
for each irrigated block. Then, the system automatically turns the appropriate irrigation
valves on or off based on the user-defined schedules. Each schedule must contain (1) valve
ID, (2) start time, (3) duration, and (4) skip days or intervals between irrigations. The
schedule module is able to apply user-defined irrigation and fertigation schedules separately
or simultaneously.

117
Since the control action relies only on user-defined schedules, there is no need to use
soil moisture sensors or a weather station in this mode of operation. However, because
irrigation and fertigation fully rely on user-defined schedules, the user must be very
knowledgeable of crop water management and nutrient requirements to specify the schedules
properly.
9.2.4 Database
The Database
module contains several
databases: (1) farm, (2)
weather, (3) irrigation
system, (4) crop, and
(5) soil. Figure 9.3 is
an example of a
database screen. Each
of the database files has
a consistent data entry screen and manipulating functions, such as add, delete, and find, etc.
These databases are used for irrigation management and simulation purposes.
9.2.5 Simulation
Computer simulation is a widely used approach for irrigation scheduling. The
simulation module, as discussed in Chapter 6, contains a water balance model that is used
to simulate soil-water content in the crop root zone. Because weather data can be obtained
from the field weather station, simulation in real time is feasible. Furthermore, a short-term
Figure 9.3. Irrigation system database.

118
prognosis of irrigation requirements can be produced by the simulation. When the system
is operated by user-defined irrigation schedules, the simulation results can be used to assist
the users in defining their irrigation schedules.
9.2.6 Tools
Because evapotranspiration (ET) is one of the major factors for irrigation
management, ET and other management utilities were included in the system.
Load weather data converts weather data from an ASCII file to a database
file. Then the user can manipulate the data using database management
capabilities included in the system.
ET calculation daily or monthly ET can be estimated by the following
methods: (1) Penman, (2) Blaney-Criddle, (3) modified Blaney-Criddle
method using solar radiation, and (4) Stephens-Stewart method.
Irrigation duration estimates duration of a microirrigation event to bring
soil water to field capacity.
Field layout map -- displays the irrigation pipe layout and sensor locations
in the field.
9.2.7 Help
This module consists of several computer tools and user's guide for the system.
These tools include (1) hypertext help on how to use the system, (2) calculator, diary, and
clock, (3) filer and editing utilities, and (4) game.

119
9,2.8 User Interface
Because the user's ability to learn an interface is crucial to software acceptance, the
user interface is a vital factor to ensure success of the system. A graphical user interface
(GUI) is commonly used for the Windows environment in software development. This is
because a GUI has advantage of window, mice, and menu to create a more user friendly
interface. The Windows environment is the current trend in software development.
Therefore, the system uses GUI developed under the Microsoft Windows environment.
9.3 Data and Message Passing of CIMS
Because CIMS is a dynamic system, the sensor data and time are decisive parameters
to the system. Defining data objects and data flows among modules is important for the
system development. Figure 9.4 shows the data flow diagram among the modules.
9,3,1 Data Flow of the RTES Module
The RTES requires input data from soil moisture sensors and a weather station.
Real-time data from tensiometers and the weather station are collected at a given time
interval. Initial facts, such as initial valve on or off status and tree status, should be provided
to the system during the system setup.
The simulation module requires soil, crop, and irrigation system data. The weather
data are also used in the crop water requirement simulation. The scheduling module requires
user-defined irrigation and fertigation schedules. Irrigation and fertigation are controlled
based on the user-defined schedules in the scheduling module.

120
Figure 9.4. Data flow of CIMS.
Control messages can be passed by the RTES and scheduling modules to the valve
control module or the control procedures. Control actions take place when the control
module receives a message from the RTES or scheduling module. All control events
implemented by the RTES are stored in the application log files.
9,3,2 Data Requirements of the Simulation Module
Data required for the water budget simulation are weather, crop, soil, and irrigation
system data (Table 9.1). The simulated results are soil-water content in the crop root zone,
and a one-week prognosis of irrigation requirements. The simulation results can be used
as a reference for the user in irrigation management.

121
Table 9.1 Input data of CIMS.
Weather
Crop
Irrigation
Soil
Solar radiation
(kW/m2)
Tree Coverage (%)
Emitter number per
tree
Soil series
Maximum daily
temperature (F)
Root depth (in)
Emitter flow rate
(gal/hr)
Permanent
wilting point
(ft/ft)
Minimum daily
temperature (F)
Young and mature
tree
Wetted diameter (ft)
Field capacity
(ft/ft)
Wind speed (mph)
Crop coefficient
Overall irrigation
efficiency (%)
Soil depth (ft)
Relative humidity
(%)
Management
allowed depletion
(%)
Rainfall (in)
93,3 Data Requirements of the Scheduling Module
The scheduling module provides the ability to activate or deactivate control valves
according to user-defined schedules. The user needs to input irrigation and fertigation
schedules to the system. Consequently, databases and data entry screens to specify irrigation
and fertigation schedules were developed.
9,4 Maintenance of CIMS
Expert system maintenance is extremely important for the success of the system. A
poorly designed system can increase the difficulty and cost of maintenance. The current

122
approach is an ad hoc development that lacks a standard or practical design approach that
can be generally used. Furthermore, expert systems often must be maintained by someone
other than the developers (Prerau et al., 1990). Therefore, planning for long-term
maintenance when designing an expert system is critical for the system's continued success.
The maintenance strategy must be considered at all design stages.
Maintenance of CIMS involves two tasks: (1) conventional software maintenance,
and (2) knowledge base maintenance. The system's structural design attempts to simplify
the maintenance task so that the user can maintain the system if necessary. The following
steps have been to facilitate system maintenance.
Modularity of rule base:
The production rules are grouped by tasks (Figure 9.5). Rules dealing with
the same event were organized into modules. Since rules in the knowledge
base may need to be changed, this modularity of the rule base can easily
identify which groups of rules need to be changed. In addition, rules are
loaded into the computer memory module by module, which reduces the
computer memory requirements.
User accessible critical values.
The user can directly modify the initial facts, such as tree status and some
critical values to start an irrigation. To do this, the user does not need to
understand the rule base.

123
System modularity:
The system is composed of individual modules. Because each module can
be executed and tested independently, the maintenance task is greatly
simplified.
9,5 System Tests of CIMS
Validation and verification of CEMS are important in the system development.
Common validation approaches include (1) face validation, (2) predictive validation, (3)
Turing tests, and (4) field tests (O'Keefe et al., 1987). For CIMS, validation approaches of
face evaluation, predictive tests, and field tests were conducted. A face validation is a
preliminary approach. Several experts and potential users evaluated the system against their
opinions. The predictive validation requires historical data or generated test cases. Expert

124
systems can be validated by running test cases and comparing results against known results
or expert opinion. Field tests place expert systems in the field, and then seek to perceive
performance errors as they occur.
9.5.1 Predictive Tests
Predictive tests must be conducted at both the system and component level. Many
test cases for a variety of control scenarios were generated to test the reasoning process and
data uncertainty management. The system was validated by running the test cases against
known results and expert opinions. The test results showed that CEMS performed the
reasoning process as was expected. Some sample test cases and reasoning results of the
system are listed in Appendix B. These test cases were also used to debug and verify the
system.
9.5.2 Field Tests
Test site description
Field tests were conducted in the Kresdorn research grove, which is funded by the
Mid-Florida Citrus Foundation and is located at the Conserv II project at Orlando, Florida.
The Kresdorn Grove is a 20-acre grove designed to conduct experiments related to irrigation
demand, fertigation, herbigation, pest management, in-row spacing, and cold protection.
Three-year-old Ambersweet orange trees are planted at this site. Irrigation is applied using
a microirrigation system (micro spray). One emitter was installed per tree. It provides a
360-degree fan pattern of approximately 5 feet diameter. The flow rate of each emitter is
about 16 gallons per hour. Soil type at this site is classified as Candler fine sand.

125
The Conserv II research site is equipped with an automated weather station, data
logger, tensiometers, solenoid valves, computer, and communication system. A fertigation
system is also installed at the site. A variety of chemicals are injected through the
irrigation system.
Tests ofCIMS
The main objective of the field tests was to test the reasoning process and hardware
interactions. The field tests were separated into three parts: (1) control modules, (2)
reasoning process, and (3) complete system. The control modules were tested to turn on or
off user-defined solenoid valves to ensure that the control actions were reliable. Then, the
reasoning process was tested by running a combination of sensor inputs. The control
actions, which are conclusions of the reasoning process, were evaluated against known
results. These tests showed that the reasoning process made decisions as expected to turn
on or off the external devices. Finally, the complete system was tested in the field.
Although long-term field tests are recommended to study the water savings and reliability
of the system, short-term field tests were conducted to avoid interrupting on-going
researches.
9,5,3 Simulated Crop Water Use
To study the possible performance of the system within the limits of this research,
a modeling approach was used. A soil-water budget model combined with Markov chain
probability of rainfall was developed using the concept on which the RTES decision process
is based. The soil-water budget model, as described in Chapter 6, was used to simulate the
soil-water content in the crop root zone with and without the effects of Markov chain

126
probability of rainfall. This simulation model only partially emulated the decision process
of the RTES. The actual system uses soil-water potential measurements in the irrigation
decision-making. Because accurate soil data were not available for the test site, irrigation
was simulated when the soil-water content in the crop root zone is depleted to the
Management Allowed Depletion (MAD) level. The following two assumptions were made
for the simulation:
(1) The irrigation system is properly managed, and the soil-water potentials
measured by tensiometers reflect the MAD listed in Table 6.2.
(2) Because the simulation is conducted at a daily interval, rainfall and irrigation
are assumed not to occur concurrently.
The simulation model applies to both full and deficit irrigation. Full irrigation is
always applied during critical growth stages. Deficit irrigation is only applied to the non-
critical growth stage if necessary. The simulation was conducted for mature trees with root
depth of 2.5 feet, 50 percent shade coverage, and sandy soil with field capacity 0.065 and
permanent wilting point 0.015.
Table 9.2 shows the simulated results of accumulated net irrigation requirements
(NIR) and number of irrigations for 22 years in central Florida. NIR and number of
irrigations were simulated for full irrigation, deficit irrigation, and irrigation affected by the
Markov chain probability of rainfall, respectively. To study the effects of Markov chain
probability of rainfall, the simulation was conducted with different rainfall probabilities
(Cases 1, 2 and 3). For these three cases, irrigation was applied to 80 percent (NIR-80) or
70 percent (NIR-70) of the field capacity when the soil-water content was greater than MAD

Table 9.2 Accumulated citrus net irrigation requirements and number of irrigations for 22 years in central Florida.
Model
SCS
AFSIRS
Soil Water Balance
Case 1 Case 2 Case 3 Case 4
60% rainfall nrobabilitv 50% rainfall probability 40% rainfall probability Deficit irrieation
NIR NIR-80 NIR-70 NIR-80 NIR-70 NIR-80 NIR-70 NIR-80 NIR-70
Irrigation
depth
(inches)
318.34
297.00
310.67
301.52
298.61
285.19
279.82
267.07
251.61
247.39
225.89
Difference
from NIR
(inches)
-7.67
13.67
0.0
9.15
12.08
25.48
30.85
43.60
59.06
63.28
84.78
Number of
deficit
irrigations
46
51
159
181
322
381
441
547
Number of
full
irrigations
596
568
569
518
524
445
447
353
355
Number of
total
irrigations
596
614
620
677
705
767
828
794
902
Note: NIR
NIR-80
NIR-70
scs
AFSIRS
Net irrigation requirements.
Irrigation applied to 80 percent of field capacity.
Irrigation applied to 70 percent of field capacity.
This value is the normal year citrus water requirement estimated by Soil Conservation
Service (SCS, 1982) and multiplied by 22.
Agricultural Field Scale Irrigation Requirements Simulation model (Smajstrla, 1990).

128
and the Markov chain probability of rainfall was greater than the threshold values (60, 50
and 40 percent, respectively). Otherwise, full irrigation was applied when an irrigation was
requiredand the Markov chain probability of rainfall was less than the threshold value. In
other words, less water was applied in anticipation of rainfall when rainfall probability was
greater than the threshold value. For deficit irrigation (Case 4), the irrigation decision ignored the Markov chain
probability of rainfall. In this case, deficit irrigation was always applied when irrigation was
required during the non-critical growth stages.
As Table 9.2 shows, for full irrigation, the SCS (SCS, 1982), AFSIRS (Smajstrla,
1990), and water-budget models computed similar water consumption for the 22-year
period. NIR was reduced and the total number of irrigations was increased when a deficit
irrigation strategy was applied. For the 60 percent Markov chain probability of rainfall,
there was no significant difference in water use (9.15 and 12.08 inches) between full
irrigation and deficit irrigation. When the threshold value of the probability of rainfall was
reduced to 50 and 40 percent (Cases 2 and 3), NIR was reduced (25.48 and 43.40 inches,
respectively) and the number of deficit irrigations was increased.
For deficit irrigation ignoring the Markov chain probability of rainfall (Case 4), even
less water was required than in Cases 1, 2, and 3. However, the total number of irrigations
was increased. This was because deficit irrigation was always applied whenever irrigation
was required. Thus, irrigation was applied more frequently than either full irrigation or
deficit irrigation when Markov chain rainfall probabilities were used. In contrast, Cases 1,
2 and 3 only applied deficit irrigation when the rainfall probability was greater than the
threshold values. Thus, there were more chances to increase effective rainfall and to save

129
water. Although high Markov chain probability of rainfall should be used in the irrigation
decision process, there might be very few rainfall sequences which can match the high
Markov chain probability of rainfall if such threshold value is too high. The results showed
that 50 percent Markov chain probability of rainfall may be an acceptable threshold value
of rainfall probability.
Although Markov chain probability of rainfall, theoretically, provides a better chance
to increase effective rainfall, this result showed that water savings are not significant if the
system only relies on the results of Markov chain probability of rainfall. Finally, this
simulation is only a preliminary study of water savings affected by the decision process used
in the RTES. Because the simulation assumes a well-managed irrigation system, higher
water savings are expected when using an RTES for a conventionally managed system.

CHAPTER 10
SUMMARY AND CONCLUSIONS
As personal computers (PC) become increasingly common, the demand for
computerized farm management tools is increasing. A citrus microirrigation management
system (CIMS) was developed using real-time expert system (RTES) and conventional
control techniques. The CIMS is a PC-based system integrated with RTES, conventional
control, simulation, database, and irrigation management utilities. The system can respond
to external environmental variations and operate continuously in order to make rapidly
decisions without human intervention on matters of irrigation, fertigation, and cold
protection. An automated weather station and soil moisture sensors were used to collect
real-time field data. A knowledge base, which represents the heuristic knowledge of experts
required for the decision-making of citrus irrigation management, was developed for the
reasoning process. Control programs were developed to turn on or off the solenoid valves
and pumps based upon the results of the reasoning process.
With the real-time weather data and soil-water potential, the RTES provides a high
level of automation for microirrigation management. The system has the potential to reduce
labor costs and to improve water, chemicals, and energy conservation. CIMS can be
operated without human presence or may be operated remotely via telephone, cellular or
130

131
radio link. Although this system was designed for citrus microirrigation management, it can
be modified to extend its use to other crops by modifying its knowledge base.
Development of the CIMS involved knowledge of expert system development,
software engineering, control techniques, and knowledge of citrus irrigation management.
This study integrated the knowledge into a single system. Because the development of an
expert system is an ad hoc approach, one of the most arduous tasks in this study was
developing the knowledge base. The complexity of knowledge acquisition was due to some
practical reasons:
Implicitness of the knowledge. The reasoning of experts could not be always
explicitly expressed in a form of heuristic rules.
Time availability of the experts.
Although the RTES is the main subject of this study and it has the potential of
providing a good tool for citrus microirrigation management, the cost of this system can be
high, mainly because of the cost of the software development and hardware requirements.
Therefore, the design philosophy of CIMS is to provide variety of irrigation management
options with different hardware requirements to satisfy the user's need. Thus, besides the
RTES, CIMS provides irrigation management options of (1) user defined irrigation and
fertigation schedules, (2) an irrigation control panel, (3) simulated crop water requirement,
and (4) a calendar for irrigation scheduling. These control and management tools can be
used without knowledge base development. Unlike the dynamic system of the RTES, the
conventional control scheme allows the users to have full control of the schedules of
irrigation and fertigation. Options (3) and (4) do not require any additional hardware to

132
operate. In addition, CIMS also provides farm databases and tools to estimate crop
evapotranspiration and irrigation duration that can be used by the system manager.
CIMS was validated by (1) face validation, (2) predictive validation, and (3) field
tests. Face validation is only a preliminary approach. Experts and potential users were
requested to evaluate the system against their opinions. For predictive validation, many test
data for variety of control scenarios were created to test the system's functionality and
reasoning behavior. Field tests were conducted mainly to test the control components and
reasoning process. The system was tested at the Kresdorn Grove which is located at the
Conserv II water reuse project in Orlando, Florida. The control routines of CIMS have been
successfully implemented and tested at the site for two years. A short-term field test of the
reasoning process of the knowledge base has been conducted at the site. Both the predictive
and field tests showed that the expert system makes decisions as expected to turn on or off
the control system in response to variations of the input data. Irrigation and fertigation
control by user defined schedules worked reliably. More field experiments are needed to
study the long-term performance of the system, especially to assess the crop responses of
different irrigation strategies.
In conclusion, CIMS integrates expert system with control techniques for citrus
microirrigation management. The system provides several options of irrigation management
and it has potential to improve citrus microirrigation management. Therefore, CIMS is new
in its design and application. This work mainly accomplished the following:

133
Integrated a variety of water management techniques into a single system
including an expert system, user defined irrigation control, crop water
requirement simulation, farm databases, and irrigation management utilities.
Developed a new technique for citrus microirrigation management in the real
time domain. This system provides a highly automated tool for irrigation
management and has potential to improve irrigation management.
Potentially, this technique can be applied to other sites or crops.
Applied Markov chain probability of rainfall in the irrigation decision
process. Although the Markov chain probability of rainfall is a well-known
approach, the system applied the approach to an actual expert system in the
process of irrigation decision-making. The simulated results showed that
irrigation decision based on the Markov chain probability of rainfall can
achieve water savings, but the amount of savings is limited. Further studies
are needed to investigate what amount of water can be saved by using the
approach.
Applied the expert system shell CLIPS to a real-time problem domain. Since
the expert system shell CLIPS cannot be directly applied to deal with a real
time problem, efforts were made to achieve reasoning with time and
continuous operation of the system. In addition, uncertainty management of
the sensor data was conducted to increase the system reliability.
In terms of irrigation management, CIMS is in an early stage of development.
Problems faced by decision makers can be much more complex than CIMS addresses.

134
These problems may include pest control, herbicide, emitter clogging, and economic factors.
However, the system as it is has provided a useful tool for irrigation decision makers where
it was tested. More importantly, the primary goal of this study was to develop a
methodology for using RTES and conventional control techniques to improve irrigation
management. CIMS showed its potential to assist the decision-making of citrus
microirrigation managers toward more efficient use of water resources. Further studies are
recommended to improve and extend the knowledge base so that it can deal with more
comprehensive problems and to apply the system to other crops.

APPENDIX A
SAMPLE SENSOR DATA
Weather and tensiometer data at Conserv II, Orlando in 1994
J. Day
Time
Mini
C
MaxT
C
RH
%
Solar
kW/m2
WS
mi/h
Rain
mm
SI-6"
cb
Sl-12"
cb
S2-6"
cb
S2-12"
cb
60
100
16.9
17.1
79.5
0
4.1
0
10.3
9.96
18.5
9.41
60
200
17
17.1
81.4
0
3.6
0
10.3
9.92
18.7
9.39
60
300
16.6
17
83.8
0
1.9
0
10.5
10.1
18.7
9.43
60
400
16.5
16.8
86.2
0
2.2
0
1.25
9.94
18.5
9.16
60
500
16.3
16.5
88.1
0
2.2
0
3.36
10.1
19.1
9.58
60
600
16.3
16.4
88.8
0
2.6
0
4
9.96
19
9.42
60
700
16.1
16.4
88.8
0
3.4
0
4.36
10
19.6
9.8
60
800
16.4
17.5
87.6
0.14
4.1
0
4.55
10
21
10.21
60
900
17.6
20.1
82.6
0.42
7.6
0
4.8
10.1
23.1
10.7
60
1000
20.1
22.1
74
0.87
11
0
4.91
9.06
20.2
8.58
60
1100
21.8
23.2
68.5
0.91
13
0
5.05
8.81
19.5
8.08
60
1200
21.8
22.4
67.49
0.5
13
0
5.1
8.56
19
7.58
60
1300
21.8
23.7
66.97
1.08
12
0
5.11
8.89
21
7.96
60
1400
22.7
23.8
63.48
0.7
13
0
5.09
8.32
19.4
7.03
60
1500
22.5
23.4
66.33
0.61
13
0
5.29
8.03
19.7
7.03
60
1600
22.5
23.2
69.25
0.51
14
0
5.39
8.1
19.6
7.12
60
1700
22
22.7
72.3
0.26
12
0
5.51
8.17
19.8
7.12
60
1800
21.5
22
74.7
0.08
11
0
5.63
8.15
19.9
7.1
60
1900
21.2
21.5
75.3
0
11
0
5.64
8.07
19.9
7.06
60
2000
20.8
21.2
78.7
0
13
0
5.7
8.23
20.3
7.11
60
2100
20.6
20.9
80.4
0
14
0
5.91
8.33
20.5
7.33
60
2200
20.6
20.9
78.2
0
14
0
5.79
8.23
20.6
7.22
60
2300
20.2
20.8
79.5
0
13
0
5.89
8.27
20.3
7.27
60
2400
19
20.5
85.4
0
12
0
5.92
8.31
19.9
7.31
61
100
18.7
19
92.5
0
10
0
5.95
8.36
20.6
7.36
61
200
18.6
18.7
95.2
0
10
1
5.99
8.43
21
7.43
61
300
18.6
18.9
96.8
0
12
9
2.24
8.41
5.22
7.29
61
400
18.9
19.4
96.8
0
14
0
2.96
8.03
5.76
4.939
61
500
19.3
19.8
97.6
0
9.8
9
1.35
4.52
4.69
3.436
61
600
19.7
20.5
98.4
0
9.8
7
2.96
5.04
6.01
4.563
61
700
20.5
20.9
98.3
0
13
3
3.3
5.27
6.19
4.961
61
800
20.9
21.6
98.2
0.02
16
2
2.7
5.17
5.41
4.86
61
900
21.5
21.7
95.8
0.1
18
2
3.23
5.3
6.13
4.905
61
1000
21.6
23.4
90.7
0.73
18
1
3.43
5.6
6.55
5.174
61
1100
22.4
23.9
85.9
0.88
18
1
3.88
5.95
6.73
5.471
61
1200
22.5
24.6
85.7
0.87
17
0
3.98
6.05
7.07
5.74
61
1300
22.2
24.5
77.5
1.52
17
0
4.67
6.57
7.94
6.037
61
1400
22
23.2
70.2
1.42
15
0
3.91
6.25
7.28
6.155
135

61
61
61
61
61
61
61
61
61
61
62
62
62
62
62
62
62
62
62
62
62
62
62
62
62
62
62
62
62
62
62
62
62
62
63
63
63
63
63
63
63
63
63
63
63
63
63
63
63
63
63
63
136
1500
21.6
23.3
66.17
1.08
15
0
4.38
6.57
7.21
6.087
1600
20.4
22
68.61
0.61
14
0
4.08
6.45
6.73
6.329
1700
19.4
21.5
68.57
0.59
15
0
4.66
7.05
7.83
6.755
1800
18.7
19.8
68.11
0.25
15
0
4.59
6.98
7.52
6.599
1900
17.3
18.7
70.7
0.02
14
0
4.74
7.1
7.73
6.723
2000
16.3
17.3
73
0
13
0
4.76
7.12
7.89
6.83
2100
15.6
16.3
76.4
0
13
0
4.73
7.12
7.86
6.859
2200
15.6
16.4
72
0
12
0
4.91
7.2
8.03
6.96
2300
15.8
16.4
70.7
0
15
0
4.95
7.25
8.05
6.993
2400
15
15.7
72.1
0
16
0
5
7.28
8.05
6.976
100
13.9
14.9
71.2
0
16
0
4.99
7.24
8.07
7.03
200
12.7
13.8
69.43
0
19
0
5.08
7.37
8.06
7.11
300
12.1
12.7
71.7
0
17
0
5.18
7.46
8.22
7.23
400
11.7
12.1
76
0
14
0
5.24
7.51
8.27
7.3
500
11.9
12.1
72.9
0
15
0
5.27
7.55
8.31
7.3
600
11.8
11.9
71.8
0
15
0
5.34
7.56
8.37
7.34
700
11.2
11.8
74.8
0
15
0
5.39
7.6
8.44
7.39
800
11.2
11.3
77.6
0.02
13
0
5.4
7.6
8.46
7.39
900
11.2
11.4
76.1
0.12
16
0
5.48
7.69
8.6
7.5
1000
11.4
11.6
75.1
0.22
14
0
5.51
7.73
8.75
7.58
1100
11.3
11.7
73.9
0.26
16
0
5.51
7.71
8.79
7.59
1200
13
14
63.01
2.14
3
0
5.97
8.09
9.15
7.8
1300
14
15.6
57.81
1.55
18
0
5.59
7.92
9.4
7.89
1400
15.4
17
52.89
2.34
17
0
5.45
7.72
8.84
7.77
1500
16.1
16.5
49.63
2.04
17
0
5.04
7.7
8.34
7.16
1600
15.8
16.7
48.39
1.57
16
0
6
8.34
9.44
8.11
1700
14.8
16
51.64
1
15
0
5.65
8.24
8.48
7.31
1800
13.5
15
55.23
0.39
15
0
5.67
8.04
8.74
7.58
1900
12.3
13.5
57.34
0.03
12
0
5.78
8.04
8.65
7.66
2000
11.6
12.2
60.5
0
9.8
0
5.94
8.15
8.72
7.79
2100
10.8
11.7
63.13
0
8.8
0
5.86
8.05
8.76
7.88
2200
9.94
10.8
64.48
0
6.5
0
6.05
8.18
8.85
8
2300
10.1
10.5
62.99
0
7.3
0
5.99
8.22
8.92
7.99
2400
9.11
10.4
61.22
0
4.6
0
5.87
8.03
8.62
7.79
100
7.42
9.06
69.02
0
1.1
0
6
8.12
8.71
7.98
200
5.74
7.54
77.2
0
0.3
0
6.04
8.16
8.75
8.09
300
5.5
7.39
76.1
0
1.1
0
6.1
8.3
9.06
8.22
400
6.95
7.83
66.09
0
3.8
0
6.12
8.27
8.86
8
500
6.55
7.88
63.78
0
2.2
0
5.86
7.91
8.73
7.88
600
4.78
6.48
72
0
0
0
6.06
8.21
8.81
8.08
700
4.71
6.33
75.1
0.01
1.1
0
6.29
8.38
9.42
8.38
800
6.26
9.65
68.19
0.27
2.6
0
6.59
8.76
10.4
8.69
900
9.73
13.6
50.61
0.76
8.4
0
6.1
8.41
9.32
8.05
1000
13.6
15.9
40.65
1.14
9.1
0
6.59
8.54
9.1
7.79
1100
15.8
17.6
36.53
2.02
8.2
0
7.12
9.16
10
8.67
1200
17.3
18.6
34.66
2.32
8.6
0
6.46
8.58
9.72
8.33
1300
18.4
20.6
30.85
2.39
9
0
5.96
8.38
9.02
7.72
1400
20.4
21.4
28.13
2.15
9.7
0
6.3
8.28
9.37
7.96
1500
21.3
22.3
27.34
2.05
11
0
6.16
8.29
9.49
8
1600
20
22.4
28.39
1.24
10
0
6.44
8.54
9.62
8.16
1700
18.8
20.3
30.69
0.76
11
0
6.66
8.87
10.3
8.52
1800
17
19.8
36.85
0.35
9.9
0
6.49
8.43
9.43
7.97

137
63
1900
15.4
16.9
52.34
0.02
7.2
0
6.6
8.5
9.6
8.1
63
2000
13.8
15.5
65.48
0
5.7
0
6.6
8.48
9.54
8.14
63
2100
12.1
13.9
76.2
0
1.9
0
6.67
8.59
9.42
8.26
63
2200
11.6
12.4
83.2
0
0
0
6.71
8.62
9.55
8.36
63
2300
11.2
12.2
86.8
0
0
0
6.6
8.47
9.42
8.27
63
2400
10.6
11.3
90.6
0
0
0
6.59
8.52
9.28
8.28
64
100
9.69
10.7
93.3
0
0
0
6.51
8.44
9.14
8.22
64
200
9.49
10.6
95.7
0
0.8
0
6.48
8.45
9.21
8.21
64
300
9.49
10
96.6
0
0.4
0
6.56
8.57
9.36
8.33
64
400
9.48
9.82
97.5
0
0.5
0
6.6
8.64
9.47
8.4
64
500
9.57
9.95
97.7
0
0
0
6.5
8.49
9.47
8.37
64
600
9.29
10.4
98
0
0
0
6.52
8.46
9.32
8.15
64
700
8.76
9.81
98.2
0.02
0
0
6.86
8.88
9.94
8.69
64
800
9.81
12.5
98.8
0.35
0.6
0
6.97
9.03
10.9
8.94
64
900
12.5
15.7
93.7
0.73
6.6
0
6.77
9.08
10.7
8.8
64
1000
15.7
18.7
75.9
1.04
9.7
0
6.52
8.67
10.5
8.67
64
1100
18.7
20.9
60.12
1.91
9.8
0
6.75
8.67
9.63
8.16
64
1200
20.8
22.7
51.09
2.13
8.8
0
6.84
8.9
9.89
8.31
64
1300
22.4
24
45.6
2.33
7
0
6.84
9.13
9.84
8.18
64
1400
23.8
24.9
42.08
2.31
7.1
0
6.63
8.59
9.69
8.35
64
1500
24.8
25.8
39.98
2.01
7.7
0
7.22
9.04
10.5
8.52
64
1600
25.4
25.8
39.15
1.52
7.7
0
6.44
8.66
9.26
7.48
64
1700
24.1
25.8
41.15
0.98
8.2
0
7.08
9
10.2
8.38
64
1800
21.7
24.1
47.13
0.42
8.1
0
6.82
8.7
9.73
7.8
64
1900
18
21.6
58.47
0.04
5.3
0
6.89
8.62
9.92
8.05
64
2000
14.9
18
71
0
1.2
0
6.92
8.64
9.89
8.14
64
2100
13.8
14.9
82.8
0
0
0
7.06
8.57
9.67
8.16
64
2200
13
13.8
87.9
0
0
0
7.17
8.85
9.78
8.44
64
2300
12.3
13
91.4
0
0
0
6.96
8.64
9.3
8.18
64
2400
12
12.3
94.5
0
0
0
7.17
8.92
9.8
8.71
65
100
11.7
12.2
96.2
0
0
0
7.27
8.89
9.77
8.66
65
200
11.3
11.7
97.3
0.01
0
0
7.19
8.84
9.74
8.63
65
300
11
11.9
98.2
0
0
0
7.33
9.03
10.2
8.89
65
400
11.9
12.3
98.8
0
0
0
7.23
8.88
10.2
8.6
65
500
11.8
12.2
99.1
0
0
0
7.09
8.76
10
8.47
65
600
11.7
12
99.3
0
0
0
7.26
8.89
10.4
8.75
65
700
12
12.5
99.5
0.01
0
0
7.24
8.91
10.6
8.8
65
800
12.5
13
99.5
0.14
0.1
0
7.26
8.99
10.8
8.89
65
900
12.9
14.2
99.4
0.62
0.3
0
7.27
8.84
11.2
8.86
65
1000
13.8
16
99.1
1.12
1.5
0
6.69
8.57
11.3
8.72
65
1100
16
20.8
83.4
1.94
2.4
0
6.61
8.83
10.5
9.29
65
1200
20.9
24.4
59.75
2.25
1.5
0
6.92
8.86
10.3
8.59
65
1300
24.3
26.9
45.16
2.4
4.3
0
6.75
8.67
9.95
8.04
65
1400
26.8
27.5
35.47
2.34
6.4
0
5.78
8.15
9.31
7.65
65
1500
27.3
28.6
31.27
2.02
3.5
0
7.71
9.19
11.4
9.27
65
1600
27.7
29.1
32.95
1.49
3.4
0
7.7
9.25
11.1
9
65
1700
28.1
29.8
32.16
1.05
3.7
0
7.1
8.78
10.9
8.7
65
1800
24.1
28.4
41.71
0.32
3.3
0
7.4
9.03
10.5
8.25
65
1900
22
24
56.56
0.03
3.1
0
7.44
9.02
10.6
8.22
65
2000
20.4
21.9
71.9
0
5.4
0
7.56
9.04
10.8
8.46
65
2100
19.4
20.4
79.7
0
1.6
0
7.6
9.1
11
8.61
65
2200
18.3
19.4
85
0
0.8
0
7.51
8.96
11
8.49

65
65
66
66
66
66
66
66
66
66
66
66
66
66
66
66
66
66
66
66
66
66
66
66
66
66
67
67
67
67
67
67
67
67
67
67
67
67
67
67
67
67
67
67
67
67
67
67
67
67
138
2300
17.4
18.2
90.1
0
0
0
7.62
9.05
11.1
8.69
2400
17.5
17.6
90.1
0
0.4
0
7.57
9
11.1
8.69
100
17.4
17.7
87.7
0
0.4
0
7.53
8.91
11.1
8.63
200
17.1
17.4
86.8
0
0.1
0
7.51
8.91
11.1
8.66
300
16.8
17.2
86.5
0
0
0
7.48
8.92
11
8.61
400
16.4
16.8
88.1
0
0
0
7.58
9.05
11.2
8.75
500
16.4
16.7
85.9
0
0
0
7.5
9
11
8.55
600
16.2
16.4
87.1
0
0
0
7.63
9.1
11.3
8.91
700
16.3
16.8
85
0.01
0.3
0
7.57
9.08
11.2
8.79
800
15.7
18.4
86.8
0.26
0
0
8.14
9.46
13
10.3
900
18.5
21.8
69.64
0.81
1.6
0
7.86
9.19
10.9
8.75
1000
21.8
23.8
61.48
1
2.1
0
7.85
9.07
11.9
9.03
1100
23.8
25.7
55.35
1.81
3.1
0
7.19
8.5
11
8.78
1200
25.2
26.6
47.71
1.97
3.4
0
7.45
9.01
11.3
9.15
1300
27.5
27.6
39.81
0.03
1.2
0
6.86
8.25
9.45
7.08
1400
27
27.2
38.62
0.27
0.9
0
8.42
9.64
11.6
9.13
1500
26.4
28.5
38.49
6.62
0.7
0
7.57
8.41
10.4
7.66
1600
25.8
28.4
39.68
4.44
0.2
0
7.92
9.22
11.3
8.5
1700
25.4
28.8
39.88
2.74
0
0
7.54
8.52
11.2
8.21
1800
25.7
28.2
40.37
2.49
0.2
0
8
9.02
11.8
8.58
1900
22
26.1
52.64
0.21
0.3
0
8.17
9.13
11.6
8.48
2000
20.2
22
68.58
0.03
0.4
0
8.27
9.15
12
8.75
2100
19.1
20.2
78.1
0.03
0.1
0
8.35
9.14
12.1
8.77
2200
18.1
19.2
84.5
0.03
0
0
8.28
9.06
12
8.78
2300
17.2
18.2
89.6
0.03
0
0
8.39
9.1
11.8
8.85
2400
16.6
17.2
92.7
0.03
0
0
8.31
9.02
11.5
8.79
100
16.1
16.7
94.8
0.03
0
0
8.23
8.99
11.4
8.78
200
15.4
16.2
95.8
0.03
0
0
7.95
8.81
10.5
8.54
300
14.4
15.6
96.8
0.05
0
0
7.89
8.9
10.7
9.11
400
14.2
14.7
98
0.06
0
0
7.94
9
10.3
8.91
500
14.2
15
98.6
0.05
0
0
8.32
9.16
10.9
9.21
600
14.5
15.4
99
0.03
0
0
8.4
9.21
11.3
9.19
700
14.8
15.5
99.1
0.11
0
0
8.62
9.36
11.6
9.59
800
15.1
17.6
99.3
2.16
0
0
9.14
9.36
13.8
10.36
900
17.5
19.6
99.2
5.07
0
0
8.33
8.92
11.9
9.42
1000
19.6
24
84.8
6.33
0.7
0
8.72
9.14
13.5
9.77
1100
24
25.7
61.45
11.7
1
0
9.63
9.73
13
9.52
1200
25.7
27.6
50.63
14
0.9
0
8.5
9.03
11.5
8.53
1300
26.8
28
46.89
14.2
0.9
0
8.54
9.1
12.7
9
1400
27.7
28.7
44.1
13.5
1.1
0
8.71
9.52
13.1
9.78
1500
27.7
30.2
41.91
11.7
0.9
0
8.37
8.47
12.6
8.95
1600
27.7
30.8
41.28
9.02
0.7
0
9.83
10.3
13.2
10.25
1700
26.7
28.2
50.14
4.18
1.1
0
8.69
9.12
12.6
9.61
1800
25.4
27.5
54.91
1.41
1
0
8.97
9.49
13.5
9.53
1900
22.3
25.4
63.6
0.16
0.4
0
8.87
9.49
13.7
9.16
2000
21.3
22.3
73.7
0.03
0.6
0
9.07
9.52
14.2
9.31
2100
20.4
21.4
82.2
0.03
0.8
0
8.99
9.41
14.2
9.14
2200
19.7
20.4
85.9
0.03
0.7
0
9
9.35
14.2
9.11
2300
18.6
19.7
90
0.03
0
0
9.14
9.48
14.1
9.27
2400
18.1
18.7
93.7
0.03
0
0
9.15
9.48
13.9
9.27

139
Rain (cm) -o- S1-6" _^S1-12" S2-6" _H_S2-12"
Figure 12.1. Tensiometer readings on Juilan day 213, 1994.
Note: SI-6"
51-12"
52-6"
S2-12"
Tensiometer SI-6"
Tensiometer at site 1 with 6" depth.
Tensiometer at site 1 with 12" depth.
Tensiometer at site 2 with 6" depth.
Tensiometer at site 2 with 12" depth,
was failed at 1400 minutes on day 213.

