adaptation of water sensitive urban … mohamed_rasheed...adaptation of water sensitive urban design...

ADAPTATION OF WATER SENSITIVE URBAN DESIGN TO CLIMATE CHANGE

ASHIQ M. RASHEED

B.Sc (Civil Engineering, Honours)

A thesis submitted in fulfilment of the requirement of the degree of

Doctor of Philosophy

Science and Engineering Faculty Queensland University of Technology

2018

This thesis is dedicated to my beloved parents, Ayna & Abdul Rasheed for their unconditional love, support and encouragement throughout.

You raise me up, so I can stand on mountains You raise me up to walk on stormy seas

I am strong when I am on your shoulders You raise me up to more than I can be

- Brendan Graham -

I

Abstract

Water Sensitive Urban Design (WSUD) is the stormwater management philosophy adopted in Australia to manage stormwater quality and quantity to minimise the impacts of urban developments on the surrounding environment. However, the treatment systems adopted in WSUD along with the stormwater quality estimation are typically designed considering static climate conditions. Changing climate in future scenarios including changes to rainfall patterns and characteristics of rainfall can reduce the effectiveness of such WSUD systems. However, there is not a robust methodology available to support the adaptation of WSUD to climate change. This is primarily due to lack of future climate data for frequent rainfall events at the small catchment scale. However, there is no appropriate methodology available for downscaling or generating such high-resolution rainfall data. In this research, methodologies were developed to generate high-resolution rainfall data for future climate change scenarios and impacts of climate change on the stormwater quality and quantities were assessed using the generated rainfall data. The research was undertaken taking southeast Queensland, Australia as the study area. Accordingly, a detailed analysis was carried out to test the rainfall homogeneity within the region so that representative meteorological stations can be selected. Long-term rainfall data sets were obtained for representative meteorological stations to facilitate the analysis. Two separate statistical downscaling models were developed to spatially and temporally downscale rainfall data from two GCMs, EC-EARTH and ACCESS 1.0 for two climate change scenarios, RCP 4.5 and RCP 8.5. The downscaled data were then used to assess the impacts of climate change on the stormwater quality and quantities. The impact assessment included at-site frequency analysis to generate Intensity-Frequency-Duration (IFD) curves and estimation of stormwater quality for the future climate change scenarios.

The analysis undertaken to evaluate the degree of rainfall homogeneity suggested that the entire southeast Queensland can be treated as homogeneous region based on the characteristics of the continuous rainfall. However, based on individual rainfall characteristics such as antecedent dry-days, maximum rainfall intensities, total rainfall and duration of the rainfall events, there were two separate homogeneous regions identified, namely, Coastal-SEQ and Inland-SEQ. Thereby, Gold Coast Seaway station (40764) and Toowoomba Airport stations (41529) were selected to represent the Coastal-SEQ and Inland-SEQ respectively.

A new spatial downscaling tool, ‘spdownscale’ was developed based on quantile-quantile bias correction approach. This tool was used to spatially downscale rainfall data for

II

future climate change scenarios at the two representative meteorological stations for southeast Queensland. Overall, the models developed by the spdownscale for spatial downscaling performed well for both GCMs. Two statistical indexes, RMSE and the gradient of observation-simulation scatter plots were used for the validation purposes. The bias-corrected GCM outputs from both the GCMs were closely comparable to the observed data. A new temporal downscaling model was developed based on first-order homogeneous Markov model to translate the 3-hour rainfall time-series into 5-minute time-series. This model was used to temporally downscale the bias-corrected (spatially downscaled) rainfall data at the two representative meteorological stations for southeast Queensland. The model performance was assessed based on an independent historical rainfall dataset. Overall, the simulated rainfall was in agreement with the observed rainfall in terms of the probability distribution and the maximum rainfall intensities at both representative meteorological stations.

IFD curves generated using at-site frequency analysis for future climate change scenarios. Overall, there was a significant increase in the IFDs for the future climate change scenarios compared to the present IFD provided by the Bureau of Meteorology (BoM). In general, smaller duration frequent rainfall and the longer duration infrequent rainfalls were expected to increase significantly in both climate change scenarios. On average, the IFDs for the Coastal-SEQ were expected to increase by 23-30% in the near future and 38-45% distant future. The IFDs for the Inland-SEQ were expected to increase by 5-15% in the near future and 37-38% in the distance future. Based on the estimated changes in the design rainfall, this research suggests an approach to estimate the design flow rates in the design of WSUD treatment system by introducing a climate change factor Cf into the Rational Method procedure. The values for the Cf for southeast Queensland is also presented and proposed to be incorporated into the city council guidelines for WSUD in order to adapt the impacts of climate change in the design of WSUD.

A stormwater quality model was developed to estimate the Event Mean Concentration (EMC) of the Total Suspended Solids (TSS) generated from urban residential catchments for future climate change scenarios. The model was designed to automatically extract all independent rainfall events from a given rainfall time-series and simulate water quality parameters based on event-based rainfall characteristics. Overall, the pollutant export showed a varying pattern in the future climate change scenario. In Coastal-SEQ, the median pollutant export was expected to increase by 15% and 9% for RCP 4.5 (2026-2045) and RCP 4.5 (2081-2100), whereas, the median pollutant export was expected to decrease by 10% and 2% for RCP 8.5 (2026-2045) and RCP 8.5 (2081-2100).In Inland-SEQ, the median pollutant export was expected to decrease by 13%, 21% and 13% for RCP 4.5 (2026-2045), RCP 8.5 (2026-2045) and

III

RCP 8.5 (2081-2100) respectively and a slight increase of 1% for RCP 4.5 (2081-2100). However, the rainfall runoff was estimated to increase significantly and thus resulting in significantly low pollutant concentrations in the future stormwater runoffs. The ensemble median of the EMC showed a decrease of 50%, 53%, 48% and 47% for RCP 4.5 (2026-2045), RCP 4.5 (2081-2100), RCP 8. 5 (2026-2045) and RCP 8.5 (2081-2100) climate change scenarios respectively in Coastal-SEQ and the ensemble median of the EMC showed a decrease of 38%, 48%, 44% and 43% for RCP 4.5 (2026-2045), RCP 4.5 (2081-2100), RCP 8.5 (2026-2045) and RCP 8.5 (2081-2100) climate change scenarios respectively in Inland-SEQ. Based on these results a set of new water quality parameters for future climate change scenarios were developed and proposed to be incorporated into the Model for Urban Stormwater Improvement Conceptualisation (MUSIC) guidelines of the southeast Queensland in order to adapt the impacts of climate change in the design of WSUD treatment systems.

V

Keywords

Water Sensitive Urban Design, pollutant process, stormwater quality, climate change, downscaling.

VII

List of publications

Rasheed AM, Egodawatta P, Goonetilleke A and McGree J (2017), spdownscale: Spatial Downscaling Using Bias Correction Approach, R package version 0.1.0. https://CRAN.R-project.org/package=spdownscale

Rasheed AM, Egodawatta P, Goonetilleke A and McGree J (2017), ‘spdownscale’ A Spatial Downscaling Tool Based on Bias Correction Approach, 7th IWA-ASPIRE Conference 2017 & Water Malaysia Exhibition 2017.

https://cran.r-project.org/package=spdownscale

IX

Statement of Authorship

The work contained in this thesis has not been previously submitted to meet requirements for an award at this or any other higher education institution. To the best of my knowledge and belief, the thesis contains no material previously published or written by another person except where due reference is made.

Ashiq M. Rasheed November 2018

QUT Verified Signature

XI

Acknowledgements

I wish to express my sincere gratitude to my principal supervisor Dr. Prasanna Egodawatta for his guidance, support and accomplished academic supervision. My appreciation is further extended to my associate supervisors Prof. Ashantha Goonetilleke and A/Prof. James McGree for their valuable expert advice and guidance during the candidature.

I would like to express my appreciation to the High-Performance Computing and Advanced Research Computing Group of QUT, particularly to Mr. Abdul Sharif for the HPC training and support provided. I also greatly acknowledge the timely support received from the IT helpdesk team of QUT.

I would like to acknowledge the support from Bureau of Meteorology, Australia for their support in the data collection for the research. A special thanks to Ms. Tamika Tihema for her support and advice. I am also grateful to the Program for Climate Model Diagnosis and Intercomparison (PCMDI) for providing CMIP5 GCM data for this research.

A special thanks to Prof. Ashish Sharma for his support and feedbacks given to improve the research methodology.

Finally, I like to express my gratitude to my siblings and friends for their encouragement and unconditional love.

XIII

Table of Content

Chapter 1 Introduction ..................................................................................... 1

1.1 Background ......................................................................................................... 1

1.2 Research problem ................................................................................................ 2

1.3 Aims and objectives ............................................................................................ 3

1.4 Justification for the research ............................................................................... 3

1.5 Description of the research .................................................................................. 4

1.6 Scope ................................................................................................................... 5

1.7 Outline of the thesis ............................................................................................ 6

Chapter 2 Water Sensitive Urban Design ........................................................... 7

2.1 Background ......................................................................................................... 7

2.2 The concept of Water Sensitive Urban Design .................................................... 8

2.2.1 Non-structural measures ............................................................................... 9

2.2.2 Structural Measures .................................................................................... 10

2.3 WSUD treatment systems and their treatment and hydraulic design ................ 13

2.3.1 Swale ........................................................................................................... 13

2.3.2 Bioretention basin ....................................................................................... 14

2.3.3 Constructed wetlands .................................................................................. 16

2.4 Stormwater pollutants ........................................................................................ 19

2.4.1 Nutrients ..................................................................................................... 19

2.4.2 Organic Carbon ........................................................................................... 20

2.4.3 Heavy Metals ............................................................................................... 20

2.4.4 Hydrocarbons .............................................................................................. 20

2.4.5 Suspended Solids ......................................................................................... 21

2.5 Pollutant process ................................................................................................ 22

2.5.1 Pollutant build-up ....................................................................................... 22

2.5.2 Pollutant wash-off ....................................................................................... 24

XIV

2.6 Stormwater quality modelling ........................................................................... 27

2.6.1 Hydrological modelling ............................................................................... 27

2.6.2 Water quality modelling ............................................................................. 28

2.7 Conclusions........................................................................................................ 30

Chapter 3 Climate Change and Downscaling ................................................... 33

3.1 Background ....................................................................................................... 33

3.2 Climate change .................................................................................................. 34

3.3 Emission scenarios ............................................................................................. 35

3.4 Global Circulation Models ................................................................................. 40

3.5 Downscaling ...................................................................................................... 45

3.5.1 Comparison of statistical and dynamic downscaling methods .................... 46

3.5.2 Statistical downscaling ............................................................................... 48

3.5.3 Components of a statistical downscaling scheme ........................................ 51

3.5.4 Statistical downscaling tools ....................................................................... 53

3.6. Uncertainties in climate change projections ..................................................... 54

3.7 Conclusions........................................................................................................ 56

Chapter 4 Research Design and Methods ......................................................... 59

4.1 Background ....................................................................................................... 59

4.2 Research design ................................................................................................. 60

4.2.1 Critical review of literature ........................................................................ 62

4.2.2 Selection of study area, study tools and analytical methods ...................... 62

4.2.3 Data collection ............................................................................................ 62

4.2.4 Modelling and analysis ............................................................................... 63

4.3 Study tools ........................................................................................................ 65

4.3.1 Programming platform ............................................................................... 65

4.3.2 Climate data operators ............................................................................... 67

4.4 Analytical methods ............................................................................................ 68

4.4.1 Homogeneous Analysis ............................................................................... 68

4.4.2 Spatial Downscaling ................................................................................... 74

XV

4.4.3 Markov model ............................................................................................. 77

4.4.4 Rainfall frequency analysis .......................................................................... 79

4.4.5 Stormwater quality modelling ..................................................................... 81

4.4.6 Classical univariate data analysis ................................................................ 84

4.5 Conclusions ........................................................................................................ 85

Chapter 5 Selection of Representative Meteorological Stations for Downscaling 87

5.1 Background ........................................................................................................ 87

5.2 Study area and data collection ........................................................................... 88

5.2.1 Study area ................................................................................................... 88

5.2.2 Data collection ............................................................................................ 89

5.3 Assessment of rainfall homogeneity in southeast Queensland ............................ 90

5.3.1 Continuous-rainfall approach ...................................................................... 91

5.3.2 Event-based rainfall approach ..................................................................... 92

5.4 Boundaries and representative meteorological station of the homogeneous regions ...................................................................................................................... 98

5.4.1 Rainfall homogeneous regions within southeast Queensland ....................... 98

5.4.2 Representative meteorological stations for southeast Queensland ............. 100

5.5 Conclusions ...................................................................................................... 101

Chapter 6 Spatial Downscaling of Rainfall Data Using Bias Correction Method .................................................................................................................... 103

6.1 Background ...................................................................................................... 103

6.2 Development of the spatial downscaling tool ................................................... 104

6.2.1 Downscaling method ................................................................................. 104

6.2.2 Architecture of the downscaling tool ......................................................... 105

6.2.3 Functions of spdownscale .......................................................................... 111

6.4 Spatial downscaling of rainfall data for southeast Queensland (SEQ) ............. 113

6.4.1 Spatial downscaling for Coastal-SEQ ........................................................ 113

6.4.2 Spatial downscaling for Inland-SEQ .......................................................... 119

6.4 Conclusions ...................................................................................................... 122

XVI

Chapter 7 Temporal Downscaling of Rainfall Data Using First-order Markov Model ........................................................................................................... 123

7.1 Background ...................................................................................................... 123

7.2 First-order homogeneous Markov model .......................................................... 124

7.2.2 Assumptions used in the model ................................................................. 124

7.2.3 Architecture of the model .......................................................................... 125

7.3 Temporal downscaling for southeast Queensland ............................................. 127

7.3.1 Calibration ................................................................................................. 128

7.3.2 Validation .................................................................................................. 128

7.3.3 Future rainfall generation .......................................................................... 135

7.4 Conclusions....................................................................................................... 135

Chapter 8 Design Rainfall for Future Climate Change Scenarios ..................... 137

8.1 Background ...................................................................................................... 137

8.2 Rainfall frequency analysis ............................................................................... 138

8.2.1 IFD generation for the historical data ....................................................... 140

8.2.2 IFD generation for future climate change scenarios .................................. 144

8.3 Adaptation of WSUD to changes in the future IFDs ....................................... 158

8.4 Conclusions....................................................................................................... 161

Chapter 9 Impact of Climate Change on Pollutant Export and Stormwater Quality ......................................................................................................... 163

9.1 Background ...................................................................................................... 163

9.2 Model setup ...................................................................................................... 164

9.2.1 Catchment ................................................................................................. 166

9.3.1 Event separations ...................................................................................... 167

9.3.2 Pollutant process modelling ....................................................................... 167

9.3.3 Runoff modelling ....................................................................................... 169

9.3 Impacts of climate change on pollutant process ............................................... 170

9.3.1 Changes in pollutant processes in the Coastal-SEQ .................................. 171

9.3.2 Changes in pollutant process in the Inland-SEQ ....................................... 175

XVII

9.4 Impacts of climate change on water quality ..................................................... 179

9.4.1 Changes in stormwater quality in the Coastal-SEQ .................................. 179

9.4.2 Changes in stormwater quality in the Inland-SEQ .................................... 180

9.5 Water quality parameters for MUSIC modelling ............................................. 182

9.7 Conclusions ...................................................................................................... 187

Chapter 10 Conclusions and Recommendations .............................................. 189

10.1 Conclusions ..................................................................................................... 189

10.1.1 Event-based rainfall homogeneity assessment for southeast Queensland . 190

10.1.2 Downscaling of rainfall data .................................................................... 191

10.1.3 Design rainfall for future climate change scenarios .................................. 192

10.1.4 Stormwater quality and quantity characteristics in the future climate change scenarios ............................................................................................................. 193

10.2 Recommendations........................................................................................... 195

References .................................................................................................... 197

Appendix A .................................................................................................. 223

Appendix B .................................................................................................. 225

B.1: Source code for ParaCal() function of spdownscale .................................... 225

B.2: Source code for ResVal() function of spdownscale ...................................... 227

B.3: Source code for downscale() function of spdownscale ................................. 232

Appendix C .................................................................................................. 235

C1: Example of First-order Markov process....................................................... 235

C2: Sample source code for temporal downscaling (station: 40476, GCM: ACCESS 1.0, RCP: RCP 4.5, period: 20206-2045) ............................................................ 237

Appendix D .................................................................................................. 245

Appendix E .................................................................................................. 253

XIX

List of Figures

Figure 2.1: Incorporation of BPPs and BMPs in WSUD (adapted and reproduced from Whelans et al. 1994; Mangangka, 2013) ........................................................................ 9

Figure 2.2: Typical stormwater treatment systems, target pollutant size and hydraulic loading (Adapted and reproduced from Wong et al., 2000) ......................................... 12

Figure 2.3: Vegetated swales ........................................................................................ 14

Figure 2.4: Typical cross section of a bioretention system (Adapted and reproduced from BCC & MWB, 2006).................................................................................................... 15

Figure 2.5: Typical section of a constructed wetland (Adapted and reproduced from VSC, 1999) ................................................................................................................... 17

Figure 2.6: Conceptual chain of pollutant process (adapted and reproduced from Goonetilleke et al., 2014) ............................................................................................. 22

Figure 2.7: Pollutant build-up in different land uses (adapted from Sartor et al., 1974) ..................................................................................................................................... 23

Figure 2.8: Hypothetical representations of surface pollutant load over time (Adapted and reproduced from Vaze and Chiew, 2002) .............................................................. 26

Figure 2.9: Hydrological process (adapted from O’Loughlin and Stacks, 2004) ........... 28

Figure 3.1: (a) CO2 emission (b) cumulative CO2 emission for SRES storyline from 1990 to 2100 (adapted from IPCC (2000) and reproduced) ................................................. 37

Figure 3.2: (a) Radiative forcing (b) Corresponding CO2 emission pathways for RCPs (adapted and reproduced from van-Vuuren et al. (2011)) ........................................... 39

Figure 3.3: Average (1986-2005) rainfall simulations from CMIP5 models for (a) summer and (b) winter (Adapted from CCIA (2015)). The regions are from the NRM cluster (see NRM cluster - see chapter 3, CCIA (2015)). ........................................................ 43

Figure 3.4: The average annual cycle of rainfall for Australia (Regions: AUS-Australia, EA- East Australia, NA- North Australia, R-Rangelands, SA – South Australia and SS- Southern Slopes) (Adapted from CCIA (2015)) .................................................... 44

Figure 3.5: Component of statistical downscaling (Adapted from Diaz-Nieto and Wilby (2005) and Wilby et al. (2004)) ................................................................................... 52

Figure 4.1: Steps of the research design ....................................................................... 61

XX

Figure 4.2: Modelling and analysis .............................................................................. 64

Figure 4.3: Basic algorithm of a K-means cluster analysis .......................................... 70

Figure 4.4: Basic algorithm of an agglomerative hierarchical cluster analysis ............. 71

Figure 4.5: Graphical definitions of cluster proximity ................................................. 72

Figure 4.6: Definition sketch for heterogeneity (Adapted from Hosking and Wallis (1997))......................................................................................................................... 73

Figure 4.7: Probability parameters for at-site frequency analysis ............................... 80

Figure 5.1: Southeast Queensland ............................................................................... 88

Figure 5.2: The locations of the selected meteorological stations ................................ 90

Figure 5.3: Dendrogram generated from the cluster analysis ...................................... 95

Figure 5.4: Geographical locations of the meteorological stations and their grouping 96

Figure 5.5: Scatterplots of the event-based rainfall characteristics (Red dots refer to the stations of Cluster 1, Black dots refer to the stations of - Cluster 2 and Green dots refer to the stations of Cluster 3) ........................................................................................ 97

Figure 5.6: Boundaries of Coastal-SEQ and Inland-SEQ rainfall homogeneous regions .................................................................................................................................... 99

Figure 6.1: Architecture of downscaling tool .............................................................. 106

Figure 6.2: Developing the statistical model for bias correction. (a) Mapping zero rainfall; (b) The Polynomial relationship between the threshold values of the observed data and GCM data. Three points have been used for the curve – red dot is found using the calibration data and the black dots refer to the lowest (0) and the largest (1) possible values; (c) Mapping non-zero rainfall. ........................................................................ 109

Figure 6.3: Calibration parameters for EC-EARTH (Gold Coast Seaway) ................ 114

Figure 6.4: Calibration parameters for ACCESS 1.0 (Gold Coast Seaway) ............... 115

Figure 6.5: Validation results for EC-EARTH (Gold Coast Seaway) ........................ 117

Figure 6.6: Validation results for ACCESS 1.0 (Gold Coast Seaway) ....................... 118

Figure 6.7: Calibration results for EC-EARTH (Toowoomba Airport) ...................... 120

Figure 6.8: Validation results for EC-EARTH (Toowoomba Airport) ....................... 121

Figure 7.1: Flow diagram of the model process .......................................................... 126

Figure 7.2: Validation outputs of the cumulative probability distribution ................ 129

Figure 7.3: Temporal patterns for rainfall data over 3-hour periods .......................... 132

XXI

Figure 7.4: Validation outputs of maximum rainfall intensities. (a) Gold Coast Seaway station and (b) Toowoomba Airport station .............................................................. 133

Figure 7.5: Validation outputs of 3-hour total rainfall. (a) Gold Coast Seaway station and (b) Toowoomba Airport station .......................................................................... 134

Figure 8.1: IFD relationship curves for the Gold Coast Seaway station. The broken lines show the curves generated using the at-site frequency analysis and the solid lines are that from BoM ........................................................................................................... 142

Figure 8.2: IFD relationship curves for the Toowoomba Airport station. The broken lines show the curves generated using the at-site frequency analysis and the solid lines are that from BoM ..................................................................................................... 143

Figure 8.3: IFD curves for Gold Coast Seaway station (40764) for RCP 4.5 climate change scenario for the period 2026-2045. The broken-lines denote the values of realizations (10 for each ARI) and solid lines denote the mean of the realizations .... 145




Figure 8.7: IFD curves for Toowoomba Airport (41529) for RCP 4.5 climate change scenario for the period 2026-2045. The broken-lines denote the values of realizations (10 for each ARI) and solid lines denote the mean of the realizations ............................. 149




XXII

Figure 9.1: Processes associated with the stormwater quality model ......................... 165

Figure 9.2: Conceptual catchment ............................................................................. 166

Figure 9.3: Probability distribution of the pollutant exports (Coastal-SEQ) ............. 174

Figure 9.4: Probability distribution of the pollutant exports (Inland-SEQ)............... 178

Figure 9.5: Water quality parameters ........................................................................ 182

Figure 9.6: EMC for future climate change scenarios for Coastal-SEQ. The dotted lines denote the actual distribution of the data and the solid lines donate the log-normal distribution fitted data (based on the mean and standard deviation)........................ 184

Figure 9.7: EMC for future climate change scenarios for Inland-SEQ. Note: The dotted lines denote the actual distribution of the data and the solid lines donate the log-normal distribution fitted data (based on the mean and standard deviation)........................ 185

XXIII

List of Tables

Table 2.1: Models for pollutant build-up ..................................................................... 24

Table 2.2: Wash-off capacity factors for different rainfall intensities (Adapted from Egodawatta, 2007) ....................................................................................................... 25

Table 3.1: Main characteristics of the four SRES storylines (adapted and reproduced from IPCC (2000)) ....................................................................................................... 36

Table 3.2: List of GCM of CMIP5 and their resolutions (Adapted from IPCC (2014) and CCIA (2015)) ........................................................................................................ 41

Table 3.3: Strengths and weaknesses of statistical and dynamical downscaling (Adapted from Wilby and Dawson (2004), Ahmed et al. (2013), Schmidli et al. (2007), Jaw et al. (2015) and Mearns et al. (1999)) ................................................................................. 47

Table 4.1: Comparison of R and Matlab against the selection criteria ........................ 66

Table 4.2: Comparison of R and Matlab against the selection criteria ........................ 67

Table 4.3: Equations for capacity factors ..................................................................... 83

Table 4.4: Summary of the comparison of Mike Urban, MUSIC and XP-SWMM models with the selection criteria ............................................................................................. 83

Table 5.1: Dispersion indexes based on event-based rainfall approach for southeast Queensland ................................................................................................................... 93

Table 5.2: Data matrix for the cluster analysis ............................................................ 94

Table 5.3: Dispersion indexes for Hosking and Wallis heterogeneity test for Coastal-SEQ and Inland-SEQ ................................................................................................... 98

Table 5.4: Local government bodies included in the identified homogeneous regions .. 99

Table 6.1: Descriptions of the functions and datasets in spdownscale ....................... 111

Table 6.2: A summary of calibration and validation periods used for the downscaling ................................................................................................................................... 114

Table 7.1: A summary of calibration and validation periods used for the temporal downscaling ................................................................................................................ 128

Table 8.1: Return periods and the corresponding frequency factors for EV-I distribution ................................................................................................................................... 140

Table 8.2: IFDs generated using at-site frequency analysis for station 40764 ............ 141

XXIV

Table 8.3: IFDs generated using at-site frequency analysis for station 41529 ............ 141

Table 8.4: Change factors for IFDs for Coastal-SEQ for RCP 4.5 (2026-2045) ......... 154




Table 8.8: Change factors for IFDs for Inland-SEQ for RCP 4.5 (2026-2045) ........... 155

Table 8.9: Change factors for IFDs for Inland-SEQ for RCP 4.5 (2081-2100) ........... 155

Table 8.10: Change factors for IFDs for Inland-SEQ for RCP 8.5 (2026-2045) ......... 156

Table 8.11: Change factors for IFDs for Inland-SEQ for RCP 8.5 (2081-2100) ......... 156

Table 8.12: Comparison of the percentages increase in the IFDs suggested by this research and interim climate change guideline of the AR&R for SEQ ....................... 157

Table 8.13: summary of the treatment and hydraulic design aspects of WSUD systems (BCC & MBW (2006) and GCCC (2005))................................................................. 159

Table 8.14: Proposed change factors for southeast Queensland (2026-2045) .............. 160

Table 8.15: Proposed change factors for southeast Queensland ................................. 162

Table 9.1: Equations for capacity factors ................................................................... 169

Table 9.2: Pollutant build-up and pollutant wash-off (Gold Coast Seaway station) . 171

Table 9.3: Estimated changes in antecedent dry-days for future climate change scenarios (Coastal-SEQ) ............................................................................................................ 172

Table 9.4: Estimated changes in maximum rainfall intensities for future climate change scenarios (Coastal-SEQ) ............................................................................................. 172

Table 9.5: Pollutant build-up and pollutant wash-off (41529) ................................... 175

Table 9.6: Estimated changes in antecedent dry-days for future climate change scenarios (Inland-SEQ) .............................................................................................................. 176

Table 9.7: Estimated changes in maximum rainfall intensities for future climate change scenarios (Inland-SEQ) .............................................................................................. 176

Table 9.8: Estimated changes in EMCs for future climate change scenarios (Coastal-SEQ) .......................................................................................................................... 179

Table 9.9: Pollutant exports and the effective rainfall for the present and future climate change scenarios for Coastal-SEQ .............................................................................. 180

XXV

Table 9.10: Estimated changes in EMCs for future climate change scenarios (Inland-SEQ) .......................................................................................................................... 180

Table 9.11: Pollutant exports and the effective rainfall for the present and future climate change scenarios for Inland-SEQ ................................................................................ 181

Table 9.12: Comparison of EMCs from this study and MUSIC guideline for SEQ .... 183

Table 9.13: Water quality parameters for MUSIC modelling .................................... 186

Table 9.14: Proposed water quality parameters for MUSIC modelling to incorporate climate change ........................................................................................................... 188

Table 10.1: Proposed change factors for southeast Queensland ................................. 193

Table 10.2: Proposed water quality parameters for MUSIC modelling to incorporate climate change ........................................................................................................... 195

XXVII

Abbreviations

AEP Annual Exceedance Probability

ANN Artificial Neural Network

AR&R Australian Rainfall & Runoff

ARI Average Recurrence Interval

BCC Brisbane City Council

BMP Best Management Practice

BoM Bureau of Meteorology

BPP Best Planning Practice

CA Cluster Analysis

CCIA Climate Change in Australia

CDF Cumulative Density Function

CDO Climate Data Operators

CF Capacity Factor

Cf Change Factor

CMIP Coupled Model Intercomparison Project

CRAN Comprehensive R Archive Network

CSIRO Commonwealth Scientific and Industrial Research Organisation

CV Coefficient of Variation

EMC Event Mean Concentration

EV-I Extreme Value Type 1 Distribution

EV-III Extreme Value Type 3 Distribution

GCM Global Climate Models/ Global Circulation Models

GPL General Public Licence

GUI Graphical User Interface

XXVIII

HMM Homogeneous Markov Model

IFD Intensity-Frequency-Duration

IPCC Intergovernmental Panel on Climate Change

KNN K-Nearest Neighbours

kT Frequency Factor

MUSIC Model for Urban Stormwater Improvement Conceptualisation

NHMM Non-Homogeneous Markov Model

NRM National Resource Management

PCA Principal Component Analysis

PDF Probability Density Function

QQ Quantile-Quantile

QUDM Queensland Urban Drainage Manual

RCM Regional Climate Model

RCP Representative Concentration Path

RMSE Root Mean Square Error

SD Standard Deviation

SDM BoM Statistical Downscaling Models, Bureau of Meteorology

SDSM Statistical DownScaling Model

SEQ South East Queensland

SEQHWP South East Queensland Healthy Waterways Partnership

SRES Special Report on Emission Scenarios

SS Suspended Solids

SSE Sum of Square Error

TP Total Phosphorous

TPM Transition Probability Matrix

TSS Total Suspended Solids

XXIX

WGCM - WCRP Working Group on Coupled Modelling, World Climate Research Programme

WSUD Water Sensitive Urban Design

1

Chapter 1 Introduction

1.1 Background

Stormwater quality is a critical concern and leads to detrimental effects on human and environmental health. Various pollutants generated by urban anthropogenic activities are deposited on urban surfaces and get washed off during storm events. These pollutants can be transported to receiving waters leading to quality degradation. The impacts on receiving water quality degradation include elevated toxicity, algal blooms and excessive sedimentation.

Thus, implementing stormwater pollution mitigation strategies to improve the quality of receiving waters has become increasingly important. Water Sensitive Urban Design (WSUD) is the stormwater management philosophy adopted in Australia. The fundamental concept of the philosophy is to manage stormwater quality and quantity to minimise the impacts of urban developments on the surrounding environment (Lloyd et al., 2002). WSUD concepts have been developed to provide technically effective, economical and less environmentally damaging solutions for stormwater management.

However, there are barriers to widespread implementation of WSUD philosophy. In this regard, the long-term viability of implemented WSUD systems is perceived with a high level of importance (Lloyd et al., 2002). The primary concern is that the treatment systems adopted in WSUD approach are typically designed considering current climate conditions. This fact is stem from the use of historical rainfall data series in stormwater quality modelling, which is an integral component of WSUD adoption. Changing climate in future scenarios including changes to rainfall patterns and characteristics of dry periods can reduce the effectiveness of such WSUD systems (Beecham and Chowdhury, 2012).

Climate change would have a significant impact on the future stormwater quality and quantity, and thus the functionality of WSUD systems. Researchers are in agreement that future climate in Australia will feature more intense rainfall events and longer dry days due to global warming (IPCC, 2014; Abbs et al., 2007; Hughes, 2003, Holper,

2

2012). On the other hand, some other studies have noted drying trends in some regions of Australia, suggesting an increased number of dry days and a decrease in total annual rainfall (Holper, 2012).

1.2 Research problem

The design and implementation of the WSUD are typically based on static climate conditions. In the design process, future changes in the rainfall characteristics and dry periods due to climate change are not taken into account appropriately. This is primarily due to the lack of future rainfall data at very fine temporal and spatial resolutions. Although many downscaling studies have been undertaken in recent years, those studies were seldom extended to a complete climate change impact assessments nor meet the resolution requirement of the impact assessment models. Thus, most climate change impact assessments were based on simple scaling of daily climate data that is consistent with global projections. For example, Burge at al. (2012) estimated the impact of climate change in the annual pollutant export in Mornington Peninsula Shrine based on a simple scaling of local climate that was roughly consistent with global projection. Similarly, the AR&R (2015) suggested interim climate change factors to adjust the current Intensity-Frequency-Duration (IFD) values for future were based on the regional temperature projections for Australia. However, these approaches are inherently subjected to many assumptions and simplifications and thus do not provide an accurate and robust assessment at the local or small catchment scale studies (Wilby et al., 2004).

On the other hand, the changes in the rainfall patterns and characteristics of dry days will have potential adverse impacts on stormwater quality. The increase in dry days may result in more pollutants accumulated (build-up) on urban catchment surfaces. The increase in rainfall intensity can wash-off a higher fraction of build-up pollutants from catchment surfaces. Such changes can result in completely different pollutant loads and concentrations from urban catchments, requiring improved stormwater treatment devices compared to their designed characteristics (Ball et al., 1998; Egodawatta et al., 2007; Sartor et al., 1974). On the other hand, changes in the stormwater quantities due to climate change would impact the hydraulic aspects of the design of WSUD due to the potential changes in the magnitude and the frequency of the rainfalls events. Thus, the WSUD treatment systems may not meet the desired objectives in the future. Therefore, it is important to assess the impacts of climate change on the stormwater quality and quantity for an effective design and implementation of WSUD for the future.

However, a robust methodology to support the adaptation of WSUD to climate change does not exist. This is primarily due to lack of predicted future climate data for frequent rainfall events at the small catchment scale. Moreover, there is no appropriate

3

methodology available for downscaling or generating such fine-scale rainfall data. The available downscaling methods and tools typically target monthly and daily climate variable for larger regions (for example, SDSM (Wilby and Dawson, 2004) and BoM-SDM (Timbal and McAvaney, 2001)). Such downscaling methods do not support the design and implementation needs of WSUD, which are informed by frequent rainfall events at small catchment scales. Therefore, developing methodologies and tools to downscale rainfall data at fine temporal and spatial resolution and assessing the impacts of climate change on stormwater quality and quantity using those high-resolution rainfall data are essential in order to make informed decisions on adapting WSUD to climate change.

1.3 Aims and objectives

The aim of this study was to develop methodologies to generate catchment scale rainfall data for future climate change scenarios and use the generated data to assess the impacts of climate change on the stormwater quality and quantity.

The primary objectives of the research were to,

1. Develop a new approach to identify frequent event-based rainfall homogeneous regions thereby facilitating the selection of representative meteorological stations for detailed analysis.

2. Develop a methodology to spatially downscale Global Circulation Model (GCM) rainfall outputs to match the observation at small catchment scales.

3. Develop methodologies to temporally downscale rainfall time-series to finer temporal resolutions so that they can be used in event-based impact assessment models.

4. Develop the Intensity-Frequency-Duration relationships (IFD) for future climate change scenarios.

5. Assess the impacts of climate change on stormwater runoff and quality for different climate change scenarios.

1.4 Justification for the research

Climate change is one of the most alarming problems that the world is currently facing. The evidence of climate change around the world is significant. According to the Intergovernmental Panel on Climate Change (IPCC), it is very likely that the hot extremes, heat waves and heavy precipitation events will occur more frequently and thus will impact the natural and man-made systems (IPCC, 2007; CCIA, 2015). Therefore, a significant global attention has been given to climate change studies in the

4

last two decades. Studies have been conducted to address the potential causes of the climate changes and their impacts on various natural and man-made systems. Among them, the impacts of climate change on the water system are perceived with a high level of importance (IPCC, 2008).

In this regard, the impacts of climate change on the WSUD are of significant interest in countries such as Australia. WSUD is the strategic approach adopted in Australia for urban planning and design to enable effective integration of water systems to eliminate the adverse impacts of urbanization. Although the use of WSUD concept in the urban design of Australia is vital, there is no national plan for adapting WSUD to climate change, which can lead to premature obsolescence of WSUD systems. This emphasises the importance of researches that aim to understand the potential impacts of climate change on the WSUD systems in order to formulate guidelines and policies to mitigate the adverse impacts of climate change.

Therefore, this research has been conducted to advance current knowledge based on the impacts of climate change on WSUD. The advancement of knowledge primarily includes developing methodologies to generate high-resolution future rainfall data and simulate the stormwater quality and quantity for future climate change scenarios. The insights developed from this study intend to play a significant role in supporting the formulation and implementation of policies and guidelines for the adaptation of WSUD to climate change.

1.5 Description of the research

The research was conducted in two phases. In Phase 1, downscaling models were developed to produce more accurate future rainfall data for different climate change scenarios. The downscaling included spatial downscaling and temporal downscaling. In Phase 2, the outputs of Phase 1 were used to assess the impacts of climate change on WSUD. The Phase 2 of the analysis consisted of an onsite frequency analysis to develop IFD relationships for future climate change scenarios and modelling of stormwater quality to assess the impact of climate change on the pollutant process, stormwater quality and quantity.

In order to perform the Phase 1 of the research, selecting appropriate meteorological stations was critical. It was required to select meteorological stations with general rainfall characteristics equivalent to the study area while having a long period of records available to facilitate the downscaling. Therefore, a set of statistical tests (cluster analysis and Hosking and Wallis heterogeneity tests) were performed to identify homogeneous regions within the study area and to select meteorological stations to

5

represent those homogeneous regions. Thereby, any analysis using the representative meteorological station data were appropriately inferred to the homogeneous regions within the study area (Objective 1).

New spatial downscaling software (as an R package) was developed to spatially downscale GCM outputs (rainfall data) based on quantile-quantile bias correction approach. The software was used to downscale rainfall data for future climate change scenarios at the selected representative meteorological stations (Objective 2).

Then, a weather generating model was developed based on first-order homogeneous Markov model to temporally downscale the rainfall data. The outputs from the spatial downscaling (Objective 2) were then temporally downscaled using the developed model and thereby developing fine-scale rainfall time-series for different climate change scenarios (Objective 3).

An at site frequency analysis was performed on the developed future rainfall data to assess the impacts of climate change on Intensity-Frequency-Duration (IFD) relationships. IFDs are considered as the primary portal to obtain design rainfall events for the design of WSUD treatment systems. The results are then appropriately inferred to be adapted in the WSUD treatment system design guidelines (Objective 4).

An event-based stormwater quality model was developed to estimate the changes in the pollutant process (primarily, pollutant build-up and pollutant wash-off), stormwater quantity and quality for different climate change scenarios. Accordingly, a fundamental understating of the future stormwater quality and quantity scenarios was developed (Objective 5).

1.6 Scope

The focus of this research was to develop detailed understating of the impact of climate change on the Water Sensitive Urban Design. With this focus, the study was confined to few dimensions. The primary scopes of the research are:

• The research was undertaken by taking southeast Queensland as a primary study area. This limits the direct use of the results of the impact assessment to other parts of the world. However, the tools developed for downscaling future rainfall data and methodologies including the models developed for the impact assessment were generic and applicable anywhere in the world.

• The uncertainties associated with the climate change projections were not addressed in detail. However, a careful and thorough investigation of the selection of most suitable GCMs and use of all recommended climate change

6

scenarios (RCP 4.5 and RCP 8.5) provide confidence to the projections of this study. These decisions were supported by in-depth literature review.

• The future stormwater quantity and quality are functions of a range of parameters including changes to the urban form, catchment properties, available management strategies and rainfall characteristics of the region. Future variabilities of all these influential factors are well established and known before the design and implementation of WSUD except for the rainfall characteristics. Therefore, this research was primarily focused on understanding the future stormwater quantity and quality due to the future changes in rainfall characteristics.

• The investigation of water quality was based only on the primary pollutant processes namely, build-up and wash-off and the research primarily focused on TSS generation from impervious urban residential catchments.

1.7 Outline of the thesis

This thesis consists of ten chapters. Chapter 1 is the introduction to the research. Chapter 2 and Chapter 3 provide a critical review of the research literature that is relevant to this research. These chapters describe the background information related to the research and identify the knowledge gaps. Chapter 4 outlines the details of the research design and method. The study area, data collection and the selection of the meteorological stations for the study are discussed in Chapter 5. The analyses of the research are presented in Chapters 6, 7, 8 and 9. Chapter 6 is focused on the spatial downscaling, whereas Chapter 7 is focused on the temporal downscaling. The objectives of the chapters are to develop methodologies (models) to downscale rainfall data at finer spatial and temporal resolution and to develop rainfall time-series for the future climate change scenarios at the selected meteorological stations. Chapter 8 discusses the development of IFDs for future climate change scenarios and Chapter 9 discusses the impacts of climate change on pollutant processes and stormwater qualities. Chapter 10 presents the conclusions of the research and provides recommendations for future research. Finally, references used throughout the thesis are listed and Appendices A to E are attached to provide the additional information referred in the main text.

7

Chapter 2 Water Sensitive Urban Design

2.1 Background

Urban catchments have a complicated water cycle involving potable water supply, wastewater disposal and stormwater drainage systems. Typically, these systems are designed and managed as independent systems (Lloyd et al., 2002). Domestic, commercial and industrial water demands are met by treated water harvested from catchments, which are located long away from the urban areas with demand. Wastewater generated from urban areas is conveyed to treatment plants, treated and discharged to the environment. Stormwater generated within urban areas is often conveyed through the drainage systems to reduce flooding. However, in recent days, increasing environmental awareness on water resources led governments and industries to consider more integrated and efficient ways of managing urban water cycle, while exploring opportunities to reuse, sustainable treatments and management (Wong, 2006; Gardiner and Hardy, 2005). This philosophical approach is known as the Water Sensitive Urban Design (WSUD) in Australia. WSUD primarily aims to minimise the hydrological and water quality impacts of urban developments, while integrating all three components of the urban water cycle in an efficient and environmentally friendly approach. Stormwater management is a subset of overall WSUD, which primarily focused on flood control, runoff management and stormwater quality improvements, and create opportunities to harvest and reuse stormwater for non-potable purposes (Lloyd et al., 2002). The primary focus of this study was on WSUD relating to stormwater management including stormwater quality, quantity and treatment options of WSUD.

Therefore, comprehensive understanding of the current scientific knowledge on stormwater pollution and treatment, with a specific focus on WSUD treatment systems was critical for the successful completion of this research. For this, a critical review of the literature focusing on pollutant types; pollutant process; structural and non-structural WSUD practices; WSUD treatment systems and their design; and stormwater quality modelling was undertaken and presented in this chapter. The literature review

8

was further extended to investigate the potential implication of climate change on WSUD and presented in detail in Chapter 3. In-depth literature review presented in this chapter helped to formulate the research objectives and to provide justification for methods adopted in this research. A detail discussion of research design and methods opted for this research is presented in Chapter 4.

This chapter presents discussions on the concept of WSUD, WSUD structural treatment systems such as swale, bioretention basin and constructed wetland, pollutant processes and stormwater quality modelling. A special focus is given to the influential rainfall characteristics and models opted in estimating event-based stormwater qualities and quantities generated from small urban catchments.

2.2 The concept of Water Sensitive Urban Design

Water Sensitive Urban Design (WSUD) is a philosophical approach of integrating the urban water cycle to minimize the environmental degradation and preserve the aesthetics of the water environment (Wong, 2006; Lloyd et al., 2002; Goonetilleke et al., 2014; VSC, 1999). Accordingly, the objectives of WSUD includes protecting and enhancing natural water systems in urban developments; integrating stormwater treatment into the landscape by incorporating multiple use corridors that maximise the recreational values of developments; protecting the quality of water draining from urban catchments; reducing runoff and peak flows from urban catchments by using local detention measures; minimising and disconnecting impervious surfaces; and adding value while reducing drainage infrastructure development costs (VSC, 1999). The objectives of the WSUD are achieved by practices that uphold the long-term success of a stormwater management scheme. The practices can be classified into two namely Best Planning Practices (BPPs) and Best Management Practices (BMPs) as illustrated in Figure 2.1. In a broader context, the BPPs and BMPs are applied in a stormwater management system to protect and enhance the receiving waters by mimicking the natural waterways and the associated processes. Both BPPs and BMPs in WSUD comprise structural and non-structural components to achieve the WSUD objectives including pollutant preventions, conveyance of stormwater, treatment, stormwater harvesting and reuse.

9

Figure 2.1: Incorporation of BPPs and BMPs in WSUD (adapted and reproduced from Whelans et al. 1994; Mangangka, 2013)

2.2.1 Non-structural measures

Non-structural WSUD measures are primarily pollution-prevention practices designed to eliminate or minimise pollutants entering stormwater runoff. They typically exist in the form of policies, guidelines, regulations, educational and enforcement programmes, primarily instrumented to change social and community behaviour (Taylor and Wong, 2002). Non-structural WSUD measures often complement the performance of structural WSUD measures, which are installed within urban stormwater drainage systems.

The effectiveness of the non-structural practices in minimising the negative impacts of urban development is difficult to quantify and document. This can be primarily due to the fact that the non-structural measures do not involve permanent solutions. They can vary based on the geographical locations, social and economic factors. However, a number of researchers in Australia, New Zealand, United States and Germany have reported that there is an increasing trend in the use of not-structural WSUD measures such as educational and enforcement programmes (Sieker and Klein, 1998; Taylor et al., 2007; Taylor and Wong, 2002). They have also suggested that the use of combined non-structural and structural stormwater management measures as solutions for stormwater management problems have extensive advantages compared to the use of structural systems alone. This implies that non-structural measures support and enhance the effectiveness of other structural measures of a stormwater management scheme.

Technology

Best Planning Practice

Design

Best Management Practice

WATER SENSITIVE

URBAN DESIGN

Efficient, sustainable and

attractive solutions

10

Several authors have attempted to classify non-structural measures (for example, Brown, 1999; ASCE & US EPA, 2002; LSRC, 2001). Though the classifications vary, Cooperative Research Centre (CRC) for catchment hydrology identified five core categories of non-structural measures featured in the Australian context (Taylor and Wong, 2002). They are:

1. Town planning controls - Controls that promote WSUDs and BMPs in the construction of new residential developments including residential housing lots, roads, and for new commercial and industrial areas.

2. Strategic planning and institutional controls - Strategic stormwater management plans and secure funding mechanisms to support the implementations.

3. Pollution prevention procedures - Onsite non-structural measures including erosion and sediment control, waste management and infrastructural maintenance including street sweeping, routine cleaning of the stormwater drainage systems.

4. Education and participation programmes - Programmes that promote awareness through media campaigns, training programmes and stormwater drain stencilling programmes.

5. Regulatory controls - Enforcement of local regulations to control erosion and sediment on development sites and the use of regulatory implements. For example, using environmental licences to manage premises likely to contaminate stormwater.

2.2.2 Structural Measures

Structural measures refer to the use of WSUD treatment systems that collect, convey and treat stormwater runoff before discharging to the receiving waters. The treatment processes in WSUD systems vary with the treatment measures. The primary treatment processes involved in WSUD treatment systems can be classified into three, namely, physical process, physicochemical process and biological process (Scholes et al., 2008; Goonetilleke et al., 2014).

Physical process primarily involves settling and filtration. Settling refers to the removal of particulate matters by means of gravity (Ellis et al., 2004). Settling of particulate matter is highly dependent on the detention time and slow flow conditions of water and thus designed accordingly (Ellis et al., 2004). Settling is one of the primary mechanisms of particulate pollutant removal in sedimentation basins, retention basins and constructed wetlands (Greenway, 2010). Filtration refers to the removal of particulate

11

matter via physical sieving as the stormwater flows through a porous filtration media, typically layers of soil substrate (Ellis et al., 2004). Filtration is one of most prominent treatment mechanism in porous media such as porous paving, filtration basins and bioretention systems (Scholes et al., 2008).

Physicochemical processes are often referred as supplementary in stormwater pollutant removal processes in WSUD treatment systems such as bioretention basins, sedimentation tanks and constructed wetlands (Ellis et al. 2004; Goonetilleke, 2014). Physicochemical processes enhance water quality by primarily treating fine particulates and dissolved pollutants, which are difficult to remove solely by physical processes. The physicochemical process facilitates pollutant to flock into large particles to facilitate settlement. The primary chemical process includes adsorption (the accumulation of pollutants at the interface between the solid surface and solution due to ion exchange) and flocculation (the separation of solids from the water column by the attachment of small particles and settling by means of gravity) (Scholes et al., 2008; Sharkey, 2006).

Similar to the physicochemical process biological processes also complement the physical process by effective removal of dissolved pollutants (Taylor et al., 2005). Biological processes occurring within WSUD treatment systems include plant and algal uptake; microbial degradation; and nitrification and denitrification processes (Scholes et al., 2008; Hong et al., 2006; Hatts et al., 2008). Plant, algal and microbial uptake facilitate the removal of pollutants such as nitrogen, phosphorous and heavy metals from the stormwater (Hatts et al., 2007). In return, this satisfies the nutrient supply for the plants and micro-organisms in the WSUD systems. However, the plant and microbial uptakes are slow processes and require dense vegetation and long retention time (Scholes et al. 2008; Greenway, 2010). The nitrification and the denitrification processes take place due to the oxidation and reduction of nitrogen in presence of plants and micro-organisms (Goonetilleke et al., 2014). The biological processes are dominant in WSUD systems such as constructed wetlands, bioretention basins and bioretention swales.

The WSUD structural treatment systems can be broadly classified as primary level treatment systems, secondary level treatment systems and tertiary level treatment systems. Primary level treatment targets removal of litter, gross pollutants and coarse sediment. Common examples of primary level treatment systems are gross pollutant traps, trash racks, sediment traps and oil traps. Secondary level treatment aims to remove sediments, particulate matters and bacteria along with the partial removal of heavy metals and hydrocarbons. Common examples of secondary level treatment systems include vegetated buffer strips, grass swales, detention basins, bioretention basins, sedimentation basin, infiltration trenches and infiltration basins. Tertiary treatment level treatment involves the removal of fine sediments, nutrients, bacteria

12

and heavy metals. Wetlands and bioretention systems are common examples of tertiary level WSUD treatment systems (VSC, 1999).

The selection of WSUD treatment systems is closely linked to the particle size range of targeted pollutants (Wong et al., 2000). Figure 2.2 illustrates the link between the particle size of the pollutants and treatment systems used. Accordingly, the treatment systems such as gross pollutant traps and sedimentation basins can operate under high hydraulic loading due to large size of targeted pollutants. As the target pollutants size reduces, required treatment processes change to include chemical and biological treatment, which essentially require considerably low hydraulic loading rate (Wong, 2006).

Figure 2.2: Typical stormwater treatment systems, target pollutant size and hydraulic loading (Adapted and reproduced from Wong et al., 2000)

A range of pollutant types is present in the stormwater runoff in dissolved and particulate form. In order to treat them effectively, suitable WSUD treatment systems are arranged in a series. The series of treatment systems is analogous to the carriages in a train and is therefore referred to as a ‘treatment train’ (Wong, 2006). The treatment train is aimed to provide better performance and eliminate problems that may limit the effectiveness of a single treatment system. The use of treatment train has often

Particle size grading

Gross solids >5000µm

Coarse-medium sized particulates 125-5000µm

Fine particulates 10-125µm

Vary fine / Colloidal particulates 0.45-10µm

Dissolved particles <0.45µm

Hydraulic loading Q/Afacility

1000000 m/yr 100000 m/yr

50000 m/yr 5000 m/yr

2500 m/yr 1000 m/yr

500 m/yr 50 m/yr

10 m/yr

Treatment Measures

Gross pollutant trap

Sedimentation basin (wet &dry) Grass

swale & Filter strips

Surface flow wetlands

Infiltration systems Sub-surface

flow wetland

13

performed better with improved stormwater treatment processes in comparison the single treatment system (Wong, 2006). Every treatment system within the treatment train complements the limitation of the other treatment measures. Jefferies et al. (2009) assessed the performance of single and multiple component stormwater treatment systems based on pollutant reduction, maintenance issues and robustness of the treatment measure. The study suggested that treatment systems in the treatment train performed significantly better than stand-alone treatment systems.

2.3 WSUD treatment systems and their treatment and hydraulic design

As discussed in Section 2.2.2, the treatment systems are the core components of stormwater management of WSUD philosophy and hence the treatment systems are often referred as WSUDs. There is a range of treatment systems used in WSUD including gross pollutant traps, grease and oil traps, swales, bioretention swale, sedimentation basin, filter strips, porous or modular pavements, infiltration trenches, infiltration systems, soak-wells, detention basins, retention basins, ponds and constructed wetland. Among them swales, bioretention basins and constructed wetlands are popularly used across Australia and therefore a particular focus is given to these treatment systems in this review.

2.3.1 Swale

Swales are vegetated parabolic or trapezoidal shallow channels designed to substitute kerbs in urban roadsides, promoting stormwater treatment (Fiener and Auerswald, 2005). Swales also serve as an aesthetic element of the urban developments. Swales are typically used in roads and parking lots (see Figure 2.3) with a contributing catchment up to 1 to 2 ha. The design of a swale consists of two components, namely, treatment design and the hydraulic design. The treatment design of swales typically targets the removal of coarse to medium particulate matters. Swales facilitate the retardation of stormwater runoff by spreading it across the landscape and thereby achieving pollutant removal by means of settling, filtration and entrapment. Therefore, the pollutant removal of a swale is primarily determined by the top surface area of the swale (BCC & MWB, 2006).

14

Figure 2.3: Vegetated swales

The dimensions of the swales depend on how much stormwater runoff that they have to pass through to achieve the water quality and hydraulic objectives. For the water quality treatment, swale focuses only on frequent rainfalls event, typically 3-month ARI (Average Recurrence Interval). However, a swale is often required to serve hydraulic functions as a part of the drainage system, which involves a safe conveyance of stormwater runoff through the WSUD units without damaging the WSUD systems or flooding the surrounding environment. Particularly, swales located in roadsides must satisfy hydraulic engineering infrastructure design requirements defined by the relevant government authorities. Typically, minor floods (2-10 year ARI) are required to be safely conveyed. The other aspect of the hydraulic design of swale is conveyance of stormwater runoff without damaging the system itself. Higher velocities can result in scouring of the swale and poor treatment while lower velocities result in clogging. The longitudinal slope of the swale is a critical factor that determines the velocities. Swales typically operate at best velocities with bed slope between 1% and 4%. Slopes steeper than 4% can potentially cause scouring whereas, slopes milder than 1% can cause water clogs. Swales are typically expected to satisfy the velocity criteria (determined by the local government authorities) for major floods (50-100 year ARIs) (BCC & MWB, 2006).

2.3.2 Bioretention basin

Bioretention basins are vegetated areas where the stormwater runoff is treated by an underlying filter media (typically sandy loams) as they percolate downward. The treated water is collected through perforated under-drains that lead to the downstream waterways. The pollutant removal in the bioretention system is both by the vegetation and the underlying filter media. The vegetation enhances the filtration process by maintaining its porosity while filter media trap, filter and adsorb sediments, suspended

15

solids (SS) and other pollutants when the stormwater passes through. Vegetation also contributes to some pollutant removal by biological uptake (Scholes et al., 2008) and reduces runoff volumes by evapotranspiration in the bioretention systems (Hunt, 2003; Hunt et al., 2006; Davis et al., 2006; Dietz and Clausen, 2005). Typically, bioretention filter media consists of three layers, namely, a drainage layer (typically sandy loam), a transition layer (typically coarse sand) and a drainage layer where the perforated pipes are fitted in (typically fine aggregates) as shown in Figure 2.4. Bioretention basins often use temporary ponding of water above the filter media to maintain the filtration process. However, flows above the design threshold are bypassed to an overflow pit.

Figure 2.4: Typical cross section of a bioretention system (Adapted and reproduced from BCC & MWB, 2006)

Bioretention systems can treat a range of pollutants including SS, nutrients and heavy metals. Among them, bioretention basins are particularly effective in removing TSS (Hatt et al., 2007). However, the outcomes stated in the research literature as to the treatment performance of the bioretention systems are inconsistent. Parker et al. (2009) reported that TSS load was reduced by 79% in the bioretention system that they monitored while a study by Hatts et al. (2007) suggested more than 90% TSS reduction in bioretention systems. Similarly, a range of nutrient removal rate is reported in different researches. For example, Davis et al. (2006) reported 70-85% removal of Total Phosphorous (TP) while Parker et al. (2009) reported 60% TP removal in bioretention systems. These variations could be potentially related to vegetation types; storm events monitored for the study; hydraulic loading, detention time, hydraulic conductivity of

Overflow pit Above-design floods spills into field

Extended Detention Design floods pond in extended detention zone increasing volume of stormwater that is captures and treated

Vegetation Functional vegetation supports nutrient removal and maintains porosity of the soil

Bypass weir Optional to divert high floods

Top soil

300 200 - 400mm

Min 400mm

100mm

150mm

Geofabric liner

Treated flood and overflows to receiving waters Drainage layer

Transition layer

Filter media

Under-drains Cleanout Stand pipes for cleaning out of under-drains

16

the filter media and size ratio; the ratio of bioretention basin area to the catchment area (Glass and Bissouma, 2005; Bratieres et al., 2008; Hatt et al., 2009; Hunt, 2003).

