<%BANNER%>

Investigating the Relationship of Scale and Resilience in Integrated Water Resource Management in the Crocodile River, S...

MISSING IMAGE

Material Information

Title:
Investigating the Relationship of Scale and Resilience in Integrated Water Resource Management in the Crocodile River, South Africa
Physical Description:
1 online resource (145 p.)
Language:
english
Creator:
Wangusi, Nathan Barasa
Publisher:
University of Florida
Place of Publication:
Gainesville, Fla.
Publication Date:

Thesis/Dissertation Information

Degree:
Doctorate ( Ph.D.)
Degree Grantor:
University of Florida
Degree Disciplines:
Agricultural and Biological Engineering
Committee Chair:
Kiker, Gregory A
Committee Co-Chair:
Munoz-Carpena, Rafael
Committee Members:
Beck, Howard W
Barnes, Grenville
Brown, Mark T

Subjects

Subjects / Keywords:
africa -- ecology -- modeling -- resilience -- water
Agricultural and Biological Engineering -- Dissertations, Academic -- UF
Genre:
Agricultural and Biological Engineering thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract:
The Crocodile River Catchment is a trans-boundary river that flows from South Africa into Mozambique traversing a variety of land use areas including forested upland areas to ecologically sensitive areas downstream in the “lowveld” region. The river also serves as a critical source of water for irrigation, mining and municipal water supply. These competing demands have put undue pressure on its water resources and the river is considered “oversubscribed” by governmental agencies. Water managers in the region are responsible for making critical decisions on water use allocations from the Crocodile River in a data-poor environment necessitating the use of models to simulate water use and management scenarios. Two major considerations in the interpretation of both observed and simulation data are the resolution of observation/prediction of hydrological events such as runoff, precipitation and groundwater movement and the resulting effects on management decisions which influence the ecological health, sustainability and resilience of the water catchment (Bloschl & Sivaplan, 1995).     The evaluation of hydrologic model behavior and performance at different scales is achievable using a variety of statistical measures that compare simulated and observed variables observed at the outlet of a catchment unit which in the South African context is at either the primary, secondary, quaternary or quinary scale.Single multi-response efficiency criteria are traditionally used by hydrologists to provide information about the closeness of simulated outputs to observed data such as coefficient of efficiency and coefficient of determination. Owing to errors due to parameterization at different scales,these methods can be misleading and ambiguous resulting in incorrect verification of results used in making critical planning decisions.  A combination of normalized goodness-of-fit measures, observation of bias and confidence intervals and graphical presentation gives a more comprehensive approach to validating hydrological model simulations and flow data. The analysis involves comparing the outputs of the model when parameterized at the quaternary and quinary scales.    Another tool for integrating environmental data,simulations and human interactions is the use of Bayesian networks in which water managers can combine and quantify variable factors such as economic output, ecological health and precipitation. While these models are inadequate in providing accurate system behavior of these outputs they have proved useful in integrating disparate sources of information to water managers that provide valuable insights in understanding the behavior of the Crocodile River system as a complex adaptive system. The process of configuring the Bayesian networks is performed iteratively. Sensitivity analysis on the different components of the system ensures that the model behavior aligns with reality. This approach presented demonstrates the application of Bayesian networks in the Crocodile River catchment of South Africa as qualitative and quantitative tool for strategic adaptive management.The Bayesian methodology approach proposed formalizes the reasoning and rules demonstrated in an influence diagram. The feasibility of this Bayesian network is shown in economic and ecological scenarios that are developed to demonstrate various system states based on management variables such as dam construction,sectoral economic contribution and ecological reserve implementation. The resulting analysis indicates that the habitat, riparian, invertebrate and fish indices have the greatest influence over the results of the ecological state. The results will be useful in ongoing catchment level ecological monitoring efforts and in the development of management scenarios for water supply and management.
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 Nathan Barasa Wangusi.
Thesis:
Thesis (Ph.D.)--University of Florida, 2013.
Local:
Adviser: Kiker, Gregory A.
Local:
Co-adviser: Munoz-Carpena, Rafael.

Record Information

Source Institution:
UFRGP
Rights Management:
Applicable rights reserved.
Classification:
lcc - LD1780 2013
System ID:
UFE0045244:00001

MISSING IMAGE

Material Information

Title:
Investigating the Relationship of Scale and Resilience in Integrated Water Resource Management in the Crocodile River, South Africa
Physical Description:
1 online resource (145 p.)
Language:
english
Creator:
Wangusi, Nathan Barasa
Publisher:
University of Florida
Place of Publication:
Gainesville, Fla.
Publication Date:

Thesis/Dissertation Information

Degree:
Doctorate ( Ph.D.)
Degree Grantor:
University of Florida
Degree Disciplines:
Agricultural and Biological Engineering
Committee Chair:
Kiker, Gregory A
Committee Co-Chair:
Munoz-Carpena, Rafael
Committee Members:
Beck, Howard W
Barnes, Grenville
Brown, Mark T

Subjects

Subjects / Keywords:
africa -- ecology -- modeling -- resilience -- water
Agricultural and Biological Engineering -- Dissertations, Academic -- UF
Genre:
Agricultural and Biological Engineering thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract:
The Crocodile River Catchment is a trans-boundary river that flows from South Africa into Mozambique traversing a variety of land use areas including forested upland areas to ecologically sensitive areas downstream in the “lowveld” region. The river also serves as a critical source of water for irrigation, mining and municipal water supply. These competing demands have put undue pressure on its water resources and the river is considered “oversubscribed” by governmental agencies. Water managers in the region are responsible for making critical decisions on water use allocations from the Crocodile River in a data-poor environment necessitating the use of models to simulate water use and management scenarios. Two major considerations in the interpretation of both observed and simulation data are the resolution of observation/prediction of hydrological events such as runoff, precipitation and groundwater movement and the resulting effects on management decisions which influence the ecological health, sustainability and resilience of the water catchment (Bloschl & Sivaplan, 1995).     The evaluation of hydrologic model behavior and performance at different scales is achievable using a variety of statistical measures that compare simulated and observed variables observed at the outlet of a catchment unit which in the South African context is at either the primary, secondary, quaternary or quinary scale.Single multi-response efficiency criteria are traditionally used by hydrologists to provide information about the closeness of simulated outputs to observed data such as coefficient of efficiency and coefficient of determination. Owing to errors due to parameterization at different scales,these methods can be misleading and ambiguous resulting in incorrect verification of results used in making critical planning decisions.  A combination of normalized goodness-of-fit measures, observation of bias and confidence intervals and graphical presentation gives a more comprehensive approach to validating hydrological model simulations and flow data. The analysis involves comparing the outputs of the model when parameterized at the quaternary and quinary scales.    Another tool for integrating environmental data,simulations and human interactions is the use of Bayesian networks in which water managers can combine and quantify variable factors such as economic output, ecological health and precipitation. While these models are inadequate in providing accurate system behavior of these outputs they have proved useful in integrating disparate sources of information to water managers that provide valuable insights in understanding the behavior of the Crocodile River system as a complex adaptive system. The process of configuring the Bayesian networks is performed iteratively. Sensitivity analysis on the different components of the system ensures that the model behavior aligns with reality. This approach presented demonstrates the application of Bayesian networks in the Crocodile River catchment of South Africa as qualitative and quantitative tool for strategic adaptive management.The Bayesian methodology approach proposed formalizes the reasoning and rules demonstrated in an influence diagram. The feasibility of this Bayesian network is shown in economic and ecological scenarios that are developed to demonstrate various system states based on management variables such as dam construction,sectoral economic contribution and ecological reserve implementation. The resulting analysis indicates that the habitat, riparian, invertebrate and fish indices have the greatest influence over the results of the ecological state. The results will be useful in ongoing catchment level ecological monitoring efforts and in the development of management scenarios for water supply and management.
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 Nathan Barasa Wangusi.
Thesis:
Thesis (Ph.D.)--University of Florida, 2013.
Local:
Adviser: Kiker, Gregory A.
Local:
Co-adviser: Munoz-Carpena, Rafael.

Record Information

Source Institution:
UFRGP
Rights Management:
Applicable rights reserved.
Classification:
lcc - LD1780 2013
System ID:
UFE0045244:00001


This item has the following downloads:


Full Text

PAGE 1

1 INVESTIGATING THE RELATIONSHIP OF SCALE AND RESILIENCE IN INTEGRATED WATER RESOURCE MANAGEMENT IN THE CROCODILE RIVER, SOUTH AFRICA By NATHAN BARASA WANGUSI A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY O F FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2013

PAGE 2

2 2013 Nathan Barasa Wangusi

PAGE 3

3 To my parents and family who have nurtured and believe d in me through the years and to my Ph.D supervisor Dr. Greg Kiker for his unwavering support, guidance and patience.

PAGE 4

4 ACKNOWLEDGEMENTS Funding for this research was generously provided for by the Department of Agricultural and Biological Engineering at the University of Florida and fieldwork studies by the Rotary Foundation. I am grateful to my committee members; Dr. Greg Kiker for giving me academic space to explore different problems and solutions, Dr. Rafael Mu oz Carpena and Dr. Howard Beck for ensu ring I maintained technical depth and rigor in my studies and Dr. Grenville Barnes and Dr. Mark Brown for giving valuable insights on natural resource management. I am indebted to Dr. Jeff Smithers, Dr. Mark Dent, David Clark and Mark Horan of the Univers ity of Kwa Zulu Natal Department of Bioresources Engineering and Environmental Hydrology; Dr. Tally Palmer of the Akili Initiative; Harry Biggs, Dr. Eddie Riddell and Craig Mc Loughlin of the Kruger National Park Service Scientific Services; Steven Mallory of IWR Water Resources; Sharon Pollard of the Association for Water and Rural Development; and Brian Jackson of the Inkomati Catchment Management Agency for their contributions and support during my fieldwork in South Africa. This dissertation reflects th eir shared knowledge, perspectives and expertise.

PAGE 5

5 TABLE OF CONTENTS Page ACKNOWLEDGEMENTS ................................ ................................ ............................... 4 LIST OF TABLES ................................ ................................ ................................ ............ 8 LIST OF FIGURES ................................ ................................ ................................ ........ 10 LIST OF OBJECTS ................................ ................................ ................................ ....... 12 LIST OF ABBREVIATIONS ................................ ................................ ........................... 13 C HAPTER 1 INTRODUCTION ................................ ................................ ................................ 18 Overview of River Modeling in South Africa ................................ ........................ 18 Research Questions, Objectives and Hypothesis ................................ ............... 19 Background ................................ ................................ ................................ ........ 21 Study Area ................................ ................................ ................................ .. 21 Delineation of Water Ma nagement Areas in the Inkomati Catchment ........ 22 Population and Land Use in the Crocodile River Catchment ...................... 23 Water Governance in South Af rica ................................ ............................. 24 2 CONCEPTUAL MODEL FOR INTEGRATING KNOWLEDGE AND INFORMATION FOR THE CROCODILE RIVER CATCHMENT, SOUTH AFRICA ................................ ................................ ................................ .............. 32 Introduction ................................ ................................ ................................ ......... 32 Methodology in the Selection of Research Tools ................................ ................ 33 A Review of Ecological Complexity and Resilie nce in the Crocodile River ................................ ................................ ................................ ...... 33 Interactions with Water Resource Institutions ................................ ............. 38 Research Outcomes ................................ ................................ ........................... 42 Development of a Conceptual Map and Integrated System Tools for Estimating Ecological Effects from Flow Regimes in the Crocodile River ................................ ................................ ................................ ...... 42 Akili Forum Results: A Gen eral Systems Diagram of the Crocodile River .. 43 The ACRU model as a hydrological model for generating flow regimes for the Crocodile River ................................ ................................ ........... 46 Bayesian Network Development for Crocodile River ................................ .. 47 Discussion ................................ ................................ ................................ .......... 50

PAGE 6

6 3 EVALUATING WATERSHED MODEL PERFORMANCE AT TWO SPATIAL SCALES TO AID ADAPTIVE MANAGEMENT IN THE CROCODILE RIVER CATCHMENT, SOUTH AFRICA ................................ ................................ ........ 59 Overview of Strategic Adaptive Management in the Crocodile River Catchment ................................ ................................ ................................ ..... 59 Study Area ................................ ................................ ................................ .......... 63 Research Hypothesis and Objectives ................................ ................................ 65 Methods and Tools ................................ ................................ ............................. 67 The Agricultural Catchment Research Unit (ACRU) model ........................ 67 Selection of Model Evaluation Indicators and Testing Criteria .................... 69 Results ................................ ................................ ................................ ............... 74 Evaluation of Model Performance ................................ ............................... 74 Comparison of Results at Quaternary and Quinary Scale .......................... 75 Effects of Uncertainty in Measurement ................................ ....................... 78 Implications of Model Performance on River Monitoring ............................ 81 Discussion ................................ ................................ ................................ .......... 82 4 MODELING WATER USE DECISIONS FOR STRATEGIC ADAPTIVE MANAGEMENT USING BAYESIAN NETWORKS IN THE CROCODILE RIVER ................................ ................................ ................................ .............. 94 Overview of Bayesian Networks in Adaptive Management ................................ 94 Methodology ................................ ................................ ................................ ....... 96 Bayesian Networks ................................ ................................ ..................... 96 Constructing a Bayesian Network ................................ ............................... 99 Problem definition ................................ ................................ ............ 99 Model inference ................................ ................................ ............. 100 Model validation ................................ ................................ ............. 101 Case Study Area Overview ................................ ................................ .............. 102 Crocodile River Bayesian Model D evelopment ................................ ................ 102 Problem Definition ................................ ................................ .................... 102 Description of management environment ................................ ...... 102 Identify management endpoints and zones ................................ ... 104 Identify management alternatives ................................ .................. 105 Identify actors and process variables ................................ ............. 105 Model Inference ................................ ................................ ........................ 106 Flow ................................ ................................ ............................... 106 Ecostatus ................................ ................................ ....................... 106 Validation of Predicted Management Endpoints ................................ ....... 108 Ecological status ................................ ................................ ............ 108 Sectoral economy ................................ ................................ .......... 108 Sensitivity Analysis ................................ ................................ ................... 109 Application of the Model ................................ ................................ ................... 111 Impacts of Managemen t on Assurance of Supply ................................ .... 111 Cost Benefit of Management Actions ................................ ....................... 112 Discussion ................................ ................................ ................................ ........ 114

PAGE 7

7 5 CONCLUSIONS ................................ ................................ ............................... 128 APPENDIX A ACRU DATA FILES ................................ ................................ .......................... 132 B BAYESIAN NETWORK MODEL AND CASE FILES ................................ ........ 133 REFERENCE LIST ................................ ................................ ................................ ...... 134 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 145

PAGE 8

8 LIST OF TABLES Table Page 2 1 Classification of variables for the Bayesian network of the Crocodile River System ................................ ................................ ................................ ............... 52 3 1 Test catchment description showing predominant land uses, mean ................... 85 3 2 Evaluation ranges of goodness of fit with respect to the Coefficient of Efficiency (C eff ) (Ritter and Mu oz Carpena, 2013) ................................ ............ 85 3 3 Comparison of quaternary scale simulations and observed monthly stream flow for 5 selected catchments of ,the coefficients of determination (R 2 ), coefficients of efficiency (C eff ), median annual observed streamflow an d the RMSE with 95% confidence intervals for C eff and RMSE ................................ ... 86 3 4 Comparison of quinary scale simulations and observed monthly streamflow for 5 selected catchments of the Crocodile River, the coefficients of determination (R 2 ) coefficients of efficiency (C eff ), median annual observed streamflow and the RMSE with 95% confidence intervals for C eff and RMSE .... 86 3 5 Evaluation of acce ptability of results from X21H and X23C based on traditional statistical measures and observation of statistical significance .......... 87 3 6 Assessment of the effect of the removal of measurement bias of 4.3% from observed measurements at monitoring points ................................ .................... 87 3 7 Effect of uncertainty correction on acceptability for C eff in X21H ........................ 87 4 1 Summary of the hydrology of the Crocodile River ................................ ............ 116 4 2 Summary of nodes in the Crocodile River Bn (Cain, 2001) .............................. 116 4 3 Selected management zones on the Crocodile River sections ......................... 116 4 4 Establish discretization states for variables ................................ ...................... 117 4 5 Ecological scores for ecostatus components with descriptions in relation to degree of modification from natural conditions ................................ ................. 119 4 6 Current ecostatus determined at the vari ous monitoring sites along the Crocodile River by RHP (2005) ................................ ................................ ........ 120 4 7 Sectoral contribution ................................ ................................ ......................... 120 4 8 Sensitivity analysis fo r posterior network to ecological status of the river ......... 120 4 9 Predicted assurance of supply for Reserve Scenario: 1 Status quo ............... 121

PAGE 9

9 4 10 Predicted assurance of supply for Reserve Scenario 2: Implement PES requirement (requires a 25% supply restriction) ................................ ............... 121 4 11 Predicted assurance of supply for Reserve Scenar io 3: Implement RES requirement (requires a 50% supply restriction) ................................ ............... 121 4 12 Total sectoral economic contributions for the damming scenarios ................... 122 4 13 Total s ectoral economic contribution for reserve implementation scenarios ..... 122

PAGE 10

10 LIST OF FIGURE S Figure Page 1 1 The Crocodile River Catchment (I nkhlakanipho Consultants, 2009) .................. 27 1 2 Secondary and quinary catchments of the Crocodile River ................................ 28 1 3 Quaternary catchments in t he Crocodile River ................................ ................... 28 1 4 Population density of the Crocodile River Catchment (Department of Environmental Affairs and Tourism, 2007) ................................ ......................... 29 1 5 Population trends in urban and rural residents in the Crocodile River Catchment (Ashton, 1995) ................................ ................................ .................. 29 1 6 Agricultural land use in the Crocodile River Catchment (Inkhlakanipho Consult ants, 2009) ................................ ................................ ............................. 30 1 7 The institutional frameworks for water resource management in the South Africa (Carmo Vas & van der Zaag, 2003; Roux, et al., 1999) ............................ 30 2 1 Adaptive Cycles in Complex Systems. A stylized representation of the four ............................ 53 2 2 Historical time perio ds in South Africa ................................ ................................ 53 2 3 Adaptive cycles in the Crocodile River since between 1600 Present Day .......... 54 2 4 Current real time decision s upport system in the Crocodile River ...................... 55 2 5 A conceptual map of tools and processes for estimating ecological status from hydrology and human decision making in the Crocodile River. .................. 56 2 6 System diagram showing interactions between components ............................. 57 2 7 The ACRU agro hydrology model developed by the School of Bioresources E ngineering and Environmental Hydrology, University of KwaZulu Natal (Schulze, 1995). ................................ ................................ ................................ 57 2 8 Network Structure of Main Categories (Cain, 2001). ................................ .......... 58 3 1 Location of the Crocodile River catchment (Inhlakanipho Consultants, 2009) .... 88 3 2 Map of selected quaternary catchments in the Crocodile River. Each of the six catchments has no upstream contributors along with multi year flow records ................................ ................................ ................................ ............... 89 3 3 The ACRU model ................................ ................................ ............................... 89

PAGE 11

11 3 4 Land use within the Crocodile Rive r Catchment (Inhlakanipho Consultants, 2009) ................................ ................................ ................................ .................. 90 3 5 Observed monthly streamflow time series graphs for X21H and corresponding comparison of simulated versus observed scatter plot and goodness of fit evaluations at quaternary and quinary simulation scales ........... 91 3 6 Observed monthly streamflow time series graphs for X23C and corresponding comparison of simulated versus observed scatter plot and goodness of fit evaluations at quaternary and quinary simulation scales ........... 92 3 7 Flow at X2H024 showing three TPC worry levels and the lags in management decisions dependent on scale of measurement. ........................... 93 4 1 Generalized framework of SAM as proposed by Pollard & Du Toit (2005) applied to management of aquatic ecosystems, their surroundings which includes management of environm ental flows ................................ .................. 123 4 2 Generalized approach to developing a Bn for application in the development of management option of a SAM framework (adapted from Pollard & Du Toit, 2005) ................................ ................................ ................................ ................ 124 4 3 Map of the Crocodile River within the Inkomati Water Management area ........ 124 4 4 System model obtained from collaborative sessions s howing interactions between socio ecological components in the Crocodile River Catchment ........ 125 4 5 Location of ecostatus monitoring sites ................................ .............................. 126 4 6 Streamflow at EWR 6 with IFR maintenance high and low flows ...................... 126 4 7 Bayesian network for the Crocodile River at EWR 6. The probabilities expressed in this diagram refl ect the discretized states outlined previously ..... 127

PAGE 12

12 LIST OF OBJECTS Object Page A 1 ACRU quaternary catchment menu files ................................ .......................... 132 A 2 ACRU quinary catchment menu files ................................ ................................ 132 A 3 ACRU quaternary climate input files ................................ ................................ 132 A 4 ACRU quinary climate input files ................................ ................................ ...... 132 A 5 ACRU quaternary catchment output files ................................ ......................... 132 A 6 ACRU quinary catchment output files ................................ ............................... 132 A 7 FITEVAL quaternary output files ................................ ................................ ....... 132 A 8 FITEVAL quinary output files ................................ ................................ ............ 132 B 1 Bayesian network model ................................ ................................ .................. 133 B 2 Bayesian network case file at EWR 5 ................................ ............................... 133 B 3 Bayesian network case file at EWR 6 ................................ ............................... 133

PAGE 13

13 LIST OF ABBREVIATIONS T ERM : Definition ACRU Agricultural Catchment Research Unit AM Adaptive management CAS Complex adaptive s ystem CMA Catchment management agency CMS Catchment management strategy CPT Cumulative probability table CEFF Coefficient of efficiency DEAT Department of Environmental Affairs and Tourism DHI Danish Hydraulic Institute DSS Decision support system DALA Department of Agriculture and Lands Administration DWAF Department of Water Affairs and Forestry ES Ecological status FAI Fish Assemblage Index IUCN EPP Ecosystems, Protected Areas and People Project of the Interna tional Union for Conservation of Nature IIWUA Tripartite Interim Inco Maputo Water Use Agreement (IIWUA) ICMA Inkomati Catchment Management Agency IFR In stream flow requirements IHI Index of habitat integrity KNP Kruger National Park KNP SS Kruger National Park Scientific Services KNPRRP Kruger National Park Rivers Research Program

