assessment of climate change impact on water

ASSESSMENT OF CLIMATE CHANGE IMPACT ON WATER RESOURCES AND HYDROPOWER IN THE JHELUM RIVER BASIN, PAKISTAN

by

Rashid Mahmood

A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Engineering in Water Engineering and Management

Examination Committee:

External Examiner:

Nationality: Previous Degree:

Scholarship Donor:

Dr. Mukand S. Babel (Chairperson) Dr. Sutat Weesakul Dr. Nitin Kumar Tripathi Dr. Sangam Shrestha

Prof. Tetsuya Sumi Water Resources Research Center Disaster Prevention Research Institute Kyoto University Japan

Pakistani Master of Engineering in Water Engineering and Management Asian Institute of Technology Thailand Higher Education Commission of Pakistan

Asian Institute of Technology School of Engineering and Technology Thailand August 2013 i

ACKNOWLEDGEMENTS In the process of the completion of my doctoral degree, many people have contributed to make it possible. I would love to express sincere gratitude to all of them. The most sincere thanks are offered to my supervisor, Professor Mukand S. Babel, who gave me invaluable comments, a great guidance, a genuine caring, and an encouragement throughout the progression of my theses. I am also very grateful to my doctoral committee members—Dr. Sutat Weesakul, Dr. Nitin Kumar Tripathi, and Dr. Sangam Shrestha—for their critical suggestions, academic guidance, and invaluable comments. I would like to acknowledge and offer gratitude to the Pakistan Meteorological Department (PMD), the Water and Power Development Authority (WAPDA) of Pakistan, the India Meteorological Department (IMD), and the National Climate Data Center (NCDC), for providing the important and valuable data for this research. Heartfelt gratitude is also extended to the Higher Education Commission (HEC) of Pakistan and Asian Institute of Technology (AIT) providing financial support to the first author for his doctoral studies at AIT. I greatly admitted that without HEC support, it might be difficult to complete my research work. I would love to express many thanks to all of my family members for giving me all kinds of support and encouragement, and to the friends who have helped me directly or indirectly carrying out this study. At the end, I want to offer special thanks to my mother and her uncountable sacrifices for me during whole of my life. To get higher education was a dream of my mother. So, I am very happy to complete my mother’s dream. I dedicate this dissertation to whole of my family especially my mother.

ii

ABSTRACT The increasing concentration of greenhouse gases in the atmosphere has been changing the climate of the world and has been causing global warming. The warming leads to changes in evapotranspiration, precipitation, soil moisture, and infiltration. Precipitation changes affect water availability in soils, rivers and lakes, domestic and industrial water supply and demand, hydropower generation, and agricultural productivity. Thus, the changes in climate cause some serious effects on the hydrological system. To date, these effects are assessed by Global Climate Models (GCMs). However, these models have course spatial resolution which is not suitable to assess the impacts of climate changes on hydrologic system at local/regional scale. Thus, many downscaling techniques have been developed to overcome the spatial mismatch between GCM and regional scale. The focus of this study is to investigate the changes in climate variables (temperature and precipitation), and their consequences on the stream flow and hydropower under the IPCC emission scenarios in the Jhelum River basin, Pakistan and India. Moreover, this study also covers the evaluation of several GCMs for selecting a suitable GCM, and the evaluation of SDSM (Statistical Downscaling Model) developed by monthly and annual sub-models for downscaling max temperature, min temperature, and precipitation. The evaluation of four GCMs (CSISRO-Mk2, CGCM2 and CCSRIES, and HadCM3), selected according to the data availability, is performed by assessing three statistical indicators; coefficient of determination (R²), root mean square error (RMSE), and standard deviation (б). These indicators are calculated by using the monthly time series (max temperature, min temperature, and precipitation) of observed and raw GCMs (A2 and B2) for the period of 1991-2009. The results show that the temperature (max and min temperature) is underestimated by all GCMs in magnitude, although the pattern is well captured by each GCM. In case of precipitation, no one of GCMs captures the variation as well as magnitude of observed precipitation. Nonetheless, HadCM3 is selected for downscaling of temperature and precipitation on the basis of above mentioned indicators, graphical visualization, and data availability. This study evaluates the statistical downscaling model (SDSM) developed by annual and monthly sub-models for downscaling maximum temperature, minimum temperature and precipitation, and assesses future changes in climate in the Jhelum River basin, Pakistan and India. Additionally, bias correction is applied on downscaled climate variables. The mean explained variances of 66, 76, and 11% for max temperature, min temperature, and precipitation respectively, are obtained during calibration of SDSM with NCEP predictors, which are selected through a quantitative procedure. During validation, average R² values by the annual sub-model (SDSM-A)—followed by bias correction using NCEP, H3A2 and H3B2—lie between 98.4 to 99.1% for both max and min temperature, and 77 to 85% for precipitation. As for the monthly sub-model (SDSM-M), followed by bias correction, average R² values lie between 98.5 to 99.5% for both max and min temperature and 75 to 83% for precipitation. These results indicate a good applicability of SDSM-A and SDSM-M for downscaling max temperature, min temperature, and precipitation under H3A2 and H3B2 scenarios for future periods of the 2020s, 2050s, and 2080s in this basin. Both submodels show a mean annual increase in max temperature, min temperature, and precipitation. Under H3A2, and according to both sub-models, changes in max temperature, min temperature and precipitation are projected as 0.91-3.15°C, 0.93-2.63°C and 6-12%, and iii

under H3B2, the values of change are 0.69-1.92°C, 0.56-1.63°C and 8-14% in 2020s, 2050s and 2080s. These results show that the climate of the basin will be warmer and wetter relative to the baseline period. SDSM-A, most of the time, projects higher changes in climate than SDSM-M. It can also be concluded that although SDSM-A performed well in predicting mean annual values, it cannot be used with regard to monthly and seasonal variations, especially in the case of precipitation unless correction is applied. The impacts of climate change on the stream flow under IPCC emission scenarios, A2 and B2, are assessed by using downscaled temperature and precipitation data as inputs to HECHMS. The model is calibrated and validated using observed daily stream flow for the period of 1982-89 and 1978-81 respectively at various stream flow gauges in the Jhelum River basin. HEC-HMS includes a loss method (Deficit and Constant), a transform method (SCSunit hydrograph), as well as a base flow method (Recession) for calculating total stream flow from the basin, and temperature index is used to take care of snow fall. Performance indicators (e.g., Nash-Sutcliffe efficiency, coefficient of determination and percent deviation) and graphical visualizations between observed and simulated stream flow indicate that the variations and magnitudes of observed data are well captured by the simulated data during calibration and validation. The changes in 2020s (2011-2040), 2050s (2041-2070), and 2080s (2071-2099) are assessed relative to baseline period (1971-2000). Thus, the simulated discharge, under A2, shows a mean annual increase in 2020s, 2050s, and 2080s at all gauging station except at Domel in 2050s, as do the simulated discharge under B2 except at Muzaffarabad in 2050s. The mean annual discharge at Azad-Pattan, the main gauging station that contributes about 87% as annual inflow to Mangla reservoir, is projected to increase about 24 to 30% under A2 and B2 at Azad-Pattan stream gauge. According to the flow changes at Azad-Pattan stream gauge under A2 and B2, the high flows in the Jhelum basin are projected to decrease in 2020s and 2080s, with 1-7%, but increase in 2050s, with 6-6.7%. The median flows are predicted to increase in all three periods, with 2036% increase w. r. to baseline. On the other hand, the low flows could be increased in 2020s and 2080s, with 2-8%, and decreased in 2050s under both A2 and B2 scenarios, with 1214%. Finally, the stream flow time series generated by HEC-HMS for the period of 2011-2099 are used as input to HEC-ResSim to find out the impacts of climate change on the hydropower from Mangla power plant. As Mangla dam has recently raised by another 12.2 m which has changed the physical characteristics of dam. For example before raising of dam the conservation level was 366.4 m but after raising it has become 378.6 m. The model is setup for the Mangla dam for two situations: before raising and after raising. The hydropower is simulated by feeding the observed inflow to Mangla reservoir for baseline period, and by storing stream flow from HEC-HMS for three future periods (2020s, 2050s, and 2080s). At the end, the hydropower simulated under before-raising and after-raising condition of dam is compared to find of the impacts of raising of dam on the hydropower production. Before-raising as well as After-raising conditions, the results show increases in all three future periods and in all seasons both under A2 and B2 but with different magnitudes. Under A2 and B2, the increase in annual hydropower ranges between 16.6-20.4% (Before-raising) and 13-15.3% (After-raising). It is also observed that the changes in hydropower under B2 are higher than A2 same like in case of before-raising conditions. In case of comparison of hydropower before and after-raising, it is seen that that the power generation is increased by 10.8%. After-raising of dam, the variations in hydropower generation are also decrease. iv

TABLE OF CONTENTS CHAPTER

TITLE

PAGE

TITLE PAGE ACKNOWLEDGEMENTS ABSTRACT TABLE OF CONTENTS LIST OF FIGURES LIST OF TABLES LIST OF ABBREVIATIONS

i ii iii v vii x xiii

1

INTRODUCTION 1.1 Background 1.2 Statement of the problem 1.3 Objectives of the study 1.4 Scope and limitations of the study 1.5 Organization of the report

2

LITERATURE REVIEW 2.1 Global Climate Model 2.2 Description of IPCC emission scenarios 2.3 Downscaling 2.4 Climate change impact on water resources 2.5 Climate change impact on the hydropower 2.6 Conclusions

8 8 9 10 20 23 26

3

STUDY AREA AND DATA 3.1 Indus River basin 3.2 Upper Jhelum River basin 3.3 Data description 3.4 Conclusions

27 27 31 39 40

4

OVERALL METHODOLOGY 4.1 Evaluation of Global Climate Models 4.2 Statistical downscaling 4.3 Evaluation of sub-models of SDSM 4.4 Climate change impact on water resources 4.5 Climate change impact on hydropower

41 41 41 42 42 43

5

EVALUATION OF GLOBAL CLIMATE MODELS 5.1 Data description 5.2 Methodology 5.3 Results and discussion 5.4 Conclusions

45 45 45 47 49

6

STATISTICAL DOWNSCALING 6.1 Introduction 6.2 Sub-basins of the study area

51 51 53

v

1 1 4 5 5 6

6.3 6.4 6.5 6.6 6.7

Data description Methodology Results and discussion Downscaling of temperature and precipitation Conclusion

56 56 59 69 76

7

EVALUATION OF SUB-MODELS OF SDSM 78 7.1 Introduction 78 7.2 Data description 80 7.3 Monthly to daily data conversion 83 7.4 Methodology 84 7.5 Results and discussion 89 7.6 Downscaling (with bias correction) of temperature and precipitation 97 7.7 Conclusions 100

8

CLIMATE CHANGE IMPACT ON WATER RESOURCES 8.1 Data description 8.2 Methodology 8.3 Results and discussion 8.4 Conclusions

102 102 105 109 131

9

CLIMATE CHANGE IMPACT ON THE HYDROPOWER 9.1 Data description 9.2 Methodology 9.3 Results and discussions 9.4 Conclusions

133 133 137 140 149

10

SUMMARY, CONCLUSIONS, AND RECOMMENDATIONS 10.1 Summary 10.2 Conclusions 10.3 Recommendations

151 151 155 156

REFERENCES

158

APPENDICES

167

vi

LIST OF FIGURES FIGURE 1-1

1-2 2-1 2-2 2-3 2-4 2-5 2-6 3-1 32 3-3 3-4 3-5 4-1 5-1 5-2

6-1 6-2

6-3

6-4

6-5

6-6

TITLE Mean annual anomalies for precipitation (%) over land for 19002005 relative to the 1961-1990, the grey areas indicate missing data Annual changes in runoff in 2090–2099 relative to 1980–1999 under A1B1 Schematic illustration of Global Climate Model (GCM) Emission scenarios and their driving forces A schematic illustration of downscaling and aggregation Comparison of statistical and dynamical downscalings Schematic illustration of weather typing approach Schematic illustration of transfer function approach Geographic location of Pakistan Location map of the Indus River basin Location map and description of the study area Monthly (a) temperature and (b) precipitation at weather stations for 1961-1990 in the Jhelum basin Mean monthly runoff, % of annul runoff, at different stream stations in the Jhelum basin Research framework of the study Grid presentation of selected GCMs Mean monthly (A) precipitation, (B) Tmax, and (C) Tmin of different GCMs against the observed data in the Jhelum River basin Location map of the Jhelum basin and climate stations used in the present study Distribution of mean monthly rainfall over (a) all weather stations, (b) TPP basin, OPP basin, and whole basin for 19611990 Distribution of mean monthly temperature at (a) all weather stations, (b) TPP basin, OPP basin, and whole basin for 19611990 Observed vs. simulated mean monthly, seasonal, and annual Tmax for (a) TPP basin (b) OPP basin, and (c) whole Jhelum basin, during validation (1991-2000) Observed vs. simulated mean monthly, seasonal, and annual Tmin for (a) OPP basin, (b) TPP basin, and (c) whole Jhelum basin, during validation (1991-2000) Observed vs. simulated mean monthly, seasonal, and annual precipitation for (a) TPP basin, (b) OPP basin, and (c) whole Jhelum basin, during validation (1991-2000) vii

PAGE 2

3 8 10 11 16 17 18 28 30 33 35 36 44 48 50

54 55

56

64

65

66

6-7

6-8

6-9

7-1 7-2 7-3

7-4

7-5

7-6

8-1 8-2 8-3 8-4 8-5 8-6 8-7 8-8 8-9 8-10 8-11

Spatial distributions of observed against downscaled mean annual (A) Tmax, (B) Tmin, and (C) precipitation for validation (1991-2000), in the Jhelum River basin Future changes in mean season and annual (A) Tmax, (B) Tmin, and (C) precipitation in 2020s, 2050s and 2080s with respect to 1961-1990 under H3A2 and H3B2 emission scenarios, in the TPP basin, OPP basin, and whole Jhelum basin Spatial changes of mean annual (A) Tmax in Celsius, (B) Tmin in Celsius, (C) precipitation in percentage in the 2020s, 2050s and 2080s relative to 1961-1990 under H3A2 and H3B2 scenarios, in the Jhelum River basin Description of MODAWEC model CRU grids representation covering study area NCEP selected predictors for calibration (1961-1990) of Tmax (red), Tmin (green) and precipitation (black) for SDSM’s submodels in the Jhelum basin104 Observed and simulated mean monthly (a) Tmax (b) Tmin and (c) precipitation for the calibration period (1961-1990) in the Jhelum basin Observed and downscaled (without bias correction) mean monthly, seasonal and annual precipitation by (a) SDSM-A (b) SDSM-M for the validation period in the Jhelum basin Observed and downscaled (with bias correction) mean monthly, seasonal and annual precipitation by (a) SDSM-A and (b) SDSM-M for the validation period (1991-20001) in the Jhelum basin Soil groups available in the Jhelum River basin Land covers in the Jhelum River basin Schematic diagram of assessing the climate change impacts on water resources in the Jhelum River basin DEM of the Jhelum River basin Sub-basins derived from the DEM for the Jhelum River basin Location map of hydro-climatic stations in the Jhelum River basin, showing main tributaries as well Observed (doted) and simulated (solid) discharges at different stations for calibration (1982-89) in the Jhelum River basin Observed (doted) and simulated (solid) discharges at different stations for validation (1978-81) in the Jhelum River basin Flow duration curves under (A) A2 and (B) B2 scenarios at different stream gauges in the Jhelum River basin Future shifts of peak flows in timing and magnitude, under A2 and B2, in the Jhelum river basin Observed and predicted linear trends in center-of-volume date (CVD), under A2 and B2 scenarios, in the Jhelum river basin

viii

68

73

75

83 84 87

92

95

97

104 105 106 110 111 111 115 118 124 127 130

8-12

9-1 9-2 9-3 9-4 9-5

9-6

Relationships between changes in mean temperature, precipitation and flow in 2020s, 2050s and 2080s in the Jhelum river basin Mean monthly energy generation from Mangla power station for the period of 1995-2000 HEC-ResSim’s setup of watershed module for Mangla watershed HEC-ResSim’s setup of reservoir network module for Mangla watershed Simulated (A) power generation and (B) reservoir water level against the observed data, for 2005-2008 Power generated (green), flow used for power (red), and max capacity of power plant (blue) for Before and After-raising of Mangla dam, for 1997-2000 Reservoir water levels between (A) before and after raising (B) baseline and three future period under A2 (C) baseline and three future period under B2

ix

131

135 139 140 142 145

149

LIST OF TABLES TABLE 2-1 2-2 2-3 3-1 3-2 3-3 3-4 3-5 3-6 3-7

3-8 3-9 5-1 6-1 6-2 6-3 6-4 6-5

6-6

6-7

7-1 7-2 7-3

TITLE Comparison of statistical and dynamical downscaling techniques Advantages and disadvantages of different types of statistical downscaling Installed electricity generation capacity (MW) in Pakistan Country areas in the Indus River basin Mean annual discharge of the major rivers of the Indus basin Main characteristics of sub-basin of the Jhelum river basin Hydraulic features of the Jhelum River and its major tributaries Some statistics about climate stations for the period of 1961-1990 in the Jhelum River basin Storage capacities of the Mangla reservoir w. r. to elevations in the 1967 and 2000 Historical max, min and mean irrigation demand for Rabi and Kharif from the Mangla reservoir calculated for the period of 19672000 Some salient features of the Mangla dam Overview of main input data required for this study R2, RMSE, and SD calculated from observed and GCMs raw data for Tmax, Tmin and precipitation for the period of 1981-2009 Some statistics about climate stations for the period of 1961-1990, in the Jhelum River basin NCEP predictors used in the screening process Selected predictors and their mean absolute partial correlation coefficient during screening Calibration results comparison with some previous studies using Explained variance (%) Statistical comparison of observed and simulated daily, monthly, and seasonal time series of Tmax during validation (1991-2000), in the Jhelum River basin Statistical comparison of observed and simulated daily, monthly, and seasonal time series of Tmin during validation (1991-2000), in the Jhelum River basin Statistical comparison of observed and simulated daily, monthly, and seasonal time series of precipitation during validation (19912000), in the Jhelum River basin Geographic, climatic, and data availability information of weather stations used in the present study NCEP predictors used in the screening process Screening of most effective predictors for precipitation at the Astore climate station x

PAGE 16 19 24 27 29 32 33 34 37 38

39 40 49 55 58 59 60 61

61

62

82 86 86

7-4 7-5 7-6

7-7

7-8

7-9 7-10 7-11 8-1 8-2 8-3 8-4

8-5 8-6 8-7 8-8

8-9

8-10

9-1 9-2 9-3

Screening of most effective predictors for precipitation at the Jhelum climate station Explained variance (E) and standard error (SE) during calibration (1961-1990) Statistical comparison of observed and downscaled mean monthly Tmax, Tmin and precipitation by two sub-models during calibration (1961-1990) Statistical comparison of observed and downscaled (without bias correction) mean monthly Tmax, Tmin and precipitation by two sub-models during validation (1991-2000) Statistical comparison of observed and downscaled (with bias correction) mean monthly Tmax, Tmin and precipitation by two sub-models during validation (1991-2001) Future changes in Tmax (°C) with respect to baseline (1961-1990) under H3A2 and H3B2 scenarios with two sub-models Future changes in Tmin (°C) with respect to baseline (1961-1990) under H3A2 and H3B2 scenarios with two sub-models Future changes in precipitation (%) with respect to baseline (19611990) under H3A2 and H3B2 scenarios with two sub-models Meteorological stations located inside and around the study area Stream flow stations located in the study area Classification of land covers in the Jhelum River basin Nash-Sutcliffe efficiency (E), coefficient of determination (R2), and percent deviation (D) for calibration (1982-89) and validation (1978-81) for different stream gauges in the Jhelum River basin Nash-Sutcliffe efficiency (E), coefficient of determination (R2), and percent deviation (D) for some previous studies Future changes (%) in stream flow at different gauges relative to baseline (1961-1990) under A2 scenarios in the Jhelum River basin Future changes (%) in stream flow at different gauges relative to baseline (1961-1990) under B2 scenarios in the Jhelum River basin Future changes (%) in low, median and high flow relative to baseline (1961-1990) at different stream gauges under A2 scenarios in the Jhelum River basin Future changes (%) in low, median and high flow relative to baseline (1961-1990) at different stream gauges under B2 scenarios in the Jhelum River basin Future changes in center-of-volume dates (CVD) with respect to baseline (1961-1990) at different stream flow stations under both scenarios, A2 and B2, in the Jhelum river basin Max discharge capacity of main outlets of Mangla dam at different elevations (A) before-raising project (B) after-raising project Mean monthly evaporation from Mangla reservoir Max and min rule curve for the Mangla reservoir, before-raising and after-raising conditions xi

90 91 91

94

96

98 99 100 103 103 105 112

113 119 120 125

125

128

133 134 136

9-4 9-5 9-6 9-7 9-8

9-9 9-10 9-11

9-12 9-13

Irrigation indent/demand used during the feasibility report of Mangla raising project Summary for the physical and operation data used in HEC-ResSim for hydropower generation Future changes (%) in hydropower relative to baseline (1961-1990) under A2 and B2 from the Mangla plant before-raising conditions Future changes (%) in hydropower relative to baseline (1961-1990) under A2 and B2 at Mangla plant after-raising conditions Future changes (%) in hydropower generation relative to baseline period under A2 scenario, increasing the irrigation demand by 10, 20, and 30%, at Mangla power station Indicators for the comparison of Mangla dam with and without raising Description of reservoir performance indicators Performance indicators for after-raising and before-raising of dam for 1961-1990, and their changes after-raising relative to beforeraising Future changes in performance indicators relative to 1961-1990 (after-raising), under A2 Future changes in performance indicators relative to 1961-1990 (after-raising), under B2

xii

136 137 141 143 143

144 146 147

147 148

LIST OF ABBREVIATIONS ABBREVIATION ANN CC CICERO DEM ECMWF GCM GHG GPH IMD IPCC KWh LR MAF MLR MSLP MW NCEP NLR PCA PMD RCM SD SDSM SH TEAM WAPDA

DESCRIPTION Artificial Neural Networks Canonical Correlation Center for International Climate and Environmental Research Oslo Digital Elevation Model European Centre for Medium-Range Weather Forecasts Global Climate Model or General Circulation Model Greenhouse Gases Geopotential Height India Meteorological Department Intergovernmental Panel on Climate Change Kilo Watt Hour Linear Regression Million Acre Feet Multiple Linear Regression Mean Sea Level Pressure Mega Watt National Centre for Environmental Protection Non Linear Regression Principal Component Analysis Pakistan Meteorological Department Regional Climate Model Statistical Downscaling Statistical Downscaling Model Specific Humidity Tool for Environmental Assessment and Management Water and Power Development Authority of Pakistan

xiii

1 1.1

INTRODUCTION

Background

A rapid increase in the concentration of greenhouse gases (CO2, methane, CFCs, water vapors, nitrous oxide, and ozone etc.) in the atmosphere due to human activities such as land use changes and extensive use of fossil fuels have caused global warming and global energy imbalance (Huang et al., 2011). The greenhouse gases (GHG) trap heat from the atmosphere and release it very slowly resulting changes in climate variables like temperature and precipitation (Gebremeskel et al., 2005). Global warming in recent decades is obviously evident in increasing global average air and ocean temperatures, rising mean sea level, and widespread melting of snow and glaciers. The net anthropogenic radiative forcing of climate is likely to increase (warming effect), with a best estimation of 1.6 Wm-2 for 2005 (relative to 1750 pre-industrial values). According to the linear trend in global surface temperature for the period of 1906 to 2005, a warming of 0.74°C (likely range 0.56 to 0.92°C) has been observed. However, a more rapid warming trend has been detected over the past 50 years, with a rising rate of 0.13°C per 10 years. This rapid increase of global mean temperature since the mid 20th century is very likely due to the observed rapid increase in GHG concentrations (Bates et al., 2008). Further global warming and changes in the climate system are projected at the rate of continued increasing concentrations of GHG or above this rate under the SRES (Special Report on Emission Scenarios) during the 21st century. These changes would likely be greater than the observed changes of the 20th century. The projected change in global average temperature ranges between 1.8°C (likely range 1.1°C to 2.9°C under B1 scenario) and 4.0°C (likely range 2.4°C to 6.4°C under A1FI scenario) in the period of 2090-2099, relative to the 1980-1999 period. This projected warming would be greater over land than the oceans and also greater over the northern latitude than southern oceans as well as some parts of the Atlantic Ocean (Bates et al., 2008). Many studies show that precipitation has generally increased over the 20th century from 30°N to 85°N, but remarkable decreases have occurred between 10°S and 30°N in the past 30 to 40 years (Figure 1-1). Precipitation, between 10°N and 30°N, increased distinctly from 1900 to the 1950s, but declined after the 1970s. No strong hemispheric-scale trends over the Southern Hemisphere’s extra-tropical land masses have been observed. Since precipitation is strongly influenced by large-scale patterns of natural variability, the attribution of changes in global precipitation is uncertain. Since 1901, annual precipitation has shown the largest negative trends over Western Africa, the Sahel, and South Asia. Since 1979, however, increasing trends have been seen in the Sahel region and in other parts of tropical Africa. Over much of North-Western India, an increase in annual precipitation of more than 20% per century was observed during the period of 1901–2005, with a strong decrease detected since 1979. Climate projections from multi-models indicate that the global average mean precipitation, evapotranspiration and water vapor would increase in the 21st century. The increase in precipitation will generally be seen in the area of regional tropical precipitation maxima (e.g., monsoon regions and tropical Pacific), and general decreases will occur in the subtropic. An annual increase in precipitation exceeding 20% will take place in most high 1

latitudes (e.g., Eastern Africa, the northern part of Central Asia and equatorial Pacific Ocean) in the period of 2080-2099. However, the Mediterranean and Caribbean regions and on the sub-tropical western coasts of each continent will receive a substantial decrease in precipitation. Overall, the increase of precipitation over the land and oceans are predicted as 5 and 4% respectively (Bates et al., 2008).

(Source: (Bates et al., 2008)) Figure 1-1 Mean annual anomalies for precipitation (%) over land for 1900-2005 relative to the 1961-1990, the grey areas indicate missing data A large number of studies have been conducted to analyze the potential trend of river discharge during the 20th century from catchment to global scale. Some studies have found significant trends and relations with precipitation or temperature. However, studies are yet to examine the considerable trends and effects of variation in temperature and precipitation on the runoff from human interferences in the catchment. A broadly coherent pattern of change in the annual discharge has been observed at the global scale. Some regions (e.g., high latitudes and large parts of the USA) are experiencing an increase in the runoff. However, the runoff is decreased in other parts of the world such as West Africa, Southern Europe and the southernmost parts of South America. Annual variations in discharge are influenced by large-scale climatic patterns (e.g., ENSO and NAO Pattern) in many parts of the world (Milly et al., 2005). One study (Labat et al., 2004) claimed that the global total runoff has increased about 4% per 1°C rise in temperature during the 20th century. There is always no consistency in the trends of runoff with the changes in precipitation. This might be because of the effects of human interventions such as reservoir impounding, the data limitations such as coverage of 2

precipitation or the contending effects of changes in temperature and precipitation. However, the more strong and widespread evidence is the significant changes in the timing of river flows in many parts of the globe where precipitation falls as snow. Due to the rise in mean temperature, a great amount of the winter precipitation occurs as rainfall instead of snow, and the snow melting season begins earlier. In New England, snowmelt shifted ahead by 1 to 2 weeks between 1936 and 2000, though this has caused little effect on the flows in summer (Hodgkins et al., 2005; Bates et al., 2008). Several studies from Europe, North America, as well as Australasia, and with a small number of studies from Asia have been conducted to find out the potential impacts of climate change on the river flows. These studies show increases in runoff at high latitudes as well as some parts of wet tropics, and point out decreases in the mid-latitudes and some parts of dry tropics as shown in Figure 1.2 (Bates et al., 2008).

(Source: (Bates et al., 2008)) Figure 1-2 Annual changes in runoff in 2090–2099 relative to 1980–1999 under A1B The hydrological system is sensitive to changes in climate. The interactions between increases in greenhouse gases and the hydrological system are very complex. Increases in temperature will result in changes in evapotranspiration, soil moisture, and infiltration. Increased atmospheric CO2 may increase global mean precipitation as indicated by all GCMs. Changes in rainfall could affect water availability in soils, rivers and lakes, with implications for domestic and industrial water supplies, hydropower generation, and agricultural productivity. Increased evapotranspiration enhances the water vapor content of the atmosphere and the greenhouse effect, and the global mean temperature rises even higher. Land use will also play a key role in increased evapotranspiration. Possible changes in temperature, precipitation and evapotranspiration may result in changes in soil moisture, ground water recharge and runoff which could intensify flooding and droughts in various parts of the world (Mirza and Ahmad, 2005).

3

Hydropower plays an important role in the sustainable contribution to meet the electricity demand worldwide. Hydropower plants had contributed to about 19% (2500 TWh) of the total world electricity demands by the middle of 1990. The importance of hydropower, along with other renewable energy sources, is expected to increase in the future. Since 1980, hydroelectricity has been increasing steadily by an average of 2.3% every year. It is expected that the average growth rate of world hydropower production will be increased by 2.4 to 3.6% per year in 2020. The developing and industrialized countries are expected to contribute highest in this growth rate of electricity. Hydroelectric generation is directly proportional to runoff, where the greater the volume of water discharged, the greater the generation potential (Lehner et al., 2005). Recently, Pakistan is in deep crisis of electricity. Pakistan has electricity generation capacity of about 19,547 MW. Installed hydropower capacity is 6,599 MW out of which 1,000 MW comes from the Mangla power station. The electricity demand will be increased by an annual compound growth rate of 7.9% in the future. (Mirza and Ahmad, 2005). 1.2

Statement of the problem

The major challenges to water management planners are the uncertainties in the availability of water demand in the future. Climate change and its impacts on the hydrological cycle are adding to these uncertainties. Climate change has affected Asia as a result of increase in sea level, along with extreme events such as flood and drought, rise in temperature, and availability of rainfall, etc. According to IPCC’s Fourth Assessment Report, the projected availability of freshwater resource in Asia is a decreasing trend by 2050 due to climate change affecting more than a billion people (Sharma, 2007). According to CICERO, the temperature is estimated to increase by 0.9°C and 1.8°C in 2020s and 2050s respectively in Pakistan, which would cause changes in precipitation by 3% (2020s) and 6% (2050s). It is also projected that the sea level will rise by 20 cm and 30 cm in 2020 and 2050 respectively. It is also predicted that summer season will be relatively wetter to present with an increase in rainfall of 17% in Pakistan. A small change in the climate would cause significant impacts on the water resources availability, as experienced in some parts of the world in recent years. Due to the increasing concentration of GHGs in the atmosphere, climate change is further causing high risk to water resource in Pakistan. The precipitation and runoff are the most affected parameters of climate change. It is reported that a small decrease in precipitation could have large impacts on the water supply. And this is a concerning issue for an arid to semi-arid region like Pakistan (Mirza and Ahmad, 2005). Recently, Archer and Fowler (2008) conducted a study to forecast the seasonal runoff in Upper Jhelum basin by using regression method. In this study, the meteorological climate data and runoff were linked together by using multiple regression approach and the inflow pattern was shown for the Mangla dam which is a major structure contributing to the Indus Basin irrigation system of Pakistan. It was concluded that seasonal forecasting to Mangla dam may still be improved by using a more sophisticated model like physical-based hydrologic model. Thus, it is of a great significance to find out the impacts of climate change on water resources in Pakistan to adopt certain suitable strategies and policies that may be helpful for a 4

sustainable development. This is possible only by simulating and analyzing some feasible future scenarios. To date, the main tools to predict these impacts are General Circulation Models (GCM) coupled with hydrologic or water management model. These models are very useful tools to see the climate scenarios but they are very coarse in spatial resolution (250-600 km) in order to assess the environmental impacts on the regional scale and to use in planning, such as flood risk planning in hydrologic modeling (Wilby et al., 2000; Gebremeskel et al., 2005). In addition, the hydrologic studies need very fine spatial and temporal resolution. To fulfill this gap the most important tool is downscaling to make the outputs of these models useful for local-scale water management (Wetterhall et al., 2006). 1.3

Objectives of the study

The overall objective of the study is to assess the impacts of climate change on the stream flow and hydropower production using a statistical downscaling in the Upper Jhelum basin. The specific objectives of the study are: 1) To evaluate different GCMs for the selection of a suitable GCM for further analysis in the study area. 2) To investigate the temporal and spatial future change in maximum and minimum temperatures, and precipitation in the Jhelum river basin. 3) To evaluate the Statistical Downscaling Model (SDSM) developed by monthly and annual sub-model. 4) To assess the impacts of climate change on the stream flow in the Jhelum river basin. 5) To assess the impacts of climate change on the hydropower production at the Mangla power station. 1.4

Scope and limitations of the study

To date, many GCMs have been developed to investigate the changes in world climate. However, all these models cannot be used for downscaling of temperature and precipitation. So, in this study, the four (4) GCMs are evaluated by using some statistics. These statistics are calculated by the simulated data of GSM and observed data. On the basis of these statistics, a suitable GCM is selected for downscaling. This study also analyzes the changes in mean temperature and precipitation in the future with respect to the baseline data (1961-1990). The temperature and precipitation data is downscaled by SDSM for the 21st century under A2 and B2, which is used in hydrologic modeling. There are two different sub-model of SDSM to downscale the climate parameters. In this study, these methods are evaluated with the help of observed data, and it was found that the monthly sub-model give better results than annual sub-model. The temperature and precipitation changes in the future are also investigated by both sub-models. This study also covers the future changes in stream flow and hydropower relative to baseline period (1961-1990) in the Jhelum basin. To find out these changes, the downscaled temperature and precipitation data from SDSM is fed into the HEC-HMS to simulate the stream data at different sites in the Jhelum river basin. In the end, the simulated stream flow is used in HEC-ResSim to generate the hydropower from the Mangla power plant. 5

The limitations of this study are given below: 1- The observed meteorological station data (14 stations) is too scarce to examine spatial variability. 2- The climate stations are located in valleys and not on higher elevations, which may affect the relationship between climate variables (temperature and precipitation) and inflow. 3- Only a few climate scenarios are possible due to time constraints. 4- Land use changes are assumed constant during the simulation periods. 5- The predictors for downscaling are available only for two GCMs (HadCM3 and CGCM2) 1.5

Organization of the report

This report consists of 10 chapters covering the objectives of the study with brief description of each of the chapter as follows: Chapter 1 covers the background of study, rationale, objectives, scope and limitations. Chapter 2 provides an overview of the GCM, concepts of downscaling, main types of downscaling, and comparative studies on downscaling. Text on SCES scenarios and impacts of climate change on the stream flow and hydropower is also reviewed. Chapter 3 gives the description about the study area, such as the location, hydro-climatic condition, main rivers, and reservoirs in the study area. Chapter 4 provides a brief summary about the overall methodology to achieve all the objectives of the study. Chapter 5 describes the selection criteria for GCMs and evaluation of GCMs with observed data to select a suitable GCM for downscaling of temperature and precipitation. Chapter 6 includes the application of downscaling techniques. The future changes in temperature and precipitation are investigated not only on temporal but also on spatial basis. Chapter 7 evaluates the annual and monthly sub-models of SDSM. The future changes in temperature and precipitation are obtained by both methods. However bias correction is applied on the data produced by both sub-models. Chapter 8 consists of extraction of basin characteristic from 30 m DEM, description and application of hydrologic model, future changes in stream flow and flow duration curves under A2 and B2. Temporal shifts of streamflow and peak flows are also explored in this chapter. Moreover, relations between temperature, precipitation, and runoff are also drawn graphically. Chapter 9 presents the physical and operational data of the Mangla dam, characteristics of the Mangla dam before and after-raising of the dam, description of reservoir simulation model, and future changes in hydropower generation. In addition, the effects of raising of dam are also assessed.

6

Chapter 10 gives the whole summary, conclusions, and recommendations taken from this study. References and appendices are given at the end of the report.

7

2

LITERATURE REVIEW

In this chapter, recent studies related to global climate models are reviewed. These include regional climate model, concepts of downscaling, downscaling techniques, SCES scenarios, and impacts of climate change on water resource and hydropower which are required to achieve the objectives of this study. 2.1

Global Climate Model

Global Climate Models (GCMs) are numerical based models. These are the most advanced and complex models among the climate models because of the description in the components of climate system in three dimensions as shown in Figure 2-1 (Viner, 2000). These are the currently available and most boosted tools for simulating the changes in global climate system which are the results of increasing concentration of greenhouse gases (GHG), changes in land use and many other anthropogenic activities. The main deriving forces of GHG are the population, socio-economic and technological changes (IPCC-TGICA, 2007). These models represent the physical processes of climate which occur in the oceans, the atmosphere, the cryosphere (frozen ice) and land surface (Fowler et al., 2007).

(Source: (Viner, 2000)) Figure 2-1 Schematic illustration of Global Climate Model (GCM) These are three dimensional models with horizontal grid resolution ranging from 250 to 600 km. It consists of 10-20 vertical layers in the atmosphere and almost 30 layers (Figure 2-1) in the oceans. Almost twenty-three (23) Global Climate Models have been developed by 8

advanced countries such as England, France, Japan, Germany, China, Canada, and the USA, etc. (IPCC-DDC, 2011). According to Fowler et al. (2007), the GCMs are numerical based coupled models representing the global systems such as sea-ice, atmosphere, oceans and very advanced tools now a day for assessing the changes and variability in climate. However, they are very coarse in temporal and spatial resolution. For example, he demonstrated that about 0.125o (lat-long) is needed for hydrologic modeling in mountainous regions but GCMs such as HadCM3 model has a spatial resolution of 2.5× 3.75o. GCMs have been extensively used for the assessment of future stream flow changes during last few decades. It is proved that GCMs operate much better on global level than local scale or regional scale due to its coarse resolution. To solve this problem downscaling techniques have been developed. These techniques are very handy tools to derive the future scenarios on very fine scale (Gagnon et al., 2005). 2.2

Description of IPCC emission scenarios

There are four main storylines A1, A2, B1, B2 defined in the Special Report on Emission Scenarios (SRES) which describe the relations between the main driving forces (population, socio-economic and technological changes, land use changes and energy, etc.) of GHG and aerosol, and their development in the 21st century. These storylines are based on demography, economy, energy, agriculture, land use, and technology as shown in Figure 2-2. Each scenario shows different technological, economic, social, and demographical changes during the 21st century. For example according to scenario A1, the population would be maximum (8.7 billion) in the middle of 21st century and then it will be decreased after 2050 and would be 7.1 billion at the end of 21st century. Figure 2-2 show two sets of four storylines, first one shows increasing globalization and regionalization, and second set presents strong environmental and economic changes. This shows that A2 and B2 emphasize much on regionalization. Therefore these two scenarios are most commonly used in regional studies. The summary of these four storylines is given below as: A1: This scenario family describes the rapid economic growth of the future world, a peak in world population by the middle of 21st century and decline thereafter, rapid changes in technologies and the introduction of new and efficient technologies. This storyline is subdivided into three sections according to the use of energy: A1FI, A1T and A1B shows the intensive use of fossil fuel, no use of fossil fuel and balance across all energy sources, respectively. A2: This scenario family presents much aspects of regionalization with a heterogeneous world showing a continuous increase in world population and less economic growth than other storylines. The economic growth and technological changes are regionally oriented. B1: This scenario family has much emphasis on globalization to achieve socio-economic and environmental sustainability. It presents the same world population growth as A1 but more clean and efficient introduction of technologies and rapid economic changes.

9

B2: This scenario reflects much emphasis on local (regionalization) solutions with sustainable development in socio-economic and the environmental structure. It presents a continuous increasing population rate lower than A2 but less than A1 and B1with intermediate economic development (Arnell, 2004). These four storylines are subdivided into four scenario families, six scenario groups and are further divided into forty scenarios according to different driving forces and GHG concentrations (IPCC-TGICA, 2007). A complete description of these forty scenarios are described in IPCC-SRES (IPCC, 2000).

(Sources: (IPCC-TGICA, 2007) Figure 2-2 Emission scenarios and their driving forces 2.3

Downscaling

Downscaling, or translation across scale, is the recent term used to describe the techniques that relate the local/regional scale variables to large scale atmospheric forcing. Downscaling is especially developed to address the present needs in global environment change research studies, by getting the temporal and spatial information from GCMs (Hewitson and Crane, 1996). The basic concept of downscaling is to derive the local-scale variables (predictands) such as temperature, precipitation, and humidity (Fowler et al., 2007) from global-scale variables (predictors) such as mean sea level pressure (MSLP), and specific humidity (SH), etc. (Wetterhall et al., 2006). There are two main techniques for downloading the large scale GCMs output to small scale resolution. One is the Dynamical Downscaling (DD), which is a process based technique focused on nested model, where a high resolution climate model called Regional Climate Model (RCM) is used. This model takes coarse resolution data from GCMs and provides information on fine resolution. The second is the Statistical Downscaling (SD) which makes an empirical relationship between variables of large-scale and local-scale. The empirical 10

relationships offer a more immediate solution and significantly lower computing requirements, consequently offering an approach that can be rapidly adopted by a wider community of scientists. The performance of both methods depends upon the quality of climate data (local and large scale) which are used during downscaling process (Wilby and Wigley, 1997; Wilby et al., 2002).