140
-+- Rain (cm) S1-6" ^_S1-12" -*-82-6" __S2-12"
Figure 12.2. Tensiometer readings on Juilan day 278, 1994.
Note:
SI-6"

Tensiometer at site 1 with 6" depth.
Sl-12"

Tensiometer at site 1 with 12" depth.
S2-6"

Tensiometer at site 2 with 6" depth.
S2-12"

Tensiometer at site 2 with 12" depth.
Tensiometer Sl-6"
had bad data at 1400 minutes on day 213 and recovered later.

141
Rain (cm)-o-S1-6" -^SI-12" S2-6" _^_S2-12"
Figure 12.3. Tensiometer reading on Juilan day 279, 1994.
Note:
SI-6"
51-12"
52-6"
S2-12"
Tensiometer SI-6"
Tensiometer at site 1 with 6" depth.
Tensiometer at site 1 with 12" depth.
Tensiometer at site 2 with 6" depth.
Tensiometer at site 2 with 12" depth,
showed low confidence and might not be failed completely.

APPENDIX B
TEST CASES AND RESULTS
B. 1 Test Data Description
The following test data are in CLIPS fact format.
Test data Data description
(TREE-STATUS YOUNG) Young trees are planted in the site
(TREE-STATUS MATURE)
(sensor-reading idl 6 13 12 12 24 16)
(sensor-reading id2 6 9 12 12 24 14)
(time-constrain no)
(weather-constrain no)
(weather-data ws 30 rh 70
rain 0.0 airtemp 38)
(FTSCH1 MONDAY START-TIME
3 10 END-TIME 4 10)
(FTSCH2 1994 11 23 START-TIME
3 20 END-TIME 3 30)
EDl-i
ID2-
Mature trees are planted in the site
Tensiometer readings at site 1 (idl)
at depth 1, (6"~12 cb), 2 (12' 12
cb), and 3 (24 "--16 cb)
Tensiometer readings at site 2 (id2)
at depth 1, (6"~9 cb), 2 (12' 12
cb), and 3 (24"14 cb)
Irrigation without time constraint
Irrigation without weather constraint
Weather data with wind speed 30
mph, relative humidity 70 %, rainfall
0.0 inches, and air temperature 38 C
Fertigation schedule by day of the
week. Fertigation is scheduled on
Monday. It starts at 3:10 a.m. and
ends at 4:10 a.m.
Fertigation schedule by date of the
year. Fertigation is scheduled on
November 23, 1994. It starts at 3:20
a.m. and ends at 3:30 a.m.
Sensor at site 1 and depth i (i = 1, 2,
and 3)
Sensor at site 2 and depth i (i = 1, 2,
and 3)
142

143
B.2. Some Critical Data for the Reasoning Process
Young trees: 12 to 15 cb with constraints and 15 cb without constraints.
Mature trees during critical growth stage: 15 to 20 cb with constraints and 20 cb
without constraints.
Mature trees during non critical growth stage: 25 to 30 cb with constraints and 30
cb without constraints.
Cold protection critical temperature: 36F.
Threshold Value of Confidence Factor is 80 percent.
B.3, Test Cases and Results
Test Cases Results
Test case 1
(TREE-STATUS YOUNG)
(sensor-reading idl 6 13 12 12 24 16)
(sensor-reading id2 6 9 12 12 24 14)
(time-constrain no)
(weather-constrain no)
Reject sensor reading at
ID 1-3 (CF = 0.75).
Turn on by sensor at ID2-3.
Test case 2
(TREE-STATUS YOUNG)
(sensor-reading idl 6 9 12 13 24 9)
(sensor-reading id2 6 9 12 9 24 9)
(time-constrain no)
(weather-constrain no)
Data validation OK.
Not turn on.
Test case 3
(TREE-STATUS YOUNG)
(sensor-reading idl 6 9 12 9 24 13)
(sensor-reading id2 6 9 12 9 24 9)
Data validation OK
Not turn on

144
(time-constrain no)
(weather-constrain no)
Test case 4
(TREE-STATUS YOUNG)
(sensor-reading idl 6 9 12 9 24 9)
(sensor-reading id2 6 13 12 9 24 9)
(time-constrain no)
(weather-constrain no)
Test case 5
(TREE-STATUS YOUNG)
(sensor-reading idl 6 9 12 9 24 9)
(sensor-reading id2 6 9 12 13 24 9)
(time-constrain no)
(weather-constrain no)
Test case 6
(TREE-STATUS YOUNG)
(sensor-reading idl 6 9 12 9 24 9)
(sensor-reading id2 6 9 12 9 24 13)
(time-constrain no)
(weather-constrain no)
Test case 7
(TREE-STATUS YOUNG)
- 0.28),
(sensor-reading idl 6 16 12 9 24 9)
(sensor-reading id2 6 16 12 9 24 9)
Test case 8
(TREE-STATUS YOUNG)
ID2-2
(sensor-reading idl 6 11 12 16 24 11)
(sensor-reading id2 6 11 12 16 24 11)
Data validation OK
Not turn on
Data validation OK
Not turn on
Data validation OK
Turn on by sensor at depth 3
Reject data from ED1-1(CF=
ID2-2 (CF = 0.28)
Not turn on
Reject data from ID 1-2 and
(CF = 0.7)
Not turn on

145
Test case 9
(TREE-STATUS YOUNG)
ID2-3
(sensor-reading idl 6 9 12 9 24 16)
(sensor-reading id2 6 9 12 9 24 16)
Test case 10
(TREE-STATUS YOUNG)
(sensor-reading idl 6 10 12 10 24 10)
(sensor-reading id2 6 19 12 10 24 10)
Test case 11
(TREE-STATUS YOUNG)
(sensor-reading idl 6 12 12 10 24 10)
(sensor-reading id2 6 12 12 10 24 19)
Test case 12
(TREE-STATUS YOUNG)
(sensor-reading idl 6 10 12 10 24 10)
(sensor-reading id2 6 10 12 10 24 19)
Test case 13
(TREE-STATUS MATURE)
(CRITICAL GROWTH)
(sensor-reading idl 6 17 12 9 24 9)
(sensor-reading id2 6 16 12 9 24 9)
(time-constrain no)
(weather-constrain no)
Test case 14
(TREE-STATUS MATURE)
(CRITICAL GROWTH)
(sensor-reading idl 6 9 12 9 24 17)
(sensor-reading id2 6 9 12 9 24 16)
(time-constrain no)
(weather-constrain no)
Reject data from ID 1-3 and
Not turn on
Reject data from ED 1-1
Not turn on
Reject data from ID2-3
(CF=0.7)
Turn on the system by ID1-1
or ID2-1
Reject data from ID2-3
Not turn on
Data OK
Turn on by sensor ID1-1 or
ED2-1
Data OK
Turn on by sensor ID 1-3 or
ID2-3

146
Test case 16
(TREE-STATUS MATURE)
(CRITICAL GROWTH)
(sensor-reading idl 6 15 12 14 24 13)
(sensor-reading id2 6 17 12 14 24 13)
(time-constrain no)
(weather-constrain no)
Test case 17
(TREE-STATUS MATURE)
(CRITICAL GROWTH)
(sensor-reading idl 6 14 12 15 24 13)
(sensor-reading id2 6 13 12 16 24 13)
(time-constrain no)
(weather-constrain no)
Test case 18
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 28 12 24 24 24)
(sensor-reading id2 6 24 12 24 24 24)
(time-constrain no)
(weather-constrain no)
Test case 19
(TREE-STATUS MATURE) Data OK
(NON-CRITICAL GROWTH) Turn on by ID 1 -1 or
(sensor-reading idl 6 28 12 24 24 24) ID2-3
(sensor-reading id2 6 24 12 24 24 26)
(time-constrain no)
(weather-constrain no)
Test case 20
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 24 12 28 24 24)
(sensor-reading id2 6 24 12 24 24 24)
Data OK
Turn on by ID 1-1
Data OK
Turn on by ID2-2
Data OK
Turn on by sensor ID1-1 or
ID2-1
Data OK
Turn on by ID 1-2

147
(time-constrain no)
(weather-constrain no)
Test case 21
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 24 12 24 24 32)
(sensor-reading id2 6 24 12 24 24 24)
(time-constrain no)
(weather-constrain no)
Test case 22
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 24 12 24 24 24)
(sensor-reading id2 6 28 12 24 24 24)
(time-constrain no)
(weather-constrain no)
Test case 23
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 24 12 24 24 24)
(sensor-reading id2 6 24 12 28 24 24)
(time-constrain no)
(weather-constrain no)
Test case 24
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 24 12 18 24 18)
(sensor-reading id2 6 18 12 18 24 18)
Test case 26
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 18 12 24 24 18)
(sensor-reading id2 6 18 12 18 24 18)
Reject data from ID 1-3
Not turn on
Data OK
Turn on by ID2-1
Data OK
Turn on by ED2-2
Data OK
Not turn on
Data OK
Not turn on

148
Test case 27
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 18 12 18 24 24)
(sensor-reading id2 6 18 12 18 24 18)
Test case 28
(TREE-STATUS MATURE)
(sensor-reading idl 6 18 12 18 24 32)
(sensor-reading id2 6 24 12 18 24 18)
(NON-CRITICAL GROWTH)
Test case 29
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 18 12 18 24 18)
(sensor-reading id2 6 18 12 24 24 18)
Test case 30
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 18 12 18 24 18)
(sensor-reading id2 6 18 12 18 24 24)
Test case 31
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 36 12 28 24 28)
(sensor-reading id2 6 28 12 28 24 28)
Test case 32
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 28 12 36 24 28)
(sensor-reading id2 6 28 12 28 24 28)
Data OK
Not turn on
Reject data from ID 1-3
(CF= -0.28)
Not turn on
Data OK
Not turn on
Data OK
Not turn on
Reject data from ED 1-1
Turn on the system
Reject data from ID1-1
Turn on the system

149
Test case 33
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 28 12 28 24 36)
(sensor-reading id2 6 28 12 28 24 28)
Test case 34
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 28 12 28 24 28)
(sensor-reading id2 6 36 12 28 24 28)
Test case 35
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 28 12 28 24 28)
(sensor-reading id2 6 28 12 28 24 36)
Test case 37
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 68 128 24 3)
(sensor-reading id2 6 8 12 8 24 5)
Test case 38
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 8 12 8 24 5)
(sensor-reading id2 6 8 12 8 24 3)
Test case 39
(MARKOV MONTH 6 W D D 0.71)
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(MONTH 6 W D D)
Reject data from ID 1-3
Turn on the system
Reject data from ID2-1
Turn on the system
Reject data from ID2-3
Turn on the system
Data OK
Not turn on
Data OK
Not turn on
Apply deficit irrigation

150
Test case 40
(weather-data ws 30 rh 70 rain 2.0 airtemp 60)
Test case 41
(weather-data ws 30 rh 70 rain 0.0 airtemp 35)
Test case 42
(weather-data ws 30 rh 70 rain 0.0 airtemp 37)
protection.
Test case 43
(weather-data ws 30 rh 70 rain 0.0 airtemp 38)
Test case 44
(FTSCH1 MONDAY START-TIME
3 10 END-TIME 4 10)
Test case 45
(FTSCH2 1994 11 23 START-TIME
3 20 END-TIME 3 30)
Not turn on cold protection
Turn on cold protection
Warning message for cold
Not turn on cold protection
Fertigation starts at 3:10 a. m.
and ends at 4:10 a m. every
Monday.
Fertigation starts at 3:20 a.m.
and ends at 3:30 a.m. on
November 11, 1994.

APPENDIX C
CIMS USER'S GUIDE

152
13.1. Brief Description of CIMS
CIMS is a computer-based citrus microirrigation management system that runs under
Microsoft Windows. CIMS provides a tool to control irrigation, fertigation, and cold
protection to improve citrus irrigation management. The system consists of several main
functions: (1) real-time expert system (RTES), (2) irrigation control panel, (3) irrigation and
fertigation control based on user defined schedules, (4) database, (5) simulated crop water
requirements, (6) tools, and (7) help. Figure 13.1.1 shows the main menu of CIMS.
Eacts Expert .Control Scheduling database Simulation Iools Help Quit
Figure 13.1.1. Main menu of CIMS.
13,1,1 System Requirements and Installation
Personal computer with 386 or higher CPU running under Microsoft
Windows 3.1 or later version
4 MB available memory
5 MB free space on hard space
VGA monitor
Mouse
Installing CIMS
Insert the disk labeled Setup into your computer drive A. Type A:SETUP,
and press . Follow the instructions on the screen to complete the
installation.

153
How to run CIMS
To launch CIMS, start Microsoft Windows and double click CIMS icon.
For information about how to use Windows, refer to Microsoft Windows' Reference. The
procedures to run each of the main modules of CIMS are described in the subsequent
sections.

154
13.2. Real-Time Expert System
The RTES uses expert knowledge and data from on-site sensors to make decisions
on citrus irrigation, fertigation, and cold protection. The RTES has two modules: Facts and
\
Facts
Expert .Control Scheduling database Simulation Tools Help .Quit
Irriqat
on
fertigation
Freeze Protection
Initial Facts
Fertigation Schedule
Figure 13.2.1. Submenu of the Facts.
Expert as Figure 13.2.1. The Facts module defines the initial facts for the expert system.
The facts can be (1) valve on or off definition for the applications of irrigation, fertigation,
or cold protection, (2) threshold values and the crop growth stage for the decision-making,
(3) fertigation schedules. The Expert module executes the expert system to perform the
following tasks: (1) read data from the soil moisture sensors and the weather station, (2) read
initial facts defined in the Facts module, (3) conduct reasoning process based on the
knowledge base it contains, and (4) take control actions to activate or deactivate irrigation
control valves and pumps.
13.2,1 Define Initial Facts
Click Facts on the main menu and submenu of the Facts showed in Figure 13.2.1.
The steps to define the initial facts are described as follows.

155
Define irrigation control valve (on or off)
Click Irrigation on the submenu of Facts, and an editing window (Figure A.2.2) is
displayed.
Enter the irrigation block and
valve numbers in the first two
columns. Then enter the valve
on or off value. Value 1
represents the valve will be
turned on and value 0 represents
that the valve will not be turned
on when an application occurs.
Thus, irrigation and fertigation
can be applied only on the
predefined blocks.
Define cold protection control valve (on or offl
Click Freeze Protection on the submenu of Facts.
The procedures to define a valve on or off for cold protection are same as the
procedures for defining the valve on or off for irrigation. For cold
protection, all valves may be specified as on (value 1) when the application
is required.
Figure 13.2.2. Irrigation block and valve
definition.

156
Define fertieation control valve (on or off)
Fertigation application is different from irrigation and cold protection. Fertigation
is applied in a sequence of preinjection, fertigation, and flush. After the irrigation pipe line
is pressurized during preinjection, fertigation is then applied. Because the potential back
flow and the chemicals may remain inside the pipe line, a flush period is needed to flush the
chemicals out of the pipe line after chemical injection is stopped. The steps to define
fertigation facts are as follows.
Click Fertigation on the submenu of Facts. An editing window is displayed
as Figure 13.2.3.
Enter the number of
preinjection time and
flush time (minutes) in
lines two and four.
Define the valve on or
off status using the same
procedures of irrigation
valve definition.
Figure 13.2.3. Fertigation block and valve
definition.
Define or modify initial facts
Click Initial Facts to open the fact editing window (Figure 13.2.4).
Select the facts to edit. Use semi-colon at the beginning of each line to
omit the line from the initial facts.

157
The initial facts include the
following data:
irrigation strategy,
threshold values to
start and stop an
irrigation,
cold protection
threshold values,
crop growth stages,
percentage of less
irrigation time,
critical and non-critical growth stages, and
Markov chain rainfall probability.
Default initial facts with
detailed comments are included in the
system. The facts are specified in the
data format of the expert system shell
CLIPS. The user can modify those
critical values to achieve his or her
control need. Facts can be eliminated
from the reasoning process by adding
a semi colon in front of the facts.
Figure 13.2.5. Fertigation schedule.
(a)
(b)
(c)
(d)
(e)
(f)
(g)
Figure 13.2.4. Initial facts of the expert system.

158
Define fertigation schedule
Fertigation schedules can be defined as two formats: by date or by day of week.
Click Fertigation Schedule to open the editing window (Figure 13.2.5).
Enter the starting and ending time. To omit a schedule, use the semi-colon in
front of each schedule.
A sample fertigation schedule by day of week is shown below.
(FTSCH1 MONDAY START-TIME 3 10 END-TIME 4 10)
FTSCH1
schedule ID,
MONDAY
day of the week,
START-TIME 3 10 -
fertigation starting hour and minute, and
END-TIME 4 10
fertigation ending hour and minute.
A sample of the fertigation schedule by date showed below.
(FTSCH2 1994 11 23 START-TIME 3 20 END-TIME 3 30)
FTSCH2
fertigation schedule ID,
1994 11 23
year, month, and day,
START-TIME 3 10 -
fertigation starting hour and minute, and
END-TIME 4 10
fertigation ending hour and minute.
Note that all the text in the facts must be upper case and each attribute must be
separated by at least one space.
13 2,2 Execution of the RTES
Before executing the RTES, initial facts must be defined. In particular, the on-off
status of control valves must be defined for each irrigated block. To run the RTES, click

159
the Expert in the main menu (Figure 13.2.6). A submenu of Execute RTES and View
Applications will be displayed. Then complete the following steps.
Figure 13.2.6. Submenu of the Expert main menu.
Ini
RTES Execute Screen
Select modules to execute
0 Irrigation
Exl Fertigation
f<] Cold Protection
1
Execute
j
E3 Initialize
1
Note:
To execute this RTFS, soil
moisture
sensor and weather station must
1
Figure 13.2.7. Execution screen of the RTES.
Click Execute RTES and a dialog window appears as Figure 13.2.7.
Select the check box if necessary from the dialog window. The check boxes include
Irrigation, Fertigation, and Cold protection. If the box is checked, the RTES will
perform the reasoning process on this subject. The check box of Initialize should
always be selected.
Click the Cancel button to exit the screen and click the Help button to view the help
messages regarding this screen.
Click the Execute button to run the RTES. The knowledge base will be loaded into
memory and the external devices are controlled by the conclusions of the decision-

160
making process. The reasoning process will continue until interrupted by a user. To
interrupt the continuous process, double click left button of the mouse. Data from
the soil moisture sensors, the weather station, and system on or off status of an
application are displayed as Figure 13.2.8 during the real-time reasoning process.
Figure 13.2.8. Screen of sensor readings and application status.
View application history
Since the RTES runs continuously, control actions can occur any time due to the
variation of input data and the reasoning process. The user may not know what has
happened in the past. Therefore, the systems automatically store control actions to a history
log file. To view the application history, click View Application on the submenu of Expert
(Figure 13.2.9). Click either Irrigation, Fertigation, or Freeze Protection to view the past
Figure 13.2.9. Submenu to view application history.

161
events on the subject. The application history files display the starting and ending time of
each event which has occurred.
13.3 Control Panel
CIMS provides an irrigation control screen so that the user can turn on or off the
control valves by simply clicking a button from the control screen. The steps to run the
control panel are described as follows.
Click the Control on the main menu. A control panel screen appears as Figure
13.3.1.
Figure 13.3.1. Control panel of the system.
Select radio button Irrigation, Fertigation, or Freeze Protection for a particular
application.
Click the Cancel button to exit the screen and click the Help button to view help
messages on the screen.
Click the Turn On or Turn Off button to turn on or off the control valves. This
procedure turns on or off the valves according to the user predefined valve on or off
status from Facts menu.

162
13 4 Irrigation and Fertigation Control Based on User Defined Schedules
Because the RTES requires soil moisture sensors and a weather station, cost of the
system can be high due to the hardware requirement. Thus, an alternative approach of
irrigation and fertigation control scheme was developed. This approach allows the users to
define their own application schedules. Thus, the control actions can be executed based on
the schedules. Since the control events rely only on the user defined schedules, soil-water
content and climate data are not required. However, the user must be knowledgeable of the
crop water and nutrient requirement to define a sound application schedule.
13,4,1 Define Irrigatipn and Furtigation Schedule
The steps to define irrigation and fertigation schedules are as follows.
Click Scheduling on the main menu of CIMS. The submenu of Scheduling shows as
Figure 13.4.1.
Facts Expert Control
Scheduling
Irrigation Schedule
Apply Irrigation
Database Simulation Tools Help Quit
Fertigation Schedule
Apply Fertigation
Apply Irrigation & Fertigation
Figure 13.4.1. Submenu of user defined control schedules.
Click Irrigation Schedule to define an irrigation schedule (Figure 13.4.2).
Enter the data showed in Figure 13.4.2. These data include following parameters.

163
Figure 13.4.2. User defined irrigation schedule screen.
Figure 13.4.3. User defined fertigation schedule screen

164
Valve number: irrigation control valve number,
Start time: irrigation start time (hour and minute) in 24-hour format,
Duration: irrigation duration (hour and minute) in 24-hour format, and
Skip day: irrigation skip days.
Fertigation schedules use the same data structure that irrigation uses. To define
fertigation schedule, click Fertigation Schedule on the submenu of Scheduling. Then, enter
the similar data that irrigation uses (Figure 13.4.3).
13.4.2 Apply Irrigation
Irrigation can be applied based upon the user defined irrigation schedule. The system
continuously checks the user defined irrigation schedule against the computer clock. If the
computer date and time satisfy the irrigation schedule, the irrigation valves, then, are
activated. The steps to run the user defined irrigation are as follows.
Figure 13.4.4. Irrigation application screen.

165
Click Apply Irrigation on the submenu of Scheduling (Figure 13.4.1). A
dialog screen appears as Figure 13.4.4.
Click the Cancel button to exit the window.
Click the Proceed button to apply irrigation based upon the user defined
irrigation schedule. A screen (Figure 13.4.5) displays the current time and
on or off status of the irrigation valves.
Double click left button of your computer mouse to interrupt the continuous
process.
Valve ON/OFF Status
it
ON
OFF
T to
e*a Sk.t>
Tto Day
¡i
m
OFF
mm¡M
lima
End Slop
Tim* Day
#
ON
OFF
Start
Tto
Cnd 6
Piiiil
kip
ip
t
OFF : f
8
8
8
2
0
¡181
OFF
3
9
9
8
8
OFF
8
8
9
9
§1
1
OFF
9
D
9
4
28
OFF
3
$
8
8
ill
l
OFF
3
9
HI
8
1

OFF
I
8
."8
i
0
29
OFF
3
0
9
8
ill
'-36-'
OFF
8
8
3
8
i
*
III
8
11
8
2$
OFF
B
8
8
a
8
38
OFF ;
3
3
9
8
¡§1

OFF
8
0
8
2
8
a*
OFF"
111
0
S
8
0
92
OFF
3
6
3
lit
fH
m
-'OFF'
8
8
3
0
8
2?
OFF
3
8
3
a
fil
93
OFF--;-;
s
3
9
s
1
m
OFF
9
C
S
8
8
28
OFF
S

fill
8

OFF
3
9
9
8
tli
1
OFF
¡111
8
3
8
23
OFF
3
a
a
n
SO
OFF
8
3
3
8

i
OFF-'
18
Hi
III
0
8
39
OFF
3
0
8
a
a
61
OFF
8
9
1
8
Pi
8
OFF
8
0
8
0
8
31
OFF '
3
8
3
8
8
$2
OFF
3
9
3
8
g
1
OFF
8
8
8
8
8
:
OFF
3
1
3
8
8
53
OFF
3
9
9
8
!!
OFF
8

mi
i
8

OFF
3

Hi
8
8
¡¡I
OFF
8
9
iff
8
¡1
m
OFF
3
111:
8'
ill
8
m
OFF
3
8
3
8
ill
S3'
OFF
3
9
9
8
¡I
m
OFF
ill
0
3
0
nil
m
I
3
8
'
8
8
S3
OFF
3
9
9
m
8
B
6
m
OFF
OFF
OFF
HI
8
3
0
ill
8
- 3
111
9
8
8
8
8
8
8

V
88
OFF
¡If
.'Off!
3
3
8
8'
¡II
8
Hi
-'3¡
3
111
0
8
H
8
8
58
se

Off
iilii
UP
#lil
3
9
9
9
3
9
5
lit
3
¡ii
3
If
ii
ii
Figure 13.4.5. Irrigation valve on or off display.
13,4,3 Apply Fertigation
Similarly, fertigation can be applied according to user defined fertigation schedules.
The program checks user defined fertigation schedule against the computer date and time
to turn on or off the control valves and fertigation pump. Fertigation is applied in a
sequence of preinjection, fertigation, and flush.

166
Figure 13.4.6. Fertigation dialog screen.
Procedures to apply the user defined fertigation schedule are
Click Apply Fertigation on the submenu of Scheduling. A dialog screen appears to
define preinjection time and flush time (Figure 13.4.6).
Click the Cancel button to exit the dialog screen.
Enter preinjection and flush time in minutes showed in Figure 13.4.6.
Click the Proceed button to apply fertigation according to the user defined schedule.
A screen display on or off status of the control valves and pump (Figure 13.4.7).
Double click left button of the mouse to interrupt the system.

167
Valve ON/OFF Status
m
ON
Start
£mt
Skip
ON
Sta*
ill
i
mrnmm.
Skip
ON
Stef
ill
Ill
11!
Hup
#
OFT
Tita
Ttata Day
Hi
OFFSSSSS
Tinta
Tft Day
ste
pins
teateisi;
Tart Day
OFF
*
8
Wk
t
mm
28
OFF
Site
wm
mm
mm
0
m
OFF
iiis
ft

9
8
't
OFF
*
8
9
t
1

OFF
9
0
9
0
9
44
OFF
s
8
8
ft
lift
m
OFF
9
0
9
8
2
24
OFF
3
8
9
0
0
<5
site!!
3
sites
9
iH
m
OFF
ft
HI
8
18
s::.*
2S
OFF
8
wm
site
mm
0
9
mwrn
9
ft
Ite
9
9
m
OFF
a
8
9
1
8
28
OFF
9
0
mm
o
0
41
OFF
3
s
9
ft
11
$
OFF
9
8
9
8
8
2?
OFF
3
A
9
a
d
48
OFF
3

ft
8
8
7
orF
9
8
i
wm
8
28
¡SI
9
9
9
8
0
4
OfF
8

ft
ft
il
9
OFF
mm

9
8
8
;S2te
OFF
8
8
9
9
A
50
OFF
9
ft
8
9
ft

OFF
ID
mm
18
8
8
38
OFF
9
A
9
8
ft
Its
OFF
3

ft
8
fill
to
OFF
<1
8
9
8
8
11
OFF

9
S'9''
o
ft
m
OFF
8

: ft
8
lift

hO!
a
mm
9
8
8
32
OFF
9
0
9
9
A
83
OFF
Ipil
9
ite
III
I
m
OFF
*
8
9
8
8
S3:
OFF
8
9
ite
A
8
m
OFF
9
ft
sites
iSSSft
I
OFF

8
8
ill

34
OFF
3
HI!
8
ft
ft
SB
OFF:
9
stess
9
sites
8

OFF
$
8
9
8
iii
35
OFF
Site
Site!
9
mm
A
SB;
OFF
sste
8
|¡¡§
9
9
IS
OFF
*
8
Wm
8
8
38
00s
iite
A
SiteS
Siftii
0
5?
OFF
te
ft
Sites:
ft
S
m
OFF
mm
mm
9
mm
8
§1
OfF
HI
III
IS!
Site:
8
86
OFF
site:
ite
ft
ft
I
*?
OFF
mm
mm
9
mm
8
38
OFF
9
9
9
ft
8
SB
OFF
site
ft
ft
8
II
m
OFF
9
0
mm
8
8
38
OFF
9
9
III
0
9
80
OFF
s
8
8
ft
ft
m
orr
9
8
mm
8
8
48
OFF
9
0
Ite
d
d
5t
OFF
9
III
ft
7
isft
m
OFF
9
8
9
8
Site
OFF
9
a
8
ft
9
82
OFF
3
ft
SSteS
8
8
fj
OFF
HI
8
9
8
8

OFF
8
9
9
8
8
Figure 13.4.7. Fertigation valve on or off display.
13,4,4 Apply Irrigation and Fertigation
After the irrigation and fertigation schedules have been defined, the program can
apply both the irrigation and fertigation schedules simultaneously. However, the user must
take care regarding the overlap between the irrigation and fertigation schedules. The steps
to apply the irrigation and fertigation schedules are as follows.
Click Apply Irrigation & Fertigation on the submenu of Scheduling. A dialog screen
appears in Figure 13.4.8.
Click the Cancel button to exit the screen or click the Proceed button to apply the
user defined irrigation and fertigation schedules.
A pop up screen displays on or off status of the valves and pump, phase of an
irrigation or fertigation (preinjection, fertigation, and flush).
Double click left button of the mouse to interrupt the continuous process.

168
Figure 13.4.8. Irrigation and fertigation dialog screen.

169
13,5 Database
Databases have been created in CIMS including (1) farm information, (2) weather
data, (3) irrigation system, (4) crop, (5) crop coefficient, and (6) soil data. These data are
used for simulation to estimate the crop irrigation requirements. The data can also be used
as the user's reference to make irrigation management decisions. Each database consists of
consistent data manipulating buttons.
13,5,1 Buttons to Manipulate Database
Generic control buttons were developed to manipulate the database (Figure 13.5.1).
Functions of each of these buttons are described as follows.
Top
Prior
Next
Bottom
Add
Delete
display the first record of a database,
move to the prior record of a database,
move to the next record of a database,
display the last record of a database,
add a new record to the database,
delete the current record from a database.
The Delete button opens a dialog window to
confer the user's deletion showed in Figure
13.5.2. Click the Cancel button to avoid the
deletion or click the Proceed button to delete
the record.
Bottom
Add
ipgi
Figure 13.5.1.
Database control
button.

170
Find
Figure 13.5.2. Delete dialog screen.
a search function to find a particular record from a database. A
search dialog window (Figure 13.5.3) will display after clicking the
Find button. The search procedure is described as follows.
Click down arrow symbol to select a search field.
Enter the actual value of string one wants to search at the
Find field.
Click the Find button to locate the record.
Click the Find Again button to repeat the search.
Click the Cancel button to exit the screen.
Search for:
Find:
¡ |
In Field:
DATE
Mi
Figure 13.5.3. Search dialog screen.

171
Browse
This routine allows the user to browse data from a database under the
user defined constraints. A browse dialog screen will open as Figure
13.5.4. The steps to browse a database are described as follows.
Figure 13.5.4. Browse dialog window.
Select a data field from the left pull down box in the Add box.
Define the browse condition at the right pull down box in the Add box.
Click the Add button to add the filter criterion to the criteria box.
Repeat steps (1), (2), and (3) to add additional criteria.
Click the Browse button to view the database filtered by the criteria.
Click the Reset button to clear the defined filter criteria.

172
To browse entire database, there is no need to create a criterion as described
in the previous step. Click the Browse button to browse the entire database.
Click the Quit button to exit from the search dialog window.
facts Expert Control Scheduling
| Simulation fools Help Quit
Weather Data
irrigation System
Crop
Crop Coefficient
Soil Data
Figure 13.5.5. Submenu of the Database main menu.
13,5,2 Databases of CIMS
Farm database
Figure 13.5.5
shows the submenu of
Database. To view the
Farm database, click
Farm on the submenu of
Database in Figure
13.5.5. The farm
database contains the
owner's information, mailing address, and phone and fax numbers (Figure 13.5.6).
Farm name
Owner or contact
First name Middle
Address
C<*y IgEftSMSS 1 stole
Phone
BB1
Farm Database
1 Last
mzmzzi
A Z'P 1
Figure 13.5.6. Farm database screen.

173
Weather database
Click Weather on the
submenu of Database to view the
weather database (Figure 13.5.7).
The weather database contains data
from the automated weather
station. The weather data are
stored in daily format.
Evapotranspiration can be estimated
based upon the data.
Irrigation database
Click Irrigation System on
the submenu of Database to view
the data of irrigation system (Figure
13.5.8). Irrigation database
contains microirrigation system
parameters such as flow rate,
emitter number, and wetted
diameter, etc.
Crop database
Click Crop on the submenu
of Database to view the crop
Figure 13.5.7. Weather database screen.
Figure 13.5.8. Irrigation database screen.
Tree Database
Block ID
Crop name
Variety
Planting date
Tree space (ft)
Tree height (ft)
Canopy diameter (ft)
Land coverage (%)
Root depth (ft)
Acreage
Top
Prior
Bottom
Figure 13.5.9. Crop database screen.

174
database (Figure 13.5.9). Crop database contains crop property such as age of the tree, crop
root depth, and crop canopy diameter.
Crop coefficients and MAD
Click Crop Coefficient
on the submenu of Database to
view the crop coefficients and
Citrus Cr)|> Gocrfficienl Mini MnnHcpiinenl Allowed Depletion
Crop name
Variety
Vegetated surface (Y/N)
Crop Coefficient
Management Allowed
Depletion
Management Allowed Depletion
(MAD) (Figure 13.5.8). In
order to calculate actual crop
ET, crop coefficients must be
available. MAD is used for the
water-budget simulation.
Munlh Kc
January
P
February
0-
March
!'
April
ms
May
m
June
m
July
mm
August
. m
September
res
October
November
December
Us
MAD
mm
m
B<5
0.13
8.3;
w
m
as?
iti*
¡MJ
i
lUi
Figure 13.5.10. Crop coefficient database
screen.
Soil database
Click Soil on the
submenu of Database to view the
soil database (Figure 13.5.11).
Soil database listed the common
soil series and soil-water holding
capacity. If the soil exists in the
site, enter YES in the field Exist
in the site (YES/NO).
Figure 13.5.11. Soil database screen.

175
13.6 Simulation
In agriculture, computer simulation is a widely used approach to estimate crop water
requirement. The CIMS uses a soil-water balance model to simulate soil-water content at
the crop root zone. Input data of the simulation come from the database including (1)
irrigation system, (2) crop and crop coefficients, (3) soil, and (4) weather data. The Penman
method was used to calculate daily evapotranspiration. The simulation model executes at
daily intervals. A prognosis of short-term irrigation schedule is given from the simulation
results.
13 61 How to Run the Simulation
Click Simulation on the CIMS main menu (Figure 13.6.1).
Figure 13.6.1. Simulation submenu.
Click Set Initial SWC to open the set initial condition dialog window (Figure
13.6.2). It is necessary to set this initial condition at first run of the simulation.
After the first run, the simulation model can automatically search its initial condition
based on the previously simulated results.

176
Click Water Budget on
the submenu of
Simulation to open the
simulation dialog window
(Figure 13.6.3). Enter
starting and ending date of
Figure 13.6.2. Screen to set initial condition of the
simulation.
the simulation from the
screen.
Click the Cancel button
to exit the screen or
click the Execute
button to run the
simulation. The
simulation model
obtains its required
data from the databases. The simulation results are stored in three files: (1) simulated
soil-water content from the simulation period, (2) the most recent simulated soil-
water content, and (3) one week prognosis of irrigation requirement, which assumes
no rainfall to occur in that week and same ET rate as the previous week.
Figure 13.6.3. Simulation dialog window.