The dimensions of the swales depend on how much stormwater runoff that they have to pass through to achieve the water quality and hydraulic objectives. For the water quality treatment, bioretention system focuses only frequent rainfalls event. However, a bioretention is often required to serve hydraulic functions in order to ensure the desired treatment outcomes, minimize damage by stormwater runoffs and sometimes to serve a part of the drainage system. The hydraulic design of the bioretention systems includes sizing of the inlet, overflow pits and sizing the systems to maintain appropriate velocities for minor floods (typically 2-5 year AIR events) (BCC & MWB, 2006).

2.3.3 Constructed wetlands

A constructed wetland is a shallow and extensively vegetated water body that uses pollutant removal processes including sedimentation, fine filtration and biological uptake for stormwater treatment. Constructed wetlands usually consist of an inlet zone, which is typically a sedimentation basin to remove the coarse sediments by settling, a macrophyte zone which is a shallow densely vegetated water body to remove fine particles and uptake dissolved pollutants, and a high flow bypass channel in order to protect the macrophyte zone from souring and damage to vegetation (VSC, 1999; BCC & MWB, 2006; Melbourne Water, 2005; Mitsch and Gosselink, 1993; DCR, 1999), as shown in Figure 2.5.

Constructed wetlands are considered as the best for treating stormwater runoff that has high concentrations of dissolved pollutants, which are difficult to be removed by other stormwater treatment devices (Bautista and Geiger, 1993; Mitsch and Gosselink, 1993; Scholz, 2006). Wetland vegetation enhances stormwater quality by facilitating sedimentation of suspended solids (SS), uptake nutrients, supporting the growth of biofilms which assimilate dissolved nutrients, hydrocarbon and heavy metals (Dierberg et al., 2002; Ellis et al., 1994; Jenkins and Greenway, 2005; Kohler et al., 2004). Having deep and shallow zones in wetlands can promote transformation and removal of nitrogen through facilitating different chemical reactions. The shallow zones are generally aerobic and therefore facilitate mineralisation (breakdown of organic nitrogen to ammonium and organic phosphorus to phosphate) and nitrification (conversion of nitrogen compounds to an oxidized state from a reduced state) (Mitsch and Gosselink, 1993). In contrast, the deeper zones are anaerobic and therefore facilitate denitrification (conversion of nitrate to gaseous nitrogen) (Martens et al., 2007). Phosphorus removal in a wetland takes place through sedimentation, adsorption, plant uptake, complexation and precipitation. Therefore, the nutrient removal efficiency in the constructed wetland

17

Inflow

Pre-treatment

Flow spreader

Flow spreader (porous wall) Inlet zone Open water

Outlet

Outflow

High flow by-pass

Wetland

Pond Pre-treatment

Open water Temporary storage

level

Pipe

Inlet Zone

Shallow marsh vegetation

Outflow

Permeant water level

Submerged vegetation

Temporary treatment storage Marsh vegetation

Deep marsh vegetation

Outlet

is superior to other WSUD treatment systems. Nonetheless, the outcomes reported in research literature regarding the treatment performance of the constructed wetlands are conflicting. For an example, Fletcher et al. (2003), Knight et al. (2000) and Rasheed et al. (2013) have suggested a significant nutrient removal efficiency by constructed wetlands (up to more than 80%) while Brydon et al. (2006) have reported as little as 20% nutrient load reduction. Constructed wetlands are also reported very effective for removing heavy metals (Kohler et al., 2004; Walker and Hurl, 2002), organic pollutants (Kohler et al., 2004; Sherrard et al., 2004; Thurston, 1999) and pathogens (Reinelt and Horner, 1995).

Figure 2.5: Typical section of a constructed wetland (Adapted and reproduced from VSC, 1999)

The design of the constructed wetland depends on how much stormwater runoff that it has to pass through to achieve the water quality and hydraulic objectives. For the water quality treatment, constructed wetlands focus only on frequent rainfalls event, typically 1-year ARI (Average Recurrence Interval). To maintain the required operation flow, the inlet and the outlet structures have to be sized appropriately. Also, a bypass

18

structure should be design to divert the above design flow to avoid damages to the macrophyte zone. Typically the bypass structure is sized to divert minor flood (2-10 year ARI rainfall flood) when the minor drainage system is directed to inlet zone or to divert major floods (50-100 year ARI rainfall flood) when the major drainage system is directed to inlet zone of the constructed wetland (BCC & MBW, 2006).

Overall, the design of WSUD treatment system primarily depends on the treatment requirement and the hydraulic requirement of the WSUD treatment systems (BCC & MBW, 2006). The treatment requirements are based on the expected pollutant reduction capacity. Typically, the sizing of the WSUD units for the pollutant treatment is derived from the standard treatment performance curves provided by Model for Urban Stormwater Improvement Conceptualisation (MUSIC). On the other hand, the hydraulic requirements involve maintaining the appropriate flows and velocities to achieve the treatment objectives and ensuring a safe conveyance of stormwater runoff through the WSUD treatment systems without damaging the WSUD systems and the surrounding developments. Also, hydraulic principles are applied in designing the adjoining structures such as inlets and outlets of the WSUD treatment systems to maintain desired flow rates and velocities to achieve the targeted pollutant treatments.

Therefore, the accurate estimation of design discharge is critical for the design of WSUD treatment systems. In context of the design of WSUD treatment systems, design discharges are typically classified as design operation flow (1 year ARI or less - typically for the treatment purposes), minor design flow (2-10 year ARI - typically for velocity checks, inlet, and outlet structure design purposes) and major design flow (50-100 year ARI - typically for bypass design purpose) (BCC & MBW, 2006).

In general, the stormwater discharges are from small urban catchments, typically with an area between 1 and 2 ha. Therefore, many local government authorities in Australia recommend the use of Rational Method to estimate the design discharge. The equation used to estimate the design discharges in the Rational Method is given by Equation 2.1.

𝑄𝑄 = 𝐶𝐶𝐶𝐶𝐶𝐶 (2.1)

Where, Q refers to the design discharge; I refers to the design rainfall; A refers to the catchment area; and C refers to the discharge coefficient.

From Equation 2.1, it is evident that the design discharge depends on the design rainfall. Typically the design rainfalls are estimated based on the rainfall Intensity-Frequency-Duration (IFD) relationship for the area of interest. However, the design rainfall data are developed based on static climate assumptions. The impact of climate change is not considered in the estimation of the design rainfall used to determine the design discharges. This is because there is limited knowledge available on the potential future

19

rainfall patterns. However, potential changes in the frequency and magnitude of the rainfall in the future have a direct impact on the functionalities of the existing WSUD systems and the design and implementation of new systems.

In addition to the hydraulic aspect (or stormwater quantity aspects) discussed above, the stormwater quality reaching the WSUD systems is imperative to the design of WSUD treatment systems. The stormwater quality depends on the quantity of pollutants generated and transported by storm events from the urban catchment. Therefore, understanding the pollutant process, factors determining the pollutant process, methods used to estimate pollutant process and resultant stormwater quality are also critical for the study. A detailed review is conducted on these topics and presented in the following sections.

2.4 Stormwater pollutants

The characteristics and concentration of pollutants in stormwater runoff vary, depending on their source of origin and site conditions including land use, surface type, climate, and traffic characteristics (Gobel et al., 2007). In general pollutants available in urban stormwater include suspended solids, organic carbon, toxic organics, nutrients, microorganisms, rubbish, debris, heavy metals, and surfactants (VSC, 1999). Among these common pollutants, nutrients, organic carbon, heavy metals, hydrocarbons and suspended solids are considered the most critical and are discussed in detail.

2.4.1 Nutrients

Nutrients, primarily nitrogen and phosphorous, are important pollutants in urban stormwater runoff. Nitrogen and phosphorous in stormwater can be present in dissolved form, particulate forms, species forms including ammonium nitrogen, oxidised nitrogen, organic form and inorganic forms (Quinton et al., 2001; Wong et al., 2000). The primary sources of nutrients includes household and industrial wastewater discharges, sewer overflows, vegetation debris, detergents and septic tank seepage (Chiew et al., 1998). Research studies have also noted that vehicle emissions are one of the most significant sources of nitrogen oxides (Sawyer et al., 2000). Excessive nutrient in receiving waters diminish the aquatic species and causes an imbalance in the water ecosystem. Moreover, nutrient boost can cause algal blooms resulting in significant decrease in the dissolved oxygen level in water and incidence of eutrophication (Oliver & Boon, 1992).

20

2.4.2 Organic Carbon

Organic carbon generally originates from plant debris, animal faeces, traffic activities and soil erosion. In addition, the degradation of organic matter, rubbish, sewage discharges and septic tank leakage are also influential (VSC, 1999).

Miguntanna (2009) reported that the concentration of organic carbon on pavement surfaces varies with land uses, suggesting that residential land use produce the higher organic carbon loads compared to other land uses. Meanwhile, Gobel et al. (2007) distinguished and categorized organic pollutants in stormwater runoff as organic macro-pollutants (leaves, bird excrement, flowers and pollen) and organic micro-pollutants (dust particles emitted from incineration of plants). Concentrations of both types of organic carbon are high in road surfaces in urban catchments.

On the other hand, microorganisms such as aerobic bacteria, decompose these organic pollutants once they accumulate in receiving water bodies. Microorganisms consume dissolved oxygen during decomposition, resulting in reduction of dissolved oxygen in the water. This produce odours and decrease the recreational value of water bodies (Warren et al., 2003). Dissolved oxygen reduction also affects the diversity of aquatic species and results in the imbalance of ecological system of the water bodies.

2.4.3 Heavy Metals

Stormwater runoff carries heavy metal pollutants including Cu, Zn, Cr, Pb, Fe, Al, Cd and Mn (Herngren et al., 2006; Davis et al., 2001; Hoffman et al., 1985; Gobel et al., 2007). Pitt et al. (2004) found that industrial activities, emissions from automobiles, combustion of fossil fuels and lubricant leakages are the prime sources of heavy metals. In addition, the wear of tyres, brake pads and asphaltic road surfaces also contribute to the accumulation of heavy metals.

Heavy metals contribute to the water quality degradation by creating an imbalance in the water ecosystem, due to their chemical properties, mobility, bioavailability and non-decomposable form. Moreover, heavy metals can effect animals easily through the food chain. Further, bio-accumulated heavy metal can be harmful effects to human health resulting in kidney damage, cancers, neurological damage, dementia and loss of memory (Devi, 1995; Chiew et al., 1997; Bowring, 2005).

2.4.4 Hydrocarbons

Hydrocarbon is a type of organic pollutant which consist of only hydrogen and carbon. Hydrocarbons are typically originated from vehicular emissions, organic waste incineration, industrial processes, power generation, and industrial spillages. Among all identified hydrocarbons, Polycyclic Aromatic Hydrocarbons (PAHs) are the most toxic.

21

PAHs are typically incorporated into particulates through attraction to suspended solids (Moilleron et al., 2002). Particulate size plays a significant role on the adsorption rate of PAHs to suspended solids in stormwater runoff. Herngren et al. (2010) noted that more than 85% of PAHs can be adsorbed to solid particle between 0.45 and 75 µm. PAHs pose serious toxic effects on plants, animals and human (Hwang and Foster, 2006).

2.4.5 Suspended Solids

The primary sources of suspended solids are pavement surfaces, lawns, parking lots, construction sites, weathering of road surfaces, construction site, and eroded soil from pervious lands. The primary source of suspended solids is from anthropogenic activities and is in the fine particulate ranges (Nelson and Booth, 2002; Goonetilleke and Thomas, 2003). Suspended solids poses both physical and chemical impacts on the receiving water bodies

The physical impacts include decreased transparency and reduction in water depth due to excessive sedimentation. High turbidity and poor of light penetration can affect the photosynthesis process, resulting in food and oxygen scarcities among aquatic plants and species. In addition, increased turbidity could result in aesthetics issues including bad colour and odour.

The chemical effects of suspended solids are largely due to their capability to adsorb other pollutants, including nutrients, heavy metals, pathogens, organic carbon and hydrocarbons and to act as carriers of contaminants. Such adsorption provides a platform for other pollutants to be transported to receiving waters (Jartun et al., 2008). Finer particles have a higher capacity of adsorbing pollutants compared to the coarser particles, due to their electrostatic attraction to and greater surface area (Herngren et al., 2006). Herngren et al. (2006) found that 0.45-75 µm size fine particles contain the highest concentrations of heavy metals in stormwater runoff in comparison to other size ranges. Similarly, nutrients are also adsorbed to suspended solids. A study conducted by Vaze and Chiew (2002) noted that 75% of the nutrients can be adsorbed to suspended solids less than 300 µm. Meanwhile, Miguntanna et al. (2010) reported that that fine particle less than 150 µm can adsorb the highest concentration of nutrients in residential, commercial and industrial land uses.

Organic pollutants has greater attraction to suspended solids. The study conducted by Sansalone and Tittlebaum (2001) noted that the organic pollutants carried by suspended solids amounted to 29% of its total concentration of organic matters in stormwater runoff. Based on above evidences, suspended solids can be regarded as prime indicator pollutant (Goonetilleke and Thomas, 2003). Therefore, pollutant process associated to

22

stormwater runoff are typically modelled based on the generation and transport of suspended solids (Egodawatta, 2007; Liu, 2011). The generation and transportation of other pollutants in stormwater runoff are estimated as a fraction of the available total suspended solids.

2.5 Pollutant process

The pollutant processes are interlinked in different pathways from the point of emission until reaching the receiving waters. Interlinked processes and pathways can be conceptualized by a chain of the process as presented in Figure 2.6. However, the most important pollutant processes associated with the stormwater quality are build-up and wash-off.

Figure 2.6: Conceptual chain of pollutant process (adapted and reproduced from Goonetilleke et al., 2014)

Urban emissions from traffic activities, industrial, commercial and residential land uses primarily contribute to pollutant build-up on urban surfaces. A fraction of the pollutants generated is also kept in suspension in the atmosphere. Atmospheric build-up contributes to the surface build-up in the form of dry deposition while wet deposition directly contributes to wash-off during storm events. The stormwater runoff mobilises the pollutants and wash-off them to the receiving water.

2.5.1 Pollutant build-up

Pollutant build-up refers to the accumulation of pollutants on pervious or impervious surfaces through deposition and redistribution (Mahbub et al., 2010). Typically, the depositions are from various anthropogenic activities such as land uses activities and vehicular traffic. While, the redistribution is the mobilisation of pollutants through wind, vehicle turbulence, erosion and street sweepings.

The pollutant build-up is influenced by a range of factors including climate conditions, land uses, vehicular traffic, pavement texture, antecedent dry day periods (days since last rainfall) and street cleaning (Gunewardena et al., 2012; Mahbub et al., 2010).

Urban emission

Atmospheric build-up

Wet and dry deposition

Pollutant build-up

Pollutant wash-off

Pollutant transport

Receiving waters

23

Industrial

All land used combined

Residential

Commercial

1400

1200

1000

800

600

400

200

0

0 1 2 3 4 5 6 7 8 9 10 11 12

Among them, the most important factor relating the pollutant build-up is the antecedent dry days (Sartor et al., 1974). Sartor et al. (1974) described pollutant build-up with time as a decreasing rate increasing function for different land use types as presented in Figure 2.7. Further, Sartor et al. (1974) reported that the rate of increase in pollutant build-up on road surfaces is not significant after few days (typically after 2 to 3 days) from the last rainfall. The influence of antecedent dry days becomes weak after few days due to the re-suspension of the particulate pollutants (Sartor et at., 1974: Deletic and Orr, 2005; Egodawatta, 2007).

Figure 2.7: Pollutant build-up in different land uses (adapted from Sartor et al., 1974)

The best fitting functions to replicate the pollutant build-up is typically influenced by land use and traffic characteristics. Consequently, they are highly site-specific (Sartor et al., 1974). Many researchers have tested pollutant build-up for different regions using different formulas (Sartor et al., 1974; Ball et al., 1998; Egodawatta, 2007; Liu, 2011) as presented in Table 2.1.

24

Table 2.1: Models for pollutant build-up

Type Equation

Reciprocal function 𝑦𝑦 = 𝑎𝑎 + 𝑏𝑏𝑥𝑥

Exponential function 𝑦𝑦 = 𝑎𝑎𝑒𝑒𝑥𝑥

Logarithmic function 𝑦𝑦 = 𝑎𝑎 + 𝑏𝑏 ln 𝑥𝑥

Power function 𝑦𝑦 = 𝑎𝑎𝑥𝑥𝑏𝑏

Hyperbolic function 𝑦𝑦 = 𝑥𝑥(𝑎𝑎 + 𝑏𝑏𝑥𝑥)

Note: x refers to the antecedent dry days; y refers to the build-up load accumulation; and a, b and c are constants.

For typical Australian suburban roads with moderate traffic intensities, Ball et al. (1997) found that the pollutant accumulation with antecedent dry days can be well replicate by a power function or a hyperbolic function. Similarly, a study by Egodawatta (2007) has also suggested a power equation for pollutant build-up for road surfaces in Gold Coast, Queensland, Australia as given by Equation 2.2.

𝑦𝑦 = min(𝑎𝑎𝐷𝐷𝑏𝑏,𝑀𝑀) (2.2)

Where, y refers to the build-up load on the road surface (g/m2); D refers to the antecedent dry days; a and b refer to the build-up coefficients; and the M is the maximum possible build-up load.

The maximum possible pollutant load is calculated assuming that the accumulation after 21 dry days is not significant (Egodawatta, 2007; Liu, 2011). The coefficient b is found to be a constant value of 0.16 for road surfaces (Egodawatta, 2007) and coefficient a primarily varies with the land use types and population density (Liu, 2011; Herngren, 2005; Egodawatta, 2007).

2.5.2 Pollutant wash-off

Pollutant wash-off refers to the removal and transport of pollutants during storm events. The pollutants accumulated over the antecedent dry day period are dislodged from the surface and get dissolved or kept in suspension in the rainwater. These pollutants are then transported with runoff, towards the receiving water bodies.

The rate at which rainfall dislodge particulate matters from impervious surfaces primarily depends on three factors. They are rainfall characteristics, characteristics of the surface and particle size (Egodawatta, 2007; Sartor et al., 1974). Among them,

25

rainfall and runoff characteristics including rainfall intensity and runoff rate are the most influential factors governing the wash-off process (Novotny et al., 1985).

Wash-off of different pollutants may be correlated with different characteristics of rainfall. For an example, Gnecco et al. (2005) reported that the total suspended solids concentration is high with the increased rainfall intensity, attributing to the fact that higher intensity rainfall can dislodge more pollutants from surfaces. Whereas, Chiew and McMohon (1999) found that the total phosphorus in the wash-off is strongly correlated to the runoff volume. However, the wash-off process is mathematically replicated as a function of rainfall intensity. Accordingly, Sartor et al. (1974) developed an exponential replication between wash-off and the influencing parameters as given in Equation 2.3.

𝑊𝑊 = 𝑊𝑊𝑖𝑖(1 − 𝑒𝑒−𝑘𝑘𝑘𝑘𝑘𝑘) (2.3)

Where, Wi refers to the initial weight of a material of a given particle size; t refers to the rainfall duration; W refers to the weight of material of a given particle size removed after time t; I refers to the rainfall intensity; and k is a constant represents the surface characteristics (k = 8.0 × 10-4 for road surface) (Egodawatta, 2007).

However, Egodawatta (2007) argued that a particular rainfall event has the potential to only mobilise a fraction of the pollutants (FW) depending on the intensity of the rainfall event. Based on this argument, Egodawatta (2007) proposed a coefficient termed as the capacity factor (CF) incorporated into the equation as given in Equation 2.4.

𝐹𝐹𝑊𝑊 = 𝑊𝑊𝑊𝑊𝑖𝑖

= 𝐶𝐶𝐹𝐹(1 − 𝑒𝑒−𝑘𝑘𝑘𝑘𝑘𝑘) (2.4)

Based on an experimental study, Egodawatta (2007) found that CF ranges from 0-1 depending on the rainfall intensity (I) and proposed capacity factors (CF) for different rainfall intensities as presented in Table 2.2.

Table 2.2: Wash-off capacity factors for different rainfall intensities (Adapted from Egodawatta, 2007)

Intensity range, I (mm/h) CF

5-40 (0.01×I)+0.1 40-90 0.5 90-133 (0.0098×I)-0.38

26

(a)

(b)

On the other hand, Vaze and Chiew (2002) argued that the pollutants wash-off can be limited by the amount of accumulated pollutant (build-up) or potential of the rainfall event. This concept is well explained by two different perspectives, namely, source limiting and transport limiting. Figure 2.8 illustrates hypothetical representations of potential source limiting and transport limiting scenarios.

Figure 2.8: Hypothetical representations of surface pollutant load over time (Adapted and reproduced from Vaze and Chiew, 2002)

Figure 2.8 (a) is a representation of pollutant load over time refers to the source limiting, where the available pollutants are completely washed off. Thus, the build-up again starts from zero. Whereas, in transport limiting (Figure 2.8 (b)), pollutants on the surface do not completely remove during the storm events. Thus, the build-up after such a storm event should start with the corresponding value.

Overall, it can be noted from Equations 2.2, 2.3 and 2.4 that the characteristics of rainfall and dry periods have strong influences on both pollutant build-up and wash-off processes. Pollutant build-up is a process that primarily varies with the antecedent dry days, while wash-off is influenced by rainfall variables such as rainfall intensity, rainfall duration and kinetic energy of rainfall (Ball et al., 1998; Egodawatta et al., 2007; Sartor et al., 1974). However, the rainfall characteristics used to estimate the pollutant processes are based on the current climate conditions. There is no knowledge available on the impacts of climate change on pollutant processes and how they ultimately change the stormwater quality scenarios in the future.

Rain

Build-up

Wash-off

Time

Surf

ace

pollu

tant

load

27

2.6 Stormwater quality modelling

Stormwater quality modelling plays an important role in informed decision making for strategic implementation of stormwater pollution mitigation. Modelling of stormwater quality generally refers to the replication of urban stormwater system, reproducing the associated processes by means of numerical simulations. These modelling approaches are subjected to many simplifications compared to the real world processes. However, the outcomes of these models are extensively used in informed decision makings of pollution mitigations and treatment designs. Stormwater quality modelling generally consists of two core components, namely, hydrological modelling and water quality modelling. The hydrological model estimates the runoff quantities resulting from storm events while the water quality models typically estimate the amount of pollutants transported during a storm event. These models are used in combination to estimate the stormwater quality resulting from rainfall events.

2.6.1 Hydrological modelling

Hydrological modelling is primarily the use of a combination of mathematical procedure to replicate relevant hydrological process to estimates the runoff for a given rainfall (Zoppou, 2001). The core of the hydrological model is to estimate rainfall losses due to processes such as evaporation, interception and reduction of surface runoff due to storage (Mansell, 2003). An illustration of the typical hydrological processes is presented in Figure 2.9. As illustrated in Figure 2.9, only a fraction of rainfall contributes to the runoff. Therefore, modelling each of these processes is a critical task. In order to model each of the hydrological processes, different mathematical equations have been proposed by several researchers (for example, Boyd et al., 2003; Laurenson and Mein, 1995; O’Loughlin and Stacks, 2004). Accordingly, different loss models are used to represent the hydrological process such as infiltration, interceptions, depression and evaporation.

Loss model enables easy and more reliable estimation of excess rainfall based on the observed rainfall-runoff relationships. The most commonly used loss models are initial loss; initial loss & continuous loss; initial loss & proportional loss; and infiltration loss models (Boyd et al., 2003; Laurenson and Mein, 1995; O’Loughlin and Stacks, 2004; Loveridge et at., 2013; Loveridge and Rahman, 2014). However, for typical urban catchment with predominantly impervious surfaces, an initial loss alone is considered to calculate the excess rainfall. For an urban catchment, the initial loss varies between 0 and 5 mm (Boyd et al., 2003; O’Loughlin and Stacks, 2004). While, for urban catchments with impervious surface initial loss and continuing loss models are most suitable and recommended for the most part of Australia (Pilgrims, 1987).

28

Figure 2.9: Hydrological process (adapted from O’Loughlin and Stacks, 2004)

Once the excess rainfall is calculated, routing models are used to develop the hydrograph based on the catchment characteristics. The main types of routing models are unit hydrograph model (Espey et al., 1969; Pilgrim et al., 1987; Sarma et al., 1973), kinematic wave model (Liu et al., 2004; Sugiyama, 1997; Zoppou, 2001), artificial storage routing models (Carroll, 2002; Laurenson 1964) and time area routing models (O’Loughlin and Stacks, 2004). Among these, time area routing model is more commonly used in hydrological studies and commercially available in hydrological software (for example, ILSAX and DRAINS) due to its simplicity and the ability to simulate urban catchment more realistically.

2.6.2 Water quality modelling

Water quality models provide estimates of the pollutant loads originating from a catchment using the rainfall and geographical data as inputs. Two basic types of models are used to estimate pollutant loads generated from a catchment, namely, long-term continuous-based models and event-based models (Akan and Houghtalen, 2003).

29

A Long-term continuous-based models

Long-term continuous-based models are focused on estimating pollutant load generated from a catchment over a long period of time (Ahyerre et al., 1998). Generally, these types of models are used in planning level studies. The concept of the long-term continuous-based models is straightforward. The models estimate the pollutant load exported from a catchment based on catchment, land use and climate parameters. For this, different models employ different mathematical procedure. For an example, the mathematical procedure given in Equation 2.5 suggested by the US Environmental Protection Agency (EPA) is widely adapted to estimate the annual pollutant loads generated from a catchment.

𝑀𝑀𝑠𝑠 = 𝛼𝛼𝛼𝛼𝛼𝛼𝛼𝛼 (2.5)

Where, Ms refers to the weight of the pollutant generated per unit land per year; α refers to the pollutant loading factor; P refers to the annual precipitation; f refers to the population density function; and s refers to the street sweeping factor.

MUSIC is one of the popular long-term continuous-based models which estimate stormwater quality output for a longer period of time based on stochastic approach (McAuley and Knights, 2009). MUSIC estimates the pollutant generation using monitored water quality datasets for the specific region of interest. The total pollutant generated from the catchment over a long period is determined by randomly selecting water quality parameters from the monitored dataset and associating them to rainfall events for the period. Other than MUSIC, XP-AQUALM is also commonly used to estimate pollutant exports for a longer time period, especially in preliminary water quality investigation.

B Event-based models

Unlike the long-term continuous-based models, event-based models simulate the stormwater quality only for a selected number of rainfall events. Event-based models are a combination of governing equations that replicates the pollutant processes, primarily, pollutant build-up and pollutant wash-off. Pollutant build-up is generally estimated as a function of antecedent dry days in the form of decreasing rate increasing function while the wash-off is estimated as a function of rainfall intensity. Typically, these mathematical equations are used to estimate suspended solids. Suspended solids are considered as an indicator pollutant that estimates other pollutants assuming a constant ratio to other pollutants (Akan and Houghtalen, 2003).

In addition, as discussed in the previous sections, simulating the pollutant process alone is not sufficient for stormwater quality investigations. Estimating the hydraulic and

30

hydrological parameters including runoff volumes and peak discharges are important for the simulation of pollutant concentrations. Accordingly, stormwater quality models are generally integrated into hydraulic and hydrological models such as MIKE URBAN, BASINS and SWMM.

2.7 Conclusions

This section summarises the important conclusions drawn from the research literature relating to the WSUD. The conclusions are mainly focused on the current state of knowledge with respect to the WSUD, the influence of climate characteristics on the design and implementation of WSUD and pollutant processes and understanding the potential impacts of climate change on the WSUD.

WSUD is a philosophical approach to urban planning and design that aims to minimise the hydrological and water quality impacts of urban development on the surrounding environment. Stormwater management is a subset of WSUD focused on flood control, flow management, water quality improvements and opportunities to harvest stormwater typically for non-potable uses. The practices that uphold the long-term success of a stormwater management scheme can be classified into two, namely, Best Planning Practices (BPPs) and Best Management Practices (BMPs). Both BPPs and BMPs in WSUD comprise structural and non-structural components to achieve the WSUD objectives including pollutant preventions, conveyance of stormwater, treatment, stormwater harvesting and reuse. Non-structural WSUD measures are primarily pollution-prevention practices designed to eliminate or minimise pollutants entering stormwater runoff while structural WSUD measures are typically stormwater treatment measures that aim to collect and convey pollutants and promote pollutant treatment to improve stormwater before discharging them to the receiving waters.

The treatment systems are the core components of stormwater management of WSUD. Swales, bioretention basins and constructed wetland are popularly used in stormwater treatment systems across Australia. The design of theses WSUD treatment units typically depends on the treatment requirement and the hydraulic requirement of the WSUD units.

The hydraulic requirements involve a safe conveyance of stormwater runoff through the WSUD units without damaging the WSUD systems and the surrounding developments. Also, hydraulic principles are applied in designing the adjoining structures such as inlets and outlets of the WSUD units to maintain desired flow rates and velocities to achieve the targeted pollutant treatments. In this regard, the estimation of the design runoff or the design rainfall is a critical requirement to the design of WSUD treatment systems.

31

However, the approach for estimating the design rainfall is based on static climate assumptions. The impacts of climate change are not considered in the estimation of the design rainfall and design discharges. There are no robust guidelines available to incorporate the impacts of climate change in the hydraulic design of WSUD systems.

The treatment requirement is typically based on the quality (pollutant concentration) of the runoff reaching WSUD systems for treatment. Therefore the water quality estimation is a critical component of the WSUD treatment design. Usually, the water quality is estimated using mathematical models to represent the pollutant processes. The primary pollutant processes considered are pollutant build-up and pollutant wash-off. Pollutant build-up and wash-off processes are primarily influenced by the climate characteristics. Pollutant build-up is a process that primarily varies with the antecedent dry days, while wash-off is influenced by rainfall variables such as rainfall intensity. However, the rainfall characteristics used to estimate the pollutant process are based on the current climate conditions. There is no knowledge available on the impacts of climate change on pollutant processes and how they ultimately change the stormwater quality scenarios in the future.

Overall, the current approach to design and implementation of WSUD is based on the static climate assumptions. The changes in the rainfall characteristics and rainfall pattern due to climate change are not considered in the design of WSUD. This is primarily due to lack of future climate data available for frequent-rainfall events at the small catchment scales. Hence, in-depth understanding of the current state of knowledge on climate change and generating reliable future climate data are essential aspects of this study.

33

Chapter 3 Climate Change and Downscaling

3.1 Background

Climate change can be defined as the changes in the long-term average weather due to natural and anthropogenic causes. However, the anthropogenic causes are predominant for climate change. The consequences of climate are global warming, sea-level rise, extreme rainfalls, drought and extinction of biodiversity (IPCC, 2014). These consequences have direct impacts on various natural systems. Among those systems, the impacts of climate change on water resources are perceived with a higher level of importance (IPCC, 2008).

As noted in Chapter 2, climate change would have a significant impact on the future functionality of Water Sensitive Urban Design (WSUD). Researchers are in agreement that future climate in Australia will feature more intense rainfall events and longer dry periods due to global warming (IPCC, 2014; Abbs et al., 2007; Hughes, 2003, Holper, 2012). Therefore, comprehensive understanding of the best available scientific knowledge on climate change, with a specific focus on impacts of climate change on the rainfall characteristics was critical for the successful completion of this research. For this, a critical review of the literature focusing on climate change and downscaling future rainfall data was undertaken and presented in this chapter. The in-depth literature review presented in this chapter helped to select downscaling methods, Global Circulation Models (GCMs) and climate change scenarios used in this research as presented in Chapter 4.

This chapter presents a detailed review of the literature on climate change, emission scenarios, Global Circulation Models, downscaling and uncertainties associated with climate change projections. In the detailed review, specific focus was given to statistical downscaling methods, components of statistical downscaling and available statistical downscaling tools due to their relevance to the methods used in this research.

34

3.2 Climate change

Climate change refers to the changes in the statistical distribution of weather observed over a long period of time (Hughes, 2003). The evidence for the occurrence of climate change around the world is significant. According to the Fifth Assessment Report of IPCC (2014), the global average surface temperature (combined land and ocean temperature) has risen by 0.85 0C over the period from 1983 to 2012. Precipitation over land has slightly increased in many regions since the 20th century. It is predicted that heavy precipitation events are very likely to increase over most areas of tropical and high-latitude regions. These areas are likely to experience an increase in both frequency and intensity of precipitation events (Hughes, 2003; IPCC, 2014 & 2008).

Australia’s climate change is eminent since the 20th century. The mean surface air temperature has been increased by 0.9 0C over the period from 1910 to 2010 (CCIA 2015). The average rainfall in Australia has slightly increased since 1900, with a large increase in north-west Australia since 1970. A declining trend of rainfall was also being observed in the southeast, southwest and eastern Australia (BoM, 2015). The trend in the changes in precipitation is less clear and highly varying across Australia (BoM & CSIRO, 2008). Though varied changes in annual rainfall are noted across different parts of Australia, the increase in the frequency of heavy precipitation events is very likely across the country (Hughes, 2003; IPCC, 2008; Abbs et al., 2007). Abbs et al., (2007) noted that the changes in the rainfall intensities would be relatively higher for infrequent events based on the projections for 2030 to 2070.

The causes of climate change can be classified into two, namely, natural causes and anthropogenic causes. The natural causes such as volcanic eruptions (Oppenheimer, 2002), changes in the solar energy (Sharma, 2002) and the ocean circulation (Clark et al., 2002) can force changes to the climate by altering the energy retained within the earth and atmosphere. However, the influence of natural causes on climate change is slow and insignificant compared to the anthropogenic causes.

The anthropogenic causes of climate change are primarily due to emissions of greenhouse gases (IPCC, 2014). Among the anthropogenic greenhouse gas emissions, the increase in CO2 levels due to fossil fuel combustion is the most significant (IPCC, 2014; US EIA, 2002). Industries, deforestation, animal husbandry and other land use related activities also play a vital role in the increase of greenhouse gases such as CO2, CH4, and N2O (FAO-UN, 2006; Stocker, 2011). These gases alter the natural radiation balance, leading to greater energy retention within earth’s atmosphere, which causes global warming (IPCC, 2014).

35

3.3 Emission scenarios

The influence of greenhouse gas emission has become a serious concern since the mid-20th century due to its long-lasting impacts on the components of the climate system (IPCC, 2014). Thus, for the mitigation and adaptation to the impacts of climate change, a detailed understanding of potential future greenhouse gas emission scenarios is important (IPCC, 2000, 2007 & 2014; Wayne, 2013). The greenhouse gas emission is a function of very complex dynamic systems such as population characteristics, socio-economic developments and technological changes. Based on different predictions of such dynamic systems, a set of storylines were developed for the future scenarios by the Intergovernmental Panel on Climate Change (IPCC). This is to facilitate climate change studies including climate modelling, assessment of impacts, adaptation and mitigation (IPCC, 2000). These standard set of storylines enable different groups of climate change researchers to complement and compare their studies among various branches of climate science.

Special Report on Emission Scenarios (SRES) in the third assessment report of IPCC (2000), which describe four different standard storylines for the future scenarios are presented in Table 3.1. The corresponding greenhouse gas emissions (primarily the global carbon dioxide emission) for these standard storylines are the main input required for climate modelling. Thus, CO2 concentrations corresponding to the SRES scenarios have been modelled for the future (1990-2100) based on 40 sets of different assumptions (IPCC, 2000). The assumptions are related to driving forces of climate change such as demographic changes, socio-economic developments and technological changes and have resulted in 40 different scenarios of CO2 emissions. Yet, these 40 SRES scenarios have been developed from the four basic storylines (6 from A1F1, 3 from A1T, 8 from A1B, 6 from A2, 9 from B1 and 8 from B2) (IPCC, 2000).

Figure 3.1 illustrates the total global annual carbon dioxide emissions (from all sources such as energy, industry and land use types) from 1990 to 2100. For scenarios such as A1B, A1T, B1 and B2, it could be noted that there is a reducing trend of CO2 emissions from mid-21st century. The CO2 concentration pathways also display significant overlapping among the storylines. Accordingly, it can be suggested that the storylines can result in similar CO2 emissions with different assumptions on the governing factors.

36

Table 3.1: Main characteristics of the four SRES storylines (adapted and reproduced from IPCC (2000))

Storyline / Family

Description

A1 A future world of very rapid economic growth, global population that peaks in the mid-century and declines thereafter, and rapid introduction of more efficient new technologies. Subclasses represent the alternative direction for the technological change in the energy system. A1F1 - Fossil intensive A1T - Non-fossil energy sources A1B - A balance across all sources

A2 Continuously increasing world population, economic

development is primarily regional oriented, economic growth and technological changes are more fragments and slower than the other storylines.

B1 A future world of very rapid economic growth, global population that peaks in the mid-century and declines thereafter similar to A1 but with rapid changes in the economies towards information economy, reduction in material intensities and the introduction of clean and resource-efficient technologies.

B2 The world with continuously increasing population, yet with a rate lower than A2, emphasis on local solutions to economic, social and environmental sustainability, intermediate level of economic development, less rapid diverse technological change than B1 and A1.

However, the latest report of the IPCC (Fifth assessment report, IPCC (2015)) established a different approach to the climate change scenarios termed as Representative Concentration Path (RCP). Instead of relying on the socio-economic storylines from which the emission trajectories and radiative forcing (a measure of the balance between incoming and outgoing energy, taking earth’s atmosphere as an enclosed system, measured in watts per square metre) are projected, RCP directly describes four possible radiative forcing, namely, RCP 2.6, RCP 4.5, RCP 6.0 and RCP 8.0.

37

Figure 3.1: (a) CO2 emission (b) cumulative CO2 emission for SRES storyline from 1990 to 2100 (adapted from IPCC (2000) and reproduced)

3000

2500

2000

1500

1000

500

0 2000 2010 2020 2030 2040 2050 2060 2070 2080 2090 1990 2100

Medium- High 1450-1800GtC

Medium-Low 1100-1450GtC

Low <1100GtC

A1F1

A2

B2

A1B

A1T

B1

YEAR

CU

M. C

O2 E

MM

ISSI

ON

(G

tC)

High >1800GtC

2000 2020 2040 2060 2080 2100

40

30

20

10

0

GLO

BA

L C

O2 E

MM

ISIO

N (

GtC

)

YEAR

2000 2020 2040 2060 2080 2100

40

30

20

10

0

GLO

BA

L C

O2 E

MM

ISIO

N (

GtC

)

YEAR

2000 2020 2040 2060 2080 2100

40

30

20

10

0

GLO

BA

L C

O2 E

MIS

SIO

N (

GtC

)

YEAR

2000 2020 2040 2060 2080 2100

40

30

20

10

0

GLO

BA

L C

O2 E

MIS

SIO

N (

GtC

)

YEAR

(b)

(a)

38

The RCPs were selected so that they can be incorporated into climate models representing different pathways for radiative forcing. This approach makes the climate modelling less time-consuming and flexible with a reduced cost of computation (IPCC, 2014, Wayne, 2013). Figure 3.2 presents the radiative forcing and corresponding CO2 emission pathways for RCPs. Accordingly, RCP 8.5 stands for the highest pathway of radiative forcing, which reaches 8.5 W/m2 by 2100 and increases further with time. RCP 6.0 and RCP 4.5 refer to intermediate stabilisation pathways, in which the radiative forces believed to be stabilised at 4.5 and 6.0 W/m2 respectively soon after 2100. In RCP 2.6, the radiative forcing peaks at 3W/m2 before 2100 and then decline as presented in Figure 3.2(a). The Figure 3.2(b) illustrates the corresponding CO2 emission for each RCP along with the four SRES scenarios marked in dotted lines. Though the RCPs and SRES emission pathways do not show a direct match, RCPs emission traceries are consistent with the range of possible emission pathways captured in the SRES scenarios (IPCC, 2014, Wayne, 2013).

A general rule applied to any climate change impact assessment is that all the climate change scenarios have taken into account when determining future impacts. However, not all GCMs provide climate variables for all climate change scenarios. Nonetheless, as a standard practice, IPCC (2014) recommend the use of a moderate and the highest climate change scenarios for impact assessments. Accordingly, RCP 4.5 and RCP 8.5 climate change scenarios were used in this research.

39

Figure 3.2: (a) Radiative forcing (b) Corresponding CO2 emission pathways for RCPs (adapted and reproduced from van-Vuuren et al. (2011))

2000 2010 2020 2030 2040 2050 2060 2070 2080 2090 2100

0

5

10

15

20

25

30

35

-5

YEAR

EMIS

SIO

N (

GtC

)

(b)

(a)

2000 2010 2020 2030 2040 2050 2060 2070 2080 2090 2100

0

2

4

6

8

10

-2

YEAR

RA

DIA

TIV

E FO

RC

ING

(W/m

2 )

RCP 6.0

RCP 2.6 RCP 4.5

RCP 8.5

(a)

40

3.4 Global Circulation Models

Climate models replicate a range of atmospheric processes including receipt of energy from the sun and exchanges of energy with underlying oceans, sea, ice and land (Donner and William, 2008). Climate models replicate the numerical representation of climate processes that are generally based on the laws of physics such as conservation of momentum, energy and mass. Some of these models also consider bio-chemical changes into account (Stocker, 2011).

General Circulation Model (GCM) is a climate model driven by a range of mathematical equations to represent circulation of the atmosphere and ocean water. The models used for atmospheric and ocean circulation are referred to as Atmospheric General Circulation Models (AGCMs) and Oceanic General Circulation Models (OGCMs) respectively. These circulation models are the primary component of Global Climate Models along with land and sea-ice components. In this basis, the term GCM is interchangeably used to infer both Global Climate Model and Global Circulation Model.

To simulate a GCM, scientists divide the planet into three-dimensional grids. The model simulates climate processes such as wind, heat transfers, radiation and surface hydrological parameters among the grids using two types of equations. The equations are prognostic equations and diagnostic equations. Prognostic equations are to replicate variation of a parameter as a function of time (e.g. variation of the wind, temperature and moisture). Diagnostic equations evaluate climate parameters for a given time period. In addition to these equations, processes occurring at scales finer than the grid size (e.g. convection) are approximated using various approaches known as parameterizations. Ultimately, the outputs of the simulations are often compared against the observed data to evaluate the performance of GCMs.

Coupled model intercomparison project (CMIP) is a prominent GCM evaluation project, undertaken by the Working Group on Coupled Modelling (WGCM) under the World Climate Research Programme (WCRP). This project was undertaken primarily to support climate model evaluation, validation, intercomparison, documentation and data access. This project has enabled a diverse community of climate researchers to analyse, compare and evaluate a range of coupled GCMs since 1995. The most current intercomparison project is CMIP5 (phase 5), which has evaluated 48 GCMs (CMIP5, 2015). The list of GCMs evaluated under CMIP5 including their country of origin and grid resolutions are presented in Table 3.2.

41

Table 3.2: List of GCM of CMIP5 and their resolutions (Adapted from IPCC (2014) and CCIA (2015))

CMIP5 model ID Institute and the country of origin

Horizontal resolution in km

Latitude Longitude ACCESS-1.0 CSIRO-BOM, Australia 210 130 ACCESS-1.3 CSIRO-BOM, Australia 210 130 BCC-CSM1-1 BCC, CMA, China 310 310 BCC-CSM1-1-M BCC, CMA, China 120 120 BNU-ESM BNU, China 310 310 CanCM4 CCCMA, Canada 310 310 CanESM2 CCCMA, Canada 310 310 CCSM4 NCAR, USA 130 100 CESM1-BGC NSF-DOE-NCAR, USA 130 100 CESM1-CAM5 NSF-DOE-NCAR, USA 130 100 CESM1-FASTCHEM NSF-DOE-NCAR, USA 130 100 CESM1-WACCM NSF-DOE-NCAR, USA 275 210 CMCC-CESM CMCC, Italy 410 410 CMCC-CM CMCC, Italy 78 78 CMCC-CMS CMCC, Italy 210 210 CNRM-CM5 CNRM-CERFACS, France 155 155 CNRM-CM4-2 CNRM-CERFACS, France 155 155 CSIRO-Mk3-6-0 CSIRO-QCCCE, Australia 210 210 EC-EARTH EC-EARTH, Europe 120 120 FIO-ESM FIO,SOA, China 310 310 GFDL-CM2p1 NOAA,GFDL, USA 275 220 GFDL-CM3 NOAA,GFDL, USA 275 220 GFDL-ESM2G NOAA,GFDL, USA 275 220 GFDL-ESM2M NOAA,GFDL, USA 275 220 GISS-E2-H NASA/GFDL, NY, USA 275 220 GISS-E2-H-CC NASA/GFDL, NY, USA 110 110 GISS-E2-R NASA/GFDL, NY, USA 275 220 GISS-E2-R-CC NASA/GFDL, NY, USA 110 110 HadCM3 MOHC, UK 410 280 HadGEM2-AO NIMR-KMA, Korea 210 130 HadGEM2-CC MOHC, UK 210 130 HadGEM2-ES MOHC, UK 210 130 INMCM4 INM, Russia 220 165 ISPL-CM5A-LR ISPL, France 410 210 ISPL-CM5A-MR ISPL, France 275 145 ISPL-CM5B-LR ISPL, France 410 210 MICRO4h JAMSTEC, Japan 60 60 MICRO5 JAMSTEC, Japan 155 155

42

CMIP5 model ID Institute and the country of origin

Horizontal resolution in km

Latitude Longitude MICRO-ESM JAMSTEC, Japan 310 310 MICRO-ESM-CHEM JAMSTEC, Japan 310 310 MPI-ESM-LR MPI-N, Germany 210 210 MPI-ESM-MR MPI-N, Germany 210 210 MPI-ESM-P MPI-N, Germany 210 210 MRI-CGCM3 MRI, Japan 120 120 MRI-ESM1 MRI, Japan 120 120 NorESM1-M NCC, Norway 275 210 NorESM1-ME NCC, Norway 275 210

Most of the CMIP5 GCMs were found successful in reproducing the large-scale climate characteristics such as rainfall and temperature (based on the averages between 1986 and 2005) across Australia. However, GISS –E2 models, MICRO-ESM models, ISPL-CM5A-LR and NorESM1-M models showed relatively poor simulation outcomes for the average climate across Australia. Deficiencies were reported in the performance of CMIP5 models in simulating the regional scale climate, especially in simulating regional rainfall (CCIA, 2015). The basic statistics of rainfall for summer periods and winter periods from 1986 to 2005, simulated using CMIP5 models are compared against the observed data for different National Resource Management (NRM) regions of Australia as presented in Figure 3.3. As shown in Figure 3.3, along the east coast of Australia, models show a drier trend than the observed rainfall. However, for in-lands Australia the CMIP5 GCMs tends to overestimate the average daily rainfall. However, it is noticeable that the rainfall simulations of CMIP5 GCMs were closed to the observed rainfall for the East Coast North regions which includes the southeast Queensland. CCIA (2015) also assessed the performance of the individual GCMs in simulating the historical rainfall and temperature based on the M statistic (Watterson et al., 2013). The M statistic, which also termed as the skill score was used as an index to evaluate the agreement between the simulated and observed climates. Based on this analysis, CCIA (2015) ranked ACCESS 1.0 as the best performing GCM for Australia. EC-EARTH and CESM1-CAM5 also performed well for eastern Australian regions.

The annual cycle of the temperature was well reproduced by the CMIP5 ensembles (mean of the CMIP5 GCMs), with a slight overestimation in east Tasmania (CCIA, 2015). However, the capabilities of GCMs to reproduce annual rainfall cycle are more varied except for northern Australia as illustrated in Figure 3.4. Nevertheless, the ensemble mean of the CMIP5 GCMs was similar to the observations.

43

Figure 3.3: Average (1986-2005) rainfall simulations from CMIP5 models for (a) summer and (b) winter (Adapted from CCIA (2015)). The regions are from the NRM cluster (see NRM cluster - see chapter 3, CCIA (2015)).

MB

AU

S C

S EC

ECS

EA

ECN

SSW

FE

SS

SST

E

SSV

E

SS

WF

SST

W

SS

VW

RS

MN

E M

NE

NA

RN

MN

W

R

SA

SSW

FW

WT

8

7

6

5

4

3

2

1

0

PREC

IPIT

AT

ION

(m

m/d

ay)

WINTER

MB

AU

C

S

EC

EC

S

EA

EC

N

SSW

FE

SS

SST

E

SSV

E

SS

WF

SST

W

SS

VW

RS

MN

M

NE

NA

RN

MN

W

R

SA

SSW

F W

T

8

7

6

5

4

3 2

1

0

12 11

10

9

PREC

IPIT

AT

ION

(m

m/d

ay)

SUMMER

Observation

Mean of the CMIP5 simulations

Range of CMIP5 simulations

Observation

Mean of the CMIP5 simulations

Range of CMIP5 simulations

NRM REGIONS

NRM REGIONS

44

Figure 3.4: The average annual cycle of rainfall for Australia (Regions: AUS-Australia, EA- East Australia, NA- North Australia, R-Rangelands, SA – South Australia and SS- Southern Slopes) (Adapted from CCIA (2015))

AUS

J M A F M J A J S N D O MONTH

150

100

50

0

PREC

IPIT

AT

ION

(m

m/m

onth

)

J M A F M J A J S N D O

150

100

50

0

EA

MONTH

PREC

IPIT

AT

ION

(m

m/m

onth

)


150

100

50

0

350

300

250

200

NA

MONTH

PREC

IPIT

AT

ION

(m

m/m

onth

)


R

MONTH

150

100

50

0

PREC

IPIT

AT

ION

(m

m/m

onth

)


SA

MONTH

80

60

40

20

PREC

IPIT

AT

ION

(m

m/m

onth

)

J M A

F M J A J S N D O

80

60

40

20

140

120

100

SS

MONTH

PREC

IPIT

AT

ION

(m

m/m

onth

)

Observation GCM simulations Mean of the GCM simulation






45

There were three models from Australia evaluated in CMIP5 project, namely, ACCESS 1.0, ACCESS 1.3 and CSIRO-mk3-6-0. These models are internationally recognised and essentially capable of simulating Australia’s historical climate (CCIA, 2015; Daohua et al., 2012). Especially, ACCESS 1.0 and ACCESS 1.3 are reported to perform well in simulating temperature and rainfall across Australia (CCIA, 2015). In addition to the Australian models, EC-EARTH (Hazeleger et al., 2012) was found highly capable of simulating the rainfall for Australia (CCIA, 2015). These GCMs have been used in various downscaling schemes including Abbs et al. (2007), Timbal et al. (2009), Yun and Smith (2002), White et al. (2013) and CCIA (2015). The ACCESS models and EC-EARTH model consist of basic climate components including atmosphere, ocean, land and sea-ice. The ACCESS 1.0 and ACCESS 1.3 differ from each other in their configuration of the atmosphere and the land component. On the other hand, and CSIRO-mk3-6-0 consists of more sophisticated climate components including dynamic sea ice simulation, vegetation and aerosol simulation. Further, the EC-EARTH and CSIRO-mk3-6-0 produce future climate data for all four RCP scenarios, while, the ACCESS model only produces future climate data for RCP 4.5 and RCP 8.5 (Rotstayn et al., 2010, Daohua et al., 2012, CMIP5, 2015). On the other hand, ACCESS and EC-EARTH model provide rainfall outputs in 3-hour time-series while CSIRO-mk3-6-0 only provides daily time-series for future scenarios. Therefore, for this research, ACCESS 1.0 and EC-EARTH models were selected.

3.5 Downscaling

Current GCMs are developed with 200 km to 400 km spatial resolution and with 3-hour to 6-hour temporal resolution. This primarily suggests that the processes occur in shorter time steps and within a fine spatial resolution (sub-grid scales processes) are not deterministic in the GCM simulation (BoM & CSIRO, 2007). These limitations in the resolution may restrict their usefulness in many regional and catchment scale applications such as water resource assessments and stormwater quality modelling (NCAR, 2009; Wilby et al., 1998 & 2000). Due to this reason, downscaling techniques have subsequently been developed as a means of bridging the gap between climate model outputs and the inputs that regional or local scale models require (Wilby and Wigley, 1997; Sato et al., 2007).

The downscaling methods are generally classified into statistical downscaling and dynamic downscaling (NCAR, 2009). Dynamic downscaling involves embedding a higher resolution Regional Climate Model (RCM) within the coarser resolution GCM. The RCM uses the GCM simulation outputs as boundary conditions around its domain. Based on the boundary, RCM physically simulates the dynamics of the atmosphere

46

within the finer grids of its domain (Leduc and Laprise, 2009; Walsh et al., 2003). Statistical downscaling relates the outputs from GCMs to local climate parameters by means of statistical relationships (Wilby et al., 1998 & 2008). Statistical downscaling uses outputs from GCMs as predictors in developing relationships with local climate variables (predictands) or corrects the GCM outputs to match the observation by statistical means (Wilby et al., 1998 & 2008).

3.5.1 Comparison of statistical and dynamic downscaling methods

Selection of suitable methods for downscaling is critical and dependent on the targeted climate parameters and scales, availability of data and models, economic considerations and the strengths and limitations of the methods (Wilby and Dawson, 2004). In general, statistical downscaling methods are less expensive, straightforward, computationally undemanding and capable of producing climate information in high resolutions. However, statistical downscaling requires high quality observed data for a long time period to support calibration and validation. In addition, the statistical downscaling schemes are underpinned by three primary assumptions as follows, (NCAR 2009; Hewitson and Crane, 1996).

• Climate variables are adequately reproduced by GCMs at the required spatial and temporal scales.

• The relationship between the GCM outputs and observed data remains valid for the periods outside the calibration period.

• GCM variables adequately capture the trends in the future climate.

On the other hand, the dynamic downscaling schemes consist of more sophisticated models embedded into the host GCMs. Therefore, this approach is physically consistent with the study area. However, the main limitations of dynamic downscaling are that they are computationally heavy, require sophisticated computers and high simulation times depending on the resolution of the RCMs. Further, the outcomes of dynamic downscaling rely on the accuracy of the projections of the host GCMs (Ahmed et al., 2013; Schmidli et al., 2007). For example, Ahmed et al. (2013) used 4 RCMs with different host GCMs and found varied projections for future rainfall. In contrast, CCIA (2015) tested a set of 22 GCMs to host a statistical downscaling scheme and found the resulting simulations were very similar. Moreover, the outputs of the dynamic downscaling schemes are often subjected to bias and generally does not lead to considerably accurate results than given by a statistical downscaling scheme (Schmidli et al., 2007; Jaw et al., 2015). Ahmed et al. (2013) suggested that statistical downscaling (using simple bias correction method) is an effective tool to derive high-resolution future data from GCM outputs compared to dynamic downscaling.

47

In addition, statistical downscaling can produce climate data for different climate scenarios and GCMs. In contrast, RCMs can produce climate data for only climate change scenarios used in the host GCMs. However, use of multiple GCMs and climate change scenarios is an important concern in various impact assessments to build confidence in the projections. The summary of the comparisons between statistical and dynamic downscaling methods discussed above is presented in Table 3.3.

Table 3.3: Strengths and weaknesses of statistical and dynamical downscaling (Adapted from Wilby and Dawson (2004), Ahmed et al. (2013), Schmidli et al. (2007), Jaw et al. (2015) and Mearns et al. (1999))

Statistical downscaling Dynamic downscaling

Stre

ngth

s

- Produce high-resolution climate data.

- Cheap, computationally undemanding and readily transferable.

- Produces readily transferable data that supports the impact assessment models.

- Less affected by the deficiencies in the projections the host GCM

- Responds in a physically consistent way to external forcing (e.g. topography)

- More consistent with the GCM data

Wea

knes

s

- Require high-quality data for model calibration

- Relationships may non-stationary

(i.e. the relationship obtained for present climate condition may not be valid for future climate condition)

- Depends on the accuracy of projections of the host GCM

- Require significant computer resource

- Initial boundary condition effects the results

- Not leading a considerably accurate result or resulting in bias

- Only produce future climate outputs for the scenarios inbuilt in the host GCM

Therefore, it could be concluded that statistical downscaling approaches are simple, more generic and are an effective means of producing climate data in fine resolutions when compared to the dynamic downscaling approach. Therefore, statistical downscaling approaches are explored in detail for this research and presented in detail in the following sections

48

3.5.2 Statistical downscaling

Statistical downscaling methods use GCMs outputs such as mean sea level pressure, surface minimum and maximum temperature, specific humidity and zonal and meridional wind components as predictor variables to develop relationships with local observed climate variables (predictands) or correct the GCM outputs to match the observation by statistical means (Wilby et al., 1998 & 2008).