(Source:(Wilby and Wigley, 1997) Figure 2-3 A schematic illustration of downscaling and aggregation 2.3.1

Dynamical downscaling

Dynamical downscaling is also called as nested downscaling or numerical downscaling. In this method, a high resolution model called Regional Climate Model (RCM) is used to provide information on fine resolution inputs from GCM. Regional Climate Model A Regional Climate Model (RCM) is a high resolution climate model that covers a limited area of the globe (e.g., 5,000 × 5,000 km) with a typical horizontal resolution of 50 km. RCMs are based on physical laws represented by mathematical equations that are solved using a three-dimensional grid. Hence, RCMs are comprehensive physical models, usually including the atmosphere and land surface components of the climate system, and containing representations of the important processes within the climate system (e.g., cloud, radiation, rainfall, and soil hydrology). Many of these physical processes take place on much smaller spatial scales than the model grid and cannot be modeled and resolved explicitly. Their effects are taken into account using parameterizations, by which the process is represented by the relationships between the areas or time averaged effect of such sub-grid scale processes and the large scale flow. Given that RCMs are limited area models, they need 11

to be driven at their boundaries by time-dependent large-scale fields (e.g., wind, temperature, water vapor and surface pressure). These fields are provided either by analyses of observations or by GCM integrations in a buffer area that is not considered when analyzing the results of the RCMs (Jones et al., 2004). GCM models respond to large scale forcings (large scale forcings are greenhouse gas concentration, ocean circulation, atmospheric circulation, etc.) while RCMs simulates on small scale. Since, RCM is mostly dependent on the GCMs boundary conditions. So, it is susceptible to any systematic errors which belong to GCM’s driving fields. There must be a strong co-ordination between the regional and global climate modeling groups to ensure that the appropriate data are available. The simulations of RCM are computationally expensive depending upon the domain size and the resolution at which it is to run. This limits the number of experiments for climate scenarios (Wilby and Wigley, 1997). This technique is a good alternative of statistical downscaling but there is no historical relationship between the global-scale and local or regional-scale. The main advantage of RCMs is that these are skillfully applicable to different areas around the world (Benestad et al., 2008). Armell et al. (2003) compared the regional climate model (HadRM3H) having a spatial resolution of 0.44° X 0.44° with two global climate model; HadCM3 (Coupled OceanAtmosphere Climate Model) and HadAM3H (Atmospheric Climate Model) with a spatial resolution of 2.5° × 3.75° and 1.875° × 1.25°. The main focus of this study was to downscale precipitation and temperature data for macro-scale hydrological model (0.5° x 0.5°) to simulate runoff scenarios in Southern Africa by using the regional climate model (HadRM3H) input data from HadAM3H. There was much difference in runoff pattern obtained by the HadCM3 and HadRM3H. Most of the features which were obtained by HadRM3H were also obtained by HadAM3H. So it is suggested that the intermediate resolution model can also be used for impact assessment studies (Arnell et al., 2003). Akhtar et al. (2008) used the PRECIS (RCM) and delta change method to downscale the temperature and precipitation for hydrological modeling under A2 scenarios for the Hindukush-Karakorum-Himalalya (HKH) region located in the Indus Basin Pakistan. PRECIS is a regional climate modeling system developed at the Hadley Centre that can run on a PC (Personal Computer) and can be applied easily to any area of the globe, to generate detailed climate change predictions. In this study, the resolution for the PRECIS was 25 × 25 km. Hydrological model HBV was run by taking inputs from delta change applied on PRECIS outputs and directly from PRECIS. The results showed that the simulation from PRECIS data directly gave better results than PRECIS data followed by delta change method. This indicates that the direct use of RCM outputs is an alternative for the regions where the observed data is of poor quality. As suggested in the guidelines presented by (Mearns et al., 2003), the main strengths and weaknesses of DD are: Advantages • • • •

These techniques may be able to provide more realistic scenarios of climate change at the regional scale than the direct application of GCM-derived scenarios. Provides very highly resolved information (spatial and temporal). Information is derived from physically-based models. Many variables available. 12

•

Better representation of some weather extremes than in GCMs.

Disadvantages • • • • 2.3.2

Computationally expensive, and thus few multiple scenarios available. Lack of two-way nesting may raise concerns regarding completeness. Dependent on (usually biased) inputs from driving GCM. Few time windows are available. Statistical downscaling

Statistical Downscaling (SD) methods are very similar to the Perfect Prog and Model Output Statistics (MOS) whose basic function is to generate weather prediction on shot-range. The basic principle behind these approaches is to produce the best correlation between synoptic variable such as MSLP and regional/local scale variables such as temperature or precipitation (Wilby et al., 2000). Statistical methods are simple methods to downscale the course resolution of Global Climate Model (GCM). These are based on producing conditions between large scale climate state and regional/local physiographic features (topography, land-sea distribution and land use, etc). According to this concept, the regional or local information can be derived by first developing a statistical model which links large-scale climate variables (predictors), such as temperature, pressure and precipitation to the small-scale variables (predictands), such as local temperature and precipitation. Predictors from GCMs are then fed into the statistical model to get the corresponding information about local climate characteristics. SD methods, initially, had their uses in synoptic climatology and numerical weather prediction but now they are currently used for a wide range of climate applications, from historical reconstruction to regional climate problem. The most important thing is that these methods are computationally very inexpensive (Hewitson and Crane, 1996) as compared to regional climate model, and thus can be easily applied to outputs from GCMs. Due to small computational demands, this method is helpful to generate different ensembles for sensitivity analysis of climate realization. Another main advantage of Statistical Downscaling is to provide the local scale information which is very helpful in climate change impact studies (Giorgi et al., 2001). The main assumption of SD is that the empirical relation between large and local scale is temporally stationary but the DD covers this assumption of stationary. SD is based on two major assumptions. • •

High quality large-scale and local-scale observed data being available for long period of time to establish a strong relationship in the current climate. The relationships derived from observed data for current climate are applicable to future climate.

The last assumption hold good for temperature. However for precipitation, the circulation is the dominant factor to establish the relationship in recent climate but humidity is important for future climate. Various statistical methods are developed and tested by “STARDEX” which are grouped into the following categories? • •

Multiple linear regression Canonical correlation analysis 13

• • • • • •

Artificial neural networks Multivariate autoregressive models Conditional re-sampling and other analogue-based methods Methods based on a ‘potential precipitation circulation index and critical circulation patterns A conditional weather generator Local scaling and dynamical scaling

These methods range from standard linear regression methods, through methods focusing on spatial patterns (such as canonical correlation analysis), to non-linear neural network methods and other novel methods, including analogue-based methods (STARDEX, 2005). Gebremeskel et al. (2005) used Statistical Downscaling Model (SDSM) for downscaling of precipitation and temperature in the assessment of climate change impacts on the flow of the Alzette basin. This basin lies in Grand-Duchy of Luxembourg which is bordered by France, Germany and Belgium. He used the NCEP global predictor data with the period of 19601990 and one point local-scale observed data for the calibration of SDSM. The seven predictors such as geopotential height at 500 and 850 hpa, relative humidity at 500 and 800 hpa, component of wind speed (uwind), maximum temperature (tmax), and specific humidity (shum) from NCEP and HdCM3A3 were used. For the calibration of temperature and precipitation, the four predictors (p500, r850, shum and tmax) for temperature and the other four predictors (uwind, shum, p850, r850) for precipitation were selected from NCEP predictors. Before using the HadCM3A2 predictors for the current and future scenarios, these predictors are normalized with respect to base line period (1960-1990). Then, the predictors from HadCM3 were fed into the SDSM and the results were verified by comparing with observed local data of the period 1997-2000. After calibration and verification, the precipitation was downscaled for the period of 2036 – 2065 and 2070 – 2099. This downscaled precipitation was then fed into the hydrologic model (WetSpa) as input to simulate stream flow scenarios for the future period of 2036 – 2065 and 2070 – 2099 (Gebremeskel et al., 2005). Gagnon et al. (2005) uses the statistical downscaling method for the analysis of stream flow in three different basins of Quebec (Canada). SDSM is used to downscale the temperature and precipitation using the predictors of CGCM1 (first generation of Coupled Canadian Climate Model) for a hydrologic model (SSARR). The calibration of SDSM model is made by using the NCEP predictors and the validation by using the CGCM1 predictors. The results of calibration show that SDSM explains the variation of 76-89% for temperature and 2848% for precipitation. It is also shown that the results are better for current climate scenarios when using the NCEP predictors in the SDSM but not better while using the CGCM1 predictors (validation process). The results showed the lagged peak flow and greater magnitude for the two northern basins. It is recommended to use other GCM model due to some discrepancies in CGCM1 during the transition of spring and fall season. Hay et al. (2000) has compared two methods, Delta change and Statistical Downscaling, to generate temperature and precipitation for three mountainous regions in the United States of America. These are (a) Colorado, Animas river basin (b) California, Carson River basin and (c) Cle Elum basin in Washington. Statistical Downscaling (SD) model is trained by using the NCEP/NCAR reanalysis gridded (2.5o × 2.5o) predictors and local-scale observed data for each basin. The NCEP predictors are first interpolated to the HadCM2 grids (2.5o×3.75o) before using in calibration. Then SD model is forced to simulate the future scenarios using 14

the HadCM2 predictors. The observed local-scale data and HadCM2 outputs (temperature and precipitation) are used in the delta change approach. In this approach, the differences in current and future outputs of HadCM2 are adjusted to the local-scale variables to get the local-scale future variables. Then, these temperature and precipitation scenarios are used in rainfall-runoff model to simulate the runoff scenarios. The results show that simulations from precipitation-runoff model for current conditions are more realistic when using the current downscaled scenarios simulated by the SD model using the NCEP outputs. But rainfall-runoff model gives questionable scenarios for current and future when using the downscaled variables (precipitation and temperature) from HadCM2’s predictors because the GCM is unable to estimate accurately the surface variable required for stream flow in these areas. He says there are uncertainties in GCMs to simulate the current scenarios using either delta change or statistical downscaling. So, the future assessment of climate should be treated very carefully. Wetterhall et al. (2006) evaluated the four SD methods (Principal Component Analysis, Teweles-Wobus Score, SDSM, and MOFRBC) in three Chinese regions using the STARDEX indices. They are located in the Southern (Jouzlou), Eastern (Baixi) and Central (Laoyukou) part of China. These areas lie in subtropical zone with heavy, intermediate and low monsoon precipitation in South, East and central regions. The main predictors used in this study are MSLP, GPH, and SH obtained from the NCEP/NCAR reanalysis data. The regional variable (predictand) daily precipitation was collected from 13 stations. The SMSD and MOFRBC were applied by using two modes of predictor variables. In the first mode, only MSLP and GPH were used and in second MSLP, GPH and specific humidity. The results demonstrate that SDSM (with MSLP and GPH) and SDSMh (with MSLP, GPH and SH) performed best in the southern basin during summer season which is a sub-tropic monsoon region, and MOFRBCh performed better during winter. It was also concluded that SDSMh gave better results in the central region of China during summer and SDSM during winter. Wilby et al. (2000) made a comparison of two main techniques by taking the outputs from SD, RCM (RegCM2), RCM (ReCM2) corrected, NCEP and NCEP corrected, and used as inputs to distributed hydrologic model (PRMS) in the Animas River basin (USA). The Animas basin is a mountainous region ranging between the elevations of 2000 to 4000 meters with an area of 1820 km2. First the NCEP data was used directly as input to hydrologic model, and then the NCEP outputs are corrected for systematic bias and used as input to hydrologic model. After this, SD outputs are used with twenty ensembles as input to hydrologic model. At last, RCM (RegCM2) and elevation corrected RCM outputs were used as input to the hydrologic model to analyze the runoff regime. The results showed that the SD and RCM explained variations better than other methods. It was concluded that SD performed better than all other as shown in Figure 2-4. It was also shown that SD has advantage over RCM to generate ensembles of small-scale variable by using very limited large-scale predictors. The scenarios from RCM are demanded computationally much more to produce than SD. Although the RCM is more sophisticated and physical based model but its results before correction were not as skillful as SD method as described in Table 2-1.

15

(Wilby et al., 2000) Figure 2-4 Comparison of statistical and dynamical downscalings Table 2-1 Comparison of statistical and dynamical downscaling techniques Data Station SDS* NCEP NCEPadj RegCM2 RegCM2adj

Precipitation E (%) D(%) 18 -6 14 -47 15 -36 26 -6 26 -5

Tmin E (%) D (ºC) 88 +0.1 75 -0.4 74 -0.7 66 +0.9 67 -0.1

Tmax E (%) D (ºC) 90 -0.5 81 -2.0 79 -1.0 72 -4.6 72 -0.9

Flow (Q) E (%) D (%) 84 -4 78 -22 75 -65 72 -54 48 +5 69 -11

E Explained variance; D mean bias (Wilby et al., 2000) SD techniques are classified into three main types; weather typing, stochastic weather generator, and transfer function (Wilby et al., 2002) which are discussed below. Weather typing In this case, weather types, or atmospheric circulation patterns are used rather than largescale predictor variables in order to find statistical relationships between the types or patterns of weather and observed station data (Figure 2-5). The first step in this process is the identification of the weather types, or atmospheric circulation patterns usually from atmospheric pressure information. These types or patterns may be based on existing subjectively-derived weather classes, e.g., the Lamb weather types in the UK or the European Grosswetterlagen, or they may be objectively derived using techniques such as principal components analysis or artificial neural networks. Once the weather types have been identified, then statistical models between the types and local station data are calibrated and then verified. If it is possible to develop models which perform satisfactorily, then they can be used in climate change studies (Wilby et al., 2002). In Figure 2-5, the blue arrows

16

shows steps based on observed climate data and the red arrows show the application of GCM data to construct the future scenarios for local scale.

(Source: http://www.cics.uvic.ca/scenarios) Figure 2-5 Schematic illustration of weather typing approach Stochastic weather generators Stochastic downscaling approaches typically involve modifying the parameters of conventional weather generators such as WGEN, LARS–WG or EARWIG. Weather generators are used for in-filling of missing data and reproducing indefinitely long synthetic series of weather data from small record of site data by simulating the key properties of observed meteorological records (such as mean, variance and covariance, frequencies and extremes). These are frequently used in water engineering, agricultural, ecosystem, hydrological impact studies because observed ground-based meteorological data are often inadequate in case of length, completeness or spatial coverage. These statistically based models are also known as models of observed daily weather consequences (Wilks and Wilby, 1999). Climate change scenarios are generated stochastically using revised parameter sets scaled in line with the outputs from a host GCM. The main advantage of this technique is that it can exactly reproduce many observed climate statistics and has been widely used, particularly for agricultural impact assessment. Furthermore, stochastic weather generators enable the efficient production of large ensembles of scenarios for risk analysis. The key disadvantages relate to the low skill at reproducing inter-annual to decadal climate variability, and to the unanticipated effects that changes to precipitation occurrence may have on secondary variables such as temperature (Wilby et al., 2002). Transfer functions (Regression methods) 17

In this method, the small scale variables (e.g., temperature and precipitation) are derived by using the large-scale variables (e.g., MLSP, 500 hPa, and SH). As show in the figure (blue arrows), the first step in this method is to define the predictors such as mean sea level pressure, 500hPa, 850hPa geopotential heights, zonal air flow, temperature, and precipitation. They should explain most of the variations of predictands at the local scale. The second step is the construction of transfer function by relating the large-scale predictors and small-scale predictands with the help of some mathematical techniques. In the third step, the transfer function is validated by using some observed data. After the validation, these transfer function can be used to predict the predictands for future climate of small-scale area by extracting the future predictors from GCMs outputs as shown in Figure 2-6(red arrows). The blue arrows show the steps based on observed data of predictors and predictands. The red arrows show the future scenarios for local-scale derived by GCMs predictors. These methods are based on statistical/empirical relationship between the large scale variables of GCMs and small scale variables. The highly used transfer methods are Linear Regression (LR), Non-Linear Regression (NLR), Multiple Linear Regressions (MLR), Canonical Correlation (CC), Principle Component Analysis (PCA), and Artificial Neural Networks (ANN) which are used to construct the relation between predictor and predictand. The big advantage of this method is its easy application to downscaling, and the weak point of this technique is that it fractionally explains the variability of climate variable like precipitation. Transfer method along with weather typing method is assumed that they are valid for future scenarios. “These downscaling techniques are very sensitive to predictor’s choice and the choice of statistical method used for downscaling.” (Wilby et al., 2002).

(Source: http://www.cics.uvic.ca/scenarios) Figure 2-6 Schematic illustration of transfer function approach The summary of these three main types, explained above, is tabulated according to their advantages and disadvantages in Table 2-2. 18

Table 2-2 Advantages and disadvantages of different types of statistical downscaling Methods

Strengths

Weather Typing (Analogue method, hybrid method, fuzzy classification, Monte Carlo Methods and Self organizing maps)

Weather Generator (e.g., Markove chain, stochastic models, spell length method, storm arrival timing, mixture modeling)

Regression Method (Linear and nonlinear, canonical correlation, principal component analysis, neural networks, kriging)

2.3.3

Weaknesses

1) Yields physically interpretable linkages to surface climate 2) Versatile ( means that it can be applied to surface climate, air quality, flooding , erosion etc) 3) Composting for analysis of extreme climate 1) Production of large ensembles for uncertainty analysis or long simulations for extreme 2) Spatial interpolation of model parameters using landscape 4) Can generate sub-daily information 1) Relatively straightforward to apply 2) Employs full range of available predictors 3) Off-the-shelf solution and software available

1) Needs additional task of weather classification 2) Circulation based scheme can be insensitive to future forcing 3) May not capture intratype variations in suture climate 1) Arbitrary adjustment of parameters for future climate 2) Unanticipated effects to secondary variables of changing precipitation parameters

1) Poor representation of observed variance 2) May assume linearity and/ or normality of data 3) Poor representation of extreme events

Delta change approach

Delta change (Change factor) is a very popular and simple procedure to assess the impacts of climate change for local scale (Diaz-Nieto and Wilby, 2005). Delta change approach is the most common method used to construct the future scenarios for climate variables (temperature, precipitation) by using the outputs of GCMs. This approach is often applied in case of the lack of RCM outputs or when multiple climate scenarios are needed. In this method differences between the future and current scenarios of climate variables is computed and then these changes are added to the observed local-scale climate variables (in case of temperature) and multiplied with observed climate variables (in case of precipitation) to find out the future scenarios for the local-scale (Salzmann et al., 2007). The U.S. National Assessment suggests that delta change is the primary method to make future scenarios. The assumption of this method is that the GCMs are more reliable to simulate the relative changes as compared to absolute values (Hay et al., 2000). The following equations are used for future scenarios of temperature and precipitation from GCMs.

19

Tfuture, daily Tobs, daily Tfuture, monthlyGCM -Tpresent,monthlyGCM 2-1

Pfuture,daily Pobs, daily ×

Pfuture, monthlyGCM Ppresent, monthlyGCM

2-2

In these equations, (#$%&'(',)*+,- and .$%&%(',)*+,- ) and (#/01,)*+,- and ./01,)*+,- ) show temperature and precipitation of future (simulated) and observed for local scale. .$%&%(',3/4&5,-678 &#$%&%(',3/4&5,-678 and #9('1'4&,3/4&5,-678 & .9('1'4&,3/4&5,-678 are the mean monthly temperature and precipitation simulated by RCM (Regional Climate Model) or GCMs for future and present (current) scenarios. The main assumption of this method is to assume that the present climate remains unchanged in the future. The spatial variability in occurrence of rainfall is remained unchanged. One main limitation of this method is that it is applicable only when equivalent observed and GCM data are available (this study observed data from1961-1990 and GCM 1961-1990 and 2040-2069). Some weaknesses and strengths of this method are given below (Diaz-Nieto and Wilby, 2005). Merits • Very simple, quick and computationally straightforward to apply for station’s future scenarios. • The small scale or local scale scenarios of climate change are directly related to GCM or RCM outputs (therefore, selection of GCM is important for this method). Demerits • Realism of this method depends upon the changing factor computed from the GCM scenarios. • For future scenarios, temporal structure remains unchanged. • It is restricted to time-slice based scenarios. 2.4

Climate change impact on water resources

There is clear evidence that global mean temperature has been increased by 0.6 °C during the past century. Scientists agree that the main cause of global warming is increasing concentration of greenhouse gases which are the resultants of human activities (e.g. burning of fossil fuel increases the carbon dioxide). It is suggested that global warming will continue in the next decades causing changes in climate system. These climate changes would lead to intensification in global water cycle changing the frequency, timing and intensity of precipitation and runoff. These climate changes would also affect the regional water resources availability which will cause increasing water demand among all sectors such as agriculture, domestic, industrial, etc. (Miller and Yates, 2006). Yu et al. (2002) explored the climate change impacts on water resources in South Taiwan. The Shin bridge station was selected in the Kao-Pen creek basin. To see the historical trends of temperature and precipitation at the Kao-Hsiung climate station, the Mann-Kandall method is selected. The results show the increasing trend in temperature and several changes in the transition probabilities of daily precipitation occurrence. After this, the weather series 20

for temperature and precipitation are generated by using the weather generator for the future. These weather series are used as input to rainfall-runoff model (HBV) to explore the change in water resources in the future. The results indicate the increasing trend of runoff in the wet season and decrease in the dry season (Yu et al., 2002). McBean and Motiee (2008) explored the climate change impacts on water resources of the Great Lakes (Lake Superior, Huron, Michigan, Erie and Ontario) of North America. In this study, seventy years of meteorological and hydrological data from 1930 to 2000 was analyzed by using some statistical analysis. A historical trend is found in temperature, precipitation and stream flow by using Mann-Kandall statistics and regression analysis in these lakes. He found that precipitation in four lakes out of five, and stream flow of the three river stations (Clair, Niagara and St. Lawrence) presents significant increase. Historical precipitation is also compared with predicted data of GCM which shows higher trend of historical precipitation than the GCM. It is a clear evidence that global warming and climate change has a great influence on the hydrologic components particularly on precipitation in the Great Lakes (McBean and Motiee, 2008). Garcia et al. (2008) applied the HEC-HMS (Hydrologic Modeling System) to estimate the water resources in twelve scarce gauges basins of the northern part of Spain. There are only 13 stream gauges in the study area. HMS is selected due to its versatility. Soil Moisture Accounting (SMA) model is selected for calculating the magnitude (volume) of water. The surface runoff is estimate by Clark’s model, base flow by two independent linear storage tanks model and flow routing by using the Lag model in short reaches and Muskingum model in long reaches. There are good relations between the observed and predicted flow data by models during the calibration and validation even though the rainfall gauges are scarce in the study area(García et al., 2008). Fujihara et al. (2008) conducted a study in the Seyhan river basin, Turkey, to explore the impacts of climate change on water resources. A newly developed method of dynamic downscaling which is known as Pseudo Global Warming Method (PGWM) is applied to connect the GCMs and Hydrologic models. Two GCMs (MRI-CGCM2 and CCSR/NIES/FRCGC-MIROC) outputs under A2 scenarios are used for downscaling purpose for two decades, 1990s and 2070s. The dynamically downscaled precipitation and temperature have a good correlation with observed data. This downscaled data followed by the bias correction method is used as input to hydrologic model (SiBUC). SiBUC is a distributed model which can be used to simulate snow melt, snow accumulation, evaporation, soil moisture, base flow, and surface runoff on each grid cell. Results obtained by both GCMs are different but both models’ projections indicate an obvious rise in temperature and decrease in precipitation in the future. It is projected that the mean annual temperature would be increased by 2.7 °C (in case of CCSR) and 2 °C (in case of MRI), and annual precipitation would be increased by 157 mm (25%) and 182 mm (29%) in case of MRI and CCSR, respectively. The annual evapotranspiration will be decreased by 36 mm (9% in case of MRI) and 39 mm (10% in case of CCSR), and the annual runoff by 118 mm (52% in MRI) and 139 mm (61% in CCSR) in the future. A simple scenario approach is used to analyze the water resource system finding the changes in water use. This analysis presents that there will be no scarcity of water in the future if demand remains constant. So, it is concluded that water management can play a significant role than climate change in maintaining water resources in the future for this basin (Fujihara et al., 2008).

21

Tolika et al. (2008) investigated the future changes in temperature and precipitation over Greece. The two statistical tools (MLRct and ANN) were applied here for downscaling of future scenarios both for temperature and precipitation scenarios. The two regionalized emission scenarios A2 and B2 of MadAM3P global climate model were selected. The two precipitation indices (90th percentile of precipitation (pq90) and maximum dry spell (pxcdd), two temperature indices (90th percentile of maximum temperature (txq90), and 10th percentile of minimum temperature (tnq10) were simulated by downscaling models. The MLRct model explores more efficient results than ANN for validation period. Both models underestimate to get variability in observed time series. ANN is not so strong to give the climate change signal and sometimes gives opposite results to MLPct. According to MLPct, there would be an increasing trend in both temperature indices in 2070-2100. In case of extreme precipitation indices, the results are spatially incoherent and too complex to give the exact trend (Tolika et al., 2008). In this study, the impacts of climate change on snowpack in Spanish Pyrenees basin are assessed by using the input data from regional climate model (HIRHAM) into the snow model (GRENBLS). The regional climate model is run at 50 km2 to produce the input data for snow model under A2 and B2 emission scenarios at different altitudes (1,500, 2,000, 2,500 and 3,000) for the periods of 1960-1990 and 2070-2100. The results give a complete picture of impacts of climate change on the thickness of snowpack covering the special and altitudinal variability. The results clearly indicate that the thickness of snowpack would decrease significantly during the 21st century. Maximum impacts (78% reduction) of climate change are seen at 1500 masl (above sea level). It is also noted that as the altitude increases the impacts decreases (López-Moreno et al., 2009). Pal and Al-Tabbaa (2009) conducted a study to find out the extreme events; drought before monsoon and floods after monsoon in Kerala, India. A rainfall data from 1954-2003 was used to examine the extreme precipitation by using the non-parametric Mann-Kendall method. The three seasons namely winter, spring and autumn are selected to examine a set of rainfall indices. Large intra-regional differences are found in trends, and in different seasons. In winter and autumn, there is an increasing tendency which causes more floods but in spring there is a decreasing trend in rainfall extreme causing increasing frequency of dry days. This can make the Kerala region more vulnerable to water scarcity in pre-monsoon time and also delaying the monsoon. The results of this study gives good climate change indicators which will help in risk management and seasonal forecasting (Pal and Al-Tabbaa, 2009). Xu et al. (2009) have conducted a study to analyze the extreme climate events for future period derived from an ensemble of GCMs under a range of emission scenarios. These are used in the Fourth Assessment Report of IPCC (IPCC AR4). The seven climate indices (CDD, R10, R5D, HWDI, TN90, SDII, and R95T) are selected for both temperature and precipitation. The study area selected for this research is the Yangtze River basin. Results show that heat wave would be of longer duration in the 21stcentury than today. Warm nights will be more frequent in the region. All the indicators are uniform except CDD and R10. A single day’s heavy rainfall events would increase in intensity in the basin (Xu et al., 2009). Göncü and Albek (2010) conducted a study which deals with the effects of the expected climate change on the hydrology of watersheds and on water resources. HSPF (Hydrological Simulation Program-Fortran) has been used to simulate stream-flow and reservoir volume as realizations of watershed response. Climate change scenarios have been prepared based 22

on trends expected in Western Turkey in the first half of the 21st century, and a hypothetical watershed with different land uses have been prepared. Changes in stream-flow due to land use, soil type, and climate change have been examined using flood frequency and low flow analysis. The simulations revealed quantitatively the difference among the responses of watersheds with no vegetative cover and with forests or pasture to trends in temperature and precipitation. It has also been found that monthly variations are very important in predicting the future response of watersheds. Significant differences have been observed in stream flows and reservoir volumes on a monthly basis between scenarios, soil types and land uses. Though the effects of temperature and precipitation act to counterbalance their effects on a long term scale, on a monthly basis they can act to reinforce their effects and create drought periods and floods(Göncü and Albek, 2010). Zhao et al. (2010) conducted a study by using the hydro-climatic data from five sub-basins in the Poyang Lake basin in the Southeast China over the past 50 years to investigate the annual and seasonal trends of stream flow, and the correlations between stream flow and climatic variables. The Theil-Sen approach and the non-parametric Mann-Kendall test were applied to identify the trends in the annual and seasonal stream flow, precipitation, and evapotranspiration time series. It was found that annual and seasonal stream flow on all the stations have increasing trends except Lijiadu station in the wet season. The trends in annual and seasonal precipitation during the whole period are generally not as significant as in evapotranspiration. The correlations between stream flow and climate variables (precipitation and evapotranspiration) are detected by the Pearson’s test. The results show that stream flow in the Poyang Lake basin are more sensitive to changes in precipitation than potential evapotranspiration (Zhao et al., 2010). A couple of studies, one by Akhtar et al.(2008) and another by Ashiq et al. (2010) using downscaling were conducted in Pakistan. Akhtar et al. (2008) implemented Providing Regional Climate for Impact Studies (PRECIS) RCM, and a simple downscaling method called “Delta Change” for downscaling mean precipitation and temperature for hydrological modeling under IPCC emission scenario A2 for the period of 2071–2100 in the region of the Hindukush – Karakorum – Himalaya ranges located in the Indus River basin, Pakistan. HBV hydrologic model was used to simulate the future stream flow. The results showed high risk of flooding under climate change. Ashiq et al. (2010) evaluated the monthly precipitation outputs of PRECIS, run by Akhtar et al. (2008), and interpolated them from coarse resolution (50×50 km) to a fine scale (250×250 m). Seven different interpolation methods were used to interpolate monthly precipitation and were validated with the observed precipitation. This study was conducted in the Northwestern Himalayan Mountains and the upper Indus plains of Pakistan, which also cover a small part of the Jhelum basin located in Pakistan. Through their study they concluded that the systematic errors associated with RCM cannot be reduced by interpolation methods(Ashiq et al., 2010). 2.5

Climate change impact on the hydropower

Hydropower plays an important role in the sustainable contribution to meet the electricity demand worldwide. Hydropower plants have contributed about 19% (2500 TWh) of the total electricity demands by the middle of 1990. The importance of hydropower, along with other renewable energy sources is expected to increase in the future. Since 1980, hydroelectricity has been increasing steadily at an average of 2.3% every year. It is expected that the average 23

growth rate of world hydropower production will be increased by 2.4 to 3.6% per year in 1990 through 2020. The developing countries and industrialized countries are expected to contribute highest in this growth rates of electricity. Hydroelectric generation is directly proportional to runoff, where the greater the volume of water discharged, the greater the generation potential ((Lehner et al., 2005). Pakistan has an electricity generation capacity of about 19,547 MW. Installed hydropower capacity is 6,599 MW out of which 1,000 MW comes from the Mangla Power station. The electricity demand will be increased by an annual compound growth rate of 7.9% in the future. The main energy sources of Pakistan and their production is described in Table 2-3 (Mirza and Ahmad, 2005) Table 2-3 Installed electricity generation capacity (MW) in Pakistan Sector Public sector Private Total

Thermal 4,629 7,707 12,336

Hydro 6,599

Nuclear 462

6,599

462

Coal 150 150

Total 11,840 7,707 19,547

Source: (Mirza and Ahmad, 2005) Schaefli et al. (2007) addressed the climate change impacts on hydropower production in the Swiss Alps power plant. The potential climate change impacts are analyzed in terms of system performance for the control period (1961-1990) and for the future period (20702099) under the range of climate change. The system performance is simulated through a set of four model types, including the production of regional climate change scenarios based on global mean warming scenarios, the corresponding discharge model, the model of glacier surface evaluation and the hydropower management model. The modeling uncertainties are examined for each model. The overall modeling uncertainties are simulated though Mont Carlo simulations of the system behavior. The results show that climate change has significant negative impacts on the system performance (Schaefli et al., 2007). Vicuna et al. (2008) indicated that climate change is likely to affect the generation of energy from California’s high-elevation hydropower systems. To investigate these impacts, this study formulates a linear programming model of an 11-reservoir hydroelectric system operated by the Sacramento Municipal Utility District in the Upper American River basin. Four sets of hydrologic scenarios are developed using the Variable Infiltration Capacity model combined with climatic output from two general circulation models under two greenhouse-gas emissions scenarios. Power generation and revenues fall under two of the four climate change scenarios, as a consequence of drier hydrologic conditions. Energy generation is primarily limited by annual volume of stream flow, and is affected more than revenues, reflecting the ability of the system to store water when energy prices are low for use when prices are high (July through September). Power generation and revenues increase for two of the scenarios, which predict wetter hydrologic conditions. In this case, power generation increases more than revenues indicating that the system is using most of its available capacity under current hydrologic conditions. Hydroelectric systems located in basins with hydrograph centroids occurring close to summer months (July through September) are likely to be affected by the changes in hydrologic timing associated with climate change (e.g., earlier snowmelts and stream flows) if the systems lack sufficient storage capacity. High Sierra hydroelectric systems with sufficiently large storage capacity should not be affected by climate-induced changes in hydrologic timing(Vicuna et al., 2008).

24

Lehner et al. (2005) presents a model-based approach for analyzing the possible effects of global change on Europe’s hydropower potential at a country scale. By comparing current conditions of climate and water use with future scenarios, an overview is provided of today’s potential for hydroelectricity generation and its mid and long term prospects. The application of the global water model WaterGAP for discharge calculations allows for an integrated assessment, taking both climate and socioeconomic changes into account. This study comprises two key parts: First, the ‘gross’ hydropower potential is analyzed, in order to outline the general distribution and trends in hydropower capabilities across Europe. Then, the assessment focuses on the ‘developed’ hydropower potential of existing hydropower plants in order to allow for a more realistic picture of present and future electricity production. For the second part, a new data set has been developed which geo-references 5991 European hydropower stations and distinguishes them into run-of-river and reservoir stations. The results of this study present strong indications that following moderate climate and global change scenario assumptions, severe future alterations in discharge regimes have to be expected, leading to unstable regional trends in hydropower potentials with reductions of 25% and more for Southern and Southeastern European countries. Robinson (1997) developed a model based on the systems of Duke Power and Virginia Power in the Southeastern USA to simulate reservoir performance. The annual maximum draw-down of the reservoir, which represents the minimum dam size needed to maintain continuous energy generation is considered here. The model was tested for four regions in Eastern USA using 1951-1995 observations. The amount of draw-down depended on the linked daily sequences of precipitation and temperature, the former dictating the water available, the latter influencing both evaporation and energy demand. The time and level of the annual extreme emphasized that small changes in the timing of a dry spell had a major impact on the draw-down. Climatic changes were simulated by uniformly increasing temperatures by 2 °C and decreasing precipitation by 10 per cent. The resultant draw-down increased from current simulated values by about 10 per cent to 15 per cent with extremes up to 50 per cent. This was of the same order, but in the opposite direction, as the change created by a 10 per cent increase in the efficiency of energy generation. Without such an efficiency increase, many utilities will face the prospect of reduced or less reliable hydroelectric generation by climate changes in the manner examined here (Robinson, 1997). Harrison and Whittington (2002) assessed the relationship between changes in climate and the viability, technical and financial of hydro development. The planned Batoka Gorge scheme on the Zambezi River is used as a case study to validate the model and to predict the impact of climate change on river flows, electricity production and scheme financial performance. A hydrological model for stream flow and a reservoir simulation model (HEC5) for hydropower production was developed. The model was found to perform well, given the inherent difficulties in the task, although there is concern regarding the ability of the hydrological model to reproduce the historic flow conditions of the upper Zambezi basin. Simulations with climate change scenarios illustrate the sensitivity of the Batoka Gorge scheme to changes in climate. They suggest significant reductions in river flows, declining power production, reductions in electricity sales revenue and consequently an adverse impact on a range of investment measures (Harrison and Whittington, 2002). The Pacific Northwest (PNW) hydropower resource, central to the region’s electricity supply, is vulnerable to the impacts of climate change. The Northwest Power and Conservation Council (NWPCC), an interstate compact agency, has conducted long term planning for the PNW electricity supply for its 2005 Power Plan. In formulating its power 25

portfolio recommendation, the NWPCC explored uncertainty in variables that affect the availability and cost of electricity over the next 20 years. The NWPCC conducted an initial assessment of potential impacts of climate change on the hydropower system, but these results are not incorporated in the risk model upon which the 2005 Plan recommendations are based. To assist in bringing climate information into the planning process, they present an assessment of uncertainty in future PNW hydropower generation potential based on a comprehensive set of climate models and greenhouse gas emissions pathways. They found that the prognosis for PNW hydropower supply under climate change is worse than anticipated by the NWPCC’s assessment. Differences between the predictions of individual climate models are found to contribute more to overall uncertainty than to divergent emissions pathways. Uncertainty in predictions of precipitation change appears to be more important with respect to impact on PNW hydropower than uncertainty in predictions of temperature change. We also they also found that a simple regression model captures nearly all the response of a sequence of complex numerical models to large scale changes in climate. This result offers the possibility of streamlining both top-down impact assessment and bottom-up adaptation planning for PNW water and energy resources (Markoff and Cullen, 2008). 2.6

Conclusions

From the literature review, it is concluded that the temperature of the globe has increased and will increase in the future which is caused the changes in climate. And this climate change has obvious impacts on the hydrology of watershed. To date, the most advanced tool to find the impacts of climate change on water resource is GCM but it cannot be used directly on basin level. To use the data from GCM, many statistical and dynamical methods have been developed. From the comparative studies on the downscaling methods, it is concluded that statistical downscaling methods are easy to use, computationally inexpensive, and relatively efficient. The dynamic downscaling is a good alternative of statistical downscaling when the observed data is not available in a watershed.

26

3

STUDY AREA AND DATA

3.1 Indus River basin 3.1.1 Geography and climate of the basin The Indus River basin is one of the largest basins in the world that covers an area of 1.1 million km2. The areas of 6, 8, 39, and 47% of the Indus basin are located in Afghanistan, China, India, and Pakistan respectively as show in Figure 3-1 and Table 3-1. The catchment of the Indus River stretches between Himalaya in the North, and dry alluvial plain of the Sindh province of Pakistan in the South. The Indus basin covers 65% area of Pakistan, including the whole of Punjab and Khyber Pakhtun Khwa (KPK) provinces as well as most territories of the Sindh province and some eastern parts of Balochistan province. Only 14% area of India is covered by the Indus basin, comprising of the states of Jammu and Kashmir, Himachal Pradesh, Punjab, Rajasthan, Haryana and Chandigarh. On the other hand, only 1 and 11% of China and Afghanistan, respectively, are located in the Indus basin as shown in in Figure 3-1 and Table 3-1. At least 300 million people are roughly estimated to live in the Indus basin. The Indus basin drainage area is ranked as the 12th largest in the world. Indus delta is ranked as 7th largest in the world, with an area of 3 × 104 km2. The sediment discharge of the river is on number 6th in the world, with 2 × 105 Kg per annual. The annual average discharge of the river is of about 2 × 1011 m3 ranking as 10th in the world. The climate of the Indus basin is very diverse in nature. The southern parts of the basin (plains of Sindh and Punjab) are under the subtropical and semi-arid temperate sub-humid zone. However, the northern parts of the basin consist of cold mountainous highlands. The mean annual precipitation ranges between 100-500 mm on lowlands and 2000 mm on mountainous areas. Snowfall at high altitude also contributes a lot in the river flow. Since the climate of the basin is arid to semi-arid, December to February in the lowland areas are considered as the cold months, with 14 to 20 °C as mean monthly temperatures. Between March and June, the mean monthly temperature varies from 42 to 44 °C. In the upper plains, it varies from 2 to 23 °C in winter and 23 to 49 °C in the summer. Table 3-1 Country areas in the Indus River basin Basin

Indus

3.1.2

Area (km2 × 1000)

1,120

Countries included

Pakistan India China Afghanistan

Area of country in basin (1000 km2) 520 440 88 72

As % of total area of the basin

As % of total area of the country

47 39 8 6

65 14 1 11

Pakistan in the Indus River basin

As discussed above, about 65% of Pakistan’s area is located in the Indus basin. So, Pakistan is mainly dependent on water resources of the Indus River basin. Pakistan, located between the 23°30'-36°45' N and 61°-75°31' E, has 2,912 km long boundary with India in the East, 523 km with China in the Northeast, 2,430 km with Afghanistan in the Northwest, and 909 27

km with Iran in the Southwest. Pakistan stretches over 1,600 km from North to South and about 885 km from East to West which covers an area of about 796,095 km2 and a population of about 172.8 million. Pakistan has land boundaries of 6,774 km and a coastal line of 1,046 km (Country-Profile, 2005; Khattak, 2011).

Figure 3-1 Geographic location of Pakistan Pakistan is a territory of great diversity with arid deserts in the East and snow covered mountains in the North. The territory consist portions of the Hindu Kush, Himalaya, and Karakoram mountain ranges. Pakistan is the home of some world’s highest peaks such as K2 (second world highest), with 8,611 m altitude above sea level. Intermountain valleys are located in much of the North-West Frontier Province, and rocky plateaus lie in the Balochistan province in the West. In the East and along the Indus River, extensive irrigated plains cover much of Punjab and Sindh provinces, which have deserts as well (Cholistan in Punjab, Thar in Sindh). Pakistan has a great diversity of climate. There are four main seasons in Pakistan: cool and dry (December-March), hot and dry (April-June), wet monsoon (July-September) and, dry post-monsoon (October-November). It receives less than 250 mm of rainfall annually, and about 75% of annual rainfall occurs during the monsoon season. The temperature in the country varies from -30 (during the coldest months in the mountainous northern part of Pakistan) to 50°C (during the warmest months in some areas of Punjab, Sindh and Balochistan provinces), with 27°C as a mean annual temperature of Pakistan (CountryProfile, 2005).

28

3.1.3

Major rivers of the basin

Five major tributaries such as Jhelum, Chenab, Ravi, Beas, and Sutlej rivers join the Indus from the Eastern side of Indus basin. Instead of these rivers, two smaller rivers Soan and Harrow also join the Indus from the Eastern side. A large number of small rivers such as Kabul, Swat, Panjkora, Kunar, Kurram, Gomal, Kohat, Tai, Tank, etc. join the Indus River on the Western side. Kabul River is the largest Western tributary of the Indus. The mean annual discharges of the major rivers are given in Table 3-2 (Ahmad, 2009; Khattak, 2011). Table 3-2 Mean annual discharge of the major rivers of the Indus basin SN

River

1 2 3 4 5 6

Indus Jhelum Chenab Ravi Sutlej Kabul

Mean Annual volume (billion m3) 74.1 28.3 31.6 7.14 7.14 27.43

29

Contribution (%) 42.1 16.1 18.0 4.1 4.1 15.6

Figure 3-2 Location map of the Indus River basin

30

3.2 Upper Jhelum River basin 3.2.1 Geography and major rivers The Upper Jhelum River basin is focused as the study area in this research work. Jhelum River shown in Figure 3-3 is one of biggest tributary of the Indus River. The Indus Basin is one of the largest basins in the world with a drainage area of 11, 65,500 sq. km. About 71% area of Pakistan lies in the Indus Basin. The Jhelum River drains West of Pir Punjal separating the provinces of Jammu and Kashmir. It originates from the Verinag Spring in the North-West side of Pir Punjal and moves parallel to the Indus River at an average elevation of 1,676 m. It moves through the area of about 5,957 km2 (2,300 mile2) alluvial soil of the Kashmir valley. On the North side of the Kashmir valley, there are large sources of glaciers which also contribute to the Jhelum River. Moving towards North-West direction and after flowing through Srinagar, which is the capital of Jammu and Kashmir, it joins at the Dal and Wular lakes near Baramula where it drops course sediments. Leaving the Baramula, it flows through a long gorge of 80 miles with mean slope of 6.28 m/km (33 feet per mile). Near Muzaffarabad, the biggest tributary of the Jhelum River called Kishanganga (Neelum) joins at the place called Domel. The Neelum originates from the great Himalayas with an elevation ranging from 4,573 m (15,000 ft) to 6,097 m (20,000 ft) covered with snow and drains about 6,734 sq. km (2,600 sq. mile) of hilly areas on the Eastern side of the Naga Parbat peak. Its length is about 150 miles and passes through very stable mountains which are covered with dense forests. The second biggest tributary of the Jhelum River is Kunhar River which joins the Jhelum at five miles below Domel, as shown in Figure 3-3. It drains in to the famous valleys of Kaghan and Naran, with an elevation ranging from 457-518 m. (1,500 to 1,700 ft). Ninety miles below Domel towards Mangla, the two tributaries, Poonch River and Kanshi River meet the Jhelum River. The Kanshi River drains in the Rawalpindi and Jhelum districts which carries most of the monsoon and seepage water. The Poonch River, which rises from the Southern side of the Pir-Panjal Mountain with an elevation of about 3,048-3,658 m (10,000 to 12,000 ft), joins the Jhelum River at the Tangrot near Mangla. The catchment area of the Poonch is mostly covered by forests having a drainage area of 3, 937 sq. km (1,520 sq. mile). It is about eighty miles long and passes through the Poonch, Kotli and Mirpur cities. The main charactristics such as basin slope, specific discharge, catchment areas, discharges, and the sediment load from Jhelum and its major tributaries are shown in Table 3-3 and Table 3-4. After passing through the Mangla reservoir, which divides the basin into upper Jhelum (33,342 sq. km) and lower Jhelum basin, the Jhelum River enters into the plains of Punjab where it joins the Chenab River at Trimmu barrage in the Jhang district and then the Ravi River, Beas River and Sutlaj River at Panjnad. It finally merges into the Indus River at Mithonkot. There is one dam at Mangla and two barrages, Rasul barrage and Trimmu barrage with a maximum upstream flow of about 24,000 m3/s and 18,000 m3/s, respectively constructed on the Jhelum River.