177
13.6.2 View the Results of Simulation
Click Browse Result on the submenu of Simulation.
Select one of the simulation results (Prognosis, Simulated SWC, and
Historical SWC). The prognosis results (Figure 13.6.4) contains the
following data:
1
[jT>-
i
Prognosis Result
Hi
I1
I1
Our.br |
I
t
1
i
i0.052
17 ;
0
0 j
0
g
2
;0.046
30 |
0
0 i
0
i:
I 3
!0.040
43 ;
0
o j
0
0
4
i0.035
57 j
0
o!
0
5
|0.060j
70 i
72
7 |
9
1
6
i 0.054i
13 j
0
0 |
0
ly
7
j0.048
27 j
0
0 1
0
T
m
T
m
11
Figure 13.6.4. Irrigation prognosis from the simulation.
Day
number of days of the prognosis,
Swc
soil-water content by volume,
Depletion
soil-water depletion at the crop root zone in
percent,
Vol_tree
volume of water need to be applied per tree (gallon),
Durhour
irrigation duration hour, and
Durmin
irrigation duration minute. The overall irrigation
duration is equal to the irrigation hour plus the
irrigation minute.

178
The prognosis result can be used as a reference for the users to define their own
irrigation schedules when irrigation is controlled by the user defined schedules.
To view the simulated soil-water content, click the Historical SWC on the submenu
of Simulation. A display appears in Figure 13.6.5. Then execute the following steps.
Click the upper left box to exit the screen.
Click Simulated SWC from the submenu of Simulation. A browse screen
will pop up as Figure 13.6.5.
Figure 13.6.5. Simulated soil-water content.
The simulated results include the following results:
Sdate date of the simulation,
Swc soil water content by volume, and
Depletion soil water depletion at the crop root zone in percent.
Click the upper left box to exit the screen or click the upper right arrow box
to maximize the window.

179
13.7 Tools of CIMS
An automated irrigation control system requires extensive hardware and can be a
major concern for the users. CIMS provides several tools in aid of the decisions on citrus
irrigation management without hardware requirements. The tools are (1) to retrieve weather
data from a text file, (2) to calculate crop evapotranspiration (ET), (3) to estimate duration
of an irrigation, and (4) to display the irrigation pipe layout. Each of these functions is
described as follows.
13 71 Aggregate Weather Data
To aggregate the weather data to daily time interval, execute the following steps.
Click Tools on the main menu of CIMS. The submenu of Tools shows in
Figure 13.7.1.
Eacts Expert Control Scheduling database Simulation
Iools
Help Quit
Aggre
gate Weather Data
ET Calculation
Estimate Irr. Duratiion
Field Layout Map
Figure 13.7.1. Submenu of Tools.
Click Load Weather Data. The program will convert the weather data from
a text file to a database file and aggregate the data into daily data (Figure
13.7.2). The weather data are used for the simulation of crop water
requirements.

180
(Inactive FoxPro Run Command)
ftread: 2271
Line written: 92
Weather Watch Processed Daily Outputs
Month
Day
Vear
Max
fiir
IcnpF
Min
Dir
TenpF
flug
Sir
Mf/n2
Us
fliph
Rain
in
ET
6 29
199-3
104.00
62.51
0.35
3.4
0.000
0.309
6 3a
199-3
102.30
64.16
0.35
3.5
0.000
0.310
7 l
1994
104.20
69.13
0.34
3.4
0.000
0.318
7 2
1994
104.30
74.30
0.34
4.8
0.000
0.332
? 3
1994
101.90
63.18
0.35
3.1
0.000
0.269
7 -3
1994
90.00
65.94
0.33
3.4
0.000
0.258
7 S
1994
96.30
67.13
0.28
3.8
0.010
0.205
7 6
1994
93.10
67.70
8.33
3.8
0.038
0.218
7 7
1994
97.10
72.40
0.34
3.8
0.800
0.243
7 8
1994
97.20
94.30
0.7?
5.8
fl .
0.857
7 9
1994
90.70
64.39
0.30
2.0
0.000
0.145
7 10
1994
92.90
65.18
0.30
3.4
0.003
0.211
7 11
1994
90.50
83.40
0.96
4.4
0.000
0.106
sailing files completed.
Figure 13.7.2. Read weather data from weather station.
13,7,2 Calculate Reference ET
To calculate reference ET, execute the following steps.
Click ET Calculation on the submenu of Tools. An ET dialog window
appears as Figure 13.7.3.
Select an ET method
Select one of the ET methods
[PenmanJ
O Blaney-Criddle
O Modified Blaney-Criddle
O Stephens-Stewart
Figure 13.7.3. ET method dialog window.
Select one of the ET methods by clicking the radio button.

181
Penman method
Select the Penman radio
button and click the Continue
button. A screen to calculate
ET shows in Figure 13.7.4.
Enter the data on the
screen and click the Calculate
button to calculate ET. Click
the Exit button to quit from the
screen.
Penman Reference ET Calculation
Date
Solar radiation (kW/m*)
Maximum air temperature (T)
Minmum air temperature (*F)
Wind speed (mph)
Albedo
08/22/94
Figure 13.7.4. Penman ET screen.
Blanev-Criddle method
Select the Blaney-Criddle radio button and click the Continue button. A
calculation screen appears as
Figure 13.7.5.
Enter the required data to
the screen and click the Calculate
button to display the result. Click
the Exit button to quit from the
screen.
Blaney-Criddle Method
Coefficient for the Blaney-Criddle method (M)
Percent of annual daylight hours in the month
Mean monthly temperature (T)
mmm
Figure 13.7.5. Blaney-Criddle ET screen.

Modified Blanev-Criddle method
Select the Modified
Blaney-Criddle radio
button and click the
Continue button. A
calculation screen appears
as Figure 13.7.6.
Enter the required
data to the screen and click
the Calculate button to display the result. Click the Exit button to quit from the
screen.
Stephens Stewart method
Select the Stephens
Stewart radio button and click
the Continue button. A
calculation screen appears as
Figure 13.7.7.
Enter the required data
to the screen and click the
Calculate button to display the result. Click the Exit button to quit from the screen.
IS
Stephens-Stewart Method
Monthly mean temperature (*F)
Monthly solar radiation (kW/m*)
M
1
Monthly reference ET {tt^menth)
S
Calculate
¡SB
Figure 13.7.7. Stephens-Stewart ET screen.
Figure 13.7.6. Modified Blaney-Criddle ET screen.

183
13.7.3 Calculate Irrigation Duration
Irrigation duration can
be estimated when one has the
available data: (1) current soil
water depletion, (2) soil
characteristics, (3) irrigation
system, and (4) the crop root
zone. To calculate irrigation
Estimated irrigation time to bring root zone to field capacity
Microirrigation system data
Emitter flow rate (gallons/hour)
Emitter number / tree
Diameter of spray pattern (feet)
Overall irrigation efficiency (%)
Crop and soil data
Crop effective root depth (feet)
Available soil water content (in/ft)
Depletion of available water
prior to irrigation (%)
Estimated irrigation duration
m
i
u
m
houf 3? mu
Figure 13.7.8. Estimate irrigation duration screen.
duration, execute the following
steps.
Click Estimation Irr.
Duration on the submenu of
Tools. A screen as Figure
13.7.8 appears for the
calculation.
Enter the required data to the
screen and click the Calculate
Help of irrigation duration utility
Enter data of irrigation system, soil and crop root depth
Click Co/cu/ata to calculate irrigation duration
Click Caoce/ to exit from the screen
Average available waler holding capacity
for various soil type
Soil type
AW holding capacity (intft)
Coarse sand
072
Fine sane
0.96
Loamy sand
1.20
Sandy loam
1.38
Fine snady loam
1.56
Very fine sandy loam
1.68
ay and day loom
1.80
Silty day and silty day loam
1.92
Silt loam
222
Peats and mucks
2.52
Martin et al.. 1990 Irrigation scheduling principles, in
Management of Form irrigation System. ASAE
Figure 13.7.9. Help screen of irrigation
duration.
button to display the result.
Click the Help button to view soil data (Figure 13.7.9).
Click the Exit button to quit from the screen.

184
13 7 4 Map of Irrigation System Layout
An irrigation system layout map can be created in the system. The soil moisture
sensors and weather station locations can also be displayed on the map.
To view the map, click the
Field Layout Map on the
submenu of Tools. Figure
13.7.10 shows a dummy
field layout map.
Click the Close button to Figure. 13.7.10. A dummy field layout
Pipe layout map
II
exit the screen.
map.

185
13 8 Help Utilities of CIMS
CIMS provides several help utilities in the system. This includes (1) hypertext
content help, (2) calculator, (3) calendar/diary, (4) clock, and (5) text editor (Figure 13.8.1).
Each of these utilities is described as follows.
13.8.1 Help
Help contains a hypertext content help screen (Figure 13.8.2). The contents are
categorized as the main menu. To inquire help message on each subject, click the underlined
Eacts Expert Control Scheduling database Simulation Tools mhim Cult
Calculator
Calendar/Diary
Clock
Editor
Puzzle
About
Figure 13.8 .1. Submenu of the Help.
content title. Click the upper left box to
exit from the window.
13.8.2 Calculator
A simple calculator was included.
Click the Calculator on the submenu of
Help to activate the calculator (Figure
13.8.3). Click the upper left button to exit
from the calculator.
Figure 13.8.2. CIMS help screen.

186
Figure 13.8.3. Calculator
Figure 13.8.4. Calendar and diary screen.
13,8,3 Irrigation Calendar and Diary
Figure 13.8.4 shows the calendar and diary. The left side of the window displays a
calendar and the right side of the window shows the diary box. This calendar and diary can
be used as a tool for calendar based irrigation management.
Click the < Month button to display previous month.
Click the Month > button to display next month.
Click the < Year button to display previous year.
Click the Year > button to display next year.
Click the Today button to display highlight today's date.
Click any day in the month to view that day's diary or message.
13.8 4 Clock
Click the Clock on the submenu of Help to display the clock in Figure 13.8.5.

187
13.8.5 Text Editor
Click the Editor on the submenu of Help to open
a text editor (Figure 13.8.6).
13.8.6 Puzzle
Click the Puzzle on the submenu of Help to
open a puzzle (Figure 13.8.7).
Figure 13.8.5. Clock.
Figure 13.8.6. Text editor.
13,8.7 About
Figure 13.8.7.
Puzzle.
Click About on the
submenu of Help to view the about
screen (Figure 13.8.8).
Click the Exit button to
quit from the screen or click the
Next button to view next screen
CIMS For Windows
Version 1.0.1994
Jionnong Xin
AgricuraJ Engineering Deportment
University ot Florida
Gainesville, FL 32611
USA
(Figure 13.8.9).
Figure 13.8.8. About the CIMS.

188
13.8.8 Quit from CIMS
To quit from the CIMS, click
the Quit button on the main menu. A
confirmation screen appears as Figure
13.8.10. Click the OK button to exit
from CIMS or click the Cancel button
Figure 13.8.9. More about the CIMS.
to stay in CIMS.
lit i*nowtoH*9imut>t te my study amuntttee, ;
O r. s mota (CfMwmttM
Dr A a Si*ttte
Or. J W tore
O P H Jymfi
Dr O, D. Oankrtl
Of.Lrt.Mote
Special thettfcs to my rfomaJe experts:
Dr T. A Wheolon cert Of. L ft. Pareeas
Cines RCC at lake ASK* lor the vofcmfcte
advices.
Exit
Figure 13.8.10.
Screen to quit from
the CIMS.

LIST OF REFERENCES
Adams, J. T. 1992. Microirrigation management for mature flatwoods citrus. The Citrus
Industry, July 7:44-45.
Anderson, C. E., H. P. Johnson, and W. L. Powers. 1978. A water-balance model for
deep loess soils. Trans, of the ASAE 19:314-320.
Augustin, B. J. and G. H. Snyder. 1984. Moisture sensor-controlled irrigation for
maintaining bermudagrass turf. Agron. J. 76:848-850.
Barfield, B. J., K. B. Perry, J. D. Martsolf, and C. T. Morrow. 1990. Modifying the
aerial environment. In Management of Farm Irrigation Systems, St. Joesph, MI,
p. 827-869.
Barrett, M. L. and A. C. Beerel. 1988. Expert System in Business: A Practical Approach,
Ellis Horwood, Chichester.
Batchelor, W. D. and R. W. McClendon. 1992. A blackboard approach for resolving
conflicting irrigation and insecticide scheduling recommendations. Trans, of ASAE
35(2):741-747.
Bobrow, D. G. and T. Winograd. 1977. An overview of KRL, a knowledge
representation language. Cognitive Science 1 (l):3-46.
Boman, B. J. 1993. A comparison of controlled-released to conventional fertilizer on
mature 'Marsh' grapefruit. Proc. Fla. State Hort. Soc. 106:1-4.
Bonissone, P. P. and R. M. Tong. 1985. Reasoning with uncertainty in expert systems.
International Journal of Man-Machine Studies 22(3):241-250.
Brachman, R. J. and H. J. Levesque (Editors). 1985. Readings in Knowledge
Representation. Morgan Kaufmann, Los Altos, CA.
Bronowski, J. 1965. The identity of man. Penguin Books, Bergenfield, ND.
Buchanan, B. G. and E. H. Shortliffe. 1984. Rule-Based Expert Systems. Addison-Wesley
Publishing Company, Reading, MA.
189

190
Burns, I. L., E. R. Muller, E. Bell, and S. Novak. 1990. A weighing lysimeter for
evapotranspiration monitoring and irrigation control. Proc. of the 3rd Int.
Conference on Computer in Agr. Extension Programs, Orlando, FL, pp. 73-78.
Cahoon, J., J. Ferguson, D. Edwards, and P. Tacker. 1990. A microcomputer-based
irrigation scheduler for the humid mid-south region. Applied Eng. in Agriculture
6(3): 289-294.
Campbell, G. S. and M. D. Campbell. 1982. irrigation scheduling using soil moisture
measurements: Theory and practice. In Advances in Irrigation, ed. D. Hillel.
New York: Academic Press, 1:25-54.
Carlisle, V. W., R. E. Caldwell, F. Sodek, III, L. C. Hammond, F. G. Calhoun, M. A.
Granger, and H. L. Breland. 1978. Characterization Data for Selected Florida
Soils. IFAS, Univ. of Florida, Gainesville, FL.
Carr-Brion, K. 1986. Moisture Sensors in Process Control, Elsevier Applied Sci.
Publishers, London.
Cary, J. W. and H. D. Fisher. 1983. Irrigation decisions simplified with electronics and
soil water sensors. Soil Sci. Soc. Am. J. 47:1219-1223.
Cassell, D. K. and A. Klute. 1986. Water potential: tensiometry, in Methods of Soil
Analysis, Part 1, Physical and Mineralogical Methods, ed. A. Klute, 2nd. edition.
Soil Society of American, Madison, Wisconsin.
Chesness, J. L., D. L. Cochran, and J. E. Hook. 1986. Predicting seasonal irrigation
water requirements on coarse-textured soils. Trans, of the ASAE 29(4): 1054-
1057.
Collins, A. G., S. J. Nix, T. Tsay, A. Gera, and M. A. Hopkins. 1990. The potential for
expert systems in water utility operation and management. Journal AWWA 9:44-
51.
Creighton, Jack, David A. Sleeper, and Calvin Hubbard. 1989. Tensiometers for
irrigation scheduling in Florida citrus grove. Proc. Fla. State Hort. Soc. 102:69-
72
Davies, F. S., T. E. Marler, and L. R. Parsons. 1989. Irrigation of young citrus trees.
Citrus Industry, April, pp. 5-9.

191
Dempster, A. P. 1967. Upper and lower probabilities induced by a multivalued mapping.
Annals Math. Statistics 28(2):325-339.
Duke, H. R., L. E. Stetson, and N. C. Ciancaglini. 1990. Irrigation system control. In
Management of Farm Irrigation Systems, ed. G. J. Hoffman, T. A. Howell and
K. H. Solomon. ASAE Monograph, St. Joseph, MI, pp. 265-312.
Duke, H. R., D. F. Heermann, and M. C. Blue. 1984. Computer control of irrigation for
electrical load management. Trans, of the ASAE 27(2):597-602,608.
Durkin, J. 1994. Expert systems: an overview of the field. PC AI, January/February, pp
37-39.
Ege, R. K. and C. Stary. 1992. Designing maintainable, reusable interfaces. IEEE
Software 11:24-32.
Erbach, D. C. 1983. Measurement of soil moisture and bulk density, ASAE Paper No.
83-1553, St. Joseph, MI.
Feigenbaum, E. A. 1979. Themes and case studies of knowledge engineering. In Expert
System in the Micro-Electronic Age, ed. D. Michie. Edinburgh Univ. Press,
Edinburgh, Scotland, pp. 3-25.
Feigenbaum, E. A. 1982. Knowledge Engineering in the 1980's. Dept, of Computer
Science, Stanford University, Stanford, CA.
Feigenbaum, E. A., P. E. Friedland, B. B. Johnson, H. P. Nil, H. Schorr, H. Shrobe,
and R. S. Engelmore. 1994. Knowledge-based systems research and applications
in Japan, 1992. AI Magazine, Summer, pp. 29-43.
Feller, W. 1969. An Introduction to Probability Theory and Its Applications. Vol. 1. 3rd
ed. New York: John Wiley & Sons.
Ferguson, J., C. Taylor, G. Israel, and B. Summerhill. 1989. Citrus production survey:
young tree care. Citrus Industry, September, pp. 38-41.
Fernald, E. A. and D. J. Patton. 1984. Water Resources Atlas of Florida, Florida State
University, Tallahassee, FL.
Fikes, R. and T. Kehler. 1985. The role of frame-based representation in reasoning.
Communications of the ACM 28(9): 904-920.

192
Fisher, J. 1990. Fertigation: microsprinklers provide new options in fertilization
programs. Citrus Industry, January, pp. 12-18.
Fitzsimmons, D. W. and N. C. Young. 1972. Tensiometer-pressure transduce system for
studying unsteady flow through soils. Trans, of the ASAE 15:272-275.
Florida Agricultural Statistics. 1994. Commercial Citrus Inventory, Orlando, FL.
Fox, M. S. 1990. AI and expert system myths, legends and facts. IEEE Expert, February,
pp. 8-20.
Gabriel, K. R. and J. Newmann. 1962. A Markov chain model for daily rainfall
occurrence at Tel Aviv. Jour, of the Royal Meteorological Society 88(375): 90-95.
Gardner, W. H. 1986. Water content. In Methods of Soil Analysis. Part 1. Physical and
Mineralogical Methods, ed. A. Klute. Agronomy Series No. 9. Am. Soc.
Agronomy, 2nd ed., pp. 493-544.
Gerber, J. F., A. H. Krezdorn, J. F. Bartholic, J. R. Conner, H. J. Reitz, R. C. J. Koo,
D.S. Harrison, E. T. Smerdon, J. G. Georg, L. H. Harris, and J. T. Bradley.
1973. Water needs for Florida Citrus. Proc. Fla. State Hort. Soc. 86:61-64.
Giarratano, J. and G. Riley. 1989. Expert Systems Principles and Programming. PWS-
Kent Publishing Company, Boston.
Gonzalez, A. and D. Dankel II. 1993. Engineering Knowledge-based Systems: Theory
and Practice. Prentice Hall, Inc., Englewood Clitts, NJ.
Harandi, M. T. and R. Lange. 1990. Model-based knowledge acquisition. In Knowledge
Engineering, ed. H. Adeli. McGraw-Hill Publishing Company, New York, vol.
I pp. 103-129.
Hardy, N. G. 1989. Using irrigation for frost protection. Citrus Industry, October, pp.
39-41.
Harrison, D. S., J. F. Gerber, and R. E. Choate. 1987. Sprinkler irrigation for cold
protection, Florida Cooperative Extension Service, Circular 348. IFAS, Univ. of
Florida, Gainesville, FL.
Harrison, P. R. and P. A. Harrison. 1994. Validating an embedded intelligent sensor
control system. IEEE Expert, June, pp. 49-53.

193
Hayes-Roth, F. 1985. Rule-based systems. Communications of the ACM 28(9):921-
932.
Hearn, C. J. 1993. The influence of cultivar and high nitrogen and potassium fertilization
on fruit quality traits of young orange trees. Proc. Fla. State Hort. Soc. 106:8-12.
Heinemann, P. H., C. T. Morrow, R. M. Crassweller, and J. D. Martsolf. 1991. A
decision support system for the protection of crop from frost. Hortscience
26(6):732.
Heinemann, P. H., C. T. Morrow, J. D. Martsolf, R. M. Crassweller, and. K. B. Perry.
1994. Decision support system for the protection of crops from frost. Proc. of the
5th Int. Con. on Computers in Agriculture, Orlando, FL, pp.375-380.
Hodgman, C. (ed) 1949. Handbook of Chemistry and Physics. Chemical Rubber,
Cleveland, Ohio.
Holland, M. 1990. Elevation of microsprinklers for cold protection. Citrus Industry
Magazine 71(8): 12-15.
Howell, T. A., D. W. meek, C. J. Phene, K. R. Davis, and R. L. McCormick. 1984.
Automated weather data collection for research on irrigation scheduling. Trans, of
ASAE 47:386-391, 396.
Ingrand, F. F., M. P. Georgeff, and A. S. Rao. 1992. An architecture for real-time
reasoning and systems control. IEEE Expert, December, pp. 34-44.
Jackson, L. K. 1991. Citrus Growing in Florida, 3rd. ed. Univ. Press of Florida,
Gainesville, FL.
Jacobson, B. K., J. W. Jones, and P. H. Jones. 1987. Tomato greenhouse environment
controller: real-time expert system supervisor. ASAE Paper No 87-5022, St.
Joseph, MI.
Jacobson, B. K., P. H. Jones, J. W. Jones, and J. A. Paramore. 1989. Real-time
greenhouse monitoring and control with an expert system. Computers and
Electronics in Agriculture 34:273-285.
Jensen, M. E., C. N. Robb, and C. E. Franzoy. 1970. Scheduling irrigation using
climate-crop-soil data. J. of Irrigation and Drainage Division, ASCE 3:26-35.
Jensen, M. E., J. L. Wright, and B. J. Pratt. 1971. Estimating soil moisture depletion
from climate, crop and soil data. Trans, of the ASAE 14(5):954-959.

194
Jones, J. W. 1985. Using expert systems in agricultural models. Agricultural Engineering
66(7) :21-23.
Jones, J. W. 1993. Decision support systems for agricultural development. In Systems
Approaches for Agricultural Development, ed. F. W. T. Penning de Vries et al.,
Kluwer Academic Publishers, Netherlands, pp. 459-471.
Jones, J. W., R. F. Colwick, and E. D. Threadgill. 1972. A simulated environmental
model of temperature, evaporation, rainfall and soil moisture. Trans, of the ASAE
15(2):366-372.
Jones, J. W. and J. T. Ritchie. 1990. Crop growth models. In Management of Farm
Irrigation Systems, ed. G. J. Hoffman, T. A. Howell and K. H. Solomon. ASAE
Monograph, St. Joseph, MI, pp. 63-85.
Jones, P. 1989a. Agricultural applications of expert systems concepts. Agricultural
Systems 31:3-18.
Jones, P. 1989b. Knowledge acquisition. Knowledge Engineering in Agriculture. ASAE,
St. Joesph, MI, pp.33-46.
Jones, P. and J. Haldeman. 1986. Management of a crop research facility with a
microcomputer-based expert system, Trans, of ASAE 29:235-242.
Jones, P. G. and P. K. Thornton. 1993. A rainfall generator for agricultural applications
in the tropics. Agricultural and Forest Meteorology 63:1-19.
Kalkar, S. and P. R. Goodrich. 1986. An expert system for animal waste management.
ASAE Paper No. 86-4540, St. Joseph, MI.
Keller, J. and R. D. Bliesner. 1990. Sprinkle and Trickle Irrigation. Van Nostrand
Reinhold, New York.
Khanjani, M. J. and J. R. Busch. 1982. Optimal irrigation water use from probability and
cost-benefit analyses. Trans, of the ASAE 42:961-968.
Kline, D. E., D. B. Bender, and B. A. McCarl. 1987. Farm-level machinery management
using intelligent decision support systems. ASAE Paper No. 87-1046, St. Joseph,
MI.
Koo, R.C.J. 1963. Effects of frequency of irrigations on yield of oranges and grapefruit.
Proc. Fla. State Hort. Soc. 76:1-5.

195
Koo, R. C. J. and A. G. Smajstrla. 1984. Effects of trickle irrigation and fertigation on
fruit production and juice quality of 'Valencia' orange. Proc. Fla. State Hort. Soc.
97:8-10.
Koo, R. C. J., C. A. Anderson, I. Stewart, D. P. H. Tucker, D. V. Calvert, and H. K.
Wutscher. 1984. Recommended fertilizers and nutritional sprays for citrus. Bui.
536D, IFAS, Univ. of Florida, Gainesville, FL, 30p.
Kumar, D., C. D. Heatwole, B. B. Ross, T. A. Dillaha. 1992. A knowledge-based
system for preliminary selection and economic evaluation of sprinkler irrigation
systems. Applied Eng. in Agriculture 8(4):441 -447.
Laffey, T. J., P. A. Cox, J. L. Schmidt, S. M. Kao, and J. Y. Read. 1988. Real-time
knowledge based systems. AI Magazine, Spring, pp. 27-45.
Lembke, W. D. and B. A. Jones, Jr. 1972. Selecting a method for irrigation scheduling
using a simulation model. Trans.of the ASAE 15(2):284-286.
Lemmon, H. 1986. COMAX: An expert system for cotton crop management. Science
233(7):29-33.
Martin, D. L., E. C. Stegman, and E. Fereres. 1990. Irrigation scheduling principles.
In Management of Farm Irrigation Systems, ed. G. J. Hoffman, T. A. Howell, and
K. H. Solomon, ASAE Monograph, St. Joseph, MI, pp. 155-203.
Martsolf, J. D. 1990. Cold protection strategies. Proc. Fla. State Hort. Soc. 103:72-78.
Martsolf, J. D., P. H. Heinemann, and C. T. Morrow. 1991. Protection from frost or
freeze with sprinklers. HortScience 26(6):729.
Martsolf, J. D. 1992. Cold protection mechanism. Proc. Fla. State Hort. Soc. 105:91-94.
McClendon, R. W. W. D. Batchelor, and J. E. Hook. 1989. An expert simulation
system for irrigation management. ASAE Paper No. 89-2460, St. Joseph, MI.
Mckim, H. L., J. E. Walsh, and D. N. Arion. 1980. Review of techniques for measuring
soil moisture in situ. United States Army Corps of Engineers, Special Report 80-
31, Cold Regions Research and Engineering Lab, Hanover, NH.
Merlo, G. 1992. Possible applications of experts systems in the management of water
distribution systems. Water Supply 10:87-95.

196
Michalski, R. S., J. G. Carbonell, and T. M. Mitchell (eds.). 1986. Machine Learning,
Vol.2, Morgan Kaufmann, Los Altos, CA.
Minsky, Marvin. 1975. A framework for representing knowledge. In The Psychology of
Computer Vision, ed. P. Winston. McGraw-Hill, New York, pp. 211-217.
Moore, D. S. and G. P. McCabe. 1989. Introduction to the Practice of Statistics. W. H.
Freeman and Company, New York.
Morey, R. V., S. A. Sargent, C. D. Baird, and M. T. Talbot. 1988. A knowledge based
system for selecting precooling methods. ASAE Paper No. 88-7539, St. Joseph,
ML
Muttiah, R. S., C. N. Thai, S. E. Prussia, R. L. Shewfelt, and J. L. Jordan. 1988. An
expert system for lettuce handling at a retail store. Trans, of ASAE 31(2):622-628.
Myers, J. M. and D. S. Harrison, 1978. Drip irrigation of orange trees in humid climate.
ASAE Paper No. 78-2018, St. Joseph, MI.
Nann, S. R., A. Ray, and S. Kumara. 1991. A decision support system for real-time
monitoring and control of dynamical processes. Inter. J. of Intelligent Systems
6:739-758.
NOAA. 1952-1992. Climatological Data, Florida. U.S. Environmental Data Service,
Asheville, NC.
O'Keefe, R. M., O. Balci, and E. P. Smith. 1987. Validating expert system performance.
IEEE Expert, Winter, pp. 81-88.
Padalkar, S., G. Karsai, C. Biegl, J. Sztipanovits, K. Okuda, and N. Miyasaka. 1991.
Real-time fault diagnostics. IEEE Expert, June, pp.75-85.
Palmer, R. G. 1986. How expert systems can improve crop production. Agricultural
Engineering, September/October, pp. 28-30.
Parsons, L. R., C. Cliff, B. Summerhill, and G. Israel. 1989. Cold protection survey
major changes in a decade. Citrus Industry, November, pp. 46-48.
Parsons, L. R. and T. A. Wheaton. 1990. Irrigation for cold protection during the
December, 1989 freeze. Citrus Industry, November, pp. 12-14.

197
Parsons, L. R., K. T. Morgan, and T. A. Wheaton. 1993. Effects of microsprinkler
precipitation rate, soil type, and water depletion on depth of soil wetting. Proc.
Fla. State Hort. Soc. 106:38-41.
Peart, R. M., F. S. Zazueta, P. Jones, J. W. Jones, and J. W. Mishoe. 1986. Expert
systems take on three tough agricultural tasks. Agricultural Engineering,
May/June, pp. 8-10.
Phene, C. J. 1986. Operation principles: automation, In Trickle Irrigation for Crop
Production: Design, Operation and Management, ed. F. S. Nakayama and D. A.
Bucks. Elsevier, Tokyo, pp. 188-215.
Phene, C. J. 1989. Techniques for computerized irrigation management. Computers and
Electronics in Agriculture 3:189-208.
Phene, C. J., C. P. Allee, and D. J. Pierro. 1989b. Soil matric potential sensor
measurements in real-time irrigation scheduling. Agrie. Water Management
16:173-185.
Phene, C. J., W. R. DeTar, and D. A. Clark. 1992. Real-time irrigation scheduling of
cotton with an automated pan evaporation system. Applied Eng. in Agriculture
8(6):787-793.
Phene, C. J., G. J. Hoffman, and R. S. Austin. 1973. Controlling automated irrigation
with soil matric potential sensor. Trans, of ASAE, Paper No. 71-230, St. Joseph,
MI.
Phene, C. J. and T. A. Howell. 1984. Soil sensor control of high-frequency irrigation
systems. Trans, of ASAE 27:392-396.
Phene, C. J., R. L. McCormick, K, D. Davis, J. D. Pierro, D. W. Meek. 1989a. A
lysimeter feedback irrigation controller system for evapotranspiration
measurements and real time irrigation scheduling, Trans, of ASAE 32:477-484.
Plant, R. E. 1989. An integrated expert decision support system for agricultural
management. Agricultural System 29:49-66.
Pleban, S., J. W. Labadie, and D. F. Heermann. 1983. Optimal short term irrigation
schedules. Trans, of ASAE 26:141-147.
Pleban, S., D. E. Heermann, J. W. Labadie, and H. R. Duke. 1984. Real time irrigation
scheduling via "reaching" dynamic programming, Water Resources Research
20:887-895.

198
Prerau, D. S., A. S. Gunderson, R. E. Reinke, and M. R. Adler. 1990. Maintainability
techniques in developing large expert systems. IEEE Expert, June, pp.71-79.
Protopapas, A. L., and A. P. Georgakakos. 1990. An optimal control method for real
time irrigation scheduling, Water Resources Research 26:647-669.
Quillian, M. R. 1968. Semantic memory. In Semantic Information Process, ed. by M.
Minsky. The MIT Press, Cambridge, pp. 227-270.
Quinlan, J. R. 1983. Learning efficient classification procedures and their application to
chess endgames. In Machine Learning: An Artificial Intelligence Approach, ed.
R. S. Michalski, J. G. Carbonell, and T. M. Mitchell. Tioga Publishing, Palo
Alto, CA, pp. 463-482.
Reitz, H. J., L. G. Albrigo, L. H. Allen, Jr., J. F. Bartholic, D. V.Calvert, H. W. Ford,
J. F. Gerber, L. C. Hammond, D. S. Harrison, R. C. J. Koo, R. S. Mansell, J.
M. Myers, J. S. Rogers, and D. P. H. Tucker. Water requirements for citrus.
WRC-4. Water Resources Council, IFAS, Univ. of Florida, Gainesville, FL.
Richardson, C. W. 1981. Stochastic simulation of daily precipitation, temperature, and
solar radiation. Water Resources Research 17(1)182-190.
Richardson, C. W. 1985. Weather simulation for crop management models. Trans, of the
ASAE 28(5): 1602-1606.
Richardson, C. W. and Wright D. A. 1984. WGEN: A model for generating daily
weather variables. ARS-8, U. S. Department of Agricultural Res. Service,
Temple, TX, 80p.
Ringland, G. A. and D. A. Duce. 1988. Approaches to Knowledge Representation: An
Introduction. John Wiley & Sons Inc., New York.
Robinson, M. R. 1990. Report on fertilizers and irrigation. Proc. Fla. State Hort. Soc.
13:140-145.
Rogers, J. S., L. H. Allen Jr., and D. V. Calvert. 1983. Evapotranspiration from a
humid-region developing citrus grove with grass cover. Trans, of ASAE
26:1778-1783, 1792.
Rogers, D. H. and R. L. Elliott. 1989. Irrigation scheduling using crop growth
simulation, risk analysis, and weather forecasts. Trans, of ASAE 32:1669-1667.

199
Rolston, D. E., R. J. Miller, and H. Schulbach. 1986. Management principles of
fertilization, in Trickle Irrigation for Crop Production: Design, Operation and
Management, ed. F. S. Nakayama and D. A. Bucks. Elsevier, Tokyo, pp. 317-
344.
Safley, J. M, Jr, L. A. Fribour, J. V. Vaiksoras, and R. H. Strand. 1974. Probability of
sequences of wet and dry days for Tennessee. NOAA Technical Memorandum,
NOAA/Environmental Data Service, Washington, DC.
Schalkoff, R. J. 1990. Artificial Intelligence: An Engineering Approach. McGraw-Hill
Publishing Company, New York.
Schmidt, G. M. 1992. Stochastic rainfall modeling and long-term climatic variability:
model parameter estimation and model evaluation. Doctoral dissertation, Agr. Eng.
Dept., Univ. of Florida, Gainesville, FL.
Schmidt, G. M., A. G. Smajstrla, and F. S. Zazueta. 1987. Estimating the frequency and
duration of wet and dry periods in a humid subtropical region. Soil Crop Sci. Soc.
Fla. 47:83-88.
Schmugge, T. J., T. J. Jackson, and H. L. McKim. 1980. Survey of methods for soil
moisture determination. Water Resources Research 16:961-979.
Schnelle, K. D. and R. S. H. Mah. 1992. A real-time expert system for quality control.
IEEE Expert, October, pp.36-42.
SCS. 1982. Florida Irrigation Guide. USDA, Soil Cons. Serv., Gainesville, FL.
Shafer, G. 1976. A Mathematical Theory of Evidence, Princeton Univ. Press, Princeton,
NJ.
Sharpies, R. A., D. E. Rolston, J. W. Biggar, and H. I. Nightingale. 1985. Evaporation
and soil water balances of young trickle-irrigated almond trees. In Drip/trickle
Irrigation in Action, Proc. of Third Int. Drip/Trickle Irrigation Congress, ASAE,
St. Joseph, MI, vol. II, pp. 792-797.
Shayya, W. H., V. F. Bralts, and T. R. Olmsted. 1990. General irrigation scheduling
package for microcomputers. Computers and Electronics in Agriculture, 5:197-
212.
Shortliffe, E. H. and B. G. Buchanan. 1975. A model of inexact reasoning in medicine.
Mathematical Biosciences 23:351-379.

200
Simpson, J. R. and J. J. Meyer. 1987. Water content measurements comparing a TDR
array to neutron scattering. International Conference on Measurement of Soil and
Plant Water Status, Centennial of Utah State Univ., Logan, pp. 111-114.
Smajstrla, A. G. 1990. Evapotranspiration based citrus irrigation scheduling. Special
series Report SS-AGE-18, IFAS, Univ. of Florida, Gainesville, FL.
Smajstrla, A. G., D. S. Harrison, and F. X. Duran. 1985b. Tensiometers for soil
moisture measurement and irrigation scheduling. Cooperative Extension Service,
Circular 487. IFAS, Univ. of Florida, Gainesville, FL.
Smajstrla, A. G., W. G. Boggess, B. J. Boman, G. A. Clark, D. Z. Haman, G. W.
Knox, S. J. Locascio, T. A. Obreza, L. R. Parsons, F. M. Rhoads, T. Yeager,
and F. S. Zazueta. 1995. Status and growth of microirrigation in Florida, Proc.
of the Fifth International Microirrigation Congress, Orlando, FL, p. 325-330.
Smajstrla, A. G., D. S. Harrison, F. S. Zazueta, L. R. Parsons, and K. C. Stone. 1987.
Trickle irrigation scheduling for Florida citrus. Florida Cooperative Extension
Service, Bulletin 208. IFAS, Univ. of Florida, Gainesville, FL.
Smajstrla, A. G. and R. C. J. Koo. 1984. Effects of trickle irrigation methods and
amounts of water applied on citrus yields. Proc. Fla. State Hort. Soc. 97:3-7.
Smajstrla, A. G. and R. C. J. Koo. 1986. Use of tensiometers for scheduling of citrus
trickle irrigation. Proc. Fla. State Hort. Soc. 99:51-56.
Smajstrla, A. G., L. R. Parsons, K. Aribi, and G. Velledis. 1985a. Response of young
citrus trees to irrigation. Proc. Fla. State Hort. Soc. 98:25-28.
Smajstrla, A. G., L. R. Parsons, F. S. Zazueta, G. Vellidis, and K. Aribi. 1986. Water
use and growth of young citrus trees. ASAE Paper No. 86-2069, St. Joseph, MI.
Smajstrla, A. G. and F. S. Zazueta. 1987. Simulation of irrigation requirements of
Florida agronomic crops, Soil and Crop Science Soc. of Florida, Proc. 47:78-82.
Smajstrla, A. G. 1990. Technical Manual of AFSIRS Model. IFAS, University of
Florida, Gainesville, FL.
Snyder, G. H., B. J. Augustin, and J. M. Davidson. 1984. Moisture sensor-controlled
irrigation for reducing N leaching in bermudagrass turf. Agron. J. 76:964-969.
Srinivasan, R., B. A. Engel, and G. N. Paudyal. 1991. Expert system for irrigation
management. Agricultural Systems 36:297-314.

