Optimization of Copper Fungicide Application Timing for Citrus Groves in Florida

MISSING IMAGE

Material Information

Title:
Optimization of Copper Fungicide Application Timing for Citrus Groves in Florida
Physical Description:
1 online resource (64 p.)
Language:
english
Creator:
Zortea, Tiago
Publisher:
University of Florida
Place of Publication:
Gainesville, Fla.
Publication Date:

Thesis/Dissertation Information

Degree:
Master's ( M.S.)
Degree Grantor:
University of Florida
Degree Disciplines:
Agricultural and Biological Engineering
Committee Chair:
FRAISSE,CLYDE WILLIAM
Committee Co-Chair:
ASSENG,SENTHOLD
Committee Members:
DEWDNEY,MEGAN M
DANKEL,DOUGLAS D,II

Subjects

Subjects / Keywords:
citrus -- copper -- decision -- florida -- fungicide -- model -- optimization -- webtool
Agricultural and Biological Engineering -- Dissertations, Academic -- UF
Genre:
Agricultural and Biological Engineering thesis, M.S.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract:
Copper fungicides are commonly used for protective applications against foliar fungal and bacterial diseases in citrus groves. Management of these products must be finely balanced between disease prevention, application costs, fruit blemishes caused by copper phytotoxicity, and toxic accumulation of copper in the soil. The traditional schedule for copper sprays in Florida is an every 21-day post-bloom application. However, our computer simulation analysis showed that this traditional schedule is inefficient; it leaves the grove unprotected in wet years and applies unnecessary copper sprays in dry years. In order to facilitate the copper management for citrus growers, a user-friendly internet-based decision support system was developed. This system is capable of estimating the copper residue on the fruit based on rainfall records and spray details. This information allows producers to plan the copper applications in order to minimize unprotected periods while avoiding unnecessary applications in dry years. For growers who are distant from weather stations or that cannot quickly adjust their schedules according to the web tool recommendations, we developed a schedule with varying application intervals or spray concentrations. These schedules were calculated with the objective of minimizing the number of unprotected days according to historic weather data.
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Bibliography:
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility:
by Tiago Zortea.
Thesis:
Thesis (M.S.)--University of Florida, 2013.
Local:
Adviser: FRAISSE,CLYDE WILLIAM.
Local:
Co-adviser: ASSENG,SENTHOLD.

Record Information

Source Institution:
UFRGP
Rights Management:
Applicable rights reserved.
Classification:
lcc - LD1780 2013
System ID:
UFE0046396:00001


This item is only available as the following downloads:


Full Text

PAGE 1

1 OPTIMIZATION OF COPPER FUNGICIDE APPLICATION TIMING FOR CITRUS GROVES IN FLORIDA By TIAGO ZORTEA A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2013

PAGE 2

2 2013 Tiago Zortea

PAGE 3

3 To my family and everyone else who have he lped me be the better person I am today

PAGE 4

4 ACKNOWLEDGMENTS I am immensely grateful to my advisor Dr. Clyde Fraisse along with Dr. Willingthon Pavan for providing me the opportunity to study in the United States which greatly broadened my perspectives. Dr. Fraisse performed well beyond his expected duties as advisor by having an intense ca re for the well being of me and other students advised by him. I also give thanks to the other committee members, Dr. Senthold Asseng, Dr. Douglas D. Dankel and especially to Dr. Megan Dewdney wh o considerably helped in improving this study. I owe my dearest thanks to my m om for being a model of perseverance and pur suit to improvement in my life W ithout her I would not be able to withstand the long journey and all the failures that inevitably happen in life. I thank my d ad for our endless anal ytical conversations about the most varied topics; these conversations are for sure the very reason why I am so interested in science which defines me today. I also thank my girlfriend, Jennifer Grace and my sisters Laisa and Cindi Zortea for bringing s o much joy and support to my life. I extend my gratitude to Leonilce G irardi for greatly improving me as a person. Finally I would like to thank my fellow graduate students for the ir assistance and encouragement and the Citrus Initiative for funding this p roject.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 7 LIST OF FIGURES ................................ ................................ ................................ .......... 8 ABSTRACT ................................ ................................ ................................ ................... 10 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 12 Objective ................................ ................................ ................................ ................. 14 Citrus Production in Florida ................................ ................................ ..................... 15 The Copper Residue Model ................................ ................................ .................... 16 Disease Management Models for Citrus ................................ ................................ 17 AgroClimate ................................ ................................ ................................ ............ 18 Summary ................................ ................................ ................................ ................ 19 2 SYSTEM DEVELOPMENT AND SIMULATION METHODS ................................ ... 20 Understanding the Copper Model Java Source Code ................................ ............. 20 Web based Interface Development on AgroClimate ................................ ............... 23 Analysis of the Copper Residue Using Historical Data ................................ ........... 26 Meteorological Data ................................ ................................ ......................... 26 Analysis of Fruit Protection Based on the Traditional 21 Day Schedule ........... 28 Analysis of Fruit Protection Based on the Web Tool Recommendations .......... 29 Sensitivity Analysis of The Model to Application Parameters ........................... 29 Copper Application Schedule Optimization ................................ ............................. 30 Optimization of Fruit Protection Using a Varying Interval Schedule ................. 31 Optimization of Fruit Protection Using a Varying Conce ntration Schedule ....... 33 Summary ................................ ................................ ................................ ................ 35 3 RESULTS AND DISCUSSION ................................ ................................ ............... 36 Web tool ................................ ................................ ................................ ................. 36 Model Evaluation ................................ ................................ .............................. 38 Sensitivity Analysis ................................ ................................ ........................... 42 System Evalu ation ................................ ................................ ............................ 44 Web tool Usage Statistics ................................ ................................ ................ 46 Dynamic Optimized Schedules ................................ ................................ ............... 50 4 CONCLUSIONS ................................ ................................ ................................ ..... 54

PAGE 6

6 APPENDIX COPPER RESIDUE MODEL TRANSLATED TO R ................................ ...................... 56 LIST OF REFERENCES ................................ ................................ ............................... 61 BIOGRAPHICAL SKETCH ................................ ................................ ............................ 64

PAGE 7

7 LIST OF TABLES Table page 2 1 Copper application residue formulas extracted from the copper spray scheduling recommendation system (CuSSRS) source code. ........................... 21 2 2 Parameters for calculation of fruit area extracted from the Copper Spray Schedul ing Recommendation System (CuSSRS) source code. ......................... 22 2 3 Copper residue reduction formulas as a function of rainfall events separated into daily cumulative levels as extracted from the copper model source code. .. 22 2 4 Traditional 21 day spray schedule with early, average, and late peak bloom scenarios. ................................ ................................ ................................ ........... 29 3 1 Number of unprotected days as determined by the copper residue simulation using 56 years of weather data for each region with a 21 day application schedule and average peak bloom date (March 20). ................................ .......... 38 3 2 Number of un protected days as determined by the copper residue simulation using 56 years of weather data for each region with a 21 day application schedule and average peak bloom date. ................................ ............................ 39 3 3 r ................................ ................................ ....... 45 3 4 Schedules resulting from the interval optimization algorithm. These results consider all years of available weather data and all locations average. ............ 51 3 5 Schedules resulting from the variable concentration optimization algorithm. These results consider all years of available weather data and all locations ave rage ................................ ................................ ................................ ............. 52

PAGE 8

8 LIST OF FIGURES Figure page 1 1 Florida citrus production by county in 2009 10 according to the Florida Department of Agriculture and Consumer Services (FDACS, 2011). ................. 16 2 1 Simulation steps of the copp er spray scheduling recommendation system (CuSSRS) showing the daily loop required to estimate daily copper residue levels. ................................ ................................ ................................ ................. 20 2 2 Components diagram of a copper residue simulation on the developed web tool. ................................ ................................ ................................ ..................... 25 2 3 Selected National Weather Service (NWS) Cooperat ive Observer Program (COOP) stations for historical data analysis in Florida. ................................ ...... 27 2 4 Steps of the program created to analyze combinations of different intervals between each copper application. ................................ ................................ ...... 33 2 5 Steps of the program created to analyze combinations of different concentrations on each copper application. ................................ ....................... 34 3 1 The citrus copper application scheduler on the AgroClimate website (July 2012). ................................ ................................ ................................ ................. 37 3 2 Copper residue simulation for Hendry County in 2008 using the 21 day schedule and typical spray parameters. The cros ses are residue on grapefruit and dots are residue on mandarins. ................................ .................. 40 3 3 Comparison between the 21 .. ................................ ................. 41 3 4 Number of unprotected days summed across 56 years of every weather station. Each data point shows the simulated results varying only the spray volume from 467 to 46 76 L ha 1 . ................................ ................................ ........ 43 3 5 Number of unprotected days summed across 56 years of every weather station. Each data point shows the simulated results varying only the spray concentration from 0.56 to 4.48 kg ha 1 by increments of 0.056 kg ha 1 ............ 44 3 6 Copper residue simulation using worst case scenario plant parameters, mandarin fruit inside the canopy, 0.84 kg ha 1 metallic copper concentration, 1170 L h a 1 volume, and P olk County weather data of 2005. ............................. 46 3 7 Map of Florida showing number of unique visitors of the Copper web tool pr areas. ................................ ................................ ................................ ................. 47