The classification of statistical downscaling methods is fairly complex due to their interconnections. Statistical downscaling methods are primarily classified based on the principles used to develop the relationships between GCM outputs and observed data (Wilby and Dawson, 2004; Wilby et al., 2004). Accordingly, statistical downscaling approaches discussed in the literature could be roughly classified into five namely, regression methods (Kim et al, 1984; Goyal and Ojha, 2012), weather classification methods (Timbal and McAvaney, 2001), weather generating method (Wilks and Wilby, 1999; Bordoy and Burlando, 2014), change factor methods (Diaz-Nieto and Wilby, 2005) and bias correction methods (Ines and Hansen, 2006; Sharma et al., 2007; Elshamy et al., 2009; Mishra and Herath 2015). Regression, weather classification and weather generation methods involve establishing statistical relationships between the coarse resolution climate variables from GCMs (predictors) and fine resolution local climate variables (predictands) (Wilby et al., 1998 & 2008). The predictands are typically the observed climate variable such as temperature, rainfall and evaporation, while the predictors are the climate data from GCMs such as mean sea level pressure, surface minimum and maximum temperature, specific humidity and zonal and meridional wind components. Statistical models are developed based on a historical period and used to simulate the future local climate variables based on the future predictor variable provided by the GCMs. In contrast, the change factor method involves determining a factor to adjust the observed data, termed as the change factor. Change factors are typically established based on the ratio between historical and future climates simulated by the GCMs (Hay et al., 2000; Diaz-Nieto and Wilby, 2005). Bias correction methods involve establishing a relationship to correct the GCM outputs to match observed data based on the probabilistic approaches (Ines and Hansen, 2006; Sharma et al., 2007; Elshamy et al., 2009).

Regression method is one of the simple and the early statistical downscaling methods (Wilby and Dawson, 2004, NCAR, 2009). Developing regression functions to relate the predictor and the predictands involve different mathematical procedures. Commonly applied regression methods includes linear regression analysis (Goyal and Ojha, 2011; Hessami et al., 2008), non-linear regression analysis (Vrac et al., 2007), artificial neural networks (Schoof and Pryor, 2001, Snell et al., 2000), canonical correlation analysis

49

(Barnett and Preisendorfer, 1987; Storch et al., 1993) and the principal component analysis (Kim et al., 1984; Wigley et al., 1990). Regression method is straightforward and widely used in various downscaling studies due to its simplicity and the availability of tools and software for analysis. For an example, SDSM (Wilby and Dawson, 2004) is a highly reputed regression based downscaling tool used in downscaling studies around the world. However, regression-based statistical downscaling methods fail to capture the full variance in the precipitations (Burger, 1996; Von Storch, 1999). This leads to poor simulations for extreme rainfall events. Moreover, regression methods typically perform well in downscaling normally distributed climate data such as monthly temperature or long-term average temperature (Wilby et al., 2004; Trzaska and Schnar, 2014). Therefore, regression models are not popularly used in downscaling daily or sub-daily rainfall data. Regression method is only used in spatial downscaling schemes.

Weather classification (or weather typing) method is based on the grouping of the days into different weather states, typically by means of cluster analysis (Huth, 2000; Corte-Real et al., 1999; Hewitson and Crane, 2002). Then, the predictands are replicated for future climate based on the weather state of the particular day (Timbal and McAvaney, 2001). This method is particularly well suited for downscaling non-normally distributed climate data such as daily rainfall. Moreover, weather classification methods are capable of downscaling different predictands at different stations in a single experiment. For example, Timbal and McAvaney (2001) used weather classification method to downscale surface air temperature at 22 stations at Southwest Corner of Western Australia and 29 stations at Murray Darling Basin of south-eastern Australia. Consequently, Timbal and McAveney (2001) developed a downscaling model (SDM BoM) to downscale local climate parameters for Australia (Timbal et al., 2009). This model is significant as it has been already optimised (e.g. predictor sets, regions) for downscaling across Australia. However, weather classification models require observed data for long period (preferably 30 years) to cover a large array of possible weather states (Trzaska and Schnar, 2014). Consequently, weather classification models only produce future data based on the historical weather states and therefore incapable of predicting new values that are outside the range of the historical data.

Weather Generator technique is based on generating a random sequence of variables following the probability distribution of the climate variable (Wilks and Wilby, 1999). Thus, stochastic weather generating models are to resemble the summary of the local climate parameter such as mean and the variance but the observed sequence (Richardson, 1981). The concept of weather generator models adapted for downscaling is based on the changes in large-scale GCM outputs for present and future scenarios (Gregory et al., 1993; Wilby and Wigley, 1997). Most weather generating models use Markov processes where, each successive day the precipitation occurrence is decided by

50

the outcomes on previous days (E.g. WGEN – Richardson, 1981). Markov models are used to describe systems that follow a chain of linked events, where the occurrence of an event depends only on the previous states of the system (Wilks, 1999). Markov models can be categorized as Homogeneous Markov Model (HMM) and Non-Homogeneous Markov Model (NHMM). HMM assumes that the state of weather at a time is a dependent only on the state of the weather at the previous time, whereas NHMM assumption based on the fact that the state of weather at a time is a dependent not only on the state of the weather at the previous time but also on the current values of some atmospheric variables (predictors). These variables are provided by different climate models simulations. The NHMM, therefore, perform well and more popularly used to temporally downscale rainfall (Hughes et al., 1999; Charles et al., 1999). However, use of NHMM for rainfall downscaling is limited only up to monthly or daily temporal resolutions due to the absence of GCM predictor variable at sub-daily temporal resolutions. On the other hand, HMM does not require any predictor variable for temporal downscaling. Moreover, HMM is a straightforward and simple approach to weather generation. The most prominent feature of this downscaling technique is that they are capable of generating sub-daily rainfall events (temporal downscaling) which is a critical input for many impact studies (e.g. Kilsby et al., 2007; Fowler et al., 2007; Burton et al., 2010; Bordoy and Burlando, 2014). Weather generating methods are capable of producing multiple numbers of time-series of climate data, which is valued for uncertainty analysis.

Change factor method, also known as the Delta method, is the simplest form of statistical downscaling of GCM. The change factor refers to the ratio between GCM outputs of future and historical data. Then, the future data is determined by multiplying the change factor by the present (observed) climate (Hay et al., 2000). Since the ratio between GCM outputs of future and historical climate is used to determine the future local climate. This method assumes that GCMs more reliably simulate relative change rather than absolute values (Hay et al., 2000). The main disadvantage of this approach is that the same change factor is applied to all regions lying within the same GCM grid, implying that local differences in future climate are not captured.

Bias correction method can be defined as the correction of systematic misrepresentation of the statistics of the GCM simulation (Teutschbein and Seibert, 2012). The systematic misrepresentation is inherent to all models in general due to reasons such as limitations in scales, and simplification of processes and equations (Haerter et al., 2010; Ehret et al., 2012). These errors can be easily identifiable and corrected. Bias correction approach is a simple and very efficient method of downscaling rainfall data (Ines and Hansen 2006; Mishra and Herath 2015). For example, Ahmed et al. (2013) used 4 RCMs and 6 bias corrected GCMs to produce daily maximum temperature, minimum temperature

51

and daily precipitation for SRES future climate change scenarios. They suggested that all bias corrected GCMs performed remarkably well compared to the outputs from RCMs. Moreover, bias correction approaches are less data-intensive and do not require predictor variable for downscaling. Therefore, bias correction methods can be applied to downscale daily and sub-daily rainfall data. Quantile-Quantile mapping (Q-Q) is a popular bias correction technique used in various recent climate change studies (Ines and Hansen, 2006; Sharma et al., 2007; Elshamy et al., 2009; Mishra and Herath 2015). For a given quantile of the model simulation, the value corresponding to the same quantile in the observation is mapped. Once the mapping is done, the ratio between the observation and the simulation is determined for that particular quantile and assumed to be a constant for that particular quantile. Accordingly, the future simulations are corrected based on the ratio to represent the future data. (Michelangeli et al., 2009; Mishra and Herath, 2015).

3.5.3 Components of a statistical downscaling scheme

This section briefly summarises the key steps associated with statistical downscaling schemes discussed in the literature. The basic process involved in a statistical downscaling can be illustrated by the illustration in Figure 3.5.

Defining the objectives is the first component of designing a downscaling scheme. This involves defining the variables required to be downscaled; defining the spatial and temporal resolution requirements; period of the projection; climate change scenarios; characteristics of the output requirements (compatibility of feeding to impact assessment models, file formats); selection of sites (single or multi-site information); the selection host GCMs; selection of the downscaling methods and the selection of predictor variables (Beersma et al., 2000, Wilby et al., 2004).

Data collection and data handling is the second step of the process. In this regard, the availability and the accessibility of high-quality observed data and GCM outputs are important concerns. Once the data are collected, the data have to be manipulated to meet the input requirement of the statistical downscaling scheme. For an example, the metrological station data (predictands) may be incomplete and may contain inaccurate data. Hence, the data have to be corrected before using them to calibrate the downscaling scheme. Similarly, the climate variables generated from different climate models such as GCM and reanalysed data have to be re-gridded to a matching scale or interpolated at specific meteorological stations of interest. Reanalysed data is a special type of climate model, also referred as operational climate model, used to simulate the predictor variables. Unlike GCMs, reanalysis data are constrained (boundary conditioned) by a range of high quality global observational, satellites and radar data (Kalnay et al., 1996). Thus, the reanalysis data are more accurate than the ordinary

52

GCM and more suitable and typically used in calibrating regression, weather classification and weather generating downscaling schemes. The reanalysis data do not necessarily correspond to the grid spacing of the GCM which is used to generate the future predictor data set in those downscaling schemes. Therefore, it is important to rearrange grid spacing of the GCMs to match the reanalysis data. (Zorita et al., 1995; Timbal and McAveney, 2001).

Figure 3.5: Component of statistical downscaling (Adapted from Diaz-Nieto and Wilby (2005) and Wilby et al. (2004))

The most important component of the statistical downscaling scheme is the calibration, which determines the relationship between the observed data and GCM outputs. Calibration of a statistical downscaling model involves developing the relationship between the large-scale GCM/reanalysis outputs and the local climate variables or correcting the GCM outputs to match the observational data by means of mathematical relationships typically with the aid of computers (Timbal and Fernandez, 2009; Wilby and Dawson, 2004; Abatzoglou and Brown, 2011).

Data collection

Data handling

Defining downscaling objectives

Downscaling scheme

Calibration

Validation

Generate future scenarios

Impact assessments

REVISE

VALUE ADDITION

53

Calibrated downscaling scheme needs to be validated using an independent historical data set, which has not been used for calibration. The common practice is to split the available historic data set into two timeframes and to use one set for calibration and the other for validation (e.g. Sachindra et al., 2014; Wilby and Wigley, 1997). However, this approach is valid only in scenarios where long period observed data is available. In addition, splitting may yield a significant error in situations where clear trend exists in the data set. In such cases, it is preferred to select a subset of data, for example, selecting the odd years for calibration and the even years for validation (Wilby et al., 2004).

After the validation of the downscaling scheme, the climate variables supplied by the GCM for required future climate scenarios are used to generate local climate variables for the future. Preferably, downscaling can be performed using a range of GCMs and emission scenarios to represent the uncertainty associated with the different emission scenarios, model structures, parameterization schemes of the host GCMs (Giorgi, 2010; IPCC, 2015).

Finally, downscaled local climate variables for future climate require evaluation to make sure that downscaling has contributed to the required impact assessment as opposed to direct use of GCM outputs (value addition). The common practice is to assess the direct GCM outputs (with suitable interpolations) and downscaled variables against the observed historical data (Hay et al., 2000) or against the outputs from the impact assessments (e.g. hydrology - Wilby et al, 2000; flood frequency - Reynard et al., 2004; agricultural - Mearns et al., 2001).

3.5.4 Statistical downscaling tools

Several of downscaling tools have been developed to downscale different variables, at different locations, for different resolutions and using different approaches. Among them, SDSM (Wilby and Dawson, 2004) is the most popular downscaling tools used in hundreds of downscaling projects around the world (for example, Hashmi et al., 2011; Gagnon et al., 2013; Wilby and Dawson, 2013). SDSM is a free decision support tool for assessing local climate change impacts using the statistical downscaling methods. The statistical downscaling method used in this model can be well described as a hybrid of regression and stochastic weather generation types. SDSM is a generic downscaling tool which can be optimised and applied anywhere in the world provided that required data is available (Wilby and Dawson, 2004).

BoM-SDM (Timbal et al., 2009) downscaling tool is developed by the BoM, Australia using the analogue downscaling approach. The BoM-SDM was initially used for downscaling the minimum and the maximum temperature across the Murray-Darling basin of Australia, which largely covers the New South Wales, Victoria and the Southern

54

parts of Queensland. Then, the model was extended to the entire Australian regions for downscaling temperature, rainfall and evaporation. This model is completely optimised for the Australian continent and ready to be used. However, the graphical user interface (GUI) for the SDM simulates on the BoM’s research machine (Gale), which can only be accessed within the BoM and not available for public use.

Non-homogeneous Hidden Markov Model (NHMM) is a set of popular downscaling models, which can be used to downscale climate variable, typically daily rainfall (Richardson, 1981; Charles et al., 1999; Mehrotra and Sharma, 2006). For example, W-GEN (Richardson, 1981) is an NHMM based downscaling model, capable of determining separate multisite daily precipitation occurrence patterns. LARS-WS (Semenov and Barrow, 2002) is another stochastic weather generating downscaling tool used across many impact assessment studies (Fakhri et al., 2013; Dibike and Coulibaly, 2005; Khan et al., 2006; Moen and Fredmen, 2007). In addition, ASD (Hessami et al, 2008), dsclim (Page et al., 2009), ClimPACT (Alexander, 2013) and ENSEMBLES Downscaling Portal (Cofiño et al., 2011) are some of the available statistical tools for downscaling.

However, there are many limitations in the existing downscaling tools. First, many of the existing downscaling tools (that are based on the regression, weather classification and weather generation method) produce monthly or daily climate data, which is not adequate for accurate analysis (for example SDSM- Wilby et al., 2002). Secondly, some tools are research-specific and developed only to downscale targeted climate characteristic. Many of the available downscaling tools are associated with GCMs from the CMIP 3 family with older emission scenarios (for example SDM-BoM - Timbal and McAvaney, (2001)), while, more accurate and feasible emission scenarios are provided in the Fifth Assessment Report of IPCC (IPCC, 2014). Moreover, many of the existing tools are not available for public use and not easily accessible.

3.6. Uncertainties in climate change projections

Any climate change impact assessment is subjected to uncertainties. The uncertainties arise from various components used in impact assessments include uncertainties associated with emission scenarios; uncertainties associated with the host GCMs; downscaling process; and impact assessment models. Therefore, quantifying the uncertainty of a climate change impact assessment is a complex task (Wilby and Harris, 2006; Bae et al, 2011).

The host GCMs and the emission scenarios incorporated in an impact assessment are the prime source of uncertainty. GCMs have inherent variations in their assumptions, approximations, temporal and spatial resolutions, and initial conditions that lead to variation in their outputs (IPCC, 2014; Yu et al., 2002). The projected increase in the

55

global mean temperature, for example, varies between 2.40C and 6.40C by 2100 based on the 23 GCMs simulations for the highest emission scenario (IPCC, 2007). Also, it is widely reported that the capabilities of GCMs to simulate different climate variables and for a different region of the world vary (CCIA, 2015; Gleckler et al., 2008; Johnson and Sharma, 2009; Mpelasoka et al., 2012). For, Australia, a comprehensive study was undertaken by CSIRO and BoM to understand the uncertainties of 50 CMIP5 GCMs in reproducing different climate variable. They conclude that the CMIP5 GCMs showed higher variabilities in reproducing the regional rainfall patterns compared to the temperature variations (CCIA, 2015). Similarly, Perkins et al. (2007) evaluated the prediction skills of 14 GCMs by comparing the probability density functions of observed and modelled temperature and rainfall over 12 regions of Australia and found that the performance of the GCMs was significantly different.

On the other hand, future climate change scenarios are underpinned on several assumptions to denote possible scenarios of the future developments of the world. However, the scenarios of the future developments are highly variable, and no scenario would perfectly represent the future and thus the future emission scenarios. Therefore, the uncertainty associated with the emission scenario is inevitable (Giorgi, 2010).

The downscaling approaches also contribute to the overall uncertainty of the impact assessment. The uncertainties associated to downscaling can be related to many factors including the downscaling methods, locations, resolutions, climate variables, quality of the GCMs, observed data, selection of the predictors and seasonal variation in the climate (Wilby et al., 2004; Frost et al., 2011). Therefore, a deterministic approach to quantify uncertainties associated is a complicated task. A study conducted by Statistical and Regional Dynamical Downscaling of Extremes for the European region (STARDEX) concluded that the skill of statistical downscaling techniques varies non-systematically from station to station, season to season, predictand to predictand and method to method (STARDEX, 2005). Similarly, Frost et al. (2011) evaluated six downscaling methods to downscale rainfall characteristics of the Murray-Darling Basin in South-Eastern Australia and concluded that their performance varied significantly in terms of reproducing observed data.

Also, the impact assessment model itself can result in uncertainties. The modelling approach; selection of models; assumptions and approximations adapted in the models and temporal and spatial resolutions are factors that could result in variation in the outputs (Wilby et al., 2005)

Therefore, evaluation of the uncertainty associated with the climate change projection is a complex and time-consuming task. However, in order to build confidence in the climate change projections, many impact assessments have used multiple GCMs for the

56

climate change projections (Wilby et al, 2004; IPCC, 2015). In addition, to account for the uncertainties in the emission scenarios, multiple emission scenarios are taken into account. The IPCC (2015) recommends using a moderate and the highest emission scenarios for impact assessments. Similarly, the uncertainties associated with the downscaling models and impact assessment models can be evaluated by using multiple realizations and multiple models to perform the analysis.

3.7 Conclusions

This section summarises the important conclusions derived from the literature review related to climate change and downscaling. Climate change is one of the serious problems facing the world and the evidence for the occurrence of climate change around the world is significant. Climate change is caused primarily due to anthropogenic emissions of greenhouse gases. Among the greenhouse gas emissions, the increase in CO2 levels due to fossil fuel combustion is the most significant. The greenhouse gas emission is a function of very complex dynamic systems such as population characteristics, socio-economic developments and technological changes. Based on different predictions of such dynamic systems, a set of storylines were developed for the future scenarios by the Intergovernmental Panel on Climate Change (IPCC) to assist in climate change studies including climate modelling, assessment of impacts, adaptation and mitigation. Fifth Assessment Report of IPCC recommends the use of a moderate and the highest emission scenarios for any impact assessments. Accordingly, RCP 4.5 and RCP 8.5 climate change scenarios were selected for this research.

General Circulation Model (GCM) is a climate model driven by a range of mathematical equations to represent circulation of the atmosphere and ocean water, forced with the emission scenarios. The coupled model intercomparison project (CMIP) is identified as the most prominent and internationally recognised GCM evaluations programme that supports evaluations, validation, intercomparison, documentation and data access for a range of GCMs. A similar study was also conducted by CSIRO and BoM to evaluate the GCMs for Australia. The Australian origin models ACCESS 1.0, ACCESS 1.3 and CSIRO-Mk3-6-0 are found well performing and widely used in various studies across Australia. In addition to the Australian models, EC-EARTH was found highly skilful in simulating rainfall for Australia. ACCESS 1.0 model was ranked as the best model in simulating the overall Australian climate. On the other hand, EC-EARTH best simulated the average rainfall in Australia while CSIRO-Mk3-6-0 best simulated the Australian monsoons. However, CSIRO-MK-6-0, unlike ACCESS and EC-EARTH models, does not provide sub-daily rainfall outputs for future climate change scenarios. Therefore, ACCESS 1.0 and EC-EARTH models are used in this research.

57

The methods of downscaling are generally classified into statistical downscaling and dynamic downscaling. Dynamic downscaling involves embedding a higher resolution Regional Climate Model (RCM) within the coarser resolution GCM, whereas, statistical downscaling relates the outputs from GCMs to local climate parameters by means of statistical relationships. Statistical downscaling methods are less expensive, straightforward, computationally undemanding and capable of producing more accurate climate information compared to dynamic downscaling. Therefore, statistical downscaling methods are explored in this research.

Several statistical downscaling tools have been developed based on different downscaling approaches (For example, SDSM - based on regression and weather generation methods; BoM-SDM - based on weather classification method; and LARS-WS based on weather generation method). However, there are many limitations in the existing downscaling tools. Many of the existing downscaling tools (that are based on the regression, weather classification and weather generation method) typically produce daily climate data which is not adequate for accurate analysis. Moreover, existing downscaling tools are research-specific and specifically developed to downscale targeted climate characteristics. Also, many of the available downscaling tools are associated with GCMs from the CMIP 3 family with older emission scenarios (for example, SDM-BoM), while, more accurate and feasible emission scenarios are provided in the 5th assessment report of IPCC. Moreover, most of the available downscaling tools are not available for public use and not easily accessible. Due to these reasons, it was required to develop a completely new downscaling tool for this research.

Accordingly, for the spatial downscaling, the bias correction approach was selected in this research. This is primarily because the bias correction approach does not require predictor variables. The other downscaling approaches require different sets of predictor variables at the same temporal resolutions of the predictand to develop the downscaling relationships. Use of those methods is restricted due to the availability of GCM outputs as predictor variables only in daily and monthly temporal resolutions. This limits the use developing models to downscale rainfall of sub-daily temporal resolutions which is the core requirement of this research. Moreover, the bias correction approach was technically robust, straightforward and simple compared to other statistical downscaling methods. For temporal downscaling, Homogeneous Markov Model (HMM) was selected. The primary reason for the selection was that HMM does not require any predictor variable for downscaling. Moreover, HMM is a straightforward and simple approach to weather generation. HMM method is capable of producing different realizations for climate change scenarios, which is essential for uncertainty analysis. Therefore, HMM weather generating technique was used to develop a temporal downscaling model in this research.

59

Chapter 4 Research Design and Methods

4.1 Background

As discussed in Chapter 2 and Chapter 3, regional rainfall characteristics play a key role in designing Water Sensitive Urban Design (WSUD) systems. Therefore, it can be claimed that the changes in the rainfall characteristics due to climate change would potentially impact the design and the performance of the WSUD systems (Egodawatta, 2007; Liu, 2011). On the other hand, researchers are in agreement that future climate in Australia will feature more intense rainfall events and longer dry days due to global warming (IPCC, 2014; Abbs et al., 2007; Hughes, 2003; Holper, 2012). These changes would have direct impacts on the design and implementation of the WSUD in the future. For example, the increase in dry days allows more pollutants to accumulate (build-up) over urban catchment surfaces while the increase in rainfall intensity can wash-off a higher fraction of pollutants that build-up on urban surfaces (Ball et al., 1998; Egodawatta et al., 2007; Sartor et al., 1974). This can convey altered stormwater quality and quantity to treatment devices compared to their designed characteristics. On the other hand, changes in the stormwater quantities due to climate change would impact the hydraulic aspects of the design of WSUD due to the potential changes in the magnitude and the frequency of the rainfalls events (BCC & MBW, 2006). Therefore, the WSUD treatment systems may not meet the desired objectives in the future. However, there is limited research available discussing the future stormwater quality and quantity scenarios to support the adaptation of WSUD to climate change. This is primarily due to lack of future climate data for frequent rainfall events at the small catchment scales. Therefore, this research was designed to develop methodologies to generate more accurate future rainfall data in finer resolutions for different climate change scenarios and assess the impacts of climate change on WSUD.

Accordingly, the methodology developed primarily provides the structure adopted in the thesis in terms of research design, data collection, modelling and analysis. The subsequent chapters therefore depend on the data collection, methods and procedures adopted and discussed in this chapter. Further, the development of research

60

methodology was informed by the literature review undertaken and presented in Chapter 2 and Chapter 3.

Accordingly, this chapter presents comprehensive discussions on research design including criteria for the selection of study area, tools and analytical methods; data collections; and modelling and analysis. This chapter also presents separate sections on

the selection of study tools and analytical methods used in this research in detail.

4.2 Research design

The research design was divided into four main steps as given below:

1. Critical review of literature 2. Selection of study area, study tools and analytical methods 3. Data collection and quality control 4. Modelling and analysis

The flow of the research process is presented in Figure 4.1.

61

Figure 4.1: Steps of the research design

Selection of study tools Selection criteria • Able to perform a wide variety of statistical analysis

on large datasets in different file formats. • Accessibility

Selection of analytical methods Selection criteria

• Technical robustness to achieve the specific research objective

• Complexity

Selection of study area Selection criteria

• Have a broad implementation of the WSUD philosophy in its developments

• Have a network (meteorological stations) of rainfall recordings for a longer period of times.

Data collection and quality control • Meteorological station data • GCM data

Critical literature review Topics covered Influence of rainfall characteristics in the stormwater quality; pollutant processes; design of WSUD treatment systems; climate change; GCMs and downscaling approaches; and stormwater quality modelling.

Modelling and analysis Modelling and analysis conducted to address each of

the objectives separately

Understanding the current state of knowledge; identifying research gaps; and thereby defining the research objectives

62

4.2.1 Critical review of literature

A critical review of the literature was conducted to gain an in-depth understanding of the past research studies in Water Sensitive Urban Design and climate change identifying knowledge gaps. Further, the review of the literature was used to develop methods adopted in the research.

Through the literature review, the current state of knowledge was gained relating to the influence of rainfall characteristics in the stormwater quality; pollutant processes; design of WSUD treatment systems; climate change; GCMs and downscaling approaches; and stormwater quality modelling. Further, the selections of the study tools and analytical methods were supported by the conclusions derived from the literature review.

4.2.2 Selection of study area, study tools and analytical methods

This research involved a comprehensive assessment of the impacts of climate change on the WSUD. Therefore, the study area selected for this research required to have a broad implementation of WSUD philosophy in its developments while having a network (meteorological stations) of rainfall recordings for a longer period of time across the area to facilitate the downscaling. Based on these criteria, a comprehensive discussion on the selection of study area and selection of meteorological stations was conducted and presented in Chapter 5.

The study tools were primarily selected for this research based on the technical viability and accessibility. The tools were required to perform a wide variety of statistical analysis including linear and nonlinear modelling, classical statistical tests, time-series analysis, and univariate and multivariate analysis of large datasets in different file formats. Also, the tools were expected to produce clear and accurate graphical outputs. A detailed discussion on the selection of the different tools for this research is further presented in Section 4.3.

A list of statistical analytical methods was used in different stages of the research in order to achieve the objectives. These analytical methods were selected primarily based on technical viability and complexity. Further, the selections of the analytical methods were supported by the literature review. A detailed discussion of various analytical methods used in the research is presented in Section 4.4.

4.2.3 Data collection

This research aimed to develop future rainfall data at finer spatial and temporal resolutions and use them to undertake impact studies relating to stormwater quality and quantity. Accordingly, different sets of data were required at different stages of the

63

research. The primary data required for the research were observed rainfall data and GCM outputs (rainfall). Observed rainfall data in pluviographic format (every minute) for the selected meteorological stations were collected from the Bureau of Meteorology (BoM) using the online Weather Station Directory. The collected data were then subjected to a careful quality control based on the quality flag description provided by BoM.

Two GCMs namely, EC-EARTH and ACCESS-1.0 were selected for this research based on a thorough literature review relating to the performance of 48 CMIP5 GCMs to simulate the Australian climate (see Section 3.4 of Chapter 3). These two GCMs were found well performing for Australia and widely used across various climate change studies in Australia (CCIA, 2015). Both the GCMs produce 3-hour time-series of rainfall data for the historic and future timeframes. The future data includes simulations for RCP4.5 and RCP8.5 climate change scenarios which are IPCC recommended climate change scenarios for impact assessments. The GCM outputs were collected from Program for Climate Model Diagnosis and Intercomparison (PCMDI) archive and extracted using Climate Data Operators (CDO) developed by the Max Planck Institute, Germany.

4.2.4 Modelling and analysis

Figure 4.2 illustrates the construct of the analysis performed to achieve the objectives of the research. Each analysis was linked to specific objectives of the research. Accordingly, the first step of the analysis (Objective 1) was to select meteorological stations to generate future rainfall data. The selected stations were expected to approximately represent the general rainfall characteristics of the study area and thereby future rainfall data generated on these meteorological stations can be appropriately inferred to the entire study area. Therefore, a set of statistical tests were performed to identify homogeneous rainfall regions within the study area, informing the selection of representative meteorological stations.

64

Figure 4.2: Modelling and analysis

Met

hodo

logi

es t

o de

velo

p re

liabl

e fu

ture

rai

nfal

l dat

a at

fine

r te

mpo

ral

and

spat

ial r

esol

utio

ns

(INPUTS) (MODELLING AND ANALYSIS) (OUTPUTS)

5-min time-series at the r.m.s*

(observed/future)

On-site frequency analysis

Modelling (future) stormwater quality

PHASE 02

OBJECTIVE 4 / Chapter 8

OBJECTIVE 5/ Chapter 9

IDFs for future climate change scenarios

Impact on stormwater quality in future climate

change scenarios

Und

erst

andi

ng t

he im

pact

s of

cl

imat

e ch

ange

on

futu

re

stor

mw

ater

qua

lity

and

quan

tity

5-min rainfall time-series at the r.m.s

(observed)

3-hr spatially downscaled rainfall

time-series at the r.m.s (observed/future)

Observed rainfall data

Homogeneous assessment and selection of the

r.m.s*

3-hr rainfall time-series at the r.m.s

(observed)

GCM Extraction of rainfall

data at the r.m.s

3-hr rainfall time-series at the r.m.s (GCM output)


Developing a model to spatially downscale rainfall using bias

correction approach

Developing a model to temporally downscale rainfall using Markov

model

PHASE 01



3-hr spatially downscaled rainfall

time-series at the r.m.s* (observed/future)

Spatial downscaling model

Temporal downscaling model

5-min temporally downscaled rainfall

time-series at the r.m.s* (observed/future)

AIMS

* r.m.s - representative meteorological station of the study area

65

The future rainfall data were obtained from GCM outputs. The GCM outputs are inherently coarse in resolutions and unsuitable for local scale investigations (Abbs et al., 2007; Wilby and Dawson, 2004). Therefore, it was a requirement to develop methodologies to downscale the GCM outputs to finer resolutions. Accordingly, separate statistical downscaling models were developed to spatially and temporally downscale rainfall data (Objective 2 and Objective 3). As discussed in Chapter 3, the bias correction approach was selected for the development of the spatial downscaling model and Markov model was selected for the development the temporal downscaling model for this research. The selection was primarily based on ability to downscale sub-daily rainfall data and computational demand.

The hydraulic design of WSUD systems directly linked to the design rainfall (BCC & MBW. 2006). Therefore, it is important to estimate the changes required in the design to handle future climate change scenarios. Hence, an at-site frequency analysis was required to generate design rainfalls for future climate change scenarios (Objective 4). On the other hand, the treatment design of the WSUD systems directly linked to the quality of stormwater runoff reaching the inlets of the WSUD systems (Mangangka, 2013; WBD, 2010; Lucke et al., 2018). Therefore, stormwater quality modelling was required to understand the future stormwater quality scenarios (Objective 5). Finally, the results from these objectives were used to derive conclusions on the adaptation of WSUD to climate change.

4.3 Study tools

4.3.1 Programming platform

This research involved computer programming. This was to develop models and sequence of commands that will automate the execution of a specific task and solve a given problem in different settings. Computer programming included the performing statistical analysis to identify homogeneous regions; developing spatial and temporal downscaling models; performing at-site frequency analysis; and stormwater quality modelling.

In order to perform these tasks, the selection of the appropriate programming platform was critical. Accordingly, the programming platform was selected based on the ability to perform a wide variety of statistical analysis including linear and nonlinear modelling, classical statistical tests, time-series analysis, and other classical univariate and multivariate analysis. In addition, free and easy accessibility, ability to manage large data matrix, convenience in importing and exporting files in different file formats, ability to produce clear and publishable graphics, and ability to develop and publish new

66

software package and source codes were considered as important criteria in selecting the software platform for this research.

Accordingly, two programming platforms namely R and Matlab were evaluated in order to select the best platform for the research. Table 4.1 shows the comparison of R and Matlab against the selection criteria. Both R and Matlab were technically sound and used in many researches across various disciplines. Similarly, both platforms were suitable for managing different input/output file formats and producing graphics. However, R is superior in managing large dataset than Matlab. Moreover, R is free and easily accessible for common users. Based on these facts, R was selected for the research (Venables and Smith, 2008; R development Core team, 2010; MATLAB, 2010).

Table 4.1: Comparison of R and Matlab against the selection criteria

Criteria R Matlab

Technical viability √√ √√

Free and easy accessibility √√ ×

Manage large data matrix √√ √

Convenience in importing/exporting files in different file formats

√√ √√

Ability to produce clear and publishable graphics

√√ √√

Ability to develop and publish new software/package/source code

√√ √

Note: √√ -very good; √ -good; and X - poor

R (www.r-projects.org) is one of the popular open-source programming environment used across a wide range of disciplines. R provides a wide variety of statistical (linear and nonlinear modelling, classical statistical tests, time-series analysis, classification and clustering) and graphical techniques. R is available as free software under the terms of the Free Software Foundation’s General Public License (GNU) in source code form and is compatible with a wide variety of operating systems such as UNIX, Windows and MacOS (R development Core team, 2010).

R is one of the most popular statistical programming tools where the users can develop packages for different analysis using the base tools or develop entirely new advanced analytical techniques for specific purposes. These tools often contain functions that are from the R based system or exist through other previously developed packages. There are currently around 10,000 packages readily available via a global network of repositories known as Comprehensive R Archive Network (CRAN). Packages submitted to CRAN undergo extensive checks in both coding and documentation before they have

http://www.r-projects.org/

67

been published for public use. This ensures that all the codes execute exactly as expected and all functions are provided with descriptions and help files (R development Core team, 2010).

4.3.2 Climate data operators

This research involves an extensive use of GCM outputs as input data at different stages of the research. GCM outputs are a set of software libraries and self-describing, machine-independent data formats that support the creation, access, and sharing of array-oriented scientific data, referred as NetCDF (Network Common Data Form) (CMIP5, 2015). Therefore, selection of software to subsample, extract and process GCM outputs was a critical task in the research. The selected software needed to have different operators including simple statistical and arithmetic functions, data selection, subsampling and spatial interpolation. In addition, the software should be freely available. Based on these selection criteria, three software, namely, Climate Data Operators (CDO), Ncview and Panoply were assessed and the summary of the evaluation is presented in Table 4.2. As presented in Table 4.2, all three software were freely available and easily accessible. However, Panoply has limited features to execute statistical and arithmetic functions while Ncview only facilitates viewing climate data (Seidenglanz and Bremen, 2012; Wong et al., 2014). In contrast, CDO can perform a range of statistical and arithmetic functions, subsampling and spatial interpolations (Schulzweida et al., 2009). Based on this, CDO was selected for this research.

Table 4.2: Comparison of R and Matlab against the selection criteria

Criteria CDO Ncview Panoply

Ability to perform statistical and arithmetic functions, data selection, sub-sampling and spatial interpolation

√√ × √

Free availability √√ √√ √√

Note: √√ -very good; √ -good; and X - poor

The Climate Data Operators (CDO) software is a compilation of several operators for standard processing of climate model data and forecasted model data. The operators include simple statistical and arithmetic functions; data selection; subsampling; and spatial interpolation. The key features of CDO include fast processing of large datasets; dataset can be processed by several operators without storing the interim results in files; handle datasets with missing values; support different grid types; and works in UNIX/Linux, Cygwin, and MacOS-X operating systems. However, the datasets used in CDO has to be consistent. In a given dataset, all the time steps need to have the same

68

variables, and within a time step, each variable may occur only once (Schulzweida et al., 2009).

4.4 Analytical methods

4.4.1 Homogeneous Analysis

Objective 1 of the research was to select meteorological stations to generate future rainfall data. The selected stations were expected to represent the general rainfall characteristics of the study area and therefore future rainfall data generated on these meteorological stations can be appropriately used for the entire study area. Therefore, a set of statistical tests were performed to identify homogeneous rainfall regions within the study area and thereby selecting meteorological stations to generate future rainfall data. Technically, homogeneous region represents a region that has statistically similar rainfall everywhere in the region in a long time period (Hosking and Wallis, 1997). A homogeneous region may have number of different metrological stations but their observations must be similar in nature.

A range of methods exist to identify rainfall homogeneous regions and can be broadly classified into four, namely, geographical convenience, subjective partitioning, objective partitioning and multivariate analysis (Hosking and Wallis, 1997). Geographical convenience refers to the demarcation of possible homogeneous regions based on administrative boundaries or based on major geographical and physical grouping. These approaches are essentially arbitrary and often considered misleading (Hosking and Wallis, 2005). Subjective partitioning refers to the demarcation of possible homogeneous regions by inspection of the rainfall characterising and previous knowledge established about the study area (Schaefer, 1990). Objective partitioning involves identifying a group of similar meteorological stations by minimizing a with-in group heterogeneity criterion. The typical heterogeneity criteria are with-in group variation of sample coefficient of variation, with-in group variation of sample L-skewness; and likelihood-ratio statistics (Pearson, 1991; Hosking and Wallis, 2005; Lin and Chen, 2006; Ngongondo et al., 2011). Multivariate analysis is another method for demarcating homogeneous regions. A data matrix is associated with the meteorological stations of the study area and stations are then grouped based on the similarities of the data matrix. Cluster Analysis (CA) and Principal Component Analysis (PCA) are popular multivariate analysis techniques used across various researches to group meteorological stations with similar observations (Lin and Chen, 2006; Munoz-Diaz and Rodrigo et al., 2004; Unal et al., 2003; Dyer, 1975).

69

Among these methods, multivariate analysis and objective partitioning methods are technically robust methods compared to subjective partitioning and geographical convenience methods (Hosking and Wallis, 2005; Lin and Chen, 2006). Therefore, a multivariate analysis (cluster analysis) and an objective partitioning method (Hosking Wallis heterogeneity test) were used in combination in this research to demarcate the homogeneous regions. Cluster analysis was used to identify potential homogeneous regions and Hosking Wallis heterogeneity test was used to assess the degree of homogeneity of the identified regions.

A Cluster analysis

A cluster can be referred as the formed group of objects with similar attributes. The objects in a particular cluster are more relatable to each other when compared to objects from different clusters. The higher the closeness of objects within the cluster and the higher the difference between any different clusters make the cluster analysis robust and accurate. A range of models and algorithms are used to perform cluster analysis. Among them, k-means clustering and hierarchical clustering are the basic and more popularly used (Tan et al., 2005).

- K-means clustering

The basic algorithm behind the k-means clustering can be explained as shown in Figure 4.3. In Figure 4.3, k refers to the number of clusters, which is generally specified by the user. Once k is assigned, k random centroids are selected and each object is assigned to the closest centroids to form clusters. The closest centroid is assigned based on the smallest Euclidean distance to the centroids as given in Equation 4.1.

𝑑𝑑𝑑𝑑𝛼𝛼𝑑𝑑(𝑥𝑥, 𝑐𝑐) = ��(𝑥𝑥𝑖𝑖 − 𝑐𝑐𝑖𝑖)2𝑛𝑛

𝑖𝑖=1 (4.1)

Where, 𝑥𝑥𝑖𝑖 = (𝑥𝑥1, 𝑥𝑥2, 𝑥𝑥3,…𝑥𝑥𝑛𝑛) refers to the objects; 𝑐𝑐𝑖𝑖 = (𝑐𝑐1, 𝑐𝑐2, 𝑐𝑐3,… 𝑐𝑐𝑛𝑛) refers to the corresponding centroids of clusters; and 𝑑𝑑𝑑𝑑𝛼𝛼𝑑𝑑(𝑥𝑥, 𝑐𝑐) refers to Euclidean distance.

70

Figure 4.3: Basic algorithm of a K-means cluster analysis

As illustrated in Figure 4.3, the centroids of the clusters are repeatedly calculated and used as the new centroids until the centroids remain unchanged. The quality of the k-mean clustering can be expressed by the proximities of the objects to the cluster centroids. Sum of Square Error (SSE), also referred as scatter, is the most common index used to measure the quality of the clustering. SSE calculates the Euclidian distances of each object to their closest centroids and computes the sum of squared errors as given in Equation 4.2.

𝑆𝑆𝑆𝑆𝑆𝑆 = ��(𝑑𝑑𝑑𝑑𝛼𝛼𝑐𝑐(𝑐𝑐𝑖𝑖, 𝑥𝑥))2

𝑥𝑥𝜖𝜖𝑐𝑐𝑖𝑖

𝑘𝑘

𝑖𝑖=1 (4.2)

Many clustering packages are able to generate the graphical representation of k-mean clustering outcomes in the form of scree plots. Scree plots show the relationship between the numbers of clusters and the SSE (higher the numbers of clusters lower the SSE).

Select random k point for the initial centroids

Assign objects to the closest centroids and form clusters

Compute the centroids of the formed clusters

Do centroids change?

END

NO

YES

71

- Hierarchical clustering Hierarchical clustering can be broadly classified as agglomerative and divisive. The agglomerative hierarchical clustering approach is the most popularly used hierarchical clustering method (Tan et al., 2005). The basic algorithm behind agglomerative hierarchical clustering is explained in Figure 4.4. The algorithm initially treats every single object as an individual cluster and successively merges the closest pair of clusters until all objects are formed into a single cluster. This formation is typically presented in a tree-diagram known as dendrogram which displays all sub-clusters and the order in which they merge.

Figure 4.4: Basic algorithm of an agglomerative hierarchical cluster analysis

In agglomerative hierarchical clustering, the proximity among the clusters is defined by three different approaches namely; (a) simple link, (b) complete link; and (c) group average. Single link determines the Euclidean distance between the closest two objects of different clusters while complete links determine the Euclidean distance between the farthest two objects as shown in Figures 4.5(a) and 4.5(b) respectively. The group average determines the average pairwise Euclidean distance to measure the proximity of clusters as shown in Figure 4.5(c).

Consider individual objects as clusters and develop the proximity matrix

Merge the closest clusters based on the proximity

Update the new proximity matrix

Number of clusters

END

=1

> 1

72

Figure 4.5: Graphical definitions of cluster proximity

B Hosking Wallis Heterogeneity test

The primary aim of this test is to estimate the degree of heterogeneity of a given group of meteorological stations and to evaluate whether the set of meteorological stations can be treated as homogeneous. This test primarily compares the variations in the L-moments of the probability distribution of the observations at the meteorological stations. L-moments are parameters that describe the location, scale and the shape of probability distributions of a given dataset. Mean, standard deviation (SD), coefficient of variation (L-CV), coefficient of skewness (L-skewness) and coefficient of kurtosis (L-kurtosis) are the primary L-moment that describes a probability distribution. These L-moments are often used to objectively assess the heterogeneity of a group of meteorological stations. Accordingly, in an ideal situation, every station in a homogeneous region should have the same L-moments. However, in practice, the L-moments are different for every meteorological station due to differences in their

(b) Complete link

(c) Group average

(a) Single link

73

observations. Nevertheless, they can be reasonably treated as homogeneous if the differences in the L-moments of the meteorological stations are statistically insignificant (Hosking and Wallis, 2005).

The Hosking and Wallis heterogeneity test estimates the degree of heterogeneity of a group of meteorological stations and assesses whether they can be treated reasonably as a homogeneous region. The test compares the between-station dispersion of L-moments for a group of stations with what would be expected for an artificially developed homogeneous region as presented in Figure 4.6. The artificial homogeneous region is developed by repeated simulations that generate synthetic rainfall data with the same record lengths of the actual meteorological stations based on the regional average L-moments. Then, the dispersions in the L-moment of the actual and simulated meteorological stations are compared using an appropriate statistical index.

Figure 4.6: Definition sketch for heterogeneity (Adapted from Hosking and Wallis (1997))

The regional weighted average of L-CV, 𝑑𝑑(𝑅𝑅), L-skewness, 𝑑𝑑3(𝑅𝑅) and L-kurtosis 𝑑𝑑4(𝑅𝑅) in Hosking Wallis heterogeneity test are calculated using Equations 4.3, 4.4 and 4.5 respectively.

𝑑𝑑(𝑅𝑅) = ∑ 𝑛𝑛𝑗𝑗𝑑𝑑𝑗𝑗

𝑁𝑁𝑗𝑗=1

∑ 𝑛𝑛𝑗𝑗𝑁𝑁𝑗𝑗=1

(4.3)

𝑑𝑑3(𝑅𝑅) = ∑ 𝑛𝑛𝑗𝑗𝑑𝑑3𝑗𝑗𝑁𝑁

𝑗𝑗=1


(4.4)

Observed data Simulated data (Artificial homogeneous region)

L-skewness L-skewness

L-C

V

L-C

V

74

𝑑𝑑4(𝑅𝑅) = ∑ 𝑛𝑛𝑗𝑗𝑑𝑑4𝑗𝑗𝑁𝑁

𝑗𝑗=1


(4.5)

Where, N refers to the number of stations in the region and nj refers to the record length of station j.

In order to measure the heterogeneity of the meteorological stations, Hosking Wallis heterogeneity test suggests three dispersion measures based on the L-CV; L-CV and L-skewness; and L-skewness and L-kurtosis as given in Equations 4.6, 4.7 and 4.8 respectively.

𝑉𝑉1 = {∑ 𝑛𝑛𝑗𝑗(𝑑𝑑(𝑗𝑗) − 𝑑𝑑𝑅𝑅)2𝑛𝑛

𝑗𝑗=1

∑ 𝑛𝑛𝑗𝑗𝑛𝑛𝑗𝑗=1

}1/2 (4.6)

𝑉𝑉2 =∑ 𝑛𝑛𝑗𝑗 {(𝑑𝑑(𝑗𝑗) − 𝑑𝑑𝑅𝑅)2 + �𝑑𝑑3(𝑗𝑗) − 𝑑𝑑3𝑅𝑅�2}1/2 𝑛𝑛

𝑗𝑗=1


(4.7)

𝑉𝑉3 =∑ 𝑛𝑛𝑗𝑗 {�𝑑𝑑3(𝑗𝑗) − 𝑑𝑑3𝑅𝑅�2 + �𝑑𝑑4(𝑗𝑗) − 𝑑𝑑4𝑅𝑅�2}1/2 𝑛𝑛

𝑗𝑗=1


(4.8)

These dispersion measures are then for an artificial homogeneous region. A four parameter kappa distribution is fitted to the regional average L-moment ratios (1, 𝑑𝑑(𝑅𝑅), 𝑑𝑑3(𝑅𝑅) and 𝑑𝑑4(𝑅𝑅)) to simulate 𝑁𝑁𝑠𝑠𝑖𝑖𝑠𝑠 realization of an artificial homogeneous region with N stations with the same record length of the actual meteorological stations. The mean 𝜇𝜇𝑣𝑣 and the standard deviation 𝜎𝜎𝑣𝑣 of the dispersion measures of the simulated homogeneous region are calculated. Then, the dispersions of the actual and simulated homogeneous regions are compared using a statistical index H, as given in Equation 4.9.

𝐻𝐻𝑖𝑖 = (𝑉𝑉𝑖𝑖 − µ𝑣𝑣)𝜎𝜎𝑣𝑣

for 𝑑𝑑 = 1,2 and 3 (4.9)

Based on the Equation 4.9, three statistical indexes, H1, H2 and H3 are calculated based on the corresponding dispersion measures V1, V2 and V3. The region is declared acceptably homogeneous if H < 1, possibly heterogeneous if 1 ≤ H < 2 and definitely heterogeneous if H ≥ 2 (Hosking and Wallis, 1997 & 2005).

4.4.2 Spatial Downscaling

As discussed in Chapter 3, the future rainfall data can be obtained from GCM outputs. However, the GCM outputs are inherently coarse in resolutions and unsuitable for local scale investigations (Abbs et al., 2007; Wilby and Dawson, 2004). Therefore, it was a

75

requirement of the Objective 2 of the research to develop methodologies to spatially downscale the GCM outputs to finer resolutions.

Several authors have proposed methodologies to downscale GCM outputs to finer spatial resolutions (Wilby and Dawson, 2004; Timbal et al., 2009; Richardson, 1981, Hughes et al., 1999; Semenov and Barrow, 2002). Accordingly, the downscaling methods can be primarily classified as dynamic downscaling and statistical downscaling (NCAR, 2009). Dynamic downscaling involves embedding a higher resolution regional climate model (RCM) within the coarser resolution GCM. The RCM uses the GCM simulation outputs as boundary conditions around its domain to physically simulate the dynamics of the atmosphere within its finer grids (Leduc and Laprise, 2009; Walsh and Syktus, 2003). Statistical downscaling uses outputs from GCMs as predictors in developing relationships with local climate variables (predictands) or corrects the GCM outputs to match the observation by statistical means (Wilby et al., 1998 & 2008). As noted by Ahmed et al. (2013), Schmidli et al. (2007), and Jaw et al. (2015), statistical downscaling methods are less expensive, straightforward, computationally undemanding and capable of producing more accurate climate information than dynamic downscaling. A detail literature review has been undertaken on the selection of downscaling method and presented in Chapter 3. Based on this review, statistical downscaling was selected for this research.

Commonly used statistical downscaling methods can be classified as regression methods (Kim et al, 1984; Goyal and Ojha, 2012), weather classification methods (Timbal and McAvaney, 2001), weather generating method (Wilks and Wilby 1999; Bordoy and Burlando, 2014), change factor methods (Diaz-Nieto and Wilby, 2005) and bias correction methods (Ines and Hansen, 2006; Sharma et al., 2007, Elshamy et al., 2009; Mishra and Herath 2015). Regression, weather classification and weather generation methods involve establishing statistical relationships between the coarse resolution climate variables (predictors) and fine resolution local climate variables (predictands) (Wilby et al., 1998 & 2008). The predictands are typically the observed climate variable such as temperature, rainfall and evaporation while the predictors are the climate data such as mean sea level pressure, surface minimum and maximum temperature, specific humidity and zonal and meridional wind components. In contrast, the change factor method involves adjusting the observed data by multiplying the ratio between future and present climates simulated by the GCM. Bias correction methods involve establishing a relationship to correct the GCM outputs to match observed data.

Several statistical downscaling tools have been developed based on downscaling approaches as discussed above (For example, SDSM - based on regression and weather generation methods (Wilby and Dawson, 2013); BoM-SDM - based on weather

76

classification method (Timbal et al., 2009); and LARS-WS based on weather generation method (Semenov and Barrow, 2002)). However, there are many limitations in the existing downscaling tools. Many of the existing downscaling tools (that are based on the regression, weather classification and weather generation method) typically produce daily climate data which is not adequate for accurate analysis (for example SDSM- Wilby and Dawson, 2004). Some of the tools are research-specific and specifically developed to downscale targeted climate characteristics. Also, they are associated with GCMs from the CMIP 3 family with older emission scenarios (for example, SDM-BoM), while, more accurate and feasible emission scenarios are provided in the 5th assessment report of IPCC (IPCC, 2014). Moreover, most of the available downscaling tools are not available for public use and not easily accessible. Due to these reasons, it was required to develop a completely new spatial downscaling tool for this research.

The bias correction approach was used to develop the spatial downscaling tool in this research. This is because the bias correction approach does not require predictor variables. The other downscaling approaches require different sets of predictor variables at the same temporal resolutions of the predictand to develop the downscaling relationships. Use of those methods is restricted due to the availability of GCM outputs as predictor variables only in daily and monthly temporal resolutions. This limits the use developing models to downscale rainfall of sub-daily temporal resolutions which is the core requirement of this research. Moreover, the bias correction approach is technically robust, straightforward and simple compared to other statistical downscaling methods.

Bias can be defined as the systematic misrepresentation of the statistics of the simulation (Teutschbein and Seibert, 2012). The systematic misrepresentation is inherent to all models in general due to reasons such as limitations on resolutions, and simplification of processes and equations (Haerter et al., 2010; Ehret et al., 2012). These errors can be easily identifiable and corrected. This approach is generally referred as the bias correction. The quantile-quantile (Q-Q) mapping is a popular bias correction technique used in various recent climate change studies (Ines and Hansen 2006, Sharma et al., 2007, Elshamy et al., 2009; Mishra and Herath 2015). It maps the GCM outputs to the observed data based on their probability distributions. Cumulative Density Functions (CDF) are constructed separately for the GCM outputs and the observed data. Then, for a given quantile of the GCM output, the value corresponding to the same quantile in the observed data is mapped by assuming that the ratio between the observed rainfall and the GCM outputs is a constant as presented in Equation 4.10.

𝑥𝑥𝑜𝑜𝑥𝑥𝑠𝑠

= 𝑐𝑐 (4.10)

77

Where, 𝑥𝑥𝑜𝑜 and 𝑥𝑥𝑠𝑠 refer to the observed rainfall and GCM rainfall output respectively and c is a constant.

This equation can be rewritten as shown in Equation 4.11 to Equation 4.14.

𝑥𝑥𝑜𝑜,𝑝𝑝𝑝𝑝𝑝𝑝𝑠𝑠𝑝𝑝𝑛𝑛𝑘𝑘

𝑥𝑥𝑠𝑠,𝑝𝑝𝑝𝑝𝑝𝑝𝑠𝑠𝑝𝑝𝑛𝑛𝑘𝑘=

𝑥𝑥𝑜𝑜,𝑓𝑓𝑓𝑓𝑘𝑘𝑓𝑓𝑝𝑝𝑝𝑝

𝑥𝑥𝑠𝑠,𝑓𝑓𝑓𝑓𝑘𝑘𝑓𝑓𝑝𝑝𝑝𝑝 (4.11)

𝑥𝑥𝑜𝑜,𝑓𝑓𝑓𝑓𝑘𝑘𝑓𝑓𝑝𝑝𝑝𝑝 =𝑥𝑥𝑜𝑜,𝑝𝑝𝑝𝑝𝑝𝑝𝑠𝑠𝑝𝑝𝑛𝑛𝑘𝑘

𝑥𝑥𝑠𝑠,𝑝𝑝𝑝𝑝𝑝𝑝𝑠𝑠𝑝𝑝𝑛𝑛𝑘𝑘 × 𝑥𝑥𝑠𝑠,𝑓𝑓𝑓𝑓𝑘𝑘𝑓𝑓𝑝𝑝𝑝𝑝 (4.12)

𝑥𝑥𝑜𝑜,𝑓𝑓𝑓𝑓𝑘𝑘𝑓𝑓𝑝𝑝𝑝𝑝 =𝐹𝐹𝑜𝑜,𝑝𝑝𝑝𝑝𝑝𝑝𝑠𝑠𝑝𝑝𝑛𝑛𝑘𝑘

−1(𝑞𝑞)𝐹𝐹𝑠𝑠,𝑝𝑝𝑝𝑝𝑝𝑝𝑠𝑠𝑝𝑝𝑛𝑛𝑘𝑘

−1(𝑞𝑞) × 𝑥𝑥𝑠𝑠,𝑓𝑓𝑓𝑓𝑘𝑘𝑓𝑓𝑝𝑝𝑝𝑝

(4.13)

Therefore,


−1 �𝐹𝐹𝑠𝑠,𝑓𝑓𝑓𝑓𝑘𝑘𝑓𝑓𝑝𝑝𝑝𝑝�𝑥𝑥𝑠𝑠,𝑓𝑓𝑓𝑓𝑘𝑘𝑓𝑓𝑝𝑝𝑝𝑝��

𝐹𝐹𝑠𝑠,𝑝𝑝𝑝𝑝𝑝𝑝𝑠𝑠𝑝𝑝𝑛𝑛𝑘𝑘−1 �𝐹𝐹𝑠𝑠,𝑓𝑓𝑓𝑓𝑘𝑘𝑓𝑓𝑝𝑝𝑝𝑝�𝑥𝑥𝑠𝑠,𝑓𝑓𝑓𝑓𝑘𝑘𝑓𝑓𝑝𝑝𝑝𝑝��

× 𝑥𝑥𝑠𝑠,𝑓𝑓𝑓𝑓𝑘𝑘𝑓𝑓𝑝𝑝𝑝𝑝 (4.14)

Where, 𝐹𝐹() and 𝐹𝐹−1() refer the cumulative density and its inverse functions respectively and subscripts o and s refer to the observed data and GCM outputs respectively.