31

Table 3-3 Main charactristics of sub-basin of the Jhelum river basin River/ flow gauge

Sub-basin

Shape Length or Pre. (Km)

Area (km2)

Annual Dis. (m3/s)

Sp. Dis. (m3/s/km2)

Basin Slope (%)

1,050 7,416 4,274 0.5764 60 Neelum/Muzaffarab Neelum ad 509 2,335 1,257 0.5382 57 Kunhar/Garihabibull Kunhar a Jhelum 2,171 14,394 4,349 0.3021 38 Jhelum/Domel 3,973 24,750 10,159 0.4105 47 Jhelum/Azad-Pattan Jhelum Poonch 677 4,404 1,605 0.3645 30 Poonch/Kotli Kanshi 273 1,283 78 0.0607 5 Kanshi/Palotate Sp. Dis. Specific discharge; Imp. Imperviousness; Dis. Discharge; CN curve number; Peri. Perimeter

Imp. (%)

Basin CN

Elevation (m ASL) Max

Min

Mean

18

74

6,266

665

3,466

26

76

5,195

803

2,999

10 13 8 6

77 76 81 81

5,385 6,266 4,697 854

671 573 316 330

3,028 3,420 2,507 592

32

Table 3-4 Hydraulic features of the Jhelum River and its major tributaries Discharge (106 m3)

Sediments (106 m3 / year)

Domel Mangla

Basin Area (km2) 13,597 34,136

14,062 30,837

13 54,361

Neelum

M.abad

6,734

7,524

6,444

Kunhar

G. Habib

2,797

2,467

3,529

Poonch

Palak

3,937

2,467

7,004

Kanshi

Gujar khan

NA

444

361

Kahan

Rohtas

1,217

46

524

Tributary

Site

Jhelum Jhelum

(Source: (Ahmad and Chaudhry, 1988; PWG, 2011))

Figure 3-3 Location map and description of the study area 3.2.2

Hydro-climate condition

Jhelum basin has very diverse climate with very hot, cool, dry, and wet seasons. In the winter, minimum temperature can drop up to -25°C on some stations like Naran, and in summer, maximum temperature rises up to 50°C on some stations like Jhelum and Kotli. Figure 3-4 (a) illustrates the mean monthly temperature distribution in all climate stations and for the whole basin (black and bold line). It is seen from Figure 3-4 (a) and Table 3-5 that Jhelum, with a mean temperature of 23.53°C, and Naran with 6.14°C, are the hottest and coldest stations respectively in the study area. The month of July with a mean temperature of 23°C and January with 2.9°C, are the hottest and coldest months respectively. A mean temperature of 13.72°C is calculated for the entire basin. 33

Figure 3-4 (b) shows the monthly precipitation distribution in all the climate stations and for the mean in the whole basin (black and bold lines). Naran, Srinagar, Kupwara, Gulmarg, Astore and Qazigund have one relatively large peak in March. Climate stations (Kotli, Jhelum, Murree, Rawlakot, Plandri, Garidopatta, Muzaffarabad and Balakot) have two peaks, that is, a small peak in March and another substantial peak in the July or August (monsoon season). Based on the entire basin’s mean precipitation, two distinct peaks can be observed, one occurring in the month of March with a mean precipitation of 165 mm, which is 13.8% of the mean annual precipitation of basin, and the other in July with a mean precipitation of 143 mm, which is 12% of the mean annual precipitation of the basin. As for the weather stations, Murree is the wettest and Astore is the driest station in the study area. The whole basin has an average annual precipitation of about 1202 mm (Table 3-5). The maximum rainfall occurs at elevation from 5,000 to 6,000 m in the Karakoram Mountains. Since the most of the climate station lies in the valleys, it means that the precipitation which occurs in the basin can be greater than measured by stations (Archer and Fowler, 2008). Table 3-5 Some statistics about climate stations for the period of 1961-1990 in the Jhelum River basin Station Annual Precipitation Mean Tmax Mean Tmin Tmean (mm) (°C) (°C) (°C) 1 Jhelum 860 30.51 16.54 23.53 2 Kotli 1,249 28.41 15.75 22.08 3 Plandri 1,459 22.73 11.30 17.01 4 Rawlakot 1,398 21.87 10.25 16.06 5 Murree 1,765 16.57 8.97 12.77 6 Garidopatta 1,586 26.06 12.58 19.32 7 Muzaffarabad 1,418 27.34 13.54 20.44 8 Balakot 1,731 25.04 12.01 18.53 9 Naran 1,217 11.14 1.15 6.14 10 Kupwara 1,314 19.64 6.30 12.93 11 Gulmarg 1,574 11.73 1.98 6.81 12 Srinagar 764 19.76 7.37 13.52 13 Qazigund 1,395 19.11 6.57 12.80 14 Astore 564 15.48 4.06 9.77 Mean 1,202 19.79 7.68 13.72

34

50

Jhelum Rawlakot Muzaffarabad Astore Gulmarg

a Mean Temperature (°C)

40

Kotli Murree Balakot Kupwara Qazigund

Plandari Garidopatta Naran Srinagar Basin mean

30 20 10 0 -10 Jan 500 450

Feb

Apr

May

Jun

Jhelum Rawlakot Muzaffarabad Astore

b

400 Precipitation (mm/month)

Mar

Jul

Aug

Sep

Kotli Murree Balakot Kupwara

Oct

Nov

Dec

Plandari Garidopatta Naran Srinagar

350 300 250 200 150 100 50 0 Jan

Feb

Mar

Apr

May

Jun

Jul

Aug

Sep

Oct

Nov

Dec

Figure 3-4 Monthly (a) temperature and (b) precipitation at weather stations for 1961-1990 in the Jhelum basin Annual runoff of the Jhelum, Kunhar, and Neelum River is 750-850 mm, 1500-1700 mm, and 1125 mm, respectively above the Neelum confluence. The mean annual flow at the Mangla is about 856 mm. The Jhelum, Neelum and Kunhar below the confluence contribute about 45, 43 and 12% to the annual total runoff, respectively. Jhelum contributes a greater amount during the spring as well as winter, and in the month of March, its contribution reaches to almost 65% of the total annual flow. The Neelum and Kunhar contribute about 53 and 14% respectively of the total annual runoff in July. The peak flow occurs in May, June and July for Jhelum, Neelum and Kunhar, respectively. The southern tributaries—Kanshi and Poonch—have a peak flow in August but with extreme outlier high flow in this month. The peak flow at Mangla is most likely between June and July. All tributaries show a high seasonal percentage of flow from April to June than July to September, even though Neelum has marginally higher flow in spring at Muzzafarabad 35

(Figure 3-5). On the other hand, the Poonch and Kanshi show quite different seasonal flow patterns with a much higher percentage of summer flow due to monsoon rainfall rather than snowmelt. The Poonch River also has a significant percentage both in winter and spring due to winter rainfall and early spring snow melt at all lower elevations (Archer and Fowler, 2008). Figure 3-5 shows that flow pattern of Poonch River is quite different than other rivers in the basin. For example, peak flows in all other rivers occur during Jun-July. However, peak flows in Poonch River come in March and August. Kunhar at Naran

30.0

Mean monthly runoff (%)

Kunhar ar G.Habib 25.0

Neelum at M.Abad Jhelum at Domel

20.0

Poonch at Kotli Jhelum at Kohala

15.0

Jhelum at Azad Pattan 10.0 5.0 0.0 Jan

Feb

Mar

Apr

May

Jun

Jul

Aug

Sep

Oct

Nov

Dec

Figure 3-5 Mean monthly runoff, % of annul runoff, at different stream stations in the Jhelum basin 3.2.3

Mangla reservoir

The Mangla reservoir is mainly divided into two parts which are seperated by a saddle at a reservoir level of 339 (1113 ft) These parts are controlled by the Mangla dam and Jari dam. Jari outlet is used for irrigation purpose with a minimum operating level of 320 m (1050 ft) (On the other hand, the minimum level of operation of the main reservoir controlled by the Mangla dam is 316 m (1040 ft) The primary purpose of the reservoir is to fulfill the irrigation requirements which are mainly done by power tunnel, with power generation as the secondary objective. Although, the minimum conservation level of the reservoir is 316 m (1040 ft) however, it tries to maintain the level of reservoir up to 320 m (1050 ft) Table 3-6 shows the incremental and cumulative storage capacities at different zones (inactive, conservation, and flood control) of the reservoir in 1967 (when constructed) and in 2000, which shows definite decrease in storage capacity due to sedimentation (WAPDA, 2001). 3.2.4

Main outlets

There are four main outlets used to release the water: 1) Main spillway 2) Emergency spillway 3) Power house/irrigation tunnels, and 4) Jari outlet The main spillway having 9 gates is a submerge orifice type, with a sill elevation of 331 m (1086 ft) Each radial gate is 10.97 m (36 ft) wide and 12.2 m(40 ft) high. These gates are controlled by electricity which can move with the speed of 0.3048 m (1 ft/min). The main spillway has a maximum discharge capacity of 3.1148 × 104 m3/s (1.1 milion cfs) at an elevation of 366 m (1,202 ft) It is used mainly for the following purposes:

36

1. To release flood water through the reservoir with minimum increase in water level in the reservoir. 2. To discharge available water inside the reservoir in advance to control the coming flood. 3. To supply water for irrigation requirements/indents when the irrigation demand is above the capacity of power tunnels. 4. To release water when the reservoir is at its conservation level and the water is in excess of the power/irrigation valves. The emergency spillway is an uncontrolled outlet with a maximum capacity of 6.5 × 103 m3/s (0.23 milion cfs) at an elevation of 374.3 m (1228 ft) This is a weir type spillway with width length of 152 m (500 ft) Its crest elevation is about 366.4 m (1,202 ft) at which it begins to operate. The discharge of emergency spillway goes into the Bara Kas canal which joins the Jhelum river at the downstream of the reservoir. It is designed to operate only when the water release exceeds the capacity of main spillway and power tunnels. Table 3-6 Storage capacities of the Mangla reservoir w. r. to elevations in the 1967 and 2000 Description

Embankment crest Design flood level Conservation level Minimum level

Elevation (m) 376 374 366 317

Storage Capacity in 1967 (106 m3) Incremental Cumulative – – 2,233 9,485 6,587 7,253 666 666

Storage Capacity in 2000 (106 m3) Incremental Cumulative – – 2,233 8,055 5,575 5,822 185 185

The Jari dam outlet includes an intake, a tunnel, and a regulating valve. The valve is located at the end of the tunnel downstream. A fixed wheel type gate is fitted at the intake of the tunnel. The main purpose of the outlet is to supplement irrigation demands. The power station has five tunnels with an internal diameter of 9 m (30 ft) There is a fixed wheel gate at the intake of each tunnel. These gate are designed to operate either fully opened or fully closed. Each tunnel is subdivided into 2 parts to operate two set of turbines. There are 10 turbines with a maximum release capacity of 1,274 m3/s (45,000 cfs). The irrigation valves are also provided to fullfil irrigation demand when power tunnel release is less than the irrigation demand (WAPDA, 2001). The installed capacity of the Mangla power station is 1,000 MW with 10 units. Each unit has a capacity of 100 MW. The Mangla power station can produce a maximum power of 1150 MW, with 15% overload. The mean annual energy production is about 6,000 GWh. The energy produced during the 2007-08 was 4,687 GWh which is 23.8% less than the energy produced in 2006 (6,151 GWh). This shows that 1,464 GWh energy is decreased only in one year which is due to less inflow to the Mangla dam (WAPDA, 2008). 3.2.5

Mangla dam after-raising

At the construction time of the Mangla dam (1967), the Government of Pakistan requested the World Bank to raise the dam by 9 m (30 ft) which was later approved. So, the construction work of the Mangla dam raising project was started in 2003, raising the dam by another 9 m (30 ft) and the completion was commenced officially in December 2009. As a result of raising the height of the dam, the conservation level was reached up to 378.6 m 37

(1242 ft), with an additional storage of 3,577 million m3 (2.9 MAF). The power is also expected to increase by 14%. This means that the power plant can produce an annual energy of about 664 GWh (Kayani, 2012). 3.2.6

Historical operations of the Mangla reservoir

The historic operations can be described by fluctuations in the water elevation, hydropower releases, irrigation uses and flood release. Irrigation release The irrigation releases are based on the 10-day irrigation indents which was developed by the IRSA (Indus River System Authority) in 1991. Before 1991, WAPDA used to make irrigation indents for Mangla by the consultation of provinces. Rabi (October-March) and Kharif (April-September) have different irrigation indents. Although, the minimum reservoir level is held at 317 m (1,040 ft) but it tries to keep the reservoir level higher than 317 (1040 ft), so that early demands of the Kharif season can be fulfilled. Table 3-7 gives information about the historical releases for irrigation in the Rabi and Kharif seasons for the period of 1967-2000. Table 3-7 Historical max, min and mean irrigation demand for Rabi and Kharif from the Mangla reservoir calculated for the period of 1967-2000 Rabi Oct Nov Dec Jan Feb Mar

Max 106 m3 2,851 2,557 2,305 2,055 2,997 4,000

Kharif April May Jun Jul Aug Sep

3,303 5,417 6,336 3,988 3,413 2,637

m /s 1,100 987 889 793 1,156 1,543

Min 106 m3 921 1,233 868 454 724 554

1,274 2,090 2,445 1,539 1,317 1,017

739 1,365 771 759 685 881

3

m /s 355 476 335 175 279 214

Mean 106 m3 1,956 1,872 1,680 1,029 1,682 2,113

m3/s 755 722 648 397 649 815

285 527 297 293 264 340

2,012 2,802 2,473 1,906 1,567 1,679

776 1,081 954 735 604 648

3

Power release Since the primary goal of the Mangla is to meet irrigation demands and power tunnels are mainly used for this purpose, no water is released for power generation over and above the irrigation demands, except in case of floods. Only that water is used for power generation which is actually the irrigation requirements. So, the irrigation requirement is released through the power tunnels with a maximum discharge capacity of 1274 m3/s. However, Mangla is being used as a peaking station.

38

Flood release Flood releases are mostly made during the rainy season (monsoon) to hold the water level below the maximum conservation level or rule curve. A mean annual flood release is 4564 million m3 calculated for the period of 1967 to 2000. The main objective of the Mangla reservoir operation is the best utilization of inflow. In the last 33 years of operation, it has been observed that the mean annual flow is 28,173 million m3 (22.84 MAF) and the mean annual irrigation release is about 23,473 m3 (19.03 MAF), which shows that about 83% water is used for irrigation. However, percentage uses obviously varies from year to year (WAPDA, 2001). 3.2.7

Main features of the Mangla dam

The Mangla dam, which is one of the biggest earth fill dam of the world was constructed on the Jhelum River in 1967. This dam is a multipurpose project used to conserve the flood water of the Jhelum for irrigation and energy generation. The gross capacity of the dam was 7,253 million m3 (5.88 MAF) at the time of its construction. However, it is now reduced to 5,945 million m3 (4.82 MAF) during the 34 years of operation due to siltation. At the start of its operation, its life was expected to 120 years but now expected life is about 170 years due to some management practices in the basin (WAPDA, 2008). Its command area is about 6 Mha (million hectares) which is about 42% of the Indus Basin Irrigation system of Pakistan (14 million hectares) (Archer, et al., 2008). Some salient features of the Mangla reservoir are given in Table 3-8: Table 3-8 Some salient features of the Mangla dam Features Net electric power output Maximum monthly power generation during July 2007 Maximum daily generation on various dates Maximum load attained on various dates Number of power units Total design capacity Crest level of embankments Conservation level Minimum operating level Gross storage capacity Live storage capacity design Live storage capacity Present Expected life (at the start of its operation) Expected life now (after watershed management project) Max. height above core trench Crest length

Values/Units 4568.445 MKWh 562.374MKWh 27.21 MKWh 1150 MW 10 1,000 MW El. 376 m El. 366.4 m El. 317 m 7,253 million m3 6,587 million m3 5,723 million m3 120 years 170 years 138.4 m 2,560 m

(Source: WAPDA 2008) 3.3

Data description

Table 4-1 shows the main input data used in this study. However, the whole data required to achieve each objective is explained in more detail under each chapter. The data is collected from Pakistan Meteorological Department (PMD), India Meteorological Department (IMD) and Water and Power Development Authority (WAPDA) of Pakistan. 39

Table 3-9 Overview of main input data required for this study Data Meteorological Precipitation Max temperature Min temperature Wind speed Sunshine hours Humidity Hydrological Stream flow Reservoir Inflow to the Mangla dam Out flow from the Mangla dam GCM data Precipitation Max temperature Min temperature 26 present predictors 26 future predictors NCEP data 26 present predictors 26 future predictors

3.4

Frequency Daily/Monthly Daily/Monthly Daily/Monthly Monthly Monthly Monthly Daily Daily/10 Daily Daily/10 Daily Monthly Monthly Monthly Daily Daily Daily Daily

Conclusions • • • •

The Jhelum River is the biggest tributary of the Indus River. In this study, Upper Jhelum basin (Mangla watershed) is taken as the study area covering an area of 33,342 km2. This basin is located between Pakistan and India. The Mangla dam constructed (1967) on the Jhelum River is the second biggest reservoir in Pakistan. The primary objective of this dam is to fulfill irrigation demand mainly through the power tunnels, with hydropower as a byproduct. The installed capacity of the power plant is 1,000 MW. Recently, the Mangla dam was raised by 12 m (40 ft) resulting in an additional storage of 3577 million m3 (2.9 MAF) and 14% hydropower generation.

40

4

OVERALL METHODOLOGY

Figure 1-1 shows the whole methodology applied in this study to achieve all objectives. So, this chapter describes the summary of methodology applied to complete this study. The detailed methodology to achieve each specific objective is explained in each chapter separately. 4.1

Evaluation of Global Climate Models

The overall objective of this study is to assess the changes in climate and its impacts on stream flow, and hydropower generation from the Mangla power station. To date, the most advanced and credible tool used to study the effects of greenhouse gases on the atmosphere and to assess the changes in climate variables is the GCM model (Gebremeskel et al., 2005). Because many GCMs are freely available for downscaling purpose, it is difficult to decide which model gives better results. The following specific steps are used to choose a suitable GCM for this study: 1. The first step of this study is to choose a suitable GCM for downscaling of precipitation and temperature for the study area. For this purpose, 4 GCMs (HadCM3, CGCM2, CCSRIES, and CSISRO) out of eight were used in the IPCC Third Assessment Report as obtained according to the criteria described by Smith and Hulme (1998) and observed data availability, as well as GCMs data availability. 2. The monthly maximum temperature, minimum temperature, and precipitation data of these models are downloaded from IPCC Data Distribution center for A2 and B2 scenarios. 3. The three indicators, namely, coefficient of determination, root mean square error, and standard deviation are calculated by using the observed and GCMs data. 4. A model that gives better results relative to other GCMs is selected for further analysis (like downscaling). 4.2

Statistical downscaling 5. After the selection of a suitable GCM (HadCM3 in this study), 26 predictors (A2 and B2) of that GCM and NCEP each are downloaded (http://www.cccsn.ec.gc.ca/?page=dst-sdi) for statistical downscaling of maximum temperature, minimum temperature, and precipitation with Statistical Downscaling Model (SDSM). 6. Predictand (maximum temperature, minimum temperature, and precipitation) input files are prepared. 7. After preparing the data, the first and most important step of statistical downscaling is the screening of predictors. For this purpose, the predictands are fed into the SDSM one by one and regressed with 26 NCEP-predictors to select the most appropriate predictors. Explained variance and standard error are the two indicators which are used in this study during the screening of predictors. 8. There are three kinds of modules (monthly, seasonal, and annual). In second objective, monthly sub-model is used for downscaling of predictands. 9. After the screening of predictors, the model is calibrated with NCEP predictors by considering the two indicators (explained variance and standard error). 41

10. The model is validated by comparing the simulated data (daily, monthly, and seasonal time series) against observed data. The four indicators - correlation coefficient (R), coefficient of determination (R²), root mean square error (RMSE), and mean (µ) - are used for validation. 11. SDSM is also validated by comparing the spatial distribution of mean annual observed data with simulated data. For this purpose, spatial maps for each variable (Tmax, Tmin, and precipitation) are built by converting the mean annual point data into raster data by the Inverse Distance Weighted (IDW) interpolation method with ArcGis 9.3. 12. After successful validation, the future time series are simulated and future changes in maximum temperature, minimum temperature, and precipitation are assessed relative to baseline period (1961-1990) for the three periods—2020s (2011-2040), 2050s (2041-2070), and 2080s (2071-2099). 4.3

Evaluation of sub-models of SDSM

There are three kinds of modules in SDSM for downscaling of predictand, and the monthly sub-model has been used in the previous studies for downscaling purposes. In this objective, the monthly and annual sub-models are evaluated to explore which model gives better results in what conditions. 13. Steps 5 to 9 are repeated for both monthly and annual sub-models of SDSM. 14. After calibration of both sub-models (annual and monthly sub-models) with NCEP predictors, the data is simulated for the period of calibration (1961-1990) and validation (1991-2000) by feeding the NCEP and HadCM3 (A2 and B2) predictors. 15. Then, the outputs of both SDSMs are compared with observed data by calculating the coefficient of determination (R²), root mean square error (RMSE), mean(µ), standard deviation (б), relative error in mean (RE_µ) and standard deviation (RE_ б) for temperature and precipitation for the periods of calibration and validation. 16. The outputs of both SDSMs are also presented graphically against observed data to explore variations of observed data captured by simulated data for both calibration and validation period. 17. In this study, the Bias Correction (BC) is also applied on the outputs of both models. 18. Before applying the BC to future time series, BC is applied first on the simulated data for the validation period. 19. The above mentioned indicators are calculated for the validation period before and after application of BC. 20. It is seen that the results are improved after application of BC during validation period, and then BC is applied on future time series from both models (annual and monthly). 21. At the end, the change in maximum temperature, minimum temperature, and precipitation are accessed for three future periods (2020s, 2050s, and 2080s). 4.4

Climate change impact on water resources 22. The first step of hydrologic modeling is extracted in the basin characteristics such as slope, sub-basin areas, length of rivers, and longest flow paths, etc. These characteristics are obtained by feeding the DEM data into HEC-GeoHMS. 23. Then, hydrologic model (HEC-HMS) is set up for the Jhelum river basin. In this study, the initial and constant loss method, SCS unit hydrograph, recession method, 42

Muskingum, gauge weight method, Penman Monteith, and temperature index are included in HEC-HMS-setup for the Jhelum River basin to calculate runoff volume, direct runoff, base flow, channel losses, meteorological modeling, evapotranspiration and snow melt, respectively. 24. Then the model is calibrated and validated with observed data at different flow gauging stations. For this purpose, Nash efficiency, coefficient of determination, and root mean square error are used to check the performance of the model. 25. After successful calibration and validation of the model, the downscaled data (temperature and precipitation) for the future period is fed into the model to simulate the future time series of stream flow in the Jhelum basin. 26. In the end, the future change in the stream flow for the three periods (2020s, 2050s, and 2080s) are explored relative to baseline period under A2 and B2 scenarios. Flow duration curves, high, median, and low flows are extracted for the future periods. Moreover, the relationships between temperature, precipitation, and flow are also graphically presented. 4.5

Climate change impact on hydropower 27. Since the objective is to assess the impacts of climate change on the hydropower generation, the HEC-ResSim (Reservoir Simulation Model) is setup for the Mangla power station located at the Jhelum River basin. 28. As the Mangla dam has been recently (2003-2009) raised by about 12 m (40 ft), the model is run for both condition (before-raising and after-raising) by feeding the simulated data from HEC-HMS as input data to generate hydropower. 29. Then, the future changes in hydropower generation are determined for three future periods (2020s, 2050s, and 2080s) relative to baseline period for both before-raising and after-raising conditions under A2 and B2 scenarios. 30. In the end, the impacts of the raising of the Mangla dam are also explored on the hydropower generation.

43

Historical hydrometeorological data

GCMs outputs

HadCM3/NCEP predictors

Evaluation of GCMs

Objective 1

Evaluation of Annual and Monthly submodels of SDSM Bias Correction

Objective 3

Statistical Downscaling with monthly sub-model of SDSM

Future temporal and spatial Changes in Climate variables under A2 and B2

44

Future Changes in Climate variables under A2 and B2

Objective 4

Objective 2

Reservoir Model (HEC-ResSim)

Hydrologic Model (HEC-HMS) Objective 5

Future Changes in Water Resources under A2 and B2

Figure 4-1 Research framework of the study

Future time series under A2 and B2

Climate Change Impacts on Hydropower under A2 and B2 Scenario

5 5.1

EVALUATION OF GLOBAL CLIMATE MODELS

Data description

The monthly observed precipitation, maximum and minimum temperature data for the period of 1991-2009 is obtained from 14 climate stations from Pakistan Meteorological Department (PMD), India Meteorological Department (IMD) and Water and Power Development Authority (WAPDA) of Pakistan. GCM’s monthly data of precipitation, Tmax (max temperature) and Tmin (min temperature) is obtained from the IPCC Data Distributed Center for A2 and B2 scenarios— the selected scenarios are more related to regional changes (Arnell, 2004) and more details about these scenarios are given in Chapter 2—for the period of 1991-2009. This data is extracted from each climate model on the basis of grids which cover the study area by using CDO (Climate Data Operators) model. The grids of each model covering the study area are shown in Figure 5-1. 5.2 Methodology 5.2.1 GCM selection criteria To date, the most advanced and credible tool used to study the effects of greenhouse gases on the atmosphere, and to assess the changes in climate variables is the GCM model (Gebremeskel et al., 2005). These numerical based models are currently available and most boosted tools for simulating the changes in global climate system which are the results of increasing greenhouse gas concentration, land use change and many other anthropogenic activities. These represent the physical processes which occur in the ocean, atmosphere, cryosphere (frozen ice) and land surface (Fowler et al., 2007). In this section, criteria will be described for the best selection of GCM for the study area. The reason is to find out the best GCM which can explain the magnitude and variability of local variables (temperature and precipitation). The best GCM will be selected on the basis of the following criteria described by (Smith and Hulme, 1998). Vintage: The most recent GCMs model and their experiments should be selected because it will be based on more recent and advanced technology and knowledge. However, it is not a hard and fast rule because the best GCM is that which gives comparable results to observed data. Resolution: A higher resolution GCM should be selected because it can capture more regional climate information than coarse resolution. For example to capture complex feature like stream flow in mountainous region (Himalayan region or Andes), a very fine resolution GCM is required. In the past some GCM had the spatial resolution about 800 km but recently fine resolution GCM such as HadCM2 with 278×417 km spatial resolution have been developed. Validity: The GCM which produces the more reliable present climate during validation should be selected. In this process, the control data (temperature or precipitation) from different GCMs is compared with observed data by using some statistical parameter like mean, RMSE, correlation coefficient, and standard deviation (Sharma et al., 2007), and the GCM which gives the closest results to observed data is selected for future scenarios. The correlation coefficient is most commonly used statistical parameter for this purpose. 45

Representativeness of Results: Different GCMs can simulate different range of the climate changes on the regional level. So, where more than one model is to be selected, then those models which can represent a range of climate change in region should be selected. For example, if one model shows better representation of dry season then it should be chosen with that model which gives better representation in wet season. In this way both models can cover the range of changes of region for both seasons, wet and dry. 5.2.2

Performance Indicators

Mostly the following statistical parameters are used for the comparison of data from two or more different sources (Sharma et al., 2007). Coefficient of Determination-R2: “Statistic correlation (often measured as a correlation coefficient, R) indicates the strength and direction of a linear relationship between two random variables”. The square of the correlation coefficient is denoted by R2 and called as coefficient of determination. R

n ∑ xy - ∑ x × ∑ y

n ∑ x 2 - ∑ x 2 × n ∑ y 2 - ∑ y

2

5-1

Root Mean Square Error (RMSE): It is used to measure the accuracy of the model. It measures the difference between the observed data and the simulated data from the model. The differences of individual values are called residuals, and the RMSE aggregates all these differences (residuals) into a single value. 1 RMSE N

N

N 1

y-x 2 5-2

Standard Deviation (б) It is a very important statistic measure to represent the variability in data. It shows the variability of data around the mean. If the б has low value, it means that the values are very close to the mean, and high value shows the large range of data. б

1 N

x-x

2 5-3

The above mentioned statistical parameters will be used for the selection of the best GCM. The model which has high correlation coefficient, less RMSE and very close standard deviation with observed data will be chosen during comparison of outputs of the GCM against the observed data. 46

5.3

Results and discussion

Four GCMs (HadCM3, CCCma, CCSRIES, and CSISRO) are selected for the statistical analyses to evaluate them for this study based on the following limitations. These models are used in the 3rd Assessment Report of IPCC. • The station’s observed data is from 1961 to 2009 and almost 30 years of observed and Global Climate Model data is necessary for making strong statistical relations for statistical downscaling. ECHAM4/OPYC3 and NCAR-CSM have outputs starting from 1990 and 2000 respectively therefore, these are not included for further statistical analyses. • For this study, A2 and B2 scenarios are taken because these scenarios are mostly related to regional studies. NCAR-CSM does not have B2 scenario output therefore it is also not included in the further analyses. • R30-USA is also excluded for further analyses because it does not have Tmax and Tmin outputs and also very limited number of outputs. • Almost all GCM models used in the AR4 have output starting from the year of 2000. Although these are most recent and high resolution than GCMs used in the 3rd Assessment report but could not be used because their outputs starts from the year 2000 but baseline observed data which is selected for statistical downscaling and other analyses has a starting date from 1961. Additionally, B2 scenario is not simulated in the AR4. As shown in Figure 5-1, one grid of CSISRO-Mk2, CCCma as well as CCSRIES and two grids of HadCM3 cover the whole study area. The characteristics of these four models are describes in Appendix A. The upper grid and lower grid are denoted by G1 and G2, respectively. The characteristics climate models used in TAR are given in Appendix A.

47

Figure 5-1 Grid presentation of selected GCMs The results of statistical analysis are described in Table 5-1.The results show great variability among all GCMs. In case of precipitation, the coefficient of determination shows low values for all GCMs except HadCM3. The R2 value for HadCM3_A2_G1and HadCM3_A2_G2 (G1 used for upper grid & G2 for lover grid) is 2.8 and 10.5%, respectively. The correlation coefficient of HadCM3_B2_G1 and HadCM3_B2_G2 is 0.9 and 1.1%, respectively, but all other models have low values for R2. On the other hand, the RMSE of HadCM3 is lower than other GCMs. The standard deviation (showing the variability of data) of HadCM3 is closer to the standard deviation of observed data rather than the other GCMs. In Figure 5-2, the raw data of Tmax, Tmin, and precipitation from GCM is drawn against observed data. Figure 5-2(A) also explores that except HadCM3 all other models produce very opposite pattern, like for precipitation. CSISRO-MK2, CCCma and CCSRNIES show low precipitation from July to August which is opposite to observed precipitation. In case of Tmax and Tmin, R2 ranges between 33 and 88%. As for Tmax, almost all GCMs follow the same kind of pattern as observed data but all produce different magnitudes. All GCMs underestimate except HadCM3 which overestimates as shown in Figure 5-2 (B). For Tmin, all GCMs also underestimate but follow the good pattern as shown in Figure 5-2 (C). SCISRO-Mk2 has higher R2 and lower RMSE

48

in case of temperature (Tmax and Tmin) but not much better than other GCMs. But in case of precipitation CSISRO-Mk2 has very low correlation. 5.4

Conclusions

According to above discussion, although all GCMs predict better results for temperature than precipitation, the outputs from almost all GCMs underestimate temperature except HadCM3 which overestimate in case of Tmax. These results also show a real need for the downscaling of climate variables which are used in regional hydrological modeling and regional trend analysis. However, HadCM3 predict much better than all other GCMs in case of precipitation, with good pattern and high R2 relative to other GCMs. Based on the above results and discussion, it is decided to use the HadCM3 for further analysis. Table 5-1 R2, RMSE, and SD calculated from observed and GCMs raw data for Tmax, Tmin and precipitation for the period of 1981-2009 Precipitation R2

RMSE

SD

(mm)

(mm)

Max Temperature

Min Temperature

R2

R2

RMSE

SD

(°C)

(°C)

RMSE

SD

(°C)

(°C)

Observed

1.00000

0.00

51.50

1.00

0.00

7.64

1.00

0.00

3.93

HadCM3_A2_G1

0.02890

50.97

86.44

0.79

3.50

9.64

0.38

3.12

8.45

HadCM3_A2_G2

0.10531

48.92

53.47

0.76

3.75

9.81

0.34

3.20

7.24

CGCM2_A2

0.00044

51.71

96.63

0.78

3.63

11.67

0.38

3.10

11.37

CSISRO_A2

0.00156

51.68

36.51

0.87

2.72

8.41

0.43

2.98

8.30

CCSR/NIES_A2

0.00011

51.72

28.25

0.87

2.75

11.44

0.33

3.24

9.19

HadCM3_B2_G1

0.00934

51.48

89.34

0.81

3.31

9.93

0.38

3.12

7.29

HadCM3_B2_G2

0.01104

51.44

45.50

0.76

3.77

9.73

0.34

3.21

12.10

CGCM2_B2

0.00394

51.62

82.49

0.79

3.48

10.85

0.38

3.10

8.42

CSISRO_B2

0.00042

51.71

31.16

0.88

2.64

8.67

0.42

3.01

9.47

CCSR/NIES_B2

0.00001

51.72

27.09

0.74

3.91

11.40

0.37

3.14

3.93

G1 upper grid of HadCM3; G2 lower grid of HadCM3

49

Obs CGCM2-A2 HadCM3-B2_G1 CSISRO-B2

300

HadCM3-A2_G1 CSISRO-A2 HadCM3-B2_G2 CCSR/NIES-B2

HadCM3-A2_G2 CCSRNIES-A2 CGCM2-B2

Precipitation (mm)

250 200 150 100 50 0 Jan

Feb

Mar

Apr

May

Jun

Jul

Aug

Sep

Oct

Nov

Dec

(A) obs CGCM2-A2 HadCM3-B2_G1

50

HadCM3-A2_G1 CSISRO-A2 HadCM3-B2_G2

HadCM3-A2_G2 CCSRNIES-A2 CGCM2-B2

40

Tmax (°C)

30 20 10 0 -10 -20 Jan

Feb

Mar

Apr

May

Jun

Jul

Aug

Sep

Oct

Nov

Dec

(B) 40 30

obs

HadCM3-A2_G1

HadCM3-A2_G2

CGCM2-A2

CSISRO-A2

CCSRNIES-A2

HadCM3-B2_G1

HadCM3-B2_G2

CGCM2-B2

CSISRO-B2

CCSRNIES-B2

Tmin (°C)

20 10 0 -10 -20 -30 Jan

Feb

Mar

Apr

May

Jun

Jul

Aug

Sep

Oct

Nov

Dec

(C) Figure 5-2 Mean monthly (A) precipitation, (B) Tmax, and (C) Tmin of different GCMs against the observed data in the Jhelum River basin

50

6 6.1

STATISTICAL DOWNSCALING

Introduction

The concentration of CO2 and the other greenhouse gases has dramatically been increasing since 1950 (Gebremeskel et al., 2005) causing global energy unbalance and global warming. It is reported in the Fourth Assessment Report of Intergovernmental Panel on Climate Change (IPCC) that the global mean surface temperature has increased by a 0.74°C during the last hundred years (1906-2005), and huge increment has been seen in the last 50 years with an increasing rate of 0.13°C per 10 years(IPCC, 2007). It is predicted that global mean surface temperature could increase by 1.1 to 6.4°C during the 21st century (Chu et al., 2010).This global warming could change the hydrological cycle of the world causing effects on the water resources, public health, industrial and municipal water demand, water energy exploitation, and ecosystem(Chu et al., 2010; Zhang et al., 2011). Recently, General Circulation Models (GCMs) are the main tools to examine the changes and variability in most important climate variables — temperature and precipitation— on continental and global scale. These are very advance and numerical-based coupled models representing the global systems, such as sea-ice, atmosphere, and oceans (Fowler et al., 2007). The climate models are very useful to investigate the climate changes in the future. However, their outputs are spatially very coarse (e.g., 250 to 600 km) (Gebremeskel et al., 2005),and these cannot give the realistic presentation of local or regional scales (0-50 km, 50 × 50, and 500 × 500 km are defined as the local, regional, and global scales, respectively (Xu, 1999)) due to parameterization (Benestad et al., 2008). In addition, the outputs of GCMs cannot be used to assess the environmental and hydrological impacts of climate change on a local scale—basin or sub-basin level—due to the huge resolution-mismatch(Wilby et al., 2000). To deal with this mismatch, many downscaling methods which makes the large-scale outputs of GCMs useful for local/regional-scales have been developed (Hay et al., 2000; Wetterhall et al., 2006) during the last two decades. These methods have mostly been used in the Europe and USA (Wilby and Wigley, 1997; Gellens and Roulin, 1998; Murphy, 1999; Zorita and Storch, 1999; Li and Sailor, 2000; Hay and Clark, 2003; Harpham and Wilby, 2005; Salzmann et al., 2007; Wetterhall et al., 2009; Yang et al., 2010; Sunyer et al., 2011) and are now applied throughout the globe to examine the changes in climate variables (Ines and Hansen, 2006; Sharma et al., 2007; Hessami et al., 2008; Timbal and Jones, 2008; Elshamy et al., 2009; Kannan and Ghosh, 2011; Souvignet and Heinrich, 2011). Broadly, the downloading methods are divided into two main categories utilizing large-scale GCM’s outputs on local scale: (1) dynamical downscaling (DD)—also called as nested downscaling or numerical downscaling, and (2) statistical downscaling (SD). In DD, a Regional Climate Model (RCM ) of high resolution —ranging from 5 to 50 km—(Chu et al., 2010)nested with a GCM gets the inputs from the GCM, and then provides high resolution outputs on local-scale. Because the RCM is dependent on the boundary conditions of the GCM, it is a great chance to any systematic errors occurred in the RCM which belong to the GCM. Thus, a strong co-ordination is required between the regional and global climate modeling groups to ensure that the appropriate data are available. The simulations of the RCM are computationally intensive, concerning the domain size and resolution at which it is to run. This restricts the number of climate scenarios(Fowler et al., 2007). 51

The Statistical methods which make empirical/statistical relationships between variables of large-scale and local-scale are much simpler than DD. In SD, the local/regional information are derived by linking large-scale climate variables such as mean sea level pressure, temperature, zonal wind, and geopotential height etc. with small-scale variables (e.g., observed temperature or precipitation etc.) (Wetterhall et al., 2006). These methods offer a more immediate solution for downscaling climate variables, and have significantly lower computing requirements than DD. Consequently, these approaches are rapidly adopted by a wider community of scientists. (Wilby et al., 2000). However, a long historical weather station data is required to establish a suitable statistical relationship with large-scale variables (Chu et al., 2010), and the relationship between large and local scales is assumedto be temporally stationary (Hay and Clark, 2003). DD is a good alternative of the SD in the basins where no historical data is available. SD is further classified into three main types: (1) regression, (2) stochastic weather generator, and (3) weather typing, which are reviewed by these studies (Wilby and Wigley, 1997; Wilby et al., 2002; Fowler et al., 2007) in more detail. The regression methods (e.g., Artificial Neural Network, Principal Component Analysis, Multiple Linear Regression, conical Correlation Analysis etc) have extensively been used in hydrological impact assessment studies under different IPCC emission scenarios (Hewitson and Crane, 1996; Chu et al., 2010). So far, many SD models have been developed for downscaling. Statistical Downscaling Model (SDSM) which is a combination of multiple linear regression and stochastic weather generator has widely been used throughout the world (Huang et al., 2011) for downscaling the most important climate variable (e.g., temperature, precipitation, and evaporation, etc.) for assessing hydrologic responses under climate change scenarios (Diaz-Nieto and Wilby, 2005; Gagnon et al., 2005; Gebremeskel et al., 2005; Wilby et al., 2006). In statistical downscaling method, the most important and central part is to select some suitable predictors for a specified predictand (Wilby et al., 2002; Huang et al., 2011). Most of the time in case of precipitation one predictor cannot be able to explain the variances of predictand and to select the other suitable predictors is much subjective, because many things such as multiple colinearity, statistical significance, physical sensitivity etc. have to be considered during selection of predictors. So, to solve this problem a more quantitative procedure is used in this study to order the most suitable predictors by using the concept of partial correlation. For statistical downscaling, a well-known Statistical downscaling model (SDSM) developed by Wilby et al., (2002) is selected. This method is considered to be better in case of capturing intra-annual variability than the other statistical techniques, e.g., weather typing and weather generator (Hashmi et al., 2011), and most widely used throughout the world (Huang et al., 2011) for downscaling the most important climate variable (temperature, precipitation, and evaporation, etc.) for assessing hydrologic responses under climate change (Diaz-Nieto and Wilby, 2005; Gagnon et al., 2005; Gebremeskel et al., 2005; Wilby et al., 2006). Some studies (Tripathi et al., 2006; Anandhi et al., 2007; Akhtar et al., 2008; Ghosh and Mujumdar, 2008; Mujumdar and Ghosh, 2008; Ashiq et al., 2010; Goyal and Ojha, 2010; Opitz-Stapleton and Gangopadhyay, 2010) have been reported, using SD and DD methods, in the South Asian region, most of these studies conducted in Indian river basins. A couple of studies such as Akhtar et al. (2008) as well as Ashiq et al. (2010) using DD and one study 52

by Mahmood and Babel (2012) using SDare reported in Pakistan. In the study by Akhtar et al. (2008), a Regional Climate model called asPRECIS (Providing REgional Climate for Impact Studies)is used downscaling mean temperature and precipitation for hydrological impacts analysis under A2 scenario for the period of 2071-2100 in the region of the Hindukush-Karakorum-Himalalya (HKH) ranges located in the Indus River basin, Pakistan. Akhtar et al. (2008) conclude that PRECIS has several uncertainty sources. Ashiq et al. (2010) interpolates monthly precipitation output of PRECIS, run by Akhtar et al. (2008), from coarse resolution (50×50 km) to a fine scale (250×250 m), in the northwestern Himalayan Mountains and the upper Indus plains of Pakistan. This also covers a small part of the Jhelum basin located in Pakistan. In this study, it is concluded that the systematic errors associated with RCM cannot be reduced by interpolation methods. Only one study by Mahmood and Babel (2012) is reported in the Jhelum River basin using statistical downscaling. In this study, they evaluate the SDSM developed by annual and monthly sub-models (there are three sub-models, monthly, annual and seasonal, in SDSM to make a bridge between local and large scale variables) in the mountainous regionwith a high influence of monsoon. A Bias Correction technique is also applied on the outputs from both sub-models because SDSM developed by annual sub-model shows lack in capturing the variability in data. The changes in future periods (2020s, 2050s, and 2080s) are also discussed from both sub-models after the application of Bias Correction. In this study, they conclude that the SDSM developed by the annual sub-model cannot be used to investigate the intra annual (monthly or seasonal time series) variations, without Bias Correction, in the climate variables (temperature and precipitation), although it can predict mean values reasonably well. On the other hand, they also conclude that the SDSM developed by monthly sub-model can be used for downscaling and to examine the intra annual variations without Bias Correction. Base on the conclusions by Mahmood and Babel (2012), in the present study, the SDSM is applied for downscaling of max temperature min temperature and precipitation using monthly sub-model for the future periods: 2020s (2011-2040), 2050s (2041-2070), 2080s (2071-2099) under A2 and B2 scenarios. For the sake of deep analysis, in the present study, the whole Jhelum basin is dived into two sub-basins based on the occurrence of precipitation. Moreover, changes in spatial distributions of mean annual max temperature, min temperature, and precipitation relative to baseline period (1961-1990) are also analyzed in the two sub-basins as well as in the whole basin. 6.2

Sub-basins of the study area

For in-depth analysis, the whole basin is divided in two parts (OPP and TPP described shown in Figure 6-1) based on the rainfall regimes. The detailed information such as topography and climatic conditions about the study area and sub-basins of the Jhelum basin is described in Chapter 3. Climate conditions in the Jhelum basin Figure 6-2(a) shows the mean monthly rainfall regimes of all climate stations used in this study for the period of 1961-1990. Some stations such as Naran, Srinagar, Kupwara, Gulmarg, Astore, and Qazigund have one big peak in the March, and other climate stations like Kotli, Jhelum, Murree, Plandri, Rawlakot, Murree, Garidopatta, Muzaffarabad, and Balakot have two peaks one small peak centered on the March, and other big peak centered 53

in the July. According to occurrence of rainfall-regimes in the basin, the whole basin is divided into two parts: (1) One Peak Precipitation (OPP) basin, and (2) Two Peak Precipitation (TPP) basin. The OPP basin contains most of the northeast parts of basin (having Srinagar, Gulmarg, Qazigund, as well as Astore weather stations) and some of northwest parts of basin ( having Naran and Kupwara weather stations) shown in Figure 6-1. On the other hand, the TPP basin consists of South-West parts of the basin (having climate stations like Jhelum, Kotli, Plandri, Rawlakot, and Murree) and North-West parts of basin (having climate stations like Garidopatta, Muzaffarabad, and Balakot) shown in Figure 6-1. It is seen in Figure 6-2(b) and Table 6-1 that the TPP basin, with a mean annual rainfall of 1,341 mm, is wetter than OPP basin, with a mean annual rainfall of 1,139 mm. The July and March are the wettest months in the TPP and OPP basin respectively, and the November and September are the driest months in TPP and OPP basin respectively. As for climate station, Murree—with a mean annual precipitation of 1,765 mm—and kotli—with a mean annual precipitation of 1,249 mm—are the wettest and driest weather stations inside the TPP basin, and Balakot (annual precipitation of 1,731 mm) and Srinagar (annual precipitation of 764 mm) are wettest and driest climate station in the OPP basin. A mean annual precipitation of 1,202 occurs in the whole basin. Figure 6-3 and Table 6-1 also describes that the TPP basin is hotter than the OPP basin with mean temperature of 19.85 and 10.93°C respectively. The hottest month is the Jun in TPP basin and July in the OPP basin as well as in whole basin. The Jan is the coldest month in both sub-basins—TPP and OPP basin— and also in whole basin. Regarding to climate stations, Kotli (with mean temperature of 22°C) and Naran (with mean temperature of 6.14°C) are the hottest and coldest stations in the study. The whole basin has a mean temperature of 13.72°C. Although the Astore (driest) and Jhelum (hottest) do not lie in the Upper Jhelum River basin, these stations are included in the present study due to lack of climate station in the basin, and both station also have good quality meteorological data.