PAGE 14

14 MAE Mean absolute error NWA National Water Act PDI Previously disadvantaged individual PES Present ecological status RES Recommended ecological status RHP Rive r Health Program RMSE Root mean square error RRS Rapid Response System RVI Riparian vegetation index SAM Strategic adaptive management SASS Macro invertebrate index SRI Shared Rivers Initiative TPC Threshold of probable concern WREMP Water Resource Modeling Platform ZAR South African Rand

PAGE 15

15 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 INVESTIGATING THE RELATIONSHIP OF SCALE AND RESILIE NCE IN INTEGRATED WATER RESOURCE MANAGEMENT IN THE CROCODILE RIVER, SOUTH AFRICA By Nathan Barasa Wangusi August 2013 Chair: Gregory A. Kiker Co chair: Rafael Munoz Carpena Major: Agricultural and Biological Engineering The Crocodile River Catchment is a trans boundary river that flows from South Africa into Mozambique traversing a variety of land use areas including forested upland areas to ecologically sensitive areas downstream in the lowveld region The river also serves as a critical source of water for irrigation, mining and municipal water supply. These competing demands have put undue pressure on its water resources and the region are responsible for making c ritical decisions on water use allocations from the Crocodile River in a data poor environment necessitating the use of models to simulate water use and management scenarios. Two major considerations in the interpretation of both observed and simulation da ta are the resolution of observation/prediction of hydrological events such as runoff, precipitation and groundwater movement and the resulting effects on management decisions which influence the ecological health, sustainability and resilience of the wate r catchment (Bloschl & Sivaplan, 1995) The evaluation of hydrologic model behavior and performance at different scales is achievable using a variety of statistical measures that compare simulated and

PAGE 16

16 observed variables obser ved at the outlet of a catchment unit which in the South African context is at either the primary, secondary, quaternary or quinary scale. Single multi response efficiency criteria are traditionally used by hydrologists to provide information about the clo seness of simulated outputs to observed data such as coefficient of efficiency and coefficient of determination. Owing to errors due to parameterization at different scales, these methods can be misleading and ambiguous resulting in incorrect verification of results used in making critical planning decisions. A combination of normalized goodness of fit measures, observation of bias and confidence intervals and graphical presentation gives a more comprehensive approach to validating hydrological model simul ations and flow data The analysis involves comparing the outputs of the model when parameterized at the quaternary and quinary scales. Another tool for integrating environmental data, simulations and human interactions is the use of Bayesian n etworks in which water managers can combine and quantify variable factors such as economic output, ecological health and precipitation. While these models are inadequate in providing accurate system behavior of these outputs they have proved useful in integrating dis parate sources of information to water managers that provide valuable insights in understanding the behavior of the Crocodile River system as a complex adaptive system. The process of configuring the Bayesian networks is performed iteratively. Sensitivity analysis on the different components of the system ensures that the model behavior aligns with reality. This approach presented demonstrates the application of B ayesian networks in the Crocodile River catchment of South Africa as qualitative and quantitati ve tool for strategic adaptive management The Bayesian methodology approach proposed formalizes the reasoning and rules

PAGE 17

17 demonstrated in an influence diagram. The feasibility of this Bayesian network is shown in economic and ecological scenarios that are d eveloped to demonstrate various system states based on management variables such as dam construction, sectoral economic contribution and ecological reserve implementation. The resulting analysis indicates that the habitat, riparian, invertebrate and fish i ndices have the greatest influence over the results of the ecological state. The results will be useful in ongoing catchment level ecological monitoring efforts and in the development of management scenarios for water supply and management.

PAGE 18

18 CHAPTER 1 INTRODUCTION Overview of River Modeling in South Africa River systems in southern Africa are highly complex and coupled social ecological systems in which a var iety of human stakeholders compete with ecological factors for limited water resources (Pahl Wostl, et al., 2007) As one of these multifaceted catchments, the Crocodile River is characterized by a diversity of connections bet ween dynamic socio ecological components. Inherent to complex systems is a characteristic that describes their ability to withstand and absorb change or disturbance and still retain their basic structure and functional attributes (Walker & Salt, 2006) This characteristic is known as resilience and has been the focus of much effort in academic study (Gunderson & Holling, 2002) and institutional management (Polla rd, Du Toit, & Biggs, 2011) A fundamental characteristic in the resilience dynamic is the concept of temporal and spatial scale and its connection with physical and biological processes. This dynamic is termed panarchy (Gunderson & Holling, 2002) and provides a useful construct for envisioning these scale interactions within an adaptive context. Another critical element to understanding complex, adaptive systems is the role of systems simulation to explore these conn ections and drivers. The literature referring to scale in hydrological models includes both a temporal and spatial description of heterogeneity which is essential in understanding underlying processes (Braun, Molnar, & Kleeberg 1997) Bergstrom and Graham (1998) maintain that the magnitude of the scale problem is related to the specific hydrologic problem to be solved and to the scientific approach and perspective of the modeler. Scale is sues confound both hydrological model developer s and river managers in different ways.

PAGE 19

19 Hydrological model developers may focus on whether the appropriate algorithms (and their corresponding input parameters) are being incorporated for the corresponding sp atial and temporal execution while river managers seek to optimize river control and abstractions to provide consistent societal benefits while mitigating adverse consequences from extreme events. Both of these perspectives must be integrated to set the c onditions for successful adaptive control. In attempting to manage the limited water resources in the catchment, environmental managers at the Inkomati Catchment Management Agency (ICMA) and the Kruger National Park (KNP) are concerned with implementing t he policies of the National Water Act (National Water Act 1998) which include the devolution of water management to catchment management agencies, implementation of the human and ecological water reserves and sustainable use o f available water resources. framework takes into account the needs of various stakeholders within the catchment (Colvin, Everard, Goss, Klarenberg, & Ncala, 2008) This has ne cessitated the use of traditional hydrological models and the development of ecological models as management tools as an integral part of water resources management in the catchment. Research Questions, Objectives and Hypothesis Fundamental questions in th e adaptive management (AM) effort addressed in this research of this riverine system are the following: 1. How can river managers functionally describe resilience to meet the policies of the National Water Act of 1998? 2. What practical solutions does eco hydr ological modeling using a distributed, catchment scale hydrological model and integrative tools such as Bayesian

PAGE 20

20 networks provide in operationalizing the proposed Adaptive Management framework in the Crocodile River ? Given the need for improved water res ource management in the Crocodile River and other South African catchments, it is important to define a method of modelling and monitoring resilience as well as investigating scale as it relates to management of water resources in the catchment. This prop osed research will utilize two systems tools, watershed scale hydrological models and Bayesian networks (Bns), to investigate the relationship between scale and resilience in the Crocodile River catchment. Within this overall objective, this effort will fo cus upon two primary efforts. Firstly, the use of statistical analysis measures to evaluate the ACRU agro hydrological model (Schulze, 1995; Schulze and Smithers, 1995) performance and hydrological resilience at two different input/catchment scales. Second ly, the use of Bayesian networks that are increasingly used in environmental modelling and management (Varis, 1997; Marcot, Holthausen, Raphael, & Rowland, 2001; Borsuk, Stow, & Reckhow, 2004; Bromley, Jackson, Clymer, Giacomello, & Jensen, 2005) Bayesian networks are used to define and analyse ecological river states within a decision framework in the Crocodile River catchment. This interdisciplinary study of water resource management in the Crocod ile River addresses the following objectives: To understand the performance of the ACRU model at different spatial and temporal scales currently used by river managers within the catchment (i.e. at quaternary and quinary scales). To design a Bayesian beli ef network for decision support that integrates diverse data types such as quantitative data types such as stochastic hydrological data with

PAGE 21

21 qualitative data types such as stakeholder view points, ecological endpoints and management goals and choices. This will include performing sensitivity analysis of Bayesian belief networks to determine effectiveness of making management choices at the different scales and the coarseness of data measurement. This research project examines the following specific hypoth esis in addressing the afore mentioned objectives. Hypothesis 1: The optimal performance of the ACRU hydrology model improves as the modeling resolution is increased (i.e. spatial, hydrological inputs are defined in greater detail). Hypothesis 2: Bayesia n networks provide a tool to test management scenarios and outcomes considering the ecological provisions of the National Water Act of South Africa To address these diverse questions, this dissertation is divided into two background sections giving brief summaries of the study site and the computational tools to be used in the research. Following this conceptual review, two chapters that will compose the body of this dissertation are described with preliminary results and research methods. Background Stu dy Area The Crocodile River is part of the Eastern Transvaal Region and falls under the jurisdiction of the trans border Komati River System with Mozambique lying in the North East and Swaziland in the South (Figure 1 1 ). The Cr ocodile River has its head water in the highveld regions east of the Drakensburg Mountains and flows steadily east over the escarpment and into the lowveld region of low rolling hills for 320 km with

PAGE 22

22 approximately 1200km of tributaries. Ultimately, the Cro codile River joins the Komati River in Mozambique and terminates into the Indian Ocean. Additionally, the Crocodile River forms the Southern border of the Kruger National Park (Department of Water Affairs and Forestry 2009) The Crocodile River catchment and its tributaries (Figure 1 2 ) is one of the largest irrigation areas in South Africa and provides the sole source of water for wide variety of users (Ashton, 1995) The river catchment area is characterized by a mixture of land uses including urban, peri urban/informal/rural settlements, national and private wildlife reserves such as KNP and production agriculture (primarily sugarcane and citrus). These various inte rests often compete for the limited water resources in the region, especially in time of low flows (Brown & Woodhouse, 2004) Ashton et al. (1995) note progressive downstream decreases in water quality, especially within the challenges, Ashton et al. resource problems currently facing South Africa a representative of the water governance and management issues facing South Africa. In South Africa, water resource management is performed at different catchment boundaries. These are discussed in the subsequent secti on. Delineation of Water Management Areas in the Inkomati Catchment The entire South Africa has been delineated into 19 catchment units at various scales for management and institutional governance by the Department of Water Affairs and Forestry (DWAF) w (Department of Water Affairs and Forestry 1999) The largest scale is the primary scale which are labeled alphabetically from A to Z then secondary scale labeled fr om 1

PAGE 23

23 to 4 then subdivided into the tertiary scale labeled numerically from 1 to 4. Catchment jurisdiction at the primary scale (Figure 1 2). Each of the tertiary catchments is fur ther disaggregated into 8 quaternary catchments labeled alphabetically from A to H (excluding I hence A,B,C,D,E,F,G,H). Each of the quaternary catchments is again subdivided into three quinary catchments labeled 1, 2 and 3 to delineate the upper, middle and lower catchment areas, respectively (Mallory, Odendaal, & Desai, 2008) In total, there are 1946 quaternary catchments or 5838 (1946 x 3) quinary catchments in South Africa. The Crocodile is in the larger X primary catchm ent which includes the Sabie, Crocodile and Komati rivers. The Crocodile River basin alone forms the X2 secondary catchment which has a total of 4 main tertiary catchments namely; Upper Crocodile (X21), Middle Crocodile (X22), the Kaap (X23) and Lower Croc odile River (X24). These are divided into a 36 quaternary catchments (X21A to X24H) with each quaternary catchment having 3 quinary catchments (e.g. X21 1, X21 2 & X21 3). Subsequent model simulations refer to these different quaternary and quinary catchm ent labels (Figure 1 3). Population and Land Use in the Crocodile River Catchment Currently, in Mpumalanga province where the Crocodile River watershed is located, a population of 1.6 million. Roughly 30% of the catchment area has 0 5 people per kilomet er, and approximately 10% of the area has the highest population density between 50 1000 people per kilometer. This heavily populated area coincides with the capital of Nelspruit as shown in Figure 1 4 (Department of Environme ntal Affairs and Tourism, 2007) .In addition to the population density, it is important to

PAGE 24

24 evaluate the growth of a population over time. As demonstrated in the following graph (Figure 1 5), growth within the Catchment has been significant over the last t wenty years. In fact, it has doubled in this 20 year span, from 342,200 in 1985 to 632,500 in 2005. Exponential increases in the population (Figure 1 5) due to urbanization and industrial growth have resulted in a marked increase in the per capita demand f or water in the Crocodile River Catchment. Consequently, there will be a need for significant improvements in the management of the limited and diminishing supply of water resources. Coupled with this is the urgency for ensuring equitable distribution of w ater biodiversity by meeting ecological water demands. Agriculture in the Crocodile River Catchment area consists of plantation forestry, irrigated sugarcane and five major irrigated subtropical fruits, including, oranges, grapefruit bananas, avocadoes, and mangoes as shown in Figure 1 6 (Crawford, 2004) Water Governance in South Africa The Department of Water Affairs and Forestry (DWAF) is t he agency tasked as a custodian of water resources in South Africa (Department of Water Affairs and Forestry 2009) The Inkomati Catchment Management Authority (ICMA) was one of the first of the 19 Catchment Management Agenci of the National Water Act of 1998 to allocate water use and development of the Inkomati Water Management Area (WMA), including the Komati, Sabie Sand, Crocodile and Nwanedzi River catchments (National Water Act 1998) Anderson (2005) mentions the differing expectations of stakeholders within the river catchment and the differing values of various participants. For instance, large scale water users such as commercial

PAGE 25

25 irrigators and forestry companies located upstream are concern about constant supply as well as having control of gauging structures giving them disproportionate control over water resources in comparison to downstream ecological w ater users and municipalities. From 2009 to 2012, the Inkomati Catchment Management Agency (ICMA) was involved in the formulation and implementation of a stakeholder inclusive Catchment Management Strategy (CMS) that ensures equity, efficiency and susta inability in the management of and exploitation of water resources in the Crocodile River catchment. The strategy involved creating forums through organized workshops that enable a decentralized and communal approach to planning and development in terms of shared responsibility on the part of the various stakeholders for water resource management and scenario planning for water use. This will ensure assurance of supply to the various stakeholders locally while meeting international obligations under the Tri partite Interim Inco Maputo Water Use Agreement (IIWUA) with Mozambique and Swaziland. The benefit of this approach being that it facilitates a collective understanding of the threats to the catchment, shared responsibility over the use of the catchments w ater resources and a common vision for the management of water resources. This approach while more tedious allows for inclusion of all stakeholders. At the national level the water management structure consists of a central water management bodies, follow ed by devolved bodies that oversee water allocation at a catchment scale (Figure 1 7). Beneath the former two bodies are the local institutions and organizations that represent the local individuals within the catchments (C armo Vas & van der Zaag, 2003) .Water resource management falls under the Department of

PAGE 26

26 Water Affairs and Forestry (DWAF) and conservation falls under the Department of Environmental Affairs and Tourism (DEAT). The central department where much of the wat er resource management occurs is the former, while the involvement of the latter, is from the perspectives of national park management (Department of Water Affairs and Forestry 2009) The Water Research Commission is responsi ble mainly for retaining institutional knowledge on water resources through research and also for the monitoring and protection of rivers through the river health program (River Health Program 2005) At the catchment scale t here are several other water interest groups such as water boards formed at a community level to lobby on behalf irrigators. The National Water Resources Infrastructure Agency is a government entity responsible for the construction and maintenance of water related infrastructure such as dams, tunnels and weirs and water services authorities which are municipal institutions involved in water supply in urban areas (Figure 1 8). After the collapse of the apartheid government in 1994 there was recognition and consensus based on majority interests and minority self preservation of the need to redistribute water resources in South Africa due to a process of institutional learning on the values such as equality in the distribution of national water resources and prioritization of ecological conservation (Biggs & Rogers, 2003) This led to a reorganization of the entire structure of the institutional framework responsible for water resource management. Other institutional adjustments su ch as a recent reorganization of the Department of Water Affairs and Forestry (DWAF) which moved the forestry docket to the Department of Agriculture and Fisheries (DAF) have occurred as priorities changed and evolved.

PAGE 27

27 Figure 1 1. The Crocodile River C atchment (Inkhlakanipho Consultants, 2009)

PAGE 28

28 Figure 1 2. Secondary and quinary catchments of the Crocodile River Figure 1 3. Quaternary catchments in the Crocodile River

PAGE 29

29 Figure 1 4. Population density of the Cro codile River Catchment (Department of Environmental Affairs and Tourism, 2007) Figure 1 5. Population trends in urban and rural residents in the Crocodile River Catchment (Ashton, 1995) 0 100 200 300 400 500 600 700 800 1985 1990 1995 2000 2005 2010 Estimated Population (Thousands) Year Urban Rural Total

PAGE 30

30 Figure 1 6. Agricultural land use in the Crocodile River Catc hment (Inkhlakanipho Consultants, 2009) Figure 1 7. The institutional frameworks for water resource management in the South Africa (Carmo Vas & van der Zaag, 2003; Roux, et al., 1999)

PAGE 31

31 Figure 1 8. The linkages between DWAF and other water institutions

PAGE 32

32 CHAPTER 2 A CONCEPTUAL MODEL FOR INTEGRATING KNOWLEDGE AND INFORMATI ON FOR THE CROCODILE RIVER CATCHMENT, SOUTH AFRICA Introduction The Crocodile River system in South Africa presents an example of a highly complex social ecological system in which a variety of stakeholders compete for limited water resources. Tradition ally, in South Africa, managing and understanding these complex systems was confined to environmental managers who were solely responsible for decision making and management (Carmo Vas & van der Zaag, 2003; Anderson, 20 05) However, with the implementation of an inclusive legislative framework following the promulgation of the National Water Act of 1998 (National Water Act 1998) stakeholders have been integrated into the decision making pr ocess. In addition to prioritizing the water needs of previously marginalized people in the catchment, the National Water Act prioritizes ecological concerns by establishing an ecological reserve for all rivers in South Africa. Two significant institutio nal changes in the legislative framework are the establishment of Catchment Management Agencies (CMAs) as jurisdictional units of water management at the catchment level and the separation of water and land rights necessitating a more integrated approach t o water management (Colvin, Everard, Goss, Klarenberg, & Ncala, 2008; Anderson, 2005; Du Toit, Biggs, & Pollard, 2011; Pollard, Du Toit, & Biggs, 2011) As a multifaceted catchment with oversubscribed water demands, the Crocodile River is characterized by a diversity of connections between dynamic socio ecological components. This system complexity coupled with significant governance changes create a unique opportunity for the study of adaptive water re source management. Accordingly, this chapter focuses on the following objectives:

PAGE 33

33 Using a resilience perspective, briefly review the ecological factors operative in the Crocodile River catchment along with the larger scale socio political forces at work in South African water governance Document the interactions with three different water resource focused institutions and highlight their primary tools for analysing complex systems Create a conceptual map that integrates the various features of the Croc odile River system for further analysis with computational modelling tools Methodology in the Selection of Research Tools A Review of Ecological Complexity and Resilience in the Crocodile River In describing the Crocodile River as a complex adaptive sys tem, a fundamental property known as resilience must be considered first. Walker and Salt (2006) describes resilience as the capacity of a complex adaptive system (CAS) to absorb disturbance and reorganize while underg oing transformation so as to retain the same function, structure, identity and feedbacks. There are two central ideas associated with the resilience of a CAS. The first is the presence of regimes that are separated by thresholds which can be thought of as tipping points. The second idea is the notion of adaptive cycles which describes how systems behave over time in transitioning through cycles of growth and collapse (Gunderson & Holling, 2002; Daedlow, B eckmann, & Arlinghaus, 2011; Norberg, Wilson, Walker, & Ostrom, 2008) These two concepts are scale relevant over space and time In the Crocodile River, regimes are applicable when looking at the behavior of water and ecological systems over short perio ds while adaptive cycles can primarily be understood when looking at the behavior of these systems over longer periods of time at different spatial scales

PAGE 34

34 Regimes in a river ecosystem such as the Crocodile River can be described in terms of the flow cat egory such as shallow riffles, rapids or dry river beds which occur at various spatial and temporal locations along the river depending on the season, sedimentation, and anthropogenic activities (e.g. damming, river scouring, sand mining, abstraction and r iver channel modifications). In river systems such as the Crocodile River regime shifts can occur when the ecological state of a river is pushed across some threshold by external or internal dynamic Ecological units within river systems can also be affect ed by these acute shifts which cause alteration in habitat types in which aquatic species thrive (e.g. changes in flow velocities in river riffles or point source pollution can cause loss of small fish species and macroinverterbrates species). This can hav e a ripple effect in the system since large fish species diversity depends on the presence of intermediate fish species and further larger aquatic carnivores such as crocodiles, leading to an overall loss of biodiversity (Norberg, Wilson, Walker, & Ostrom, 2008) Recently in the KNP, there ha ve been mass crocodile kills caused by bioaccumulation of heavy metals possibly from consuming large aquatic fish lower in the food chain which are not consumed by juvenile crocodiles. Crocodi les were found to have pansteatitis a hardening of body fat tissue which was also found in large fish species in the river (Govender D. 2010; Woodborne, et al., 2012) On a catchment scale, river systems are vulnera ble to flow as a driver can easily shift state in the event of an environmental perturbation such as a flooding event such as the catastrophic floods in 2000 and 2013 that radically altered landscapes socio economic systems and flood regimes The Crocodil e River is largely supplied by water from precipitation hence is vulnerable to changes in flow hence can quickly shift states

PAGE 35

35 in the event of extreme events such as floods or interruptions in flow caused by sedimentation, damming or drought. Adaptive cycl es constitute a heuristic to understand a CAS in response to disturbance and change (Daedlow, Beckmann, & Arlinghaus, 2011; Gunderson & Holling, 2002) This cycle, shown in Figure 2 1, defines the dynamics be hind the transitions in ecosystems over times as they go through four distinct phases: exploitation(r phase), conservation (K phase), rapid release phase ) and phase ) and rapidly back to exploitation. The two key dimensions of the changing shapes of these four phases in the adaptive phases are the degree of connectedness and the range of potential in the system (Gunderson & Holling, 2002; Walker & Salt, 2006) These adaptive cycles have been observed in complex adaptive systems such as terrestrial, aquatic and arboreal ecosystems within the Kruger National Park. The sequenced transi tions from exploitation (r) to conservation (K) where biomass, nutrients and resources are gradually accumulated. Competitive processes lead a few species to become dominant. Within savanna ecosystems common in the KNP the accumulat ion occurs in biomass an d nutrients and within aquatic ecosystems this refer s to an increase in biodiversity of fish, macro invertebrates and riparian vegetation. The shift to the K phase proceeds with accumulated capital and resources become more difficult to extract as they bec ome more tightly bound within existing networks, vegetation or biota. At this point agents of change intensity from slight perturbations to catastrophic events such as floods, earthquakes,

PAGE 36

36 fires, insect outbrea ks, disease and drought. These change agents can be natural or artificial. The system then reorganizes either back to its original form through this phase or into an alternative stable state. Resilient systems have the ability to swing between multiple sta ble and functional states that retain productive function (Gunderson & Holling, 2002) Adaptive cycles result in adaptations in different components of the ecological systems (such as species composition, population structures and behavior) can occur as a coping mechanism to these changes. Modification in ecological characteristics can also be human induced or constructed. In the KNP, various interventions such as animal population control programs (e.g. culling and breeding programs) aim at improving the diversity of species, fencing for control of spatial and temporal variability of animals and artificial watering points have been implemented. For instance between 1967 1996 the KNP have implemented elephant popula tion management maintaining the population at around 7000 a policy which was later rescinded (van Aarde, Whyte, & Pimm, 1999; Whyte, 2004) In the KNP, routine fire management is also performed to reduce accumulated bi omass (Govender N. 2010) These modifications are introduced to make ecological systems more resilient in the face of variability in the environmental or sudden or catastrophic changes in dominant system drivers such as water availability in the event there is a drought and diseases in the case of epidemics (Norberg, Wilson, Walker, & Ostrom, 2008) Artificially causing a decline in biomass by burning or biodiversity by population management is aimed at enhancing natural ecosystem cycles. A distinct advantage of the resilience perspective is its use over multiple temporal scales in the same CAS (Gunderson & Pritchard, 2002; Norberg, Wilson, Walker, &

PAGE 37

37 Ostrom, 2008 ) A historical review of Crocodile River gives us a perspective of two distinct adaptive cycles that have occurred over the last three centuries. South Africa has transitioned through several phases within the four defined transitions during this perio d (Figure 2 2). Events that have defined these cycles have varied from administrative changes, political upheavals such as the Anglo establishment of South Africa as a Republic in 1910, the apartheid government 1948 and to its event ual collapse in 1994 when a democratic rainbow nation was established (Figure 2 3). Concurrently, with the socio political changes, climatic variations catastrophic floods of 2000 compound water management challenges Water resource management in this region can best be understood by categorizing history into periods: African customary law prior to colonialism; the colonial period under Dutch rule; the colonial period under Br itish rule through apartheid under Afrikaners; and lastly, democratic post apartheid period, from 1991 to the present. During Dutch rule, legal courts usually favored the company who was given all control of water rights over any water that was considered public. The second period during colonial and apartheid era were aimed at satisfying the needs of the dominant communities in the society at the expense of the majority of the native society, as demonstrated through British rule when water rights were priv atized and the individual was favored (Tewari, 2002) populations, were actively taken int o consideration with revisions to water resource access for human consumption, an ecological reserve and then to all commercial uses.