PAGE 9

9 3 8 Plot of a Gaussian kernel density estimate o f the 1460 bloom dates recorded by the web tool. The vertical line marks March 20 th which is the suggested average bloom date. ................................ ................................ ........................... 48 3 9 Plot of a Gaussian kernel density estimate of 3259 spray volumes recorded by the web tool. The vertical line marks 1170 L ha 1 (125 gal ac 1 ) concentration which is the current recommendation. ................................ .......... 49 3 10 Plot of a Gaussian kernel density estimate of 3259 spray concentrations recorded by the web tool. The vertical line marks 0.84 kg ha 1 (0.75 lb ac 1 ) concentration which is the curren t recommendation. ................................ .......... 50 3 11 Plot the average unprotected days of each schedule produced by both interval optimization (continuous lin e) and concentration optimization (dashed line) for the average bloom date scenario. ................................ ......................... 53

PAGE 10

10 Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science OPTIMIZATION OF COPPER FUNGICIDE APPLICATION TIMING FOR CITRUS GROVES IN FLORIDA By Tiago Zortea December 2013 Chair: Clyde William Fraisse Major: A gricultural and B iological E ngineering Copper fungicides are commonly used for protective applications against foliar fungal and bacterial diseases in citrus groves. Management of these products must be finely balanced between disease prevention, application costs, fruit blemishes caused by cop per phytotoxicity, and toxic accumulation of copper in the soil. The traditional schedule for copper sprays in Florida is an every 21 day post bloom application. However, our computer simulation analysis showed that this traditional schedule is inefficient ; it leaves the grove unprotected in wet years and applies unnecessary copper sprays in dry years. In order to facilitate the copper management for citrus growers, a user friendly internet based decision support system was developed. This system is capable of estimating the copper residue on the fruit based on rainfall records and spray details. This information allows producers to plan the copper applications in order to minimize unprotected periods while avoiding unnecessary applications in dry years. For growers who are distant from weather stations or that cannot quickly adjust their schedules according to the web tool recommendations we developed a schedule with varying application intervals or spray concentrations. These schedules were calculated

PAGE 11

11 with the objective of minimizing the number of unprotected days according to historic weather data.

PAGE 12

12 CHAPTER 1 INTRODUCTION Copper (Cu) compounds are the most widely used fungicides or bactericides in Florida citrus for the management of foliar diseases (Albrigo et al. 2005 ; Graham et al., 2010; Graham et al., 2011). Traditionally, copper fungicides have been used to manage diseases such as melanose (c aused by Diaporthe citri ), Alternaria brown spot (caused by Alternaria alternata ), citrus scab (caused by Elsino fawcettii ) and greasy spot rind blotch (caused by Mycosphaerella citri ) (Dewdney et al., 2012a). Particularly for melanose, sufficient copper residue must be present to protect fruit from petal fall until mid July, when the fruits are no longer susceptible to these diseases (Albrigo et al., 2005). More recently, two new diseases were introduced to Florida; Asian citrus canker, caused by the bacterium Xanthomonas citri subsp. citri and citrus black spot, caused by the fungus Guignardia citricarpa ( Spann, T.M., 2008; Schubert et al., 2012). Copper applications are an essential part of the management programs for these diseases (Sc hutte et al., 1997; Graham et al., 2010; Graham et al., 2011; Dewdney et al., 2012a; Dewdney et al., 2012b). However, citrus is susceptible to black spot and citrus canker until September and October, respectively, and not much is known about how copper re sidue decay is affected by the high summer rainfall common in Florida. Copper as any agricultural input has to be correctly dimensioned. If the concentration of copper is too high, it can frequently cause or accentuate market value reducing blemishes due to copper stippling (phytotoxic burn) from excessive copper ion uptake by the fruit rind cells, especially at temperatures above 34.5C (Schutte et al., 1997; Timmer and Zitko, 1998). Furthermore, toxic levels of copper can build up in soil due to multipl e, high concentration applications of copper over many years (Alva et al.,

PAGE 13

13 1993; Graham et al., 1986). In older groves where copper has been used for many years, the soil can contain up to 370 kg ha 1 metallic copper (Timmer and Zitko, 1996). High copper c oncentrations in the soil can slow growth, thin canopies, darken fibrous roots and cause foliar iron deficiency, particularly on acid soils (Alva et al., 1993; Graham et al., 1986). Historically, it was recommended to use one or two applications of 9 kg ha 1 metallic copper for foliar disease management (Timmer and Zitko, 1996). It was shown that lower rates of copper fungicides could give the same disease management efficacy as the higher rates (Timmer et al., 1998) and that splitting the applications, wit hout increasing the total copper used per year improved disease management (Timmer and Zitko, 1998). These findings and other studies were used to better understand the behavior of copper as a fungicide or bactericide in Florida citrus groves (Albrigo et a l., 1997; Timmer and Zitko, 1996; Timmer et al., 1998) and to develop a copper spray scheduling recommendation system (CuSSRS) (Albrigo et al., 2005) to aid growers in scheduling copper fungicide sprays for early season disease management. The CuSSRS was e valuated by comparing predicted residue levels to actual copper residue levels in the field. Disease severity in plots sprayed following the CuSSRS predictions were compared with a standard 21 day calendar schedule and an unsprayed treatment (Albrigo et al 2005). The traditional 21 day schedule is the currently recommended application schedule (Dewdney and Graham, 2012) and is commonly used by Florida citrus growers, especially for melanose and/ or canker management. Although the CuSSRS was shown to effect ively reduce cost and improve coverage (Albrigo et al 2005), it was not widely used by citrus growers. The reasons

PAGE 14

14 given by growers for not adopting the system include d a confusing interface, too many inputs, difficult to install, unclear output and lack of updates. A probl em in the routine which connected to the Florida Automated Weather Network (FAWN; http://fawn.ifas.ufl.edu/) for real time weather information across the state was another problem that made the copper system difficult to use. To revive this valuable tool, a project was initiated to develop a web based version of CuSSRS with a simple and self explanatory interface to allow growers to estimate copper residue in their groves and analyze the results without external aid. The connection betwe en weather data and disease models is an essential step in order to enhance the contribution of these models to the producers ( Guillespie and Sentelhas 2008 ). Objective Our primary objective was to help Florida citrus growers better schedule copper applications and maximize fruit protection while reducing environmental impacts and production costs. This objective has to be fulfilled with a pr actical interface aiming to require the least possible time commitment from the producer. Specific objectives included: To understand and review the algorithms used in the CuSSRS model and the sensitivity of the model to the various inputs To translate the original CuSSRS model to the R statistical language (R Developme nt Core Team, 201 1 ). To analyze the variab ility of copper residue coverage using historical meteorological data for citrus producing areas in Florida and current schedule recommendations. Develop a practical web tool which allows producer s to simulate the remaining copper residue with minimal effo rt. To analyze the potential benefits of the developed web tool based on simulations.

PAGE 15

15 Develop optimizations for the current recommendations using historical weather data. Citrus Production i n Florida Citrus production is important to the Florida economy. D uring the 2010 2011 season, Florida produced more than 63% of United States citru s in a combined area of 203,799 ha for all types (USDA, 2011). Approximately 84% of the Florida citrus production is processed for juice. The estimated production value of Flo rida Citrus in the 2010 2011 season was US$ 1,573,116 ,000 which represents 52% of total citrus production value from United States (USDA, 2011). Florida Citrus production is principa lly located i n central and south western Florida. Florida citrus growers in 2009 2010 produced 133.7 million boxes (40.8 kg box 1 ) of sweet oranges 96 % being used for juice and 20.3 million boxes (38.5 kg box 1 ) of grapefruit of which 54 % were used for grapefruit juice. Other citrus types grown in Florid a include specialty fruit like mandarin (tangerine) hybrids and Navel orange. The specialty fruit industry is concentrated on the Central Ridge of Florida in Lake, Orange and Polk Counties and many grapefruit plantings are on the East coast in Indian River and St. Lucie Counties (Fig 1 1)

PAGE 16

16 Figure 1 1. Florida citrus production by county in 2009 10 according to the Florida Department of Agriculture and Consumer Services (FDACS, 2011). The Copper Residue Model The CuSSRS model was created to simulate the copper residue decay on citrus fruit over time. It was developed with the Java programming language as a stand alone application and integrated within a larger citrus planning and scheduling program called DISC (Decision Information System for Citrus) (Beck, 2006). The residual copper is calculated on a daily basis from the most recent application relying on inputted spray details and daily rainfall. The required input information includes bloom date, cultivar, details on copper spray applications such as concentration and spray volume, fruit position, and daily