4.4.3 Markov model

It is a requirement to select suitable tools to achieve Objective 3 of developing temporal downscaling models. Temporal downscaling models are commonly developed based on different stochastic data generation techniques such as Artificial Neural Network (ANN), K-Nearest Neighbours (KNN) and Markov models. All the methods are data-intensive, require long sequences of data, and are sensitive to missing or erroneous data in the calibration set (Wilby et al., 2004). Among them, Markov models are more commonly used and proven in downscaling rainfall to finer temporal resolutions (Richardson, 1981, Hughes et al., 1999; Charles et al., 1999; Mehrotra and Sharma, 2006). Markov models are probabilistic models with a particular set of conditional assumptions. However, Artificial Neural Network and K-Nearest Neighbours do not link to any probabilistic conditions. In addition, KNN and ANN techniques are data-driven and therefore expected to perform poorly when working with a limited period of data in comparison to a Markov model. Therefore, Markov model was used to develop a temporal downscaling model in this research.

Markov models are used to describe systems that follow a chain of linked events, where the occurrence of an event depends only on the previous states of the system (Wilks, 1999). Markov models can be classified into two classes, namely, Homogeneous Markov Model (HMM) and Non-Homogeneous Markov Model (NHMM). HMM assumes that the state of weather at a time is a dependent only on the state of the weather at the previous time, whereas NHMM assumption based on the fact that the state of weather at a time is a dependent not only on the state of the weather at the previous time but

78

also on the current values of some atmospheric variables (predictors). These variables are provided by different climate models simulations. The NHMM, therefore, perform well and more popularly used to temporally downscale rainfall (Hughes et al., 1999). However, use of NHMM for rainfall downscaling is limited only up to monthly or daily temporal resolutions due to the absence of GCM predictor variable at sub-daily temporal resolutions. However, this research required downscaling 3-hour rainfall time-series into 5-minute time-series and therefore NHMM is not a suitable method. Furthermore, there is no established approach to downscale such finer resolution climate data. In contrast, HMM does not require any predictor variable for temporal downscaling. Moreover, HMM is a straightforward and simple approach to weather generation. Therefore, HMM weather generating technique was used to develop a temporal downscaling model in this research. Accordingly, the temporal downscaling model was developed based on the following principles.

1. The first order dependency The model assumes that the occurrence of a rainfall event at any given time-step is only depended on the rainfall event of the previous time step. This can be mathematically expressed as given in Equation 4.15.

𝛼𝛼(𝑥𝑥𝑘𝑘|𝑥𝑥𝑘𝑘−1, 𝑥𝑥𝑘𝑘−2,… , 𝑥𝑥0) = 𝛼𝛼(𝑥𝑥𝑘𝑘|𝑥𝑥𝑘𝑘−1) (4.15)

Where, x=(x0,x1,x2,…,xt,…,xn) refers to the rainfall sequence; 𝛼𝛼(𝐶𝐶|𝐵𝐵) refers to the probability of event A occurring given the previous event is B; xt refers to the rainfall at time t; and xt-1 refer to the rainfall at time t-1.

2. The transition probability remains constant with time Transition probability refers to the probability of a particular event (for example, 𝛼𝛼𝑗𝑗) occurring, given that a particular event (for example,𝛼𝛼𝑖𝑖) has occurred

in previous time step. The transition probabilities for all possible transitions are determined based on a long historical rainfall records. These transition probabilities are assumed independent of time. This can be mathematically expressed as given in Equation 4.16

𝛼𝛼𝛼𝛼�𝑥𝑥𝑘𝑘 = 𝛼𝛼𝑗𝑗�𝑥𝑥𝑘𝑘−1 = 𝛼𝛼𝑖𝑖� = 𝛼𝛼�𝑥𝑥(𝑘𝑘+𝜏𝜏) = 𝛼𝛼𝑗𝑗�𝑥𝑥(𝑘𝑘+𝜏𝜏)−1 = 𝛼𝛼𝑖𝑖� (4.16)

Where, S=(s1,s2,s3,…,si,…,sm) refers to the discrete weather states (all possible rainfall recordings of the station) and 𝑥𝑥 ∈ 𝑆𝑆.

3. Data generation A Transition Probability Matrix (TPM) can be developed by considering all possible transitions based on a historical rainfall data. Then, the rainfall data can be generated using the TPM and a sequence of random numbers. For example, if

79

the current state of rainfall is xt-1=si (initial condition) and Ru be a random number between 0 and 1 and if,

�𝛼𝛼𝑖𝑖𝑖𝑖

𝑖𝑖−1

𝑛𝑛=1 < 𝑅𝑅𝑓𝑓 ≤ �𝛼𝛼𝑖𝑖𝑖𝑖

𝑖𝑖

𝑛𝑛=1

Then the model determines, 𝑥𝑥𝑘𝑘 = 𝛼𝛼𝑖𝑖

And, for the next event, xt-1=sl is considered as the current state and such the process continues.

4.4.4 Rainfall frequency analysis

The hydraulic design of WSUD systems directly linked to the design rainfall (BCC & MBW. 2006). Therefore, it is important to estimate the changes in the design rainfalls for future climate change scenarios, forming the Objective 4 of this research. Hence, an at-site frequency analysis was conducted to generate Intensity-Frequency-Duration (IFD) curves for future climate scenarios.

IFD relationships can be developed by performing a regional frequency analysis to a large number of stations in the region (for example, IFDs for Australia - AR&R, (2015)). This approach involves a collective use of the statistics of the recorded rainfall data at different meteorological stations to estimate the IFD relationship for a large region. Regional frequency analysis recognises that meteorological stations with rainfall records for a shorter period would considerably impose uncertainties and bias into the estimation of the IFDs, and therefore, those meteorological station data are used together with the neighbouring stations. Hence, regional frequency analysis is expected to produce more robust results and therefore popularly used in the development of the national IFD relationships (AR&R, 2015; Green et al., 2012; Madsen et al., 2009; Hosking and Wallis, 1997). However, regional frequency analysis essentially requires rainfall records at several meteorological stations. But, in this research, the future data was developed only for representative meteorological stations. Further, regional frequency analysis is a complex process that involves a number of challenging phases including comprehensive data screening, regionalising and estimation of the frequency distribution (Hosking and Wallis, 2005). Such complex analyses are beyond the scope of this research. In this research, a conventional at-site frequency analysis was undertaken for the selected representative meteorological stations. The percentage changes in the future IFDs developed at these stations were then inferred to the homogeneous regions represented by those stations.

80

Figure 4.7: Probability parameters for at-site frequency analysis

The IFDs are developed based on the probability distributions of the annual series of maximum rainfall of a particular duration. Figure 4.7 illustrates the estimation of the magnitude of the rainfall for a given return period (expected time between the occurrences of such similar events or the frequency of the event) based on the probability distribution.

Accordingly, the magnitude of the rainfall of a particular return period can be given by Equation 4.17.

𝑥𝑥𝑇𝑇 = 𝜇𝜇+ 𝑘𝑘𝑇𝑇𝜎𝜎 (4.17)

Where, 𝑥𝑥𝑇𝑇 refers to the magnitude of the rainfall for a return period T and kT refers to frequency factor.

Frequency factor only depends on the return period for a given probability distribution. Therefore, frequency factors can be determined based on return period of interest and the fitted probability distribution. Two primary probability distribution families, namely, Generalised Extreme Value (GEV) distribution, Pearson type distributions are used in rainfall frequency analysis. For example, Hajani and Rahman (2018) used GEV and Pearson type distribution (LP3) for design rainfall estimation and reported that both distribution performed were suitable for the design rainfall estimation. However, probability distributions from GEV family including Extreme Value type I (EV-I), also referred as the Gumbel distribution and extreme value type III (EV-III), also referred as Weibull distributions are the most commonly used probability distributions to fit the maximum rainfall events. EV-I is most suitable for modelling extreme rainfalls and maximum flowrates whereas EV-III is suitable for modelling minimum flows (Koutsoyiannis, 2004). Therefore, in this research, extreme value type I (EV-I) probability distribution is used to fit the annual series of the maximum rainfalls. The general form of the EV-I distribution is given by Equation 4.18.

𝜇𝜇

𝑘𝑘𝑇𝑇𝜎𝜎

𝑥𝑥𝑇𝑇

𝛼𝛼(𝑥𝑥)

0

The probability of an event exceeds xT, 𝛼𝛼(𝑥𝑥 ≥ 𝑥𝑥𝑇𝑇 ) = � 𝛼𝛼(𝑥𝑥)𝑑𝑑𝑥𝑥

∞

𝑥𝑥𝑇𝑇

81

𝛼𝛼(𝑥𝑥) = 1𝛽𝛽

𝑒𝑒−(𝑥𝑥−𝛼𝛼)

𝛽𝛽 𝑒𝑒−𝑝𝑝−(𝑥𝑥−𝛼𝛼)

𝛽𝛽 (4.18)

Where, 𝛼𝛼 and 𝛽𝛽 are location and scale parameters respectively. 𝛼𝛼 and 𝛽𝛽 can be mathematically expressed by Equations 4.19 and 4.20 respectively.

𝛼𝛼 =√

6𝜎𝜎𝜋𝜋

(4.19)

𝛽𝛽 = 𝜇𝜇 − 0.5772𝛼𝛼 (4.20)

Where, 𝜇𝜇 and 𝜎𝜎 refer to the mean and standard deviation of the distribution. The relationship between the frequency factor and return period for an EV-I distribution is given by Equation 4.21.

𝑘𝑘𝑇𝑇 = −√

6𝜋𝜋

�0.5772 + ln �ln� 𝑇𝑇𝑇𝑇 − 1

�� (4.21)

Where, 𝑘𝑘𝑇𝑇 and 𝑇𝑇 refer to the frequency factor and return period respectively.

4.4.5 Stormwater quality modelling

Stormwater quality modelling can be strategically used in estimating the future stormwater quality and informed decision making of future stormwater pollution mitigation. The stormwater quality model replicates the urban stormwater by replicating the build-up, wash-off, rainfall and runoff properties by means of mathematical models. Therefore, stormwater quality modelling essentially consists of two major components, namely, pollutant process modelling and hydrologic and hydraulic modelling.

A Pollutant process modelling

Pollutant process modelling refers to the mathematical representation of the pollutant processes. The primary pollutant processes that constitute the stormwater quality are pollutant build-up and pollutants wash-off (Egodawatta, 2007). Therefore pollutant build-up and pollutant wash-off were considered in this research to estimate the pollutant generated and transported from urban catchments.

Reliable estimation of pollutant generation primarily depends on the mathematical functions and the parameters used to estimate the pollutant build-up. Several authors have proposed different mathematical relationships to model the pollutant build-up process such as reciprocal; logarithmic; exponential and power functions (Sartor et al., 1974; Ball et al., 1998; Egodawatta, 2007; Liu, 2011). A comprehensive study conducted by Egodawatta (2007) in areas of Gold Coast, Queensland suggested that a power

82

function could better replicate the observed pollutant build-ups compared to other mathematical functions. This study was also consistent with a previous investigation by Ball et al. (1998). Hence in this research, a power equation is used to estimate the pollutant build-up as presented in Equation 4.22,

𝐵𝐵 = 𝑎𝑎𝐷𝐷𝑏𝑏 (4.22)

Where, B refers to build-up load on road surfaces (in g/m2); D refers to the antecedent dry-days; and a and b refer to the build-up coefficients.

On the other hand, the most common mathematical function used to replicate the pollutant transport (wash-off) from the road surface is based on exponential functions. Different forms of exponential equations have been suggested by several researchers. Also, different rainfall and runoff parameters have been suggested as independent variables in the exponential equations proposed. For an example, Chiew et al. (1997) used runoff volume to estimate the wash-off whereas Sartor et al. (1974) used rainfall intensities to estimate the wash-off. However, Egodawatta (2007) concluded that the wash-off can be well replicated using the equation proposed by Sartor (1974) based on studies conducted in southeast Queensland and hence used in this research. The generic format of the equation suggested by Sartor (1974) is presented in Equation 4.23.

𝑊𝑊 = 𝑊𝑊𝑜𝑜(1 − 𝑒𝑒−𝑘𝑘𝑘𝑘𝑘𝑘) (4.23)

Where, W refers to the weight of mobilised material after time t; Wo refers to the initial weight of the material on the surface; I refers to the rainfall intensity; and k refers to the wash-off co-efficient.

Equation 4.23 suggests that the weight of material mobilised by a storm event is a simple function of the rainfall intensity, initial weight of the material and the surface characteristics of the catchment, assuming that any storm event would have the potential to wash-off all available pollutants on the surface. However, Egodawatta (2007) argued that the fraction of pollutant wash-off during a storm is always less than 1, suggesting a capacity factor, CF into the wash-off equation as given by Equation 4.24 and the capacity factors can be calculated using equations presented in Table 4.3.

𝑊𝑊𝑊𝑊𝑜𝑜

= 𝐶𝐶𝐹𝐹 (1 − 𝑒𝑒−𝑘𝑘𝑘𝑘𝑘𝑘) (4.24)

83

Table 4.3: Equations for capacity factors

Intensity Range, I, (mm/hr) Capacity Factor

5-40 0.01 𝐶𝐶 + 0.1

40-90 0.5

90-133 0.0098 𝐶𝐶 − 0.38

B Hydrological and hydraulic modelling

Hydrologic and hydraulic models are often used to estimate the stormwater quantities generated from catchments to support the stormwater quality modelling (Zoppou, 2001). Hydrologic and hydraulic models supply essential runoff information to the water quality modelling to calculate the pollutant concentration resulting from catchments during storm events.

In this research, in order to estimate the hydrological and hydraulic parameters, three commercially available and popularly used models namely, Mike URBAN, MUSIC and XP-SWIMM were evaluated based on the three criteria listed below,

1. Highly compatible with the adopted methodology of estimating stormwater quality. The model expected to estimate event-based Event Mean Concentrations (EMC) from a long period (20-years) of rainfall data.

2. Ability to manage large data. The research intended to simulate around 80 sets of 5-minute rainfall time-series for 20-year data (generated at two different stations, for two different GCMs, two different climate change scenarios, two different future timeframes and each of them having 10 realizations)

3. The capability of importing data in different file formats

A comparison of the three models against the selection criteria is presented in Table 4.4.

Table 4.4: Summary of the comparison of Mike Urban, MUSIC and XP-SWMM models with the selection criteria

Models Criteria 1 Criteria 2 Criteria 3

Mike URBAN × × √

MUSIC × √ ×

XP-SWMM × × × Note: √√ -very good; √ -good; and X - poor

84

Accordingly, none of the models was able to satisfy all the selection criteria. In particular, no model was compatible to estimate event-based EMCs for a long period of rainfall which is a primary requirement of the research. This is primarily because that the commercially available models are not capable of separating independent events for a rainfall time-series. Therefore, in this research, a new model was developed to simulate the hydrological and hydraulic to support the stormwater quality estimation. A detailed discussion on the development of the stormwater quality model for this research is presented in Chapter 9.

4.4.6 Classical univariate data analysis

A Mean and Standard Deviation

Mean and Standard Deviation (SD) are the commonly used statistics to describe characteristics of a univariate data set. The mean is primarily the most representative single value to describe the central tendency of a data set, whereas, SD describes the spread of data with respect to the mean. A smaller SD refers to the concentrated dataset and larger SD refers to a broadly distributed dataset (Hamburg, 1970).

B Quantiles

In statistics and probability theory, quantiles are cut-points dividing the range of a probability distribution into attached intervals or dividing the observations in a sample similarly. Typically, quartiles are the three cut points that will divide a dataset into four equal-sized groups. The cut-points are referred as Q1 (25% value), Q2 (median) and Q3 (75% value).

C Root Mean Square Error (RMSE)

Mathematical models were developed in different stages of this research. These equations required to be best fit the observed data of the validation periods. Therefore, it was critical to establish a statistics to define the goodness of fit. Among different statistics, RMSE is one of the most commonly applied techniques (Hamburg, 1970) and used appropriately in this research.

RMSE is the measures of the average of the difference between the observation and the simulated values using the developed models. Smaller RSME refers to good simulation and large RMSE refers to poor simulation. The equation used to calculate Root Mean Square Error is given by Equation 4.25.

85

𝑟𝑟𝑟𝑟𝛼𝛼𝑒𝑒 = 1𝑛𝑛

��𝑥𝑥𝑜𝑜,𝑖𝑖 − 𝑥𝑥𝑠𝑠,𝑖𝑖�2�𝑛𝑛

𝑖𝑖=1 (4.25)

Where, 𝑥𝑥𝑜𝑜 and 𝑥𝑥𝑠𝑠 refer to the observed and the simulated data.

4.5 Conclusions

This chapter provides a detailed discussion on the construct of the research methodology, selection of research tools and analytical methods to achieve the research objectives. The research methodology includes a comprehensive literature review; selection of study area, study tools and analytical methods; data collection; and modelling and analysis. A particular focus was given to the selection of study tools and analytical methods in this chapter.

Based on a detailed review on the ability to perform various statistical analysis, availability and accessibility, managing large data and ability to develop new software, R has been selected as the programming platform for the research. On the other hand, different analytical methods have been selected to achieve the objectives of the research. Hosking Wallis heterogeneity test and cluster analysis were selected to identify homogeneous regions and select representative meteorological stations for the research thereby achieve Objective 1 of the research. A spatial downscaling model based on Q-Q bias correction method and a temporal downscaling model based on homogeneous Markov model were decided to be developed to generate future rainfall data at finer spatial and temporal resolutions in order to achieve Objective 2 and 3 of the research. An at-site frequency analysis was expected to be performed to develop IFDs for future climate change scenarios to achieve Objective 4. Finally, a stormwater quality model was expected to be developed to estimate the stormwater qualities for future climate change scenarios and thereby achieve the Objective 5 of the research.

87

Chapter 5 Selection of Representative Meteorological Stations for Downscaling

5.1 Background

Developing statistical downscaling models requires high quality observed rainfall data for long periods (Wilby and Dawson, 2004). Such observed rainfall data are typically obtained from meteorological stations. However, the discrepancy in recorded time steps and missing data in a given area can diminish the quality of data needed for downscaling studies. Moreover, when a region with multiple meteorological stations is concerned, all stations do not necessarily require to be considered in developing the downscaling models due to similarities in data sets. Performing downscaling based on data belonging to multiple meteorological stations is time-consuming. Therefore, selecting the appropriate meteorological stations for downscaling studies was a critical task.

The scientific approach adopted in this research was to identify meteorological stations with similar rainfall characteristic by identifying rainfall homogeneous regions within the study area and thereby selecting representative meteorological stations. Analytical tools used for identification of homogeneous regions and selecting representative rainfall stations are presented in Section 4.4.1 in detail. Development of downscaling models was undertaken based on the observations at the representative meteorological stations and the future rainfall data generated at these stations were appropriately used for the respective homogeneous regions of the study area. Development of downscaling models and their technical attributes are presented in Chapter 6 and Chapter 7.

This chapter first presents the details of the selected study area and methods and criteria adopted for data collection. Then, detail discussions are presented on the delineation of homogeneous regions within the study area and selection of representative meteorological stations for downscaling.

88

5.2 Study area and data collection

5.2.1 Study area

Based on the selection criteria established in Chapter 4, southeast Queensland (SEQ) was selected as the study area for this research. SEQ is a political, biogeographical and administrative region of the state of Queensland in Australia. SEQ comprises of 3.4 million people out of the state's population of 4.8 million, which is around 70% of the total population of Queensland (ABS, 2014). The area covered by SEQ varies, depending on the definition of the region, though it tends to include Queensland's three largest cities: the capital city Brisbane; the Gold Coast; Logan City and the Sunshine Coast (see Figure 5.1). SEQ covers 22,420 square kilometres and incorporates 11 local government areas, extending 240 kilometres from Noosa in the north to the Gold Coast and New South Wales border in the south, and 140 kilometres west to Toowoomba. SEQ consists of highly urban areas, especially along the coast with rapid urban developments; population growth; tourism and industrial development.

Figure 5.1: Southeast Queensland

SEQ has a subtropical climate. SEQ experiences warm temperate summers with the average daily temperature varying between 200C and 280C. The winters are slightly cooler with the average daily temperature varying between 90C and 200C. While the seasonal rainfall pattern of SEQ is relatively consistent with wet summers and dry

89

winters, there is a considerable spatial variation in the average rainfall across the region. In average, the mean annual rainfall in SEQ varies between 800mm and 1700mm.

SEQ consists of a widespread network of waterways, creeks, and rivers spread across the region. However, studies on the waterways of SEQ have shown that the waterways are highly polluted by the stormwater runoff originating from the urban areas of the region (WBD, 2010; BCC & MBM, 2006; Goonetilleke et al., 2005). To counter this, SEQ has adopted the holistic approach of Water Sensitive Urban Design in the planning and design of urban developments in order to eliminate the negative impacts of urbanisation on the natural waterways. As a result, SEQ has a broad implementation of the WSUD philosophy in its developments.

SEQ is one of the most vulnerable regions in Australia to climate change due to its growing population and coastal location. The Intergovernmental Panel on Climate Change (IPCC, 2007) identified SEQ region as one of the ‘hot spots’ in Australia to be affected by climate change. The average annual temperature in SEQ has increased 0.4 °C (from 19.4 °C to 19.8 °C) between 1998 and 2007, showing a trend for further increase in the future (CCIA, 2015; Syktus et al., 2009). Studies have indicated an increase of up to 4°C in the average annual temperature by 2070 (Syktus et al., 2009; DEHP, 2016; CCIA, 2015). However, the influence of the climate change on the average annual rainfall is less clear (CCIA, 2015).

5.2.2 Data collection

SEQ has extensive metrological station network with measurements taken in daily, 3-hour, 30-minute and 1-minute (pluviography) formats. Among them, 17 meteorological stations were used for this research based on the availability of longer period of observation (at least for 5-years) in pluviographic format. The locations of the stations with their station number are presented in Figure 5.2, while additional information of the selected meteorological stations is presented in Table A1 in Appendix A.

Rainfall events were extracted from the time-series of data from each of the meteorological stations and variables such as antecedent dry day period, maximum rainfall intensity, total rainfall depth and the rainfall duration of rainfall events were determined. The following criteria were considered in extracting individual rainfall characteristics.

• An event was considered independent only if the consecutive event was separated by 3-hour antecedent dry duration. Otherwise, those events were treated as a single event.

90

• An event that constitutes less than 1-mm total rainfall for a period greater than 1-hour was not considered as a storm event (but a drizzle) and not considered for the analysis.

• The maximum rainfall intensities (in mm/hr) of the events were estimated by calculating the moving total of 1-hour rainfall throughout the storm duration.

• Any event having data entries with poor quality was discarded for the analysis.

Figure 5.2: The locations of the selected meteorological stations

5.3 Assessment of rainfall homogeneity in southeast Queensland

Among 17 meteorological stations identified, not all stations have substantially long recordings of high quality observed data, which is a prerequisite for developing statistical downscaling models. Furthermore, the observed data at these meteorological stations

91

may be similar in nature, and therefore, not all meteorological stations required to be considered in developing the downscaling models. Therefore, the strategy adopted in this research was to identify homogeneous rainfall regions within the study and selecting one meteorological station that has a high-quality observation for long period for each of the identified homogeneous regions as the representative meteorological station. By this way, development of downscaling models can be focused on the observations at the representative meteorological stations and the future rainfall data generated at these stations can be inferred to the entire rainfall homogeneous region.

The delineation of homogeneous regions for SEQ was carried out based on two different approaches:

1. Based on the characteristics of the continuous rainfall data (continuous-rainfall approach); and

2. Based on the characteristics of the event-based rainfall data (event-based rainfall approach).

5.3.1 Continuous-rainfall approach

In the continuous rainfall approach, the rainfall data series of the selected meteorological stations are tested for regional homogeneity purely based on the probabilistic characteristics of the rainfall recordings. The degrees of homogeneity of the stations are objectively determined using Hosking and Wallis heterogeneity test. Technical details of the Hosking and Wallis heterogeneity test are presented in Section 4.4.1 in Chapter 4. The Hosking and Wallis heterogeneity test estimates the degree of homogeneity of a group of meteorological stations based on the dispersion of L-moments among the given group of stations with what would be expected from an artificially developed homogeneous region (refer Section 4.1.1). In order to quantify the homogeneity of a selected region, the test uses three dispersion indexes, namely, H1, H2 and H3. The region is declared acceptably homogeneous if H1,H2, H3 < 1, possibly heterogeneous if 1 ≤ H1,2,3 < 2 and definitely heterogeneous if H1,2,3 ≥ 2 (Hosking and Wallis, 1997 & 2005). Further, Hosking and Wallis (2005) suggested that dispersion index H1 alone can be used to identify homogeneous regions as it has higher discriminatory power than other indexes. However, many studies use all three dispersion indexes or at least first two dispersion indexes to identify homogeneous regions (Mishra and Herath, 2015; Viglione, 2012). Therefore, in this research, the first two indexes, H1 and H2 have only been used to assess the degree of homogeneity of the region. The dispersion indexes were generated using an R package called ‘homtests’ (Viglione, 2012).

Hosking and Wallis heterogeneity test was performed using ‘homtests’ package on the selected 17 meteorological stations. All stations consisted of high quality and

92

substantially complete rainfall data in a pluviographic format for the period between 2011 and 2015. Therefore, rainfall data from this period was used to test the homogeneity of the meteorological stations. The results suggested that the entire SEQ could be treated as homogeneous with H1 = 0.6483 and H2 =0.9142.

5.3.2 Event-based rainfall approach

Although the overall rainfall characteristics of the entire SEQ were homogeneous, the event-based rainfall characteristics of the region may be different. For example, an event with intense rainfall for a short period of time and an event with less intense rainfall for a longer period of time may result in similar overall (average) rainfall characteristics. However, these events can potentially produce completely different stormwater quality and quantity scenarios. Therefore, it was important to consider the homogeneity of meteorological stations based on event-based rainfall characteristic when selecting the representative meteorological stations for this research.

In the context of WSUD, the event-based rainfall characteristics are perceived with more importance than the overall rainfall characteristics. Rainfall event characteristics such as antecedent dry days, rainfall intensities, total rainfall depth and rainfall durations are key characteristics of rainfall that are directly related to the stormwater quality and quantity. For example, pollutant build-up is primarily influenced by the antecedent dry days (Sartor et al., 1974; Ball et al., 1998; Egodawatta, 2007; Liu, 2011). In terms of pollutant wash-off, some researchers have identified the rainfall intensities (Sartor et al., 1974; Egodawatta, 2007) as the primary influence on stormwater quality, while others have indicated that the total rainfall depth (Chiew et al., 1997) is the primary influence. In addition, the total rainfall depth and the duration of rainfall events jointly provide indications to the quantities of stormwater generated from catchments (Zoppou, 2001; Egodawatta, 2007). Therefore, the homogeneity assessment required for this study needs consideration of event-based rainfall characteristics such as antecedent dry days, rainfall intensities, total rainfall depth and rainfall durations.

Similar to the continuous-rainfall approach, Hosking and Wallis heterogeneity test was performed to objectively assess the degree of rainfall homogeneity of the selected 17 meteorological stations of SEQ based on the event-based rainfall characteristics. Rainfall events were extracted from the time-series of data from these meteorological stations for a period between 2011 and 2015 based on the criteria established in Section 5.2.2. Rainfall characteristics such as antecedent dry day period, maximum rainfall intensity, total rainfall depth and the rainfall duration of each event were determined and Hosking and Wallis heterogeneity tests were performed based on each rainfall characteristics separately. The results of the Hosking and Wallis heterogeneity test based on the event-based rainfall characteristic are presented in Table 5.1.

93

Table 5.1: Dispersion indexes based on event-based rainfall approach for southeast Queensland

𝑯𝑯𝟏𝟏 𝑯𝑯𝟐𝟐 Remarks

Antecedent dry day periods

3.37 1.16 Heterogeneous

Maximum rainfall intensity

1.95 0.59 Possibly heterogeneous

Total rainfall 0.81 0.32 Homogeneous

Duration 0.25 0.10 Homogeneous

Based on the first two dispersion indexes (H1 and H2,) antecedent dry day periods showed a higher level of heterogeneity compared to other event-based rainfall characteristics. The maximum rainfall was found to be potentially heterogeneous across SEQ. In contrast, the total rainfall and the rainfall duration were homogeneous across the SEQ. Overall these results suggest that the entire SEQ cannot be considered homogeneous based on the event-based rainfall characteristics. Accordingly, homogeneous regions suggested based on the continuous-rainfall approach may not necessarily be homogeneous based on individual rainfall characteristics. In addition, it was also noticeable that antecedent dry days and the maximum rainfall intensity showed heterogeneity among the meteorological stations while the total duration and the total rainfall of the events were homogeneous. Therefore, it can be concluded that antecedent dry day periods and maximum rainfall intensity have higher spatial variation and thus should be the deciding rainfall characteristics in delineations of the homogeneous regions.

Accordingly, the next step was to identify all potential homogeneous regions inside SEQ and to assess the degree of homogeneity of identified potential homogeneous regions (using the Hosking and Wallis heterogeneity test). In order to identify the potential homogeneous regions, cluster analysis was performed based on the rainfall characteristics. A detailed discussion on the cluster analysis and the algorithm associated with different cluster analyses are presented in Section 4.4.1. Agglomerative hierarchical cluster analysis was performed using R package ‘stats’ (R Development Core Team, 2016). The parameters used for the analysis included 3rd quartile (Q3) values of antecedent dry day periods, maximum rainfall intensity, total rainfall and duration of the individual events of each station as given in Table 5.2. This was because the average or the median of the rainfall characteristics at the considered meteorological stations were expected to be similar in nature and therefore may not be grouped into

94

discrete clusters. In contrast, selecting a higher quantile value may result in too many unrealistic clusters. Therefore, the 3rd quartile values were used in this analysis.

Table 5.2: Data matrix for the cluster analysis

Serial number

Station number

3rd quartile values – Q3 Antecedent dry-days (days)

Maximum intensities (mm/h)

Total rainfall (mm)

Duration (h)

1 40004 6.4 7.1 12.0 5.3 2 40043 5.4 6.7 13.4 6.1 3 40082 8.4 7.6 14.4 5.9 4 40093 6.0 6.0 13.5 5.9 5 40211 5.9 6.8 11.6 4.8 6 40717 3.8 6.4 12.3 5.7 7 40764 5.0 7.0 12.4 6.0 8 40842 6.0 6.6 12.1 5.4 9 40861 4.3 8.2 13.8 7.2 10 40908 3.3 5.9 10.4 5.3 11 40913 6.1 7.4 13.2 5.8 12 40922 9.8 8.2 12.4 6.2 13 40958 6.1 7.3 12.6 5.3 14 40983 7.1 7.6 12.8 5.3 15 40988 4.2 6.2 11.9 6.5 16 41525 7.9 6.6 14.1 5.9 17 41529 8.2 7.8 13.2 5.7

95

9

7 11

16

12 3

14 17 2 4

6 15

1 13 8

5

10

6

5

4

2

3

1

0

Dist

ance

Figure 5.3 presents the dendrogram developed using the ‘stats’ package. The dendrogram illustrates the merging of similar meteorological stations forming clusters at different levels and successively merging into a single cluster. Based on Figure 5.3, it can be seen that there are three discrete branches of the dendrogram forming three clusters. Accordingly, Cluster 1 comprised of Stations 9, 7 and 11. Cluster 2 comprised of Stations 3, 12, 14, 16 and 17 and Cluster 3 comprised of Stations 1, 5, 6, 8, 10, 13 and 15.

Figure 5.3: Dendrogram generated from the cluster analysis

The geographical locations of the meteorological stations of the clusters are presented in Figure 5.4. Figure 5.4 suggests that the meteorological stations of Cluster 1 and Cluster 2 were located in close proximity to the coast of the SEQ and the meteorological stations of Cluster 3 were located inland. However, Cluster 1 and Cluster 2 stations do not show clear geographical separation.

96

Figure 5.4: Geographical locations of the meteorological stations and their grouping

Further, scatter plots were produced to examine of meteorological stations of the three clusters based on the considered rainfall characteristics and presented in Figure 5.5. It can be observed from Figure 5.5 that the meteorological stations of the Cluster 3 have a clear distinction between the meteorological stations of Cluster 1 and Cluster 2. In contrast, the meteorological station of Cluster 1 and Cluster 2 showed no clear separations. These results suggest that the Cluster 1 and Cluster 2 can be treated as single clusters representing the coastal SEQ while the Cluster 3 representing the inland SEQ. Accordingly, two potential homogeneous regions were identified within the study area, namely, Coastal-SEQ and Inland-SEQ.

Cluster 1

Cluster 2

Cluster 3

97

Figure 5.5: Scatterplots of the event-based rainfall characteristics (Red dots refer to the stations of Cluster 1, Black dots refer to the stations of - Cluster 2 and Green dots refer to the stations of Cluster 3)

Coast

Inland

14000

12000

10000

8000

6000

Ant

eced

ent

dry-

days

(m

in)

300 340 380 420 Duration (min) Intensity (mm/min)

Inland

Coast

14000

12000

10000

8000

6000

0.6 0.7 0.8 0.9 1.0

Total rainfall (mm)

Inland

Coast

14000

12000

10000

8000

6000

Ant

eced

ent

dry-

days

(m

in)

11 12 13 14

98

The results of the cluster analysis provided insight for selecting the potential homogeneous regions, and the rainfall homogeneousness were tested objectively for the selected groups. The degree of homogeneousness of the identified regions of Coastal-SEQ and Inland-SEQ were evaluated by performing Hosking and Wallis heterogeneity tests using event-based rainfall characteristics. The summary of the result is presented in Table 5.3. As shown in Table 5.3, the dispersion indexes H1 and H2 were found less than one for all the rainfall characteristics for both Coastal-SEQ and Inland-SEQ. Therefore, Coastal-SEQ and Inland-SEQ were identified as two separate homogeneous regions within SEQ based on the event-based rainfall characteristics.

Table 5.3: Dispersion indexes for Hosking and Wallis heterogeneity test for Coastal-SEQ and Inland-SEQ

Coastal-SEQ Inland-SEQ

H1 H2 H1 H2

Antecedent dry day periods

-0.20 1.18 -0.35 0.06

Maximum rainfall intensity

0.78 0.97 -0.61 -0.65

Total rainfall -0.55 0.65 -0.54 -0.10

Duration 0.26 0.32 -1.29 -1.00

5.4 Boundaries and representative meteorological station of the homogeneous regions

5.4.1 Rainfall homogeneous regions within southeast Queensland

Based on the Hosking and Wallis Heterogeneity test and cluster analysis presented in Section 5.3, two homogeneous regions, namely Coastal-SEQ and Inland-SEQ were identified within the study area based on event-based rainfall characteristics. The Coastal-SEQ includes the greater Brisbane regions, north Brisbane and south Brisbane areas whereas the Inland-SEQ includes the western regions. The boundaries for the two regions were determined by forming polylines along the perpendicular bisectors of two neighbouring stations from the two different regions. Then, these borderlines were smoothened and adjusted slightly to fit the nearest local government boundaries for consistency as presented in Figure 5.6. The local government bodies included in the identified homogeneous regions is presented in Table 5.4.

99

Table 5.4: Local government bodies included in the identified homogeneous regions

Region Local government bodies

Coastal-SEQ

Brisbane City Council, Gold Coast City Council, Logan City Council, Redland City Council, Moreton Bay Regional Council and Sunshine Coast Regional Council.

Inland-SEQ Ipswich City Council, Scenic Rim Regional Council, Lockyer Regional Council and Somerset Regional Council

Figure 5.6: Boundaries of Coastal-SEQ and Inland-SEQ rainfall homogeneous regions

Inland-SEQ

Coastal-SEQ

100

The results are partially consistent with the adopted climate regions for SEQ in the WSUD conceptual design (BCC & MBW, 2006). However, BCC & MBW (2006) further separated coastal SEQ into three regions namely, North Coast, Greater Brisbane and South Coast. However, the partitioning was based on a subjective approach. Two climate characteristics, average number of rain days per year and mean annual rainfall were only considered in this approach. Even though the costal-SEQ has been identified as three separate homogeneous regions by BCC & MBW (2006), the entire Coastal-SEQ showed the similar average number of rain days per year and mean annual rainfall. On the other hand, based on the National Resource Management (NRM) homogeneous regions, the entire SEQ is considered as a single homogeneous region called East Coast which contradicts the results of this research. NRM regions defined largely correspond to the broad-scale climate and biophysical regions of Australia. However, climate change impacts and adaptation approach of Australia is broadly based on NRM regions (Dowdy et al., 2015, CCIA, 2015). For example, the interim climate change factors suggested by AR&R (2015) were based on the NRM regions, implying the changes in the temperatures and rainfall for the entire region is uniform. However, many climate change impact assessments such as WSUD are sensitive to changes in the local-scale rainfall. Therefore, the homogeneous regions identified based on event-based rainfall characteristics in this research are more appropriate for local-scale investigations.

5.4.2 Representative meteorological stations for southeast Queensland

Meteorological stations within the identified homogeneous regions have similar rainfall characteristics such as antecedent dry day periods, rainfall intensities, total rainfall and rainfall durations. Based on this, it can be hypothesised that the future rainfall data generated (using downscaling models) based on the data from any of the stations can be inferred to the respective homogeneous regions. However, in order to develop downscaling models, suitable metrological stations within each homogeneous region should be selected as representative meteorological stations. In this regard, the criteria adopted for selection is primarily based on the availability of high quality and complete data sets for a long period of time. In addition to this selection criterion, stations used in previous studies relating to WSUD were given more preference and stations along the borderlines were given less preference in the selection of representative meteorological stations.

Accordingly, for Coastal-SEQ, Gold Coast Seaway station (40764) was selected as the representative meteorological stations. Gold Coast Seaway station has rainfall recording in pluviographic format since 2000 with almost 100% completeness. Moreover, Gold Coast Seaway station has been used in several previous research studies including Ma (2016), Chowdhury (2018), Liu (2011) and Egodawatta (2007). For Inland-SEQ,

101

Toowoomba Airport station (41529) was selected as the representative meteorological station since the station has the longest period of the rainfall records in pluviographic format (since 2009 with around 100% completeness) compared to other meteorological station in Inland-SEQ.

The rainfall data of the representative meteorological stations were used to develop downscaling models and the future data generated based on these downscaling models were inferred to the respective homogeneous regions. The procedures of developing downscaling models are presented in Chapter 6 and Chapter 7 in detail.

5.5 Conclusions

The following conclusions were derived from the analysis of this chapter:

• The entire southeast Queensland can be treated as a homogeneous region based on the continuous-rainfall approach. However, based on individual rainfall characteristics such as antecedent dry-days, maximum rainfall intensities, total rainfall and duration of the rainfall events, there were two separate homogeneous regions identified. This implies that although the characteristics of the continuous rainfall data between stations were statistically similar, the event-based characteristics have a significant difference among stations.

• Antecedent dry-days and maximum rainfall intensities of the rainfall events have significant variation between the coastal and the inland areas of SEQ compared to the total rainfall and duration of the event.

• Two homogeneous regions were identified based on the event-based rainfall characteristics and named as Coastal-SEQ and Inland-SEQ. Coastal-SEQ includes Brisbane City Council, Gold Coast City Council, Logan City Council, Redland City Council, Moreton Bay Regional Council and Sunshine Coast Regional Council areas and the Inland-SEQ includes Ipswich City Council, Scenic Rim Regional Council, Lockyer Regional Council and Somerset Regional Council areas.

• Two representative meteorological stations Gold Coast Seaway station (40764) and Toowoomba Airport stations (41529) were selected to represent the Coastal-SEQ and Inland-SEQ respectively based on a set of selection criteria.

103

Chapter 6 Spatial Downscaling of Rainfall Data Using Bias Correction Method

6.1 Background

As noted in Chapter 3, statistical downscaling models are popularly used in many climate change impact assessments as they are less expensive, straightforward, computationally undemanding and capable of producing more accurate climate information than dynamic downscaling (Ahmed et al., 2013; Schmidli et al., 2007; Jaw et al., 2015). Several statistical downscaling tools have been suggested in the literature based on different downscaling methods (Timbal and McAvaney, 2001; Mehrotra and Sharma (2006); Wilks and Wilby 1999; Mishra and Herath, 2015). However, there are many limitations in the existing downscaling tools. The limitations include the ability to produce only daily climate data, which is not adequate for accurate impact assessment including the assessment of future stormwater quantities and qualities. Some of the tools are research-specific and developed only to downscale targeted climate characteristics. Moreover, some tools are associated with Global Circulation Models (GCM) from the CMIP 3 family with older emission scenarios and not available for public use.

Therefore, a more robust spatial downscaling tool was developed in this research to eliminate the limitations in the existing downscaling tools. The tool was designed to develop Quantile-Quantile (Q-Q) bias correction models to downscale rainfall data from Global Circulation Models. The reason for the selection of Q-Q bias correction approach for this tool is presented in detail in Section 4.4.2. Further, the new spatial downscaling tool was used to produce Q-Q bias correction models to downscale 3-hour rainfall time-series from two GCMs, namely, ACCESS 1.0 and EC-EARTH for southeast Queensland (SEQ). The models were developed based on the rainfall data from selected representative meteorological stations discussed in Chapter 5. The approach adopted for temporal downscaling of the bias-corrected data is explained in Chapter 7.

104

This chapter presents a detailed discussion on the development of the new downscaling tool including the downscaling method used in developing the tool, its architecture and the capabilities. The chapter also presents the characteristics of spatially downscaled rainfall data for SEQ.

6.2 Development of the spatial downscaling tool

6.2.1 Downscaling method

As noted in Chapter 4, the bias correction method was selected for the development of the spatial downscaling tool. The primary reason for this selection is that the bias correction method does not require predictors, unlike other downscaling methods. Most of the other methods require reliable predictor variable from the GCMs at the same temporal resolutions of the predictands. However, GCMs, in general, do not produce predictor variables at the sub-daily temporal resolution (CMIP5, 2015). Therefore, statistical downscaling methods which require predictor variables (such as regression method, weather classification methods and weather generating methods) are not suitable for this research. In contrast, bias correction methods do not require predictor variables and therefore, more suitable for this research. In addition, the bias correction method is one of the most effective statistical downscaling methods gaining popularity among researchers working on climate change impact assessments (Mishra and Herath, 2015).

In general, all GCMs are inherent to systematic misrepresentations during their simulation stage (Teutschbein and Seibert, 2012), due to limitations in scales and simplification of processes and equations (Haerter et al., 2010; Ehret et al., 2012). These misrepresentations in the GCM simulations can be easily identified and corrected. This identification and correction of misrepresentation in GCM data are referred to as the bias correction. Quantile-Quantile (Q-Q) bias correction is a popular bias correction method used in a range of recent climate change studies (Ines and Hansen, 2006; Sharma et al., 2007; Elshamy et al., 2009; Mishra and Herath 2015). The Q-Q bias correction method compares the probability distribution of the GCM data and the observed data and accordingly corrects the bias in the GCM data. A detailed discussion on the step-by-step procedure for the Q-Q bias correction approach used in the development of the new statistical downscaling tool is presented below.

105

6.2.2 Architecture of the downscaling tool

A step-by-step procedure adopted to develop a downscaling tool using the Q-Q bias correction method is discussed in this section. The discussion utilises the schematic provided in Figure 6.1.

Step 1: Input data

The tool was designed to operate with two types of input data, namely, observed data and GCM data. Both of these data sets are used for calibration and validation of the tool. A separate future data set obtained from GCMs for the required future climate scenario is also needed for this tool. Observed data used for calibration and validate is typically from a meteorological station, while the GCM data is from the Global Climate Model simulations for the same historical period as for the observed data. Both datasets used for calibration and validation required to be recorded at the same time steps. The data for the historical period was required to be further split into two sets for the use of calibration and the validation of the models as shown in Figure 6.1.

Step 2: Determination of probability distribution parameters and construction of the cumulative density functions

The tool was designed to determine the basic probability distribution parameters such as mean, standard deviation, shape factor and scale factors of the data sets in order to construct the cumulative density functions. The cumulative density functions are developed assuming a two-parameter gamma distribution. The two-parameter gamma distribution was adopted to model the rainfall data in this tool due to its common use in studies relating to rainfall data analysis (Mirsha and Herath, 2015; Hanson and Vogal, 2008; Sharma and Singh, 2010; Husak et al., 2007). Furthermore, the gamma distribution is invertible and readily available on the R platform. The general form of the cumulative density function of the two-parameter gamma distribution is shown in Equation 6.1.

𝐹𝐹(𝑥𝑥,𝛼𝛼, 𝛽𝛽) = � 𝛼𝛼 � 1𝛽𝛽𝛼𝛼Г(𝛼𝛼)

𝑥𝑥𝛼𝛼−1𝑒𝑒�−𝑥𝑥𝛽𝛽 ��𝑑𝑑𝑥𝑥

𝑥𝑥

0 (6.1)

Where, x represents the rainfall data (0 ≤ x <∞); α = (µσ)2 refers to the shape factor;

and β = σ2

µ refers to the scale factor. µ and σ are mean and standard deviation of the

rainfall data.

106

Figure 6.1: Architecture of downscaling tool

(CALIBRATION) (VALIDATION) (FUTURE DATA)

3. Developing Q-Q bias correction models 2.1 - Correcting zero rainfall 2.2 - Correcting non-zero rainfall

4. Validation

𝑑𝑑𝑜𝑜 = 𝑑𝑑𝑠𝑠𝑛𝑛



𝐹𝐹𝑠𝑠,𝑝𝑝𝑝𝑝𝑝𝑝𝑠𝑠𝑝𝑝𝑛𝑛𝑘𝑘−1 �𝐹𝐹𝑠𝑠,𝑓𝑓𝑓𝑓𝑘𝑘𝑓𝑓𝑝𝑝𝑝𝑝�𝑥𝑥𝑠𝑠,𝑓𝑓𝑓𝑓𝑘𝑘𝑓𝑓𝑝𝑝𝑝𝑝��

× 𝑥𝑥𝑠𝑠,𝑓𝑓𝑓𝑓𝑘𝑘𝑓𝑓𝑝𝑝𝑝𝑝

MODEL

Validation outputs

……….....................................................

5. Downscaling future rainfall

time-series

2. Determination of probability distribution parameters and construction of the CDF

……….....................................................

……….....................................................

Observed GCM

……….....................................................

……….....................................................

Observed GCM

……….....................................................

GCM

Probability parameters and CFD



Rai

nfal

l tim

e-se

ries

1. Input data

Historical data Historical data

107

Step 3: Developing Q-Q bias correction models

The tool adapts the Q-Q bias correction approach for assessing and quantifying the bias between GCM simulations and the observed data for a historical period by means of mathematical models. Then, the quantified bias is used for correcting future GCM simulations. This process was undertaken in two sub-steps, namely, developing a model to correct zero rainfall (also known as the frequency correction) and developing a model to correct non-zero rainfall (also known as the intensity correction).

- Developing a model to correct zero rainfall (frequency correction)

It is expected that the number of zero rainfall time-steps in the GCM data should be approximately equal to that in the observed data. By satisfying this criterion, it can be postulated that the frequency of rainfall events are approximately similar for observed data and GCM outputs. This feature in rainfall data can be assessed by comparing the Cumulative Frequency (CF) corresponding to the zero rainfall, F(x, 𝛼𝛼, 𝛽𝛽 =0), in the GCM data and the observed data. As noted by a range of researchers, GCMs underestimate the proportion of zero rainfall compared to the observations (Mirsha and Herath, 2015; Ines and Hansen 2006; Elshamy et al., 2009). Therefore, it is important to introduce this correction to the zero rainfall in the GCM simulations.

Generally, correction for zero rainfall is applied by calculating a CF to truncate the gamma distribution of the GCM data such that the mean frequency of rainfall above the determined CF matches the observed mean rainfall frequency. For this, CF corresponding to zero rainfall in the observed data for the historical period, which is commonly termed as the threshold is calculated as presented in Figure 6.2(a). The calculated threshold is then used as the CF corresponding to the zero rainfall for the future period. (Mirsha and Herath, 2015; Sharma and Singh, 2010), implying that the ratio of zero rainfall to non-zero rainfall is the same for both historical and future scenarios. However, many studies have suggested otherwise (Almazroui et al., 2017, Mehrotra and Sharma, 2006; Singh et al, 2014). For example, Almazroui et al. (2017), has suggested an increased number of rainy days (wet spell) over western Saudi Arabia in the future compared to the historical climate.

Therefore, this tool was designed to estimate the CF corresponding to the zero rainfall for the future period by defining a relationship between the CF for zero rainfall in the GCM simulations and observed data. A simple mathematical model was developed between CF corresponding to the zero rainfall in the GCM data and CF corresponding to the zero rainfall in the observed data using a polynomial function (based on the historical data) as shown in Figure 6.2(b). Accordingly, the equation developed was assumed to be true for future scenarios and intended for use

108

in correcting the CF for the zero rainfall of the future GCM output. The polynomial equation used is given in Equation 6.2.

𝑑𝑑𝑜𝑜 = 𝑑𝑑𝑠𝑠𝑛𝑛 (6.2)

Where, to refers to the threshold (corrected CF of the zero rainfall) and ts refers to the CF of the zero rainfall from GCM data. n is a coefficient established based on the historical data (calibration) as given in Equation 6.3.

𝑛𝑛 = ln( 𝑑𝑑0,𝑐𝑐𝑐𝑐𝑖𝑖𝑖𝑖𝑏𝑏𝑝𝑝𝑐𝑐𝑘𝑘𝑖𝑖𝑜𝑜𝑛𝑛)ln( 𝑑𝑑𝑠𝑠,𝑐𝑐𝑐𝑐𝑖𝑖𝑖𝑖𝑏𝑏𝑝𝑝𝑐𝑐𝑘𝑘𝑖𝑖𝑜𝑜𝑛𝑛)

(6.3)

- Developing model to correct non-zero rainfall (intensity correction)

It is expected that the rainfall intensities of the GCM data are closely similar to that of observed data for each quantile considered. Based on this, the tool was designed to develop models to correct the non-zero rainfall in the future GCM data by mapping the differences in the observed data and GCM simulations for a historical period. Accordingly, for a given rainfall in the GCM data, the rainfall corresponding to the same quantile in the observed data was mapped as presented in Figure 6.2(c). The ratio between the rainfall from the observed data and the GCM output was determined for every quantile. This ratio is then used to correct the GCM outputs for the future period, implying that for the particular quantile, the ratio is a constant even outside of the calibration period (Michelangeli et al., 2009; Mishra and Herath, 2015). The development of the model to correct the non-zero rainfall can be explained using Equation 6.4 to 6.8.

For a given quantile, 𝑥𝑥𝑜𝑜𝑥𝑥𝑠𝑠

= 𝑐𝑐𝑐𝑐𝑛𝑛𝛼𝛼𝑑𝑑𝑎𝑎𝑛𝑛𝑑𝑑 (6.4)

Where, x0 refers to the rainfall from the observed data and xs refers to the rainfall data from the GCM data for a given quantile.

Based on Equation 6.4, the relationship between data from historical period and future period can be established in the form of Equations 6.5 and 6.6.

109

Figure 6.2: Developing the statistical model for bias correction. (a) Mapping zero rainfall; (b) The Polynomial relationship between the threshold values of the observed data and GCM data. Three points have been used for the curve – red dot is found using the calibration data and the black dots refer to the lowest (0) and the largest (1) possible values; (c) Mapping non-zero rainfall.

(a)

CF

corr

espo

ndin

g ze

ro r

ainf

all i

n th

e ob

serv

ed d

ata

(Thr

esho

ld)

( 𝑡𝑡 𝑜𝑜

)

CF corresponding zero rainfall in the GCM data (𝑡𝑡𝑠𝑠)

(b)

CD

F

Threshold

10 20 30 40 Rainfall (mm)

0.2

0.4

0.6

0.8

1.0

GCM data

Observed data

x1 x2

(c)

Observed data

GCM output

q

For quantile q,

x1/x2 = constant

Rainfall (mm)

10 20 30 40 0 50

1.0

0.9

0.8

CD

F

110

𝑥𝑥𝑜𝑜,ℎ𝑖𝑖𝑠𝑠𝑘𝑘𝑜𝑜𝑝𝑝𝑖𝑖𝑐𝑐𝑐𝑐𝑖𝑖

𝑥𝑥𝑠𝑠,ℎ𝑖𝑖𝑠𝑠𝑘𝑘𝑜𝑜𝑝𝑝𝑖𝑖𝑐𝑐𝑐𝑐𝑖𝑖=

𝑥𝑥𝑜𝑜,𝑓𝑓𝑓𝑓𝑘𝑘𝑓𝑓𝑝𝑝𝑝𝑝

𝑥𝑥𝑠𝑠,𝑓𝑓𝑓𝑓𝑘𝑘𝑓𝑓𝑝𝑝𝑝𝑝 (6.5)

𝑥𝑥𝑜𝑜,𝑓𝑓𝑓𝑓𝑘𝑘𝑓𝑓𝑝𝑝𝑝𝑝 =𝑥𝑥𝑜𝑜,ℎ𝑖𝑖𝑠𝑠𝑘𝑘𝑜𝑜𝑝𝑝𝑖𝑖𝑐𝑐𝑐𝑐𝑖𝑖

𝑥𝑥𝑠𝑠,ℎ𝑖𝑖𝑠𝑠𝑘𝑘𝑜𝑜𝑝𝑝𝑖𝑖𝑐𝑐𝑐𝑐𝑖𝑖 × 𝑥𝑥𝑠𝑠,𝑓𝑓𝑓𝑓𝑘𝑘𝑓𝑓𝑝𝑝𝑝𝑝

(6.6)

Where, x0, historical and xs, historical refer to the rainfall for the historical period from the observed data and GCM data respectively; xs, future refers to the rainfall for the future period from GCM data; and x0, future refers to the bias-corrected rainfall for the future period.

Since the gamma distribution is invertible, the Equation 6.6 can be rewritten as shown in Equation 6.7. 𝑥𝑥𝑜𝑜,𝑓𝑓𝑓𝑓𝑘𝑘𝑓𝑓𝑝𝑝𝑝𝑝 =

𝐹𝐹𝑜𝑜,ℎ𝑖𝑖𝑠𝑠𝑘𝑘𝑜𝑜𝑝𝑝𝑖𝑖𝑐𝑐𝑐𝑐𝑖𝑖−1(𝑞𝑞)

𝐹𝐹𝑠𝑠,ℎ𝑖𝑖𝑠𝑠𝑘𝑘𝑜𝑜𝑝𝑝𝑖𝑖𝑐𝑐𝑐𝑐𝑖𝑖−1(𝑞𝑞)


Where, q refers to the quantile considered; xs, future refers to the rainfall for the future period from GCM data; x0, future refers to the bias-corrected rainfall for the future period; and 𝐹𝐹𝑜𝑜,ℎ𝑖𝑖𝑠𝑠𝑘𝑘𝑜𝑜𝑝𝑝𝑖𝑖𝑐𝑐𝑐𝑐𝑖𝑖

−1() and 𝐹𝐹𝑠𝑠,ℎ𝑖𝑖𝑠𝑠𝑘𝑘𝑜𝑜𝑝𝑝𝑖𝑖𝑐𝑐𝑐𝑐𝑖𝑖−1() refer to the inverse of cumulative

density function of the historical data from the observation and GCM data respectively. Therefore,

𝑥𝑥𝑜𝑜,𝑓𝑓𝑓𝑓𝑘𝑘𝑓𝑓𝑝𝑝𝑝𝑝 =𝐹𝐹𝑜𝑜,ℎ𝑖𝑖𝑠𝑠𝑘𝑘𝑜𝑜𝑝𝑝𝑖𝑖𝑐𝑐𝑐𝑐𝑖𝑖


𝐹𝐹𝑠𝑠,ℎ𝑖𝑖𝑠𝑠𝑘𝑘𝑜𝑜𝑝𝑝𝑖𝑖𝑐𝑐𝑐𝑐𝑖𝑖−1 �𝐹𝐹𝑠𝑠,𝑓𝑓𝑓𝑓𝑘𝑘𝑓𝑓𝑝𝑝𝑝𝑝�𝑥𝑥𝑠𝑠,𝑓𝑓𝑓𝑓𝑘𝑘𝑓𝑓𝑝𝑝𝑝𝑝��


Where, xs, future refers to the rainfall for the future period from GCM data; x0, future refers to the bias-corrected rainfall for the future period; 𝐹𝐹𝑜𝑜,ℎ𝑖𝑖𝑠𝑠𝑘𝑘𝑜𝑜𝑝𝑝𝑖𝑖𝑐𝑐𝑐𝑐𝑖𝑖

−1() and

𝐹𝐹𝑠𝑠,ℎ𝑖𝑖𝑠𝑠𝑘𝑘𝑜𝑜𝑝𝑝𝑖𝑖𝑐𝑐𝑐𝑐𝑖𝑖−1() refer to the inverse of cumulative density function of the historical

data from the observation and GCM data respectively; and 𝐹𝐹𝑠𝑠,𝑓𝑓𝑓𝑓𝑘𝑘𝑓𝑓𝑝𝑝𝑝𝑝() refers to the

cumulative density function of the future GCM data.