Figure 6-1 Location map of the Jhelum basin and climate stations used in the present study 54

Table 6-1 Some statistics about climate stations for the period of 1961-1990, in the Jhelum River basin

1 2 3 4 5 6 7 8 9 10 11 12 13 14

Jhelum Kotli Plandri Rawalakot Murree G.Dupatta M. abad Balakot TPP basin Naran Kupwara Gulmarg Srinagar Qazigund Astore OPP basin Whole basin

Precipitation (mm)

500 400 300

a

Annual Precipitation (mm) 860 1,249 1,459 1,398 1,765 1,586 1,418 1,731 1,341 1,217 1,314 1,574 764 1,395 564 1,139 1,202 Jhelum Plandari Murree M.abad Balakot Astore Srinagar Qazigund

Kotli Rawalakot Gari-dup Shinkiari Naran Kupwara Gulmarg

200 100

Mean Tmax (°C) 30.51 28.41 22.73 21.87 16.57 26.06 27.34 25.04 26.21 11.14 19.64 11.73 19.76 19.11 15.48 16.88 19.79

Mean Tmin (°C) 16.54 15.75 11.30 10.25 8.97 12.58 13.54 12.01 13.50 1.15 6.30 1.98 7.37 6.57 4.06 5.04 7.68

Tmean(°C) 23.53 22.08 17.01 16.06 12.77 19.32 20.44 18.53 19.85 6.14 12.93 6.81 13.52 12.80 9.77 10.93 13.72

500 Precipitation (mm)

Station

0

TPP Basin OPP Basin Whole Basin

400 300

b

200 100 0

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

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

Figure 6-2 Distribution of mean monthly rainfall over (a) all weather stations, (b) TPP basin, OPP basin, and whole basin for 1961-1990

55

Mean temperature (°C)

40 30 20

Kotli Rawalakot Gari-dup Shinkiari Naran Kupwara Gulmarg

a

10

50 Mean temperature (°C)

Jhelum Plandari Murree M.abad Balakot Astore Srinagar Quazi_Gund

50

0

30

b

20 10

TPP Basin OPP Basin Whole Basin

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

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

40

-10

Month

Month

Figure 6-3 Distribution of mean monthly temperature at (a) all weather stations, (b) TPP basin, OPP basin, and whole basin for 1961-1990 6.3

Data description

Measured daily, and monthly historical data—for period of 1961-2000—of max temperature (Tmax), min temperature (Tmin), and precipitation is obtained from the Water and Power Development Authority of Pakistan (WAPDA), Pakistan Meteorological Department (PMD), and India Meteorological Department (IMD). The historical data of meteorological stations like Kupwara, Srinagar, Gulmarg, and Qazigund is obtained from the IMD, Rawlakot, Plandri, Kotli, and Naran from the WAPDA, and Muzaffarabad, Jhelum, Garidopatta, Balakot, Murree, and Astore from the PMD. All stations are described in Figure 6-1. Some daily precipitation data for Srinagar and Qazigund for the period of 1961-1970 is obtained from the National Climate Data Center (NCDC). The 26 predictors each of NCEP, H3A2, and H3B2 are obtained from the Canadian website http://www.cccsn.ec.gc.ca/?page=dst-sdi for the period of 1961-2001 and 1961-2099, respectively. H3A2 and H3B2 are the IPCC emission scenarios of HadCM3 Global Climate Model. These predictors are specially prepared for SDSM model. During the preparation, the NCEP predictors (2.5° × 2.5°) are first interpolated to HadCM3 grid resolution (2.5° × 3.75°) to eliminate the spatial difference. Then, the NCEP and HadCM3 predictors are normalized by long term (e.g., 1961-1990) mean and standard deviation. Only HadCM3, CGCM2, and CGCM3 predictors are available in such form that can be downloaded from the above mentioned website according to the coordinates of study area and can be used directly into SDSM. 6.4 Methodology 6.4.1 Description of SDSM The Statistical Downscaling Model (SDSM) is a combination of two statistical downscaling techniques: (1) Stochastic Weather Generator (SWG), and (2) Multiple Linear Regression (MLR). In MLR, a statistical/empirical relationship between NCEP predictors and predictands is established which leads to produce some regression parameters. The regression parameters are, then, used by SWG—along with NCEP and GCM predictors—to generate daily time series (Wilby et al., 2002).

56

In SDSM, different kinds of indicators such as correlation matrix, partial correlation, Pvalue, histograms, and scatter plots are used to select some suitable predictors from the atmospheric predictors, through a multiple linear regression model. There must be considered a multiple co-linearity during the selection of predictors. Ordinary Least Squares (OLS) and Dual Simplex (DS) are two optimization methods. The OLS is faster than DS and produces comparable results with DS. Three kinds of sub-models (monthly, seasonal, and annual) are used to establish the statistical/empirical relationships between the predictands (temperature and precipitation) and predictors. In the monthly sub-model, 12 regression equations are developed, one for each of the 12 months, during the calibration process, giving different calibrated parameters. SDSM has two kinds of sub-models (1) conditional used for condition variable such as precipitation and precipitation, (2) unconditional used for independent local scale variable—temperature— during the calibration(Wilby et al., 2002; Liu et al., 2009; Chu et al., 2010; Mahmood and Babel, 2012). Mostly, unlike temperature, precipitation data is not distributed normally. So, the precipitation data can be made normal by SDSM before using the data in regression equations (Khan et al., 2006). For example, Khan et al. (2006), Huang et al. (2011), and Mahmood and Bablel (2012) used the fourth root to turn the precipitation into normal before using it in a regression equation. Developing SDSM, two kinds of daily time series, daily historical site data and NCEP daily predictors, are required. SDSM simulates daily time series as outputs forcing the NCEP or HadCM3 predictors (Huang et al., 2011). A full mathematical detail is presented in (Wilby et al., 1999). 6.4.2

Selection of predictors

In all statistical downscaling methods, the first and most important process is the screening of large scale variables (Wilby et al., 2002; Huang et al., 2011). Different kinds of methods such as stepwise regression, ridge regression,(Hessami et al., 2008)and multiple linear regression, along with partial correlation forthe selection of predictors. The four main indicators such as explained variance, correlation matrix, partial correlation, and P-value are used during the selection of predictors in SDSM. The combination of partial correlation and p-value is generally used for screening process like in these studies (Wilby et al., 2002; Chu et al., 2010; Hashmi et al., 2011; Huang et al., 2011; Souvignet and Heinrich, 2011) . The selection of predictors is more subjective and dependant on user judgement. Two main things which must be considered during the predictor selection are the muliple co-linearity and physical sense beween predictors and predictands. In SDSM, selection of the first and most prominent predictor is relatively easy which can be done with simple correlation matrix. However, the selection of the second, third, fourth and so on is more subjective. So, in this study, a more quantitative procedure used by Mahmood and Babel (2012) is applied for screening of predictors. In this procedure, a combination of correlation matrix, partial correlation, p-value, and percentage reduction in partial correlation is used. Twenty six NCEP predictors described in Table 6-2 are used during screening of suitable predictors for each of the predictors (Tmax, Tmin, and Precipitation).

57

Table 6-2 NCEP predictors used in the screening process Predictors 1 p_f 2 p_u 3 p_v 4 p_z 5 p_th 6 p_zh 7 Rhum 8 p5_f 9 p5_u 10 p5_v 11 p5_z 12 p5th 13 p5zh

6.4.3

Description Surface airflow strengh Surface zonal velocity Surface meridional velocity Surface vorticity Surface wind direction Surface divergence Surface relative humidity 500 hPa airflow strengh 500 hPa zonal velocity 500 hPa meridional velocity 500 hPa vorticity 500 hPa wind direction 500 hPa divergence

Predictors 14 r500 15 p8_f 16 p8_u 17 p8_v 18 p8_z 19 p8th 20 p8zh 21 r850 22 p500 23 p850 24 temp 25 shum 26 mslp

Description 500 hPa relative humidity 850 hPa airflow strengh 850 hPazonal velocity 850 hPa meridional velocity 850 hPa vorticity 850 hPa wind direction 850 hPa divergence 850 hPa relative humidity 500 hPa geopotential height 850 hPa geopotential height Mean temperature at 2m height Surface specific humidity Mean sea level pressure

Calibration and validation

According to the daily available data, three data sets such —961-1990, 1969-1990, and 1970-1990— are used to calibrate the model. Nonetheless, a data period from 1991 to 2000 is used for validation of all climate stations. In this study, monthly sub-model is used to calibrate the SDSM utilizing the selected NCEP predictors for each of the predictands (Tmax, Tmin, and precipitation), at each site. The unconditional sub-model is used for temperature without any transformation because temperature follows almost normal distribution, and conditional sub-model is set for precipitation with fourth root transformation to make the precipitation normal before using in regression equations. To check performance of SDSM during calibration, two statistical indicators (percentage of explained variance and standard error) are usedin the present study, the same as in these studies (Wilby et al., 2002; Huang et al., 2011). After calibration, Tmin, Tmax and precipitation data is simulated, using NCEP, H3A2, ad H3B2 predictors, for the period of 1991-2000 for validation. Then, the outputs of SDSM are compared with observed data by calculating the correlation coefficient, coefficient of determination (R²), root mean square error (RMSE), and mean (µ). These indicators are first calculated for each weather station, and then mean values of each indicator are calculated from all weather stations. The simulated data is also compared graphically with observed data to check the temporal pattern and variability during validation. SDSM is also validated by comparing the spatial distribution of mean annual observed data with simulated data. For this purpose, spatial maps for each variable (Tmax, Tmin, and precipitation) are built by converting the mean annual point data into raster data by the Inverse Distance Weighted (IDW) interpolation method with ArcGis 9.3, as done by (Huang et al., 2011). After satisfactory results of calibration and validation, the Tmax, Tmin, and precipitation data is simulated for the period of 2011-2099 under H3A2 and H3B2 scenarios. Then, the simulated data is divided into three spells: 2020s (2011-2040), 2050s (2041-2070), and 2080s (2071-2099), and compared with the baseline period (1961-1990) to analyze the future changes in the basin. For the illustration purpose, the temporal and spatial changes are also presented graphically, as in these studies (Chu et al., 2010). Gulmarg and Qazigund weather 58

stations are lack of data starting from 1961. For these stations, the data is projected back by SDSM using NCEP predictors. In this study, 1961-1990 is taken as base period. This period has been using as the base period worldwide in the majority of climate change studies (Huang et al., 2011).A 30 years period is also considered long enough to define local climate because it is likely to have dry, wet, cool and warm period. The IPCC also recommends such data period length for the baseline (Gebremeskel et al., 2005). For the illustration purpose, the temporal and spatial changes are presented graphically, as in these studies (Chu et al., 2010; Huang et al., 2011). Gulmarg and Qazigund weather stations are lack of data starting from 1961. For these stations, the data is projected back by SDSM using NCEP predictors. 6.5 Results and discussion 6.5.1 Screening of predictors Table 6-3shows the selected predictors, with mean absolute partial correlation (P.r), for Tmax, Tmin and precipitation with significance level of 0.05. It is seen that temp (temperature at 2m height) is a supper predictor for Tmax and Tmin in both basins, TPP and OPP. A supper predictor is the most prominent and has max correlation with predictand. In case of precipitation, surface specific humidity (shum) and Meridional wind velocity at 500 hPa (p5_v) are the most dominant predictors in TPP and OPP basin, respectively, for almost all weather stations. These predictors also express physical sense with local variables, Tmax, Tmin and precipitation. The selected predictors for the present study are also match with the predictors used in several other studies (Wilby et al., 2002; Chu et al., 2010; Hashmi et al., 2011; Mahmood and Babel, 2012). Different combinations of predictors—along with super predictor—are used for each predictand, on each site, to improve the performance of SDSM during the calibration. Table 6-3 Selected predictors and their mean absolute partial correlation coefficient during screening TPP basin

OPP basin

Tmax temp r500 P8_z

Abs P.r 0.73 0.22 0.19

Tmin temp rhum r500 P8_z

Abs P.r 0.82 0.18 0.12 0.32

temp p_u p_z P8_z r500

0.76 0.38 0.32 0.25 0.17

temp p_u p_z p8_z

0.79 0.37 0.32 0.26

Precipitation shum p5_v r850 p8_v mslp p5_v p8_v p5_z p8_z

Abs P.r 0.21 0.14 0.10 0.15 0.12 0.23 0.14 0.11 0.08

Bold are supper predictors 6.5.2

Calibration of SDSM

Percentage of explained variances (E) used as SDSM performance indicator ranges from 60 to 72% for Tmax, 67 to 85% for Tmin, and 8 to 32% for precipitation, and the mean standard deviations (SE) for Tmax, Tmin and precipitation are 4.6°C, 3.4°C, and 0.45 mm/day, 59

respectively. The E of precipitation is much lower than both, Tmax and Tmin, because the precipitation is a heterogeneous climate variable and difficult to simulate. Therefore, the E of precipitation is more likely lower than 40%. Nonetheless, the E for temperature is more likely greater than 70% (Wilby et al., 2002). Table 6-4 shows that the calibration results (Tmax, Tmin, and precipitation) of the present study are satisfactory relative to some previous studies. Table 6-4 Calibration results comparison with some previous studies using Explained variance (%) Study

Region

(Wilby et al., 2002) (Liu et al., 2008)

Toronto, Canada Upper-middle Yellow River, China Yangtze River basin, China Quebec, Canada Coquimbo Region

(Huang et al., 2011) (Nguyen, 2005) (Souvignet and Heinrich, 2011) This study

Upper Jhelum River basin, Pakistan

Temperature Tmax Tmin 73 72 N/A N/A

Precipitation Prec. 28 8-20

N/A 71-78* 28-65

39-66

18.5-32.4 6.2-9.8 19-56

60-72

67-85

8-32

N/A

Tmean Mean temperature; N/A Not available, * mean temperature 6.5.3

Validation of SDSM

For the validation of the SDSM, three sets of data are simulated by forcing the NCEP, H3A2, and H3B2 large scale variables, along with calibrated parameters, for the period of 19912000, and for each of local scale variable (Tmax, Tmin, and precipitation). Then, the SDSM performance indicators such as correlation coefficient (R), coefficient of determination (R²), root mean square error (RMSE), and mean (µ) are calculated to compare the observed and simulated data. These indicators are calculated using daily, monthly, and season time series of observed and simulated data (Tmax, Tmin, and precipitation), and the results are described in Table 6-5, Table 6-6, and Table 6-7. In case of Tmax, the mean R², the most important indicator, values obtained from daily, monthly and seasonal time series of NCEP are 76, 93 and 95%; the R² values from monthly as well as seasonal time series of H3A2 and H3B2 are 92 and 95%, and 92 and 94.6%, respectively. As for Tmin, the mean R² values obtained by comparing the observed and NCEP, H3A2, and H3B2 time series are: 85 (daily), 95 (monthly), and 96.8% (seasonal); 93.7 (monthly), and 96.4% (seasonal); 93.3(monthly) and 95.9% (seasonal), respectively. For precipitation, the the mean R² values obtained by NCEP data are 9.6 (daily), 42 (monthly), and 61.7% (seasonal), by H3A2 data are 28.1 (monthly) and 55.2% (seasonal), and by H3B2 are 22.3(monthly) and 50.8 (seasonal). On the whole, the results of Tmax and Tmin are much better than precipitation regarding all indicators in daily, monthly, and seasonal time series, and the Tmin results are slightly better than Tmax. It is seen that the R2 values obtained by NCEP daily precipitation (ranging 815%) are much lower than monthly (21-73%) and seasonal (25-87%). However these results are comparable with the previous study (Huang et al., 2011). The main reason of lower results of daily precipitations is that the occurrence/amount of precipitation is a stochastic process. Therefore, the simulation of precipitation is always a difficult task (Huang et al., 60

2011). Several previous studies (Khan et al., 2006; Fealy and Sweeney, 2007)show worse results of daily precipitation than monthly or/and seasonal. Thus, the SDSM shows better applicability, when downscaling monthly and seasonal precipitation. It is also observed that the results obtained by NCEP simulated data are better than the H3A2 and H3B2 in each of the predictands (Tmax, Tmin, and precipitation), and the simulated data from H3A2 gives slightly better results than H3B2. Because the SDSM is calibrated with NCEP predictors, thus, the calibrated parameters give biases when the model is run by the H3A2 and H3B2 predictors. Table 6-5 Statistical comparison of observed and simulated daily, monthly, and seasonal time series of Tmax during validation (1991-2000), in the Jhelum River basin Correlation (%) Range Mean Daily

Monthly

Seasonal

R2 (%) Range Mean

RMSE (°C) Range Mean

µ (°C) Range

Mean 21.03 21.02

Obs NCEP

84-92

87.48

71-85

76.5

3.08-3.98

3.52

10.2-30.41 11.5-30.55

Obs NCEP H3A2 H3B2

95-98 94-98 93-98

96.48 95.96 96.00

89-97 88-96 89-96

93.1 92.1 92.0

1.55-2.41 1.70-3.33 1.73-3.25

1.88 2.12 2.17

10.12-30.01 11.47-30.52 11.29-30.49 11.32-30.52

20.99 20.98 21.09 21.16

1.40 1.47 1.57

10.12-30.01 11.47-30.52 11.29-30.49 11.32-30.52

20.99 20.98 21.09 21.16

Obs NCEP H3A2 H3B2

96-99 95-99 95-99

97.67 97.65 97.30

92-98 92-98 91-98

95.4 95.3 94.6

1.06-2.04 1.11-2.75 1.22-2.74

Table 6-6 Statistical comparison of observed and simulated daily, monthly, and seasonal time series of Tmin during validation (1991-2000), in the Jhelum River basin

Daily

Monthly

Seasonal

Correlation (%) Range Mean

R2 (%) Range Mean

RMSE (°C) Range Mean

µ (°C) Range

Mean

Obs NCEP

88-95

92.27

77-89

85.2

2.32-3.26

2.64

1.54-16.85 1.18-16.51

9.15 8.98

Obs NCEP H3A2 H3B2

93-99 93-98 93-98

97.49 96.81 96.59

87-98 86-97 86-97

95.1 93.7 93.3

1.09-2.41 1.23-3.15 1.26-3.16

1.45 1.75 1.82

1.48-16.82 1.13-16.47 1.08-16.51 1.10-16.52

9.11 8.94 9.05 9.06

Obs NCEP H3A2 H3B2

95-99 94-99 94-99

98.37 98.20 97.90

90-99 89-99 88-99

96.8 96.4 95.9

0.56-2.00 0.60-2.50 0.63-2.58

1.05 1.18 1.27

1.48-16.82 1.13-16.47 1.08-16.51 1.10-16.52

9.11 8.94 9.05 9.06

61

Table 6-7 Statistical comparison of observed and simulated daily, monthly, and seasonal time series of precipitation during validation (1991-2000), in the Jhelum River basin

Daily

Monthly

Seasonal

Correlation (%) Range Mean

R2 (%) Range

Obs NCEP

10-39

25.045

Obs NCEP H3A2 H3B2

43-85 29-72 25-68

Obs NCEP H3A2 H3B2

50-93 40-81 26-77

Mean

RMSE (mm) Range Mean

µ (mm) Range

Mean

8-15

9.6

1.9-5.5

3.65

1.74-5.69 1.69-5.53

3.919 3.620

62.70 51.40 43.91

21-73 18-51 13-46

42.0 28.1 22.3

24-82 43-12 47-13

50.67 86.64 89.89

53-174 50-166 50-154 48-149

119.77 110.41 109.87 105.13

70.90 66.00 55.27

25-87 30-67 22-60

61.7 55.2 50.8

17-55 27-85 33-95

35.40 53.71 57.88

53-174 50-165 50-154 48-149

119.30 110.21 109.87 105.13

Temporal comparison of observed with downscaled data Since the whole basin is dived into two sub-basin (TPP and OPP) based on the precipitation regimes, the mean monthly, seasonal, and annual downscaled data (Tmax, Tmin, and precipitation)is compared with the observed data in TPP, OPP and whole basin as done in several studies (Wilby et al., 2002; Gebremeskel et al., 2005; Zhao and Xu, 2008; Chu et al., 2010; Huang et al., 2011). For this purpose, three datasets from each of the variables (Tmax, Tmin and precipitation) are simulated by forcing the NCEP, H3A2, and H3B2 predictors into SDSM and compared with the observed data as shown. Temperature (Tmax and Tmin) and precipitation Figure 6-4 and Figure 6-5show that the mean monthly, seasonal, and annual temperature (Tmax and Tmin) are well simulated with NCEP, H3A2, and H3B2 in the TPP, OPP as well as in the whole basin. The patters and seasonal variations of observed data is best captured by NCEP simulated data in both sub-basins. However, the outputs simulated by H3A2 and H3B2 are also comparable with NCEP. Figure 6-6 describes the graphical comparison of mean monthly, seasonal, and annual simulated precipitation against observed in the TPP, OPP, and whole basin. In the TPP basin, Figure 6-6(a) shows that the big precipitation-peak occurred inJuly-August (monsoon season) is reasonably well simulated by all three data sets (NCEP, H3A2, and H3B2). Small peak centered in March is underestimated by all three datasets. But overall, patter is well followed by all three data sets. In the seasonal analysis, it is seen that NCEP dataset underestimates in all seasons with average seasonal precipitation of7% with respect to observed. H3A2 dataset overestimates in all season with average precipitation of 6% except in spring where it underestimates, with an average precipitation of 0.04%. H3B2 underestimates in winter and spring and overestimates in summer and autumn. On the whole, H3B2 underestimate with an average precipitation of 0.89%. In case of predicting mean annual values for the validation period, NCEP underestimates a 5.7% mean annual

62

precipitation as compared to observed. However, H3A2 and H3B2 overestimate 4 and 1.2% respectively. In OPP basin (Figure 6-6 (b)) mean monthly precipitation is not well simulated by the SDSM as in TPP basin though the patter is well followed by all three data sets. The NCEP predicts much better than other two data set in case of peak months. The NCEP, H3A2 and H3B2 datasets underestimate in all seasons with an average seasonal precipitation of 6.9, 11.8, and 12%, and as do in case of predicting mean annual precipitation, with mean annual precipitation of 6.6, 13, and 15%, respectively. As for the whole basin,Figure 6-6 (c) shows that the mean monthly precipitation is well simulated in Monsoon season by all three data sets. On the other hand, simulated small peak (March) by all three dataset is well underestimated. The NCEP and H3B2 data sets underestimate with an average monthly precipitation of 5.5 and 2.8%, respectively, and H3A2 overestimate with mean monthly precipitation of 0.9%. The NCEP, H3A2 and H3B2 data sets clearly underestimate with an average seasonal precipitation of 6.5, 2.3, and 6.2% respectively. The mean annual precipitation predicted for the period of validation by NCEP, H3A2 and H3B2 data sets are all underestimated with an average precipitation of 6, 3.45, and 5.75% respectively. On the whole, the SDSM plays reasonably well in getting the variation of observed precipitation pattern in case of mean monthly, seasonal, and annual precipitation. These results are also comparable with these studies (Chu et al., 2010; Huang et al., 2011)

63

35

a Max Temperature (⁰⁰C)

30 25

Obs NCEP

20

H3A2

15

H3B2 Obs

10

NCEP H3A2

5

Ann

Aut

Sum

Spr

Win

Dec

Nov

Oct

Sep

Aug

Jun

May

Apr

Mar

b

Max Temperature (⁰⁰C)

30

Feb

Jan 35

Jul

H3B2

0

25

Obs NCEP H3A2 H3B2 Obs NCEP H3A2 H3B2

20 15 10 5

Sep

Oct

Nov

Dec

Win

Spr

Sum

Aut

Ann

Sep

Oct

Nov

Dec

Win

Spr

Sum

Aut

Ann

Aug

Jul

Jun

May

Apr

Mar

Feb

Jan

0

Max Temperature (⁰⁰C)

35 30

c Obs NCEP H3A2 H3B2 Obs NCEP H3A2 H3B2

25 20 15 10 5

Aug

Jul

Jun

May

Apr

Mar

Feb

Jan

0

Figure 6-4 Observed vs. simulated mean monthly, seasonal and annual Tmax for (a) TPP basin (b) OPP basin, and (c) whole Jhelum basin, during validation (1991-2000)

64

Min temperature (⁰⁰C)

20

a

15

20

Ann

Aut

Sum

Spr

Win

Obs NCEP H3A2 H3B2 Obs NCEP H3A2 H3B2

b 15

Min temperature (⁰⁰C)

Dec

Nov

Oct

Jun

May

Apr

Mar

Feb

Jan

0

Sep

5

Aug

10

Jul

Obs NCEP H3A2 H3B2 Obs NCEP H3A2 H3B2

10 5

Sep

Oct

Nov

Dec

Win

Spr

Sum

Aut

Ann

Sep

Oct

Nov

Dec

Win

Spr

Sum

Aut

Ann

Aug

Jul

Jun

Apr

May

-5

Mar

Feb

Jan

0

-10 20

c Min temperature (⁰⁰C)

15

Obs NCEP H3A2 H3B2 Obs NCEP H3A2 H3B2

10 5

Aug

Jul

Jun

May

Apr

Mar

Feb

Jan

0

-5

Figure 6-5 Observed vs. simulated mean monthly, mean seasonal and mean annual Tmin for (a) OPP basin, (b) TPP basin, and (c) whole Jhelum basin, during validation (1991-2000)

65

Precipitation (mm)

300

1600

Obs NCEP H3A2 H3B2 Obs NCEP H3A2 H3B2

a

250 200 150 100

1400 1200 1000 800 600 400

50

200

Precipitation (mm)

300

Ann

Aut

Sum

Spr

Win

Dec

Nov

Oct

Sep

Aug

Jul

Jun

May

Apr

Mar

Feb

0 Jan

0

1400

Obs NCEP H3A2 H3B2 Obs NCEP H3A2 H3B2

b

250 200 150

1200 1000 800 600

100

400

50

200

Precipitation (mm)

300

1600

Obs NCEP H3A2 H3B2 Obs NCEP H3A2 H3B2

c

250 200 150

Ann

Aut

Sum

Spr

Win

Dec

Nov

Oct

Sep

Aug

Jul

Jun

May

Apr

Mar

Feb

0 Jan

0

1400 1200 1000 800 600

100

400 50

200 Ann

Aut

Sum

Spr

Win

Dec

Nov

Oct

Sep

Aug

Jul

Jun

May

Apr

Mar

Feb

0 Jan

0

Figure 6-6 Observed vs. simulated mean monthly, mean seasonal and mean annual precipitation for (a) TPP basin, (b) OPP basin, and (c) whole Jhelum basin, during validation (1991-2000) Spatial comparison of observed and downscaled data To see the capability of SDSM spatially, three groups of maps (Figure 6-7) for each variable (Tmax, Tmin and precipitation) are built by converting the point into raster data by ArcGis using the Inverse Distance Weighted (IDW) interpolation method. Each group has four maps; one is produced by using observed data and other three by simulated data from NCEP, H3A2 and H3B2 predictors. It is seen that the spatial variation in observed mean annual 66

temperature and precipitation are well reflected by all three downscaled data sets as shown inFigure 6-7. Spatial distributions of temperature and precipitation show that the temperature decreases from South to North according to elevated area in the basin and precipitation decreases from West to East. It is seen that spatial variability is reasonably well simulated by NCEP data than H3A2 and H3B2, and also in TPP basin than OPP basin. It is seen that the overall simulated results from NCEP data are better than H3A2 and H3B2 data. However, the results from H3A2 and H3B2 are also comparable with NCEP dataset. Thus, it is concluded that the SDSM has good credibility to simulate mean monthly, season and annual Tmax, Tmin, and precipitationrather than daily data in Upper River Jhelum basin. So, the calibrated SDSM with NCEP data can be used for H3A2 and H3B2 scenarios to predict the future changes in Tmax, Tmin, and precipitation with respect to baseline period.

(A)

67

(B)

(C) Figure 6-7 Spatial distributions of observed against downscaled mean annual (A) Tmax, (B) Tmin, and (C) precipitation for validation (1991-2000), in the Jhelum River basin 68

6.6 Downscaling of temperature and precipitation 6.6.1 Temporal future changes After satisfactory results of calibration and validation, the Tmax, Tmin, and precipitation are simulated for the period of 2020s (2011-2040), 2050s (2041-2070) and 2080s (2071-2099) usingH3A2 and H3B2 scenarios. Then, the simulated monthly, seasonal and annual Tmax, Tmin and precipitation data is compared with the baseline to analyze the future changes in 2020s, 2050s and 2080s (Chu et al., 2010; Huang et al., 2011) for each site. In this study, the seasonal and annual changes are presented in Figure 6-8. Some weather stations (e.g. Gulmarg and Qazigund) are lack of data starting from 1961 to 1968. For these stations the data projected by SDSM using NCEP predictors is used to get future changes. Temperature Figure 6-8(A) shows the mean annual and seasonal changes of Tmax in the TPP basin, OPP basin, and whole Jhelum basin under the H3A2 and H3B2 scenarios, in 2020s, 2050s and 2080s. The mean annual Tmax under H3A2 scenario is projected to increase by 0.3°C (2020s), 0.74°C (2050s) and 1.2°C (2080s) in the TPP basin, 0.14°C (2020s), 0.38°C (2050s), and 0.74°C (2080s) in the OPP basin, and 0.24 (2020s), 0.60 (2050s) and 1.02°C in whole Jhelum basin. Under H3B2 scenario, the mean annual Tmax changes in 2020s, 2050s, and 2080s would be 0.34°C, 0.58°C ,and 0.89°C in the TPP basin, 0.15°C, 0.31°C and 0.44°C in the OPP basin, and 0.26°C, 0.48°C and 0.74°C in the whole basin, respectively. The overall mean annual changes in Tmax in the Upper Jhelum basin show an increasing trend under both H3A2 and H3B2 scenarios. However, the H3A2 predicted more distinct changes than H3B2. The seasonal future changes of Tmax are projected different in different seasons. All seasons give positive (increasing) changes under both H3A2 and H3B2 scenarios in the future periods (2020s, 2050s and 2080s) except in winter of the OPP basin. Under H3A2 scenario, in the TPP basin and OPP basin, the most effected seasons are the spring and autumn, with a projected increase of 1.7°C and1.14°C in 2080s, respectively. As for whole basin, the spring (1.4°C in 2080s) has distinct change as compared to other seasons. The same kinds of future changes are found in Tmax under H3B2 scenario but with relatively less in magnitude than H3A2. Figure 6-8(B) shows the mean annual and seasonal changes of Tmin in the TPP basin, OPP basin and whole basin under the H3A2 and H3B2 scenarios in 2020s, 2050s and 2080s. Under H3A2 scenario,the changes in mean annual Tmin are predicted as 0.12 (2020s), 0.54(2050s), and 0.87°C (2080s) in the TPP basin; -0.03 (2020s), 0.1 (2050s) and 0.31°C (2080s) in the OPP basin; and 0.04 (2020s), 0.35 (2050s), and 0.653°C in the whole basin. Under H3B2 scenario, the same kind of mean annual patter is seen but with less in magnitude than H3A2. In case of seasonal changes, the spring and autumn are affected more in the TPP and OPP basins, respectively, the same as in case of Tmax. The summer shows almost decreasing trend in Tmin in 2020s and 2050s, and a little increase in 2080s under both,H3A2 and H3B2 scenarios. The future changes in Tmin are predicted to be more distinct under H3A2 than H3B2, as do Tmax.

69

Precipitation Figure 6-8(C) presents the percentage changes in mean annual and seasonal precipitation in the TPP, OPP and whole basin under H3A2 and H3B2 scenario. The mean annual precipitation is simulated to increase (about 1 to 3%) in the TPP basin and decrease (about 2 to 5%) in OPP basin under both scenarios, with overall decrease in the whole basin. In case of seasonal changes, in the TPP basin, the summer (monsoon season in basin) and autumn seasons show a definite increase about 10 to 13% and 1 to 6%, respectively, under both scenarios in all three future periods. Under H3B2 scenario, the summer precipitation presents increase in 2020s and 2050s but decrease in 2080s. In winter the precipitation is projected to decrease by 7 to 12% and 5 to 8% under H3A2 and H3B2, respectively, in the TPP basin. On the whole, in the TPP basin, the summer and autumn could receivemore precipitation in the future with respect to baseline, and the winter as well as the spring could receive lesser in the future relative to baseline. In OPP basin, only autumn presents definite increase in precipitation by 5 to 8% and 8 to 12% in all three periods under H3A2 and H3B2, respectively. In the summer, the changes are positive in 2020s and 2050s but negative in 2080s. The spring is projected to decrease about 11 to 15% and 9 to 12% under H3A2 and H3B2, respectively. It is observed that the peak season (spring) in the OPP basin is going toward dryness in future. As a whole basin, the summer and autumn would be wetter than before, and the winter and spring would be dryer in the future as compared to baseline period under both, H3A2 and H3B2, scenarios.

70

1.5 1 0.5 0 Sum

Aut

2020s 2050s 2080s

A2-OPPB

0.8 0.6 0.4 0.2 0

1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0

2020s 2050s 2080s

B2-TPPB

1 0.8 0.6 0.4 0.2

Ann

Win

Spr

Sum

Aut

Win

1 Max Temperature (°C)

Max Temperature (°C) Max Temperature (°C)

Spr

1

-0.2

1.2

0 Win

1.2

1.4 Max Temperature (°C)

2020s 2050s 2080s

A2-TPPB

Spr

Sum

Aut

Ann 2020s 2050s 2080s

B2-OPPB

0.8 0.6 0.4 0.2 0

Ann

Win

Spr

Sum

Aut

Ann

-0.2 2020s 2050s 2080s

A2-Whole Basin

1 Max Temperature (°C)

Max Temperature (°C)

2

2020s 2050s 2080s

B2-Whole Basin

0.8 0.6 0.4 0.2 0

Win

Spr

Sum

Aut

Ann

Win

(A)

71

Spr

Sum

Aut

Ann

A2_TPPB

1

1

2020s 2050s 2080s

Min temperature (°C)

Min temperature (°C)

1.5

0.5 0

-0.5 Spr

Sum

Aut

0

Ann 0.6

A2_OPPB 2020s 2050s 2080s

0.5

Win

Min temperature (°C)

Min temperature (°C)

0.5

-0.5 Win

1

0

-0.5

Spr

Sum

Aut

Ann

Aut

Ann

Aut

Ann

B2_OPPB 2020s 2050s 2080s

0.4 0.2 0 -0.2 -0.4 -0.6

Win

Spr

Sum

Aut

Ann

Win

1.5

0.8 A2_whole Basin

1

2020s 2050s 2080s

Min temperature (°C)

Min temperature (°C)

2020s 2050s 2080s

B2-TPPB

0.5 0

-0.5

0.6

Spr

Sum

B2_whole Basin 2020s 2050s 2080s

0.4 0.2 0 -0.2 -0.4

Win

Spr

Sum

Aut

Ann

Win

(B)

72

Spr

Sum

2020s 2050s 2080s

A2_TPPB

10 5 0 -5

15 Precipitation (%)

Precipitation (%)

15

-10

10

0 -5 Win

Sum

Aut

15

A2_OPPB

5 0 -5 2020s 2050s 2080s

-15

Spr

Sum

-20 Spr

Sum

Aut

B2_OPPB

5 0 -5 -10

Ann

Win

Precipitation (%)

0 -5 2020s 2050s 2080s

-15 Spr

Sum

2020s

B2_Whole Basin

5

Win

Spr

2020s 2050s 2080s Aut Ann

15

A2_Whole Basin

-10

Sum

Aut

Ann

10

-15 Win

Aut

Ann

Precipitation (%)

Precipitation (%)

Spr

-10

Precipitation (%)

2080s

-10 Win

10

2050s

5

-15

10

2020s

B2_TPPB

2050s

10

2080s 5 0 -5 -10

Ann

Win

Spr

Sum

Aut

Ann

(C) Figure 6-8 Future changes in mean season and annual (A) Tmax, (B) Tmin, and (C) precipitation in 2020s, 2050s and 2080s with respect to 1961-1990 under H3A2 and H3B2 emission scenarios, in the TPP basin, OPP basin, and whole Jhelum basin 6.6.2

Spatial future changes

In this study, the mean annual changes in Tmax, Tmin, and precipitation are also analyzed spatially in Upper Jhelum River basin under H3A2 and H3B2 scenarios with help of ArcGis 9.3 shown in Figure 6-9. This shows remarkable differences in spatial changes in basin. In this figure the Tx, Tm and P are used for max temperature, min temperature and precipitation, respectively. Figure 6-9(A) shows the spatial distribution of mean annual changes of Tmax in the 2020s, 2050s and 2080s under H3A2 and H3B2 scenarios. It is seen that the North-West part of the basin is projected to be the most effected in all three future periods, with increasing in Tmax. 73

The South-East part of the basin is predicted to be the least effected (small increase). It is seen that the positive changes are almost increasing when moving from South-East to NorthWest side under both scenarios. Spatial distribution of changes in mean annual Tmin are presented in Figure 6-9(B) which shows that in some patches of basin Tmin is projected to decrease compared with base period, but the most parts of the basin shows increase in Tmin in all three future periods under both scenarios. Figure 6-9(C) gives spatial distribution of percentage changes in mean annual precipitation compared with base period under H3A2 and H3B2 scenarios. As whole the changes are spread between -13 to 16% and -10 to 15% over whole basin under H3A2 and H3B2, respectively, in all three future periods. The simulated decrease in precipitation is reported from the East part of the basin, and increase is reported from the West side of the basin. The increase in precipitation in the in the Jhelum basin (especially in Kunhar basin) might be due to increase in temperature over that area as in study by Mirza and Ahmad, (2005). The same is reported in 4th Assessment Report of IPCC (IPCC, 2007) that area-mean precipitation over Southeast Asia (Figure 6-10) increases in most MMD (Multi Model Data) model simulations, with a median change of about 7% in all seasons, but the projected seasonal changes vary strongly within the region. It is seen that almost half of the basin shows decreasing precipitation in 2020s, but in 2080s the more than half of the basin is projected to decrease in precipitation under both scenarios. Both scenarios present a similar kind of spatial distribution pattern of mean annual precipitation changes in all three future periods, but the changes (positive or negative) are higher reported by H3A2 than H3B2.