PAGE 17

17 rainfall data. The different copper concentrations and volumes produce dissimilar residue levels on the fruit. Fruit growth and rainfall events are used to simulate the amount of copper residue d ecreased by fruit surface expansion and weathering of the residue layer. The model simulates the copper residue for fruit located inside and outside of the tree canopy. The copper deposition and rainfall events affect the outer fruit more intensely than th e inner fruit (Albrigo et al., 2005). Interior fruit and fruit on the top of the tree receive lower copper deposits from commonly used spray equipment. On the other hand, there is less rainfall removal of copper for interior fruit or less disease pressure for the top of the tree when considering melanose (Albrigo et al., 2005). Very low spray diluent volumes can lead to excessive deposits on exterior surfaces of outer fruit which can cause copper stippling in addition to poor disease management on interior fruit (Albrigo et al., 1997). Increasing diluent rate produced a more uniform coverage along the tree but also increased the copper lost by run off (Albrigo et al., 1997). The copper residue threshold for reapplication was based on the recommended values n eeded to provide a complete protection safety margin. CuSSRS adopts a default warning threshold of 0.5 g cm 2 and a danger threshold of 0.25 g cm 2 The minimum residue in which the grove is still considered protected is 0.1 g cm 2 (Albrigo et al., 2005). Disease M anagement M odels for Citrus Other systems have been proposed to help citrus growers better manage diseases. The A L TER RATER is a weather based model with the objective of help producers correctly time fungicide sprays for Alternari a management ( Bhatia, A., 2002; Timmer et al., 2001) It is based on a cumulative score which is influenced by

PAGE 18

18 rain fall leaf wetness and temperature. The system does not have a web interface and requires the producer to manually fill the weather data in a table. Also a model was developed fo r management of Post bloom fruit drop, caused by Colletotrichum acutatum (Timmer et al., 1996). This model has shown to produce accurate predictions but requires considerably more information beyond weather data. AgroCl imate AgroClimate is a web based climate information and decision support system (http://www.agroclimate.org) (Fraisse et al., 2006) developed to help agricultural producers reduce risks associated with climate variability in the southeastern U.S.A. (Frais se et al., 2006). It is periodically updated and maintained to ensure up to date information and the simplest possible interface. A mobile version is also available when the AgroClimate website is accessed from a mobile device. It was designed and implemen ted by the Southeast Climate Consortium (SECC http://seclimate.org) in partnership with the Florida Cooperative State Extension Service. The system was developed to be hosted in Linux/Unix platforms but can easily be transferred to others. The dynamic tool s were developed using the PHP (Hypertext Preprocessor) web programming language, Javascript language, HTML, Cascading Style Sheets (CSS) and MySQL database (Pavan et al., 2011). Decision support tools available in AgroClimate include: (a) Climate risk tools: expected (probabilistic) and historical climate information as well as freeze risk at the county level; (b) Crop yield tools: expected yield based on soil type, planting date, and basic management practices for corn, cotton, peanut, potato, and tomato, and historical county and regional yield databases; (c) Crop disease tools: disease risk monitoring and forecasting for anthracnose and botrytis fruit rot in strawberry, peanut l eaf spot, and the

PAGE 19

19 citrus copper application scheduler; (d) Crop development tools: monitoring and forecasting of growing degree days and chill accumulation; (e) Drought monitoring tools: monitoring and forecasting of the Agricultural Reference Index for Dr ought (ARID), Keetch Byram (KBDI), and the Lawn and Garden (LGMI) drought indices; and (f) Footprint tools: carbon footprint of selected fruits and water footprint of cereal crops. AgroClimate provides climate forecasts and outlooks, monthly climate summar ies, crop management options to mitigate climate associated risks for pasture, forestry as well as certain crops and fruits. It also includes background information about the main drivers of climate variability and basic information about climate change in the Southeast USA. Summary This chapter described the objectives of this study and introduced background information about the citrus production in Florida. T raditional copper management practices and the copper model used in the simulations contained in this study were also introduced Chapter 2 details the copper model inner equations and the methods used for the simulation s Also in Chapter 2, it is described the development process of the proposed web tool for copper residue management. Simulation r esu lts and the copper residue web based tool are discussed in Chapter 3 Chapter 4 includes our main conclusions and recommendations for future developments.

PAGE 20

20 CHAPTER 2 SYSTEM DEVELOPMENT AND SIMULATION METHODS Understanding the Copper Model Java Source Cod e F igure 2 1 shows that the first step i n the extracted copper residue model was the calculation of the copper deposition provided by the first spray application. The copper residue is always zero at the beginning of the season because the previous s fruit have been harvested by time of application in the spring. The applied copper residue depends on the fruit area available, volume and concentration of the copper suspension, and the position of the fruit, inside or outside the tree canopy ( T able 2 1). Figure 2 1 Simulation steps of the copper spray scheduling recommendation system (CuSSRS) showing the daily loop required to estimate daily copper residue levels.

PAGE 21

21 Table 2 1 Copper application residue formulas extracted from the copper spray scheduling recommendation system (CuSSRS) source code. Fruit p osition Volume (L ha 1 ) Formula to estimate c opper residue [a] Inside <1,169 Inside >=1,169 and <=2,338 Inside 2,338 Outside <1,169 Outside >=1,169 and <=2,338 Outside >2,338 [a ]The volume (V) is in L ha 1 and concentration (C) is in kg ha 1. Area on the day of application (A) is provided by A Gompertz growth function using the parameters described in T able 2 1 was used to estimate fruit growth. Because it is an empirical approximation, the model does not use weather data to provide a more precise estimation of fruit area. An idealized growth curve for each scion is given by Equation 2 1 using the variables and parameters found in T able 2 2 The idealized growth curve is calculated as follows: ( 2 1) w here AREA is fruit surface area in mm 2 T is the sum of the current Julian day with the regression offset MAX is the maximum measured AREA M IN is an arbitrarily small value and B is the parameter for each scion type ( T able 2 2 ).

PAGE 22

22 Table 2 2 Parameters for calculation of fruit area extract ed from the C opper S pray S cheduling R ecommendation S ystem (CuSSRS) source code. Cultivar Regression offset (Julian Day) MAX (mm 2 ) MIN (mm 2 ) B (unitless) Grapefruit 73 22,650 645 X 10 12 0.0220 69 14,949 645 X 10 12 0.0222 Mandarin 77 14,263 645 X 10 12 0.0198 64 19,856 645 X 10 12 0.0214 Each day post application, the residue is reduced proportionally to the fruit surface area. The copper residue in g cm 2 of a given day is calculated as follows: ( 2 2) W h ere DEPO is the initial residue from the most recent application and AREA is the Equation 2 1. Rainfall events are responsible for rapid copper residue loss. The loss is proportional to the rainfall amount on a given day an d the remaining residue. Table 2 3 shows how rainfall intensities have different effects on residue loss. The residue also slowly decreases over time because of the increase in the AREA value in Equation 2 2 while DEPO remains constant over time until there is a new application. Table 2 3 Copper residue reduction formulas as a function of rainfall events separated into daily cumulativ e levels as extracted from the c opper m odel source code. Rainfall (mm) Reduction [a] > 0 and <= 127 > 127 and <= 508 > 508 [a]Where R is daily rainfall in millimeters and RESIDUE is calculated by R esidue thresholds can be adjusted by the user in the CuSSRS model For this study a residue threshold of 0.25 g cm 2 to Albrigo et al. (2005), at the 0.25 g cm 2 till ha s complete

PAGE 23

23 reached since a strong weathering event could leave the grove unprotected. When the residue falls under 0.1 g cm 2 a grove is considered unprotected as th e remaining residue is not sufficient to keep the fruit protected (Albrigo et al. 1997; 2005). Web based Interface Development on AgroClimate The copper residue model was extracted from the CuSSRS J ava source code and translated into the R language with th e objective of facilitating the production of high quality graphs, its integration into the AgroClimate web server, and the execution of statistical analysis. The R statistical analysis software system (http://www.r project.org) is a language and environme nt for statistical computing and graphics generation. It is an open source implementation of the S language developed at Bell Laboratories by John Chambers and colleagues (R Development Core Team, 201 1 ). R is multiplatform, which means that it can be run o n all modern operating systems such as UNIX, Linux, Windows, and Macintosh. Being able to run under Linux is an important feature when integrating code with websites as the majority of web servers operate in Linux. R also provides a well developed programm ing language and a self contained environment to perform a wide range of statistical analyses. As a programming language, R is highly expansible allowing it to be easily adapted to new tasks that are not part of the built in functionality ( R Development Co re Team, 201 1 ). Along with correcting the communication issues that prevented the CuSSRS to retrieve real time weather data, the Citrus Copper Application Scheduler was intended to have a more functional and easier to understand interface consequently bro adening its use by citrus producers. With daily copper residue information, it is possible for growers to better time application decisions and reduce the number of unprotected

