Water Demand and Supply Analysis under Changing Climatic Conditions

Transcription

1 Water Demand and Supply Analysis under Changing Climatic Conditions by Md Mahmudul Haque B.Sc (Civil Engg.), M.Sc (Civil and Environmental Engg.) A thesis submitted in fulfilment for the degree of Doctor of Philosophy School of Computing, Engineering and Mathematics University of Western Sydney Australia November 2014

2 Dedicated To My Wonderful Parents, Beloved Wife and Sweet Sister Who are My Inspirations

3 ABSTRACT ABSTRACT Water is an essential natural resource, which plays a vital role in supporting human life, environmental and ecological systems. Water is increasingly being viewed as a severely stressed resource. This important resource is likely to be affected by climate change conditions in a negative way at many locations which will have direct impact on the ability of a water supply system to ensure adequate water supply to meet customer demands in the future. Both water demand and catchment water yield (i.e. runoff) are the two vital components in estimating adequacy of a water supply system. Since these two components of a water supply system are likely to be affected by changing climate conditions in future, it is essential to estimate climate change impact on them to get a reliable estimate of the yield of a water supply system. Moreover, the acknowledgement and proper quantification of uncertainties in catchment water yield as well as in water demand forecasting are crucial to facilitate decision making by the policy makers to manage water resources effectively and to ensure adequate water supply to the cities. However, despite the emerging concern of climate change issue, application of Global Climate Models (GCMs ) data (which is an important tool in climate change impact studies and which provides future climate scenarios under different greenhouse gas emission conditions) in forecasting water demand is limited in the scientific literature. Moreover, there is a lack of knowledge on the estimation of uncertainty in the water demand forecasting by accounting for the stochastic nature of the predictor variables and their inter correlations. In addition, exploration of the performance of a water supply system under combined effect of uncertain climate, water demand and catchment water yield scenarios are limited. Therefore, this thesis has investigated the impacts of climate change on water demand, catchment yield and water supply system reliability using a suite of statistical techniques, hydrological modelling, uncertainty analysis and outputs from GCMs. In this thesis, the Blue Mountains region and the Blue Mountains Water Supply System in western Sydney, New South Wales, Australia were selected as the study area and water supply system, respectively. University of Western Sydney Page i

4 ABSTRACT In this research, impact of climate change on future water demand has been investigated by using the climate projections from a GCM and uncertainties in demand projections being estimated by developing a long term probabilistic water demand forecasting model considering stochastic nature of the predictor variables and correlation structures. The probabilistic water demand forecasting model has been developed by adopting a Monte Carlo simulation technique with multivariate normal distribution. Climate change impact on future catchment water yield (i.e. runoff) and their associated uncertainties are estimated by coupling the GCMs projections with the rainfall-runoff models. Four different GCMs (MIROC, ECHAM 5, CSIRO Mk. 3 and CCCMA) and two rainfall-runoff models (AWBM and SIMHYD) have been used to estimate future catchment water yield scenarios. This research has also developed an integrated methodology to examine the performance of a water supply system under future climate, water demand and runoff scenarios in the future periods. It has been found that the impacts of potential future climate change on water demand are negligible. On the other hand, it has been found that future catchment runoff/water yield scenarios will be notably affected by the climate change conditions. Moreover, it has been found that consideration of future climate change scenarios on water demand and catchment yield in an integrated modelling framework can provide important insights on the reliability and resilience of a water supply system i.e. when a water supply system may not be able to provide the desired water supply. Furthermore, it has been found that the choice of GCM is the largest source of uncertainty in the forecasted runoff among other possible sources. The uncertainty due to the internal variability of a GCM (i.e. realisation uncertainty) has also been found to be notably high. The ranking of various sources of uncertainties are found to be as: GCM uncertainty > realisation uncertainty > rainfall runoff model uncertainty > rainfall-runoff model parameter uncertainty. From this study, the main recommendation for water authorities/policy makers is to consider a number of possible estimates of future water demand and water yield scenarios in investigating the performance of a water supply system under changing climate regime. Consequently, a number of potential assessment scenarios, GCMs and rainfall-runoff models and associated uncertainties should be considered in estimating the future water demand and water yield scenarios. University of Western Sydney Page ii

5 ABSTRACT The developed methods along with the outcomes of the research would provide vital knowledge about the possible climate change impact on future water demand and runoff, and future performance of a water supply system for better planning and management of a water supply system. This will also help to develop appropriate adaptive strategies to supply adequate water to the communities. The methodologies developed in this thesis can be adopted to other regions and to other water supply systems in Australia and elsewhere in the world. University of Western Sydney Page iii

6 Climate change impact on water demand and Supply STATEMENT OF AUTHENTICATION STATEMENT OF AUTHENTICATION I, Md Mahmudul Haque, declare that all the materials presented in the PhD thesis entitled Water demand and supply analysis under changing climatic conditions are of my own work, and that any work adopted from other sources is duly cited and referenced as such. This thesis contains no material that has been submitted previously, in whole or in part, for any award or degree in other university or institution. Md Mahmudul Haque November, 2014 University of Western Sydney Page iv

7 ACKNOWLEDGEMENTS ACKNOWLEDGEMENTS I would like to express heartiest gratitude and thankfulness to my Principal Supervisor, Associate Professor Ataur Rahman for his scholastic guidance, continued encouragement, invaluable support and suggestions throughout my study at University of Western Sydney (UWS). He has been a great source of confidence and inspiration for me, continuously providing insightful comments with detailed attention to my arguments and timely advice on my work. He offered comprehensive comments and suggestions in reviewing my writings at the same time respecting my voice. He has often taken time to introduce me to the scholarly people within the discipline, kept me focused and carefully listened to my concerns. I am greatly indebted to Dr Rahman for his valuable time and efforts throughout this thesis, and I am truly blessed and honoured to work with such a great supervisor. I also like to thank Dr Rahman for providing me the research assistanceship and opportunity to work in the Australian Rainfall-Runoff revision project. I would also like to give special thanks to Dr Dharma Hagare who has been a strong and supportive Associate Supervisor throughout my study. He has been always helpful and taken his time to thoroughly review my writings and give his valuable suggestions to improve my work. I am also grateful to him for providing financial assistance by giving the teaching assistance throughout my study. I would like to thank Dr Khaled Haddad and late Dr Mohammad Ashrafuz Zaman for their assistance in some of the statistical techniques adopted in this thesis. I would also like to express my gratitude and convey thanks to the Research Director at UWS, Professor Wei Xing Zheng for his wonderful support and guidance during my study. I would like to express my appreciation to Sydney Catchment Authority, and especially Mr Golam Kibria and Mr Mahes Maheswaran, who have been an excellent industry guide for their wonderful support and cooperation to provide the necessary data to carry out the research. I would also like to give thanks to Ms Lucinda Maunsell and Ms Pei Tillman from Sydney Water for their nice support to get the valuable data in relation to water demand in the Blue Mountains region. Finally, I would especially like to thank to my wonderful parents, loving wife and sweet sister for their endless love, care and patience. Moreover, I would like to give University of Western Sydney Page v

8 ACKNOWLEDGEMENTS thanks to my fellow PhD colleagues and friends for the fun times, and being supportive and providing courage and motivation during the study period. University of Western Sydney Page vi

9 LIST OF PUBLICATIONS PUBLICATIONS MADE FROM THE RESEARCH PRESENTED IN THIS THESIS Journal papers 1. Haque, M.M., Rahman, A., Hagare, D., Kibria, G. and Karim, F Estimation of catchment yield and associated uncertainties due to climate change in a mountainous catchment in Australia. Hydrological Processes, published online, DOI: /hyp (ERA 2010 ranking: A, Impact factor: 2.69). 2. Haque, M.M., Hagare, D., Rahman, A., and Kibria, G Quantification of water savings due to drought restrictions in water demand forecasting models. Journal of Water Resources Planning and Management, 140(11), (ERA 2010 ranking: A*, Impact factor: 1.76). 3. Haque, M.M., Rahman, A., Hagare, D. and Kibria, G Parameter uncertainty of the AWBM model when applied to an ungauged catchment. Hydrological Processes, published online, DOI: /hyp (ERA 2010 ranking: A, Impact factor: 2.69). 4. Haque, M.M., Rahman, A., Hagare, D. and Kibria, G Probabilistic water demand forecasting using projected climatic data for Blue Mountains Water Supply System in Australia. Water Resources Management, 28(7), (ERA 2010 ranking: B, Impact factor: 2.46). 5. Haque, M.M., Rahman, A., Hagare, D. and Kibria, G Impact of climate change on future water demand: A case study for the Blue Mountains Water Supply System, NSW, Australia. Water Journal of the Australian Water Association, 41(1), (ERA 2010 ranking: C). 6. Haque, M.M., Rahman, A., Hagare, D. and Kibria, G Principal component regression analysis in water demand forecasting: An application to the Blue Mountains, NSW, Australia. Journal of Hydrology and Environment Research, 1(1), (not ERA ranked). University of Western Sydney Page vii

10 LIST OF PUBLICATIONS 7. Haque, M.M., Egodawatta, P., Rahman, A. and Goonetilleke, A Assessing the significance of climate and community factors on urban water demand. International Journal of Sustainable Built Environment, under review. (not ERA ranked). Full length conference papers 1. Haque, M.M., Rahman, A., Hagare, D. and Kibria, G A comparison of linear and nonlinear regression modelling for forecasting long term urban water demand: A Case Study for Blue Mountains Water Supply System in Australia. In proceedings of the presented in 6 th International Conference on Water Resource and Environmental Research (ICWRER 2013), June 3-7, 2013, Koblenz, Germany. 2. Haque, M.M., Rahman, A., Hagare, D. and Kibria, G Climate change impact assessment on water resources in the Blue Mountains, Australia, In proceedings of the 6 th International Conference on Water Resource and Environmental Research (ICWRER 2013), June 3-7, 2013, Koblenz, Germany. 3. Haque, M.M., Rahman, A., Hagare, D. and Kibria, G Identification of suitable predictor variables for water demand forecasting model by principal component analysis: An application to the Blue Mountains, NSW, Australia. In proceedings of the 35 th IAHR World Congress, September 8-13, 2013, Chengdu, China. 4. Haque, M.M., Haddad, K., Rahman, A., Hossain, M., Hagare, D. and Kibria, G Long term water demand forecasting: Use of Monte Carlo crossvalidation for the best model selection. In proceedings of the 20th International Congress on Modelling and Simulation (MODSIM2013, December 1-6, 2013, Adelaide, Australia. University of Western Sydney Page viii

11 LIST OF PUBLICATIONS Book Chapter 1. Haque, M.M., Ahmed, A. and Rahman, A. (2013). Impacts of water price and restriction on water demand: A case study for Australia. In Water Conservation: Practices, Challenges and Future Implications, edited by Monzur A. Imteaz, published by Nova, ISBN: Haque, M.M., Rahman, A., Goonetilleke, A., Hagare, D. and Kibria, G. (2015). Impact of climate change on urban demand in future decades: An Australian case study, in Advances in Environmental Research, Nova Publishers, USA, under review. University of Western Sydney Page ix

12 TABLE OF CONTENTS TABLE OF CONTENTS ABSTRACT i STATEMENT OF AUTHENTICATION iv ACKNOWLEDGEMENTS v PUBLICATIONS MADE FROM THE RESEARCH PRESENTED IN THIS THESIS vii TABLE OF CONTENTS x LIST OF TABLES xvii LIST OF FIGURES xxi LIST OF ABBREVIATIONS xxxiv CHAPTER 1 : INTRODUCTION 1.1 Overview Background Need for this research Research questions Summary of research undertaken in this thesis Contributions to knowledge Outline of the thesis CHAPTER 2 : REVIEW OF CLIMATE CHANGE IMPACT ANALYSIS ON WATER DEMAND AND YIELD ESTIMATION IN URBAN WATER SUPPLY SYSTEMS 2.1 Overview Climate change issues relevant to water security Linkage of climate change/variables with urban water demand Climate change analysis/studies on future water demand University of Western Sydney Page x

13 TABLE OF CONTENTS 2.5 Impact of water restrictions on urban water demand Urban water demand forecasting Temporal scales/types of urban water demand Forecasting Deterministic vs. probabilistic water demand Forecasting Climate change impact on water resources Uncertainties in climate change impact analysis on catchment yield Uncertainty due to GCM Downscaling uncertainty Emission scenario uncertainty Realisation uncertainty Hydrological model uncertainty Climate change impact analysis on ungauged catchments Uncertainty owing to input data Uncertainty owing to observed gauged/output data Uncertainty owing to choice of optimization technique Uncertainty owing to choice of calibration and validation data length Climate and demand uncertainty on yield of urban water supply systems Summary CHAPTER 3 : STUDY AREA AND DATA 3.1 Overview Case study area and its importance Catchments and dams in the BMWSS University of Western Sydney Page xi

14 TABLE OF CONTENTS 3.4 Climate conditions of the study area Water conservation programs and water restrictions Data collection and future projections of the variables Historical water demand in the BMWSS Water price Number of dwellings Water conservation programs Climate projections Runoff Summary CHAPTER 4 : IMPACT OF WATER RESTRICTIONS ON URBAN WATER DEMAND 4.1 Overview Methodology Multiple regression analysis Estimation of total water savings Yearly base difference method (YBDM) Before and after method (BAM) Expected use method (EUM) Weighted average method (WAM) Model evaluation criteria Leave-One-Out (LOO) cross validation Water demand variables Results Summary University of Western Sydney Page xii

15 TABLE OF CONTENTS CHAPTER 5 : PROBABILISTIC FORECASTING OF LONG TERM URBAN WATER DEMAND 5.1 Overview Methodology Multivariate normal distribution Results Water demand forecasting results in the single dwelling sector Water demand forecasting results in the multiple dwelling sector Summary CHAPTER 6 : IMPACT OF CLIMATE CHANGE ON URBAN WATER DEMAND 6.1 Overview Methodology Principal component analysis Impact of climate change on urban water demand Results Relative influence of variables on urban water demand Impact of climate change on urban water demand Summary CHAPTER 7 : ESTIMATION OF PARAMETER SETS AND EVALUATION OF UNCERTATINTIES IN CALIBRATION OF A RAINFALL-RUNOFF MODEL 7.1 Overview Rainfall - runoff models University of Western Sydney Page xiii

16 TABLE OF CONTENTS AWBM model structure SIMHYD model structure Methodology Model results evaluation criteria Uncertainty due to variability in rainfall time series Uncertainty due to variability in optimization methods Uncertainty due to variability in calibration data lengths Estimation of runoff Results Uncertainty due to input rainfall data Uncertainty due to optimization methods Uncertainty due to calibration data length Runoff estimation at the Katoomba and Blackheath (ungauged) catchments Calibration and runoff estimation results using the SIMHYD model Summary CHAPTER 8 : ESTIMATION OF FUTURE RUNOFF, UNCERTAINTIES AND FUTURE PERFORMANCE OF A WATER SUPPLY SYSTEM UNDER CHANGING CLIMATE CONDITIONS 8.1 Overview Methodology Forecasting runoffs Estimating uncertainties Assessing the reliability of a water supply system Results Rainfall projections University of Western Sydney Page xiv

17 TABLE OF CONTENTS Uncertainty due to internal variability of a GCM (Realisation uncertainty) Uncertainty due to choice of GCMs (GCM uncertainty) Uncertainty due to choice of rainfall-runoff models Uncertainty due to choice of rainfall-runoff model parameter Comparison of uncertainties Forecasting of runoffs (Median projections) Forecasting of runoffs (5 th and 95 th Performance assessment of the Blue Mountains Water Supply System Summary CHAPTER 9 : SUMMARY, CONCLUSIONS AND RECOMMENDATIONS 9.1 Summary Selection of the study area and data collection Assessing the impacts of water restriction on urban water demand Forecasting the long term urban water demand in a probabilistic way Assessing the impact of climate change on urban water demand Estimating the calibrated parameter sets and uncertainties in the calibration of a rainfall-runoff model Estimating future runoff, uncertainties and future performance of a water supply system under changing climate conditions Conclusions Recommendation for further study University of Western Sydney Page xv

18 TABLE OF CONTENTS REFERENCES APPENDIX A : ADDITIONAL TABLES AND FIGURES FROM CHAPTER APPENDIX B : ADDITIONAL TABLES AND FIGURES FROM CHAPTER APPENDIX C : ADDITIONAL TABLES AND FIGURES FROM CHAPTER APPENDIX D : ADDITIONAL TABLES AND FIGURES FROM CHAPTER University of Western Sydney Page xvi

19 LIST OF TABLES LIST OF TABLES Table 2.1 Levels, scope and timing of water restrictions imposed in Sydney during the drought periods ( ) Table 2.2 List of Global Climate Models (GCMs), (Randall et al. 2007) Table 2.3 Future growth patterns of population, economy and technology owing to four sets of scenario storylines (A1, A2, B1 and B2) (Nakićenović et al. 2000) Table 3.1 Monthly mean maximum temperature of the Blue Mountains region for the period of Table 3.2 Water price data for the Blue Mountains region of the period 1997 to Table 3.3 Average water savings from the water conservation programs Table 3.4 List of the GCMs used in this study and their spatial resolution Table 4.1 List of dependent and independent variables used in developing water demand models Table 4.2 Performance statistics of the developed models for the single dwelling sector 68 Table 4.3 Performance statistics of the developed models for the multiple dwelling sector Table 4.4 Percentage of water savings due to water restrictions during the drought periods ( ) in the single dwelling sector in the Blue Mountains region Table 4.5 Percentage of water savings due to water restrictions during the drought periods ( ) in the multiple dwelling sector in the Blue Mountains region Table 4.6 Performance statistics of the developed Semi-Log model for the forecasting period (July 2009 to September 2011) in the single and multiple dwelling sectors Table th percentile (most expected value) of the forecasted water demand values for the single dwelling sector in the Blue Mountains region in the period of under twelve water demand scenarios Table th percentile (most expected value) of the forecasted water demand values for the multiple dwelling sector in the Blue Mountains region in the period of under twelve water demand scenarios Table 6.1 Description of the dependent and independent variables used in the Principal Component Analysis (PCA) Table 6.2 Water demand forecasting results of the decades of and in the single dwelling sector (Bracketed results indicate percentage changes in the forecasting results in comparison to the predicted water demand under current climate condition ( )) University of Western Sydney Page xvii

20 LIST OF TABLES Table 6.3 Projection of future climate by the CSIRO Mk. 3 global climate model (GCM) under three emission scenarios (A1B, A2 and B1) of the two decades and (Bracketed results indicate percentage changes in the forecasting temperature and rainfall values in comparison to the observed climate data of ) Table 6.4 Projection of future water demand under three hypothetical climate change scenarios in the year, 2040 in the single dwelling sector Table 6.5 Water demand forecasting results of the decades of and in the multiple dwelling sector (Bracketed results indicate percentage changes in the forecasting results in comparison to the predicted water demand under current climate condition ( )) Table 6.6 Projection of future water demand under three hypothetical climate change scenarios in the year, 2040 in the multiple dwelling sector Table 7.1 Descriptions of the AWBM and SIMHYD model parameters Table 7.2 Comparison of calibration results of the AWBM model with five different rainfall inputs Table 7.3 Effect of rainfall scaling factor on the calibration results of the AWBM model Table 7.4 Performance statistics of the AWBM model based on different optimization methods Table 7.5 Values of NSE (total), average ratio, MBIAS (%) and V (%) for the 23 tests due to different calibration and validation data lengths adopting the AWBM model Table 7.6 Estimated runoff of the Blue Mountains catchments (Katoomba and Blackheath) by the AWBM model for the period by transposing the calibrated parameter from the nearby catchment Table 7.7 Estimated annual average runoff for the Katoomba and Blackheath catchments on the basis of the regional methods by Boughton (2009) and Boughton and Chiew (2007) for the period of Table 7.8 Performance statistics of the SIMHYD model based on different optimization methods Table 7.9 Values of NSE (total), average ratio, MBIAS (%) and V (%) for the 23 tests due to different calibration and validation data lengths adopting the SIMHYD model Table 7.10 Estimated runoff of the Blue Mountains catchments (Katoomba and Blackheath) by the SIMHYD model for the period by transposing the calibrated parameter from the nearby catchment Table 8.1 Twelve combinations of future water demand and runoff scenarios to assess the performance of the Blue Mountains Water Supply System University of Western Sydney Page xviii

21 LIST OF TABLES Table 8.2 Percentage of rainfall changes in the future decades projected by the four GCMs compared to annual average rainfall during the period Table 8.3 Percentage changes of median projections of mean annual runoff in comparison to the reference period ( ) in the Blue Mountains catchments Table 8.4 Percentage changes in the 5 th and 95 th percentiles projections of the mean annual runoff adopting the AWBM model in comparison to the reference period ( ) Table 8.5 Forecasted values of reliability and security criteria for the Blue Mountains Water Supply System under A1B-No water demand and four runoff scenarios in the periods Table 8.6 Forecasted values of reliability and security criteria for the Blue Mountains Water Supply System (BMWSS) under A1B-L1 water demand and four runoff scenarios in the periods Table 8.7 Forecasted values of reliability and security criteria for the Blue Mountains Water Supply System (BMWSS) under A1B-L2 water demand and four runoff scenarios in the periods Table 8.8 Forecasted values of reliability and security criteria for the Blue Mountains Water Supply System (BMWSS) under A1B-L3 water demand and four runoff scenarios in the periods Table A.4.1 Calculation of total water savings by yearly base difference method (YBDM) in 2004 for the single dwelling sector Table A.4.2 Monthly average base water use ( ) in the single dwelling sector Table A.4.3 Calculation of total water savings by before and after method (BAM) in 2004 for the single dwelling sector Table A.4.4 Calculation of total water savings by expected use method (EUM) in 2004 for the single dwelling sector Table A.4.5 Calculation of total water savings by weighted average method (WAM) in 2004 for the single dwelling sector Table A.4.6 Calculation of water savings from conservation programs implemented in the Blue Mountains region in January, Table A.4.7 Results of leave-one-out cross validation of the developed model for the single dwelling sector Table A.4.8 Results of leave-one-out cross validation of the developed model for the multiple dwelling sector Table C.7.1 AWBM model parameter values for the selected three parameter sets Table C.7.2 SIMHYD model parameter values for the selected three parameter sets Table D.8.1 Forecasted values of reliability and security criteria for the Blue Mountains Water Supply System under A2-No water demand and four runoff scenarios in the periods University of Western Sydney Page xix

22 LIST OF TABLES Table D.8.2 Table D.8.3 Table D.8.4 Table D.8.5 Table D.8.6 Table D.8.7 Table D.8.8 Forecasted values of reliability and security criteria for the Blue Mountains Water Supply System under A2-L1 water demand and four runoff scenarios in the periods Forecasted values of reliability and security criteria for the Blue Mountains Water Supply System under A2-L2 water demand and four runoff scenarios in the periods Forecasted values of reliability and security criteria for the Blue Mountains Water Supply System under A2-L3 water demand and four runoff scenarios in the periods Forecasted values of reliability and security criteria for the Blue Mountains Water Supply System under B1-No water demand and four runoff scenarios in the periods Forecasted values of reliability and security criteria for the Blue Mountains Water Supply System under B1-L1 water demand and four runoff scenarios in the periods Forecasted values of reliability and security criteria for the Blue Mountains Water Supply System under B1-L2 water demand and four runoff scenarios in the periods Forecasted values of reliability and security criteria for the Blue Mountains Water Supply System under B1-L3 water demand and four runoff scenarios in the periods University of Western Sydney Page xx

23 LIST OF FIGURES LIST OF FIGURES Figure 1.1 Illustration of the major steps conducted in this research study Figure 2.1 Different pathways of downscaling of GCM outputs (Fowler et al. 2007) Figure 3.1 Location of the Blue Mountains region in the New South Wales, Australia Figure 3.2 Water supply zone (Mt Victoria to Faulconbridge) of the Blue Mountains water supply system (City of Blue Mountains 2007) Figure 3.3 Blue Mountains water supply system (Sydney Catchment Authority 2009a) Figure 3.4 Location maps of the Blue Mountains dams in the New South Wales, Australia (Sydney Catchment Authority 2014) Figure 3.5 Location maps of the study catchments (Katoomba, Blackheath and Narrow Neck catchments) (NSW Office of Water 2014) Figure 3.6 Annual rainfall of the Blue Mountains region for the period of (red line represents annual average rainfall) - 45 Figure 3.7 Composition of total water consumption in the Blue Mountains region, NSW, Australia for the period of Figure 3.8 Yearly total water consumption ( ) in the Blue Mountains region, Australia Figure 3.9 Per dwelling monthly water consumption of the residential sector ( ) in the Blue Mountains region, Australia Figure 3.10 Number of total dwellings in the Blue Mountains region during the period of 1997 to Figure 3.11 Number of participated dwellings in the water conservation programs (1: WaterFix, 2: Rainwater tank, 3: DIY, 4: Washing machine and 5: Toilet replacement) in the Blue Mountains region for the period of 2000 to Figure 4.1 Framework for quantifying water savings and developing water demand forecasting models Figure 4.2 Framework of calculating total water savings by expected use method (EUM) Figure 4.3 Comparison of the observed and modelled water use for the period of January 1999 to December 2002 in the Blue Mountains region using climate water demand model Figure 4.4 Validation results of the observed and modelled water use for the period of January 1997 to December 1998 in the Blue Mountains region using climate water demand model University of Western Sydney Page xxi

24 LIST OF FIGURES Figure 4.5a Figure 4.5b Figure 4.5c Figure 4.5d Figure 4.6a Figure 4.6b Figure 4.6c Figure 4.6d Figure 4.7 Figure 4.8 Comparison of modelled versus observed water consumption by the best model (Semi-Log) for the single dwelling sector during water restriction periods under yearly base difference method (YBDM) of water savings calculation Comparison of modelled versus observed water consumption by the best model (Raw-Data) for the single dwelling sector during water restriction periods under expected use method (EUM) of water savings calculation Comparison of modelled versus observed water consumption by the best model (Log-Log) for the single dwelling sector during water restriction periods under weighted average method (WAM) of water savings calculation Comparison of modelled versus observed water consumption by the best model (Raw-Data) for the single dwelling sector during water restrictions periods under before and after method (BAM) of water savings calculation Comparison of modelled versus observed water consumption by the best model (Semi-Log) under yearly base difference method (YBDM) of water savings calculation for the multiple dwelling sector during water restriction periods Comparison of modelled versus observed water consumption by the best model (Log-Log) under expected use method (EUM) of water savings calculation for the multiple dwelling sector during water restriction periods Comparison of modelled versus observed water consumption by the best model (Raw data) under weighted average method (WAM) of water savings calculation for the multiple dwelling sector during water restriction periods Comparison of modelled versus observed water consumption by the best model (Log-Log) under before and after method (BAM) of water savings calculation for the multiple dwelling sector during water restriction periods Comparison of monthly forecasted versus observed water demand by the Semi-Log model coupled with YBDM for the forecasting period (2009 to 2011) in the single dwelling sector Comparison of yearly forecasted versus observed water demand by the Semi-Log model coupled with YBDM for the forecasting period (2009 to 2011) in the single dwelling sector University of Western Sydney Page xxii

25 LIST OF FIGURES Figure 4.9 Figure 4.10 Figure 5.1 Figure 5.2a Figure 5.2b Figure 5.2c Figure 5.2d Figure 5.3a Figure 5.3b Comparison of monthly forecasted versus observed water demand values using the Semi-Log model coupled with YBDM for the forecasting period (2009 to 2011) in the multiple dwelling sector Comparison of yearly forecasted versus observed water demand values using the Semi-Log model coupled with YBDM for the forecasting period (2009 to 2011) in the multiple dwelling sector Framework of estimating future water demand scenarios adopting a probabilistic method % confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for A1B climate scenario and no water restriction condition for the single dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) % confidence intervals and 50th percentile of the forecasted total yearly water demands from 2015 to 2040 for A1B climate scenario and Level 1 water restriction condition for the single dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) % confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for A1B climate scenario and Level 2 water restriction condition for the single dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) % confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for A1B climate scenario and Level 3 water restriction condition for the single dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) % confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for A1B climate scenario and no water restriction condition for the multiple dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) % confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for A1B climate scenario and Level 1 water restriction condition for the multiple dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) University of Western Sydney Page xxiii

26 LIST OF FIGURES Figure 5.3c 90% confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for A1B climate scenario and Level 2 water restriction condition for the multiple dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) Figure 5.3d 90% confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for A1B climate scenario and Level 3 water restriction condition for the multiple dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) Figure 6.1 Framework of the climate change impact assessment on urban water demand ( T refers to monthly maximum temperature and R refers to monthly total rainfall) Figure 6.2 Resulting PCA biplot (PC 1 vs. PC 2) of variables (8 independent variables) influencing water demand, (Data labels (e.g.08_12; Year_Month) indicate the corresponding year and month in the data matrix) Figure 6.3 Resulting PCA biplot (PC 1 vs. PC 2) on modified data matrix of 9 variables including dependent variable, PDWC (Data labels (e.g.08_12; Year_Month) indicate the corresponding year and month in the data matrix) Figure 6.4 Resulting PCA biplot (PC 1 vs. PC 2) on modified data matrix of 6 variables (one dependent and five independent variables) after removing highly correlated variables from similar kinds (Data labels (e.g.08_12; Year_Month) indicate the corresponding year and month in the data matrix) Figure 6.5 Projection of water demand under A1B, A2, B1 and current climate conditions for the period in the single dwelling sector Figure 6.6 Projection of water demand under A1B, A2, B1 and current climate conditions for the period in the multiple dwelling sector Figure 7.1 Structure of the AWBM model (Boughton 2004) Figure 7.2 Structure of the SIMHYD model (Podger 2004) Figure 7.3 Total, calibration and validation NSE values for the 23 tests due to different calibration and validation data lengths adopting the AWBM model Figure 8.1 Framework of forecasting runoff adopting the AWBM and SIMHYD model using the projected climate data Figure 8.2 Framework of estimating uncertainty due to choice of Global Climate Models (i.e. GCM uncertainty) University of Western Sydney Page xxiv

27 LIST OF FIGURES Figure 8.3 Framework of estimating uncertainty due to internal variability of a Global climate model (i.e. realisation uncertainty) Figure 8.4 Framework of estimating uncertainty due to choice of rainfall-runoff models (i.e. rainfall-runoff model uncertainty) Figure 8.5 Framework of estimating uncertainty due to choice of rainfall-runoff model parameter sets (i.e. rainfall-runoff model parameter uncertainty) 152 Figure 8.6 Forecasted 36 total water demand scenarios for the period of Figure 8.7 Forecasted 12 runoff scenarios for the period of Figure 8.8 Coefficient of variation (C V ) values of the simulated median runoffs using data from the four GCMs (i.e. CSIRO, CCCMA, ECHAM 5, MIROC). The red horizontal line represents the average CV value Figure 8.9 Coefficient of variation (C V ) values of the simulated runoffs using ECHAM 5 model data (i.e. realisation uncertainty). The red horizontal line represents the average C V value Figure 8.10 Coefficient of variation (C V ) values of the simulated median runoffs using the CSIRO global climate model data adopting the AWBM and SIMHYD models (i.e. rainfall-runoff model uncertainty). The red horizontal line represents the average C V value Figure 8.11 Coefficient of variation (C V ) values of the simulated median runoffs by the AWBM model using the CCCMA GCM data adopting three different calibrated parameter sets (i.e. rainfall-runoff model parameter uncertainty). The red horizontal line represents the average C V value Figure 8.12 Comparison of four types of uncertainties by presenting the average C V (%) values of the forecasted runoff Figure 8.13 Projected status of the Blue Mountains storage under the most probable scenario (Demand: 50 th percentile + Runoff: 50 th using the forecasted water demand (A1B-No) and runoff (MIROC) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Figure 8.14 Projected status of the Blue Mountains storage under the most favourable scenario (Demand: 5 th percentile + Runoff: 95 th using the forecasted water demand (A1B-No) and runoff (MIROC) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Figure 8.15 Projected status of the Blue Mountains storage under the worst scenario (Demand: 95 th percentile + Runoff: 5 th using the forecasted water demand (A1B-No) and runoff (MIROC) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply University of Western Sydney Page xxv

28 LIST OF FIGURES Figure 8.16 Figure 8.17 Figure 8.18 Figure 8.19 Figure 8.20 Figure 8.21 Figure 8.22 Figure 8.23 Figure 8.24 Projected status of the Blue Mountains storage under the most probable scenario (Demand: 50 th percentile + Runoff: 50 th using the forecasted water demand (A1B-L1) and runoff (MIROC) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most favourable scenario (Demand: 5 th percentile + Runoff: 95 th using the forecasted water demand (A1B-L1) and runoff (MIROC) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the worst scenario (Demand: 95 th percentile + Runoff: 5 th using the forecasted water demand (A1B-L1) and runoff (MIROC) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most probable scenario (Demand: 50 th percentile + Runoff: 50 th using the forecasted water demand (A1B-L2) and runoff (MIROC) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most favourable scenario (Demand: 5 th percentile + Runoff: 95 th using the forecasted water demand (A1B-L2) and runoff (MIROC) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the worst scenario (Demand: 95 th percentile + Runoff: 5 th using the forecasted water demand (A1B-L2) and runoff (MIROC) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most probable scenario (Demand: 50 th percentile + Runoff: 50 th using the forecasted water demand (A1B-L3) and runoff (MIROC) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most favourable scenario (Demand: 5 th percentile + Runoff: 95 th using the forecasted water demand (A1B-L3) and runoff (MIROC) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the worst scenario (Demand: 95 th percentile + Runoff: 5 th using the forecasted water demand (A1B-L3) and runoff (MIROC) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply University of Western Sydney Page xxvi

29 LIST OF FIGURES Figure B.5.1(a) Figure B.5.1(b) Figure B.5.1(c) Figure B.5.1(d) Figure B.5.2(a) Figure B.5.2(b) Figure B.5.2(c) Figure B.5.2(d) 90% confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for A2 climate scenario and no water restriction condition for the single dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) % confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for A2 climate scenario and no water restriction condition for the single dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) % confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for A2 climate scenario and Level 2 water restriction condition for the single dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) % confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for A2 climate scenario and Level 3 water restriction condition for the single dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) % confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for B1 climate scenario and No water restriction condition for the single dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) % confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for B1 climate scenario and Level 1 water restriction condition for the single dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) % confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for B1 climate scenario and Level 2 water restriction condition for the single dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) % confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for B1 climate scenario and Level 3 water restriction condition for the single dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) University of Western Sydney Page xxvii