Step 4: Validation

Once the Q-Q bias correction models were developed, the tool was designed to validate the performance of the developed models using an independent historical dataset. This was undertaken by assessing the similarity between the bias-corrected GCM data and the observed data for the validation period. In order to assess the similarities of the data, the tool uses two statistical indexes namely Root Mean Square Error (RMSE) and Gradient (m) of the observation and GCM data scatter plots. The tool is designed to provide numerical and graphical outputs of the validation outcomes.

111

Step 5: Downscaling Future rainfall

The final step of the tool is to correct the future GCM data for any given climate change scenarios using the developed models. For this process, the developed and validated models described in step 1 to 4 were used to downscale the GCM data for the required future scenarios.

This downscaling tool has already been developed and published as an R package by the name ‘spdownscale’ for public use (Rasheed et al., 2017). spdownscale is licensed under the General Public License, version 2 (GPL 2), where the public is permitted to copy and distribute verbatim copies of this source code with restrictions to change it. spdownscale is freely available as an R package (https://CRAN.R-project.org/ package=spdownscale) and details on installing the R packages can be found at R Development Core Team (2016). A detailed description of the functions of the spdownscale is given below.

6.2.3 Functions of spdownscale

The spdownscale package has three functions, namely, ParaCal(), ResVal() and downscale(). The package also contains three datasets, namely, data_mod, data_mod_fut and data_obs, primarily for demonstration purposes. A summary description of these functions and datasets are presented in Table 6.1.

Table 6.1: Descriptions of the functions and datasets in spdownscale

Functions Descriptions

ParaCal() Displays the shape factors, scale factors and the threshold values of the observed data and GCM outputs.

ResVal() Displays the summary of the validation results.

downscale() Generating the future climate data (rainfall)

data_obs

Observed rainfall data at the Gold Coast Seaway meteorological station, Australia (station number - 40764, Period- 1/4/2000 to 12/31/2012, Latitude/longitude - 27.9390/153.4283).

data_mod EC-EARTH rainfall data at the Gold Coast Seaway meteorological station, Australia (station number - 40764, Period- 1/1/2000 to 12/31/2012, Lat/Lon - 27.9390/153.4283)

data_mod_fut

EC-EARTH future (RCP 4.5) rainfall data at the Gold Coast Seaway meteorological station, Australia (station number - 40764, period- 1/1/2026 to 12/31/2045, Lat/Lon - 27.9390/153.4283)

https://cran.r-project.org/%20package=spdownscale

https://cran.r-project.org/%20package=spdownscale

112

The user has to input five sets of data in order to execute any of the functions; The obs_cal (observed data for calibration), mod_cal (GCM outputs for calibration), obs_val (observed data for validation), mod_val (GCM outputs for validation) and mod_fut (GCM outputs for the chosen future climate scenarios). All the dataset are required to be in the vector or data.frame or matrix format.

The sample data sets provided in spdownscale were observed rainfall data from Gold Coast Seaway station and EC-EARTH GCM as presented in Table 6.1. This data was used for analytical purposes in this research and included in the spdownscale tool for demonstration purpose.

A ParaCal()

Executing the function ParaCal() provides a list of calibration parameters such as shape factors, scale factors, thresholds and n as described in Table 6.1. This function is the execution of Step 2 and 3 in Section 6.2.2. The shape and scale factors are used to determine the model parameters for Equation 6.8. The threshold and n are used to determine the model parameters for Equation 6.2. Source codes for ParaCal() function is presented in Appendix B1.

B ResVal()

ResVal() is the execution of Step 4 in Section 6.2.2, which validates the established models given in Equations 6.2 and 6.8, using rainfall data for a user-defined independent historical period. Bias-corrected data are then compared against the observed data for the same period. ResVal() function displays all the validation results such as percentage error in the threshold, RMSE and the gradient in numeric and graphical formats. The gradient of observation and GCM data scatter plots and RMSE provide similarity measure between the observation and GCM data. The equation and methods used to estimate these parameters are explained in Section 4.4.6. Source codes for ResVal() function is presented in Appendix B2.

C downscale ()

downscale() is the execution of Step 5 in Section 6.2.2. Once the user is confident about the models from the validation outcomes, downscaling can be performed using the downscale() function. The downscale() function corrects the GCM outputs for the chosen future climate change scenarios and generates a corrected time-series of rainfall time-series in “data.frame” format. The output file would not be saved automatically in the system, but be readily available in the R global environment by the name crt_mod_fut after executing the function. Source codes for downscale() function is presented in Appendix B3.

113

6.4 Spatial downscaling of rainfall data for southeast Queensland (SEQ)

The developed downscaling tool, spdownscale was used to spatially downscale rainfall data for southeast Queensland. Two GCMs, namely, EC-EARTH and ACCESS-1.0 were used to downscale rainfall for this task based on the critical review of literature presented in Chapter 3. These two GCMs were found to perform well in simulating the historical climate for Australia and widely used in various climate change studies in Australia (Hazeleger et al., 2012; CCIA, 2015; Daohua et al., 2012). Both the GCMs produce 3-hour time-series of rainfall data for the historic and future timeframes including RCP 4.5 and RCP 8.5 climate change scenarios, which are the IPCC recommended climate change scenarios for impact assessments (IPCC, 2014). The GCM outputs were collected from the CMIP5 project (CMIP5, 2015) and extracted using Climate Data Operators (CDO) developed by the Max Planck Institute, Germany (Schulzweida et al., 2009).

6.4.1 Spatial downscaling for Coastal-SEQ

Based on the analysis presented in Chapter 5, Gold Coast Seaway station was selected as the representative station for Coastal-SEQ. The selection was primarily due to having a longer period of high-quality rainfall records compared to other suitable stations. Accordingly, rainfall data from Gold Coast Seaway station (40764) was used for the downscaling as the representative meteorological station for Coastal-SEQ. Rainfall records are available for Gold Coast Seaway station since 2000 in 3-hour time steps. However, rainfall records from 2000 to 2002 are mostly erroneous and contain numerous missing data. Therefore, as part of quality control, these years were eliminated from the analysis. On the other hand, the GCMs selected for this research, EC-EARTH and ACCESS 1.0 also contain gaps in simulated historical data. Simulated historical data from EC-EARTH are only available from 1960 to 2012. Historical data availability of ACCESS 1.0 is from 1960 to 2005. Therefore, there were limited overlapping years between observed rainfall data and simulated GCM data. Overlapping data was divided into calibration and validation of the downscaling scheme as presented in Table 6.2. As demonstrated in Table 6.2, data from 10 overlapping years were available for EC-EARTH, while data availability for ACCESS 1.0 is only for four year. Odd years were used for calibration and even years were used for validation. This avoids the bias in the data due to trends of potential La Nina and El Nino effects.

114

Table 6.2: A summary of calibration and validation periods used for the downscaling

Years used for calibration

Years used for validation

EC-EARTH 2003, 2005, 2007, 2009 & 2011

2004, 2006, 2008, 2010 & 2012

ACCESS 1.0 2003 & 2005 2002 & 2004

The calibration outputs were obtained using ParaCal() function of the spdownscale tool for both EC-EARTH and ACCESS 1.0 separately. Figures 6.3 and 6.4 present the calibration outputs for the EC-EARTH and ACCESS 1.0, respectively.

Observed GCM Shape factor 0.0210 0.1056 Scale factor 20.9364 3.0682 F(x=0) 0.8560 0.6790 n 2.4984

Figure 6.3: Calibration parameters for EC-EARTH (Gold Coast Seaway)

0 10 20 30 40 Rainfall (mm)

LEGEND

Model based on Gamma distribution (observed data) Model based on Gamma distribution (GCM data) Actual cumulative probability distribution (observed data) Actual cumulative probability distribution (GCM data)

1.00

0.98

0.96

0.94

0.92

0.90

0.88

0.86

CD

F

115

Observed GCM Shape factor 0.0158 0.0939 Scale factor 29.5511 4.8113 F(x=0) 0.8530 0.7110 n 2.1536

Figure 6.4: Calibration parameters for ACCESS 1.0 (Gold Coast Seaway)

It is evident from Figures 6.3 and 6.4 that the cumulative distribution functions of the rainfall data generated using both EC-EARTH and ACCESS 1.0 data have significant differences from that of the observed data. Particularly, rainfalls corresponding to higher quantiles (above 98%) were significantly underestimated in the GCM simulations compared to the observed data. In contrast, rainfalls corresponding to relatively smaller quantiles (below 95%) were slightly overestimated for both GCMs. On the other hand, the cumulative frequency corresponding to zero rainfall was considerably underestimated in both GCM simulations. The cumulative frequency corresponding to zero rainfall in EC-EARTH was 0.6790 and cumulative frequency corresponding to zero rainfall in ACCESS 1.0 was 0.7110, whereas, the observed value was approximately 0.85. It is also noticeable from Figures 6.3 and 6.4 that the cumulative density function constructed based on the actual data and the modelled cumulative density function using gamma distribution were closely comparable for both GCMs and observed data.

0 10 20 30 40 50 Rainfall (mm)

1.00

0.95

0.90

0.85

CD

F

LEGEND


116

Therefore, it can be argued that the two-parameter gamma distribution very closely replicates the actual probability distribution of the observed and GCM rainfall data.

The validation outputs were obtained by executing the ResVal() function. ResVal() generated the cumulative density functions for both raw and bias-corrected GCM data for the validation period. The percentage error in the threshold was only 2.12% and 0.10% for EC-EARTH and ACCESS, respectively. This validates the calibration parameters used in the models to calculate the CF of the zero rainfall (threshold). On the other hand, Figures 6.5(a) and 6.6(a) present the RMSE for the raw and bias-corrected rainfall data of EC-EARTH and ACCESS 1.0, respectively, for Coastal-SEQ. The raw rainfall outputs from both GCMs have relatively larger RMSEs. The RMSE for raw rainfall outputs from EC-EARTH was 0.34 mm and that from ACCESS 1.0 was 0.27 mm. However, RMSE for the bias-corrected rainfall data were significantly lower for both GCMs. The RMSE for the bias-corrected EC-EARTH and ACCESS 1.0 were 0.05 mm and 0.07 mm, respectively. It also noticeable that bias-corrected data produced rainfall corresponding to higher quantiles more accurately compared to the raw GCM data. ResVal() function also provides a gradient of the observation and GCM data scatter plots. The gradient is expected to be one when the GCM data is equivalent to the observations. Figures 6.5(b) and 6.6(b) present the gradient of observation and GCM data scatter plots for raw and bias-corrected rainfall data of EC-EARTH and ACCESS 1.0, respectively, in both numerical and graphical formats. It can be seen from Figures 6.5(b) and 6.6(b) that the raw rainfall outputs from both GCMs had significantly smaller gradients. The gradient for the raw outputs from EC-EARTH was 0.34 and that from ACCESS 1.0 was 0.33. However, the gradients for bias corrected rainfall data for both GCMs had improved significantly. The gradient for bias corrected EC-EARTH was 0.97 and the gradient for bias corrected ACCESS 1.0 was 0.79. This provided validity for the models and calibration parameters established to correct the rainfall intensities.

Overall, it can be established that the models developed by the spdownscale performed well for both GCMs. The bias-corrected GCM outputs from both GCMs were closely comparable to the observed data. Therefore, the developed models were used to correct the bias in the future GCM data using the downscale() function. As established in Chapter 4, four sets of future rainfall data, which included 2 climate change scenarios (RCP 4.5 and RCP 8.5), each having 2 time-frames (near future- 2026 to 2045 and distant future - 2081 to 2100) were downscaled from both EC-EARTH and ACCESS 1.0. Each set of data comprised of 3-hour rainfall time-series for a 20 year period.

117

Raw GCM data

Bias-corrected GCM data

RMSE 0.3439 0.0496 Gradient 0.3374 0.9743 Percentage error in threshold (%) 2.12

Figure 6.5: Validation results for EC-EARTH (Gold Coast Seaway)

LEGEND

m = 1 (ideal line) Raw GCM data Bias corrected GCM

m=0.9743

m=0.3374

m=1.00

0 10 20 30 40 GCM rainfall (mm)

40

30

10

20

0

Obs

erve

d ra

infa

ll (m

m)

0 10 20 30 40 50 Rainfall (mm)

1.00

0.95

0.90

0.85

CD

F

LEGEND

Observed data Raw GCM data Bias corrected GCM

RMSE=0.0496 mm

RMSE =0.3439 mm

(a)

(b)

118

Raw GCM data



Figure 6.6: Validation results for ACCESS 1.0 (Gold Coast Seaway)


40

30

10

20

0

Obs

erve

d ra

infa

ll (m

m)

LEGEND


m=0.7914

m=0.3293

m=1.00

LEGEND


CD

F

1.00

0.98

0.96

0.94

0.92

0.90

RMSE=0.0654 mm

RMSE=0.2685 mm

0 10 20 30 40 Rainfall (mm)

(a)

(b)

119

6.4.2 Spatial downscaling for Inland-SEQ

Based on the results presented in Chapter 5, Toowoomba Airport station was used as the representative meteorological station for Inland-SEQ. The selection was primarily based on the availability of long record of high-quality rainfall observations. Rainfall records are available for Toowoomba Airport station since 2009 in 3-hour time steps. However, as discussed in the Section 6.4.1, the simulated historical data from EC-EARTH is only available from 1960 to 2012 and historical data availability of ACCESS 1.0 is from 1960 to 2005. Therefore, there were limited overlapping years between observed rainfall data and EC-EARTH. ACCESS had no overlapping years of rainfall records. Hence, EC-EARTH only was used for downscaling future rainfall data at the Toowoomba station. Accordingly, overlapping data for EC-EARTH was subdivided into two sets, for calibration and validation. The data from 2009 and 2011 were used for calibration, and 2010 and 2012 were used for validation of the downscaling scheme.

Similar to the process discussed in the Section 6.4.1, the calibration outputs were obtained using ParaCal() function in the spdownscale tool. Figure 6.7 presents the calibration parameters for EC-EARTH. Accordingly, the cumulative distribution functions of the rainfall data constructed using the raw EC-EARTH had a significant difference from that of the observed data. In particular, rainfall data corresponding to higher quantiles (above 98%) was significantly underestimated in the GCM simulations compared to the observed data. In contrast, rainfall events corresponding to relatively smaller quantiles (below 96%) were slightly overestimated. On the other hand, the cumulative frequency corresponding to zero rainfall was considerably underestimated. The cumulative frequency corresponding to zero rainfall in EC-EARTH was 0.78 while the observed value was approximately 0.90. It is also noticeable from Figure 6.7 that the CDF constructed based on the actual data and the modelled CDF using gamma distribution were almost identical for both GCM and observed data. These observations were in general similar to the calibration outputs of the Coastal-SEQ. ParaCal () function also provided the shape factors, scale factors and n for the observed and GCM data separately as presented in Figure 6.7. These parameters were included in Equations 6.2 and 6.8 to build the specific models required to correct the bias associated with EC-EARTH. The developed models were then validated using independent historical data. The validation outputs were produced using ResVal() function in spdownscale and presented in Figure 6.8. The percentage error in the threshold estimation was only 1.06%, which validates the calibration parameters used in the models to calculate the CF of the zero rainfall (threshold).

120

Observed GCM Shape factor 0.0166 0.0790 Scale factor 18.2353 3.2523 Threshold 0.8998 0.7771 n 2.3890

Figure 6.7: Calibration results for EC-EARTH (Toowoomba Airport)

On the other hand, Figure 6.8(a) presents the RMSE for the raw and bias-corrected rainfall data of EC-EARTH for Inland-SEQ. RMSE for direct rainfall outputs from EC-EARTH was 0.14mm while RMSE for the bias-corrected rainfall output was 0.08mm. Also, bias-corrected data produced rainfall corresponding to higher quantiles more accurately compared to the direct GCM outputs. In addition, Figure 6.8(b) presents the gradient of the observation and GCM data scatter plots for the raw and bias-corrected rainfall data of EC-EARTH for Inland-SEQ in both numerical and graphical formats. It can be seen from Figure 6.8(b) that the direct rainfall outputs from EC-EARTH had significantly smaller gradient compared to the bias-corrected EC-EARTH. The gradient for the direct outputs from EC-EARTH was 0.59 while the gradient for bias-corrected EC-EARTH was 1.25.

0 5 15 20 30 10 25 35 Rainfall (mm)

1.00

0.98

0.96

0.92

0.90

0.94

CD

F

LEGEND


121

Raw GCM data



Figure 6.8: Validation results for EC-EARTH (Toowoomba Airport)


20

15

5

10

0

Obs

erve

d ra

infa

ll (m

m) LEGEND


m=1.2569

m=0.5924

m=1.00

0 10 20 30 40 50 Rainfall (mm)

1.00

0.94

0.90

0.86

0.98

0.96

0.92

0.88 LEGEND


RMSV=0.0765mm

RMSV=0.1435 mm CD

F (a)

(b)

122

These outcomes assured the validity of the models and calibration parameters established at calibration stage. Therefore, the models developed were used to correct the bias in the future GCM data using the downscale() function. As noted in Section 6.4.1, four sets of future rainfall data, which included 2 climate change scenarios (RCP 4.5 and RCP 8.5), each having 2-time frames (near future - 2026 to 2045 and distant future - 2081 to 2100) were downscaled from EC-EARTH.

6.4 Conclusions


• A new spatial downscaling tool, ‘spdownscale’ was developed based on quantile-quantile bias correction approach. The ‘spdownscale’ tool is published and licensed under the General public license, version 2 (GPL 2), where anyone is permitted to copy and distribute verbatim copies of this document (source code), but changes are not allowed. The spdownscale tool is freely available as an R package. Details for installing R packages can be found at R Development Core Team (2016).

• This tool was used to spatially downscale EC-EARTH and ACCESS-1.0 for RCP 4.5 and RCP 8.5 climate change scenarios at the two representative meteorological stations in southeast Queensland.

• Overall, the models developed by the spdownscale for spatial downscaling performed well for both GCMs. Two statistical indexes, RMSE and the gradient of the observation-simulation line were used for validation purposes. The bias-corrected GCM outputs from the both GCMs were closely comparable to the observed data. However, bias-corrected EC-EARTH performed better compared to the ACCESS 1.0.

123

Chapter 7 Temporal Downscaling of Rainfall Data Using First-order Markov Model

7.1 Background

Temporal downscaling models, typically based on stochastic weather generation techniques (Wilks, 1999) are broadly used in assessing the potential implications of climate change in terms of water resources at national or regional scales (Prudhomme et al., 2002). Nevertheless, these studies generally involved temporal downscaling models operating at monthly to daily temporal resolutions. However, as noted in Chapter 2, continuous rainfall simulation at sub-daily to sub-hourly time steps is a necessity for small catchment scale studies such as Water Sensitive Urban Design (WSUD) related studies. Nonetheless, there is no established methodology available to temporally downscale rainfall data at such fine resolutions. This is primarily due to fact that there is no predictor variable available from Global Circulation Models at such fine temporal resolution to construct the statistical models.

Therefore, developing a temporal downscaling model to downscale the rainfall data to sub-hourly temporal resolution was essential for this study. This chapter primarily discusses the development of the temporal downscaling tool capable of downscaling 3-hour spatially downscaled rainfall time-series to 5-minute time-series. The tool was designed to operate using the outputs of spatial downscaling tool discussed in Chapter 6. The tool was developed based on Homogeneous Markov Model (HMM) due to its unique capability of operating without predictor variables. The data generated using the temporal downscaling tool played a critical role in the impact assessment as presented in Chapter 8 and Chapter 9.

Accordingly, this chapter presents detailed discussions on the development of the temporal downscaling models including the basic assumptions used in the model and

124

architecture of the model. This chapter also presents discussions on the downscaled rainfall data for SEQ with their key attributes.

7.2 First-order homogeneous Markov model

As discussed in Chapter 4, among different temporal downscaling methods, Markov models are popularly used and proven in downscaling rainfall data (Richardson, 1981, Hughes et al., 1999; Charles et al., 1999; Mehrotra and Sharma, 2016; Zucchini and Guttorp, 1991). Markov Models can be classified into two, namely, Homogeneous Markov Model (HMM) and Non-homogeneous Markov Model (NHMM). HMM operates on the basis that the state of weather at a given time depends only on the state of the weather at the previous time, whereas NHMM operates on the basis that the state of weather at a given time depends not only on the state of the weather at the previous time but also on the current states of predictor variables supplied by the GCMs (Hughes et al., 1999). Therefore, NHMM can be used to downscale rainfall data typically up to daily temporal resolutions due to the absence of GCM predictor variables at sub-daily temporal resolutions. Therefore, in this research, HMM was used to develop a temporal downscaling model. A detail discussion on the selection of temporal downscaling methods is presented in Section 4.4.3. Accordingly, a first-order homogeneous weather generating model was developed to translate the bias-corrected 3-hour rainfall time-series to 5-minute time-series.

7.2.2 Assumptions used in the model

The first-order homogeneous weather generating model was developed to translate the bias-corrected 3-hour rainfall time-series to 5-minute time-series. The model was developed based on two primary assumptions as follows:

Assumption 1: The first-order dependency

It was assumed that the weather states are first-order dependent, which means that every weather state is depended only on the previous weather state. The first-order dependency is used in many studies relating to stochastic rainfall generation (Gabriel and Neumann, 1962; Yoo et al., 2016; Feyerherm and Bark, 1967). For example, rainfall depth in a particular time step is assumed to be dependent only on the rainfall depth in the previous time step (Hughes et al., 1999; Charles et al., 1999; Wilks, 1999). This can be mathematically expressed in the form of Equation 7.1.

𝛼𝛼(𝑥𝑥𝑘𝑘|𝑥𝑥𝑘𝑘−1, 𝑥𝑥𝑘𝑘−2,… , 𝑥𝑥0) = 𝛼𝛼(𝑥𝑥𝑘𝑘|𝑥𝑥𝑘𝑘−1) (7.1)

Where, x=(x0,x1,x2,…,xt,…,xn) refers to the rainfall sequence; n refers to the total time steps of the data; 𝛼𝛼(𝐶𝐶|𝐵𝐵) refers to the probability of event A occurring given the previous event is B; and xt refers to the rainfall at time t.

125

Assumption 2: The transition probabilities remain constant with time

Transition probability refers to the probability of occurrence of a particular event (for example, 𝛼𝛼𝑗𝑗), given that a particular event (for example,𝛼𝛼𝑖𝑖) has occurred in the previous

time step. The probabilities for all possible transitions are determined based on long historical rainfall records. These transition probabilities are assumed to be constant and independent of time (Hughes et al., 1999; Charles et al., 1999). This can be mathematically expressed in the form of Equation 7.2.

𝛼𝛼�𝑥𝑥𝑘𝑘 = 𝛼𝛼𝑗𝑗�𝑥𝑥𝑘𝑘−1 = 𝛼𝛼𝑖𝑖� = 𝛼𝛼�𝑥𝑥(𝑘𝑘+𝜏𝜏) = 𝛼𝛼𝑗𝑗�𝑥𝑥(𝑘𝑘+𝜏𝜏)−1 = 𝛼𝛼𝑖𝑖�

𝛼𝛼𝑖𝑖𝑗𝑗𝑘𝑘 = 𝛼𝛼𝑖𝑖𝑗𝑗

𝑘𝑘+𝜏𝜏 (7.2)

Where, S=(s1,s2,s3,…,si,…,sm) refers to the discrete weather states (all possible rainfall recordings of the station) and 𝑥𝑥 ∈ 𝑆𝑆.

7.2.3 Architecture of the model

Based on the assumptions discussed in Section 7.2.2, the temporal downscaling model was developed in two steps as shown in Figure 7.1. The first step involved the development of the Transition Probability Matrix (TPM) based on the historical rainfall dataset. The second step involved the generation of a 5-minute rainfall time-series for any given 3-hour period based on the developed TPM. A detailed discussion explaining these two steps are provided below:

Step 1: Developing the Transition Probability Matrix

In the first step in temporal downscaling, the model was designed to identify all possible discrete weather states from the 5-minute rainfall time-series of observed data as illustrated in Figure 7.1. The weather states, s, refer to all discrete weather recordings from the observed data used to develop the TPM. The model was designed to determine the conditional probability of occurrence of every weather state with different previous weather states. These probabilities are referred to as the transition probabilities. The model then develops a matrix that contains all cumulative transition probabilities for all weather states and it is referred to as the Transition Probability Matrix.

126

Figure 7.1: Flow diagram of the model process

5 min rainfall time-series

TPM

S1 S2 S3 . . . . . . . . . . . . . . . . . . . . . .

S1 S2

S3 . . . . . . . Sm

Initi

al s

tate

Current state

………...............................................

Step 1: Developing the Transition Probability Matrix

Step 2: Generates 5-minute time-series of rainfall based on the developed TPM

TPM

………...............................................

3 hr rainfall time-sereis

Random number generation

R= [ri, r2, r3 ………. r36]

3 hr rainfall

Xobs

5 min rainfall series

Xsim= [xi, x2, x3 …. x36] FALSE

5 min rainfall series

Xsim= [xi, x2, x3 ………. x36]

∑ 𝑋𝑋𝛼𝛼𝑑𝑑𝑟𝑟 = Xobs ± 0.1 Xobs

TRUE

127

Step 2: Generates 5-minute time-series of rainfall based on the developed TPM

In the second step, the model was designed to transform the 3-hour rainfall time-series to 5-minute time-series based on the developed TPM. For this, the model first generates a random number between 0 and 1 and an initial weather state. The initial weather state is set to be zero in this model and the random number is considered as the cumulative transition probability for the first time step of the 5-minute time-series. Based on the initial state and the cumulative transition probability, a random draw for rainfall in the second time step is chosen using the TPM established in Step 1. The chosen rainfall for the second time step becomes the current state of weather for the next time step, and with a new random number for the cumulative transition probability, the process continues for 36 time steps (3-hours consists of 36, 5-minute intervals). Once the 36 iterations are performed, the model checks whether the total of the generated rainfall time-series is equal to the actual 3-hour rainfall (with a tolerance of 10% error). If not, the model reproduces the process until it satisfies this criterion as illustrated in Figure 7.1. A step-by-step example of these processes is explained in Appendix C1 and the source code for the model is provided in Appendix C2. In addition, the model was designed to produce 10 different realizations of 5-min time-series for any given 3-hour time-series.

7.3 Temporal downscaling for southeast Queensland

The developed temporal downscaling model was used to translate 3-hour rainfall time-series obtained from the spatial downscaling tool as discussed in Chapter 6 into 5-minute time-series. The processes involved in this task were in three stages, namely, calibration, validation and future rainfall data generation. Calibration of the model primarily involves the development of the TPM based on historical rainfall records at the selected rainfall stations in SEQ. The validation process involved verifying the performance of the developed TPM by simulating a rainfall time-series for an independent historical period and comparing those with the observed rainfall data. The comparison was primarily to investigate whether the simulations captured the overall probabilistic distributions of the rainfall time-series. In addition, the simulated temporal patterns of the rainfall time-series and maximum rainfall intensities were compared with that of the observations. The 3-hour total rainfall depth was also checked in the validation process as a further check. The future data generation involved generating 5-minute rainfall time-series using the developed TPM and spatially downscaled (bias corrected) GCM outputs in 3-hour time-steps. For every future scenario, 10 different realizations were generated to cover potentially different temporal patterns of the same total rainfall. This provides confidence in the impact assessments performed based on these rainfall data.

128

7.3.1 Calibration

The calibration primarily involved developing the TPM based on historical rainfall records. Based on the comprehensive study on the selection of the meteorological station presented in Chapter 5, two selected representative meteorological stations, namely, Gold Coast Seaway and Toowoomba Airport stations were used to represent the Coastal-SEQ and the Inland-SEQ, respectively. The selection was primarily based on the availability of high-quality rainfall over a long time period. Similar to the processes adopted in Chapter 6, the data from these stations were divided into two, for calibration and validation purposes and the summary of the data used are presented in Table 7.1. Two separate TPMs were developed based on the calibration datasets of the respective meteorological stations.

Table 7.1: A summary of calibration and validation periods used for the temporal downscaling

Years used for calibration

Years used for validation

Gold Coast Seaway 2003, 2005, 2007, 2009, 2011, 2013 & 2015

2004, 2006, 2008, 2010, 2012 & 2014

Toowoomba Airport 2009, 2011, 2013 & 2015

2010, 2012 & 2014

7.3.2 Validation

The performances of the models were validated for both stations using independent historical rainfall data. For this, a historical rainfall time-series in 3-hour time-steps was used as the input to the model so that it generates a time-series in 5-minute time-steps for the validation period. The results from the simulations were then compared against the observed 5-minute rainfall data for the same validation period. Figure 7.2 shows the validation outputs in the form of cumulative probability distributions for simulated 5-minute rainfall time-series and the observed 5-minute rainfall for the Gold Coast Seaway station (Figure 7.2 (a)) and the Toowoomba Airport station (Figure 7.2 (b)). As shown in Figure 7.2, the simulated rainfalls were in agreement with the observed rainfall in terms of the probability distribution for both stations. The agreement was also estimated using the Root Mean Square Error (RMSE) of the simulated rainfall. Accordingly, The RMSE for Gold Coast Seaway station was 0.67 mm and for Toowoomba Airport station was 0.59 mm.

129

Figure 7.2: Validation outputs of the cumulative probability distribution

Rainfall (mm)

Rainfall (mm)

CD

F

CD

F a. Gold Coast Seaways

b. Toowoomba Airport

Simulated data

Observed data

Simulated data

Observed data

130

The temporal patterns of the simulated rainfall were also compared against the observed rainfall patterns. In this research, for each simulation, 10 different realizations were generated to avoid the uncertainties associated with the stochastic process of the model. Each of these realizations results in different temporal patterns for any given 3-hour rainfall total. The temporal patterns are essentially consistent with historical rainfall patterns (technically, Transition Probability Matrix) and thereby produce 10 different most likely sequence of rainfall. The use of these temporal patterns covers different hydrological outcomes and thereby produces more plausible results. A similar approach has been adopted in a number of studies, including Herron et al. (2010) and AR&R (2015). Two examples of simulated temporal patterns against the observed temporal patterns are presented in Figure 7.3 for explanation purposes. Figure 7.3(a) and Figure 7.3(b) are ensembles (collection of different realizations) of simulated rainfall patterns for 3-hour total rainfall of 8.4 mm and 30.2 mm, respectively. It can be seen from Figure 7.3 that the 10 temporal patterns of the ensembles were different from each other. However, in general, there was at least one temporal pattern that closely represents the observed temporal patterns. For example, the 2nd and the 8th realizations were similar to the observation in Figure 7.3(a) and the 3rd and 6th realizations were close to the observation in Figure 7.3(b). It was also noticeable that the rainfall distribution over time was less intense in the observation compared to most of the simulated temporal patterns. Consequently, the peak rainfalls were also overestimated in most of the simulated temporal patterns. However, the maximum intensities (maximum rainfall recorded during any given 1-hour duration) of the simulated rainfall were reasonably similar with the observed data at both stations as presented in Figure 7.4.

131

0

0.5

1

1.5

2

2.5

3

3.5

4

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35

0

0.5

1

1.5

2

2.5

3

3.5

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35

0

0.5

1

1.5

2

2.5

3

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35

0

0.5

1

1.5

2

2.5

3

3.5

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35

0

0.5

1

1.5

2

2.5

3

3.5

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 350

0.5

1

1.5

2

2.5

3

3.5

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35

0

0.5

1

1.5

2

2.5

3

3.5

4

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35

00.5

11.5

22.5

33.5

44.5

5

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35

00.5

11.5

22.5

33.5

44.5

5

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35

0

0.5

1

1.5

2

2.5

3

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35

Date: 08-05-2004 Time: 09:00-12:00 Station: Gold Cost Seaway 3-hour total: 8.4 mm

Observation

Realization 1 Realization 1 Realization 2

Realization 3

Realization 5

Realization 4

Realization 6

Realization 7 Realization 8


(a)

132

0

1

2

3

4

5

6

7

8

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35

0

1

2

3

4

5

6

7

8

9

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 350123456789

10

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35

0

2

4

6

8

10

12

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35

0

2

4

6

8

10

12

14

16

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35

0

2

4

6

8

10

12

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 350

2

4

6

8

10

12

14

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35

0

1

2

3

4

5

6

7

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35

0123456789

10

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35

0

0.5

1

1.5

2

2.5

3

3.5

4

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35

0

2

4

6

8

10

12

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35

Figure 7.3: Temporal patterns for rainfall data over 3-hour periods

Date: 29-01-2010 Time: 03:00-06:00 Station: Gold Cost Seaway 3-hour total: 30.2 mm

Observation


Realization 3

Realization 5

Realization 4

Realization 6



(b)

133

Figure 7.4: Validation outputs of maximum rainfall intensities. (a) Gold Coast Seaway station and (b) Toowoomba Airport station

Observed

Observed

Sim

ulat

ed

Sim

ulat

ed

(a)

(b)

134

On the other hand, 3-hour total of the simulated rainfalls were expected to be consistent with the observed data as it is a criterion used in the model to simulate the rainfall. However, the model allows 10% error in the 3-hour total rainfall in the simulations. Figure 7.5 presents the scatter plots of the simulated and observed 3-hour total rainfall. It can be seen that the simulated 3-hour total rainfall were closely comparable to the observed 3-hour total rainfalls at both stations.

Figure 7.5: Validation outputs of 3-hour total rainfall. (a) Gold Coast Seaway station and (b) Toowoomba Airport station

Observed

Observed

Sim

ulat

ed

Sim

ulat

ed

(a)

(b)

135

Overall, the model was able to reproduce the overall probabilistic distribution of the rainfall time-series, temporal patterns of the rainfall time-series, and maximum rainfall intensities accurately for both meteorological stations. Therefore, the model was used to generate 5-minute time-series for future climate change scenarios using the bias-corrected rainfall data discussed in Chapter 6.

7.3.3 Future rainfall generation

The developed model was used to temporally downscale the spatially downscaled (bias-corrected) future rainfall data (results of Chapter 6). Bias-corrected 3-hour rainfall time-series from two GCMs, EC-EARTH and ACCESS 1.0 were used as inputs to the model. Each GCM consisted rainfall data for two future periods. These periods include rainfall data for a near future (2026-2041) and a distant future (2081-2100). Data from two climate change scenarios, RCP 4.5 and RCP 8.5, were used for downscaling. For each of the future dataset, 10 different realizations were generated using the model. Characteristics of the downscaled rainfall and the impact assessments undertaken using the generated future rainfall data are discussed in Chapter 8 and Chapter 9.

7.4 Conclusions


• A new temporal downscaling model was developed based on the first-order homogeneous Markov model. The model is used to translate the 3-hour rainfall time-series into 5-minute time-series.

• The model was used for temporal downscaling at the two representative meteorological stations. The model performance was then assessed based on independent historical rainfall data. For both stations, the simulated rainfall was in agreement with the observed rainfall in terms of the probability distribution. 10 realizations (temporal pattern) were generated for every simulation. Although the temporal patterns were different to each other, there was at least one temporal pattern that closely represented the observed temporal patterns. Further, the maximum rainfall intensities of the simulated rainfall were reasonably similar to the observed data at both stations.

137

Chapter 8 Design Rainfall for Future Climate Change Scenarios

8.1 Background

The WSUD treatment systems such as bioretention basins and constructed wetlands are common features in urban developments in Australia (Cottus et al., 2012; Mangangka, 2013). Various factors have to be taken into account when designing and implementing WSUD treatment systems. Most critical factors include design rainfall and design flows; landscaping of the areas; urban form; targeted pollutants for removal; treatment sequence; components of the WSUD system such as inlets, outlets and by-pass structures; and vegetation (BCC & MBW, 2006). Among them, most of the factors are established and known at the planning stages of the WSUD developments. The most critical of them and the factor that makes WSUDs location specific is design rainfall. Moreover, the rainfall characteristics determine the treatment performance and hydraulic performance of WSUD systems. The design of the WSUD systems is typically based on the required percentage of pollutant reduction capacity derived from standard treatment performance curves provided by Model for Urban Stormwater Improvement Conceptualisation (MUSIC). The pollutant treatment typically targets very frequent events. For example, most treatment systems are focused on treating rainfall events less than 1-year Average Recurrence Interval (ARI). On the other hand, the design of WSUD systems is also underpinned by hydraulic criteria. All systems are required to safely convey the stormwater runoff without damaging the systems and the surrounding developments. The hydraulic design is typically based on frequent rainfall. Bioretention basins for example, typically focus on 2, 5 or 10-year ARIs. Accordingly, both treatment design and hydraulic design of the WSUD system are primarily governed by the design rainfall characteristics and thus, design rainfall primarily decides the size and hydraulic characteristics of the system (BCC & MBW, 2006). However, design rainfall is estimated based on static climate assumptions. Climate change and its impact on rainfall patterns including changes to the Intensity-Frequency-Duration (IFD) are generally not considered in the design of the WSUD systems.

138

This chapter primarily focuses on developing IFD relationships for different future climate change scenarios. At-site frequency analyses were performed to develop the IFD relationships for southeast Queensland based on the future rainfall data developed in Chapter 6 and 7. Further, the changes to the future and the present IFDs are discussed in detail in this chapter. In addition, a methodology was developed to incorporate the changes in determining the design rainfall, which is primarily applicable for the design of WSUD systems.

Accordingly, this chapter presents detailed discussions on rainfall frequency analysis, development of IFDs for southeast Queensland for present and future climate change scenarios and the approach for incorporating the changes in future IFDs in the design of WSUD systems.

8.2 Rainfall frequency analysis

Frequency analysis is a statistical approach for estimating the frequency of occurrence of a specified event. Frequency analysis estimates the probability of occurrence of an event with given magnitude and duration, and thereby develops Intensity-Frequency-Duration relationships for rainfall events for a given region (or a meteorological station) of interest (Hosking and Wallis, 1997; Fowler and Kilsby, 2003; Trefry et al., 2005; Raes et al., 2004). Rainfall frequency analysis can be classified into two, namely, regional frequency analysis and at-site frequency analysis (Hansen, 2015; Goudenhoofdt et al., 2017). Regional frequency analysis essentially considers observations from a group of meteorological stations to estimate the IFDs. The use of observations from a group of neighbouring stations eliminates the uncertainties associated with stations that have observations for a short period of time and accordingly, produces more accurate IFD estimations for large regions (Hosking and Wallis, 2005; Hansen, 2015). However, this approach is limited to regions with an extensive network of meteorological stations with long-term observed data (Hassan and Ping, 2012). In contrast, at-site frequency analysis estimates IFDs based on rainfall records from a single meteorological station. Moreover, at-site frequency analysis is simple and straightforward compared to regional frequency analysis. Therefore, in this research, at-site frequency analyses were performed for the selected representative meteorological stations as the data were only available at those stations for the future climate change scenarios. The complete approach for developing IFDs using at-site frequency analysis is presented in Section 4.4.4, in detail.

Accordingly, at-site rainfall frequency analyses were undertaken for the selected representative meteorological stations, namely, Gold Coast Seaway (-27.9390, 153.4283) and Toowoomba Airport (-27.5425, 151.9134). The Gold Coast Seaway station represented the Coastal-SEQ while the Toowoomba Airport station represented the

139

Inland-SEQ regions. Both stations consisted of 5-minute time-series of rainfall data for historic and future climate change scenarios. As presented in Chapter 6 and Chapter 7, the future climate change scenarios included data for RCP 4.5 and RCP 8.5 for 2026-2045 (near future) and 2081-2100 (distant future) timeframes. Each dataset consisted of 10 realizations in order to account for the uncertainties associated with the stochastic process in the downscaling models.

Annual maximum rainfall series of 5-minute, 10-minute, 15-minute, 30-minute, 1-hour, 2-hour, 3-hour, 6-hour, 12-hour and 24-hour durations were extracted from the representative meteorological stations. These durations are the standard set of durations used by Bureau of Meteorology (AR&R, 2015). The annual maximum series was then separately (for each duration) modelled assuming Extreme Value Type-I (EV-I) (also known as Gumbel distribution) probability distributions. EV-I distribution is the most suitable probability distribution to model the annual maximum rainfall series, used across various projects including AR&R (2015), Green et al. (2012), Khan et al. (2017) and Hosking and Wallis (1997). ARIs considered were 1, 2, 5, 10, 20, 50 and 100 year and the corresponding frequency factors for the EV-I distribution were calculated for each ARI using Equation 8.1 as discussed in Chapter 4.

𝑘𝑘𝑇𝑇 = −√

6𝜋𝜋

�0.5772 + ln �ln� 𝑇𝑇𝑇𝑇 − 1

�� (8.1)

Where, 𝑘𝑘𝑇𝑇 and 𝑇𝑇 refer to the frequency factor and return period respectively.

Table 8.1 tabulates the frequency factors calculated for the corresponding return periods. It also provides the corresponding Average Recurrence Interval (ARI) and Annual Exceedance Probability (AEPs). Technically, ARI refers to the expected value of periods between exceedance of a given rainfall event and AEP refers to the probability of a given rainfall to occur (over a given duration) in any given year (AR&R, 2015).

After frequency factors were established, the magnitudes of the rainfall for the corresponding return periods were calculated using Equation 8.2. The magnitude of rainfall events for every ARI was determined for all considered rainfall duration to construct the IFDs for the stations.

𝑥𝑥𝑇𝑇 = 𝜇𝜇 + 𝑘𝑘𝑇𝑇 𝜎𝜎 (8.2)

Where, 𝜇𝜇 and 𝜎𝜎 mean and the standard deviation of the annual maximum series.

140

Table 8.1: Return periods and the corresponding frequency factors for EV-I distribution

ARI (years)

AEP (%)

Return periods AEP

(1 in T) kT

1 63.2 1.58 -0.4517

2 39.35 2.54 0.0889

5 18.13 5.52 0.7195

10 10 10 1.3046

20 5 20 1.8658

50 2 50 2.5923

100 1 100 3.1367

The IFDs were generated based on the historical rainfall records and future climate change scenarios separately for the two representative meteorological stations. The historical data was used to develop the present IFDs and the IFDs for the future climate change scenarios were developed based on the downscaled rainfall data as discussed in Chapter 6 and 7. The ratios between the IFDs of the future and the present, termed as the change factors, Cf, were calculated for all durations and return periods separately using Equation 8.3. The change factor is a novel term defined in this research to translate the currently available IFDs to the future climate change scenarios.

𝐶𝐶𝑓𝑓(𝑌𝑌 ,𝐷𝐷) =

𝐶𝐶𝑓𝑓𝑓𝑓𝑘𝑘𝑓𝑓𝑝𝑝𝑝𝑝(𝑌𝑌 ,𝐷𝐷)

𝐶𝐶ℎ𝑖𝑖𝑠𝑠𝑘𝑘𝑜𝑜𝑝𝑝𝑖𝑖(𝑌𝑌 ,𝐷𝐷) (8.3)

Where, 𝐶𝐶𝑓𝑓(𝑌𝑌 ,𝐷𝐷) is the change factor for a return period Y and duration D and I is the

rainfall intensities.

Accordingly, the IFDs for the future climate change scenarios are proposed to be estimated by multiplying the respective change factors with the current IFDs provided by the Bureau of Meteorology (based on the historical data). This approach is consistent with the interim climate change guidelines in the Australian Rainfall & Runoff (AR&R, 2015).

8.2.1 IFD generation for the historical data

IFDs for the two representative meteorological stations, Gold Coast Seaway station and Toowoomba Airport station were developed based on the at-site frequency analysis using historical rainfall data and compared against the IFDs provided by BoM. The IFDs generated (using at-site frequency analysis) for the historical time-period for the two representative meteorological stations are given in Table 8.2 and Table 8.3.

141

Table 8.2: IFDs generated using at-site frequency analysis for station 40764

ARI (years) 1 2 5 10 20 50 100

5-minute 8.30 10.6 15.6 18.9 22.0 26.1 29.1 10-minute 14.0 15.9 20.0 22.7 25.3 28.6 31.2 15-minute 17.7 19.3 22.6 24.8 27.0 29.7 31.8 30-minute 25.1 27.5 32.7 36.2 39.4 43.7 46.9 1-hour 33.9 39.3 50.7 58.2 65.5 74.8 81.8 2-hour 42.7 53.5 76.2 91.2 106 124 138 3-hour 48.5 64.5 98.2 121 142 170 191 6-hour 63.5 86.8 136 168 200 240 270 12-hour 73.8 113 195 250 302 370 421 24-hour 97.3 136 218 273 325 392 443

Note: All values given in mm.

Table 8.3: IFDs generated using at-site frequency analysis for station 41529

ARI (years) 1 2 5 10 20 50 100

5-minute 8.25 8.96 10.5 11.4 12.4 13.6 14.5 10-minute 12.0 13.5 16.9 19.1 21.2 23.9 26.0 15-minute 16.2 18.4 23.0 26.0 29.0 32.8 35.6 30-minute 21.5 25.0 32.6 37.5 42.3 48.5 53.1 1-hour 24.2 31.0 45.3 54.7 63.8 75.5 84.3 2-hour 28.9 37.6 56.1 68.4 80.1 95.4 107 3-hour 33.0 40.7 57.0 71.2 82.1 98.5 109 6-hour 40.3 51.0 73.4 88.3 103 121 135 12-hour 55.4 69.1 98.0 117 135 159 177 24-hour 64.6 86.7 133 164 194 232 261

Note: All values given in mm.

142

The generated IFDs were then compared against the IFDs provided by BoM. Figure 8.1 and Figure 8.2 present the IFD curves generated using at-site frequency analysis developed by this research against the curves provided by BoM for Gold Coast Seaway station and Toowoomba Airport station, respectively.

(a)

Figure 8.1: IFD relationship curves for the Gold Coast Seaway station. The broken lines show the curves generated using the at-site frequency analysis and the solid lines are that from BoM

DEP

TH

/ (

mm

)

1000

500 400

300

200

100

50

10

5

0

Minutes Hours

DURATION

1 year ARI 2 year ARI 5 year ARI 10 year ARI 20 year ARI 50 year ARI 100 year ARI

5 10 15 30 1 2 3 6 12 24

LEGEND

143

Figure 8.2: IFD relationship curves for the Toowoomba Airport station. The broken lines show the curves generated using the at-site frequency analysis and the solid lines are that from BoM

As presented in Figure 8.1 and Figure 8.2, the IFD curves generated from the at-site frequency analysis were overall consistent with the IFDs provided by BoM at both stations. However, the at-site frequency analysis showed a trend of slight under-estimation for short-duration rainfall events and slight over-estimation for the longer-duration rainfall events compared to BoM IFDs at both stations. Further, the discrepancy between the IFDs of these approaches was higher for large return periods. A similar trend was observed in a study conducted by Hansen (2015) in southern

DEP

TH

/ (

mm

)

1000

500 400

300

200

100

50

10

5

0

Minutes Hours

DURATION


5 10 15 30 1 2 3 6 12 24

LEGEND

(b)

144

Alberta, Canada. Hansen (2015) suggested that the conventional at-site frequency analysis tends to under-estimate the IFDs compared to regional frequency analysis, especially for shorter durational rare events. This may be due to the nature of the at-site frequency analysis and regional frequency analysis. At-site frequency analysis only considers the data from a specific meteorological station whereas regional frequency analysis considers a collective use of the statistics of the recorded rainfall data at different meteorological stations with approximations to cover a large area (Hansen, 2015; AR&R, 2015; Hosking and Wallis, 1997).

In the context of estimating the change factors of the future IFDs against the historical periods, the approach to the frequency analysis does not play an influential role. However, it is important to maintain a consistent approach to the development of both, historic and future IFDs when generating data for change factor calculations.

8.2.2 IFD generation for future climate change scenarios

Two climate change scenarios, RCP 4.5 and RCP 8.5 from EC-EARTH were used to develop IFDs for future scenarios. Both RCPs consisted of rainfall data for two timeframes, specifically 2026-2045 (near future) and 2081-2100 (distant future). In this way, there were 4 sets of data per meteorological station available for the analysis. Each of these 4 sets of data comprised of 10 realizations.

The results of the analysis are presented in Figure 8.3 to Figure 8.10. Overall, it can be seen that the IFDs generated for the future climate change scenarios at both stations had similar shapes in comparison to the present IFDs. The shorter duration rainfalls (5-60 minute) and the long duration rainfalls (6-24 hour) show a sharp increasing trend compared to that of the mid-range durations (1-3 hour). It was also notable that the variations between the 10 realizations of each ARI were insignificant. However, in general, mid-range (1-3 hour) duration rainfalls of the 20 to 50-year ARIs show higher variation between realizations. The 95% confidence intervals for the IFDs for the future climate change scenarios were calculated and presented in Appendix D.

145

Figure 8.3: IFD curves for Gold Coast Seaway station (40764) for RCP 4.5 climate change scenario for the period 2026-2045. The broken-lines denote the values of realizations (10 for each ARI) and solid lines denote the mean of the realizations

Minutes Hours

DURATION

5 10 15 30 1 2 3 6 12 24

DEP

TH

/ (

mm

)

1000

500 400

300

200

100

50

10

5

1


LEGEND

146


Minutes Hours

DURATION

5 10 15 30 1 2 3 6 12 24

DEP

TH

/ (

mm

)

1000

500 400

300

200

100

50

10

5

1


LEGEND

147


Minutes Hours

DURATION

5 10 15 30 1 2 3 6 12 24

DEP

TH

/ (

mm

)

1000

500 400

300

200

100

50

10

5

1


LEGEND

148


Minutes Hours

DURATION

5 10 15 30 1 2 3 6 12 24

DEP

TH

/ (

mm

)

1000

500 400

300

200

100

50

10

5

1


LEGEND

149

Figure 8.7: IFD curves for Toowoomba Airport (41529) for RCP 4.5 climate change scenario for the period 2026-2045. The broken-lines denote the values of realizations (10 for each ARI) and solid lines denote the mean of the realizations

Minutes Hours

DURATION

5 10 15 30 1 2 3 6 12 24

DEP

TH

/ (

mm

)

1000

500 400

300

200

100

50

10

5

1


LEGEND

150


Minutes Hours

DURATION

5 10 15 30 1 2 3 6 12 24

DEP

TH

/ (

mm

)

1000

500 400

300

200

100

50

10

5

1


LEGEND

151


Minutes Hours

DURATION

5 10 15 30 1 2 3 6 12 24

DEP

TH

/ (

mm

)

1000

500 400

300

200

100

50

10

5

1


LEGEND

152


Minutes Hours

DURATION

5 10 15 30 1 2 3 6 12 24

DEP

TH

/ (

mm

)

1000

500 400

300

200

100

50

10

5

1


LEGEND

153

Table 8.4 and Table 8.5 present the estimated change factors for the IFDs for Coastal-SEQ for RCP 4.5 climate change scenarios. As shown in Table 8.4 and Table 8.5, the change factors varied for different return periods and for different durations of rainfall events. For the near future (2026-2045), the change factors varied between 0.84 and 2.10 while for the distant future (2081-2100), the change factors vary between 0.84 and 2.31. In general, the IFDs of the distant future are expected to be slightly higher than the IFDs of near future for RCP 4.5 climate change scenarios.

The estimated change factors for RCP 8.5 climate change scenarios for Coastal-SEQ is presented in Table 8.6 and Table 8.7. Similar to the observation for RCP 4.5 scenario, the estimated changes in the IFDs vary across the return periods and the duration of the rainfall. The change factors were between 0.81 and 1.99 for near future and between 0.82 and 2.27 for distant future. It was also noticeable that the IFDs projected for the RCP 4.5 climate change scenarios were slightly higher than IFDs projected for RCP 8.5 climate change scenario.

Table 8.8 and Table 8.9 present the estimated change factors for the IFDs for Inland-SEQ for RCP 4.5 climate change scenarios. The change factors were estimated between 0.77 and 1.60 for near future and between 0.88 and 2.03 for distant future. The IFDs for the distant future were expected to be considerably higher than the IFDs for near future, especially for long duration rainfall.

Change factors for IFDs for Inland-SEQ for RCP 8.5 are provided in Table 8.10 and Table 8.11. The change factors varied between 0.85 and 1.61 for near future and between 0.90 and 2.30 for distant future. Consistent with the observation for Coastal-SEQ, the IFDs projected for the RCP 4.5 climate change scenarios were slightly higher than IFDs projected for RCP 8.5 climate change scenario.

154

Table 8.4: Change factors for IFDs for Coastal-SEQ for RCP 4.5 (2026-2045)

ARI (years) 1 2 5 10 20 50 100

5-minute 1.71 1.44 1.14 1.03 0.95 0.88 0.84 10-minute 1.53 1.48 1.41 1.38 1.35 1.33 1.31 15-minute 1.54 1.61 1.72 1.78 1.83 1.88 1.92 30-minute 1.36 1.50 1.73 1.85 1.94 2.04 2.10 1-hour 1.10 1.20 1.34 1.40 1.44 1.49 1.52 2-hour 1.00 1.03 1.06 1.08 1.09 1.10 1.10 3-hour 0.96 0.92 0.87 0.86 0.85 0.85 0.84 6-hour 1.14 1.08 1.02 1.00 0.99 0.97 0.97 12-hour 1.42 1.28 1.16 1.13 1.11 1.09 1.08 24-hour 1.48 1.45 1.41 1.40 1.39 1.38 1.38


ARI (years) 1 2 5 10 20 50 100



ARI (years) 1 2 5 10 20 50 100


155


ARI (years) 1 2 5 10 20 50 100


Table 8.8: Change factors for IFDs for Inland-SEQ for RCP 4.5 (2026-2045)

ARI (years) 1 2 5 10 20 50 100



ARI (years) 1 2 5 10 20 50 100


156


ARI (years) 1 2 5 10 20 50 100



ARI (years) 1 2 5 10 20 50 100


Overall, there was a significant increase in the IFDs for the future climate change scenarios compared to the IFD provided by BoM. Also, the variations in the change factors across the ARI and durations were high. In general, frequent shorter duration rainfall events and the infrequent longer duration rainfall events were expected to increase significantly for both climate change scenarios. On average, the IFDs for the Coastal-SEQ were expected to increase by 23-30% for the near future and 38-45% for the distant future. The IFDs for the Inland-SEQ were expected to increase by 5-15% for the near future and 37-38% for the distant future (the percentage covers both RCPs). These results suggest that the interim climate change factors suggested by the revised AR&R project (AR&R, 2015) could be underestimated. A comparison of the percentages increase in the IFDs suggested by this research and interim climate change guideline of the AR&R project (AR&R, 2015) is presented in Table 8.12.

157

Table 8.12: Comparison of the percentages increase in the IFDs suggested by this research and interim climate change guideline of the AR&R for SEQ


Near future

Distance future

Near future

Distance future

Results from this research

23-30% 38-45% 5-15% 37-38%

AR&R guidelines 4.5-8.8% 8.9-18.6% 4.9-9.8% 9.9-20.8%

The values provided by AR&R (AR&R, 2015) was based on a ‘broad brush’ approach due to the lack of accurate regional future rainfall data developed based on the best science available. AR&R guideline provides adjustment factors for IFD curves informed primarily by the temperature projections for the climate change scenarios, suggesting a 5% increase in rainfall intensity per 1oC of local warming (AR&R, 2015). However, various studies suggested that the future rainfall is not only a function of the temperature indices, but also influenced by a set of other independent climate variables such as mean sea level pressure, minimum and maximum temperature, humidity and zonal and meridional wind velocities (Timbal and McAvaney, 2001 & 2008; Wilby and Dawson, 2004). Moreover, the AR&R projection was based on very coarse dynamic models and primarily focused on the National Resource Management (NRM) regions which are very large in area (AR&R, 2015). For example, AR&R (2015) suggests the same variation for future IFDs for almost the entire East Coast North region, which includes Gold Coast in the south and Mackay in the north. In addition, there is no insight provided in relation to the changes in the scaling for different frequencies and durations, which is critical when selecting the design rainfall for any infrastructure development.

In contrast, change factors suggested by this research were based on downscaled rainfall data for small identified homogeneous regions and the results focus on separate change factors for all standard frequencies and durations that are typically used for selecting the design rainfall. Furthermore, the outcomes presented in this study were consistent with the previous work conducted by Abbs et al (2007) for the coastal belt of southeast Queensland. Abbs et al (2007) suggested that IFDs (for the 24-hour duration) are expected to increase by 30% by 2040. However, Abbs et al (2007) study was based on the older climate change scenarios (SRES B2, SRES A2) and limited to the Coastal-SEQ. Furthermore, the study did not consider smaller duration and frequent rainfall events that are essential for the design of WSUD systems.