PAGE 38

38 Interactions with Water Resource Institutions This section documents the interactions with three different water foc used interest groups and their primary tools for addressing system complexity over the 2009 2010 academic year. These water resource groups span a range from a long term, research institute (University of KwaZulu Natal Department of Bioresources Engineeri ng and Environmental Hydrology), to a newly established Catchment Management Agency (Inkomati CMA), to a temporary academic stakeholder interaction program (Akili Network). Each of these groups focused on different aspects of water resources and used diff erent tools to explore socio ecological dynamics. The first institution was the School of Bioresources Engineering and Environmental Hydrology (BEEH) at the University of KwaZulu Natal (UKZN Pietermaritzburg campus). This surface water hydrology group has existed since the late 1970s primarily through long term support of the Water Research Commission (WRC); a governmental funding agency established under the Water Research Act No.34 of 1971 that is focused on all levels South African water resources topic s from computational model development to water governance (Water Research Act, 1971) The primary focus of the BEEH group was the development and use of the Agricultural Catchments Research Unit (ACRU) model (Figure 2 7 ). Th e second institutional interaction was conducted over a three month period at the Inkomati Catchment Management Agency (ICMA) in Nelspruit, Mpumalanga, South Africa. The ICMA was a newly created agency to implement the local considerations of the National Water Act (1998) in terms of allocation of water for human consumption, ecological reserve and for all other commercial uses. During this period, the national government through its primary agency the Department of Water Affairs and Forestry

PAGE 39

39 (DWAF), was initiating a comprehensive real time Decision Support System (DSS) of the Crocodile River (CROCDSS) ( http://crocdss.inkomaticma.co.za/Website/Index.html ) to enhance the practical operation a nd management of water resources. The CROC DSS is designed to enable efficient allocation of water resources from the Kwena Dam to supply water to urban, rural and agricultural users for providing water to satisfy environmental demands downstream in the KNP ; and to fulfill international water agreements under the IIWUA with Mozambique and Swaziland. The real time CROCDSS selected by DWAF centers on a software framework developed by the Danish Hydraulic Institute (DHI) called Mike FLOOD WATCH (Figure 2 4). The implementation was contracted to a local engineering firm called Clear Pure Water. Mike FLOOD WATCH integrates spatial data, real time data, forecast models and dissemination tools in a GIS environment in addition to running third party model engines Long term operations are determined with a combination of two models. Annual operating rules will and monthly restriction rules are determined using the Water Resources Modeling Platform (WReMP) (Mallory et al., 2011 ). The WReMP system integrates the inc remental flows for simulated sub catchments to simulate water user abstractions, return flows and reservoir yields. The WReMP results are implemented and published using the Mike BASIN model. Short term operations water quantity and quality management wer e achieved using the Mike 11 model system which is a dynamic one dimensional water resources planning modeling tool (Clear Pure Water, 2008) While the CROCDSS was primarily focused on water quantity and allocation in the Cro codile River, there is a significant need to integrate potential ecological effects into the CROCDSS. As such the primary focus of interaction time at the ICMA was on

PAGE 40

40 obtaining hydrological, ecological and geographical data on the catchment from the DWAF and related consulting firms involved in the catchment. Given the lack of ecological information within the flow focused CROCDSS for the Crocodile River, additional interactions were initiated with South African National Parks (SANP arks ) specifically wit h the Kruger National Park Science Services Section (KNP SSS). The Crocodile River forms the southern boundary of the KNP and quarterly meetings are held between the ICMA, agricultural stakeholders and the KNP to discuss monitored flow levels and water qu ality concerns. These quarterly network meetings focused on addressing the need for the consideration of ecological flows and the development of tools for active ecological monitoring and responses. While several spreadsheet based monitoring tools for in dividual ecosystem units and locations were developed for these meetings, there were no tools that integrated throughout the entire river system. This lack of system wide ecological tools was expressed as a need for this forum. Through these interaction s with the KNP SSS, additional interactions were initiated with an academic community of practice in the science of complexity and resilience through the Akili Complexity and Integration Colloquia, funded by the South African National Research Foundation ( NRF). The Akili Colloquia provided a forum for academic researchers and stakeholders to both exchange knowledge on the natural resource systems in South Africa as well as to provide a forum for the development of both theoretical knowledge and applications for the complexity theory and resilience. This initiative explicitly aimed at using :

PAGE 41

41 1. knowledge building, knowledge in relation to values, knowledge hierarchy and ways of knowin g (Wickson, Carew, & Russell, 2006) 2. Concepts of social ecological and complex adaptive systems that, emphasises the (Gunderson & Pritchard, 2002) 3. Integration and implementation science (Bammer, 2005) methods, or skill sets, required to engage in the application of knowledge in a way that engage s with complexity and integration. These include complexity analysis, knowledge management, participatory methods, systems thinking and uncertainty appreciation 4. Ecosystem services (Millenium Ecosystem Assessment, 2003) a not ion which creates practical a translative space between natural resource management and the often science based knowledge domain. During the Akili network meetings, the Crocodile River was selected through a competitive process as a case study of a complex adaptive system and used to demonstrate the process of performing a complexity analysis of the system. This was done collaboratively between a group of ten researchers, facilitated by Dr H. Biggs of KNP SSS. The outputs for these discussions are summarize d in the results section and formed the basis for further integrative modeling in this dissertation. The last catchment interaction was focused on gaining broader stakeholder perspectives as a participant of a series Catchment Management Strategy meeting s organized by DWAF and the KNP. The Crocodile River Catchment Management

PAGE 42

42 Strategy Workshops were ordered by the Minister of Water and Environmental Affairs in response by a new government directive to implement the provision of the inclusivity and stake holder participation mandated by the National Water Act. They were established as forums for collective action and management for various stakeholder and interest groups in the catchment. The Catchment Management Strategy workshops were conceived to enable all stakeholder groups including irrigation farmers, rural water users, mining companies, forestry companies and the Kruger National Park to have an opportunity to contribute towards the formulation of The c ollaborative approach used was facilitated by Dr Sharon Pollard of the Association for Rural Water and Development and Professor Kevin Rodgers of the University of Witwatersrand, primary experts in adaptive management and systems thinking in South Africa. These forums allowed researchers, managers and stakeholders to construct a visual representation of the systems and in making reliable inferences of the responses in the system in addition to increasing their collective understanding of the structure of th e system (Shachter R. 1986) In keeping with the requirement for stakeholder participation mandated by law, IWR Water Resources Director Stephen Mallory was also tasked with engaging stakeholders in the determination of rese rve requirements for human and ecological use for the various stakeholders. Research Outcomes Development of a Conceptual Map and Integrated System Tools for Estimating Ecological Effects from Flow Regimes in the Crocodile River Each of the above mentione d collaborations helped to both formulate a rich picture of the many dimensions to the water resource challenges in the Crocodile River

PAGE 43

43 but also helped to identify practical gaps needed for further analysis. In such complex adaptive systems, information a nd knowledge of the system su ch as hydrological characteristics of the system related to water quantity, climatic information, ecological data and stakeholder perceptions is acquired from various sources and methods. These sources include databases and i nformation libraries, elicitation through expert consultation, stakeholder interviews and the use of computational models information (Brugnach, Dewulf, Pahl Wostl, & Tai, 2008) Figure 2 5 was developed by the researcher as an integrative description of all the different institutional experiences. The products from each of these experiences provide critical components to the overall system diagram and subsequent dissertation research. The information gained from this process i s then interpreted and communicated back to experts who inform policy making, water managers who make short and long term water allocation decisions, water users who rely on the catchments resources and ecologists focused on maintaining the health of ecolo gical units. The sections that follow will clarify the Figure 2 5 and its components in further detail. The computational elements of Figure 2 5 (ACRU hydrological modeling and Bayesian Networks) are given even greater detail in subsequent dissertation ch apters. Akili Forum Results : A General Systems Diagram of the Crocodile River In modeling socio ecological systems, it is imperative to consider a variety of environmental constants, stressors and drivers that determine the overall state of given system. Natural systems such as wetland, riverine, marine and terrestrial ecosystems are characteristically dynamic with different factors coming into play in determining their functional characteristics such as species biodiversity, and

PAGE 44

44 ecological productivity. In cases where human systems interface with natural systems, it affects their ability and efficiency in providing ecosystem goods and services. Therefore, the stability, productivity and health of these ecosystems depend on their ability to remain resilien t in the face of multiple stressors both naturals and artificial; such as pollution, eutrophication, invasive alien species, natural disasters, encroachment and seasonal variations in climate. Within the Akili forum, the initial step in creating a systems diagram involves defining the level of complexity that will be considered for analysis of the system which allows for setting of the boundaries. It was found that stakeholders and collaborators have different concerns and different worldview hence a broad based consultation was used to establish these limits. Collaborative sessions between researchers and stakeholders were the bases for the final system diagram as shown in Figure 2 6 System components were classified into two broad categories based on whe ther they were functional in a human or natural system. These were then sub categorized according to the following regime: Abiotic drivers and variables in the system such as climate, water quality, sediment and flow. Anthropogenic factors which are co nstructed primarily for economic activity or socio ecological management at various scales. These include management institutions, government, mining and forestry companies, tourist establishments such as the Kruger National Park and other natural reservat ions and individual stakeholder or stakeholder groups

PAGE 45

45 Ecological habitats and locations such as riparian zones, climatic zones such as the high veld to the East of the catchment and the low veld further west; resources that these various classification u tilize and compete for such as water, food, land and economic activity. The principal focus in human systems was the economic activity while in ecological systems they are sustenance of biodiversity, ecosystem health and functional integrity. The main eco nomic activities in the Crocodile River that drive the economy were determined to be forestry, mining, tourism and commercial agriculture in citrus and sugarcane. These are also major competitors for both land and water resources to the ecological systems such as riparian regions and water ways that require these resources. Furthermore, other system processes such as sedimentation and water pollution that affects water quantity and quality are controlled significantly by these activities. Commercial farming and forestry have been established as a source of both point and non point pollution along the Crocodile River in addition to being responsible for significant stream flow reductions. The interactions between them were iteratively discussed amongst the s takeholder after establishing the components. This allowed the process of identifying sub systems within the overall picture that interacted and reinforced negative and positively, feedbacks within the system and balancing loops. Moreover, the process rev ealed latent relationships and interplay between these subsystems e.g. on a larger scale, the interaction between the ecological and human systems and on a smaller scale the positive or negative interactions between agricultural and tourist activities,

PAGE 46

46 eco logical health and degradation and agricultural activities and human infrastructure and ecological health. A systems development exercise on the Crocodile River catchment played a role in helping the various stakeholder groups understand the broader funct ion of the catchment and most importantly highlighted the role that their activities especially in the consumptive use of physical resources played in the overall function of the system. This was done collaboratively during stakeholder meetings. This facil itates a more collaborative regime for management of the catchment as each stakeholder group is cognizant of the ripple effect their activities can have in the system. It also improved the knowledge of ecological managers about the system. This approach i mpresses attention on balancing of resource allocation and decision tradeoffs. Aligned to this it sets a stage for the development of computer based models and decision support tools that are useful in the overall decision framework used within the catchme nt (Shachter R. 1988) In this respect, it is necessary to develop decision support systems and tools that enable environmental managers and policy makers to make informed choices and to enhance the overall health of the eco systems while catering to the variety of and services becomes relevant. The ACRU model as a hydrological model for generating flow regimes for the Crocodile River A full analysis of the myriad of problems faced in water resources requires the use of tools such as hydrological models, expert systems, GIS and statistical analysis techniques. Hydrological considerations are important component information required

PAGE 47

47 for catchme nt managers. This information is available from historical streamflow and climate data. In areas where long term streamflow data is unavailable and cases where future flow predictions are required, models are used to transform climate data (precipitation a nd evaporation) into streamflow which takes into consideration land surface and soil characteristics of a catchment (Jewitt, Garratt, Calder, & Fuller, 2004) The ACRU model is a daily time step model that was selected specifically because it was developed for southern Africa hydrological conditions such as the unique soils and vegetation (Jewitt, Garratt, Calder, & Fuller, 2004) Since its initial design, the model has undergone several updates and revisions including addition of groundwater modeling, water quality and reprogramming in an object oriented form (Kiker & Clark, 2006) It is widely used and has been validated and calibrated under various conditions internationally. Since the 1980s, the ACRU model has been used in a wide variety of catchment scale projects from irrigation demand (Hendricks, Shukla, Martinez, & Kiker; Jewitt, Garratt, Calder, & Fuller, 2004) to urban/peri urban issues to climate change (Schulze, 1989) The model (Figure 2 7 ) has been coupled with several interfaces to allow practical catchment modeling at multiple scales (Kiker G. 1998; Walburton, 201 0; Kienzle, 2011) Bayesian Network Development for Crocodile River Bayesian networks are decision tools that use conditional probabilities about the occurrence of events in socio ecological systems. The structure consists of a directed acyclic graph of a joint probability distribution over a set of statistical variables made up of nodes that represent the variables. These variables are connected by causal dependencies based on direct connections, relationships, mathematical or statistical

PAGE 48

48 associations. T his allows to account for the impact of uncertainty in the decision making process by balancing the desirability of outcomes against that the management option may fail or succeed (Cain, 2001; A mes, Neilson, Stevens, & Lall, 2005; Barton, Saloranta, Moe, Eggestad, & Kuikka, 2008; Pearl, 1988) Bayesian networks are generally constructed by considering management system variables in a system which are represented by nodes or variables each with a finite set of mutually exclusive states with variable probability. These nodes are connected by directional links that represent casual relationships between the elements. Each node in this network is assigned set of probabilities each representative of the uncertainty of a state (Batchelor & Cain, 1999; Bashari, Smith, & Bosch, 2008) can be used to express how the relationships betwee n the nodes operate. In complex socio ecological systems such as the Crocodile River, the various management nodes can be classified to allow for a structured view. The categories allow for separation of nodes according to whether they are implementing fac tors, interventions, intermediate factors, controlling factors, objectives and additional impacts. The relationships between these components are explained and described graphically in Figure 2 7 and explained in Table 2 6. The process of assessing the im pacts of water availability on ecological health and economic state of this catchment begins with identifying important endpoints, drivers, variables and systems by consulting with stakeholders, experts and managers involved in the management and use of th e Crocodile catchment. This conceptual

PAGE 49

49 model described previously, formed the basis for the construction of the Bayesian network. Of interest to water managers in the Department of Water Affair is the water resource sustainability in terms of its consump tion, allocation and supply. Water supply in this catchment is dependent on precipitation hence is driven by seasonal variation in rainfall. The Kruger National Park focuses mainly on monitoring the ecological status in the interest of ensuring the integ rity and health of the river and its ability to sustain biodiversity, ecosystem integrity and tourism activities which are an income earner for the region. The regional governments at various scales are focused on the economic performance of the various se ctors in the catchment area, specifically, commercial forestry, mining, agriculture and tourism. Of importance to the ecologists is the ecological status of the river which provides a system of monitoring the health of the river As a multifaceted catchm ent, the Crocodile River is characterized by a diversity of connections between dynamic socio ecological components. As a result, the ecological state of the river system is l state varies due to the availability of water within the system. Naturally, aquatic and river systems depend largely on water flow for their continued sustenance and health (Kleynhans, Thirion, Louw, & Rowntree, 2008) The c ausal connections between the variables are all probabilistic and are defined using a combination of aggregated data, expert knowledge and stakeholder perspectives. The Bayesian network is an abstraction and simplified view of the variables and the relatio nships that influence the economy, water sustainability and ecological health. It is sufficient to explore various

PAGE 50

50 management issues and natural resource scenarios. Bayesian networks were preferred as they can be used to incorporate different quantitative and qualitative data types essential for such an integrated water management study. Discussion Decision support tools are an important component of decision support systems and an adaptive management framework. The Crocodile River is a complex socio ecol ogical system where the management system requires consideration for the diversity of actors, stakeholders, systems and perspectives in the catchment. The development of a decision support tool in this context is an iterative and consultative process. In this respect, development of decision support systems and tools that enable environmental managers and policy makers to make informed choices to enhance the natural resource extraction and benefit from ecosystem goods and services becomes relevant. In so far as sustainable policy and management choices are concerned, environmental decision support systems and models t o provide predictive tools for consultation with st akeholders while taking into account the effect of an ever changing social, political and natural environment, and consider the consequences of allocation of resources to stakeholders versus leaving them within ecosystems. Lastly the effects of alterati ons of the ecological systems on their ability to remain productive and diverse. A description of the catchment from a complexity and resilience gives a broad understanding of the ecosystem characteristics in terms of it social, political and environmenta l characteristics. In addition to highlighting vulnerabilities and emergent issues in the catchment, this allows for the development of an informed conceptual model of the catchment. The process of the forming a conceptual model also provides a

PAGE 51

51 forum for t he incorporation of expert opinions and stakeholder values and perspectives. This then translated into a Bayesian network which provides a structured method of both quantifying stakeholder/management perspectives and in accounting for uncertainty (Cheng, Bell, & Liu, 1997; Ni, Philips, & Hanna, 2010; Nyberg, Marcot, & Sulyma, 2006) these decision support tools in the overall adaptive ma nagement framework currently being implemented in the Crocodile River Catchment.

PAGE 52

52 Table 2 1 Classification of variables for the Bayesian network of the Crocodile River System Categories Description Example Ob jectives Factors that management actions seek to change improve or prevent from deteriorating. Economy growth Sectoral water supply Ecological integrity Interventions Factors that are implemented in order to achieve your objectives. Also can be though t of as management options. Alter forest cover Implement reserve Curtail water use Dam construction Intermediate Factors Factors that link objectives to interventions. River Flow Agricultural productivity Tourism revenues Controlling Factors Factors w hich cannot be changed by interventions but control or are drivers in the environmental system Rainfall Government Implementation Factors which affect whether an intervention can be successfully implemented both immediately and in the future. Stakeholder consensus Government funding Institutional program implementation Additional Factors Factors which are changed as a result of interventions that do not affect anything else in the environmental system Crocodile Population Bird population

PAGE 53

53 Figure 2 1. Adaptive Cycles in Complex Systems. A stylized representation of the four (Kiker G. A., 2001) Figure 2 2. Historical time periods in South Africa

PAGE 54

54 Events: ( 1) Pre colonial period (African Customary Law) ;(2) Dutch colonial period (mineral rush and est. of urban centers);(3) Formation of the Union of South Africa ;(4) British colonial period;(5) National Party comes to power;(6) Apartheid period;(7) Completion of Kwena Dam;(8) Democratic government established under Nelson Mandela;(9) Post apartheid period Figure 2 3. Adaptive cycles in the Crocodile River since between 1600 Present Day

PAGE 55

55 Figure 2 4.Current real time decision support system in the Crocodile River

PAGE 56

56 Figure 2 5 A conceptual map of tools and processes for estimating ecological status from hydrology and human decision making in the Crocodile River.

PAGE 57

57 Figure 2 6 System diagram showing interactions between components Figure 2 7 Th e ACRU agro hydrology model developed by the School of Bioresources Engineering and E nvironmental Hydrology, U niversity of KwaZulu Natal (Schulze, 1995).