30 LIST OF FIGURES Figure B.5.3(a) Figure B.5.3(b) Figure B.5.3(c) Figure B.5.3(d) Figure B.5.4(a) Figure B.5.4(b) Figure B.5.4(c) Figure B.5.4(d) 90% confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for A2 climate scenario and No water restriction condition for the single dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) % confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for A2 climate scenario and Level 1 water restriction condition for the single dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) % confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for A2 climate scenario and Level 2 water restriction condition for the single dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) % confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for A2 climate scenario and Level 3 water restriction condition for the single dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) % confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for B1 climate scenario and No water restriction condition for the single dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) % confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for B1 climate scenario and Level 1 water restriction condition for the single dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) % confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for B1 climate scenario and Level 2 water restriction condition for the single dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) % confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for B1 climate scenario and Level 3 water restriction condition for the single dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) University of Western Sydney Page xxviii

31 LIST OF FIGURES Figure D.8.1 Figure D.8.2 Figure D.8.3 Figure D.8.4 Figure D.8.5 Figure D.8.6 Figure D.8.7 Figure D.8.8 Figure D.8.9 Figure D.8.10 Coefficient of variation (C V ) values of the simulated runoffs using MIROC model data (i.e. realisation uncertainty). The red horizontal line represents the average C V value Coefficient of variation (C V ) values of the simulated runoffs using CCCMA model data (i.e. realisation uncertainty). The red horizontal line represents the average C V value Coefficient of variation (C V ) values of the simulated runoffs using CSIRO model data (i.e. realisation uncertainty). The red horizontal line represents the average C V value Projected status of the Blue Mountains storage under the most probable scenario (Demand: 50 th percentile + Runoff: 50 th using the forecasted water demand (A1B-No) and runoff (ECHAM 5) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most probable scenario (Demand: 50 th percentile + Runoff: 50 th using the forecasted water demand (A1B-No) and runoff (CSIRO) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most probable scenario (Demand: 50 th percentile + Runoff: 50 th using the forecasted water demand (A1B-No) and runoff (CCCMA) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most favourable scenario (Demand: 5 th percentile + Runoff: 95 th using the forecasted water demand (A1B-No) and runoff (ECHAM 5) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most favourable scenario (Demand: 5 th percentile + Runoff: 95 th using the forecasted water demand (A1B-No) and runoff (CSIRO) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most favourable scenario (Demand: 5 th percentile + Runoff: 95 th using the forecasted water demand (A1B-No) and runoff (CCCMA) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the worst scenario (Demand: 95 th percentile + Runoff: 5 th using the forecasted water demand (A1B-No) and runoff (ECHAM 5) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply University of Western Sydney Page xxix

32 LIST OF FIGURES Figure D.8.11 Figure D.8.12 Figure D.8.13 Figure D.8.14 Figure D.8.15 Figure D.8.16 Figure D.8.17 Figure D.8.18 Figure D.8.19 Projected status of the Blue Mountains storage under the worst scenario (Demand: 95 th percentile + Runoff: 5 th using the forecasted water demand (A1B-No) and runoff (CSIRO) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the worst scenario (Demand: 95 th percentile + Runoff: 5 th using the forecasted water demand (A1B-No) and runoff (CCCMA) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most probable scenario (Demand: 50 th percentile + Runoff: 50 th using the forecasted water demand (A1B-L1) and runoff (ECHAM 5) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most probable scenario (Demand: 50 th percentile + Runoff: 50 th using the forecasted water demand (A1B-L1) and runoff (CSIRO) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most probable scenario (Demand: 50 th percentile + Runoff: 50 th using the forecasted water demand (A1B-L1) and runoff (CCCMA) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most favourable scenario (Demand: 5 th percentile + Runoff: 95 th using the forecasted water demand (A1B-L1) and runoff (ECHAM 5) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most favourable scenario (Demand: 5 th percentile + Runoff: 95 th using the forecasted water demand (A1B-L1) and runoff (CSIRO) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most favourable scenario (Demand: 5 th percentile + Runoff: 95 th using the forecasted water demand (A1B-L1) and runoff (CCCMA) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the worst scenario (Demand: 95 th percentile + Runoff: 5 th using the forecasted water demand (A1B-L1) and runoff (ECHAM 5) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply University of Western Sydney Page xxx

33 LIST OF FIGURES Figure D.8.20 Figure D.8.21 Figure D.8.22 Figure D.8.23 Figure D.8.24 Figure D.8.25 Figure D.8.26 Figure D.8.27 Figure D.8.28 Projected status of the Blue Mountains storage under the worst scenario (Demand: 95 th percentile + Runoff: 5 th using the forecasted water demand (A1B-L1) and runoff (CSIRO) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the worst scenario (Demand: 95 th percentile + Runoff: 5 th using the forecasted water demand (A1B-L1) and runoff (CCCMA) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most probable scenario (Demand: 50 th percentile + Runoff: 50 th using the forecasted water demand (A1B-L2) and runoff (ECHAM 5) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most probable scenario (Demand: 50 th percentile + Runoff: 50 th using the forecasted water demand (A1B-L2) and runoff (CSIRO) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most probable scenario (Demand: 50 th percentile + Runoff: 50 th using the forecasted water demand (A1B-L2) and runoff (CCCMA) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most favourable scenario (Demand: 5 th percentile + Runoff: 95 th using the forecasted water demand (A1B-L2) and runoff (ECHAM 5) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most favourable scenario (Demand: 5 th percentile + Runoff: 95 th using the forecasted water demand (A1B-L2) and runoff (CSIRO) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most favourable scenario (Demand: 5 th percentile + Runoff: 95 th using the forecasted water demand (A1B-L2) and runoff (CCCMA) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the worst scenario (Demand: 95 th percentile + Runoff: 5 th using the forecasted water demand (A1B-L2) and runoff (ECHAM 5) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply University of Western Sydney Page xxxi

34 LIST OF FIGURES Figure D.8.29 Figure D.8.30 Figure D.8.31 Figure D.8.32 Figure D.8.33 Figure D.8.34 Figure D.8.35 Figure D.8.36 Figure D.8.37 Projected status of the Blue Mountains storage under the worst scenario (Demand: 95 th percentile + Runoff: 5 th using the forecasted water demand (A1B-L2) and runoff (CSIRO) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the worst scenario (Demand: 95 th percentile + Runoff: 5 th using the forecasted water demand (A1B-L2) and runoff (CCCMA) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most probable scenario (Demand: 50 th percentile + Runoff: 50 th using the forecasted water demand (A1B-L3) and runoff (ECHAM 5) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most probable scenario (Demand: 50 th percentile + Runoff: 50 th using the forecasted water demand (A1B-L3) and runoff (CSIRO) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most probable scenario (Demand: 50 th percentile + Runoff: 50 th using the forecasted water demand (A1B-L3) and runoff (CCCMA) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most favourable scenario (Demand: 5 th percentile + Runoff: 95 th using the forecasted water demand (A1B-L3) and runoff (ECHAM 5) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most favourable scenario (Demand: 5 th percentile + Runoff: 95 th using the forecasted water demand (A1B-L3) and runoff (CSIRO) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most favourable scenario (Demand: 5 th percentile + Runoff: 95 th using the forecasted water demand (A1B-L3) and runoff (CCCMA) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the worst scenario (Demand: 95 th percentile + Runoff: 5 th using the forecasted water demand (A1B-L3) and runoff (ECHAM 5) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply University of Western Sydney Page xxxii

35 LIST OF FIGURES Figure D.8.38 Figure D.8.39 Figure D.8.40 Figure D.8.41 Figure D.8.42 Figure D.8.43 Figure D.8.44 Figure D.8.45 Figure D.8.46 Figure D.8.47 Projected status of the Blue Mountains storage under the worst scenario (Demand: 95 th percentile + Runoff: 5 th using the forecasted water demand (A1B-L3) and runoff (CSIRO) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the worst scenario (Demand: 95 th percentile + Runoff: 5 th using the forecasted water demand (A1B-L3) and runoff (CCCMA) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most probable scenario (Demand: 50 th percentile + Runoff: 50 th using the forecasted water demand (A2-No) and runoff (MIROC) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most probable scenario (Demand: 50 th percentile + Runoff: 50 th using the forecasted water demand (A2-L1) and runoff (MIROC) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most probable scenario (Demand: 50 th percentile + Runoff: 50 th using the forecasted water demand (A2-L2) and runoff (MIROC) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most probable scenario (Demand: 50 th percentile + Runoff: 50 th using the forecasted water demand (A2-L3) and runoff (MIROC) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most probable scenario (Demand: 50 th percentile + Runoff: 50 th using the forecasted water demand (B1-No) and runoff (MIROC) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most probable scenario (Demand: 50 th percentile + Runoff: 50 th using the forecasted water demand (B1-L1) and runoff (MIROC) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most probable scenario (Demand: 50 th percentile + Runoff: 50 th using the forecasted water demand (B1-L2) and runoff (MIROC) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply Projected status of the Blue Mountains storage under the most probable scenario (Demand: 50 th percentile + Runoff: 50 th using the forecasted water demand (B1-L3) and runoff (MIROC) scenarios: (a) without the FRWS water supply, (b) with the FRWS water supply University of Western Sydney Page xxxiii

36 LIST OF ABBREVIATIONS LIST OF ABBREVIATIONS AARE AWBM BAM BH BMWSS CWDM DIY EUM EVP FRWS GA GCM IPCC KT LOO MBIAS MLR MMT MVN NARCliM NRD NSE NSW PBIAS PC PCA PDF PS PSMS PWDC RCM RF Absolute average relative error Australia water balance model Before and after method Blackheath Blue Mountains water supply system Climate water demand model Do-it-yourself Expected use method Evaporation Fish river water scheme Genetic algorithm Global climate model Intergovernmental panel for climate change Katoomba Leave-one-out Median of bias Multiple linear regression Monthly mean maximum temperature Multivariate normal distribution NSW/ACT regional climate modelling Number of rain days Nash-Sutcliffe efficiency New South Wales Percentage of bias Principal component Principal component analysis Probability density function Pattern search Pattern search multi start Per dwelling water consumption Regional climate model Rainfall University of Western Sydney Page xxxiv

37 LIST OF ABBREVIATIONS RMSS RRL RS SCE-EU SE SRES URS UWS WAM WCS WP WRS YBDM Rosenbrock multi start search Rainfall-runoff library Rosenbrock search Shuffle complex evolution Solar exposure Special report on emission scenarios Uniform random search University of Western Sydney Weighted average method Water conservation savings Water price Water restriction savings Yearly base difference method University of Western Sydney Page xxxv

38 CHAPTER 1: Introduction CHAPTER 1 INTRODUCTION 1.1 Overview This thesis focuses on the estimation of climate change impact on urban water demand and water supply with associated uncertainties. Climate change impact on water demand is estimated by developing a probabilistic long term water demand forecasting model that allows consideration of the stochastic nature and the inter correlation structure of the independent variables. Probabilistic water demand forecasting model is developed by adopting a Monte Carlo simulation technique with multivariate normal distribution. Then projections of future climate from a Global Climate Model (GCM) under different emission scenarios are fed into the developed water demand model to simulate future water demand scenarios. Climate change impact on future catchment water yield (i.e. runoff) and their associated uncertainties are estimated by coupling the GCM projections with the rainfall-runoff models. Four different GCMs and two rainfall-runoff models are used to estimate future catchment water yield scenarios. Then forecasted water demand and catchment water yield scenarios are integrated to assess the performance of a water supply system under climate change regime. This chapter of the thesis begins by presenting a background to this study, need for this research, research questions, summary of the research tasks, research contributions and an outline of the thesis. 1.2 Background Climate change issue has been emerged as an increasing concern among the water planners and managers around the globe as it is likely to change water management tasks by altering the availability of fresh water resources and by changing the water demand pattern. Rises in temperature and changes in rainfall patterns are expected to occur in many parts of the world due to the potential changes in future climatic conditions (IPCC 2007). These changes are likely to affect the water balance at local, regional and global scales in a negative way at many regions, which would make water supply a challenging task for many cities. Moreover, some other factors such University of Western Sydney Page 1

39 CHAPTER 1: Introduction as increasing population, rise in water demand, rapid urbanization and water pollution are likely to affect water availability at a given location in future. Climate change is expected to have some impacts on urban water demand as water demand varies with climate variables to some extent especially with temperature and rainfall. Rainfall and temperature have influence on outdoor activities particularly gardening. During the hot days and low rainfall periods, more water is required in the gardens. Moreover, use of water in the swimming pool and for personal hygiene (e.g. bathing) increases during the hotter periods. Influence of the climate variables, especially influence of rainfall and temperature on water demand has been reported by several studies (e.g. Babel et al. 2007, Gato et al. 2007, Corbella and Sauri Pujol 2009, Xiao-jun et al. 2013). In Australian cities, water supply is more vulnerable to the changes in climatic conditions as it is highly dependent on rainfall and storage capacity of surface water reservoirs (ABS 2010, ABS 2012). However, rainfall in Australia is highly variable (Sahin et al. 2013) and about 50 to 70% of the country are in the semi-arid and arid regions where rainfall is very low (Zaman et al. 2012). During the recent droughts in Australia ( ), most of the major reservoirs were reached at a critical low water level thereby making water supply at risk. Consequently, different levels of water restrictions based on the severity of the drought conditions were imposed by the water authorities in many Australian cities to limit residential water consumption to deal with inadequate water supply (Queensland Water Commission 2010, Sydney Water 2010). Annual average temperature in Australia has increased by C from 1910 to 2011 (CSIRO 2012) which is higher than the global average increase of C for the same period (Cleugh et al. 2011). Majority of this increment in temperature has occurred since 1950 s with the highest increment in the eastern part of Australia by 2 0 C and lowest change in the northwest part by C (Head et al. 2014). Moreover, from 1957 numbers of hot days and nights have increased, and just opposite has been observed for the number of cold days and nights (Nicholls and Collins 2006). In addition, projection of temperature in Australia indicates that the annual average temperature may go higher by approximately 1 0 C by 2030 relative to 1990 with an increment of about C in coastal and C in inland areas (CSIRO 2012). University of Western Sydney Page 2

40 CHAPTER 1: Introduction By 2050 and 2070, the increment may go up to C and 5 0 C, respectively, under high emission scenarios (CSIRO 2012). Rainfall in Australia generally shows significant variability from year to year (Hennessy et al. 2008). Notable changes in rainfall have also been observed in Australia since 1950 s, mainly in northwest Australia, southwest Western Australia, southeast Australia and northeast Australia (Keenan and Cleugh 2011). Increase in annual rainfall has been observed only in northwest region whereas southwest Western Australia has experienced a steady decline in rainfall over the past 30 years, and southeast and eastern parts of Australia have become drier since the mid 1990s including a reduction in March-May rainfall by 61% (Murphy and Timbal 2008, Cleugh et al. 2011). Moreover, annual average rainfall is expected to be altered by around -10% to + 5% in northern areas and -10% to little change in southern areas by 2030, though a high level of uncertainty is present in the prediction results due to the variation in the results of different GCMs (CSIRO 2012). Projected changes in the rainfall are larger in the later part of the century. These past changes and probable projections of the climatic conditions have raised a concern about meeting the necessary requirements of water supply to the current and future population of Australia. Therefore, impacts of the climate change on water demand and water supply need to be identified in order to plan for appropriate measures (e.g. expansion of existing water supply systems, sourcing new water supply catchments, building desalination plants and managing water demand) to supply water to the community with a desired level of security and satisfaction. Impact of climate change on water demand can be estimated by forecasting long term water demand adopting future plausible climatic scenarios. Long term water demand can be forecasted by the deterministic and probabilistic models (Froukh 2001, Almutaz et al. 2012). Deterministic models generally forecast single value of water demand without considering the stochastic nature of the independent variables. As water demand depends on different independent variables which are stochastic in nature and are correlated among themselves such as population, household size, income, water usage price, rainfall, temperature and conservation measures (Babel et al. 2011, Qi and Chang 2011), the usefulness of deterministic models in forecasting urban water demand may be limited. If the associated uncertainties in the University of Western Sydney Page 3

41 CHAPTER 1: Introduction independent variables are ignored, the forecasted water demands may not be realistic and adequate for efficient planning and management of water supply systems. Therefore, uncertainties associated with the independent variables should be explicitly incorporated into demand forecasting models to allow decision makers to understand how uncertainties in the independent variables may affect the future water demand. Impact of climate change on catchment water yield is generally estimated through the combination of climate and rainfall-runoff models. GCMs are extensively used to generate future climate scenarios to be used in the climate change impact studies on catchment water yield (Boé et al. 2007, Chen et al. 2007, Fowler et al. 2007). GCM outputs are downscaled first to obtain the catchment scale/appropriate scale climate projection data and then these downscaled climate projections are taken as input into the rainfall-runoff models to estimate climate change impact on catchment runoff (Chen et al. 2011). However, different types of uncertainties are associated with the climate change impact studies due to choice of GCMs, greenhouse gas emission scenarios, downscaling methods, rainfall-runoff model structures and rainfall-runoff model parameters (Kay et al. 2009, Teutschbein et al. 2011). It has been recognised by many studies (Graham et al. 2007, Jiang et al. 2007, Teutschbein and Seibert 2010) that uncertainties should be taken into account during the investigation of climate change impact on water resources in order to produce reliable estimates/forecasts. Both water demand and catchment water yield are the two vital components in estimating adequacy of a water supply system. Since these two components of a water supply system are likely to be affected by changing climatic conditions in future, it is essential to estimate climate change impact on them to get a reliable estimate of the yield of a water supply system under changing climatic conditions. Moreover, the acknowledgement and proper quantification of uncertainties in catchment water yield as well as in water demand forecasting are crucial to facilitate decision making by the policy makers to manage water resources effectively and to ensure adequate water supply to the cities. Hence, this thesis is devoted to develop a modelling framework to assess the impacts of climate change on both future water demand and catchment water yield with associated uncertainties to evaluate the University of Western Sydney Page 4

42 CHAPTER 1: Introduction performance of a water supply system under plausible future climate, water demand and catchment water yield scenarios. It should be noted that rainwater harvesting has been indirectly taken into account in the thesis by considering water saving associated it. However, alternative water supplies such as groundwater and greywater recycling have not been considered in this thesis in the catchment yield analysis, only the runoff is considered as the source of reservoir water. 1.3 Need for this research Water is an essential natural resource, which plays an important role in supporting human life and ecological systems. Water is increasingly being viewed as a severely stressed resource. This important resource is likely to be affected by climate change conditions in a negative way at many locations (IPCC 2007, McFarlane et al. 2012) which will have direct impact on the ability of a water supply system to ensure adequate water supply to meet customer demands. Forecasting of long term water demand and catchment water yield are the critical factors of a water supply system to undertake adaptive management strategies to maintain a reliable supply of water to the communities/cities. A number of research studies have been conducted to forecast long term water demand (e.g. Babel et al 2007, Mohamed and Mualla 2010, Polebitski et al. 2010). However, despite the emerging concern of climate change issue, application of GCMs data (which is an important tool in climate change impact studies and which provides future climatic scenarios under different greenhouse gas emission conditions) in forecasting water demand is limited. As a result, a knowledge gap exists in linking GCM projection with water demand modelling in forecasting future water demand. Moreover, there is a lack of knowledge on the estimation of uncertainty in the water demand forecasting by accounting for the stochastic nature of the independent variables and their inter correlation. Quantification of uncertainties is an integral part in the climate change impact studies on catchment water yield as different types of uncertainties are generally associated with runoff projections. Several studies have investigated the different sources of uncertainties (e.g. choice of GCMs, hydrological models and emission scenarios) associated with the projected runoff in changing climatic regime (Kay et al. 2009, Chen et al. 2011, Teutschbein et al. 2011). However, uncertainty due to many University of Western Sydney Page 5

43 CHAPTER 1: Introduction realisations (arising from repetitive simulations for a given time step during downscaling of the GCM data to a catchment scale) of a GCM has not been given much attention in the literature. In addition, exploration of the performance of a water supply system under combined effect of uncertain climate, water demand and catchment water yield scenarios are limited. 1.4 Research questions Based on the research need identified in Section 1.3, this thesis seeks to answer the following research hypotheses/questions in relation to the assessment of climate change impact on water demand and supply: How can water restrictions on residential water consumption affect water demand? What are the most important independent variables affecting future residential water demand and how the uncertainties associated with these variables can be estimated? How do the long term future water demand predictions vary due to inherent uncertainties in the independent variables? How can climate change affect the water demand in future? How can the uncertainty in water balance model be ascertained for ungauged catchments? How is catchment water yield expected to be affected by changing climatic conditions in future? How can the uncertainty in water demand and yield under changing climatic conditions affect the performance of a water supply system? 1.5 Summary of research undertaken in this thesis This thesis investigates the impacts of climate change on future water demand by using the climate projections from a GCM and uncertainties in demand projections being estimated by developing a long term probabilistic water demand forecasting University of Western Sydney Page 6

44 CHAPTER 1: Introduction model considering stochastic nature of the independent variables and correlation structures. Moreover, realisation uncertainty is investigated along with other types of uncertainties during the forecasting of catchment water yield using climate projections from several GCMs. In addition future performance of a water supply system is examined by adopting the projected water demand and catchment yield scenarios along with their uncertainties. The research tasks undertaken in this thesis to answer the research questions presented in Section 1.4 are outlined below: Select a urban water supply system (with adequate data in terms of quantity and quality) from the state of New South Wales, Australia and collate metered water consumption, water price, water savings, water restrictions, rainfall, temperature, evaporation, runoff and relevant catchment characteristics data for the proposed research. Identify water savings due to the implementation of different levels of water restrictions in a water supply system. Develop a long term water demand forecasting model by (a) including climate variables and (b) adopting water savings as a continuous independent variable. Develop a probabilistic water demand forecasting model adopting Monte Carlo Simulation technique assuming a multivariate normal distribution for the independent variables to ascertain the degree of uncertainty associated with the water demand projections due to the stochastic nature of the independent variables. Identify the relative influence of climate variables and other independent variables on water demand in qualitative term. Estimate climate change impact on future water demand by (a) forecasting water demand incorporating future climatic scenarios from a GCM, and then (b) comparing the projected water demand under different future climate conditions with that of the selected reference period. University of Western Sydney Page 7

45 CHAPTER 1: Introduction Estimate calibrated parameters for water balance models for ungauged catchments using a regionalisation technique and assess the uncertainties associated with the calibration of a water balance model. Forecast runoff under future climate conditions using data from several GCMs and estimate uncertainties from different sources (i.e. GCM uncertainty, realisation uncertainty, rainfall-runoff model uncertainty and rainfall-runoff model parameter uncertainty) associated with the projected runoff. Assess the performance of a water supply system under changing climatic conditions using the projected water demand and water yield values in the future periods. The major outcomes of this research include a new modelling framework and a body of scientific knowledge that can be applied by water supply authorities to enhance the reliability and resilience of their water supply systems in future under changing climate. 1.6 Contributions to knowledge This research study has developed a modelling framework to investigate the climate change impacts on future water demand and water supply with their associated uncertainties. The major contributions made in this thesis to the knowledge are summarised below: 1. A method has been proposed to quantify the water savings in an effective way from the implementation of water restriction in a water supply system. 2. The numerical representation of the water savings variables in the water demand forecasting model has been formulated. 3. A probabilistic long term water demand forecasting model has been developed to account for the stochastic nature of the independent variables and the inter correlation structure of the variables. 4. The linking of GCMs projections with the water demand forecasting model has been established. University of Western Sydney Page 8

46 CHAPTER 1: Introduction 5. A modelling framework has been developed to estimate the uncertainties due to different types of uncertainties in the calibration of the rainfall-runoff models and in forecasting runoff. 6. An integrated methodology has been developed and applied to examine the performance of a water supply system under various climate, water demand and runoff scenarios. These methods along with the outcomes of the research would provide vital knowledge about the possible climate change impact on future water demand and runoff changes, and future performance of a water supply system for better planning and management of a water supply system. This will also help to develop appropriate adaptive strategies to supply necessary water to the communities. The methodologies developed in this thesis can be adopted to other region and to other water supply system in Australia and elsewhere in the world. Figure 1.1 Illustration of the major research tasks undertaken in this thesis University of Western Sydney Page 9

47 CHAPTER 1: Introduction 1.7 Outline of the thesis The research undertaken in this study is presented in this thesis consisting of nine chapters and four appendices, as outlined below: Chapter 1 presents a brief introduction to the proposed research, including the background and need for this research. The research questions that are to be investigated and the research tasks to be undertaken to answer the identified research questions and the research contributions are also presented in this chapter. Chapter 2 presents a literature review on water demand modelling and forecasting, variables of water demand, water restrictions, climate change issue in water demand, future climate scenarios, use of GCMs, downscaling methods, water balance models and impact of climate change on catchment water yield and water supply systems. Research studies on uncertainty estimation of water demand forecasting and catchment yield are also reviewed. Furthermore, review of various regionalisation methods to calibrate a water balance model for use in an ungauged catchment is presented to select an appropriate method to be adopted in this study. At the end, this chapter summarises the findings from the literature and identify gaps in the current state of the knowledge on assessing the climate change impact on water demand and supply, and formulate the research problems that are to be investigated in this thesis. Chapter 3 presents the selection of study area and data. The chapter starts with discussion of the selection of the study region, which is followed by a quantitative summary of the data used in this thesis. Chapter 4 presents the assessment of water savings for the implementation of water restrictions in the residential sector. This chapter commences with presenting the methodologies to quantify the water savings for different levels of imposed water restriction. This is followed by the development of water demand forecasting models adopting water savings as a continuous independent variable in the model along with other water demand variables. The chapter then presents the comparison of the developed models to select the reliable estimates of water savings. Chapter 5 presents the quantification of uncertainties in water demand projection due to stochastic nature of the independent variables. This chapter commences with University of Western Sydney Page 10

48 CHAPTER 1: Introduction presenting the methodology to forecast long term water demand adopting a Monte Carlo simulation technique based on multivariate normal distribution. This is followed by presenting the forecasted water demand results and the estimation of uncertainty band in the projection of water demand. Chapter 6 presents the principal component biplot technique to evaluate the relative influence of independent variables qualitatively on water demand. Then it presents the quantitative assessment of climate change impact on future water demand using three future emission scenarios from a GCM. This chapter also presents the impact of climate change on future water demand using three different hypothetical future climate scenarios. Chapter 7 presents the calibration and validation of two selected rainfall-runoff models (i.e. AWBM and SIMHYD) which are needed to estimate catchment yield. This covers the selection of proper rainfall time series data and rainfall factor, proper calibration and validation data length, and effective optimisation technique to obtain reliable calibrated parameter sets of the models to be used in ungauged catchment situations. This chapter also presents the estimation of uncertainty in the calibration of a rainfall-runoff model. Chapter 8 presents the forecasting of catchment yield under future climate conditions. This chapter commences with the estimation of future catchment yield incorporating the data from several GCMs using the two selected rainfall-runoff models. This is followed by quantifying the uncertainties in the projection of water yield. Afterwards, the chapter presents the assessment of a water supply system using the forecasted water demand and water yield scenarios in the future changing climate regime. Chapter 9 presents the summary and conclusions of the research undertaken in this thesis, and provides recommendations for further research. Appendix A presents some additional tables and figures from Chapter 4. Appendix B presents some additional tables and figures from Chapter 5. Appendix C presents some additional tables and figures from Chapter 7. Appendix D presents some additional tables and figures from Chapter 8. University of Western Sydney Page 11

49 CHAPTER 2: Literature review CHAPTER 2 REVIEW OF CLIMATE CHANGE IMPACT ANALYSIS ON WATER DEMAND AND YIELD ESTIMATION IN URBAN WATER SUPPLY SYSTEMS 2.1 Overview Chapter 1 has presented the background, motivation of the research, the research questions to be investigated, the research contributions and outline of the thesis. This chapter provides a review of the issues relevant to climate change impact analysis on urban water demand, catchment water yield and water supply systems. It also summarise the knowledge gaps in the existing literature. In the first part, this chapter provides a review of the studies related to the linkage of climate variables and climate change with urban water demand, impact of water restrictions on urban water demand and forecasting of long term urban water demand. In the second part, this provides a review of the uncertainties associated with the climate change impact analysis on future runoff estimates and the uncertainties associated with the calibration of a rainfall-runoff model. In the third part, this discusses the issues relevant to the reliability assessment of a water supply system considering uncertainties associated with both the water demand and yield in the context of changing climate. Finally, this chapter concludes by summarising the knowledge gaps in the existing literature and how this thesis can contribute to fill these gaps. 2.2 Climate change issues relevant to water security In a research on water security in global perspective, it has been reported that 80% of world s population is in vulnerable conditions in regards to receiving necessary water for their use and survival (Vörösmarty et al. 2010). Over extraction of groundwater, inadequate flow in the major river systems, increase in water demand due to growing population and rapid urbanisation, water pollution and economic development are placing unsustainable demands on fresh water resources at many locations (Postel 2000, Vörösmarty et al. 2000, Güneralp and Seto 2008, Beck and Bernauer 2011). University of Western Sydney Page 12

50 CHAPTER 2: Literature review Moreover, changing climate conditions are likely to exacerbate the existing pressures on water supplies and would exert increased impacts on water resources around the globe in negative ways (Bates et al. 2008, House-Peters and Chang 2011, Xiao-jun et al. 2013). The limited availability of fresh water resources for many urban cities around the world has become a crucial concern in recent years. Fourth assessment report of the Intergovernmental Panel for Climate Change (IPCC) states that alteration of water resources are happening around the world due to changing climate (Rosenzweig et al. 2007). Though a range of sectors are likely to be affected by water shortages including industry and agriculture, predictions by the IPCC has suggested that water demand and supply in residential sector would need more attention in the changing climate conditions (Bates et al. 2008). It has been predicted by several studies that frequency and severity of drought events are likely to be increased in future as a result of climate change (Gergis and Fowler 2009). Consequently, availability of water resources are expected to be declined more and demand for water is expected to be higher in future. Hence, a sound understanding and quantification of the impacts of climate change on urban water demand and supply are critical in order to maximize the efficiency of urban water demand management and to ensure adequate water supplies to the cities in future Linkage of climate change/variables with urban water demand Climate change is expected to have impacts on water demand as it is generally influenced by the climate variables such as temperature and rainfall. Rainfalls are likely to have an effect on water demand in outdoor activities, particularly watering garden. In an urban environment, rainfall regime determines how much and when water is required to the plants and lawns that have to be met by the supply water (Corbella and Sauri Pujol 2009). Temperature has also some effects on water demand; the rational is that during hot days more water is required in gardens, in swimming pools and for personal hygiene. Influence of the climate variables on water demand has been reported by a number of studies in different parts of the world. For example, Babel et al. (2007) found that rainfall was one of the significant demand variables to predict domestic water University of Western Sydney Page 13

51 CHAPTER 2: Literature review demand in Kathmandu, Nepal. They demonstrated that 10% increase in rainfall would lead to reduction of water use by about 2.1%. However, they reported that temperature had no effect on water demand in Kathmandu, Nepal. Gato et al. (2007) found that temperature and rainfall had a statistically significant correlation with residential water usage in Melbourne, Australia. In a review of the significant variables influencing domestic water demand, Corbella and Sauri Pujol (2009) found that climate variables (i.e. temperature and rainfall) were among the major drivers of domestic water demand. Praskievicz and Chang (2009) analysed the water consumption data with some climate variables, temperature, daylight length, precipitation, wind speed, relative humidity, and identified the variables that play a significant role in determining water consumption in Seoul, South Korea. In a study in identifying the determinants of residential water demand in Germany, Schleich and Hillenbrand (2009) found that rainfall had an effect on water consumption while temperature had no impact. In Bangkok, Thailand, Babel et al. (2011) demonstrated that climate variables (temperature, rainfall and relative humidity) had influence on medium term water demand (6 months lead time). The results of these studies indicate that climate change may have impacts on future water demand as climate change will affect climate variables such as temperature and rainfall which influence water demand. Hence, climate change has important implications for management of urban water resources under potential future climate change conditions. These results also highlight the necessity of identifying the impacts of climate change on future water demand to ensure water security in the future. 2.4 Climate change analysis/studies on future water demand Impact of climate change on future water demand can be identified in a number of steps: (i) develop water demand forecasting model based on the climate variables along with other influential water demand variables, (ii) input the future climate scenarios/probable future values of the climate variables to the developed forecasting model and (iii) compare the predicted future water demand with that of the selected reference period. Hence, incorporation of probable future climate scenarios is an University of Western Sydney Page 14

52 CHAPTER 2: Literature review integral part to identify the impacts of climate change on future water demand. Projections of future climate conditions can be made by using hypothetical scenarios (e.g. assume a reasonable change in the future climate conditions, for example assume an increase of 1 0 C in temperature and 10% decrease in rainfall amount in future time from the reference period). Another way of getting the future projections of climate scenarios is to use the climate prediction from global climate models (GCMs). GCMs are generally considered to be the most effective and reasonable tool to get the estimate of future climate scenarios and they are extensively used in estimating the impacts of climate change on future runoff/streamflow conditions (IPCC 2007, van Roosmalen et al. 2010, Andersson et al. 2011, Bastola et al. 2011). Very few studies (e.g. Babel et al. 2007, Khatri and Vairavamoorthy 2009) are available in the scientific literature that include future climate scenarios in the water demand forecasting model and identify the impacts of climate change on future water demand. In addition, incorporation of future climate scenarios using GCMs is quite limited in forecasting urban water demand. Hence, there is a gap in knowledge in regards to the use of future climate variables in forecasting future water demand based on GCM output and to identify the impacts of potential climate change conditions on the future water demand. Babel et al. (2007) included rainfall (mm/year) variable in their water demand forecasting model along with other three different water demand variables and forecasted water demand up to the year, However, future value of annual rainfall amount was taken to be a single value based on the historical data (i.e mm/year) and assumed to be constant during the forecasting period in their study. Khatri and Vairavamoorthy (2009) used precipitation and temperature data from the HadRM3 global climate model with four different emission scenarios (Low, Medium-Low, Medium-High, and High) to assess the sensitivity of the climate variables on future water demand. However, they could not identify and assess the impact of climate change on future water demand due to inadequate data. 2.5 Impact of water restrictions on urban water demand Water supply to large metropolitan cities has emerged as a challenge due to global climate change, and ever increasing water demand and size of the cities. Water shortage has become a common problem in many urban water supply schemes. University of Western Sydney Page 15