158

8.3 Adaptation of WSUD to changes in the future IFDs

As discussed in Section 8.1, the sizing of the WSUD treatment systems is typically dependent on treatment and hydraulic requirements (BCC & MBW, 2006). The treatment requirements are based on the expected pollutant reduction capacity. In general, the design of WSUD units for the pollutant removal is determined based on standard treatment performance curves provided by MUSIC. The hydraulic design typically involves the safe conveyance of stormwater through the WSUD systems without damaging key components and the surrounding developments. Further, hydraulic principles are applied in designing the adjoining structures such as inlets and outlets of the WSUD systems to maintain desired flow rates and velocities to achieve the targeted pollutant treatments. Table 8.13 shows a summary of the treatment and hydraulic design aspects of typical WSUD systems. It can be seen that the sizing of all WSUD treatment systems is primarily governed by the design discharge. Therefore, estimating the design discharge more accurately is critical for determining the size of WSUD systems (BCC & MBW, 2006).

The catchments for the WSUD systems are typically small. Therefore, many WSUD guidelines recommend the use of Rational Method for estimating design discharges. In addition, Queensland Urban Drainage Manual (QUDM, 2017) recognises Rational Method as the suitable for estimating the design discharge for small catchments (<500 ha) in Queensland. The equation for the Rational Method for the estimation of the design discharge can be expressed as given in Equation 8.4,

𝑄𝑄 = 𝐶𝐶𝑑𝑑𝐶𝐶𝑖𝑖𝑘𝑘𝐶𝐶 (8.4)

Where, Q refers to the design discharge; Cd refers to the discharge coefficient; Iyt refers

to the design rainfall intensity for y return period and t time of concentration; and A refers to the catchment area.

However, the approach for estimating the design discharge and thereby determining the size of the WSUD treatment systems are based on a static climate assumption. The impact of climate change is not considered in the estimation of the design discharge and there are no robust guidelines available to incorporate the impacts of climate change in the design of WSUD systems.

159

Table 8.13: summary of the treatment and hydraulic design aspects of WSUD systems (BCC & MBW (2006) and GCCC (2005))

WSUD systems

Treatment aspects Hydraulic aspect

Swale

Comply with the treatment performance curve from the conceptual model (MUSIC). The pollutant load reduction is derived using the percentage of the top area of the swale to the catchment area Typically focus on 3-month AIR for the water quality improvement

Minor floods (2-10 year ARI) should be conveyed safely Major floods (50-100 year ARI) velocities should be kept low to avoid scouring of the system

𝑣𝑣 = 𝑄𝑄𝐶𝐶

Sedimentation basin

Comply with the treatment performance curve from the conceptual model (MUSIC). The sediment capture efficiency is derived using the sediment basin area and design discharge typically for 1 year ARI floods. Alternatively, required sediment basin area, (A) can be calculated using, 𝐶𝐶 = 𝑄𝑄 {𝑛𝑛((𝑅𝑅 − 1)𝑛𝑛 − 1)𝑣𝑣𝑠𝑠

�𝑑𝑑𝑝𝑝 + 𝑑𝑑𝑝𝑝�(𝑑𝑑𝑝𝑝 + 𝑑𝑑∗)

}

Where, Q- design discharge; n-turbulence parameter; vs- settling velocity; de- detention depth above the pool level; dp- depth of the pool level; and d*- depth below the pool level

Should have a volume (V)to ensure the desilting is not frequent (typically once in 5-year)

𝑉𝑉 = 𝐶𝐶𝐶𝐶𝑅𝑅𝐿𝐿𝑜𝑜𝐹𝐹𝑐𝑐 Where, Ac- catchment area; R- de- capture efficiency; Lo- sediment loading rate; and Fc- cleanout frequency Outlet structure (1-year ARI) floods. Bypass structures for 2- 10-year ARI flood (for minor drainage systems) and 50-100 year ARI (for major drainage systems)

Bioretention basin

Comply with the treatment performance curve from the conceptual model (MUSIC). The pollutant load reduction is derived using the percentage of bioretention basin surface area to the catchment area

Inflow and outflow structures for minor flood (typically 2-10 year ARI) Major floods (50-100 year ARI) velocities should be kept low to avoid scouring of the system

Constructed wetland

Comply with the treatment performance curve from the conceptual model (MUSIC). The pollutant load reduction is derived using the percentage of the wetland macrophyte zone surface area system surface area to the catchment area. Designed typically for 1 year ARI.

Inflow and outflow structures for operation flow, typically 1-year ARI Bypass (macrophyte zone) for 2- 10-year ARI flood (for minor drainage systems) and 50-100 year ARI (for major drainage systems)

160

However, the impact of climate change on the estimation of design rainfall is evident in this research. Therefore, an additional factor, Cf is suggested to the Rational Method equation to account for the changes in the design rainfall intensities due to climate change, as given in Equation 8.5.

𝑄𝑄 = 𝐶𝐶𝑑𝑑𝐶𝐶𝑓𝑓𝐶𝐶𝑖𝑖𝑘𝑘𝐶𝐶 (8.5)

The change factors estimated for southeast Queensland for different durations and ARIs for RCP 4.5 and RCP 8.5 climate change scenarios can be derived from Table 8.4 to Table 8.11.

In general, the time of concentrations (duration of the rainfall event) considered for WSUD design are small (typically < 2-hour). Further, in the context of WSUD, design discharges are typically classified as design operation flow (1 year ARI or less - typically for the treatment purposes), minor design flow (2-10 year ARI - typically for velocity checks, inlet, and outlet design purpose) and major design flow (50-100 year ARI, - typically for bypass design purpose) (BCC & MBW, 2006). Therefore, in this research, change factors proposed for the estimation of the design discharge were averaged across the design discharges and duration of the rainfall (only durations less than 2-hour were considered) and presented in Table 8.14.

Table 8.14: Proposed change factors for southeast Queensland (2026-2045)


RCP 4.5 RCP 8.5 RCP 4.5 RCP 8.5 Operation flow (≤1 year ARI)

1.37 1.31 1.18 1.14

Minor flow (2-10 year ARI)

1.40 1.34 1.08 1.02

Major flow (50-100 year ARI)

1.46 1.40 1.00 1.00*

* The calculated value was 0.89, implying a reduced future discharge. However, this research suggests using 1.00 for safety.

These averaged change factors can be used in the modified Rational Method equation proposed in Equation 8.5 in order to estimate the design discharges for future climate change scenarios. The design discharges should be used in the design of WSUD systems in order to withstand climate change. These factors are proposed to be incorporated into the WSUD guidelines of the southeast Queensland city councils in order to adapt the impacts of climate change in the design of WSUD treatment units.

161

8.4 Conclusions


• The IFD curves generated using at-site frequency analysis was overall consistent with the IFDs provided by the Bureau of Meteorology at both stations. However, the at-site frequency analysis showed a trend of slight under-estimation for low-duration rainfalls and slight over-estimation for the higher-duration rainfalls compared to the AR&R IFDs. This is attributed to the fact that the approach adopted in this study was different from what had been used in the AR&R project. AR&R IFDs were based on a regional frequency analysis that covers the entire Australia, which involved many approximations to cover a large area.

• Overall, there was a significant increase in the rainfall intensities for the future climate change scenarios compared to the present IFD provided by BoM. In general, smaller duration frequent rainfall and the longer duration infrequent rainfall are expected to increase significantly for both RCP 4.5 and RCP 8.5 climate change scenarios.

• On average, the IFDs for the Coastal-SEQ is expected to increase by 23-30% for the near future and 38-45% distant future. The IFDs for the Inland-SEQ is expected to increase by 5-15% for the near future and 37-38% for the distant future (the percentage covers both RCPs).

• This study proposes an approach to consider the impacts of climate change in the design of a WSUD system, by introducing a climate change factor Cf into the Rational Method procedure to determine the design flow rate as given by,

𝑄𝑄 = 𝐶𝐶𝑑𝑑𝐶𝐶𝑓𝑓𝐶𝐶𝑖𝑖𝑘𝑘𝐶𝐶

The values for the Cf for southeast Queensland is also presented and proposed to incorporate into the city council guidelines for WSUD in order to adapt the impacts of climate change in the design of WSUD. The proposed Cf for southeast Queensland is presented in Table 8.15.

The climate change factor Cf can be also applied to any regional flood frequency estimation models for southeast Queensland including the ARR RFFE (2016).

162

Table 8.15: Proposed change factors for southeast Queensland

Coastal-SEQ Inland- SEQ

RCP 4.5 RCP 8.5 RCP 4.5 RCP 8.5 Operation flow (≤1 year ARI)

1.37 1.31 1.18 1.14


1.40 1.34 1.08 1.02


1.46 1.40 1.00 1.00

163

Chapter 9 Impact of Climate Change on Pollutant Export and Stormwater Quality

9.1 Background

WSUD treatment systems are primarily designed to achieve pollutant concentration reductions to pre-specified levels (Wong, 2006; BCC & MBW, 2006; Goonetilleke et al., 2014; Mangangka, 2013). The design for pollutant concentration reductions is generally based on the treatment performance curves provided by local government guidelines (BCC & MBW, 2006). These performance curves are typically generated using the Model for Urban Stormwater Improvement Conceptualisation (MUSIC). MUSIC curves can be used to size the WSUD treatment systems based on the conceptual design in order to meet local government requirements (Wong, 2006). Designs based on the MUSIC treatment curves are primarily for smaller developments. However, for larger developments, detailed numerical modelling is recommended to demonstrate compliance with the treatment objectives (Wong, 2006; BCC & MBW, 2006). The treatment performance curves or the detailed numerical models are typically based on observed rainfall data from meteorological stations in the region of interest. The observed rainfall events are used to estimate the pollutant concentrations of the stormwater received at the inlets of the treatment systems (Goonetilleke et al., 2014; Zoppou, 2001). This essentially implies that the estimation of stormwater quality is undertaken for static climate conditions. The impacts of climate change on rainfall characteristics and rainfall patterns are not considered during the WSUD treatment system design. However, as discussed in Chapter 2, the change in the rainfall characteristics and rainfall pattern would have a direct and significant impact on pollutant export and the stormwater quality reaching the inlets of these treatment systems. Consequently, this can compromise the targeted treatment efficiency or the amount of pollutants removed by the treatment systems.

164

The primary focus of this chapter was to estimate the changes in pollutant export and the stormwater quality characteristics for different future climate change scenarios. As discussed in Chapter 4, stormwater quality modelling was used as a tool to estimate the changes in future pollutant exports and stormwater quality. Stormwater quality modelling can be classified as event-based modelling and continuous-based (lumped time-based) modelling (Akan and Houghtalen, 2003; Liu, 2011; Ahyerre et al., 1998). Among them, the event-based stormwater quality modelling is perceived as having greater accuracy and reliability (Akan and Houghtalen, 2003; Liu, 2011; Egodawatta, 2007). However, most event-based stormwater quality models are typically based on a few selected rainfall events. The selection of events is often determined by the availability of corresponding water quality monitoring data (Ma, 2016; Egodawatta, 2007). Therefore, stormwater quality estimations using selected rainfall events do not completely represent actual scenarios. Nevertheless, using all possible rainfall events for water quality estimation is a time-consuming task. This is because that there are no stormwater quality models available to extract all events from a given time-series of rainfall data and to estimate the stormwater quality (for every single event) simultaneously. However, it was important in this research to consider all rainfall events over a long period of time for robust assessment. Therefore, an event-based stormwater quality model was developed to estimate the stormwater quality based on long-term rainfall time-series. This model was used to estimate the future event-based pollutant exports and pollutant concentrations for southeast Queensland. The analysis was based on future rainfall data generated using the downscaling models discussed in Chapter 6 and Chapter 7.

Accordingly, this chapter provides detailed discussions on the model setup including catchment characteristics, event separation, pollutant process modelling and runoff modelling. Further, discussions are presented on the impacts of climate change on pollutant exports and stormwater quality in southeast Queensland.

9.2 Model setup

The estimation of stormwater quality primarily involves the estimation of pollutant export and the amount of runoff produced during storm events. Pollutant export was primarily determined by pollutant build-up and pollutant wash-off, whereas runoff was determined by the catchment properties (Liu, 2011; Akan and Houghtalen, 2003).

A customised stormwater quality model was developed to estimate Event Mean Concentration (EMC) of the Total Suspended Solids (TSS) for future climate change scenarios. TSS is considered as the most important pollutant since most of the other pollutants are absorbed to its surface. TSS is also referred to as the indicator pollutant

165

as other pollutant concentrations are estimated as a percentage of TSS (Ball at al., 2000; Jartun et al., 2008; Kayhanian et al., 2008; Clark and Pitt, 2009; Vaze et al., 2000).

The model consisted of four components, namely, catchment, event separation, pollutant process modelling, and runoff modelling. A schematic diagram indicating the components and the key functionality of the model is presented in Figure 9.1. A detailed discussion of each of the components is presented in the following sections. The model was developed in the R platform and the source code for the model is provided in Appendix E. A detail discussion on R is presented in Section 4.3.1.

Figure 9.1: Processes associated with the stormwater quality model

EVENT SEPERATION

Antecedent dry-days

Rainfall intensity

Total rainfall

Rainfall time-series

POLLUTANT PROCESS MODELLING

Pollutant Build-up

Pollutant Wash-off

RUNOFF MODELLING

EMC

Pollutant export Total runoff

CATCHMENT

166

9.2.1 Catchment

The first step in the investigation of the impacts of climate change on stormwater quality involved establishing the catchment. This included defining the catchment properties such as catchment area, urban form, impervious fractions, and the detailed drainage network in the catchment. Based on the catchment properties selected, the resultant stormwater quality was expected to differ. Therefore, in general, there was no exemplar catchment available for the study. In addition, the study was intended to estimate the changes in stormwater quality due to future climate change compared to the current stormwater quality. To achieve this, a set of historical rainfall data was used to simulate stormwater quality and to act as the baseline. Then, the future stormwater quality data obtained by simulating the same model with future rainfall scenarios were compared against the baseline. Accordingly, this approach does not influence by the catchment used. Therefore, for this study, a simpler conceptual catchment was used for the stormwater quality simulations as presented in Figure 9.2.

Figure 9.2: Conceptual catchment

The catchment had a unit area of 1 m2 road surface with a unit length (1 m) of kerb alongside. The catchment drainage consisted of a single node, where runoff from this catchment flowed to this node.

Rainfall

1 m 1 m

Area (1 m2)

Road

Node

Curb

167

9.3.1 Event separations

Event separation is one of the important components of the model. The model was designed to identify independent rainfall events from the time-series of rainfall data. The events were identified based on the following assumptions:

• An event was considered independent only if the consecutive event was separated by 6-hour antecedent dry period. Otherwise, those events were treated as a single event.

• An event that constituted less than 1 mm total rainfall during a period longer than 1-hour was not considered as a rainfall event (but a drizzle) and not considered for the analysis

• Any event having erroneous data entries were discarded from the analysis.

Once the independent events were identified, the antecedent dry-days, intensities and the total rainfall were determined for each of the rainfall events.

9.3.2 Pollutant process modelling

The pollutant processes considered in the model development were pollutant build-up and pollutant wash-off.

A Pollutant build-up

Reliable estimation of TSS build-up on catchment surface primarily depends on the mathematical function and underlying parameters used to replicate the build-up process in the model. Several authors have proposed different mathematical functions to replicate the pollutant build-up process such as reciprocal; logarithmic; exponential and power functions. A comprehensive study conducted by Egodawatta (2007) in Gold Coast, Queensland suggested that a power function could replicate the observed pollutant build-up loads compared to the other mathematical functions. This study was also consistent with a previous investigation by Ball et al. (1998). Hence, in this study, a power equation was used to estimate the pollutant build-up as presented in Equation 9.1.

𝐵𝐵 = 𝑎𝑎𝐷𝐷𝑏𝑏 (9.1)

Where, B is build-up load on road surfaces (g/m2); D is the antecedent dry-days; and a and b are the build-up coefficients.

The coefficient a depends on the urban form of the catchment and coefficient b depends on the surface properties of the catchment (Ball et al., 1998). For this research study, a and b were determined based on the results from Egodawatta (2007). Egodawatta (2007) conducted a small plot-scale study on residential road surfaces and concluded

168

that the build-up coefficients, a and b can be estimated as 1.65 and 0.16, respectively, for residential catchments. Therefore, in this research, these values were used in the model to estimate the pollutant build-up.

B Pollutant Wash-off

The most common mathematical function used to replicate the wash-off process on a road surface is an exponential function (Sartor et al., 1974; Vaze and Cheiw, 2002; Egodawatta, 2007). However, different rainfall and runoff parameters have been suggested as independent variables in the exponential equation. For example, Chiew et al. (1997) used runoff volume to estimate the wash-off, whereas Sartor et al. (1974) used rainfall intensities to estimate the wash-off. However, Egodawatta (2007) concluded that wash-off can be well replicated using the equation proposed by Sartor (1974) based on the studies conducted in southeast Queensland and accordingly, was used in this study. The generic format of the equation suggested by Sartor (1974) is presented in Equation 9.2.

𝑊𝑊 = 𝑊𝑊𝑜𝑜(1 − 𝑒𝑒−𝑘𝑘𝑘𝑘𝑘𝑘) (9.2)

Where, W refers to the weight of mobilised material after time t; Wo refers to the initial weight of the material on the surface; I refers to the rainfall intensity; and k refers to the wash-off co-efficient.

The Equation 9.2 had been developed based on the fundamental knowledge which expresses the direct relationship of the weight of material mobilised by a storm event to the rainfall intensity, initial weight of the material and the surface characteristics of the catchment, assuming that any storm event would have the potential to wash-off all available pollutants on the surface. However, Egodawatta (2007) argued that the fraction of pollutant wash-off during a storm is always less than 1, suggesting a capacity factor, CF into the wash-off equation as given by Equation 9.3.

𝑊𝑊𝑊𝑊𝑜𝑜

= 𝐶𝐶𝐹𝐹 (1 − 𝑒𝑒−𝑘𝑘𝑘𝑘𝑘𝑘) (9.3)

The capacity factor can be calculated using the equations presented in Table 9.1. However, in the context of climate change, it was expected that for a few storm events, the rainfall intensities may exceed 133 mm/hr, which was the highest intensity suggested by Egodawatta (2007) to determine the capacity factors. In such cases, it was assumed that the equation given for the 90 to 133 mm/hr range is valid even beyond 133 mm/hr rainfall intensity, also suggesting that a complete wash-off of available pollutant is possible for such high rainfall intensities.

169

Table 9.1: Equations for capacity factors

Intensity Range, I, (mm/hr) Capacity Factor

5-40 0.01 𝐶𝐶 + 0.1

40-90 0.5

90-133 0.0098 𝐶𝐶 − 0.38

The wash-off coefficient k is an empirical parameter with no physical significance with units mm-1. Wash-off coefficient k is typically a site-specific constant (Millar, 1999). However, studies have found the wash-off coefficient is also a function of pollutant type, catchment area, rainfall intensity, and catchment slopes (Alley, 1981; Millar, 1999; Alley and Smith, 1981). Nevertheless, a constant value for k is used to avoid the complexities of the wash-off models. Further, it has been noted that the performance of the wash-off models is not significantly compromised with a constant k value. Accordingly, the wash-off coefficient for road surfaces is considered as 0.0008 mm-1 (Egodawatta 2007; Liu, 2011) and was also used in this study.

9.3.3 Runoff modelling

Runoff models generally involved two primary processes, namely hydrologic, simulation and hydraulic simulation (Zoppou, 2001). However, in this research, a hydraulic simulation was not applicable as there was no drainage network in the conceptual catchment. Hydrological simulation refers to the estimation of runoff hydrograph at the nodes of the drainage network based on the rainfall data provided. The hydrologic simulation primarily involves estimation of the losses, calculation of the effective rainfall, and calculations to develop the runoff volume based on a set of basic assumptions and approximations (Mansell, 2003).

In order to estimate the losses, different loss models are suggested by different researchers including initial loss model; initial loss and continuous loss models; initial loss and proportional loss model; and infiltration models (Boyd et al., 2003; O’Loughlin and Stacks, 2004). In this study, an initial loss model was used as the catchment was predominantly an impervious surface. The initial loss for impervious surface generally varies from 0 to 5 mm (Boyd et al., 2003; O’Loughlin and Stack, 2004). However, Chowdhury (2018) suggested that the median initial loss in a small urban catchment varies between 2.3 - 2.5 mm based on a study conducted in southeast Queensland. Therefore, in this study, an initial loss of 2.4 mm was used to estimate the effective

170

rainfall. Once the effective rainfall was calculated, the runoff volume was estimated using Equation 9.4.

𝑉𝑉 = 𝑅𝑅𝛼𝛼𝑝𝑝𝑓𝑓𝑓𝑓𝐶𝐶 (9.4)

Where, V refers to the runoff volume; Rfeff refers to the effective rainfall; and A refers to the area of the catchment.

Finally, the EMC was estimated for each of the storm events using the outputs from the pollutant process modelling and runoff modelling. The equation used to calculate the EMC is presented in Equation 9.5.

𝑆𝑆𝑀𝑀𝐶𝐶 = 𝛼𝛼𝑆𝑆𝑉𝑉

(9.5)

Where, EMC refers to the event mean concentration; PE refers to the pollutant export; and V refers to the total runoff from the storm event.

9.3 Impacts of climate change on pollutant process

The primary focus of this analysis was to estimate pollutant export from a unit impervious catchment surface using build-up and wash-off replications for different future climate change scenarios. These values were compared against the values for present conditions. Accordingly, rainfall data from a historical period was used in the model to simulate the pollutant build-up and pollutant wash-off for all rainfall events. The rainfall data (5-minute time-series) from 2002 to 2012 and from 2010 to 2015 were used for Gold Coast Seaway station (representing Coastal-SEQ) and Toowoomba Airport station (representing Inland-SEQ) respectively.

Build-up and wash-off were calculated for each of the rainfall events for the historical periods. The first (Q1), second (Q2 -median) and third (Q3) quantiles of the pollutant build-up and pollutant wash-off were estimated for the historical periods. Quantiles define the positions of the frequency distribution of the estimated pollutant build-up and pollutant wash-off. These indices were determined for the historical period and considered as baselines and compared against the future pollutant build-up and wash-off for different climate change scenarios.

Two climate change scenarios, RCP 4.5 and RCP 8.5 from EC-EARTH were used to estimate the pollutant build-up and pollutant wash-off for the future. Both RCPs consisted of rainfall data for near future (2026-2045) and distant future (2081-2100), making 4 sets of data per meteorological station. Each of the 4 sets of data comprised of 10 realizations. The quantiles Q1, Q2 and Q3 were estimated for each of the ensembles.

171

9.3.1 Changes in pollutant processes in the Coastal-SEQ

The results of the estimated pollutant build-up, pollutant wash-off and the percentages changes compared to the historical data for Coastal-SEQ is presented in Table 9.2.

Table 9.2: Pollutant build-up and pollutant wash-off (Gold Coast Seaway station)

a. Pollutant build-up (g/m2)

Q1 Q2 Q3 Historical (2002-2015) 1.57 1.82 2.15 RCP 4.5 (2026-2045) 1.69 (8) 2.10 (15) 2.49 (16) RCP 4.5 (2081-2100) 1.71( 9) 2.13 (17) 2.47 (15) RCP 8.5 (2026-2045) 1.74 (11) 2.14 (18) 2.50 (16) RCP 8.5 (2081-2100) 1.74 (11) 2.16 (19) 2.52 (17)

b. Pollutant wash-off (Pollutant export) (g/m2)

Q1 Q2 Q3 Historical (2002-2015) 0.28 0.68 2.15 RCP 4.5 (2026-2045) 0.20 (-29) 0.78 (15) 2.81 (31) RCP 4.5 (2081-2100) 0.21 (-25) 0.74 (9) 2.89 (34) RCP 8.5 (2026-2045) 0.21 (-25) 0.61 (-10) 2.63 (22) RCP 8.5 (2081-2100) 0.18 (-36) 0.67 (-2) 2.73 (27)

Note: The values provided in the parenthesis denote the percentage change compared to the historical values.

Table 9.2 provides averaged quantiles (across 10 realizations) of pollutant build-up and pollutant wash-off for the historical and future climate change scenarios. The pollutant build-up was expected to increase across the Coastal-SEQ for both climate change scenarios in the future. The ensemble median of the build-up showed an increase of 15%, 17%, 18% and 19% for RCP 4.5 (2026-2045), RCP 4.5 (2081-2100), RCP 8.5 (2026-2045) and RCP 8.5 (2081-2100) climate change scenarios, respectively. The events with smaller pollutant build-up (Q1) showed an increase between 8% and 11%, whereas the events with higher pollutant build-up (Q3) showed an increase between 16% and 17%. This can be attributed to the increase in antecedent dry days between storm events in the future as presented in Table 9.3.

172

Table 9.3: Estimated changes in antecedent dry-days for future climate change scenarios (Coastal-SEQ)

a. Antecedent dry-days (days)

Q1 Q2 Q3 Historical (2002-2015) 0.74 1.91 5.45 RCP 4.5 (2026-2045) 1.26 4.81 13.75 RCP 4.5 (2081-2100) 1.36 5.26 12.80 RCP 8.5 (2026-2045) 1.60 5.55 14.34 RCP 8.5 (2081-2100) 1.75 5.89 14.51

A substantial increase was expected in the antecedent dry days for the future storm events. There was 15% to 21% increase in the future median antecedent dry days compared to the present climatic conditions. The antecedent dry day period of a storm event is the primary contributor to the pollutant build-up (Ball et al., 1998; Egodawatta, 2007). Therefore, the increase in the pollutant build-up for the future climate change scenarios can be clearly related to the increase in the antecedent dry days.

On the other hand, the changes in the pollutant wash-off (total pollutant export for the storm event) showed different trends between RCP 4.5 and RCP 8.5 climate change scenarios. The median pollutant export was expected to increase by 15% and 9% in the near and distant future, respectively, for RCP 4.5 climate change scenario, whereas, the median pollutant export was expected to decrease by 10% and 2% in the near and distant future, respectively, for RCP 8.5 climate change scenario. However, it can be noted that the first quantile events (events with pollutants exports within the first 25% of the all events considered), showed a substantial decrease while the third quantile events (events with pollutant exports above the first 75% of the all events considered) showed a substantial increase. The changes in pollutant export can be explained by the changes in the rainfall patterns, primarily, the changes in the rainfall intensities.

Table 9.4: Estimated changes in maximum rainfall intensities for future climate change scenarios (Coastal-SEQ)

Maximum intensity (mm/hr)


Table 9.4 shows the changes in the maximum intensities for different climate change scenarios for Coastal-SEQ. The median of the maximum rainfall intensities showed an

173

increase (9% in the near future and 13% in the distant future) for RCP 4.5 and showed a decrease (15% in the near future and 2% in the distant future) for RCP 8.5. However, the first quartile events showed a substantial decrease and the third quartile events showed a substantial increase in the rainfall intensities compared to that of the present climatic conditions.

Figure 9.3 shows the probability distribution of the pollutant exports of the 10 realizations for different climate change scenarios. All realizations have produced similar distributions for pollutant exports. The distribution of the pollutant export was intrinsically positive and highly skewed. It can be seen from Figure 9.3 that the future pollutant exports for all climate change scenarios will have higher variation compared to the historical pollutant exports. In general, the proportions of the events with smaller pollutant exports were expected to decrease while the events with the higher pollutant export are expected to increase in all future climate change scenarios.

174

Figure 9.3: Probability distribution of the pollutant exports (Coastal-SEQ)

0 5 10 15 20

0

0.1

0.2

0.3

0.4

0.5

0.6

PDF

0 5 10 15 20

0

0.1

0.2

0.3

0.4

0.5

0.6

PDF

0 5 10 15 20

0

0.1

0.2

0.3

0.4

0.5

0.6

PDF

0 5 10 15 20

0

0.1

0.2

0.3

0.4

0.5

0.6

PDF

Pollutant export (g/m2)

RCP 4.5 (2026-2045) RCP 4.5 (2081-2100)

RCP 8.5 (2026-2045) RCP 8.5 (2081-2100)

Historical data

Future data

Historical data

Future data

Historical data

Future data

Historical data

Future data

Pollutant export (g/m2)

Pollutant export (g/m2) Pollutant export (g/m2)

175

9.3.2 Changes in pollutant process in the Inland-SEQ

The results of the estimated pollutant build-up and pollutant wash-off for the Inland-SEQ are presented in Table 9.5.

Table 9.5: Pollutant build-up and pollutant wash-off (41529)

a. Pollutant build-up (g/m2)

Q1 Q2 Q3 Historical (2010-2015) 1.66 2.03 2.34 RCP 4.5 (2026-2045) 1.59 (-4) 2.01 (-1) 2.41 (3) RCP 4.5 (2081-2100) 1.71 (3) 2.13 (5) 2.47 (6) RCP 8.5 (2026-2045) 1.74 (5) 2.14 (6) 2.50 (7) RCP 8.5 (2081-2100) 1.65 (-1) 2.07 (2) 2.44 (4)

b. Pollutant wash-off (Pollutant export) (g/m2)

Q1 Q2 Q3 Historical (2010-2015) 0.33 1.01 2.63 RCP 4.5 (2026-2045) 0.26 (-21) 0.88 (-13) 2.87 (9) RCP 4.5 (2081-2100) 0.26 (-21) 1.03 (1) 2.74 (4) RCP 8.5 (2026-2045) 0.23 (-30) 0.80 (-21) 2.72 (3) RCP 8.5 (2081-2100) 0.28 (-15) 0.88 (-13) 2.69 (2)


Table 9.5 provides quantiles of the pollutant build-up and pollutant wash-off averaged across 10 realizations for the historical and future climate change scenarios. It can be noted from Table 9.5 that pollutant build-up is expected to slightly increase across the Inland-SEQ for all climate change scenarios in the future except for a slight decrease in RCP 4.5 (2026-2045). The ensemble median of the build-up showed an increase of 5%, 6% and 2% for RCP 4.5 (2081-2100), RCP 8.5 (2026-2045) and RCP 8.5 (2081-2100) climate change scenarios respectively, and a decrease of less than 1% for RCP 4.5 (2026-2045). Similar to the discussion in Section 9.3.1, this consequence can be linked to the changes in antecedent dry days of the storm events in the future as presented in Table 9.6. A slight increase was expected in the antecedent dry days in the future storm events with 2% to 25% increase in the median compared to the present climatic conditions. Therefore, the increase in the pollutant build-up in the future climate change scenarios can be clearly related to the increase in the antecedent dry days. However, it can be noted that the antecedent dry period expected in the Inland-SEQ was significantly higher than that of the Coastal-SEQ.

176

Table 9.6: Estimated changes in antecedent dry-days for future climate change scenarios (Inland-SEQ)

Antecedent dry-days (days)


On the other hand, the changes in pollutant wash-off (total pollutant export for the storm event) were estimated to have a significant decrease except for a slight increase for RCP 4.5 (2081-2100). The median pollutant export was expected to decrease by 13%, 21% and 13% in for RCP 4.5 (2026-2045), RCP 8.5 (2026-2045) and RCP 8.5 (2081-2100) respectively, and a slight increase for RCP 4.5 (2081-2100). However, it can be noted that the first quantile events showed a significant decrease while the third quantile events showed a considerable increase. The changes in the pollutant export can be explained by the changes in the rainfall patterns, primarily the magnitude of the rainfall intensities.

Table 9.7: Estimated changes in maximum rainfall intensities for future climate change scenarios (Inland-SEQ)

Maximum intensity (mm/hr) Q1 Q2 Q3

Historical (2010-2015) 1.66 4.75 8.03 RCP 4.5 (2026-2045) 1.61 4.22 10.00 RCP 4.5 (2081-2100) 1.72 4.93 10.83 RCP 8.5 (2026-2045) 1.52 4.16 10.58 RCP 8.5 (2081-2100) 1.66 4.15 10.72

Table 9.7 shows the changes in the maximum intensities for different climate change scenarios for Inland-SEQ. The median of the maximum rainfall intensities shows a decrease of 15%, 12% and 13% for RCP 4.5 (2026-2045), RCP 8.5 (2026-2045) and RCP 8.5 (2081-2100), respectively, and a slight increase of 4% for RCP 4.5 (2081-2100). However, in general, the first quartile events show a decrease and the third quantile events show a substantial increase in the rainfall intensities compared to that for the present climate conditions.

Figure 9.4 shows the probability distribution of the 10 realizations of the pollutant exports for different climate change scenarios. As noted in the analysis of Coastal-SEQ

177

data, the distribution of the pollutant export was similar among realizations and was intrinsically positive and highly skewed. However, in contrast to the Coastal-SEQ, the distributions of pollutant exports for the future climate change scenarios for Inland-SEQ were nearly similar to the present characteristics.

178

Figure 9.4: Probability distribution of the pollutant exports (Inland-SEQ)

0 5 10 15 20

0

0.1

0.2

0.3

0.4

0.5

PDF

0 5 10 15 20

0

0.1

0.2

0.3

0.4

0.5

PDF

0 5 10 15 20

0

0.1

0.2

0.3

0.4

0.5

PDF

0 5 10 15 20

0

0.1

0.2

0.3

0.4

0.5

PDF

RCP 4.5 (2026-2045) RCP 4.5 (2081-2100)

RCP 8.5 (2026-2045) RCP 8.5 (2081-2100)



179

9.4 Impacts of climate change on water quality

In the impact study undertaken, EMCs were calculated for all of the rainfall events for the historical periods and future climate change scenarios. The first (Q1), second (Q2 -median) and third (Q3) quantiles of the pollutant EMCs were determined and compared.

9.4.1 Changes in stormwater quality in the Coastal-SEQ

The results of the estimated stormwater quality (EMCs) for Coastal-SEQ are presented in Table 9.8.

Table 9.8: Estimated changes in EMCs for future climate change scenarios (Coastal-SEQ)

EMC (mg/L)

Q1 Q2 Q3 Historical (2002-2015) 113.88 170.78 259.63 RCP 4.5 (2026-2045) 56.14 (-51) 86.15 (-50) 136.14 (-48) RCP 4.5 (2081-2100) 50.02 (-56) 80.82 (-53) 126.84 (-57) RCP 8.5 (2026-2045) 53.44 (-53) 89.14 (-48) 142.58 (-45) RCP 8.5 (2081-2100) 52.72 (-54) 89.89 (-47) 134.67 (-48)


Table 9.8 provides averaged (across 10 realizations) quantiles (Q1, Q2, and Q3) of event-based stormwater quality for the historical and future climate change scenarios. The stormwater quality for Coastal-SEQ was estimated to decrease significantly for both RCP 4.5 and RCP 8.5 climate change scenarios. The ensemble median of the EMC showed a decrease of 50%, 53%, 48% and 47% for RCP 4.5 (2026-2045), RCP 4.5 (2081-2100), RCP 8.5 (2026-2045) and RCP 8.5 (2081-2100) climate change scenarios, respectively. This can be linked to the estimated pollutant export and the total effective rainfall (that directly constitutes the runoff volume) for the future climate change scenarios. Table 9.9 presents the pollutant exports and the effective rainfall for the present and future climate change scenarios for Coastal-SEQ. Although the future pollutant exports were estimated to increase (except for a slight decrease for RCP 8.5 climate change scenarios), the changes were not significant compared to the effective rainfall. The effective rainfall, in general, was expected to roughly double for both climate change scenarios producing high runoff. Consequently, the future EMCs were expected to reduce due to the dilution effect.

180

Table 9.9: Pollutant exports and the effective rainfall for the present and future climate change scenarios for Coastal-SEQ

a. Pollutant export (g/m2) Q1 Q2 Q3


b. Effective rainfall (mm) Q1 Q2 Q3


9.4.2 Changes in stormwater quality in the Inland-SEQ

The results of the estimated stormwater quality (EMCs) for the Inland-SEQ is presented in Table 9.10

Table 9.10: Estimated changes in EMCs for future climate change scenarios (Inland-SEQ)

EMC (mg/L)

Q1 Q2 Q3 Historical (2010-2015) 135.28 191.74 301.80 RCP 4.5 (2026-2045) 78.46 (-42) 117.94 (-38) 167.93 (-44) RCP 4.5 (2081-2100) 65.22 (-52) 99.99 (-48) 144.70 (-52) RCP 8.5 (2026-2045) 72.65 (-46) 107.78 (-44) 163.77 (-46) RCP 8.5 (2081-2100) 74.78 (-45) 110.46 (-42) 150.83 (-50)


Table 9.10 provides averaged (across 10 realizations) quantiles (Q1, Q2, and Q3) of event-based stormwater quality for the historical and future climate change scenarios. Similar to the Coastal-SEQ, the stormwater quality for Inland-SEQ was estimated to decrease significantly for both RCP 4.5 and RCP 8.5 climate change scenarios. The ensemble median of the EMC showed a decrease of 38%, 48%, 44% and 43% for RCP 4.5 (2026-2045), RCP 4.5 (2081-2100), RCP 8.5 (2026-2045) and RCP 8.5 (2081-2100)

181

climate change scenarios, respectively. Similar to the analysis undertaken for Coastal-SEQ, pollutant export and total rainfall were separately analysed to understand the reasons for the decrease in EMCs for future scenarios.

Table 9.11: Pollutant exports and the effective rainfall for the present and future climate change scenarios for Inland-SEQ

a. Pollutant export


b. Effective rainfall


Table 9.11 presents the pollutant exports and the effective rainfall for the present and future climate change scenarios for Inland-SEQ. The future pollutant exports were estimated to decrease while effective rainfall was estimated to increase significantly. The effective rainfall, in general, was expected to increase by approximately 40% producing high runoff, and consequently, resulting in low pollutant concentrations. However, compared to the Coastal-SEQ, the Inland-SEQ is expected to produce higher EMCs in the future climate change scenarios.

182

9.5 Water quality parameters for MUSIC modelling

MUSIC modelling is considered to be the standard in Australia for the design of WSUD treatment units. The water quality input parameters (the mean and the standard deviation of the EMCs) are the primary input for the MUSIC modelling. These parameters are based on different water quality monitoring programs and are provided in the current MUSIC modelling guidelines (BCC, 2001 & 2003; GCCC, 2005; WBD, 2010; Duncan, 1999). In this section, the results of this study are compared against the MUSIC water quality parameters (for TSS) and insights are given for the adaptation of these parameters to account for climate change.

In general, the water quality dataset collected from urban residential and commercial land uses across Australia have been confirmed to fit a log-normal distribution (Lucke, 2018; Blackwood, 1992) and therefore, the mean and standard deviations of these distributions are often presented in the log-normal form. Accordingly, the water quality outcomes of this study were converted to log-normal distribution and compared against the current MUSIC parameters provided for southeast Queensland. Figure 9.5 compares the results from this study to the MUSIC guidelines for southeast Queensland. The mean and the standard deviations of the distributions are presented in Table 9.12.

Figure 9.5: Water quality parameters

It can be seen from Figure 9.5 and Table 9.12 that the mean of the distribution of water quality for this study and the MUSIC guidelines are similar. However, there are slight

Den

sity

EMC (log scale)

Coastal-SEQ

Inland-SEQ

Results of this study

MUSIC guidelines

183

differences in the standard deviations. This can be due to many reasons. The MUSIC parameters were developed based on varied study outcomes from different timeframes, whereas in this study, model outcomes from recent rainfall data were used. In addition, the MUSIC values were based on water quality resulting from different catchments across SEQ, whereas this study was based on a single hypothetical catchment. Moreover, the values suggested by the MUSIC guidelines were from monitoring programs, whereas this study equipped a custom-made water quality model to determine these parameters. However, these values compare well and fall into the data range suggested in the review by Duncan (1999).

Table 9.12: Comparison of EMCs from this study and MUSIC guideline for SEQ

Parameters This study MUSIC guidelines

(WBD, 2010) Coastal-SEQ Inland-SEQ Number of events (n) 770 243 209 Mean 2.21 2.27 2.18 Std. Dev. 0.29 0.30 0.39 Note: Costal-SEQ had more rainfall events (770) recorded compared to the Inland-SEQ (243) during the period considered in the study. Mean and standard deviations are based on log10 (mg/l). Mean of the log-transformed dataset is not mathematically equivalent to the log-transformation of the raw dataset mean.

Figure 9.6 and Figure 9.7 present the log-normal distributions of the water quality (EMCs) for the present and the future climate change scenarios for Coastal-SEQ and Inland-SEQ, respectively. The future water quality was expected to be considerably different compared to the present. Although there was a slight increase in the pollutant export during storm events, the pollutant concentration was expected to reduce due to higher runoff volumes. This feature was attributed to all climate change scenarios in both, Coastal-SEQ and the Inland-SEQ as shown in Figure 9.6 and Figure 9.7. However, variation in the stormwater quality was expected to increase in the Coastal-SEQ in contrast to the decrease in variation for Inland-SEQ.

184

Figure 9.6: EMC for future climate change scenarios for Coastal-SEQ. The dotted lines denote the actual distribution of the data and the solid lines donate the log-normal distribution fitted data (based on the mean and standard deviation)

EMC (log mg/l)

EMC (log mg/l)

EMC (log mg/l) EMC (log mg/l)

Den

sity

Den

sity

Den

sity

Den

sity

RCP 4.5 (2026-2045) RCP 4.5 (2081-2100)

RCP 8.5 (2026-2045) RCP 8.5 (2081-2100)

Historical

Future

Historical

Future

Historical

Future

Historical

Future

185

Figure 9.7: EMC for future climate change scenarios for Inland-SEQ. Note: The dotted lines denote the actual distribution of the data and the solid lines donate the log-normal distribution fitted data (based on the mean and standard deviation)

Den

sity

Den

sity

Den

sity

Den

sity

RCP 4.5 (2026-2045) RCP 4.5 (2081-2100)

RCP 8.5 (2026-2045) RCP 8.5 (2081-2100)

Historical

Future

Historical

Future

Historical

Future

Historical

Future

EMC (log mg/l)

EMC (log mg/l)

EMC (log mg/l) EMC (log mg/l)

186

However, these changes in the water quality parameters in the future have not been incorporated into the MUSIC guidelines for adaptation of WSUD treatment design incorporating climate change. In this study, water quality parameters for future climate change scenarios were determined and presented in Table 9.13. These parameters are proposed to be incorporated into the MUSIC guidelines of the southeast Queensland in order to adopt the impacts of climate change in the design of WSUD treatment systems.

Table 9.13: Water quality parameters for MUSIC modelling

Scenarios Parameters Coastal-SEQ Inland-SEQ Historical Number of events (n) 770 243 (2002-2015) Mean (log10 mg/l) 2.21 2.27 Std. Dev. (log10 mg/l) 0.29 0.30 RCP 4.5 (2026-2045)

Number of events (n) 5498* 6262*

Mean (log10 mg/l) 1.89 2.02 Std. Dev. (log10 mg/l) 0.33 0.28 RCP 4.5 (2081-2100)






Mean (log10 mg/l) 1.89 1.98 Std. Dev. (log10 mg/l) 0.33 0.27

* These events are generated from 10 realizations 20year periods

187

9.7 Conclusions


• A completely new and simplified stormwater quality model was developed to estimate Event Mean Concentration (EMC) of the Total Suspended Solids (TSS) generated from urban residential catchments for future climate change scenarios. The model was designed to automatically extract all independent rainfall events from a given rainfall time-series and simulate water quality parameters based on event-based rainfall characteristics.

• For Coastal-SEQ, the median pollutant export were expected to increase by 15% and 9% for RCP 4.5 (2026-2045) and RCP 4.5 (2081-2100) respectively, whereas, the median pollutant export for RCP 8.5 (2026-2045) and RCP 8.5 (2081-2100) were expected to decrease by 10% and 2% respectively. The stormwater quality for Coastal-SEQ was estimated to decrease significantly for both climate change scenarios. The ensemble median of the EMC showed a decrease of 50%, 53%, 48% and 47% for RCP 4.5 (2026-2045), RCP 4.5 (2081-2100), RCP 8.5 (2026-2045) and RCP 8.5 (2081-2100) climate change scenarios, respectively.

• In Inland-SEQ, the median pollutant export was expected to decrease by 13%, 21% and 13% in for RCP 4.5 (2026-2045), RCP 8.5 (2026-2045) and RCP 8.5 (2081-2100) and a slight increase of 1% for RCP 4.5 (2081-2100). The stormwater quality for Inland-SEQ was estimated to decrease significantly for both climate change scenarios. The ensemble median of the EMC showed a decrease of 38%, 48%, 44% and 43% for RCP 4.5 (2026-2045), RCP 4.5 (2081-2100), RCP 8.5 (2026-2045) and RCP 8.5 (2081-2100) climate change scenarios, respectively.

• Overall, the pollutant export showed a slight change (increase in general) in the future climate change scenario due to changes in the rainfall characteristics. However, the runoff was estimated to increase significantly and thus, resulting in significantly low pollutant concentrations in the future rainfall events. Therefore, though the pollutant concentration was expected to be low, the total pollutant carried by the stormwater could be higher than the present conditions.

• Water quality parameters for future climate change scenarios were developed and

presented in Table 9.14. These parameters are proposed to be incorporated into the MUSIC guidelines of the southeast Queensland in order to adopt the impacts of climate change in the design of WSUD treatment units.

188

Table 9.14: Proposed water quality parameters for MUSIC modelling to incorporate climate change

Scenarios Parameters

Coastal-SEQ

Inland-SEQ

Historical Number of events (n) 770 243 (2002-2015) Mean (log10 mg/l) 2.21 2.27 Std. Dev. (log10 mg/l) 0.29 0.30 RCP 4.5 (2026-2045) Number of events (n) 5498 6262 Mean (log10 mg/l) 1.89 2.02 Std. Dev. (log10 mg/l) 0.33 0.28 RCP 4.5 (2081-2100) Number of events (n) 5405 6038 Mean (log10 mg/l) 1.86 1.97 Std. Dev. (log10 mg/l) 0.33 0.26 RCP 8.5 (2026-2045) Number of events (n) 5305 5536 Mean (log10 mg/l) 1.91 2.00 Std. Dev. (log10 mg/l) 0.33 0.28 RCP 8.5 (2081-2100) Number of events (n) 4979 5388 Mean (log10 mg/l) 1.89 1.98 Std. Dev. (log10 mg/l) 0.33 0.27

189

Chapter 10 Conclusions and Recommendations

10.1 Conclusions

This research primarily developed methodologies for downscaling GCM rainfall outputs to a finer temporal and spatial resolution. The research further demonstrates the applicability of downscaled data in stormwater quality and quantity impact assessments for different future climate change scenarios. A detailed discussion also provided on the characteristics of possible rainfall scenarios in near future and distance future comparing them to current rainfall conditions.

Developing the future rainfall data was based on the state-of-the-art statistical methods that translate the coarse resolution GCM outputs to a fine resolution. In this study, downscaling was performed in two steps. First was spatial downscaling using quantile-quantile bias correction approach and the second was temporal downscaling using first-order homogeneous Markov weather generation model. The future rainfall data generated using these downscaling procedures were then used for the impact assessments.

In the impact assessment, IFD curves were developed for future climate change scenarios based on an at-site rainfall frequency analysis. Stormwater qualities for future climate change scenarios were also assessed based on a newly developed stormwater quality model that simulates water quality parameters based on event-based rainfall characteristics. The model is designed to automatically extract all independent rainfall events based on a set of criteria from a given rainfall time-series.

These outcomes from the impact assessment provided an in-depth knowledge on the stormwater quality and the quantity for different future climate change scenarios, thereby providing guidance for the adaptation of Water Sensitive Urban Design to climate change.

190

10.1.1 Event-based rainfall homogeneity assessment for southeast Queensland

An event-based rainfall homogeneity test was carried out to define the homogeneous regions inside the study area and thereby selecting representative meteorological stations for downscaling. The analyses were conducted (delineation of homogeneous regions in South East Queensland) based on two different approaches, namely, continuous-rainfall approach and event-based rainfall approach. Hosking Wallis heterogeneity test was performed to evaluate the degree of homogeneity of the meteorological stations. The major conclusions derived from this analysis are as follows:

• A novel methodology was developed for identification of rainfall homogeneous regions based on event-based rainfall characteristics. This methodology is generic and applicable in any part of the world.

• The entire southeast Queensland can be treated as a homogeneous region based on the characteristics of the continuous rainfall. However, based on individual rainfall event characteristics such as antecedent dry-days, maximum rainfall intensities, total rainfall and duration of the rainfall events, there are two separate homogeneous regions identified within the region. This implies that although the characteristics of the continuous rainfall data between stations were statistically similar, the event-based characteristics have significant differences within the region.

• Antecedent dry-days and maximum rainfall intensities of the rainfall events have significant variation between the coastal and the inland areas of southeast Queensland compared to the total rainfall and duration of the events.

• Two homogeneous regions were identified based on the event-based rainfall characteristics namely, Coastal-SEQ and Inland-SEQ. Coastal-SEQ includes Brisbane City Council, Gold Coast City Council, Logan City Council, Redland City Council, Moreton Bay Regional Council and Sunshine Coast Regional Council areas and the Inland-SEQ includes Ipswich City Council, Scenic Rim Regional Council, Lockyer Regional Council and Somerset Regional Council areas.

• Two representative meteorological stations, namely Gold Coast Seaway station (40764) and Toowoomba Airport stations (41529) were selected to represent the Coastal-SEQ and Inland-SEQ respectively based on diverse selection criteria. The criteria were primarily based on the availability of considerably long period of rainfall recording, quality and the completeness of the rainfall data.

191

10.1.2 Downscaling of rainfall data

Downscaling was performed in two steps. Firstly, GCM rainfall outputs (3-hour time-series) were spatially downscaled to match the observed data. Then these data were temporally downscaled to 5-minute time-series. Two separate statistical downscaling models were developed to perform the downscaling in this research.

For spatial downscaling, a new downscaling tool, spdownscale was developed based on quantile-quantile bias correction approach. spdownscale is a generic downscaling tool that can be used in spatial downscaling of daily and sub-daily rainfall data in any part of the world. The spdownscale tool is licensed under the General public license, version 2 (GPL 2), where the general public is permitted to copy and distribute verbatim copies of this document (source code) with no permission to edit. The spdownscale tool is freely available as an R package (https://CRAN.R-project.org /package=sp downscale). Details on installing R packages can be found at R development core team. The spdownscale tool was used to spatial downscale EC-EARTH and ACCESS-1.0 data for RCP4.5 and RCP8.5 climate change scenarios at the two representative meteorological stations for southeast Queensland. Overall, the models developed by the spdownscale for spatial downscaling performed well for both GCMs. Two statistical indexes, RMSE and the gradient of observation-simulation scatter plots were used for the validation purposes. The bias-corrected GCM outputs from the both GCMs were closely comparable to the observed data.

For temporal downscaling, a new model was developed based on first-order homogeneous Markov model. The model is capable of translating 3-hour rainfall time-series into 5-minute time-series. The approach of downscaling opted in the model does not require predictor variable enabling downscaling to 5-minute temporal resolutions. The approach used in the model is generic and can be adapted for downscaling rainfall at any part of the world. Moreover, the model is capable of producing multiple realizations of the rainfall time-series for future climate change scenarios, which is essential for uncertainty assessments. The developed model was used for temporal downscaling of rainfall data in southeast Queensland. The model performance was then assessed based on an independent historical rainfall data. For both Coastal-SEQ and Inland-SEQ the simulated rainfall was in agreement with the observed rainfall in terms of the probability distribution. The RMSE for Gold Coast Seaway station was 0.67 mm and for Toowoomba Airport station was 0.59 mm. The maximum intensities of the simulated rainfall were reasonably similar with observed data in both Coastal-SEQ and Inland-SEQ.

192

10.1.3 Design rainfall for future climate change scenarios

Rainfall statistical models in the form of Intensity - Frequency - Duration (IFD) tables are imperative to obtain design rainfalls required for WSUD treatment system designs. IFDs can be developed by regional frequency analysis or at-site frequency analysis. Regional frequency analysis considers observations from a group of meteorological stations to estimate the IFDs. The use of observations from a group of neighbouring station incorporates the uncertainties associated with stations that have observations for a short period of time and produces IFD estimations for large regions. In contrast, at-site frequency analysis estimates IFDs based on rainfall records from a single meteorological station with exactly reflects the region of interest.

An at-site frequency analysis was undertaken based on the downscaled data at the representative meteorological stations in order to develop IFDs for future climate change scenarios. In order to translate the currently available IFDs to the future climate change scenarios, a novel term namely, change factors (Cf) was introduced in this research. The change factor is determined for all standard durations and Average Recurrence intervals for both RCP 4.5 and RCP 8.5 climate change scenarios. The major conclusions derived from this analysis are as follows:

• The IFD curves generated based on the historical rainfall data using at-site frequency analysis was overall consistent with the IFDs of the Bureau of meteorology for both stations. However, the IFDs calculated based on the at-site frequency analysis were slightly lower than the IFDs provided by the Bureau of Meteorology for low-duration rainfalls. In contrast, the IFDs calculated based on the at-site frequency analysis were slightly higher than the IFDs provided by the Bureau of meteorology for the higher-duration rainfalls.

• Overall, there is a significant increase in the IFDs for the future climate change scenarios compared to the present IFD provided by the Bureau of Meteorology. In general, the intensities of smaller duration frequent rainfall and the longer duration infrequent rainfalls are expected to increase significantly in both climate change scenarios.

• On average, the IFDs for the Coastal-SEQ are expected to increase by 23-30% in the near future and 38-45% distant future. The IFDs for the Inland-SEQ is expected to increase by 5-15% in the near future and 37-38% in the distant future (the percentage covers both RCPs).

• This study suggests an approach to consider the impacts of climate change on the stormwater quantities by introducing the climate change factor Cf into the Rational Method procedure to find the design flow rate as given by,

193

𝑄𝑄 = 𝐶𝐶𝑑𝑑𝐶𝐶𝑓𝑓𝐶𝐶𝑖𝑖𝑘𝑘𝐶𝐶

Where, Q refers to the design discharge, Cd refers to the discharge coefficient, Iyt

refers to the design rainfall intensity for y return period and t time of concentration, A refers to the catchment area and Cf is the change factor. The values for the Cf for southeast Queensland is also presented and proposed to be incorporated into the city council guidelines for WSUD in order to adapt the impacts of climate change in the hydraulic design of WSUD. The proposed Cf for southeast Queensland is presented in Table 10.1.

Table 10.1: Proposed change factors for southeast Queensland

SEQ-Coast SEQ-Inland

RCP 45 RCP 85 RCP 45 RCP 85 Operation flow (≤1 year ARI)

1.37 1.31 1.18 1.14


1.40 1.34 1.08 1.02


1.46 1.40 1.00 1.00

10.1.4 Stormwater quality and quantity characteristics in the future climate change scenarios

A customised stormwater quality model was developed to estimate Event Mean Concentration (EMC) of the Total Suspended Solids (TSS) for future climate change scenarios. The novel feature of this model is that the model has a unique component for separating independent rainfall events from a given time-series of rainfall data. Therefore, this model is capable of simulation event-based water quality parameters for a long period of time. The developed model was used to simulate the future water quality parameters for Coastal-SEQ and Inland-SEQ. The major conclusions derived from this study are as follows:

• For Coastal-SEQ, the median pollutant export were expected to increase by 15% and 9% for RCP 4.5 (2026-2045) and RCP 4.5 (2081-2100), whereas, the median pollutant export for RCP 8.5 (2026-2045) and RCP 8.5 (2081-2100) were expected to decrease by 10% and 2%, respectively. The stormwater quality for Coastal-SEQ was estimated to decrease significantly for both climate change scenarios. The ensemble median of the EMC showed a decrease of 50%, 53%, 48% and 47% for RCP 4.5 (2026-2045), RCP 4.5 (2081-2100), RCP 8.5 (2026-2045) and RCP 8.5 (2081-2100) climate change scenarios, respectively.

194

• In Inland-SEQ, the median pollutant export was expected to decrease by 13%, 21%

and 13% in for RCP 4.5 (2026-2045), RCP 8.5 (2026-2045) and RCP 8.5 (2081-2100) respectively and a slight increase of 1% for RCP 4.5 (2081-2100). The stormwater quality for Inland-SEQ was estimated to decrease significantly for both climate change scenarios. The ensemble median of the EMC showed a decrease of 38%, 48%, 44% and 43% for RCP 4.5 (2026-2045), RCP 4.5 (2081-2100), RCP 8.5 (2026-2045) and RCP 8.5 (2081-2100) climate change scenarios, respectively.

• Overall, the pollutant export showed a slight change (increase in general) in the future climate change scenario due to changes in the rainfall characteristics. However, the runoff was estimated to increase significantly and thus, resulting in significantly low pollutant concentrations in the future rainfall events. Therefore, though the pollutant concentration was expected to be low, the total pollutant carried by the stormwater could be higher than the present conditions.