PAGE 58

58 Figure 2 8 Network Structure of Main Categories (Cain, 2001)

PAGE 59

59 CHAPTER 3 EVALUATING WATERSHED MODEL PERFORMANCE AT TWO SPATIAL SCALES TO AID ADAPTIVE MANAGEMENT IN THE CROCODILE RIVER CATCHMENT, SOUTH AFRICA Over view of Strategic Adaptive Management in the Crocodile River Catchment South African watersheds are severely water stressed with competing stakeholders such as irrigation, forestry, mining and municipal water users. Various researchers have noted that inc reased agricultural abstraction, afforestation and mining are regarded as the primary driver for the increasing demands on water resources (Pollard, Du Toit, & Biggs, 2011) Over a decade ago, the post apartheid government pass ed the National Water Act of 1998 to begin to address both the historic inequities in the water resource allocation to stakeholders and the sustainability in the exploitation of future water needs for agricultural, industrial, ecological and human water us e. The central foundations of the National Water Act of 1998 were the prioritization of human and ecological water use by formation of the reserve the separation of water and land rights to address historical inequities and the formation of catchment leve l water management agencies to implement consensus based adaptive management strategies (National Water Act 1998; Stone Jovicich, Lynam, Leitch, & Jones, 2011; Du Toit, Biggs, & Pollard, 2011) The Inkomati Cat chment Management Authority (ICMA) was one of the first of 19 catchment management agencies (CMA) to be developed under the authority of the National Water Act to allocate water use and development of the Inkomati Water Management Area (IWMA) that includes Komati, Sabie Sand, Crocodile and Nwanedzi River catchments (Figure 3 1). The Kruger National Park Rivers Research Program (KNPRRP) (Breen, et al., 2000) which pre dated and helped shape the drafting of the

PAGE 60

60 National Water Act wa s initiated to address concerns about water quantity and quality of the KNP rivers. More recently this legislative framework has resulted in the implementation of initiatives such as the Ecosystems, Protected Areas and People Project of the International Union for Conservation of Nature (IUCN EPP) and more recent work under the Shared Rivers Initiative (SRI).To date, progress towards reconciling user needs with significantly limited water flows has been a challenge especially in following an adaptive manag ement approach where learning by doing is reinforced with history, precedent and evolving user rights. Open, collaborative and participatory approaches to water management are seen as a way forward for parks, conservation and science by various leading res earchers in South Africa (Etienne, Du Toit, & Pollard, 2011; Pollard, Du Toit, & Biggs, 2011; Venter & Deacon) In the context of a Strategic Adaptive Management (SAM) framework (Pollard, Du Toit, & Biggs, 2011) water within a catchment is a common pool resource for individuals, industrial and municipal consumers, agricultural enterprises, and the environment. Allocation decisions are made collaboratively at various management scales under consideration as propos ed by Noss (1990) .Management scales refer to the spatial resolution at which allocation decisions on demand and supply are executed. These are sub catchment allocations for individual farmers, basin scale allocations f or municipal water supply and inter basin transfers at a catchment scale. Hydrological models provide a means for water managers to develop test management scenarios that consider the climatic, biophysical and socio economic drivers within the catchment at these various scales (Rogers & Biggs, 1999)

PAGE 61

61 Models are often used to manage water resources, forecast streamflow, and guide allocation decisions therefore; any improvement in model performance should decrease the uncertainty in management decisions (Rogers & Biggs, 1999) Model performance is judged by comparing the simulated values to the corresponding observed data as a benchmark. Hydrological performance measures and a discussion of their suitability can have been describe d in various publications (Legates & McCabe Jr., 1999; Krause, Boyle, & Bse, 2005; Willmott & Matsuura, 2005; Willmott, et al., 1985; Ritter & Muoz Carpena, 2013) In the Crocodile River hydro logical models are used by water managers to achieve the following water resource management goals: Guide allocation decisions in real time to the various stakeholders and water users Forecast streamflow for future catchment wide planning The developme nt and testing of an ecological reserve mandated under the National Water Act of 1998 (National Water Act 1998) which is a recommended flow levels for the sustenance of sensitive ecological units that depend on the rivers waters (Hughes, Louw, & Mallory, Methods and software for the real time implementation of the ecological reserve : explanations and user manual (No. 1582/1/08), 2008) In moving toward these objectives, significant progress has been made to date within the C rocodile River and greater Inkomati basin. As a significant stakeholder in the IWMA, the Kruger Park has formulated well organized and documented objectives towards ecological monitoring in the Crocodile River (Rogers & Bestbier, 1997; Rogers & Biggs, 199 9) Science and policy integration efforts provide functional and CMA relevant, adaptive learning/management frameworks that include systematic use of discipline specific and scale focused insights (Kiker & Clark, 2006) Thes e frameworks

PAGE 62

62 show that different levels of discussion require diverse types of information to create adaptive, watershed level policies. (Breen, et al., 2000) South African national parks adopted SAM as a mean ecosystems and over exploited water resources (McLoughlin, Deacon, Ababio, & Sithole, 2011) The SAM framework is implemented in collaboration with the Inkomati Catchment Management Agency. Under the KNPRRP, various thresholds of probable concern (TPCs) have been adopted by the KNP and other entities to provide specific criteria for analysis and action (Rogers & Bestbier, 1997) Specific TPCs for the Crocodile River have been proposed for w ater quantity (Kruger National Park, 2006; McLoughlin et al., 2011 ) and for water quality (Rogers & Bestbier, 1997; McLoughlin, Deacon, Ababio, & Sithole, 2011) ecosystem comp onents such as river flow, water quality, geomorphology and biotic components such as fish, macro invertebrates and riparian vegetation based on influences from scale dependent natural and anthropogenic influences such as large scale irrigation operations and forestry that affect the water quantity and quality of the river system (Foxcroft, 2009; Kleynhans, 2007; Department of Water Affairs and Forestry, 2009; Rogers & Biggs, 2003; Rountree, McLoughlin, Mackenzie, Deacon, & Sithole, 2008) Biological integr ity and adaptive assessment/management studies have been conducted on the Crocodile/Elands River (Roux, et al., 1999) a Rapid Response System (RRS) has been integrated into the SAM framework that serves a

PAGE 63

63 levels relative to the in to sustain healthy river systems) which have traditionally been used by the KNP to monitor rivers (Council of Scientific and Industrial Research, 2002) In the South Africa, hydrological models such as the SPATSIM (Hughes, Louw, & Mallory, 2008; Hughes & Hannart, 2003) and the MIKE FLOOD Watch and WReMP system (Department of Water Affairs and Forestry, 2008) by mimicking natural flow variability (McLoughlin, Deacon, Ababio, & Sithole, 2011) Therefore it is reasonable to conclude that any improvement in model performance should decrease the uncertainty in management Study Area The Crocodile River Catchment and its tributaries (Figure 3 1) which falls under the jurisdiction of the ICMA is one of the la rgest irrigation areas in South Africa and provides the sole source of water for wide variety of users (Roux, et al., 1999) The river catchment area is characterized by a mixture of land uses including urban, peri urban/informal/rural settlements, nation al and private wildlife reserves and production agriculture (primarily sugarcane and citrus). These various interests have been vying for the limited water resources in the region. The Crocodile River study area falls within the boundaries of the Inkoma ti Water Management Area and covers the entire X2 secondary catchment with a total area of 10,446 km 2 (Figure 3 1). The catchment is divided into four tertiary catchments namely the Upper Crocodile (X21), Middle Crocodile (X22), the Kaap (X23) and Lower Cr ocodile River (X24).The Upper Crocodile which covers 3090 km 2 has three main sections of the river which are the Crocodile River upstream of Kwena Dam, the

PAGE 64

64 Crocodile River Downstream of Kwena and the Elands. The Middle Crocodile that covers 1573 km 2 has tw o main tributaries namely the Nelspruit and White River. The Kaap which covers 1640 km 2 has three namely the Noordkaap, Suidkaap and Queens River. The Lower Crocodile which covers 3349 km 2 is the largest and borders the Kruger National Park to the North. E ach of the tertiary catchments are further disaggregated into eight quaternary catchments labeled from A, B,C,D,E,F,G,H,J and K (Table 3 1). Recently, each of the quaternary catchments was again subdivided into three quinary catchments labeled 1, 2 and 3 (Mallory S. 2010) For the purpose of this study, several externally located quaternary catchments that had no upstream contributing catchments were selected (Figure 3 2). These catchments were also determined to have accepta ble, multi year, observed monthly flow data (Department of Water Affairs and Forestry, 2008) to enable statistical comparison with simulated results. Currently, the DWAF are using monitoring stations on the Crocodile River to implement a comprehensive a re al time Decision Support System (DSS) to supervise efficient allocation of water resources from the Kwena Dam to ;(1) supply water to urban, rural and agricultural users; (2) provide water to satisfy environmental demands downstream in the KNP; (3) to fulf ill international water agreements under the IIWUA with Mozambique and Swaziland (Inkomati Catchment Management Agency, 2012) The DSS centers on a real time software framework based on the MIKE FLOOD WATCH system (Hughes, Louw, & Mallory, 2008; Danish Hyd raulic Institute, 2008) MIKE FLOOD WATCH integrates spatial data, real time flow data, forecast models and dissemination tools in a GIS environment in addition to running third party model

PAGE 65

65 engines for further assessment and impact. Longer term operations are determined with a combination of two models. Annual operating rules and monthly restriction rules are determined using the Water Resource Modeling Platform (WReMP) (Mallory, Odendaal, & Desai, 2008) and the results implemented and published using the M ike BASIN model (Department of Water Affairs and Forestry, 2008) Short term operations water quantity and quality management will be achieved using the MIKE 11 model system which is a dynamic one dimensional water resources planning modeling tool (Danish Hydraulic Institute, 2008) Given the significant investment and monitoring for efficient water allocation in the basin, there are lingering questions concerning the best scale and use of observed data and hydrological models for building confidence betwee n projected flows and performance metrics. Research Hypothesis and Objectives Within South African watersheds, streamflow simulation of potential management scenarios and climatic events is crucial to successful water management. Often, models must be cons tructed with limited data. Results from models calibrated under these conditions have increased uncertainty, which may have unexpected consequences for water management decisions. As technologies and data access have been implemented in South Africa, water shed data has been developed at finer spatial resolutions to ostensibly improve the quality and accuracy of simulated results. The disparity at the different scales of measurement can occur due to variable rainfall patterns, changes in vegetation and soil surface profiles. Given the availability of different scales of watershed data, specific questions occur as to the proper scale of (Noss, 1990) needed by water resource planners.

PAGE 66

66 The following hypo theses were established to address the problems associated with modeling and subsequent management of streamflow s by investigating uncertainty at different scales in the Crocodile River with the second goal of providing a comprehensive approach to evaluate hydrological models. Modeling basin streamflow s at a finer catchment resolution (quinary scale) improves model performance in comparison to observed flow data than a coarser (quaternary) scale. An expanded statistical analysis of simulated and observed f lows will help to highlight both uncertainty in the observed data and the overall model performance. Quinary scale models will have less bias than the quaternary scale. Calibrated models at finer spatial resolution can better inform management decisions an d regulation of water resources. In addressing these hypotheses, the following objectives provided a framework for this study: Execute the ACRU model (Schulze et al., 1995; Kiker et al., 2006) in selected Crocodile River catchments at the quaternary and q uinary scales. Evaluate the predictive ability of ACRU by evaluating the a series of statistical time series analysis techniques ( Ritter, A., and Muoz Carpena, 2013) including the coefficient of efficiency ( C eff ), root mean squared of the error ( RMSE ) a nd more traditional methods such as the coefficient of determination ( R 2 ). Discuss the implications of uncertainty in measurements due to systematic or random errors

PAGE 67

67 Discuss the implications of simulation scale for water management decisions, TPC develop ment and regulatory frameworks. Methods and Tools The Agricultural Catchment Research Unit (ACRU) model The Agricultural Catchments Research Unit model (ACRU) is a versatile watershed scale model that can be used to simulate various hydrological processe s such as runoff, streamflow, reservoir yield, irrigation water demand and supply, sediment management, regional water resources assessment and water resource utilization ( Schulze et al., 1995; Kiker & Clark, 2006) ACRU can represent large catchments by s imulating lumped sub catchments or as distributed cell type model to account for spatial and climatic variability. ACRU was built to model watershed conditions in southern Africa where slopes are high, the water table is deep, and soils are not particula rly sandy. The hydrology portion of the model has been calibrated and validated as a useful model for various watersheds particularly the Umgeni Catchment (Kiker & Clark, 2006) and the Luvuvhu River of South Africa in which it has been used for the assessment of land use changes (Jewitt, Garratt, Calder, & Fuller, 2004) Previous verification studies have been done shallow silts and clays with low water tables (Jewitt & Schulze, 1999) ACRU has been widely used for watershed modeling in South Africa. Precipitation data is provided by the DWAF (Department of Water Affairs and Forestry, 2008) at both quaternary and quina ry scales. Evaporation takes place from previously intercepted water as well as simultaneously from the various soil horizons The data for pan evaporation if available is used to account for water loss by

PAGE 68

68 evapotranspiration. Evaporation from soil surfaces is determined using a soil water balance equation as described by Ritchie (1972) Later the runoff is determined using the modified curve number equation and storm flow is generated using the SCS equation adapted for Southern African soil and land use conditions (Schulze et al., 1995) which were developed for empirical analysis of water catchment areas after accounting for initial abstrations such as infiltration, depression storage and interception (United States Department of Agriculture, 1972) The model was parameterized based on detailed land use and soil map information obtained from the ICMA. This data formed the basis for configuration of soil, land use and surface characteristics. The ACRU system can use parameters from preset databases that incorporate the parameters about the catchments surface characteristics and constants that describe elevation, slope and base vegetation (Water Research Commission, 2006) The parameters used in this study are based on a default quaternary catchment database that is contained in a system of menu file options that have been created for all of South Africa (Schulze, 1995) The two main climate inputs in ACRU are the rainfa ll and evapotranspiration. Rainfall and evaporation data is required for the yield modeling at the two selected scales. The quaternary scale catchment rainfall and evaporation data was obtained from the DWAF hydrology database while quinary scale climate d ata derived from the gridded South African Atlas of Agro hydrology (Water Research Commission, 2006) Monthly correction factors were applied to each rainfall data to account for spatial variability at this scale (Schulze, 1995) The monthly pan evaporatio n data is also corrected for evaporation at the quinary scale using a monthly factor.

PAGE 69

69 Detailed land use and land cover data was provided by the ICMA based on a detailed land use study (Inkhlakanipho Consultants, 2009) Land us e in the Crocodile River Catchment area consists of plantation forestry, irrigated sugarcane and five major irrigated subtropical fruits, including, oranges, grapefruit bananas, avocadoes, and mangoes (Crawford, 2004) Several la nd use categories within quaternary and quinary catchments were summarized and translated into ACRU land cover categories that are linked to model parameters such as crop coefficients, root depth, interception loss and root depth and distribution constant s by month (Figure 3 4) (Schulze, 1995) The ACRU model has a complementary, quaternary scale database from the Institute for Soil, Climate and Water (ISCW) with soil data and classifications for the whole of South Africa which provides default values for soil types and characteristics for the area of study (Schulze, 1995) However, GIS soil maps made available by the ICMA and the Department of Agriculture and Lands Administration (DALA) were used to provide a more detailed description of the catchments soi l types for the quinary scale model. Selection of Model Evaluation Indicators and Testing Criteria One objective of this study was to determine whether the performance of the ACRU model was influenced by different spatial resolutions. The simulation runs were conducted at quaternary and quinary scales. Monthly flow results from each ACRU/scale simulation were compared to the observed flow data for six test catchments. The primary differences in the two modeling scales are the resolution of the parame ter values for climate, land use and soil. This section presents and describes the statistical tests conducted by comparing the simulated and observed data sets from the test catchments. The model was

PAGE 70

70 parameterized to reasonably depict the actual geomorp hic, agricultural and land use characteristics based on historical maps. The simulations for the six test quaternary catchments are run for between 5 to 10 years depending on the observed data availability for different time periods between 1950 and 1993. The streamflow values were simulated at a monthly time step and compared with monthly observed streamflow s. As each observation point had a different period of operation, each of the test catchments was simulated results was selected to compare with observ ed data. Comparison of the simulated and observed results was performed by regression analysis of the simulated and observed data. The performance of the ACRU model in terms of the predictive ability and accuracy was evaluated for monthly flows by cond ucting a statistical comparison of the simulated model results with the observed data. of used to assess the model results as it compares to reality. The traditional two measures th at were selected were the coefficient of determination ( R 2 ) which measures the degree of co linearity between model simulated variates and the coefficient of efficiency ( C eff ) which is the ratio of the mean square error to the variance in the observed data subtracted from unity (Legates & McCabe Jr., 1999) Additionally, the root mean squared error of the residuals ( RMSE ) was calculated for each data set to provide an estimate of the accuracy of model predictions. The RMSE expresses the average model predic tion error in the units of the variable of interest (Willmott, et al., 1985) Regression analysis is accomplished by fitting an equation of the form y = a + b x fitted

PAGE 71

71 to the one to one regression line between the observed and simulated streamflow s on the x and y axes respectively. The coefficient of determination or simply the R 2 value describes the total variance in the observed data that can be explained by the model. The value of R 2 ranges from 0.0 to 1.0 with higher values indicating better agreement. The coefficient of determination is given by (3 1) where Oi and Pi represent monthly observed and predicted streamflow s respectively of sample size n and represent mean observed streamf low and mean predicted streamflow s respectively. Coefficient of determination ( R 2 ) describes the total variance in the observed data that can be explained by the model. Although useful in indicating of R 2 has the disadvantage of being insen sitive to additive and proportional differences between simulated and observed data (Willmott, et al., 1985) As a correlation measure, the R 2 value is also sensitive to outliers in data which can lead to inaccurate interpretation of model performance. For instance, a model with several extreme events like spikes in rainfall or abstraction will have an artificially high value of R 2 Similarly, several equally deviated events will have high R 2 values since by the law of averages; errors in measurements will cancel leading to a false conclusion on model correctness. Generally, R 2 values ranging from 0.7 0.9 are viewed as favorable in that most of the variance in the observed data is explained by the model (Legates & McCabe Jr., 1999; Krause, Boyle, & Bse, 200 5)

PAGE 72

72 The RMSE is used to represent average difference in the observed and simulated data. It can be interpreted as the standard deviation of the unexplained variance. It is a good predictor of how accurately the model predicts the response. Lower values of the RMSE indicate a better fit. The RMSE is given as (3 2) Average error between observed and predicted streamflow is squared in the calculation of RMSE to remove the negative sign if it exists hence only the magnitude of the error influence the average error measure. Willmott et al. (2005) have demonstrated that it is impossible to discern to what extent RMSE reflects central tendency (average error) and variability within the distribution of sq uare errors in simulated data (Krause, Boyle, & Bse, 2005) The coefficient of efficiency ( C eff ) has been employed widely to evaluate model performance (Wilcox, Rawls, & Brakensiek, 1990; Leavesley, Lichty, Troutman, & Saindon, 1983) In this study it is used in conjunction with the R 2 to validate the simulated results from the ACRU model. The coefficient of efficiency also referred to as the Nash Sutcliffe Index or efficiency index in literature has values that range between infinity and +1. Typical valu es range between 0 and 1 for unbiased model results and for biased models the efficiency index may be negative (McCuen, Knight, & Cutter, 2006) The coefficient of efficiency is calculated as follows

PAGE 73

73 (3 3) SD represents the standard deviation of the observations. In a validation exercise where the square of the differences between the model simulation and the observed data is as large as the variability in the observed data, the value of efficiency i ndex will be closer to zero. In this case, the observed mean is just as good a predictor as the simulation results. Values lower than zero therefore mean that the observed mean is a better predictor. The disadvantage as is the R 2 value is that C eff is sens itive to outliers and extreme values (Legates & McCabe Jr., 1999) In order to efficiently compare time series data from simulated and observed sources, the FITEVAL tool was constructed (Ritter A. 2013) This model independen t software application has been developed in Matlab 2007b and requires input files in the ASCII format that contains two vectors or columns (Ritter A. 2013) FITEVAL provides a graph of a time series of the observed and simulated data, the 1:1 graph of observed and simulated data with calculations of the C eff and R 2 with corresponding 95% confidence intervals and cumulative probability graph of the C eff The output also contains a goodness of fit evaluation (Ritter & Muoz Carpena, 2013). The goodness of fit is achieved by using the FITEVAL software which uses C eff and RMSE graphical comparisons, and bias and outliers tests. Based on the probability distribution of the calculated C eff and RMSE values it is possible to generate a criteria for the evaluati on of the goodness of fit as presented in Table 3 2 below. In the case where this distribution is unknown, a bootstrapping method as described by Efron (1979) is

PAGE 74

74 used to generalize criteria (Ritter & Muoz Carpena, 2013) .The 95% confidence interval is calc ulated for both the C eff and RMSE as a method of correcting for model bias. Additionally, the model bias is explicated as a percentage of the error from the mean and the mean absolute error (MAE) calculated to quantify and explain the presence of outliers in the data. Results Evaluation of Model Performance Yield modeling in the catchment has an impact on catchment level decisions as the simulated discharges are the basis for assessment of design flows and water use at the outlet of each sub catchment (Smit hers, Schulze, Pike, & Jewitt, 2001). The results from the model evaluation of the quaternary scale and quinary scales are presented in Table 3 3 and 3 4 respectively. Observing the statistical measures for both model tests reveals the following trends. Fir st, that the results obtained using FITEVAL for the simulations at the quinary scale have a generally higher C eff and R 2 values than the simulations at the quaternary scale compared against the same observed benchmark. Secondly, the results in both cases a re significantly better correlated Upper Crocodile versus those in the Middle Crocodile and in the Kaap River. The resulting goodness of fit evaluation provided from running the FITEVAL software supports the significant correlation in upper basin catchmen ts indicated by C eff and R 2 values. FITEVAL analyses show that the C eff values are more adequate at a quinary scale as they have a higher probability in most cases of falling within higher ranges of acceptability than in the quaternary case. The C eff value s in the upper basin catchments similarly show more statistical relevance based on the goodness of fit evaluation (Table 3 4). This evaluation also reveals that the model bias trends that