(A)

74

(B)

(C) Figure 6-9 Spatial changes of mean annual (A) Tmax in Celsius, (B) Tmin in Celsius, (C) precipitation in percentage in the 2020s, 2050s and 2080s relative to 1961-1990 under H3A2 and H3B2 scenarios, in the Jhelum River basin 75

6.7

Conclusion

The primary tools to examine the climate changes are Global Climate Models. However, these have very course resolution and cannot give the realist presentation of regional or local scale. To construct a bridge between local and large scale variable, many statistical downscaling techniques have been developed and used throughout the world. In this study, Statistical Downscaling Model (SDSM) is applied to downscale the max temperature, min temperature and precipitation in sub-basins—TPP and OPP—and whole Upper Jhelum River basin under H3A2 and H3B2 scenarios. Downscaling of these variables is very important to study the impacts of climate change on local scale hydrology of this basin in the future. The big challenge of using SDSM model is the selection of suitable predictors for this complex mountainous area having a strong influence of monsoon season. For this purpose, a more quantitative approach is used by which the most prominent predictors can be ranked for calibration process reducing the effect of multiple co-linearity. It is concluded that the near surface large-scale (atmospheric) variables are the most suitable for downscaling of temperature. The most prominent predictor for the present study is the temp (large-scale mean temperature at 2 m height) in both sub-basins. As for precipitation, in the OPP basin (North-East and South-East part), the surface meridional velocity and surface vorticity at 500 hpa are the most influential atmospheric variables. In the TPP basin (South-West part), local precipitation is mostly effected by surface specific humidity and surface meridional velocity at 500 hpa. During the calibration and validation of SDSM, it is seen that the SDSM shows better capability to simulate temperature (Tmax and Tmin) in all three formats (daily, monthly, and seasonal), with R2 ranging from 0.71 to 0.98—daily to seasonal. As for precipitation, SDSM downscale good results in case of monthly and seasonal, with mean R2 ranging 0.42-0.62 (monthly to seasonal), but in daily format, it gives very poor results, with R2 ranging between 0.8-0.15. It is concluded that, in both (temperature and precipitation), the seasonal time series could be best simulated by SDSM than daily or monthly in the Jhelum River basin. The simulated changes in mean annual and seasonal temperate and precipitation for the periods of 2020s, 2050s and 2080s compared with base period (1961-1990) show obvious different patterns under H3A2 and H3B2 scenarios, in the TPP and OPP basin. The main finding is that the mean annual Tmax and Tmin are projected to increase in both parts of the basin, TPP and OPP, under both scenarios and in all three future periods—2020s, 2050s, and 2080s. This increase in temperature is simulated to be higher under H3A2 scenario than H3B2 in the both sub-basins and higher in the TPP basin than OPP basin. Moreover, the increasing trend in mean annual temperature is found in the future from 2020s to 2080s. As for seasonal changes in mean annual temperature (Tmax and Tmin), the most increase is simulated in the spring under both, H3A2 and H3B2 scenario in the whole basin. However, the spring in TPP basin and autumn in OPP basin are most affected seasons in case of rise in temperature. The mean annual precipitation is predicted to increase (1 to 3%) in the TPP basin and decrease (2 to 5%) in the OPP basin under both scenarios, with an overall decrease in the whole basin. Seasonal changes are different in all seasons in both parts of the basin. The summer (with 10 to 13% rise in precipitation) and winter (with 5 to 12% decrease in precipitation) are the most affected seasons in the TPP basin under both scenarios and in all 76

three future periods. In the OPP basin, the autumn shows a 5 to 12% increase in precipitation, and the spring—peak season —shows 9 to15% decreases under both scenarios. As for the whole basin, the summer and autumn are projected to receive more precipitation, and the winter as well as the spring could receive lesser amount of precipitation in the future, as compared to baseline period. The spatial distribution of mean annual Tmax shows a rise, in almost all part of the basin, in the future periods relative to baseline. However, the North-West parts are projected to face higher increase than South-East parts of the basin under both scenarios. The min temperature is projected to decrease in some patches of basin but the major parts of the basin show rise in min temperature with respect to the baseline under both scenarios. In case of the precipitation, the percentage changes are spread between -13 to 16% and -10 to 15% over whole basin under H3A2 and H3B2 respectively. It is seen that almost half of basin shows decreasing precipitation in 2020s, but in 2080s the most parts of the basin are projected to decrease in precipitation under both scenarios. Both scenarios present a similar kind of spatial distribution patterns of mean annual Tmax, Tmin, and precipitation changes in all three future periods but with different magnitude. However, these changes reported by H3A2 are higher than H3B2.

77

7 7.1

EVALUATION OF SUB-MODELS OF SDSM

Introduction

The increasing concentration of greenhouse gases in the atmosphere due to human activities such as land use changes and the dependence upon fossil fuels has resulted in global warming and a global energy imbalance (Wentz et al., 2007; Chu et al., 2010; Huang et al., 2011). According to the Fourth Assessment Report (4AR) of the Intergovernmental Panel on Climate Change (IPCC), a 0.74°C rise in the global mean surface temperature was reported in the last hundred years, specifically between 1906 and 2005. Significant increase has been reported in the last 50 years, with an increasing rate of 0.13°C every 10 years, and the global mean surface temperature is projected to increase approximately from 1.1 to 6.4°C during the 21st century (IPCC, 2007). This increased global warming can impact the hydrological cycle, affecting water resources, public health, industrial and municipal water demands, water energy exploitation and the ecosystem(Chu et al., 2010; Zhang et al., 2011). To date, the main tools to predict the variability and changes in climate variables, on global and continental levels, are Global Climate Models that are also called General Circulation Models (GCMs). These advanced and numerical based coupled models interpret global systems such as sea-ice, the oceans and atmosphere (Fowler et al., 2007). Although these models are very helpful in the investigation and predictions regarding future changes in climate, the outputs of these model are based on a large grid scale (250 to 600 km) (Gebremeskel et al., 2005). Because of their coarse resolution, the outputs cannot be used successfully to investigate the environmental and hydrological impacts of climate change on a regional scale (Hay et al., 2000; Wilby et al., 2000). The most important tool to create a bridge between a regional/local scale and GCM scales (a coarse scale), is downscaling (Wetterhall et al., 2006). The local and regional scales are defined as 0-50 km and 50×50 km respectively (Xu, 1999). Many downscaling methods have been developed during the last two decades which consider the temporal and spatial mismatch between regional scales and coarse scales. In the beginning, these methods were mostly applied in Europe and in the United States (Gellens and Roulin, 1998; Wilby et al., 1999; Huth, 2002; Hay and Clark, 2003; Diaz-Nieto and Wilby, 2005; Wilby et al., 2006; Salzmann et al., 2007; Wang and Zhang, 2008; Wetterhall et al., 2009; Yang et al., 2010; Sunyer et al., 2011). These are now implemented throughout the world to investigate climate change, utilizing the outputs of GCMs to downscale climate change impact at regional and local levels (Akhtar et al., 2008; Elshamy et al., 2009; Huang et al., 2011; Kannan and Ghosh, 2011). Generally, there are two main approaches—statistical and dynamical—for downscaling outputs of a GCM. In dynamical downscaling, a high resolution numerical model, or Regional Climate Model (RCM), with a resolution of about 5 to 50 km (Chu et al., 2010) is coupled with the GCM. The RCM drives the lateral and large-scale boundary condition from the GCM and provides detailed information or high resolution outputs at the regional level. The GCM responds to large scale forces such as greenhouse gases, atmospheric circulation, and oceanic circulation etc. The RCM, on the other hand, simulates small scale climatic variables such as extreme climate events, orographic precipitation, and regional scale anomalies. Since the RCM is dependent on the GCM’s boundary conditions, it is susceptible to any systematic errors which belong to the GCM’s driving fields. The skill of the RCM is strongly dependent on both, the GCM’s driving forces and information about regional scale forcing (land use data, land-sea 78

contrast and orography etc.). There must be a strong co-ordination between the global and regional climate modeling groups to ensure that the appropriate data is available. The RCMs are computationally intensive, depending upon the resolution and domain size. This confines the number of experiments for climate scenarios (Wilby and Wigley, 1997; Hay and Clark, 2003; Fowler et al., 2007). Statistical downscaling (SD) methods produce empirical/statistical links among the large-scale and local-scale variables. These methods are faster and computationally inexpensive, and consequently offer approaches that have been rapidly adopted by a wider community of scientists (Wilby et al., 2000). SD methods are much simpler than DD methods to downscale the outputs of a GCM. Using SD methods, global-scale climate variables such as mean sea level pressure, zonal wind, temperature, geo-potential height etc. are linked with local-scale variables (regional scale variables) such as observed temperature and precipitation, and this is done by producing some statistical/empirical relationships (Wetterhall et al., 2006). SD is not only useful in numerical weather prediction and synoptic climatology, but is also applied for a wide range of climate applications. The main advantage of SD is that it provides local scale information, which is very useful in climate change impact assessment studies (Giorgi et al., 2001). On the downside, the main disadvantage of this approach is that it requires long historical meteorological weather station data to construct an appropriate link with large scale variables. The main assumption of SD is that the empirical relationship between larger and small scales is temporally stationary (Hay and Clark, 2003). DD is a good alternative for SD in the case of basins which have no historical data (Benestad et al., 2008). There are three main types of SD: 1) Stochastic Weather Generator, 2) Regression (Yaoming et al., 2004), and 3) Weather typing. These are reviewed in more detail by (Hewitson and Crane, 1996; Wilby and Wigley, 1997; Wilby et al., 2002; Fowler et al., 2007). Although SD is classified into different types, the main concept of this method is to provide some kind of regression relationship. There are different kinds of regression methods in use: Principal Component Analysis, Artificial Neural Networks, Multiple Linear Regression and Canonical Correlation Analysis (Schoof and Pryor, 2001; Huth, 2002). In impact assessment studies, these methods are widely used to develop the relationship between predictors and predictand, and to assess hydrologic responses under different climate change scenarios (Chu et al., 2010). To date, many statistical models have been developed and are available. SDSM is being used widely throughout the world (Huang et al., 2011) to downscale the most important climate variables such as temperature, precipitation and evaporation etc. for assessing hydrologic responses in climate change scenarios. This SDSM model is developed through a combination of Multiple Linear Regression and the Stochastic Weather Generator (Diaz-Nieto and Wilby, 2005; Gagnon et al., 2005; Gebremeskel et al., 2005; Wilby et al., 2006). Some studies (Tripathi et al., 2006; Anandhi et al., 2007; Akhtar et al., 2008; Ghosh and Mujumdar, 2008; Ashiq et al., 2010; Goyal and Ojha, 2010; Opitz-Stapleton and Gangopadhyay, 2010) have been conducted in the South Asian region using SD and DD methods, with the majority of these studies carried out in Indian river basins. A couple of studies, one by Akhtar et al. (2008) and anotherby Ashiq et al. (2010) using downscalingwere conducted in Pakistan. Akhtar et al. (2008) implemented PRECIS (Providing REgional Climate for Impact Studies) RCM, and a simple downscaling method called “Delta Change” for downscaling mean precipitation and temperature for hydrological modelling under IPCC 79

emission scenario A2 for the period of 2071-2100 in the region of the Hindukush-KarakorumHimalalya (HKH) ranges located in the Indus River basin, Pakistan. Ashiq et al. (2010) evaluated the monthly precipitation outputs of PRECIS, run by Akhtar et al. (2008), and interpolated them from coarse resolution (50×50 km) to a fine scale (250×250 m). Seven different interpolation methods are used to interpolate monthly precipitation and are validated with the observed precipitation. This study was conducted in the northwestern Himalayan Mountains and the upper Indus plains of Pakistan, which also cover a small part of the Jhelum basin located in Pakistan. Through their study they concluded that the systematic errors associated with RCM cannot be reduced by interpolation methods. Based on the review of literature and as far as authors are aware, it was found that not a single study has been reported that uses SDSM in Jhelum River basin, or even throughout South Asia, although the SDSM model has been proven to be skillful and is widely used for the downscaling of precipitation and temperature elsewhere in the world (Liu et al., 2011). There are three sub-models: monthly, seasonal, and annual to develop SDSM, but only the monthly sub-model is used in previous studies such as (Wilby et al., 2002; Gagnon et al., 2005), to develop SDSM for downscaling local climate variables. Authors are not aware of any study evaluating these sub-models thoughout the world. Although, some studies (mentioned above) have been carried out in or around the Jhelum River basin in Pakistan using the DD approach and interpolation methods for downscaling temperature and precipitation, Akhtar et al. (2008) conclude that PRECIS has several uncertainty sources, and Ashiq et al. (2010) state that interpolation methods are not able to improve the systimatic errors inherent to PRECIS. Moreover, in these studies mentioned above, the temperature and precipitation data is only projected for the future period of 2071-2100 not for entire century (2011-2100) as done in this study. So, the present study is based on the following objectives: 1) to investigate the applicability of SDSM for downscaling max temperature, min temperature and precipitation in the Jhelum River basin, 2) to evaluate annual and monthly developed sub-models of SDSM, 3) to investigate future climate changes under IPCC emission scenarios (A2 and B2) for the 21st century which will be used for assessing the impact of climate change on water resources and hydropower in the Jhelum River basin, Pakistan as part of an on-going research project. 7.2

Data description

Observed daily and monthly historical data of 14 weather stations in the basin for max temperature (Tmax), min temperature (Tmin) and precipitation was collected from the Water and Power Development Authority of Pakistan (WAPDA), Pakistan Meteorological Department (PMD), and the India Meteorological Department (IMD). The daily data of Kupwara, Srinagar, Gulmarg and Qazigund weather stations was obtained from IMD. Daily precipitation data for Srinagar and Qazigund was also obtained from the National Climate Data Center (NCDC-NOAA) for the period of 1961-1970. The basic information (geographic and climatic, for the period of 1961–1990) and data availability for all these meteorological stations are presented in Table 7-1. Missing data was filled with multilevel regression (multiple imputation method) using winMice software version 0.1 (Jacobusse, 2005). The 26 predictors of NCEP and HadCM3 (H3A2 and H3B2) were obtained from a Canadian website: http://www.cics.uvic.ca/scenarios/index.cgi?Scenarios, for the periods of 1961-2001 80

and 1961-2099 respectively. H3A2 and H3B2 are respectively the IPCC emission scenarios A2 and B2 of HadCM3. These predictors are especially processed for the SDSM model. The NCEP predictors (2.5×2.5°) are first interpolated to HadCM3 grid resolution (2.5×3.75°) to eliminate spatial differences. Subsequently, the NCEP and HadCM3 predictors were normalized by utilizing long-term mean and standard deviations of 1961-1990. The normalized predictors are only available for HadCM3 and CGCM2 in such a form that can be downloaded according to the coordinates of the study area and used directly for SDSM. HadCM3 was selected because of two reasons: 1) it showed better agreement during evaluation of various GCMs (CGCM2, CSISRO, CCSR/NIES, and HadCM3), with observed data in the Jhelum basin (results not included in this paper), 2) it is mostly applied around the Jhelum basin such as in studies like (Akhtar et al., 2008; Chu et al., 2010; Huang et al., 2011).

81

Table 7-1 Geographic, climatic, and data availability information of weather stations used in the present study No.

82

1 2 3 4 5 6 7 8 9 10 11 12 13 14

Station

Jhelum Kotli Plandri Rawlakot Murree Garidopatta Muzaffarabad Balakot Naran Kupwara Astore Srinagar Gulmarg Qazigund Whole Basin

Location Lat (deg.) 32.94 33.50 33.72 33.87 33.91 34.22 34.37 34.55 34.90 34.51 35.34 34.08 34.00 33.58

Long (deg.) 73.74 73.90 73.71 73.68 73.38 73.62 73.48 73.35 73.65 74.25 74.90 74.83 74.33 75.08

Precipitation (mm/yr) Elev (m, MSL) 287 614 1402 1,676 2,213 814 702 995 2,362 1,609 2,168 1,587 2,705 1,690

Tmax (°C)

Tmin (°C)

Tmean (°C)

860 1,249 1,459 1,398 1,765 1,586 1,418 1,731 1,217 1,314 564 764 1,574 1,395

30.51 28.41 22.73 21.87 16.57 26.06 27.34 25.04 11.14 19.64 15.48 19.76 11.73 19.11

16.54 15.75 11.30 10.25 8.97 12.58 13.54 12.01 1.15 6.30 4.06 7.37 1.98 6.57

23.53 22.08 17.01 16.06 12.77 19.32 20.44 18.53 6.14 12.93 9.77 13.52 6.81 12.80

1,202

19.79

7.68

13.72

N/A Not available; Tmean Mean temperature; Elev Elevation; Temperature* Tmax and Tmin

Precipitation

Temperature*

Daily

Monthly

Daily

Monthly

1970-00 1970-00 1962-00 N/A 1970-00 1970-00 1961-00 1970-00 N/A 1976-00 1953-00 1961-00 1969-00 1961-00

1950-09 1952-09 1962-09 1960-09 1960-09 1955-09 1955-09 1960-09 1961-09 1976-09 1953-09 1950-09 1969-09 1961-09

1970-00 1970-00 1970-00 1970-00 1970-00 1970-00 1970-00 1970-00 1970-00 1976-00 1961-00 1969-00 1969-00 1969-00

1950-09 1952-09 1961-09 1961-09 1960-09 1955-09 1955-09 1960-09 1961-09 1976-09 1954-09 1950-09 1969-09 1969-09

7.3

Monthly to daily data conversion

The daily data for most of the meteorological stations is starting from 1970 and ending at 2009 but monthly data is starting from 1960 or earlier for most of the climate stations as shown in Table 7-1. Since for statistical downscaling and hydrological modeling, the daily rainfall and temperature data is needed for SDSM model and HEC-HMS model. For this purpose the data of climate stations which have only monthly data from 1961-1990 is converted into daily data by using the Modawec model. This model is used for generating the daily rainfall, maximum and minimum temperature from the monthly rainfall, max and min temperature as described in Figure 7-1. The main inputs for this model are mean monthly maximum and minimum temperature, number of wet days in each month, daily standard deviation (if available), extreme (highest and lowest) temperature values for each month for the specified period for which the data is to be generated and monthly rainfall. The precipitation generation model is based on the first order Markove chain and temperature generation model is of multi variant type. The model functions and equations used in this model are explained in more detail in this study (Liu et al., 2009) Since the only monthly rainfall, max and min temperature are available for the study area. So the other input data (like wet days) is extracted from the Climate Research Unit (CRU). The latest data version of CRU-TS3.0 (Mitchell and Jones, 2005) is used here. This dataset is constructed by the Climate Research Unit at the University of East Anglia. Presently the dataset is hold by the British Atmospheric Data Centre (BADC) for the period of 1901-2006 but originally produced by CRU. This data is high resolution with grid distribution of 0.5° × 0.5° which covers the whole world by 360 rows and 750 columns. The CRU data is extracted in such way that it covers completely the whole study area. Almost 30 grids cover the whole Mangla watershed as shown in Figure 7-2. Then the CRU data is separated and arranged according to the corresponding climate station. Then all input files (monthly rainfall, max and min temperature, wet days etc.) are put into the Modawec to generate daily time series for the period of 1961-69. Monthly-Prec Daily-Tmin Monthly-Tmin Monthly-Tmax

MODAWEC

Daily-Tmax

Monthly-Wet days Daily-Prec  

If Std. of Tmax and Tmin of each month or Long-term Max and Min temp. extremes

Figure 7-1 Discription of MODAWEC model

83

Figure 7-2 CRU grids representation covering study area 7.4 Methodology 7.4.1 Description of SDSM SDSM developed by Wilby et al. (2002) is a hybrid of Multiple Linear Regression (MLR) and the Stochastic Weather Generator (SWG).MLR establishes a statistical/empirical relationship between NCEP, large-scale variables and local-scale variables, and produces some regression parameters. These calibrated parameters, along with NCEP and GCM predictors, are then used by SWG to simulate up to 100 daily time series in order to create a better correlation with the observed time series(Wilby et al., 2002; Liu et al., 2009). In SDSM, some suitable predictors from the atmospheric predictors are selected through a multiple linear regression model, utilizing the combination of the correlation matrix, partial correlation, P-value, histograms, and scatter plots. Multiple co-linearity must be considered during the selection of predictors. There are two kinds of optimization methods: 1) Ordinary Least Squares (OLS) and 2) Dual Simplex (DS). The OLS produces comparable results with DS and is also faster than DS (Huang et al., 2011). There are three kinds of sub-models—monthly, seasonal, and annual— that comprise the statistical/empirical relationship between the regional scale variables (temperature and precipitation), and large-scale atmospheric variables. Annual sub-models drive the same kind of regression parameters for 12 months and the monthly sub-model represents 12 regression equations, giving different calibrated parameters for each of the 12 months. There are also two kinds of sub-models, conditional and unconditional; any of them can be used according to the local scale variables. The unconditional sub-model is used for independent or unconditional variables such as temperature. The conditional sub-model is used for variables such as precipitation and evaporation (Wilby et al., 2002; Chu et al., 2010). 84

Most of the time, precipitation data is not distributed normally, but in the case of temperature, the data is distributed normally. SDSM can transform the data to make it normal before using the data in regression equations (Khan et al., 2006). For example, Khan et al. (2006) and Huang et al. (2011) used the fourth root for precipitation to render it normal before using it in a regression equation. Two kinds of daily time series, namely 1) daily historical site data and 2) NCEP daily predictors, are used to develop SDSM. The outputs of this model are daily time series, which can be produced by forcing the NCEP or HadCM3 predictors (Huang et al., 2011). The mathematical details are presented in (Wilby et al., 1999). 7.4.2 Screening of predictors The screening of predictors is central to all statistical downscaling (Wilby et al., 2002; Huang et al., 2011). In this study, a combination of the correlation matrix, partial correlation, and Pvalue is used (Gagnon et al., 2005; Huang et al., 2011). To select the first and most suitable large-scale variable is relatively easy, but the selection of the second, third, fourth and so on is much more subjective. Therefore, a more quantitative procedure is applied for screening largescale variables for each local scale variable at each of the climate stations. The following steps describe the entire procedure along with an example where predictors are selected for precipitation at the Astore, northwest part of basin, (Table 7-4). 1. First, a correlation matrix between 26 NCEP predictors (Table 7-2) and the predictand is made, and then the predictors of high correlation coefficient (12 in this case) are selected out of the 26 and arranged in descending order. The first ranked predictor, having the highest correlation coefficient among all predictors, is selected and defined as a super predictor (SP). The correlation between them is defined as a super correlation coefficient (SCC). The predictor ncepp5_v is the super predictor for precipitation at Astore. 2. Following this, the absolute correlation coefficient between the predictor and predictand, the absolute correlation coefficient between individual predictors, absolute partial correlation and p-value are obtained by regressing (11 in this case) the remaining highly correlated predictors individually in the presence of SP (ncepp5_v), as shown in Table 7-4. 3. Then the predictors which have a P-value greater than α (0.05) are removed to render the results statistically significant, and the predictors which are highly correlated (0.5 in this study) are taken out in order to remove any multi-co-linearity. The correlation coefficient up to 0.7 between two predictors is acceptable (Pallant, 2007). 4. Percentage reduction in an absolute partial correlation (PRP) with respect to absolute correlation is calculated for each predictor using the following equation. The PRP values for precipitation at Astore are shown in Table 7-4’sfinal column. 𝑷𝑹𝑷 = (

𝑷. 𝒓 − 𝑹𝟏 ) 𝑹𝟏

7-1

Where PRP is the percentage reduction in partial correlation with respect to the correlation coefficient, P.r is the partial correlation coefficient and R1 is the correlation coefficient between the predictor and predictand. 5. The predictor which has a min PRP (ncepp8_v in this case) in partial correlation (Table 7-4) is selected as the second most suitable predictor. As a result, this predictor has almost no, or a very insignificant multi-co-linearity with the super predictor. 85

6. The 3rd, 4th and following predictors can be obtained by repeating step 2 to step 5. In the second repetition, there will be two super predictors. It is seen that mostly 1 to 3 predictors are enough to explain the predictand during calibration and without multico-linearity. Although some predictors such as ncepp5zh and ncepp8_z have a higher correlation coefficient (Table 7-4) than ncepp8_v, they do not have any significant effect on the performance parameters (explained variance and standard error) of the model during calibration. This is because these predictors might have a high co-linearity with SP. Conversely, the predictor selected through this process indicates a significant effect on the model’s performance parameters. Since Table 7-3 shows the screening of predictors for precipitation at Astore climate station located upper (North) part of basin, Table 7-3 shows the screening of predictors for precipitation at Jhelum station located lower (South) part of basin (Figure 7-3). Table 7-2 NCEP predictors used in the screening process No. 1 2 3

Predictor p_f p_u p_v

No. 14 15 16

Predictor r500 p8_f p8_u

Description 500 hPa relative humidity 850 hPa airflow strengh 850 hPazonal velocity

17 18 19 20 21 22 23

p8_v p8_z p8th p8zh r850 p500 p850

850 hPa meridional velocity 850 hPa vorticity 850 hPa wind direction 850 hPa divergence 850 hPa relative humidity 500 hPa geopotential height 850 hPa geopotential height

p5_z

Description Surface airflow strength Surface zonal velocity Surface meridional velocity Surface vorticity Surface wind direction Surface divergence Surface relative humidity 500 hPa airflow strengh 500 hPa zonal velocity 500 hPa meridional velocity 500 hPa vorticity

4 5 6 7 8 9 10

p_z p_th p_zh rhum p5_f p5_u p5_v

11

24

temp

p5th p5zh

500 hPa wind direction 500 hPa divergence

25 26

shum mslp

Mean temperature at 2m height Surface specific humidity Mean sea level pressure

12 13

Table 7-3 Screening of most effective predictors for precipitation at the Jhelum climate station SN 1 2 3 4 5 6 7

Predictor ncepshum ncepr850 ncepp_th ncepp500 ncepP_f ncepP_v nceptemp

R1 (%) 25.6 18.2 11.0 16.9 9.7 13.0 19.2

R2 (%)

P .r (%)

P-value

PRP (%)

30.1 60.3 78.2 46.3 10.7 82.4

10.9 6.4 9.7 5.2 6.6 8.4

0.00 0.00 0.01 0.31 0.40 0.00

40.10 41.8 42.60 46.39 49.23 56.25

86

Figure 7-3 NCEP selected predictors for calibration (1961-1990) of Tmax (red), Tmin (green) and precipitation (black) for SDSM’s sub-models in the Jhelum basin 7.4.3 Calibration and validation Based on the available observed daily data, two datasets, 1961-1990 and 1970-1990, are used for the calibration of Tmax, Tmin and precipitation. In this study, SDSM is developed with selected NCEP predictors using annual and monthly sub-models. The developed annual and monthly sub-models of SDSM are denoted by SDSM-A and SDSM-M respectively. While calibrating, the same predictors are used to calibrate both SDSM-A and SDSM-M. Both models are developed individually for each of the predictands (Tmax, Tmin, and precipitation) at each site. The conditional sub-model is used for Tmax and Tmin without any transformation and unconditional sub-model for precipitation with fourth root transformation. Optimization of the best fit is done by OLS. The explained variance and standard error are used to evaluate the performance of SDSM (Huang et al., 2011). With the developed models, Tmax, Tmin and precipitation are simulated for 1961-2000 using the NCEP, H3A2 and H3B2 predictors. A total of 20 ensembles are generated using the annual and monthly SDSMs and the mean of these ensembles is used in this study. The outputs of SDSM are compared with observed data by calculating the coefficient of determination (R²), root mean square error (RMSE), mean(µ), standard deviation (б), relative error in mean (RE_µ) and standard deviation (RE_ б) for temperature and precipitation for the periods of calibration and validation. R² and RMSE show the accuracy of the model in predicting data. µ andRE_µ are used to check that how well the model predicts the mean values, and б and RE_ б are used to observe the variability of data predicted by the model. The above performance indicators are 87

first calculated for each weather station and then the mean values of each indicator are obtained from all weather stations. As can be seen, many users of SDSM (Wilby et al., 2002; Dibike and Coulibaly, 2005; Fealy and Sweeney, 2007; Chu et al., 2010)plot the observed and simulated data in order to observe the variation and pattern of data. In this study, however, first the monthly mean simulated data (Tmax, Tmin and precipitation) computed by both models utilizing NCEP predictors, is plotted against the observed data for the calibration period. Following this, the mean monthly, seasonal and annual simulated data of both models utilizing NCEP and HadCM3 predictors is then compared graphically with the observed data for the validation period. However, only the graphical presentation of precipitation is shown in this study to take note of the pattern and variations because precipitation is a heterogeneous climate variable and is difficult to capture variations. In this study, bias correction (BC), which is discussed in detail below, is also applied to the downscaled data obtained from the two SDSMs using HadCM3 predictors, in order to obtain a more realistic and unbiased data of future climate. Before applying it on the future downscaled data, BC was first validated for the period of 1991-2000. For this purpose, the mean monthly biases are obtained from the period of 1981-1990 because these biases have to be adjusted for the validation period that is also of 10 year duration by utilizing downscaled data of SDSM and observed data at each site. These biases are then adjusted to the downscaled daily data by SDSM in the period of 1991-2000. The corrected downscaled data (Tmax, Tmin and precipitation) is compared with the observed data by calculating the above mentioned statistical indicators, and only the corrected downscaled precipitation is plotted for graphical visualization. After successful validation, BC is applied to the future downscaled data (Tmax, Tmin and precipitation) by both models. For this, more recent data, from 1980 to 2009 is used to calculate the 30-year mean monthly biases, and then these biases are adjusted to the downscaled daily time series according to their respective months. The reason of using a recent period is that during validation, the mean biases are obtained from two periods: 1971-1980 (offline) and 1981-1990, and applied to 1991-2000. It is seen that the results from a recent period (19811990) are better than those from an earlier period (1971-80), and this is quite an obvious finding. In the end, the changes in the 2020s (2011-2040), 2050s (2041-2070) and 2080s (20712099) are obtained with respect to the baseline (1961-1990).The period from 1961-1990 is taken as the baseline period in this study as this period has been utilized worldwide in most climate change studies (Huang et al., 2011). A 30-year period is considered long enough to define local climate because it is likely to have dry, wet, cool, and warm periods. The length of the period (30 years) is also recommended by the IPCC for use as a baseline period (Gebremeskel et al., 2005). 7.4.4 Bias correction The Bias correction (BC) approach is used to eliminate the biases from the daily time series of downscaled data (Salzmann et al., 2007). In this method, the biases are obtained by subtracting (in the case of temperature) the long-term monthly mean (30 years) of observed data, from the mean monthly simulated control data (downscaled data by SDSM for the period of 1980-2009), and dividing (in the case of precipitation) the long term observed monthly mean data with simulated control data. The biases are then adjusted with the future downscaled daily time

88

series according to their respective months. Equations 2 and 3 are used to de-bias daily temperature and precipitation data. ̅̅̅̅̅̅̅̅ ̅̅̅̅̅̅ 𝑻𝒅𝒆𝒃 = 𝑻𝑺𝑪𝑬𝑵 − (𝑻 𝑪𝑶𝑵𝑻 − 𝑻𝒐𝒃𝒔 ) 𝑷𝒅𝒆𝒃 = 𝑷𝑺𝑪𝑬𝑵 × (

7-2

̅̅̅̅̅̅ 𝑷𝒐𝒃𝒔 ) ̅̅̅̅̅̅̅̅ 𝑷 𝑪𝑶𝑵𝑻

7-3

Where Tdeb and Pdeb are the de-biased (corrected) daily time series of temperature and precipitation respectively for future periods. SCEN represents the scenario data downscaled by SDSM for future periods (e.g. 2011-2099), and CONT represents downscaled data by SDSM for the present period (e.g. 1980-2009). TSCEN and PSCEN are the daily time series of temperature ̅̅̅̅̅̅̅̅ and precipitation generated by SDSM for future periods respectively. ̅̅̅̅̅̅̅̅ TCONT and P CONT are the long-term mean monthly values for temperature and precipitation respectively for the control period simulated by SDSM. ̅̅̅̅̅ Tobs and ̅̅̅̅̅ Pobs represent the long-term mean monthly observed values for temperature and precipitation. The bar on T and P shows the long-term average. The frequency and intensity of precipitation are the two main factors affecting precipitation variability (Sharma et al., 2007). The application of this method of study is to correct the precipitation amount and not the frequency, and also to remove any systematic errors belonging to SDSM during downscaling. It is assumed that the frequency is accurately simulated by SDSM. 7.5 Results and discussion 7.5.1 Screening of predictors The selected first, second and third predictors for Tmax, Tmin and precipitation are shown in Table 7-4. It is observed that ncep temperature (temperature at 2m height) is the super predictor for both, local Tmax and Tmin. As for precipitation, there are two super predictors, Surface Specific Humidity (shum) and Meridional Wind Velocity at 500 hPa (P5_V) for almost all weather stations. Shum is the dominant predictor in the southwest part of the basin and P5_v in the rest of the area of the basin. These predictors also convey a physical relationship regarding the local scale Tmax, Tmin and precipitation. The monsoon from the Indian Ocean also enters into the country from the southwest from June to September (PWP, 2011). The predictors selected for Tmax, Tmin, and precipitation in this study are mostly the same ones used in several other studies (Wilby et al., 2002; Chu et al., 2010; Hashmi et al., 2011).

89

Table 7-4 Screening of most effective predictors for precipitation at the Astore climate station SN 1

Predictor ncepp5_v

R1 (%) 27.3

R2 (%)

P .r (%)

P-value

PRP (%)

2 3 4 5 6 7 8 9 10 11 12

ncepp5zh ncepp8_z ncepp5_f ncepp_f ncepp_v noepr850 nceprhum ncepp8_v ncepp_u ncepp500 ncepr500

26.5 21.4 20.7 20.0 19.3 18.6 18.6 17.0 16.0 15.7 15.1

95.8 49.6 47.6 41.8 28.7 46.0 46.0 5.6 44.3 37.2 55.5

1.4 9.2 9.9 9.9 12.5 7.1 7.1 16.1 4.5 6.2 0.0

0.38 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.56

94.7 57.0 52.2 50.5 35.2 61.8 61.8 5.3 71.9 60.5 100..0

R1 Absolute correration between Pedictand and Predictor; R2 Absolute correration between predictors; P.r Absolute partial correration between predictor and predictand in the presence of supper predictor; Bold are the selected predictors for Astore climate station 7.5.2 Calibration of SDSM The data simulated by SDSM-A and SDSM-M, using the NCEP variables, is denoted as NCEP_A and NCEP_M respectively. The results shown in Table 7-5 are comparable to some previous studies (Wilby et al., 2002; Nguyen, 2005; Souvignet and Heinrich, 2011). The explained variance of precipitation is much lower than temperature. According to Wilby et al. (2002), precipitation is a heterogeneous climate variable and is difficult to simulate accurately. Furthermore, the percentage of explained variance of temperature is most likely greater than 70% while precipitation is most likely lower than 40%. Tmax, Tmin and precipitation are simulated for the period of calibration, through both submodels, and compared with the observed data, as shown in Table 7-6. Although both models, SDSM-A and SDSM-M, performed well in the case of Tmax and Tmin, SDSM-M produces better results than SDSM-A as reflected by the values of performance indicators, except the µ of Tmin as µ produced by SDSM-A is much closer to the observed mean than µ produced by SDSM-M. With regard to precipitation, the SDSM-M model performed much better than SDSM-A. The R² value of SDSM-A is much lower and its RMSE value much greater than those of SDSM-M. The predicted mean precipitation by SDSM-A and SDSM-M is about 4 mm lower and 4 mm higher respectively. The percentage of relative error in standard deviation of SDSM-A and SDSM-M is -40.25 and 3.46% respectively. The monthly simulated mean outputs of SDSMs against the observed data are plotted in Figure 7-4. In the case of SDSM-A, Tmax is overestimated in the months of December, January, February, March, July and August, and underestimated in May, June, September, October, and November. The Tmin simulated by SDSM-A is overestimated in December, January, February, and March, but underestimated in the remaining months. SDSM-A underestimates precipitation in rainy season (July-August) and overestimates in the dry season (SeptemberJanuary) in the basin. As presented in Table 7-6 and Figure 7-4, although SDSM-A does not perform well in predicting the variation in data, it does perform well in predicting mean values.

90

Table 7-5 Explained variance (E) and standard error (SE) during calibration (1961-1990) Variable Tmax Tmin Precipitation

E (%) Max 71.5 85.2 32.0

Min 59.7 67.6 8.1

SE (°C or mm/day) Mean 66.2 75.9 10.9

Mean 4.69 3.47 0.45

Table 7-6 Statistical comparison of observed and downscaled mean monthly Tmax, Tmin and precipitation by two sub-models during calibration (1961-1990)

Tmax

Tmin

Precipitation

Observed NCEP-A NCEP-M Observed NCEP-A NCEP-M

Observed NCEP-A NCEP-M

1.05 0.19 RMSE (mm)

µ (°C) 21.03 21.03 21.02 8.93 8.94 8.89 µ (mm)

б (°C) 7.53 6.58 7.54 7.17 6.40 7.07 б (mm)

35.75 11.42

107.1 103.9 110.9

63.48 38.54 65.60

R²

RMSE (°C)

0.950 0.999

1.72 0.07

0.985 0.999 R²

0.688 0.992

RE-µ (°C)

RE-б (°C)

0.014 0.009

-0.950 0.002

0.008 -0.036 RE-µ (%)

-0.704 -0.092 R E-б (%)

-3.01 3.67

-40.25 3.460

R² Coefficient of determination; RMSE Root mean square error; µ Mean; б Standard deviation; RE_µ Relative error in mean; RE_ б Relative error in standard deviation

91

35

a

30

Tmax (°C)

25 20 15

Obs NCEP_A NCEP_M

10 5 0 Jan Feb Mar Apr May Jun

35

Jul

Aug Sep Oct Nov Dec

b

30

Tmin (°C)

25 20 15

Obs NCEP_A NCEP_M

10 5 0 Jan Feb Mar Apr May Jun

Jul

Aug Sep Oct Nov Dec

-5

Precipitation (mm/month)

250

c

200 150 100 Obs NCEP_A NCEP_M

50 0 Jan Feb Mar Apr May Jun

Jul

Aug Sep Oct Nov Dec

Figure 7-4 Observed and simulated mean monthly (a) Tmax (b) Tmin and (c) precipitation for the calibration period (1961-1990) in the Jhelum basin

92

7.5.3 Validation of SDSM without bias correction To validate SDSM-A) and SDSM-M, six sets of data (three from SDSM-A and SDSM-M each) are generated for the period of 1991 to 2000, forcing NCEP, H3A2, and H3B2 variables. The three datasets generated by SDSM-A are denoted as NCEP_A, H3A2_A, and H3B2_A, and the three datasets generated by SDSM-M are denoted as NCEP_M, H3A2_M, and H3B2_M. The results presented in Table 7-7 are the average values calculated using the data at all the weather stations. It is seen that in the case of temperature (both max and min), the R² and RMSE of SDSM-A range between 93.6 to 97.3% and 1.68 to 2.25°C respectively, and the absolute relative error in predicting mean (RE_µ) and standard deviation (RE_б), range between 0.04 to 0.29°C and 0.91 to 1.56°C, respectively. In the case of precipitation, the R², RMSE, abs RE_µ and abs RE_б range from 46 to 68%, 46 to 52 mm, 4 to 13%, and 45 to 53% respectively based on the observed and simulated monthly data. Conversely, the R², RMSE, abs RE_µ and, abs RE_б values obtained by SDSM-M lie between 98 to 99%, 0.91 to 1.14°C, 0.02 to 0.19°C, and 0.04 to 0.28°C for temperature (max and min considered together), and 71 to 90%, 22 to 49 mm, 3 to 6%, and 3 to 10% for precipitation. The results described in Table 7-7 indicate that both models (SDSM-A and SDSM-M) produce comparable results insofar as the coefficient of determination and predicting mean values regarding temperature are concerned. Despite this, SDSM-M produces much better results than SDSM-A in all other statistical parameters. In the case of precipitation, SDSM-A produces inferior results in all six parameters, especially in case of RE_б. This indicates that SDSM-A is not able to produce variations in observed precipitation. Nevertheless, the mean precipitation simulated by SDSM-A is comparable with SDSM-M. The simulated mean monthly, seasonal and annual precipitation are compared graphically with the observed data, as shown in Figure 7-5, to take note of pattern and variations captured by both sub-models. Figure 7-5(a) indicates that SDSM-A significantly underestimates precipitation in the peak precipitation months and overestimates it in the dry months in the basin in all three data sets. The variability or pattern of mean monthly observed precipitation is not well represented by this model. In the drier months, SDSM-A is overestimated; however, the simulated mean seasonal and also annual precipitation amounts by SDSM-A using NCEP, H3A2, and H3B2 data sets are well underestimated. Figure 7-5(b) indicates that the big peak of precipitation (July) is well captured by SDSM-M. However, the small peak occurring in March is not well predicted. The mean seasonal and annual amount of precipitation by this model has a much better agreement with the observed data in all three data sets than that of the monthly data set. It is also seen (Table 7-7) that most of the time, the Tmax, Tmin and precipitation predicted by using the NCEP predictors produces better results than H3A2 and H3B2.