201
Stafford, J. V. 1988. Remote, non-contact and in-situ measurement of soil moisture
content: a review. J. Argic. Engng. Res. 41:151-172.
Stanley, J. M., C. Taylor, W. R. Summerhill, Jr, L. J. Beaulieu. 1980. Citrus energy
survey-use estimates and conservation. IFAS Energy Report 2. Univ. of Florida,
Gainesville, FL.
Stombaugh, T. S., P. H. Heinemann, C. T. Morrow, B. L. Goulart. 1992. Automation of a
plused irrigation system for frost protection of strawberries. Applied Eng. in
Agriculture 8(5):597-602.
Stone, K. C., A. G. Smajstrla, and F. S. Zazueta. 1986. Entrapped air and ceramic cup
effects on temperature response times. Soil and Crop Sci. Soc. of Florida, Proc.
46:26-29.
Strosnider, J. K., and C. J. Paul. 1994. A structured view of real-time problem solving. AI
Magazine, Summer, pp. 45-66.
Swaney, D. P., J. W. Jones, W. G. Boggess, G. G. Wilkerson, and J. W. Mishoe. 1983.
Real-time irrigation decision analysis using simulation. Trans, of ASAE 26:562-
568.
Taylor, S. A. 1955. Field determinations of soil moisture. Agr. Engineering 26:654-659.
Thomson, S. J., R. M. Peart, J. W. Mishoe. 1989. Expert system coupling of model and
sensor based techniques for irrigation scheduling. ASAE Paper No. 89-7585, St.
Joseph, MI.
Todorovic, P. and D. A. Woolhiser. 1975. A stochastic model of n-day precipitation. J.
of Applied Meteorology 14:17-24.
Topp, G. C. and J. L. Davis. 1985. Measurement of soil water content using time-domain
reflectometry (TDR): a field evaluation. Soil Sci. Soc. Am. J. 49:19-24.
Towner, G. D. 1980. Theory of time response of tensiometers. J. of Soil Sci. 31:607-
621.
Tucker, D. P. H. 1983. Citrus irrigation management. Florida Cooperative Extension
Service, Circular 444. IFAS, Univ. of Florida, Gainesville, Florida.
Turner, M. 1986. Real time experts. Systems International 14(l):55-57.

202
Vellidis, G., A. G. Smajstrla, F. S. Zazueta. 1990. Continuous soil water potential
measurement with a microcomputer-based data acquisition system. Applied Eng.
in Agriculture 6(6):733-738.
Villalobos, F. J. and E. Fereres. 1989. A simulation model for irrigation scheduling
under variable rainfall. Trans, of ASAE 32(1): 181-188.
Wheeler, P. A. and G. L. Duncan. 1984. Electromagnetic detection of soil moisture.
ASAE Paper No. 84-2078, St. Joseph, MI.
Wierenga, P. J., J. L. Fowler, and D. D. Davis. 1987. Use of tensiometer for scheduling
drip-irrigated cotton. Int. Con. on Measurement of Soil and Plant Water Status.
Centennial of Utah State Univ., Logan, pp. 157-161.
Wiersum, L. K. and K. Harmanny. 1983. Changes in the water permeability of roots of
some trees during drought stress and recovery, as related to problems of growth in
an urban environment. Plant and Soil 75:443-448.
Willis, L. E. and F. S. Davies. 1991. Fertigation and growth of young 'Hamlin' orange
trees in Florida. HortScience 26:106-109.
Wright, M. L., M. W. Green, G. Fiegl, and P. F. Cross. 1986. An expert system for
real-time control. IEEE Software 3:16-24.
Xin, J. N, F. S. Zazueta, and A. G. Smajstrla. 1993. Simulated effects of a rain-activated
irrigation shut-off-device on water use in turf irrigation. Soil Crop Sci. Soc.
Florida Proc. 52:107-114.
Zazueta, F, S., R. M. Lokers, and A. G. Smajstrla. 1989. Real-time expert control of
irrigation systems, ASAE Paper No. FCS 89-022, St. Joseph, MI.
Zazueta, F. S., S. Park-Brown, A. G. Smajstrla, and D. S. Harrison. 1984b. Computer
control of irrigation systems for nurseries. Proc. Fla. State Hort. Soc. 97:285-286.
Zazueta, F. S. and A. G. Smajstrla. 1992. Microcomputer based control of irrigation
systems. Applied Eng. in Agriculture 8(5):593-596.
Zazueta, F. S., A. G. Smajstrla, and G. A. Clark. 1986. Irrigation scheduling of peanut
in northern Florida. Soil and Crop Sci. Soc. of Florida, Proc. 46:60-71.
Zazueta, F. S., A. G. Smajstrla, and D. S. Harrison. 1984a. Microcomputer control of
irrigation systems I: hardware and software considerations. Soil and Crop Sci.
Soc. of Florida, Proc. 43:123-129.

203
Zazueta, F. S. and J. N. Xin. 1992. Soil-moisture and rainfall sensing devices: a review.
Final Report to South Florida Water Management District. Florida, Univ. of
Florida, Gainesville, 130pp.
Zazueta, F. S., J. N. Xin, and A. G. Smajstrla. 1993. Simulated computer-controlled
irrigation using soil water-content sensors with linear error functions. Soil Crop
Sci. Soc. Florida Proc. 52:102-107.
Zazueta, F.S., J. N. Xin, A. Wheaton, J. Jackson, and M. Almedo. 1994. Development
of a computer-based control system for reclaimed water citrus irrigation. Proc. of
the 5th Int. Con. on Computers in Agriculture, Orlando, FL.
Zoldoske, D. 1988. Computer irrigation scheduling. Fruit Grower, April, pp. 6-7.
Zur, B., U. Ben-Hanan, A. Rimmer, and A. Yardeni. 1994. Control of irrigation
amounts using velocity and position of wetting front. Irrigation Science 14:207-
212.

BIOGRAPHICAL SKETCH
The author was bom on February 16, 1961, in Yulin County, Shaanxi, China. After
receiving his high school diploma from Middle School of Yulin in 1978, he studied at the
Beijing University of Aeronautics and Astronautics (BUAA), China. In 1982, he received
a Bachelor of Science degree from BUAA. After graduating from BIAA, he was an
assistant engineer employed at the Research Institute of CFTRC, Xi'an, China. In 1987, he
was sent to the United States through the visiting scholar program at the University of
Florida. In 1988, he studied in the Agricultural and Biological Engineering Department at
the University of Florida and received a Master of Engineering degree in May, 1990.
204

I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and ig"fidly adequate in scope and quality,
as a dissertation for the degree of Doctor of Philos >phy\ f \ [ \ *
44-A^ -
Fetro S. Zazuetal Chair
Professorlof Agricultural and Biological
Engineering
I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is fully adequate, in scope and quality,
as a dissertation for the degree of Doctor of Philosophy.
Douglas D. Dankel, II
Assistant Professor of
Computer and Information Sciences
I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is fully adequate, in scope and quality,
as a dissertation for the degree of Doctor of Philosophy.
'rofessor of Agricultural and Biological
Engineering
I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is fully adequate, in pcope and quality,
as a dissertation for the degree of Doctor of Philosophy^
Pierce H. Jones
Associate Professor of Agricultural and
Biological Engineering

I certify that I have read this study and that in my opinion it Conforms to
acceptable standards of scholarly presentation and is fully adequate, in scope and quality,
as a dissertation for the degree of Doctor of Philosoffyy.
Louis H. Motz
Associate Professor of Civil
Engineering
I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is fully adequate, in scope and quality,
as a dissertation for the degree of Doctor of Philosophy.
AULft.^^
Allen G. Smajstrla
Professor of Agricultural and Biological
Engineering
This dissertation was submitted to the Graduate Faculty of the College of
Engineering and to the Graduate School and was accepted as partial fulfillment of the
requirements for the degree of Doctor of Philosophy.
August, 1995 f £ -\
h" Winfred M. Phillips
Dean, College of Engineering
Karen A. Holbrook
Dean, Graduate School



174
database (Figure 13.5.9). Crop database contains crop property such as age of the tree, crop
root depth, and crop canopy diameter.
Crop coefficients and MAD
Click Crop Coefficient
on the submenu of Database to
view the crop coefficients and
Citrus Cr)|> Gocrfficienl Mini MnnHcpiinenl Allowed Depletion
Crop name
Variety
Vegetated surface (Y/N)
Crop Coefficient
Management Allowed
Depletion
Management Allowed Depletion
(MAD) (Figure 13.5.8). In
order to calculate actual crop
ET, crop coefficients must be
available. MAD is used for the
water-budget simulation.
Munlh Kc
January
P
February
0-
March
!'
April
ms
May
m
June
m
July
mm
August
. m
September
res
October
November
December
Us
MAD
mm
m
B<5
0.13
8.3;
w
m
as?
iti*
¡MJ
i
lUi
Figure 13.5.10. Crop coefficient database
screen.
Soil database
Click Soil on the
submenu of Database to view the
soil database (Figure 13.5.11).
Soil database listed the common
soil series and soil-water holding
capacity. If the soil exists in the
site, enter YES in the field Exist
in the site (YES/NO).
Figure 13.5.11. Soil database screen.


118
prognosis of irrigation requirements can be produced by the simulation. When the system
is operated by user-defined irrigation schedules, the simulation results can be used to assist
the users in defining their irrigation schedules.
9.2.6 Tools
Because evapotranspiration (ET) is one of the major factors for irrigation
management, ET and other management utilities were included in the system.
Load weather data converts weather data from an ASCII file to a database
file. Then the user can manipulate the data using database management
capabilities included in the system.
ET calculation daily or monthly ET can be estimated by the following
methods: (1) Penman, (2) Blaney-Criddle, (3) modified Blaney-Criddle
method using solar radiation, and (4) Stephens-Stewart method.
Irrigation duration estimates duration of a microirrigation event to bring
soil water to field capacity.
Field layout map -- displays the irrigation pipe layout and sensor locations
in the field.
9.2.7 Help
This module consists of several computer tools and user's guide for the system.
These tools include (1) hypertext help on how to use the system, (2) calculator, diary, and
clock, (3) filer and editing utilities, and (4) game.


CHAPTER 8
CONSTRUCTION OF THE KNOWLEDGE BASE
8.1 Introduction
The crucial step in the development of an expert system is the construction of the
knowledge. The knowledge base deals mainly with the issues of uncertainty management
of the sensor data, irrigation, fertigation, and cold protection. In particular, decisions on
citrus irrigation management may involve (1) when and how long an irrigation should be
applied, (2) when cold protection should be applied, and (3) when and how long fertigation
should be applied. To make management decisions in real time, the system needs on-site
Figure 8 .1. Inputs and outcomes of the expert system.
88


190
Burns, I. L., E. R. Muller, E. Bell, and S. Novak. 1990. A weighing lysimeter for
evapotranspiration monitoring and irrigation control. Proc. of the 3rd Int.
Conference on Computer in Agr. Extension Programs, Orlando, FL, pp. 73-78.
Cahoon, J., J. Ferguson, D. Edwards, and P. Tacker. 1990. A microcomputer-based
irrigation scheduler for the humid mid-south region. Applied Eng. in Agriculture
6(3): 289-294.
Campbell, G. S. and M. D. Campbell. 1982. irrigation scheduling using soil moisture
measurements: Theory and practice. In Advances in Irrigation, ed. D. Hillel.
New York: Academic Press, 1:25-54.
Carlisle, V. W., R. E. Caldwell, F. Sodek, III, L. C. Hammond, F. G. Calhoun, M. A.
Granger, and H. L. Breland. 1978. Characterization Data for Selected Florida
Soils. IFAS, Univ. of Florida, Gainesville, FL.
Carr-Brion, K. 1986. Moisture Sensors in Process Control, Elsevier Applied Sci.
Publishers, London.
Cary, J. W. and H. D. Fisher. 1983. Irrigation decisions simplified with electronics and
soil water sensors. Soil Sci. Soc. Am. J. 47:1219-1223.
Cassell, D. K. and A. Klute. 1986. Water potential: tensiometry, in Methods of Soil
Analysis, Part 1, Physical and Mineralogical Methods, ed. A. Klute, 2nd. edition.
Soil Society of American, Madison, Wisconsin.
Chesness, J. L., D. L. Cochran, and J. E. Hook. 1986. Predicting seasonal irrigation
water requirements on coarse-textured soils. Trans, of the ASAE 29(4): 1054-
1057.
Collins, A. G., S. J. Nix, T. Tsay, A. Gera, and M. A. Hopkins. 1990. The potential for
expert systems in water utility operation and management. Journal AWWA 9:44-
51.
Creighton, Jack, David A. Sleeper, and Calvin Hubbard. 1989. Tensiometers for
irrigation scheduling in Florida citrus grove. Proc. Fla. State Hort. Soc. 102:69-
72
Davies, F. S., T. E. Marler, and L. R. Parsons. 1989. Irrigation of young citrus trees.
Citrus Industry, April, pp. 5-9.


122
approach is an ad hoc development that lacks a standard or practical design approach that
can be generally used. Furthermore, expert systems often must be maintained by someone
other than the developers (Prerau et al., 1990). Therefore, planning for long-term
maintenance when designing an expert system is critical for the system's continued success.
The maintenance strategy must be considered at all design stages.
Maintenance of CIMS involves two tasks: (1) conventional software maintenance,
and (2) knowledge base maintenance. The system's structural design attempts to simplify
the maintenance task so that the user can maintain the system if necessary. The following
steps have been to facilitate system maintenance.
Modularity of rule base:
The production rules are grouped by tasks (Figure 9.5). Rules dealing with
the same event were organized into modules. Since rules in the knowledge
base may need to be changed, this modularity of the rule base can easily
identify which groups of rules need to be changed. In addition, rules are
loaded into the computer memory module by module, which reduces the
computer memory requirements.
User accessible critical values.
The user can directly modify the initial facts, such as tree status and some
critical values to start an irrigation. To do this, the user does not need to
understand the rule base.


171
Browse
This routine allows the user to browse data from a database under the
user defined constraints. A browse dialog screen will open as Figure
13.5.4. The steps to browse a database are described as follows.
Figure 13.5.4. Browse dialog window.
Select a data field from the left pull down box in the Add box.
Define the browse condition at the right pull down box in the Add box.
Click the Add button to add the filter criterion to the criteria box.
Repeat steps (1), (2), and (3) to add additional criteria.
Click the Browse button to view the database filtered by the criteria.
Click the Reset button to clear the defined filter criteria.


183
13.7.3 Calculate Irrigation Duration
Irrigation duration can
be estimated when one has the
available data: (1) current soil
water depletion, (2) soil
characteristics, (3) irrigation
system, and (4) the crop root
zone. To calculate irrigation
Estimated irrigation time to bring root zone to field capacity
Microirrigation system data
Emitter flow rate (gallons/hour)
Emitter number / tree
Diameter of spray pattern (feet)
Overall irrigation efficiency (%)
Crop and soil data
Crop effective root depth (feet)
Available soil water content (in/ft)
Depletion of available water
prior to irrigation (%)
Estimated irrigation duration
m
i
u
m
houf 3? mu
Figure 13.7.8. Estimate irrigation duration screen.
duration, execute the following
steps.
Click Estimation Irr.
Duration on the submenu of
Tools. A screen as Figure
13.7.8 appears for the
calculation.
Enter the required data to the
screen and click the Calculate
Help of irrigation duration utility
Enter data of irrigation system, soil and crop root depth
Click Co/cu/ata to calculate irrigation duration
Click Caoce/ to exit from the screen
Average available waler holding capacity
for various soil type
Soil type
AW holding capacity (intft)
Coarse sand
072
Fine sane
0.96
Loamy sand
1.20
Sandy loam
1.38
Fine snady loam
1.56
Very fine sandy loam
1.68
ay and day loom
1.80
Silty day and silty day loam
1.92
Silt loam
222
Peats and mucks
2.52
Martin et al.. 1990 Irrigation scheduling principles, in
Management of Form irrigation System. ASAE
Figure 13.7.9. Help screen of irrigation
duration.
button to display the result.
Click the Help button to view soil data (Figure 13.7.9).
Click the Exit button to quit from the screen.


97
sensors are installed at different soil depths, several combinations of sensor output
differences can be checked. For sensors SI and S3, the combination is showed in Figure 8.6
and Table 8.1. Sensor S2 has two adjacent sensors. The combinations checked are shown
in Figure 8.7 and Table 8.1. The "R" notation in Figures 8.6 and 8.7 represents rules used
for data uncertainty management. Sensor readings are also compared among sensors at
different locations with the same installation depth. Table 8.2 shows the comparison pairs
to check the possible failure. This possible failure is denoted as Failure C.
Table 8.1 Criteria for checking possible sensor Failure B.
Sensor S1
Sensor S2
Sensor S3
|S1 -S2| > EGV
|S2 SI | > EGV
| S2 S31 > EGV
| S2 S31 < EGV
| S3 S21 > EGV
|S1 -S2| < EGV
Table 8.2 Criteria for checking possible sensor Failure C for sensors
at the same depth from different locations.
Sensor S1
Sensor S2
Sensor S3
iSi-Sil > EGV
| S¡ Si | > EGV
|Si- Sil > EGV
'lote: i and j represent sensors at same depth but in difieren
Propagation of CF
locations.
The three possible failures (Failures A, B and C) reduce confidence in sensor
readings. Overall confidence in sensor readings relies on the final propagation of CF. The
decision that a sensor failure has occurred depends on what CF value is assigned to the


98
Table 8.3 A sample propagation of CF.
Sensors 1 or 3
Instance
Initial CF
F ailure A
Failure B
Failure C
Combined CF
1
0.95
-0.9
i
o
-0.8
-0.75
2
0.95
-0.9
-0.4
N/A
0.17
3
0.95
N/A

1
l
o
bo
0.55
4
0.95
N/A
N/A
-0.8
0.75
5
0.95
N/A

1
N/A
0.91
6
0.95
-0.9
N/A
N/A
0.50
Sensor 2
Instance
Initial CF
Failure A
Failure B
Failure C
Combined CF
7
0.95
-0.9
i
o
4^
-0.4
-0.28
8
0.95
l

VO
-0.4
N/A
0.17
9
0.95
N/A

1
i
o
0.86
10
0.95
N/A
N/A
-0.4
0.92
11
0.95
N/A
i
o
N/A
0.92
12
0.95
-0.9
N/A
N/A
0.50
possible failure and at what confidence level the sensor reading is considered valid. In other
words, the value of the final propagation of CF and the threshold of valid CF values
determine whether a sensor reading is valid or not. For example, if an initial CF value is
assigned as 0.95 and the CF values of Failures A, B, and C are -0.8, -0.7, and -0.4,
respectively, Table 8.3 shows the likely combined CF results. If the value 0.88 was used as
the valid threshold, instances 5, 10, and 11 are valid readings. Instances 4 and 9 would also
be valid if the threshold was reduced to 0.75.


197
Parsons, L. R., K. T. Morgan, and T. A. Wheaton. 1993. Effects of microsprinkler
precipitation rate, soil type, and water depletion on depth of soil wetting. Proc.
Fla. State Hort. Soc. 106:38-41.
Peart, R. M., F. S. Zazueta, P. Jones, J. W. Jones, and J. W. Mishoe. 1986. Expert
systems take on three tough agricultural tasks. Agricultural Engineering,
May/June, pp. 8-10.
Phene, C. J. 1986. Operation principles: automation, In Trickle Irrigation for Crop
Production: Design, Operation and Management, ed. F. S. Nakayama and D. A.
Bucks. Elsevier, Tokyo, pp. 188-215.
Phene, C. J. 1989. Techniques for computerized irrigation management. Computers and
Electronics in Agriculture 3:189-208.
Phene, C. J., C. P. Allee, and D. J. Pierro. 1989b. Soil matric potential sensor
measurements in real-time irrigation scheduling. Agrie. Water Management
16:173-185.
Phene, C. J., W. R. DeTar, and D. A. Clark. 1992. Real-time irrigation scheduling of
cotton with an automated pan evaporation system. Applied Eng. in Agriculture
8(6):787-793.
Phene, C. J., G. J. Hoffman, and R. S. Austin. 1973. Controlling automated irrigation
with soil matric potential sensor. Trans, of ASAE, Paper No. 71-230, St. Joseph,
MI.
Phene, C. J. and T. A. Howell. 1984. Soil sensor control of high-frequency irrigation
systems. Trans, of ASAE 27:392-396.
Phene, C. J., R. L. McCormick, K, D. Davis, J. D. Pierro, D. W. Meek. 1989a. A
lysimeter feedback irrigation controller system for evapotranspiration
measurements and real time irrigation scheduling, Trans, of ASAE 32:477-484.
Plant, R. E. 1989. An integrated expert decision support system for agricultural
management. Agricultural System 29:49-66.
Pleban, S., J. W. Labadie, and D. F. Heermann. 1983. Optimal short term irrigation
schedules. Trans, of ASAE 26:141-147.
Pleban, S., D. E. Heermann, J. W. Labadie, and H. R. Duke. 1984. Real time irrigation
scheduling via "reaching" dynamic programming, Water Resources Research
20:887-895.


53
data as initial facts to conduct its reasoning process. Then, the irrigation control valves can
be turned on or off through the irrigation control board according to the results of the
reasoning process.
4.7 Paradigm for the Real-Time Expert System
Since CLIPS provides a built-in inference engine for forward reasoning, the
development is focused mainly on the formation of the knowledge base (KB), external
control procedures, and the user interface. Figure 4.7 shows the structure of the system.
Input data from the sensors are collected and stored by external devices. Users can select
the frequency of downloading data from the data logger to the CLIPS fact base (FB). Before
the data are transferred to FB, a data pre-process procedure is needed to reorganize the data
format into the CLIPS data format.
Figure 4.7. Paradigm of the real-time expert system.


3
Villalobos and Fereres, 1989; Rogers and Elliott, 1989). Irrigation scheduling approaches
based on crop growth models and soil water budget components have been developed by
researchers (Jensen et al., 1971; Chesness et al., 1986; Smajstrla and Zazueta, 1987; Jones
and Ritchie, 1990). Models have also been developed to determine optimal irrigation
strategies using stochastic and probabilitic models of weather variables (Khanjani and
Busch, 1982). Long-term historical weather data are commonly used by such simulation
models. Many agricultural simulation models, which use historical weather data, have
succeeded in planning and long-term prediction of irrigation management. However, factors
such as the difficulty of model development, uncertainty of future conditions, and limitations
of available data combine to make the use of simulation models difficult for real-time
application.
With increasing competition for the use of water and high energy costs associated
with irrigation, microirrigation systems have become common in Florida, particularly for
high cash value crops like citrus. Because a microirrigation system wets only the soil
volume around the emitters, microirrigation systems must apply water at a high frequency.
Water should be applied at a rate equal to plant uptake (Phene et al., 1992). The soil water
potential can be maintained reasonably constant under high frequency irrigation scheduling.
Irrigation can be applied at least daily at a rate equal to the ET requirement; consequently,
there is a need for "real-time" irrigation scheduling and control systems (Phene et al.,
1989b).
Cold protection is an important issue for citrus growers. Cold weather has caused
severe economic damage to the Florida citrus industry in the past, particularly in 1989. The


99
Since soil moisture sensors are
installed in at least two locations to
reduce the effects of non-uniform soil
characteristics and crop root
distributions, only the valid sensor
readings from the two sites are
averaged and then used in the
reasoning process. As Figure 8.8
shows, sensor readings (Sx, x=1, 2,
and 3) at the same depth from
different locations (ID1 and ID2) are
averaged if the readings are valid. If
both sensor readings at the same depth
are valid, the average value is used
(Fig. 8.8 (a)). When one of the sensors is considered to have failed, only the valid sensor
reading is used (Fig. 8.8 (b) and (c)). The invalid reading is discarded and a warning
message is displayed for maintenance purposes. When both sensors at the same depth
from different locations are considered to have failed (Fig. 8.8 (d)), both of the readings
at this depth are discarded from the reasoning process. Thus, a dependable decision can
be made when one sensor has failed or when sensor data are missing. Some sample sensor
data are included in Appendix A.
(d)
Note: ID1-Sx = Sensor readngs at different sol depths at location A
ID2-Sx = Sensor readings at afferent sol depths at location B
Figure 8.8. Process of selecting valid sensor
readings from different locations.


162
13 4 Irrigation and Fertigation Control Based on User Defined Schedules
Because the RTES requires soil moisture sensors and a weather station, cost of the
system can be high due to the hardware requirement. Thus, an alternative approach of
irrigation and fertigation control scheme was developed. This approach allows the users to
define their own application schedules. Thus, the control actions can be executed based on
the schedules. Since the control events rely only on the user defined schedules, soil-water
content and climate data are not required. However, the user must be knowledgeable of the
crop water and nutrient requirement to define a sound application schedule.
13,4,1 Define Irrigatipn and Furtigation Schedule
The steps to define irrigation and fertigation schedules are as follows.
Click Scheduling on the main menu of CIMS. The submenu of Scheduling shows as
Figure 13.4.1.
Facts Expert Control
Scheduling
Irrigation Schedule
Apply Irrigation
Database Simulation Tools Help Quit
Fertigation Schedule
Apply Fertigation
Apply Irrigation & Fertigation
Figure 13.4.1. Submenu of user defined control schedules.
Click Irrigation Schedule to define an irrigation schedule (Figure 13.4.2).
Enter the data showed in Figure 13.4.2. These data include following parameters.


61
probabilities. The one sample t statistic (Moore and McCabe, 1989) has the t distribution
with n-1 degrees of freedom:
t
5-4
Within each season, a paired t-test between monthly rainfall probability was conducted.
Subjects were matched in pairs, and the outcomes were compared within each matched pair.
The results of the paired t-test are shown in Table 5.2. In winter and spring, there are no
significant differences in rainfall probability among the months at the 95 percent confidence
level. The same test showed a significant difference in rainfall probabilities for summer and
fall at the 95 percent confidence level. Because wet-day frequencies during winter and
spring showed no significant differences, the average value of the seasonal rainfall
probability can be used to represent the seasonal rainfall probability. However, probabilities
of rainfall during summer and fall must be treated as monthly because they are significantly
different among months within the seasons.
Table 5.2 Results of paired t-test for rainfall probabilities within each season.
Winter
Spring
Summer
Fall
Jan vs Nov
NS
Feb vs Mar
NS
May vs Jun
S
Aug vs Sep
S*
Jan vs Dec
NS
Feb vs Apr
NS
May vs Jul
S
Aug vs Oct
S
Nov vs Dec
NS
Mar vs Apr
NS
Jun vs Jul
S
Sep vs Oct
S
NS = not significant different at 95% confidence level.
S = significant different at 95% confidence level.
S* = significant different at 90% confidence level.


155
Define irrigation control valve (on or off)
Click Irrigation on the submenu of Facts, and an editing window (Figure A.2.2) is
displayed.
Enter the irrigation block and
valve numbers in the first two
columns. Then enter the valve
on or off value. Value 1
represents the valve will be
turned on and value 0 represents
that the valve will not be turned
on when an application occurs.
Thus, irrigation and fertigation
can be applied only on the
predefined blocks.
Define cold protection control valve (on or offl
Click Freeze Protection on the submenu of Facts.
The procedures to define a valve on or off for cold protection are same as the
procedures for defining the valve on or off for irrigation. For cold
protection, all valves may be specified as on (value 1) when the application
is required.
Figure 13.2.2. Irrigation block and valve
definition.


A REAL-TIME EXPERT SYSTEM
FOR CITRUS MICROIRRIGATION MANAGEMENT
By
JIANNONG XIN
A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
UNIVERSITY OF FLORIDA
1995


58
Thus, a Markov chain of wet day probability with order 3 can be formed with the
conditional probability:
P{W|D1D2D3} = P{Wt= xJXw- x, X= x,2) Xt_3~ x,_3} 5-3
where W = wet day,
D, X = daily sequence of rainfall event, and
xt = random variable of a wet and dry day at day t.
The occurrence of the previous three consecutive days (D,D2D3) could be {000 001
010011 100 101 110 111}, where 0 represents a dry day and 1 represents a wet day. This
third-order Markov chain was used to describe rainfall occurrence only. The amount and
intensity of rainfall is not described by this equation. The probability of a wet day in the
immediate future is based on past long-term weather data rather than the changing
meteorological conditions.
5.3 Rainfall Data
Forty years of daily rainfall data from 1952 to 1992 were used to study the sequence
of wet and dry days. This weather station is located in Orlando, Florida (latitude 28:27:00,
longitude 81:19:00). These data were obtained from the NOAA weather station and the
Earthlnfo CD-ROM disk (NOAA, 1952-1992).


104
A range of soil-water potentials is specified to increase the possibility of satisfying these
constraints simultaneously.
As Table 8.4 shows, the soil-water potentials of -10 cb, -15 cb, and -25 cb
approximately correspond to 25 percent, 45 percent, and 55 percent soil-water depletion for
Candler fine sand, respectively. These values can be easily modified by the user to satisfy
a unique soil type or management requirements.
Starting irrigation by Criteria I considers both crop water requirements and reduction
of water loss due to evaporation. However, these conditions may not be satisfied
simultaneously even when the sensor readings are over the maximum allowed values.
Irrigation starting Criteria II (Figure 8.11) is used when Criteria I is not satisfied and the
Figure 8.11. Decision process (criteria II) to start an irrigation and critical sensor
readings for trees during different growth stages.


59
5.4 Frequency of Rainfall
A third-order Markov chain analysis was conducted using the forty-year rainfall data
in Orlando, Florida. The prediction derived here is a long term estimation of rainfall
occurrence. The probability of rainfall occurrence of each day is considered only dependent
upon the wet-dry sequence of the three previous days. Each day can be either wet or dry.
Days with a trace of rainfall (less than 0.01 inch) are considered to be a dry day. Table 5.1
shows the wet-day probabilities.
Table 5.1 Markov chain wet-day frequency.
Previous
Case Jan
Feb
Mar
Apr
Month of year
May Jun Jul
Aug
Sep
Oct
Nov
Dec
000
0.26
0.21
0.23
0.15
0.20
0.38
0.45
0.39
0.33
0.19
0.18
0.22
001
0.20
0.30
0.29
0.19
0.22
0.33
0.40
0.53
0.37
0.27
0.20
0.20
010
0.22
0.29
0.22
0.21
0.35
0.45
0.53
0.47
0.46
0.22
0.18
0.20
Oil
0.23
0.20
0.15
0.22
0.35
0.37
0.51
0.50
0.45
0.24
0.10
0.17
100
0.40
0.54
0.43
0.39
0.50
0.65
0.59
0.70
0.54
0.56
0.44
0.35
101
0.44
0.37
0.49
0.47
0.52
0.62
0.65
0.63
0.69
0.59
0.44
0.53
110
0.36
0.41
0.47
0.45
0.56
0.70
0.70
0.70
0.70
0.59
0.51
0.37
111
0.35
0.46
0.45
0.47
0.57
0.70
0.75
0.69
0.65
0.61
0.39
0.43
Rainfall data is for Orlando, Florida from 1952 to 1990.
0-Dry day
1-Wet day
This probability indicates the likelihood or frequency of rainfall during a particular month
and previous wet-dry day sequence. As the results show, the rainfall frequency ranges from
10 percent to 70 percent. Summer and fall have a higher wet day frequency than winter and
spring. Higher probability of rainfall during June through September is due to the higher
rainfall occurrence during these months. Figure 5.1 shows the annual rainfall distribution
in Orlando over forty-year average data. Approximately 70 percent of the annual rainfall
occurs during the summer and fall seasons. Thus, higher wet-day frequencies (> 0.6 from


172
To browse entire database, there is no need to create a criterion as described
in the previous step. Click the Browse button to browse the entire database.
Click the Quit button to exit from the search dialog window.
facts Expert Control Scheduling
| Simulation fools Help Quit
Weather Data
irrigation System
Crop
Crop Coefficient
Soil Data
Figure 13.5.5. Submenu of the Database main menu.
13,5,2 Databases of CIMS
Farm database
Figure 13.5.5
shows the submenu of
Database. To view the
Farm database, click
Farm on the submenu of
Database in Figure
13.5.5. The farm
database contains the
owner's information, mailing address, and phone and fax numbers (Figure 13.5.6).
Farm name
Owner or contact
First name Middle
Address
C<*y IgEftSMSS 1 stole
Phone
BB1
Farm Database
1 Last
mzmzzi
A Z'P 1
Figure 13.5.6. Farm database screen.


101
Figure 8.9. Decision process to use a full or deficit irrigation strategy.
Expert reasoning
Expert reasoning determines a full or deficit irrigation strategy according to age of
tree, crop growth stages, and probability of rainfall (Figure 8.9). Since the system can
obtain real-time climate data from the automated weather station, the wet-dry day sequence
for the most recent three days can be used to estimate today's rainfall probability based on
the Markov chain results as discussed in Chapter 6. A high probability of rainfall was
assumed when the rainfall probability was greater than a specified value (for instance, > 60


102
percent). If a weather station is not available, the user can enter probability of rainfall to the
system according to a weather forecasting network.
During the non-critical growth stages, mature trees can sustain some water stress
without causing significant yield losses. A deficit (partial) irrigation strategy may be
suitable for this situation. For mature trees during the non-critical growth stage, deficit
irrigation may be applied when the probability of rainfall is high. Irrigation can be delayed
or less water may be applied if a high probability of rainfall occurs at a given time. Full
irrigation is always suggested for the rest of the growing season.
8,4,2 Criteria for Starting an Irrigation
Irrigation is usually scheduled during a period of low evaporative demand. Thus,
both time and weather constraints are attached to the criteria. Time constraints can be
imposed by regulatory requirements, and weather constraints may result from favorable
weather conditions such as low wind speed and high relative humidity. Overall, time and
weather constraints ensure that irrigation is applied during a low evaporative demand period.
Soil-water potential plays an important role in irrigation management. The critical
value of soil-water potential, related to the maximum allowed depletion, varies with the crop
growth stage. Young trees and mature trees during the critical growth stage require a higher
soil-water content than mature trees during the non-critical growth stage. Irrigation may be
applied when the soil-water potential is in the trigger irrigation range (Figure 8.10).
Irrigation starting criteria based on soil-water potential and constraints of time and weather
are denoted as Criteria I. A set of sample values of Criteria I is shown in Table 8.4.


201
Stafford, J. V. 1988. Remote, non-contact and in-situ measurement of soil moisture
content: a review. J. Argic. Engng. Res. 41:151-172.
Stanley, J. M., C. Taylor, W. R. Summerhill, Jr, L. J. Beaulieu. 1980. Citrus energy
survey-use estimates and conservation. IFAS Energy Report 2. Univ. of Florida,
Gainesville, FL.
Stombaugh, T. S., P. H. Heinemann, C. T. Morrow, B. L. Goulart. 1992. Automation of a
plused irrigation system for frost protection of strawberries. Applied Eng. in
Agriculture 8(5):597-602.
Stone, K. C., A. G. Smajstrla, and F. S. Zazueta. 1986. Entrapped air and ceramic cup
effects on temperature response times. Soil and Crop Sci. Soc. of Florida, Proc.
46:26-29.
Strosnider, J. K., and C. J. Paul. 1994. A structured view of real-time problem solving. AI
Magazine, Summer, pp. 45-66.
Swaney, D. P., J. W. Jones, W. G. Boggess, G. G. Wilkerson, and J. W. Mishoe. 1983.
Real-time irrigation decision analysis using simulation. Trans, of ASAE 26:562-
568.
Taylor, S. A. 1955. Field determinations of soil moisture. Agr. Engineering 26:654-659.
Thomson, S. J., R. M. Peart, J. W. Mishoe. 1989. Expert system coupling of model and
sensor based techniques for irrigation scheduling. ASAE Paper No. 89-7585, St.
Joseph, MI.
Todorovic, P. and D. A. Woolhiser. 1975. A stochastic model of n-day precipitation. J.
of Applied Meteorology 14:17-24.
Topp, G. C. and J. L. Davis. 1985. Measurement of soil water content using time-domain
reflectometry (TDR): a field evaluation. Soil Sci. Soc. Am. J. 49:19-24.
Towner, G. D. 1980. Theory of time response of tensiometers. J. of Soil Sci. 31:607-
621.
Tucker, D. P. H. 1983. Citrus irrigation management. Florida Cooperative Extension
Service, Circular 444. IFAS, Univ. of Florida, Gainesville, Florida.
Turner, M. 1986. Real time experts. Systems International 14(l):55-57.