PAGE 75

75 provide insights on the presence of outliers in the data which is im portant as it provides clues on sources of uncertainty. For illustrative purposes the graphical results show the results for catchments X21H (Figure 3 6) and X23C (Figure 3 7). This rendering provides a comprehensive method to evaluate model performance t hrough graphical outputs, plots of observed and simulated streamflow, goodness of fit correlation indicators, and statistical inference results to determine the model bias and effect of outliers on the indicators. The figures show a comparison of the one t o one graph of the simulation results at a quaternary and quinary scale with a corresponding comparison of time series streamflow simulation results. Comparison of Results at Quaternary and Quinary Scale Simulation at quaternary scale are satisfactory wit h a R 2 ranging from 0.49 in the Queens River at Sassenheim (X23E) to 0.74 in the Crocodile River at Badfonteins with a corresponding C eff from 0.24 bounded at a 95% confidence interval of ( 0.21 0.55) to 0.73 bounded at a 95% confidence interval of (0.49 0.89). The results for X21C, X21H,X22A and X23C in Table 3 3 show satisfactory correlation of model and observed data. The FITEVAL evaluation indicates that the C eff may be influenced by model bias. This demonstrates that it is important to evaluate these outputs on multiple criteria. Regression slope values are generally larger than 1.0 which shows a constant over prediction of observed streamflow s by the ACRU model when modeled at this scale. This indicates that the simulated streamflow s are consistently under predicted for this catchment. The RMSE values range from 0.73 Mm 3 /month to 6.61 Mm 3 /month. The results in Table 3 4 for the simulation at the quinary scale gave a R 2 ranging from 0.47 in the Queens River at Sassenheim (X23E) to 0.67 in the Wit Riv er at Goede

PAGE 76

76 Hoop (X23C) with a C eff ranging from 0.05 bounded at a 95% confidence interval of ( 0.62 0.44) to 0.64 bounded at a 95% confidence interval of (1.31 2.73). Regression slope values are consistently under 1.0 which shows a constant over pre diction of observed streamflow s by the ACRU model when modeled at this scale. The RMSE values range from 0.87 Mm 3 /month to 4.78 Mm 3 /month. At the quinary simulation scales, the model efficiency exceeds the 0.5 threshold in most cases which according to Le gates and McCabe (1999) indicates that the model is a better predictor than the average. However, at the quaternary scale in X22A, X23C and X23E the simulation falls below 0.5 indicating that mean of the observations would have been a better predictor of t he streamflow than the model. Ritter and Carpena (2013) however propose a higher value of 0.65 as acceptable based on statistical testing on a 95% confidence interval that the C eff is within a range. The simulation results some improvement in the results a t the quinary scale. The variance between simulated and observed data is less evident in the reduction in spread of the correlated data. There is significantly more bias and outlying data effects detected at the quinary level. This may be because the mode l parameterization at a smaller scale provides more information on uncertainty such as information on discharge loss due to unsanctioned withdrawals and hydrological events such as localized rainfall due to microclimates. The bias in the quaternary scal e simulations in all the six catchments a range of between 15.9% to 20.7% bias from the mean while the quinary scale simulations show a lower bias in the range between 15.5% to 15.4% from the mean. The bias is

PAGE 77

77 insufficient as a summary of model performa nce in both management scales since it is ambiguous and inconsistent with our expected hypothesis which assumed that larger bias would be expected at the quaternary scale than at the quinary scale with a consistent positive or negative bias indicating over prediction or under prediction. A positive bias can indicate a systematic under prediction or both under prediction and over prediction with preponderance for under prediction while a negative bias has an analogous ambiguity (W allach, Makowski, & Jones, 2006) Examining the bias however improves our understanding of the quality of the observed data. FITEVAL detects an obvious outlier in X21C when modeled at the quinary scale which affects the indicators. The results indicate the presence or absence of outliers in the observed data with the mean absolute error values showing a wider range in the quinary versus the quaternary simulations. Large positive or negative bias may indicate the presence of outliers in the observed data. It also helps explain the behavior of the other measures such as the coefficients of determination which are sensitive to outliers. Positive bias also indicates predominantly positive residuals which indicate that the model systematically under predicts a t both quaternary and quinary scales. This necessitated investigation into the source of this uncertainty caused by either outliers or systematic errors which is discussed in the next section. For illustrative purposes, the evaluation of statistical signi ficance for two cases X21H and X23C shown in Figure 3 5 and 3 6 is used to demonstrate that while traditional measure show that results as acceptable an evaluation of the statistical significance of the modeled results demonstrates the acceptability of the se results in addition to looking at the goodness of fit measures. At a quaternary scale in both X21H

PAGE 78

78 and X23C, the results have a C eff of 0.61 ( 0.2 0.85) and 0.48 ( 0.20 0.74) respectively. At a quinary scale this improves with a C eff of 0.66 ( 0.26 0.91) and 0.64 (0.48 0.74) respectively. This would indicate an improvement fit in both cases. The 1 1 scatters indicates that the calculated values in both scale deviate from the observations both positively and negatively. Observing the statistical significance of the outputs for X21H shows that at the quinary scale there is a 31.7% chance of obtaining C eff <0.65 0.799 which is deemed Acceptable according to the criteria outlined in Table 3 2. At the quaternary scale the probability of obtaining the same range of in the C eff is 29.5 %. We would therefore conclude that the results at quaternary scale are less acceptable for this scale of measurement. In the X23C catchment, the probability to obtain an Acceptable rating at C eff <0.65 0.799 is higher at the quaternary scale at 41.1% than at the quinary scale at 15.7%. Comparing the two scales indicates that the measure at the quinary scale is less significant. The model bias in the quinary measurement over predicts the mean by 9.6% and in the quaternary it under predicts the mean by 15.9%. This evaluation gives more detailed criteria to determine the acceptability of results as shown in Table 3 5. Effects of Uncertainty in Measurement Uncertainty in measurements can either be caused by random or systema tic errors in the data. Random errors can include outliers in the data which can either be representative of the actual hydrological conditions in the catchment often caused by sudden and extreme weather such as storm events, water theft causing sudden red uctions in discharge or indicate data collection errors. While it is unethical and

PAGE 79

79 inaccurate to correct these points to improve correlation data based on the expected trend in the data, it is valuable to understand and explain the model bias as well as id entify outliers. This diagnosis pinpoints possible errors in the data which can then be either verified as valid deviations or actual errors in the observed data. Systematic errors occur during data collection such as those caused by flow alteration by gau ging structures that create a discrepancy in data quality. The South African Directorate of Hydrological Services (SADHS) in DWAF has a policy of building gauging structures to pre calibrate the discharge at flow stations. The gauging structure creates an artificial control in the river channel that creates a determinable relationship between stage and discharge. The most commonly used gauging structures in South Africa are Crump weirs. For the purpose of evaluation of the effect of uncertainty in observed data for the several test stations, we will assume that the weirs used in the Crocodile River system are Crump weirs. A recent study by SADHS found that the observed flow at the Crump weirs overestimated actual discharge by 4.3% (Wessels & Rooseboom, 2009) To illustrate the effect of this systematic bias in results from two catchments (X21H and X23C) at both quaternary and quinary scale are re evaluated. Once the observed data and its constituent error are modified the C eff v alues improve consistently as is demonstrated an improvement of the goodness of fit measurements and the ratings in the table below. For the purpose of this narrative we will only include improvements at the probability of the fit being evaluated as a perf ormance class of Good Statistical testing is performed at a 0.1 significance level.

PAGE 80

80 The results show a slight improvement on each of the goodness of fit measurements and ratings in each of the test cases. There is a reduced error indicated by a consiste nt reduction in the RMSE and improvement of correlation efficiency with higher C eff values. Various studies have however demonstrated that considering uncertainty simply by using stage discharge relationships as proposed by Wessel (2009) which are tested i n this section (Table 3 6) are insufficient in accounting for measurement errors. Harmel et al. (2006) summarizes several additional error estimates that (Table 3 7) have been proposed by various authors depending on t he data quality and method of calculation ranging from ideal, average to poor quality data with corresponding values of 6.1%, 10% and 20% (Slade, 2004; Pelletier, 1988) Simply put ideal data has less error while p oor quality data is assumed to have significantly more error associated with reported values To explore this concept further the acceptability level of each catchment/scale combination was tested using the same methodology which shows some improvement in all the indicators namely improvement of C eff with progressive correction of the data. The highest acceptability as shown in Table 3 7 is found with the assumption of poor data quality with the C eff improving to a high of >0.8 and a progressive improve ment in acceptability in the goodness of fit. Th ese results show that model performance and thus its acceptability for contributing to management decisions, is strongly associated with the quality of observed hydrological data. The difference in model per formance due to quinary or quaternary scales is generally smaller than the difference in model performance given the different error levels in the observed data.

PAGE 81

81 Given the results of these statistical tests, f urther work is justified to more fully explore the potential errors of observed flow data in the Crocodile River system. Implications of Model Performance on River Monitoring As previous mentioned, the Crocodile is currently managed under the RRS monitoring system which assesses the state of flow in the river real time using worry levels of flow (McLoughlin, Deacon, Ababio, & Sithole, 2011) An example of this worry levels is shown in Figure 3 7 at a monitoring point X2H024 at the mouth of X23C. The point at which worry levels intersect with the f low curves estimate points in time when management decisions are taken. The worry levels are set at high worry the reserve and inform DWAF as shown in Figure 3 7. Assuming these estimated worry levels, a water manager would observe the flow draws down be low the reserve on the first day of August if using the quinary model outputs or observed data. This draw down is reflected nearly 2 months later if relying on the quaternary data on at the beginning of October. In this case, simulated data used to run thi s system hence would benefit greatly by utilizing quinary versus quaternary data which mimic observed data more accurately hence providing for more fine tuned decision points both in space and time. The methodology outlined in this paper therefore not only quantifies errors in model outputs at the two management scales but also provides an implicitly measure of errors in the timing of management decisions depending on the scale of measurement. Yield modeling in the catchment has an impact on catchment leve l decisions as outlet of each sub catchment (Hughes & Hannart, 2003; Smithers, Schulze, Pike, & Jewitt, 2001) In the Crocodile River while estimated runoffs are used with confidence,

PAGE 82

82 phase shifts and peak differences with actual observed flows can affect various management decisions by river managers. Depending on the scale of modeling the shifts vary hence affect management. Phase sh ifts in runoff would affect timing of allocation or restriction decisions which in turn influence demand, application rates for irrigation water and ultimately crop yields. In a practical setting, management flows are used to at various flow levels to take necessary actions depending on ecological reserve requirements. Scale dictates the accuracy of model results that is used to determine the accuracy of endpoints and timeliness of management decisions. Discussion There were two objectives of this study 1) to demonstrate that of the model parameterization at a higher resolution produced better simulation results and 2) to illustrate that model through management scenarios implications for management at the finer quinary scale. Results of this study do not indicate that parameterization at a quinary scale gives significantly better outputs than at a quaternary scale. In comparison with observed flow data, t he simulations are somewhat biased when compared to the observed data. Schulze and Smithers (2001) desc ribe this as an indication of that the base flows are over simulated with a simultaneous under simulation of stormflows. This trend may explain the inconsistency of the bias measurements across different test catchments The variations in the residual betw een the mean and simulated measurements while consistent in terms of directionality vary in terms of quantity leading to ambiguity in the bias measured. An examination of the mean, slope and intercept values of the observed versus simulated regression lin es reveals a recurrent trend in the catchments simulated. The model consistently under predicts base flow with a simultaneous under simulation of

PAGE 83

83 storm flows especially during high rainfall seasons (Schulze, 1995) Intercepts are shown to be too high with slopes being low. This indicates that the area with low intensity winters or consistent with weather conditions within the catchment. It also points to a poor distribution of rainfall over the catchment. However, these results improve with better resolutio n of the catchments rainfall distribution. The basis for this simulation study is the hypothesis that model accuracy is scale dependent. It is assumed that finer spatial and temporal parameterization of the ACRU model will result in better simulation outpu ts. By comparing model results at two different scales, it has been shown that at a finer parameter resolution the model results are generally a better representation of reality. Following the proposed method by Ritter and Muoz Carpena (2013) the goodne ss of fit evaluation performed proved an improvement over the traditional methods of simply comparing the C eff R 2 and RMSE values as it explains and accounts for the statistical ranges of confidence for these indicators. The model behavior in terms of the bias is also explained and tested. The information from the bias led a further investigation on the source of uncertainty due to the presence of systematic bias or outliers in the data. The effect of this uncertainty is further evaluated. The results show ed that correcting measurement error significantly improves the correlation between model outputs and observed data. Cumming et al. (2008) point out that the process by which we learn about the world has two scale dependent components, the actual scale a t which patterns and processes occur and the scales at which we obtain data about them that is used for decision making. In the interest of catchment level hydrological modeling, these two

PAGE 84

84 considerations play an important role in informing the decision mak ing at various scales. Different scales and levels of complexity are required when planning to account for various small scale processes that might affect management decisions at different resolutions. This is particularly important in the current South Af rican climate with the implementation of the National Water Act of 1998 (National Water Act, 1998) In integrating hydrological with ecological models to estimate impacts of environmental flows, it is important to harmonize model inputs at both a spatial a nd temporal scale. The challenge however remains to translate the statistical measures and methods in a way that informs the decision environment in Crocodile River catchment. Scale of measurement therefore dictates the effectiveness of the entire SAM fram ework. Pollard et.al. (2011) notes that adaptive management is not an end in itself but a process that evolves as new learning is brought to bear and promotes the idea of linking science and management. Scale sensitivity introduces new scientific perspecti ves and lessons on uncertainty that should be systematically included in managing the Crocodile River as a complex adaptive system.

PAGE 85

85 Table 3 1 Test catchment description showing predominant land uses, mean Tributary Quaternary catchment Quin ary catchment Flow Station ID Area (km 2 ) MAP (mm) MAR (Mm 3 /a) Land use UPPER CROCODILE Krokodil@ Badfonteins X21C X21C 1,X21C 2, X21C 3 X2H070 311 761 121.8 Irrigation Agriculture Ngodwane River @ Coetzeestroom X21H X21H 1,X21H 2, X2 1H 3 X2H034 228.8 1069 51.6 Forest Plantation MIDDLE CROCODILE Houtbosloop @ Sudwalaskraal X22A X22A 1,X22A 2, X22A 3 X2H014 251.4 990 71.5 Forest Plantation KAAP Wit River @ Goede Hoop X23C X23C 1,X23C 2, X23C 3 X2H02 4 81.3 1134 25.4 Sugarcane Queens River @ Sassenheim X23E X23E 1,X23E 2,X23E 3 X2H008 180.4 1024 36.7 Forest Plantation [a] MAP = Mean annual precipitation [b] MAR = Mean annual runoff Table 3 2 Evaluation ranges of goodness of fit with resp ect to the Coefficient of Efficiency (C eff ) (Ritter and Mu oz Carpena, 2013) Evaluation Range of C eff Very good 0.90 1.00 Good 0.89 0.90 Acceptable 0.65 0.79 Unsatisfactory <0.65

PAGE 86

86 Table 3 3. Comparison of quaternary scale sim ulations and observed monthly stream flow for 5 selected catchments of ,the coefficients of determination (R 2 ), coefficients of efficiency (C eff ), median annual observed streamflow and the RMSE with 95% confidence intervals for C eff and RMSE Catchment Per iod n [a] R 2 C eff with 95% CI Median Flow [b] RMSE with 95% CI [c] Outliers [d] Model Bias [e] X21C 1980 1985 67 0.53 0.50 [ 0.03 0.76] 4.19 6.61 [4.61 8.89] No Yes X21H 1972 1982 121 0.63 0.61 [ 0.2 0.85] 3.11 2.97 [2.22 4.04] No No X22A 19 60 1970 141 0.54 0.49 [ 0.19 0.69] 3.66 2.29 [1.86 2.88] No Yes X23C 1973 1983 132 0.57 0.48 [ 0.20 0.74] 1.16 0.87 [ 0.67 1.18] No Yes X23E 1965 1975 129 0.47 0.05 [0.62 0.44] 0.65 1.87 [1.31 2.73] No Yes [a] Number of observations [b ] Median annual observed streamflow (Mm 3 /month) [c] Root mean squared error of the residuals (Mm 3 /month) [e] Influence of model bias on C eff Table 3 4. Comparison of quinary scale simulations and observed monthly streamflow for 5 selected catchments of the Crocodile River, the coefficients of determination (R 2 ) co efficients of efficiency (C eff ), median annual observed streamflow and the RMSE with 95% confidence intervals for C eff and RMSE Catchment Period n [a] R 2 C eff with 95% CI Median Flow [b] RMSE with 95% CI [c] Outliers [d] Model Bias [e] X21C 1980 1985 67 0.74 0.73 [0.49 0.89] 4.19 4.78 [3.28 7.53] Yes Yes X21H 1972 1982 121 0.67 0.66 [ 0.26 0.91] 3.11 2.78 [2.22 3.61] Yes Yes X22A 1960 1970 141 0.56 0.54 [0.40 0.76] 3.66 2.28 [1.72 3.32] No No X23C 1973 1983 132 0.67 0.64 [0.48 0.74] 1 .16 0.73 [0.67 1.18] No Yes X23E 1965 1975 129 0.49 0.24 [ 0.21 0.55] 0.65 1.68 [ 1.42 2.06] No Yes [a] Number of observations [b] Median annual observed streamflow (Mm 3 /month) [c] Root mean squared error of the residuals (Mm 3 /month) [d] Prese n [e] Influence of model bias on C eff

PAGE 87

87 Table 3 5. Evaluation of acceptability of results from X21H and X23C based on traditional statistical measures and observation of statistical significance Quaternar y Quinary Evaluation Criteria > > GOF (Ceff and R 2 ) Statistical Significance GOF (Ceff and R 2 ) Statistical Significance X21H Acceptable Unsatisfactory Acceptable Acceptable X23C Acceptable Unsatisfactory Acceptable Unsatisfactory Table 3 6. Assessment of the effect of the removal of measurement bias of 4.3% from observed measurements at monitoring points Scale Catchment Reduction in RMSE Improvement in C eff Evaluation improvement in P(0.65>C eff >0.79) Quaternary X21C 6.61 to 6.27 0.5 0 to 0.54 Acceptable (14.9% to 11.8%) X21H 2.97 to 2.76 0.61 to 0.67 Acceptable (31.7% to 31.1%) X22A 2.29 to 2.12 0.49 to 0.57 Acceptable (8.1% to 4.3%) X23C 0.87 to 0.82 0.48 to 0.54 Acceptable (15.7% to 18.0%) X23E 1.87 to 1.8 0 0.05 to 0.12 Acceptable (0 %) Quinary X21C 4.78 to 4.38 0.73 to 0.78 Acceptable (57.7% to 62.8%) X21H 2.78 to 4.39 0.66 to 0.70 Acceptable (29.5% to 33.5%) X22A 2.28 to 4.40 0.54 to 0.66 Acceptable (28.6% to 54.3%) X23C 0.73 to 4.4 1 0.64 to 0.70 Acceptable (41.1% to 53.2%) X23E 1.68 to 4.42 0.24 to 0.30 Acceptable (0%) Table 3 7. Effect of uncertainty correction on acceptability for C eff in X21H Condition Method Uncertainty Reference C eff Goodness o f fit evaluation Very good Good Acceptable Unsatisfactory None 0 0.66 1.6% 16.7% 29.5% 52.2% Crump weir Stage discharge method 4.3% Wessel (2009) 0.70 3.2% 20% 27.7% 49.1% Ideal conditions Direct discharge method 6.1% Pelletier (1988) 0.71 4.0% 21.6% 27.4% 47.0% Average conditions Direct discharge method 10.0% Slade (2004) 0.74 6.0 % 25.2% 25.9% 42.9% Poor conditions Stage discharge method 20.0% Slade (2004) 0.81 14.7% 31.5% 18.6% 35.2%

PAGE 88

88 Figure 3 1 Location of the Crocodile River catchment (Inhlakanipho Consultants, 2009)

PAGE 89

89 Figure 3 2 Map of selected quaternary catchments in the Crocodile River. Each of the six catchments has no upstream contributors along with multi year flow rec ords Figure 3 3. The ACRU model

PAGE 90

90 Figure 3 4. Land use within the Crocodile River Catchment (Inhlakanipho Consultants, 2009)

PAGE 91

91 Figure 3 5. Observed monthly streamflow time series graphs for X21H and corresponding comparison of simulated versus obse rved scatter plot and goodness of fit evaluations at quaternary and quinary simulation scales

PAGE 92

92 Figure 3 6. Observed monthly streamflow time series graphs for X23C and corresponding comparison of simulated versus observed scatter plot and goodness of fit evaluations at quaternary and quinary simulation scales

PAGE 93

93 Figure 3 7. Flow at X2H024 showing three TPC worry levels and the lags in management decisions dependent on scale of measurement.

PAGE 94

94 CHAPTER 4 MODELING WATER USE DECISIONS FOR STRATEGIC ADAPTIVE MANAGEMENT USING BAYESIAN NETWORKS IN THE CROCODILE RIVER Overview of Bayesian Networks in Adaptive Management In order to effectively make water allocatio n and management decisions in any complex socio ecological system, it is imperative that ecological perspectives and stakeholder concerns be incorporated into the overall decision framework. This has encouraged the development of approaches t hat include c ommunity perspectives in water resource, environmental and land use management and decision making (Campbell & Luckert, 2002; Varis, 1997; Stassopoulou, Petrou, & Kittler, 1998) One of these approaches is Ad aptive Management. Adaptive management (AM) is widely advocated and cited by practicioners and scientists involved in the area of water resource management in multifacted catchments as it provides an explicit framework for motivating, designing and interpr eting the results of monitoring so as to improve management of limited environmental resources in water stressed catchments (Holling & Sanderson, 1996; Rogers & Biggs, 1999; Walters, 1986; Ringold, et al., 1996; Parma, 1998; Yoccoz, Nichols, & Boulinier, 2001; Schreiber, Bearlin, Nicol, & Todd, 2004; Nicols & Williams, 2006; Duncan & Wintle, 2008) Despite its growing popularity there are limited examp les of successful adaptive management implementation due to the various challenges such as complexity and uncertainty of natural system, data poor conditions and conflicting structural inequalities that are associated with managing complex socio ecological systems. In dealing with these uncertainties, South Africa ecologists and water resource managers (Biggs & Rogers, 2003) have applied this approach in a process dubbed Strategic Adaptive Management (SAM). SAM is achieved in four steps as follows;

PAGE 95

95 and; (4) evaluation and learning (Pollard & Du Toit, 2005) as river flow, water quality, geomorphology and biotic components such as fish, macro invertebrates and riparian vegetation (Foxcroft, 2009; Department of Water Affairs and Forestry 1999; Biggs & Rogers, 2003) A key element to the successful implementation of a SAM framework is the inclusi on of appropriate dynamic process models that provide a framework for the identification of potential management actions (Figure 4 1) in characteristically ambiguous, complex and data poor decision environments. This is done in four distinct steps which in clude the development of a static system model which includes the component habitats and species and how the ecosystem functions and intersects with social systems. The second step which we focus on in this paper involves predicting potential management ou tcomes and building management scenarios (Walters, 1986; Duncan & Wintle, 2008; Nyberg, Marcot, & Sulyma, 2006; Aalders & Aitkenhead, 2009; Bromley, Jackson, Clymer, Giac omello, & Jensen, 2005; Colvin, Everard, Goss, Klarenberg, & Ncala, 2008; Rogers & Biggs, 1999) We propose using dynamic Bayesian networks (Bns) to provide a method to test and select management options. The third step involves prioritizing, planning an d implementing management actions. The last step entails measuring the outputs of these actions, communicating the results to concern stakeholders and reviewing the effectiveness of these actions.