PAGE 24

24 periods. Lapses in residue protection can allow fruit infection to occur by pla nt pathogens decreasing their market value. The tool was designed to be as simple as possible using only free and open source components following the AgroClimate website guidelines. The website runs in PHP, HTML ( Hyper Text Markup Language ), CSS ( Cascading Style Sheets ) and JavaScript using the jQuery framework ( http://jquery.com/ ). Since it is available at a central server, it is possible to deliver updates and corrections with minimal delay. The model and plots ru n in the R language. The inputs are stored in a MySQL (http://www.mysql.com/) database. To further assist decision making, the tool is linked to the Citrus Pesticide Application Tool (http://fawn.ifas.ufl.edu/tools/pesticide/) available on FAWN that provid es rainfall forecasts and application conditions for the next 45 hours. Figure 2 2 shows a diagram of the different components of the copper residue simulation web tool The simulation starts with the grower providing spray and grove information ( F ig. 2 2 A ) on the PHP form. The user provided information is stored in the MySQL database ( F ig. 2 2 B) by the PHP algorithm which also generates a system call to the R program containing the simulation ( F ig. 2 2 C). The simulation information is then retrieved by th e R program from the database ( F ig. 2 2 D) which afterwards generates a HTTP call ( F ig. 2 2 E) to the FAWN web service retrieving ( F ig. 2 2 F) information from the weather station selected by the user. The copper residue simulation is then executed based on the information retrieved. The resulting graph is saved as an image in a specific folder on the server ( F ig. 2 2 H), the numeric results are saved back on the database ( F ig. 2 2 G) After the simulation is finished the PHP code

PAGE 25

25 which was halted by the system call continues to run ( F ig. 2 2 I), it reads the results as well as the location of the graph on the server ( F ig. 2 2 J K ) and presents the simulation information to the user ( F ig. 2 2 M). Figure 2 2 Components diagram of a copper residue simulation on the developed web tool. All the experiments executed by the growers in the website stay anonymously logged in the database. Additionally the Google Analytics tool was installed in the website T his tool allows tracking of the website traff ic and users behavior. Using the logged information it was possible to create plots of the average bloom date, average concentration and average volume inputted in the webs ite. It was also used to generate a map of the number of user accesses to the website from Florida This map was grouped by a political division called m etropolitan area These divisions are related to

PAGE 26

26 densely populated areas which share the same infrastructure, it might encompas s several counties Analysis of the Copper Residue Using Historical D ata Having the Copper residue model translated to R enables the possibility of simulating scenarios using the historical meteorological data. These scenarios can produce valuable information regarding how different application approaches affected the copper re sidue in past years More specifically, it is possible to count the number of days in which the grove was unprotected, the number of sprays necessary to achieve optimal protection, the effect of each model parameter in the results and how different rainfa ll patterns affect the copper residue It is also possible to test if the current spray recommendations provide d adequate amounts of copper residue t h rough the whole vulnerability period in different locations. Meteorological D ata Daily precipitation data are a key input for the copper model as rainfall significantly reduces current residue levels on the fruit In this study two different meteorological data sources were used. The website tool uses observed data from FAWN weather stations located in citrus producing areas of the state FAWN data were used for in season simulation of copper residue levels and application scheduling. Historical analysis of copper application regimens needed a longer time series of daily weather data than the ones availabl e fr om FAWN stations. For that reason, f ifty six years of historical daily weather data from the National Weather Service (NWS) Cooperative Observer Program (COOP) were used ( http://www.nws.noaa.gov/om/coop ). Five locations in the Florida counties of Highlands, Hendry, Lake, Indian River, and Polk were selected for the historical analysis aiming for

PAGE 27

27 a representative picture of the major citrus producing regions ( F ig. 2 3 ). These counties annual citrus production (FDACS 2012). Figure 2 3. Selected National Weather Service (NWS) Cooperative Observer Program (COOP) stations for historical data analysis in Florida. Google and the Google logo are registered trademarks of Google Inc., used with permission. Each year of observation in a location w as classified according to its yearly accumulated rainfall as a wet, average, or dry year based on the first and third quartiles of the entire series Every year with an accumulated rainfall greater than 1500 mm was classified as wet for that given location. Conversely, years with less than 1130 mm of accumulated rainfall were classified as dry and all remaining years were considered to have had an average rainfall.

PAGE 28

28 Analysis of Fruit Protection Base d o n the Traditional 21 Day Schedule S everal simulation experiments were conducted using the copper residue model and observed weather data t o better understand the dynamics of copper residue decay and how different parameters affect c opper residue levels. It was important to de termine which combination of fruit position and scion define d the worst case scenario. These parameters have a small but noticeable effect o n the copper residue and knowing the worst case scenario allow ed the simulations to be restri cted by assuming that other scenarios would have superior protection. The criteria used to determine the worst case scenario wa s the number of unp rotected days (residue < 0.1 g cm 2 ) for different scions and fruit positions using the traditional 21 day ap plication schedule This approach slightly reduced t he precision of the analysis but greatly simplified the results by avoiding the need for different recommendation combination s for each fruit position and scion. T he simulation experiments produced in thi s study use d the typical spray parameters for Florida : 0.84 kg ha 1 (0.75 lb ac 1 ) metallic copper concentration and 1170 L ha 1 (125 gal ac 1 ) volume (Dewdney et al., 2012a) The period in which the residue level wa s evaluated for unprotected days extended from the first spray 3 weeks post bloom to the last day of July The data for CuSSRS stopped in early July because the fruit are no longer susceptible to melanose (Albrigo et al. 2005) the main disease of concern p rior to citrus canker and black spot. For simulation standardization, t h ree different peak bloom date scenarios were used to simulate early, average, and late bloom ( T able 2 4)

PAGE 29

29 Table 2 4. Traditional 21 day spray schedule with early, average, and late peak bloom scenarios. Event Early bloom Average bloom Late bloom [a] Bloom date 10 Mar. 20 Mar. 30 Mar. 1 st spray 31 Mar 10 Apr 20 Apr 1 st scheduled spray 21 Apr 1 May 11 May 2 nd scheduled spray 12 May 22 May 1 Jun e 3 rd scheduled spray 2 Jun e 12 Jun e 22 Jun e 4 th scheduled spray 23 Jun e 3 Jul y 13 Jul y 5 th scheduled spray 14 Jul y 24 Jul y N/A End of vulnerability period 31 Jul y 31 Jul y 31 Jul y [a] N/A not applicable Analysis o f Fruit Protection Based o n t he Web Tool Recommendations An R program was developed to compare the amount of copper used by the traditional 21 day copper schedule and the hypothetical case in which producers are able to spray whenever necessary (copper residue under 0.25 g cm 2 ) according to the citrus copper a pplication scheduler recommendations. For this comparison we used an average bloom period, copper concentration of 0.84 kg ha 1 spray volume of 1170 L ha 1 the mandarin scion, and 5 6 years of rainfall data for all the studied locations. The objective of the comparison was to measure how many copper applications would be necessary per year to achieve no unprotected days assuming that the web tool recommendations were followed to the le tter. Additionally, the difference in residue levels between dry, average and wet years were calculated. Sensitivity A nalysis of T he M odel to A pplication P arameters Aside from application dates, the spray concentration and volume are the only parameters th at can be easily changed by the producer to achieve better protection. Although recommendations for these parameters already exist (Dewdney et al., 2012a), it was important to understand the model sensitivity to them. The model was run by varying both meta llic copper concentration and diluen t volume one at a time small

PAGE 30

30 increments while all other parameters were kept the same. This approach is known as one parameter at a time Many other more complex sensitivity analysis approac hes have been proposed. These more complex approaches better cover the search space as in the Morris method (Morris et al., 1991) or use variance decomposition as in the FAST method (Schaibly et al., 1973) but these are better suited for large models with unpredictable interactions between the inputs. one parameter at a time range in which the parameters can vary. The spray volume range was varied from 467 to 4676 L ha 1 by increments of 9.3 L ha 1 and the concentration range was varied from 0.56 to 4.48 kg ha 1 by increments of 0.056 kg ha 1 These are the maximum and minimum values that were used by producers on the web tool. It was also necessary to define the default values for the other model parame ters while only one is varied in each simulation. The typical 21 day spray schedule, average bloom date, inside canopy fruit position, copper concentration of 0.84 kg ha 1 spray volume of 1170 L ha 1 the mandarin scion, and 5 6 years of rainfall data were used for all the studied locations. This sensitivity analysis approach also required relevant output to be selected. The sum of the number of unprotected days from the first spray to the end of July across all years of weather data and locations was selected. C opper A pplication S chedule Optimization Disease control is not always satisfactory with the traditional 21 day schedule and many Florida citrus farms encompass thousands of hectares or are scattered over wide areas. In the se cases, it is not always possible to quickly move equipment according to output from a daily model and a compromise was sought to improve the copper coverage over the traditional schedule for these operations on a set schedule. It