93

Table 7-7 Statistical comparison of observed and downscaled (without bias correction) mean monthly Tmax, Tmin and precipitation by two sub-models during validation (19912000)

Tmax

Tmin

Precipitation

Observed NCEP_A H3A2_A H3B2_A NCEP_M H3A2_M H3B2_M Observed NCEP_A H3A2_A H3B2_A NCEP_M H3A2_M H3B2_M

Observed NCEP_A H3A2_A H3B2_A NCEP_M H3A2_M H3B2_M

R²

RMSE (°C)

0.945 0.946 0.936 0.993 0.990 0.990

1.92 2.16 2.25 0.96 1.14 1.06

0.973 0.959 0.949 0.990 0.987 0.983 R²

1.68 2.07 2.11 0.91 1.04 1.10 RMSE (mm)

0.683 0.491 0.462 0.897 0.737 0.708

46.98 50.06 52.23 22.34 39.08 39.77

94

µ (°C) 21.14 21.11 21.22 21.31 21.02 21.07 21.12 9.00 8.70 9.15 9.25 8.80 8.92 8.96 µ (mm) 117.2 100.0 111.7 107.6 109.6 113.1 110.4

б (°C) 7.49 6.63 6.07 6.04 7.52 7.48 7.58 7.33 6.23 5.81 5.98 7.12 7.05 7.08 б (mm) 68.95 37.44 35.20 34.85 65.80 64.14 64.50

RE_µ (°C)

RE_б (°C)

0.04 0.15 0.24 -0.12 -0.07 -0.02

-0.91 -1.53 -1.56 0.04 0.12 0.10

-0.29 0.15 0.26 -0.19 -0.07 -0.04 RE_µ (%)

-1.11 -1.52 -1.36 -0.21 -0.28 -0.25 RE_б (%)

-12.97 -4.02 -7.58 -5.64 -2.80 -5.65

-45.87 -50.37 -53.02 -3.00 -8.40 -9.60

300

Precipitation (mm/month)

1600

Obs NCEP_A H3A2_A H3B2_A Obs NCEP_A H3A2_A H3B2_A

a 250 200 150

1400 1200 1000 800 600

100

400 50

200

Ann

Aut

Sum

Spr

Win

Dec

Nov

Oct

Sep

Aug

Jul

Jun

May

Apr

Mar

Feb

0

Jan

0

300

1600 Obs NCEP_M H3A2_M H3B2_M Obs NCEP_M H3A2_M H3B2_M

Precipitation (mm/month)

b 250 200 150

1400 1200 1000 800 600

100

400 50

200

Ann

Aut

Sum

Spr

Win

Dec

Oct

Nov

Sep

Aug

Jul

Jun

Apr

May

Mar

Feb

0

Jan

0

Figure 7-5 Observed and downscaled (without bias correction) mean monthly, seasonal and annual precipitation by (a) SDSM-A (b) SDSM-M for the validation period in the Jhelum basin 7.5.4 Validation of SDSM with bias correction From the above results and discussion, there are large biases which should be removed to improve the validation results. The study applies a simple Bias Correction method discussed in detail by Salzmann et al. (2007). The six sets (NCEP_A_BC, H3A2_A_BC, H3B2_A_BC from SDSM-A, and NCEP_M_BC, H3A2_M_BC, and H3B2_M_BC from SDSM-M) are downscaled and corrected for biases. The statistical results for Tmax, Tmin and precipitation are described in Table 7-8, and the downscaled (with bias correction) mean monthly, seasonal and annual data (only precipitation) by SDSM-A and SDSM-M using NCEP, H3A2, and H3B2 predictors is compared with the observed data in Figure 7-6. These results indicate that almost all statistical indicators are quite improved when bias correction is applied, especially in the case of precipitation downscaled by SDSM-A. The R2 values from SDSM-A increase from 46-68% to 75-84% with bias correction. RMSE decreases up to 28-33 mm from 46-52 mm, and abs RE_б decreases from 95

46-53% to 1-6%, as presented in Table 7-8. Additionally, patterns and variations of simulated precipitation by both models are also far improved in all the three data sets, as can be seen in Figure 7-6. On the whole, results obtained by using the NCEP variables are relatively superior to the results obtained from H3A2 and H3B2, because both models (SDSM-A and SDSM-M) are calibrated using the NCEP data set. Nonetheless, the results of H3A2 and H3B2 are satisfactory now. These results indicate a strong applicability of both SDSM-A and SDSM-M to downscaling Tmax, Tmin and precipitation under emission scenarios H3A2 and H3B2. Table 7-8 Statistical comparison of observed and downscaled (with bias correction) mean monthly Tmax, Tmin and precipitation by two sub-models during validation (19912001)

Tmax

Tmin

Precipitation

Observed NCEP_A_BC H3A2_A_BC H3B2_A_BC NCEP_M_BC H3A2_M_BC H3B2_M_BC Observed NCEP_A_BC H3A2_A_BC H3B2_A_BC NCEP_M_BC H3A2_M_BC H3B2_M_BC

Observed NCEP_A_BC H3A2_A_BC H3B2_A_BC NCEP_M_BC H3A2_M_BC H3B2_M_BC

R²

RMSE (°C)

0.991 0.984 0.985 0.991 0.985 0.990

1.14 1.21 1.18 1.10 1.21 1.09

0.990 0.991 0.984 0.992 0.993 0.990 R²

1.01 0.90 1.02 0.92 0.79 0.85 RMSE (mm)

0.834 0.780 0.754 0.845 0.814 0.774

28.31 33.50 32.39 27.00 30.44 30.80

96

µ (°C) 21.14 20.82 20.87 21.00 20.79 20.82 20.86 9.00 8.70 8.97 9.04 8.75 8.90 8.91 µ (mm) 117.2 115.4 123.8 121.1 113.1 119.8 116.9

б (°C) 7.49 7.55 7.45 7.42 7.51 7.37 7.46 7.33 7.11 6.93 7.05 7.18 7.01 7.04 б (mm) 68.95 69.07 73.11 70.64 67.12 74.32 71.48

Error_µ (°C)

Error_б (°C)

-0.32 -0.27 -0.14 -0.35 -0.32 -0.28

0.08 -0.04 -0.07 0.02 0.001 -0.03

-0.30 -0.02 0.04 -0.24 -0.10 -0.08 Error_µ (%)

-0.22 -0.41 -0.28 -0.15 -0.32 -0.29 Error_б (%)

-0.92 6.61 3.81 -2.79 3.59 0.75

1.39 6.09 2.46 -0.34 8.64 3.77

Precipitation (mm)/month

300

1600

Obs NCEP_A_BC H3A2_A_BC H3B2_A_BC Obs NCEP_A_BC H3A2_A_BC H3B2_A_BC

a 250 200 150

1400 1200 1000 800 600

100

400 50

200

300

Aut

Ann

Spr

Win

Sum

1600

Obs NCEP_M_BC H3A2_M_BC H3B2_M_BC Obs NCEP_M_BC H3A2_M_BC H3B2_M_BC

b Precipitation (mm/month)

Dec

Nov

Oct

Sep

Jul

Aug

Jun

May

Apr

Mar

Feb

0

Jan

0

250 200 150

1400 1200 1000 800 600

100

400 50

200

Ann

Aut

Sum

Spr

Win

Dec

Nov

Oct

Sep

Aug

Jul

Jun

May

Apr

Mar

Jan

0

Feb

0

Figure 7-6 Observed and downscaled (with bias correction) mean monthly, seasonal and annual precipitation by (a) SDSM-A and (b) SDSM-M for the validation period (1991-20001) in the Jhelum basin 7.6

Downscaling (with bias correction) of temperature and precipitation

SDSM-A and SDSM-M are used to simulate Tmax, Tmin and precipitation for the future periods of the 2020s (2011-2040), 2050s (2041-2070) and 2080s (2071-2099) under H3A2 and H3B2 scenarios. The downscaled data from the two models is corrected for biases. The simulated (with bias correction) monthly, seasonal and annual Tmax, Tmin and precipitation data is compared with the baseline period (1961-1990) to analyze future changes in the 2020s, 2050s, and 2080sin the upper Jhelum River basin, and the results are presented in Tables 8, 9, and 10 respectively. Since Gulmarg and Qazigund do not have daily or even monthly data from 1961 to 1968, for these stations data is estimated for the missing period by SDSM using NCEP. Max temperature Table 7-9 presents the changes in Tmax in the 2020s, 2050s and 2080s, relative to the baseline period (1961-1990), under H3A2 and H3B2 scenarios from both models. It is seen that the changes in Tmax predicted by SDSM-A and SDSM-M are quite different in magnitude (amount), but identical in pattern. Both SDSM-A and SDSM-M models indicate incremental 97

changes with respect to the baseline period in the 2020s, 2050s and 2080s under both scenarios. For H3A2 and according to SDSM-A, the mean annual Tmax could increase by 3.15°C compared to a 0.91°C increase in last three decades of this century with SDSM-M. Under the H3B2 scenario, based on both SDSM-A and SDSM-M, these increments in mean annual Tmax could be 1.9 and 0.7°C respectively in the last three decades of this century. All seasons show increments in Tmax in the 2020s, 2050s and 2080s, presenting an increasing trend in the entire period from 2011 to 2099 in both scenarios, but with different magnitudes. According to SDSM-A, under the H3A2 scenario, the season with maximum increase in Tmax could be spring in all the three future periods, but as per the SDSM-M, maximum increase in Tman will occur in autumn in the 2020s, in winter in the 2050s, and in spring in the 2080s. Under the H3B2 scenario, both models project increased Tmax in the winter seasons for all three periods. Table 7-9 Future changes in Tmax (°C) with respect to baseline (1961-1990) under H3A2 and H3B2 scenarios with two sub-models H3A2 SDSM-A SDSM-M 2020s 2050s 2080s 2020s 2050s 2080s Winter 0.68 1.82 3.16 0.26 0.78 0.96 Spring 1.10 1.95 3.90 0.03 0.53 1.13 Summer 0.34 1.31 2.58 0.01 0.23 0.57 Autumn 0.50 1.64 2.96 0.30 0.49 1.02 Annual 0.65 1.68 3.15 0.14 0.50 0.91 H3B2 SDSM-A SDSM-M 2020s 2050s 2080s 2020s 2050s 2080s Winter 0.71 1.64 2.44 0.38 0.59 0.86 Spring 0.48 1.21 1.81 0.18 0.42 0.73 Summer 0.18 1.00 1.71 0.05 0.25 0.46 Autumn 0.64 1.23 1.73 0.34 0.56 0.72 Annual 0.50 1.27 1.92 0.23 0.45 0.69 Min temperature Table 7-10 shows the changes in Tmin in the 2020s, 2050s, and 2080s with respect to the baseline period under the H3A2 and H3B2 scenarios obtained from both SDSM-A and SDSMM. It is recognized that the changes in Tmin predicted by both models are different in magnitude, but similar in their patterns as is the case with Tmax. Under the H3A2 scenario, according to SDSM-A, the mean annual Tmin can increase by 2.63°C, and according to SDSMM, the increment might be 0.93°C in the 2080s. Under the H3B2 scenario, as per SDSM-A and SDSM-M, these increments in the mean annual Tmin could be 1.63 and 0.56°C respectively in the last three decades of this century. Under the H3A2 scenario, spring is predicted to be the maximum increase in the 2020s and 2080s with an increment of 0.71 and 3.22°C respectively, and autumn, with a mean rise of 1.62°C in the 2050s with SDSM-A. Interestingly, SDSM-M predicts autumn with maximum increase in Tmin for all three future periods.

98

Under the H3B2 scenario, SDSM-A projected the autumn of the 2020s as witnessing an increment of 0.92°C. Winter in the 2050s and also in 2080s, with increments of 1.47 and 2.06°C respectively, is the season with maximum increase in warming. SDSM-M projected autumn as the season with most warming, with a rise of mean Tmin of 0.7, 0.85 and 0.97°C in the 2020s, 2050s, and 2080s respectively under H3B2. Table 7-10 Future changes in Tmin (°C) with respect to baseline (1961-1990) under H3A2 and H3B2 scenarios with two sub-models H3A2 SDSM-A SDSM-M 2020s 2050s 2080s 2020s 2050s 2080s Winter 0.60 1.51 2.53 0.33 0.77 0.80 Spring 0.71 1.41 3.22 0.08 0.37 0.90 Summer 0.23 1.05 2.12 0.05 0.24 0.54 Autumn 0.68 1.62 2.65 0.62 0.93 1.39 Annual 0.55 1.39 2.63 0.27 0.57 0.93 H3B2 SDSM-A SDSM-M 2020s 2050s 2080s 2020s 2050s 2080s Winter 0.79 1.47 2.06 0.34 0.47 0.63 Spring 0.32 0.88 1.44 0.06 0.12 0.32 Summer 0.14 0.82 1.39 0.00 0.18 0.33 Autumn 0.92 1.40 1.66 0.70 0.85 0.97 Annual 0.54 1.14 1.63 0.24 0.40 0.56 Precipitation Table 7-11 presents the percentage change in seasonal and annual mean precipitation in the 2020s, 2050s and 2080s with respect to the baseline period under the H3A2 and H3B2 scenarios obtained from both SDSM-A and SDSM-M. The changes in precipitation predicted by both models are different from each other with regard to magnitude. Both models show mean annual increments with respect to the baseline period in the 2020s, 2050s and 2080s for both scenarios. Under the H3A2 scenario, SDSM-A indicates an increase in the mean annual precipitation by 7.9, 9.6 and 11.7% in the 2020s, 2050s and 2080s respectively in the basin. SDSM-M projects an increase of about 5.9, 6.8 and 6% in the 2020s, 2050s, and 2080s respectively. Under the H3B2 scenario, according to SDSM-A, there will be a rising trend in precipitation in the future and there will be a slight decreasing trend according to SDSM-M. According to the two sub-models, most of seasons (except spring) indicate an increased precipitation in the 2020s, 2050s, and 2080s relative to the baseline in both scenarios but with different magnitudes (amounts) of increase. Only spring, projected by SDSM-A, indicates a decrease precipitation in 2080s, while SDSM-M shows a decrease in all the three future periods under the H3A2 scenario. Under the H3A2 scenario and according to SDSM-A, autumn is expected to have highest percentage increase in the 2020s and 2080s whereas winter will have a maximum percentage increase in the 2050s. In case of SDSM-M, winter will possibly have maximum percentage increase in precipitation for all three future time periods.

99

Under the H3B2 scenario, it is projected by SDSM-A that autumn may be the most affected with a rise in precipitation approximately by 24, 28 and 31% in 2020s, 2050s and 2080s respectively. However, SDSM-M projects that winter could have maximum increase in precipitation of about 13, 14 and 19% in 2020s, 2050s and 2080s respectively. Spring is expected to have minimum change in precipitation with both sub-models. Table 7-11 Future changes in precipitation (%) with respect to baseline (1961-1990) under H3A2 and H3B2 scenarios with two sub-models H3A2 SDSM-A SDSM-M 2020s 2050s 2080s 2020s 2050s 2080s Winter 12.19 14.71 13.31 16.56 16.21 12.29 Spring 0.19 4.35 -4.82 -7.17 -0.34 -1.60 Summer 10.57 10.32 14.30 7.24 9.34 9.27 Autumn 14.10 12.19 24.74 7.08 2.73 6.43 Annual 7.87 9.58 11.75 5.86 6.85 5.98 H3B2 SDSM-A SDSM-M 2020s 2050s 2080s 2020s 2050s 2080s Winter 14.05 11.59 17.62 13.76 14.34 18.59 Spring 4.29 5.29 5.17 3.66 2.00 1.18 Summer 9.17 11.81 9.18 9.78 8.46 3.74 Autumn 24.01 27.85 31.86 10.99 11.35 13.04 Annual 10.31 12.23 13.78 8.61 8.28 7.98 7.7

Conclusions

A widely used decision support tool known as the “SDSM” was applied to downscale the Tmax, Tmin and precipitation in the upper Jhelum River basin in Pakistan and India under IPCC SRES A2 and B2 scenarios. The downscaling of these parameters is very important in order to study the impact of future climate on the hydrology of the basin. SDSM was developed in this study using the annual (SDSM-A) and monthly (SDSM-M) submodels. Historical daily data of 30 years (1961-1990) was used to construct a strong statistical relationship between large scale and local scale variables in conjunction with a 10-year (19912000) data period for validation. The main challenge in the application of this method, apart from the mountainous topography and dominant monsoons in the study area, was the screening of appropriate predictors. A quantitative approach was used by which the most prominent predictors to be used in the calibration process could be ordered. Six statistical indicators were used to verify the performance of downscaled data from both sub-models for calibration and validation periods. The validation results are found to be greatly improved with the application of bias correction on the downscaled data. The results (downscaled and bias corrected) of validation show that both SDSM-A and SDSM-M can be applied to simulate future climate under the H3A2 and H3B2 scenarios. The downscaled and bias corrected data (2011-2099) was divided into three periods: 20112040 (2020s), 2041-2070 (2050s) and 2071-2099 (2080s) and compared with the baseline period (1961-1990) to observe the changes. The comparison results show that the changes in Tmax and Tmin predicted by the two sub-models are different in magnitude but similar in their patterns. Both sub-models show an increase in Tmax and Tmin in the 2020s, 2050s and 2080s 100

under the two scenarios. The results on seasonal basis also indicate an increasing trend in Tmax and Tmin in all four seasons under both scenarios. Both SDSM-A and SDSM-M show an increase in mean annual precipitation under H3A2 and H3B2 for all three future periods in the basin. All seasons except spring indicate increments in the 2020s, 2050s and 2080s relative to the baseline under both scenarios with the two submodels. Autumn, according to SDSM-A, and winter, according to SDSM-M are projected to experience maximum percentage changes in the precipitation under both H3A2 and H3B2 scenarios. The spring season in the basin is predicted to have minimum changes in precipitation. SDSM-A most often projects higher changes in Tmax, Tmin and also in precipitation, than SDSM-M under both scenarios. The changes in Tmax and Tmin simulated under H3A2 scenario by both sub-models are higher than H3B2. It is also seen that the increment in Tmax is higher than the increment in Tmin in mean annual and seasonal values under both scenarios, except in autumn. It is interesting to note that the mean annual changes in precipitation simulated by both sub-models are higher under H3B2 than H3A2. It is concluded that although SDSM developed by annual sub-model (SDSM-A) performed well in predicting mean annual values, it cannot be used to find out the monthly and seasonal variations, especially with regard to precipitation unless bias correction is applied.

101

8

CLIMATE CHANGE IMPACT ON WATER RESOURCES

This chapter describes the process of finding the impacts of climate change on the water resources in the Jhelum basin. The main sections of this chapter consist of data required for hydrologic model (HEC-HMS), watershed delineation using HEC-GeoHMS and whole methodology. Climate change impacts are found out on stream flow, flow duration curves, low, high, and medium flow, and temporal shifts of peak flows 8.1 Data description 8.1.1 DEM data In this study, a 30 m ASTER-GDEM developed by NASA (United States National Aeronautics and Space Administration) and METI (Ministry of Economy, Trade, and Industry) of Japan is used to extract the physical feature of the Jhelum river basin. NASA and METI jointly have released the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) Global Digital Elevation Model Version 2 (GDEM V2) on October 17, 2011.The first version of the ASTER GDEM was released in June 2009. It was generated using stereo-pair images collected by the ASTER instrument onboard Terra. The coverage span of ASTER GDEM range between 83° North and 83° South, covering 99% of the Earth's landmass. The improved GDEM V2 includes 260,000 additional stereo-pairs, refining coverage and reducing the occurrence of artifacts. In this version, spatial resolution, horizontal and vertical accuracy, and superior water body coverage and detection are improved greatly. The ASTER GDEM V2 keeps the GeoTIFF format and the same gridding and tile structure as V1, with 30 m postings and 1 x 1° tiles. As a contribution from METI and NASA to the Global Earth Observation System of Systems (GEOSS), ASTER GDEM V2 data are available free of charge to users worldwide from the Land Processes Distributed Active Archive Center (LP DAAC) and J-spacesystem. 8.1.2 Hydro-meteorological data The observed hydro-meteorological data needed for hydrologic modeling is obtained from the Water and Power Development Authority of Pakistan (WAPDA), India Meteorological Department (IMD), and Pakistan Meteorological Department (PMD). The whole data is illustrated in Table 8-1 and Table 8-2. The gauging stations mainly are maintained and operated by WAPDA and the meteorological climate stations by PMD. Some meteorological stations located in Pakistan are also maintained by WAPDA but mostly by PMD. The downscaled data of precipitation and temperature used to predict the future time series of flow is downscaled in Chapter 5. Table 8-2 shows the main characteristics of stream gauges available in the Jhelum River basin. Azad-Pattan is the main gauge of the Jhelum basin where about 87% of the Jhelum basin area contributes.

102

Table 8-1 Meteorological stations located inside and around the study area Sr. No. 1 2 3 4 5 6 7 8 9 10 11 12 13 14

Station Jhelum Kotli Plandri Rawlakot Murree Garidopatta Muzaffarabad Balakot Naran Kupwara Astore Srinagar Gulmarg Qazigund

Lat 32.94 33.50 33.72 33.87 33.91 34.22 34.37 34.55 34.90 34.51 35.34 34.08 34.00 33.58

Long 73.74 73.90 73.71 73.68 73.38 73.62 73.48 73.35 73.65 74.25 74.90 74.83 74.33 75.08

Elev. (m) 287 614 1402 1676 2213 814 702 995 2362 1609 2168 1587 2705 1690

Daily data 1970-00 1970-00 1962-00 N/A 1970-00 1970-00 1961-00 1970-00 N/A 1976-00 1953-00 1961-00 1969-00 1961-00

Monthly data 1950-09 1952-09 1962-09 1960-09 1960-09 1955-09 1955-09 1960-09 1961-09 1976-09 1953-09 1950-09 1969-09 1961-09

Table 8-2 Stream flow stations located in the study area Sr. No.

Rivers

Stations

Lat

Long

Elev. (m)

1 2 3 4 5 6 7 8

Jhelum Jhelum Jhelum Kanshi Kunhar Kunhar Neelum Poonch

Domel Kohala Azad-Pattan Palote Garihabibulla Naran Muzaffarabad Kotli

34.35 34.11 33.73 33.22 34.45 34.90 34.370 33.489

73.46 73.49 73.60 73.43 73.36 73.65 73.467 73.885

2,198 1,917 1,139 1,281 2,657 7,993 2,254 1688

Daily data 1976-09 1965-96 1961-95 1970-09 1960-98 1960-05 1963-09 1960-09

Monthly data 1976-09 1965-96 1961-95 1970-09 1960-98 1960-05 1963-09 1960-09

8.1.3 Soil data There are four hydrologic soil groups (A, B, C, and D) described below which are assigned on the basis of measured rainfall, runoff, and infiltrometer data. These soils groups along with land cover data are used to determine the runoff curve number (USDA, 2007). Group A: 1. Low runoff potential in this group 2. Free water transmission through the soil means high infiltration rate ( greater than 7.62 mm/h) 3. This group has 10% clay and 90% sand or gravel 4. Loamy sand, sandy loam or silt loam texture comes under this group Group B: 1. Moderately low runoff potential 2. Unimpeded water transmission through the soil with infiltration rate ranges between 3.8 and 7.62 mm/h 3. This group has 10-20% clay and 50-90% sand 4. A texture of loam, silt loam, silt, or sandy loam belong to this group Group C: 1. The soils under this group have moderately high runoff potential 103

2. Somewhat restricted water transmission through the soil and infiltration rates ranges between 1.25 to 3.8 mm/h 3. This soil group has 20-40% clay and less than 50% sand 4. Clay, silt clay and sandy clay texture comes under this group Group D 1. High runoff potential 2. Very restricted water movement through the soil and the infiltration rates lie between 0-1.25 mm/h 3. The soils under this group have greater than 40% clay and less than 50% sand 4. Having clayey texture Figure 8-1 shows the soil groups derived for the Jhelum basin from the Harmonized World Soil Database v 1.21 which was updated on 7 March, 2012 and having a resolution of 1 km.

Figure 8-1 Soil groups available in the Jhelum River basin 8.1.4 Land cover data Land cover can importantly affect the hydrologic processes. These processes are mainly influenced by the density of plant cover and morphology of plant species. Table 8-3 describes the 18 classification of land covers in the Jhelum basin, derived from the global land cover data (1 km resolution) developed by the Joint Research Center (JRC) of European Commission, which are then reclassified into 7 main classes as shown in Figure 8-2. Percentage area of each class is described as: 29% (forest), 45% (Agriculture), 16% (Alpine meadow/grassland) 8% (snow), 1% water body and 0.04% residential area.

104

Table 8-3 Classification of land covers in the Jhelum River basin Class Code 6 7 8 10 12 23 24 25 26

LU_types Temperate Conifer Subtropical Conifer Tropical Moist Deciduous Junipers Degraded Forest Slope Grasslands Desert Grasslands Alpine Meadow Alpine Grasslands

Class Code 26 27 29 32 33 34 39 40 45

LU_types Alpine Grasslands Sparse vegetation (cold) Gobi Irrigated Intensive Agriculture Irrigated Agriculture Slope Agriculture Water Bodies Snow Settlement

LU = Land Use

Figure 8-2 Land cover in the Jhelum River basin 8.2 Methodology 8.2.1 Model selection criteria There are various criteria for choosing a suitable hydrological model. Since every study (research or project) has its own specific objectives, so these criteria are mostly depended upon the research or project type. Some criteria are related to user’s choice and therefore subjective, such as personal preference for GUI (graphical user interface), input-output management and structure or computer operation system. But the following four are the most important project depended criteria that have to be fulfilled: 105

1- Outputs required for the project which must be estimated by the model (Is the model able to predict the data required by the research? For example, peak flow, hydrographs, event volume, etc.). 2- Input data availability (Is all the input data needed for model to estimate the required output available within the time and cost constraints of the research?) 3- Hydrologic processes that require to be modeled to estimate the desired outputs adequately (Is the model able of simulate regulated reservoir operation, singleevent or continuous processes, snow accumulation and melt ?) 4- Price (Should the investment seem to be useful for the objectives of the research) The whole methodology developed for hydrological modeling is schematically presented in Figure 8-3 and explained in the following sections

 DEM  Land use data  Soil Data

HEC-GeoHMS

Basin Characteristics  Elevations  Slopes  Sub-basin areas  Stream flow lines  Longest Flow paths  Basin Lag times  Soil groups  Curve Numbers  Basin centroids

Hydrological Modeling System (HEC-HMS 1. Runoff volume model (Initial and constant loss method) 2. Direct runoff model (SCS unit hydrograph) 3. Base flow model (Recession) 4. Channel routing (Muskingum) 5. Meteorological model (Gage weight) 6. Evapotranspiration model (Penman Monteith) 7. Snowmelt model (Temp Index)

 Calibration  Validation

Hydro-Climatic data  Rainfall  Temperature  Evapotranspiration  Stream flow

Downscaled data (A2 and B2)  Rainfall  Temperature

Climate change impacts on stream flow under A2 and B2 Scenarios

Simulated stream flow series

Figure 8-3 Schematic diagram of assessing the climate change impacts on water resources in the Jhelum River basin 8.2.2 HEC-GeoHMS In the recent years, the hydrologic modeling systems have improved due to the advancements in Geographical Information Systems (GIS). GIS has developed many extensions to support hydrologic models. HEC-GeoHMS is one of them, which mainly supports the HEC-HMS (Hydrologic modeling system). This extension is used to delineate sub-basins and stream lines, to visualize the spatial information, construct watershed characteristics and to prepare all input data for the hydrologic model.

106

8.2.3 HEC-HMS The Hydrological Modeling system (HEC-HMS) is a rainfall-runoff simulation software, used for a wide range of watersheds from large river basins to small urban areas, formulated by the US Army Corps of Engineer belonging to Hydrologic Engineering Center (HEC). This system consists of different kinds of loss methods such as SCS curve number, Initial and Constant, Green Ampt, One-layer Deficit-Constant (which is used for simple continuous modeling), Smith Parlange, and Five-layer Soil Moisture Accounting which can be used for complex infiltration and evapotranspiration environments to estimate excess rainfall. So, this model can be used for both event and continuous modeling. This also compasses seven methods such as SCS, Clark, Snyder, and ModClark for transforming excess precipitation to surface runoff, five base flow methods—recession method, constant monthly method, and linear reservoir method, and six channel routing methods like Muskingum and modified pulse methods. There are six kinds of meteorological models such as Thiessen and inverse distance methods to analyze the meteorological data like precipitation, evapotranspiration and snowmelt. The meteorological model extracts the precipitation for each sub-basin in the watershed. Currently, only two methods (Temperature Index and Gridded Temperature Index) are available to deal with snowfall in this modeling system (HEC-HMS, 2000, 2010). A complete basin model setup simulating a rainfall-runoff process comprises of a basin model, meteorological model, control specification and input time series. The basin model describes the physical properties (areas of sub basins and river lengths, etc.) of a watershed. Each basin model in HEC-HMS consists of a loss, transforming (unit hydrograph), base flow, and channel routing methods (Verma et al., 2010). The control specification is one of the main components of the model setup which controls the simulation period. For example, when the model is to start and stop, and what should be the time interval for simulation. The input time series encompasses the precipitation, temperature, evapotranspiration, observed stream flow, etc., which have a direct link with the basin model and meteorological model. The detailed description of the model formulation and various processes are explained in the HEC-HMS’s User’s Manual and Technical Reference Manual (HEC-HMS, 2000, 2010). For the present research work, the basin model includes the Deficit and Constant-rate Loss (DCL) method, the SCS unit hydrograph, Musking for channel routing and recession base flow, and the meteorological model comprises of gauge weight method for precipitation calculation, simple temperature index method for snowmelt modeling, and monthly evapotranspiration method. The DCL model computes the excess precipitation for watershed. It is a single layer continuous method used for calculating the changes in soil moisture content. It is similar to initial and constant-rate loss method but this method recovers the initial losses after a long period of no precipitation. This model contains four main parameters: maximum deficit, initial deficit, constant rate, and impervious percentage. To transfer the flow from one point to other, Musking method is used here which is a simple mass conservation scheme for routing flow through the channels. There are two main parameters for this method: travel time K and Muskingum coefficient X. The coefficient ranges between 0 and 0.5. The excess precipitation from DCL is transformed into direct surface runoff by SCS unit hydrograph method. Basin lag is only one parameter which needs to be determined during calibration. It can also be estimated as an initial value for calibration by multiplying the time of concentration with 0.6. In this study, recession method is used to calculate the base flow 107

which contributes to total flow from the watershed. Three parameters such as initial discharge, recession constant, and threshold are determined during calibration in this method. Thiessen polygon method is used to assign the weights to each gauge in the watershed during the development of meteorological model. In this study, different elevation bands are used for each sub-basin in temperature index method during snow melt process. A 7.0° km-1 is calculated for this study area and considered to remain constant for the whole Jhelum river basin. FAO Penman-Monteith method—recommended as the standard method for computation of potential evapotranspiration (Yimer et al., 2009)—is carried out to calculate the potential evapotranspiration. 8.2.4 Model calibration and validation Calibration of a model is a process in which the model parameters are adjusted in such a way that the simulated flow captures the variations of the observed flow(García et al., 2008). In this study, a split sample method is used for calibration and validation. In this method, the calibration period does not overlap with the validation period. So, a data period of eight years from 1982 to 1989 is chosen as the calibration period and from 1978 to 1981 for validation because these periods have minimum missing values in case of both precipitation and stream flow. The physical characteristics—land use cover and soil properties— of the watershed are considered constant during the simulation period. There are two automated algorithms—Nelder Mead and Univariate Gradient— to optimize the objective function. There are seven different kinds of objective functions in HEC-HMS, and the sum of squared residual is taken for this study which is minimized by Nelder Mead algorithm to explore the optimized model parameters which gives best results of simulation. The simulated flow is compared with the observed flow using the coefficient of determination (R2), Nash-Sutcliffee efficiency (E), and percent deviation (D). The R2 value describes how well the variation in the observed data are captured by the simulated data, ‘E’ shows how well the observed plot fits to the simulated plot, and ‘D’ represents the mean percent deviation between observed and simulated flow (Meenu et al., 2012). For more illustrative purposes, the simulated data is also compared with observed data graphically to explore how well the low and high observed flows are captured by simulated flow. In this study, the model is calibrated and validated at seven different gauging stations, as shown in Figure 8-6. The above mentioned model performance parameters (R, E, and D) are calculated by the following equations: 𝑅2 =

∑(𝑄𝑜𝑏𝑠 − ̅̅̅̅̅̅ 𝑄𝑜𝑏𝑠 ) × (𝑄𝑠𝑖𝑚 − ̅̅̅̅̅̅ 𝑄𝑠𝑖𝑚 ) 2 ̅̅̅̅̅̅ ̅̅̅̅̅̅ 2 √∑(𝑄𝑜𝑏𝑠 − 𝑄 𝑜𝑏𝑠 ) × (𝑄𝑠𝑖𝑚 − 𝑄𝑠𝑖𝑚 )

8-1

The value of R2 should be closer to 1 (best if exactly 1) for good correlation between simulated and observed flow. 𝐸 = 1−

∑(𝑄𝑠𝑖𝑚 − 𝑄𝑜𝑏𝑠 )2 2 ̅̅̅̅̅̅ ∑(𝑄𝑜𝑏𝑠 − 𝑄 𝑜𝑏𝑠 )

8-2

108

Values of ‘E’ lie between 0 and 1. A positive and close to 1 is the indication of good calibration but a value which is negative and close to 0 is not acceptable. If the ‘E’ value is greater than 0.75 then the results are considered to be good and satisfactory if between 0.36 and 75 (Van Liew and Garbrecht, 2003). 𝐷 (%) = 100 ×

𝑄𝑠𝑖𝑚 − 𝑄𝑜𝑏𝑠 𝑄𝑜𝑏𝑠

8-3

‘D’ value should be 0% or very close to 0% for best results. Positive and negative values are the indication of over and underestimation of model respectively. 8.2.5 Projected changes in stream flow and flow duration curve After the successful calibration and validation, the downscaled daily time series (A2 and B2) of precipitation and temperature described in Chapter 0 for the period of 2011 to 2099 are stored into the HEC-HMS, and daily flow time series for each flow gauge are simulated. The physical characteristics of the Jhelum basin are considered constant throughout the simulation period. However, these characteristics vary along the time. Then the simulated data is divided into three periods 2020s (2011-2040), 2050s (2041-2070), and 2080s (20712099) and compared with base period (1961-1990) to assess the changes in the future flow. During the analysis of the impacts of climate change on the stream flow, two questions are more important: 1) how often the stream flow will occur in future? And what will be the magnitude? To give answers of these two questions, flow duration curves are the main tools. These curves present the percentage of time a flow in a stream is likely to exceed or equal to a specified value of flow. These curves can be applied in different kinds of studies such as hydropower management, reservoir sedimentation, water quality management, low and high flow studies, etc. (HEC-ResSim, 2007). The following equation is used to construct the flow duration curves: 𝑃 (%) =

𝑀 × 100 (𝑛 + 1)

8-4

P= probability of flow equal or exceeded from a specified value (% of time) M= ranks of events N= the number of events in the specified period of time In the present study, daily time series are used to construct the flow duration curves for the base period (1961-1990) and three future periods; 2020s, 2050s, and 2080s. Three indicators Q5 (high flow), Q50 (median flow), and Q95 (low flow) are used to analyze the occurrence of high, low and median flows in the Jhelum river basin at different sites. 8.3 Results and discussion 8.3.1 Watershed delineation Basin delineation is the first step in any kind of hydrologic modeling for analyzing some basic properties of watershed such as slope, area, flow directions, flow lengths and stream network density. These days Digital Elevation Models (DEM) are mostly used to delineate the watersheds. In this study, a 30 m ASTER-GDEM developed by NASA is loaded into the HEC-GeoHMS to extract the topographic features of watershed as show in Figure 8-4. The elevation in the Jhelum River basin lies between 214 m to 6,000 m above mean sea level (MSL) as also reported in this study (Shabeh ul et al., 2012). The two elevation zones, one 109

above 5,000 masl and other below 3,000 are important in the sense that the area below 3,000 masl is considered as no snow occurrence and above 5,000 masl is considered as permanent glaciers(Khattak, 2011). An area of 33,308 km2 is derived from the DEM data which is quite similar to the area described by Archer and Fowler (2008). The whole basin is divided into 21 sub-basins, as show in Figure 8-5, for hydrologic analysis and their main characteristics are described in Appendix CError! Reference source not found..

Figure 8-4 DEM of the Jhelum River basin

110

Figure 8-5 Sub-basins derived from the DEM for the Jhelum River basin

Figure 8-6 Location map of hydro-climatic stations in the Jhelum River basin, showing main tributaries as well

111

8.3.2 Calibration and validation of HEC-HMS Table 8-4 shows the calculated model performance parameters from the observed and simulated stream flow for calibration and validation periods. Ranges of E and R values are 0.65-0.74 and 0.66-0.75 for calibration and 0.62-0.78 and 0.63-0.81 for validation respectively which are quite acceptable according to Van-Liew and Garbrecht (2003) except Kotli with 0.31 (E) and 0.36 (R). On the other hand, the values of D lie between -11.73 -1.21 for calibration and -3.04-12.4 for validation respectively. The poor calibration of Kotli could be due to less number of rain gauges in the Jhelum basin and abrupt changes in stream flow at that point. Poonch River contributes to only 10-14% flow to the Jhelum River. The validation results are slightly better than the calibration results on some gauges. The graphical comparison of observed against simulated for both calibration and validation periods is shown in Figure 8-7 and 7-7 respectively. At Naran and Garihabibulla gauging stations, patterns and variations of observed flow are well followed by simulated flow during the calibration. However the peaks and low flows are not well captured by the simulated flow which is why the total simulated volume is underestimated by about 10%. In case of validation, the peaks are comparatively better captured by the simulated flow especially at Garihabibulla station. At Muzaffarabad, Kohala, Domel and Azad-Pattan gauges, the peaks and low flows are comparatively well captured by the simulated flow except at Muzaffarabad where low flows are overestimated by the model in both calibration and validation periods. On the other hand, at the Kotli gauge, the results are poor and the patterns and variations in observed flow are not well followed by simulated flow. Nevertheless, the good thing is that, the total simulated flow is overestimated by only 1.2% (calibration) and 0.14% (validation) which is much better than other gauges. Table 8-4 Nash-Sutcliffe efficiency (E), coefficient of determination (R2), and percent deviation (D) for calibration (1982-89) and validation (1978-81) for different stream gauges in the Jhelum River basin Gauge (River) Naran (Kunhar) Gari-habibullah (Kunhar) Muzaffarabad (Neelum) Domel (Jhelum) Kohala(Jhelum) Azad-pattan (Jhelum) Kotli (Poonch)

E 0.74 0.65 0.72 0.66 0.72 0.70 0.31

Calibration R2 D (%) 0.75 -11.00 0.67 -9.92 0.73 -11.73 0.66 -2.38 0.73 -8.54 0.71 -8.44 0.36 1.21

112

E 0.71 0.71 0.75 0.62 0.78 0.75 0.32

Validation R2 D (%) 0.78 7.62 0.74 2.53 0.81 12.40 0.63 -3.04 0.80 6.34 0.78 8.71 0.35 0.14

Table 8-5 Nash-Sutcliffe efficiency (E), coefficient of determination (R2), and percent deviation (D) for some previous studies Study (Cal. Yrs/Val. Yrs)

Country

Calibration E R2

D (%)

Validation E R2

(Meenu et al., 2012) (20/10) (García et al., 2008) (1/1)

India

0.48

0.07

0.59

Spain

0.52-0.79

(Verma et al., 2010) (1/1)

India

0.70-0.83

0.8-0.84

(Yimer et al., 2009) (5/3)

Ethiopia

0.62

0.63

400

Discharge (m³/s)

0.77

0.73-0.75

0.7-0.8

6-9

0.72

0.73

0.25-0.72

-08-31

NARAN

Obs

350

0.72

D (%) 0.07

Sim

300 250 200 150 100 50

01/Jan/87

01/Jan/88

01/Jan/89

01/Jan/88

01/Jan/89

Sim

400 300 200 100

01/Jan/86

01/Jan/85

01/Jan/84

01/Jan/83

0

01/Jan/82

Discharge (m3/s)

01/Jan/86

G.HABIB

Obs 500

01/Jan/87

600

01/Jan/85

01/Jan/84

01/Jan/83

01/Jan/82

0

113

3500 Sim

1000

500

0

114 01/Jan/89

1500

01/Jan/89

2000

01/Jan/88

2500

01/Jan/88

Obs

01/Jan/87

KOHALA

01/Jan/87

01/Jan/86

Sim

01/Jan/86

1600

01/Jan/85

01/Jan/84

01/Jan/89

01/Jan/88

01/Jan/87

01/Jan/86

01/Jan/85

01/Jan/84

Sim

01/Jan/85

3000

01/Jan/84

1400

01/Jan/83

01/Jan/82

Discharge (m3/s) 1400

01/Jan/83

01/Jan/82

Discharge (m3/s)

1600

01/Jan/83

01/Jan/82

Discharge (m3/s)

1800

M.ABAD

Obs

1200

1000

800

600

400

200

0

DOMEL

1200 Obs

1000

800

600

400

200

0

4500 Obs

3500

Discharge (m3/s)

AZAD-PATTAN

Sim

4000

3000 2500 2000 1500 1000 500

2500

01/Jan/89

01/Jan/88

01/Jan/87

KOTLI

Sim Obs

2000

Discharge (m3/s)

01/Jan/86

01/Jan/85

01/Jan/84

01/Jan/83

01/Jan/82

0

1500 1000 500

01/Jan/89

01/Jan/88

01/Jan/87

01/Jan/86

01/Jan/85

01/Jan/84

01/Jan/83

01/Jan/82

0

Figure 8-7 Observed (doted) and simulated (solid) discharges at different stations for calibration (1982-89) in the Jhelum River basin

115

700 Sim

116 01/Jan/81

Sim

01/Jan/80

1600

01/Jan/81

600

01/Jan/81

01/Jan/80

01/Jan/79

01/Jan/78

Discharge (m3/s)

Sim

01/Jan/80

1400

01/Jan/79

01/Jan/78

Discharge (m3/s) 350

01/Jan/79

01/Jan/78

Discharge (m3/s)

400

NARAN

300

Obs

250

200

150

100

50

0

DOMEL

1200 Obs

1000

800

600

400

200

0

G.HABIB

Obs

500

400

300

200

100

0

3500 Sim

117 01/Jan/81

Sim

01/Jan/80

3000

01/Jan/81

3000

01/Jan/81

01/Jan/80

01/Jan/79

01/Jan/78

Discharge (m3/s)

Sim

01/Jan/80

2500

01/Jan/79

01/Jan/78

Discharge (m3/s) 1400

01/Jan/79

01/Jan/78

Discharge (m3/s)

1600

M.ABAD

1200 Obs

1000

800

600

400

200

0

KOHALA

Obs

2000

1500

1000

500

0

AZAD-PATTAN

Obs

2500

2000

1500

1000

500

0

1800 Obs

1400

Discharge (m3/s)

KOTLI

Sim

1600 1200 1000 800 600 400 200

01/Jan/81

01/Jan/80

01/Jan/79

01/Jan/78

0

Figure 8-8 Observed (doted) and simulated (solid) discharges at different stations for validation (1978-81) in the Jhelum River basin 8.3.3 Projected changes in stream flow Table 8-6 describes the future changes in seasonal flow with respect to baseline period under A2. In 2020s and 2050s, the flow in the spring season (MAM) is projected to decrease on all gauging stations. However, the flow in summer (JJA) and autumn (SON) is predicted to increase but with different magnitudes. On the whole, the mean annual flow could be increased in these two periods. The main stream gauge in the basin is Azad-Pattan where almost 87% of the Jhelum area contributes. At this gauge the flow is predicted to increase by 30 and 26% in 2020s and 2050s respectively. In 2080s, the same kinds of changes have been observed on all gauges, as in 2020s and 2050s, except at Domel in the summer. However, the mean annual changes could be increased in 2080s with 34% increase at Azad-Pattan. Table 8-7 shows the future flow changes in all the three periods on different sites in the Jhelum basin under B2 scenario. Almost similar in case of pattern and kinds of flow changes have been seen in all three periods but with different magnitudes. It is observed that in 2080s, the mean annual changes on all gauges are higher than A2 especially at Azad-Pattan but in the other two periods these are reverse.