53 CHAPTER 2: Literature review Moreover, rapid population growth, economic expansion and changing climatic conditions have increased water demands in some urban areas beyond the capacity of local water supplies. Historically, urban water management strategies have relied on supply-side management, which increases the availability of water through the implementation of various expensive and engineering measures such as construction of new dams and desalination plants, expansion of existing water supply systems and sourcing new water supply catchments (Marvin et al. 1999, Galán et al. 2009). However, these options have some notable limitations such as high cost and wider environmental impacts. Moreover, sourcing of new water supply catchments may not be possible in many cases. In an effort to ensure water security in the urban areas, demand side management has emerged as an important strategy in recent years to bring a stressed water supply system into balance condition and as a complement of more traditional water supply side management (Arbués et al. 2003, Brooks 2006, Jeffrey and Gearey 2006). In urban water supply, residential water consumption is the major component of the total water demand (Wong et al. 2010). Hence, reduction in residential consumption is viewed as an important means to reduce the total water demand (Suero et al. 2012, Cahill and Lund 2013) and thereby to manage the shortages in water supply. The water demand management strategies that assist to reduce residential demand generally include implementation of water restrictions, incentive schemes, setting of water pricing policies and promoting water-efficient appliances. Since these programs have the potential to play an important role in reducing the vulnerability of fresh water supplies, their effectiveness in saving water are required to be evaluated to make an effective and efficient water demand management plan. Some water authorities rely on water restrictions to manage shortages in water supply during the periods of droughts (Kanta and Zechman 2014). For examples, due to recent prolonged droughts in Brisbane, Melbourne and Sydney, the three major cities in Australia, water authorities imposed water restrictions of varying severity to their customers to reduce water demand as the dam water storage levels dropped quite low (Queensland Water Commission 2010, Sydney Water 2010). Cooper et al. (2011) also mentioned that water restrictions remain the dominant demand management strategy to reduce urban water demand during drought periods in most University of Western Sydney Page 16

54 CHAPTER 2: Literature review urban cities across Australia. Some studies such as Brennan et al. (2007) and MacDonald et al. (2010) stated that due to the changing climate condition, which may result in increased frequency of droughts and associated water scarcity, water restrictions may become more frequent in the different parts of the world in future. Given the importance of water restrictions as a policy mechanism to restrict urban water consumption, quantification of water savings from water restrictions is necessary to evaluate the effectiveness of these programs. These water restrictions normally target outdoor residential water use and can be imposed in several forms. They are typically implemented by specifying the time of day for garden watering, the maximum length of the watering period and allowances for hand watering (MacDonald et al. 2010). Moreover, prohibition for using hoses to wash paved areas, limits on car washing and filling or refilling swimming pools are some of the common forms of these restrictions. For example, hosing of lawns and gardens were not permitted in any time during the Level 2 water restrictions in Sydney. Only handheld hosing was permitted before 10 am and after 4 pm on Wednesdays, Fridays and Sundays. The severity and timing of water restrictions are largely determined based on the dam storage levels. For example, in Sydney, Level 1, Level 2 and Level 3 water restrictions (Table 2.1) were imposed when the dam levels dropped below 55%, 50% and 40%, respectively (Spaninks 2010, Sydney Water 2014). Level 1 and Level 3 were the most liberal and most stringent water restrictions, respectively, in terms of imposing the restriction rules. Despite the fact that water authorities adopt water restrictions to manage urban water demand during drought periods, limited number of studies exist in the literature that systematically quantify the water savings derived from the imposed water restrictions. Such studies include Anderson et al. (1980), Moncur (1987), Shaw and Maidment (1988), Shaw et al. (1992), Renwick and Archibald (1998), Michelsen et al. (1999), Renwick and Green (2000), Kenney et al. (2004), Jacobs et al. (2007) and Kenney et al. (2008). Most of these studies, such as Anderson et al. (1980), Moncur (1987), Renwick and Archibald (1998), Michelsen et al. (1999), Renwick and Green (2000) and Kenney et al. (2008) used a binary variable to represent the water restrictions in their water demand models. The value of binary variable was considered as 1 (one) when the water restrictions were taken in place, otherwise its University of Western Sydney Page 17

55 CHAPTER 2: Literature review value was considered zero in the models. The estimated coefficients of the binary variables were used to anticipate the water savings due to water restrictions. For example, Anderson et al. (1980) estimated that the drop in water use in the City of Fort Collins, Colorado could be million litres per day (6.97 million gallons/ day) due to water restrictions as they found as a coefficient value of the binary variable in their multiple linear regression model. Kenney et al. (2008) found that water restrictions could reduce water demand by 31% in Aurora, Colorado, as they found as the coefficient of the binary variable for water restrictions in their Log-Log multiple regression model. Kenney et al. (2004) estimated water savings in the several cities in Colorado due to the water restrictions in the summer of 2002 by two approaches. In the first approach, water savings were calculated by Before and After Method (BAM) which compared daily water usage during the drought periods (May to August 2002) to the average daily usage over the same months (years 2000 to 2001). In the second approach, which can be termed as Expected Use Method (EUM), daily water use during the drought periods was compared to an estimate of the water use that would have occurred in the absence of any restrictions. However, their study period was relatively short (only four months). Jacobs et al. (2007) investigated the water savings in Cape Town from the water restrictions by the similar approach as of Kenney et al. (2004). Water savings were calculated by comparing meter readings of the summer periods of 1 October 2004 to 1 April 2005 with the period of 1 October 2003 to 1 April However, in this study the data period was also relatively short (i.e. six months). University of Western Sydney Page 18

56 CHAPTER 2: Literature review Table 2.1 Levels, scope and timing of water restrictions imposed in Sydney during the drought periods ( ) Restriction rules I. No hosing of hard surfaces and vehicles. Restriction rules belonging to restrictions levels Level 1 (I +II) Introduction date 1-Oct-03 II. No use of sprinklers or other watering systems. III. No hosing of lawns and gardens, only hand-held hosing was allowed for three days in a week (before 10 am and after 4 pm on Wednesdays, Fridays and Sundays). IV. No filling of new or renovated pools over 10,000 L except with a permit from Sydney Water. Level 2 (I + II + III + IV) Level 3 (I + II + III + IV + V +VI) 1-Jun-04 1-Jun-05 V. No hosing of lawns and gardens, only hand-held hosing was allowed for two days in a week (before 10 am and after 4 pm on Wednesdays and Sundays). VI. Fire hoses are allowed only for firefighting purpose and not for cleaning. 2.6 Urban water demand forecasting Water is generally considered to be the most vital resource in any urban development program (Nasseri et al. 2011). Most of the decisions in urban planning and development programs are highly dependent on the availability of water resources and the forecasting of future water demand. Moreover, estimation of future water demand is vital to the planning of water supply systems as it allows the water authorities to know the demand for long term periods in the future to develop new water sources and to extend the capacity of the existing systems. One of the main University of Western Sydney Page 19

57 CHAPTER 2: Literature review purposes of urban water demand forecasting is to supply necessary water to the communities corresponding to the demand and to keep the balance between supply and demand (Zhou et al. 2002). Accurate projection of future water demand plays an important role in optimum utilization of available water resources and in efficient allocation of water between competing users. In addition, as 25-30% of total operating costs is generally incurred from energy use, forecasting of water demand can help to optimise energy use, which is beneficial to both the environmental and economic sectors (Ghiassi et al. 2008, Herrera et al. 2010). Thus forecasting of water demand plays a crucial role in socially, economically and environmentally sustainable water resources planning and management Temporal scales/types of urban water demand forecasting Three types of temporal resolution of urban water demand forecasting are generally found in the literature based on their uses and differing modelling techniques. (a) Short term forecasting: Prediction resolution of this forecasting type generally varies from 1 day to several weeks (Billings and Agthe 1998, Gato et al. 2007, Ghiassi et al. 2008). Short term prediction of water demand is generally required for operation and management of existing water supply systems within a specified time period (Qi and Chang 2011). (b) Medium term forecasting: Monthly forecast of water demand with up to one year lead time are generally considered as medium term forecasting (Maidment and Parzen 1984, Nasseri et al. 2011), which is required for planning improvements to distribution and water supply systems and implementing technological changes. (c) Long term forecasting: The prediciton resolution of this type of forecasting is usually greater than one year, mostly annual and decadal (Tiwari and Adamowski 2013). Long term projection of water demand is mainly required for the development, planning and design of water supply systems and infrastructures (Jain and Ormsbee 2002, Ghiassi et al. 2008, Firat et al. 2009, Herrera et al. 2010). University of Western Sydney Page 20

58 CHAPTER 2: Literature review Deterministic vs. probabilistic water demand forecasting Long term adequacy of water supply is needed in order to ensure water requirements for the current and future population with the desired level of satisfaction, which is a major national concern in many countries. Hence, it is necessary to determine the current water demand and future water demand in order to assess the future adequacy of water supplies. In order to do this, suitable tools are needed to estimate the future water demands and to assess the effects of future climate and other factors on both water demand and water availability. Long term water demand can be forecasted by the deterministic and probabilistic models (Froukh 2001, Almutaz et al. 2012). Deterministic models generally forecast single value of water demand without considering the stochastic nature of the independent variables. As water demand depends on different independent variables which are stochastic in nature and are correlated among themselves such as population, household size, income, water usage price, climate conditions and conservation measures (Babel et al. 2011, Qi and Chang 2011), the usefulness of deterministic models in forecasting urban water demand may be limited. If the associated uncertainties in the independent variables are ignored, the forecasted water demands may not be realistic and adequate for efficient planning and management of water supply systems, as decisions based on deterministic (single-point) forecasts do not accommodate possible variations in demand. Therefore, uncertainties associated with the independent variables should be explicitly incorporated into demand forecasting models to allow decision makers to understand how uncertainties in the independent variables may affect the future water demand. Incorporation of such uncertainties can be achieved by developing a probabilistic water demand forecast model using a Monte Carlo simulation. Another important aspect of water demand forecasting is to account for the correlations among the independent variables as the independent variables are often correlated. In the literature, most of the long term water demand forecasting studies estimated future water demand by a deterministic approach (e.g. Babel et al. 2007, Mohamed and Mualla 2010). On the contrary, there has been a limited research on the probabilistic forecast of long term urban water demand. Examples include studies by Khatri and Vairavamoorthy (2009), and Almutaz et al. (2012) who adopted a University of Western Sydney Page 21

59 CHAPTER 2: Literature review probabilistic forecasting method; however, the correlations among the independent variables were not accounted for. 2.7 Climate change impact on water resources Water supply has emerged as a major issue in many counties in the world due to factors such as increasing population, rapid urbanisation and water pollution. This problem is intensified in many regions that are experiencing rapid changes in climate conditions (e.g. increase in temperature, decrease in rainfall, and frequent droughts and floods) (Parajuli 2010). For example, south-western Australia has experienced an increase in temperature of about 1 0 C through the 20 th Century and there has seen a reduction in rainfall by 16% since the mid 1970s, resulting in a reduction of about 50% in streamflow into major reservoirs (McFarlane et al. 2012, Silberstein et al. 2012). Moreover, the IPCC stated that changes in climate conditions would have noticeable impacts, mostly negative, on water resources due to plausible changes in precipitation, temperatures and evaporation in the future (IPCC 2007). Projections of changes in the water resources conditions (mainly runoff which is a measure of catchment yield) due to plausible climate change conditions have been reported for many countries around the world. For example, in the western USA, Thomson et al. (2005) demonstrated that water yield would reduce by around 50% under changing climate conditions where there are already shortages in the water supply. In southeast Australia, Chiew et al. (2009) projected that mean annual runoff would be changed by -17% to +7% resulting from a global warming of C. In southern Italy, D Agostino et al. (2010) projected a 16-23% reduction in streamflow by 2050 resulting from a reduction in rainfall of about 5 10%. In North-Algeria Jean-Pierre et al. (2010) demonstrated that 40% reduction in surface water resources would be possible if rainfall would reduce by 15%. In the Mono Lake basin, western United States, Ficklin et al. (2013) found that annual streamflow would be reduced by 15% by the end of 21 st century ( ), compared to average historical value ( ). In recent years, extreme events such as droughts and floods are occurring frequently worldwide. A small change in precipitation and temperature may lead to higher percentage change in runoff in arid and semiarid regions (Gan 2000). Changes in the University of Western Sydney Page 22

60 CHAPTER 2: Literature review available water resources conditions challenge water authorities to look for new water sources (e.g. sourcing new catchments and building desalination plants) and alternatives (e.g. water restrictions and water re-use) to meet increasing water demand. That is why, understanding of potential impacts of climate change on runoff/catchment water yield is crucial to deal with the effects of climate change. It will also help water managers to make more coherent decisions on water allocation and management. It is, therefore, important to formulate a new vision for the future water resources in a given catchment based on the investigation of climate change impacts on water resources availability Uncertainties in climate change impact analysis on catchment yield Impact analysis of climate change on future catchment yield generally involves use of different tools such as hydrological and global climate models in a number of steps as outlined below: Step 1: Calibration and validation of hydrological models using the historical climate and observed runoff data to obtain the calibrated model parameter set. Step 2: Generation of future climate scenarios using the global climate models (GCMs) and downscaling the climate scenarios to the regional/catchment scale using different downscaling techniques. Step 3: Estimation of future runoff quantity adopting the future climate scenarios and using the calibrated model parameter set. Step 4: Comparison of the future runoff estimates with the observed runoff in the adopted reference period to estimate the changes in the future runoff conditions. During the process of estimating the climate change impact on future runoff, different types of uncertainties are associated with the climate change impact studies due to choice of GCMs, greenhouse gas emission scenarios, downscaling methods and hydrological models (Kay et al. 2009, Chen et al. 2011, Teutschbein et al. 2011). It has been recognised by many studies (Graham et al. 2007, Jiang et al. 2007, University of Western Sydney Page 23

61 CHAPTER 2: Literature review Teutschbein and Seibert 2010) that uncertainties should be taken into account in the investigation of climate change impact on water resources in order to produce reliable estimates. Therefore, acknowledgement and proper quantification of uncertainties are crucial to facilitate decision making by policy makers. Brief descriptions of different types of uncertainties are provided in the following section Uncertainty due to GCM Global climate models (GCMs) have come to the light as the vital tool for estimating future responses to changes in atmospheric composition and land-surface properties (Gober et al. 2010). Usually GCMs are used to estimate global warming by predicting the impact of amplified CO 2 concentration on climate variables (Yu et al. 2002). GCMs, a type of numerical model, are used in long-term climate change experiments, which represent various earth systems including the atmosphere, oceans, land surface and sea-ice. To assess the impacts of future climate change, outputs from GCMs are the primary source of information (IPCC 2007, Chiew et al. 2009). Randall et al. (2007) described the confidence in GCM estimates from different points of view, such as fundamental of the models which are based on established physical principles, ability of climate models to simulate important aspects of the current climate and ability of models to reproduce features of past climates and climate change. Many modelling advances have been accomplished over the past several years to the GCMs and the development programs are continuing to improve the predictions of the GCMs. Twenty three GCMs are available globally to predict the climate change scenarios using a three dimensional grid over the globe (Randall et al. 2007). Model ID along with the calendar year ( vintage ) of the first publication of results from each model, the respective sponsoring institutions and the horizontal resolution of the models are presented in Table 2.2. As the models continue to develop and their resolution continues to improve, they are representing more physical and biophysical processes and interactions which are important for climate change, and thus they are becoming increasingly useful for investigating important climate features. However, despite their popularity and prevalent use in the climate change impact studies, choice of GCMs is considered to be one of the major source of uncertainty as the results predicted by different GCMs are quite different and they vary widely in their University of Western Sydney Page 24

62 CHAPTER 2: Literature review projections especially for precipitation (Wilby et al. 2006, Graham et al. 2007, Exbrayat et al. 2014). The uncertainty in GCMs projections normally comes from their course spatial resolution, model structure and parameterization, and the way GCMs respond to changes in atmospheric forcing. Table 2.2 List of Global Climate Models (GCMs), (Randall et al. 2007) SN Model Vintage Sponsor(s), Country Horizontal Resolution (~km) 1 BCC-CM1, Beijing Climate Center, China BCCR-BCM CCSM CCCMA-CGCM3.1 (T47) CCCMA-CGCM3.1 (T63) CNRM-CM CSIRO-MK ECHAM5/MPI-OM ECHO-G FGOALS-g GFDL-CM Bjerknes Centre for Climate Research, Norway National Center for Atmospheric Research, USA Canadian Centre for Climate Modelling and Analysis, Canada Canadian Centre for Climate Modelling and Analysis, Canada Météo-France/Centre National de Recherches Météorologiques, France Commonwealth Scientific and Industrial Research Organisation (CSIRO) Atmospheric Research, Australia Max Planck Institute for Meteorology, Germany Meteorological Institute of the University of Bonn, Meteorological Research, Institute of the Korea, Meteorological Administration (KMA), and Model and Data Group, Germany/Korea National Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics (LASG)/Institute of Atmospheric Physics, China U.S. Department of Commerce/National Oceanic and Atmospheric Administration (NOAA)/Geophysical Fluid Dynamics Laboratory (GFDL), USA University of Western Sydney Page 25

63 CHAPTER 2: Literature review Table 2.2 Global Climate Models (GCMs) (continued) SN Model 12 GFDL-CM GISS-AOM 2004 Vintage Sponsor(s), Country U.S. Department of Commerce/National Oceanic and Atmospheric Administration (NOAA)/Geophysical Fluid Dynamics Laboratory (GFDL), USA National Aeronautics and Space Administration (NASA)/Goddard Institute for Space Studies (GISS), USA Horizontal Resolution (~km) GISS-EH 2004 National Aeronautics and Space Administration (NASA)/Goddard Institute for Space Studies (GISS), USA GISS-ER 2004 NASA/GISS, USA INM-CM Institute for Numerical Mathematics, Russia IPSL-CM Institut Pierre Simon Laplace, France MIROC3.2(hires) 2004 Center for Climate System Research (University of Tokyo), National Institute for Environmental Studies, and Frontier Research Center for Global Change (JAMSTEC), Japan MIROC3.2(medres) 2004 Center for Climate System Research (University of Tokyo), National Institute for Environmental Studies, and Frontier Research Center for Global Change (JAMSTEC), Japan MRI-CGCM Meteorological Research Institute, Japan PCM UKMO-HadCM UKMO-HadGEM National Center for Atmospheric Research, USA Hadley Centre for Climate Prediction and Research/Met. Office, UK Hadley Centre for Climate Prediction and Research/Met. Office, UK Several studies have investigated the uncertainty due to the choice of GCMs in the climate change impact studies on water resources. Wilby et al. (2006) investigated the climate change impact in the River Kennet, UK using three different GCMs (HadCM3, CGCM2, and CSIRO Mk2) and found large variations in the projected future river flows driven by these GCMs. Prudhomme and Davies (2009) used the University of Western Sydney Page 26

64 CHAPTER 2: Literature review same GCMs as Wilby et al. (2006) to investigate the uncertainties in the river flows in four British catchments and concluded that GCMs were the main contributors to monthly mean flow uncertainty among others. GCMs uncertainty was found to be the most dominant uncertainty among the other sources of uncertainty in climate change impact analysis by several recent studies (e.g. Kay et al. 2009, Chen et al. 2011, Gosling et al. 2011). In an investigation on climate change impact on runoff across southeast Australia by 15 different GCMs, Teng et al. (2012) found that 28 to 35% variation in the future runoff estimates occurred due to the use of these different GCMs Downscaling uncertainty Many climate change impact studies require future climate information at scales of 50 km or less. To achieve this, appropriate techniques for downscaling GCM data to smaller-scale are needed, which then can be used to predict the climate conditions of the locality of interest. Due to the spatial resolution limitation of the GCMs, many downscaling methods have been developed to study regional and local-scale climate change. Two principal approaches for the downscaling of large-scale GCM output to a finer spatial resolution are: (a) Dynamical downscaling (b) Statistical downscaling The dynamical downscaling approach uses RCMs (Regional Climate Models) where a higher resolution climate model embedded within a GCM. Statistical downscaling approaches use statistical methods to establish empirical relationships between GCM-resolution and local climate variables (Fowler et al. 2007). Several techniques have been developed for statistical downscaling (i.e. linear regression, weather types, and stochastic weather generators). These downscaling approaches can be applied in different ways as illustrated in Figure 2.1. A source of uncertainty is associated with the choice of the downscaling methods by which global scale climate output are transferred to regional scale climate outputs. Uncertainty associated with the different downscaling methods has been reported by several studies. In an assessment of climate change impacts on alpine discharge regimes, Horton et al. (2006) used climate projections from 19 RCMs driven by three University of Western Sydney Page 27

65 CHAPTER 2: Literature review different GCMs and concluded that downscaling uncertainty could be comparable to the GCM uncertainty. In a comparative study on modelled future runoff in eight Australian catchments using five different downscaling methods, Chiew et al. (2010) demonstrated that differences in the modelled runoff could be significant owing to different downscaling methods. In a hydrological impact study in response to climate change in Denmark, van Roosmalen et al. (2010) investigated the variability in the modelled results using different dynamical downscaling models and found notable uncertainty in the modelling results due to the climate outputs from different downscaling models. Several recent studies also reported the same that downscaling uncertainty could be as high as that of the GCMs in climate change analysis on runoff estimates (Chen et al. 2011, Teutschbein et al. 2011). Global Climate Models (GCMs) Dynamical Downscaling Regional Climate Models (RCMs) Statistical Downscaling Change Factors Regression Methods Weather Classification Stochastic Weather Generators Climate Outputs Figure 2.1 Different pathways of downscaling of GCM outputs (Fowler et al. 2007) Emission scenario uncertainty The IPCC has developed a set of long term greenhouse gas emission scenarios, and the Special Report on Emission Scenarios (SRES), to predict changes in future climate conditions. Based on the possible future development and growth in demographic, economic and technological sectors (Table 2.3), four sets of scenario storylines, namely A1, A2, B1 and B2 were generated (Parry et al. 2004, Chowdhury and Al-Zahrani 2013 ). The scenarios are ordered as A1 > A2 > B2 > B1 based on the levels of CO 2 emissions in the atmosphere where A1 represent the maximum University of Western Sydney Page 28

66 CHAPTER 2: Literature review (2189 GtC in 2100) and B1 represent the minimum (983 GtC in 2100) levels of CO 2 emissions, respectively. The B2 and A2 scenarios are generally considered to be the most likely scenarios in the future by the IPCC models as these can indicate the reasonable lower and upper limits of CO 2 emission in the atmosphere ( GtC in 2100) (Nakićenović et al. 2000). Based on the technological use, the A1 scenario has been divided into three more subdivision, A1F1 (fossil intensive), A1T (non-fossil intensive) and A1B (balance across all sectors). Under these different emission scenarios future global temperatures are expected to increase in between 1 0 C and 5 0 C in 2100 in comparison to the temperature of the year 1990 (Arnell 2004). Due to the differences in the projection of future emission scenarios, greenhouse gas emission scenarios are considered to be another source of uncertainty in climate change impact studies. However, the effects of this kind of uncertainty in the estimated results were found to be much smaller than that of GCM and downscaling methods by several studies. Wilby and Harris (2006) found that uncertainty owing to emission scenarios contributed least amount in the total uncertainty and they suggested that uncertainty might be ranked in the decreasing order as follows: GCMs > downscaling methods > hydrological model structure > hydrological model parameters > emission scenario. Prudhomme and Davies (2009) used two emission scenarios (A2 and B2) in their studies and demonstrated that uncertainty owing to emission scenarios was smaller than that of GCMs. Chen et al. (2011) also found that uncertainty owing to emission scenarios was less than that of GCM and downscaling methods Realisation uncertainty Besides the uncertainty owing to the choice of different GCMs in climate change impact studies, another source of uncertainty can affect the estimated results, which is due to the variability within a GCM output. The variability in the outputs from a single GCM (within-gcm) can be occurred due to multiple runs of that GCM under a emission scenario that produce slightly different but equally plausible outcome from each run. This kind of uncertainty can be called as realisation uncertainty since the climate projects (e.g. NARCliM 2014 (Evans et al. 2014)) generally name the different outputs of a GCM as realisation. Despite the importance of this type of University of Western Sydney Page 29

67 CHAPTER 2: Literature review uncertainty as highlighted by several studies (Tebaldi and Knutti 2007, Hawkins and Sutton 2011, Deser et al. 2012) in climate change impact assessment, very little attention has been given to quantify this uncertainty in future runoff estimates. One recent study (Peel et al. 2014) has investigated the impact of this type of uncertainty on future runoff estimates under changing climate conditions and found that within- GCM uncertainty can affect the estimated results by about 10% around the mean value. Table 2.3 Future growth patterns of population, economy and technology owing to four sets of scenario storylines (A1, A2, B1 and B2) (Nakićenović et al. 2000) Criteria A1 A2 B1 B2 Population growth low high low medium Economic growth very high medium high medium Energy use very high high high medium Rate of changes in technology Environmental awareness rapid slow medium medium low varied by region high high Scale global local/regional global local/regional Hydrological model uncertainty Hydrological/rainfall-runoff models are an important tool to assess the impacts of future climate change scenarios on water resources by which future runoff and climate change impact can be estimated through inputting the future climate scenarios into the hydrological models. A number of hydrological models are available and have been used in climate change impact studies. However, uncertainty owing to the choice of hydrological models can put some uncertainty in the climate change impact assessment on runoff estimates as different models have different inherent assumptions and they vary in their parameters. Uncertainty owing to the choice of hydrological models has been reported by several studies and they found that potential uncertainty would present in the future runoff estimates. For example, Jiang et al. (2007) compared the hydrological impacts of University of Western Sydney Page 30

68 CHAPTER 2: Literature review climate change simulated by six hydrological models in the Dongjiang Basin, South China and found large differences in the model results. Two recent studies, Gosling et al. (2011) and Haddeland et al. (2011) demonstrated that differences in results obtained by different hydrological models could be substantial for the same GCM data and that the choice of hydrological models could be a major contributor to uncertainly. Hagemann et al. (2013) compared eight hydrological models to assess the hydrological response due to climate change and demonstrated that uncertainties linked to hydrological models could be comparable to the GCMs uncertainty in many regions Climate change impact analysis on ungauged catchment Rainfall-runoff modelling plays an important role in many areas of hydrology including estimation of design floods, analysis of catchment yield and evaluation of the impacts of land use changes on water resources. Rainfall-runoff models are also used in assessing climate change impacts on water resources (Yilmaz et al. 2011, Islam et al. 2014). A rainfall-runoff model needs to be calibrated and validated using the observed climate and runoff data; however, in ungauged catchments, the calibration and validation cannot be undertaken directly due to unavailability of some or all of these data. Researchers in many countries attempted to develop rainfallrunoff models for ungauged catchments but with limited success (Boughton 2009). A number of initiatives including Prediction in Ungauged Basins (Sivapalan et al. 2003) and the Model Parameter Estimation Experiment (Duan et al. 2006) coordinated multi-national efforts to enhance the accuracy of runoff prediction in ungauged catchments. Generally, regional relationships are used to estimate the parameters of a rainfall runoff model for application in an ungauged catchment. Mainly two regionalisation principles are reported in the scientific literature for this purpose (Merz et al. 2006): (i) (ii) calibrate the hydrological model in the nearby gauged catchments and transpose the model parameters to the ungauged catchment; and derive relationship between the model parameters and catchment attributes based on gauged catchments and use these relationship to predict model parameters at the ungauged catchment. University of Western Sydney Page 31

69 CHAPTER 2: Literature review Estimation of runoff with a reasonable accuracy in an ungauged catchment is regarded as a challenging task as notable uncertainties are involved in the regionalisation technique (Sivapalan 2003, Goswami et al. 2007). In order to transpose the optimized parameter sets from the geographically nearest gauged catchments to the ungauged ones, the rainfall-runoff models need to be calibrated using the data from the gauged catchments. Different sources of uncertainties are associated with the model parameter estimation during the calibration of a rainfallrunoff model such as selection of appropriate input data for the gauged catchments, quality of observed data against which the model is calibrated, choice of calibration data length, selection of calibration technique and objective function. These uncertainties need to be quantified to assess the relative accuracy of the prediction made by a rainfall-runoff model (Beven 2006) Uncertainty owing to input data Uncertainty in the input data to a rainfall-runoff model is mainly associated with the measurement error, and spatial and temporal sampling error. Rainfall and evaporation data are the two most important inputs to the rainfall-runoff models. As rainfall is the most important driving variable in the rainfall-runoff modelling, the uncertainty in the input rainfall data is considered to be one of the most prevalent sources of uncertainties in the rainfall-runoff modelling. Evaporation has a much smaller spatial and temporal variability than rainfall and hence rainfall-runoff modelling results are likely to be less influenced by the errors in evaporation data compared with rainfall data (Paturel et al. 1995, Andréassian et al. 2004, Boughton 2006). Chapman (2003) found that monthly average evaporation can be used as a replacement for daily evaporation in the case of missing data without any significant loss of accuracy in the outcomes of a rainfall-runoff model. Measurement error in rainfall generally occurs due to faulty instruments. Another uncertainty in rainfall data is associated with the limited spatial and temporal representation of rainfall due to insufficient density of rain gauges. In most of the cases rainfalls are measured at discrete intervals in time and at a limited number of points but rainfall are highly variable in both space and time. Therefore, selection of rainfall data can have a large impact on the calibration of a rainfall-runoff model. Oudin et al. (2006) investigated the effects of rainfall data error on streamflow University of Western Sydney Page 32

70 CHAPTER 2: Literature review estimation in twelve watersheds in the United States using GR4J and TOPMODEL and found that the rainfall data error could introduce 35% to 50% uncertainty in the predicted runoff. Yoo et al. (2012) simulated seven rainfall events considering the error in the actual rainfall events and found that these rainfall events could lead to uncertainty in the Clark instantaneous unit hydrograph model parameters by about 30% in the Chugnju Dam Basin, Korea. The results of these studies indicate that error in the input rainfall data can introduce a notable degree of uncertainty in the outputs of a rainfall-runoff model Uncertainty owing to observed gauged/output data Another problem with the calibration of a rainfall-runoff model is the presence of error in the runoff data. In many cases, runoff data in Australia is not measured directly rather it is estimated from a rating curve during the event of large storms due to practical difficulties including risk of life, high cost and access to gauging stations. A rating curve is constructed in most cases by correlating measured discharges with the corresponding observed stages at a particular gauged station (Petersen-Øverleir and Reitan 2009, Haddad et al 2010). However, due to many ranges of extrapolation involved during the construction of rating curve, the discharges that are estimated by rating curve are subject to high degree of uncertainty specially during large flows (Kuczera 1996, Pappenberger et al 2006, Di Baldassarre and Montanari, 2009). Nevertheless, runoff data from the gauged catchments are assumed to be of good quality in most of the rainfall-runoff studies Uncertainty owing to choice of optimization technique A rainfall-runoff model estimates runoff by simulating the physical processes in a catchment that represent the movement of water over the surface and through the soil during or after a rainfall event. A large number of parameters are generally associated with a rainfall-runoff model which are conceptual in nature and cannot be measured directly (Kim and Lee 2014). They are estimated through a calibration procedure which involves matching simulated runoff values with the corresponding observed values as closely as possible. Consequently, the calibration identifies an optimum parameter set of a rainfall-runoff model by minimizing the deviations between the observed and simulated runoff values. University of Western Sydney Page 33

71 CHAPTER 2: Literature review Over the last couple of decades, notable research has been carried out to develop reasonable and effective calibration methods based on optimization techniques (Yu and Yang 2000, Vrugt et al 2003). Several studies have been conducted to compare the relative performance of different optimization techniques for the calibration of rainfall-runoff models. For example, Gan and Biftu (1996) assessed the performance of three optimization techniques: (i) the shuffle complex evolution method; (ii) the multiple start simplex and (iii) the local simplex during the calibration of four rainfall-runoff models in the eight catchments selected from different parts of the world. They found that even though the performances of three different optimization techniques were comparable with each other; they produced different parameter sets for the same catchment. Franchini and Galeati (1997) compared two optimization techniques in the calibration of a rainfall-runoff model and found that the pattern search optimizer performed slightly better than the genetic algorithm optimizer. Madsen et al. (2002) compared three different optimization techniques during the calibration of the NAM rainfall-runoff model in a Danish catchment and found that all the methods produced comparable results. Due to the variable performance of the optimization techniques, an uncertainty analysis is very much needed during the calibration of a rainfall-runoff model by adopting a number of different optimization techniques Uncertainty owing to choice of calibration and validation data length Length of calibration and validation data set is another source of uncertainty in the calibration of a rainfall-runoff model. Generally, model users tend to use longer periods of data for model calibration to achieve a good calibration result. However, many researchers have demonstrated that longer periods of calibration data may not necessarily produce better calibration result and suggested different lengths of calibration data set to produce an optimum parameter set for different rainfall-runoff models and study regions. For example, Yapo et al. (1996) demonstrated that approximately eight years of calibration data was adequate to obtain an optimum parameter set for NWSRFS- SMA conceptual rainfall-runoff flood forecasting model. They noted that the effect of using more than eight years of calibration data could not improve the modelling results. Lidén et al. (2001) applied the HBV-SED model for a Zimbabwean basin and University of Western Sydney Page 34

72 CHAPTER 2: Literature review found that the model could give satisfactory results even using a single year s of data in the calibration. Xia et al. (2004) tested the Chameleon Surface model to a Russian basin and found that at least three years of calibration data was necessary to obtain a parameter set that was independent of the selected period. However, some studies reported that a shorter period of calibration data length may produce erroneous model results. For example, Boughton (2007) demonstrated by using the Australian Water Balance Model (AWBM) in the Snowy Creek catchment in Victoria, Australia that use of short periods of data (2 to 5 years) in the calibration could produce -21% to 30% error in the runoff estimation. From these studies, it is difficult to generalize the minimum data length required for model calibration or to identify the maximum length beyond which no improvement can be achieved in the modelling output. Therefore, it is important to identify the uncertainty in the model parameters due to the variability in calibration data length before using the parameter set to the ungauged catchments or climate impact studies. 2.8 Climate and demand uncertainty on yield of urban water supply systems Yield of an urban water supply system is generally defined as the maximum volume of water that can be adequately supplied to the communities/cities from the system over a given period. Yield is generally subject to several factors, such as, climate change, operating rules, demand pattern and adopted level of service. It is a key indicator of the performance of a water supply system which plays important roles in water supply system management, policy development and enforcement, expansion studies and decision making strategies. Water supply systems and its yields (potable water) are increasingly being viewed as vital resources throughout Australia and the rest of the world. Enormous pressures in supplying adequate water to the communities have been experienced by many water supply systems throughout the world due to changing climate and increasing population conditions (i.e. increase in water demand), sometimes being required to supply water close to or exceeding its sustainable yield level (Queensland Water Commission 2010, Sydney Water 2010). Most Australian urban water systems have been gone through such pressures during the recent droughts in the decade of , which resulted in the enforcement of record water restriction periods and permanent water savings measures to some of the urban cities (Queensland Water Commission 2010, Sydney Water 2010). Hence University of Western Sydney Page 35