PAGE 31

31 was hypothesized that h istorical weather data along with the fruit growth and copper residue decay estimates from the CuSSRS could be used to develop a dynamic copper application schedule with fewer coverage gaps to improve disease management in the spring and early summer. Cons idering that e ach location had slightly different historical rain f all pattern a more optimized schedule for each independent region would exist This region specific optimization would produce better results than one optimized for all the regions. Howeve r there would be a total of 15 optimized schedules, one for each region and bloom date It was decided not to publish these results as they were overly complic ated for the producers and the additional protection benefit small. Optimization of F ruit P rotection U sing a V arying I nterval S chedule Based on initial observations from the model output, it was determined that the driving factors behind copper residue loss, fruit growth and rainfall, vary throughout the growing season. It was then possible to d evelop a fixed schedule that adjusted the interval between sprays during the growing season to account for different growing conditions and weather patterns that could result in more uniform protection. To further study this hypothesis, an R schedule optim ization algorithm was developed with the objective of generating and testing different scheduling strategies aimed at the best protection with model outputs and historical weather data. The generated schedules were ranked by the resulting fruit protection. More specifically the fruit protection is given by t he sum of unprotected days over every year from 1956 to 201 2 across the 5 studied locations. The simulations used the worst case scenario for fruit position and scion typical spray parameters and the f irst spray started 21 days

PAGE 32

32 after bloom (Fig. 2 4 ) The number of unprotected days was calculated also for the traditional 21 day schedule with the same parameter s to serve as a baseline scenario The number of schedule simulations that could be tested was limited by the substantial computational processing required for this task. The ideal scenario would be to vary each date for at least +6 to 6 days, but this approach for a 5 spray date season would create 371,293 different schedules. Considering it would be necessary to run all th ese schedules for each of the 56 ye ars for each of the 5 locations, it would create an impractical total of over 100 million simulations. A more manageable and sophisticated approach was to vary each schedule for +2 to 2 days re quiring only 3,125 whole year simulations T hen use this local minima result and re run it again starting out from the optimized schedule. This process was repeated each time using the b est result from the previous schedules until the resulting optimum schedule was the same as the starting schedule provided to the algorithm. It is very likely that the result of this approach is also the global minima because the best schedules usually are similar However, it can only be guar anteed to be the local minima in the last run range. For example, the first experiment for the early bloom date tested schedules with spray intervals ranging from (19, 19, 19, 19, 19) to (23, 23, 23, 23, 23). These extreme schedules were obviously not sui table as the first one would produce a period of low residue in July and the last one simply has too many days between applications leaving the grove vulnerable A mong all the possibilities there is likely a more efficient schedule than the 21 day fixed sc hedule. There was little difference between the residual copper losses of the different scion types which supported the contention that it was acceptable to create the

PAGE 33

33 optimized schedules based on the worst case scenario. With the worst case scenario as ou r test subject, the residue decay predictions for the other scion types are likely to be conservative. This means that if there is an error in the prediction, it is likely to encourage an application before needed but not allow a lapse in coverage. Figu re 2 4. Steps of the program created to analyze combinations of different intervals between each copper application. The proposed approach was not supposed to increase amount of copper applied compared to the amount that would otherwise be applied with t he 21 day schedule. For this reason in the case of late bloom, only 4 copper applications were simulated, 5 applications would have extended the protection past the vulnerable period (Table 2 4 ). Optimization of F ruit P rotection U sing a V arying C oncentrati on S chedule It is a common practice of Florida growers to combine several products in the spraying equipment tank in order to minimize costs. These products also have set schedules which might complicate the usage of the prop osed varying interval schedule

PAGE 34

34 For these cases a dynamic schedule was proposed which varies the concentration of each application and still has the fixed 21 day schedule. This approach is less effective because the residue lost per day is proportional to the current residue on the fruit. Consequently greater concentrations also prod uce greater loss of residue reducing the overall possible gain in protection. Figure 2 5 Steps of the program created to analyze combinations of different concentrations on each copper application. For the varying concentration schedule it was necessary to also vary the concentration of the first application which was fixed in the varying interval schedule. This approach increased the number of possible schedules by 5 times, however the number of valid proposed schedules was greatly reduced b y the fact that this approach was not suppos ed to increase the amount of copper applied consequently limiting the valid schedules to only those which fit this criteria. The concentrations were varied by

PAGE 35

35 0.122 kg ha 1 (0.10 lb ac 1 ) with a range of 0.224 k g ha 1 (0.20 lb ac 1 ) in each run of the optimization algorithm. Summary In this chapter it was presented a detailed description of the copper model which is used in this study including equations and overall structure. It was also described the methodologies and parameters used in the simulations for accessing copper protection performance in past scenarios. Additionally, the approach for optimizing both interval of application and copper concentration in each application optimizations was descri bed In chapter 3 it will be presented optimization results analysis of the current recommendations for Florida and the developed and implemented web tool for simulating copper residue levels.

PAGE 36

36 CHAPTER 3 RESULTS AND DISCUSSION Web t ool The web based ap plication developed for this study works as a practical interface for the producer with the citrus copper application scheduler. With the daily residue information, it is possible for producer s to make accurate and timely decisions regarding copper applica tions. T o operate t he system the user inputs the spray concentration and volume scion bloom date and weather data source ( F ig. 3 1 ) This source can be either the FAWN weather station closest to the grove or a comma separated value ( CSV ) file containing the user rain measurements. T button creates a graph that show s the daily copper residue from 3 days before first spray until a week after the current day. The blue bars indicate rainfall events and the red/ye llow areas are the danger and warning thresholds respectively. When the residue reaches the warning level, the grower is advised to plan for a n application. W hen the residue reaches the danger zone the grower is advised to make an application as soon as p ossible. Additional information ( Dewdney et al. 2012a).

PAGE 37

37 Figure 3 1 The c itrus c opper a pplication s cheduler on the AgroClimate website (July, 2012). The user also can see the model result s in table format or download them as a CSV file. A detailed screencast on default quantity of metallic copper per area and spray volume. To calculate the kilograms of metallic copper used it is necessary to multiply the percent metallic copper in a product (found on the label) by kg ha 1 used. The fruit position is not requested because it would not make sense to ask tha t question to a producer; instead the worst case scenario is always assumed (fruit inside the canopy).

PAGE 38

38 Model Evaluation Simulation results of the traditional 21 day schedule ( T able 3 1 ) indicate that of all scion combinations and fruit positions, the frui t inside the canopy of a mandarin type scion were the least protected with an average of 3.12 unprotected, under 0.1 g of copper cm 2 days per year. This combination of parameters was then considered to be the worst case scenario and used in all other s imulations in this study. Table 3 1 Number of unprotected days as determined by the copper residue simulation using 5 6 years of weather data for each region with a 21 day application schedule and average peak bloom date (March 20). Florida County Grapefruit Valencia M andarin T ot. [a ] M ax Avg. [a] Tot. Max Avg. Tot. Max Avg. Fruit inside the canopy Hendry 257 26 4.59 271 27 4. 84 282 2 7 5. 04 Highlands 153 21 2.73 167 21 2.98 1 7 1 21 3. 05 Indian River 102 13 1.82 1 1 6 14 2.0 7 1 1 9 1 4 2. 12 Lake 114 14 2.04 125 15 2. 23 138 1 5 2. 46 Polk 153 11 2.73 167 11 2. 98 174 11 3.11 Average 1 5 6 2. 78 1 70 3.02 1 7 7 3. 16 Fruit on the canopy surface Hendry 150 23 2.68 160 2 3 2.86 166 23 2.96 Highlands 74 12 1.32 78 12 1.39 83 15 1.48 Indian River 49 11 0.88 52 1 1 0.93 55 11 0.98 Lake 41 10 0.73 49 1 0 0.88 49 10 0.88 Polk 67 8 1.20 72 8 1.29 71 8 1.27 Average 76 1.36 84 1.47 85 1.51 [a] Tot. total Avg. average per year Table 3 2 shows the simulated maximum and average unprotected days using the worst case scenario parameters for each yearly accumulated rainfall classification. The minimum is not displayed because i t was 0 for all the proposed scenarios. There is

PAGE 39

39 a large difference in the number of unprotected days in wet years compared to dry years, with more unprotected days occurring in wet years. Table 3 2 Number of unprotected days as determined by the copper residue simulation using 5 6 years of weather data for each region with a 21 day application schedule and average peak bloom date. Florida County Dry [a] Average [a] Wet [a] Max Avg./year Max Avg./year Max Avg./year Hendry 4 1. 07 16 5. 58 27 7.18 Hig hlands 7 0 71 13 3. 22 21 4.62 Indian River 6 0. 81 13 2. 18 14 3.46 Lake 8 1 15 11 2. 21 15 4. 55 Polk 7 1. 29 10 2. 68 1 1 5.40 Average 1.00 3. 17 5. 04 [a] Classification based on the first and third quartiles of the yearly accumulated rainfall. Years with accumulated rainfall greater than 1500 mm were classified as wet. Years with accumulated rainfall less than 1130 mm were classified as dry and all remaining years were considered to have had an average rainfall. Figure 3 2 shows a residue simulation of a typical wet year 2008, for Hendry County showing the differ ence in residue decay between g rapefruit and mandarin. In this scenario, the grove would have been unprotected from June 26 th to July 2 nd and from July 15 th to the end of the season with the traditional 21 day application schedule. This was a typical case in which a varied application schedule would have provided much more efficient protection with the same amount of copper by delaying the first 3 applications. There were little differences between the residual copper decay of the different scions.