118

Table 8-6 Future changes (%) in stream flow at different gauges relative to baseline (19611990) under A2 scenarios in the Jhelum River basin Season\Station

Naran

Garihab ibulla Stream flow (m³/s), 1961-1990 Winter 9.9 25.4 Spring 36.5 110.0 Summer 121.2 225.5 Autumn 23.8 58.1 Annual 47.9 104.7 Future changes (%) in 2020s Winter -70.4 0.0 Spring -56.0 -13.9 Summer 52.7 38.3 Autumn 103.7 109.1 Annual 7.5 33.3 Future changes (%) in 2050s Winter -71.4 -5.6 Spring -52.0 -25.8 Summer 50.7 45.7 Autumn 98.2 107.3 Annual 6.4 30.4 Future changes (%) in 2080s Winter -69.9 8.4 Spring -40.0 -11.9 Summer 61.3 49.2 Autumn 110.5 125.7 Annual 15.5 42.9

Muzaffar abad

Domel

Kohala

AzadPattan

Kotli

74.1 503.7 682.0 164.9 356.2

124.9 586.2 521.0 217.6 362.4

241.1 1212.5 1394.5 395.8 811.0

263.2 1259.0 1446.3 417.9 846.6

65.6 166.5 215.6 87.3 133.8

-15.0 -11.1 19.0 96.8 22.4

-12.7 -38.4 2.5 52.6 1.0

4.6 -21.6 22.6 115.4 30.2

6.9 -21.3 21.6 116.2 30.9

75.2 -34.9 -33.1 83.5 22.7

-14.8 -15.0 27.9 95.2 23.3

-11.5 -43.1 18.0 59.5 5.7

-6.5 -26.7 35.9 111.2 28.5

-10.5 -26.3 36.7 106.3 26.5

-36.0 -20.8 19.9 82.0 11.3

-7.9 -12.3 16.5 100.2 24.1

2.6 -39.0 -1.2 59.5 5.5

13.4 -23.3 22.0 121.2 33.3

16.5 -23.4 20.8 123.3 34.3

137.7 -53.9 -32.2 173.1 56.2

119

Table 8-7 Future changes (%) in stream flow at different gauges relative to baseline (19611990) under B2 scenarios in the Jhelum River basin Season\Station

Naran

Stream flow (m³/s), 1961-1990 Winter 9.9 Spring 36.5 Summer 121.2 Autumn 23.8 Annual 47.9 Future changes (%) in 2020s Winter -73.1 Spring -61.9 Summer 58.4 Autumn 88.4 Annual 2.9 Future changes (%) in 2050s Winter -72.5 Spring -57.5 Summer 56.1 Autumn 92.3 Annual 4.6 Future changes (%) in 2080s Winter -70.4 Spring -48.8 Summer 53.9 Autumn 96.8 Annual 7.9

Garihabib ulla

Muzaffar abad

Domel

Kohala

AzadPattan

Kotli

25.4 110.0 225.5 58.1 104.7

74.1 503.7 682.0 164.9 356.2

124.9 586.2 521.0 217.6 362.4

241.1 1212.5 1394.5 395.8 811.0

263.2 1259.0 1446.3 417.9 846.6

65.6 166.5 215.6 87.3 133.8

-2.2 -15.6 37.9 97.2 29.3

-15.8 -10.9 20.6 86.9 20.2

-9.2 -34.9 3.6 53.5 3.2

5.0 -19.7 23.9 108.6 29.5

7.1 -18.5 23.5 109.8 30.5

112.5 -11.3 -27.0 77.0 37.8

-11.2 -27.5 43.2 90.8 23.8

-23.7 -9.5 27.1 78.8 18.2

-14.0 -39.6 18.9 63.5 7.2

-12.0 -23.8 35.1 106.6 26.5

-14.6 -23.3 35.9 104.3 25.6

-38.3 -33.9 17.8 70.9 4.1

1.7 -10.2 45.9 130.8 42.1

-9.2 -11.2 18.0 102.5 25.0

-0.9 -42.1 0.4 74.5 8.0

11.7 -24.0 22.9 131.5 35.5

14.6 -23.8 22.5 132.6 36.5

143.8 -41.1 -18.6 213.1 74.3

8.3.4 Projected changes in flow duration curve Figure 8-9 shows the comparison of flow duration curves of baseline period against three future periods under A2 and B2 scenarios. Three indicators such as Q5 (high flow), Q50 (median flow), and Q95 (low flow) are used to find out the changes in low, median and high flow described in Table 8-8 and 7-6. According to the flow changes at Azad-Pattan stream gauge which contributes about 87% to the Mangla reservoir, under A2 and B2, the high flows in the Jhelum basin are projected to decrease in 2020s and 2080s by 1-7%, but increase in 2050s by 6-6.7%. The median flows are predicted to increase in all three periods by 2036% increase w. r. to baseline. On the other hand, the low flows could be increased in 2020s and 2080s by 1.7-7.75%, and decrease in 2050s under both A2 and B2 scenarios by 12-14%.

120

2000

G.Habibulla

1800

Stream flow (m³/s)

1600 1400

Baseline 2020s 2050s 2080s

1200 1000 800 600 400 200 0 0

10

20

30 40 50 60 70 Time equalled or exceeded (%)

80

90

2500

M.abad

Stream flow (m³/s)

2000 Baseline 2020s 2050s 2080s

1500 1000 500 0 0

10

20

2500

30 40 50 60 70 Time equalled or exceeded (%)

80

90

Domel

Stream flow (m³/s)

2000 Baseline 2020s 2050s 2080s

1500 1000 500 0 0

10

20

30 40 50 60 70 Time equalled or exceeded (%)

121

80

90

5000

Kohala

4500

Stream flow (m³/s)

4000 3500

Baseline 2020s 2050s 2080s

3000 2500 2000 1500 1000 500 0 0

10

20

5000

30 40 50 60 70 Time equalled or exceeded (%)

80

90

80

90

80

90

Azad Pattan

4500

Stream flow (m³/s)

4000 3500

Baseline 2020s 2050s 2080s

3000 2500 2000 1500 1000 500 0 0

10

20

3000

30 40 50 60 70 Time equalled or exceeded (%)

Kotli

Stream flow (m³/s)

2500 Baseline 2020s 2050s 2080s

2000 1500 1000 500 0 0

10

20

30 40 50 60 70 Time equalled or exceeded (%)

(A) 122

1400

G. Habibulla

Stream flow (m³/s)

1200 1000

Baseline 2020s 2050s 2080s

800 600 400 200 0 0

10

20

2500

30 40 50 60 70 Time equalled or exceeded (%)

80

90

80

90

80

90

M.abad

Stream flow (m³/s)

2000 Baseline 2020s 2050s 2080s

1500 1000 500 0 0

10

20

3000

30 40 50 60 70 Time equalled or exceeded (%)

Domel

Stream flow (m³/s)

2500 Baseline 2020s 2050s 2080s

2000 1500 1000 500 0 0

10

20

30 40 50 60 70 Time equalled or exceeded (%)

123

4000

Kohala

Stream flow (m³/s)

3500 3000 Baseline 2020s 2050s 2080s

2500 2000 1500 1000 500 0 0

10

20

4500

30 40 50 60 70 Time equalled or exceeded (%)

80

90

80

90

80

90

Azad Pattan

Stream flow (m³/s)

4000 3500 Baseline 2020s 2050s 2080s

3000 2500 2000 1500 1000 500 0 0

10

20

2000

30 40 50 60 70 Time equalled or exceeded (%)

Kotli

1800

Stream flow (m³/s)

1600 1400

Baseline 2020s 2050s 2080s

1200 1000 800 600 400 200 0 0

10

20

30 40 50 60 70 Time equalled or exceeded (%)

(B) Figure 8-9 Flow duration curves under (A) A2 and (B) B2 scenarios at different stream gauges in the Jhelum River basin 124

8.3.5 Changes in low, high and median flow Table 8-8 Future changes (%) in low, median and high flow relative to baseline (19611990) at different stream gauges under A2 scenarios in the Jhelum River basin Naran

Garihabibulla

Stream flow (m³/s) (1961-1990) Q5 170 316 Q50 19 54 Q95 8 21 Changes in 2020s w. r. to Baseline Q5 40.7 16.6 Q50 42.9 67.1 Q95 -98.8 -25.0 Changes in 2050s w. r. to Baseline Q5 38.7 22.1 Q50 43.5 54.7 Q95 -98.8 -22.2 Changes in 2080s w. r. to Baseline Q5 51.6 27.5 Q50 60.5 75.6 Q95 -98.8 -18.3

Muzaffarabad

Domel

Kohala

Azad Pattan

Kotli

1078 194 53

848 306 64

2142 606 151

2205 687 166

359 81 19

-4.5 72.1 -34.0

-16.5 1.2 0.1

-5.1 40.5 6.4

-5.5 30.1 6.0

8.5 -6.4 -42.6

1.3 65.2 -33.3

-3.4 1.8 1.2

4.1 32.1 -8.6

5.0 20.4 -12.5

11.5 17.1 -27.4

-4.6 66.0 -32.3

-22.9 5.7 4.0

-5.5 39.1 2.1

-6.0 29.6 1.7

18.5 14.3 -10.5

Q5 high flow, Q50 median flow, Q50 low flow Table 8-9 Future changes (%) in low, median and high flow relative to baseline (19611990) at different stream gauges under B2 scenarios in the Jhelum River basin Naran Garihabibulla Stream flow (m³/s) (1961-1990) Q5 170 316 Q50 19 54 Q95 8 21 Changes in 2020s w. r. to Baseline Q5 42.3 18.9 Q50 30.9 65.2 Q95 -98.8 -23.1 Changes in 2050s w. r. to Baseline Q5 46.0 21.0 Q50 42.9 50.6 Q95 -98.8 -24.6 Changes in 2080s w. r. to Baseline Q5 42.5 23.7 Q50 45.0 86.2 Q95 -98.8 -20.7

Muzaffarabad

Domel

Kohala

Azad Pattan

Kotli

1078 194 53

848 306 64

2142 606 151

2205 687 166

359 81 19

1.1 59.0 -35.5

-8.7 3.9 -3.2

-0.9 31.9 6.7

-1.2 23.4 7.7

3.9 32.7 -8.4

4.4 59.0 -38.8

-0.6 2.2 -2.3

6.0 32.9 -12.0

6.7 22.2 -14.1

14.1 2.9 -36.3

-6.7 72.5 -32.7

-17.7 7.7 5.1

-6.9 47.0 3.9

-7.3 36.6 5.9

25.3 47.2 1.1

8.3.6 Temporal shift and changes in magnitude of stream flow Temporal shifts of peak flows In Figure 8-10, mean daily flow (mean of 30 years) of baseline (1961-1990) is plotted against the mean daily flows of future periods (2020s, 2050s, and 2080s) to explore the temporal shifts and magnitudes of peak flows at different streamflow stations in the Jhelum river basin. At G. Habib, Domel, and Azad-pattan stream flow stations, a definite delay has been 125

predicted in all three future periods. A definite increase in magnitude of peak flows has been seen on the G. Habib, M.abad, and Azad-pattan in the all three future periods, under both A2 and B2 scenarios. At domel, peak flow magnitude is projected to less than the baseline peak flow. At Kotli stream gauge, no definite shift of peak has been observed. Nonetheless, the peak flow magnitude is projected to increase in 2050s under A2 and B2 and projected to decrease in other periods under both scenarios.

KUNHAR RIVER AT G.HABIB

Discharge (m³/s)

400

Baseline 2020s_A2 2050s_A2 2080s_A2 2020s_B2 2050s_B2 2080s_B2

300 200

100 0 Jan

Feb Mar Apr May Jun

Jul

Aug

Sep

Oct

Nov Dec

NEELUM RIVER AT M.ABAD 1200 Baseline 2020s_A2 2050s_A2 2080s_A2 2020s_B2 2050s_B2 2080s_B2

Discharge (m³/s)

1000 800 600 400

200 0 Jan

Feb Mar Apr May Jun

126

Jul

Aug Sep

Oct Nov Dec

JHELUM RIVE AT DOMEL

Discharge (m³/s)

800

Baseline 2020s_A2 2050s_A2 2080s_2 2020s_B2 2050s_B2 2080S_B2

600 400 200 0 Jan

Feb Mar Apr May Jun

Jul

Aug

Sep

Oct

Nov Dec

JHELUM RIVER ATAZAD PATTAN 2500

Baseline 2020s_A2 2050s_A2 2080s_A2 2020S_B2 2050s_B2 2080s_B2

Discharge (m³/s)

2000 1500 1000 500

0 Jan

Jul

Aug Sep Oct Nov Dec

POONCH RIVER AT KOTLI

400

Discharge (m³/s)

Feb Mar Apr May Jun

300

Baseline 2020s_A2 2050s_A2 2080s_A2 2020s_B2 2050s_B2 2080sB2

200 100 0 Jan

Feb Mar Apr May Jun

Jul

Aug Sep

Oct Nov Dec

Figure 8-10 Future shifts of peak flows in timing and magnitude, under A2 and B2, in the Jhelum river basin

127

Temporal shifts in center-of-volume date (CVD) To find out the impacts of climate change on the timing of stream flows, an indicator such as center-of-volume date—a date at which half of the total volume of stream flow passed though at a gauging station for a specific time period— is used in the present study and calculated according to the equation described in this study (Stewart et al., 2005). Table 8-10 shows the changes in CVD relative to baseline, under A2 and B2 scenarios, in three future periods at different gauging stations located in the Jhelum river basin. The positive values show delay flow, and negative values show early flows. The delay flows have been projected at all stations in the Jhelum river basin under A2 and B2 in all three future periods. The delay flows range between 7 (M. Abad) and 35 (Kotli) days at different stations. The column 3 and 4 (Table 8-10) show that about half of the flow, an averagely, of each year, in the Jhelum basin, passes through at different gauges by 22 June for the period of baseline (1961-1990). Nonetheless, it is predicted to shift in July. Figure 8-11 shows linear trends in center-of-volume date (CVD) at different streamflow stations for the baseline period (1961-1990) and three future periods (2020s, 2050s, and 2080s), under A2 and B2 scenarios, in the Jhelum river basin. At two stations (G. Habib and M. Abad), the linear trends are positive (upward slope) showing delay in flows. Conversely, linear trends on the other stations are negative (downward slope) showing earlier in flow, for the baseline period. In 2020s and 2080s, the linear trends are negative at almost all station under both scenarios, A2 and B2, and trends are positive at all stations in 2050s period under both scenarios. Table 8-10 Future changes in center-of-volume dates (CVD) with respect to baseline (1961-1990) at different stream flow stations under both scenarios, A2 and B2, in the Jhelum river basin River

Station

Kunhar Neelum Jhelum Jhelum Poonch

G. Habib M. Abad Domel Azad Pattan Kotli

CVD of Baseline Day of year 183 2-Jul 173 22-Jun 169 17-Jun 169 17-Jun 171 20-Jun

128

2020s 15(14) 13(12) 22(22) 20(19) 9(10)

2050s A2 (B2) Day 17(16) 14(11) 26(24) 22(21) 27(31)

2080s 9(16) 7(14) 18(27) 16(23) 32(35)

Kunhar river at G.Habib

Baseline 2020s_B2 2050s_B2 2080s_B2 2020s_A2 2050s_A2 2080s_A2 Linear (Baseline) Linear (2020s_B2) Linear (2050s_B2) Linear (2080s_B2) Linear (2020s_A2) Linear (2050s_A2) Linear (2080s_A2)

230 220

Day of year

210 200 190 180 170 160 150 1940

1960

1980

2000

2020

2040

2060

2080

2100

2120

Year

Neelum river at M. ABAD 230 220

Day of year

210 200 190 180 170 160 150 140 1940

1960

1980

2000

2020

2040

2060

2080

2100

2120

Baseline 2020s_B2 2050s_B2 2080s_B2 2020s_A2 2050s_A2 2080s_A2 Linear (Baseline) Linear (2020s_B2) Linear (2050s_B2) Linear (2080s_B2) Linear (2020s_A2) Linear (2050s_A2) Linear (2080s_A2)

Year

Jhelum river at Azad pattan 230 220

Day of year

210 200 190 180 170 160 150 140 1940

1960

1980

2000

2020

2040

2060

Year

129

2080

2100

2120

Azad Pattan 2020s_B2 2050s_B2 2080s_B2 2020s_A2 2050s_A2 2080s_A2 Linear (Azad Pattan) Linear (2020s_B2) Linear (2050s_B2) Linear (2080s_B2) Linear (2020s_A2) Linear (2050s_A2) Linear (2080s_A2)

Jhelum river at Domel 230 220

Day of year

210 200 190 180 170 160 150 140 130 1940

1960

1980

2000

2020

2040

2060

2080

2100

2120

Baseline 2020s_B2 2050s_B2 2080s_B2 2020s_A2 2050s_A2 2080s_A2 Linear (Baseline) Linear (2020s_B2) Linear (2050s_B2) Linear (2080s_B2) Linear (2020s_A2) Linear (2050s_A2) Linear (2080s_A2)

Year

Poonch river at Kotli 300

Day of year

250 200 150 100 50 1940

1960

1980

2000

2020

2040

2060

2080

2100

2120

Baseline 2020s_B2 2050s_B2 2080s_B2 2020s_A2 2050s_A2 2080s_A2 Linear (Baseline) Linear (2020s_B2) Linear (2050s_B2) Linear (2080s_B2) Linear (2020s_A2) Linear (2050s_A2) Linear (2080s_A2)

Year

Figure 8-11 Observed and predicted linear trends in center-of-volume date (CVD), under A2 and B2 scenarios, in the Jhelum river basin 8.3.7 Relations between changes in mean temperature, precipitation and flow Figure 8-12 shows the relationships between future changes in mean temperature, precipitation and flow for the three future periods under A2 and B2 scenarios. In this study, the flow at Azad-Pattan is used for comparison. It is seen that temperature is projected to increase in all seasons except in the winter (2020s). Precipitation is projected to decrease in winter and spring and increase in summer and autumn under both scenarios. The flow is predicted to increase in summer and autumn in all three periods under both scenarios. The annual temperature, precipitation and flow are almost projected to increase in the future. However, the increase in annual precipitation is negligible. In the summer and autumn, temperature has direct increasing relations in the future except in 2020s, under A2 and B2, and in spring the relationship is inverse in all future period as well as under both scenarios. On the other hand, precipitation has a directly increasing relationship with the flow in 130

summer and autumn but directly decreasing in spring in all future periods, under both scenarios.

40

0.4

0

0.0 -0.4 Aut

PREC FLOW TEMP

80

-40

120

0.0

-40

-0.4

A2_2080s

120

PREC FLOW TEMP

80

Spr

Sum

Aut

0.8

-40

-0.4 Win

1.2

PREC FLOW TEMP

0.0

120

0.0

1.6

0

1.2

0

Ann

0.4

160

0.4

Aut

40

1.6

40

Sum

B2_2050s

-0.4 Win

0.8

-40

-0.4 Spr

80

Ann

Prec and Flow (%)

160

Aut

0.0

1.2

0

Sum

0

160

0.4

Spr

0.4

1.6

0.8

0.8

40

Win

40

Win

80

Ann

A2_2050s

120

Prec and Flow (%)

Sum

Prec and Flow (%)

160

Spr

Tmean (°C)

Win

1.2

PREC FLOW TEMP

Spr

Sum

Aut

Ann 1.6

B2_2080s

1.2

PREC FLOW TEMP

80

0.8

40

0.4

0

0.0

-40

Ann

Tmean (°C)

0.8

-40

Prec and Flow (%)

120

Tmean (°C)

80

1.2

1.6

B2_2020s

Tmean (°C)

PREC FLOW TEMP

160

Tmean (°C)

Prec and Flow (%)

120

1.6

Prec and Flow (%)

A2_2020s

Tmean (°C)

160

-0.4 Win

Spr

Sum

Aut

Ann

Figure 8-12 Relationships between changes in mean temperature, precipitation and flow in 2020s, 2050s and 2080s in the Jhelum river basin 8.4

Conclusions

To assess the future changes in stream flow in the Jhelum river basin, downscaled temperature and prepetition data by SDSM, under A2 and B2 scenarios for the period of 2011-2099, was fed into the HEC-HMS to produce the future time series of stream flow. Then, these time series were divided into three periods: 2020s, 2050s, and 2080s and 131

compared with the baseline period (1961-1990) to get future changes. The main conclusions obtained from this study are given below: 

  

The upper Jhelum River basin is delineated by 30 m DEM feeding into HECGeoHMS. The whole basin is divided into 21 sub-basins. The characteristics of the sub-basins (areas, stream lengths, basin slopes etc.) are derived and fed into hydrologic model. A total of 14 meteorological and 7 stream gauges are used during the calibration and validation in HEC-HMS. The model is calibrated and validated for the period of 1982-1989 and 1978-1981 respectively. Nash-efficiency for calibration and validation ranges between 0.62-0.78 except at Kotli where E is less than 0.4. The stream flows in summer and autumn are projected to increase both under A2 and B2 and winter and spring are almost predicted to decrease in the future. However, the annual flow is projected to increase about 25 to 36% under A2 and B2.

132

9

CLIMATE CHANGE IMPACT ON THE HYDROPOWER

This chapter presents the data required for reservoir simulation. This data consists of discharge capacities of main outlets of Mangla reservoir, rule curves of reservoir, irrigation demand from reservoir etc. HEC-ResSim is used to find out the impacts of climate change on the hydropower produced from Mangla power plant, under raised and unraised conditions, as Mangla dam has been recently raised. The impacts of climate change on hydropower are also investigated under A2 and B2. Reservoir performance is investigated both, under raised and unraised conditions of dam. In addition, reservoir performance is also explored under A2 and B2 scenarios. 9.1

Data description

The physical data shown in Appendix D, Table 9-1, and the operational data shown in Table 9-2 and Table 9-3 are obtained from WAPDA. The inflow and outflow time series for the period of 1961-2008 are also provided by WAPDA. Table 9-1(A) shows the maximum discharge capacity of the main components of the Mangla reservoir before raising dam and Table 9-1 (B) gives discharge capacities after the raising of dam. Table 9-2 describes the maximum and minimum rule curves for the reservoir before and after raising dam respectively. Figure 9-1 shows the mean monthly generated power. Table 9-3 describes the irrigation demand required downstream of the Mangla reservoir which is used during the feasibility report of the Mangla Raising project (MJV 2001). Table 9-4 showing the mean monthly evaporation is also obtained from WAPDA. Table 9-5 describes the whole summary of data used in setting up the HEC-ResSim. It includes physical characteristics of the dam before and after raising. Table 9-1 Max discharge capacity of main outlets of Mangla dam at different elevations (A) before-raising project (B) after-raising project (A) 12 (m) 317 323 331 335 341 347 351 354 357 360 363 366 366 369 372 374 375 376

Main Spillway 9-gates

Emergency Spillway Uncontrolled Outlet

0 0 0 1076 4248 9486 12743 15348 18066 20360 22370 24211 24636 25825 27411 28600 28770 29336

0 0 0 0 0 0 0 0 0 0 0 0 0 991 3568 6513 7277 8693

133

Jari Outlet 1-gate m3/s

Power plant 10-gates 0 14 34 35 38 41 42 44 45 46 47 49 49 50 51 51 52 52

1274 1274 1274 1274 1274 1274 1274 1274 1274 1274 1274 1274 1274 1274 1274 1274 1274 1274

Total

1274 1289 1308 2386 5560 10801 14059 16666 19386 21680 23692 25534 25959 28141 32304 36439 37373 39356

(B)

(m) 317 323 333 335 341 347 351 354 357 360 363 366 366 367 369 370 372 373 375 376 378 379 379 381 383 384

Main Spillway 9-gates

Emergency Spillway Unct. Outlet

0 0 0 574 3,801 8,935 11,361 13,714 16,068 17,702 19,336 20,971 21,236 21,634 22,298 22,962 23,626 24,290 24,954 25,618 26,282 26,547 26,890 27,460 28,030 28,600

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 263 1,719 3,860 6,513

Jari Outlet Cont. m3/s 0 14 34 35 38 41 42 44 45 46 47 49 49 50 50 51 51 51 52 52 53 53 54 54 54 55

Power plant 10-gates

Total

1,274 1,274 1,274 1,274 1,274 1,274 1,274 1,274 1,274 1,274 1,274 1,274 1,274 1,274 1,274 1,274 1,274 1,274 1,274 1,274 1,274 1,274 1,274 1,274 1,274 1,274

1,274 1,289 1,308 1,883 5,114 10,250 12,678 15,032 17,387 19,022 20,658 22,294 22,559 22,958 23,623 24,287 24,951 25,616 26,280 26,945 27,609 27,875 28,481 30,507 33,218 36,442

WAPDA operation rule The operation rules for the Mangla reservoir are shown in the Table 9-2, and some important points are described below:    

The water level in the Mangla reservoir should be lower than 320 m MSL by 10 February. The water level should be held between 317 m and 320 m MSL till 31 March every year. The water level in the Mangla reservoir should reach its maximum conservation level 317.6 m (1042 ft) before 1st September if allowed by the inflow and irrigation demands. The main spillway should only be activated when the water is more than the maximum capacity of the power plant (1274 m3/s) which is only used to fulfill the irrigation demands (WAPDA, 2003; Ahmad, 2009).

Table 9-2 Max and min rule curve for the Mangla reservoir, before-raising and afterraising conditions Conservation level 10-Day Period

Before-raising 366 m Min (m) Max (m)

134

After-raising 379 m Min (m) Max (m)

Power (MW)

Oct. I Oct. II Oct. III Nov.I Nov.II Nov.III Dec.I Dec.II Dec.III Jan.I Jan.II Jan.III Feb.I Feb.II Feb.III Mar.I Mar.II Mar.III Apr.I Apr.II Apr.III May.I May.II May.III Jun.I Jun.II Jun.III Jul.I Jul.II Jul.III Aug.I Aug.II Aug.III Sep.I Sep.II Sep.II

1000 900 800 700 600 500 400 300 200

354 353 351 349 347 346 344 342 341 340 339 338 337 332 317 317 317 317 317 321 326 330 334 338 341 343 345 348 350 353 355 357 360 360 358 356

366 366 365 364 362 361 360 359 358 356 355 354 353 353 353 350 350 350 356 358 359 360 361 363 363 364 365 365 366 366 366 366 366 366 366 366

354 353 351 349 347 346 344 342 341 340 339 338 337 332 317 317 317 317 317 321 326 330 334 338 341 343 345 348 350 353 355 357 360 360 358 356

379 379 377 376 375 373 372 371 370 368 367 366 360 360 360 360 363 367 369 370 371 372 374 375 376 376 377 377 378 379 379 379 379 379 379 379

Power (MW)

Jan Feb Mar Apr May Jun

Jul Aug Sep Oct Nov Dec

Figure 9-1 Mean monthly energy generation from Mangla power station for the period of 1995-2000

135

Table 9-3 Irrigation indent/demand used during the feasibility report of Mangla raising project Rabi 10-Day Period Oct-I II III Nov-I II III Dec-I II III Jan-I II III Feb-I II III Mar-I II III

Adopted Indent (m3/s) 850 765 736 708 793 736 793 765 566 340 340 368 623 793 991 991 991 906

Kharif 10-Day Period Apr-I II III May-I II III Jun-I II III Jul-I II III Aug-I II III Sep-I II III

Adopted Indent (m3/s) 906 991 1,133 1,274 1,557 1,841 1,699 1,133 991 991 991 906 793 736 736 708 680 651

(MJV, 2001) Table 9-4 Mean monthly evaporation from Mangla reservoir Month Jan Feb Mar Apr May Jun

Evaporation(mm) 48 68 108 158 226 229

Month Jul Aug Sep Oct Nov Dec

136

Evaporation(mm) 157 123 111 94 66 46

Table 9-5 Summary of the physical and operational data used in HEC-ResSim for hydropower generation Before raising Physical Pool data 1) Flood level (374 m) data 2) Conservation level (366 m) 3) Inactive level (317 m) 4) Evaporation (Table 9-4) 5) Seepage (NA) 6) Elevation-storage-area curve (Appendix D) Dam 1) Elevation at the top of the dam (376 m) 2) Length of the top of the dam (2560 m) For controlled outlet 3) Number of gate (9 gates) 4) Elevation vs Max Capacity (Table 9-1) For uncontrolled outlet 5) Elevation (379.5 m) 6) Elevation vs Max Capacity (Table 9-1) Power 1) No. of outlets (10) Plant 2) Installed capacity (1000 MW) 3) Efficiency (90%) 4) Station use (NA) 5) Hydraulic losses (NA) 6) Tail water level (244 m) Operational Reservoir 1) Reservoir max and min guide data curves (Table 9-2) Power 1) Monthly or seasonally power Plant requirement—hydropower schedule (Figure 9-1)

9.2

After raising 1) Flood level (384 m) 2) Conservation level (379 m) 3) Inactive level (317 m) 4) Evaporation (Table 9-4) 5) Seepage (NA) 6) Elevation-storage-area curve (Appendix D) 1) Elevation at the top of the dam (385 m) 2) Length of the top of the dam (2560 m) For controlled outlet 3) Number of gate (9 gates) 4) Elevation vs Max Capacity (Table 9-1) For uncontrolled outlet 5) Elevation (379.5 m) 6) Elevation vs Max Capacity (Table 9-1) 1) No. of outlets (10) 2) Installed capacity (1000 MW) 3) Efficiency (90%) 4) Station use (NA) 5) Hydraulic losses (NA) 6) Tail water level (244 m 1) Reservoir max and min guide curves (Table 9-2) 1) Monthly or seasonally power requirement—hydropower schedule (Figure 9-1)

Methodology

To assess the impacts of climate change on the hydropower production under A2 and B2 scenarios, the simulated stream flow from HEC-HMS is used as input for HEC-ResSim. 9.2.1 Description of HEC-ResSim HEC-ResSim (Reservoir System Simulation) is the computer software designed by U.S. Army Corps of Engineer to help engineers and planners in projecting the behavior of the reservoir systems in water management studies. It is one of the advanced decision support tools available for water management systems. This model can be used for many reservoir process such flood control, irrigation, hydropower simulation, water supply, and environmental quality control. There are three separate modules—Watershed Setup, Reservoir Network, Simulation—which are used to specify different type of data within the basin.

137

Watershed Setup Module (WSM) is used to provide a common framework to create and define the watershed. This module encompasses all the stream line, reservoir, levees, impact areas, locations of gauging stations, locations of time series data, and hydrologic and hydraulic data. All these things together make a watershed framework once configured. Reservoir Network, the second main module, isolates the development of the reservoir model from the output analysis. This module is used to create a schematic network to describe the physical and operation data of the reservoir model, and to define different alternatives which have to be analyzed. Using the configurations as template which are built in WSM, the basis for reservoir network are created. Then different elements like routing reaches, diverted outlets, and diversions are added to complete the whole schematic network. The physical and operation data for each element of reservoir network is then stored. After that, the alternatives are defined which specify the reservoir network, initial conditions, operation sets, and assignments of the time series data using DSS files. After completion of the reservoir module, the reservoir model is needed to prepare for simulation which is done by the Simulation Module in HEC-ResSim. In this module, a simulation time window, a computation interval (e.g. hourly, daily or monthly etc.), and the alternatives are specified. Then the computations are performed and the results are analyzed in this module (HEC-ResSim, 2007). 9.2.2 HEC-ResSim setup for the Mangla dam As discussed above, the Mangla dam has recently been (2003-2011) raised by another 12 m (40 ft) that changed its physical characteristics. Therefore, in the present study, HEC-ResSim is setup for two different physical characteristics of the Mangla dam; 1) before raising and 2) after raising. However the operational rules are held as same. The following are the main steps to setup the HEC-ResSim for the Mangla dam: 1. First of all, the Watershed setup for the Mangla dam is configured by adding map layers of the Mangla watershed, stream gauges and stream lines (rivers). Then stream lines, junctions, and a reservoir (Mangla reservoir) are added to the watershed setup as shown in Figure 9-2.

138

Figure 9-2 HEC-ResSim’s setup of watershed module for Mangla watershed 2. The second step is the development of reservoir network for the Mangla watershed. In the present study, two reservoir networks are developed. One, before raising of the Mangla dam and second, after raising of the Mangla dam. However, schematic diagram for both the cases is same but the physical characteristics are different. First, the computation points (red dot) are defined as shown in Figure 9-3, and then the physical and operational data described in Table 9-5 is stored. The pool of the Mangla reservoir is divided into six zones: 1) Flood control (Table 9-5) 2) Conservation level (Table 9-5) 3) Upper rule curve 4) Average rule curve 5) Lower rule curve and 6) Inactive zone (Table 9-5). The upper and lower rule curves are shown in Table 9-2. The upper rule curve is used as guide curve for the reservoir. In this study, the hydraulic losses through the power plant are assumed to be negligible and the generation efficiency for the power plant is considered to be 0.9. The four main outlets (Main spillway, Emergency spillway, Jari dam, and Power plant) of the Mangla dam are added to the physical section of the Reservoir network module and their discharge capacities as shown in Table 9-1 are populated. The elevation-storage data (Appendix D) is also loaded for the pool of reservoir. As the preliminary objective of the Mangla reservoir is to fulfill the irrigation demands through the power tunnels, power production is only a byproduct. The two main rules (given below) are used to operate the reservoir. In the present study, 10 day time series are used to simulate the power generation because the Mangla dam is operated on a 10day basis. 139

 

So, the main rule to operate the reservoir is to fulfill irrigation demands which is describe in Table 9-3. Power tunnel cannot release water more than 1274 m3/s.

Figure 9-3 HEC-ResSim’s setup of reservoir network module for Mangla watershed 3. The simulation module is prepared for the Mangla dam to simulate the data both for before-raising and after-raising separately. 4. In the end, the percentage future changes in the three periods ((2020s, 2050s, and 2080s) relative to the baseline period (1961-1990) are obtained by comparing the baseline data with three future periods. 9.3

Results and discussions

The main objective of this chapter is to find out the impacts of climate change on the hydropower generation of the Mangla power plant, and the main objective of the Mangla dam operation is to fulfill the irrigation demand downstream of the Mangla. The power generation is the byproduct produced by fulfilling the irrigation demands. Since, the observed hydropower data is only available for the period of 1995-2000, therefore, the hydropower data is first simulated for the baseline period (1961-1990) and then, for the three future periods (2020s, 2050s, and 2080s) under A2 and B2 scenarios. In addition, since the 140

Mangla dam is raised by 12 m which changed the characteristics of the Mangal dam. So, the percentage changes in the future (under A2 and B2) relative to baseline are obtained separately for before-raising and after-raising of the Mangla dam. In the end, the power generation for the period of 1997-2000 from both before-raising and after-raising is compared to explore the benefits of raising the dam in term of hydropower only. 9.3.1 Before raising the Mangla dam Table 9-6 shows the percentage seasonal changes in hydropower in three future periods with respect to baseline (1961-1990) under A2 and B2 scenarios. These results show increases in all three future periods and in all seasons both under A2 and B2 but with different magnitudes. Under A2, the increase in annual hydropower ranges between 16.6 and 19.6%, and 16.7 and 20.4% under B2. These increases are the results of predicted increases in flow described in the previous chapter. It is also observed that the changes in hydropower under B2 are higher than A2 same like in case of flow (Chapter 7). Table 9-6 Future changes (%) in hydropower relative to baseline (1961-1990) under A2 and B2 from the Mangla plant before-raising conditions Seasons

1961-1990 (MW)

DJF MAM JJA SON Ann

416 799 927 688 707

2020S 39.7 18.0 6.1 14.6 19.6

A2 2050S 35.7 9.6 6.8 14.7 16.7

2080S 41.2 14.8 5.0 14.6 18.9

2020S 39.3 20.7 7.1 14.3 20.4

B2 2050S 34.8 10.2 7.1 14.7 16.7

2080S 41.1 17.2 6.4 14.7 19.9

9.3.2 After raising of the Mangla dam Comparison of simulated with observed power and reservoir water level To check the performance of model, the model was run for the period of 2005-2008, because only observed power was available for this period, and power and water level were simulated and compare with the observed data shown in Figure 9-4. The variations of the observed data are not well followed by the simulated data, because insufficient information (such as hydraulic loses, overload factor and efficiency of power plant etc.) was available for the power station. In addition, that was also the period of raising of Mangla dam. Nonetheless, the pattern is well followed by the model.

141

1400

Power (MW)

1200 1000 800 600 Obs

400

Sim

200 0

Month

(A) Reservoir water level (m)

380

360

340 Obs 320

Sim

300

Month

(B) Figure 9-4 Simulated (A) power generation and (B) reservoir water level against the observed data, for 2005-2008 Future changes in hydropower generation Table 9-7 described the percentage seasonal changes in hydropower in 2020s, 2050s and 2080s with respect to baseline (1961-1990) under A2 and B2 scenarios, under raised conditions of the Mangla dam. These results indicate projected increases in all three future periods and in all seasons both under A2 and B2 but with different magnitudes. Under A2, the increase in annual hydropower ranges between 13.6-15%, and 13.8-15.3% under B2. It is also observed that the changes in hydropower under B2 are higher than A2 similar to the case of before raising conditions.

142

Table 9-7 Future changes (%) in hydropower relative to baseline (1961-1990) under A2 and B2 at Mangla plant after-raising conditions Season DJF MAM JJA SON ANN

1961-1990 (MW) 494 916 989 764 791

A2 2050s 26.8 7.8 6.7 13.2 13.6

2020s 29.8 11.0 6.1 13.0 15.0

2080s 30.7 9.5 4.8 12.8 14.5

B2 2050s 26.5 8.1 7.3 13.4 13.8

2020s 29.3 11.9 7.3 12.8 15.3

2080s 30.6 10.4 6.0 13.2 15.1

Future changes in hydropower generation, increasing irrigation demand by 10, 20, 30% Table 9-8 shows the future changes in hydropower relative to baseline period by increasing irrigation demand by 10, 20, and 30% of existing irrigation demand under A2 scenario. By comparing the Table 9-7 (existing irrigation demand) and Table 9-8, it is observed that increasing irrigation demand by10, 20, and 30% results in an annual increase in percentage change by almost 8, 14, and 18%. Table 9-8 Future changes (%) in hydropower generation relative to baseline period under A2 scenario, increasing the irrigation demand by 10, 20, and 30%, at Mangla power station Changes in Power Generation (%) Season

19611990 (MW)

10% increase in ID 2020s

20% Increase in ID

30% Increase in ID

2050s

2080s

2020s

2050s

2080s

2020s

2050s

2080s

DJF

494

40.8

38.5

42.1

50.4

49.8

51.6

55.9

58.7

59.3

MAM

916

15.6

11.4

13.8

15.8

12.5

14.9

14.5

10.8

13.0

JJA

989

11.3

12.3

9.9

14.4

16.1

13.3

16.9

19.1

16.1

SON

764

23.6

24.4

23.1

33.4

35.3

32.7

41.9

45.1

41.1

ANN

791

22.8

21.7

22.2

28.5

28.4

28.1

32.3

33.4

32.4

ID: Irrigation demand 9.3.3 Comparison of power generation, before and after-raising To compare the power production before-raising and after-raising of the Mangla dam, the hydropower is simulated for the period of 1997-2000 for both conditions (Before-raising and After-raising) and the year 1996 is taken as the reference year. For comparison, five indicators shown in Table 9-9 are used in this study. These indicators show that by raising the Mangla dam, the power generation is increased by 10.8% and plant factor by 14%. However, in the feasibility report of the Mangla Raising Project (2001), it was described that the power generation will be increased by 10-15% by raising of the Mangla dam with the same reservoir operation used for before-raising. For more illustrative purposes, the graphical presentation of power generated, flow used for power, maximum power which can be generated, inflow, and outflow time series are shown in Figure 9-5. It is seen that after the raising of the dam, power generation has become smoother per time step (e.g., circles on Figure 9-5) especially when the reservoir receives low inflow. This is only because of increased reservoir capacity. 143

Table 9-9 Indicators for the comparison of Mangla dam with and without raising Before 93 14,468 603 0.6 709

Power Head (m) Energy Generated per Time Step (MWH) Power Generated (MW) Plant Factor Flow Power (m3/s)

Plant Factor power generated/Install capacity of Plant

144

After 101 16,212 676 0.7 732

% 7.9 10.8 10.8 14.3 3.1

Figure 9-5 Power generated (green), flow used for power (red), and max capacity of power plant (blue) for Before and After-raising of Mangla dam, for 1997-2000

145

9.3.4 Performance of reservoir under unraised, raised and climate change conditions In the context of the present study, the climate change impacts on the water resources and hydropower are evaluated in terms of relative changes. There are two types of indicators, quantitative and qualitative, used in different studies (Schaefli et al. 2007) to evaluate the performance of reservoir. However, in this study, a set of quantitative indicators is used for evaluating the hydropower production and its seasonal distribution. These indicators defined by Schaefli et al. (2007) are described in Table 9-10. Table 9-10 Description of reservoir performance indicators Indicator Name Total inflow Total turbine flow Efficiency Spill volume

Significance Reservoir inflow volume Turbine outflow volume Water use efficiency Volume passing through main spillway

Spill Dam overtop

SAI Production

Winter production

Spillway activation index Mean annual production

Mean winter production

Definition Sum of water inflow into the reservoir for the entire simulation period Sum of water released through turbine for the entire simulation period Total turbine outflow/Total inflow Sum spill water passing though the turbines over the entire simulated period Number of days with main spillway activation for entire simulated period Number of days with dam overtopping or emergency spill activation for entire simulated period Number of days with main spillway activation/ number of simulated years Sum of produced hydropower/number of simulated years Sum of hydropower production in winter/sum of hydropower production through entire simulated period

Change in performance indicators after raising Table 9-11shows changes in some quantitative performance indicators for reservoir. These indicators were obtained from simulated flows and hydropower data from HEC-ResSim for the period of 1961-1990, and changes in after-raising of the dam were calculated with respect before-raising. Some indicators such as spill volume, number of spills and SAI are decreased and other like efficiency and power production are increased after raising of the dam. This shows increasing performance of the reservoir after raising. Table 9-12 and Table 9-13 show the future changes in performance indicators relative to 1961-1990 (After raising of the dam) under A2 and B2 scenarios respectively. Almost all indicators show increments in all three future periods (2020s, 2050s, and 2080s) except efficiency. The main reason of decreasing efficiency is same power rule applied for future periods, however, the inflow is projected to increase. In addition, power capacity of the reservoir is also kept same for the future. The major thing to increase the efficiency of water use is to increase the water flow through the turbines. From these tables we can conclude that, we can increase the efficiency and power production by increasing the installed capacity of power plant or increasing the outflow (demand) through the turbines.