73 CHAPTER 2: Literature review it is crucial to estimate the effects of changing climate and demand scenarios on the future yield of a water supply system to facilitate in the processes, practices, management and operation of urban water supply systems in an efficient and reliable way. Analysing the impacts of climate change on runoff is increasingly well recognised (Gosling et al. 2011, Teng et al. 2012). However, climate change impact analysis on urban water supply system has not been given much attention in the literature. In addition, translating the impacts of climate change on runoff cannot be used as a substitute of the quantification of potential impacts of climate change on water supply system, as a water supply system is a complex system whose performance/reliability in supplying adequate water to the community depends on many factors, specially on the balance between inflow (i.e. catchment yield) and outflow (i.e. water demand). Climate change is expected to affect not only the catchment yield but also the water demand pattern. Hence, incorporation of demand uncertainty due to changing climate and population growth along with catchment yield uncertainty are vital factors in identifying the reliability/performance of a water supply system to supply adequate water in the future. Few investigations have been done on the impact analysis of climate change on water supply system, for example, in Japan (Islam et al. 2005), in USA (Wiley and Palmer 2008), and in Australia (Paton et al. 2013). However, most of them concentrated on the identification of the probable impacts of climate change on the water supply systems due to several GCMs and greenhouse gas emission scenarios while ignoring the water demand uncertainty. For example, Wiley and Palmer (2008) investigated the climate change impact on a municipal water supply system in Puget Sound Region in the U.S using projections from several GCMs and a fixed water demand condition. But, water demand is one of the key factors that determine the reliability of the performance of a water supply system. By ignoring the effects of demand uncertainty due to potential changes in future climate and population conditions on water supply system, the reliability estimation of a water supply system under changing climatic conditions may not give accurate results. These results would negatively affect the water planning and management decisions and may pose threat in providing adequate water supply to the community in the University of Western Sydney Page 36

74 CHAPTER 2: Literature review future. In a recent study (i.e. Paton et al. 2013) of uncertainty assessment of a water supply system under changing future climatic conditions in Adelaide, Australia, the authors demonstrated that the demand uncertainty could play a major role in determining the reliability of a water supply system. However, this study considered a scenario based approach where six demand projections were considered (very low, low, medium low, medium high, high and very high) during the assessment of the magnitudes of the uncertainties of a water supply system. Hence there is a lack of knowledge in the field of estimating future performance of a water supply system by incorporating both the future catchment yield and demand uncertainties. 2.9 Summary This chapter has discussed the issues of climate change impact analysis on urban water demand, catchment water yield and their associated uncertainties in estimating yield of a water supply system. It has been found that despite the growing concern about future climate change and the affects of climate variables on water demand, very little attention has been given in finding the impacts of climate change on future urban water demand. In addition, incorporation of future climate scenarios through the adoption of global climate models has largely been unexplored. Hence a knowledge gap exists in finding the climate change impact on future urban water demand and in linking GCM projection with the water demand forecasting model. Due to the high importance of estimating long term urban water demand to plan for future water supply systems, more focus is needed to estimate the long term urban water demand in probabilistic way to consider the possible scenarios in designing, planning and management of future water resources. It has been found that most of the earlier studies have estimated long term water demand in a deterministic way (i.e. single forecast of future water demand). Few studies have estimated the urban demand in a probabilistic way, and correlation structures of the input variables were not considered by those studies, which is an important limitation as variables in urban water demand are often correlated with each other. Therefore, a methodology needs to be developed to estimate future water demand by considering the stochastic nature of the independent variables as well as their correlation structures. University of Western Sydney Page 37

75 CHAPTER 2: Literature review To overcome these knowledge gaps, this thesis estimates the climate change impact on future water demand by linking GCM projections with the water demand forecasting model. In addition, this thesis presents a methodology to estimate future water demand scenarios in a probabilistic way by considering the stochastic nature and inter correlation of the independent variables. It has been found that use of water restrictions has emerged as an important tool to manage shortages in water supply during drought periods. Some studies have also mentioned that use of water restrictions would be more frequent in future as frequency and severity of drought periods are expected to increase in future. Hence, there is a need for evaluating the effectiveness of the water restriction programs to manage water demand. There are limited studies in this regard in the literature. Another issue present in regards to inclusion of water restrictions and other water conservation programs in the water demand forecast modelling as continuous independent variables. Most of the studies incorporate the water saving programs as binary variables. From the reviewed literature, it has been found that till date no study has investigated the inclusion of water savings as continuous independent variables in the water demand forecasting model. The inclusion of the water saving variables offers distinct advantage over binary variables as in the future conditions quantity of water savings from each program/number of household participants of each program can be put in the forecasting model instead of considering only the presence or absence of a given water saving programs. Therefore, the inclusion of water savings variables as continuous independent variable in the water demand forecasting model needs to be investigated. Hence, this thesis investigates the effectiveness of incorporating water savings programs in the water demand forecasting model by taking them as continuous independent variables. In climate change studies on runoff/catchment yield estimates, it has been found that a number of uncertainties (e.g. GCM uncertainty, downscaling uncertainty, emission scenario uncertainty, realisation uncertainty and hydrological model uncertainty) are associated with the future runoff projections under changing climate conditions. Several studies have been conducted on estimating uncertainty in the climate change impact studies on runoff estimates with the major focus on GCM, downscaling, hydrological model and scenarios uncertainty. In spite of being a potential source of University of Western Sydney Page 38

76 CHAPTER 2: Literature review uncertainty, the realisation uncertainty (within-gcm uncertainty) has received a very little attention in the literature and has largely been unexplored. Hence this thesis investigates and quantifies the impacts of this uncertainty during estimating the future runoff under changing climatic conditions. It has been also found that a number of uncertainties (selection of proper input data, quality of observed output data, choice of optimization algorithms and selection of appropriate calibration and validation data lengths) are associated with the calibration of a rainfall-runoff model in order to be used in ungauged catchments to estimate future runoff. Several studies have been investigated these issues separately and the studies on systematic evaluation of these uncertainties are limited. Hence this thesis contributes to the existing literature by evaluating these sources of uncertainty in an integrated way during the calibration of a rainfall-runoff model. In regards to assessing the impacts of climate change on water supply systems, it has been found that most studies have considered only the climate change impact on water yield while ignoring the impact of climate change on urban water demand and uncertainty in the demand estimates. Despite the importance of considering future water demand scenarios under plausible changing climate conditions, scant attention has been given to the combined contribution of future water demand and yield scenarios in evaluating the reliability of a water supply system in supplying adequate water to the community in the context of changing climate. Hence, this thesis incorporates both water demand and yield scenarios under changing climate conditions in assessing the reliability of a water supply system. University of Western Sydney Page 39

77 CHAPTER 3: Study area and Data STUDY AREA AND DATA CHAPTER Overview Chapter 2 has discussed the issues of climate change impact studies on urban water demand, catchment water yield and water supply systems, and summarised the knowledge gaps in the existing literature. This chapter describes the study area and data used for climate change impact analysis on water demand and supply. In particular, this chapter discusses the selection of the study region and an urban water supply system, the importance of the study area and the data collation undertaken in this study. 3.2 Case study area and its importance The Blue Mountains region has been selected as the case study area (Figure 3.1). It is a mountainous region in the state of New South Wales (NSW), Australia located in west of Sydney. It has latitude of 33.7 S and a longitude of E. In the Blue Mountains region, residents get water from the Blue Mountains Water Supply System (BMWSS). The BMWSS provides water to about 48,000 people residing between Faulconbridge and Mount Victoria (Figure 3.2) through its two demand zones, upper (Mount Victoria to Leura) and middle (Wentworth Falls to Faulconbridge) Blue Mountains. The BMWSS consists of three major water sources: (i) Blue Mountains dams at Katoomba and Blackheath, (ii) The Fish river water scheme, originates in Oberon (Miller 2012), and (iii) Warragamba dam. In the case of emergency the BMWSS is supplemented with additional water from the Warragamba dam up to the area of Wentworth Falls (Figure 3.3). However, beyond the Wentworth Falls, the residents in the upper Blue Mountains region (Mount Victoria to Leura) solely depend on the BMWSS to get supply water. During the recent drought ( ) in Sydney, the risks of relying on the storages University of Western Sydney Page 40

78 CHAPTER 3: Study area and Data of the Blue Mountains dams in the BMWSS were demonstrated, as storage levels became critical to supply water (Sydney Catchment Authority 2009a). During that period, the NSW government had to implement water restriction rules in the greater Sydney to reduce the demand for water in order to manage the shortages in the water supply. The same restriction rules were also applied to the Blue Mountains region. Hence, the need for forecasting long term water demand, estimating future catchment water yield and assessing the performance of the BMWSS with associated uncertainties under plausible changing climate conditions is deemed to be very high. This necessary forecasting and assessment will enable the decision makers and the authorities to take appropriate management decisions and adaption strategies to supply adequate water to the communities. Figure 3.1 Location of the Blue Mountains region in the New South Wales, Australia 3.3 Catchments and dams in the BMWSS There are total six dams (Figure 3.4) in the Blue Mountains regions. These dams receive water from the three small catchments (Figure 3.5) namely, Katoomba, Woodford and Blackheath. Among the six dams, three dams are located in the University of Western Sydney Page 41

79 CHAPTER 3: Study area and Data Katoomba catchment, namely, Lower, Middle and Upper Cascade dams on Cascade Creek. Greaves Creek dam on Greaves Creek and Lake Medlow dam on Adams Creek are located in the Blackheath catchment. Woodford dam at the junction of Bulls Creek and Woodford Creek is located in Woodford catchment (Sydney Catchment Authority 2014a). Woodford dam is currently decommissioned which is not used to supply water. Katoomba and Blackheath catchments are ungauged catchments with an area of about 2.81 km 2 and 7.32 km 2, respectively. The nearest gauged catchment, Narrow Neck catchment (26 km 2 ), is located in the Megalong area, which is around 6.2 km away from Katoomba and 10 km away from the Blackheath catchment. 3.4 Climate conditions of the study area The climate of the Blue Mountains region is normally moderate compared with the lower Sydney region. As Mount Victoria is over 1,000 meters above sea level, the temperature is generally 7 0 C lower on average than the coastal Sydney. Monthly mean maximum temperature and monthly total rainfall of the study area (Katoomba weather station) for the period of 1997 to 2011 are presented in Table 3.1 and Figure 3.6, respectively. The monthly mean maximum temperature in the Blue Mountains area is found to be around 11 0 C and 23 0 C in winter (June to August) and summer months (December to February), respectively. The annual average rainfall in the Blue Mountains area is found to be around 1320 mm per year. Table 3.1 Monthly mean maximum temperature of the Blue Mountains region for the period of Month Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Mean maximum temperature University of Western Sydney Page 42

80 CHAPTER 3: Study area and Data Figure 3.2 Water supply zone (Mt Victoria to Faulconbridge) of the Blue Mountains water supply system (City of Blue Mountains 2007) University of Western Sydney Page 43

81 Water can be supplied up to Wentworth Falls from Warragamba dam Climate change impact on water demand and supply CHAPTER 3: Study area and Data Oberon Dam Upper Blue Mountains (Mt Victoria to Leura) supply zone Lower Cascade Dam Greaves Creek Dam Middle Cascade Dam Upper Cascade Dam Medlow Dam Boundary limit of Warragamba dam supply Middle Blue Mountains (Wentworth Falls to Faulconbridge) supply zone Woodford Dam Penrith supply zone Orchard Hills water filtration pump Legend Dam Pumping station Warragamba Dam River channel Raw water pipeline Treated water pipeline Figure 3.3 Blue Mountains water supply system (Sydney Catchment Authority 2009a) Figure 3.4 Location maps of the Blue Mountains dams in the New South Wales, Australia (Sydney Catchment Authority 2014b) University of Western Sydney Page 44

82 Annual rainfall (mm) Climate change impact on water demand and supply CHAPTER 3: Study area and Data Figure 3.5 Location maps of the study catchments (Katoomba, Blackheath and Narrow Neck catchments) (NSW Office of Water 2014) Year Figure 3.6 Annual rainfall of the Blue Mountains region for the period of (red line represents annual average rainfall) University of Western Sydney Page 45

83 CHAPTER 3: Study area and Data 3.5 Water conservation programs and water restrictions Sydney Water has initiated some water conservation programs in greater Sydney including the Blue Mountains region to encourage the residents to use the water efficient appliances in their houses to save water, in other words, to reduce the demand for water. These programs have been introduced at the beginning of the year, These programs mainly consist of the following five programs: (i) (ii) (iii) (iv) (v) WaterFix (installation of new showerheads, flow restrictors and minor leak repairs undertaken by a licensed plumber); DIY (Do-It-Yourself) kits (self-installed flow restrictors); Replacement of washing machines with more water efficient one; Installation of water efficient toilet flushes; and Installation of a rainwater tank. Voluntary water restrictions were introduced by NSW Government in October 2002 when Sydney s reservoirs were at a combined capacity of 67.4% (Sydney Water 2014). After voluntary water restrictions, NSW Government imposed mandatory Level 1 water restrictions on 1 October 2003 when dam levels dropped below 55%. A penalty of $220 was set for any breach of restriction rules. More rigid Level 2 restrictions were introduced on 1 June 2004 when the dam levels dropped below 50%. At Level 2, the restrictions on watering garden became more restricted and the days on which watering can be allowed were reduced. On 1 June 2005, more stringent Level 3 restrictions were applied when the dam levels dropped further below 40% (Sydney Water 2014). In June 2009, Level 3 restrictions were lifted as dam storage levels had improved to around 60%. Restriction rules of these different levels of water restriction that was imposed in the Blue Mountains region can be found in Table 2.1 (Chapter 2). 3.6 Data collection and future projections of the variables Historical water demand in the BMWSS Monthly metered water consumption data of the Blue Mountains region (Mount Vitoria to Faulconbridge) were obtained from Sydney Water for the period of January 1997 to September 2011 for the BMWSS. It has been found that around 80% of total water is used by residential sector and the remaining 20% is consumed by University of Western Sydney Page 46

84 Total yearly consumption (kl) Climate change impact on water demand and supply CHAPTER 3: Study area and Data non-residential (commercial and industrial) sector (Figure 3.7). It has also been found that the single dwelling sector (i.e. free standing houses/semi-detached houses) is responsible for about 94% of the total residential water consumption, while the multiple dwelling sector (i.e. apartment blocks/units) for the remaining 6%. Multiple dwelling sector consumption 5% Non residential sector consumption 20% Single dwelling sector consumption 75% Figure 3.7 Composition of total water consumption in the Blue Mountains region, NSW, Australia for the period of Year Figure 3.8 Yearly total water consumption ( ) in the Blue Mountains region, Australia Total yearly metered water consumption of all sectors and per dwelling monthly water consumption data of the residential sector comprising of single and multiple University of Western Sydney Page 47

85 Jan Jul Jan Jul Jan Jul Jan Jul Jan Jul Jan Jul Jan Jul Jan Jul Jan Jul Jan Jul Jan Jul Jan Jul Jan Jul Jan Jul Water Consumption (kl/dwelling/month) Climate change impact on water demand and supply CHAPTER 3: Study area and Data dwellings sectors are presented in Figures 3.8 and 3.9. Both the figures show the gradually decreasing trend in the water consumption during the period of 2003 to 2010 which is more likely to be attributed to the combined effect of imposed water restrictions and conservation programs adopted during that period in the Blue Mountains region. It should be noted here that no mandatory water restrictions were applied in the Blue Mountains region before Month & Year Figure 3.9 Per dwelling monthly water consumption of the residential sector ( ) in the Blue Mountains region, Australia Water price Water usage price data were obtained from Sydney Water for the period of , which is presented in Table 3.2. Two water consumption bands (tiers) are applied in the BMWSS to determine the water use price, tier 1 (0 to 400 kl/year) and tier 2 (400 + kl/year). In the Blue Mountains region, water consumption band falls under the tier 1; therefore, price of tier 1 has been used throughout the study. From the water price data, it has been found that price of water has been increasing in every year by AUD/kL. Based on this growth rate, water price for the forecasting period is estimated to be used in the water demand forecasting. At 2040, the water price is forecasted to be 4.48 AUD/kL. University of Western Sydney Page 48

86 CHAPTER 3: Study area and Data Table 3.2 Water price data for the Blue Mountains region of the period 1997 to 2012 Year Water price ($/kl) Consumption band tier - 1 (kl/year) Number of dwellings Dwellings data of the single and multiple dwelling residential sectors in the Blue Mountains region (Mount Vitoria to Faulconbridge) were obtained from Sydney Water for the period of 1997 to Total number of dwellings of the Blue Mountains region is presented in Figure From the data, it has been found that majority of the dwellings are in the single dwelling category. Composition of the dwellings in the single and multiple dwelling sectors demonstrate that 91% of the dwellings belong to single dwelling sector and the rest 9% belongs to multiple dwelling sector. Figure 3.10 shows that the number of dwellings is gradually increasing. The monthly growth rates of single and multiple dwelling sectors have been found to be around 0.07% and 0.17%, respectively. Based on this growth rate, numbers of single and multiple dwellings have been forecasted for the periods of to be used in the water demand forecasting models to predict future water demand. University of Western Sydney Page 49

87 Jan June Jan June Jan June Jan June Jan June Jan June Jan June Jan June Jan June Jan June Jan June Jan June Jan June Jan June Jan June Number of dwellings Climate change impact on water demand and supply CHAPTER 3: Study area and Data Single Residential Dwellings Multi Residential Dwellings Month, Year Figure 3.10 Number of total dwellings in the Blue Mountains region during the period of 1997 to Water conservation programs Several water conservation programs (mentioned in section 3.5) have been introduced to the greater Sydney including the Blue Mountains region in 2000 by Sydney Water to reduce water demand. Number of participated dwellings in those programs in monthly time steps was obtained from Sydney Water for the period of 2000 to In addition, approximate yearly average water savings from those programs were obtained from Sydney Water (Table 3.3). It should be noted here that not all the programs started in 2000; WaterFix started in 2000, and installation of rainwater tank and washing machine programs started in 2003 (Figure 3.11). DIY and replacement programs started in 2005 and 2009, respectively. From the recent two years (September 2009 to September 2011) available data on the number of participated dwellings in the water conservation programs, it has been found that in each month 7, 13, 2, 8 and 11 dwellings were added in the WaterFix, toilet replacement, DIY, washing machine and rainwater tank programs, respectively. Future values of the participating dwellings in the water conservation programs have been estimated based on this monthly growth rate till University of Western Sydney Page 50

88 Number of participated dwelling Climate change impact on water demand and supply CHAPTER 3: Study area and Data Table 3.3 Average water savings from the water conservation programs Water conservation programs Average water savings per program (kl/year) Average water savings per program (kl/month) WaterFix a Residential toilet replacement DIY (Do-it-Yourself) kits b Washing machine Residential rainwater tank a Installation of new showerheads, flow restrictors and minor leak repairs undertaken by a licensed plumber b Self-installed flow restrictors 7000 WaterFix Rainwater tank DIY Waching machine Toilet replacement Month, Year Figure 3.11 Number of participated dwellings in the water conservation programs (1: WaterFix, 2: Rainwater tank, 3: DIY, 4: Washing machine and 5: Toilet replacement) in the Blue Mountains region for the period of 2000 to 2011 University of Western Sydney Page 51

89 CHAPTER 3: Study area and Data Climate projections The future projections of rainfall, temperature and evaporation data for the period of 2021 to 2040 were obtained from Sydney Catchment Authority. Katoomba weather station, which is located in the Blue Mountains region, has been used to obtain future projections of rainfall and temperature data. Richmond (UWS Hawkesbury weather station) has been used to obtain evaporation data as it is the closest weather station to the study area for which downscaled evaporation data for the future period was available. These future projections of climatic data were generated using CSIRO Mk. 3 GCM under three emission scenarios, A1B, A2 and B1, and downscaled by a statistical downscaling method (Mehrotra and Sharma 2010). These data were generated for an inter-governmental project called Climate change and its impacts on supply and demand in Sydney (2009), more details of the project can be found in the technical report produced by Sydney Catchment Authority (2009b). Table 3.4 List of the GCMs used in this study and their spatial resolution No Global climate model 1 CCCMA 2 CSIRO Mk 3 3 MIROC 4 ECHAM5 Modelling group and Country Canadian Centre for Climate Modeling and Analysis, Canada CSIRO Atmospheric Research, Australia Center for Climate System Research (The University of Tokyo), National Institute for Environmental Studies, and Frontier Research Center for Global Change, Japan Max Planck Institute of Meteorology, Germany Horizontal resolution (km) (approx.) Reference 175 Flato and Boer (2001) 175 Gordon et al. (2002) 250 K-1 model developers (2004) 175 Jungclaus et al. (2006) More recent and up to date future projections of rainfall data for the Katoomba weather station were also obtained from Sydney Catchment Authority for four GCMs (CSIRO, MIROC, CCCMA and ECHAM 5) for the period of 2021 to 2040 under A2 emission scenario. The horizontal resolutions and originating countries of the four GCMs that are used in this study are given in Table 3.4. These rainfalls projections were downscaled by MMM-KDE stochastic downscaling model (Mehrotra and University of Western Sydney Page 52

90 CHAPTER 3: Study area and Data Sharma 2007) and 100 realisations of downscaled rainfall were made available for each of the GCM. These data are the outputs of the NARCliM (NSW/ACT Regional Climate Modelling) project that is producing an ensemble of regional climate projections for New South Wales and the Australian Capital Territory, Australia. More details on the NARCliM project can be found on the website, Runoff As the Blue Mountains catchments (Katoomba and Blackheath) are ungauged catchments, historical runoff data that is required to calibrate and validate the rainfall-runoff models in order to be used in the climate impact analysis was not available for those catchments. Therefore, runoff data were obtained from the nearest gauged catchment (Narrow Neck) of the Blue Mountains catchments to be used in the calibration of the rainfall-runoff models for the period of 1990 to 2012 in monthly time steps. The streamflow data of the Narrow Neck catchment was obtained from NSW Department of Water, which indicated a good quality data set based on the reported data quality. 3.7 Summary In this thesis, the Blue Mountains region and the Blue Mountains water supply system have been selected as the study region and as an urban water supply system, respectively, to carry out the research. Forecasting of the future changes in the water demand and supply scenarios due to the changing climatic conditions is of great importance in this study area and the water supply system to ensure efficient water supplies to the communities. The necessary data to develop long term water demand forecasting models and to identify the climatic impacts on water demand and supply were obtained from Sydney Water and Sydney Catchment Authority. These data have been used in the analyses and modelling in the subsequent chapters. University of Western Sydney Page 53

91 CHAPTER 4: Impact of water restriction CHAPTER 4 IMPACT OF WATER RESTRICTIONS ON URBAN WATER DEMAND This chapter is partial reproduction of the following refereed paper: Haque, M.M. 1, Hagare, D. 1, Rahman, A. 1, and Kibria, G Quantification of water savings due to drought restrictions in water demand forecasting models. Journal of Water Resources Planning and Management, 140(11), (ERA 2010 ranking: A*, Impact factor: 1.76). 1 School of Computing, Engineering and Mathematics, University of Western Sydney, Australia 2 Sydney Catchment Authority, Penrith, Australia Abstract This chapter presents a technique to quantify water savings due to implementation of water restrictions by adopting water restriction indices as a continuous numerical independent variable in a regression analysis. The adopted modelling technique compares four methods: Yearly Base Difference Method (YBDM), Weighted Average Method (WAM), Before and After Method (BAM) and Expected Use Method (EUM). These methods are applied to the residential sectors in the Blue Mountains region, Australia, which consists of single and multiple dwelling sector. Three forms of multiple regression techniques are adopted: Raw-Data/Linear, Semi- Log and Log-Log. The model performances are evaluated by a number of statistics such as absolute average relative error, Nash-Sutcliffe efficiency and percentage bias. Moreover, the potential of using the water restriction savings and water conservation savings as continuous independent variables in the water demand forecasting model is investigated. The performances of different modelling techniques are evaluated using split-sample and leave-one-out cross validation methods. The YBDM method is found to quantify the water savings more accurately than the other adopted methods. The water savings due to Levels 1, 2 and 3 water restrictions are found to be approximately 9%, 18% and 20%, respectively, for the University of Western Sydney Page 54

92 CHAPTER 4: Impact of water restriction single dwelling sector and approximately 4%, 8% and 9%, respectively, for the multiple dwelling sector. The Semi-Log model coupled with the YBDM method is found to perform the best in predicting water demand for both the single and multiple dwelling sectors with an absolute average relative error of about 3%. University of Western Sydney Page 55

93 CHAPTER 4: Impact of water restriction 4.1 Overview Chapter 3 has discussed the selection of the study area and its importance, and data collation for the study area. This chapter estimates water savings due to the implementation of water restrictions in the Blue Mountains water supply system during the drought periods ( ). This chapter also develops the long term water demand forecasting models by adopting water savings as continuous independent variables (i.e. numeric representation) in the models. This chapter commences with presenting the methodologies developed to quantity water savings and the methodologies adopted to develop water demand forecasting models. It then discusses the variables used in the water demand forecasting models. This is followed by the results and discussion, and the chapter concludes by summarising the findings. 4.2 Methodology The methodology developed to quantify water savings from water restrictions is illustrated in Figure 4.1. First, total water savings due to water restrictions and water conservation programs were calculated for the period of by four different methods: Yearly Base Difference Method (YBDM), Weighted Average Method (WAM), Before and After Method (BAM) and Expected Use Method (EUM) (detailed descriptions of these methods are given in Section 4.3.2). Then, water savings due to water restrictions were identified by separating the water conservation savings from the total savings. The evaluation of the water savings calculation methods was done in two steps: (i) by including the water restriction and water conservation savings variables in the water demand models as independent variables along with other climatic, demographic and water price variables; and (ii) by comparing the performance of the developed models to simulate historical water demand for the water restrictions periods ( ) in the Blue Mountains region. University of Western Sydney Page 56

94 CHAPTER 4: Impact of water restriction From these results, the best water demand model and the best water savings calculation method were identified. Thereafter, the identified water demand model was used to forecast water demand for the period of July 2009 to September 2011 to test its suitability as a forecasting model. In addition, leave-one-out cross validation was undertaken with the identified water demand forecasting model to test the reliability of the model. These water savings and demand forecasting calculations were done for both the single and multiple dwelling sectors separately. Step 1: Calculation of total water savings By YBDM, BAM, EUM & WAM Step 2: Estimation of water savings due to water conservation programs Step 3: Estimation of water savings due to restrictions from step 1(value)-step 2(value) for YBDM, BAM, EUM & WAM Step 4: Modelling of water demand for the period Demand model (YBDM) Demand model (BAM) Demand model (EUM) Demand model (WAM) Step 5: Evaluating the performances of the models Step 6: Identifying the best model and best water savings calculation method Step 7: Forecasting of water demand by the best model for the period of July 2009 to September 2011 Step 8: Leave-one-out cross validation of the considered forecasting model Figure 4.1 Framework for quantifying water savings and developing water demand forecasting models University of Western Sydney Page 57

95 CHAPTER 4: Impact of water restriction Multiple regression analysis In this chapter, multiple linear regression (MLR) techniques were adopted to model and forecast water demand for both the single and multiple dwelling sectors in the Blue Mountains region. A MLR technique attempts to model the relationship between two or more independent variables with a dependent variable by fitting a linear equation to the observed data. In water demand literature, three forms of multiple regression techniques: Raw-Data, Semi-Log, and Log-Log are widely used to model long term water demand (e.g. Hoffmann et al. (2006), Babel et al. (2007), Dziegielewski and Chowdhury (2011)). In this chapter, these three forms of multiple regression techniques were adopted to develop the water demand models. In the Raw-Data model, the relationship between the dependent variable and the independent variables are assumed to be linear. The following represents a multiple linear regression equation (Montgomery et al. 2001): Y X X k X k (4.1) where is the model intercept, 1,2,3... k are the slope coefficients, and k is the number of independent variables. In the Semi-Log model, only the dependent variable is normally taken in 10-base logarithmic form whereas in the Log-Log models both the independent and dependent variables are entered as 10-base logarithmic form in the regression equation. The functional forms of the Semi-Log and Log-Log model are given in equations 2 and 3, respectively (Babel et al. 2007). log Y X X k X k (4.2) logy 1 log( X1) 2 log( X 2)... k log( X k ) (4.3) Estimation of total water savings Yearly base difference method (YBDM) This approach defines a base consumption period for which no restriction is implemented and estimates the average water consumption for that period. Then total University of Western Sydney Page 58

96 CHAPTER 4: Impact of water restriction water savings for a period due to water restrictions and water conservation programs are calculated by comparing the average base water use to the water use of the drought periods. For illustration, let s say the base period ( ) and WU b is the average monthly water use for WU ij is the water use in the month j of the drought year ( ), where j = Jan,..., Dec. Then total monthly water savings ( WS ) can be calculated by the following equation: ij WS ij WU WU (4.4) b ij In this study, the period was chosen as the base consumption period as during these periods no water restriction was imposed in the study area. Moreover, no water conservation programs were implemented during However, a very little amount of water savings (2% of total water use) was achieved during the period due to the introduction of the water conservation programs, which was quite negligible. Average water use per dwelling per month was found to be kl/dwelling/month for the base consumption period for the single dwelling sector. Total monthly water savings was calculated by comparing this value with the monthly value of the water consumption per dwelling for the period of January 2003 to June As an example, calculation of total water savings by the YBDM method of the year 2004 for the single dwelling sector is presented in Table A.4.1 in Appendix A Before and after method (BAM) In this method, the total monthly water savings are calculated by comparing the monthly base ( ) water use value with the corresponding monthly water use of the drought periods. For illustration, if ( WU ) is the average water use of the b j base consumption period of month j where j =Jan,..., Dec and WU ij is the water use in the month of j of the drought year i, then total monthly water savings ( WS ij ) can be calculated by the following equation: WS ij ( WU ) WU (4.5) b j ij University of Western Sydney Page 59

97 CHAPTER 4: Impact of water restriction As an example, monthly base water use and calculation of total water savings by the BAM method of the year 2004 for the single dwelling sector are presented in Tables A.4.2 and A.4.3, respectively in Appendix A Expected use method (EUM) Water savings calculation by the EUM method normally consists of estimating the level of water demand (expected use) that would have occurred due to climate conditions of a particular year under no restrictions and comparing that estimated water demand value with the observed water demand for that period (Kenney et al. 2004, Spaninks 2010). For illustration, if ( WU e) ij is the expected water use in the month of j under climate conditions of drought year i assuming no water restrictions and ( WU ) is the observed water use of the same period. Then total o ij water savings ( WS ) can be calculated by the following equation: ij WS ( WU ) ( WU ) (4.6) ij e ij o ij Define water demand as a function of climatic variables (i.e. Climate water demand model) Estimate the expected water use assuming no water restriction for the periods of Calculate the difference between the expected and observed water consumption University of Western Sydney Page 60

98 Water Use (kl/dwelling/month) Jan Apr Jul Oct Jan Apr Jul Oct Jan Apr Jul Oct Jan Apr Jul Oct Climate change impact on water demand and supply CHAPTER 4: Impact of water restriction Figure 4.2 Framework of calculating total water savings by expected use method (EUM) The EUM method for calculating total water savings is illustrated in Figure 4.2. First step of the EUM is to define the water demand function with respect to climatic variables. In this study, monthly total rainfall and monthly total evaporation data were used as independent variables in a multiple linear regression equation to predict monthly per dwelling water use that would have been demanded in the absence of water restrictions. To easily refer this equation in the next few paragraphs, this model has been named as Climate Water Demand Model (CWDM). The coefficients of this CWDM model were estimated using the data from the year 1999 to 2002 and it was validated against the year of The results of these cross-validations demonstrated that this multiple linear regression equation (R 2 = 0.71) had substantial accuracy in predicting water use as presented in Figures 4.3 and 4.4. For example, Kenney et al. (2004) also got the R 2 values for their models ranged from 0.62 to They argued that this level of accuracy was adequate for the purpose of describing drought response of water use Observed Modelled Month & Year Figure 4.3 Comparison of the observed and modelled water use for the period of January 1999 to December 2002 in the Blue Mountains region using climate water demand model After doing the multiple regression analysis, the Log-Log model was found better to model the water demand to be used in the EUM. The developed CWDM model was applied to the data set (January 2003 to June 2009) to estimate the expected water University of Western Sydney Page 61

99 Water Use (kl/dwelling/month) Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Nov Climate change impact on water demand and supply CHAPTER 4: Impact of water restriction use during the periods of water restrictions. The difference between the expected water use and observed water use provides an estimate of the total water savings. As an example, calculation of total water savings by the EUM method of the year 2004 for the single dwelling sector is presented in Table A.4.4 in Appendix A Observed Modelled Month & Year Figure 4.4 Validation results of the observed and modelled water use for the period of January 1997 to December 1998 in the Blue Mountains region using climate water demand model Weighted average method (WAM) Water savings by the WAM method was estimated by assigning weights to YBDM and EUM methods. In this study, three pairs of weightages were considered for the YBDM and EUM methods to be able to make a better comparison of the results of the WAM with other methods. In this regard, the three pairs of adopted weightages were: (0.75, 0.25), (0.5, 0.5) and (0.25, 0.75). Total water savings ( was calculated by the following equation: WS WAM ) by WAM WS WAM K WS K WS (4.7) YBDM YBDM EUM EUM where, K YBDM = 0.75, 0.5, 0.25 and corresponding K EUM = 0.25, 0.5, As an example, calculation of total water savings by the EUM method of the year 2004 for the single dwelling sector is presented in Table A.4.5 in Appendix A Model evaluation criteria University of Western Sydney Page 62

100 CHAPTER 4: Impact of water restriction The performance of each of the developed models was estimated using three evaluation statistics: the absolute average relative error (AARE), the percentage of bias (PBIAS) and the Nash-Sutcliffe efficiency (NSE). AARE, expressed as a percentage, measures the relative magnitude of the error of the predicted values in relation to the observed data. AARE value can be calculated with the following equation: n ( Oi Pi ) 1 AARE (%) 100 (4.8) n O 1 i where O i is the observed value at time i and P i is the predicted value by the model at time i. A smaller AARE value indicates better performance of the model with an ideal value of 0%. PBIAS measures the average tendency of the modeled value to be larger or smaller than their observed value (Gupta et al. 1999). PBIAS is expressed in percentage and can be calculated with the following equation: n ( O ) 1 i Pi PBIAS *100 (4.9) n O 1 i where, n is the number of observations. The most favorable value of PBIAS is zero. Negative value of PBIAS indicates overestimation bias whereas positive value indicates underestimation. The Nash-Sutcliffe efficiency is a normalized measure (- to 1), that estimates the relative magnitude of the residual variance compared to the observed data variance (Nash and Sutcliffe 1970). It can be calculated by the following equation: NSE n n 2 ( Oi Pi ) 1 1 (4.10) 2 ( Oi O 1 mean) An ideal value of NSE is one, which indicates a perfect model performance. A NSE value of zero indicates that the model results are as accurate as the mean of the observation. University of Western Sydney Page 63