PAGE 96

96 The advantage of the use of Bayesian network as noted by Ru mpff and Duncan (2011) is that they meet the following three requirements: (1) quantitative predictive capability; (2) representation of uncertainty in states and transitions; (3) amenability to updating as new knowled ge accumulated. Additionally, Bns also incorporate both quantitative such hydrological data and qualitative data types such as stakeholder views and opinions. This research describes the integration of Bns as a tool to facilitate Strategic Adaptive Managem ent in the Crocodile River. We describe the critical role of presented here is the implantation of a realistically complex and dynamic model in the SAM framework. We pre sent an initial review of the construction and implementation of Methodology Bayesian Networks Bayesian networks (Bns) provide a graphical method for representing relationships between variables in an un certain, probabilistic, imprecise and unpredictable environment (Batchelor & Cain, 1999) The outcomes of events occurring variables were established either deterministically or probabilistically using different data sources such as empirical field data from models, surveys and databases or expert opinion. They are constructed by considering management system variables in a system which are represented by nodes or variables each with a finite set of mutually exclusive states with variable probability. These nodes are connected by directional links that represent casual relationships between the elements. Links have direction from cause to ef fect in which parent nodes precede child nodes which are assigned a

PAGE 97

97 set of probabilities each representative of the uncertainty of a state (Batchelor & Cain, 1999) These are contained in underlying conditional probability ta be used to express how the relationships between the nodes operate. Bayesian reflects the knowledge or belief about how a system functions. This belief can be updated with time as the understanding of the system increases based on data or experiences (Nyberg, Marcot, & Sulyma, 2006; Bashari, Smith, & Bosch, 2008) Bayesian nets have three types of nodes based on th e functionality. These are decision nodes, which typically represent management points with measureable impacts on the system. Utility nodes represent impacts of decisions (costs and benefits) and nature nodes which represent internal and external variable s with conditional states (Barton, Saloranta, Moe, Eggestad, & Kuikka, 2008) that shows how uncertainties are updated with reference to new informati on. According to Aitkinhead and Alders (2009) probability distribution being available (i.e. all possible permutations of the variables have been described in terms of their outcomes) which is then updated to the likely outcome which is known as the posterior probability theorem is to capture the probability of an uncertain event or quantity in the form of prior probabilities ; summarize the data, observe the results and summarize the results as a conditional probability or likelihood prior probability by the likelihood (i.e. prior probability x likelihood = posterior probability ) (Ni, Philips, & Hanna,

PAGE 98

98 2010) For S i which denotes one of n mutually exclusive events and D which denotes some diagnostic data where P(S i ) is the probability of S i occurring and P(D) is the probability of D occurring then in its most rudimenta (4 1) P (S i ) is the prior probability of S i and is the posterior probability which is the probability that S i occurs when D is true. Conversely, is the c onditional probability of D when S i is true. The joint probability is obtained by multiplying and which indicates how often S i theorem can be written in different forms with varyi ng levels complexity depending on the number of variables involved. For example in an ecological system, the management context consists of states of a state variable S (ecostatus) describing ecological state as a management endpoint and the process variab le X(flow) that drives changes in the system and whether the intervention D (implement reserve) is operationalized or not. Prior knowledge of the ecological state can be expressed as a probability of a state S given intervention D and process variable X: P (S| D, X). In an impact analysis an ecological manager might be interested in evaluating the posterior probability for a state variable given driver pressure and the mitigating interventions (Barton, Saloranta, Moe, Eggestad, & Kuikka, 2008) In this case we use open source software Netica (Norsys Software Corp., 2012) to implement a Bayesian net for the Crocodile River catchment to evaluate management outcomes under different scenarios.

PAGE 99

99 Const ructing a Bayesian Network The development of Bns for water management comprises four main steps as demonstrated by Ames (2005) shown in Figure 4 2. This procedure and its various sub steps is described below and has been adapted for the development of a Bayesian network for adaptive management of the Crocodile River Catchment as described in the next section Problem definition Defining the problem enables the stakeholders and decision makers input in producing a p reliminary assessment of the important variables, decisions, outcomes and relationships in the system. It is achieved with the following steps: Describe management environment: The initial step involves defining your study area and management environment. A useful approach in describing the study area is the development of a system model which is an explicit graphical representation of how an ecological unit functions that includes components such as habitats, species, actors, institutions and markets which was achieved by using knowledge from expert consultation and stakeholder meetings It describes the connectivity of information. Key natural processes such as ecosystem dynamics, hydrology and climate and anthropogenic processes such as market dynamics, i nstitutions and policy making underpin the development of these connections (Pollard & Du Toit, 2005) Identify management endpoints: Identifying endpoints (e.g. ecostatus and economy) at the onset keeps the Bn focused on the important variables to the decision problems under investigation. The geographic locations at which these endpoints will be assessed are also selected (Ames, Neilson, Stevens, & Lall, 2005)

PAGE 100

100 Identify management alternatives: Ma nagement endpoints are classified into short term and long term planning alternatives that are geared to improving the overall conditions in the catchment. Each management alternative is represented in the net as a separate node. Identify process variable s, drivers and actors: The nodes which represent variables identified by stakeholders and experts in the previous steps are selected. These will be represented as nodes within Bn. These include dynamic process variables which are dynamic and represent or c ontain a consumable resource, actors which consume this resource and drivers that affect the system but cannot be directly controlled or influenced. Establish discretization states for variables: The discretization is done to reflect states that are releva nt to the stakeholders and managers. In the South Africa, river states and flow requirements to maintain ecosystem health have been already defined and are commonly used (River Health Program 2005) Model inference Bayesian between the variables. The inferred from a variety of sources such as observed or simulated flow data, expert judgment and stakeholder consultation Observed and simulated data: Observed streamflow data and results for ecological (1997) propose a procedure for inferring Bn structure and the CPT from these data sets hence: Tabulate the observations of each variable sorting each by parent variable. Observations are discretized at various levels (High, Low or True, False, etc).

PAGE 101

101 Count the child node states ( S c ) for every combination of parent states ( S p ). Probabilities are calc ulated as: P = (4 2) Expert elicitation and stakeholder surveys: Certain qualitative cases arise where should be infe rred from stakeholder elicitation and expert opinions. Within the Netica platform used to build the Bn in this exercise, this procedure is done automatically and rendered in graphical form. Model validation Validation determines if the model gives correct predictions and mirrors the observed phenomena in question. This is particularly challenging in Bayesian models for corroborating outcomes of management options that have never been tested in a particular area. This is particularly true in South African w atersheds where there is a lag in implementation of the National Water Act of 1998 with major delays in promulgation of provisions such as the establishment of the Catchment Management Agencies (Colvin, Everard, Goss, Klarenberg & Ncala, 2008) To validate this model we used results from a previous study by the River Health Program (River Health Program 2005) to reconstruct scenarios and test the predictive ability of the constructed Bn. This is the only river monitoring program that proved useful for validation. Within the Netica platform used to build the Bn in this process is automated from CPT tables and rendered in graphical form. The calculated probabilities for each individual outcome are dynamically rendered as percent bars.

PAGE 102

102 Case Study Area Overview The Crocodile River catchment is part of the Eastern Transvaal Region and falls under the jurisdiction of the trans border Komati River System with Mozambique lying in the North East and Swazil and in the South. The Crocodile River falls under the jurisdiction of the Inkomati Catchment Management Agency (ICMA) (Carmo Vas & van der Zaag, 2003) The Crocodile River has its head water in the Highveld regions east of th e Drakensburg Mountains into the Lowveld and flows steadily east for 320 km with approximately 1200km of tributaries. The Crocodile River joins the Komati River in Mozambique terminates in the Indian Ocean. The Crocodile River forms the Southern border of the Kruger National Park (Department of Water Affairs and Forestry 2009) The mean annual rainfall falls between 600 1100 mm/a with higher rainfall recorded in the Highveld areas. The Crocodile River catchment forms the X2 catchment which is further subdivided into 2 river systems. The Crocodile (East) River which consists of the Elands (X21), Middle Crocodile (X22) and Lower Crocodile (X24) that form the main stem of the river and the Kaap River (X23) joins from the south e ast. These two stems (i.e. the Crocodile and Kaap) form the management zones for this study (Zone 1 and Zone 2). Crocodile River Bayesian Model Development Problem Definition Description of management environment The initial system model was developed d uring the Akili Complexity and Integration Initiative Workshop in which the researcher was a facilitator, was led by Harry Biggs with a group of academics, researchers and environmental managers (Akili, 2010) This was done during a session in which participants collaborated in creating a

PAGE 103

103 system model to describe the Crocodile River Catchment. This model was refined with perspectives and inputs from Catchment Management Strategy (CMS) meetings which were a series stakeho lder meetings moderated by the ICMA. The CMS meetings were part of a catchment wide strategy by the Department of Water Affairs and Forestry to encourage participation of stakeholders in the co management of the Crocodile Rivers The process of development of the system model as determined during the collaborative workshops involved identifying components in the system which were classified according into two broad categories based on whether they were functional in a human or natural system. These were then sub categorized according to the following regime: (1) process variables in the system such as flow and biodiversity that hold the resources consumed in the system; (2) anthropogenic actors which are primarily focused on economic activities. These include municipalities, agricultural water users, forestry companies, tourist establishments such as the Kruger National Park; (3) management endpoints whose state or resource exploitation is monitored. Over exploitation of these resources result in state changes or shifts; (4) management alternatives that are implemented to control the rate of resource exploitation. The principal focus in human systems was economic activity while in ecological systems is sustenance of biodiversity, ecosystem health and functiona l integrity. The economic activities in the Crocodile River were established to be forestry, tourism and commercial agriculture in citrus and sugarcane. These were also major competitors for both land and water resources to the ecological systems such as riparian regions and water ways that require these resources.

PAGE 104

104 This system model developed during the Akili forum formed the foundation for constructing the Bayesian networks as it provided a common mental model or understanding of the system (Stone Jovicich, Lynam, Leitch, & Jones, 2011) by representing various stakeholder perspectives and management scales. These preliminary structures (shown in Figure 4 4) allow users to change the hydrological or socio economic variables in the networks to explore various management outcomes to conform to the provisions of the National Water Act of 1998 (National Water Act 1998) This is also the basis for the organization of the various nodes in the Bn describe d in Table 4 2. Identify management endpoints and zones Selection of management endpoints: The goals for the SAM framework as they relate to the provision of the National Water Act are: (1) water resource sustainability for water users from the perspectiv e of the ICMA and in the KNP context, (2) to maintain biodiversity to promote tourism. The nodes in the Bn associated with these endpoints shown in Figure 4 Selection of management zones : The monitoring of m anagement endpoints is performed on the two main stems of the Crocodile River in two defined zones i.e. Zone 1 and Zone 2 (see Table 4 3.). A different Bn will be built for each of these management zones. Ecostatus monitoring at these management zones is d one under the River Health Program (RHP) initiated by the DWAF (River Health Program, 2005) through sub catchment X21, X22 and X24. This s ection of the river already has a major impoundment at the Kwena Dam. The RHP program has previously established

PAGE 105

105 ecostatus monitoring site at the Tenbosch (EWR 6) which is located at the far west end of the river just before the border with Mozambique. catchment X23 and does not have any major impoundments. RHP has a monitoring site (EWR 5) which is at the before the point of confluence of the Kaap with the main stem of the Crocodile Ri ver at Malelane (Table 4 3 and Figure 4 5). Identify management alternatives The management actions that can be implemented by the water management institutions include implementation of the ecological reserve mandated under the NWA which would entail cu rtailing stakeholder water use through licensing, monitoring of water use and farm dams and enforcement of restrictions by irrigators. Additionally, damming schemes have been proposed to improve water storage (Mallory S. 2010) These alternatives can be viewed as resource consumption control actors. The two considered based in ICMA long range water management studies are reserve implementation and damming the river to restrict water use. The reserve implementation node is con sidered at the Status Quo in which there is no curtailment, Present Ecological Status [PES] implementation requires 25% curtailment and Recommended Ecological Status [RES] requires 50% curtailment). The damming node is based on the deficit or surplus of wa ter supply after damming the river shown in Table 4 4 (Mallory S., 2010). Identify actors and process variables The actors are variables in the socio ecological systems that are affected by management actions. In this preliminary study they are classifie d as those that contribute to the economy while impacting the river (i.e. agriculture, domestic, tourism

PAGE 106

106 and mining) and those that impact the river directly without necessarily making any direct economic contribution (i.e. municipalities and international water users). Process variables, as previously explained, include those that retain resources that are exploited by actors in the system. These are flow and biodiversity. Biodiversity is disaggregated into its various components by observing ecological in dices designed by ecologists (i.e. macro invertebrate, fish, habitat and riparian vegetation indices) to monitor the state of biodiversity in the river ecosystem (Table 4 4). Model Inference Flow Daily records of historical flow at the flow stations at th e EWR sites between January 1960 and December 1980 are used to determine the breakpoint percentages at the IFR low and high flows. This is done by calculating the probability that the observed flow at the gauge is below the IFR low flow at the lower breakp oint percentage and high flow at the higher breakpoint percentage (Oregon State University, 2002 2005) This percentage represents the probability that a given streamflow will fall within a particular range as shown in Figure 4 6. The three ranges established based on these probabilities are used to discretize the flow levels as follows: Low 0 52%, Medium 53 77% and High 77 100%. Ecostatus The process of determining the ecological status or state (or simply ecostatus) of a ri ver was a step by step process known as the Building Block Method (BBM) as elaborated by Kleynhans (2007) This method assesses the in stream habitat integrity and riparian habitat integrity according to a number of ke y modifiers (Table 4 2). Each is ranked as a relative measure of state compared to what the state would have been

PAGE 107

107 under unperturbed conditions to indicate the degree of change from natural habitat integrity. The sum of the modifiers for riparian component, water abstraction, flow modification, bed modification, channel modification, water quality and inundation modifiers for the in stream component are weighted according to whether they were large, serious or critical. These are then used to classify the in stream and riparian zone components according to a descriptive integrity class (Table 4 5). The first step entails determining what the undisturbed reference condition of the river ecosystem which is the original ecological condition in terms of hydrolog ical properties, biological diversity and geomorphologic characteristics. The reference condition serves as a baseline from which the measurement of the ecological status at the time of observation is determined. The second step involves establishing the e costa on biophysical surveys, historical data, land cover information and aerial photographs. Expert knowledge is essential in determining the degrees of change. Ecostatus is a relat ivistic measure of modification on the ecological characteristics of a river from its original form either in terms of improvement or degradation. Once the ecological status has been established an assessment is done to determine the trends in case there i s a change, the causes for the variation as determined. In river ecosystems the changes are predominantly flow related. For instance changes in in stream habitats can occur if flow regimes change. Lower flows can cause greater sedimentation which can alter both channel characteristics and the riparian vegetation. The RHP program has defined several monitoring points along the river channel where ecostatus measurements are done.

PAGE 108

108 Validation of Predicted Management Endpoints Ecological status The RHP determin ation of ecostatus focuses of in stream and riparian biological communities and habitats (e.g. fish, invertebrates, vegetation) to characterize the response of the aquatic environment to multiple disturbances and monitors the state of river systems. The cu rrent states provided are benchmarks which are used as measures of the present ecological states. To develop the Bayesian networks, these benchmark were extrapolated within ranges of values to describe hypothetical states both above and below the given re ference points in relation to other factors in the system such as assurance of water supply (Table 4 6). The states defined within this model were where possible are based on published values from a previous RHP study but also obtained from expert opinions and published literature (Mallory S. 2010; River Health Program 2005) The outcome of this test showed that that the model is consistent with expected with past observations and expected observations under variou s scenarios. Sectoral economy Water has economic value either as an input towards production in agriculture and industry (mining and forestry), in providing income generating ecosystem services in tourism and as a commodity for trade in municipal/domestic water markets (Nieuwoudt, Backeberg, & Du Plessis, 2004) This value addition provides the basis for the development of cost indices for the value of water which is obtained as a function of water supply (Table 4 7). However, in the case of tourism where value addition is implicitly linked to ecosystem services the sectoral contribution is obtained from the

PAGE 109

109 regional Department of Environmental Affairs and Tourism (Department of Environmental Affairs and Tourism, 2007) These simplified assumption s form the basis for the development of supply scenarios and evaluation of management endpoints. Priority is given in supply to the ecological and human reserve, then agricultural users and finally industri al users according to the National Water Act (National Water Act 1998) Sensitivity Analysis Sensitivity analysis is used to measure the sensitivity of changes in probabilities of query nodes when parameters and inputs are cha nged. The query nodes of interest in this study were the management nodes (alternatives and endpoints). The metric used to evaluate sensitivity is the entropy variable which considers how the Bns posterior distributions change under different conditions an d parameter values. The modeling platform Netica has a sensitivity analysis tool that is used to evaluate the variables that contribute the most uncertainty in the net. Sensitivity analysis would be useful if a manager wants to determine what economic se ctor is most affected by changes in the ecological state of the river. Tests on the Bn indicate that uncertainty is reduced with additional information to the hypothesis variable of interest such as ecostatus (S) in the previous example. This is quantified by looking at the changes in the probability distribution of the hypothesis variable of interest (Barton, Saloranta, Moe, Eggestad, & Kuikka, 2008) Mutual information or entropy reduction function is used to measure uncerta inty (Shannon & Warren, 1949; Norsys Software Corp., 2012; Pearl, 1988) It is based on the assumption that the uncertainty of the hypothesis variable S characterized by a probability function P(S) can be rep resented by the entropy function:

PAGE 110

110 H(S) = log { P(S) } (4 3) A higher value of the function indicates more uncertainty about the state of the variable of interest. The reduction in uncertainty about S given a conditional mitigating variable re presented by D such as will therefore be represented as: H(S|D) = log { P(S \ D) } (4 4) The difference between H(S) and H(S|D) is a measure known as entropy reduction given by: I(S,D) = H(S) H(S|D) (4 5) In the previous example if I(S 1 ,D) > I(S 2 ,D) would indicate that the expected value of the ecostatus is reduced more with the effect of reserve implementation (S 1 ) than another conditional variable such as dam construction (S 2 ). This conclusion would r eflect the higher likelihood for implementation of the reserve versus dam construction as an effective management option in that it indicates the variable which has a larger effect on the uncertainty in the variable of interest. The resulting analysis (Ta ble 4 8) at the site of interest which is in the uppermost catchments where ecological units are highly stressed indicate that the habitat, riparian, invertebrate and fish indices have the greatest influence over the results of the ecological state. The fi ndings also show that tourist sector is most sensitive to the ecological status. Outcomes of the sensitivity to findings reflect the expected dynamics within the catchment. Consumptive water users have significantly less dependence on ecological status. T he calculated entropy expressed as percent of the entropy of ecosystem for instance of tourism is 32% as compared to the agricultural sector (0.0023%) and

PAGE 111

111 industry (~1%).This indicates that nodes that dependent on ecosystem services are severely affected b y decline in the ecological status whereas those dependent on consumptive water use are not. Application of the Model In this section we demonstrate how the resultant model can be used to prioritize and test management interventions and set targets as part of an integrated approach to management of water resources in the Crocodile River. Case study data based on projected ecostatus targets under the various management options were tested which includes the various damming scenarios and implementation of the reserve. The management endpoints observed provided projections for economic contributions to the regional economy. This enabled us to make some conclusions about how the Bn can be used to achieve the following goals related to the overall Strategic Adap tive Management goals: (a) prioritizing management actions under various reserve scenarios; (b)monitoring of thresholds of probable concern; (c) forecasting of ecological status of the river at the The Bn (Figure 4 7) provides a tool to evaluate the prob ability of exceeding a particular management conditions designed to improve water and ecosystem management and the evaluation of ecosystem health targets. Several test scenarios are simulated to demonstrate the application of the Bn to make decisions under uncertainty. Impacts of Management on Assurance of Supply Assurance of supply denotes the percentage of time that the water user gets all the water that they require. Water availability is estimated in the Inkomati catchment as a function of the amount of time that they are able to supply water to users (Mallory S. 2010)

PAGE 112

112 The results indicate that Scenario 1 shown in Table 4 9 which is the status quo is the most favorable in terms of achieving the water supply goals for the catchment. Within this reserve scenario addition of two dams improves storage hence water availability (Table 4 10). The model however indicates that this comes at the cost of degradation of the ecological systems resulting in a higher likelihood for a cl ass C river which would negatively impact both aquatic flora and fauna. This subsequent degradation would most likely affect regional tourism. Scenario 3 (Table 4 11) presents a full implementation of the RES which while good for the ecological health si gnificantly limits water availability. The assurance of supply falls below 50% which will result in severe water shortages and a corresponding reduction in productivity. Scenario 2 presents a balanced tradeoff between ecological health and water supply to stakeholders in which the assurance of supply is maintained at an acceptable level (above 50%) and the ecostatus of the river is kept at a Class B river. The two dam solution while maintaining the PES reserve levels provides the best case scenario in terms of providing the highest services to all sectors Cost Benefit of Management Actions Figure 4 7 shows the Bn at EWR 6 with the probability of all states under status quo in management actions in which there is no dam construction to improve storage or re serve implementation to prioritize water supply to human and environmental use. This scenario also assumes normal flow within the river. Under these conditions, the resultant states in ecological state reflect a higher likelihood for moderate degradation o f the river system due to resultant low flows. The economic impact under the various dam and reserve scenarios is evaluated by looking at the percentage sectoral contribution to the economy. The Bn provides a

PAGE 113

113 the changes in proportions based on water suppl y in supply driven sectors such agriculture and ecosystem service driven sectors such as tourism. The scenarios are considered individually as presented in Table 4 9 and Table 4 10. By looking at the total sectoral contribution it is possible to evaluate t he best scenarios which provide a balanced approach to management and favorable outcomes in terms of economic contribution and ecosystem health. This analysis reveals that upstream damming increases water supply to sectoral users but significantly degrades the ecological system. While more comprehensive decision making methodologies might result in a different decision using the simplified Bayesian network determines that the best option is to dam the river downstream while maintaining a Class B river whic h prevents damage of sensitive ecological systems while providing adequate sectoral water supply (Table 4 12). This analysis shows that reserve implementation should be maintained at the PES level which has a 25% restriction as opposed to the RES level whi ch is not feasible economically while maintaining a Class B river. While restricting supply at this level significantly improves the ecological state of the river to pristine levels it will lead to significant shortfalls in water supply to other stakeholde rs and is economically not viable (Table 4 13). These results reflect the strong emphasis on traditional economic analysis through direct sectoral contributions This traditional decision framework is however insufficient as it simply maximizes on econom ic performance. A better decision framework should be constructed that prioritize sustainability of water use, equity in water access and improved human health as envisioned in the new water laws in South Africa.