APPENDIX C
CIMS USER'S GUIDE


173
Weather database
Click Weather on the
submenu of Database to view the
weather database (Figure 13.5.7).
The weather database contains data
from the automated weather
station. The weather data are
stored in daily format.
Evapotranspiration can be estimated
based upon the data.
Irrigation database
Click Irrigation System on
the submenu of Database to view
the data of irrigation system (Figure
13.5.8). Irrigation database
contains microirrigation system
parameters such as flow rate,
emitter number, and wetted
diameter, etc.
Crop database
Click Crop on the submenu
of Database to view the crop
Figure 13.5.7. Weather database screen.
Figure 13.5.8. Irrigation database screen.
Tree Database
Block ID
Crop name
Variety
Planting date
Tree space (ft)
Tree height (ft)
Canopy diameter (ft)
Land coverage (%)
Root depth (ft)
Acreage
Top
Prior
Bottom
Figure 13.5.9. Crop database screen.


22
contains the representation of domain specific knowledge. The essence of the knowledge
base must fit the structure of the knowledge representation scheme.
The strength of an ES lies in the knowledge base because the knowledge base
contains a representation of the human expertise of the problem-solving decision. Therefore,
the whole process of knowledge acquisition is crucial in the system's development. The
transferring of domain expertise to an ES's knowledge base has proved difficult and time
consuming, in part because the process requires the interposition of a knowledge engineer
between the human expert and a computer.
3.1.3 User Interface
Because ESs are generally interactive and involve users with little or no computer
experience, the user interface should be designed to be friendly, explainable, and easy to use.
A clear definition of the user interface requirements for an ES is essential to the success of
the system. In particular, for users to accept the interface, it must accomplish the task in a
straightforward way and still meet the entire range of problem solving requirements. To
develop a friendly and explainable user interface, Ege and Stary (1992) suggested that the
designers need to provide a global system perspective to create task-oriented, or user-
centered user interfaces.
3 2 Real-Time Expert Systems
Historically, AI researchers have focused on problems in which the time response
is not a concern, such as the medical diagnostic ES (i.e., MYCIN, Shortliffe and Buchanan,
1975). This kind of system is asking humans to supply necessary inputs, and the response


68
vegetation. Consequently, conventional estimation of consumptive use, which assumes
wetting the entire field surface, needs to be modified for microirrigation.
The transpiration rate under microirrigation is a function of the conventionally
computed consumptive use rate and the extent of the plant canopy (Sharpies et al., 1985).
Keller and Biliesner (1990) used a simple equation for estimating the average daily
transpiration rate:
ETm ETa[0.\ (.PJ)05] 6-2
where
ETm =
average daily transpiration rate for a crop under
microirrigation, in/day,
ETa =
conventionally estimated average daily consumptive use,
in/day, and
Pa
percentage of soil surface area shaded by crop canopies at
midday (solar noon), %.
In Florida, citrus crop coefficients (Rogers et al., 1983), Kc, with grass coverage and citrus
irrigation Management Allowed Depletion (Koo, 1963), MAD, are given in Table 6.2.
MAD is used to express the amount of water that can be depleted in the crop root zone
without adversely affecting the plant.
Table 6.2 Citrus crop coefficients and recommended MAD in Florida.
Month
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
Kc
0.90
0.90
0.90
0.90
0.95
1.0
1.0
1.0
1.0
1.0
1.0
1.0
MAD (%)
67
67
33
33
33
33
67
67
67
67
67
67


65
65
66
66
66
66
66
66
66
66
66
66
66
66
66
66
66
66
66
66
66
66
66
66
66
66
67
67
67
67
67
67
67
67
67
67
67
67
67
67
67
67
67
67
67
67
67
67
67
67
138
2300
17.4
18.2
90.1
0
0
0
7.62
9.05
11.1
8.69
2400
17.5
17.6
90.1
0
0.4
0
7.57
9
11.1
8.69
100
17.4
17.7
87.7
0
0.4
0
7.53
8.91
11.1
8.63
200
17.1
17.4
86.8
0
0.1
0
7.51
8.91
11.1
8.66
300
16.8
17.2
86.5
0
0
0
7.48
8.92
11
8.61
400
16.4
16.8
88.1
0
0
0
7.58
9.05
11.2
8.75
500
16.4
16.7
85.9
0
0
0
7.5
9
11
8.55
600
16.2
16.4
87.1
0
0
0
7.63
9.1
11.3
8.91
700
16.3
16.8
85
0.01
0.3
0
7.57
9.08
11.2
8.79
800
15.7
18.4
86.8
0.26
0
0
8.14
9.46
13
10.3
900
18.5
21.8
69.64
0.81
1.6
0
7.86
9.19
10.9
8.75
1000
21.8
23.8
61.48
1
2.1
0
7.85
9.07
11.9
9.03
1100
23.8
25.7
55.35
1.81
3.1
0
7.19
8.5
11
8.78
1200
25.2
26.6
47.71
1.97
3.4
0
7.45
9.01
11.3
9.15
1300
27.5
27.6
39.81
0.03
1.2
0
6.86
8.25
9.45
7.08
1400
27
27.2
38.62
0.27
0.9
0
8.42
9.64
11.6
9.13
1500
26.4
28.5
38.49
6.62
0.7
0
7.57
8.41
10.4
7.66
1600
25.8
28.4
39.68
4.44
0.2
0
7.92
9.22
11.3
8.5
1700
25.4
28.8
39.88
2.74
0
0
7.54
8.52
11.2
8.21
1800
25.7
28.2
40.37
2.49
0.2
0
8
9.02
11.8
8.58
1900
22
26.1
52.64
0.21
0.3
0
8.17
9.13
11.6
8.48
2000
20.2
22
68.58
0.03
0.4
0
8.27
9.15
12
8.75
2100
19.1
20.2
78.1
0.03
0.1
0
8.35
9.14
12.1
8.77
2200
18.1
19.2
84.5
0.03
0
0
8.28
9.06
12
8.78
2300
17.2
18.2
89.6
0.03
0
0
8.39
9.1
11.8
8.85
2400
16.6
17.2
92.7
0.03
0
0
8.31
9.02
11.5
8.79
100
16.1
16.7
94.8
0.03
0
0
8.23
8.99
11.4
8.78
200
15.4
16.2
95.8
0.03
0
0
7.95
8.81
10.5
8.54
300
14.4
15.6
96.8
0.05
0
0
7.89
8.9
10.7
9.11
400
14.2
14.7
98
0.06
0
0
7.94
9
10.3
8.91
500
14.2
15
98.6
0.05
0
0
8.32
9.16
10.9
9.21
600
14.5
15.4
99
0.03
0
0
8.4
9.21
11.3
9.19
700
14.8
15.5
99.1
0.11
0
0
8.62
9.36
11.6
9.59
800
15.1
17.6
99.3
2.16
0
0
9.14
9.36
13.8
10.36
900
17.5
19.6
99.2
5.07
0
0
8.33
8.92
11.9
9.42
1000
19.6
24
84.8
6.33
0.7
0
8.72
9.14
13.5
9.77
1100
24
25.7
61.45
11.7
1
0
9.63
9.73
13
9.52
1200
25.7
27.6
50.63
14
0.9
0
8.5
9.03
11.5
8.53
1300
26.8
28
46.89
14.2
0.9
0
8.54
9.1
12.7
9
1400
27.7
28.7
44.1
13.5
1.1
0
8.71
9.52
13.1
9.78
1500
27.7
30.2
41.91
11.7
0.9
0
8.37
8.47
12.6
8.95
1600
27.7
30.8
41.28
9.02
0.7
0
9.83
10.3
13.2
10.25
1700
26.7
28.2
50.14
4.18
1.1
0
8.69
9.12
12.6
9.61
1800
25.4
27.5
54.91
1.41
1
0
8.97
9.49
13.5
9.53
1900
22.3
25.4
63.6
0.16
0.4
0
8.87
9.49
13.7
9.16
2000
21.3
22.3
73.7
0.03
0.6
0
9.07
9.52
14.2
9.31
2100
20.4
21.4
82.2
0.03
0.8
0
8.99
9.41
14.2
9.14
2200
19.7
20.4
85.9
0.03
0.7
0
9
9.35
14.2
9.11
2300
18.6
19.7
90
0.03
0
0
9.14
9.48
14.1
9.27
2400
18.1
18.7
93.7
0.03
0
0
9.15
9.48
13.9
9.27


18
(4) processing feedback data to evaluate the irrigation process, and (5) maintaining a
complete record of all applications.
Development of this RTES requires extended knowledge including (1) artificial
intelligence, (2) software engineering, (3) citrus irrigation and fertigation management, and
(4) control engineering. Studies are needed to integrate the technology of water
management into effective control engineering. An automatic weather station and soil
moisture sensors are essential for the system. An RTES could offer a better tool for the
management of irrigation, fertigation, and cold protection. RTESs are a feasible and
necessary approach for citrus microirrigation management.


75
water depletion for Candler fine sand. One-third (33 percent) of the depletion approximately
corresponds to 11 cb, and 50 percent depletion approximately corresponds to 20 cb of soil-
water tension (Figure 6.2). Then, irrigation can be applied when soil-water potential or soil-
water depletion reaches a threshold value.
Table 6.4 Estimated soil-water tension in corresponding to soil-water depletion for
Candler fine sand.
Depletion, %
AWC %
SWC-7.5, %
SWC-9.5, %
ASWT, %
ESWT, CB
100
0
1.40
1.40
1.40
-1500
90
10
2.21
2.01
2.11
-
80
20
3.02
2.62
2.82
-
70
30
3.83
3.23
3.53
-
67
33
4.07
3.41
3.74
-38
60
40
4.64
3.84
4.24
-33
50
50
5.45
4.45
4.95
-20
40
60
6.26
5.06
5.66
-13
33
67
6.83
5.49
6.16
-11
20
80
7.88
6.28
7.08
-9
10
90
8.69
6.89
7.79
-8
0
100
9.50
7.50
8.50
-7.5
Note: AWC = available water content,
SWC-7.5 = soil-water content for field capacity at 7.5 percent by volume,
SWC-9.5 = soil-water content for field capacity at 9.5 percent by volume,
ASWC = average soil-water content of above two, and
ESWT = estimated soil-water tension.


169
13,5 Database
Databases have been created in CIMS including (1) farm information, (2) weather
data, (3) irrigation system, (4) crop, (5) crop coefficient, and (6) soil data. These data are
used for simulation to estimate the crop irrigation requirements. The data can also be used
as the user's reference to make irrigation management decisions. Each database consists of
consistent data manipulating buttons.
13,5,1 Buttons to Manipulate Database
Generic control buttons were developed to manipulate the database (Figure 13.5.1).
Functions of each of these buttons are described as follows.
Top
Prior
Next
Bottom
Add
Delete
display the first record of a database,
move to the prior record of a database,
move to the next record of a database,
display the last record of a database,
add a new record to the database,
delete the current record from a database.
The Delete button opens a dialog window to
confer the user's deletion showed in Figure
13.5.2. Click the Cancel button to avoid the
deletion or click the Proceed button to delete
the record.
Bottom
Add
ipgi
Figure 13.5.1.
Database control
button.


5
1.2 Objective of the Dissertation
Agricultural production is related to many factors including crop, soil, and biological
conditions, and management decisions. Many complex decisions must be made daily. New
techniques are needed to assist farm managers. With the complexity of modem farm
management and available computer technology, it appears that an RTES is a means to assist
farm management. The primary goal of this research was to develop a methodology using
an RTES to improve the management of citrus microirrigation systems. The specific
objectives of this dissertation were
1) To acquire expert knowledge on citrus microirrigation management.
2) To develop control routines and a control panel to turn on or off user
specified valves from a local or remote computer.
3) To develop a user-friendly RTES for citrus irrigation, fertigation, and cold
protection management.
4) To provide alternative control functions so that irrigation and fertigation can
be applied according to user defined schedules.
5) To use farm databases, crop water requirement simulation, and other
computer tools to assist the decision-making of farm managers.
6) To demonstrate the use of an RTES as an operational tool to improve
management of an irrigation system.
Acquiring expert knowledge is crucial in the development of an expert system.
Experts need to be identified to acquire their knowledge in the problem domain: citrus


72
Soil-water depletion at
any time is the amount of water
needed to irrigate the current
crop root zone to field capacity
(FC). Normally, irrigation is
applied when the depletion
exceeds MAD or when the
managed soil-water content is
less than a threshold value
(SWCo). Thus, soil-water content is maintained at a certain level (Figure 6.1). Because
each irrigation raises soil-water content from SWCo to FC, the amount of applied water is
the same for all irrigations. However, the application frequency varies between irrigation
events. On the other hand, irrigation can also be applied at a fixed interval (every day or two
days), but irrigating with different amounts of water each time.
6.6 Irrigation Scheduling Using Tensiometers
Researchers reported that tensiometers can be used effectively to schedule irrigation
by measuring soil-water potential, but proper installation and maintenance are required for
the application (Smajstrla et al., 1985a; Smajstrla and Koo, 1986; Fitzsimmons and Young,
1972; Creighton et al., 1989). Two practical issues should be resolved in using tensiometers
for citrus irrigation scheduling:
S FC
-2
3
\i\A/\/\A
m SkVCO
i
3
Irrigation Events
Time
Figure 6.1. Irrigation by threshold of soil-water
content.


35
3.5.2 Uncertainty Management
The quality of an ES is related to the knowledge base. Each piece of knowledge
acquired from experts may involve some kind of uncertainty. Bronowski (1965) stated that
in trying to formalize a rule, we look for truth, but what we find is
knowledge, and what we fail to find is certainty, (p. 32)
Uncertainty arises from a variety of sources: (1) unreliable information, (2) imprecise
descriptive languages, (3) inferences with incomplete information, and (4) a poor
combination of knowledge from different experts (Bonissone and Tong, 1985). Uncertainty
may prevent a system from making the best decision and may even cause a bad decision to
be made. The basic numeric approaches to deal with uncertainty are Bayesian probability,
Dempster-Shafer theory (Dempster, 1967; Shafer, 1976), and certainty factors. Bayesian
and Dempster-Shafer approaches were developed before expert systems became popular.
Because their practical implementation is complex and the approach requires a prodigious
amount of data that are usually not available, these approaches are not widely applied in the
development of expert systems.
Shortliffe and Buchanan (1975) developed an uncertainty approach to represent
uncertain information in MYCIN. A certainty factor (CF) is calculated from a measure of
belief (MB) and a measure of disbelief (MD):
MB MD
1 min [MB, MD]
3-1


175
13.6 Simulation
In agriculture, computer simulation is a widely used approach to estimate crop water
requirement. The CIMS uses a soil-water balance model to simulate soil-water content at
the crop root zone. Input data of the simulation come from the database including (1)
irrigation system, (2) crop and crop coefficients, (3) soil, and (4) weather data. The Penman
method was used to calculate daily evapotranspiration. The simulation model executes at
daily intervals. A prognosis of short-term irrigation schedule is given from the simulation
results.
13 61 How to Run the Simulation
Click Simulation on the CIMS main menu (Figure 13.6.1).
Figure 13.6.1. Simulation submenu.
Click Set Initial SWC to open the set initial condition dialog window (Figure
13.6.2). It is necessary to set this initial condition at first run of the simulation.
After the first run, the simulation model can automatically search its initial condition
based on the previously simulated results.


57
Let Nw denote the number of wet days in the n day period.
- E c, 5-2
ui
The possible values of the random value Nw are 0, 1, n.
Gabriel and Neumann (1962) studied a first-order Markov chain model for daily
rainfall occurrence in Tel Aviv, Israel. Their assumption was that the probability of rainfall
on any day depends only on whether the previous day was wet or dry. Such a probability
model is a Markov chain with two conditional probabilities:
P! = Pr {wet day | previous day wet}
p0 = Pr (wet day | previous day dry}
Although this model obtained satisfactory results in Tel Aviv (Gabriel and Neumann,
1962) and other regions, previous studies (Schmidt et al., 1987; Jones and Thornton, 1993)
showed that using a first-order Markov chain to estimate rainfall probability may not be
adequate for a subtropical or tropical region such as Florida. Their studies suggested that
a higher order Markov chain should be used for tropical and subtropical weather conditions.
Jones and Thornton (1993) have applied a third-order Markov chain for tropical and
subtropical regions. For a third-order Markov chain model, the probability of a rainfall
event on any given day is assumed only depending on the states of the three previous days.


87
Table 7.2 Solubility of common fertilizers in water.
Fertilizer Formula
Temperature range
(kg fertilizer per m3 water)
Cold
Lukewarm
Hot
Ammonium chloride
NH4CL
297 (0)
758 (100)
Ammonium nitrate
nh4no3
1183 (0)
1950 (20)
3440 (50)
Mono ammonium
phosphate
nh4h2po4
227 (0)
282 (20)
417 (50)
Diammonium phosphate
(NH4)2HP04
429 (0)
575 (10)
1060 (70)
Ammonium sulfate
(NH4)2so4
706 (0)
760 (20)
850 (50)
Potassium chloride
KC
280 (0)
347 (20)
430 (50)
Potassium nitrate
KNOj
133 (0)
316 (20)
860 (50)
Potassium sulfate
k2so4
69 (0)
110 (20)
170 (50)
Monopotassium
phosphate
kh2po4
- -
330 (25)
835 (90)
Dipotassium
phosphate
k2hpo4
- -
1670 (20)
- -
Calcium nitrate
Ca(N03)2
1020 (0)
3410 (25)
3760 (100)
Magnesium nitrate
Mg(N03)2
423 (18)
578 (90)
Monocalcium phosphate
Ca(H2P04)2
18 (30)
Phosphoric acid
h3po4
5480 (25)
Urea
(NH2)2CO
780 (5)
1193 (25)
Source: Hodgman, C. (ed) 1949. Handbook of Chemistry and Physics. Chemical Rubber, Cleveland, Ohio.
* Numbers in parentheses are solution temperature, *C.


188
13.8.8 Quit from CIMS
To quit from the CIMS, click
the Quit button on the main menu. A
confirmation screen appears as Figure
13.8.10. Click the OK button to exit
from CIMS or click the Cancel button
Figure 13.8.9. More about the CIMS.
to stay in CIMS.
lit i*nowtoH*9imut>t te my study amuntttee, ;
O r. s mota (CfMwmttM
Dr A a Si*ttte
Or. J W tore
O P H Jymfi
Dr O, D. Oankrtl
Of.Lrt.Mote
Special thettfcs to my rfomaJe experts:
Dr T. A Wheolon cert Of. L ft. Pareeas
Cines RCC at lake ASK* lor the vofcmfcte
advices.
Exit
Figure 13.8.10.
Screen to quit from
the CIMS.


CHAPTER 7
CITRUS COLD PROTECTION AND FERTIGATION
7.1 Introduction
Cold protection refers to methods used to prevent cold damage to the crop. This
term is typically used for (1) frost protection, (2) freeze protection, (3) frost/freeze
protection, and (4) chilling protection (Barfield et al., 1990). Cold protection is always
important to citrus production. Cold weather has caused severe economic damage in
Florida's citrus industry in January, 1985 and February, 1989. Although several cold
protection approaches are available, such as tree wraps, heaters, and wind machines,
irrigation is the primary means of cold protection in Florida citrus. Microirrigation is a
valuable tool for cold protection. Major cold protection in Florida (estimated over
100,000 acres of citrus) is accomplished with microsprinkler irrigation (Parsons et al.,
1989; Parsons and Wheaton, 1990). Experience has indicated that micro spray jet systems
are effective for cold protection (Harrison et al., 1987; Hardy, 1989; Parsons and
Wheaton, 1990), particularly for young trees. Studies of cold protection methods with
computer aided decision systems have been conducted by researchers (Holland, 1990;
Heinemann et al., 1991, 1994; Martsolf et al., 1991). Their results showed that
computerized systems could improve decision-making on cold protection. Irrigation
equipment must be specifically designed for cold protection purposes. The irrigation
77


93
8.3 The Sensor Data
8.3.1 Download the Sensor Data
To increase reliability of the sensor data, tensiometers should be properly calibrated
and installed. In this design, three tensiometers (depths 6", 12", and 24") were used at each
of two locations. EDI and ID2 are used to denote sensors at the two locations. SI, S2 and
S3 denote sensor readings at the three depths (6", 12", and 24", respectively). The sensor
installation depth can be adjusted based on field conditions without affecting the knowledge
base.
Figure 8.4 illustrates
the process of downloading
sensor data. A data logger
(CRIO) was used to acquire
sensor data. Rules were
developed to access the data
logger at a specified time
interval. The system keeps
track of the elapsed time since
the last data downloading
time. The sensor and weather
data are downloaded at the
specified time intervals. This
Figure 8.4. Process of downloading weather and soil
moisture sensor data.


Ill
conflict, fertigation is always arranged at the end of the irrigation. The fertigation should
serve as a part of irrigation whenever possible.


29
As a result, knowledge acquisition is an art. Each problem may require specific
acquisition strategies and could sometimes involve psychological issues. To reduce the
errors caused by human intervention, more efficient and reliable approaches for acquiring
knowledge are required. These should automate the elicitation process based on a
representation scheme that will completely and efficiently denote all the domain traits and
encompass all the essential knowledge.
3.4 Knowledge Representation
Knowledge representation consists of encoding real-world expert knowledge into a
format both readable and understandable by a computer. Some way to represent knowledge
is needed that allows the computer to derive new conclusions about its environment by
manipulating the representation.
In the process of knowledge representation, the primary problem is to find a kind of
format or knowledge representation language. Usually, knowledge representation is not
straightforward. First, the knowledge engineer should understand the concepts required to
solve the problem. Second, these concepts should be represented precisely and
unambiguously at all granularity levels. Third, these concepts should be easy to understand
and applicable to many systems. Ringland and Duce (1988) stated some issues that should
be raised in knowledge representation:
Is the approach expressively adequate to the domain?
Is reasoning efficient enough to allow the inference to perform in an
acceptable time?


196
Michalski, R. S., J. G. Carbonell, and T. M. Mitchell (eds.). 1986. Machine Learning,
Vol.2, Morgan Kaufmann, Los Altos, CA.
Minsky, Marvin. 1975. A framework for representing knowledge. In The Psychology of
Computer Vision, ed. P. Winston. McGraw-Hill, New York, pp. 211-217.
Moore, D. S. and G. P. McCabe. 1989. Introduction to the Practice of Statistics. W. H.
Freeman and Company, New York.
Morey, R. V., S. A. Sargent, C. D. Baird, and M. T. Talbot. 1988. A knowledge based
system for selecting precooling methods. ASAE Paper No. 88-7539, St. Joseph,
ML
Muttiah, R. S., C. N. Thai, S. E. Prussia, R. L. Shewfelt, and J. L. Jordan. 1988. An
expert system for lettuce handling at a retail store. Trans, of ASAE 31(2):622-628.
Myers, J. M. and D. S. Harrison, 1978. Drip irrigation of orange trees in humid climate.
ASAE Paper No. 78-2018, St. Joseph, MI.
Nann, S. R., A. Ray, and S. Kumara. 1991. A decision support system for real-time
monitoring and control of dynamical processes. Inter. J. of Intelligent Systems
6:739-758.
NOAA. 1952-1992. Climatological Data, Florida. U.S. Environmental Data Service,
Asheville, NC.
O'Keefe, R. M., O. Balci, and E. P. Smith. 1987. Validating expert system performance.
IEEE Expert, Winter, pp. 81-88.
Padalkar, S., G. Karsai, C. Biegl, J. Sztipanovits, K. Okuda, and N. Miyasaka. 1991.
Real-time fault diagnostics. IEEE Expert, June, pp.75-85.
Palmer, R. G. 1986. How expert systems can improve crop production. Agricultural
Engineering, September/October, pp. 28-30.
Parsons, L. R., C. Cliff, B. Summerhill, and G. Israel. 1989. Cold protection survey
major changes in a decade. Citrus Industry, November, pp. 46-48.
Parsons, L. R. and T. A. Wheaton. 1990. Irrigation for cold protection during the
December, 1989 freeze. Citrus Industry, November, pp. 12-14.


66
several researchers (Gerber et al., 1973; Reitz et al., 1977). They estimated that annual ET
for citrus is about 48 inches in Florida. Monthly mean ET rates vary from a low of 0.08
inches per day in the winter to a peak of 0.17 to 0.2 inches per day in the summer (Tucker,
1983). In Florida, it was reported that annual water use is about 47.6 inches for ridge citrus
(Koo, 1963), and 44.6 inches for flatwood citrus (Rogers et al., 1987). Citrus irrigation
requirements have been recommended by researchers based upon crop ET requirement and
effective rainfall.
Citrus irrigation requirements are different for young trees and mature trees. Young
trees are usually managed to grow as quickly as possible for early production. Moreover,
young trees are less able to resist water stress than mature trees. Therefore, adequate
irrigation is especially important for young trees.
For mature trees, irrigation management should be different for critical and non-
critical growth stages. The critical growth stage for mature trees refers to the months of leaf
expansion, bloom, fruit set, and fruit enlargement. This occurs mostly during the spring
months. Irrigation during this stage is very important to both fruit quality and yield. The
spring in Florida usually has the lowest rainfall, thus the greatest moisture stress occurs. A
sufficient amount of water is essential for mature trees during the critical growth stage.
Sound irrigation practices should be emphasized during this critical state (Tucker, 1983).
The remaining months are considered to be a non-critical growth stage. Irrigation
application during the non-critical stage should be considered only when tree stress is
imminent.


150
Test case 40
(weather-data ws 30 rh 70 rain 2.0 airtemp 60)
Test case 41
(weather-data ws 30 rh 70 rain 0.0 airtemp 35)
Test case 42
(weather-data ws 30 rh 70 rain 0.0 airtemp 37)
protection.
Test case 43
(weather-data ws 30 rh 70 rain 0.0 airtemp 38)
Test case 44
(FTSCH1 MONDAY START-TIME
3 10 END-TIME 4 10)
Test case 45
(FTSCH2 1994 11 23 START-TIME
3 20 END-TIME 3 30)
Not turn on cold protection
Turn on cold protection
Warning message for cold
Not turn on cold protection
Fertigation starts at 3:10 a. m.
and ends at 4:10 a m. every
Monday.
Fertigation starts at 3:20 a.m.
and ends at 3:30 a.m. on
November 11, 1994.


8.5 Critical sensor readings (Criteria II) to start an irrigation 105
9.1 Data input of QMS 121
9.2 Accumulated citrus net irrigation requirements and number of
irrigations for 22 years in central Florida 127
IX


117
Since the control action relies only on user-defined schedules, there is no need to use
soil moisture sensors or a weather station in this mode of operation. However, because
irrigation and fertigation fully rely on user-defined schedules, the user must be very
knowledgeable of crop water management and nutrient requirements to specify the schedules
properly.
9.2.4 Database
The Database
module contains several
databases: (1) farm, (2)
weather, (3) irrigation
system, (4) crop, and
(5) soil. Figure 9.3 is
an example of a
database screen. Each
of the database files has
a consistent data entry screen and manipulating functions, such as add, delete, and find, etc.
These databases are used for irrigation management and simulation purposes.
9.2.5 Simulation
Computer simulation is a widely used approach for irrigation scheduling. The
simulation module, as discussed in Chapter 6, contains a water balance model that is used
to simulate soil-water content in the crop root zone. Because weather data can be obtained
from the field weather station, simulation in real time is feasible. Furthermore, a short-term
Figure 9.3. Irrigation system database.


131
radio link. Although this system was designed for citrus microirrigation management, it can
be modified to extend its use to other crops by modifying its knowledge base.
Development of the CIMS involved knowledge of expert system development,
software engineering, control techniques, and knowledge of citrus irrigation management.
This study integrated the knowledge into a single system. Because the development of an
expert system is an ad hoc approach, one of the most arduous tasks in this study was
developing the knowledge base. The complexity of knowledge acquisition was due to some
practical reasons:
Implicitness of the knowledge. The reasoning of experts could not be always
explicitly expressed in a form of heuristic rules.
Time availability of the experts.
Although the RTES is the main subject of this study and it has the potential of
providing a good tool for citrus microirrigation management, the cost of this system can be
high, mainly because of the cost of the software development and hardware requirements.
Therefore, the design philosophy of CIMS is to provide variety of irrigation management
options with different hardware requirements to satisfy the user's need. Thus, besides the
RTES, CIMS provides irrigation management options of (1) user defined irrigation and
fertigation schedules, (2) an irrigation control panel, (3) simulated crop water requirement,
and (4) a calendar for irrigation scheduling. These control and management tools can be
used without knowledge base development. Unlike the dynamic system of the RTES, the
conventional control scheme allows the users to have full control of the schedules of
irrigation and fertigation. Options (3) and (4) do not require any additional hardware to


113
requirements, but also associated with factors such as ease of use and maintenance. Figure
9.1 shows the main modules of CIMS, which can be categorized as (1) expert system, (2)
control panel, (3) scheduling, (4) database, (5) simulation, (6) tools, (7) help, and (8)
graphical user interface (GUI). Specification of each module is described as following.
Figure 9.1. Program modules of CIMS.


143
B.2. Some Critical Data for the Reasoning Process
Young trees: 12 to 15 cb with constraints and 15 cb without constraints.
Mature trees during critical growth stage: 15 to 20 cb with constraints and 20 cb
without constraints.
Mature trees during non critical growth stage: 25 to 30 cb with constraints and 30
cb without constraints.
Cold protection critical temperature: 36F.
Threshold Value of Confidence Factor is 80 percent.
B.3, Test Cases and Results
Test Cases Results
Test case 1
(TREE-STATUS YOUNG)
(sensor-reading idl 6 13 12 12 24 16)
(sensor-reading id2 6 9 12 12 24 14)
(time-constrain no)
(weather-constrain no)
Reject sensor reading at
ID 1-3 (CF = 0.75).
Turn on by sensor at ID2-3.
Test case 2
(TREE-STATUS YOUNG)
(sensor-reading idl 6 9 12 13 24 9)
(sensor-reading id2 6 9 12 9 24 9)
(time-constrain no)
(weather-constrain no)
Data validation OK.
Not turn on.
Test case 3
(TREE-STATUS YOUNG)
(sensor-reading idl 6 9 12 9 24 13)
(sensor-reading id2 6 9 12 9 24 9)
Data validation OK
Not turn on


168
Figure 13.4.8. Irrigation and fertigation dialog screen.


193
Hayes-Roth, F. 1985. Rule-based systems. Communications of the ACM 28(9):921-
932.
Hearn, C. J. 1993. The influence of cultivar and high nitrogen and potassium fertilization
on fruit quality traits of young orange trees. Proc. Fla. State Hort. Soc. 106:8-12.
Heinemann, P. H., C. T. Morrow, R. M. Crassweller, and J. D. Martsolf. 1991. A
decision support system for the protection of crop from frost. Hortscience
26(6):732.
Heinemann, P. H., C. T. Morrow, J. D. Martsolf, R. M. Crassweller, and. K. B. Perry.
1994. Decision support system for the protection of crops from frost. Proc. of the
5th Int. Con. on Computers in Agriculture, Orlando, FL, pp.375-380.
Hodgman, C. (ed) 1949. Handbook of Chemistry and Physics. Chemical Rubber,
Cleveland, Ohio.
Holland, M. 1990. Elevation of microsprinklers for cold protection. Citrus Industry
Magazine 71(8): 12-15.
Howell, T. A., D. W. meek, C. J. Phene, K. R. Davis, and R. L. McCormick. 1984.
Automated weather data collection for research on irrigation scheduling. Trans, of
ASAE 47:386-391, 396.
Ingrand, F. F., M. P. Georgeff, and A. S. Rao. 1992. An architecture for real-time
reasoning and systems control. IEEE Expert, December, pp. 34-44.
Jackson, L. K. 1991. Citrus Growing in Florida, 3rd. ed. Univ. Press of Florida,
Gainesville, FL.
Jacobson, B. K., J. W. Jones, and P. H. Jones. 1987. Tomato greenhouse environment
controller: real-time expert system supervisor. ASAE Paper No 87-5022, St.
Joseph, MI.
Jacobson, B. K., P. H. Jones, J. W. Jones, and J. A. Paramore. 1989. Real-time
greenhouse monitoring and control with an expert system. Computers and
Electronics in Agriculture 34:273-285.
Jensen, M. E., C. N. Robb, and C. E. Franzoy. 1970. Scheduling irrigation using
climate-crop-soil data. J. of Irrigation and Drainage Division, ASCE 3:26-35.
Jensen, M. E., J. L. Wright, and B. J. Pratt. 1971. Estimating soil moisture depletion
from climate, crop and soil data. Trans, of the ASAE 14(5):954-959.


8
2.2 Soil Moisture Sensors
The ability to measure soil moisture in-situ is important for irrigation management.
Irrigation water can be saved by using soil moisture sensors (Zazueta et al., 1993).
However, the choice of soil moisture sensor is crucial to the success of irrigation control and
management. Usually, "the most intractable barrier to the full implementation of automatic
process control is the lack of adequate on-line sensors (p. v)" (Carr-Brion, 1986). A poor
choice of a sensor at the design stage is commonly caused by lack of adequate appreciation
of the limitations of the type of sensor used or by lack of knowledge of what is available.
Many literature reviews can be found on soil-moisture measurement by a variety of
techniques (Taylor, 1955; Schmugge et al., 1980; McKim et al., 1980; Erbach, 1983;
Wheeler and Duncan, 1984; Gardner, 1986; Stafford, 1988; Zazueta and Xin, 1992). The
techniques commonly used in soil moisture sensors include (1) electromagnetic, (2) nuclear,
(3) remote sensing, (4) hygrometric, (5) tensiometric, (6) optical, and (7) time domain
reflectometry (TDR). Not all these soil moisture sensing techniques are suitable for
automation. The sensor must have the capability of interfacing with a computer or other
electronic devices. Sensor cost is another major concern for agricultural applications. Some
soil moisture sensors, such as TDR and neutron probes, can achieve high accuracy (Topp
and Davis, 1985; Simpson and Meyer, 1987), but costs of the devices are also high.
Tensiometers are relatively inexpensive and are easy to use.
Tensiometers measure the matric potential (capillary tension) directly, which is
related to the energy required for plants to extract water from the soil. Tensiometers are the


CHAPTER 5
PROBABILITY OF RAINFALL
5 1 Introduction
Many agricultural operations and activities are affected by rainfall frequency and
amount. To improve water use, an irrigator must consider probable occurrence of rainfall.
Ideally, irrigation should be managed to maximize effective rainfall while satisfying crop
water demands. Although rainfall cannot be predicted with certainty, estimated probability
of rainfall is useful for irrigation management.
Rainfall probability is commonly predicated through weather forecasting according
to meteorological observations. Stochastic modeling is another available approach to
generate daily weather data from the use of observed weather data. Estimation of daily or
seasonal rainfall sequences can be obtained by examining past precipitation records. The
rainfall sequence can be estimated by using rainfall occurrence models (Schmidt, 1992): (1)
alternating wet and dry interval models, (2) wet and dry day models (Markov-chain models),
and (3) point process models. Such prediction is based upon the assumptions that sequences
will tend to be the same in the future as they were during the period of record.
Markov chain models are widely used because of their simplicity, flexibility,
seasonality, and number of states. Markov chain probability models for daily precipitation
occurrences have been studied extensively. This approach has been implemented with
55


199
Rolston, D. E., R. J. Miller, and H. Schulbach. 1986. Management principles of
fertilization, in Trickle Irrigation for Crop Production: Design, Operation and
Management, ed. F. S. Nakayama and D. A. Bucks. Elsevier, Tokyo, pp. 317-
344.
Safley, J. M, Jr, L. A. Fribour, J. V. Vaiksoras, and R. H. Strand. 1974. Probability of
sequences of wet and dry days for Tennessee. NOAA Technical Memorandum,
NOAA/Environmental Data Service, Washington, DC.
Schalkoff, R. J. 1990. Artificial Intelligence: An Engineering Approach. McGraw-Hill
Publishing Company, New York.
Schmidt, G. M. 1992. Stochastic rainfall modeling and long-term climatic variability:
model parameter estimation and model evaluation. Doctoral dissertation, Agr. Eng.
Dept., Univ. of Florida, Gainesville, FL.
Schmidt, G. M., A. G. Smajstrla, and F. S. Zazueta. 1987. Estimating the frequency and
duration of wet and dry periods in a humid subtropical region. Soil Crop Sci. Soc.
Fla. 47:83-88.
Schmugge, T. J., T. J. Jackson, and H. L. McKim. 1980. Survey of methods for soil
moisture determination. Water Resources Research 16:961-979.
Schnelle, K. D. and R. S. H. Mah. 1992. A real-time expert system for quality control.
IEEE Expert, October, pp.36-42.
SCS. 1982. Florida Irrigation Guide. USDA, Soil Cons. Serv., Gainesville, FL.
Shafer, G. 1976. A Mathematical Theory of Evidence, Princeton Univ. Press, Princeton,
NJ.
Sharpies, R. A., D. E. Rolston, J. W. Biggar, and H. I. Nightingale. 1985. Evaporation
and soil water balances of young trickle-irrigated almond trees. In Drip/trickle
Irrigation in Action, Proc. of Third Int. Drip/Trickle Irrigation Congress, ASAE,
St. Joseph, MI, vol. II, pp. 792-797.
Shayya, W. H., V. F. Bralts, and T. R. Olmsted. 1990. General irrigation scheduling
package for microcomputers. Computers and Electronics in Agriculture, 5:197-
212.
Shortliffe, E. H. and B. G. Buchanan. 1975. A model of inexact reasoning in medicine.
Mathematical Biosciences 23:351-379.