101 CHAPTER 4: Impact of water restriction Leave-One-Out (LOO) cross validation In this study, the developed water demand forecasting model was tested by using the leave-one-out cross validation technique. This method consists of developing a model using all the data except one and then the model is tested on the omitted data point. Next, the first data record is replaced and a second one is removed and a new model is then developed with these data and it is now tested on that second data record that has been omitted. This procedure is normally repeated for all the data records. 4.3 Water demand variables Water demand can be influenced by many variables such as socio-economic, demographic, climatic and demand management programs (Babel and Shinde 2011, Odan and Reis 2012). In this study, water demand models were developed using per dwelling monthly metered water consumption as a dependent variable and using water usage price, rainfall, temperature, water savings from conservation programs, and water savings from water restrictions as independent variables (Table 4.1). Many earlier water demand studies (e.g. Hoffmann et al. (2006), Mazzanti and Montini (2006), Abrams et al. (2012)) have included water price as an independent variable because of its potential to be a short and long term water demand management tool. Climate variables, especially rainfall and temperature have some influence on residential water demand (Gato et al. 2007, Polebitski et al. 2010). A number of researchers, such as Adamowski (2008), Polebitski and Palmer (2009), Franczyk and Chang (2009) have included rainfall and temperature as independent variables in their models. Residential water use is expected to be positively and negatively correlated with temperature and rainfall, respectively as with higher temperature and/or lower rainfall, residents tend to use more water for garden watering and personal use. As discussed in Chapter 2 (Section 2.5), earlier studies included water conservation programs and water restrictions as binary variables in the water demand models. Incorporation of water savings variables as continuous independent variables has not been investigated in the literature till now. Hence, in this study, estimates of water University of Western Sydney Page 64

102 CHAPTER 4: Impact of water restriction savings were included as continuous independent variables in the water demand models instead of using binary variables. From the number of participant dwelling data in water conservation programs in the Blue Mountains regions, it has been found that the number increases in each month. It implies that the residents in that area are gradually adopting the water conservation programs in their houses. Therefore, inclusion of binary variable to represent the effect of water conservation programs may not capture the increasing effect of those programs properly, as binary variable only represent the existence or non-existence of the event. Table 4.1 List of dependent and independent variables used in developing water demand models Symbol Description Dependent variable Y Monthly metered water consumption of a dwelling in kl Independent variables X 1 X 2 X 3 X 4 Monthly total rainfall in mm Monthly mean maximum temperature in 0 C Water usage price ($/kl) Water savings from conservation programs in kl/dwelling/month X 5 Water savings from water restrictions in kl/dwelling/month As mentioned in Chapter 3 (Section 3.6.4), data on approximate average yearly water savings for each of the water conservation programs implemented in the study area during the study period were obtained from Sydney Water (Table 3.3). Some of the water conservation savings are weather dependent (e.g. rainwater tank), and some are mostly weather independent (e.g. washing machine use and toilet flushing). The University of Western Sydney Page 65

103 CHAPTER 4: Impact of water restriction monthly savings arising from a given water conservation program are ideally different from month to month; however, in this study, the annual water savings from a given water conservation program were equally distributed over all the months. This is mainly due to the fact that monthly water conservation data from a given conservation program was not available and it is unlikely to affect the outcome of this study. Data on the numbers of household that was participated in the programs were also obtained in the monthly steps from Sydney Water. Then total monthly water savings from the conservations programs were estimated by multiplying the average monthly savings with monthly participated household number. These monthly total savings were divided by the total number of household in that month to get the average per dwelling saving from all of the conservation programs. For an illustration, calculations of per dwelling water savings in the months of January 2007 from the conservation programs are presented in Table A.4.6 in Appendix A. These monthly per dwelling water savings which was termed as water conservation savings (WCS) was taken as one of the independent variables in the regression models. Per dwelling monthly water savings from water restrictions which was termed as water restriction savings (WRS) were calculated by deducting monthly per dwelling water conservation savings from monthly per dwelling total water savings (as detailed in Section 4.3.2), which can be expressed by the following equation. ( WRS ) ( WS ) ( WCS ) (4.11) ij T ij ij where, WRS = Per dwelling monthly water restrictions savings (kl/month/dwelling); WS T = Total water savings (kl/month/dwelling); WCS = Per dwelling monthly water conservation savings (kl/month/dwelling); i = drought year ( ); and j = month (Jan,..., Dec). University of Western Sydney Page 66

104 CHAPTER 4: Impact of water restriction As mentioned in Chapter 3 (Section and 3.6.2), data on per dwelling monthly metered water consumption and water usage price were obtained from Sydney Water for the study area for the period of January 1997 to September Rainfall and temperature data were obtained from Sydney Catchment Authority. 4.4 Results Water demand modelling was done by the three forms (Raw Data, Semi-Log, Log- Log) of multiple regression techniques using the data from 1997 to Each model was developed by taking WRS as an independent variable under four different methods of savings calculation along with rainfall, temperature, water usage price and WCS as other independent variables. A total of 12 models (3 multiple regression forms times 4 water savings calculation methods) were developed for both the single and multiple dwelling sectors separately. Comparative performances of the developed models for the single dwelling sector are presented in Table 4.2. In the table, only the results of developed WAM (0.5, 0.5) model have been reported among the three WAM models (WAM (0.75, 0.25), WAM (0.5, 0.5), WAM (0.25, 0.75)), as WAM (0.5, 0.5) model found to model the water demand in a better way than the other two WAM models. It can be seen in Table 4.2 that the Semi-Log - YBDM model performed better than all the other models, the AARE and PBIAS values of this model were 2.49% and 0.61%, respectively, the lowest among all the models. Also, NSE and R 2 values of the Semi-Log model were found to be 0.95 and 74.80%, respectively, which were the highest values among all the developed models, which again signified the superiority of this model. Regression coefficients associated with few candidate models were found not to be as per physical intuition, and hence these were not further examined. For example, in the case of Raw-Data model coupled with WAM method, the regression coefficient associated with the independent variable water conservation savings came out as positive, which does not make sense since a water savings program should reduce the overall water demand. The performance statistics associated with these candidate models are not reported in Table 4.2. University of Western Sydney Page 67

105 CHAPTER 4: Impact of water restriction Table 4.2 Performance statistics of the developed models for the single dwelling sector Water savings calculation method YBDM EUM WAM BAM Performance statistics Raw Data model Semi-Log model Log-Log model AARE (%) PBIAS (%) NSE R 2 (%) AARE (%) PBIAS (%) NSE R 2 (%) AARE (%) * * 3.21 PBIAS (%) * * 0.50 NSE * * 0.94 R 2 (%) * * AARE (%) PBIAS (%) NSE R 2 (%) Note: Model values for which the signs of the coefficients of the independent variables were found to be inconsistent with the expected behaviour of the dependent variable are not presented in the table and marked by * The relatively better multiple regression models under YBDM, EUM, WAM and BAM method of calculating water savings were found to be Semi-Log, Raw-Data, Log-Log and Raw-Data, respectively for the single dwelling sector. Their relative performances in modelling water use during the restriction periods ( ) are presented in Figures 4.5 (a to d). As can be seen in Figures 4.5 (a to d), water use during the restrictions periods was simulated better by the Semi-Log model coupled with YBDM of water savings calculation among the other developed water demand models. Performance statistics of the developed models for the multiple dwelling sector are presented in Table 4.3. Likewise single dwelling sector, results of the models are not given here for which sign of the coefficients of the independent variables were found to be inappropriate. As can be seen in Table 4.3, three developed models: the Semi- Log model with YBDM, Raw-Data and Semi-Log model coupled with the WAM University of Western Sydney Page 68

106 CHAPTER 4: Impact of water restriction method of savings calculation were found to perform better among all the developed water demand models. However, AARE of the Semi-Log model coupled with the YBDM method was found to be 2.38% which was the lowest among the models indicating its better performance. The relatively better multiple regression models under the YBDM, EUM, WAM and BAM method of calculating water savings were found to be in the order of Semi- Log, Log-Log, Raw-Data and Log-Log models, respectively for the multiple dwelling sector. Their relative performances in modelling water use during the restriction periods ( ) are presented in Figures 4.6 (a to d). As can be seen in Figures 4.6 (a to d), water use during the restrictions periods were simulated more accurately by the Semi-Log model coupled with the YBDM method of water savings calculation among the other developed water demand models. Water savings estimated by the YBDM method for the single dwelling sector during the water restriction periods ( ) is presented in Table 4.4. Around 9.13%, 18.10% and 20.09% water savings were achieved during the Levels 1, 2 and 3 restrictions periods, respectively. In Table 4.4, as expected, it can be seen that Levels 2 and 3 restrictions achieved greater water savings than Level 1 as they were more stringent in restriction rules. Moreover, water savings achieved during Level 3 restriction period were approximately 2% higher than the one achieved under Level 2. Similar results in relation to incremental effect of Level 3 restriction on Level 2 restriction were reported by Abrams et al. (2012) for the greater Sydney area in Australia. Water savings estimated by the YBDM method for the multiple dwelling residential sector during the drought restrictions periods are presented in Table 4.5. Around 3.90%, 7.62% and 8.79% water savings were achieved during the Levels 1, 2 and 3 restrictions, respectively. The effects of water restriction are higher in the single dwelling sector than that of multiple dwelling sector. These results of water savings from water restrictions in both the single and multiple dwelling sector in the Blue Mountains region indicated that the water restriction programs were quite successful in reducing water usage to cope up with the limited water supply during the drought periods. University of Western Sydney Page 69

107 Total yearly water consumption (kl) Total yearly water consumption (kl) Climate change impact on water demand and supply CHAPTER 4: Impact of water restriction Observed consumption Modelled consumption NSE = 0.95 R 2 = 74.8% Year a)ybdm-semi Log Figure 4.5.a Comparison of modelled versus observed water consumption by the best model (Semi-Log) for the single dwelling sector during water restriction periods under yearly base difference method (YBDM) of water savings calculation Observed consumption Modelled consumption NSE = 0.93 R 2 = 66.5% Year b)eum-raw data Figure 4.5.b Comparison of modelled versus observed water consumption by the best model (Raw-Data) for the single dwelling sector during water restriction periods under expected use method (EUM) of water savings calculation University of Western Sydney Page 70

108 Total yearly water consumption (kl) Total yearly water consumtion (kl) Climate change impact on water demand and supply CHAPTER 4: Impact of water restriction Observed consumption Modelled consumption NSE = 0.94 R 2 = 74.4% Year c)wam-log Log Figure 4.5.c Comparison of modelled versus observed water consumption by the best model (Log-Log) for the single dwelling sector during water restriction periods under weighted average method (WAM) of water savings calculation Observed consumption Modelled consumption NSE = 0.93 R 2 = 64.3% Year d)bam-raw data Figure 4.5.d Comparison of modelled versus observed water consumption by the best model (Raw-Data) for the single dwelling sector during water restrictions periods under before and after method (BAM) of water savings calculation University of Western Sydney Page 71

109 CHAPTER 4: Impact of water restriction Table 4.3 Performance statistics of the developed models for the multiple Water savings calculation method YBDM EUM WAM BAM Performance criteria dwelling sector Raw-Data model Semi-Log model Log-Log model AARE (%) * PBIAS (%) * NSE * R 2 (%) * AARE (%) * * 4.41 PBIAS (%) * * 0.10 NSE * * 0.84 R 2 (%) * * AARE (%) PBIAS (%) NSE R 2 (%) AARE (%) * * 4.31 PBIAS (%) * * 0.22 NSE * * 0.85 R 2 (%) * * Note: Model values for which the signs of the coefficients of the independent variables were found to be inconsistent with the normal behaviour of the dependent variable are not presented in the table and marked by * Table 4.4 Percentage of water savings due to water restrictions during the drought periods ( ) in the single dwelling sector in the Blue Mountains region Level of restrictions and periods Level 1 (Oct 03 May 04) Level 2 (June 04 May 05) Level 3 (June 05 June 09) Total water use (a) (kl) Conservation savings (b) (kl) Restrictions savings by YBDM (c) (kl) Restrictions savings in % by YBDM c ( a+b+c ) University of Western Sydney Page 72

110 CHAPTER 4: Impact of water restriction Table 4.5 Percentage of water savings due to water restrictions during the drought periods ( ) in the multiple dwelling sector in the Blue Mountains region Level of restrictions and periods Level 1 (Oct 03 May 04) Level 2 (June 04 May 05) Level 3 (June 05 June 09) Total water use (kl) Conservation savings (kl) Restrictions savings in kl by YBDM Restrictions savings in % by YBDM c a+b+c ) During the last one year of Level 3 water restrictions (July 08 to June 09), average water use was found to be kl/dwelling/month for the single dwelling sector. However, during the post restrictions periods, average water usage was found to be kl/dwelling/month and kl/dwelling/month for the period July 09 to June 10 and July 10 to June 11, respectively. Comparing with water consumption during drought periods, it seems that residents have largely chosen to retain the water use practice established during the drought restriction periods. Similar finding was reported by Sydney Water in their water consumption and recycling implementation report ( ) that water use was only increased by less than 3% in as compared to despite the non-existence of mandatory water restrictions. University of Western Sydney Page 73

111 Total Yearly Water Demand (kl) Total yearly water consumption (kl) Climate change impact on water demand and supply CHAPTER 4: Impact of water restriction Observed consumption Modelled consumption NSE = 0.92 R 2 = 71.2% Year a)ybdm-semi Log Figure 4.6.a Comparison of modelled versus observed water consumption by the best model (Semi-Log) under yearly base difference method (YBDM) of water savings calculation for the multiple dwelling sector during water restriction periods Observed consumption Modelled consumption NSE = 0.84 R 2 = 50.9% Year b)eum-log Log Figure 4.6.b Comparison of modelled versus observed water consumption by the best model (Log-Log) under expected use method (EUM) of water savings calculation for the multiple dwelling sector during water restriction periods University of Western Sydney Page 74

112 Total Yearly Water Demand (kl) Total Yearly Water Demand (kl) Climate change impact on water demand and supply CHAPTER 4: Impact of water restriction Observed consumption Modelled consumption NSE = 0.92 R 2 = 68.0% Year c)wam-raw data Figure 4.6.c Comparison of modelled versus observed water consumption by the best model (Raw data) under weighted average method (WAM) of water savings calculation for the multiple dwelling sector during water restriction periods Observed consumption Modelled consumption NSE = 0.85 R 2 = 50.1% Year d)bam-log Log Figure 4.6.d Comparison of modelled versus observed water consumption by the best model (Log-Log) under before and after method (BAM) of water savings calculation for the multiple dwelling sector during water restriction periods University of Western Sydney Page 75

113 CHAPTER 4: Impact of water restriction As the Semi-Log model with YBDM was found to be the best model for both the single and multiple dwelling sectors, this was finally adopted to forecast the water demand for the period of July 2009 to September Then the forecasted values were compared with the corresponding observed values to check the model reliability as a long term water demand forecasting tool. The developed equations of the Semi- Log model for the single and multiple dwelling sectors are given in equations 4.12 and 4.13, respectively: log( log( Y s Y m ) X ) X X X X X X X X 5 (4.12) X 5 where (4.13) Y s and Y m represent per dwelling monthly single and multiple dwelling demand, respectively, and X 1, X 2, X3, X 4, X5 represent monthly total rainfall (mm), monthly mean maximum temperature ( 0 C), water usage price ($/kl), water conservation savings (kl/dwelling/month) and water restriction savings (kl/dwelling/month), respectively. Performance statistics of the Semi-Log model for the forecasted period (2009 to 2011) for both the single and multiple dwelling sectors are given in Table 4.6. The values of AARE, PBIAS and NSE were found to be in the accepted margin indicating a very good agreement between the model forecast and observed values. The comparison of monthly and yearly observed and modelled demand by the Semi- Log model is presented in Figures 4.7 and 4.8 for the single dwelling sector, respectively, and in Figures 4.9 and 4.10 for the multiple dwelling sector, respectively. The simulated monthly and yearly water demand values for both the sectors were found to be quite close to the observed values. However, some variations were observed for few months. These might be attributed to high variation in temperature and rainfall values and/or changes in water consumption pattern. A daily demand model might capture these variations more efficiently. However, average AARE values for all of the predicting months were only 4% and 2.62% for University of Western Sydney Page 76

114 Jul Sep Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Monthly water demand (kl) Climate change impact on water demand and supply CHAPTER 4: Impact of water restriction the single and multiple dwelling sectors, respectively. This indicates the developed models are quite accurate and hence can be used for predicting both monthly and yearly water demand. Table 4.6 Performance statistics of the developed Semi-Log model for the forecasting period (July 2009 to September 2011) in the single and multiple dwelling sectors Criteria Single dwelling sector Multiple dwelling sector AARE (%) PBIAS (%) NSE Observed water demand Modelled water demand Month & Year Figure 4.7 Comparison of monthly forecasted versus observed water demand by the Semi-Log model coupled with YBDM for the forecasting period (2009 to 2011) in the single dwelling sector University of Western Sydney Page 77

115 Jul Sep Nov Jan Mar May Jul Sep Nov Jan Mar May Jul Sep Monthly water demand (kl) Yearly water demand (kl) Climate change impact on water demand and supply CHAPTER 4: Impact of water restriction Observed water demand Modelled water demand (July-Dec) 2010 (Jan-Dec) 2011 (Jan-Sept) Year Figure 4.8 Comparison of yearly forecasted versus observed water demand by the Semi-Log model coupled with YBDM for the forecasting period (2009 to 2011) in the single dwelling sector Observed water demand Modelled water demand Year & Month Figure 4.9 Comparison of monthly forecasted versus observed water demand values using the Semi-Log model coupled with YBDM for the forecasting period (2009 to 2011) in the multiple dwelling sector University of Western Sydney Page 78

116 Yearly water demand (kl) Climate change impact on water demand and supply CHAPTER 4: Impact of water restriction Observed water demand Modelled water demand (July-Dec) 2010 (Jan-Dec) 2011 (Jan-Sept) Year Figure 4.10 Comparison of yearly forecasted versus observed water demand values using the Semi-Log model coupled with YBDM for the forecasting period (2009 to 2011) in the multiple dwelling sector The finally accepted model (Semi-Log model) was also tested by the leave-one-out cross validation procedure. The average AARE values from this validation were found to be 1.5% and 1.33% for the single and multiple dwelling residential sectors, respectively, for the period of 2003 to These AARE values are quite small indicating the developed models are acceptable. The R 2 values were found to be in the range of 77-79% and 68-73% for the single and multiple dwelling sectors, respectively. These are also considered to be within an acceptable range. Also, the regressions coefficients do no vary notably from run to run, which indicate the models are quite stable. Results of the leave-one-out cross validation of the developed models are presented in Tables A.4.7 and A.4.8 in Appendix A for the single and multiple dwelling sectors, respectively. 4.5 Summary In this chapter, two new methods to simulate water savings (Yearly Base Difference Method (YBDM) and Weighted Average Method (WAM)) and two existing methods (Before and After Method (BAM) and Expected Use Method (EUM)) were investigated in the Blue Mountains region, New South Wales, Australia. Evaluation of the proposed methods was done by simulating the water use during the water restriction periods using the water savings calculated by these four methods. University of Western Sydney Page 79

117 CHAPTER 4: Impact of water restriction Modelling was done by the three forms of multiple regression techniques: Raw-Data, Semi-Log and Log-Log models. It was found that Semi-Log model coupled with the YBDM method of water savings calculation provided more accurate results than the other methods and models that were tested for both the single and multiple dwelling sectors. Moreover, this method provided better simulation results of water use during the water restrictions periods ( ) in comparison to the other methods. This YBDM of water savings calculation offers a greater advantage as the relevant information is readily available to the authorities. Thus, this method can be used by water authorities to quantify water restriction savings. Water savings from water restrictions during the Levels 1, 2 and 3 restrictions periods were found to be 9.13%, 18.10% and 20.09% by the YBDM method, respectively, for the single dwelling sector. For multiple dwelling residential sectors, these values were found to be 3.9%, 7.62% and 8.79% by the YBDM method for Levels 1, 2 and 3 water restrictions, respectively. These results indicate that the water restrictions were effective in reducing water demand during the drought periods in the Blue Mountains region, Australia. It was also found that Levels 2 and 3 restrictions were more stringent than Level 1, as expected. However, effect of water restrictions was found to be lesser in the multiple dwelling sector than that of the single dwelling sector. This may be attributed to the fact that the higher proportion of the supplied water is normally used for outdoor purposes in the case of single dwelling sector and the water use restrictions mainly target outdoor water use. The potential of water restriction savings (WRS) and water conservation savings (WCS) variables to be included as continuous independent variables with numerical representation in the water demand forecasting model was also investigated in this study. Quantitative measurement of monthly WRS and WCS were taken as two separate independent variables along with rainfall, temperature and water price variables in the water demand forecasting models. The model was used to forecast water demand for the period of July 2009 to September It was found that the developed models were capable of forecasting monthly and yearly water demand with a high degree of accuracy for both the single and multiple dwelling sectors in the Blue Mountains region in Australia, which can be used to forecast long term water demand. University of Western Sydney Page 80

118 CHAPTER 5: Water demand forecasting CHAPTER 5 PROBABILISTIC FORECASTING OF LONG TERM URBAN WATER DEMAND This chapter is partial reproduction of the following refereed journal paper: Haque, M.M. 1, Rahman, A. 1, Hagare, D. 1 and Kibria, G Probabilistic water demand forecasting using projected climatic data for Blue Mountains Water Supply System in Australia. Water Resources Management, 28(7), ERA 2010 ranking: B, Impact factor: School of Computing, Engineering and Mathematics, University of Western Sydney, Australia 2 Sydney Catchment Authority, Penrith, Australia Abstract Long term water demand forecasting is needed for the efficient planning and management of water supply systems. A Monte Carlo simulation approach is adopted in this chapter to quantify the uncertainties in long term water demand prediction due to the stochastic nature of independent variables and their correlation structures. Three future climatic scenarios (A1B, A2 and B1) and four different levels of water restrictions are considered in the demand forecasting for single and multiple dwelling sectors in the Blue Mountains region, Australia. It is found that future water demand in 2040 would rise by 2% to 33% (median rise by 11%) and 72% to 94% (median rise by 84%) for the single and multiple dwelling sectors, respectively under different climatic and water restriction scenarios in comparison to water demand in 2010 (base year). From the 90% confidence intervals of the forecasted values, it is found that the uncertainty band varies about 11 to 13% and 6% around the median forecasted demand for the single and multiple dwelling sectors, respectively. It is found that the increase in future water demand is not notably affected by the projected climatic conditions but by an increase in the dwelling numbers in future i.e. an increase in the total population. The modelling approach presented in this paper can provide realistic scenarios of forecasted water University of Western Sydney Page 81

119 CHAPTER 5: Water demand forecasting demands which would assist water authorities in devising appropriate management strategies to enhance the resilience of the water supply systems. The developed method can be adapted to other water supply systems in Australia and other countries. University of Western Sydney Page 82

120 CHAPTER 5: Water demand forecasting 5.1 Overview Chapter 4 has discussed the impact of water restriction on urban water demand and developed a methodology to estimate water savings from water restriction. This chapter estimates a range of long term water demands in the Blue Mountains region for the period of 2015 to 2040 in a probabilistic fashion. It commences with presenting the probabilistic method adopted to forecast long term water demand considering the stochastic nature of the independent variables and their correlation structures. It is followed by the results and discussion, and then summary of the findings. 5.2 Methodology In this chapter, future water demand for the Blue Mountains water supply system (BMWSS) was estimated for time period for both the single and multiple dwelling sectors by adopting a Monte Carlo simulation technique. Monte Carlo simulation is a well-established method which plays an important role in many scientific applications and has been widely used to evaluate overall uncertainty in modelling and forecasting tasks (Rahman et al. 2002, Nash and Hannah 2011). It is a method for iteratively evaluating a deterministic model using sets of random numbers as inputs and producing a range of possible outcomes allowing for better decision making under uncertainty. Using three different future climatic scenarios (A1B, A2 and B1) (discussed in Chapter 2 in Section ) and four possible water restriction conditions (No water restriction, Level 1, Level 2 and Level 3) (discussed in Chapter 2 in Section 2.5), 12 possible water demand scenarios were simulated. In the generated 12 scenarios, water usage price and water conservation savings variables were kept the same. The uncertainty in the forecasted demand was expressed by developing 90% confidence intervals from the generated 10,000 forecasts. From the 90% confidence intervals, it can be interpreted that 90% of all the possible forecasts would fall within this interval for a given forecast year. Finally, the forecasted demands were compared with the observed water demand in 2010 (considered as base water demand in this chapter) to get the estimates of relative changes. In this chapter, estimation of future water demand scenarios in a probabilistic way was done in four steps: University of Western Sydney Page 83

121 CHAPTER 5: Water demand forecasting 1) First, a deterministic long term water demand forecasting model was developed for both the single and multiple dwelling sectors, separately. 2) Then, the plausible future values of the independent variables were estimated. 3) Afterwards, a multivariate normal distribution (MVN) (described in Section 5.2.1) was adopted to generate stochastic independent variables data maintaining the correlation structure among the independent variables. In applying the MVN, it was assumed that each of the independent variables data can be described by a univariate normal distribution. The pair-wise correlations of the independent variables to be used in the MVN were obtained from the observed independent variables data set. 4) Thereafter, a Monte Carlo simulation was carried out to obtain the distribution of total water demand for the forecasted period ( ). A total of 10,000 simulations per scenario were undertaken. The simulation was carried out in two steps: (i) first, per dwelling monthly water demands (10,000 values) were estimated from the generated independent variables; and (ii) then, estimated per dwelling monthly demands were multiplied by the projected values of the monthly dwellings (10,000 values) to get the monthly demands. The methodology developed to estimate future water demand scenarios using a probabilistic method is illustrated in Figure 5.1. The deterministic models for both the single and multiple dwelling sectors were developed and discussed in Chapter 4, which were the Semi-Log multiple regression models. The independent variables were monthly total rainfall (mm), monthly mean maximum temperature ( 0 C), water usage price ($/kl), water conservation savings (kl/dwelling/month) and water restriction savings (kl/dwelling/month) and the dependent variable was per dwelling monthly water consumption (kl/dwelling/month) in the developed deterministic water demand forecasting model (discussed in Chapter 4 in Section 4.5). University of Western Sydney Page 84

122 CHAPTER 5: Water demand forecasting In order to forecast future water demand, the plausible future values of the independent variables are needed. In this chapter, population growth is considered in the modelling through the growth in the number of dwellings in future. It is assumed that lifestyle of residents would remain unchanged in the forecast period. Number of single and multiple dwellings were estimated for the period of in the BMWSS based on the monthly growth rate found in the dwellings data during the period (discussed in Chapter 3 in Section 3.6.3). As mentioned in Chapter 3 in Sections and 3.6.4, the future values of water usage price and number of participating dwelling in water conservation programs were estimated based on the growth rate found in the observed data. Thereafter, per dwelling monthly water savings from water conservation programs were estimated by the method described in Chapter 4 (Section 4.4). In this study, four different water restriction conditions were considered being Level 1, Level 2, Level 3 and no restrictions in the water demand forecasting. Per dwelling water savings due to imposed water restrictions were estimated by the method mentioned in Chapter 4 (Section 4.4). The average per dwelling water restriction savings for the single and multiple dwelling sectors for different levels of water restrictions were used in the forecasting period. Projections of future climate are needed to estimate future water demand under various plausible climatic conditions. Due to highly uncertain future emission growth, a series of potential greenhouse gas emission scenarios were developed by the Intergovernmental Panel on Climate Change (IPCC) (described in Chapter 2 in Section ). In this study, future water demand values were estimated under three future climate scenarios being B1, A1B and A2, which represent low, medium and high future emission scenarios, respectively. Climate projections by CSIRO Mark 3.0 global climate model (GCM) were used in this study. The statistically downscaled temperature and rainfall data under three emission scenarios were taken as input to the water demand forecasting model to estimate the future water demand scenarios. These downscaled future climatic data of Katoomba weather station were obtained from Sydney Catchment Authority for the period of 2021 to In this chapter, water demand was forecasted by the Monte Carlo simulation technique for the period 2021 to Projection of water demand for 2015 was estimated by interpolation method using the observed demand from 2000 to 2010 and predicted University of Western Sydney Page 85

123 CHAPTER 5: Water demand forecasting demand from 2021 to 2040, because the future meteorological data during 2014 to 2020 were not available during the study. Develop a long term deterministic water demand forecasting model Estimate plausible future values of the predictor variables Generate 12 scenarios using 3 future climate conditions (B1, A1B and A2) and 4 (No, L1, L2 and L3) water restriction conditions (B1-No; B1-L1; B1-L2; B1-L3; A1B-No; A1B-L1; A1B-L2; A1B-L3; A2-No; A2-L1; A2-L2 and A2-L3) Apply multivariate normal distribution to generate stochastic predictor variables data for each scenario Initiate Monte Carlo simulation to generate 10,000 per dwelling demand forecasts for each scenario Estimate monthly demands by multiplying generated 10,000 per dwelling demand values with the projected values of monthly dwelling (10,000 values) Develop 90% confidence intervals from the generated 10,000 demand forecasts Figure 5.1 Framework of estimating future water demand scenarios adopting a probabilistic method Multivariate normal distribution The multivariate normal distribution is a generalization of the one dimensional (univariate) normal distribution to higher dimensions (i.e. two or more variables). One of the key characteristics of the multivariate normal distribution is that it can capture correlations between different random variables which is a crucial factor for many simulations. It is a distribution for random vectors of correlated variables when every linear combination of its components has a univariate normal distribution. The multivariate normal distribution is often used to model the correlations among the University of Western Sydney Page 86

124 CHAPTER 5: Water demand forecasting stochastic time series and can be adopted to explore the effects of these correlations in Monte Carlo simulations (Thomas and Luk 2008). The multivariate normal distribution has N normal distributions (or N dimensions) within itself. It is parameterized with two elements: a mean vector, μ, and a variancecovariance matrix,. The diagonal elements of the matrix, contains the variances for each variable, while the off-diagonal elements of contain the covariances between the variables. If x ε R be a random variable, the univariate normal distribution x ~ N(μ, σ 2 ) can be described by its Probability Density Function (PDF): p(x) = 1 (x μ) 2 2πσ 2 e 2σ 2 (5.1) where p(x) is the probability density function of x, μ and σ 2 are the mean and variance of x, respectively. In the case of a multivariate normal distribution, let x ε R N, then the distribution x ~ N(, ) can be described by the PDF of a vector of length N: T 1 ( x) 2 ( x) 1 p ( x) e / 2 1/ 2 (5.2) N (2 ) where p(x) is the probability density function of x, x is the vector of x N values, μ is the vector of the means of the N distributions, T represents the transpose of the matrix, the -1 represents the inverse of the matrix, denotes the variance-covariance matrix and denotes the determinants of the variance-covariance matrix. Covariance for two variables x i and x j can be defined as: Cov(x i, x j ) = σ ij = ρ ij σ i σ j (5.3) where σ = var(x) and ρ ij is the correlation coefficients between x i and x j. University of Western Sydney Page 87

125 CHAPTER 5: Water demand forecasting University of Western Sydney Page 88 Subsequently, the variance-covariance matrix is defined as follows: N N N N N N N N N (5.4) The required multivariate normal random samples x N are generated by multiplying a vector containing univariate normal random numbers, r~n(0,1) with the lower triangular matrix A where T AA, to achieve the desired correlation structure. The mean of the components are then adjusted by adding the vector. Hence, the generation of the N-th vector x N can be calculated as follows (Barr and Slezak 1972): x N = Ar N + μ (5.5) After expanding equation 5.5, the structure of the computation becomes as follows: N N N N N N N r r r a a a a a a x x x ,,2,1 2,2 2,1 1,1 2 1 (5.6) 5.3 Results Water demand forecasting results in the single dwelling sector The 50 th percentile of the forecasted water demand from the Monte Carlo simulation is presented in Table 5.1 for the single dwelling sector. As can be seen in Table 5.1, for A1B climatic scenario and under different restriction levels, forecasted water demands vary from 2.66 to 3.39 GL/year in In comparison to water demand in 2010 (base year) (2.55 GL/year), the water demand is expected to rise by 4 to 33% in 2040 under A1B climatic scenario and different restriction levels (i.e. under no water restriction and Levels 1, 2 & 3 water restrictions conditions water demand is

126 CHAPTER 5: Water demand forecasting expected to rise by 33, 18, 7 and 4%). For A2 climatic scenario under different restriction levels, forecasted water demands would be in the range of 2.63 and 3.34 GL/year in 2040, which is 3 to 31% rise as compared to the base year. The water demand is expected to rise by 2 to 30% in 2040 under B1 climatic scenario and different restriction levels as compared to the water demand in 2010 as forecasted water demand ranges 2.60 to 3.31 GL/year. Across the twelve scenarios, the rise in water demand in 2040 was found to be in the range of 2 to 33% with a median rise of 11% in comparison to water demand in Though water demand characteristics generally vary from city to city across different countries, the results obtained in this study are found to be quite comparable with other similar studies. For example, Babel et al. (2007) predicted a 20% increase in water demand in Kathmandu, Nepal by 2015 as compared with Mohamed and Al-Mualla (2010) predicted that water demand would rise by 50% in 2020 in Umm Al-Quwain, UAE. Dziegielewski and Chowdhury (2011) found that water demand would rise by 36 54% in 2050 in North-eastern Illinois, USA. As can be seen in Table 5.1, the effect of water restriction on future demand is rather more important than various climatic scenarios. For example, the projected water demands in 2040 were found to be 3.39, 3.34 and 3.31 GL/year for A1B, A2 and B1 climatic scenarios, respectively under no water restriction (representing 29 to 33% increase in forecasted demands as compared with the base year, 2010). On the other hand, the forecasted water demands in 2040 were found to be 3.39, 3.00, 2.72 and 2.66 GL/year for no restriction and Levels 1, 2 & 3 restrictions, respectively under A1B climatic scenarios (representing 4 to 33% increase in the forecasted demands as compared with the base year 2010). Hence, it can be stated that the variations in the forecasted water demands are much higher due to the different levels of water restrictions than those for different climatic scenarios. A similar result was noted by Khatri and Vairavamoorthy (2009) who showed that future climatic scenarios would have a minimal impact on future water demand in Birmingham, UK. Also, Slavíková et al. (2013) found that the future climatic scenarios would not have any significant effect in explaining water demand variability in two municipalities located in Central Bohemia, Czech Republic. University of Western Sydney Page 89