PAGE 40

40 Figure 3 2 Copper residue simulation for Hendry County in 2008 using the 21 day schedule and typical spray parameters (0.84 kg ha 1 metallic copper concentration and 1170 L ha 1 volume). The crosses are residue on grapefruit and dots are residue on manda rins. The red threshold is 0.25 g cm 2 of copper and the black line is 0.1 g cm 2 The blue bars are daily total rainfall. In contrast to wet years, a substantial amount of copper is often wasted in dry years. For example, F igure 3 3 shows the copper re sidue continuously accumulated from April to July, 1998. Yet in July even a small amount of rain washed off most of the copper residue because the residue loss is proportional to the residue present ( T able 2 1 ). This is in agreement with the current theor y that there is no reason to apply large amounts of copper in fewer sprays (Timmer et al., 1998). Based on the model, only 3 timed applications were able to keep sufficient copper residue levels thus avoiding 2 unnecessary copper applications ( F igure 3 3 ).

PAGE 41

41 Figure 3 3 Comparison between the 21 using 1998 Lake County weather data for mandarin types and typical spray parameters (0.84 kg ha 1 metallic copper concentration and 1170 L ha 1 volume) and the traditional 21 day schedule. The red threshold is 0.25 g cm 2 of copper and the black line is 0.1 g cm 2

PAGE 42

42 The citrus copper application scheduler can be used to estimate summer copper resid ue decay but it may not be as accurate as for early season estimates. The causes of these potential inaccuracies include the fruit growth curves which estimate growth between petal fall and early July, when growth is faster than in the summer After this period, our preliminary data showed a slower fruit expansion over the summer and fall. In addition, rainfall becomes more scattered with thunderstorms and potentially more intense during the summer compared to spring and early summer. It is not known how t he slowing of fruit growth along with the seasonal change in rainfall patterns can affect copper residue levels. Summer copper residue predictions will eventually be improved with data generated by on going experiments. Sensitivity Analysis With F ig. 3 4 ), it was possible to demonstrate that different volumes change the number of unprotected days. The recommended volume of 1170 L ha 1 volume is located exactly on the optimal protection point

PAGE 43

43 Figure 3 4 Number of unprotected days summed across 56 years of every weather station. Each data point shows the simulated results varying only the spray volume from 467 to 4676 L ha 1 by increments of 9.3 L ha 1 All the other inputs for the model were kept fixed according to the defined worst case scenario. T he dashed line shows the current recommendation of 1170 L ha 1 result. Figure 3 5 parameter had a logarithmic impact on t he number of unprotected days. The recommended concentration of 0.84 kg ha 1 is a well balanced value between a reasonable amount of protection and an economical use of copper. It was also shown that concentrations greater than 1.5 kg ha 1 should be avoided as they provide little increase in protection while increasing the chance of fruit blemish es caused by copper phytotoxicity

PAGE 44

44 Figure 3 5 Number of unprotected days summed across 56 years of every weather station. Each data point shows the simulated results varying only the spray concentration from 0.56 to 4.48 kg ha 1 by increments of 0.056 kg ha 1 All the other inputs for the model were kept fixed according to the defined worst case scenario. The dashed line shows the current recomme ndation of 0.84 kg ha 1 result. System Evaluation The citrus copper application scheduler allows citrus producers to achieve copper residue. For instance, in some exceptionally dry years, even with the worst case scenario of mandarin fruit inside the canopy, only 3 standard copper applications were able to keep the residue levels over the danger threshold for the entire season ( F ig. 3 3 ) in the simulated sc enarios. On the other hand, F igure 3 6 shows an extreme case of a year with intense rainfalls where seven applications were necessary to hold the residue always above the danger threshold. With the traditional 21 day schedule, there would have been several critical gaps in copper coverage.

PAGE 45

45 The same approach of applying copper when the danger threshold is reached was then simulated for all 5 6 years of weather data and all stations. These simulations showed that across all locations in average years it would be necessary to use 6.2 standard copper sprays to keep the residue above the danger zone with the average bloom scenario. Also, the average period between needed applications was 20.6 days; this number suggests the current recommended 21 day schedule is a reasonable excess applications for very wet or dry years, respectively ( T able 3 3 ). Table 3 3 reac 6 years. Typical spray parameters and average bloom period were used. The worst case scenario, mandarin fruit inside the canopy, was used as plant parameters. Florida County Average number of applications (wet years) Average number of applications (average years) Average number of applications (dry years) Average number of days between applications (all years) Average yearly rainfall (mm) (all years) [a] Hendry 6.4 6.2 5.7 19.2 587 Highlands 6.4 5.9 5.4 20.4 537 Indian River 5.8 5.5 5.0 21.9 444 Lake 6.4 5.9 5.0 21.0 512 Polk 6.3 5.8 5.4 20.5 536 All Counties Averaged 6.2 5.8 5.3 20.6 523 [a] In the considered period, from first spray to 31 July.

PAGE 46

46 Figure 3 6 Copper residue simulation using worst case scenario plant parameters, mandarin fruit inside the canopy, 0.84 kg ha 1 metallic copper concentration, 1170 L ha 1 volume, and Polk County weather data of 2005. The spray threshold is 0.25 g cm 2 of copper and the black line is 0.1 g cm 2 Web tool Usage S tatistics Figure 3 7 shows the number of unique visitors from Florida to the created web tool. This figure is restricted to on ly visits incoming from Florida; o ther areas with significant number of visits include California, Brazil and China. The central Florida region ha d a greater number of visitors, this is an expected result as th is area has the most Citrus production ( Figure 1 1 ) However the large number o f visitors from the Gainesville area is explained by tests by the staff of University of Florida and is not representative of the Citrus producers.

PAGE 47

47 Figure 3 7 Map of Florida showing number of unique visitors of the Copper web tool produced using Goog le Analytics The visitors are grouped by metropolitan areas. Google and the Google logo are registered trademarks of Google Inc., used with permission. Figure 3 8 shows a Gaussian kernel density estimate of the bloom dates recorded by the web tool. A peak of bloom dates around mid March can be observed this peak is in agreement with proposed bloom dates of the executed experiments (Table 2 4). The peak in the beginning of the year however is explained by the fact that the website suggests a valid dat e up to 21 days before the current date as default value for bloom date, consequently most of the users visiting the website in January will have the date automatically set to January 1 st increasing the odds of these dates.

PAGE 48

48 Figure 3 8 Plot of a Gaussi an kernel density estimate of the 1460 bloom dates recorded by the web tool The vertical line marks March 20 th which is the suggested average bloom date. Figure 3 9 shows a Gaussian kernel density estimate of 3259 spray volumes recorded by the web tool The peak at 125 gal ac 1 shows that most of producers do not modify the suggested default value, but there are also noticeable peaks at 200 and 250 gal ac 1 This result suggests that some producers use more diluted spray applications.

PAGE 49

49 Figure 3 9 Plo t of a Gaussian kernel density estimate of 3259 spray volumes recorded by the web tool. The vertical line marks 1170 L ha 1 (125 gal ac 1 ) concentration which is the current recommendation. Figure 3 10 shows a Gaussian kernel density estimate of 3259 spray concentrations recorded by the web tool The peak at 0.75 lb ac 1 shows that most of producers do not modify the suggested default value however there are several high concentration peaks at 1.5, 2.0 and 3.0 lb ac 1 These results s uggest that many producers still use high concentration applications which were shown to provide little increase in protection.

PAGE 50

50 Figure 3 10 Plot of a Gaussian kernel density estimate of 3259 spray concentrations recorded by the web tool. The vertical line marks 0.84 kg ha 1 (0.75 lb ac 1 ) concentration which is the current recommendation. It is important to realize that web sites statistics have strong biases towards default values and towards the region in which the website was produced. This subsection would have more realistic information if it was possible to exclude all the t raffic incoming from University of Florida and if ev ery field on the web tool had no default values. Yet default values are crucial for the web tool usability and help the producers to be aware of the recommended parameters. Dynamic O ptimized S chedule s For each peak bloom date scenario, the R schedule inter val optimization algorithm was run until the dates converged to a point at which no bette r schedule was found. On average, 3 executions of the optimizing algorithm were needed. Table 3 4 shows the result of each algorithm execution. It is important to reme mber that each

PAGE 51

51 peak bloom date scenario has its own schedule. Consequently for scions with large differences in the peak bloom date or different vulnerability period, these schedules should be modified to reflect the new conditions. The reduction column of Table 3 4 refers to the reduction in the number of unprotected days when compared to the usual 21 day in terval schedule. The average number of unprotected days per year/station using the 21 day interval schedule for early, average and late bloom were resp ectively 2.38 3.10 and 3.78. Table 3 4 Schedules resulting from the interval optimization algorithm. These results consider all years of available w eather data and all locations average, the worst case scenario as plant parameters and typical spray volume and concentration. Peak Bloom Interval to spray [b] Average of unprotected days per year/station Reduction % [c] 1st 2nd 3rd 4th 5th Early 20 24 21 19 19 1.9 18.9 Average 19 2 4 16 17 17 1.5 50.9 Late 22 22 20 19 N / A [a] 3 3 11. 7 [a] Not applicable. [b ] Number of days after the last spray. The first spray is always 21 days after peak bloom (Table 2 4 ), the was applied. [c] The percent reduction of the mean of unprotected days across all years and stations compared to the 21 day schedule using the traditional spray parameters. The improved protection given by the optimized schedule is due to a better distri bution of copper applications over time. The average bloom schedule got the greatest benefit because in the 21 day schedule the last spray (Table 2 4 ) would provide most of its protection outside the proposed vulnerability period which extends until end of July. These results show how a variable interval spray schedule can increase the fruit protection by distributing the residue more evenly according to the rainfall pattern and vulnerability pressure.