82
Nitrogen needed = 0.4 lbs. N/box x 500 boxes = 200 lbs.
Potash needed = 0.4 lbs. K20/box X 500 boxes = 200 lbs.
Other nutrients applied on basis of need.
Table 7.1 shows the nitrogen requirement for orange and grapefruit under normal
conditions (Koo et al., 1984). Nitrogen requirement for young citrus trees is approximately
0.16 pound per tree per year in Florida (Fisher, 1990). Although some general
recommendations have been given for citrus fertigation, with current knowledge it is
difficult to develop a general and precise fertigation strategy. There are many different
opinions about rates, concentrations, and times at which fertigation should be applied.
Research is needed to determine the proper amount of fertilizer and application frequency
when microirrigation is used.
For this application, a set of fertigation schedules is created in the knowledge base.
These schedules were determined by interviewing citrus fertigation experts at the University
of Florida. The knowledge base continuously checks the current time and the fertigation
schedules. Fertigation is applied when the computer time matches the predefined fertigation
schedules. The user should modify the schedules based upon expert recommendation only.
Chemical injection rate can be computed by the following equation (Keller and Bliesner,
1990):
Fr
C'fr Ta
7-2


120
Figure 9.4. Data flow of CIMS.
Control messages can be passed by the RTES and scheduling modules to the valve
control module or the control procedures. Control actions take place when the control
module receives a message from the RTES or scheduling module. All control events
implemented by the RTES are stored in the application log files.
9,3,2 Data Requirements of the Simulation Module
Data required for the water budget simulation are weather, crop, soil, and irrigation
system data (Table 9.1). The simulated results are soil-water content in the crop root zone,
and a one-week prognosis of irrigation requirements. The simulation results can be used
as a reference for the user in irrigation management.


43
be higher than with other developing tools because the development needs to be started from
scratch in many cases.
The GPRL are the languages which are especially written for expert system
development, such as OPS5 and UNITS. These languages have relatively higher flexibility
in comparison with shells and development environments. Expert system development
environments (such as KEE, LEVEL5, and ART) are complete development environments
which provide sophisticated features such as knowledge representation, debugging routines,
and development facilities. Although these development environments provide many
features, the use of these software packages may require a substantial amount of time to
learn the package to take the advantage of their features. Also, the cost of the packages is
high.
4.5.1 Expert System Shells
The introduction of expert system shells has made expert system development much
faster and attracted more new developers. A range of expert system shells have been
developed to match the cost and application requirements. Some of these shells are CLIPS,
EXSYS, and VP-EXPERT. The shells have the advantages that they are low cost, easy
to learn, and save development time, but their flexibility and capability are the major
constraints for development of expert systems.
Choosing the right tool for implementing a particular application is difficult because
there is almost no absolutely one 'right' choice. One needs to ponder the advantages of any
selection against its limitations to determine the most suitable tool for a particular problem.
If a shell exists and satisfies the application requirements, the shell may be the better choice


115
a decision to irrigate is made after the reasoning process. Data uncertainty management as
discussed in Chapters 3 and 8 is conducted during the reasoning process to handle the
missing and unreliable sensor data. A warning message is displayed when a sensor is
identified to have failed.
Valve control
After a decision is made to start the irrigation system, the external hardware (pumps
and solenoid valves) must be activated. The valve control module consists of three control
procedures: (1) irrigation, (2) fertigation, and (3) cold protection (Figure 9.1). These three
procedures are conventional programs used to turn on or off pumps and electronic valves.
The valve control module is shared by the: (1) RTES, (2) control panel, and (3) scheduling
modules.
Application log
Because the RTES can be operated continuously without a human presence, the user
may not know what control events have been implemented. Application log files are created
to store the detailed information of each control event that was conducted by the system.
From the user's point of view, this function allows past events to be viewed and necessary
reports to be created. For the developer, this log file can be used as a debugging tool.
9.2.2 Control Panel
Besides the RTES, the control panel provides an additional means to control the
irrigation system. This module uses the control screen (Figure 9.2 ) so that the user can turn
control valves on or off by simply clicking a button. The control panel also displays the on
or off status of irrigation, fertigation, and cold protection. Each of these control functions


Copyright 1995
by
Jiannong Xin


11
Cassell and Klute (1986) studied soil effects on tensiometers. They found that the
use of tensiometers for irrigation scheduling was more successful in coarse textured soils
than it was in fine textured soils. This is because a greater percentage of the water available
to a plant is retained by coarse textured soils at suctions less than 0.8 bar than is the case for
fine-textured soils. Tensiometers only operate from zero to about 0.8 bar. Tensiometers are
an effective tool to assist irrigation decision-making, but soil water potential must be
maintained within their operational range.
2.3.2 Computer Simulation
As computer systems have become widespread, simulation-based approaches have
been developed (Lembke and Jones, 1972; Swaney et al., 1983; Villalobos and Fereres,
1989; Rogers and Elliott, 1989). Soil water balance and crop growth simulation models are
two common approaches.
Soil water balance
This method applies the principle of continuity to the root zone. It describes soil
moisture change in the root zone over time. Using this approach to manage irrigation
involves estimating the amount of water in the crop root zone. To maintain the soil water
content in the crop root zone within a desired range, irrigation should be applied to satisfy
evapotranspiration (ET) demands.
Jensen et al. (1970) reported a scheduling method based upon soil-crop-climate data.
Soil water balance simulation models were developed by researchers (Jensen et al., 1971;
Zazueta et al., 1986; Smajstrla and Zazueta, 1987; Anderson et al., 1978; Cahoon et al.,


2
the energy used in irrigation. Currently, close to 16,000 billion BTUs (the energy equivalent
of approximately 128 million gallons of gasoline) (Zazueta et al., 1994) are needed by about
750,000 acres of citrus production. If 32.2 percent of this energy is used in irrigation, about
5,152 billion BTUs are consumed only for irrigation. A conservative 10 percent reduction
in energy use by using better control systems would result in a savings of 515 billion BTUs.
Fertilizer and chemical applications are common in agricultural practice. Because
water is used to convey many of the chemicals through microirrigation systems, the
efficiencies of fertigation and chemigation are directly related to the water application
efficiency. Furthermore, improper chemical applications may result in adverse
environmental impacts. Ground and surface water can be contaminated if applied chemicals
are transported out of the crop root zone due to inadequate management.
It is generally accepted that improved management techniques are necessary to
increase water use efficiency, particularly in the scheduling and application of irrigation.
Current irrigation management practice attempts to satisfy crop water demands and relies,
for the most part, on manual operation or timers for control of the irrigation system. Many
irrigation schedules rely on calendars, evapotranspiration (ET) estimation, and the grower's
experience. Water can be wasted due to poor irrigation management, and irrigation may
even be applied during rainfall when a preset timer is used to control the system (Xin et al.,
1993).
Computer-based irrigation scheduling has received much attention. Simulation
models have been developed for various irrigation strategies and to assist irrigation
scheduling using historical weather records (Lembke and Jones, 1972; Swaney et al., 1983;


153
How to run CIMS
To launch CIMS, start Microsoft Windows and double click CIMS icon.
For information about how to use Windows, refer to Microsoft Windows' Reference. The
procedures to run each of the main modules of CIMS are described in the subsequent
sections.


46
Pressure transducer
Tensiometers continuously
measure the soil-water tension. The
tension must be converted into an
electrical signal to output to a
computer. Tensiometers with micro
pressure transducers have been
constructed at the Soil and Water
Hydraulics Laboratory, Agricultural
and Biological Engineering
Department, University of Florida
Figure 4.2. Pressure transducer (Model
141PC) from Micro Switch.


63
forecasting data, such as probability of rainfall for the next few days, those data can be input
to the system to assist the irrigation decision-making process. Because computer networks
are widely used nowadays, the system can link to a network to retrieve weather forecasts and
historical weather data. The computer network at the Institute of Agricultural and Food
Science (IF AS), University of Florida, supports a weather forecasting database.


95
Checking the ranee of sensor readings
A common irrigation
management strategy is to
maintain soil-water content in a
certain range. Irrigation is
applied when sensor outputs
indicate that the maximum
allowable depletion has been
exceeded. A possible sensor
failure may have occurred when
its output is out of the normal
range. This possible failure is
denoted as Failure A. Figure
8.5 shows the propagation of
CF values. When the sensor output is out of the predefined range, a lower value of CF' is
assigned to the sensor. The combined new confidence value (CF") is calculated through CF
propagation from equation 3-1. Thus, confidence in the sensor reading is reduced because
of the lower CF value. However, this reduction of the CF value does not definitely indicate
sensor failure. Only the final propagation of CF determines whether a sensor has failed at
Figure 8.5. CF propagation by checking range of
sensor data.
a certain confidence level.


13.7.9 Help screen of irrigation duration 183
13.7.10 A dummy field layout map 184
13.8.1 Submenu of the Help 185
13.8.2 CIMS help screen 185
13.8.3 Calculator 186
13.8.4 Calendar and diary screen 186
13.8.5 Clock 187
13.8.6 Text editor 187
13.8.7 Puzzle 187
13.8.8 About the CIMS 187
13.8.9 More about the CIMS 188
13.8.8 Screen to quit from the CIMS 188
xiv


73
Depth and number of tensiometers that should be installed, and
Soil water tension at which irrigation should be initiated.
6.6.1 Tensiometer Installation Depth
Because tensiometers measure soil-water potential in only a small volume of soil
immediately surrounding the ceramic cup, installation sites should be representative of the
surrounding field conditions and water content in the effective root zone. Citrus root zone
moisture extraction depths in unsaturated soils range from 3 to 5 feet and the minimum root
zone moisture extraction depth required is 1.5 to 2 feet (SCS, 1982). The maximum
effective rooting depths for citrus are 3.0 to 5.4 feet (Martin et al., 1990). In Florida, Tucker
(1983) reported that citrus rooting depths extend to 5 feet for well-drained sandy soils;
groves on flatwood soils rarely exceeded 2 feet in rooting depth. For citrus under
microirrigation, the effective root zone should be defined as the upper 1.5 to 3 feet of the
root zone for ridge citrus; and 1 to 1.5 feet for flatwoods citrus (Smajstrla et al., 1987).
How many tensiometers should be installed in the field is a compromise between cost
and accuracy. One set of tensiometers, in general, is desirable for every five acres (Smajstrla
et al., 1985b). At least two tensiometers should be used per location in order to check soil-
water depletion in the effective root zone (Smajstrla et al., 1986). One tensiometer should
be located near the soil surface (6 to 12 inches from soil surface) where most of the roots are
located. The second one should be located near the bottom of the effective zone (24 to 36
inches from soil surface) that will be irrigated. Because the upper portion of the effective
root zone contains the most roots actively involved in water uptake, it is important to


23
time can be slow or is not considered a major factor. This scenario is different from the
environment where real-time expert systems (RTES) are used. In RTESs, data change
rapidly and the input data are often collected automatically. RTESs typically need to
respond to changing task environments, timely handling of data, and execution of diverse
functions. This may involve an asynchronous flow of events and dynamically changing
requirements with limitations on time, hardware, and other factors. Figure 3.2 shows
additional components that may be required for an RTES. Sensors may be used to provide
facts to the knowledge base, and external hardware can be controlled through an ES. Thus,
the control process can be accomplished through conventional logic and procedures.
Applications of RTESs in different areas have been reported by many researchers (Wright
et al., 1986; Laffey et al., 1988; Nann et al., 1991; Ingrand et al., 1992).
Figure 3.2. Major components of an RTES.


89
data and expert knowledge on the subject. Figure 8.1 shows the major inputs and outcomes
of the expert system. Inputs to the system consist of soil-water content, time of day, weather
data, crop information, fertigation requirements, and Markov chain rainfall probabilities.
Soil moisture sensors are used to monitor the soil-water potential in the crop root zone. An
automated weather station provides wind speed, air temperature, solar radiation, relative
humidity, and rainfall data. With the continuous monitoring of soil-water potential and
weather data, the reasoning process can be conducted to make decisions on citrus irrigation,
fertigation, and cold protection. The decision-making process may involve (1) irrigation
strategy, (2) uncertainty management of the sensor data, and (3) application of irrigation,
fertigation, and cold protection. Figure 8.2 shows the decision flow diagram of the system.
8.2 The Process of Control and Reasoning
In development of the real-time expert system (RTES), two essential tasks are (1)
to control external devices, and (2) to represent the expert knowledge required to realize
such control. The system provides the following control functions:
Monitoring sensor and weather data at a given time interval,
Analyzing sensor data to identify possible sensor failure, and
Controlling electronic valves and pumps for irrigation, fertigation, and cold
protection applications.
These control actions are conventional control problems. Thus, conventional
software programs were developed for each of the control tasks. The control tasks can


CHAPTER 2
REVIEW OF THE LITERATURE
2.1 Citrus Irrigation in Florida
Citrus is one of the major crops in Florida. The total acreage of citrus was 853,742
acres in 1994 (Florida Agricultural Statistics, 1994). The citrus industry is a significant
contributor to the economy of Florida. Its annual economic impact on the state's economy
has been estimated at billions of dollars.
Although the average yearly rainfall in Florida varies from 50 to 62 inches, irrigation
is required to achieve maximum production and improve the quality of citrus fruit (Koo,
1963; Tucker, 1983). Citrus irrigation systems are also used for cold protection purposes.
Microirrigation systems are common today in Florida for citrus irrigation. Of Florida's
1,855,390 irrigated acres, 19 percent is citrus (Smajstrla et al., 1995). Billions of gallons of
water are required for the industry each year. Thus, even modest increases in water use
efficiency will result in substantial water savings.
In Florida, microirrigation is the preferred method for citrus irrigation.
Microirrigation is an efficient and convenient means of supplying water directly to a crop
root zone. It provides an effective means for utilizing small continuous streams of water for
irrigation. Furthermore, microirrigation systems more easily realize computerized control
than do other types of irrigation systems.
7


Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy
A REAL-TIME EXPERT SYSTEM
FOR CITRUS MICROIRRIGATION MANAGEMENT
By
Jiannong Xin
August, 1995
Chairman: Dr. Fedro S. Zazueta
Major Department: Agricultural and Biological Engineering
Elaborate techniques are commonplace in modern farm management and
microirrigation scheduling for citrus. Water management practices involve complex
decisions and daily operations that are affected by water and nutrient requirements of the
trees, temporal distribution of rainfall, and extreme weather conditions. A computer-based
system (CIMS) was developed using a real-time expert system (RTES) and conventional
control techniques to assist citrus microirrigation, cold protection, and fertigation
management. The system integrates water management technologies into an effective
control technique that can be used as a tool by farm managers. CIMS combines the RTES,
conventional control, and irrigation management tools into a single system to help the
decision-making of irrigators. On-site soil moisture sensors and an automated weather
station provide data to the system. CIMS activates or deactivates the control devices of an
xv


17
Although many expert systems have been developed for agricultural applications,
these systems were successful to some extent from a purely pedagogical viewpoint; but very
few of the systems are considered to be successful from a commercial viewpoint (Jones,
1989a). RTESs appear to have the potential to be successful because of their well-defined
domains. However, developing such systems can be extremely difficult due to factors such
as critical timing, knowledge acquisition, temporal reasoning, and uncertainty management.
2,6 Summary
Modem farm management and irrigation scheduling are complex tasks. There are
many factors that need to be considered to achieve successful farm management. There is
a need for more research in the field of farm management, which includes (1) irrigation
management, (2) chemigation, (3) maintaining water quality, (4) pest control, (5)
environment impact, (6) labor and energy conservation, and (7) cold protection. The
"optimal" management of an agricultural farm involves complex daily operation and
management decisions because of the temporal distribution of rainfall and extreme
microclimate.
One way of dealing with the problems is by the introduction of expert systems
integrated with control engineering techniques into irrigation management. With an RTES,
a computer can be used to implement these tasks: (1) decision-making on irrigation,
fertigation, and cold protection, (2) monitoring the performance of the irrigation system, (3)
adjusting irrigation applications as climatic or other conditions change during the irrigation,


181
Penman method
Select the Penman radio
button and click the Continue
button. A screen to calculate
ET shows in Figure 13.7.4.
Enter the data on the
screen and click the Calculate
button to calculate ET. Click
the Exit button to quit from the
screen.
Penman Reference ET Calculation
Date
Solar radiation (kW/m*)
Maximum air temperature (T)
Minmum air temperature (*F)
Wind speed (mph)
Albedo
08/22/94
Figure 13.7.4. Penman ET screen.
Blanev-Criddle method
Select the Blaney-Criddle radio button and click the Continue button. A
calculation screen appears as
Figure 13.7.5.
Enter the required data to
the screen and click the Calculate
button to display the result. Click
the Exit button to quit from the
screen.
Blaney-Criddle Method
Coefficient for the Blaney-Criddle method (M)
Percent of annual daylight hours in the month
Mean monthly temperature (T)
mmm
Figure 13.7.5. Blaney-Criddle ET screen.


80
pipe lines can be frozen at this temperature. A survey report (Ferguson et al., 1989)
indicated that 45 percent of the growers initiated irrigation for cold protection at 33-35 F
and turned it off at 35-39 F (43 percent). The Institute of Food and Agricultural Science
(EFAS) at the University of Florida recommends initiating cold protection when air
temperature reaches 36F and finishing when air temperature reaches 36 to 40 F. Thus,
air temperature at 36F was used as the critical application temperature in this system.
7.2.3 Water Application Rate
The water that is most effective for cold protection is that which covers the foliage,
and not the ice formed on the ground (Harrison et al., 1987). Uniformity of application is
very important to cold protection. Thus, uniformity design criteria of the irrigation system
must be strictly met. The application rate must be high enough to provide sufficient energy
to the system and guarantee plant coverage. A previous study (Parsons et al., 1989) showed
33.3 gal/acre/min (12 gal/hr/tree) was recommended for frost protection. Young trees may
potentially use less water. Furthermore, in Florida, emitters should be installed properly.
Parsons and Wheaton (1990) reported that emitters placed north or northwest (upwind) of
the tree provided better cold protection.
7,3Fertigation
Microirrigation offers the opportunity for precise application of fertilizer to the soil.
Fertigation is the addition of soluble nutrients or agricultural chemicals through irrigation
systems to crops. Fertilizer application through irrigation systems is desirable because of
labor and energy savings, flexibility in timing of application, and easy and precise control


194
Jones, J. W. 1985. Using expert systems in agricultural models. Agricultural Engineering
66(7) :21-23.
Jones, J. W. 1993. Decision support systems for agricultural development. In Systems
Approaches for Agricultural Development, ed. F. W. T. Penning de Vries et al.,
Kluwer Academic Publishers, Netherlands, pp. 459-471.
Jones, J. W., R. F. Colwick, and E. D. Threadgill. 1972. A simulated environmental
model of temperature, evaporation, rainfall and soil moisture. Trans, of the ASAE
15(2):366-372.
Jones, J. W. and J. T. Ritchie. 1990. Crop growth models. In Management of Farm
Irrigation Systems, ed. G. J. Hoffman, T. A. Howell and K. H. Solomon. ASAE
Monograph, St. Joseph, MI, pp. 63-85.
Jones, P. 1989a. Agricultural applications of expert systems concepts. Agricultural
Systems 31:3-18.
Jones, P. 1989b. Knowledge acquisition. Knowledge Engineering in Agriculture. ASAE,
St. Joesph, MI, pp.33-46.
Jones, P. and J. Haldeman. 1986. Management of a crop research facility with a
microcomputer-based expert system, Trans, of ASAE 29:235-242.
Jones, P. G. and P. K. Thornton. 1993. A rainfall generator for agricultural applications
in the tropics. Agricultural and Forest Meteorology 63:1-19.
Kalkar, S. and P. R. Goodrich. 1986. An expert system for animal waste management.
ASAE Paper No. 86-4540, St. Joseph, MI.
Keller, J. and R. D. Bliesner. 1990. Sprinkle and Trickle Irrigation. Van Nostrand
Reinhold, New York.
Khanjani, M. J. and J. R. Busch. 1982. Optimal irrigation water use from probability and
cost-benefit analyses. Trans, of the ASAE 42:961-968.
Kline, D. E., D. B. Bender, and B. A. McCarl. 1987. Farm-level machinery management
using intelligent decision support systems. ASAE Paper No. 87-1046, St. Joseph,
MI.
Koo, R.C.J. 1963. Effects of frequency of irrigations on yield of oranges and grapefruit.
Proc. Fla. State Hort. Soc. 76:1-5.


126
probability of rainfall. This simulation model only partially emulated the decision process
of the RTES. The actual system uses soil-water potential measurements in the irrigation
decision-making. Because accurate soil data were not available for the test site, irrigation
was simulated when the soil-water content in the crop root zone is depleted to the
Management Allowed Depletion (MAD) level. The following two assumptions were made
for the simulation:
(1) The irrigation system is properly managed, and the soil-water potentials
measured by tensiometers reflect the MAD listed in Table 6.2.
(2) Because the simulation is conducted at a daily interval, rainfall and irrigation
are assumed not to occur concurrently.
The simulation model applies to both full and deficit irrigation. Full irrigation is
always applied during critical growth stages. Deficit irrigation is only applied to the non-
critical growth stage if necessary. The simulation was conducted for mature trees with root
depth of 2.5 feet, 50 percent shade coverage, and sandy soil with field capacity 0.065 and
permanent wilting point 0.015.
Table 9.2 shows the simulated results of accumulated net irrigation requirements
(NIR) and number of irrigations for 22 years in central Florida. NIR and number of
irrigations were simulated for full irrigation, deficit irrigation, and irrigation affected by the
Markov chain probability of rainfall, respectively. To study the effects of Markov chain
probability of rainfall, the simulation was conducted with different rainfall probabilities
(Cases 1, 2 and 3). For these three cases, irrigation was applied to 80 percent (NIR-80) or
70 percent (NIR-70) of the field capacity when the soil-water content was greater than MAD


CHAPTER 9 SYSTEM IMPLEMENTATION AND TESTS 112
9.1 Function Requirements of CEMS 112
9.2 Module Design of CEMS 112
9.2.1 Expert System Module 114
9.2.2 Control Panel 115
9.2.3 Scheduling 116
9.2.4 Database 117
9.2.5 Simulation 117
9.2.6 Tools 118
9.2.7 Help 118
9.2.8 User Interface 119
9.3 Data and Message Passing of QMS 119
9.3.1 Data Flow of the RTES Module 119
9.3.2 Data Requirements of the Simulation Module 120
9.3.3 Data Requirements of the Scheduling Module 121
9.4 Maintenance of CEMS 121
9.5 System Tests of CEMS 123
9.5.1 Predictive Tests 124
9.5.2 Field Tests 124
9.5.3 Simulated Crop Water Use 125
CHAPTER 10 SUMMARY AND CONCLUSION 130
APPENDIX A SAMPLE SENSOR DATA 135
APPENDIX B SAMPLE TEST CASES OF KNOWLEDGE BASE 142
APPENDIX C USER'S GUIDE OF CIMS 151
LIST OF REFERENCES 189
BIOGRAPHICAL SKETCH 204
vii


42
application, many inputs are acquired automatically from sensors. The decision-making
progress is from initial facts (data from the sensors), to intermediate facts, and finally to a
conclusion. Thus, a development shell with forward reasoning should be selected.
4.4.2 System Performance Requirements
As an expert system which operates in the real-time domain, the system imports
initial data (facts) from sensors and the sensor data are varied with time. The system must
respond to this variation to realize real-time performance. In particular, the system should
achieve the following performance:
Must be operated within a fixed time constraint,
Should have the ability to react to the changing external environment and
must be operated continuously and as information is updated,
Must deal with incomplete and faulty input data from external devices, and
Must allow procedure calls to other systems to bring back the necessary
information for reasoning.
4.5 Development Tools
The software tools available for development of expert systems can be categorized
into four classes: General-Purpose Programming Language (GPPL), General-Purpose
Representation Language (GPRL), expert-system building frameworks (shells), and expert
system development environments (Collins et al., 1990).
The GPPL includes computer languages such as C, Lisp, and Prolog. The GPPL
languages have high programming flexibility, but the development time and the cost may


119
9,2.8 User Interface
Because the user's ability to learn an interface is crucial to software acceptance, the
user interface is a vital factor to ensure success of the system. A graphical user interface
(GUI) is commonly used for the Windows environment in software development. This is
because a GUI has advantage of window, mice, and menu to create a more user friendly
interface. The Windows environment is the current trend in software development.
Therefore, the system uses GUI developed under the Microsoft Windows environment.
9.3 Data and Message Passing of CIMS
Because CIMS is a dynamic system, the sensor data and time are decisive parameters
to the system. Defining data objects and data flows among modules is important for the
system development. Figure 9.4 shows the data flow diagram among the modules.
9,3,1 Data Flow of the RTES Module
The RTES requires input data from soil moisture sensors and a weather station.
Real-time data from tensiometers and the weather station are collected at a given time
interval. Initial facts, such as initial valve on or off status and tree status, should be provided
to the system during the system setup.
The simulation module requires soil, crop, and irrigation system data. The weather
data are also used in the crop water requirement simulation. The scheduling module requires
user-defined irrigation and fertigation schedules. Irrigation and fertigation are controlled
based on the user-defined schedules in the scheduling module.


179
13.7 Tools of CIMS
An automated irrigation control system requires extensive hardware and can be a
major concern for the users. CIMS provides several tools in aid of the decisions on citrus
irrigation management without hardware requirements. The tools are (1) to retrieve weather
data from a text file, (2) to calculate crop evapotranspiration (ET), (3) to estimate duration
of an irrigation, and (4) to display the irrigation pipe layout. Each of these functions is
described as follows.
13 71 Aggregate Weather Data
To aggregate the weather data to daily time interval, execute the following steps.
Click Tools on the main menu of CIMS. The submenu of Tools shows in
Figure 13.7.1.
Eacts Expert Control Scheduling database Simulation
Iools
Help Quit
Aggre
gate Weather Data
ET Calculation
Estimate Irr. Duratiion
Field Layout Map
Figure 13.7.1. Submenu of Tools.
Click Load Weather Data. The program will convert the weather data from
a text file to a database file and aggregate the data into daily data (Figure
13.7.2). The weather data are used for the simulation of crop water
requirements.


203
Zazueta, F. S. and J. N. Xin. 1992. Soil-moisture and rainfall sensing devices: a review.
Final Report to South Florida Water Management District. Florida, Univ. of
Florida, Gainesville, 130pp.
Zazueta, F. S., J. N. Xin, and A. G. Smajstrla. 1993. Simulated computer-controlled
irrigation using soil water-content sensors with linear error functions. Soil Crop
Sci. Soc. Florida Proc. 52:102-107.
Zazueta, F.S., J. N. Xin, A. Wheaton, J. Jackson, and M. Almedo. 1994. Development
of a computer-based control system for reclaimed water citrus irrigation. Proc. of
the 5th Int. Con. on Computers in Agriculture, Orlando, FL.
Zoldoske, D. 1988. Computer irrigation scheduling. Fruit Grower, April, pp. 6-7.
Zur, B., U. Ben-Hanan, A. Rimmer, and A. Yardeni. 1994. Control of irrigation
amounts using velocity and position of wetting front. Irrigation Science 14:207-
212.


106
Rainfall
base checks rainfall and
irrigation events to prevent these two events from occurring at the same time.
Predefined duration
In practice, irrigations are stopped after certain predefined durations. The duration
of an irrigation could be suggested by irrigation experts. When deficit irrigation is applied
at the user's request or due to a high probability of rainfall, the duration is reduced to a
predefined percentage of full irrigation. In addition, the system also provides a simulation
model to estimate soil-water content at the crop root zone. The simulation model is a simple
water-budget as discussed in Chapter 6. By running the simulation model, a prognosis of
crop water requirements and irrigation duration is given. Thus, the simulation results can
be used as a reference to determine irrigation schedules.


74
concentrate on both water applications and observations in this zone. In practice,
tensiometers at three depths are desirable for deep rooting crops.
To account for sensor failures and different soil characteristics, tensiometers should
be installed at several different sites adequately to represent the water status in large areas.
How many and at what depth tensiometers should be installed at each site needs to be
justified for each specific application.
6,6,2 Soil-Water Potential and Allowable Water Depletion
Irrigation scheduling is usually determined by allowable soil-water depletion.
However, tensiometers measure soil-water potential. Hydraulic characteristics for the
irrigated soil are needed to establish the relationship between the soil-water potential and the
amount of soil-water depletion. Table 6.3 shows the soil-water tension versus the soil-water
content for Candler fine sand at different soil depths (Carlisle et al., 1978).
Table 6.3 Average soil-water content for Candler fine sand by volume.
Depth
Soil Water Tension, Centibar
()
0
2
3
4.5
6
8
10
IS
20
33
1500
0-1.7
36.8
34.8
31.4
17.7
11.9
8.0
6.8
5.7
5.1
4.5
1.5
0-4
36.3
34.7
32.0
17.6
11.3
7.5
6.3
5.3
4.7
4.2
1.4
Source: Carlisle et al. (1978)
For Candler fine sand, assuming the permanent wilting point (PWP) is at -15 bars,
the available water-holding capacity ranges from 7.5 to 9.5 percent by volume (Martin et al.,
1990). These values approximately correspond to -8 and -7 cb according to Table 6.3.
Thus, the relationship between soil-water potential and soil-water depletion can be
approximately established. Table 6.4 shows the relationship of soil-water potential and soil-


202
Vellidis, G., A. G. Smajstrla, F. S. Zazueta. 1990. Continuous soil water potential
measurement with a microcomputer-based data acquisition system. Applied Eng.
in Agriculture 6(6):733-738.
Villalobos, F. J. and E. Fereres. 1989. A simulation model for irrigation scheduling
under variable rainfall. Trans, of ASAE 32(1): 181-188.
Wheeler, P. A. and G. L. Duncan. 1984. Electromagnetic detection of soil moisture.
ASAE Paper No. 84-2078, St. Joseph, MI.
Wierenga, P. J., J. L. Fowler, and D. D. Davis. 1987. Use of tensiometer for scheduling
drip-irrigated cotton. Int. Con. on Measurement of Soil and Plant Water Status.
Centennial of Utah State Univ., Logan, pp. 157-161.
Wiersum, L. K. and K. Harmanny. 1983. Changes in the water permeability of roots of
some trees during drought stress and recovery, as related to problems of growth in
an urban environment. Plant and Soil 75:443-448.
Willis, L. E. and F. S. Davies. 1991. Fertigation and growth of young 'Hamlin' orange
trees in Florida. HortScience 26:106-109.
Wright, M. L., M. W. Green, G. Fiegl, and P. F. Cross. 1986. An expert system for
real-time control. IEEE Software 3:16-24.
Xin, J. N, F. S. Zazueta, and A. G. Smajstrla. 1993. Simulated effects of a rain-activated
irrigation shut-off-device on water use in turf irrigation. Soil Crop Sci. Soc.
Florida Proc. 52:107-114.
Zazueta, F, S., R. M. Lokers, and A. G. Smajstrla. 1989. Real-time expert control of
irrigation systems, ASAE Paper No. FCS 89-022, St. Joseph, MI.
Zazueta, F. S., S. Park-Brown, A. G. Smajstrla, and D. S. Harrison. 1984b. Computer
control of irrigation systems for nurseries. Proc. Fla. State Hort. Soc. 97:285-286.
Zazueta, F. S. and A. G. Smajstrla. 1992. Microcomputer based control of irrigation
systems. Applied Eng. in Agriculture 8(5):593-596.
Zazueta, F. S., A. G. Smajstrla, and G. A. Clark. 1986. Irrigation scheduling of peanut
in northern Florida. Soil and Crop Sci. Soc. of Florida, Proc. 46:60-71.
Zazueta, F. S., A. G. Smajstrla, and D. S. Harrison. 1984a. Microcomputer control of
irrigation systems I: hardware and software considerations. Soil and Crop Sci.
Soc. of Florida, Proc. 43:123-129.


157
The initial facts include the
following data:
irrigation strategy,
threshold values to
start and stop an
irrigation,
cold protection
threshold values,
crop growth stages,
percentage of less
irrigation time,
critical and non-critical growth stages, and
Markov chain rainfall probability.
Default initial facts with
detailed comments are included in the
system. The facts are specified in the
data format of the expert system shell
CLIPS. The user can modify those
critical values to achieve his or her
control need. Facts can be eliminated
from the reasoning process by adding
a semi colon in front of the facts.
Figure 13.2.5. Fertigation schedule.
(a)
(b)
(c)
(d)
(e)
(f)
(g)
Figure 13.2.4. Initial facts of the expert system.


177
13.6.2 View the Results of Simulation
Click Browse Result on the submenu of Simulation.
Select one of the simulation results (Prognosis, Simulated SWC, and
Historical SWC). The prognosis results (Figure 13.6.4) contains the
following data:
1
[jT>-
i
Prognosis Result
Hi
I1
I1
Our.br |
I
t
1
i
i0.052
17 ;
0
0 j
0
g
2
;0.046
30 |
0
0 i
0
i:
I 3
!0.040
43 ;
0
o j
0
0
4
i0.035
57 j
0
o!
0
5
|0.060j
70 i
72
7 |
9
1
6
i 0.054i
13 j
0
0 |
0
ly
7
j0.048
27 j
0
0 1
0
T
m
T
m
11
Figure 13.6.4. Irrigation prognosis from the simulation.
Day
number of days of the prognosis,
Swc
soil-water content by volume,
Depletion
soil-water depletion at the crop root zone in
percent,
Vol_tree
volume of water need to be applied per tree (gallon),
Durhour
irrigation duration hour, and
Durmin
irrigation duration minute. The overall irrigation
duration is equal to the irrigation hour plus the
irrigation minute.


86
applied by fertigation. This is because potassium oxide is so soluble that the fertilizer moves
freely into the soil. However, potassium molecules are adsorbed on the soil complex and
are not readily leached away (Keller and Bliesner, 1990).
Phosphorus fertigation is difficult because the treble-superphosphate (0-45-0) is only
moderately water-soluble (Keller and Bliesner, 1990). The quality of irrigation water must
be considered for phosphorous fertigation. If the water contains considerable amount of
calcium, it can cause clogging because any form of phosphorus will precipitate as
dicalcium phosphate in the pipeline and emitters.
Although there are general recommendations on fertigation, such as how many
pounds of nitrogen should be applied each year for citrus, precise knowledge on how long,
how much, and when to fertigate is not currently available. This is mainly due to the
complexity of crop development. However, management decisions on fertigation are
available. Fertigation may be applied as a part of irrigation, and irrigation application
should always avoid leaching the nutrients out of the crop root zone.


20
Figure 3.1. Major components of an expert system.
3,1,1 Inference Engine
The inference engine is the heart of an ES used for drawing conclusions based upon
a knowledge base. Schalkoff (1990) stated that the inference engine can be considered to
be a finite state machine with states representing typical actions such as (1) match rules, (2)
select rules, (3) execute rules, and (4) check stopping conditions (i.e., goal satisfaction).
Following are some commonly used reasoning approaches.
Forward chaining
Forward chaining is also called data driven. It is a reasoning method from facts
(data) to conclusions. For instance, if a traffic light is green (fact), then one can drive
through the traffic light (conclusion). Because of the nature of the forward reasoning
process, this reasoning method is suitable for problem domains such as monitoring and real
time control systems, where data or facts are continuously acquired or updated.