127 CHAPTER 5: Water demand forecasting Table th percentile (most expected value) of the forecasted water demand values for the single dwelling sector in the Blue Mountains region in the period of under twelve water demand scenarios Climate scenario: A1B (50 th percentile water demand in GL/year) No Restriction Level 1 Level 2 Level Climate scenario: A2 No Restriction Level 1 Level 2 Level Climate scenario: B1 No Restriction Level 1 Level 2 Level The 90% confidence intervals of forecasted total yearly water demands for A1B climatic scenario for the single dwelling sector under different restriction levels are presented in Figures 5.2 (a-d). Similar results were obtained for the A2 and B1 climatic scenario that are presented in Figures B.5.1 (a-d) and B.5.2 (a-d) in Appendix B. As can be seen in Figures 5.2 (a-d), there is 90% possibility that water demands would be in the range of 2.8 GL to 3.2 GL, 2.6 GL to 2.9 GL, 2.5 GL to 2.8 GL and 3.2 GL to 3.6 GL in 2040 under Levels 1, 2 & 3 and no restriction conditions, respectively for A1B climatic scenario. From the forecasted results of all of the 12 scenarios, uncertainty bands were found to be in the range of 0.3 to 0.4 University of Western Sydney Page 90

128 Yearly Water Demand in ML/year Climate change impact on water demand and supply CHAPTER 5: Water demand forecasting GL/year in 2040, which represent 11 to 13% variation around the median forecasted demand. On the contrary, the deterministic model predicted a single water demand value. For example, the forecasted water demands in 2040 from the deterministic models were found to be 3.05, 2.98 and 2.95 GL/year under Level 1 restriction and A1B, A2 & B1 climatic conditions (representing median 17% increase from the base year, 2010). Under the same conditions, the Monte Carlo simulation predicted an increase in water demand in 2040 by 9 to 25% as opposed to a fixed increase by 17% by the deterministic model. Since the Monte Carlo simulation accounts for the stochastic nature of the independent variables and their correlation structures, it provides a more realistic demand prediction under different plausible conditions that might arise in future th Percentile Year A1B-No Restriction Figure 5.2(a) 90% confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for A1B climate scenario and no water restriction condition for the single dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) University of Western Sydney Page 91

129 Yearly Water Demand in ML/year Yearly Water Demand in ML/year Climate change impact on water demand and supply CHAPTER 5: Water demand forecasting th Percentile Year A1B-Level 1 Figure 5.2(b) 90% confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for A1B climate scenario and Level 1 water restriction condition for the single dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) th Percentile Year A1B-Level 2 Figure 5.2(c) 90% confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for A1B climate scenario and Level 2 water restriction condition for the single dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) University of Western Sydney Page 92

130 Yearly Water Demand in ML/year Climate change impact on water demand and supply CHAPTER 5: Water demand forecasting th Percentile Year A1B-Level 3 Figure 5.2(d) 90% confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for A1B climate scenario and Level 3 water restriction condition for the single dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) It is also found that the highest forecasted value of water demand for single dwelling sector is 3.6 GL in 2040 under no water restriction condition and A1B climatic scenario. The lowest forecasted value is 2.4 GL in 2040 under Level 3 restrictions and B1 climatic scenario. As with more strict water restriction, the residents will tend to conserve more water, and hence smaller value of forecasted water demand was obtained under Level 3 water restriction. Moreover, B1 climatic scenario is generally considered to be low impact emission scenario by the IPCC that has been reflected in the lowest value of forecasted demand Water demand forecasting results in the multiple dwelling sector The 50 th percentile of the forecasted water demands in the multiple dwelling sector from the Monte Carlo simulation under twelve water demand scenarios are presented in Table 5.2. As can be seen in Table 5.2, forecasted water demands vary 0.31 to 0.35 GL/year in 2040 for A1B climatic scenario under different restriction levels. For both the A2 and B1 climatic scenarios, forecasted water demand in 2040 also falls in the range of 0.31 and 0.35 GL/year. In comparison to water demand in 2010 (the University of Western Sydney Page 93

131 CHAPTER 5: Water demand forecasting base year) (0.18 GL/year), the water demand is expected to rise by 72 to 94% in 2040 for the twelve scenarios with a median rise of 84%. As can be seen in Table 5.2, there is no remarkable variation in the forecasted values for all of the scenarios in any year for multiple dwelling sector since water demand of this sector is very low in comparison to water demand of single dwelling sector. Moreover, multiple dwellings normally consume less water in outdoor purpose due to smaller outdoor area therefore effect of climate variables are likely to be less in this sector than single dwelling sector. In addition, effect of water restriction is less in the multiple dwelling sector than single dwelling sector as discussed in Chapter 4 (Section 4.4). The 90% confidence intervals of forecasted total yearly water demand for A1B climatic scenario under different restriction levels for multiple dwelling sector are presented in Figures 5.3 (a-d). Similar results were obtained for other scenarios; the graphs of confidence intervals for A2 and B1 climatic scenarios under different restriction levels are presented in Figures B.5.3 (a-d) and B.5.4 (a-d) in Appendix B. As can be seen in Figures 5.3 (a-d), there is 90% possibility that water demands would be in between 323 and 344 ML/year, 308 and 328 ML/year, 304 and 323 ML/year, and 340 and 362ML/year in 2040 under Level 1, Level 2, Level 3 and no restriction conditions, respectively for A1B climatic scenario. From the forecasted results of the 12 water demand scenarios, uncertainty bands were found to be about 20 ML/year in 2040, which represent 6% variation around the median forecasted demand. From the confidence intervals of all the scenarios, the highest forecasted value of water demand for multiple dwelling sector was found to be 362 ML/year in 2040 under no water restriction condition and A1B climatic scenario, and the lowest forecasted value were found to be 300 ML/year in 2040 under Level 3 restriction and B1 climatic scenario. The estimated increases in the forecasted total monthly water demand for both the single and multiple dwelling sectors were found to be mainly associated with an increase in the dwelling numbers in future i.e. the increase in population. For example, 50 th percentile of the forecasted per dwelling monthly water demands in 2040 (average across the twelve months) under Level 1 water restriction and A1B, A2 and B1 climatic scenarios were found to be 11.73, and kl/dwelling/month for single dwelling sector, representing an average across the University of Western Sydney Page 94

132 CHAPTER 5: Water demand forecasting three scenarios of kl/dwelling/month. In comparison to the observed average water demand (11.65 kl/dwelling/month) in 2010, these results indicate no increase in per dwelling monthly demand in Table th percentile (most expected value) of the forecasted water demand values for the multiple dwelling sector in the Blue Mountains region in the period of under twelve water demand scenarios Climate scenario: A1B (50 th percentile water demand in GL/year) No Restriction Level 1 Level 2 Level Climate scenario: A2 No Restriction Level 1 Level 2 Level Climate scenario: B1 No Restriction Level 1 Level 2 Level University of Western Sydney Page 95

133 Yearly Water Demand in ML/year Yearly Water Demand in ML/year Climate change impact on water demand and supply CHAPTER 5: Water demand forecasting th Percentile Year A1B-No Restriction Figure 5.3(a) 90% confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for A1B climate scenario and no water restriction condition for the multiple dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) th Percentile Year A1B-Level 1 Figure 5.3(b) 90% confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for A1B climate scenario and Level 1 water restriction condition for the multiple dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) University of Western Sydney Page 96

134 Yearly Water Demand in ML Yearly Water Demand in ML Climate change impact on water demand and supply CHAPTER 5: Water demand forecasting th Percentile Year A1B-Level 2 Figure 5.3(c) 90% confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for A1B climate scenario and Level 2 water restriction condition for the multiple dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) th Percentile Year A1B-Level 3 Figure 5.3(d) 90% confidence intervals and 50 th percentile of the forecasted total yearly water demands from 2015 to 2040 for A1B climate scenario and Level 3 water restriction condition for the multiple dwelling sector in the Blue Mountains region (grey area in the plot refers to the 90% confidence band) University of Western Sydney Page 97

135 CHAPTER 5: Water demand forecasting However, total water demand in single dwelling sector in 2040 was found to be about 3 GL/year and total observed water demand in 2010 was found to be 2.55 GL/year. These results indicate about 17.6% increase in water demand in 2040 in comparison to 2010 water demand value. This increments is likely to be associated with the increment in dwelling number as in 2010 total number of single dwelling was and the forecasted total single dwelling number in 2040 was found to be (based on the growth rate found in the observed dwelling data as discussed in Chapter 3 in Section 3.6.3), representing an increase of about 28%. In addition, from the demand equations (4.12 and 4.13) in Chapter 4, it was found that rainfall had a decreasing effect on the demand, temperature had an increasing effect, and water usage price, water conservation and water restrictions had decreasing effects on water demand. This result indicate that future water demand in the BMWSS would not be significantly affected by the projected climatic conditions as the increasing effects of the climatic variables on water demand are likely to be minimized by the decreasing effects of increasing water usage price and savings variables (Impact of climate variables on water demand is presented in Chapter 6 in more details). As discussed in the earlier, median rise in water demand is expected to be around 11% and 84% for the single and multiple dwelling sectors, respectively in comparison to water demand in These results indicate that growth rate in water demand in the multiple dwelling sector is much higher than that of single dwelling. This higher demand growth rate in the multiple dwelling sector is happened due to the higher dwelling number growth rate found in the multiple dwelling sector than the single dwelling sector. In 2040, the forecasted total number of multiple dwelling was found to be 38,547 (based on the growth rate found in the observed dwelling data as discussed in Chapter 3 in Section 3.6.3), and the total observed multiple dwelling numbers was 20,406. These values represent an increase of dwelling number by 89% in 2040 in comparison to On the other hand, growth in single dwelling sector was found to be 28% in 2040 in comparison to 2010 as mentioned earlier, which is quite small in comparison to the growth rate of multiple dwelling sector. However, the composition of water consumption in single and multiple dwelling sectors was found to remain same as found in the water consumption data (i.e. 94% and 6% water consumption in single and multiple dwelling sectors, respectively as mentioned in Chapter 3 in Section 3.6.1). For example, water University of Western Sydney Page 98

136 CHAPTER 5: Water demand forecasting demand in single and multiple dwelling sector was found to be 3.39 GL/year and 0.35 GL/year in 2040 under no restriction and A1B climate conditions (Tables 5.1 and 5.2). These values indicate that single and multiple dwelling sector would consume around 91% and 9% of total residential water in 2040, respectively, which is quite comparable to the composition in Summary This chapter develops a methodology to incorporate uncertainty in the independent variables and various climatic scenarios explicitly into the water demand forecasting model using a Monte Carlo simulation technique. The method is applied to the Blue Mountains water supply system in New South Wales, Australia. Using five independent variables, three different future climatic scenarios (i.e. A1B, A2 and B1) and four different water restriction conditions (Levels 1, 2 & 3 and no restriction), twelve different plausible scenarios are generated. It is found that the median water demand for the Blue Mountains water supply system in 2040 is expected to rise by 2 to 33%, and 72 to 94% for the single and multiple dwelling sectors, respectively in comparison to water demand in 2010 (base year). It is also found that growth rate of water demand in multiple dwelling sector is much higher than the single dwelling sector as growth of multiple dwelling sector is higher than the single dwelling sector in the study area. Forecasted water demand values are found to be the highest during no water restriction condition and the lowest during Level 3 water restriction as expected. It is found that the effects of different levels of water restriction conditions in water demand forecasting are more significant than the effects of various climatic scenarios. It is also found that the increase in future water demand is not notably affected by the projected climatic conditions but by the increase in the dwelling numbers in future i.e. the increase in the total population. From the 90% confidence intervals of the forecasted values, it is found that the uncertainty band varies about 11 to 13% and 6% around the median forecasted demand for single and multiple dwelling sectors, respectively. The highest and lowest forecasted water demands for both the single and multiple dwelling sector are found to be for A1B - no restriction and B1 - Level 3 restriction scenario, respectively. University of Western Sydney Page 99

137 CHAPTER 5: Water demand forecasting The probabilistic modelling approach considering the correlation structures of the independent variables presented in this chapter can provide a range of realistic scenarios of forecasted water demands as opposed to a single forecast value by the deterministic model. These ranges of realistic water demand forecasts would assist water authorities in devising appropriate management strategies to enhance the resilience of the water supply systems. The developed method can be adapted to other water supply systems in Australia and other countries. University of Western Sydney Page 100

138 CHAPTER 6: Climate change & demand CHAPTER 6 IMPACT OF CLIMATE CHANGE ON URBAN WATER DEMAND This chapter is partial reproduction of the following two refereed journal papers: Paper 1 Haque, M.M. 1, Egodawatta, P. 2, Rahman, A. 1 and Goonetilleke, A Assessing the significance of climate and community factors on urban water demand, Urban Water Journal, under review (ERA 2010 ranking: C, Impact factor: 0.91). 1 School of Computing, Engineering and Mathematics, University of Western Sydney, Australia 2 Science and Engineering Faculty, Queensland University of Technology, Australia Paper 2 Haque, M.M. 1, Rahman, A. 1, Hagare, D. 1 and Kibria, G Impact of climate change on future water demand: A case study for the Blue Mountains Water Supply System, Water, 41(1), (ERA 2010 ranking: C). 1 School of Computing, Engineering and Mathematics, University of Western Sydney, Australia 2 Sydney Catchment Authority, Penrith, Australia Abstract Ensuring adequate water supply to urban areas is a challenging task due to factors such as rapid urban growth, increasing water demand and climate change related impacts. In developing a sustainable water supply system, it is important to identify the dominant water demand factors for any given water supply scheme. This chapter applies principal component analysis to identify the factors that dominate residential water demand using the Blue Mountains Water Supply System in New South Wales University of Western Sydney Page 101

139 CHAPTER 6: Climate change & demand (NSW), Australia. The results show that the influence of community factors (e.g. use of water efficient appliances and rainwater tanks) on water demand are among the most significant. The result also confirmed that the community programs and water pricing policy together can play a noticeable role in reducing the overall water demand. On the other hand, influence of rainfall on water demand is found to be very limited, while temperature shows some degree of correlation with water demand. The results also demonstrate that the water demand is mostly influenced by community factors rather than the climate factors. This chapter also evaluates the impact of climate change on residential water demand in the Blue Mountains Water Supply System in NSW, Australia. Forecasting is done by a long term water demand forecasting model developed using a multiple linear regression technique for the period of Here, three climatic scenarios are considered during the water demand forecasting including B1 (low), A1B (medium) and A2 (high) scenarios. The results suggest that the future climate will have a minor impact on future water demand in the study area. However, water demand projections show an increasing trend, which is mainly attributed to the rise in the number of households (dwellings) due to the increasing population in the area. University of Western Sydney Page 102

140 CHAPTER 6: Climate change & demand 6.1 Overview Chapter 5 has developed a probabilistic long term water demand forecasting model to incorporate stochastic nature of the independent variables and their correlation structures in the water demand forecasting. This chapter evaluates the relative influence of climate variables on urban water demand and assesses the impacts of plausible climate change on future water demand. It commences with presenting the principal component analysis adopted to evaluate the relative influence of the governing variables on urban water demand. It then presents the methods adopted to quantify the impact of climate change on future water demand. This is followed by the results and discussion, and then summary of the findings. This chapter is based on two journal papers, paper 1 is linked to evaluate the relative influences of the climate variables on water demand and paper 2 is linked to assess the impacts of climate change on future water demand. 6.2 Methodology In this chapter, both qualitative and quantitative analyses were conducted to estimate the climate change impact on urban water demand. Qualitative analysis was conducted to evaluate the relative importance of the climate variables on urban water demand adopting principal component analysis (PCA). Quantitative assessment was conducted to estimate the impact of climate change on future water demand by forecasting water demand using projected climate variables by CSIRO Mk. 3 global climate model (GCM) and three different hypothetical climate change scenarios Principal component analysis PCA is a popular pattern recognition technique which is capable of illustrating correlations among variables and clusters of objects in a graphical format. The PCA transforms the original data set of n factors to a new data set containing n number of orthogonal principal components (PCs). The PCs are linear functions of the original factors. Though the number of PCs is same as original variables, PCA transforms PCs such that most of the useful data variances are explained by the first few PCs. Hence, the first few PCs can be selected for interpretations reducing the number of variables without losing much information contained in the original data set (Mahbub et al. 2010). University of Western Sydney Page 103

141 CHAPTER 6: Climate change & demand When a PCA is applied to a data matrix, it generates a loading value for each of the variables and a score for each object on the PCs. Therefore, data can be presented graphically by plotting the loading value in the form of a vector and the score in the form of a data point. This kind of plot is generally termed as a biplot (Gabriel 1971, Gabriel and Odoroff 1990). In this chapter, PCA was conducted on the data matrix containing the water demand variables: monthly total rainfall, numbers of rain days in a month, monthly mean maximum temperature, monthly total evaporation, monthly mean global solar exposure, water conservation savings, water restriction savings and water price (Table 6.1). The variables are represented by the vectors, and the consumption months are represented by the points in the resulting biplot. Degree of correlation between the variables can be explained by the angle between the vectors and the objects with similar characteristics can be indicated by the clustered data points in a biplot. A small angle between two vectors indicates that the variables are highly correlated with each other, and they represent similar behaviour. Two vectors with opposite direction indicate that they are highly correlated in an inverse way. If the two vectors are perpendicular, they are considered as independent to each other (i.e. no correlation between them) (Gardner et al. 2005, Blasius et al. 2009). In this study, statistixl (Robert and Wither 2007) software was used to perform the PCA Impact of climate change on urban water demand In this chapter, assessment of the impacts of climate change on urban water demand was conducted in four steps as outlined below and illustrated in Figure First, a set of future climate scenarios were developed to be taken as input into the water demand forecasting models. The generated future climate scenarios were as below: a. Three future climate scenarios projected by the CSIRO Mk. 3 GCM (discussed in Chapter 3 in Section 3.6.5) under three emission scenarios (A1B, A2 and B1). b. A current climate scenario which was obtained from the average of monthly rainfall and monthly maximum temperature during the period of University of Western Sydney Page 104

142 CHAPTER 6: Climate change & demand c. Three hypothetical future climate scenarios: (i) 1 0 C rise in temperature and 10% decrease in rainfall, (ii) 2 0 C rise in temperature and 20% decrease in rainfall, (iii) 3 0 C rise in temperature and 30% decrease in rainfall from current climate conditions. 2. Afterwards, future water demands were estimated in the single and multiple dwelling sector separately for the period of using the water demand forecasting models developed in Chapter 4 (i.e. Semi- Log multiple regression model). In the water demand forecasting models, generated future climate scenarios were taken as input along with water price and water conservation savings variables. Water demand was forecasted under no restriction conditions. 3. Thereafter, projections of water demand under generated future climate scenarios were compared with the projection of water demand under current climate conditions to estimate the relative impacts of climate change on urban water demand. Water demand model (Developed in Ch.4) Future values of predictor variables (Estimated in Ch.3) Forecast water demand using projected climate variables by CSIRO Mk. 3 Global Climate Model under three emission scenarios (A1B, A2 and B1) (Conducted in this chapter) Forecast water demand using current climate (average of ) (Conducted in this chapter) Forecast water demand using three hypothetical climate change scenarios: (i) 1 0 C T rise; 10 % R decrease, (ii) 2 0 C T rise; 20% R decrease, (iii) 3 0 C T rise; 30% R decrease (Conducted in this chapter) Estimate relative changes in forecasted water demand due to climate change by comparing forecasted water demand under different scenarios with the forecasted demand under current climate condition (Conducted in this chapter) Figure 6.1 Framework of the climate change impact assessment on urban water demand ( T refers to monthly maximum temperature and R refers to monthly total rainfall) University of Western Sydney Page 105

143 CHAPTER 6: Climate change & demand Table 6.1 Description of the dependent and independent variables used in the Principal Component Analysis (PCA) Variables Description Unit Dependent variable PDWC Per dwelling water consumption in a month kl/dwelling/month Independent variables RF Monthly total rainfall mm NRD Number of rain days in a month Day MMT Monthly mean maximum temperature 0 C EVP Monthly total evaporation mm SE Monthly mean global solar exposure MJ/m 2 WCS Water conservation savings kl/dwelling/month WRS Water restriction savings kl/dwelling/month WP Water price AUD/kL 6.3 Results Relative influence of variables on urban water demand In this study, principal component analysis (PCA) was undertaken to assess qualitatively the influence of the independent variables (Table 6.1) on the residential water demand comprising of the single and multiple dwelling sector. Analysis was first performed to assess the correlations of the likely influential variables on water demand. For this, PCA was initially performed using a data matrix of 8 independent variables covering a 14 year period (1997 to 2011). The resulting biplot is shown in Figure 6.2 (PC 1 vs. PC 2) that explains 71.2% of the data variance suggesting its sufficiency to interpret the water demand variables. Moreover, objects in Figure 6.2 did not show any obvious clustering and were spread uniformly across the biplot. University of Western Sydney Page 106

144 PC 2 (34.1%) Climate change impact on water demand and supply CHAPTER 6: Climate change & demand Thereafter, the dependent variable (PDWC) was introduced into the data matrix to observe the relationship between the PDWC and the water demand variables. The resulting PCA biplot on the updated data matrix is presented in Figure 6.3. It was found that the inclusion of the dependent variable had minimal influence on the clustering of the data set. Moreover, it did not adversely affect the percentage of variance explained by the two PCs that indicated Figure 6.3 could be used for direct interpretation of the influence of water demand variables on PDWC _1 10_1 MMT 09_12 10_12 10_2 09_1 09_2 09_11 11_2 10_11 05_1 08_1 07_1 07_11 11_3 08_12 10_3 04_1 NRD 08_11 10_10 03_1 02_1 09_10 01_12 99_1 01_1 03_12 04_12 06_2 06_1 RF 07_12 07_2 04_2 09_3 08_2 00_12 02_11 05_11 03_11 05_2 98_2 98_1 02_12 06_11 06_12 08_10 08_3 07_3 05_12 04_11 97_12 04_10 11_4 WRS 97_11 98_12 02_2 05_10 07_10 06_3 11_9 09_4 03_2 97_1 00_1 01_2 00_2 97_2 00_3 06_10 08_9 10_4 99_12 01_3 09_9 02_10 99_2 04_3 00_11 98_3 97_10 01_10 99_11 01_11 03_3 02_3 05_3 07_4 08_4 10_9 06_9 09_5 10_8 99_3 98_1199_10 03_10 07_9 10_5 11_5 98_10 11_8 11_6 04_9 05_4 05_9 09_8 10_7 97_3 00_10 06_4 07_8 07_610_6 09_6 02_9 03_9 01_4 01_9 04_4 11_7 07_5 00_9 98_4 99_4 02_4 03_4 08_508_8 09_7 06_8 97_9 98_9 98_8 04_8 08_7 08_6 99_9 00_4 05_8 06_5 97_4 03_5 05_6 98_5 05_5 07_7 06_7 99_8 97_5 02_8 01_8 01_5 00_5 02_5 03_8 04_5 05_7 06_6 04_604_7 97_8 99_5 00_8 01_7 98_7 99_703_6 01_6 00_7 03_7 97_6 98_6 97_7 02_7 99_600_6 02_6 EVP SE WP WCS PC 1 (37.1%) Figure 6.2 Resulting PCA biplot (PC 1 vs. PC 2) of variables (8 independent variables) influencing water demand, (Data labels (e.g.08_12; Year_Month) indicate the corresponding year and month in the data matrix) As evident in Figure 6.3, MMT, EVP and SE were found to be highly correlated with each other as the angles between the vectors of these variables were small. In addition, vectors of RF and NRD variables were found to be closer to each other which indicated a high correlation between them. Presence of such highly correlated factors among the similar kinds of variables may create multicollinearity problem in University of Western Sydney Page 107

145 PC 2 (30.1%) Climate change impact on water demand and supply CHAPTER 6: Climate change & demand the regression-based model, leading to unrealistic and biased results. To avoid the multicollinearity problem, one variable from each set was selected for further analysis. In this regard, MMT was retained, as the temperature variable can be easily measured and monitored, and the other two variables (SE and EVP) were removed from the data matrix. NRD was removed from the data matrix, and RF was retained _1 11_1 MMT SE 09_12 10_12 10_2 09_1 09_2 09_11 05_1 08_1 11_2 10_11 EVP 07_1 07_11 11_3 08_12 04_1 10_3 NRD 08_11 07_2 02_1 03_1 09_10 10_10 99_1 01_12 01_1 03_12 04_12 04_2 06_2 06_1 07_12 02_11 05_11 08_10 09_3 08_2 RF 00_12 03_11 98_102_12 06_11 05_2 05_12 06_12 08_3 07_3 04_11 97_12 98_12 97_11 98_2 04_10 02_203_2 06_3 11_4 05_10 07_10 97_1 00_1 01_2 11_9 09_4 00_2 97_2 00_306_10 99_12 99_2 01_3 08_9 10_4 02_10 00_11 03_3 09_9 04_3 98_3 02_3 05_3 10_9 07_408_4 97_10 01_10 01_11 99_11 06_9 09_5 10_8 98_1199_3 99_10 03_10 07_9 10_5 98_10 11_5 04_9 05_405_9 09_8 11_8 10_7 PDWC 97_3 00_10 06_4 07_8 07_610_6 02_9 01_9 01_4 03_9 04_4 03_4 09_6 07_5 11_7 00_9 98_4 99_4 02_4 08_5 08_8 09_7 98_9 97_9 98_8 04_8 03_5 06_8 99_9 08_7 00_4 97_4 05_8 06_5 08_6 98_5 03_8 05_6 99_8 05_5 06_7 07_7 97_5 02_8 01_8 02_5 01_5 00_5 04_5 05_7 06_6 04_6 97_8 99_5 00_8 01_7 03_604_7 99_7 98_7 03_7 01_6 00_7 97_6 97_7 98_600_6 02_7 02_699_6 WP WCS WRS PC 1 (41.7%) Figure 6.3 Resulting PCA biplot (PC 1 vs. PC 2) on modified data matrix of 9 variables including dependent variable, PDWC (Data labels (e.g.08_12; Year_Month) indicate the corresponding year and month in the data matrix) Though WP, WCS and WRS variables showed a high correlation with each other, no variable was removed from this set as these three variables were coming from three different sources (i.e. WP is related to economic perspective, WCS is related to community participation in saving water and WRS is related to policy issues and drought response plan) that may have important implication in water demand in the future. Due to the removal of highly correlated variables of the similar kinds, the revised data matrix contained only five independent variables (RF, MMT, WP, WCS University of Western Sydney Page 108

146 PC 2 (20.8%) Climate change impact on water demand and supply CHAPTER 6: Climate change & demand and WRS) and was subjected to PCA. The outcomes of this PCA on the revised data matrix demonstrated that the elimination of the correlated variables did not put any adverse impact on the PCA results rather it improved the percentage of variance (77.3%) explained by the first two principal components (Figure 6.4). This results indicated that the refined set of variables (i.e. five variables) were sufficient for further analysis PDWC 00_3 MMT 06_1 09_2 01_2 07_2 07_11 02_2 05_1 01_1 98_1 99_1 97_2 00_11 04_2 08_1 10_1 10_12 11_1 97_1 03_2 02_1 98_2 99_10 98_8 02_3 04_10 05_2 11_3 99_2 01_3 07_1 07_3 09_12 97_12 02_11 02_12 00_12 03_1 97_11 98_12 04_1 03_11 09_1 08_2 03_3 07_12 10_3 10_11 11_2 01_12 03_12 05_12 09_11 08_12 98_399_1200_2 04_12 05_11 07_6 09_4 06_2 09_10 01_11 99_3 06_11 04_11 06_12 09_3 98_4 06_3 09_5 97_398_11 05_3 08_11 08_10 10_10 02_10 98_5 97_1000_1 04_3 05_10 08_3 WP 07_4 01_4 08_4 11_4 06_10 06_9 01_10 99_4 02_4 04_4 05_4 07_10 08_9 11_9 97_4 99_11 97_9 98_1000_10 10_4 99_9 02_9 00_4 03_10 03_4 07_8 01_9 00_9 03_5 99_7 06_4 07_5 08_6 09_9 10_5 10_9 11_6 98_9 02_5 05_9 05_6 07_9 11_8 11_5 97_5 99_5 00_5 04_9 05_5 10_6 02_8 99_8 98_6 01_5 03_9 04_5 06_5 09_8 97_6 04_8 05_7 06_8 06_7 08_5 10_8 03_8 06_6 09_6 10_7 97_8 01_8 98_7 01_6 01_7 97_7 99_6 00_8 02_7 02_6 03_6 05_8 08_7 00_7 08_8 09_7 11_7 00_603_7 04_7 04_6 07_7 RF 10_2 WCS WRS PC 1 (56.5%) Figure 6.4 Resulting PCA biplot (PC 1 vs. PC 2) on modified data matrix of 6 variables (one dependent and five independent variables) after removing highly correlated variables from similar kinds (Data labels (e.g.08_12; Year_Month) indicate the corresponding year and month in the data matrix) As evident in Figure 6.4, the dependent variable, PDWC shows strong negative correlation with WP, WCS and WRS variables as they are in the opposite direction and the angles between them are large (approximately close to ). This suggests a decrease in water consumption with an increase in WP, WCS and WRS values. These strong negative correlations of these variables with water demand indicate that University of Western Sydney Page 109

147 CHAPTER 6: Climate change & demand water price and water saving variables are the most dominant in describing water consumption in the Blue Mountains Water Supply System, and these variables in combination would be able to help in reducing water demand. This conclusion is in accordance with the findings of a recent study in Las Vegas Valley in southern Nevada by Dawadi and Ahmad (2013). They concluded that conservation and water pricing policies in combination would help to reduce water demand and to delay an imminent water shortage in Las Vegas Valley. It can also be seen in Figure 6.4 that climate variables, MMT and RF show weak positive and negative correlations with PDWC, respectively. The angles between the vectors of MMT, RF and PDWC, are comparatively smaller than that of WP, WCS, WRS, and PDWC. This result indicates that the climate variables are less influential on water demand than the water savings and water price variables. RF variable was found to be perpendicular to the PDWC, which indicated that rainfall had no impact on residential water demand. Another climate variable, MMT showed some degree of correlation with the PDWC. Out of these two climate variables, influence of temperature on water demand was found to be greater than that of the rainfall variable, indicating an increase in water demand with an increase in temperature. The results of this study in relation to the influence of climate on water demand are somewhat comparable with other studies. For example, the effect of temperature was found to be important than that of rainfall on water demand in Bangkok and Seoul (Praskievicz and Chang 2009, Babel et al. 2014), while the effect of rainfall was found to be significant than that of temperature in Germany (Schleich and Hillenbrand 2009). Moreover, both the rainfall and temperature were found to be significant in the residential water use in Phoenix, Arizona (Balling Jr and Gober 2007). These results indicate that the drivers of water consumption vary between different geographic areas, highlighting the necessity of finding the influence of water demand variables specific to a given area Impact of climate change on urban water demand Projections of water demand under three climatic scenarios (i.e. A1B, A2 and B1) for the single dwelling sector are presented in Figure 6.5, where it can be seen that water demand projections under the current climate are quite close to the projections under three other future climatic scenarios, which indicate that future water demand University of Western Sydney Page 110

148 Water demand (ML/year) Climate change impact on water demand and supply CHAPTER 6: Climate change & demand conditions in the Blue Mountains regions would not be affected appreciably by the future climate. Average values of the forecasted results of two decades, and and corresponding relative changes in the projections with current climate conditions are presented in Table 6.2. Percentage changes in average decadal projections of for A1B, A2 and B1 scenarios in comparison to the demand projections with the current climate were found to be 0.10%, 0.45% and 0.03%, which implies that the future water demand would be higher due to the changed climatic conditions but the impact would be quite negligible. These small impacts might be associated with the minor changes in the future climate projected by the CSIRO Mk. 3 GCM. The relative changes in the climate conditions (projected by the CSIRO model) of the two future decades ( and ) in comparison to the average of the current climate are presented in Table 6.3. It can be seen that the projected future climate conditions are expected to be changed by a little margin A1B A2 B1 Current cliamte 2200 Year Figure 6.5 Projection of water demand under A1B, A2, B1 and current climate conditions for the period in the single dwelling sector In order to verify the climate change impact on future water demand in the Blue Mountains region, three hypothetical climatic scenarios were also considered to input to the water demand forecasting models. University of Western Sydney Page 111

149 CHAPTER 6: Climate change & demand Table 6.2 Water demand forecasting results of the decades of and in the single dwelling sector (Bracketed results indicate percentage changes in the forecasting results in comparison to the predicted water demand under current climate condition ( )) Description Water demand under "Current climate" condition ML/year Water demand under "A1B" climate scenario ML/year Water demand under "A2" climate scenario ML/year Water demand under "B1" climate scenario ML/year Forecasting decade Forecasting decade (-0.33%) 3108 (0.10%) 3067 (0.56%) 3119 (0.45%) 3069 (0.62%) 3106 (0.03%) Table 6.3 Projection of future climate by the CSIRO Mk. 3 global climate model (GCM) under three emission scenarios (A1B, A2 and B1) of the two decades and (Bracketed results indicate percentage changes in the forecasting temperature and rainfall values in comparison to the observed climate data of ) Current climate ( ) Average maximum temperature ( 0 C) Average rainfall (mm/year) CSIRO Mk. 3 A1B A2 B1 A1B A2 B (-0.05) (2.67) (2.24) 1591 (12.42) 1355 (-4.24) 1288 (-9) (1.68) (3.37) (1.32) 1499 (5.88) 1447 (2.24) 1504 (6.22) University of Western Sydney Page 112

150 CHAPTER 6: Climate change & demand Forecasted water demand results under these three hypothetical climatic scenarios were compared with the forecasted water demand under current climate condition for the year 2040 and the results are presented in Table 6.4. It was found that the changes in the water demand projections were 1.21%, 2.42% and 3.64% for the hypothetical climactic scenario 1, 2 and 3, respectively compared to the water demand projections under the current climate condition. These results also indicate that there would be a minor impact in the water demand conditions due to possible future climatic scenarios in the Blue Mountains region. Similar results were obtained for the multiple dwelling sector in relation to climate change impact on water demand. Projections of water demand for the period of under three different climatic scenarios (i.e. A1B, A2 and B1) and current climate conditions for the multiple dwelling sector are presented in Figure 6.6. Water demand projections with the three climatic scenarios were found to be quite close to the projection under the current climate condition. Moreover, only a little variation was found among the projections with these three climatic scenarios. These results indicate a minor influence of future climate change on water demand in the multiple dwelling sector. Percentage changes in average decadal projections for for A1B, A2 and B1 scenarios (Table 6.5) in comparison to the demand projections with current climate were found to be 0.36%, 0.36% and 0.03%, which also indicate that impact of future climatic conditions on water demand would be negligible. In addition, forecasted results under three hypothetical climate change scenarios (Table 6.6) also support the above findings that impact of climate change would be minimal on water demand. University of Western Sydney Page 113