PAGE 52

52 Table 3 5 Schedules resulting from the variable concentration optimization algorithm. These results consider all years of available w eather data and all locations average, the worst case scenario as plant parameters and typical spray volume and concentration and 21 day application schedule. Peak Bloom C oncentration of spray [b] Average of unprotected days per year/station Reduction % [c] 1st 2nd 3rd 4th 5th 6th Early 0 .75 0 .55 0 .75 0 95 0 .9 5 0 55 2. 1 11.6 Average 0 .85 0 .55 0 .85 0 .95 0 .75 0 .55 2. 7 13.6 Late 0 .75 0 .65 0 .85 0 .95 0 .55 NA [a] 3.5 7. 5 [a] Not applicable. [b] Copper concentration of each spray including the first one after peak bloom (Table 2 4). [c] The percent reduction of the mean of unprotected days across all years and stations compared to the 21 day schedule using the traditional s pray parameters. However, it is important to keep in mind that the optimized schedule s are an empirical analysis of the copper residue influen ced by past rain distribution. C onsequently in the majority of years, the optimized schedule will perform better, but there can be years in which the benefit will be minimal.

PAGE 53

53 Figure 3 11 Plot the average unprotected days of each schedule produced by both interval optimization ( continuous line) and concentration opt imization (dashed line) for the average bloom date scenario. The horizontal line marks the average protection of the traditional approach of 21 days interval and 0.84 kg ha 1 (0.75 lb ac 1 ) concentration for all copper applications. Each mark in the x axis correspond s to one simulated optimization The concentration optimization has less satisfactory results as shown in Figure 3 11. Increasing the concentration of the applications has diminishing effects with larger values consequently reducing the margin for improvement. It is also possible to see how inadequately planned schedules can greatly decrease the protection The worst tested schedules had an increase of more than 100% in the number of unprotected days.

PAGE 54

54 CHAPTER 4 CONCLUSIONS The citrus copper ap plication scheduler, a web based decision support system, enables citrus growers to easily access information to make decisions concerning the timing of copper applications. By using the web based tool, growers can reduce copper applications in dry years a nd minimize unprotected periods in wet years. The results of the copper model with a threshold of 0.25 g cm 2 and historical weather data showed that the traditional 21 day schedule at recommended copper application rates did not provide enough residual c opper to protect groves in wet years. Conversely, it was shown that it is possible to avoid unnecessary copper applications in dry years by optimizing the timing of the sprays. Two approaches for optimizing the copper schedules were proposed with the obje ctive of evenly distributing the copper protection according to the historic weather data. The optimized schedule with varying intervals between application s was able to attain 50% fewer unprotected days for the simulated period using the average bloom dat e scenario The optimized schedules with varying concentrations provide an alternative for producers wh o must have a fixed interval spray schedule, but it provides smaller gains in protection. Lastly, this study provides a documentation of the algorithm used in the source code of the copper model as well as a study of the model sensitivity to change in parameters. It was found that the current recommendations for spray volume and concentration are a good tradeoff between protection and amount of copper a pplied. As future developments, there exists an ongoing effort for allowing the user to automatically use the Real Time Mesoscale Analysis (RTMA, http://www.nco.ncep.noaa

PAGE 55

55 .gov/pmb/products/rtma ) rainfall data for a FAWN weather stations. Since RTMA is calculated on a 5 km wide grid, it would likely have more precise rainfall information when the grove is distant from a weather station. Also the fruit growth functions for the model end in early July. With the arrival of citrus canker and black spot, copper applications are needed throughout the summer. Fruit growth and copper residue loss data are being gathered. These data will be used to improve the residue predictions from July to October and be added to the model.

PAGE 56

56 APPENDIX COPPER RESIDUE MODEL TRANSLATED TO R # Translated from Java to R by Tiago Zortea # Tiago Zortea (zortea@ufl.edu ) # 05/16/2011 # The comments are from the original Java code toJulian = function (pdate){ pdate=as.POSIXlt(pdate) return(pdate$yday + 1) } residue = function(init_depo, daysAfterBloom, model){ # convert to ug/cm2 return (init_depo/area(daysAfterBloom, model)) } init_depo = function (volume, concRatio, sprayDat e, bloomDate, model){ # compute initial deposition from spray volume and concentration # unit: ug = ug / cm2 cm2 inside=T # 1 for insider 2 for outsider if(inside){ if(volume < 125) return ((0.7167 + 0.058 volume) area(sprayD ate bloomDate, model) concRatio) else if(volume >= 125 && volume <= 250) return ((9 0.01133 volume) area(sprayDate bloomDate, model) concRatio) else return ((6 + .0022*volume) area(sprayDate bloomDate, model ) concRatio) } else { if (volume < 125) return ((0.5792 + 0.08017 volume) area(sprayDate bloomDate, model) concRatio) else if(volume >= 125 && volume <= 250) return ((13

PAGE 57

57 0.016 volume) area(sprayDate bloomDate, model) concRatio) else return ((11 .0078 volume) area(sprayDate bloomDate, model) concRatio) } } reduceRatio = function(rain){ if (rain >= 0 && rain <= 0.5) return (.48 rain) if (rain > 0.5 && rain <= 2) return (.12 (rain 0.5) + .24) return (0.42) } area = function (daysAfterBloom, model){ if (tolower(model) == "grapefruit") return (gompertz(daysAfterBloom,73,22650,0.0220)) if (tolower(model) == "valencia") return (gompert z(daysAfterBloom,69,14949,0.0222)) if (tolower(model) == "mandarin") return (gompertz(daysAfterBloom,77,14263,0.0198)) if (tolower(model) == "navel") return (gompertz(daysAfterBloom,64,19856,0.0214)) #Orange (generic, use valencia p arameters) return (gompertz(daysAfterBloom,69,14949,0.0222)) } residue = function(init_depo, daysAfterBloom, model){ # convert to ug/cm2 return (init_depo / area(daysAfterBloom, model)) } gompertz = function (daysAfterBloom, originalBloomDate, max, b){ # Gompertz equation... # AREA = MAX*EXP(LN(MIN/MAX)*EXP( B*T))

PAGE 58

58 # # AREA = fruit surface area in square millimeters # T = date (julian, Jan 1 = 1) # MIN = Minimum Size (always 0) # MAX = Maximum Size (esti mated for each experiment) # B = parameter (estimated for each experiment) # originalBloomDate is date regression was based on # daysAfterBloom is days since bloom in current year # add to get adjusted julian date (accounts for the fact # that bloom date this year might not be same as bloom # date in original year) # Max B #Control Block/Yellow valencia 14949 0.0222 #Shade/Pink valencia 14611 0.0213 #Plastic Block/Pink valencia 14763 0.0215 #Control Block/Yellow grapefruit 22650 0.0220 #Shade/Pink grapefruit 22733 0.0217 #Plastic Block/Pink grapefruit 24116 0.0216 #Fallglo 14263 0.0198 #Navel 19856 0.0214 julianDate = originalBloomDate + daysAfterBloom min = 0.000000000645 # very small return (max exp( log(min / max) exp( 1 b julianDate))) } simulateResidue = function(experiment) { sql = paste ("SELECT simID, CONVERT( scion, CHAR ) as scion bloom_date FROM simCtrl WHERE simID=",experiment) rs = dbSendQuery(con,statement=sql) experimentPar = na.omit(fetch(rs,n= 1))