90
Figure 8.2. Decision flow of the expert system.


54
The KB was developed using the CLIPS language which contains expert knowledge
represented in rules, logical analysis, uncertainty management, and calls to external
procedures. After the knowledge is implemented into the rule base, it needs to be verified
by an domain expert. The KB development is one of the main tasks for the development.
External control procedures were designed to activate or deactivate the irrigation and
fertigation devices. Irrigation control procedure can turn on or off user specified irrigation
valves for a desired water application. Thus, irrigation can be applied block by block for
different irrigation management strategies.
Fertigation control procedure was implemented with the capability of turning on or
off both the injection pump and the irrigation valves. Cold protection control procedure is
similar to the irrigation control procedure in that they turn the irrigation valves on or off.
However, all the electronic valves should be turned on or off at once to cover the entire field
for a cold protection application.
These control actions are accomplished by the results of the reasoning process of the
KB. Each control action is displayed on the screen so that the user can view the current
event. In addition, all control events, including application types and duration, are saved in
a control event data log file.


49
Agricultural and Biological Engineering Department, University of Florida (Smajstrla,
Personal communication). Tensiometers are placed in a chamber which is partially filled
with water and sealed. A vacuum pump is used to alter the pressure inside the chamber.
Outputs (voltages) from the pressure sensors vary with the chamber pressure (tension), as
indicated by a pressure gauge. The sensor outputs can be recorded through a data logger to
a computer or one can simply use a multimeter. The relationship between the tension and
the sensor output voltage is linear. Figure 4.4 is one of the typical calibration curves. Soil-
water tension can be calculated from the pressure sensor output based upon this calibration
curve.
4.6.2 Personal Computer
An IBM or any compatible PC with 386 or higher CPU is recommended to run the
expert system. Four MB of RAM and five megabytes free space on a hard disk are required.
The PC must have at least one communication port.
4.6.3 Automated Weather Station
The weather station used in this project was a Weather 2000 system from Campbell
Scientific, Inc. This station is an automated system designed for commercial, agricultural,
and irrigation scheduling applications. This weather station measures meteorological
conditions that affect crop water consumption. The station provides the following data:
rainfall, air temperature, solar radiation, wind speed, relative humidity, and soil temperature.
4.6.4 Data Logger
The Campbell Scientific data logger (CRIO) is a fully programmable module, which
provides sensor measurement, time keeping, communications, data reduction, data/program


44
than an AI language. Barrett and Beerel (1988) stated "use a shell if you can, an
environment where you should, and an AI language when you must (p. 69)." This study
selected an expert system shell, C Language Integrated Production System (CLIPS), as the
development tool.
4.5.2 CLIPS
CLIPS, developed by NASA's Johnson Space Center, is a forwarding chaining rule-
based language which uses the Rete Algorithm for pattern matching. CLIPS can be used as
an embedded application or child process. The tool can easily call an external executable
routine. Because a real-time system requires access to external devices, these features are
important for this application. CLIPS is delivered with a complete source code so that the
user can modify and re-compile the program for special purposes.
4,6 Hardware Specification
Agricultural decisions are highly related to climate data such as rainfall and air
temperature. Because the system is designed for real-time performance, hardware is
required to acquire external data and to realize control actions. Sensors are needed for
measuring climate parameters and soil-water content. The following hardware is necessary:
Soil moisture sensor,
Personal computer,
Automated weather station,
Data logger,


5.5 Statistical Test 60
5.6 Irrigation Decision with Rainfall Probability 62
CHAPTER 6 CITRUS IRRIGATION SCHEDULING 64
6.1 Introduction 64
6.2 Citrus Water Requirements 64
6.3 Evapotranspiration and Management Allowed Depletion 67
6.4 Irrigation Depth and Duration 69
6.5 Soil-Water Budget 71
6.6 Irrigation Scheduling Using Tensiometers 72
6.6.1 Tensiometer Installation Depth 73
6.6.2 Soil-Water Potential and Allowable Water Depletion 74
CHAPTER 7 CITRUS COLD PROTECTION AND FERTIGATION 77
7.1 Introduction 77
7.2 Cold Protection Application 78
7.2.1 Principle of Cold Protection 78
7.2.2 Critical Application Temperature 79
7.2.3 Water Application Rate 80
7.3 Fertigation 80
7.3.1 Application of Fertigation 81
7.3.2 System Components of Fertigation 84
7.3.3 Fertilizer Materials 85
CHAPTER 8 CONSTRUCTION OF THE KNOWLEDGE BASE 88
8.1 Introduction 88
8.2 The Process of Control and Reasoning 89
8.3 The Sensor Data 93
8.3.1 Download the Sensor Data 93
8.3.2 Uncertainty Management of the Sensor Data 94
8.4 Irrigation Management 100
8.4.1 Irrigation Strategies 100
8.4.2 Criteria for Starting an Irrigation 102
8.4.3 Criteria for Stopping an Irrigation 105
8.5 Cold Protection 108
8.5.1 When to Turn On 109
8.5.2 When to Turn Off 109
8.6 Fertigation 110
vi


16
help facilitate the use of simulation models in several ways (Jones, 1985): (1) estimate
model parameters, (2) provide input for models, and (3) restrict scenarios for model
analyses.
As expert system technology has evolved, applications to irrigation scheduling and
operation have been developed. Wright et al. (1986) developed a real-time expert control
system (Hexscon). They concluded that the important issue in developing an RTES is
combining the best of conventional and expert-system controllers. Their results suggested
that real-time expert control can be built on a microcomputer and has enough sophistication
and capacity to be effective for real-world problems. Jacobson et al. (1987) developed an
RTES that supervised a tomato greenhouse environment controller. Jacobson et al. (1989)
implemented real-time greenhouse monitoring and control, linking a conventional expert
system with a set of utilities for data acquisition and control. Conventional expert systems
have been linked with models and data acquisition to make management recommendations.
Thomson et al. (1989) reported on an expert system that was coupled with simulation of a
peanut growth model and databases. This system evaluated moisture sensor readings
combined with a crop growth model to make estimates of irrigation timing. An RTES has
been applied to turf irrigation by Zazueta et al. (1989). In tests, it was found that the RTES
can apply irrigation in response to crop water demand. A major defect of the system was
the lack of heuristic knowledge available for irrigation control. An expert system for
irrigation management was developed in Thailand (Srinivasan et al., 1991). The system
demonstrated its effectiveness in improving water management decisions.


84
where qc =
rate of injection of liquid fertilizer solution into the system,
gph,
Fr
fertilizer application rate (quantity of nutrients to be applied) per
irrigation cycle, lb/A,
A
area irrigated in T ha,
T. =
irrigation application or set time, hr,
c'
concentration of actual nutrients in the liquid fertilizer, lb/gal, and
tr
ratio between fertilizer time and irrigation application time.
Because of the uncertainty of rainfall, a risk is always present that fertigation may
be applied immediately before a rain. Thus, plant nutrients can be leached out of the crop
root zone. To reduce this risk, the Markov chain probability of rainfall can be applied to
fertigation management. Fertigation may be delayed when a high probability of rainfall
occurs. When irrigation and fertigation schedules conflict, an algorithm is needed to merge
the two schedules and to satisfy both the application requirements. Fertigation should serve
as a part of irrigation whenever possible.
7.3.2 System Components of Fertigation
Fertigation systems consist of several components including an irrigation pumping
station, a fertilizer injection device, an injection port, a solute fertilizer reservoir, a backflow
prevention system, and calibration devices. Figure 7.1 is a common arrangement for
fertilizer injection equipment. The electrical control board controls the operation of the
fertigation pump (valve 64) and water supply valves (valve 1 to 63). Fertigation can be
applied based upon the expert knowledge or user defined schedules.


85
Figure 7.1. Major components of a fertigation system.
7.3.3 Fertilizer Materials
Many soluble materials are suitable for application through microirrigation systems.
Selection of a fertilizer material should consider the solubility, convenience, and cost of the
desired nutrients. Table 7.2 shows some of the soluble fertilizer formulas. Liquid fertilizers
can contain a single nutrient or combinations of nitrogen (N), phosphate (P), and potash (K).
Nitrogen injection is relatively easy because anhydrous ammonia (82-0-0) and aqua
ammonia (24-0-0) are completely soluble and can be injected directly into irrigation water.
However, some nutrients of the fertilizer may be lost because gaseous ammonia is likely to
volatilize (Keller and Bliesner, 1990). In addition, nitrate-nitrogen tends to persist in the soil
in solution and to move with water. Thus, this material is highly susceptible to loss due to
leaching if excessive water is applied. Materials like nitrogen and potassium are easily


30
How do we construct meta-knowledge representation?
What are the primitives and how does one manage incomplete knowledge?
The commonly applied knowledge representation methods in production systems are (1)
semantic network, (2) frame, (3) objects, and (4) rules.
3.4.1 Semantic Network
The semantic network was initially applied by Quillian (1968) to analyze words and
sentences. Since then, the approach has become widely used. A semantic net is a formal
graphic language for representing facts about entities. Its structure is shown graphically in
terms of nodes (objects) and the arcs (links) connecting them. The directed arcs that connect
the nodes represent relationships between objects. A semantic net can virtually represent
any relationship that holds among the objects or concepts in some domain of interest.
The graphical relational representation of a semantic net is explicit and succinct to
the state of knowledge. In addition, because nodes are directly connected with related nodes,
the search can be efficient. However, the nets offer no standard definition of link names
among nodes; and there are some practical difficulties in performing computer reasoning
with completely general semantic nets.
3.4.2 Frame
Frame-based representation was developed to manage information overload inherent
in large semantic nets without sacrificing their expressive power (Minsky, 1975; Bobrow
and Winograd, 1977; Fikes and Kehler, 1985; Brachman and Levesque, 1985). The basic
characteristics of a frame are that it maintains the fundamental notions of abstraction
hierarchies and inheritance of properties from superclasses, but it packages the descriptive


Modified Blanev-Criddle method
Select the Modified
Blaney-Criddle radio
button and click the
Continue button. A
calculation screen appears
as Figure 13.7.6.
Enter the required
data to the screen and click
the Calculate button to display the result. Click the Exit button to quit from the
screen.
Stephens Stewart method
Select the Stephens
Stewart radio button and click
the Continue button. A
calculation screen appears as
Figure 13.7.7.
Enter the required data
to the screen and click the
Calculate button to display the result. Click the Exit button to quit from the screen.
IS
Stephens-Stewart Method
Monthly mean temperature (*F)
Monthly solar radiation (kW/m*)
M
1
Monthly reference ET {tt^menth)
S
Calculate
¡SB
Figure 13.7.7. Stephens-Stewart ET screen.
Figure 13.7.6. Modified Blaney-Criddle ET screen.


91
be an initialization of the hardware, and turning on or off the control valves according to
predefined values. These control programs were integrated into the knowledge base so that
the control actions take place when the proper rule is fired. In other words, when a rule
"condition" is satisfied, the "action" executes external control programs. To realize these
control capabilities and real-time performance, two basic technical problems were overcome
in developing the knowledge base:
How to deal with constantly changing data in real time, and
How to reason about the behavior of the decision process.
Figure 8.3. Paradigm of the knowledge base.


ACKNOWLEDGEMENTS
I would like to express my sincere gratitude to the supervisory committee. Special
thanks go to Dr. Fedro S. Zazueta, supervisory committee chairman, for his advice and
encouragement during the study and for the opportunity to study as a graduate research
assistant. Special thanks go to Dr. Allen G. Smajstrla, supervisory committee member, for
making himself available on many occasions and for contributing his expertise toward this
study. I would also like to thank the supervisory committee members, Dr. James W. Jones,
Dr. Pierce H. Jones, Dr. Douglas D. Dankel, II, and Dr. Louis H. Motz, for their advice and
support during this study.
Special thanks also go to Dr. Thomas A. Wheaton and Dr. Lawrence R. Parsons of
the Lake Alfred Citrus Research Center for their expertise, advice, and time toward this
study.
I would like to extend my appreciation to the following people for their evaluation
and comments on the software: Dr. Robert M. Peart and Dr. Dorota Z. Haman of the
Agricultural and Biological Engineering Department, University of Florida; Dr. James J.
Ferguson and Dr. J. David Martsolf Jr. of the Horticultural Department, University of
Florida; and Ms. Cynthia Moore, St. Johns River Water Management District, Florida.
Finally, I would like to thank my wife, Qiuping Jian, for her support during this
study.
iii


123
System modularity:
The system is composed of individual modules. Because each module can
be executed and tested independently, the maintenance task is greatly
simplified.
9,5 System Tests of CIMS
Validation and verification of CEMS are important in the system development.
Common validation approaches include (1) face validation, (2) predictive validation, (3)
Turing tests, and (4) field tests (O'Keefe et al., 1987). For CIMS, validation approaches of
face evaluation, predictive tests, and field tests were conducted. A face validation is a
preliminary approach. Several experts and potential users evaluated the system against their
opinions. The predictive validation requires historical data or generated test cases. Expert


152
13.1. Brief Description of CIMS
CIMS is a computer-based citrus microirrigation management system that runs under
Microsoft Windows. CIMS provides a tool to control irrigation, fertigation, and cold
protection to improve citrus irrigation management. The system consists of several main
functions: (1) real-time expert system (RTES), (2) irrigation control panel, (3) irrigation and
fertigation control based on user defined schedules, (4) database, (5) simulated crop water
requirements, (6) tools, and (7) help. Figure 13.1.1 shows the main menu of CIMS.
Eacts Expert .Control Scheduling database Simulation Iools Help Quit
Figure 13.1.1. Main menu of CIMS.
13,1,1 System Requirements and Installation
Personal computer with 386 or higher CPU running under Microsoft
Windows 3.1 or later version
4 MB available memory
5 MB free space on hard space
VGA monitor
Mouse
Installing CIMS
Insert the disk labeled Setup into your computer drive A. Type A:SETUP,
and press . Follow the instructions on the screen to complete the
installation.


78
system must have a sufficient capacity so that the entire crop area being protected can be
simultaneously watered to achieve adequate cold protection. With aid of an on-site
weather station and a real-time expert control system, cold protection management can be
implemented automatically. The computer can be used to turn on the irrigation system when
critical environmental conditions occur.
1.2 Cold Protection Application
7,2,1 Principle of Cold Protection
Cold protection is based on thermodynamic principles, which have been discussed
by Harrison et al. (1987), Martsolf (1990, 1992), and Barfield et al. (1990). A plant gains
or looses heat from its surroundings through a heat transfer process. Heat transfer can occur
as conduction and convention, evaporation and transpiration, and radiant energy exchange
(Harrison et al., 1987). Irrigation water provides cold protection because the heat loss from
the plant to its surroundings is replaced by the sensible heat and the heat of fusion of water.
The latent heat of fusion is released when water changes from liquid to ice. The total latent
heat input to the air, ignoring heat from the soil or atmosphere, is equal to the heat lost in
cooling to wet bulb temperature, plus that lost as a portion of the drops freeze. The latent
heat flux released from the water can be expressed as equation 7-1 (Barfield et al., 1990):
Ql = 0.27 x 10-6 Pw I [C (Tws Twb) + Yf Ff] 7-1
where QL = total latent heat flux in W m'2,
pw = density of water in kg m'3,
irrigation application rate in mm h'1,
I


38
4.2 Requirements Specification
4.2.1 Goal of the System
The overall goal is to develop a computer-based tool integrated with expert
knowledge as a decision aid for irrigation, fertigation, and cold protection of citrus. The
system should have the ability in response to current weather and soil-water moisture
changes to realize real-time performance. The system must also meet the following
requirements:
Be easy to use,
Achieve irrigation system automation with minimum maintenance,
Handle missing data or unreliable sensor data,
Record each event conducted by the system,
Develop several control schemes to minimize hardware requirements, and
Be accessible through telecommunication systems.
4.2.2 System Inputs
Table 4.1 shows the basic input and output requirements of the system. Because soil
water status is a crucial factor for crop growth, sensors are needed to monitor soil-water
content in the crop root zone. Rainfall and evapotranspiration (ET) are two important
parameters that determine crop water requirements. An automated weather station is
essential to obtain weather data for real-time control. The weather data are also used for
crop ET estimation. Crop, soil, and irrigation system data are required in the decision
process. Crop data include effective root depth, percentage of crop land coverage, and age


92
The solution of these problems involves the development of a special architecture
for real time and the use of reasoning with time. Figure 8.3 shows the paradigm used for
the knowledge base that performs the following tasks:
Recognize events that may indicate a problem or need to take an action,
Determine priority of these events or actions,
Execute the action based upon the priority, and
Explain the reasoning to the user.
A scheduler is used to determine when to execute a specific group of rules. Rules
can be scheduled to fire at given time intervals or may be event driven. Data from soil
moisture sensors and the weather station should be accessed at given time intervals. The
validity of the data decreases with time after a reading. Rules to perform a task are valid or
expire in a certain period of time depending on the validity of the sensor data.
Event detection recognizes the current state of the system. Event detection involves
determining the current status of sensor readings, whether a valve (irrigation, fertigation, or
cold protection) is on or off, and rainfall occurrence. The event priority is used to determine
priority of rules. Rules having higher priority are executed first in the reasoning process.
Then, execution of the control actions is based upon the conclusion of the reasoning process.
The explanation facility displays control actions that have occurred. In addition, all
previous control actions are stored in an application log file. This decision-making process
must operate continuously to deal with constantly changing data and maintain real-time
performance.


110
and cold protection is on
Then stop cold protection
Rule 3: If air temperature > critical temperature
and air temperature < critical temperature + 2F
Then display warning message
8 6 Fertigation
Fertigation is used to provide the necessary nutrients to plants. Fertigation requires
proper irrigation to maintain an adequate soil-water content and to enable the plant to utilize
the nutrients. In practice, factors that affect the fertigation application include fertilizer
concentration, application rate, type of fertilizer, and crops. Each irrigation system may
have different application rates and different management strategies. As discussed in
Chapter 7, expert decisions on fertigation are difficult because of the complexity of plant
development. Specific knowledge on crop nutrient requirements is not currently available,
but knowledge on fertigation management is available. For this application, a set of
fertigation schedules was created in the knowledge base. Although the scheduling approach
is not a perfect solution, this approach allows the user to modify the schedule according to
expert recommendation for a specific crop and soil type. By using the schedule, the rule
base continuously checks current time and the fertigation schedule. Fertigation is applied
at the appropriate time in the scheduled period. When irrigation and fertigation schedules


41
Sensor failure can be a main cause of making an unreliable or poor decision. Thus,
safety measures or data validation should be used to evaluate the sensor data. One approach
is to install a redundant sensor; then sensor readings from the redundant sensors can be
evaluated to increase the data reliability. However, this approach increases the cost of system
hardware. An alternative approach is to use data uncertainty analysis, such as the certainty
factor (CF) discussed in Chapter 3. Sensor readings with low CFs should be discarded from
the input of the decision process. In addition, a simulation approach can be used to verify
the actual irrigation duration to avoid excess water application.
4 4 Knowledge Representation Paradigm
Commonly used knowledge representation paradigms are rules, logic, frame, objects,
or semantic networks. The choice of the paradigm should be suitable to represent the
domain knowledge and be in the consideration of the selection of development tools.
Decision processes in citrus irrigation management are heuristic in nature. The knowledge
required in this application can be considered shallow knowledge. Rule-based systems are
the best currently available means for codifying the problem-solving knowledge of human
experts (Hayes-Roth, 1985). Because of the heuristic nature of the decision-making process
and available development tools, a rule-based knowledge representation paradigm is used
in the system.
4,4,1 Reasoning Method
Since rules are selected as the knowledge representation paradigm, reasoning
methods can be either forward, backward, or opportunistic (bi-directional). For this


21
Backward chaining
Backward chaining is a goal driven method. In contrast to forward chaining,
backward chaining reverses the process. It reasons from a hypothesis, a potential goal to be
proved, to the facts which support the hypothesis. This reasoning approach is more
applicable to problems having many more inputs than possible conclusions, such as
diagnosis and classification problems. The approach was used in Prolog and the medical
expert system MYCIN.
Opportunistic chaining
Opportunistic chaining combines the forward and backward reasoning methods. For
applications with many inputs and many possible conclusions, neither forward nor backward
reasoning is an efficient approach. Thus, the two reasoning methods are applied together
in one system to achieve efficiency. However, such a method may expand the difficulty of
development.
Advanced reasoning methods
Advanced reasoning approaches (Gonzalez and Dankel, 1993) are model-based
reasoning, qualitative reasoning, case-based reasoning, temporal reasoning, and artificial
neural networks.
3,1,2 Knowledge Base
A knowledge base contains expert-level information required to solve problems in
a specific domain. A knowledge base consist of a human expert's knowledge acquired by
a knowledge engineer and encoded into the system. In other words, a knowledge base


134
These problems may include pest control, herbicide, emitter clogging, and economic factors.
However, the system as it is has provided a useful tool for irrigation decision makers where
it was tested. More importantly, the primary goal of this study was to develop a
methodology for using RTES and conventional control techniques to improve irrigation
management. CIMS showed its potential to assist the decision-making of citrus
microirrigation managers toward more efficient use of water resources. Further studies are
recommended to improve and extend the knowledge base so that it can deal with more
comprehensive problems and to apply the system to other crops.


I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and ig"fidly adequate in scope and quality,
as a dissertation for the degree of Doctor of Philos >phy\ f \ [ \ *
44-A^ -
Fetro S. Zazuetal Chair
Professorlof Agricultural and Biological
Engineering
I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is fully adequate, in scope and quality,
as a dissertation for the degree of Doctor of Philosophy.
Douglas D. Dankel, II
Assistant Professor of
Computer and Information Sciences
I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is fully adequate, in scope and quality,
as a dissertation for the degree of Doctor of Philosophy.
'rofessor of Agricultural and Biological
Engineering
I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is fully adequate, in pcope and quality,
as a dissertation for the degree of Doctor of Philosophy^
Pierce H. Jones
Associate Professor of Agricultural and
Biological Engineering


185
13 8 Help Utilities of CIMS
CIMS provides several help utilities in the system. This includes (1) hypertext
content help, (2) calculator, (3) calendar/diary, (4) clock, and (5) text editor (Figure 13.8.1).
Each of these utilities is described as follows.
13.8.1 Help
Help contains a hypertext content help screen (Figure 13.8.2). The contents are
categorized as the main menu. To inquire help message on each subject, click the underlined
Eacts Expert Control Scheduling database Simulation Tools mhim Cult
Calculator
Calendar/Diary
Clock
Editor
Puzzle
About
Figure 13.8 .1. Submenu of the Help.
content title. Click the upper left box to
exit from the window.
13.8.2 Calculator
A simple calculator was included.
Click the Calculator on the submenu of
Help to activate the calculator (Figure
13.8.3). Click the upper left button to exit
from the calculator.
Figure 13.8.2. CIMS help screen.


45
PC digital input/output board, and
Irrigation control system.
4.6.1 Soil Moisture Sensor
Numerous soil moisture sensors are available commercially. The types include
electrical, magnetic, nuclear, optical, and tensiometric sensors. A literature review of soil
moisture sensors was conducted by researchers (Schmugge et al., 1980; Mckim et al., 1980;
Zazueta and Xin, 1992). Among the many available soil moisture sensors, the tensiometer
is one of the most widely used sensors mainly because of its performance and its low cost.
In this study, tensiometers are used to measure soil-water potential at the crop root zone.
Tensiometers
Figure 4.1 shows a vacuum gauge tensiometer and a tensiometer with a micro
pressure transducer. A tensiometer consists of a ceramic porous cup, plastic body tube, a
gauge, and a service cap. The tube is filled with water. In the field the ceramic cup is
installed in the active root zone of the soil. As the soil dries, water in the tube is pulled
through the ceramic cup and the tension is displayed on the vacuum gauge or the pressure
transducer. This tension is equivalent to the soil-water tension when equilibrium is reached.
The tensiometer, in this way, measures the force exerted by the soil to extract water from
the ceramic cup. Thus, the tensiometer measures the soil-water potential or the energy status
of water for a plant rather than the quantity of water in the soil. The range of soil-water
potential that tensiometers measure is from 0 to about -0.8 bar.


116
can activate a pump or electronic valves according to the system's pre-defined on or off
status. The user can activate the pump and control valves from a local or remote computer.
ON/OFF Control Screen
ON/OFF
() Irrigation
OFF
O Fertigation
OFF
O Freeze Protection
OFF
Figure 9.2. Control panel of CIMS.
9,2,3 Scheduling
The cost of an RTES can be high because of the hardware and complexity of the
software development, particularly knowledge engineering. Consequently, CIMS provides
an alternative approach to assist irrigation managers. The scheduling module is a
conventional program that enables users to define their irrigation and fertigation schedules
for each irrigated block. Then, the system automatically turns the appropriate irrigation
valves on or off based on the user-defined schedules. Each schedule must contain (1) valve
ID, (2) start time, (3) duration, and (4) skip days or intervals between irrigations. The
schedule module is able to apply user-defined irrigation and fertigation schedules separately
or simultaneously.


79
C = specific heat of water in J kg'1 "C1,
Tws = water temperature in C,
Twb = the wet bulb temperature in C,
Yf = the latent heat of fusion in J kg'1, and
Ff = the fraction of water that has become ice (fused) when it strikes the
ground. Ff depends on drop size and environmental conditions.
As equation 7-1 shows, cold protection is mainly related to (1) air temperature, (2)
wind speed, which affects evaporation, and (3) irrigation application rate. In this
application, it was assumed that there were adequate water and energy supplies and the
irrigation system was properly designed for cold protection. This implies that the system
can irrigate an entire citrus grove simultaneously with an adequate application rate.
7,2,2 Critical Application Temperature
Knowledge of weather data, particularly air temperature, is crucial for cold
protection. Most citrus growers in Florida receive weather data from sources such as the
National Weather Service, commercial radio and TV, and county extension offices. During
the winter season, growers carefully track weather changes to make decisions related to cold
protection.
Since air temperature is a crucial factor for cold protection, it is important to
determine the critical air temperature to start cold protection. Applying cold protection too
early or too late may result in water waste or crop damage. Harrison et al. (1987) reported
that the freezing temperature for Tngelo is about 30.1 F in Orlando, Florida. However,
irrigation must be initiated before the temperature reaches freeze point because irrigation


200
Simpson, J. R. and J. J. Meyer. 1987. Water content measurements comparing a TDR
array to neutron scattering. International Conference on Measurement of Soil and
Plant Water Status, Centennial of Utah State Univ., Logan, pp. 111-114.
Smajstrla, A. G. 1990. Evapotranspiration based citrus irrigation scheduling. Special
series Report SS-AGE-18, IFAS, Univ. of Florida, Gainesville, FL.
Smajstrla, A. G., D. S. Harrison, and F. X. Duran. 1985b. Tensiometers for soil
moisture measurement and irrigation scheduling. Cooperative Extension Service,
Circular 487. IFAS, Univ. of Florida, Gainesville, FL.
Smajstrla, A. G., W. G. Boggess, B. J. Boman, G. A. Clark, D. Z. Haman, G. W.
Knox, S. J. Locascio, T. A. Obreza, L. R. Parsons, F. M. Rhoads, T. Yeager,
and F. S. Zazueta. 1995. Status and growth of microirrigation in Florida, Proc.
of the Fifth International Microirrigation Congress, Orlando, FL, p. 325-330.
Smajstrla, A. G., D. S. Harrison, F. S. Zazueta, L. R. Parsons, and K. C. Stone. 1987.
Trickle irrigation scheduling for Florida citrus. Florida Cooperative Extension
Service, Bulletin 208. IFAS, Univ. of Florida, Gainesville, FL.
Smajstrla, A. G. and R. C. J. Koo. 1984. Effects of trickle irrigation methods and
amounts of water applied on citrus yields. Proc. Fla. State Hort. Soc. 97:3-7.
Smajstrla, A. G. and R. C. J. Koo. 1986. Use of tensiometers for scheduling of citrus
trickle irrigation. Proc. Fla. State Hort. Soc. 99:51-56.
Smajstrla, A. G., L. R. Parsons, K. Aribi, and G. Velledis. 1985a. Response of young
citrus trees to irrigation. Proc. Fla. State Hort. Soc. 98:25-28.
Smajstrla, A. G., L. R. Parsons, F. S. Zazueta, G. Vellidis, and K. Aribi. 1986. Water
use and growth of young citrus trees. ASAE Paper No. 86-2069, St. Joseph, MI.
Smajstrla, A. G. and F. S. Zazueta. 1987. Simulation of irrigation requirements of
Florida agronomic crops, Soil and Crop Science Soc. of Florida, Proc. 47:78-82.
Smajstrla, A. G. 1990. Technical Manual of AFSIRS Model. IFAS, University of
Florida, Gainesville, FL.
Snyder, G. H., B. J. Augustin, and J. M. Davidson. 1984. Moisture sensor-controlled
irrigation for reducing N leaching in bermudagrass turf. Agron. J. 76:964-969.
Srinivasan, R., B. A. Engel, and G. N. Paudyal. 1991. Expert system for irrigation
management. Agricultural Systems 36:297-314.


105
soil-water potential is over the maximum allowed value. Only two conditions, soil-water
potentials and growth stages of trees, are required to start an irrigation. An irrigation starts
when the soil-water potentials are over the maximum allowed values. Table 8.5 shows a set
of sample data for Criteria D.
Table 8.5. Critical sensor readings (Criteria II) to start an irrigation.
Young
Mature (critical)
Mature (non critical)
Sensor;
S¡ < R1 (-10 cb)
Si S; Where S¡ denotes sensor outputs at depth i (i = 6", 12", and 24" .
Irrigation Criteria I has higher priority than irrigation Criteria II. Criteria II plays
a role of preventing crop water stress due to extremely low soil-water potential.
Furthermore, sensor readings at different soil depths have the same priority to start an
irrigation. This implies that any one of the sensors can trigger an irrigation if the starting
criteria are satisfied.
8,4,3 Criteria for Stopping an Irrigation
After an irrigation system is turned on, a decision needs to be made on when to turn
it off. Figure 8.12 exhibits alternative approaches to stopping irrigation. Irrigation is
stopped by the following criteria: (1) rainfall, (2) predefined duration, and (3) readings of
soil moisture sensors.


28
any formal way; and they may have difficulty in explicitly describing their reasoning. These
problems were summarized by Harandi and Lange (1990):
Vocabulary. Knowledge engineering is virtually impossible unless the
knowledge engineer has a basic understanding of the problem domain. An
essential part of that understanding is familiarity with domain terminology.
Completeness. A knowledge engineer must be able to identify pieces of
information or knowledge that are missing from the knowledge base.
Integration. A knowledge engineer should find out how new information fits
into the current knowledge base because the new information could interact
with already available information in an undesirable way.
Analysis. Usually, experts have difficulty in explaining exactly how and
why they reach certain conclusions. Therefore, knowledge engineers may
have to conduct an interview that may require substantial communication
skills.
3,3,3 Practical Issues
In the real world, the process of knowledge acquisition should consider many
practical issues; and there is no standard approach to follow. The practical considerations
of knowledge acquisition were discussed by Jones (1989b), and Gonzalez and Dankel
(1993). For instance, how to find a "real" expert who is articulate and very knowledgeable
in the problem domain, how to plan and to conduct an interview, and how to capture the
detailed knowledge are problems that must be addressed.


128
and the Markov chain probability of rainfall was greater than the threshold values (60, 50
and 40 percent, respectively). Otherwise, full irrigation was applied when an irrigation was
requiredand the Markov chain probability of rainfall was less than the threshold value. In
other words, less water was applied in anticipation of rainfall when rainfall probability was
greater than the threshold value. For deficit irrigation (Case 4), the irrigation decision ignored the Markov chain
probability of rainfall. In this case, deficit irrigation was always applied when irrigation was
required during the non-critical growth stages.
As Table 9.2 shows, for full irrigation, the SCS (SCS, 1982), AFSIRS (Smajstrla,
1990), and water-budget models computed similar water consumption for the 22-year
period. NIR was reduced and the total number of irrigations was increased when a deficit
irrigation strategy was applied. For the 60 percent Markov chain probability of rainfall,
there was no significant difference in water use (9.15 and 12.08 inches) between full
irrigation and deficit irrigation. When the threshold value of the probability of rainfall was
reduced to 50 and 40 percent (Cases 2 and 3), NIR was reduced (25.48 and 43.40 inches,
respectively) and the number of deficit irrigations was increased.
For deficit irrigation ignoring the Markov chain probability of rainfall (Case 4), even
less water was required than in Cases 1, 2, and 3. However, the total number of irrigations
was increased. This was because deficit irrigation was always applied whenever irrigation
was required. Thus, irrigation was applied more frequently than either full irrigation or
deficit irrigation when Markov chain rainfall probabilities were used. In contrast, Cases 1,
2 and 3 only applied deficit irrigation when the rainfall probability was greater than the
threshold values. Thus, there were more chances to increase effective rainfall and to save


24
3.2.1 Whv Use an RTES?
ESs have great potential value as control devices for many applications. In this role,
it is important to make provisions for an easy-to-use interface of sensor inputs, system
outputs, and control actuators. In the real world, systems that are designed to control
complex and dynamic processes (such as on-line monitoring) require fast handling of data
and execute diverse functions. The control decisions, which may require deep knowledge
or expertise, should be made based upon timely data. RTESs might be useful for domains
where conventional ES approaches have failed or are impractical. These may include
situations in which humans fail to effectively monitor data, make costly mistakes, miss
optimizing opportunities, are unable to solve conflicting constraints, or suffer from cognitive
over load. Turner (1986) pointed out that the main reason for using an RTES is to reduce
the cognitive load on users to enable them to increase their productivity without the
cognitive load on them increasing.
3.2.2 Characteristics of RTESs
Three factors are of main concern for RTESs. First, conclusions must be reached
and actions must be taken in real-time to respond to the sensor's perceptions and
environmental change. Second, the system must be able to provide tentative conclusions
based on initial evidence if not all of the data are available at once. Third, the system must
operate safely and reasonably on inaccurate and uncertainty of data input.
Laffey et al. (1988) perceived a series of characteristics of RTESs that differ from
conventional ESs. Those major characteristics are as follows.


34
3 5.1 Rule-Based Architectures
Rule-based systems or production systems have three main components: (1) working
memory, (2) rule memory, and (3) inference engine. The architecture and execution cycle of
rule-based systems is illustrated in Figure 3.4.
The working memory functions as a storage facility of these objects representing facts
about the world. The rule memory contains rules or the knowledge base of the system. The
inference engine is the active element in the system. It selects rules from the rule memory that
matches the contents of the working memory and executes the associated actions. If a rule
is matched with the content of the working memory, the rule is said to be fired. The conflict
resolution strategy will affect system behavior (Gonzalez and Dankel, 1993). It should be
chosen with care.
Figure 3.4. The architecture and execution cycle of rule-based systems.


141
Rain (cm)-o-S1-6" -^SI-12" S2-6" _^_S2-12"
Figure 12.3. Tensiometer reading on Juilan day 279, 1994.
Note:
SI-6"
51-12"
52-6"
S2-12"
Tensiometer SI-6"
Tensiometer at site 1 with 6" depth.
Tensiometer at site 1 with 12" depth.
Tensiometer at site 2 with 6" depth.
Tensiometer at site 2 with 12" depth,
showed low confidence and might not be failed completely.