151 Water demand (ML/year) Climate change impact on water demand and supply CHAPTER 6: Climate change & demand Table 6.4 Projection of future water demand under three hypothetical climate change scenarios in the year, 2040 in the single dwelling sector Description Projection under scenario 1 (+1 0 C T & -10% R) Projection under scenario 2 (+2 0 C T & -20% R) Water demand (ML/year) Year : 2040 % changes in comparison to water demand under current climate Projection under scenario 3 (+3 0 C T & -30% R) Projection under current climate 3130 conditions (avg. of ) * T refers to monthly maximum temperature and R refers to monthly total rainfall A1B A2 B1 Current cliamte 0 Year Figure 6.6 Projection of water demand under A1B, A2, B1 and current climate conditions for the period in the multiple dwelling sector University of Western Sydney Page 114

152 CHAPTER 6: Climate change & demand Table 6.5 Water demand forecasting results of the decades of and in the multiple dwelling sector (Bracketed results indicate percentage changes in the forecasting results in comparison to the predicted water demand under current climate condition ( )) Description Water demand under "Current climate" condition ML/year Water demand under "A1B" climate scenario ML/year Water demand under "A2" climate scenario ML/year Water demand under "B1" climate scenario ML/year Forecasting decade Forecasting decade (0%) 279 (0.36%) 234 (0.43%) 279 (0.36%) 234 (0.43%) 278 (0.03%) Table 6.6 Projection of future water demand under three hypothetical climate change scenarios in the year, 2040 in the multiple dwelling sector Description Projection under scenario 1 (+1 0 C T & -10% R) Projection under scenario 2 (+2 0 C T & -20% R) Projection under scenario 3 (+3 0 C T & -30% R) Water demand (ML/year) Year : 2040 % changes in compare to water demand under current climate Projection under current climate conditions (avg. of ) * T refers to monthly maximum temperature and R refers to monthly total rainfall 301 University of Western Sydney Page 115

153 CHAPTER 6: Climate change & demand 6.4 Summary In this chapter, principal component analysis (PCA) was conducted to evaluate the relative influence of independent variables including climate variables (i.e. temperature and rainfall) on urban water demand. Moreover, impact of future climate change conditions on urban demand was also assessed by forecasting water demand under different future climate conditions and current climate condition (average rainfall and maximum temperature values for the period of 1960 to 2010). These forecasted results under different climate scenarios were compared with those obtained assuming current climate condition to estimate the probable impacts. The future climate scenarios considered in this chapter were (a) three scenarios of CSIRO Mk. 3 global climate model (GCM): A1B, A2 and B1, and (b) three hypothetical climate change scenarios: (i) 1 0 C rise in temperature and 10% decrease in rainfall, (ii) 2 0 C rise in temperature and 20% decrease in rainfall, (iii) 3 0 C rise in temperature and 30% decrease in rainfall from current climate conditions. Water demand forecasting was done for the period 2021 to 2040 for the single and multiple dwelling sectors, separately. The results of PCA biplot show that the water savings and water price variables are the dominant drivers in urban water demand modelling. Moreover, it is found that water savings measures in conjunction with water pricing policy can play an important role in managing water demand in order to maintain the balance between water demand and supply. As for the impact of climate variables, the results show that rainfall has no impact on water demand while temperature has some degree of positive influence on water demand. These results indicate that water demand are likely to be less affected by climate change conditions in the future and can be expected to rise by a little margin when there is a rise in temperature. The forecasting results indicate that the future water demand under climate change scenarios would increase by just 0.62% and 0.43% (maximum increase in the decade of ) above the current climatic condition for the single and multiple dwelling sectors, respectively. Furthermore, water demand projections with the hypothetical climate conditions by 2040 show that the water demand may increase by 1.21%, 2.42% and 3.64%, respectively, under the hypothetical climactic scenario 1, 2 and 3, above the forecasted water demand under current climate condition. University of Western Sydney Page 116

154 CHAPTER 6: Climate change & demand Under the same criteria, the increases in water demand in the multiple dwelling sector are found to be 0.57%, 1.14% and 1.72% for the hypothetical climate scenarios 1, 2 and 3, respectively. The above results indicate that the impact of potential future climate change on water demand would be negligible for the Blue Mountains region. The results of these quantitative assessments of climate change impact on urban water demand also support the results found in the PCA analysis. The overall findings in this chapter highlight the fact that urban water demand is likely to be less influenced by the climate variables and there will be a minor impact on urban water demand due to change in the climate conditions in the future. University of Western Sydney Page 117

155 CHAPTER 7: Calibration & Uncertainty CHAPTER 7 ESTIMATION OF PARAMETER SETS AND EVALUATION OF UNCERTAINTIES IN CALIBRATION OF A RAINFALL-RUNOFF MODEL This chapter is partial reproduction of the following refereed journal paper: Haque, M.M. 1, Rahman, A. 1, Hagare, D. 1 and Kibria, G Parameter uncertainty of the AWBM model when applied to an ungauged catchment. Hydrological Processes, published online, DOI: /hyp (ERA 2010 ranking: A, Impact factor: 2.69). 1 School of Computing, Engineering and Mathematics, University of Western Sydney, Australia 2 Sydney Catchment Authority, Penrith, Australia Abstract This chapter focuses on catchment yield estimation using rainfall runoff models. In this regard, a quantitative assessment of uncertainty was made in connection with the calibration of two rainfall-runoff models: Australian Water Balance Model (AWBM) and SIMHYD model for both gauged and ungauged catchment cases. For the gauged catchment, five different rainfall data sets, twenty three different calibration data lengths and eight different optimization techniques were adopted. For the ungauged catchment case, the optimum parameter sets obtained from the nearest gauged catchment were transposed to the ungauged catchments, and two regional prediction equations were used to estimate runoff. Uncertainties were ascertained by comparing the observed and predicted runoffs by the models on the basis of different combinations of methods, model parameters and input data. The main finding from this investigation was that the uncertainties in the modelling outputs could vary from -1.3% to 70% owing to different input rainfall data, -5.7% to 11% owing to different calibration data lengths and -6% to 0.2% owing to different optimization techniques University of Western Sydney Page 118

156 CHAPTER 7: Calibration & Uncertainty adopted in the calibration of the AWBM and SIMHYD model. The performance of the models was found to be dominated mainly by the selection of appropriate rainfall data followed by the selection of an appropriate calibration data length and optimization algorithm. Use of relatively short data length (e.g. 3 to 6 years) in the calibration was found to generate relatively poor results. Effects of different optimization techniques on the calibration were found to be minimal. The uncertainties reported here in relation to the calibration and runoff estimation by the models are relevant to the selected study catchments, which are likely to differ for other catchments. The methodology presented in this chapter can be applied to other catchments in Australia and other countries using the similar rainfall runoff models. University of Western Sydney Page 119

157 CHAPTER 7: Calibration & Uncertainty 7.1 Overview Chapter 6 has evaluated the influence of climate variables on urban water demand adopting Principal Component Biplot technique and assessed the impacts of plausible climate change scenarios on future water demand by forecasting future water demand under different plausible future climate conditions. This chapter focuses on catchment yield estimation using rainfall runoff models. In this regard this evaluates the uncertainties due to variability in input rainfall time series, variability in calibration data lengths and variability in optimization methods during calibration of two rainfall-runoff models: the Australian Water Balance Model (AWBM) and the SIMHYD model. In addition, it identifies three sets of optimized parameters sets for each of the models to estimate runoff in the two ungauged catchments (Katoomba and Blackheath) in the Blue Mountains region. These optimized parameters sets will be used in Chapter 8 to forecast future runoff in the Blue Mountains catchments. This chapter commences with presenting the structure of the AWBM and SIMHYD model. It then presents the methods developed in this chapter to quantify the uncertainties and to estimate the runoffs. This is followed by the results and discussion, and summary of the findings. 7.2 Rainfall-runoff models In this chapter, two widely used daily conceptual rainfall-runoff models; Australian Water Balance Model (AWBM) (Boughton 2004) and SIMHYD (Chiew et al. 2010) were used to evaluate the uncertainties during calibration and to estimate the calibrated parameter sets to be used in the two ungauged catchments in the Blue Mountains region to estimate runoff. These models are included in the Rainfall- Runoff Library (RRL), a software product in the Catchment Modelling Toolkit in Australia (more details of the RRL can be found in AWBM model structure The AWBM model is a conceptual rainfall runoff model, which generates runoff in daily time scale from the input data of rainfall and evapotranspiration (Boughton 2004). It consists of three surface moisture stores, C1, C2 and C3 that occupy partial areas of the catchments A1, A2 and A3, respectively. The average surface storage capacity is the single parameter that determines the amount of runoff, which is the University of Western Sydney Page 120

158 CHAPTER 7: Calibration & Uncertainty sum of three products of individual surface store capacity and respective partial area, i.e. C1 A1 + C2 A2 + C3 A3. In water balance calculation, rainfall (P) and evapotranspiration (E) are added and subtracted, respectively, to each of the stores at each time step, whereas water remains in the store. When the amount of water in any store reaches higher than the capacity of that store, the excess becomes runoff, and the amount in the store is reset to the capacity (Boughton 2004). This runoff is then divided between surface runoff and baseflow recharge. The model structure is presented in Figure 7.1, and the descriptions of the AWBM model parameters are given in Table 7.1. The net surface runoff and baseflow are estimated by the baseflow index (BFI), which varies between 0 and 1. This BFI can be estimated from a streamflow record by using any of the established techniques for segregation of total streamflow into net surface runoff and baseflow (Chapman 1999). The recharge of the baseflow (Q br ) and surface runoff store (Q sr ) is estimated by the following equations: Q br = BFI Excess (7.1) Q sr = (1 BFI) Excess (7.2) The daily discharge from the baseflow ( Q bd ) and surface store ( Q sd ) into streamflows are estimated by equations 7.3 and 7.4, respectively. Q bd = (1 K b ) BS (7.3) Q sd = (1 K s ) SS (7.4) where BS and SS are the amount of moisture in the baseflow and surface store, respectively and, K b and K s are the daily baseflow and surface runoff recession constant, respectively. These recessions constant can be estimated from the streamflow record. The AWBM model has nine parameters (Table 7.1); three of them represent areas of three different surface stores, and another three represent storage capacity of each surface store. The magnitude of runoff primarily depends on the storage capacity of the surface stores. Three partial areas of the surface stores must sum to 1.0; therefore, University of Western Sydney Page 121

159 CHAPTER 7: Calibration & Uncertainty only two areas are evaluated during the calibration, and the third one is automatically determined. The other three parameters (BFI, K b and K s ) control the timing of runoff. The AWBM model is one of the few rainfall runoff models that have the auto calibration capability. In the auto-calibration option, the model self-calibrates to a data set of daily rainfall, evapotranspiration and runoff. In this auto-calibration, fixed pattern of the surface storage capacities and their partial areas are used to disaggregate the average surface storage capacity in the individual values needed to run the model. Average surface storage capacity is determined by matching the total calculated runoff with the total actual runoff. Trial and error adjustments are used to calibrate baseflow parameters to match the calculated daily runoff with the observed daily runoff over the calibration period (Boughton and Chiew 2007). More details description of the auto calibration method in the AWBM model can be found in Boughton (2006). Figure 7.1 Structure of the AWBM model (Boughton 2004) SIMHYD model structure SIMHYD model has nine parameters and it is capable of estimating runoff values for both the daily and monthly time steps (Chiew et al. 2010). The structure of the University of Western Sydney Page 122

160 CHAPTER 7: Calibration & Uncertainty SIMHYD model is presented in Figure 7.2 and the description of the model parameters are given in Table 7.1. There is an interception store in the SIMHYD model which is filled in with rainfall at the first step and then is emptied each day by evaporation. The excess rainfall is then subjected to an infiltration function that determines the infiltration capacity. If the amount of excess rainfall is higher than the infiltration capacity, it becomes infiltration excess runoff. The amount of water that infiltrates is then subjected to a soil moisture function that diverts the water to the stream (interflow), groundwater store (recharge) and soil moisture store. Interflow and groundwater recharge are estimated in the order of first and second, respectively, as a linear function of soil wetness (soil moisture level divided by soil moisture capacity). The remaining water then flows into the soil moisture store which has a finite capacity. If the water exceeds the soil moisture store after evapotranspiration from that store, the water overflows into the groundwater store. From the groundwater store, base flow is simulated as a linear recession. In summary, the model generates runoff from three sources: (i) infiltration excess runoff, (ii) interflow and (iii) base flow. Table 7.1 Descriptions of the AWBM and SIMHYD model parameters AWBM parameter Description SIMHYD parameter Description A1 Partial area of the smallest store BC Baseflow coefficient A2 Partial area of middle store IT Impervious threshold A3 Partial area of the largest store IC Infiltration coefficient C1 Surface storage capacity of the smallest store IS Infiltration shape C2 Surface storage capacity of middle store IC Interflow coefficient C3 Surface storage capacity of the largest store PF Pervious fraction BFI Baseflow index RISC Rainfall interception store capacity K b Baseflow recession constant RC Recharge coefficient K s Surface runoff recession constant SMSC Soil moisture store capacity University of Western Sydney Page 123

161 CHAPTER 7: Calibration & Uncertainty 7.3 Methodology As discussed in Chapter 2 (Section 2.7.2), a rainfall-runoff model needs to be calibrated and validated using the observed data (e.g. runoff, rainfall and evaporation) before using it to climate change impact analysis on water yield or to forecast runoff. However, in ungauged catchments, the calibration and validation of a rainfall-runoff model cannot be undertaken directly due to unavailability of some or all of these observed data. As mentioned in Chapter 3 (Section 3.3), both the Blue Mountains catchments (Katoomba and Blackheath) are ungauged catchments. Hence, a nearest neighbour regionalisation technique (i.e. calibrate a rainfall-runoff model in the nearby gauged catchments and transpose the model parameters to the ungauged catchment) was adopted in this chapter to calibrate and validate the rainfall-runoff models and to estimate the calibrated parameter sets to be used in the Katoomba and Blackheath catchments. Rainfall Evapotranspiration Interception store Rainfall interception store capacity Infiltration coefficient Infiltration shape Infiltration excess runoff Interflow coefficient Recharge coefficient Saturation excess runoff Interflow Soil Moisture Store Soil moisture store capacity Runoff Groundwater Store Baseflow coefficient Baseflow Figure 7.2 Structure of the SIMHYD model (Podger 2004) As discussed in Chapter 2 (Section 2.7.2), different sources of uncertainties are associated with the calibration of a rainfall runoff model that need to be quantified to University of Western Sydney Page 124

162 CHAPTER 7: Calibration & Uncertainty assess the relative accuracy of the prediction made by a rainfall runoff model. In this chapter, three sources of uncertainties were quantified during the estimation of the calibrated parameter sets of the rainfall-runoff models, which were as follows: 1. Uncertainty due to the variability in input data, mainly uncertainty in rainfall time series. Rainfall and evaporation are the two primary variables taken into a rainfall-runoff model to estimate runoff. Between these two variables, evaporation has a much smaller spatial and temporal variability than rainfall and hence, rainfall-runoff modelling results are likely to be less influenced by the errors in evaporation data compared with rainfall data. In addition, in case of missing data, monthly average evaporation can be used as the replacement without any significant loss of accuracy in the outcomes of a rainfall-runoff model (Chapman 2003). Therefore, only uncertainty in calibration due to the variability in rainfall time series data was examined in this chapter. 2. Uncertainty due to variability in calibration data length. 3. Uncertainty due to different optimization methods to calibrate the rainfall-runoff models. As mentioned in Chapter 3 (Section 3.3), Narrow Neck catchment is the only nearest gauged catchment of both the Blue Mountains catchment selected in this chapter. Hence, the rainfall-runoff models were calibrated using the observed runoff data from the Narrow Neck catchment. These observed runoff data were assumed to be of good quality. As mentioned before, in this chapter, analysis was done with two rainfall-runoff models (i) Australian Water Balance Model (AWBM) and (ii) SIMHYD model. The following tasks were conducted in this chapter to obtain the calibrated parameter sets and to estimate the uncertainties in the calibration of the models. Descriptions of the methods to perform these tasks are given in the following sub-sections. 1. Estimation of uncertainty due to the variability in rainfall data during calibration of the AWBM model by using five different rainfall time series. University of Western Sydney Page 125

163 CHAPTER 7: Calibration & Uncertainty 2. Selection of appropriate rainfall time series and rainfall factor. 3. Evaluation of the impacts of different calibration and validation data length on the AWBM model performance. 4. Estimation of the impacts of using different optimisation methods on the calibration of the AWBM model. 5. Selection of the optimum parameter sets of the AWBM based on the results of the above Tasks 1, 2, 3 and Estimation of the monthly runoff in the Blue Mountains catchments using the selected optimum parameter sets. 7. Estimation of the monthly runoffs in the Blue Mountains catchments using two regional methods developed by Boughton (2009) and Boughton and Chiew (2007), details of these two methods are given in Section in this chapter. 8. Examination of the uncertainty in the calibrated parameter sets and the AWBM modelling outputs on the basis of the results found in the above Task 6 and Execution of the Task 3, 4, 5 and 6 using the SIMHYD model and compare the results obtained using the AWBM model Model results evaluation criteria In this chapter, the evaluation of modelling results was carried out using a number of performance statistics, the Nash Sutcliffe efficiency (NSE) (described in Section in Chapter 4), the median of bias in percentage (MBIAS) (described in Section in Chapter 4), the percentage difference between the total modelled and observed runoff (V%) and the average ratio of the yearly modelled runoff to the yearly observed runoff using the simulated and observed monthly and annual runoff values of the gauged catchment (Narrow Neck). Normally, NSE values greater than 0.6 indicate reasonable agreement, and NSE values greater than 0.8 indicate good agreement between observed and modelled values in the catchment yield studies (Chiew and McMahon 1993). The ideal value of BIAS is zero with low values of BIAS indicating better modelling results, where positive and negative values represent overestimation and underestimation bias, respectively in the modelled results. University of Western Sydney Page 126

164 CHAPTER 7: Calibration & Uncertainty Uncertainty in the model outputs is reported in this chapter by estimating the percentage difference between the total observed and modelled runoff that indicates the model performance in simulating the total observed runoff during the calibration and validation process, which can be represented by V%. The perfect value of V% is 100, which indicates that the total modelled runoff is equal to the total observed runoff. It can be calculated by the following equation: PT OT V (%) 100 (7.5) O T where O T is the total observed runoff and P T is the total modelled runoff. The ratio of the yearly modelled runoff to the yearly observed runoff represents the variation in simulating yearly runoff which can be calculated by the following equation: P O Y Ratio (7.6) Y where P Y is the yearly modelled runoff and O Y is the yearly observed runoff. An optimum value of the ratio is one that indicates that yearly modelled runoff is equal to the observed runoff value Uncertainty due to variability in rainfall time series In order to identify the impact of input rainfall data on model calibration, the AWBM model was calibrated with the observed monthly runoff values for the whole period of available data set ( ) adopting auto calibration feature available in the AWBM model for the Narrow Neck catchment using five different rainfall inputs: (i) rainfall of Katoomba weather station; (ii) rainfall of Blackheath weather station; (iii) simple average of rainfall values from Katoomba (KT) and Blackheath (BH) weather station; and (iv) two pairs of factors were assigned to the Katoomba and Blackheath rainfall values on the basis of the distance from the catchment. Distances between the Narrow Neck catchment and Katoomba and Blackheath weather stations are 6.57 km and km, respectively. Two factors (a and b) were calculated as follows: University of Western Sydney Page 127

165 CHAPTER 7: Calibration & Uncertainty a 0.36; b 0.64 (7.7) This pair of factor (a, b) was assigned alternatively to the rainfall values of Katoomba and Blackheath stations. In one case, a was assigned to the Katoomba station and b was assigned to the Blackheath station; and in another case, a was assigned to the Blackheath station and b was assigned to the Katoomba station. The best rainfall station was selected on the basis of the performance measures discussed above. After identifying the best rainfall station/rainfall time series to calibrate the AWBM model, some scaling was carried out to the rainfall values of the identified station to check whether it improves the calibration results. Scaling of the input data is common in water balance studies to improve the model results (Boughton 2009). Annual average rainfall in Blackheath weather station is smaller than that of Katoomba weather station. Hence, if the Katoomba weather station would come as appropriate station, then few rainfall factors greater than 1 (one) (e.g. 1.05, 1.1, 1.5 and 1.2) would be multiplied with the Katoomba rainfall time series, otherwise if Blackheath weather station would come as appropriate station, then few rainfall factors smaller than 1 (one) (e.g. 0.95, and 0.8) would be multiplied with the Blackheath rainfall time series Uncertainty due to variability in optimization methods In order to quantify the uncertainty due to the different optimization techniques, the rainfall-runoff models were calibrated using eight different optimization methods including Uniform random search (URS), Pattern search (PS), Pattern search multi start (PSMS), Rosenbrock search (RS), Rosenbrock multi start search (RMSS), Genetic Algorithm (GA), Shuffle complex evolution (SCE-UA) and auto calibration option using the total data lengths in the calibration period. The models were calibrated adopting the above mentioned optimization methods to maximise the objective function, Nash-Sutcliffe efficiency. In the URS method, each parameter is divided into a specified number of intervals considering the minimum and maximum limits. Then parameter value is randomly selected from the parameter space to run the model and assess the objective function. The procedure is repeated for a specified University of Western Sydney Page 128

166 CHAPTER 7: Calibration & Uncertainty number of times, and the parameter set with the best value of the objective function is retained as the optimum solution. The PS method starts with an initial value and evaluates the objective function for an incremental decrease and increase in the initial value to find the optimum value. The PSMS method divides the parameter values into a specified number of increments between the specified limits, and then PS is carried out for each of these starting points. The RS method is similar to the PS method to some extent where it returns at each step a point at least as good as the previous one in the parameter space. The RBMS works similar to PSMS by dividing the parameter values into a specified number of increments to avoid bias due to the pre-specified starting points in the RS method. The GA method searches among a population of randomly generated points, and then each point is evaluated to find out the maximum value of the objective function. The SCE-UA method is a probabilistic search method, which was developed at the University of Arizona. More details description of these optimization methods can be found in Podger (2004) Uncertainty due to variability in calibration data lengths After selecting the appropriate rainfall time series and optimization method (as discussed earlier), the rainfall-runoff models were calibrated by varying the calibration data length to find the optimum data length to calibrate the model and to estimate the uncertainty due to the different calibration data periods. The 25 years of data period was split into two segments: one data set was used for calibration, and the remaining data set was used for validation purpose. In the first test, the first 3 years of data period was considered for calibration, and the remaining 22 years was considered for validation. Then in the next test, 1 year was added to calibration data length, and validation was carried out for the remaining data. In this way, a total of 23 (T1 to T23) tests were carried out, and model performance statistics were estimated for each of the tests. NSE was calculated for monthly runoff values for both the calibration and validation data sets for all the tests. A total NSE value was calculated for the whole of the data set by assigning equal importance factor (0.5) to the calibration and validation NSE values. University of Western Sydney Page 129

167 CHAPTER 7: Calibration & Uncertainty Estimation of runoff Runoff values in the Blue Mountains catchments were estimated by transposing the optimum parameters of the Narrow Neck catchment to the Katoomba and Blackheath catchments adopting the two rainfall-runoff models, AWBM and SIMHYD. Average yearly runoff values in the Katoomba and Blackheath catchments were also estimated by the method described in Boughton (2009) and Boughton and Chiew (2007). Boughton (2009) produced a single set of parameter values for each of the five states (e.g. one parameter set for the catchments in New South Wales (NSW) and another parameter sets for the catchments in Western Australia) for the AWBM model to estimate runoff on ungauged catchments. He regionalized the parameter set from the calibration results of 121 catchments comprising five states of Australia [NSW, Queensland, South Australia, Tasmania and Western Australia]. As the Blue Mountains catchments are located in NSW region, the parameter values from the NSW region was used to estimate runoff in the Katoomba and Blackheath catchments. The parameter values were taken as follows: surface storage capacity (SS) = 145 mm, BFI = 0.33, baseflow recession constant (K b ) = 0.98 and surface runoff recession constant (K s ) = Boughton and Chiew (2007) developed regression equations to relate average annual runoff to average annual rainfall and potential evapotranspiration adopting data from 213 catchments across Australia. They developed regional prediction equations for each of the six major drainage divisions of Australia. Then they used these regional equations to estimate annual average runoff at the ungauged catchments. Thereafter, the AWBM model was adopted to estimate daily and monthly runoff using the rainfall and evapotranspiration data by calibrating its surface storage parameters to match with the estimated average annual runoff. In this chapter, the developed regression equation for Australian Drainage Division II was used to estimate annual average runoff for the Katoomba and Blackheath catchment as these catchments are located in Drainage Division II. The developed equations by Boughton and Chiew (2007) are given as follows: If P > 1000 mm/year; Q P E 361 (7.8) University of Western Sydney Page 130

168 CHAPTER 7: Calibration & Uncertainty If P ranges mm/year; Q P E 206 (7.9) where Q is runoff in mm/year, P is rainfall in mm/year and E is evapotranspiration in mm/year. 7.4 Results Uncertainty due to input rainfall data The AWBM model calibration results using the five different input rainfall data sets (represented by T1 to T5) are presented in Table 7.2. The results showed that the NSE value was the highest when rainfall values from the Blackheath station were used, and the NSE value was the lowest when rainfall values from Katoomba station were used. The second highest NSE value was obtained when more weighting was assigned to the rainfall data of the Blackheath station. In respect to MBIAS, the value was found to be the highest (30.67%) for the Katoomba station, which indicated that the monthly estimated runoff was about 30% higher than the recorded runoff. For two cases, the MBIAS values were found to be the lowest and close to each other (- 6.98% and 5.96%) when using the rainfall data from the Blackheath station and simple average value on the basis of the Katoomba and Blackheath stations. Total estimated runoff was found to be 69% higher than the total recorded runoff in T1 (Table 7.2). The lowest value of V was found to be -1.29% for T2, which indicated that the total estimated runoff is only 1.29% smaller than the total modelled runoff. The second best result was found to be 12.35% for T5. For T3 and T4, the total estimated runoffs were within 25% and 37% of the recorded runoff. From these results, it might be noted that the AWBM model was calibrated well when the rainfall data were taken from the Blackheath station and the worst when the rainfall data were taken from the Katoomba station. It was also found that when more weighting was given to the rainfall values of the Blackheath station, the calibration results were found to be better than the other test results (i.e. T1, T3 and T4). The results favour the selection of the Blackheath rainfall station for calibration of the AWBM model for the Narrow Neck catchment. These results also demonstrated that the uncertainty (in terms of variation between the total modelled runoff and recorded University of Western Sydney Page 131

169 CHAPTER 7: Calibration & Uncertainty runoff) in the model outputs could be between -1.29% and 70% owing to the different input rainfalls to the model. After identifying the appropriate rainfall station to calibrate the AWBM model, some scaling was done to the rainfall values of the Blackheath station to check whether it improves the calibration results. Since the calibration results with rainfall stations tended to be better with the lower rainfall values (Blackheath rainfall < Katoomba rainfall), four decreasing scaling factor 0.95, 0.90, 0.85 and 0.80 was assigned to the rainfall values of the Blackheath station and run the model with these rainfall values. Results of incorporating different rainfall scaling factors are presented in Table 7.3; it can be seen that NSE value increases with the decreasing rainfall factor. Moreover, all the NSE results of these runs were found to be better than the results of using the original rainfall values of the Blackheath station indicating the necessity of adopting rainfall factor. MBIAS value was found to be within the close range of the previous results with the Blackheath station for the first three tests. However, it went higher for T4 when 0.80 was used as the rainfall scaling factor. The agreements between total modelled runoff and total recorded runoff were found to be satisfactory for all the tests. However, the value of V(%) increased with the decreasing rainfall factor, which indicated that a higher decreasing factor would make the total modelled runoff being more underestimated. Moreover, R 2 value (Table 7.3) of the trend line between the monthly observed and modelled runoff was found to increase for the first three tests, and it started to decrease with the T4. Therefore, considering all of these results, rainfall scaling factor of 0.85 was found to be the best option and hence was selected for further analysis. The aforementioned results due to different rainfall inputs demonstrate the importance of selecting proper rainfall station(s)/appropriate rainfall data in order to obtain good calibration results for a rainfall-runoff model. It also indicates that the rainfall runoff model calibrated with inappropriate or poor quality rainfall data would produce less effective parameter set, which would eventually affect the runoff estimation in the ungauged catchments even with the high-quality rainfall data in the ungauged catchments. Similar conclusions were made by Post et al. (2008) and Vaze et al. (2008) when investigating the impact of rainfall data quality on the calibration University of Western Sydney Page 132

170 CHAPTER 7: Calibration & Uncertainty results of the SIMHYD and Sacramento rainfall runoff models for ten catchments in the Murray Darling Basin of Australia. They found a large variation in the estimated parameter sets due to different rainfall inputs and showed that the models performance increased with improved rainfall data. Table 7.2 Comparison of calibration results of the AWBM model with five different rainfall inputs SN Rainfall series NSE MBIAS (%) V (%) T1 Katoomba T2 Blackheath T3 Katoomba (0.5)+Blackheath (0.5) T4 Katoomba (0.64)+Blackheath (0.36) T5 Katoomba (0.36)+Blackheath (0.64) Table 7.3 Effect of rainfall scaling factor on the calibration results of the AWBM model SN Rainfall station Rainfall factor NSE MBIAS (%) V (%) R 2 T1 Blackheath T2 Blackheath T3 Blackheath T4 Blackheath Uncertainty due to optimization methods The performance statistics of eight different optimization methods adopting the AWBM model are presented in Table 7.4. The results showed that the GA performed the best in terms of average ratio, but it performed the worst when considering the agreement between the total modelled runoff and total observed runoff. In terms of V value, the RS performed the best, but it performed poorer in terms of MBIAS. In terms of MBIAS, the SCE-UA performed the best, but value of V in this case was University of Western Sydney Page 133

171 CHAPTER 7: Calibration & Uncertainty higher than the other results. In terms of NSE value, the URS performed the worst, and rest of the methods were found to perform similarly. This result is likely since GA is based on random number generation, which may need a greater number of simulations than 500 (as adopted in this study) to achieve comparable results. The relatively poor performance of the SCE-UA and RBMS methods cannot be readily explained; further study might have unfolded the reason for this; which however was not undertaken in this thesis. On the basis of these results, it might be noted that not a single method could produce the best result with respect to all the four performance statistics adopted here. Nevertheless, all of the methods were found to be performing in a comparable manner. The uncertainty due to the calibration methods was found to be in the range of -6.01% to 0.80%, which is much smaller than the uncertainty due to the variability in rainfall time series data. Table 7.4 Performance statistics of the AWBM model based on different optimization methods Optimization methods Average ratio V (%) MBIAS (%) NSE 1) GA ) PS ) PSMS ) URS ) RBMS ) RS ) SCE-UA ) Auto *Bold marked value represents the best value in the table Uncertainty due to calibration data length The overall calibration and validation NSE values for monthly runoffs adopting different calibration data lengths using auto calibration methods of the AWBM model are presented in Figure 7.3. The calibration NSE values were found to be above 0.8 for the first ten tests (T1 T10), and then it gradually declined. Calibration University of Western Sydney Page 134

172 CHAPTER 7: Calibration & Uncertainty NSE values were found to be in the range of 0.87 (T2) and 0.63 (T16). It was found that when the calibration data length was equal or close to half of the data set, the calibration results were satisfactory. When considering the full data set for calibration, the NSE value was found to be 0.68 (lower than that of using half of the data set), which indicated that using the full dataset to calibrate the model might not necessarily give the better calibration results. The validation NSE values varied between 0.82 (T19 and T20) and 0.51 (T9 and T10), and the results were found to be better for T13 to T22. Hence, it may be noted that the validation NSE values were found to be better when less than or equal to half of the data set was used in the validation. In terms of total NSE, not so much difference was found for the tests using the different calibration and validation data lengths as it varied between 0.65 (T10) and 0.74 (T13). The best value was found to be 0.74 for T13, where calibration data period was (15 years) and validation data period was (10 years). It should be noted that better results were obtained for T19, T20, T21 and T22. Values of yearly average ratio (estimated using Equation (7.6)), monthly MBIAS and V (estimated using Equation (7.5)) for all of the adopted tests due to different calibration data lengths are presented in Table 7.5. It showed that better average ratio values were obtained for T7 and T8. In terms of MBIAS, better results were obtained for T7, T8, T17, T19, T20, T21, T22 and T23. In terms of an agreement (V%) between the total modelled runoff and recorded runoff, the results for T17, T19 and T23 were found to be better. From these results, it may be noted that no single test was found to be the best with respect to all the three statistics adopted here. Moreover, it was found that the uncertainty due to the different calibration data lengths was in the range of -5.69% to 11%. Since no single test was found to be the best, it was decided to select few different sets of plausible parameters on the basis of the test results to apply on the ungauged catchments. Finally, three sets of parameters were selected for transposing to the ungauged catchments, which corresponded to T13 (on the basis of the highest NSE value), T8 (on the basis of the lowest average ratio) and T23 (on the basis of the value of MBIAS and V, and consideration of using full data length). The values of University of Western Sydney Page 135

173 T1 T2 T3 T4 T5 T6 T7 T8 T9 T10 T11 T12 T13 T14 T15 T16 T17 T18 T19 T20 T21 T22 T23 NSE Value Climate change impact on water demand and supply CHAPTER 7: Calibration & Uncertainty the AWBM parameters for the selected three sets are presented in Table C.7.1 in Appendix C Runoff estimation at the Katoomba and Blackheath (ungauged) catchments Average yearly runoff values in the Katoomba and Blackheath catchments for the period of 1988 to 2012 were estimated by transposing three different sets of calibrated parameters (i.e. T8, T13 and T23). The calculated runoffs are presented in Table 7.6; it can be seen that average yearly runoff varied between 592 to 607 mm/year for the Katoomba catchment and 381 to 401 mm/year for the Blackheath catchment. Total runoff and average yearly runoff estimated by the three different sets of parameters were found to be relatively close to each other, which indicated that any one of the selected three calibrated parameter sets could be used to estimate runoff in the two ungauged catchments. However, a single set of parameter values should not be considered in the rainfall-runoff modelling as a diverse set of possible parameter values can lead to similar model performance, which would give higher confidence on the model output. In this regard, Bárdossy (2007) mentioned that model parameters are generally correlated with each other; changes in one parameter can be compensated for by changes in one or others, which eventually could result in similar model outputs. NSE (Calibration) NSE (Total) NSE (Validation) Test No University of Western Sydney Page 136