PAGE 114

114 Thus, a different Bayesian network structu re focusing on alternative quality of life indicators rather than simple economic/monetary returns may give significantly different results given similar inputs. Discussion Bayesian networks have shown to be useful in this case study in providing a rob ust and adaptable tool for modeling management scenarios required within a SAM framework. Bns facilitate a process of identifying vulnerabilities enabling a description of more resilient interventions to the management of the catchment. It also aims at inv estigating inter connections between stakeholders and the associated feedbacks of information/knowledge for an effective SAM framework using the Crocodile River catchment as a prototype but with wider implementation in water management areas in South Afric a. Strategic adaptive management includes learning outcomes which are used to improve the subsequent phases of management. On one level the Bn model helps in providing valuable insights into the character of this socio ecological system as a complex adapti ve system but can incorporate new information as this understanding increases or as it becomes available. The important finding in this research is that economic analysis of the various proposed scenarios indicates that the intrinsic value of the environment that dictates ecosystem value is lower compared to economic interests such as agricultural and industrial water users. While the system has demonstrable resilience degradation of ecological integrity in the river does not have a proport ional decline in economic productivity of the socio ecological system. This means that stakeholder valuation systems for ecological systems need to be reassessed by emphasizing the cultural value of ecosystems rather than simply their economic value. A dis tinct advantage in

PAGE 115

115 using Bns to integrate such disparate data is their versatility in constructing several alternate valuations and testing them for their concomitant sensitivities. Thus, in analyzing a multiple set of structures and beliefs, a richer pic ture and analysis can be formed. This prototype Bn is the first of many potential constructs and can be used to inform future management decisions while taking into account uncertainties such as the effectiveness of management actions, climate changes an d policy changes. Further model validation along with alternative network structuring would be useful to further determine its effectiveness in predicting management interventions.

PAGE 116

116 Table 4 1. Summary of the hydrology of the Crocodile Rive r Management zone Sub catchment Natural MAR (Mm 3 /a) Zone 1 X21 (Elands River) 467 Zone 1 X22 (Middle Crocodile) 362 Zone 1 X24 (Lower Crocodile) 204 Zone 2 X23 (Kaap River) 107 Zone 2 Total 1140 Table 4 2. Summary of nodes in the Crocodile River Bn (Cain, 2001) Categories Description Example Management endpoints These are internal characteristics of the socio ecological system that indicate its health, stability or productivity. Ecostatus Regional economy Management Alternatives Actions taken t o control the rate and timing of water consumption Implement reserve Dam river Restrict water use Actors Water resource consumers and users Agriculture Tourism Domestic Industry International water users Process variables Abiotic and biotic variabl es that control the environmental system at the nominated scale which management actions seek to change improve or prevent from deteriorating. Biodiversity Flow Land use Drivers Factors which cannot be changed by interventions but control the environment al system Precipitation Government policies & laws Table 4 3. Selected management zones on the Crocodile River sections Management zone RHP Site Location Flow station Sub catchments River Zone 1 IFR 6 Tenbosch X2H016 X21, X22, X24 Crocodile Z one 2 IFR 5 Malelane X2H022 X23 Kaap

PAGE 117

117 Table 4 4. Establish discretization states for variables Category Variable Node type Discretization method States Management endpoints Ecostatus Decision Based on scoring system from River Health Program (RH P) study (River Health Program, 2005) A: Natural (90 100%) B: Good (80 89%) C: Fair (60 79%) D: Poor (40 59%) E: Unacceptable (0 39%) Sectoral Economic Contribution Utility Based on study on USAID economic value o f water in South Africa that provides sectoral indices in R/m 3 The % economic contribution of tourism to the region is based on indices from a study on the economic value of water in tourism (Conningarth Consultants, 2001) Agricultural sector Industrial sector Tourism sector Management Alternatives Reserve Implementation Decision Based on ICMA water management scenarios study (Present Ecological Status [PES] implementation requires 25% curtailment and Recommended Ecologica l Status [RES] requires 50% curtailment). These curtailments necessary to attain PES and RES create a further deficit to sectoral supply (Mallory S. 2010) Status quo Implement PES (25% deficit) Implement RES (50% deficit) Dam Decision Based on ICMA long range water management scenarios study. Each scenario presents an increase or decrease in the supply. Scenarios are based on deficit or surplus of water supply after damming of river (Mallory S 2010) Scenario 1 : Status quo (12% deficit) Scenario 2 :Dam Upstream (7 % surplus ) Scenario 3 :Dam Downstream (14% deficit) Scenario 4 Dam Upstream & Downstream(4 % surplus) Actors Agriculture Nature Based on ICMA water management scenario study Demand is measured as an assurance of supply which is the percentage of time users get the water demand. Meets demand (> 400 M m 3 /a) Fails to meet demand (<0 400 M m 3 /a) Mining Nature Based on ICMA water management scenario study. Demand is measure d as an assurance of supply which is the percentage of time users get the water demand (Mallory S. 2010) Meets demand (> 23.5 M m 3 /a) Fails to meet demand (0 23.5 M m 3 /a) International Obligations Nature Flow level at th e border with Mozambique should be maintained at 28 M m 3 /a to meet the Inco Maputo Tripartite Agreement (Department of Water Affairs and Forestry 2010) (Department of Water Affairs and Forestry 2 010) Fulfills agreement : >28 M m 3 /a Fails to fulfill agreement: 0 27 M m 3 /a Domestic Nature Based on ICMA water management scenario study. Demand is measured as an assurance of supply which is the percentage of time users get the water demand (Mallory S. 2010) Meets demand (> 48 M m 3 /a) Fails to meet demand (0 48 M m 3 /a) Tourism Based on study on the value of wat er (Nieuwoudt, Backeberg, & Du Plessis, 2004) Thriving : 77 100% Status quo: 53 77% Failing : 0 52%

PAGE 118

118 Table 4 4. Continued Category Variable Node Type Discretization method States Process variables Seasonal Flow Decision Calculated as a the % of observed flow exceeding IFR high and low flows at EWR monitoring sites High : 77 10 0% Medium: 53 77% Low: 0 52% Macro invertebrate Index (SASS) Nature Based on ecostatus scoring system from RHP study. Natural: 90 100% Good : 80 89% Fair: 60 79% Poor: 40 59% Unacceptable: 0 39% Fish Assemblage Index (FAI) Nature Based on ecostatus scoring system from RHP study. Natural: 90 100% Good : 80 89% Fair: 60 79% Poor: 40 59% Unacceptable: 0 39% Riparian Vegetation Index (RVI Score) Nature Based on ecostatus scoring system from RHP study. Natural: 90 100% Good : 80 89% Fair: 60 79% Poor: 4 0 59% Unacceptable: 0 39% Habitat Integrity Index (IHI) Nature Based on ecostatus scoring system from RHP study. Natural: 90 100% Good : 80 89% Fair: 60 79% Poor: 40 59% Unacceptable: 0 39%

PAGE 119

119 Table 4 5. Ecological scores for ecostatus components with des criptions in relation to degree of modification from natural conditions Ecostatus Description Score (% of the total) A Unmodified, close to natural conditions 90 100 B Largely natural with few modifications. A small change in natural habitats and biota may have taken place but the ecosystem functions are essentially unchanged 80 89 C Moderately modified. A loss and change of natural habitat and biota have occurred but the basic ecosystem functions are still predominantly unchanged. 60 79 D Largely mod ified. A large loss of natural habitat, biota and basic ecosystem function have occurred 40 59 E Seriously modified. The loss of natural habitat, biota and basic ecosystem function are extensive 20 39 F Critically/Extremely modified. Modifications have r eached a critical level and the system has been modified completely with an almost complete loss of natural habitat and biota. Ba 0 sic ecosystem function destroyed and the changes are irreversible 0 19

PAGE 120

120 Table 4 6. Current ecostatus determined a t the various monitoring sites along the Crocodile River by RHP (2005) Index Zone 1:EWR 5 (Malelane) Zone 2: EWR 6 (Tenbosch) FAI 52% 72% SASS 74% 82% RVI 40% 49% IHI 75% 39% ES Score 60% 61% PES C C Table 4 7. Sectoral contribution Water Use r Cost Index ( ZAR /m 3 ) Supply (Mm 3 /annum) Sectoral Contribution/annum ( ZAR ) Sectoral Contribution (%) Agriculture 1.5 344.00 ZAR 516,000,000.00 9.8% Industry 196.9 23.40 ZAR 4,607,460,000.00 87.7% Tourism 44.4 n/a ZAR 84,625,000.00 1.6% Domestic 1 46.60 ZAR 46,600,000.00 0.9% Table 4 8. Sensitivity analysis for posterior network to ecological status of the river Node Entropy reduction Percent Entropy of ecological state 1.873 100% Habitat Index 1.157 61.8% Ripar ian Index 1.157 61.8% Invertebrate Index 1.157 61.8% Fish Index 1.157 61.8% Flow 0.948 50.6% Tourism 0.615 32.8% Implement Reserve 0.195 10.4% Industry 0.016 0.88% Domestic Water Use 0.006 0.32% Dam River 0.006 0.298% Agricultural Sector 0.006 0.2 95% Sectoral Economic Contribution 0.000 0.0023% International Water Obligation 1.873 0

PAGE 121

121 Table 4 9. Predicted assurance of supply for Reserve Scenario: 1 Status quo Dam Scenario Agriculture Industry Domestic No New Dam 68% 99% 95% Dam Upstream 76% 96% 80% Dam Downstream 62% 98% 68% Dam Up & Downstream 86% 98% 82% Table 4 10. Predicted assurance of supply for Reserve Scenario 2: Implement PES requirement (requires a 25% supply restriction) Dam Scenario Agricu lture Industry Domestic No New Dam 51% 74% 71% Dam Upstream 57% 75% 60% Dam Downstream 47% 74% 51% Table 4 11. Predicted assurance of supply for Reserve Scenario 3: Implement RES requirement (requires a 5 0 % supply restriction) Dam Scenario Agriculture Industry Domestic No New Dam 3 4% 49% 48% Dam Upstream 38% 50% 40% Dam Downstream 31% 49% 34% Dam Up & Downstream 43% 49% 41%

PAGE 122

122 Table 4 12. Total sectoral economic contributions for the damming scenarios TOTAL SECTOR ECONOMIC CONTRIBUTION Scenario/ Ecostatus A B C D No New Dam ZAR 5,198,183,071 ZAR 5,202,299,076 ZAR 5,201,243,639 ZAR 5,199,661,672 Dam Upstream ZAR 5,195,594,858 ZAR 5,201,266,931 ZAR 5,200,211,494 ZAR 5,202,827,750 Dam Downstream ZAR 5,196,098,329 ZAR 5,200,722,092 ZAR 5 ,199,666,655 ZAR 5,198,611,218 Dam Up & Dow nstream ZAR 5,196,153,951 ZAR 5,199,731,547 ZAR 5,199,731,547 ZAR 5,198,147,204 Table 4 13. Total sectoral economic contribution for reserve implementation scenarios TOTAL SECTOR ECONOMIC CONTRIBUTION Scenario/Ecostatus A B C D No reserve implementation X ZAR 5,204,351,023 ZAR 5,203,297,962 ZAR 5,206,972,030 Implement PES ZAR 5,196,107,367 ZAR 5,196,107,367 ZAR 5,195,580,836 ZAR 5,194,001,245 Implement RES X X X X

PAGE 123

123 Figure 4 1. Generalized framework of SAM as proposed by Pollard & Du Toit (2005) applied to management of aquatic ecosystems, their surroundings which includes management of environmental flows

PAGE 124

124 Figure 4 2. Generalized approach to developing a Bn for application in the development of management option of a SAM framework (adapted from Pollard & Du Toit, 2005) Figure 4 3. Map of the Crocodile River within the Inkomati Water Management area

PAGE 125

125 Figure 4 4. System mo del obtained from collaborative sessions showing interactions between socio ecological components in the Crocodile River Catchment

PAGE 126

126 Figure 4 5. Location of ecostatus monitoring sites Figure 4 6. Streamflow at EWR 6 with IFR maintenance hi gh and low flows

PAGE 127

127 Figure 4 7. Bayesian network for the Crocodile River at EWR 6. The probabilities expressed in this diagram reflect the discretized states outlined previously

PAGE 128

128 CHAPTER 5 CONCLUSIONS This research project involved the use of complexity and resilience theory as a foundation for integrating several tools to explore ecological impacts from water management decisions in the Crocodile River. The t ools varied from (1) system envisioning frameworks that involved multiple stakeholder groups and subject matter experts to (2) computational modeling using the ACRU agro hydrological model at two catchment scales analyzed with the FITEVAL tool for evaluati on to (3) Bayesian networks to model water use scenarios to provide practical insights into the ecological status of river reaches. The Crocodile River System presents an example of a highly complex social ecological system in which a variety of stakehold ers compete for common and shared water resources. As a multifaceted catchment, the Crocodile River is characterized by a diversity of connections between dynamic socio ecological components. The complexity and resilience framework used in this research h elped to better understand and describe these levels of interaction between the ranges of actors that exist within the catchment such as the interaction between the human and eco logical system The integrated modeling framework used in this research hypot hesizes that the ability of a river system to provide ecosystem goods and services is a useful measure of the resilience. As such, identifying key ecosystem goods and services provided by the catchment will facilitate a process of understanding potential s ystem vulnerabilities which will enable the design of more resilient interventions to the management of the catchment. It also aims at investigating inter connections between stakeholders and the associated feedbacks of information/knowledge for an effecti ve adaptive management

PAGE 129

129 framework using the Crocodile River Catchment as a prototype but with wider implementation in water management areas in South Africa. Hydrological information obtained from modeling the catchment using ACRU provided valuable insights both on the behavior of the model at the different management scales and the quality of observed data. The improved evaluation methods presented allowed for the analysis of data collection errors at monitoring sites and help to determine the effect of da ta quality and uncertainty on adaptive water planning management decisions. While it was initially presumed that that at a finer scale, model outputs would be more accurate this hypothesis did not anticipate that uncertainties such as data collection erro rs, friction losses and increased effects of microclimate would magnify the effect of errors at a smaller scale This begs the question as to whether water managers should be concerned with appropriate temporal and spatial scale s of measurement depending o n the decision environment versus simply assuming that more detail provides better answers. Bayesian networks have been used to prioritize and test management interventions and set targets as part of an integrated approach to management of water resources in the Crocodile River. Case study data based on projected ecostatus targets under the various management options were tested which includes the various damming scenarios and implementation of the reserve. The challenge that water managers in the Crocodil e River experience is understanding the cumulative effects of their management actions on the ecological health and economic productivity of activities that depend on the Crocodile River. Bns brought the ability to bridge these two critical endpoints. The management endpoints observed provided projections for

PAGE 130

130 economic contributions to the regional economy. This enabled us to make some conclusions about how the Bn can be used to achieve the following goals related to the overall Strategic Adaptive Management goals: (a) prioritizing management actions under various reserve scenarios; (b ) monitoring of thresholds of probable concern; (c) forecasting of ecological status of the river at the monitoring points The Bn provides a tool to evaluate the probability of exceeding a particular management condition designed to improve water and ecosystem management and the evaluation of ecosystem health targets. Several test scenarios were simulated to demonstrate the application of the Bn to make decisions under uncertain ty. Management scenarios were adopted from previous studies that took into consideration various levels of enforcement of the provisions of the National Water Act (National Water Act, 1998). Ecostatus measurement at the two monitoring sites EWR 5 and EWR 6 along the Crocodile were based on a previous study by the River Health Program (River Health Program 2005) to monitor the health of rivers based on various ecological indicators such as FAI, RVI, SASS and IHI that cumulative ly determine the ecostatus both in the present (PES) and for recommended levels (RES). Economic indicators were based on the cost indices for the economic value of water. The model outputs enabled a broader understanding of the interaction of ecosystem he alth and economic activities within the catchment. This analysis shows that at the site of interest which is in the uppermost catchments where ecological units are highly stressed indicate that the habitat, riparian, invertebrate and fish indices have the greatest influence over the results of the ecological state. The findings also show that tourist sector is most sensitive to the ecological status. Industrial use of water

PAGE 131

131 shows low impact on both endpoints owing to the fact they have a larger impact on w ater quality than quantity. We also conclude that nodes that are dependent on ecosystem services are severely affected by decline in the ecological status whereas those dependent on consumptive water use are not. In summary this study shows that process b y which we learn about the world has two scale dependent components, the actual scale at which patterns and processes occur and the scales at which we obtain data about them that is used for decision making. In the interest of catchment level hydrological modeling, these two considerations play an important role in informing the decision making at various scales. Different scales and levels of complexity are required when planning to account for various small scale processes that might affect management de cisions at different resolutions. This is particularly important in the current South African climate with the implementation of the National Water Act of 1998 (National Water Act 1998) In integrating hydrological with ecolog ical models, it is important to harmonize model inputs at both a spatial and temporal scale. The challenge however remains to translate the statistical measures and methods in a way that effectively informs the decision environment in Crocodile River catch ment.

PAGE 132

13 2 APPENDIX A ACRU DATA FILES Object A 1. AC RU quaternary catchment menu files Object A 2. ACRU quinary catchment menu files Object A 3. ACRU quaternary climat e input files Object A 4. ACRU quinary climate input files Object A 5. ACRU quaternary catchment output files Object A 6. ACRU quinary catchment output files Object A 7. FITEVAL quaternary output files Object A 8. FITEVAL quinary output files

PAGE 133

133 APPENDIX B BAYESIAN NETWORK MODE L AND CASE FILES Object B 1. Bayesian network model Object B 2. Bayesian network case file at EWR 5 Object B 3. Bayesian network case file at EWR 6

PAGE 134

134 REFERENCE LIST Aalders, I. H., & Aitkenhead, M. J. (2009). Predicting land cover using GIS, Bayesian and evolutionary algorithm methods. Journal of Environmental Management, 90 (1), 236 250. doi:10.1016/j.jenvman.2007.09.010 Akili. (2010). Akili Complexity and Integration Initiative. Ames, D., Neilson, B., Stevens, D., & Lall, U. (2005). Using Bayesian networks to model watershed management decisions: an East Canyon Creek case study. Journal of Hydroinformatics 267 283. Anderson, J. A. (2005). Engaging disadvantaged communities: Lessons from the Inkomati. frameworks for rural water'. Johannesburg, South Africa. Ashton, P. J. (1995). Water quality management in the Crocodile River Eastern Transvaal, South Africa. Water Science and Technology, 32 (5 6), 201 208. doi:10.1016/0273 1223(95)00664 8 Bammer, G. (2005). Integration and implementation sciences: Building a new specialization. Ecology and Society, 10 (2). Retrieved from http://www.ecologyandsociety.org/vol10/iss2/art6/ Barton, D. N., Saloranta, T., Moe, S. J., Eggestad, H. O., & Kuikka, S. (2008). Bayesian belief networks as a meta modelling tool in integrated river basin management Pros and cons in evaluating nutrient abatement decisions under undertainty in a Norwegian river basin. Ecological Economics, 66 91 104. doi:10.1016/j.ecolecon.2008 Bashari, H., Smith, C., & Bosc h, O. J. (2008, December). Developing decision support tools for rangeland management by combining state and transition models and Bayesian belief networks. Agricultural Systems, 99 (1), 23 24. doi:10.1016/j.agsy.2008.09.003 Batchelor, C., & Cain, J. (1999) Applications of Bayesian belief networks to water management studies. Agricultural Water Management, 40 (1), 51 57. doi:10.1016/S0378 3774(98)00103 6 Bergstrom, S., & Graham, L. (1998, November). On the scale problem in hydrological modelling. Journal of Hydrology, 211 (1 4), 253 265. doi:10.1016/S0022 1694(98)00248 0

PAGE 135

135 Biggs, H., & Rogers, K. (2003). An adaptive system to link science, monitoring and management in practice. In H. Biggs, K. Rogers, J. du Toit, H. Biggs, & K. Rogers (Eds.), The Kruger experien ce: ecology and management of savanna heterogeneity (pp. 59 80). Washington: Island Press. Bloschl, G., & Sivaplan, M. (1995). Scale issues in hydrological modeling: a review. Hydrological Processes, 9 251 290. doi:10.1002/hyp.3360090305 Borsuk, M., Stow, C., & Reckhow, K. (2004, April 1). A Bayesian network of eutrophication models for synthesis, prediction, and uncertainty analysis. Ecological Modeling, 13 (2 3), 219 239. doi:10.1016/j.ecolmodel.2003.08.020. Braun, P., Molnar, T., & Kleeberg, H. B. (1997) The problem of scaling in grid related hydrological process modelling. Hydrological Processes, 11 (9), 1219 1230. doi:10.1002/(SICI)1099 1085(199707)11:9<1219::AID HYP553>3.0.CO;2 S Breen, C. .., Dent, M. .., Jaganyi, J. B., Madikizela, B., Maganbeharie, J., & Ndlovu, A. (2000). 130/00). Pretoria: Water Research Commission. Bromley, J., Jackson, N., Clymer, O., Giacomello, A., & Jensen, F. (2005). The use of Hugin to develop Bayesian networks asan aid to integrated water resource planning (234 242 ed., Vol. 20). Environmental Modeling Software. Brown, J., & Woodhouse, P. (2004). Pioneering redistributive regulatory reform. A study of implementation. University of Manchester, Institute for Devel opment Policy and Management (IDPM). Retrieved from http://purl.umn.edu/30601 Brugnach, M., Dewulf, A., Pahl Wostl, C., & Tai, T. (2008). Toward a relational concept of uncertainty: about knowing too little, knowing too differently, and accepting not to kn ow. Ecology and Society, 13 (2). Retrieved from http://www.ecologyandsociety.org/vol13/iss2/art30/ Cain, J. (2001). Planning improvements in natural resource management. Guidelines for using Bayesian networks to support the planning and management of develo pment programmes in the water sector and beyond. Wallingford, Centre for Ecology and Hydrology,. Campbell, B. M., & Luckert, M. K. (2002). Towards understanding the role of forests in rural livelihoods. Retrieved from http://www.cifor.org/nc/online library /browse/view publication/publication/998.html Carmo Vas, A., & van der Zaag, P. (2003). Sharing the Incomati waters: Cooperation and competition in the balance. UNESCO, IHP. World Water Assessment.