• Water quality parameters for future climate change scenarios were developed and

presented in Table 9.14. These parameters are proposed to be incorporated into the MUSIC guidelines of the southeast Queensland in order to adopt the impacts of climate change in the design of WSUD treatment units.

195

Table 10.2: Proposed water quality parameters for MUSIC modelling to incorporate climate change

Scenarios Parameters Coastal-

SEQ Inland-SEQ

Historical Number of events (n) 770 243 (2002-2015) Mean (log10 mg/l) 2.21 2.27 Std. Dev. (log10 mg/l) 0.29 0.30 RCP 4.5 (2026-2045) Number of events (n) 5498 6262 Mean (log10 mg/l) 1.89 2.02 Std. Dev. (log10 mg/l) 0.33 0.28 RCP 4.5 (2081-2100) Number of events (n) 5405 6038 Mean (log10 mg/l) 1.86 1.97 Std. Dev. (log10 mg/l) 0.33 0.26 RCP 8.5 (2026-2045) Number of events (n) 5305 5536 Mean (log10 mg/l) 1.91 2.00 Std. Dev. (log10 mg/l) 0.33 0.28 RCP 8.5 (2081-2100) Number of events (n) 4979 5388 Mean (log10 mg/l) 1.89 1.98 Std. Dev. (log10 mg/l) 0.33 0.27

10.2 Recommendations

This research created new knowledge including methodologies to develop future climate data at finer resolutions. The study also presents the impacts of climate change on stormwater quality and quantities and thereby proving insight for the adaptation of Water Sensitive Urban Design to climate change. Also, the research opens up new potential investigations that can be beneficial for creating further knowledge in relation to this research. The following are recommendations for further investigations:

• In this research, two GCMs EC-EARTH and ACCESS 1.0 had been selected based on an in-depth literature review on the performance of all CMIP5 GCMs. Also, both IPCC recommended climate change scenarios, RCP 4.5 and RCP 8.5, were used in this research in order to build confidence in the projections. However, the uncertainty associated with the climate change projection was not quantitatively assessed. Therefore, it is recommended to expand similar studies to many GCMs (preferably all CMIP5 GCMs) to predict future rainfall at finer resolutions thereby assessing uncertainties is possible.

• Both spatial and temporal downscaling models could be transformed into full-fledged downscaling software with all GCMs data inbuilt, user-friendly interface and documentation. Further, incorporation and synchronisation of these two models into a single hybrid downscaling package is recommended.

196

• An approach to minimize the overall variations in among realizations in the temporal downscaling model is recommended.

• This research investigated water quality based on TSS urban residential catchments. Extending the knowledge in relation to different pollutants and different land uses are recommended.

• This research used EMC as the indicator for stormwater quality. However, instantaneous stormwater quality data provides information on the variation of stormwater quality with time. This can strengthen the detailed analysis of the future stormwater quality scenarios. Therefore, further analysis using instantaneous stormwater quality based on the developed future rainfall data from this study is recommended.

• It is recommended to use the future rainfall data (or the methodologies to develop the future rainfall data) to simulate the performance existing WSUD treatment systems across southeast Queensland for future climate change scenarios and thereby appropriate retrofitting can be adapted to overcome the negative impact.

• It is recommended to develop new conceptual pollutant removal curves for different WSUD treatment units for different climate change scenarios based on the future rainfall data developed in the study.

197

References

Abatzoglou T.,John and Brown Timothy J. 2012. A comparison of statistical downscaling methods suited for wildfire applications. International Journal of Climatology 32 (5): 772-780.

Abbs DJ, McInnes KL, Rafter R. 2007. The impact of climate change on extreme rainfall and coastal sea levels over south-east Queensland.

ABS. 2014. Regional population Growth, Australia, 2013-14 Australian Bureau of Statistics.

Ahmed, Kazi Farzan, Guiling Wang, John Silander, Adam M. Wilson, Jenica M. Allen, Radley Horton and Richard Anyah. 2013. Statistical downscaling and bias correction of climate model outputs for climate change impact assessment in the U.S. northeast. Global and Planetary Change 100: 320-332.

Ahyerre, M., G. Chebbo, B. Tassin and E. Gaume. 1998. Stormwater quality modelling, an ambitious objective? Water Science and Technology 37 (1): 205-213.

Akan, A.O. and R.J. Houghtalen. 2003. Urban Hydrology, Hydraulics, and Stormwater Quality: Engineering Applications and Computer Modeling: Wiley.

Alexander, Lisa, Hongang Yang and Sarah Perkins. 2013. ClimPACT: The Commission for Climatology (CCl) Expert Team on Climate Risk and Sector-Specific Climate Indices (ET CRSCI), World climate Programme.

Alley, W.M and P.E Smith. 1981. Estimation of accumulation parameters for urban runoff quality modeling. Water Resources Research 17 (6): 1657-1664.

Alley, W.M. 1981. Estimation of impervious‐area Wash-off Parameters. Water Resources Research 17 (4): 1161-1166.

Almazroui, Mansour, Khaled S. Balkhair, M. Nazrul Islam and Zekai Sen. 2017. Climate Change Impact on Monthly Precipitation Wet and Dry Spells in Arid Regions: Case Study over Wadi Al-Lith Basin. Advances in Meteorology 2017: 13.

AR&R. 2015. Australian Rainfall & Runoff - Development of Intensity-Frequency-Duration Information Across Australia: CSIRO and Bureau of Meteorology, Australia.

198

ARR RFFE. 2016. Australian Rainfall Runoff – Regional Flood Frequency Estimation Model. https://rffe.arr-software.org/

Bae, Deg-Hyo, Il-Won Jung and Dennis P. Lettenmaier. 2011. Hydrologic uncertainties in climate change from IPCC AR4 GCM simulations of the Chungju Basin, Korea. Journal of Hydrology 401 (1): 90-105.

Ball, J. E., R. Jenks and D. Aubourg. 1998. An assessment of the availability of pollutant constituents on road surfaces. Science of the Total Environment 209 (2–3): 243-254.

Barnett, T. P. and R. Preisendorfer. 1987. Origins and Levels of Monthly and Seasonal Forecast Skill for United States Surface Air Temperatures Determined by Canonical Correlation Analysis. Monthly Weather Review 115 (9): 1825-1850.

Bautista, M. F. and N. S Geiger. 1993. Wetlands for Stormwater Treatment. Water Environment & Technology 5 (7): 50-55.

BCC. 2001. Stormwater Quality Monitoring Program 2000/2001: Brisbane City Council.

BCC. 2003. Guidelines for pollutant export modelling in Brisbane. 7 ed: Brisbane City Council.

Beecham, Simon and Rezaul Chowdhury. 2012. Effects of changing rainfall patterns on WSUD in Australia. Proceedings of the Institution of Civil Engineers: Water Management 165 (5): 285-298.

Beersma, J., Maureen Agnew, D. Viner and Mike Hulme. 2000. Climate Scenarios for Water-Related and Coastal Impacts.

Blackwood, Larry G. 1992. The lognormal distribution, environmental data, and radiological monitoring. Environmental Monitoring and Assessment 21 (3): 193-210.

BoM & CSIRO. 2007. Climate change in Australia. Accessed 04/05. http://www.climatechangeinaustralia.gov.au/en/.

BoM & CSIRO. 2008. The Bureau of Meteorology Statistical - Downscaling Model Graphical User Interface: user manual and software documentation: Bureau of Metrology, government of Australia, CSIRO.

BoM. 2015. Australian climate variablitiy and change. Accessed 06 May. http://www.bom.gov.au/climate/.

199

Bordoy, R. and P. Burlando. 2014. Stochastic downscaling of precipitation to high-resolution scenarios in orographically complex regions: 1. Model evaluation. Water Resources Research 50 (1): 540-561.

Bowring, J. 2005. Heavy metal toxicity and the unborn child. Midwifery today with international midwife,(76), 48-67.

Boyd, M.J, E.H Rigby and R Van Drie. 2003. WBNM 2003, User Guide.

Bratieres, K., T. D. Fletcher, A. Deletic and Y. Zinger. 2008. Nutrient and sediment removal by stormwater biofilters: A large-scale design optimisation study. Water Research 42 (14): 3930-3940.

Brown, R. 1999. The stormwater source control: Facing the challenges. In Comprehensive Stormwater and the Aquatic Ecosystem Conference 2, edited, 67-74.

Brydon, J., M. C. Roa, S. J. Brown and H. Schreier. 2006. Integrating Wetlands into Watershed Management: Effectiveness of Constructed Wetlands to Reduce Impacts from Urban Stormwater, Dordrecht, edited, 143-154: Springer Netherlands.

Burge, K, D Browne, P Breen and J Wingad. 2012. Water Sensitive Urban Design in a changing climate: Estimating the performance of WSUD treatment measures under various climate change scenarios. Paper presented at the 7th international conference on Water Sensitive Urban Design, Barton, A.C.T. Engineers Australia.

Bürger, Gerd. 1996. Expanded downscaling for generating local weather scenarios. Vol. 7.

Burton, A., H. J. Fowler, S. Blenkinsop and C. G. Kilsby. 2010. Downscaling transient climate change using a Neyman–Scott Rectangular Pulses stochastic rainfall model. Journal of Hydrology 381 (1–2): 18-32.

Carrol G D. 2002. URBS a catchment management & flood forecasting rainfall runoff routing model- user manual.

CCIA. 2015. Climate change in Australia. Australia: CSIRO.

Charles, Stephen P., Bryson C. Bates, Peter H. Whetton and James P. Hughes. 1999. Validation of downscaling models for changed climate conditions case study of southwestern Australia. Climate Research 12 (1): 1-14.

Chiew, F. H. S. and T. A. McMahon. 1999. Modelling runoff and diffuse pollution loads in urban areas. Water Science and Technology 39 (12): 241-248.

200

Chiew, F. H., Mudgway, L. B., Duncan, H. P., & McMahon, T. A. 1998. Urban stormwater pollution: Cooperative Research Centre for Catchment Hydrology, Melbourne. Industry report 97/5.

Chiew, F.H.S, H.P Duncan and W Smith. 1997. Modelling pollutant build-up and wash-off: keep it simple. Paper presented at the Proceedings of the 24th International and Water Resources Symposium Auckland New Zealand Hydrological Society.

Chowdhury, Anupam. 2018. Statistical Framework for Determining Rainfall Parameters for Stormwater Quality Treatment Design. Brisbane, Australia: Queensland University of technology.

Christel, Prudhomme, Reynard Nick and Crooks Sue. 2002. Downscaling of global climate models for flood frequency analysis: where are we now? Hydrological Processes 16 (6): 1137-1150.

Clark Shirley, E. and Robert Pitt. 2009. Solids Removal in Storm-Water Filters Modeled Using a Power Equation. Journal of Environmental Engineering 135 (9): 896-899. Accessed 2018/05/24.

Clark, Peter U., Nicklas G. Pisias, Thomas F. Stocker and Andrew J. Weaver. 2002. The role of the thermohaline circulation in abrupt climate change. Nature 415: 863.

CMIP5. 2015. Coupled Model Inter-comparison Project: Coupled Model Inter-comparison Project.

Cofiño, Antonio, Daniel San-Martin and J. Gutiérrez. 2010. The ENSEMBLES Statistical Downscaling Portal. Vol. 12.

Corte-Real, J., B. Quian and H. Xu. 1999. Circulation patterns, daily precipitation in Portugal and implications for climate change simulated by the second Hadley Centre GCM. Climate Dynamics 15 (12): 921-935.

Coutts, Andrew M., Nigel J. Tapper, Jason Beringer, Margaret Loughnan and Matthias Demuzere. 2012. Watering our cities: The capacity for Water Sensitive Urban Design to support urban cooling and improve human thermal comfort in the Australian context. Progress in Physical Geography: Earth and Environment 37 (1): 2-28. Accessed 2018/05/24.

Daohua Bi,, Martin Dix, S. Marsland, Siobhan O'Farrell, Harun Rashid, P. Uotila, Hirst, et al. 2012. The ACCESS coupled model: Description, control climate and evaluation. Vol. 63.

201

Davis, A. P., Shokouhian, M., & Ni, S. 2001. Loading estimates of lead, copper, cadmium, and zinc in urban runoff from specific sources. Chemosphere, 44(5), 997-1009.

Davis, Allen P., Mohammad Shokouhian, Himanshu Sharma and Christie Minami. 2006. Water Quality Improvement through Bioretention Media: Nitrogen and Phosphorus Removal. Water Environment Research 78 (3): 284-293.

DCR. 1999. Virginia Stormwater Management Handbook. Vol. 1. Richmond Virginia: Department of Conservation and Recreation, Division of Soil and Water Conservation.

DEHP. 2016. Climate change in South East Queensland region. Brisbane, Queensland: Department of Environment and Heritage Protection, Queensland Government.

Deletic, Ana and David Orr. 2005. Pollution Build-up on Road Surfaces. Vol. 131.

Devi, G. K., Ganasri, B. P., & Dwarakish, G. S. 2015. A Review on Hydrological Models. Aquatic Procedia, 4, 1001-1007.

Diaz-Nieto, Jacqueline and Robert L. Wilby. 2005. A comparison of statistical downscaling and climate change factor methods: impacts on low flows in the River Thames, United Kingdom. Climatic Change 69 (2): 245-268.

Dibike, Yonas B. and Paulin Coulibaly. 2005. Hydrologic impact of climate change in the Saguenay watershed: comparison of downscaling methods and hydrologic models. Journal of Hydrology 307 (1): 145-163.

Dierberg, F. E., T. A. DeBusk, S. D. Jackson, M. J. Chimney and K. Pietro. 2002. Submerged aquatic vegetation-based treatment wetlands for removing phosphorus from agricultural runoff: response to hydraulic and nutrient loading. Water Research 36 (6): 1409-1422.

Dietz, Michael E. and John C. Clausen. 2005. A Field Evaluation of Rain Garden Flow and Pollutant Treatment. Water, Air, and Soil Pollution 167 (1): 123-138.

Donner, Leo J. and William G. Large. 2008. Climate Modeling. Annual Review of Environment and Resources 33: 1-17.

Dowdy, Andrew, Debbie Abbs, Jonas Bhend, Francis Chiew, John Church, Marie Ekström, Dewi Kirono, et al. 2015. East Coast Cluster Report, Climate Change in Australia Projections for Australia’s Natural Resource Management Regions: Cluster Reports: CSIRO and Bureau of Meteorology, Australia.

202

Duncan, H. 1999. Urban stormwater quality : a statistical overview / Hugh P. Duncan, edited by Hydrology Cooperative Research Centre for Catchment, Report (Cooperative Research Centre for Catchment Hydrology) ; 99/3. Clayton, Vic: CRC for Catchment Hydrology.

Dyer T. G. 1975. The assignment of Rainfall stations into homogeneous groups: An application of principal component analysis. Quarterly Journal of the Royal Meteorological Society 101 (430): 1005-1013.

Egodawatta, Prasanna, Evan Thomas and Ashantha Goonetilleke. 2007. Mathematical interpretation of pollutant wash-off from urban road surfaces using simulated rainfall. Water research 41 (13): 3025-3031.

Egodawatta, Prasanna. 2007. Translation of small plot pollution mobilisation and transport measurement from impermeable urban surface to urban catchment scale. PhD, Faculty of built environment and engineering, Queensland University of Technology.

Ehret, U., E. Zehe, V. Wulfmeyer, K. Warrach-Sagi and J. Liebert. 2012. HESS Opinions Should we apply bias correction to global and regional climate model data?. Hydrol. Earth Syst. Sci. 16 (9): 3391-3404.

EIA, US. 2002. Annual Energy Review 2002. United States: Energy Information Administration.

Ellis, Bryan, B Chocat, S Fujita, W Rauch and Marsalek J. 2004. Urban drainage a multilingual glossary. London: IWA.

Ellis, J. B., R. B. Shutes, D. M. Revitt and T. T. Zhang. 1994. Use of macrophytes for pollution treatment in urban wetlands. Resources, Conservation and Recycling 11 (1): 1-12.

Elshamy, M. E., I. A. Seierstad and A. Sorteberg. 2009. Impacts of climate change on Blue Nile flows using bias-corrected GCM scenarios. Hydrol. Earth Syst. Sci. 13 (5): 551-565.

EPA, ASCE & US. 2002. Urban Stormwater Best Management Practice (BMP) Performance Monitoring: A Guidance Manual for Meeting the National Stormwater BMP Database Requirements: American Society of Civil Engineers and US Environment Protection Agency.

Espey, W H, D E Winslow and C W Morgan. 1969. Urban effects on the unit hydrograph. In Second Water Resources Symposium, edited, 215-228.

Fakhri, Masoomeh , Mohammad Reza Farzaneh, Saeid Eslamian and Mohammad Javad Khordadi. 2013. Confidence Interval Assessment to Estimate Dry

203

and Wet Spells under Climate Change in Shahrekord Station, Iran. Journal of Hydrologic Engineering 18 (7): 911-918.

FAO-UN. 2006. Livestock's Long Shadow. Rome: Food and Agriculture Organization of The United Nations.

Feyerherm, A. M. and L. Dean Bark. 1967. Goodness of Fit of a Markov Chain Model for Sequences of Wet and Dry Days. Journal of Applied Meteorology 6 (5): 770-773.

Fiener, P. and K. Auerswald. 2005. Measurement and modeling of concentrated runoff in grassed waterways. Journal of Hydrology 301 (1): 198-215.

Fletcher, T. D., H. P Duncan, P Poelsma and S. D Lloyd. 2003. Stormwater Flow, Quality and Treatment: literature review, gap analysis and recommendations report. Melbourne, Victoria: NSW Environmental Protection Authority and Institute for Sustainable Water Resources, Department of Civil Engineering, Monash University.

Fowler H. and Kilsby C. G. 2003. A regional frequency analysis of United Kingdom extreme rainfall from 1961 to 2000. International Journal of Climatology 23 (11): 1313-1334.

Fowler H., Blenkinsop S. and Tebaldi C. 2007. Linking climate change modelling to impacts studies: recent advances in downscaling techniques for hydrological modelling. International Journal of Climatology 27 (12): 1547-1578.

Frost, Andrew J., Stephen P. Charles, Bertrand Timbal, Francis H. S. Chiew, R. Mehrotra, Kim C. Nguyen, Richard E. Chandler, et al. 2011. A comparison of multi-site daily rainfall downscaling techniques under Australian conditions. Journal of Hydrology 408 (1): 1-18.

Gabriel R., K. and Neumann J. 1962. A Markov chain model for daily rainfall occurrence at Tel Aviv. Quarterly Journal of the Royal Meteorological Society 88 (375): 90-95.

Gagnon, Sébastien, Bhawan Singh, Jean Rousselle and Luc Roy. 2005. An Application of the Statistical DownScaling Model (SDSM) to Simulate Climatic Data for Streamflow Modelling in Québec. Canadian Water Resources Journal / Revue canadienne des ressources hydriques 30 (4): 297-314.

Gardiner, Anne and Matthew Hardy. 2005. Beyond demonstration mode: The application of WSUD in Australia. Australian Planner 42 (4): 16-21.

204

GCCC. 2005. MUSIC Modelling Guidelines. Gold Coast: Gold Coast City Council.

Giorgi, F. 2010. Uncertainties in climate change projections, from the global to the regional scale. EPJ Web of Conferences 9: 115-129.

Glass, Charles and S. Bissouma. 2005. Evaluation of a parking lot bioretention cell for removal of stormwater pollutants. Vol. 81.

Gleckler P., Taylor K. E. and Doutriaux C. 2008. Performance metrics for climate models. Journal of Geophysical Research: Atmospheres 113 (D6).

Gnecco, I., C. Berretta, L. G. Lanza and P. La Barbera. 2005. Stormwater pollution in the urban environment of Genoa, Italy. Atmospheric Research 77 (1–4): 60-73. Accessed 2005/10.

Gobel, P., Dierkes, C., &Coldewey, W. 2007.Storm water runoff concentration matrix for urban areas.Journal of contaminant hydrology, 91(1), 26-42.

Goonetilleke, A, T Yingitcanlar, GA Ayoko and P Egodawatta. 2014. Sustainable Urban Water Envirinment - climate, pollution and adaption: Edward Elgar Publishing limited. Book.

Goonetilleke, A., & Thomas, E. C. 2003. Water quality impacts of urbanisation: Evaluation of current research. Technical Report, Centre for Built Environment and Engineering Research, Faculty of Built Environment and Engineering, pp. 1-93.

Goonetilleke, Ashantha, Evan Thomas, Simon Ginn and Dale Gilbert. 2005. Understanding the role of land use in urban stormwater quality management. Journal of Environmental Management 74 (1): 31-42.

Goudenhoofdt, Edouard, Laurent Delobbe and Patrick Willems. 2017. At-site and regional frequency analysis of extreme precipitation from radar-based estimates.

Goyal, Manish Kumar and C. S. P. Ojha. 2012. Downscaling of surface temperature for lake catchment in an arid region in India using linear multiple regression and neural networks. International Journal of Climatology 32 (4): 552-566.

Green, J., Karin Xuereb, Fiona Johnson, G. Moore and C. The. 2012. The revised intensity-frequency- duration (IFD) design rainfall estimates for Australia - An overview.

Greenway, Margaret. 2010. Wetlands and Ponds for Stormwater Treatment in Subtropical Australia: Their Effectiveness in Enhancing Biodiversity and

205

Improving Water Quality? Journal of Contemporary Water Research & Education 146 (1): 22-38.

Gregory, J. M., T. M. L. Wigley and P. D. Jones. 1993. Application of Markov models to area-average daily precipitation series and interannual variability in seasonal totals. Climate Dynamics 8 (6): 299-310.

Gunawardana, Chandima, Ashantha Goonetilleke, Prasanna Egodawatta, Les Dawes and Serge Kokot. 2012. Source characterisation of road dust based on chemical and mineralogical composition. Chemosphere 87 (2): 163-170.

Haerter, J., Stefan Hagemann, C. Moseley and C. Piani. 2010. Climate model bias correction and the role of timescales. Vol. 7.

Hajani, E., Rahman, A. 2018. Design rainfall estimation: Comparison between GEV and LP3 distributions and at-site and regional estimates. Natural Hazards, 93, 67-88.

Hamburg, M. 1970. Statistical analysis fotr decision making. 6 ed. Philadephia, USA: The Drydeb Press, Harcourt Brace college Publishers.

Hansen, Colin R. 2015. Comparison of regional and at-site frequency analysis methods for the estimation of southern Alberta extreme rainfall. Canadian Water Resources Journal / Revue canadienne des ressources hydriques 40 (4): 325-342.

Hanson Lars, S. and Richard Vogel. 2008. The Probability Distribution of Daily Rainfall in the United States.

Hashmi, Muhammad Zia, Asaad Y. Shamseldin and Bruce W. Melville. 2011. Comparison of SDSM and LARS-WG for simulation and downscaling of extreme precipitation events in a watershed. Stochastic Environmental Research and Risk Assessment 25 (4): 475-484.

Hassan, Badreldin G. H. and Feng Ping. 2012. Regional Rainfall Frequency Analysis for the Luanhe Basin – by Using L-moments and Cluster Techniques. APCBEE Procedia 1: 126-135.

Hatt, B.E., A. Deletic and T.D. Fletcher. 2007. Stormwater reuse: designing biofiltration systems for reliable treatment. Water Science and Technology 55 (4): 201-209.

Hatt, Belinda E., Tim D. Fletcher and Ana Deletic. 2008. Hydraulic and Pollutant Removal Performance of Fine Media Stormwater Filtration Systems. Environmental Science & Technology 42 (7): 2535-2541.

206

Hatt, Belinda E., Tim D. Fletcher and Ana Deletic. 2009. Hydrologic and pollutant removal performance of stormwater biofiltration systems at the field scale. Journal of Hydrology 365 (3): 310-321.

Hay, Lauren E., Wilby Robert L. and Leavesley George H. 2000. A Comparison of Delta Change and Downscaled GCM Scenarios for Three Mountainous Basins in the United States1. JAWRA Journal of the American Water Resources Association 36 (2): 387-397.

Hazeleger, W., X. Wang, C. Severijns, S. Ştefănescu, R. Bintanja, A. Sterl, K. Wyser, et al. 2012. EC-Earth V2.2: description and validation of a new seamless earth system prediction model. Climate Dynamics 39 (11): 2611-2629.

Herngren, L., Goonetilleke, A., &Ayoko, G. A. 2006. Analysis of heavy metals in road-deposited sediments. AnalyticaChimicaActa, 571(2), 270-278.

Herngren, L., Goonetilleke, A., Ayoko, G. A., &Mostert, M. M. 2010.Distribution of polycyclic aromatic hydrocarbons in urban stormwater in Queensland, Australia.Environmental pollution, 158(9), 2848-2856.

Herngren, Lars. 2005. Build-up and wash-off process kinetics of PAHs and heavy metals on paved surfaces using simulated rainfall. PhD, Queensland University of Technology.

Herron, A; , D; Stephens, R Nathan and Jayatilaka. 2011. Monte Carlo Temporal Patterns for Melbourne. Paper presented at the Proceedings of the 34th World Congress of the International Association for Hydro- Environment Research and Engineering: 33rd Hydrology and Water Resources Symposium and 10th Conference on Hydraulics in Water Engineering, Barton, A.C.T.: Engineers Australia.

Hessami, Masoud, Philippe Gachon, Taha B. M. J. Ouarda and André St-Hilaire. 2008. Automated regression-based statistical downscaling tool. Environmental Modelling & Software 23 (6): 813-834.

Hewitson, Bruce and Robert Crane. 1996. Climate Downscaling: Techniques and Application. Vol. 07.

Hoffman, E. J., Latimer, J. S., Hunt, C. D., Mills, G. L., & Quinn, J. G. 1985. Stormwater runoff from highways. Water, Air, and Soil Pollution, 25(4), 349-364.

Holper, Paul N. 2012. Climate Change Science, edited by Department of climate change and energy efficiency; and BoM: CSIRO.

207

Hong, Eunyoung, Eric A. Seagren and Allen P. Davis. 2006. Sustainable Oil and Grease Removal from Synthetic Stormwater Runoff Using Bench-Scale Bioretention Studies. Water Environment Research 78 (2): 141-155.

Hosking, J.R.M. and J.R. Wallis. 1997. Regional Frequency Analysis: An Approach Based on L-Moments: Cambridge University Press.

Hosking, J.R.M. and J.R. Wallis. 2005. Regional Frequency Analysis: An Approach Based on L-Moments: Cambridge University Press.

Hughes, James P., Peter Guttorp and Stephen P. Charles. 1999. A Non-Homogeneous Hidden Markov Model for Precipitation Occurrence. Journal of the Royal Statistical Society. Series C (Applied Statistics) 48 (1): 15-30.

Hughes, Lesley. 2003. Climate change and Australia: Trends, projections and impacts. Austral Ecology 28 (4): 423-443.

Hunt, W. F., A. R. Jarrett, J. T. Smith and L. J. Sharkey. 2006. Evaluating Bioretention Hydrology and Nutrient Removal at Three Field Sites in North Carolina. Journal of Irrigation and Drainage Engineering 132 (6): 600-608.

Hunt, William F. 2003. Pollutant removal evaluation and hydraulic characterization for bioretention stormwater treatment devices Department of Agricultural and Biological Engineering, The Pennsylvania State University.

Husak Gregory, Michaelsen Joel and Funk Chris. 2007. Use of the gamma distribution to represent monthly rainfall in Africa for drought monitoring applications. International Journal of Climatology 27 (7): 935-944.

Huth, R. 2000. A circulation classification scheme applicable in GCM studies. Theoretical and Applied Climatology 67 (1): 1-18.

Hwang, H.-M., & Foster, G. D. 2006. Characterization of polycyclic aromatic hydrocarbons in urban stormwater runoff flowing into the tidal Anacostia River, Washington, DC, USA. Environmental Pollution, 140(3), 416-426.

Ines, Amor V. M. and James W. Hansen. 2006. Bias correction of daily GCM rainfall for crop simulation studies. Agricultural and Forest Meteorology 138 (1): 44-53.

IPCC. 2000. IPCC SPECIAL REPORT EMISSIONS SCENARIOS: Intergovernmental Panel on Climate Change.

208

IPCC. 2007. Climate Change 2007: Synthesis Report. Contribution of Working Groups I, II and III to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Geneva, Switzerland: IPCC.

IPCC. 2008. Climate Change and Water, edited by Bryson Bates; Zbigniew W. Kundzewicz; Shaohong Wu; Jean Paulutokof. Geneva: Intergovenmental Panel on Climate Change.

IPCC. 2014. Summary for Policymakers. In Climate Change 2014: Impacts, Adaptation, and Vulnerability. Part A: Global and Sectoral Aspects. Contribution of Working Group II to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change, edited by C. B. Field, V. R. Barros, D. J. Dokken, K. J. Mach, M. D. Mastrandrea, T. E. Bilir, M. Chatterjee, K. L. Ebi, Y. O. Estrada, R. C. Genova, et al., 1-32. Cambridge, United Kingdom, and New York, NY, USA: Cambridge University Press.

Jartun, Morten, R. T. Ottesen, Eiliv Steinnes and Tore Volden. 2008. Runoff of particle bound pollutants from urban impervious surfaces studied by analysis of sediments from stormwater traps. Vol. 396.

Jaw, Tsou, Jialun Li, Kuo-lin Hsu, Soroosh Sorooshian and Fatima Driouech. 2015. Evaluation for Moroccan dynamically downscaled precipitation from GCM CHAM5 and its regional hydrologic response. Journal of Hydrology: Regional Studies 3: 359-378.

Jefferies, C., A. Duffy, N. Berwick, N. McLean and A. Hemingway. 2009. Sustainable Urban Drainage Systems (SUDS) treatment train assessment tool. Water Science and Technology 60 (5): 1233-1240.

Jenkins, Graham A. and Margaret Greenway. 2005. The hydraulic efficiency of fringing versus banded vegetation in constructed wetlands. Ecological Engineering 25 (1): 61-72.

Johnson, Fiona and Ashish Sharma. 2009. Measurement of GCM Skill in Predicting Variables Relevant for Hydroclimatological Assessments. Journal of Climate 22 (16): 4373-4382. Accessed 2018/05/31.

Kalnay, E., M. Kanamitsu, R. Kistler, W. Collins, D. Deaven, L. Gandin, M. Iredell, et al. 1996. The NCEP/NCAR 40-Year Reanalysis Project. Bulletin of the American Meteorological Society 77 (3): 437-472. Accessed 2018/05/31.

Kayhanian, Masoud, Ehsan Rasa, Akshay Vichare and E. Leatherbarrow Jon. 2008. Utility of Suspended Solid Measurements for Storm-Water Runoff

209

Treatment. Journal of Environmental Engineering 134 (9): 712-721. Accessed 2018/05/24.

Khana, Said, Ijaz Hussain, Muhammad Faisal, Yousaf shad Muhammad, Alaa Mohamd Shoukryc and Tajammal Hussain. 2017. Regional Frequency Analysis of Extremes Precipitation Using L-Moments and Partial L-Moments. Vol. 2017.

Kilsby, C. G., P. D. Jones, A. Burton, A. C. Ford, H. J. Fowler, C. Harpham, P. James, A. Smith and R. L. Wilby. 2007. A daily weather generator for use in climate change studies. Environmental Modelling & Software 22 (12): 1705-1719.

Kim, J-W., J-T. Chang, N. L. Baker, D. S. Wilks and W. L. Gates. 1984. The Statistical Problem of Climate Inversion: Determination of the Relationship between Local and Large-Scale Climate. Monthly Weather Review 112 (10): 2069-2077.

Knight, Robert L., Victor W. E. Payne, Robert E. Borer, Ronald A. Clarke and John H. Pries. 2000. Constructed wetlands for livestock wastewater management. Ecological Engineering 15 (1): 41-55.

Kohler, E. A., V. L. Poole, Z. J. Reicher and R. F. Turco. 2004. Nutrient, metal, and pesticide removal during storm and nonstorm events by a constructed wetland on an urban golf course. Ecological Engineering 23 (4): 285-298.

Koutsoyiannis, Demetris. 2004. Statistics of extremes and estimation of extreme rainfall: I. Theoretical investigation / Statistiques de valeurs extrêmes et estimation de précipitations extrêmes: I. Recherche théorique. Hydrological Sciences Journal 49 (4): null-590.

Kumar Mishra, Binaya and Srikantha Herath. 2015. Assessment of Future Floods in the Bagmati River Basin of Nepal Using Bias-Corrected Daily GCM Precipitation Data. Journal of Hydrologic Engineering 20 (8): 05014027. Accessed 2018/05/23.

Laurenson, E. M. 1964. A catchment storage model for runoff routing. Journal of Hydrology 2 (2): 141-163.

Laurenson, EM and RG Mein. 1995. RORB- Version 4- Runoff Routing program, User Manual 4. Victoria, Australia: Department of Civil Engineering, Monash University.

Leduc, Martin and René Laprise. 2009. Regional climate model sensitivity to domain size. Climate Dynamics 32 (6): 833-854.

210

Li, Yun and Ian Smith. 2009. A Statistical Downscaling Model for Southern Australia Winter Rainfall. Journal of Climate 22 (5): 1142-1158.

Lin, Gwo-Fong and Lu-Hsien Chen. 2006. Identification of homogeneous regions for regional frequency analysis using the self-organizing map. Journal of Hydrology 324 (1): 1-9.

Liu, An, Environment Queensland University of Technology. Faculty of Built and Engineering. 2011. Influence of rainfall and catchment characteristics on urban stormwater quality. Dissertation/Thesis.

Liu, Q. Q., L. Chen, J. C. Li and V. P. Singh. 2004. Two-dimensional kinematic wave model of overland-flow. Journal of Hydrology 291 (1): 28-41.

Lloyd, Sara, T. H. F. Wong and Christopher J. Chesterfield. 2002. Water Sensitive Urban Design: a stormwater management perspective. Clayton, Vic: CRC for Catchment Hydrology.

Loveridge, M., Rahman, A. 2014. Quantifying uncertainty in rainfall-runoff models due to design losses using Monte Carlo simulation: A case study in New South Wales, Australia, Stochastic Environment Research & Risk Assessment, 28, 8, 2149-2159.

Loveridge, M., Rahman, A., Hill, P. and Babister, M. 2013. Investigation into probabilistic losses for design flood estimation: A case study for the Orara River catchment, NSW, Australian Journal of Water Resources, 17(1), 13-24.

LSRC. 2001. Stormwater Control Principles and Practices, Stormwater Fact Sheet No. 2. http://www.landofsky.org.

Lucke, Terry, Darren Drapper and Andy Hornbuckle. 2018. Urban stormwater characterisation and nitrogen composition from lot-scale catchments — New management implications. Science of the Total Environment 619-620: 65-71.

Ma, Y. 2016. Human Health Risk of Toxic Chemical Pollutants Generated From Traffic and Land Use Activities. PhD, Queensland University of Technology, Australia.

Madsen, Henrik, Karsten Arnbjerg-Nielsen and Peter Steen Mikkelsen. 2009. Update of regional intensity–duration–frequency curves in Denmark: Tendency towards increased storm intensities. Atmospheric Research 92 (3): 343-349.

211

Mahbub, Parvez, Godwin A. Ayoko, Ashantha Goonetilleke, Prasanna Egodawatta and Serge Kokot. 2010. Impacts of Traffic and Rainfall Characteristics on Heavy Metals Build-up and Wash-off from Urban Roads. Environmental Science & Technology 44 (23): 8904-8910.

Mangangka, Isri Ronald. 2013. Role of hydraulic factors in constructed wetland and bioretention basin treatment performance.

Mansell, M. G. 2003a. Rural and urban hydrology, Rural and urban hydrology.

Mansell, M. G. 2003b. Rural urban hydrology.

Martens, S., R. Perrigo, A. Torre, L Chalmers, E. Monk, J MacKay and B Till. 2007. 'Stormwater Management Manual for Western Australia: Structural Controls: Department of Water, Government of Western Australia.

MATLAB. 2010. MATLAB. R2010a ed. Natick, MA: The MathWorks.

MBW, BCC &. 2006. Water Sensitive Urban Design Technical Design Guidelines for South East Queensland. Vol. 1: South East Queensland Healthy Waterways Partnership and Brisbane City Council.

McAuley, A and Knights D. 2009. Water sensitive urban design Stormwater quality modelling guide. Darwin NT: Northern Territory Department of Planning and Infrastructure.

Mearns, L. O., I. Bogardi, F. Giorgi, I. Matyasovszky and M. Palecki. 1999. Comparison of climate change scenarios generated from regional climate model experiments and statistical downscaling. Journal of Geophysical Research: Atmospheres 104 (D6): 6603-6621.

Mearns, L. O., W. Easterling, C. Hays and D. Marx. 2001. Comparison of Agricultural Impacts of Climate Change Calculated from High and Low Resolution Climate Change Scenarios: Part I. The Uncertainty Due to Spatial Scale. Climatic Change 51 (2): 131-172.

Mehrotra R., and Sharma Ashish. 2006. A nonparametric stochastic downscaling framework for daily rainfall at multiple locations. Journal of Geophysical Research: Atmospheres 111 (D15).

Michelangeli P.‐A., Vrac M. and Loukos H. 2009. Probabilistic downscaling approaches: Application to wind cumulative distribution functions. Geophysical Research Letters 36 (11).

Miguntanna, N. P. 2009. Nutrients build-up and wash-off processes in urban land uses. Queensland University of technology, Brisbane. PhD Thesis.

212

Millar Robert, G. 1999. Analytical Determination of Pollutant Wash-Off Parameters. Journal of Environmental Engineering 125 (10): 989-992. Accessed 2018/05/25.

Mitsch, W. J. and J. G. Gosselink. 1993. Wetlands New York: Van Nostrand Reinhold.

Moen, Jon and Peter Fredman. 2007. Effects of Climate Change on Alpine Skiing in Sweden. Vol. 15.

Moilleron, R., Gonzalez, A., Chebbo, G., & Thévenot, D. R. 2002. Determination of aliphatic hydrocarbons in urban runoff samples from the “Le Marais” experimental catchment in Paris centre. Water Research, 36(5), 1275-1285.

Mpelasoka, F. S., S. Charles, F. H. Chiew, G. Fu and S. Beecham. 2012. Assessment of CMIP5 GCM daily predictor variables for statistical downscaling. Paper presented at the EGU General Assembly 2012, Vienna, Austria.

Muñoz-Diaz, D. and F. S. Rodrigo. 2004. Spatio-temporal patterns of seasonal rainfall in Spain (1912-2000) using cluster and principal component analysis: comparison. Annales Geophysicae 22 (5): 1435-1448.

NCAR. 2009. Methods of Downscaling Future Climate Information and Applications. In NARCCAP Users’ Meeting edited by Linda O. Mearns: National Center for Atmospheric Research

Nelson, E. J., & Booth, D. B. (2002). Sediment sources in an urbanizing, mixed land-use watershed. Journal of Hydrology, 264(1), 51-68.

Ngongondo, Cosmo S., Chong-Yu Xu, Lena M. Tallaksen, Berhanu Alemaw and Tobias Chirwa. 2011. Regional frequency analysis of rainfall extremes in Southern Malawi using the index rainfall and L-moments approaches. Stochastic Environmental Research and Risk Assessment 25 (7): 939-955.

Novotny, Vladimir, Hung-Ming Sung, Roger Bannerman and Ken Baum. 1985. Estimating Nonpoint Pollution from Small Urban Watersheds. Journal (Water Pollution Control Federation) 57 (4): 339-348.

O’Loughlin, G and B Stack. 2004. DRAINS user Manual. Sydney, Australia.

Oliver, R. L., & Boon, P. I. 1992. Bioavailability of nutrients in turbid waters. Land and Water Resources Research and Development Corporation, Project No. 88/74, pp. 1-6.

213

Oppenheimer, Clive. 2002. Limited global change due to the largest known Quaternary eruption, Toba ≈74kyr BP? Quaternary Science Reviews 21 (14): 1593-1609.

Page, Christian, Christian, Laurent Terray, Laurent, Boé and Julien. 2009. dsclim: A software package to downscale climate scenarios at regional scale using a weather-typing based statistical methodology.

Parker, Nathaniel , Ted Gardner, Ashantha Goonetilleke, Prasanna K Egodawatta and Daniel Giglio. 2009. Effectiveness of WSUD in the ‘real world’. Paper presented at the 6th International Water Sensitive Urban Design Conference and the 3rd Hydropolis : Towards Water Sensitive Cities and Citizens, Parmelia Hilton, Perth, Western Australia.

Pearson, C. P. 1991. New Zealand Regional Flood Frequency Analysis Using L-Moments. Journal of Hydrology (New Zealand) 30 (2): 53-64.

Perkins, S. E., A. J. Pitman, N. J. Holbrook and J. McAneney. 2007. Evaluation of the AR4 Climate Models’ Simulated Daily Maximum Temperature, Minimum Temperature, and Precipitation over Australia Using Probability Density Functions. Journal of Climate 20 (17): 4356-4376.

Pilgrim, D.H. and Australia Institution of Engineers. 1987. Australian Rainfall and Runoff: A Guide to Flood Estimation: Institution of Engineers, Australia.

Pitt, R., Bannerman, R., Clark, S., & Williamson, D. 2004. Sources of Pollutants in Urban Areas (Part 2)–Recent Sheetflow Monitoring, Effective Modeling of Urban Water Systems, Monograph 13. pp. 485-506..

QUDM. 2017. Queensland Urban Drainage Manual. Australia: Institute of Public Works Engineering Australasia, Queensland (IPWEAQ).

Quinton, J. N., Catt, J. A., & Hess, T. M. 2001. The selective removal of phosphorus from soil. Journal of Environmental Quality, 30(2), 538-545

R Development Core Team. 2010. An Introduction to R.

R Development Core team. 2016. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing. Vienna, Austria.

Raes, Dirk, K.U.Leuven and Inter-University Programme in Water Resources Engineering (IUPWARE). 2004. Frequency analysis of rainfall data Leuven, Belgium.

Rasheed, A.M. , M.M.A. Mansoor, M.H.A. Ahmath and S.M. Shameer. 2014. Nutrient Removal in Hybrid Constructed Wetlands. International Journal of Scientific & Engineering Research 5 (2): 1004-1006.

214

Rasheed, AM, P Egodawatta, A Goonetilleke and J McGree. 2017. spdownscale: Spatial Downscaling Using Bias Correction Approach, R package 0.1.

Reinelt, Lorin E. and Richard R. Horner. 1995. Pollutant removal from stormwater runoff by palustrine wetlands based on comprehensive budgets. Ecological Engineering 4 (2): 77-97.

Reynard, N S, Crooks S M and A L Kay. 2004. Impact of climate change on flood flows in river catchments.

Richardson. 1981. Stochastic simulation of daily precipitation, temperature, and solar radiation. Water Resources Research 17 (1): 182-190.

Rotstayn D. Leon, Collier Mark A., Dix Martin R., Feng Yan, Gordon Hal B., O'Farrell Siobhan P., Smith Ian N. and Syktus Jozef. 2010. Improved simulation of Australian climate and ENSO‐related rainfall variability in a global climate model with an interactive aerosol treatment. International Journal of Climatology 30 (7): 1067-1088.

Sachindra A., D., Huang F., Barton A. and Perera B. J. C. 2014. Statistical downscaling of general circulation model outputs to precipitation—part 1: calibration and validation. International Journal of Climatology 34 (11): 3264-3281.

Sajjad, Khan Mohammad, Coulibaly Paulin and Dibike Yonas. 2006. Uncertainty analysis of statistical downscaling methods using Canadian Global Climate Model predictors. Hydrological Processes 20 (14): 3085-3104.

Sansalone, J., J. & Tittlebaum, M., E. 2001. Storm water transport ofparticulate matter from elevated urban transportation corridors intowaterways of Louisiana: The role partitioning and implications for treatment,Lousiana. Lousiana, USA: Water Resources Research Institute, Annual Technical Report.

Sarma, P. B. S., J. W. Delleur and A. R. Rao. 1973. Comparison of rainfall-runoff models for urban areas. Journal of Hydrology 18 (3): 329-347.

Sartor, James D., Gail B. Boyd and Franklin J. Agardy. 1974. Water Pollution Aspects of Street Surface Contaminants. Journal (Water Pollution Control Federation) 46 (3): 458-467.

Sato, Tomonori, Fujio Kimura and Akio Kitoh. 2007. Projection of global warming onto regional precipitation over Mongolia using a regional climate model. Journal of Hydrology 333 (1): 144-154.

215

Sawyer, R. F., Harley, R. A., Cadle, S., Norbeck, J., Slott, R., & Bravo, H. 2000. Mobile sources critical review: 1998 NARSTO assessment. Atmospheric Environment, 34(12), 2161-2181.

Schaefer, M. G. 1990. Regional analyses of precipitation annual maxima in Washington State. Water Resources Research 26 (1): 119-131. Accessed 2018/05/23.

Schmidli J.,, Goodess C. M., Frei C., Haylock M. R., Hundecha Y., Ribalaygua J. and Schmith T. 2007. Statistical and dynamical downscaling of precipitation: An evaluation and comparison of scenarios for the European Alps. Journal of Geophysical Research: Atmospheres 112 (D4).

Scholes, Lian, D. Michael Revitt and J. Bryan Ellis. 2008. A systematic approach for the comparative assessment of stormwater pollutant removal potentials. Journal of Environmental Management 88 (3): 467-478.

Scholz, Miklas. 2006. Chapter 2 - Water treatment. In Wetland Systems to Control Urban Runoff, edited by Miklas Scholz, 7-9. Amsterdam: Elsevier.

Schoof, J. T. and S. C. Pryor. 2001. Downscaling temperature and precipitation: a comparison of regression-based methods and artificial neural networks. International Journal of Climatology 21 (7): 773-790.

Schulzweida U, L. Kornblueh and R. Quast. 2009. CDO User's Guide: Climate Data Operators Version 1.4.1.

Seidenglanz, A and Uni Bremen. 2012. Panoply-A Tool for Visualizing NetCDF-Formatted Model Output.

Semenov, Mikhail A and Elaine M Barrow. 2002. LARS-WG: A Stochastic Weather Generator for Use in Climate Impact Studies 3.0 Rothamsted Research, Harpenden, Hertfordshire, AL5 2JQ, UK and Canadian Climate Impacts Scenarios (CCIS) Project, Canada.

Semenov, Mikhail and Elaine M Barrow. 2002. LARS-WG A Stochastic Weather Generator for Use in Climate Impact Studies.

Shamshad, A., M. A. Bawadi, W. Wanhussin, Taksiah Majid and Mohd Sanusi S. Ahamad. 2005. First and second order Markov chain models for synthetic generation of wind speed time-series. Vol. 30.

Sharkey, Lucas John. 2006. The Performance of Bioretention Areas in North Carolina: A Study of Water Quality, Water Quantity, and Soil Media, Biological and Agricultural Engineering, North Carolina State Univ., Raleigh, N.C.

216

Sharma, Ashish, Rajeshwar Mehrotra, Jingwan Li and Sanjeev Jha. 2016. A programming tool for nonparametric system prediction using Partial Informational Correlation and Partial Weights. Vol. 83.

Sharma, D., A. Das Gupta and M. S. Babel. 2007. Spatial disaggregation of bias-corrected GCM precipitation for improved hydrologic simulation: Ping River Basin, Thailand. Hydrology and Earth System Sciences Discussions 11 (4): 1373-1390.

Sharma, Mohita and J. B. Singh. 2010. Use of probability distribution in rainfall analysis. Vol. 3.

Sharma, Mukul. 2002. Variations in solar magnetic activity during the last 200 000 years: is there a Sun–climate connection? Earth and Planetary Science Letters 199 (3–4): 459-472.

Sherrard, R. M., J. S. Bearr, C. L. Murray-Gulde, J. H. Rodgers and Y. T. Shah. 2004. Feasibility of constructed wetlands for removing chlorothalonil and chlorpyrifos from aqueous mixtures. Environmental Pollution 127 (3): 385-394.

Sieker, H. and M. Klein. 1998. Best Management Practices for stormwater-runoff with alternative methods in a large urban catchment in Berlin, Germany. Water Science and Technology 38 (10): 91-97.

Singh, Deepti, Michael Tsiang, Bala Rajaratnam and Noah Diffenbaugh. 2014. Observed changes in extreme wet and dry spells during the South Asian summer monsoon season. Vol. 4.

Snell, Seth E., Sucharita Gopal and Robert K. Kaufmann. 2000. Spatial Interpolation of Surface Air Temperatures Using Artificial Neural Networks: Evaluating Their Use for Downscaling GCMs. Journal of Climate 13 (5): 886-895.

STARDEX. 2005. Downscaling climate extremes. Climatic Research Unit, School of Environmental Sciences, University of East Anglia. UK.

Stocker, Thomas. 2011. Introduction to Climate Modelling: Springer-Verlag.

Storch, Hans von, Eduardo Zorita and Ulrich Cubasch. 1993. Downscaling of Global Climate Change Estimates to Regional Scales: An Application to Iberian Rainfall in Wintertime. Journal of Climate 6 (6): 1161-1171.

Sugiyama, Hironobu, Mutsumi Kadoya, Akihiro Nagai and Kevin Lansey. 1997. Evaluation of the storage function model parameter characteristics. Vol. 191.

217

Syktus, Jozef I. 2009. Climate change in the South East Queensland Region: Queensland Government.

Tan, P. N., Michael Steinbach and Vipin Kumar. 2005. Cluster Analysis: Basic Concepts and Algorithms.

Taylor, Andre and Tony Wong. 2002. Non-structural stormwater quality bestmanagement practices - an overview oftheir use, value, cost and evaluation. 02 vols: Cooperative research centre for catchment hydrology.

Taylor, André, Rob Curnow, Tim Fletcher and Justin Lewis. 2007. Education campaigns to reduce stormwater pollution in commercial areas: Do they work? Journal of Environmental Management 84 (3): 323-335.

Taylor, Geoff D., Tim D. Fletcher, Tony H. F. Wong, Peter F. Breen and Hugh P. Duncan. 2005. Nitrogen composition in urban runoff—implications for stormwater management. Water Research 39 (10): 1982-1989.

Teutschbein, Claudia and Jan Seibert. 2012. Is Bias Correction of Regional Climate Model (RCM) Simulations Possible for Non-Stationary Conditions? Vol. 9.

Thurston, KerryA. 1999. Lead and petroleum hydrocarbon changes in an urban wetland receiving stormwater runoff. Ecological Engineering 12 (3): 387-399.

Timbal, B, E Fernandez and Z Li. 2009. Generalization of a statistical downscaling model to provide local climate change projections for Australia. Environmental Modelling & Software 24 (3): 341-358.

Timbal, B. and B. J. McAvaney. 2001. An analogue-based method to downscale surface air temperature: application for Australia. Climate Dynamics 17 (12): 947-963.

Trefry Christopher, M., W. Watkins David and Dennis Johnson. 2005. Regional Rainfall Frequency Analysis for the State of Michigan. Journal of Hydrologic Engineering 10 (6): 437-449. Accessed 2018/05/24.

Trzaska, Sylwia and Emilie Schnarr. 2014. A Review of Downscaling Methods for Climate Change Projections.

Unal, Yurdanur, Tayfun Kindap and Mehmet Karaca. 2003. Redefining Climate Zones for Turkey Using Cluster Analysis. Vol. 23.

van Vuuren, Detlef P., Jae Edmonds, Mikiko Kainuma, Keywan Riahi, Allison Thomson, Kathy Hibbard, George C. Hurtt, et al. 2011. The

218

representative concentration pathways: an overview. Climatic Change 109 (1): 5.

Vaze, J. and Francis H. S. Chiew. 2002. Experimental study of pollutant accumulation on an urban road surface. Urban Water 4 (4): 379-389.

Vaze, Jai;, Francis HS; Chiew, Lynda; Suryadi and Khagendra Khanal. 2000. Pollutant Accumulation on an Urban Road Surface. In National conference publication (Institution of Engineers, Australia), Barton, A.C.T, edited: Institution of Engineers, Australia

Venables N., W. and D. M. Smith. 2008. An introduction to R: a programming environment for data analysis and graphics.

Viglione, Alberto. 2012. homtest: Homogeneity tests for Regional Frequency Analysis.

Von Storch, Hans 1999. On the Use of Inflation in Statistical Downscaling. Journal of Climate 12 (12): 3505-3506.

Vrac, M., P. Marbaix, D. Paillard and P. Naveau. 2007. Non-linear statistical downscaling of present and LGM precipitation and temperatures over Europe. Climate of the Past 3 (4): 669-682.

VSC. 1999. Urban stormwater: best practice environmental management guidelines. Melbourne, Australia: Victorian Stormwater Committee -CSIRO Publishing.

Walker, David J. and Sigrid Hurl. 2002. The reduction of heavy metals in a stormwater wetland. Ecological Engineering 18 (4): 407-414.

Walsh Kevin J.E., and Syktus Jozef. 2003. Simulations of observed interannual variability of tropical cyclone formation east of Australia. Atmospheric Science Letters 4 (1‐4): 28-40.

Warren, N., Allan, I., Carter, J., House, W., & Parker, A. 2003. Pesticides and other micro-organic contaminants in freshwater sedimentary environments—a review. Applied Geochemistry, 18(2), 159-194.

Water, Melbourne. 2005. WSUD Engineering Procedures: Stormwater: CSIRO Publishing.

Watterson, Ian;, Tony; Hirst and Leon Rotstayn. 2013. A skill-score based evaluation of simulated Australian climate. Australian Meteorological and Oceanographic Journal 63: 181-190.

219

Wayne, Graham. 2013. The Beginner's Guide to Representative Concentration Pathways.

WBD. 2010a. MUSIC Modelling Guidelines. Vol. 1.0. Brisbane, Queensland: South East Queensland Healthy Waterways Partnership.

WBD. 2010b. Water by Design - A Business Case for Best Practice Urban Stormwater Management Vol. 1.1. Brisbane, Queensland: South East Queensland Healthy Waterways Partnership.

Whelans, C, Halpern Glick Maunsell and Thompson Palmer. 1994. Planning & management guidelines for water sensitive urban (residential) design : consultants report prepared for the Department of Planning and Urban Development, the Water Authority of Western Australia and the Environmental Protection Authority. Mount Hawthorn, W.A: Whelans and Halpern Glick Maunsell in association with Thompson Palmer and Institute for Science and Technology Policy, Murdoch University.

White, Christopher J., Kathleen L. McInnes, Robert P. Cechet, Stuart P. Corney, Michael R. Grose, Gregory K. Holz, Jack J. Katzfey and Nathaniel L. Bindoff. 2013. On regional dynamical downscaling for the assessment and projection of temperature and precipitation extremes across Tasmania, Australia. Climate Dynamics 41 (11): 3145-3165.

Wigley, T. M. L.;, P. D. Jones;, K. R. Briffa; and G. Smith. 1990. Obtaining sub‐grid‐scale information from coarse‐resolution general circulation model output. Journal of Geophysical Research: Atmospheres 95 (D2): 1943-1953.

Wilby L.R. and Dawson C. W. 2013. The Statistical DownScaling Model: insights from one decade of application. International Journal of Climatology 33 (7): 1707-1719.

Wilby L.R. and Harris I. 2006. A framework for assessing uncertainties in climate change impacts: Low‐flow scenarios for the River Thames, UK. Water Resources Research 42 (2).

Wilby L.R., Beven K. J. and Reynard N. S. 2008. Climate change and fluvial flood risk in the UK: more of the same? Hydrological Processes 22 (14): 2511-2523.

Wilby L.Robert, Hay Lauren E., Gutowski William J., Arritt Raymond W., Takle Eugene S., Pan Zaitao, Leavesley George H. and Clark Martyn P. 2000. Hydrological responses to dynamically and statistically downscaled climate model output. Geophysical Research Letters 27 (8): 1199-1202.

220

Wilby R. L., T. M. L. Wigley, D. Conway, P. D. Jones, B. C. Hewitson, J. Main and D. S. Wilks. 1998. Statistical downscaling of general circulation model output: A comparison of methods. Water Resourse Research 34 (11): 2995-3008.

Wilby, R.L. and T.M.L. Wigley. 1997. Downscaling general circulation model output: a review of methods and limitations. Progress in physical geography 21 (4): 530-548.

Wilby, Robert L. and Christian W. Dawson. 2004. Using SDSM Version 3.1 — A decision support tool for the assessment of regional climate change impacts

Wilby, Robert, Stephen Charles, Eduardo Zorita, Bertrand Timbal, Penny Whetton and Linda Mearns. 2004. Guidelines For Use of Climate Scenarios Developed From Statistical Downscaling Methods.

Wilks, D. S. 1999. Interannual variability and extreme-value characteristics of several stochastic daily precipitation models. Agricultural and Forest Meteorology 93 (3): 153-169.