3.3 Knowledge Acquisition 25
3.3.1 Basic Approaches 26
3.3.2 Potential Problems 27
3.3.3 Practical Issues 28
3.4 Knowledge Representation 29
3.4.1 Semantic Network 30
3.4.2 Frame 30
3.4.3 Objects 31
3.4.4 Rules 32
3.5 Rule-Based Expert Systems 33
3.5.1 Rule-Based Architectures 34
3.5.2 Uncertainty Management 35
CHAPTER 4 SYSTEM SPECIFICATION AND DESIGN 37
4.1 Domain of the Problem 37
4.2 Requirements Specification 38
4.2.1 Goal of the System 38
4.2.2 System Inputs 38
4.2.3 System Outputs 39
4.3 Knowledge Specification 40
4.4 Knowledge Representation Paradigm 41
4.4.1 Reasoning Method 41
4.4.2 System Performance Requirements 42
4.5 Development Tools 42
4.5.1 Expert System Shells 43
4.5.2 CLIPS 44
4.6 Hardware Specification 44
4.6.1 Soil Moisture Sensor 45
4.6.2 Personal Computer 49
4.6.3 Automated Weather Station 49
4.6.4 Data Logger 49
4.6.5 PC Digital Input/Output Board 50
4.6.6 Irrigation Control Board 50
4.6.7 Overview of the Hardware 52
4.7 Paradigm of the Real-Time Expert System 53
CHAPTER 5 PROBABILITY OF RAINFALL 55
5.1 Introduction 55
5.2 Markov Chain 56
5.3 Rainfall Data 58
5.4 Frequency of Rainfall 59
v


48
Figure 4.3. Tensiometer calibration equipment.


114
9.2.1 Expert System Module
The RTES consists of (1) data input, (2) knowledge base, (3) valve control, and (4)
application log modules. The RTES reads the sensor data and conducts a reasoning process
according to the rule base to make application decisions on irrigation, fertigation, and cold
protection.
Data input modules
The data input modules (read sensor data and reformat data structure) deal with
reading sensor data from external devices, and reorganizing it into the CLIPS data format.
The read sensor data module reads real-time data from the soil moisture sensors and the
weather station at user-specified time intervals. Sensor data are stored in a data logger and
retrieved through a communication link between the computer and data logger. Because
CLIPS requires its own data format, the downloaded sensor data are processed before they
are used by the knowledge base (KB).
Knowledge base IKB1
The KB is the brain of the system. Expert knowledge on citrus irrigation, fertigation,
and cold protection is acquired to build the KB. After the knowledge is acquired, production
rules are used to represent the expert knowledge. These rules are created using the CLIPS
IF-THEN rule-based language and stored in ASCII text format.
The RTES also maintains fact base s (FBs) (Figure 9.1). FBs are used to define facts
such as valve on or off position specifications for each application. The user can modify
these facts to achieve a specific control platform. Control actions are executed based on the
reasoning process. The irrigation system's pump and electronic valves are activated when


109
8.5.1 When to Turn On
The dry bulb air temperature is measured during the decision-making process.
Critical temperature indicates that the crop is in risk of cold damage. This critical
temperature has been determined through cold protection trials. The Institute of Food and
Agricultural Science (IFAS) at the University of Florida recommends initiating cold
protection at a temperature 36F and stopping at 36 to 40 F. Thus, the critical air
temperature used here was assigned the value of 36F.
8.5.2 When to TurnOff
The decision of when to stop cold protection is more difficult to determine and less
certain than that of when to start the cold protection. The rate of air temperature increase
is generally much greater than the rate of fall (Barfield et al., 1990). If the plant part
remains at or below the plant critical temperature for a sufficient length of time, the plant
can be damaged. The decision to stop cold protection must balance the risk and resource
conservation. Cold protection should last long enough to achieve effective protection. The
stop temperature (36F to 40F) recommended by IFAS is used for this system. The critical
temperature to start or stop a cold protection can easily be modified by the user according
to the crop's cold resistance and the environment. Three dummy rules have been developed
based on the decision process.
Rule 1: If air temperature < critical temperature
and cold protection is not on
Then apply cold protection
Rule 2: If air temperature > critical temperature


39
of trees. Soil data include soil depth and soil-water-holding capacity. Microirrigation
(micro-spray) is assumed to be used in this application. The following microirrigation
system data are required:
Number of emitters used per tree,
Emitter flow rate,
Wetted diameter, and
Irrigation application efficiency.
Table 4.1 System input and output requirements.
Inputs
Outputs
Soil-water tension
Turn on or off specified irrigation valves
Weather data
Turn on or off fertigation pump
ET coefficients
Apply pre-injection and flush for fertigation
Tree status (age of tree)
Apply cold protection
Soil characteristics
Simulate crop water requirement
Irrigation application rate
Display sensor maintenance messages
4,2,3 System Outputs
Outputs of the system need to be specified to apply irrigation, fertigation, and cold
protection. These outputs can also be a message displayed on the screen or a control signal
sent to an external device. After the decision-making process is accomplished, the system
control procedures activate or deactivate irrigation control valves and pumps. Thus,
irrigation, fertigation, and cold protection are applied according to the decision of the expert


180
(Inactive FoxPro Run Command)
ftread: 2271
Line written: 92
Weather Watch Processed Daily Outputs
Month
Day
Vear
Max
fiir
IcnpF
Min
Dir
TenpF
flug
Sir
Mf/n2
Us
fliph
Rain
in
ET
6 29
199-3
104.00
62.51
0.35
3.4
0.000
0.309
6 3a
199-3
102.30
64.16
0.35
3.5
0.000
0.310
7 l
1994
104.20
69.13
0.34
3.4
0.000
0.318
7 2
1994
104.30
74.30
0.34
4.8
0.000
0.332
? 3
1994
101.90
63.18
0.35
3.1
0.000
0.269
7 -3
1994
90.00
65.94
0.33
3.4
0.000
0.258
7 S
1994
96.30
67.13
0.28
3.8
0.010
0.205
7 6
1994
93.10
67.70
8.33
3.8
0.038
0.218
7 7
1994
97.10
72.40
0.34
3.8
0.800
0.243
7 8
1994
97.20
94.30
0.7?
5.8
fl .
0.857
7 9
1994
90.70
64.39
0.30
2.0
0.000
0.145
7 10
1994
92.90
65.18
0.30
3.4
0.003
0.211
7 11
1994
90.50
83.40
0.96
4.4
0.000
0.106
sailing files completed.
Figure 13.7.2. Read weather data from weather station.
13,7,2 Calculate Reference ET
To calculate reference ET, execute the following steps.
Click ET Calculation on the submenu of Tools. An ET dialog window
appears as Figure 13.7.3.
Select an ET method
Select one of the ET methods
[PenmanJ
O Blaney-Criddle
O Modified Blaney-Criddle
O Stephens-Stewart
Figure 13.7.3. ET method dialog window.
Select one of the ET methods by clicking the radio button.


xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID EVQHBTM4I_Y1BZW2 INGEST_TIME 2015-03-31T18:52:54Z PACKAGE AA00029824_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES


147
(time-constrain no)
(weather-constrain no)
Test case 21
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 24 12 24 24 32)
(sensor-reading id2 6 24 12 24 24 24)
(time-constrain no)
(weather-constrain no)
Test case 22
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 24 12 24 24 24)
(sensor-reading id2 6 28 12 24 24 24)
(time-constrain no)
(weather-constrain no)
Test case 23
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 24 12 24 24 24)
(sensor-reading id2 6 24 12 28 24 24)
(time-constrain no)
(weather-constrain no)
Test case 24
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 24 12 18 24 18)
(sensor-reading id2 6 18 12 18 24 18)
Test case 26
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 18 12 24 24 18)
(sensor-reading id2 6 18 12 18 24 18)
Reject data from ID 1-3
Not turn on
Data OK
Turn on by ID2-1
Data OK
Turn on by ED2-2
Data OK
Not turn on
Data OK
Not turn on


167
Valve ON/OFF Status
m
ON
Start
£mt
Skip
ON
Sta*
ill
i
mrnmm.
Skip
ON
Stef
ill
Ill
11!
Hup
#
OFT
Tita
Ttata Day
Hi
OFFSSSSS
Tinta
Tft Day
ste
pins
teateisi;
Tart Day
OFF
*
8
Wk
t
mm
28
OFF
Site
wm
mm
mm
0
m
OFF
iiis
ft

9
8
't
OFF
*
8
9
t
1

OFF
9
0
9
0
9
44
OFF
s
8
8
ft
lift
m
OFF
9
0
9
8
2
24
OFF
3
8
9
0
0
<5
site!!
3
sites
9
iH
m
OFF
ft
HI
8
18
s::.*
2S
OFF
8
wm
site
mm
0
9
mwrn
9
ft
Ite
9
9
m
OFF
a
8
9
1
8
28
OFF
9
0
mm
o
0
41
OFF
3
s
9
ft
11
$
OFF
9
8
9
8
8
2?
OFF
3
A
9
a
d
48
OFF
3

ft
8
8
7
orF
9
8
i
wm
8
28
¡SI
9
9
9
8
0
4
OfF
8

ft
ft
il
9
OFF
mm

9
8
8
;S2te
OFF
8
8
9
9
A
50
OFF
9
ft
8
9
ft

OFF
ID
mm
18
8
8
38
OFF
9
A
9
8
ft
Its
OFF
3

ft
8
fill
to
OFF
<1
8
9
8
8
11
OFF

9
S'9''
o
ft
m
OFF
8

: ft
8
lift

hO!
a
mm
9
8
8
32
OFF
9
0
9
9
A
83
OFF
Ipil
9
ite
III
I
m
OFF
*
8
9
8
8
S3:
OFF
8
9
ite
A
8
m
OFF
9
ft
sites
iSSSft
I
OFF

8
8
ill

34
OFF
3
HI!
8
ft
ft
SB
OFF:
9
stess
9
sites
8

OFF
$
8
9
8
iii
35
OFF
Site
Site!
9
mm
A
SB;
OFF
sste
8
|¡¡§
9
9
IS
OFF
*
8
Wm
8
8
38
00s
iite
A
SiteS
Siftii
0
5?
OFF
te
ft
Sites:
ft
S
m
OFF
mm
mm
9
mm
8
§1
OfF
HI
III
IS!
Site:
8
86
OFF
site:
ite
ft
ft
I
*?
OFF
mm
mm
9
mm
8
38
OFF
9
9
9
ft
8
SB
OFF
site
ft
ft
8
II
m
OFF
9
0
mm
8
8
38
OFF
9
9
III
0
9
80
OFF
s
8
8
ft
ft
m
orr
9
8
mm
8
8
48
OFF
9
0
Ite
d
d
5t
OFF
9
III
ft
7
isft
m
OFF
9
8
9
8
Site
OFF
9
a
8
ft
9
82
OFF
3
ft
SSteS
8
8
fj
OFF
HI
8
9
8
8

OFF
8
9
9
8
8
Figure 13.4.7. Fertigation valve on or off display.
13,4,4 Apply Irrigation and Fertigation
After the irrigation and fertigation schedules have been defined, the program can
apply both the irrigation and fertigation schedules simultaneously. However, the user must
take care regarding the overlap between the irrigation and fertigation schedules. The steps
to apply the irrigation and fertigation schedules are as follows.
Click Apply Irrigation & Fertigation on the submenu of Scheduling. A dialog screen
appears in Figure 13.4.8.
Click the Cancel button to exit the screen or click the Proceed button to apply the
user defined irrigation and fertigation schedules.
A pop up screen displays on or off status of the valves and pump, phase of an
irrigation or fertigation (preinjection, fertigation, and flush).
Double click left button of the mouse to interrupt the continuous process.


62
5.6 Irrigation Decision with Rainfall Probability
One of the irrigation management goals is to maximize the use of effective rainfall.
For an automated irrigation management system, obtaining real-time rainfall data is
important to achieve this goal. The results of the Markov chain wet-dry day probability of
rainfall were integrated into the knowledge base to aid irrigation decision-making. Because
an automated weather station was installed in the field, daily rainfall for the past three
consecutive days can be used to estimate today's rainfall probability based on the Markov
chain probability of rainfall. The irrigation decision, then, is coupled with probability levels
of rainfall. Thus, irrigation may be delayed or less water may be applied if a high
probability of rainfall occurs and an irrigation is required. A practical issue related to
irrigation scheduling using the wet day frequencies is what value of wet-day frequency
should be considered a threshold for high probability of rainfall occurrence. This threshold
is a critical value that can affect an irrigation decision. When 60 percent is considered the
high rainfall probability, only summer and fall could have the possible values that are larger
than 0.6. For mature citrus trees, summer and fall are not critical growth stages. This
implies that it may be feasible to maintain the soil moisture at a lower level during these
seasons without causing yield loss. To maximize utilization of effective rainfall and to
reduce cost, irrigation may be delayed or less water may be applied when rainfall probability
is greater than the threshold value.
In addition to the stochastic model, another way to obtain rainfall probability is to
directly access a weather forecasting database. If the user can obtain a short-term weather


4
primary cold protection method for Florida citrus is irrigation (Parsons et al., 1989; Parsons
and Wheaton, 1990). Effective irrigation management for cold protection can reduce tree
loss and increase profitability. However, cold protection management requires timely and
accurate climatic data so that adequate protection measures can be taken. On-site real-time
monitoring of weather data and expert knowledge on cold protection are necessary for farm
management.
As personal computers have become increasingly common, the potential for
computer-based decision support systems for farm water management has also increased.
Computerized irrigation scheduling systems have been developed by Cahoon et al. (1990),
Phene et al. (1992), and Zazueta et al. (1984a, 1994). Expert systems techniques can be used
to represent the heuristic knowledge required for decision making. Unlike simulation
systems, which are based on mechanistic biological or mathematical models, expert systems
use expert knowledge in the decision process like that used by human decision-makers.
Real-time expert systems (RTES) operate in a real-time domain and deal with dynamic data
and time critical responses, applying expert systems technology to control engineering.
In an RTES, most of the inputs come from sensors, while many of the outputs go to
effectors. Soil moisture sensors and weather stations can monitor soil water content and
climatic conditions, respectively. Expert knowledge can be acquired to develop several
alternative strategies and apply the one most suited to a specific problem. With the real-time
soil and weather data monitoring integrated with expert knowledge on farm management,
the system can be operated in real-time.


8.2 Decision flow of the expert system 90
8.3 Paradigm of the knowledge base 91
8.4 Process of downloading weather and soil moisture sensor data 93
8.5 CF propagation by checking range of sensor data 95
8.6 CF propagation for sensors SI and S3 96
8.7 CF propagation for sensor S2 96
8.8 Process of selecting valid sensor readings from different locations .... 99
8.9 Decision process to use a full or deficit irrigation strategy 101
8.10 Decision process to start an irrigation (criteria I) and sensor readings
to trigger an irrigation for trees during different growth stages 103
8.11 Decision process (criteria II) to start an irrigation and critical sensor
readings for trees during different growth stages 104
8.12 Decision process to stop an irrigation 106
8.13 Cold protection decision processes based on the critical air
temperature 108
9.1 Program modules of CIMS 113
9.2 Control panel of CIMS 116
9.3 Irrigation system database 117
9.4 Data flow of CIMS 120
9.5 Structure of the knowledge base and initial facts 123
12.1 Tensiometer readings on Juilan day 213, 1994 139
12.2 Tensiometer readings on Juilan day 277, 1994 140
12.3 Tensiometer readings on Juilan day 278, 1994 141
xi


107
Soil moisture sensor reading
Theoretically, irrigation should be turned off when soil-water content reaches field
capacity. Using tensiometers to turn off an irrigation is more difficult than using it to turn
on an irrigation system. This is because of variation of soil characteristics, root distribution,
wetted front of irrigation water, and lag time of tensiometer response. Two practical
questions need to be answered to use tensiometers for this purpose. First, at what value of
soil-water potential should irrigation be turned off? Ideally the value of soil-water potential
should be field capacity. Second, which sensor at the different soil depths should be used
to stop irrigation?
Because tensiometers have a certain response time to irrigation water, irrigation
should be turned off before they indicate that field capacity has been reached. Unlike the
"turn on" process, tensiometers at different root depths should have different priorities when
turning off an irrigation. Deeper sensors should have higher priorities to turn off irrigations.
Since there are some practical problems with using tensiometers to turn off an irrigation, this
is an optional approach in the system.
A new device called wetting depth probe (Zur et al., 1994) may be an alternative
means that can be used to stop irrigations. The probe measures the movement of the wetting
front, and irrigation can be stopped when the wetting front reaches a certain depth.
However, more field tests are needed for practical application of the device.


15
2.5 Expert Systems in Agriculture
In the past few years there has been significant interest among researchers in the
concept of expert systems. Expert systems have been widely applied in medicine, military,
industry, and agriculture (Laffey et al., 1988; Jones, 1989a; Feigenbaum et al., 1994). As
the development of expert system technology continues, expert systems are increasingly
being used in applications that sense the environment and directly influence it through
action. Techniques of real-time problem solving have been studied (Strosnider and Paul,
1994). In practice, real-time expert systems (RTESs) have been successfully developed for
control, monitoring, and diagnosis applications (Padalkar et al., 1991; Schnelle and Mah,
1992; Harrison and Harrison, 1994). Advanced personal computers and commercially
available easy-to-use expert system shells allow many people from different disciplines to
develop expert systems (Durkin, 1994). The use of expert system design methodology in
building agricultural decision support systems has shown great potential in recent years.
In agriculture, expert systems have been developed to assist the transfer of
technology from agricultural researchers and extension services to producers. Many expert
systems have been developed for management of nutrients, irrigation machinery, insect and
weed control, disease diagnosis, harvesting, and marketing (Jones and Haldeman, 1986;
Peart et al., 1986; Kalkar and Goodrich, 1986; Lemmon, 1986; Kline et al., 1987; Morey et
al., 1988; Muttiah et al., 1988; McClendon et al., 1989; Batchelor and McClendon, 1992;
Merlo, 1992; Kumar et al., 1992). Expert systems for crop management integrated with
simulation models have been developed (Plant, 1989; Palmer, 1986). Expert systems can


149
Test case 33
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 28 12 28 24 36)
(sensor-reading id2 6 28 12 28 24 28)
Test case 34
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 28 12 28 24 28)
(sensor-reading id2 6 36 12 28 24 28)
Test case 35
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 28 12 28 24 28)
(sensor-reading id2 6 28 12 28 24 36)
Test case 37
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 68 128 24 3)
(sensor-reading id2 6 8 12 8 24 5)
Test case 38
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 8 12 8 24 5)
(sensor-reading id2 6 8 12 8 24 3)
Test case 39
(MARKOV MONTH 6 W D D 0.71)
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(MONTH 6 W D D)
Reject data from ID 1-3
Turn on the system
Reject data from ID2-1
Turn on the system
Reject data from ID2-3
Turn on the system
Data OK
Not turn on
Data OK
Not turn on
Apply deficit irrigation


187
13.8.5 Text Editor
Click the Editor on the submenu of Help to open
a text editor (Figure 13.8.6).
13.8.6 Puzzle
Click the Puzzle on the submenu of Help to
open a puzzle (Figure 13.8.7).
Figure 13.8.5. Clock.
Figure 13.8.6. Text editor.
13,8.7 About
Figure 13.8.7.
Puzzle.
Click About on the
submenu of Help to view the about
screen (Figure 13.8.8).
Click the Exit button to
quit from the screen or click the
Next button to view next screen
CIMS For Windows
Version 1.0.1994
Jionnong Xin
AgricuraJ Engineering Deportment
University ot Florida
Gainesville, FL 32611
USA
(Figure 13.8.9).
Figure 13.8.8. About the CIMS.


LIST OF TABLES
Table Page
4.1 System input and output requirements 39
4.2 Characteristics of pressure transducer model 141PC 47
5.1 Markov chain wet-day frequency 59
5.2 Results of paired t-test for rainfall probabilities within each season ... 61
6.1 Citrus irrigation water requirements in central Florida 65
6.2 Citrus crop coefficients and recommended MAD in Florida 68
6.3 Average soil-water content for Candler fine sand by volume 74
6.4 Estimated soil-water tension in corresponding to soil-water
depletion for Candler fine sand 75
7.1 Pounds of nitrogen fertilizer to be applied to furnish nitrogen
requirement of orange and grapefruit trees under normal conditions 83
7.2 Solubility of common fertilizers in water 87
8.1 Criteria for checking possible sensor Failure B 97
8.2 Criteria for checking possible sensor Failure C for sensors at the
same depth from different locations 97
8.3 A sample propagation of CF 98
8.4 Sensor readings and constraints to start an irrigation 103
Vlll


40
system. Because the real-time system operates continuously, the system should display all
sensor readings and warning messages to request maintenance when a sensor failure is
identified. In addition, each control action should be saved to a file for future reference.
4.3 Knowledge Specification
Expert knowledge is the heart and power of an expert system. Knowledge required
for the system development, in general, involves citrus irrigation management, fertigation
application, cold protection, and sensor behavior.
For irrigation management, irrigation strategies and criteria for turning on or off the
irrigation system must be resolved. These criteria relate to the soil-water content in the crop
root zone and weather conditions. Irrigation management decisions should maintain the soil-
water content between certain levels in the crop root zone and should avoid applying too
little or too much water.
Fertigation should be scheduled at the right time and be applied for the proper
duration. Application of fertigation should be in a sequence of pre-injection, nutrient
application, and flush period. Expert decisions on fertigation and irrigation are needed to
achieve water savings and to avoid environmental pollution. For cold protection, factors that
affect cold damage of citrus trees should be understood. These factors include the principle
of cold damage, air temperature, and wind speed effect. In particular, critical air
temperatures for application of cold protection need to be specified. Knowledge of when
to start and when to stop an application need to be acquired for the reasoning process.


36
The combined CF for the hypothesis is calculated:
CF
combintd
CF, CF2(1 -CF,)
CF, CF2
1 minflCF,|, |CF2|)
CF, CF2(1 CF,)
both >0
one <0
both <0
3-2
One of the advantages the CF has in comparison with Bayesian theory is that the CF
avoids the need to establish prior probability. Moreover, the CF represents and combines
the effects of multiple sources of evidence in terms of joint beliefs or disbeliefs in each
hypothesis. Consequently, the CF is an easy to apply and widely used approach for
uncertainty management.


CHAPTER 1
INTRODUCTION
1.1 Statement of the Problem
Modem farm management involves complex decisions and daily operations that are
affected by water and nutrient requirements of crops, temporal distribution of rainfall,
environmental protection, and extreme weather conditions. In recent years, increasing costs
of energy, increasing water demands from non-agricultural users, and adverse weather cycles
are forcing the agricultural industry to use new technologies to improve water management
capabilities and to increase the efficiency of resources used in production.
Irrigation is the largest consumer of fresh water in the world. In Florida, agriculture
accounts for over 40 percent of total fresh water use (Fernald and Patton, 1984) -- about
3,000 million gallons per day. Citrus, one of the major crops in Florida, is a billion dollar
industry and consumes millions of gallons of water for irrigation each year. Thus, even
modest increases in water use efficiency will result in substantial water savings and reduce
energy cost.
Operational costs for irrigation are increasingly due to increasing energy costs.
Agricultural energy consumption varies greatly among the different commodities and
agricultural practices in Florida. For citrus, an energy survey (Stanley et al., 1980) showed
that 11,588.90 billion BTUs were used in 530,000 acres of production with 32.2 percent of
1


65
Table 6.1 Citrus irrigation water requirements in central Florida.
ET
Normal year
Dry year
Month
(in/month)
ER
NIR
ER
NIR
Jan
1.68
1.03
0.65
0.88
0.80
Feb
1.75
1.26
0.49
1.08
0.67
Mar
2.54
1.72
0.82
1.48
1.06
Apr
3.33
1.42
1.91
1.21
2.12
May
4.29
1.68
2.61
1.44
2.85
Jun
4.84
3.42
1.42
2.93
1.91
Jul
5.11
4.01
1.10
3.45
1.66
Aug
4.88
3.66
1.22
3.14
1.74
Sep
4.16
3.16
1.00
2.71
1.45
Oct
3.24
2.03
1.21
1.74
1.50
Nov
2.19
0.89
1.30
0.77
1.42
Dec
1.73
0.99
0.74
0.85
0.88
Total
39.74
25.27
14.27
21.68
18.06
Note: Nil
R. = net irrigation requirement (inches),
ER = effective rainfall (inches), and
ET = evapotranspiration.
Source: SCS (1982) (p.4-30)


CHAPTER 6
CITRUS IRRIGATION SCHEDULING
6.1 Introduction
Irrigation scheduling is important to maintain adequate soil-water content for high
productivity and the resulting economic benefits. Studies have shown that citrus irrigation
can increase fruit production (Myers and Harrison, 1978; Koo and Smajstrla, 1984;
Smajstrla and Koo, 1984; Adams, 1992). Irrigation scheduling involves decisions on when
to irrigate and how much water to apply. Irrigation scheduling methods can be based on (1)
soil properties, (2) plant properties, or (3) a soil-water balance modeling approaches. Each
method has advantages and disadvantages. In this study, soil properties and a soil-water
balance model were used.
6,2 Citrus Water Requirements
Citrus water use involves a process of soil-water extraction by the roots and
transpiration from leaves. Irrigation should provide water to crops to meet the
evapotranspiration (ET) demand imposed by climate. Citrus water requirements have been
studied by researchers (Rogers and Tucker, 1978; SCS, 1982; Smajstrla et al., 1986). Table
6.1 shows the citrus water requirements for central Florida (SCS, 1982). The estimated
citrus annual ET is 39.74 inches in central Florida. Similar results have been obtained by
64


81
of application rate. Because of the high efficiency and centralized control of microirrigation
systems, fertilizer placement through microirrigation systems can improve its efficiency of
application (Keller and Bliesner, 1990). For this application it was assumed that the
irrigation system was properly designed and it was adequate for cold protection and
fertigation.
Fertigation management decisions are affected by available fertilizer concentrations,
desired application rates, types of fertilizer, and the crop. Application of too little fertilizer
may not obtain the desired results, and excessive applications of chemicals may result in
unnecessary expenses and potential crop or environmental damage.
7,3,1 Application of Fertigation
Fifteen chemical elements have been found to be essential nutrients to satisfactory
growth and functioning of citrus trees (Jackson, 1991). Among the fifteen chemical
elements, three elements (carbon, hydrogen, and oxygen) are adequately provided in the
environment suited to tree growth and are largely beyond the control of the grower. The
other twelve are fertilizer elements or "plant food." The major chemical elements are
nitrogen, phosphorus, and potassium. Numerous researches have been conducted to study
fertigation application and fertigation effects on fruit quality and growth (Koo et al., 1984;
Rolston, et al., 1986; Robinson, 1990; Willis and Davies, 1991; Fleam, 1993; Boman,
1993). However, because of the complexity of the crop nutrient requirement, it is difficult
to obtain a generally accepted fertigation schedule. A typical orange grove might require
fertilizer in the following amounts (Jackson, 1991), This assumes a yield of 500 boxes per
acre.


121
Table 9.1 Input data of CIMS.
Weather
Crop
Irrigation
Soil
Solar radiation
(kW/m2)
Tree Coverage (%)
Emitter number per
tree
Soil series
Maximum daily
temperature (F)
Root depth (in)
Emitter flow rate
(gal/hr)
Permanent
wilting point
(ft/ft)
Minimum daily
temperature (F)
Young and mature
tree
Wetted diameter (ft)
Field capacity
(ft/ft)
Wind speed (mph)
Crop coefficient
Overall irrigation
efficiency (%)
Soil depth (ft)
Relative humidity
(%)
Management
allowed depletion
(%)
Rainfall (in)
93,3 Data Requirements of the Scheduling Module
The scheduling module provides the ability to activate or deactivate control valves
according to user-defined schedules. The user needs to input irrigation and fertigation
schedules to the system. Consequently, databases and data entry screens to specify irrigation
and fertigation schedules were developed.
9,4 Maintenance of CIMS
Expert system maintenance is extremely important for the success of the system. A
poorly designed system can increase the difficulty and cost of maintenance. The current


I certify that I have read this study and that in my opinion it Conforms to
acceptable standards of scholarly presentation and is fully adequate, in scope and quality,
as a dissertation for the degree of Doctor of Philosoffyy.
Louis H. Motz
Associate Professor of Civil
Engineering
I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is fully adequate, in scope and quality,
as a dissertation for the degree of Doctor of Philosophy.
AULft.^^
Allen G. Smajstrla
Professor of Agricultural and Biological
Engineering
This dissertation was submitted to the Graduate Faculty of the College of
Engineering and to the Graduate School and was accepted as partial fulfillment of the
requirements for the degree of Doctor of Philosophy.
August, 1995 f £ -\
h" Winfred M. Phillips
Dean, College of Engineering
Karen A. Holbrook
Dean, Graduate School


LIST OF REFERENCES
Adams, J. T. 1992. Microirrigation management for mature flatwoods citrus. The Citrus
Industry, July 7:44-45.
Anderson, C. E., H. P. Johnson, and W. L. Powers. 1978. A water-balance model for
deep loess soils. Trans, of the ASAE 19:314-320.
Augustin, B. J. and G. H. Snyder. 1984. Moisture sensor-controlled irrigation for
maintaining bermudagrass turf. Agron. J. 76:848-850.
Barfield, B. J., K. B. Perry, J. D. Martsolf, and C. T. Morrow. 1990. Modifying the
aerial environment. In Management of Farm Irrigation Systems, St. Joesph, MI,
p. 827-869.
Barrett, M. L. and A. C. Beerel. 1988. Expert System in Business: A Practical Approach,
Ellis Horwood, Chichester.
Batchelor, W. D. and R. W. McClendon. 1992. A blackboard approach for resolving
conflicting irrigation and insecticide scheduling recommendations. Trans, of ASAE
35(2):741-747.
Bobrow, D. G. and T. Winograd. 1977. An overview of KRL, a knowledge
representation language. Cognitive Science 1 (l):3-46.
Boman, B. J. 1993. A comparison of controlled-released to conventional fertilizer on
mature 'Marsh' grapefruit. Proc. Fla. State Hort. Soc. 106:1-4.
Bonissone, P. P. and R. M. Tong. 1985. Reasoning with uncertainty in expert systems.
International Journal of Man-Machine Studies 22(3):241-250.
Brachman, R. J. and H. J. Levesque (Editors). 1985. Readings in Knowledge
Representation. Morgan Kaufmann, Los Altos, CA.
Bronowski, J. 1965. The identity of man. Penguin Books, Bergenfield, ND.
Buchanan, B. G. and E. H. Shortliffe. 1984. Rule-Based Expert Systems. Addison-Wesley
Publishing Company, Reading, MA.
189


146
Test case 16
(TREE-STATUS MATURE)
(CRITICAL GROWTH)
(sensor-reading idl 6 15 12 14 24 13)
(sensor-reading id2 6 17 12 14 24 13)
(time-constrain no)
(weather-constrain no)
Test case 17
(TREE-STATUS MATURE)
(CRITICAL GROWTH)
(sensor-reading idl 6 14 12 15 24 13)
(sensor-reading id2 6 13 12 16 24 13)
(time-constrain no)
(weather-constrain no)
Test case 18
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 28 12 24 24 24)
(sensor-reading id2 6 24 12 24 24 24)
(time-constrain no)
(weather-constrain no)
Test case 19
(TREE-STATUS MATURE) Data OK
(NON-CRITICAL GROWTH) Turn on by ID 1 -1 or
(sensor-reading idl 6 28 12 24 24 24) ID2-3
(sensor-reading id2 6 24 12 24 24 26)
(time-constrain no)
(weather-constrain no)
Test case 20
(TREE-STATUS MATURE)
(NON-CRITICAL GROWTH)
(sensor-reading idl 6 24 12 28 24 24)
(sensor-reading id2 6 24 12 24 24 24)
Data OK
Turn on by ID 1-1
Data OK
Turn on by ID2-2
Data OK
Turn on by sensor ID1-1 or
ID2-1
Data OK
Turn on by ID 1-2


178
The prognosis result can be used as a reference for the users to define their own
irrigation schedules when irrigation is controlled by the user defined schedules.
To view the simulated soil-water content, click the Historical SWC on the submenu
of Simulation. A display appears in Figure 13.6.5. Then execute the following steps.
Click the upper left box to exit the screen.
Click Simulated SWC from the submenu of Simulation. A browse screen
will pop up as Figure 13.6.5.
Figure 13.6.5. Simulated soil-water content.
The simulated results include the following results:
Sdate date of the simulation,
Swc soil water content by volume, and
Depletion soil water depletion at the crop root zone in percent.
Click the upper left box to exit the screen or click the upper right arrow box
to maximize the window.


165
Click Apply Irrigation on the submenu of Scheduling (Figure 13.4.1). A
dialog screen appears as Figure 13.4.4.
Click the Cancel button to exit the window.
Click the Proceed button to apply irrigation based upon the user defined
irrigation schedule. A screen (Figure 13.4.5) displays the current time and
on or off status of the irrigation valves.
Double click left button of your computer mouse to interrupt the continuous
process.
Valve ON/OFF Status
it
ON
OFF
T to
e*a Sk.t>
Tto Day
¡i
m
OFF
mm¡M
lima
End Slop
Tim* Day
#
ON
OFF
Start
Tto
Cnd 6
Piiiil
kip
ip
t
OFF : f
8
8
8
2
0
¡181
OFF
3
9
9
8
8
OFF
8
8
9
9
§1
1
OFF
9
D
9
4
28
OFF
3
$
8
8
ill
l
OFF
3
9
HI
8
1

OFF
I
8
."8
i
0
29
OFF
3
0
9
8
ill
'-36-'
OFF
8
8
3
8
i
*
III
8
11
8
2$
OFF
B
8
8
a
8
38
OFF ;
3
3
9
8
¡§1

OFF
8
0
8
2
8
a*
OFF"
111
0
S
8
0
92
OFF
3
6
3
lit
fH
m
-'OFF'
8
8
3
0
8
2?
OFF
3
8
3
a
fil
93
OFF--;-;
s
3
9
s
1
m
OFF
9
C
S
8
8
28
OFF
S

fill
8

OFF
3
9
9
8
tli
1
OFF
¡111
8
3
8
23
OFF
3
a
a
n
SO
OFF
8
3
3
8

i
OFF-'
18
Hi
III
0
8
39
OFF
3
0
8
a
a
61
OFF
8
9
1
8
Pi
8
OFF
8
0
8
0
8
31
OFF '
3
8
3
8
8
$2
OFF
3
9
3
8
g
1
OFF
8
8
8
8
8
:
OFF
3
1
3
8
8
53
OFF
3
9
9
8
!!
OFF
8

mi
i
8

OFF
3

Hi
8
8
¡¡I
OFF
8
9
iff
8
¡1
m
OFF
3
111:
8'
ill
8
m
OFF
3
8
3
8
ill
S3'
OFF
3
9
9
8
¡I
m
OFF
ill
0
3
0
nil
m
I
3
8
'
8
8
S3
OFF
3
9
9
m
8
B
6
m
OFF
OFF
OFF
HI
8
3
0
ill
8
- 3
111
9
8
8
8
8
8
8

V
88
OFF
¡If
.'Off!
3
3
8
8'
¡II
8
Hi
-'3¡
3
111
0
8
H
8
8
58
se

Off
iilii
UP
#lil
3
9
9
9
3
9
5
lit
3
¡ii
3
If
ii
ii
Figure 13.4.5. Irrigation valve on or off display.
13,4,3 Apply Fertigation
Similarly, fertigation can be applied according to user defined fertigation schedules.
The program checks user defined fertigation schedule against the computer date and time
to turn on or off the control valves and fertigation pump. Fertigation is applied in a
sequence of preinjection, fertigation, and flush.


100
8.4 Irrigation Management
Irrigation management involves daily decision-making on (1) irrigation strategies,
(2) when to start an irrigation, and (3) when to stop an irrigation.
8.4.1 Irrigation Strategies
Irrigation strategies can be classified as full or deficit irrigation. Full irrigation
provides 100 percent of irrigation demands. In practice, most irrigations are managed as
full irrigation. Deficit irrigation refers to a strategy under which crops are deliberately
allowed to sustain some degree of water deficit. The fundamental goal of deficit irrigation
is to increase water use efficiency and reduce energy costs, while controlling water stress
and yield losses. Irrigation can be managed by user selected full or deficit irrigation, or by
an irrigation strategy obtained from the expert reasoning process. Figure 8.9 illustrates the
decision-making process required to determine an irrigation strategy.
Full irrigation
Young trees are normally managed to grow rapidly for early production. For mature
trees, irrigation during the critical growth stage is crucial to yield. Thus, full irrigation is
applied for young trees and mature trees during the critical growth stages.
Deficit irrigation
Although deficit irrigation potentially increases water and energy saving, this
strategy increases difficulty of irrigation management. In particular, uncertainty of rainfall
is one of the factors.


159
the Expert in the main menu (Figure 13.2.6). A submenu of Execute RTES and View
Applications will be displayed. Then complete the following steps.
Figure 13.2.6. Submenu of the Expert main menu.
Ini
RTES Execute Screen
Select modules to execute
0 Irrigation
Exl Fertigation
f<] Cold Protection
1
Execute
j
E3 Initialize
1
Note:
To execute this RTFS, soil
moisture
sensor and weather station must
1
Figure 13.2.7. Execution screen of the RTES.
Click Execute RTES and a dialog window appears as Figure 13.2.7.
Select the check box if necessary from the dialog window. The check boxes include
Irrigation, Fertigation, and Cold protection. If the box is checked, the RTES will
perform the reasoning process on this subject. The check box of Initialize should
always be selected.
Click the Cancel button to exit the screen and click the Help button to view the help
messages regarding this screen.
Click the Execute button to run the RTES. The knowledge base will be loaded into
memory and the external devices are controlled by the conclusions of the decision-


9
primary method for measuring matric potential in soil. They have a fairly fast response time
when used for irrigation (Towner, 1980; Stone et al., 1986). A pressure transducer can be
installed on a tensiometer and interfaced to a data acquisition or readout system to realize
automation. The use of tensiometers with pressure transducers for soil-water potential
measurement has been successful in many applications (Fitzsimmons and Young, 1972).
The advantages of tensiometers are (1) low cost and easy construction, (2) easy
installation and maintenance, (3) long periods of operation if properly maintained, and (4)
adaptable to automatic measurement with pressure transducers. The disadvantages are (1)
a limited range of 0 to -0.8 bar that is not adequate for some soils, (2) hysteresis, and (3)
potential breakage during installation and cultural practices.
2,3 Irrigation Scheduling
Irrigation scheduling requires making decisions on when to irrigate and how much
water to apply. The main techniques used for scheduling include (1) monitoring of soil
moisture, (2) physiological indicators, and (3) soil water balance models. Proper irrigation
scheduling should result in savings of water and energy without yield reduction. Irrigation
scheduling decisions may relate to crop response to water stress, management objectives,
water quality control, system constraints, and public policies. Maintaining adequate soil
moisture levels in the crop root zone is critical for crop growth. Inadequate soil moisture
not only limits water supply to the roots, but also reduces root conductivity directly
(Wiersum and Harmanny, 1983).