174 CHAPTER 7: Calibration & Uncertainty Figure 7.3 Total, calibration and validation NSE values for the 23 tests due to different calibration and validation data lengths adopting the AWBM model Table 7.5 Values of NSE (total), average ratio, MBIAS (%) and V (%) for the 23 tests due to different calibration and validation data lengths adopting the AWBM model Test NSE (total) Average ratio MBIAS (%) V (%) T T T T T T T T T T T T T T T T T T T T T T T NSE (total) = 0.5 NSE of calibration data set NSE of validation data set The estimated average yearly runoff values by the regional prediction equations developed by Boughton (2009) and Boughton and Chiew (2007) are presented in Table 7.7. Annual average rainfall values in the Katoomba and Blackheath weather stations are 1406 and 1149 mm/year, respectively. Therefore, Equation (7.8) was used to estimate the annual average runoff for the catchments. Comparing Tables 7.6 and 7.7, it is apparent that there is considerable disagreement between the estimated runoffs of the Katoomba and Blackheath catchments (which are ungauged) using University of Western Sydney Page 137

175 CHAPTER 7: Calibration & Uncertainty three different methods [i.e. transposing of the model parameters from the nearby gauged catchment as carried out in this chapter, and regional methods by Boughton (2009) and Boughton and Chiew (2007)]. As these two catchments are ungauged, it is difficult to find out which of the adopted methods provide the best results. However, these results present a range of most probable estimates of the runoff at these ungauged catchments. It should be noted that some studies have reported that nearby catchments are likely to have similar catchment characteristics and hence the transpose method (i.e. use of the calibrated parameters from the nearest gauged catchment to the ungauged catchment) should produce better results (Merz and Blöschl 2004, Young 2006, Oudin et al. 2008). Boughton (2009) also found good results in an experiment with 18 Australian catchments when using the calibrated AWBM parameters with data from one catchment to estimate the runoff in the other 17 catchments. In the Boughton and Chiew (2007) method, linear relationship was assumed between the variables; however, the underlying relationship between these variables and other hydrological variables is more likely to be nonlinear. Another drawback of the method is that the form of the relationship needs to be assumed to estimate the average yearly runoff before calibration of the AWBM model. Table 7.6 Estimated runoff of the Blue Mountains catchments (Katoomba and Blackheath) by the AWBM model for the period by transposing the calibrated parameter from the nearby catchment Katoomba Blackheath Parameter set Total runoff (ML) Relative difference Average runoff (mm/year) Total runoff (ML) Relative difference Average runoff (mm/year) T T8 & T13: 2.89% T8 & T13: 5.25% 381 T T13 & T23: 0.3% T13 & T23: 0.25% 401 T T23 & T8: 2.52% T23 & T8: 4.75% 400 University of Western Sydney Page 138

176 CHAPTER 7: Calibration & Uncertainty In Boughton (2009), a single set of parameter values was used to estimate annual average runoff from a group of catchments, which is more likely to give less confidence in the runoff estimation in the ungauged catchments as catchment characteristics vary from place to place. Although Boughton (2009) demonstrated good results with the single set of parameter values and some rainfall adjustments to estimate runoff, this method may not produce good results for the smaller catchments (e.g. less than 10 km 2 ), as for smaller catchments the average surface capacity would be much less than the value (145 mm) used in Boughton (2009) to generate runoff. These issues need to be investigated further with data from additional catchments to make any firm conclusion. These issues may be investigated as a future work arising from this PhD study. Table 7.7 Estimated annual average runoff for the Katoomba and Blackheath catchments on the basis of the regional methods by Boughton (2009) and Boughton and Chiew (2007) for the period of Methods Katoomba Average runoff (mm/year) Blackheath Average runoff (mm/year) Boughton (2009) Boughton and Chiew (2007) Calibration and runoff estimation results using the SIMHYD model The SIMHYD model was run adopting seven different optimization methods to find the best methods and to estimate uncertainties. The best rainfall time series found during the AWBM model calibration was used during the calibration of the SIMHYD model. Similar to the results of the AWBM model, the results using different optimization methods (Table 7.8) showed that not a single method could produce the best result with respect to all the performance statistics. The uncertainty due to the choice of calibration methods was found to be in the range of -9.04% to 6.35%. Based on the results found in Table 7.8, the PS method was selected for University of Western Sydney Page 139

177 CHAPTER 7: Calibration & Uncertainty further analysis as the value of V (%) for this method was the lowest among all others, which indicated that it simulated total runoff better than others. In addition, the NSE value was the highest for this method and the MBIAS (%) value was comparable to others. Adopting the PS optimization method, the SIMHYD model was run with varying calibration data lengths to estimate the uncertainty due to the choice of calibration data lengths and to find the optimized parameters sets for the SIMHYD model. Total NSE values, values of yearly average ratio (estimated using Equation (7.6)), monthly MBIAS and V (estimated using Equation (7.5)) for all the tests due to different calibration data lengths are presented in Table 7.9. It can be seen that the highest NSE value was obtained for T19, the lowest average ratio value was obtained for T10, the lowest MBIAS value was obtained for T9 and the lowest V (%) value was obtained for T14. Though T9 and T10 were found to be the best in terms of MBIAS and average ratio values, respectively, there was considerable disagreement found between the total observed runoff and the total modelled runoff as the value of V(%) was much higher than other tests. The uncertainty due to the different calibration data lengths was found to be in the range of % to 6.78%. Table 7.8 Performance statistics of the SIMHYD model based on different optimization methods Optimization methods Average Ratio V (%) MBIAS (%) NSE 1) GA ) PS ) PSMS ) URS ) RBMS ) RBSS ) SCE-UA *Bold marked value represents the best value in the table University of Western Sydney Page 140

178 CHAPTER 7: Calibration & Uncertainty These results demonstrated that no single test was the best with respect to all the statistics adopted here. Nevertheless, it can be seen in Table 7.9 that the results became stable during the T14 to T17 indicating the consideration of half or close to half of the calibration data length for achieving better results. Since no single test was found to be the best similar to the AWBM model three sets of parameters were selected from the SIMHYD calibration for transposing to the ungauged catchments. The selected parameter sets were T14 (based on the lowest V(%) value), T19 (based on the highest NSE value) and T23 (based on using the total data length). The values of the three sets of parameter are presented in Table C.7.2 in Appendix C. Table 7.9 Values of NSE (total), average ratio, MBIAS (%) and V (%) for the 23 tests due to different calibration and validation data lengths adopting the SIMHYD model Test NSE (total) Average ratio MBIAS (%) V (%) T T T T T T T T T T T T T T T T T T T T T T T *NSE (total) = 0.5 NSE of calibration data set NSE of validation data set. *Bold marked value represents the best value in the table University of Western Sydney Page 141

179 CHAPTER 7: Calibration & Uncertainty Total runoff and average yearly runoff values in the Katoomba and Blackheath catchments for the period of 1988 to 2012 were estimated by transposing three different sets of calibrated parameters of the SIMHYD model (i.e. T14, T19 and T23). The calculated runoffs are presented in Table 7.10; it can be seen that average yearly runoffs varied between 585 and 601 mm/year for the Katoomba catchment, and 388 to 407 mm/year for the Blackheath catchment. Total runoff and average yearly runoff estimated by the three different sets of parameters were found to be relatively close to each other (relative differences between the results were small), which indicated that any one of the selected three calibrated parameter sets could be used to estimate runoff in the two ungauged catchments. In addition, the estimated runoffs by the SIMHYD model were found to be quite comparable with the results found adopting the AWBM model (Table 7.6) indicating the uncertainty due to the choice of models are negligible. Table 7.10 Estimated runoff of the Blue Mountains catchments (Katoomba and Blackheath) by the SIMHYD model for the period by transposing the calibrated parameter from the nearby catchment Katoomba Blackheath Parameter set Total runoff (ML) Relative difference (Abs. value) Average runoff (mm/year) Total runoff (ML) Relative difference (Abs. value) Average runoff (mm/year) T T14 & T19: 1.64% T14 & T19: 4.26% 390 T T19 & T23: 2.65% T19 & T23: 4.51% 407 T T23 & T14: 1.07% T23 & T14: 0.44% Summary This chapter has focused on catchment yield estimation using rainfall runoff models. In this regard, this chapter has examined the degree of uncertainties associated with the calibration and runoff estimation by the AWBM and SIMHYD model for both University of Western Sydney Page 142

180 CHAPTER 7: Calibration & Uncertainty gauged and ungauged catchment cases. For the gauged catchment, five different rainfall data sets, twenty three different calibration data lengths and eight different optimization techniques were adopted. For the ungauged catchment case, the optimum parameter sets obtained from the nearest gauged catchment (located in the Blue Mountains region, Australia) were transposed to two ungauged catchments (Katoomba and Blackheath) in the Blue Mountains region, Australia, and two regional prediction equations were used to estimate runoff. Uncertainties are ascertained by comparing the observed and modelled runoffs by the models on the basis of different combinations of methods, model parameters and input data. The main finding from this chapter is that the uncertainties in the modelling outputs may vary from -1.3% to 70% owing to different input rainfall data, % to 11% (AWBM: -5.69% to 11%; SIMHYD: % to 6.78%) owing to different calibration data lengths and -9.04% to 6.35% (AWBM: -6.01% to 0.8%; SIMHYD: % to 6.35%) owing to different optimization methods adopted in the calibration of the rainfall-runoff models. The performance of the rainfall-runoff models is found to be dominated mainly by the selection of appropriate rainfall data, followed by the selection of an appropriate calibration data length and optimization algorithm. It is also found that using full data period in the calibration may not produce better results than using half of the data length. It is also found that when using smaller length of data (3 to 6 years) in the calibration, it is likely to generate poorer model outputs. Results of runoff estimation show that any of the optimized parameter sets may be used as they produce similar results. However, due to the existence of uncertainties, single set of parameter should not be used to forecast runoff and to study climate change impact. It is also found that the runoffs estimated by the AWBM and the SIMHYD model are comparable to each other indicating negligible uncertainty due to the choice between these two models. The uncertainties reported here in relation to the calibration and runoff estimation by the AWBM and SIMHYD model are relevant to the selected study catchments, which are likely to differ for other catchments. However, the methodology presented in this chapter can be applied to other catchments in Australia and other countries to estimate optimized parameter sets and uncertainties using similar rainfall runoff models. University of Western Sydney Page 143

181 CHAPTER 8: Performance of BMWSS CHAPTER 8 ESTIMATION OF FUTURE RUNOFF, UNCERTAINTIES AND FUTURE PERFORMANCE OF A WATER SUPPLY SYSTEM UNDER CHANGING CLIMATE CONDITIONS This chapter is partial reproduction of the following paper: Haque, M.M. 1, Rahman, A. 1, Hagare, D. 1, Kibria, G. 2 and Karim, F Estimation of catchment yield and associated uncertainties due to climate change in a mountainous catchment in Australia. Journal of Hydrology, under review (ERA 2010 ranking: A*, Impact factor: 3.68). 1 School of Computing, Engineering and Mathematics, University of Western Sydney, Australia 2 Sydney Catchment Authority, Penrith, Australia 3 CSIRO Land and Water, Commonwealth Scientific and Industrial Research Organisation, Canberra, ACT 2601, Australia Abstract This chapter examines the impacts of climate change on future runoff and estimates uncertainties in the forecasted runoff in a mountainous catchment (Blue Mountains) in the state of New South Wales in Australia. It also assesses the future performance of the Blue Mountains Water Supply System with the forecasted water demand and runoff scenarios in the periods. The uncertainties associated with the prediction of runoff were estimated using a multi-model approach based on four global climate models (GCMs), 200 realisations (50 realisations from each GCM) of downscaled rainfalls, two rainfall-runoff models and six sets of model parameters. The four GCMs used were, CSIRO, ECHAM 5, CCCMA and MIROC, and two rainfall-runoff models were the Australian Water Balance Model and the SIMHYD model. The ensemble results of runoff projections show that the mean annual runoff University of Western Sydney Page 144

182 CHAPTER 8: Performance of BMWSS is expected to be reduced in future periods ( ) by 34% in comparison to that of However, considerable uncertainty in the runoff estimates were found as the ensemble results projected changes of the 5 th (dry scenario) and the 95 th (wet scenario) percentile by -73% to +27% and -73% to +12% in the decades of and , respectively. Median projection of runoff changes using the four GCMs were found to be in the same direction (i.e. decrease in runoff). However, significant differences were noticed in the magnitudes of the runoff calculated using the four GCMs. On the other hand, the runoffs calculated using the two rainfall-runoff models were found to be quite similar. Results of uncertainty estimation demonstrate the uncertainty rank as: GCM uncertainty > realisation uncertainty > rainfall runoff model uncertainty > rainfall-runoff model parameter uncertainty. The future performance assessment results of the Blue Mountains Water Supply System show that the Blue Mountains storages with the future runoff would not be sufficient to provide water to the community. Water from other sources and implementation of water restriction will be needed to ensure necessary water supply. The results of this chapter provide important insights about the possible runoff changes and future performance of the Blue Mountains Water Supply System in the future decades due to changing climate, which would assist the water authorities for better planning and management of the water supply system. University of Western Sydney Page 145

183 CHAPTER 8: Performance of BMWSS 8.1 Overview Chapter 7 has evaluated the uncertainties associated with the calibration of a rainfallrunoff model and estimated model parameter sets for the Australian Water Balance Model and the SIMHYD model to use in forecasting runoff. This chapter forecasts runoff in the Blue Mountains catchments, evaluates uncertainties in the forecasted results and assesses the performance of the Blue Mountains Water Supply System under future climate conditions. It commences with presenting the methodologies to conduct the above mentioned tasks. It then presents the results of estimated uncertainties, forecasted runoff and projected performance of the Blue Mountains Water Supply System in the future periods. This is followed by a summary of the findings. 8.2 Methods In this chapter, three tasks were carried out, which are as follows: 1. Forecasting the runoffs in the two ungauged catchments in the Blue Mountains region. 2. Estimating the uncertainties in forecasting of runoffs due to four different sources of uncertainties (e.g. choice of Global Climate Models (GCMs), internal variability of the GCMs, choice of rainfall-runoff models and choice of model parameters). 3. Assessing the future performance of the Blue Mountains Water Supply System (BMWSS) Forecasting runoffs To estimate possible scenarios of future runoff under changing climate conditions, two rainfall-runoff models: Australian Water Balance Model (AWBM) and SIMHYD model (discussed in Chapter 7 in Section 7.2) were run using the selected three sets of parameters (estimated in Chapter 7) for each model with the future climate projection data. As discussed in Chapter 3 (Section 3.6.5), the future rainfall scenarios were obtained from the NARCliM (NSW/ACT Regional Climate Modelling) 2014 project for four Global Climate Models (GCMs) (i.e. CSIRO, CCCMA, ECHAM 5 and MIRPC). In addition, the evaporation data were obtained University of Western Sydney Page 146

184 CHAPTER 8: Performance of BMWSS from an intergovernmental project for CSIRO global climate model. The fifty realisations (number of repetitive simulation results for a given time step within a single GCM) of the downscaled rainfall data from each of the four GCMs and fifty realisations of evaporation data from the CSIRO GCM were taken as input to the rainfall-runoff models. Forecasting of runoff was made for the period of 2021 to 2040, and then the mean annual runoff was calculated for the two decades: and The range of forecasting results from the combinations of four GCMs, two hundred realisations of downscaled rainfall (i.e. 50 realisations from each GCMs), two rainfall-runoff models and six set of parameters (3 sets for each rainfall-runoff model) were compared with the mean annual runoff of the reference period ( ) to estimate the changes relative to the reference period. The adopted framework of forecasting runoff is presented in Figure 8.1. Rainfall-runoff models AWBM SIMHYD Parameter set 1 Parameter set 2 Parameter set 3 Projection with CSIRO rainfall data Projection with CSIRO rainfall data Projection with CSIRO rainfall data Projection with CCCMA rainfall data Projection with CCCMA rainfall data Projection with CCCMA rainfall data Projection with ECHAM 5 rainfall data Projection with ECHAM 5 rainfall data Projection with ECHAM 5 rainfall data Projection with MIROC rainfall data Projection with MIROC rainfall data Projection with MIROC rainfall data Figure 8.1 Adopted framework of forecasting runoff based on the AWBM and SIMHYD model using the projected climate data University of Western Sydney Page 147

185 CHAPTER 8: Performance of BMWSS Estimating uncertainties In this chapter, four different types of uncertainties were estimated in the forecasting of runoffs, which are as follows: 1. Uncertainty due to choice of GCMs (i.e. GCM uncertainty). 2. Uncertainty due to internal variability of a GCM (i.e. realisation uncertainty). 3. Uncertainty due to choice of rainfall-runoff models (i.e. rainfall-runoff model uncertainty). 4. Uncertainty due to choice of rainfall-runoff model parameter sets (i.e. parameter uncertainty). Uncertainties were reported by calculating the ensemble means of the forecasted runoffs and the spread around these mean values due to different sources of uncertainties. Coefficient of variation (C V ) (the ratio of mean and standard deviation) was calculated to represent the spread with respect to the mean. Quantification of the uncertainties was done in the following way and the flow charts of the uncertainty estimation methods are presented in Figures 8.2, 8.3, 8.4 and 8.5. (i) (ii) (iii) The spread due to the choice of GCMs was estimated by taking the median value from the fifty forecasting simulations of annual runoffs using each GCM, and thereafter calculating the C V among the four GCMs (Figure 8.2). The spread due to the fifty realisations of the downscaled rainfall data (i.e. realisation uncertainty) was estimated by taking the ensemble mean and standard deviation of the fifty simulations using one GCM and one rainfall-runoff model (Figure 8.3). Uncertainty due to the choice of rainfall-runoff models was estimated by calculating median value from the fifty forecasting simulations using a GCM and then calculating the C V for the two median values estimated from the two rainfall-runoff models (Figure 8.4). University of Western Sydney Page 148

186 CHAPTER 8: Performance of BMWSS (iv) Parameter uncertainty was calculated for the AWBM and SIMHYD models, separately. Median values of the simulated runoffs were calculated from the fifty simulations using a GCM and rainfallrunoff model, and adopting three sets of parameters. Thereafter, the C V value was calculated from the three median values (Figure 8.5). Rainfall-runoff models (AWBM/SIMHYD) Model parameter sets (Set1/Set2/Set3) 50 simulations using 50 simulations using 50 simulations using 50 simulations using CSIRO data CCCMA data ECHAM 5 data MIROC data Estimation of median Estimation of median Estimation of median Estimation of median Calculation of Cv Figure 8.2 Framework of estimating uncertainty due to choice of GCM (i.e. GCM uncertainty) Assessing the reliability of a water supply system In this chapter, assessment of the performance of the Blue Mountains Water Supply System (BMWSS) under future climate conditions was conducted adopting three scenarios: (i) the most probable scenario [50 th percentile of forecasted water demand (i.e. median demand) + 50 th percentile of forecasted runoff (i.e. median catchment yield)] and (ii) the most favourable scenario [5 th percentile of forecasted water demand (i.e. low demand) + 95 th percentile of forecasted runoff (i.e. high catchment yield)] and (iii) the worst scenario [95 th percentile of forecasted water demand (i.e. high demand) + 5 th percentile of forecasted runoff (i.e. low catchment yield)]. University of Western Sydney Page 149

187 CHAPTER 8: Performance of BMWSS Two criteria were used to evaluate the performance of the water supply system: (i) reliability, and (ii) security. Reliability is defined as the percentage of months when water restrictions will not need to be applied more than 3% of total time (restriction will be introduced when total storage becomes 50%). Security is defined as the percentage of months when the total storage will not be lower than 5% for more than 0.001% of time (one month in 100,000 months). These two criteria are used by Sydney Catchment Authority along with another criterion (i.e. robustness) to ensure the effective performance of the water supply systems they maintained (Sydney Catchment Authority 2009b). Rainfall-runoff models (AWBM/SIMHYD) Model parameter sets (Set1/Set2/Set3) 50 simulations using CSIRO/CCCMA/ECHAM 5/MIROC data Estimation of median Calculation of Cv Figure 8.3 Framework of estimating uncertainty due to internal variability of a GCM (i.e. realisation uncertainty) Assessment of the water supply system was conducted for the period of which corresponds to 240 months. For each of the month, water balance was calculated using equation 8.1 and thereafter total number of months was counted for each of the performance criteria. During the water balance calculation, total storage volume of the dams was assumed to be full in the first month (i.e ML). As discussed in Chapter 3 (Section 3.2), other than the Blue Mountains dams, the BMWSS gets water from two other sources: the Fish River Water Scheme (FRWS) University of Western Sydney Page 150

188 CHAPTER 8: Performance of BMWSS and the Warragamba dam during dry conditions and when required. During the performance assessment of the water supply system in the future, two cases were considered: (i) no water supply from the FRWS and (ii) a fixed supply of 285 ML/month, which is close to the upper limit of water allocation from the FRWS supply to the BMWSS (Sydney Catchment Authority that maintains the Blue Mountains dams is licensed to get 300 ML water per month from the FRWS when needed to supplement the BMWSS). In the assessment calculation, no water supply from the Warragamba dam to the BMWSS was assumed during the forecasting periods. WB = I + TS + FRWS + R WD E (8.1) where WB = water balance ML/month; I = runoff in ML/month; TS = total storage in ML; FRWS = fish river water scheme; R = monthly total rainfall in the dams in ML/month; WD = total water demand in ML/month and E = total monthly evaporation from the dams. Rainfall-runoff models AWBM SIMHYD Model parameter sets (Set1/Set2/Set3) Model parameter sets (Set1/Set2/Set3) 50 simulations using CSIRO/CCCMA/ECHAM 5/MIROC data 50 simulations using CSIRO/CCCMA/ECHAM 5/MIROC data Estimation of median Estimation of median Calculation of Cv Figure 8.4 Framework of estimating uncertainty due to choice of rainfall-runoff models (i.e. rainfall-runoff model uncertainty) University of Western Sydney Page 151

189 CHAPTER 8: Performance of BMWSS Rainfall-runoff models (AWBM/SIMHYD) Model parameter set (Set1) Model parameter set (Set2) Model parameter set (Set3) 50 simulations using CSIRO/CCCMA/ECHAM 5/MIROC data 50 simulations using CSIRO/CCCMA/ECHAM 5/MIROC data 50 simulations using CSIRO/CCCMA/ECHAM 5/MIROC data Estimation of median Estimation of median Estimation of median Calculation of Cv Figure 8.5 Framework of estimating uncertainty due to choice of rainfall-runoff model parameter sets (i.e. rainfall-runoff model parameter uncertainty) After calculating the water balance for each simulation month, the reliability and the security criteria were estimated using equations 8.2 and 8.3, which are given below: Reliability = number of months of water restriction number of total simulation months 100 (8.2) Security = number of months total storage falls below 5% number of total simulation months 100 (8.3) As discussed in Chapter 5, residential water demand (single and multiple dwelling sector) was estimated with the projected climate data adopting a Monte Carlo simulation technique. Water demand was forecasted under 12 plausible future scenarios considering four water restriction levels (No restriction, Level 1, Level 2 and Level 3 water restrictions) and three climate scenarios (A1B, A2 and B1). From the probabilistic water demand forecasting, three values of future water demand (5 th, 50 th and 95 th percentiles) were picked under each twelve water demand scenarios. Thereafter, water demand for the commercial sector was calculated by multiplying the total residential demand by 20/80 as the commercial water demand is University of Western Sydney Page 152

190 CHAPTER 8: Performance of BMWSS approximately 20% (as mentioned in Chapter 3, Section 3.6.1) of the total water demand in the Blue Mountains region. Thereafter, residential and commercial water demands were added to get the total water demand. In this way total 36 water demand scenarios were generated as illustrated in Figure 8.6. (i) A1B - L1 : (5 th,50 th, 95 th ) (ii) A1B L2 : (5 th,50 th, 95 th ) (iii) A1B L3 : (5 th,50 th, 95 th ) (iv) A1B No : (5 th,50 th, 95 th ) (i) A2 - L1 : (5 th,50 th, 95 th ) (ii) A2 L2 : (5 th,50 th, 95 th ) (iii) A2 L3 : (5 th,50 th, 95 th ) (iv) A2 No : (5 th,50 th, 95 th ) (i) B1 - L1 : (5 th,50 th, 95 th ) (ii) B1 L2 : (5 th,50 th, 95 th ) (iii) B1 L3 : (5 th,50 th, 95 th ) (iv) B1 No : (5 th,50 th, 95 th ) 12 Scenarios 12 Scenarios 12 Scenarios Figure 8.6 Forecasted 36 total water demand scenarios for the period of In this chapter, future runoffs were estimated using the projected climate data from the four GCMs. From the range of forecasting results, three forecasting values: 5 th, 50 th and 95 th percentiles were picked as the likely water demand scenarios under each GCMs projection. In this way, total twelve runoff scenarios were generated as illustrated in Figure 8.7. (i) CSIRO: (5 th,50 th, 95 th ) (ii) CCCMA: (5 th,50 th, 95 th ) (iii) ECHAM 5: (5 th,50 th, 95 th ) (iv) MIROC: (5 th,50 th, 95 th ) Figure 8.7 Forecasted 12 runoff scenarios for the period of Assessment of the performance of the BMWSS under changing climate conditions was conducted by combining the forecasted water demand and runoff scenarios using equation 8.1. Total 144 simulations were considered using the 36 water demand and 12 runoff scenarios (e.g. each demand scenario with four runoff scenarios). As an example, twelve simulation scenarios with A1B-L1 water demand and four runoff scenarios to assess the performance of the BMWSS are presented in Table 8.1. University of Western Sydney Page 153

191 CHAPTER 8: Performance of BMWSS Table 8.1 Twelve combinations of future water demand and runoff scenarios to assess the performance of the Blue Mountains Water Supply System SN Water demand scenario Runoff scenario Rainfall Evaporation Scenario description 1 CSIRO (95 th CSIRO CSIRO Most favourable 2 3 A1B-L1 (5 th CCCMA (95 th ECHAM 5 (95 th CCCMA CSIRO Most favourable ECHAM 5 CSIRO Most favourable 4 MIROC (95 th MIROC CSIRO Most favourable 5 CSIRO (50 th CSIRO CSIRO Most probable 6 7 A1B-L1 (50 th CCCMA (50 th ECHAM 5 (50 th CCCMA CSIRO Most probable ECHAM 5 CSIRO Most probable 8 MIROC (50 th MIROC CSIRO Most probable 9 CSIRO (5 th CSIRO CSIRO Worst A1B-L1 (95 th CCCMA (5 th ECHAM 5 (5 th CCCMA CSIRO Worst ECHAM 5 CSIRO Worst 12 MIROC (5 th MIROC CSIRO Worst 8.3 Results Rainfall projections A comparison between the annual average rainfall for the period and the projected rainfall changes for the Katoomba weather station using the four GCMs are presented in Table 8.2. Three statistics for the projected rainfall changes, the 5 th percentile (dry scenario), the median and the 95 th percentile (wet scenario), are presented for the two decades: and The final row of the table University of Western Sydney Page 154

192 CHAPTER 8: Performance of BMWSS (marked in bold) contains the ensemble results from all of the four GCMs. These results indicate that rainfall would be reduced by 5% and 6% for the decades of and , respectively. However, considerable uncertainty were found in the rainfall projections with the ranges of 5 th and 95 th percentiles being -27% to 33% and -29% to 25% for the decades of and , respectively. Table 8.2 Percentage of rainfall changes in the future decades projected by the four GCMs compared to annual average rainfall during the period Decade Decade GCMs 5 th percentile Media n 95 th percentile 5 th percentile Media n 95 th percentile CSIRO CCCMA ECHAM MIROC Ensemble median Uncertainty due to internal variability of a GCM (Realisation uncertainty) Fifty realisations of the downscaled rainfall data from the ECHAM 5 GCM were taken into the AWBM model to generate 50 runoff simulations for the periods to estimate the uncertainty due to the internal variability of a GCM. The results show that the variability is high among the simulated results. The C V values were calculated from the estimated 50 runoff values and are presented in Figure 8.8, it can be seen that the simulated results vary considerably around the mean, with the mean C V value of around 36-38% for the forecasted periods. The forecasted results show that the 50 simulated runoffs are not in good agreement indicating a large uncertainty within simulated runoffs due to the realisation uncertainty. Similar results were obtained for the other GCM predictions (i.e. MIROC, CSIRO and CCCMA); these results are presented in Figures D.8.1, D.8.2 and D.8.3 in Appendix D. University of Western Sydney Page 155

193 Coefficient of Variation Climate change impact on water demand and supply CHAPTER 8: Performance of BMWSS Uncertainty due to choice of GCMs (GCM uncertainty) In order to estimate the uncertainty due to the choice of GCMs, the AWBM model was run with one set of calibrated parameters using 50 realisations of downscaled rainfall data from the four GCMs (i.e. CSIRO, CCCMA, ECHAM 5 and MIROC). Finally, 200 runoff simulation values (50 runoff values using each GCM) were obtained by running the AWBM with the above configuration. Thereafter median projections of the runoffs using the four GCMs were compared with each other to estimate the uncertainty and the C V values were calculated using 4 median runoff values. The C V values are presented in Figure 8.9, it can be seen that the variations in the simulated runoffs are noticeably high among the results based on the data from the four GCMs. C V values of the simulated runoffs were sometimes found to be as high as 80% with a mean value of 50%, indicating a considerable difference between the simulated results. These results demonstrated that a significant uncertainty is associated with the runoff estimates due to the differences in the GCMs predictions Year Figure 8.8 Coefficient of variation (C V ) values of the simulated runoffs using ECHAM 5 model data (i.e. realisation uncertainty). The red horizontal line represents the average C V value University of Western Sydney Page 156

194 Coefficient of Variation Climate change impact on water demand and supply CHAPTER 8: Performance of BMWSS Year Figure 8.9 Coefficient of variation (C V ) values of the simulated median runoffs using data from the four GCMs (i.e. CSIRO, CCCMA, ECHAM 5, MIROC). The red horizontal line represents the average C V value Uncertainty due to choice of rainfall-runoff models To estimate the uncertainty due to choice of rainfall-runoff models, 50 realisations of the downscaled rainfall data from the CSIRO GCM were taken as inputs to the AWBM and SIMHYD models to simulate runoffs by these two models. Finally, 100 runoff simulation values (50 runoff values using each rainfall-runoff model) were obtained using CSIRO GCM. Then the median projections by these two models were compared and C V values were calculated based on the two median runoff values obtained from the estimated 50 runoff values by each rainfall-runoff model to assess the uncertainty. Figure 8.10 presents the uncertainty due to the rainfall-runoff model by showing C V values for the periods. It can be seen that the average C V values are in the range of 3 to 4%, indicating less variation among the simulated results by the models. The results demonstrate that uncertainty associated with the simulated runoffs due to the choice of rainfall-runoff models is relatively small Uncertainty due to choice of rainfall-runoff model parameter In order to estimate the uncertainty due to the choice of calibrated rainfall-runoff model parameter sets, the AWBM model was run with the fifty realisations of the downscaled rainfall data from the CCCMA GCM using three sets of calibrated parameters. Finally, 150 runoff simulation values (50 runoff values using each University of Western Sydney Page 157

195 Coefficient of Variation Climate change impact on water demand and supply CHAPTER 8: Performance of BMWSS parameter set) were obtained by running AWBM and using CCCMA GCM. Thereafter, median projections of the simulated runoff by the three sets of parameters were compared and C V values were calculated based on the three median runoff values obtained from the estimated 50 runoff values using each parameter set to estimate the uncertainty. The C V values of the simulated runoffs for the periods are presented in Figure 8.11; the C V values were found to be quite low with an average value of around 1%. The results indicate that the simulated runoffs by the three different parameter sets are close to each other. Similar results were found adopting the SIMHYD model with the calibrated SIMHYD parameter sets. These results demonstrate that the uncertainty associated with the simulated runoffs due to the different parameter sets is relatively small Year Figure 8.10 Coefficient of variation (C V ) values of the simulated median runoffs using the CSIRO global climate model data adopting the AWBM and SIMHYD models (i.e. rainfall-runoff model uncertainty). The red horizontal line represents the average C V value Comparison of uncertainties Relative magnitudes (in terms of C V values in %) of the four different types of estimated uncertainties in the forecasting of runoffs are presented in Figure 8.12, which shows that uncertainty due to different GCMs is remarkably higher than the University of Western Sydney Page 158

196 Coefficient of Variation Climate change impact on water demand and supply CHAPTER 8: Performance of BMWSS other three types. In this chapter, the realisation uncertainty was also found to be considerably high being in the second position in terms of C V value. The rainfallrunoff models demonstrated some differences in the simulated results but those were quite minor. The uncertainty due to rainfall-runoff model parameters was found to be less than that of the rainfall-runoff models and the lowest among all the four sources of uncertainties. These results are comparable to the findings of other recent global studies on climate change impact analysis such as Prudhomme and Davies (2009), Kay et al. (2009), Chen et al. (2011) and Gosling et al. (2011). These studies concluded that uncertainties linked to GCMs were the most dominant uncertainty in the climate change impact studies on water resources. A similar conclusion in regards to the rainfall-runoff model and parameter uncertainty was also made by Chen et al. (2011) and Poulin et al. (2011) that these two sources of uncertainties were not significant Year Figure 8.11 Coefficient of variation (C V ) values of the simulated median runoffs by the AWBM model using the CCCMA GCM data adopting three different calibrated parameter sets (i.e. rainfall-runoff model parameter uncertainty). The red horizontal line represents the average C V value University of Western Sydney Page 159

197 Average CV (%) Climate change impact on water demand and supply CHAPTER 8: Performance of BMWSS GCMs uncertainty Realisation uncertainty Rainfall-runoff models uncertainty Rainfall-runoff models parameter uncertainty Type of uncertainties Figure 8.12 Comparison of four types of uncertainties by presenting the average C V (%) values of the forecasted runoff Similar results were also found by some recent Australian studies. For example, Chiew et al. (2009), Chiew et al. (2010) and Teng et al. (2012) found that the uncertainties in runoff projections were mostly dominated by the choice of the GCMs. They also found that the choice of hydrological model was the least significant source of uncertainty. Crosbie et al. (2011) demonstrated that the greatest source of uncertainty was linked with the GCMs projections during investigating the impact of climate change on groundwater recharge at three locations across southern Australia. They found the differences between the highest and lowest projections to be 53%, 44% and 24% for the GCM, downscaling and hydrological model uncertainty, respectively. Teng et al. (2012) also found that the uncertainty linked to GCM was much larger than the uncertainty in the rainfall-runoff models during the investigation of climate change impact on runoff across southeast Australia. They found 28 to 35% variations in the results based on 15 different GCMs and one rainfall-runoff model. On the other hand, they found less than 7% variation in runoff results based on five different rainfall-runoff models and one GCM Forecasting of runoffs (Median projections) Table 8.3 presents the % changes of median projections in mean annual runoff of the Blue Mountains catchments in the forecasting decades (i.e , ) with that of the reference period (i.e ). The results show that runoff would be reduced by 9% (CCCMA) to 49% (CSIRO) and 25% (ECHAM 5) to 48% University of Western Sydney Page 160