Wilks, D. S. and R. L. Wilby. 1999. The weather generation game: a review of stochastic weather models. Progress in Physical Geography: Earth and Environment 23 (3): 329-357.

Wong, Pak Chung, Shen Han-Wei, R. Leung, S. Hagos, T. Y. Lee, Tong Xin and K. Lu. 2014. Visual analytics of large-scale climate model data. In 2014 IEEE 4th Symposium on Large Data Analysis and Visualization (LDAV), 9-10 Nov. 2014, edited, 85-92.

Wong, T. H. F., Peter F. Breen and Sara Lloyd. 2000. Water sensitive road design : design options for improving stormwater quality of road runoff / Tony H. F. Wong, Peter F. Breen and Sara D. Lloyd, Report (Cooperative Research Centre for Catchment Hydrology) ; 00/1. Clayton, Vic: CRC for Catchment Hydrology.

Wong, THF. 2006. An Overview of Water Sensitive Urban Design Practices in Australia. In Water Practice & Technology © CSIRO 2006, edited.

Yoo, Chulsang, Jinwook Lee and Yonghun Ro. 2016. Markov Chain Decomposition of Monthly Rainfall into Daily Rainfall: Evaluation of Climate Change Impact. Vol. 2016.

Yu, Pao-Shan, Tao-Chang Yang and Chih-Kang Wu. 2002. Impact of climate change on water resources in southern Taiwan. Journal of Hydrology 260 (1): 161-175.

221

Zoppou, Christopher. 2001. Review of urban stormwater models. Environmental Modelling & Software 16 (3): 195-231.

Zorita, Eduardo, James P. Hughes, Dennis P. Lettemaier and Hans von Storch. 1995. Stochastic Characterization of Regional Circulation Patterns for Climate Model Diagnosis and Estimation of Local Precipitation. Journal of Climate 8 (5): 1023-1042. Accessed 2018/05/31.

Zucchini, W. and P Guttorp. 1991. A Hidden Markov Model for Space‐Time Precipitation. Water Resources Research 27 (8): 1917-1923.

223

Appendix A Table A.1: The summary of the selected meteorological stations for the regional rainfall homogeneity assessment.

No Station No

Station Latitude Longitude No of events (2011-2015)

1 40004 Amberley AMO -27.6297 152.7111 265 2 40043 Cape Moreton Lighthouse -27.0314 153.4661 350 3 40082 University Of Queensland -27.5436 152.3375 274 4 40093 Gympie -26.1831 152.6414 379 5 40211 Archerfield Airport -27.5717 153.0078 358 6 40717 Coolangatta -28.1681 153.5053 501 7 40764 Gold Coast Seaway -27.939 153.4283 455 8 40842 Brisbane Aero -27.3917 153.1292 377 9 40861 Sunshine Coast Airport -26.6006 153.0903 471

10 40908 Tewantin RSL Park -26.3911 153.0403 426 11 40913 Brisbane -27.4808 153.0389 380 12 40922 Kingaroy Airport -26.5737 151.8398 249 13 40958 Redcliffe -27.2169 153.0922 385 14 40983 Beaudesert Drumley St -27.9707 152.9898 297 15 40988 Nambour Daff - Hillside -26.6442 152.9383 506 16 41525 Warwick -28.2061 152.1003 213 17 41529 Toowoomba Airport -27.5425 151.9134 275

225

Appendix B

B.1: Source code for ParaCal() function of spdownscale

#' @title Calibration Parameters #' @description Displays the shape factors, scale factors and the threshold values of the observation and GCM/RCM data set which ultimately define the model #' @param obs_c vector of observational climate data (rainfall) used for calibrating the model #' @param mod_c vector of GCM/RCM climate data (rainfall) used for calibrating the model #' @param obs_v vector of observational climate data (rainfall) used for validating the model #' @param mod_v vector of GCM/RCM climate data (rainfall) used for validating the model #' @param mod_fut vector of GCM/RCM future climate data (rainfall) need to be downscaled #' @details #' #'1) Dry-days correction / Defining threshold values #' #' The relationship between the cumulative frequencies (thresholds) corresponding to the dry days of GCM/RCM data and that of the observational data is defined by a polynomial function given by; #' #'threshold_obs = (threshold_mod)^n #' #'n = ln(threshold_obs_c) / ln(threshold_mod_c) #' #' #'2) wet-days correction / Correcting the intensity of the GCM/RCM data #' #'Two parameter (shape and scale factors) gamma distribution function was used to model the frequency distributions of the rainfall data. The GCM/RCM rainfall above the threshold were corrected using unique correction factors for different cumulative frequencies. #' #'corrected_mod_fut = mod_fut * F-1(F.mod_fut, sh_obs_c,,sc_obs_c)/ F-1 (F.mod_fut,sh_mod_c,,sc_mod_c) #' #'where obs - observational data; mod - GCM/RCM data; n - constant; c - calibration; v - validation; fut - future data; sh - shape factor; sc- scale factor; F. - cumulative density function and F-1 - inverse of cumulative density function #' @export #' @examples #' #' #subsetting dat_model #' mod_calibration=subset(data_model,(year==2003|year==2005|year==2007|year==2009|year==2011)) #' mod_validation= subset(data_model,(year==2004|year==2006|year==2008|year==2010|year==2012)) #' #subsetting data_observation

226

#' obs_calibration=subset(data_observation,(year==2003|year==2005|year==2007|year==2009|year==2011)) #' obs_validation=subset(data_observation,(year==2004|year==2006|year==2008|year==2010|year==2012)) #' #creating the input vectors #' obs_c=obs_calibration$pr #' mod_c=mod_calibration$pr #' obs_v=obs_validation$pr #' mod_v=mod_validation$pr #' mod_fut= data_model_future$pr #' #' ParaCal(obs_c,mod_c,obs_v,mod_v,mod_fut) #' @return NULL ParaCal = function(obs_c,mod_c,obs_v,mod_v,mod_fut) { m_obs_c=mean(obs_c) m_mod_c=mean(mod_c) m_obs_v=mean(obs_v) m_mod_v=mean(mod_v) s_obs_c=stats::sd(obs_c) s_mod_c=stats::sd(mod_c) s_obs_v=stats::sd(obs_v) s_mod_v=stats::sd(mod_v) sh_obs_c=(m_obs_c/s_obs_c)^2 sh_mod_c=(m_mod_c/s_mod_c)^2 sh_obs_v=(m_obs_v/s_obs_v)^2 sh_mod_v=(m_mod_v/s_mod_v)^2 sc_obs_c=s_obs_c^2/m_obs_c sc_mod_c=s_mod_c^2/m_mod_c sc_obs_v=s_obs_v^2/m_obs_v sc_mod_v=s_mod_v^2/m_mod_v ### finding thresholds for calibration data (for both model and obsevartion) p=stats::ecdf(obs_c) thr_obs_c= p(0.0) # return the probabily at zero} threshold_obs_c= (round(p(0.0),3))*1000# return the probabily at zero in 2 digit # for calibration_obs p=stats::ecdf(mod_c) thr_mod_c= p(0.0) threshold_mod_c= (round(p(0.0),3))*1000 # for calibration_mod p=stats::ecdf(obs_v) thr_obs_v= p(0.0) threshold_obs_v= (round(p(0.0),3))*1000 # for validation_obs p=stats::ecdf(mod_v) thr_mod_v= p(0.0) threshold_mod_v= (round(p(0.0),3))*1000 # for validation_mod n=log(thr_mod_c)/log(thr_obs_c) #returns shape_factors=matrix(c(sh_obs_c,sh_obs_v,sh_mod_c,sh_mod_v),nrow = 2) colnames(shape_factors)=c("obs","mod") rownames(shape_factors)=c("calibration","validation") shape_factors=round(shape_factors,4) scale_factors=matrix(c(sc_obs_c,sc_obs_v,sc_mod_c,sc_mod_v),nrow = 2) colnames(scale_factors)=c("obs","mod") rownames(scale_factors)=c("calibration","validation")

227

scale_factors=round(scale_factors,4) thresholds=matrix(c(threshold_obs_c,threshold_obs_v,threshold_mod_c,threshold_mod_v),nrow = 2) colnames(thresholds)=c("obs","mod") rownames(thresholds)=c("calibration","validation") thresholds=round(thresholds/1000,4) n=n parameter_list = list(shape_factors=shape_factors, scale_factors=scale_factors, thresholds=thresholds,n=n) parameter_list print(parameter_list) return (NULL) }

B.2: Source code for ResVal() function of spdownscale

#' @title Validation Summary #' @description Displays the summary of the validation. #' @param obs_c vector of observational climate data (rainfall) used for calibrating the model #' @param mod_c vector of GCM/RCM climate data (rainfall) used for calibrating the model #' @param obs_v vector of observational climate data (rainfall) used for validating the model #' @param mod_v vector of GCM/RCM climate data (rainfall) used for validating the model #' @param mod_fut vector of GCM/RCM future climate data (rainfall) need to be downscaled #' @details #' #'1) Dry-days correction / Defining threshold values #' #' The relationship between the cumulative frequencies (thresholds) corresponding to the dry days of GCM/RCM data and that of the observational data is defined by a polynomial function given by; #' #'threshold_obs = (threshold_mod)^n #' #'n = ln(threshold_obs_c) / ln(threshold_mod_c) #' #' #'2) wet-days correction / Correcting the intensity of the GCM/RCM data #' #'Two parameter (shape and scale factors) gamma distribution function was used to model the frequency distributions of the rainfall data. The GCM/RCM rainfall above the threshold were corrected using unique correction factors for different cumulative frequencies. #' #'corrected_mod_fut = mod_fut * F-1(F.mod_fut, sh_obs_c,,sc_obs_c)/ F-1 (F.mod_fut,sh_mod_c,,sc_mod_c) #' #'where obs - observational data; mod - GCM/RCM data; n - constant; c - calibration; v - validation; fut - future data; sh - shape factor; sc- scale factor; F. - cumulative density function and F-1 - inverse of cumulative density function

228

#' @export #' @examples #' #' #subsetting dat_model #' mod_calibration=subset(data_model,(year==2003|year==2005|year==2007|year==2009|year==2011)) #' mod_validation= subset(data_model,(year==2004|year==2006|year==2008|year==2010|year==2012)) #' #subsetting data_observation #' obs_calibration=subset(data_observation,(year==2003|year==2005|year==2007|year==2009|year==2011)) #' obs_validation=subset(data_observation,(year==2004|year==2006|year==2008|year==2010|year==2012)) #' #creating the input vectors #' obs_c=obs_calibration$pr #' mod_c=mod_calibration$pr #' obs_v=obs_validation$pr #' mod_v=mod_validation$pr #' mod_fut= data_model_future$pr #' #' ResVal(obs_c,mod_c,obs_v,mod_v,mod_fut) #' @return NULL ResVal= function(obs_c,mod_c,obs_v,mod_v,mod_fut) { m_obs_c=mean(obs_c) m_mod_c=mean(mod_c) m_obs_v=mean(obs_v) m_mod_v=mean(mod_v) s_obs_c=stats::sd(obs_c) s_mod_c=stats::sd(mod_c) s_obs_v=stats::sd(obs_v) s_mod_v=stats::sd(mod_v) sh_obs_c=(m_obs_c/s_obs_c)^2 sh_mod_c=(m_mod_c/s_mod_c)^2 sh_obs_v=(m_obs_v/s_obs_v)^2 sh_mod_v=(m_mod_v/s_mod_v)^2 sc_obs_c=s_obs_c^2/m_obs_c sc_mod_c=s_mod_c^2/m_mod_c sc_obs_v=s_obs_v^2/m_obs_v sc_mod_v=s_mod_v^2/m_mod_v ### finding thresholds for calibration data(for both model and obsevartion) p=stats::ecdf(obs_c) thr_obs_c= p(0.0) # return the probabily at zero} threshold_obs_c= (round(p(0.0),3))*1000# return the probabily at zero in 2 digit # for calibration_obs p=stats::ecdf(mod_c) thr_mod_c= p(0.0) threshold_mod_c= (round(p(0.0),3))*1000 # for calibration_mod p=stats::ecdf(obs_v) thr_obs_v= p(0.0) threshold_obs_v= (round(p(0.0),3))*1000 # for validation_obs p=stats::ecdf(mod_v) thr_mod_v= p(0.0) threshold_mod_v= (round(p(0.0),3))*1000 # for validation_mod

229

n=log(thr_mod_c)/log(thr_obs_c) ### Bias correction_validation xx_mod_v=mod_v ff_mod_v=c(1) crtd_mod_v=c(1) p=stats::ecdf(mod_v) thr= p(0.0) threshold=thr^(1/n) # this is new from the convensional bias correction # threshold # turn on/off to view it for (i in 1:length(xx_mod_v)) ff_mod_v[i]= stats::pgamma(xx_mod_v[i],sh_mod_v,,sc_mod_v) # ff_mod_v # turn on/off to view it for (i in 1:length(ff_mod_v)) if (ff_mod_v[i] >= threshold) { crtd_mod_v[i]= xx_mod_v[i] * stats::qgamma(ff_mod_v [i],sh_obs_c,,sc_obs_c) /stats::qgamma(ff_mod_v [i],sh_mod_c,,sc_mod_c) } else{ crtd_mod_v[i]=0.0} # crtd_mod_v # turn on/off to view it ## display results of the validation/summary # plotting /generating gamma curves for validation # modelling data using gamma distribution (above thresold)_ Observation # selecting threshold to plot thresholds (here, I have used the threshold of same data) p=stats::ecdf(obs_v) threshold_obs_v= (round(p(0.0),3))*1000 # threshold_obs_v # turn on/off to view it # modeling obs_val f_obs_v=c(1) x_obs_v=c(1) # defining variables for (i in threshold_obs_v:999) f_obs_v[i-threshold_obs_v+1]=i/1000 for (i in 1:length(f_obs_v)) x_obs_v[i]=round((stats::qgamma(f_obs_v[i],sh_obs_v,,sc_obs_v)),1) # f_obs_v # turn on/off to view it # x_obs_v # turn on/off to view it p.plot= graphics::plot(x_obs_v,f_obs_v,xlim=c(0,50),type="l",col= "green") #x=x_obs_v x= obs_v p=stats::ecdf(x) graphics::lines(p,col="green") # turn on/off to display actual data # modelling data using gamma distribution (above thresold)_model # selecting threshold to plot thresholds (here, I have used the threshold of same data) p=stats::ecdf(mod_v) threshold_mod_v= round(p(0.0),3)*1000 # return the probabily at zero in 2 digit # threshold_mod_v

230

# turn on/off to view it f_mod_v=c(1) x_mod_v=c(1) # defining variables for (i in threshold_mod_v:999) f_mod_v[i-threshold_obs_v+1]=i/1000 for (i in 1:length(f_mod_v)) x_mod_v[i]=round((stats::qgamma(f_mod_v[i],sh_mod_v,,sc_mod_v)),1) #f_mod_v # turn on/off to view it #x_mod_v # turn on/off to view it graphics::lines(x_mod_v,f_mod_v,type="l",col= "red") #x=x_mod_v x= mod_v p=stats::ecdf(x) graphics::lines(p,col= "red") # fitting true values on to modeled curve, turn on/off to display actual data # Modeling corrected data using gamma distribution # selecting threshold to plot thresholds (here, I have used the threshold of same data) p=stats::ecdf(crtd_mod_v) threshold_crtd_mod_v= round(p(0.0),3)*1000 # return the probabily at zero in 2 digit # threshold_crtd_mod_v # turn on/off to view it #parameters m_crtd_mod_v= mean(crtd_mod_v) s_crtd_mod_v= stats::sd(crtd_mod_v) sh_crtd_mod_v= (m_crtd_mod_v/s_crtd_mod_v)^2 sc_crtd_mod_v= s_crtd_mod_v^2/m_crtd_mod_v f_crtd_mod_v=c(1) x_crtd_mod_v=c(1) # defining variables for (i in threshold_crtd_mod_v:999) f_crtd_mod_v[i-threshold_obs_v+1]=i/1000 for (i in 1:length(f_mod_v)) x_crtd_mod_v[i]=round((stats::qgamma(f_crtd_mod_v[i],sh_crtd_mod_v,,sc_crtd_mod_v)),1) # f_crtd_mod_v # turn on/off to view it # x_crtd_mod_v # turn on/off to view it graphics::lines(x_crtd_mod_v,f_crtd_mod_v,type="l",col= "blue") #x=x_crtd_mod_v,f_crtd x= crtd_mod_v p=stats::ecdf(x) graphics::lines(p,col="blue") # fitting true values on to modeled curve, turn on/off to display actual data #VALIDATION SUMMARY/RESULT ### creating th A matrix for validation index f=c(1) for (i in 1:1000) f[i]=(i-1)/1000 #f #crtd_mod_v #obs_v$pr #mod_v$pr xx_obs_v=c(1) xx_mod_v=c(1) xx_crtd_mod_v=c(1) for(i in 1:length(f)) {

231

xx_obs_v[i]=round((stats::qgamma(f[i], sh_obs_v,,sc_obs_v)),1) xx_mod_v[i]=round((stats::qgamma(f[i], sh_mod_v,,sc_mod_v)),1) xx_crtd_mod_v[i] =round((stats::qgamma(f[i], sh_crtd_mod_v,,sc_crtd_mod_v)),1)} f= matrix(f) xx_crtd_mod_v= matrix(xx_crtd_mod_v) xx_obs_v=matrix(xx_obs_v) xx_mod_v= matrix(xx_mod_v) A=cbind(f,xx_obs_v,xx_mod_v,xx_crtd_mod_v) #A ### RMSV rmsv_obs_vs_mod=c(1) rmsv_obs_vs_crmod=c(1) for (i in 1:length(xx_obs_v)) rmsv_obs_vs_mod[i]=((xx_mod_v[i]-xx_obs_v[i])^2)^.5 for (i in 1:length(xx_obs_v)) rmsv_obs_vs_crmod[i]=((xx_crtd_mod_v[i]-xx_obs_v[i])^2)^.5 rmsv1=mean(rmsv_obs_vs_mod) rmsv2=mean(rmsv_obs_vs_crmod) # rmsv1 # rmsv2 # has to be transformed into summmary ### gradient # plot (xx_obs_v,xx_crtd_mod_v,type = "l",col="blue") # lines(xx_obs_v,xx_obs_v,type = "l",col="green") # lines(xx_obs_v,xx_mod_v,type = "l",col="red") # turn on/off to view it graphics::plot (xx_obs_v,xx_obs_v,col="green") graphics::lines(xx_obs_v,xx_obs_v,type = "l",col="green") #lines (xx_obs_v,xx_crtd_mod_v,col="blue") fit1=stats::lm(xx_crtd_mod_v~xx_obs_v) graphics::abline(fit1, col="blue") #fit1 # turn on/off to view it # summary(fit1) # turn on/off to view it # lines (xx_obs_v,xx_mod_v) # turn on/off to view it fit2=stats::lm(xx_mod_v~xx_obs_v) graphics::abline(fit2,col="red") # fit2 # summary(fit2) # turn on/off to view it #coef(fit1) #coef(fit2) err=round((((threshold_obs_v- threshold_crtd_mod_v)^2)^.5)/threshold_obs_v*100,2) validation_output_list=list(percentage_error_in_the_thresholds =err,root_mean_squre_error_in_mod=rmsv1 ,root_mean_squre_error_in_crtdmod=rmsv2,gradient_of_obs_vs_mod_line=fit2$coefficients[2],gradient_of_obs_vs_crtdmod_line=fit1$coefficients[2]) print(validation_output_list) return(NULL) }

232

B.3: Source code for downscale() function of spdownscale

#' @title Spatial Downscaling #' @description Generating the future climate data (rainfall) #' @param obs_c vector of observational climate data (rainfall) used for calibrating the model #' @param mod_c vector of GCM/RCM rainfall data (rainfall) used for calibrating the model #' @param obs_v vector of observational climate data (rainfall) used for validating the model #' @param mod_v vector of GCM/RCM climate data (rainfall) used for validating the model #' @param mod_fut vector of GCM/RCM future climate data (rainfall) need to be downscaled #' @details #'1) Dry-days correction / Defining threshold values #' #' The relationship between the cumulative frequencies (thresholds) corresponding to the dry days of GCM/RCM data and that of the observational data is defined by a polynomial function given by; #' #'threshold_obs = (threshold_mod)^n #' #'n = ln(threshold_obs_c) / ln(threshold_mod_c) #' #' #'2) wet-days correction / Correcting the intensity of the GCM/RCM data #' #'Two parameter (shape and scale factors) gamma distribution function is used to model the frequency distributions of the rainfall data. The GCM/RCM rainfall above the threshold were corrected using unique correction factors for different cumulative frequencies. #' #'corrected_mod_fut = mod_fut * F-1(F.mod_fut, sh_obs_c,,sc_obs_c)/ F-1 (F.mod_fut,sh_mod_c,,sc_mod_c) #' #'where obs - observational data; mod - GCM/RCM data; n - constant; c - calibration; v - validation; fut - future data; sh - shape factor; sc- scale factor; F. - cumulative density function and F-1 - inverse of cumulative density function #' @export #' @examples #' #subsetting dat_model #' mod_calibration=subset(data_model,(year==2003|year==2005|year==2007|year==2009|year==2011)) #' mod_validation= subset(data_model,(year==2004|year==2006|year==2008|year==2010|year==2012)) #' #subsetting data_observation #' obs_calibration=subset(data_observation,(year==2003|year==2005|year==2007|year==2009|year==2011)) #' obs_validation=subset(data_observation,(year==2004|year==2006|year==2008|year==2010|year==2012)) #' #creating the input vectors #' obs_c=obs_calibration$pr

233

#' mod_c=mod_calibration$pr #' obs_v=obs_validation$pr #' mod_v=mod_validation$pr #' mod_fut= data_model_future$pr #' #' downscale(obs_c,mod_c,obs_v,mod_v,mod_fut) #' #' @return NULL downscale = function(obs_c,mod_c,obs_v,mod_v,mod_fut) { m_obs_c=mean(obs_c) m_mod_c=mean(mod_c) m_obs_v=mean(obs_v) m_mod_v=mean(mod_v) s_obs_c=stats::sd(obs_c) s_mod_c=stats::sd(mod_c) s_obs_v=stats::sd(obs_v) s_mod_v=stats::sd(mod_v) sh_obs_c=(m_obs_c/s_obs_c)^2 sh_mod_c=(m_mod_c/s_mod_c)^2 sh_obs_v=(m_obs_v/s_obs_v)^2 sh_mod_v=(m_mod_v/s_mod_v)^2 sc_obs_c=s_obs_c^2/m_obs_c sc_mod_c=s_mod_c^2/m_mod_c sc_obs_v=s_obs_v^2/m_obs_v sc_mod_v=s_mod_v^2/m_mod_v ### finding thresholds for calibration data (for both model and obsevartion) p=stats::ecdf(obs_c) thr_obs_c= p(0.0) # return the probabily at zero} threshold_obs_c= (round(p(0.0),3))*1000 # return the probabily at zero in 2 digit # for calibration_obs p=stats::ecdf(mod_c) thr_mod_c= p(0.0) threshold_mod_c= (round(p(0.0),3))*1000 # for calibration_mod p=stats::ecdf(obs_v) thr_obs_v= p(0.0) threshold_obs_v= (round(p(0.0),3))*1000 # for validation_obs p=stats::ecdf(mod_v) thr_mod_v= p(0.0) threshold_mod_v= (round(p(0.0),3))*1000 # for validation_mod n=log(thr_mod_c)/log(thr_obs_c) # FUTURE # parameters m_mod_fut= mean(mod_fut) s_mod_fut= stats::sd(mod_fut) sh_mod_fut= (m_mod_fut/s_mod_fut)^2 sc_mod_fut= s_mod_fut^2/m_mod_fut ### Bias correction xx_mod_fut=mod_fut ff_mod_fut=c(1) crtd_mod_fut=c(1) p=stats::ecdf(mod_fut) thr= p(0.0) threshold=thr^(1/n) # this is new from the convensional bias correction #threshold # turn on/off to view it

234

for (i in 1:length(xx_mod_fut)) ff_mod_fut[i]= stats::pgamma(xx_mod_fut[i],sh_mod_fut,,sc_mod_fut) #ff_mod_fut # turn on/off to view it for (i in 1:length(ff_mod_fut)) if (ff_mod_fut[i] >= threshold) { crtd_mod_fut[i]= xx_mod_fut[i] * stats::qgamma(ff_mod_fut[i],sh_obs_c,,sc_obs_c)/stats::qgamma(ff_mod_fut [i],sh_mod_c,,sc_mod_c) } else{ crtd_mod_fut[i]=0.0} crtd_mod_fut=round(crtd_mod_fut,1) # turn on/off to view it crtd_mod_fut=matrix(crtd_mod_fut) crtd_mod_fut<<-crtd_mod_fut print(crtd_mod_fut) return(NULL) }

235

Appendix C

C1: Example of First-order Markov process

1. Development of the TPM

• Let a time-series of rainfall to construct the TPM (calibration data) be,

Time step Rainfall (mm) 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0.2 9 0.6 10 0.4 11 0 12 0.2 13 0 14 0 15 0

• The discrete weather states are 0, 0.2, 0.4 and 0.6. • In order to calculate the conditional probability of each weather states, the total

number of occurrence of any event with a known previous event is determined.

Current event

Prev

ious

eve

nt

0 0.2 0.4 0.6 0 7 2 0 0

0.2 1 0 0 1 0.4 1 0 0 0 0.6 0 0 1 0

Then the probability of those transitions are calculated

236

Current event

Prev

ious

eve

nt

0 0.2 0.4 0.6 0 0.78 0.22 0.00 0.00

0.2 0.50 0.00 0.00 0.50 0.4 1.00 0.00 0.00 0.00 0.6 0.00 0.00 1.00 0.00

The cumulative probabilities of those transitions are,

Current event

Prev

ious

eve

nt

0 0.2 0.4 0.6 0 0.78 1.00 1.00 1.00

0.2 0.50 0.50 0.50 1.00 0.4 1.00 1.00 1.00 1.00 0.6 0.00 0.00 1.00 1.00

2. Data generation

• The data generation involves generation of random numbers between 0 and 1. The model takes the random number with a known initial condition (previous event) to seek the current event. Model searches the (cumulative) transition probability that equals to the generated random number to find the current event. Let the initial value be 0 and the generated random number be 0.75,

Current event

Prev

ious

eve

nt 0 0.2 0.4 0.6

0 0.78 1.00 1.00 1.00 0.2 0.50 0.50 0.50 1.00 0.4 1.00 1.00 1.00 1.00 0.6 0.00 0.00 1.00 1.00

Then the next rainfall event will be 0. This value will become the initial value for the next event with a new random variable. This process continues for 36 times (3-hour has 36 5-minute time steps).

• Once the 36 events are generated the total of it will be compared against the observed 3-hour total. If these values are more or less the same the process ends. Otherwise, new sets of 36 events will be generated to meet this condition.

Initial value

237

C2: Sample source code for temporal downscaling (station: 40476, GCM: ACCESS 1.0, RCP: RCP 4.5, period: 20206-2045)

#SELECTING TIME EVERY THREE HOUR C <- read.csv("W:/WORK/DATA COLLECTION/Observation/40764/qc/qc40764_3h_obsv_apr2002_dec2012.txt", sep="") # subsetting for calibration and validation cal_C= subset(C,(year==2003|year==2005|year==2007|year==2009|year==2011|year==2013|year==2015),select = c(year,month,date,hr,min,pr)) val_C= subset(C,(year==2002|year==2004|year==2006|year==2008|year==2010|year==2012|year==2014),select = c(year,month,date,hr,min,pr)) write.table(cal_C,"C:/All/cal_threehour_40764_with_date.txt") write.table(val_C,"C:/All/val_threehour_40764_with_date.txt") #SELECTING I MIN DATA B<- read.csv("W:/WORK/DATA COLLECTION/Observation/40764/qc/every_min_quality_checked_2002-2005.txt", sep="") # subsetting for calibration and validation cal_B= subset(B,(year==2003|year==2005|year==2007|year==2009|year==2011|year==2013|year==2015),select = c(year,month,date,hr,min,pr)) val_B= subset(B,(year==2002|year==2004|year==2006|year==2008|year==2010|year==2012|year==2014),select = c(year,month,date,hr,min,pr)) write.table(cal_B,"C:/All/cal_oneminute_4076_with_date.txt") write.table(val_B,"C:/All/val_oneminute_4076_with_date.txt") #creating matrix for MRK cal_oneminute_4076_with.date <- read.csv("//qut.edu.au/Documents/StaffHome/StaffGroupR$/rasheeda/Desktop/aa/.Rproj.user/Temp/qc/files/cal_oneminute_4076_with_date.txt", sep="") cal_threehour_40764_with_date <- read.csv("//qut.edu.au/Documents/StaffHome/StaffGroupR$/rasheeda/Desktop/aa/.Rproj.user/Temp/qc/files/cal_threehour_40764_with_date.txt", sep="") A=cal_oneminute_4076_with.date B=cal_threehour_40764_with_date # maximum rainfall X=matrix(A$pr,nrow=180) X=t(X) max_onemin=matrix(apply(X,1,max)) #creating full matrix with max and tot R=cbind(X,B$pr,max_onemin) R=data.frame(R) #creating the matrix to develop the markov matrix Y=R Y=subset(Y,Y$X182 >0.0) # removing rows that are zeros or 0.2 Y$X182=NULL Y$X181=NULL write.table(Y,"//qut.edu.au/Documents/StaffHome/StaffGroupR$/rasheeda/Desktop/aa/.Rproj.user/Temp/qc/files/matrix_for_MRK.txt") #Making of MRK

238

matrix_for_MRK <- read.csv("//qut.edu.au/Documents/StaffHome/StaffGroupR$/rasheeda/Desktop/aa/.Rproj.user/Temp/qc/files/matrix_for_MRK.txt", sep="") M=matrix_for_MRK l=length(M$X1) #Creating markov matrix #MRK MRK=matrix(,nrow = 51,ncol = 51) a=c(seq(0,10,by=0.2)) b=c(seq(0,10,by=0.2)) for (aa in 1:51){ for (bb in 1:51){ d1=a[aa] d2=b[bb] count=0 for (row in 1:l){ for (col in 1:179){ if (M[row,col]==d1 & M[row,col+1]==d2) {count=count+1} } } MRK[aa,bb]=count } } write.table(MRK,"//qut.edu.au/Documents/StaffHome/StaffGroupR$/rasheeda/Desktop/aa/.Rproj.user/Temp/qc/files/MRK.txt") #without smoothing********************************************** MRK <- read.csv("//qut.edu.au/Documents/StaffHome/StaffGroupR$/rasheeda/Desktop/aa/.Rproj.user/Temp/qc/files/MRK.txt", sep="") tot=rowSums(MRK) MRK_P=matrix(,nrow = 51,ncol = 51) for (r in 1:51){ for (c in 1:51){ MRK_P[r,c]=MRK[r,c]/tot[r] } } # creating probability MRK_CP = matrix(,nrow = 51,ncol = 51) for (r in 1:51) { MRK_CP[r,1]=MRK_P[r,1] } for (r in 1:51){ for (c in 2:51){ MRK_CP[r,c]=MRK_P[r,c]+MRK_CP[r,c-1] } } # creating cululative probability MRK_CP[is.nan(MRK_CP)]=1 write.table(MRK_CP,"//qut.edu.au/Documents/StaffHome/StaffGroupR$/rasheeda/Desktop/aa/.Rproj.user/Temp/qc/files/MRK_CP.txt") # data generation & Validation MRK_CP <- read.csv("//qut.edu.au/Documents/StaffHome/StaffGroupR$/rasheeda/Desktop/aa/.Rproj.user/Temp/qc/files/MRK_CP.txt", sep="") A=MRK_CP E=seq(0,10,0.2) val_threehour_40764_with_date <- read.csv("//qut.edu.au/Documents/StaffHome/StaffGroupR$/rasheeda/Desktop/aa/.Rproj.user/Temp/qc/files/val_threehour_40764_with_date.txt", sep="") B=val_threehour_40764_with_date

239

B=subset(B,B$pr>0.0) #remove 0.0 l=length(B$pr) simulated=matrix(,nrow = l,ncol = 180) for (w in 1:l){ X=matrix(,nrow = 1,ncol = 180) X[1,1]=0 x=X[1,1] tot=B[w,6] totnew=9999999999 while (totnew > tot*1.05 | totnew < tot*.95){ random=runif(180,0,1) for (i in 1:180) { j=1 while (E[j]<x) {j=j+1} r=j rand=random[i] z=1 while (A[r,z]<=rand) {z=z+1} e=z x=E[e] X[1,i]=x } totnew= rowSums(X) } simulated[w,]=X } S_data=matrix(simulated,nrow = l) simulated_data_with_date= cbind(B,S_data) write.table(simulated_data_with_date,"//qut.edu.au/Documents/StaffHome/StaffGroupR$/rasheeda/Desktop/aa/.Rproj.user/Temp/qc/files/simulated_data_with_date.txt") #prepareing for validation Summary #checking total and peak -1min vs 3 hours(Dont USe; use * instread) simulated_data_with_date <- read.csv("//qut.edu.au/Documents/StaffHome/StaffGroupR$/rasheeda/Desktop/aa/.Rproj.user/Temp/qc/files/simulated_data_with_date.txt", sep="") X_simulation=subset(simulated_data_with_date,,select = -c(year,month,date,hr,min,pr)) simulation_tot=data.frame(rowSums(X_simulation)) simulation_peak=data.frame(apply(X_simulation,1,max)) val_oneminute_4076_with_date <- read.csv("//qut.edu.au/Documents/StaffHome/StaffGroupR$/rasheeda/Desktop/aa/.Rproj.user/Temp/qc/files/val_oneminute_4076_with_date.txt", sep="") A= matrix(val_oneminute_4076_with_date$pr,nrow =180) A=t(A) Rsum= matrix(rowSums(A),ncol=1) A=cbind(A,Rsum) A=data.frame(A) A=subset(A,A$X181>0.0) X_obs=subset(A,,select=-c(X181)) obs_tot=data.frame(rowSums(X_obs)) obs_peak=data.frame(apply(X_obs,1,max)) tot=cbind(obs_tot,simulation_tot) peak=cbind(obs_peak,simulation_peak) plot(tot$rowSums.X_obs.,tot$rowSums.X_simulation.) plot(peak$apply.X_obs..1..max.,peak$apply.X_simulation..1..max.)

240

#checking total and peak -5 minvs 3 hours simulated_data_with_date <- read.csv("//qut.edu.au/Documents/StaffHome/StaffGroupR$/rasheeda/Desktop/aa/.Rproj.user/Temp/qc/files/simulated_data_with_date.txt", sep="") X_simulation=subset(simulated_data_with_date,,select = -c(year,month,date,hr,min,pr)) X_simulation=t(X_simulation) X_simulation=matrix(X_simulation,ncol=1) X_simulation=matrix(X_simulation,nrow = 5) X_simulation=t(X_simulation) X5min_simulation= matrix(rowSums(X_simulation),ncol = 1) X5min_simulation=matrix(X5min_simulation,nrow=36) X5min_simulation=t(X5min_simulation) sim5_tot=data.frame(rowSums(X5min_simulation)) sim5_peak=data.frame(apply(X5min_simulation,1,max)) val_oneminute_4076_with_date <- read.csv("//qut.edu.au/Documents/StaffHome/StaffGroupR$/rasheeda/Desktop/aa/.Rproj.user/Temp/qc/files/val_oneminute_4076_with_date.txt", sep="") X_obs=val_oneminute_4076_with_date$pr X_obs=matrix(X_obs,nrow = 5) X_obs=t(X_obs) X5min_obs= matrix(rowSums(X_obs),ncol = 1) X5min_obs=matrix(X5min_obs,nrow=36) X5min_obs=t(X5min_obs) X5min_obs=data.frame(cbind(rowSums(X5min_obs),X5min_obs)) X5min_obs=subset(X5min_obs,X5min_obs$X1>0) X5min_obs$X1=NULL obs5_tot=data.frame(rowSums(X5min_obs)) obs5_peak=data.frame(apply(X5min_obs,1,max)) plot(obs5_tot$rowSums.X5min_obs.,sim5_tot$rowSums.X5min_simulation.) fit=lm(obs5_tot$rowSums.X5min_obs.~sim5_tot$rowSums.X5min_simulation.) abline(fit,col="blue") plot(obs5_peak$apply.X5min_obs..1..max.,sim5_peak$apply.X5min_simulation..1..max.,xlim=c(0,10),ylim=c(0,10)) lines(x = c(0,20), y = c(0,20)) #Validation x1=t(X5min_obs) x1=matrix(x1,ncol=1) p1=ecdf(x1) plot(p1,col="green") m_x1=mean(x1) s_x1=sd(x1) sh_x1=(m_x1/s_x1)^2 sc_x1=s_x1^2/m_x1 f_x1=c(1) x_x1=c(1) # defining variables for (i in 0:999) f_x1[i]=i/1000 for (i in 1:length(f_x1)) x_x1[i]=round((qgamma(f_x1[i],sh_x1,,sc_x1)),1) lines(x_x1,f_x1,type="l",col= "green") x2=t(X5min_simulation) x2=matrix(x2,ncol=1) p2=ecdf(x2) lines(p2,col="blue") m_x2=mean(x2) s_x2=sd(x2) sh_x2=(m_x2/s_x2)^2

241

sc_x2=s_x2^2/m_x2 f_x2=c(1) x_x2=c(1) # defining variables for (i in 0:999) f_x2[i]=i/1000 for (i in 1:length(f_x2)) x_x2[i]=round((qgamma(f_x2[i],sh_x2,,sc_x2)),1) lines(x_x2,f_x2,type="l",col= "blue") T=cbind(f_x1,x_x1,x_x2) T=data.frame(T) q=sqrt(sum((T$f_x1 - T$x_x2)^2)/nrow(T)) #creating the futute climate data_aceess_rcp4.5 crtd40674_Access_rcp45_jan2026_dec2045 <- read.csv("//qut.edu.au/Documents/StaffHome/StaffGroupR$/rasheeda/Desktop/40764_access/NEW+QC/crtd40674_Access_rcp45_jan2026_dec2045.txt", sep="") D=crtd40674_Access_rcp45_jan2026_dec2045 MRK_CP <- read.csv("//qut.edu.au/Documents/StaffHome/StaffGroupR$/rasheeda/Desktop/aa/.Rproj.user/Temp/qc/files/MRK_CP.txt", sep="") A=MRK_CP E=seq(0,10,0.2) B=D l=length(B$CRTD_mod_fut) indicator=1 for (q in 1:l){ if (B[q,6]==0.1 & indicator==1){ B[q,6]=0.0 indicator=2 } else if(B[q,6]==0.1 & indicator==2){ B[q,6]=0.2 indicator=1 } } indicator=1 for (q in 1:l){ if (B[q,6]==0.3 & indicator==1){ B[q,6]=0.2 indicator=2 } else if(B[q,6]==0.3 & indicator==2){ B[q,6]=0.4 indicator=1 } } indicator=1 for (q in 1:l){ if (B[q,6]==0.5 & indicator==1){ B[q,6]=0.4 indicator=2 } else if(B[q,6]==0.5 & indicator==2){ B[q,6]=0.6 indicator=1 } } indicator=1 for (q in 1:l){

242

if (B[q,6]==0.7 & indicator==1){ B[q,6]=0.6 indicator=2 } else if(B[q,6]==0.7 & indicator==2){ B[q,6]=0.8 indicator=1 } } indicator=1 for (q in 1:l){ if (B[q,6]==0.9 & indicator==1){ B[q,6]=0.8 indicator=2 } else if(B[q,6]==0.9 & indicator==2){ B[q,6]=1.0 indicator=1 } } indicator=1 for (q in 1:l){ if (B[q,6]==1.1 & indicator==1){ B[q,6]=1.0 indicator=2 } else if(B[q,6]==1.1 & indicator==2){ B[q,6]=1.2 indicator=1 } } indicator=1 for (q in 1:l){ if (B[q,6]==1.3 & indicator==1){ B[q,6]=1.2 indicator=2 } else if(B[q,6]==1.3 & indicator==2){ B[q,6]=1.4 indicator=1 } } indicator=1 for (q in 1:l){ if (B[q,6]==1.5 & indicator==1){ B[q,6]=1.4 indicator=2 } else if(B[q,6]==1.5 & indicator==2){ B[q,6]=1.6 indicator=1 } } indicator=1 for (q in 1:l){ if (B[q,6]==1.7 & indicator==1){ B[q,6]=1.6 indicator=2 } else if(B[q,6]==1.7 & indicator==2){ B[q,6]=1.8

243

indicator=1 } } indicator=1 for (q in 1:l){ if (B[q,6]==1.9 & indicator==1){ B[q,6]=1.8 indicator=2 } else if(B[q,6]==1.9 & indicator==2){ B[q,6]=2.0 indicator=1 } } simulated=matrix(,nrow = l,ncol = 180) for (w in 1:l){ if(B[w,6]>0 & B[w,6]<90) { X=matrix(,nrow = 1,ncol = 180) X[1,1]=0 x=X[1,1] tot=B[w,6] totnew=9999999999 while (totnew > tot*1.1 | totnew < tot*.9){ random=runif(180,0,1) for (i in 1:180) { j=1 while (E[j]<x) {j=j+1} r=j rand=random[i] z=1 while (A[r,z]<=rand) {z=z+1} e=z x=E[e] X[1,i]=x } totnew= rowSums(X) } simulated[w,]=X } } for (w in 1:l){ if(B[w,6]>=100){ X=matrix(,nrow = 1,ncol = 180) X[1,1]=0 x=X[1,1] f= B[w,6]/100 tot=100 totnew=9999999999 while (totnew > tot*1.1 | totnew < tot*.9){ random=runif(180,0,1) for (i in 1:180) { j=1 while (E[j]<x) {j=j+1} r=j rand=random[i] z=1 while (A[r,z]<=rand) {z=z+1} e=z x=E[e]

244

X[1,i]=x } totnew= rowSums(X) } simulated[w,]=(round((f*X)*5))/5 } } sim=simulated sim[is.na(sim)]=0.0 S_data=matrix(sim,nrow=l) simulated_data_with_date= cbind(B,S_data) write.table(simulated_data_with_date,"//qut.edu.au/Documents/StaffHome/StaffGroupR$/rasheeda/Desktop/aa/.Rproj.user/Temp/qc/files/access_rcp45_R1.txt")

245

Appendix D Table D.1: IFDs for Gold Coast Seaway (40764) for RCP 4.5 climate change scenario for the period 2026-2045. The values are given with 95% confidence interval.

ARI (years)

1 2 5 10 20 50 100

5 Min 14.2±0.22 15.3±0.18 17.8±0.20 19.4±0.27 20.9±0.36 22.9±0.48 24.5±0.57

10 Min 21.4±0.45 23.6±0.37 28.2±0.22 31.3±0.21 34.2±0.27 38.0±0.40 40.9±0.52

15 Min 27.2±0.44 31.0±0.43 39.0±0.49 44.3±0.59 49.4±.70 56.0±0.87 60.9±1.00

30 Min 34.1±0.51 41.4±0.46 56.7±0.89 66.9±1.30 76.6±1.71 89.3±2.25 98.7±2.66

1 Hr 37.3±0.31 47.1±0.56 67.7±1.70 81.4±2.48 94.5±3.23 111±4.21 124±4.94

2 Hr 42.6±1.39 55.0±1.34 81.1±3.24 98.3±4.79 115±6.32 136±8.32 152±9.83

3 Hr 46.6±1.91 59.1±2.16 85.4±3.31 104±4.83 121±5.99 144±7.64 160±8.41

6 Hr 72.1±0.23 93.4±0.34 138±1.13 168±1.68 197±2.21 234±2.90 262±3.41

12 Hr 105±0.68 144±0.57 227±1.63 282±2.48 335±3.31 404±4.40 455±5.22

24 Hr 144±0.76 197±0.71 308±1.82 381±2.72 451±3.60 543±4.76 611±5.63

246

Table D.2: IFDs for Gold Coast Seaway (40764) for RCP 4.5 climate change scenario for the period 2081-2100. The values are given with 95% confidence interval.

ARI (years)

1 2 5 10 20 50 100

5 Min 15.3±0.30 16.3±0.26 18.5±0.24 19.9±0.29 21.2±0.37 23.0±0.49 24.3±0.59

10 Min 23.3±0.48 25.4±0.43 29.8±0.45 32.8±0.55 35.6±0.68 39.3±0.87 42.0±1.03

15 Min 29.8±0.49 33.6±0.45 41.6±0.67 46.9±0.92 52.0±1.19 58.5±1.55 63.5±1.83

30 Min 39.1±0.78 46.9±0.99 63.3±2.02 74.2±2.80 84.6±3.57 98.1±4.58 108±5.34

1 Hr 44.4±0.69 55.4±0.96 78.7±2.37 94.1±3.40 109±4.39 128±5.69 142±6.66

2 Hr 52.5±1.28 65.0±1.50 91.5±2.56 109±3.43 126±4.29 148±5.44 164±6.30

3 Hr 54.8±1.35 68.6±1.67 96.0±2.71 115±3.67 132±4.72 156±5.79 173±6.38

6 Hr 85.6±0.45 108±0.66 154±1.29 185±1.74 215±2.18 254±2.76 282±3.19

12 Hr 142±0.51 179±0.73 259±1.31 312±1.73 363±2.13 428±2.66 478±3.06

24 Hr 187±0.41 250±0.50 382±1.16 469±1.66 553±2.14 661±2.78 743±3.26

247


ARI (years)

1 2 5 10 20 50 100

5 Min 14.1±0.38 15.3±0.32 17.7±0.24 19.4±0.26 20.9±0.33 23.0±0.45 24.5±0.56

10 Min 20.4±0.31 22.7±0.33 27.4±0.42 30.6±0.50 33.6±0.60 37.6±0.73 40.5±0.83

15 Min 25.0±0.33 28.6±0.37 36.4±0.63 41.5±0.85 46.4±1.07 52.7±1.36 57.5±1.58

30 Min 31.9±0.51 38.8±0.52 53.4±0.97 63.1±1.37 72.4±1.78 84.5±2.32 93.5±2.72

1 Hr 34.7±0.43 44.3±0.80 64.5±2.21 77.9±3.18 90.8±4.13 107±5.35 120±6.26

2 Hr 41.5±1.30 53.1±1.21 77.7±1.40 94.0±1.75 110±2.16 130±2.75 145±3.22

3 Hr 43.1±0.85 55.6±0.43 81.6±1.42 99.2±1.50 116±1.59 138±1.71 155±2.15

6 Hr 67.9±0.30 87.4±0.26 129±0.69 156±1.04 182±1.39 216±1.85 241±2.19

12 Hr 112±0.23 140±0.21 198±0.50 237±0.74 275±0.98 323±1.30 360±1.54

24 Hr 154±0.62 197±0.42 288±0.55 349±0.91 406±1.29 481±1.79 537±2.17

248


ARI (years)

1 2 5 10 20 50 100

5 Min 14.4±0.39 15.5±0.30 17.7±0.22 19.2±0.31 20.7±0.44 22.5±0.63 23.9±0.78

10 Min 21.4±0.29 23.7±0.27 28.6±0.31 31.8±0.36 34.8±0.43 38.8±0.54 41.8±0.62

15 Min 26.5±0.37 30.5±0.30 38.9±0.46 44.5±0.68 49.9±0.90 56.9±1.21 62.1±1.44

30 Min 34.6±0.54 42.7±0.57 59.8±1.21 71.1±1.73 82.0±2.25 96.1±2.93 107±3.44

1 Hr 37.9±0.37 50.1±0.45 75.8±0.81 92.8±1.09 109±1.36 130±1.73 146±2.01

2 Hr 44.1±1.12 59.4±0.97 91.5±4.13 113±6.35 133±8.48 160±11.25 179±13.33

3 Hr 46.7±1.56 63.6±1.38 98.4±3.36 122±5.28 142±6.77 166±11.81 186±13.19

6 Hr 71.2±0.35 95.2±0.61 146±1.30 179±1.77 212±2.23 253±2.83 284±3.28

12 Hr 109±0.21 146±0.50 226±1.33 279±1.90 330±2.44 396±3.15 445±3.68

24 Hr 147±0.53 217±0.78 363±2.04 460±2.95 553±3.82 674±4.96 764±5.82

249

Table D.5: IFDs for Toowoomba Airport (41529) for RCP 4.5 climate change scenario for the period 2026-2045. The values are given with 95% confidence interval.

ARI (years)

1 2 5 10 20 50 100

5 Min 9.28±0.04 9.50±0.03 10.0±0.04 10.3±0.05 10.5±0.06 10.9±0.08 11.2±0.10

10 Min 16.3±0.22 17.1±0.16 18.9±0.09 20.0±0.14 21.1±0.22 22.6±0.33 23.7±0.41

15 Min 20.4±0.42 22.3±0.31 26.3±0.25 29.0±0.38 31.5±0.55 34.8±0.79 37.3±0.97

30 Min 24.9±0.29 29.5±0.24 39.1±0.61 45.5±0.92 51.6±1.22 59.5±1.63 65.5±1.93

1 Hr 26.7±0.35 32.7±0.53 45.5±1.30 54.0±1.86 62.1±2.40 72.6±3.10 80.5±3.62

2 Hr 30.7±0.92 38.9±1.41 56.3±3.43 67.8±4.88 78.8±6.29 93.1±8.12 104±9.49

3 Hr 33.6±1.25 41.8±1.87 59.8±3.11 71.6±4.61 83.2±6.44 97.3±7.54 108±8.37

6 Hr 51.5±0.48 64.6±0.36 92.3±0.79 111±1.21 128±1.63 151±2.19 168±2.61

12 Hr 77.4±0.48 98±0.47 142±0.93 170±1.34 198±1.74 234±2.27 260±2.67

24 Hr 104±0.40 131±0.46 190±0.95 229±1.35 266±1.74 314±2.25 350±2.63

250


ARI (years)

1 2 5 10 20 50 100

5 Min 9.31±0.04 9.69±0.03 10.5±0.03 11.1±0.05 11.6±0.08 12.3±0.11 12.8±0.13

10 Min 16.7±0.16 17.8±0.10 19.9±0.10 21.3±0.18 22.7±0.27 24.5±0.39 25.8±0.48

15 Min 21.7±0.29 23.8±0.24 28.3±0.21 31.2±0.27 34.1±0.36 37.8±0.49 40.5±0.60

30 Min 28.7±0.38 34.3±0.35 46.1±0.77 53.9±1.14 61.4±1.51 71.1±1.99 78.3±2.36

1 Hr 32.0±0.35 39.9±0.33 56.5±0.69 67.5±1.00 78.1±1.32 91.8±1.73 102±2.05

2 Hr 37.3±0.93 46.5±1.17 65.9±2.08 78.7±2.78 91.0±3.47 107±4.38 119±5.06

3 Hr 38.9±0.75 49.5±1.39 70.3±1.85 84.1±2.16 97.3±2.46 114±2.86 127±3.16

6 Hr 61.6±0.29 76.1±0.28 107±0.58 127±0.84 146±1.10 171±1.44 190±1.70

12 Hr 101±0.47 129±0.53 188±1.00 228±1.38 266±1.76 315±2.26 351±2.64

24 Hr 131±0.57 170±0.63 252±1.11 306±1.51 358±1.91 425±2.45 475±2.86

251


ARI (years)

1 2 5 10 20 50 100

5 Min 9.34±0.04 9.47±0.03 9.74±0.06 9.91±0.09 10.1±0.12 10.3±0.16 10.5±0.19

10 Min 16.1±0.17 16.8±0.13 18.4±0.14 19.4±0.20 20.4±0.27 21.7±0.38 22.6±0.46

15 Min 20.2±0.29 21.8±0.24 25.1±0.24 27.2±0.31 29.3±0.41 32.0±0.55 34.0±0.66

30 Min 23.4±0.50 27.1±0.44 34.9±0.63 40.1±0.89 45.1±1.16 51.5±1.53 56.4±1.82

1 Hr 24.6±0.37 29.9±0.49 41.2±1.13 48.6±1.60 55.8±2.06 65.0±2.66 71.9±3.12

2 Hr 29.9±1.30 36.7±1.09 51.0±2.07 60.5±3.07 69.6±4.09 81.4±5.43 90.2±6.45

3 Hr 31.8±1.57 39.0±1.44 53.4±2.70 64.1±3.42 73.6±3.97 85.1±4.70 94.5±5.64

6 Hr 47.2±0.35 57.3±0.38 78.8±0.67 93.0±0.91 107±1.15 124±1.48 138±1.73

12 Hr 76.2±0.33 91.7±0.38 125±0.67 146±0.91 167±1.15 194±1.46 215±1.70

24 Hr 104±0.47 128±0.47 178±0.95 211±1.37 243±1.78 284±2.33 315±2.74

252

Table D.8: IFDs for Toowoomba Airport (41529) for RCP 8.5 climate change scenario for the period 2081-2100. The values are given with 95% confidence interval

ARI (years)

1 2 5 10 20 50 100

5 Min 9.03±0.04 9.48±0.03 10.4±0.04 11.1±0.05 11.7±0.06 12.5±0.08 13.1±0.10

10 Min 15.7±0.22 17.0±0.16 19.7±0.09 21.5±0.14 23.2±0.22 25.4±0.33 27.1±0.41

15 Min 20.0±0.42 22.4±0.31 27.6±0.25 31.0±0.38 34.3±0.55 38.6±0.79 41.8±0.97

30 Min 26.2±0.29 32.1±0.24 44.5±0.61 52.7±0.92 60.6±1.22 70.8±1.63 78.4±1.93

1 Hr 28.9±0.35 37.5±0.53 55.7±1.30 67.7±1.86 79.3±2.40 94.3±3.10 105±3.62

2 Hr 34.2±0.92 45.9±1.41 70.5±3.43 86.8±4.88 102±6.29 123±8.12 138±9.49

3 Hr 36.4±1.25 48.3±1.87 73.9±3.11 91.9±4.61 107±6.44 128±7.54 143±8.37

6 Hr 56.5±0.48 74.7±0.36 113±0.79 139±1.21 163±1.63 195±2.19 219±2.61

12 Hr 77.2±0.48 110±0.47 179±0.93 224±1.34 268±1.74 325±2.27 367±2.67

24 Hr 84.9±0.40 143±0.46 265±0.95 346±1.35 423±1.74 524±2.25 599±2.63

253

Appendix E Table E1: The estimated percentage changes in the future pollutant build-up and pollutant wash-off for Coastal-SEQ

a. Pollutant build-up

Q1 Q2 Q3

Historical (2002-2015) - - -

RCP 4.5 (2026-2045) 8 15 16

RCP 4.5 (2081-2100) 9 17 15

RCP 8.5 (2026-2045) 11 18 16

RCP 8.5 (2081-2100) 11 19 17

b. Pollutant wash-off (Pollutant export)

Q1 Q2 Q3

Historical (2002-2015) - - -

RCP 4.5 (2026-2045) -29 15 31

RCP 4.5 (2081-2100) -25 9 34

RCP 8.5 (2026-2045) -25 -10 22

RCP 8.5 (2081-2100) -36 -1 27

254

Table E2: The estimated percentage changes in the future pollutant build-up and pollutant wash-off for Inland-SEQ

a. Pollutant build-up

Q1 Q2 Q3

Historical (2010-2015) - - -

RCP 4.5 (2026-2045) -4 -1 3

RCP 4.5 (2081-2100) 3 5 6

RCP 8.5 (2026-2045) 5 5 7

RCP 8.5 (2081-2100) -1 2 4

b. Pollutant wash-off (Pollutant export)

Q1 Q2 Q3

Historical (2010-2015) - - -

RCP 4.5 (2026-2045) -21 -13 9

RCP 4.5 (2081-2100) -21 2 4

RCP 8.5 (2026-2045) -30 -21 3

RCP 8.5 (2081-2100) -15 -13 2

255

Table E3: The estimated percentage changes in the future stormwater quality for Coastal-SEQ

EMC

Q1 Q2 Q3

Historical (2002-2015) - - -

RCP 4.5 (2026-2045) -51 -50 -48

RCP 4.5 (2081-2100) -56 -53 -51

RCP 8.5 (2026-2045) -53 -48 -45

RCP 8.5 (2081-2100) -54 -47 -48

Table E4: The estimated percentage changes in the future stormwater quality for Inland-SEQ

EMC

Q1 Q2 Q3

Historical (2010-2015) - - -

RCP 4.5 (2026-2045) -42 -38 -44

RCP 4.5 (2081-2100) -52 -48 -52

RCP 8.5 (2026-2045) -46 -44 -46

RCP 8.5 (2081-2100) -45 -42 -50