PAGE 59

59 experimentPar[,'bloom_date ']=toJulian(experimentPar[,'bloom_date']) sql = paste ("SELECT date inches FROM simRain WHERE simID=", experiment," order by date") rs = dbSendQuery(con,statement=sql) experimentRain = na.omit(fetch(rs,n= 1)) experimentRain$date=toJulian( experimentRain[,1]) sql = paste ("SELECT date, volume, concentration FROM simSpray WHERE simID=",experiment," order by date") rs = dbSendQuery(con,statement=sql) experimentSprays = na.omit(fetch(rs,n= 1)) experimentSprays$date=toJulian(exp erimentSprays[,1]) concRatio = experimentSprays[1,'concentration'] / 4 startDate=experimentSprays[1,'date'] scion=experimentPar[1,'scion'] results=rep(0,366) depo =init_depo(experimentSprays[1,'volume'],concRatio, startDate,experimentPar[1,'bl oom_date'],scion) results[1]=residue(depo,startDate experimentPar[1,'bloom_date'],scion) change=0 for (x in 2:366){ idx=which.max(match(experimentRain$date,startDate + (x 1))) if(length(idx)>0){ change = reduceRatio(experimentRain[idx,'inches' ]) depo; } idx=which.max(match(experimentSprays$date,startDate + (x 1))) if(length(idx)>0){ concRatio = experimentSprays[idx,'concentration'] / 4 change =init_depo(experimentSprays[idx,'volume'],concRatio, experimentSprays[idx,'date'],

PAGE 60

60 experimentPar[1,'bloom_date'],scion) } # rainfall occurs, Cu reduced, re calculate the coefficient of # the exp function: data=new_co e ^ (x) (x starts from zero...) depo=depo + change results[x]=residue(depo,startDate + (x 1) experimentPar[1,'bloom_date'],scion) change = 0; } return(results) }

PAGE 61

61 LIST OF REFERENCES Albrigo, L.G., H.W. Beck, L.W. Timmer, and E. Stover 2005. Development and testing of recommendation system to schedule copper sprays for citrus disease control. Journal of ASTM International 2, 1 12. Albrigo, L.G., L.W. Timmer, K. Townsend, and H.W. Beck 1997. Copper fungicides residues for disease control and potential for spray burn. Proceedings of the Flor ida State Horticultural Society 110, 67 70. Alva A.K., J.H. Graham and D.P.H. Tucker 1993. Role of calcium in amelioration of copper phytotoxicity for citrus. Soil Science 155, 211 8. Bhatia, A., P.D. Roberts, and L.W. Timmer. 2003. Evaluation of the Al ter Rater model for timing of fungicide applications for control of Alternaria brown spot of citrus. Plant Dis. 87:1089 1093. Beck, H.W., L.G. Albrigo, and S. Kim 2006. DISC citrus planning and scheduling program. Acta Hort. (ISHS) 707, 25 32. Available at http://www.actahort.org/books/707/707_2.htm. Accessed 14 January 2013. Dewdney, M.M., and J.H. Graham 2012. Citrus canker. In 2012 Florida Citrus Pest Management Guide, edited by M.E. Rogers, M.M. Dewdney, T.M. Spann, Gainesville, FL: University of Flo rida, IFAS. pp. 6. Available at http://edis.ifas.ufl.edu/cg040. Accessed 14 January 2013. Dewdney, M.M., C.W. Fraisse, T. Zortea, and J.D. Burrow 2012a. A Web based tool for timing copper applications in Florida citrus. Publ. No. PP289. Univ. Florida, IF AS, EDIS, Gainesville. (4 pp.) Available at http://edis.ifas.ufl.edu/pp289. Accessed 14 January 2013. Dewdney, M.M., T.S. Schubert, M.R. Estes, and N.A. Peres 2012b. Citrus black spot. In 2012 Florida Citrus Pest Management Guide, edited by M.E. Rogers, M.M. Dewdney, T.M. Spann, Gainesville, FL: University of Florida, IFAS. pp. 6. Available at http://edis.ifas.ufl.edu/cg088. Accessed 14 January 2013. FDACS (Florida Department of Agriculture and Consumer Services) 2011. Florida Agriculture By The Numbers 2011. http://www.florida agriculture.com/pubs/pubform/pdf/florida_agricultural_statistical_directory.pdf FDACS. 2012. Florida Citrus Statistics 2010 2011. Florida Department of Agriculture and Consumer Services. Available at: http://www.nass.usda.gov/Stati stics_by_State/Florida/Publications/Citrus/fcs/201 0 11/fcs1011.pdf. Accessed 14 January 2013.

PAGE 62

62 Fraisse, C.W., N.E. Breuer, D. Zierden, J.G. Bellow, J. Paz, V.E. Cabrera, A. Garcia y Garcia, K.T. Ingram U. Hatch, G. Hoogenboom, J.W. Jones, and J.J. O'Brien 2006. AgClimate: A climate forecast information system for agricultural risk management in the southeastern USA. Computers and Electronics in Agriculture 53, 13 27. Guillespie, T.J ., and P.C. Sentelhas 2008. Agrometeorology and plant disease mana gement A happy marriage. Sci. Agric. 65: 71 75. Graham J.H., M.M. Dewdney and M.E. Meyers 2010. Streptomycin and copper formulations for control of citrus canker on grapefruit. Proceedings of the Florida State Horticultural Society 123, 92 99. Graham J.H., M.M. Dewdney H.D. Yonce 2011. Comparison of copper formulations Horticultural Society 124, 79 84. Graham J.H., L.W. Timmer and D. Fardelmann 1986. Toxicity of fung icidal copper in soil to citrus seedlings and vesicular arbuscular mycorrhizal fungi. Phytopathology 76, 66 70. Morris, M.D. 1991. Factorial sampling plans for preliminary computational experiments. Technometrics 33, 161 174. Pavan, W., C.W. Fraisse, and N .A. Peres 2011. Development of a web based disease forecasting system for strawberry. Computers and Electronics in Agriculture 75, 169 175. R Development Core Team 2011. R: A language and Environment for Statistical Computing. R Foundation for Statistica l Computing, Vienna, Austria, ISBN: 3 900051 07 0. http://www.R project.org. Schaibly, J.H. and K.E. Shuler 1973. Study of the sensitivity of coupled reaction systems to uncertainties in rate coefficients. II Applications. Journal of Chemical Physics, 59, 3879 3888. Schubert, T.S., M.M. Dewdney, N.A. Peres, M.E. Palm, A. Jeyaprakash, B. Sutton, S.N. Mondal, N.Y. Wang, J. Rascoe, and D.D. Picton, 2012. First report of Guignardia citricarpa associated with citrus black spot on sweet orange ( Citrus si nensis ) in North America. Plant Disease 96, 1225. Schutte G.C., K.V. Beeton and J.M. Kotze 1997. Rind stippling on Valencia oranges by copper fungicides used for control of citrus black spot in South Africa. Plant Disease 81, 851 4. Spann, T.M., R.A. At wood, J.D. Yates, R.H. Brlansky, and K.R Chung, 2008. Dooryard citrus production: Citrus Diseases Exotic to Florida. Publication No. HS1132, University of Florida, IFAS, EDIS, Gainesville, Florida.

PAGE 63

63 Sutton, J.C., T.J. Gillespie, and P.D. Hildebrand 198 4. Monitoring weather factors in relation to plant disease. Plant Dis. 68:78 84. Timmer, L.W., H. M. Darhower, and A. Bhatia 2001. The Alter Rater, a new weather based model fo r timing fungicide sprays for alternaria control. Publication No. SP 175, University of Florida, IFAS, EDIS, Gainesville, Florida. Timmer, L.W., and S.E. Zitko 1996. Evaluation of a model for prediction of postbloom fruit drop of citrus. Plant Dis ease 80 :380 383. Timmer, L.W., and S. E. Zitko 1996. Evaluation of copper fungicides and rates of metallic copper for control of melanose on grapefruit in Florida. Plant Disease 80, 166 169. Timmer L.W., S.E. Zitko and L.G. Albrigo 1998. Split applications of copper fungicides improve control of melanose on grapefruit in Florida. Plant Disease 82, 983 6. USDA 2011. Citrus Fruits 2011 Summary September 2011. Available at: http://usda01.library.cornell.edu/usda/nass/CitrFrui//2010s/2011/CitrFrui 09 22 2011.pdf Accessed 14 January 2013.

PAGE 64

64 BIOGRAPHICAL SKETCH Tiago Zortea was born in Passo Fundo, Brazil on October, 1983. He obtained an degree in Electronics from the Cecy Leite Costa I nstitute along with his high school degree in 2004. After, he enrolled in the Electri cal Engineering program at University of Passo Fundo, but soon he realized his real passion was i n the computing sciences In 2005 he changed his major to Computer Science obtaining the degree of of Compu ter Science in 2011. To be able to afford the high educational costs, he began searching for a job as soon as he began his college studies Still in 2005 he started working as Assistant Developer at the Caixa Federal Bank. In 2007 he was hired by Compass o S.A. as a Java web developer. In Compasso he quickly ascended positions; by 2011 he was a Senior Software Engineer lead ing a team of developers. Even tho ugh he had a fulfilling job by this time, Tiago wanted to be able to exercise his scientific knowledg e also for that matter he contacted his advisor Dr. Willingthon Pavan. His advisor, who had attended University of Florida as a research scholar, contacted Dr. Clyde Frais se which ultimately led to a position offer to Tiago as research scholar at Universi ty of Florida. In 2011 Tiago was offered a full assistantship to pursue his graduate education at the Agricultural and Biological Engineering Department at University of Florida, under the supervision of Dr. Clyde Fraisse