PAGE 136

136 Cheng, J., Bell, D. A., & Liu, W. (1997). An algorithm for Bayesian belief network construction from data. In Proc. Workshop on Artificial Intelligence and Statistics (pp. 83 90). New York: Florida. ACM Press. Clear Pure Water. (2008). The development of a real time decision support system (DSS) for the Crocodile east river system Inception report. Department of Water Affairs. Colvin, J., Everard, M., Goss, J., Klarenberg, G., & Ncala, D. (2008). Building capacity for adaptive and integrated water resources governance in the Inkomati. Inkomati Catchment Manageme nt Agency. Water S.A., 34 (6). Conningarth Consultants. (2001). Regional comparative advantage of water use: The Orange River case study. Commissioned by USAID Regional Centre for Southern Africa. Crawford, J. (2004). An analysis of the social, economic and environmental direct and indirect costs and benefits of water use in irrigated agriculture and forestry. A case study of the Crocodile River Catchment Mpumalanga Province: Report to the Water Research Commission. Daedlow, K., Beckmann, V., & Arlinghaus, K. (2011). Assessing an Adaptive Cycle in a Social System under External Pressure to Change: the Importance of Intergroup Relations in Recreational. Ecology and Society, 16 (2). Department of Environmental Affairs and Tourism. (2007). Maps and mapping. Mpum alanga. Department of Water Affairs and Forestry (2010). Tripartite Interim Agreement between the Republic of Mozambique and the Republic of South Africa and the Kingdom of Swaziland for co operation on the protection and sustainable utilisation of the w ater resources of the Incomati and Maputo watercourses. Johannesburg. Retrieved from http://www.dwaf.gov.za/Docs/Other/IncoMaputo/INCOMAPUTO_AGREEMENT2 9082002.pdf Department of Water Affairs and Forestry (1999). Department of Water Affairs: Fresh water Retrieved from http://www.dwa.gov.za Department of Water Affairs and Forestry (2009). Inkomati water availability asessment study. Hydrology report for the Crocodile (East). Report number PWMA 05/X22/00/1508.

PAGE 137

137 Department of Water Affairs and Forestry ( 2009). Operationalise the reserve: Rapid Habitat Assessment Model manual. DWA. Pretoria, South Africa: ed. D. Louw & C .J.Kleyhans. doi:Report No. RDM/Nat/00/C ON/0709 Du Toit, D. R., Biggs, H., & Pollard, S. (2011). The potential role of mental model meth odologies in multistakeholder negotiations: integrated water resources management in South Africa. Ecology and Society, 16 (3). Retrieved from http://dx.doi.org/10.5751/ES 04237 160321 Duncan, D. H., & Wintle, B. A. (2008). Towards adaptive managementof nat ive vegetation in regional landscapes. Springer Verlag. Etienne, M., Du Toit, D., & Pollard, S. (2011). ARDI: a co construction method for participatory modeling in natural resources management. Ecology and Society, 16 (1). Retrieved from URL: http://www.ec ologyandsociety.org/vol16/iss1/art44/ Foxcroft, L. C. (2009). Developing thresholds of potential concern for invasive alien species:Hypotheses and concepts. Koedoe, 51 1 6. doi:10.4102/koedoe.v51i1.157 Govender, D. (2010). What is killing South African cr ocs? (N. Lubick, Interviewer) Scientific American Retrieved from http://www.scientificamerican.com/article.cfm?id=what is killing crocs Govender, N. (2010). Fire management in the KNP. (N. Wangusi, Interviewer) Gunderson, L., & Holling, C. (2002). Panarc hy: Understanding transformations in human and natural System. Washington D.C.: Island Press. Gunderson, L., & Holling, C. (2002). Panarchy: Understanding Transformations in Human and Natural System. Washington D.C.: Island Press. Gunderson, L., & Pritchar d, L. (2002). Resilience and behavior of large scale systems. Washington D.C.: Island Press. Harmel, R. D., Cooper, R. J., Slade, R. M., Haney, R. L., & Arnold, J. G. (2006). Cumulative uncertainty in measured streamflow and water quality data for small wa tersheds. Transactions of the ASABE, 49 (3), 689 701. Hendricks, G., Shukla, S., Martinez, C., & Kiker, G. (n.d.). A modified model for simulating hydrologic processes for plastic mulch production systems. Journal of Irrigation Drainage Engineering doi:10. 1061/(ASCE)IR.1943 4774.0000615

PAGE 138

138 Holling, C., & Sanderson, S. (1996). Dynamics of (dis)harmony in ecological and social systems. In S. Hanna, C. Folke, & K. Mler (Eds.), Rights to nature: Ecological, economic, cultural, and political principles of institu tions for the environment (pp. 57 85). Washington DC: Island Press. Hughes, D., & Hannart, P. (2003). A desktop model used to provide an initial estimate of the ecological instream flow requirements of rivers in South Africa. Journal of Hydrology, 270 (3 4) 167 181. doi:10.1016/S0022 1694(02)00290 1 Hughes, D., Louw, D., & Mallory, S. (2008). Methods and software for the real time implementation of the ecological reserve : explanations and user manual (No. 1582/1/08). Water Research Commission, Pretoria. Re trieved from http://www.wrc.org.za Inkhlakanipho Consultants. (2009). The development of the Inkomati catchment management strategy: Book of maps for status quo report. Nelspruit. Jewitt, G. P., Garratt, J. A., Calder, I. R., & Fuller, L. (2004). Water res ources planning and modelling tools for the assessment of land use change in the Luvuvhu Catchment, South Africa. Water, Science, Technology and Policy Convergence and Action by All (A Meeting Point for Action leading to Sustainable Development), 29 (15 18) 1233 1241. Jewitt, G., & Schulze, R. (1999, October). Verification of the ACRU model for forest hydrology. Water SA, 25 (4), 483 490. Retrieved from http://www.wrc.org.za Kienzle, S. W. (2011). Effects of area under estimations of sloped mountain terrain on simulated hydrological behaviour: a case study using the ACRU model. Hydrological Process, 25 (8), 1212 1227. doi: 10.1002/hyp.7886 Kiker, G. (1998). Development and comparison of savanna ecosystem models to explore the concept of carrying capacity. Corn ell University Kiker, G. A. (2001). Testing and Validation of a Java based, Object Oriented Modeling System in the Mgeni River Watershed, KwaZulu Natal, South Africa. ASAE Paper No. 012031. St. Joseph, MI Kiker, G. A., & Clark, D. J. (2006). Testing a nd validation of a java based, object oriented modeling system in the Mgeni river watershed, KwaZulu Natal, South Africa. Transactions of the ASABE, 49 (5), 1419 1433. Kleynhans, C. (2007). Module D: Fish response assessment index in river eco classificatio n: Manual for ecostatus determination (version 2). Joint Water Research Commission and Department of Water Affairs and Forestry report.

PAGE 139

139 Kleynhans, C., Thirion, C., Louw, M. D., & Rowntree, K. (2008). River eco classification: Manual for ecostatus determina tion. Joint Water Research Commission and Department of Water Affairs and Forestry report. doi:WRC Report No. KV 168/05, WRC Pretoria. Krause, P., Boyle, D. P., & Bse, F. (2005). Comparison of different efficiency criteria for hydrological model assessm ent. Advances in Geosciences, 5 89 97. doi:doi:10.5194/adgeo 5 89 2005 Legates, D. R., & McCabe Jr., G. J. (1999, January). Evaluating the use of "goodness of fit" measures in hydrologic and hydroclimatic model validation. Water Resources Research, 35 (1), 233 241. doi:DOI: 10.1029/1998WR900018 Mallory, S. (2010). Inkomati catchment strategy: Water management scenarios. unpublished report. Mallory, S., Odendaal, P., & Desai, A. (2008). The Water Resources Modelling Platform Retrieved from www.waterforafrica.com Marcot, B., Holthausen, R., Raphael, M., & Rowland, M. (2001). Using Bayesian belief networks to evaluate fish and wildlife population viability under land managment alternat ives from an environmental impact statement. 153 (1 3), 29 42. doi:10.1016/S0378 1127(01)00452 2 McCuen, R. H., Knight, Z., & Cutter, A. G. (2006, November). Evaluation of the Nash Sutcliffe Efficiency Index. Journal of Hydrological Engineering 11 (6), 597 602. doi:10.1061/(ASCE)1084 0699(2006)11:6(597) McLoughlin, C. A., Deacon, A., Ababio, T. G., & Sithole, H. (2011). History, rationale, and lessons learned: thresholds of potential concern in Kruger National Park river adaptive management. Koedoe, 53 (2). doi:10.4102/koedoe.v53i2.996 Millenium Ecosystem Assessment. (2003). Ecosystems and human well being: A framework for assessment. World Resource Institute. Island Press. National Water Act No.36 (August 20, 1998). Ni, Z., Philips, L. D., & Hanna, G. B. ( 2010). The use of Bayesian networks in decision making. Key Topics in Surgical Research and Methodology 351 359. Nicols, J. D., & Williams, B. K. (2006). Monitoring for conservation. Trends in Ecology, 21 (12), 669 673. doi:10.1016/j.tree.2006.08.007

PAGE 140

140 Nieuw oudt, W., Backeberg, G., & Du Plessis, H. (2004, June). The value of water in the South African economy: Some implications. Agrekon, Agricultural Economics Association of South Africa (AEASA), 43 (2). Norberg, J., Wilson, J., Walker, B., & Ostrom, E. (2008) Diversity and Resilience of Social Ecological Systems. In J. Norberg, & G. Cummings, Complexity Theory for a Sustainable Future (pp. 46 79). New York: Columbia University Press. Norsys Software Corp. (2012). Retrieved May 4, 2012, from Norsys Software Co rp. Website: http://www.norsys.com/netica.html Noss, R. F. (1990). Indicators for monitoring biodiversity: a hierarchical approach. Conservation Biology 4 (4), 355 364. Retrieved from http://www.jstor.org/stable/2385928 Nyberg, J. B., Marcot, B. G., & Sul yma, R. (2006). Using Bayesian belief networks in adaptive management. Canadian Journal of Forest Research, 36 3104 3116. doi:Using Bayesian belief networks in adaptive management Oregon State University. (2002 2005). Streamflow evaluation for watershed r estoration planning and design Retrieved August 28, 2012, from http://streamflow.engr.oregonstate.edu/examples/yachats/index.htm Pahl Wostl, C., Sendzimir, J., Jeffrey, P., Aerts, J., Berkamp, G., & Cross, K. (2007). Managing change toward adaptive water management through social learning. Ecology and Society, 12 (2). Retrieved from http://www.ecologyandsociety.org/vol12/iss2/art30/ Parma, A. M. (1998). What can adaptive management do for our fish, forests, food and biodiversity? Integrative Biology, 1 (1), 16 26. doi:10.1002/(SICI)1520 6602(1998)1:1<16::AID INBI3>3.0.CO;2 D Pearl, J. (1988). Probabilistic reasoning in intelligent systems: Networks of plausible inference. San Mateo, CA: Morgan Kaufmann. Pelletier, P. M. (1988). Uncertainties in the single det ermination of river discharge: A literure review. Canadian Journal of Civil Engineering, 15 (5), 834 850. Pollard, S., & Du Toit, D. (2005). Recognizing heterogeneity and variability as key characteristics of savannah systems: the use of Strategic Adaptive Management as an approach to river management within the Kruger National Park, South Africa. IUCN, Gland: Report of UNEP/GEF Project No. GF/2713 03 4679, Ecosystems, Protected Areas and People Project.

PAGE 141

141 Pollard, S., Du Toit, D., & Biggs, H. (2011). River ma nagement under transformation: The emergence of strategic adaptive management of river systems in the Kruger National Park. Koedoe African Protected Area Conservation and Science, 53 (2), 14. doi:10.4102/koedoe. v53i2.1011 ReplaceXXX. (2001, Jul 3). A hyd rological perspective of the February 2000 floods:. Water SA, 27 (3), 321 328. Ringold, P. L., Alegria, J., Czaplewski, R. L., Mulder, B. S., Tolle, T., & Burnett, K. (1996). Adaptive monitoring design for ecosystem management. Ecological Applications, 6 7 45 747. Retrieved from http://dx.doi.org/10.2307/2269479 Ritchie, J. T. (1972). Model for predicting evaporation from a row crop with incomplete cover. Water Resour. Res., 8 (5), 1204 1213. doi:10.1029/WR008i005p01204 Ritter, A. (2013, March 12). Available Software Retrieved from Software: http://webpages.ull.es/users/aritter/software.html Ritter, A., & Muoz Carpena, R. (2013). Predictive ability of hydrological models: objective assessment of goodness of fit with statistical significance. Journal of Hydro logy 480 (1), 33 45. doi:10.1016/j.jhydrol.2012.12.004 River Health Program (2005). State of rivers report: Monitoring and managing the ecological state of rivers in the Crocodile (West) Marico Water Management Area. South Africa. United Nations Environ mental Programme. River Health Program. (2005). State of rivers report: Monitoring and managing the ecological state of rivers in the Crocodile (West) Marico Water Management Area. South Africa. United Nations Environmental Programme. Rogers, K., & Biggs, H. (1999). Integrating indicators, endpoints and value systems in strategic management of the Kruger National Park. Fresh Water Biology, 41 439 451. doi:doi:10.1046/j.1365 2427.1999.00441.x Rogers, K., & Biggs, H. (2003). An adaptive system to link scienc e, monitoring and management in practice. in J.T du Toit, K.H. Rogers and H.C.Biggs (eds.),The Kruger Experience Ecology and Management of savanna heterogeneity. Washington: Island Press. Roux, D. J., Kleynhans, C. J., Thirion, C., Hill, L., Engelbrecht, J S., Deacon, A. R., & Kemper, N. (1999). Adaptive assessment and management of riverine ecosystems: the Crocodile/Elands River case study. Water SA, 25 (4), 501 511.

PAGE 142

142 Rumpff, L., Duncan, D. H., Vesk, P. A., Keith, D. A., & Wintle, B. A. (2011). State and tr ansition modelling for Adaptive Mnagement of native woodlands. Biological Conservation, 144 (4), 1224 1236. doi:10.1016/j.biocon.2010.10.026 Schreiber, E. S., Bearlin, A. R., Nicol, S. J., & Todd, C. R. (2004, December). Adaptive management: a synthesis of current understanding and effective application. Ecological Management and Restoration, 5 (3), 177 182. doi:10.1111/j.1442 8903.2004.00206.x Schulze, R. E. (1989). "ACRU: Background, concepts and theory." ACRU Report 35 University of Natal, Dept. of Agri cultural Engineering, Pietermaritzburg Schulze, R. E. (1989). "ACRU: Background, concepts and theory." ACRU Report 35 University of Natal, Dept. of Agricultural Engineering, Pietermaritzburg Schulze, R. E. (1995). Hydrology and agrohydrology: A text to accompany the Acru 3.00 agrohydrological modelling system. WRC Report No. TT69/95. ACRU Report No. 43 Shachter, R. (1986, November). Evaluating influence diagrams. Operations Research, 34 (6), 871 882. Retrieved from http://www.stanford.edu/dept/MSandE /cgi bin/people/faculty/shachter/pdfs/evalid.pdf Shachter, R. (1988). Probabilistic inference and influence diagrams. Operations Research, 36 (4), 589 604. Retrieved from http://www.stanford.edu/dept/MSandE/cgi bin/people/faculty/shachter/pdfs/probinf.pdf S hannon, C. E., & Warren, W. (1949, July). A mathematical model of communication. The Bell System Technical Journal, 27 379 423. Slade, R. M. (2004). General method, information, and sources for collecting and analyzing water resources data. CDROM Copright 2004 Raymond M. Slade Jr. Smithers, J., Schulze, R., Pike, A., & Jewitt, G. (2001, Jul 3). A hydrological perspective of the February 2000 floods:A case study in the Sabie River Catchment. 27 (3), pp. 321 328. Retrieved from http://www.wrc.org.za Stassop oulou, A., Petrou, M., & Kittler, J. (1998). Application of Bayesian network in a GIS based decision making system. International Journal of Geographical Information Science, 12 (1), 23 46. doi:10.1080/136588198241996

PAGE 143

143 Stone Jovicich, S. S., Lynam, T., Lei tch, A., & Jones, N. A. (2011). Using consensus analysis to assess mental models about water use and management in the Crocodile River Catchment, South Africa. 16 (1). Retrieved from http://www.ecologyandsociety.org/vol16/iss1/art45/ Tewari, D. D. (2002). A n Evolutionary History of Water Rights in South Africa. United States Department of Agriculture. (1972). National engineering handbook: section 4 hydrology. United States Department of Agriculture Soil Conservation Se rvice, Washington DC. Retrieved from ftp://ftp.wcc.nrcs.usda.gov/wntsc/H&H/NEHhydrology/ch21.pdf van Aarde, R., Whyte, I., & Pimm, S. (1999). Culling and the dynamics of the Kruger National Park African elephant population. Animal Conservation 287 294. do i:10.1111/j.1469 1795.1999.tb00075.x Varis, O. (1997). Bayesian decision analysis for environmental and resource management. 12 (2 3), 177 185. doi:10.1016/S1364 8152(97)00008 X Venter, F. J., & Deacon, A. R. (n.d.). Managing rivers for conservation and ec otourism in the Kruger National Park. Water Science and Technology, 32 (5 6), 227 233. doi:10.1016/0273 1223(95)00667 2 for applications in land use and climate change studie s. Hydrology and Earth Systems Sciences Discussions, 7 C3031 C3036. Retrieved from http://www.hydrol earth syst sci discuss.net/7/C3031/2010/hessd 7 C3031 2010.pdf Walker, B., & Salt, D. (2006). Making Sense of Resilience. In Resilience Thinking: Sustaini ng Ecosystems and People in a Changing World (p. 174). Washington, D.C., USA: Island Press. Wallach, D., Makowski, D., & Jones, J. W. (2006). Working with Dynamic Crop Models (1st ed.). Amsterdam: Elsevier. Walters, C. J. (1986). Adaptive management of ren ewable resources. New York: Macmillian. Water Research Act. (1971). Water Research Act No. 34 of 1971. Retrieved from http://www.wrc.org.za/Knowledge%20Hub%20Documents/Other/WaterResearch Act34 71.pdf

PAGE 144

144 Water Research Commission. (2006). South African atlas of agrohydrology and climatology (S. D. Lynch, Editor) Retrieved March 14, 2011, from Table of contents: http://planet.uwc.ac.za/NISL/Invasives/Assignments/GARP/atlas/atlas.htm Wessels, P., & Rooseboom, A. (2009). Flow gauging structures in South African rivers part 2: calibration. Water SA, 35 Retrieved from http://hdl.handle.net/10019.1/10804 Whyte, I. J. (2004). Ecological basis of the new elephant management policy for Kruger National Park and expected outcomes. Pacherdym, 36 99 108. Retrieved from http://african elephant.org/pachy/pdfs/pachy36.pdf#page=102 Wickson, F., Carew, A. L., & Russell, A. W. (2006). Transdisciplinary research: Characteristics, quandaries and quality. Futures, 38 (9), 1046 1059. doi:10.1016/j.futures.2006.02.011 Willmott, C. J., & Matsuura, K. (2005). Advantages of the mean absolute error (MAE) over the root mean square error (RMSE) in assessing average model performance. Climate Res., 30 (1), 79 82. doi:10.3354/cr030079 Willmott, C., Ackleson, S. G., Davis, R. E., Feddema, J. J., Klink, K. M., Legates, D. R., . Rowe, C. M. (1985, September 10). Statistics for the evaluation and comparison of models. J. Geophys. Res., 90 (C5), 8995 9005. doi:10.1029/JC090iC05p08995 Woodborne, S., Huchzermeyer, K. D., Govender, D., Piennar, D. J., Hall, G., Myburgh, J. G., . Lubcker, N. (2012). Ecosystem change and the Olifants River crocodile mass mortality events. Ecosphere, 3 (10). Retrieved from http://dx.doi.org/10.1890/ES12 00170.1 Yoccoz, N. G., Nichols, J. D., & Boulinier, T. (2001, August 1). Monitoring of biological diversity in space and time. Trends in Ecology and Evolution, 16 (8), 446 453. doi:10.1016/S0169 5347(01)02205 4

PAGE 145

145 BIOGRAPHICAL SKETCH sector. He began his career as a Water Resources Engineer with Carollo Engineering P.C. a full service environmental engineering firm based in Kansas. He has consulted for Strathmore University in Kenya in establishing a program agenda for their Water Governance Cen ter and in the Department of Water Affairs and Forestry in South Africa in water resources planning. He was a founding member for Engineers Without Borders at the University of Florida and served as a Graduate Senator for two terms. He has been the recipie nt of numerous awards and recognition for his environmental work most notably the Rotary International Ambassadorial Research Fellowship from Rotary International, 100 Projects for Peace Grant from the Shelby Davis Found ation and the National Research Foun dation Grant of South Africa. Nathan earned a Ph.D from the University of Florida in Agricultural and Biological Engineering in the summer of 2013.