146

Table 9-11 Performance indicators for after-raising and before-raising of dam for 19611990, and their changes after-raising relative to before-raising Indicator

Before raising After raising Changes (%) 1961-1990 1961-1990 3 9 Total inflow (m ×10 ) 877 877 0.0 3 9 Total turbine flow (m ×10 ) 751 760 1.2 Efficiency (%) 85.6 86.9 1.5 3 9 Spill volume (m ×10 ) 174 109 -37.3 Spill (days) 2,940 1,947 -33.8 Dam over top (day) 0.0 0.0 0.0 SAI (day/year) 98 65 -33.8 Production (GW) 258 298 15.5 WinProd (%) 15.9 14.4 -9.2 SprProd (%) 29.8 28.4 -4.6 SumProd (%) 30.8 32.9 6.8 AutProd (%) 23.5 24.2 3.2 WinProd Winter production; SprProd Spring production; SumProd Summer production; AutProd Autumn production

Table 9-12 Future changes in performance indicators relative to 1961-1990 (after-raising), under A2 Indicator 9

3

Total inflow (10 m ) Total turbine flow (109 m3) Efficiency (%) Spill volume (109 m3) Spill (days) Dam over top (day) SAI (day/year) Production (GW) WinProd (%) SprProd (%) SamProd (%) AutProd (%)

After raising 1961-1990 877 760 87 109 1947 0.0 65 298 14.4 28.4 32.9 24.2

147

2020s 23.5

A2 (%) 2050s 28.4

2080s 21.3

5.0 -15.0

4.8 -18.4

1.5 -16.3

140 169 0.0 169 8.9 20.9 0.7 -9.8 0.0

180 182 0.0 182 8.0 20.4 -1.3 -8.6 1.0

147 152 0.0 152 4.5 24.4 2.0 -5.5 4.4

Table 9-13 Future changes in performance indicators relative to 1961-1990 (after-raising), under B2 Indicator 9

After raising 1961-1990 877 760 87 109 1947 0.0 65 298 14.4 28.4 32.9 24.2

3

Total inflow (10 m ) Total turbine flow (109 m3) Efficiency (%) Spill volume (109 m3) Spill (days) Dam over top (day) SAI (day/year) Production (GW) WinProd (%) SprProd (%) SamProd (%) AutProd (%)

2020s 26.1

B2 (%) 2050s 27.9

2080s 21.7

5.0 -16.8

4.8 -18.1

-1.8 -19.4

159.7 190.2 0.0 190.2 9.4 19.9 1.1 -9.2 -0.6

176.1 182.4 0.0 182.4 8.3 19.8 -1.2 -8.3 0.8

174.4 163.9 0.0 163.9 1.7 22.1 0.2 -10.0 0.1

Future Changes in water level after raising Figure 9-6a shows the comparison between reservoir water levels of before-raising and afterraising conditions of the dam, simulated for 1961-1990. This shows that the reservoir afterraising has better capability to store and release water. Variations in water level are also decrease after raising of the dam. For example, the number times water level goes towards dead are decreased after-raising of the dam. Figure 9-6b and Figure 9-6c show the comparison of water levels between baseline (afterraising) and three future periods under A2 and B2 scenarios respectively. Since, the inflow to the reservoir is projected to increase by 20-30%. So, the water level is projected to increase under both scenarios. So, power generation from the reservoir can be produce more in future by increasing the installed capacity of the power plant. 390

Reservoir water level (m)

380 370 360 350 340 330 320 310

Before raising

After raising

300 1

25

49

73

97

121

145

169

193

Month

(A)

148

217

241

265

289

313

337

390

A2 Scenarios

Reservoir water level (m)

380 370 360 350 340 330 320 310

1961-90 (Ater raising)

2020s

2050s

2080s

300 1

25

49

73

97

121

145

169

193

217

241

265

289

313

337

Month

(B) 390

B2 Scenario

Reservoir water level (m)

380 370 360 350 340 330 320 310

1961-90 (After raising)

2020s

2050s

2080s

300 1

25

49

73

97

121

145

169

193

217

241

265

289

313

337

Month

(C) Figure 9-6 Reservoir water levels between (A) before and after raising (B) baseline and three future period under A2 (C) baseline and three future period under B2 9.4

Conclusions

Mangla power plant is one of the biggest in Pakistan. Its installed capacity is 1000 MW, which is 15% of the total installed capacity (6599 MW) of hydropower generation of Pakistan. Recently, Pakistan has been under high energy crises. So, in this study, impacts of climate change are assessed on the hydropower generation from the Mangla power plant. Simulated stream flow data—for the period of 2020s, 2050s, and 2080s under A2 and B2 scenarios—was fed into the HEC-ResSim to generate the hydropower generation. Then the simulated future hydropower generation from HEC-ResSim was compared with the baseline simulated from the observed data to assess the changes in the hydropower. The main conclusions are given below:  

HEC-ResSim is used for the reservoir simulation modeling. Fulfillment of irrigation demand is the main operational rule for the Mangla dam. 149

    

The hydropower is simulated for future period as a byproduct of irrigation release for two different physical characteristics of dam; before raising (conservation level 366 m), after raising (conservation level 379 m). All seasons show definite increase in hydropower under both scenarios. After raising of the dam, the hydropower generation is increased by 11%. Performance of the reservoir is increased after raising of the dam Since the inflow to the reservoir is projected to increase by 20-30% in 21st century, it is concluded that more hydropower can be generated by the reservoir by increasing the installed capacity of power plant or by increasing the flow through the tubines.

150

10 SUMMARY, CONCLUSIONS, AND RECOMMENDATIONS 10.1 Summary Climate change has significant effects on the hydrologic cycle and causing considerable uncertainties in the availability of water at the regional, national, and basin levels. South Asia is more vulnerable to climate change impacts than other continents. Pakistan, presently, is facing water and electricity problem and will be more serious in the future. So, this study is based on assessing the impacts of climate change on stream flow and hydropower as an overall objective. The specific objectives cover the evaluation of GCMs, statistical downscaling of temperature (tmax and tmin) and precipitation, evaluation of two sub-models of SDSM, and impact of climate change on the stream flow and hydropower in the Jhelum River basin. The Jhelum River is the biggest tributary of the Indus River. The Jhelum basin has an area of 33,342 km2 located in Pakistan and India and the basin has very diverse climate with very hot, cool, dry, and wet seasons. In the winter, minimum temperature can drop down to 25°C, and in the summer, the max temperature rises up to 50°C. The Jhelum (with a mean temperature of 23.53°C) and Naran (with mean temperature of 6.14°C) are the hottest and coldest stations respectively in the study area. The month of July with a mean temperature of 23°C and January with 2.9°C, are the hottest and coldest months respectively. A mean temperature of 13.72°C is calculated for the entire basin. The observed daily hydrometeorological data, for the period of 1961-2009, (14 climate stations and 7 stream gauges) needed for this study was obtained from the Water and Power Development Authority of Pakistan (WAPDA), India Meteorological Department (IMD), and Pakistan Meteorological Department (PMD). Many GCMs have been developed so far but all these are impossible to use in this study for downscaling with SDSM. So, a suitable GCS is selected on the basis of vintage, validity, resolution, and representation of results, data availability, and some statistical analysis. Out of 8 GCMs, four GCMs (HadCM3, CCCma, CCSRIES, and CSISRO) are selected for the statistical analyses for evaluation. Monthly data of precipitation, Tmax and Tmin is obtained from the IPCC Data Distributed Center for A2 and B2 scenarios for the period of 1991-2009. Then, the three statistical indicators, root mean square error, coefficient of determination, and standard deviation are calculated from observed and GCM’s data. The results show great variability among all GCMs. In case of temperature, all models give comparable results. However in case of precipitation, the coefficient of determination shows low values for all GCMs except HadCM3. The R2 value for HadCM3_A2_G1and HadCM3_A2_G2 (G1 used for upper grid & G2 for lower grid) is 2.8 and 10.5%, respectively. The correlation coefficient of HadCM3_B2_G1 and HadCM3_B2_G2 is 0.9 and 1.1%, respectively, but all other models have low values for R2. On the other hand, the RMSE of HadCM3 is lower than other GCMs. The standard deviation (showing the variability of data) of HadCM3 is closer to the standard deviation of observed data relative to other GCMs. Therefore HadCM3 is selected for the downscaling of temperature and precipitation for further analysis. In this study, Statistical Downscaling Model (SDSM) is applied to downscale the maximum temperature, minimum temperature and precipitation in sub-basins—TPP and OPP—and the whole upper Jhelum River basin under A2 and B2 scenarios. 151

The big challenge of using SDSM model is the selection of suitable predictors for this complex mountainous area having a strong influence of monsoon season. For this purpose, a more quantitative approach is used by which the most prominent predictors can be ranked for calibration process reducing the effect of multiple co-linearity. It is concluded that the near surface large-scale (atmospheric) variables are the most suitable for downscaling of temperature. The most prominent predictor for the present study is the temp (large-scale mean temperature at 2 m height) in both sub-basins. As for precipitation, in the OPP basin (North-East and South-East part), the surface meridional velocity and surface vorticity at 500 hpa are the most influential atmospheric variables. In the TPP basin (South-West part), local precipitation is mostly effected by surface specific humidity and surface meridional velocity at 500 hpa. During the calibration and validation of SDSM, it is seen that the SDSM shows better capability to simulate temperature (Tmax and Tmin) in all three formats (daily, monthly and seasonal), with R2 ranging from 0.71 to 0.98—daily to seasonal. As for precipitation, SDSM downscales good results in case of monthly and seasonal with mean R2 ranging 0.42-0.62 (monthly to seasonal), but in the daily format it gives very poor results with R2 ranging between 0.8-0.15. It is concluded that in both (temperature and precipitation), the seasonal time series could be best simulated by SDSM than daily or monthly in the Jhelum River basin. The simulated changes in mean annual and seasonal temperate and precipitation for the periods of 2020s, 2050s and 2080s compared with the base period (1961-1990) show obvious different patterns under H3A2 and H3B2 scenarios in the TPP and OPP basin. The main finding is that the mean annual Tmax and Tmin are projected to increase in both parts of the basin, TPP and OPP, under both scenarios and in all three future periods—2020s, 2050s and 2080s. This increase in temperature is simulated to be higher under H3A2 scenario than H3B2 in the both sub-basins and higher in the TPP basin than OPP basin. Moreover, the increasing trend in mean annual temperature is found in the future from 2020s to 2080s. As for seasonal changes in mean annual temperature (Tmax and Tmin), the most increase is simulated in the spring season under both H3A2 and H3B2 scenario in the whole basin. However, the spring season in TPP basin and autumn in OPP basin are most affected seasons in case of rise in temperature. The mean annual precipitation is predicted to increase (1 to 3%) in the TPP basin and decrease (2 to 5%) in the OPP basin under both scenarios, with an overall decrease in the whole basin. Seasonal changes are different in all seasons in both parts of the basin. Summer (with 10 to 13% rise in precipitation) and winter (with 5 to 12% decrease in precipitation) are the most affected seasons in the TPP basin under both scenarios and in all three future periods. In the OPP basin, autumn shows a 5 to 12% increase in precipitation, and spring— peak season —shows 9 to15% decreases under both scenarios. As for the whole basin, summer and autumn are projected to receive more precipitation, and winter as well as spring could receive lesser amount of precipitation in the future, as compared to the baseline period. The spatial distribution of mean annual Tmax shows a rise in almost all part of the basin in the future periods relative to the baseline. However, the North-West parts are projected to face higher increase than South-East parts of the basin under both scenarios. The minimum temperature is projected to decrease in some patches of the basin but the major parts of the basin show rise in minimum temperature with respect to the baseline under both scenarios. In case of precipitation, the percentage changes are spread between -13 to 16% and -10 to 152

15% over the whole basin under H3A2 and H3B2 respectively. It is seen that almost half of the basin shows decreasing precipitation in 2020s, but in 2080s most parts of the basin are projected to decrease in precipitation under both scenarios. Both scenarios present a similar kind of spatial distribution patterns of mean annual Tmax, Tmin and precipitation changes in all three future periods but with different magnitude. However, these changes reported by H3A2 are higher than H3B2. In the next step, SDSM is developed in this study using the annual (SDSM-A) and monthly (SDSM-M) sub-models. The historical daily data of 30 years (1961-1990) is used to construct a strong statistical relationship between large scale and local scale variables in conjunction with a 10-year (1991-2000) data period for validation. The main challenge in the application of this method, apart from the mountainous topography and dominant monsoons in the study area is the screening of appropriate predictors. A quantitative approach is used by which the most prominent predictors to be used in the calibration process could be ordered. Six statistical indicators are used to verify the performance of downscaled data from both sub-models for calibration and validation periods. The validation results are found to be greatly improved with the application of bias correction on the downscaled data. The results (downscaled and bias corrected) of validation show that both SDSM-A and SDSM-M can be applied to simulate future climate under the H3A2 and H3B2 scenarios. The downscaled and bias corrected data (2011-2099) is divided into three periods: 20112040 (2020s), 2041-2070 (2050s) and 2071-2099 (2080s) and compared with the baseline period (1961-1990) to observe the changes. The comparison results show that the changes in Tmax and Tmin predicted by the two sub-models are different in magnitude but similar in their patterns. Both sub-models show an increase in Tmax and Tmin in the 2020s, 2050s and 2080s under the two scenarios. The results on seasonal basis also indicate an increasing trend in Tmax and Tmin in all four seasons under both scenarios. Both SDSM-A and SDSM-M show an increase in mean annual precipitation under H3A2 and H3B2 for all three future periods in the basin. All seasons except spring indicate increments in the 2020s, 2050s and 2080s relative to the baseline under both scenarios with the two sub-models. Autumn, according to SDSM-A, and winter, according to SDSM-M are projected to experience maximum percentage changes in the precipitation under both H3A2 and H3B2 scenarios. Spring season in the basin is predicted to have minimum changes in precipitation. SDSM-A most often projects higher changes in Tmax, Tmin and also in precipitation than SDSM-M under both scenarios. The changes in Tmax and Tmin simulated under H3A2 scenario by both sub-models are higher than H3B2. It is also seen that the increment in Tmax is higher than the increment in Tmin in mean annual and seasonal values under both scenarios except in autumn. It is interesting to note that the mean annual changes in precipitation simulated by both sub-models are higher under H3B2 than H3A2. It is concluded that although SDSM developed by annual sub-model (SDSM-A) performed well in predicting mean annual values, it cannot be used to find out the monthly and seasonal variations, especially with regard to precipitation unless bias correction is applied. In hydrologic modeling, delineation of physical characteristics of a basin is the initial step which is done by HEC-GeoHMS for the Jhelum basin creating 21 sub-basins from 30 m ASTER-DEM. Soil and land use data of 1km2 resolution is obtained by Harmonized World Soil Database v 1.2 and European Commission Joint Research Centre respectively. In this study, Penman-Monteith method is used for calculating the potential evapotranspiration for 153

both observed and future period and used in hydrologic model as input. In this study, HECHMS is used for simulation of stream flow for the future period (2011-2099), under A2 and B2. The HEC-HMS developed for this study includes initial and constant method as a loss method, SCS-unit hydrograph method as direct runoff, recession method as a base flow, Muskingum as a channel routing, and temperature index model as snow transfer method. The model is calibrated and validated for the period of 1982-89 and 1978-81 respectively. Three statistical indicators namely coefficient of determination, Nash-Sutcliffe efficiency (NC), and percent volume deviation (D) are used to check the performance model. The results are satisfactory comparing with other studies, with E ranging between 60-80% for all gauging sites. After this, future time series of stream flow for all sites are simulated for the period of 2011-2099. The stream flow is divided into three periods (2020s, 2050s, and 2080s) and compared with the baseline period (1961-1990) to assess the future seasonal changes in stream flow in the Jhelum River basin. The flow duration curves are also developed for three future periods and compared with the baseline. The high flow (Q5), low flow (Q95), and median flow (Q50) are also assessed in this study. The results show that the stream flows in summer and autumn are projected to increase both under A2 and B2 and winter and spring are almost predicted to decrease in future. However, the annual flow is projected to increase about 24 to 30% under A2 and B2 at Azad-Pattan stream gauge which contributes about 87% to the Mangla reservoir. According to the flow changes at Azad-Pattan stream gauge under A2 and B2, the high flows in the Jhelum basin are projected to decrease in 2020s and 2080s by 1-7%, but increase in 2050s by 6-6.7%. The median flows are predicted to increase in all three periods by 20-36% increase w. r. to baseline. On the other hand, the low flows could be increased in 2020s and 2080s by 1.7-7.75%, and decrease in 2050s under both A2 and B2 scenarios by 12-14%. Finally, the stream flow time series generated by HEC-HMS for the period of 2011-2099 are used as input to HEC-ResSim to find out the impacts of climate change on the hydropower from the Mangla power plant. HEC-ResSim (Reservoir System Simulation) is the computer software designed by U.S. Army Corps of Engineer to help engineers and planners in projecting the behavior of reservoir systems in water management studies. It is one of the advanced decision support tools available for water management systems. This model can be used for many reservoir process such flood control, irrigation, hydropower simulation, water supply, and environmental quality control. There are three separate modules— Watershed Setup, Reservoir Network, Simulation—which are used to specify different types of data within the basin. As the Mangla dam has recently been raised by another 12 m (40 ft) which has changed the physical characteristics of the dam. For example, before the raising the conservation level was 366 m (1202 ft) but after raising it has become 379 m (1242 ft) The model is setup for the Mangla dam for two situations: before raising and after raising. The pool of reservoir is divided into six operation zones: flood zone, conservation level, maximum rule curve, mean rule curve, minimum rule curve, and inactive level. In this study, the maximum rule curve is used as guide curve for the reservoir. Then all physical and operational data stored into the HEC-ResSim. The main operational objective is to fulfill the irrigation demands downstream. However, the hydropower is a byproduct of the irrigation demand. Since observed hydropower data is not available for the period of 1961-1990 for comparison with future simulated hydropower. Therefore, first the hydropower for the baseline period is simulated by feeding the observed inflow to the Mangla reservoir, and then for three future 154

periods (2020s, 2050s, and 2080s) feeding the stream flow simulated by HEC-HMS. The hydropower of the three periods is compared with the baseline to investigate the changes in hydropower in the future. In the end, the hydropower simulated under before-raising and after-raising of dam are compared to determine the impacts of the raising on the hydropower production for the period of 1997-2000. In case of before raising, the results show increases in all three future periods and in all seasons under both A2 and B2 but with different magnitudes. Under A2, the increase in annual hydropower ranges between 16.6 and 19.6%, and 16.7 and 20.4% under B2. In case of after-raising, the results indicate projected increases in all three future periods same as before-raising but with different magnitudes. Under A2, the increase in the annual hydropower ranges between 13.6-15%, and 13.8-15.3% under B2. It is also observed that the changes in hydropower under B2 are higher than A2 similar to the case of before raising conditions. In case of comparison of hydropower before and afterraising, it is seen that the power generation is increased by 10.8%. And after-raising of the dam, the variations in hydropower generation is decreased. 10.2 Conclusions 





 



None of GCMs is capable to capture the variations of observed data completely especially in case of precipitation. However, they give good correlation in case of temperature relative to precipitation. Based on the calculated statistics, vintage, validity, resolution, representation of results, and data availability, the HadCM3 model is selected for further analysis. The big challenge of using SDSM model is the selection of suitable predictors for this complex mountainous area having a strong influence of the monsoon season. For this purpose, a more quantitative approach is used. It is concluded that the near surface large-scale (atmospheric) variables are most suitable for downscaling of temperature. The most prominent predictor for the present study is the temp (largescale mean temperature at 2 m height) in both sub-basins. As for precipitation, in the OPP basin (North-East and South-East parts), the surface meridional velocity and surface vorticity at 500 hpa are the most influential atmospheric variables. In the TPP basin (South-West parts), local precipitation is mostly effected by surface specific humidity and surface meridional velocity at 500 hpa. The mean annual Tmax and Tmin are projected to increase in both parts of the basin, TPP and OPP, under both scenarios and in all three future periods—2020s, 2050s and 2080s. This increase in temperature is simulated to be higher under H3A2 scenario than H3B2 in both the sub-basins and higher in the TPP basin than OPP basin. The mean annual precipitation is predicted to increase (1 to 3%) in the TPP basin and decrease (2 to 5%) in the OPP basin under both scenarios with an overall decrease in the whole basin. The spatial distribution of mean annual Tmax shows a rise in almost all parts of the basin in the future periods relative to baseline. However, the North-West parts are projected to face higher increase than South-East parts of the basin under both scenarios. In case of the spatial distribution of mean annual precipitation, it is seen that almost half of the basin shows decreasing precipitation in 2020s, but in 2080s most parts of the basin are projected to decrease in precipitation under both scenarios. 155

 

   

 

Both scenarios present a similar kind of spatial distribution patterns of mean annual Tmax, Tmin, and precipitation changes in all three future periods but with different magnitudes. However, these changes reported by H3A2 are higher than H3B2. During the evaluation of annual and monthly sub-models of SDSM, it is concluded that monthly sub-model capture much better monthly and seasonal variations of observed data (temperature and precipitation) than annual sub-model. However, the annual sub-model gives comparable results in case of mean annual values. It is also concluded that the combination of annual sub-model and bias correction gives comparable results to monthly sub-model. The performance of HEC-HMS is acceptable to simulate the future projections of stream flow for the Jhelum River basin. The mean annual flow is projected to increase under A2 and B2 in the Jhelum River basin. The stream flow in summer and autumn are almost projected to increase but spring and winter are predicted to decrease in the Jhelum River basin. The high flows in the Jhelum basin are projected to decrease in 2020s and 2080s, but increase in 2050s. The median flows are predicted to increase in all three periods. On the other hand, the low flows could be increased in 2020s and 2080s, decreased in 2050s under both A2 and B2 scenarios. The mean annual and seasonal changes in hydropower are projected to increase under both scenarios, and under both situations (before-raising and after-raising of the Mangla dam). After-raising of the Mangla dam, the hydropower generation of the Mangla power plant has been increased by 10.8%.

10.3 Recommendations 10.3.1 Based on the study results  

It is suggested to downscale precipitation before simulation of stream flow instead of using precipitation from GCM directly for simulation of stream flow at the basin level. As there are obvious evidence of changes in temperature, precipitation, and stream flow, the water management planners must consider the climate change scenarios during water resources planning and managements.

10.3.2 For future research     

In this study, only one GCM is used. It is recommended to use different GCMs to consider the range of uncertainties. In this study, two sub-models of SDSM out of three are evaluated, so it is suggested to consider all three sub-models for evaluation. In this study, mean values are used to find out the impacts of climate change. The impacts of climate change on the extreme event are suggested to analyze in the Jhelum River basin. Thirty years period is used to find the mean changes in future. It is suggested to use 10 years period to analyze the changes under both scenarios to explore more details. In this study, A2 and B2 scenarios of HadCM3 which is used in 3rd Assessment Report are used. The results can be updated using the more recent data from CMIP5.

156



In this study, only future changes in hydropower are analyzed. It is recommended to run model under changing climate conditions by further raising of the dam or by further increasing of installed capacity of the dam.

157

REFERENCES Ahmad, I. (2009). Optimal control of multiple reservoir system under water scarcity. University of the Punjab. Ahmad, N. & Chaudhry, G. R. (1988). Irrigated agriculture of Pakistan. Lahore: S. Nazir. Akhtar, M., Ahmad, N., Booij, M. J. (2008). The impact of climate change on the water resources of Hindukush–Karakorum–Himalaya region under different glacier coverage scenarios. J Hydrol, 355 (1-4), 148-163. doi:10.1016/j.jhydrol.2008.03.015 Anandhi, A., Srinivas, V. V., Nanjundiah, R. S., Nagesh Kumar, D. (2007). Downscaling precipitation to river basin in India for IPCC SRES scenarios using support vector machine. Int J Climatol, 28, 401–420. doi:10.1002/joc.1529 Archer, D. R. & Fowler, H. J. (2008). Using meteorological data to forecast seasonal runoff on the River Jhelum, Pakistan. J Hydrol, 361 (1-2), 10-23. doi:10.1016/j.jhydrol.2008.07.017 Arnell, N. W. (2004). Climate change and global water resources: SRES emissions and socio-economic scenarios. Global Environ Change, 14 (1), 31-52. doi:10.1016/j.gloenvcha.2003.10.006 Arnell, N. W., Hudson, D. A., Jones, R. G. (2003). Climate change scenarios from a regional climate model: Estimating change in runoff in southern Africa. J Geophys Res-Atmos, 108 (D16), 4519. doi:10.1029/2002jd002782 Ashiq, M., Zhao, C., Ni, J., Akhtar, M. (2010). GIS-based high-resolution spatial interpolation of precipitation in mountain–plain areas of Upper Pakistan for regional climate change impact studies. Theor Appl Climatol, 99 (3), 239-253. doi:10.1007/s00704-009-0140-y Bates, B. C., Kundzewicz, Z. W., Wu, S., Palutikof, J. P. E. (2008) Climate Change and Water. Technical Paper of the Intergovernmental Panel on Climate Change. IPCC Secretariat, Geneva Benestad, R. E., Bauer, H. I., Chen, D. (2008). Empirical-statistical downscaling. New Jersey: World Scientific. Chu, J., Xia, J., Xu, C. Y., Singh, V. (2010). Statistical downscaling of daily mean temperature, pan evaporation and precipitation for climate change scenarios in Haihe River, China. Theor Appl Climatol, 99 (1), 149-161. doi:10.1007/s00704009-0129-6 Country-Profile (2005). Country Profile - Pakistan. United States Library of Congress. http://www.unhcr.org/refworld/docid/46f913520.html. Accessed 27 September 2012 2012 158

Diaz-Nieto, J. & Wilby, R. L. (2005). A comparison of statistical downscaling and climate change factor methods: impacts on low flows in the River Thames, United Kingdom. Clim Change, 69 (2), 245-268. doi:10.1007/s10584-005-1157-6 Dibike, Y. B. & Coulibaly, P. (2005). Hydrologic impact of climate change in the Saguenay watershed: comparison of downscaling methods and hydrologic models. J Hydrol, 307 (1-4), 145-163. doi:DOI 10.1016/j.hydrol.2004.10.012 Elshamy, M. E., Seierstad, I. A., Sorteberg, A. (2009). Impacts of climate change on Blue Nile flows using bias-corrected GCM scenarios, . Hydrol Earth Syst Sci, 13, 551565. doi:0.5194/hess-13-551-2009 Fealy, R. & Sweeney, J. (2007). Statistical downscaling of precipitation for a selection of sites in Ireland employing a generalised linear modelling approach. Int J Climatol, 27 (15), 2083-2094. doi:10.1002/Joc.1506 Fowler, H. J., Blenkinsop, S., Tebaldi, C. (2007). Linking climate change modelling to impacts studies: recent advances in downscaling techniques for hydrological modelling. Int J Climatol, 27 (12), 1547-1578. doi:10.1002/joc.1556 Fujihara, Y., Tanaka, K., Watanabe, T., Nagano, T., Kojiri, T. (2008). Assessing the impacts of climate change on the water resources of the Seyhan River Basin in Turkey: Use of dynamically downscaled data for hydrologic simulations. J Hydrol, 353 (1–2), 33-48. doi:http://dx.doi.org/10.1016/j.jhydrol.2008.01.024 Gagnon, S., Singh, B., Rousselle, J., Roy, L. (2005). An application of the Statistical DownScaling Model (SDSM) to simulate climatic data for streamflow modelling in Québec. Canad Water Resour J 30 (4), 297-314. doi:doi:10.4296/cwrj3004297 García, A., Sainz, A., Revilla, J. A., Álvarez, C., Juanes, J. A., Puente, A. (2008). Surface water resources assessment in scarcely gauged basins in the North of Spain. J Hydrol, 356 (3–4), 312-326. doi:10.1016/j.jhydrol.2008.04.019 Gebremeskel, S., Liu, Y. B., de Smedt, F., Hoffmann, L., Pfister, L. (2005). Analysing the effect of climate changes on streamflow using statistically downscaled GCM scenarios. Int J River Basin Manag, 2 (4), 271-280. doi:10.1080/15715124.2004.9635237 Gellens, D. & Roulin, E. (1998). Streamflow response of Belgian catchments to IPCC climate change scenarios. J Hydrol, 210 (1-4), 242-258. doi:10.1016/s00221694(98)00192-9 Ghosh, S. & Mujumdar, P., P (2008). Statistical downscaling of GCM simulations to streamflow using relevance vector machine. Advances in Water Resources, 31 132– 146 Giorgi, F., Christensen, J., Hulme, M., Storch, H. v., Whetton, P., Jones, R. et al. (2001) Regional climate information- evaluation and projections, Climate Change 2001: The Scientific Basis. Contribution of working group to the Third Assessment Report of the Intergouvernmental Panel on Climate Change. 159

Göncü, S. & Albek, E. (2010). Modeling Climate Change Effects on Streams and Reservoirs with HSPF. Water Resour Manag, 24 (4), 707-726. doi:10.1007/s11269009-9466-6 Goyal, M. K. & Ojha, C. S. P. (2010). Robust weighted regression as a downscaling tool in temperature projections. Int J Global Warming, 2 (3), 234-251. doi:10.1504/IJGW.2010.036135 Harpham, C. & Wilby, R. L. (2005). Multi-site downscaling of heavy daily precipitation occurrence and amounts. J Hydrol, 312 (1-4), 235-255. doi:10.1016/j.jhydrol.2005.02.020 Harrison, G. P. & Whittington, H. W. (2002). Susceptibility of the Batoka Gorge hydroelectric scheme to climate change. J Hydrol, 264 (1–4), 230-241. doi:http://dx.doi.org/10.1016/S0022-1694(02)00096-3 Hashmi, M. Z., Shamseldin, A. Y., Melville, B. W. (2011). Comparison of SDSM and LARS-WG for simulation and downscaling of extreme precipitation events in a watershed. Stoch Env Res Risk A, 25 (4), 475-484. doi:DOI 10.1007/s00477-0100416-x Hay, L. E. & Clark, M. P. (2003). Use of statistically and dynamically downscaled atmospheric model output for hydrologic simulations in three mountainous basins in the western United States. J Hydrol, 282 (1-4), 56-75. doi:10.1016/s00221694(03)00252-x Hay, L. E., Wilby, R. L., Leavesley, G. H. (2000). A comparison of delta change and downscaled GCM scenarios for three mountainous basins in the United States. J Am Water Resour As, 36 (2), 387-397. doi:10.1111/j.1752-1688.2000.tb04276.x HEC-HMS (2000) Hydrologic Modeling System HEC-HMS Techinical Reference Manual. Institute for Water Resources, Davis, USA HEC-HMS (2010) Hydrologic Modeling System HEC-HMS. User's Manual. Institute for Water Resources, Davis, USA HEC-ResSim (2007) Reservoir system simulation User's Manual. User's Manual, vol 3.0. Hydrologic Engineering Center, Davis, USA Hessami, M., Gachon, P., Ouarda, T. B. M. J., St-Hilaire, A. (2008). Automated regression-based statistical downscaling tool. Environ Model Software, 23 (6), 813834. doi:10.1016/j.envsoft.2007.10.004 Hewitson, B. C. & Crane, R. G. (1996). Climate downscaling: techniques and application. Clim Res, 07, 97-110. doi:citeulike-article-id:6985199 Hodgkins, G. A., Dudley, R. W., Huntington, T. G. (2005). SUMMER LOW FLOWS IN NEW ENGLAND DURING THE 20TH CENTURY1. J Am Water Resour As, 41 (2), 403-411. doi:10.1111/j.1752-1688.2005.tb03744.x

160

Huang, J., Zhang, J., Zhang, Z., Xu, C., Wang, B., Yao, J. (2011). Estimation of future precipitation change in the Yangtze River basin by using statistical downscaling method. Stoch Env Res Risk A, 25 (6), 781-792. doi:10.1007/s00477-010-0441-9 Huth, R. (2002). Statistical Downscaling of Daily Temperature in Central Europe. J Climate, 15 (13), 1731-1742. doi:10.1175/15200442(2002)0152.0.co;2 Ines, A. V. M. & Hansen, J. W. (2006). Bias correction of daily GCM rainfall for crop simulation studies. Agric Forest Meteor, 138 (1-4), 44-53. doi:10.1016/j.agrformet.2006.03.009 IPCC-DDC (2011). Global circulation model. IPCC. http://www.ipccdata.org/ddc_gcm_guide.html. 2012 IPCC-TGICA (2007) General guidelines on the use of scenario data for climate impact and adaptation assessment. vol 2. IPCC (2000) IPCC Special reposrt on the emission scenarios. IPCC (2007) Climate Change 2007: The physical science basis. In: Solomon S, Qin D, Manning M, Chen Z, Marquis M, Averyt KB, Tignor M, Miller HL (eds) Contribution of Working Group I to the fourth assessment report of the intergovernmental panel on climate change. Cambridge University Press, Cambridge Jacobusse, G. (2005) WinMICE User’s manual. Jones, R. G., Noguer, M., Hassell, D. C., Hudson, D., Wilson, S. S., Jenkins, G. J. et al. (2004) Generating high resolution climate change scenarios using PRECIS. Met Office Hadley Centre, UK Kannan, S. & Ghosh, S. (2011). Prediction of daily rainfall state in a river basin using statistical downscaling from GCM output. Stoch Env Res Risk A, 25 (4), 457-474. doi:10.1007/s00477-010-0415-y Kayani, S. (2012). Magla dam raising project (Pakistan): General review and socio-spatial impact assessment. Khan, M. S., Coulibaly, P., Dibike, Y. (2006). Uncertainty analysis of statistical downscaling methods. J Hydrol, 319 (1-4), 357-382. doi:10.1016/j.jhydrol.2005.06.035 Khattak, M. S. K. (2011). Impacts of Climate Change on the Water Resources and Irrigation Water Demand in the Upper Indus River Basin, Pakistan. Bangkok: Asian Institute of Technology. Labat, D., Goddéris, Y., Probst, J. L., Guyot, J. L. (2004). Evidence for global runoff increase related to climate warming. Ad Water Resour, 27 (6), 631-642. doi:10.1016/j.advwatres.2004.02.020 161

Lehner, B., Czisch, G., Vassolo, S. (2005). The impact of global change on the hydropower potential of Europe: a model-based analysis. Energy Policy, 33 (7), 839-855. doi:http://dx.doi.org/10.1016/j.enpol.2003.10.018 Li, X. & Sailor, D. (2000). Application of tree-structured regression for regional precipitation prediction using general circulation model output. Climate Research, 16 (1), 17-30. doi:10.3354/cr016017 Liu, J., Williams, J. R., Wang, X., Yang, H. (2009). Using MODAWEC to generate daily weather data for the EPIC model. Environ Model Software, 24 (5), 655-664. doi:10.1016/j.envsoft.2008.10.008 Liu, Z.-f., Xu, Z.-x., Liu, L.-l., Chen, Y.-n. (2008). Temporal trends of future maximum and minimum air temperature in the Tarim River Basin(in Chinese with English abstract). Arid Land Geogr, 31, 822-829. doi: CNKI:SUN:GHDL.0.200806-004 Liu, Z., Xu, Z., Charles, S. P., Fu, G., Liu, L. (2011). Evaluation of two statistical downscaling models for daily precipitation over an arid basin in China. Int J Climatol, n/a-n/a. doi:10.1002/joc.2211 López-Moreno, J. I., Goyette, S., Beniston, M. (2009). Impact of climate change on snowpack in the Pyrenees: Horizontal spatial variability and vertical gradients. J Hydrol, 374 (3–4), 384-396. doi:http://dx.doi.org/10.1016/j.jhydrol.2009.06.049 Mahmood, R. & Babel, M. (2012). Evaluation of SDSM developed by annual and monthly sub-models for downscaling temperature and precipitation in the Jhelum basin, Pakistan and India. Theor Appl Climatol, 1-18. doi:10.1007/s00704-012-0765-0 Markoff, M. & Cullen, A. (2008). Impact of climate change on Pacific Northwest hydropower. Clim Change, 87 (3-4), 451-469. doi:10.1007/s10584-007-9306-8 McBean, E. & Motiee, H. (2008). Assessment of impact of climate change on water resources: a long term analysis of the Great Lakes of North America. Hydrol Earth Syst Sci, 12 (1), 239-255. doi:10.5194/hess-12-239-2008 Mearns, L. O., Giorgi, F., Whetton, P., Pabon, D., Hulme, M., Lal, M. (2003) Guidelines for use of climate scenarios developed from regional climate model experiments Meenu, R., Rehana, S., Mujumdar, P. P. (2012). Assessment of Hydrologic Impacts of Climate Change in Tunga-Bhadra River Basin, India with HEC-HMS and SDSM. Hydrol Proces, n/a-n/a. doi:10.1002/hyp.9220 Miller, K. & Yates, D. (2006) Climate change and water resources: Aprimer for munucipal water providers. National Center for the Atmospheric Research, Boulder Milly, P. C. D., Dunne, K. A., Vecchia, A. V. (2005). Global pattern of trends in streamflow and water availability in a changing climate. Nature, 438 (7066), 347350. doi:10.1038/nature04312

162

Mirza, M. M. Q. & Ahmad, Q. K. (2005). Climate change and water resources in South Asia. 1 edn. London: Taylor & Francis. Mitchell, T. D. & Jones, P. D. (2005). An improved method of constructing a database of monthly climate observations and associated high-resolution grids. Int J Climatol, 25, 693-712. doi:10.1002/joc.1181 MJV (2001) Storage Sites Upstream of Mangla. Lahore Mujumdar, P., P & Ghosh, S. (2008). Modeling GCM and scenario uncertainty using a possibilistic approach: Application to the Mahanadi River, India. Water Resour Res, 44 (W06407). doi:10.1029/2007WR006137 Murphy, J. (1999). An Evaluation of Statistical and Dynamical Techniques for Downscaling Local Climate. Journal of Climate, 12 (8), 2256-2284. doi:10.1175/1520-0442(1999)0122.0.co;2 NCDC-NOAA. http://www.ncdc.noaa.gov/oa/climate/ghcn-daily/. 2011 Nguyen, V. (2002). Downscaling methods for evaluating the impacts of climate change and variability on hydrological regime at basin scale. In: International Symposium on Role of Water Sciences in Transboundary River Basin Management, Thailand, 2005. pp 1-8 Opitz-Stapleton, S. & Gangopadhyay, S. (2010). A non-parametric, statistical downscaling algorithm applied to the Rohini River Basin, Nepal. Theor Appl Climatol 103 (3-4), 375-386. doi:10.1007/s00704-010-0301-z Pal, I. & Al-Tabbaa, A. (2009). Trends in seasonal precipitation extremes – An indicator of ‘climate change’ in Kerala, India. J Hydrol, 367 (1–2), 62-69. doi:http://dx.doi.org/10.1016/j.jhydrol.2008.12.025 Pallant, J. (2007). SPSS Survival Manual: A step by step guide to data analysis using the SPSS for windows. PWG (2011). Pakistan Water Gateway (Indus basin). http://cms.waterinfo.net.pk/pdf/indusbasin.PDF. Accessed 05-10-2012 2012 PWP (2011). Monsoon of Pakistan. http://pakistanweatherportal.com/monsoon-ofpakistan/. Robinson, P. J. (1997). Climate change and hydropower generation. Int J Climatol, 17 (9), 983-996. doi:10.1002/(sici)1097-0088(199707)17:93.0.co;2-i Salzmann, N., Frei, C., Vidale, P.-L., Hoelzle, M. (2007). The application of Regional Climate Model output for the simulation of high-mountain permafrost scenarios. Global Planet Change, 56 (1-2), 188-202. doi:10.1016/j.gloplacha.2006.07.006

163

Schaefli, B., Hingray, B., Musy, A. (2007). Climate change and hydropower production in the Swiss Alps: quantification of potential impacts and related modelling uncertainties. Hydrol Earth Syst Sci, 11 (3), 1191-1205 Schoof, J. T. & Pryor, S. C. (2001). Downscaling temperature and precipitation: a comparison of regression-based methods and artificial neural networks. Int J Climatol, 21 (7), 773-790. doi:10.1002/joc.655 Shabeh ul, H., Mobushir Riaz, K., Valerio, L. (2012). XXI Century Climatology of Snow Cover for the Western River Basins of the Indus River System. CORD Conference Proceedings, Sharma, D. (2007). Downscaling of general circulation model precipitation for assessment of impact on water resurces at basin level. Bangkok: Asian Institute of Technology. Sharma, D., Gupta, A. D., Babel, M. S. (2007). Spatial disaggregation of bias-corrected GCM precipitation for improved hydrologic simulation: Ping River Basin, Thailand. Hydrol Earth Syst Sci, 4, 35-74 Smith, J. B. & Hulme, M. (1998) Climate Change Scenarios. Handbook on Methods for Climate Change Impact Assessment and Adaptation Strategies, 2 edn. UNEP/IES, Amsterdam Souvignet, M. & Heinrich, J. (2011). Statistical downscaling in the arid central Andes: uncertainty analysis of multi-model simulated temperature and precipitation. Theor Appl Climatol 1-16. doi:10.1007/s00704-011-0430-z Stewart, I. T., Cayan, D. R., Dettinger, M. D. (2005). Changes toward Earlier Streamflow Timing across Western North America. Journal of Climate, 18 (8), 1136-1155. doi:10.1175/jcli3321.1 Sunyer, M. A., Madsen, H., Ang, P. H. (2011). A comparison of different regional climate models and statistical downscaling methods for extreme rainfall estimation under climate change. Atmos Res. doi:10.1016/j.atmosres.2011.06.011 Timbal, B. & Jones, D. (2008). Future projections of winter rainfall in southeast Australia using a statistical downscaling technique. Clim Change, 86 (1), 165-187. doi:10.1007/s10584-007-9279-7 Tolika, K., Anagnostopoulou, C., Maheras, P., Vafiadis, M. (2008). Simulation of future changes in extreme rainfall and temperature conditions over the Greek area: A comparison of two statistical downscaling approaches. Global Planet Change, 63 (2–3), 132-151. doi:http://dx.doi.org/10.1016/j.gloplacha.2008.03.005 Tripathi, S., Srinivas, V. V., Nanjundiah, R. S. (2006). Downscaling of precipitation for climate change scenarios: A support vector machine approach. J Hydrol, 330 (3-4), 621-640. doi:10.1016/j.jhydrol.2006.04.030 USDA (2007) Nationa engineering handbook hydrology part 630. Nationa engineering handbook hydrology part 630. 164

Van Liew, M. W. & Garbrecht, J. (2003). Hydrologic simulation of the little Washita River Experimental watershed using SWAT1. J Am Water Resour As, 39 (2), 413-426. doi:10.1111/j.1752-1688.2003.tb04395.x Verma, A., Jha, M., Mahana, R. (2010). Evaluation of HEC-HMS and WEPP for simulating watershed runoff using remote sensing and geographical information system. Paddy and Water Environ, 8 (2), 131-144. doi:10.1007/s10333-009-0192-8 Vicuna, S., Leonardson, R., Hanemann, M. W., Dale, L. L., Dracup, J. A. (2008). Climate change impacts on high elevation hydropower generation in California’s Sierra Nevada: a case study in the Upper American River. Clim Change, 87 (1), 123-137. doi:10.1007/s10584-007-9365-x Viner, D. (2000). Modelling climate change. Climate Research Unit. http://www.cru.uea.ac.uk/cru/info/modelcc/. 2012 Wang, J. & Zhang, X. (2008). Downscaling and Projection of Winter Extreme Daily Precipitation over North America. J Climate, 21 (5), 923-937. doi:10.1175/2007jcli1671.1 WAPDA (2001) Mangla dam raising project: Feasibility study report. WAPDA (2003) Reservoir operation Mangla dam raising project. WAPDA (2008) WAPDA Annual Report 2007-2008. Wentz, F., J, Ricciardulli, L., Hilburn, K., Mears, C. (2007). How much more rain will global warming bring? Science, 317, 233-235 Wetterhall, F., A, Bárdossy, D., Chen, S. H., C.-Y, X. (2006). Daily precipitationdownscaling techniques in three Chinese regions. Water Resour Res, 42 (aa), W11423. doi:10.1029/2005WR004573. Wetterhall, F., Bárdossy, A., Chen, D., Halldin, S., Xu, C.-y. (2009). Statistical downscaling of daily precipitation over Sweden using GCM output. Theor Appl Climatol 96 (1), 95-103. doi:10.1007/s00704-008-0038-0 Wilby, R. L., Dawson, C. W., Barrow, E. M. (2002). SDSM — a decision support tool for the assessment of regional climate change impacts. Environ Model Software, 17 (2), 145-157. doi:10.1016/s1364-8152(01)00060-3 Wilby, R. L., Hay, L. E., Gutowski, W. J., Arritt, R. W., Takle, E. S., Pan, Z. et al. (2000). Hydrological responses to dynamically and statistically downscaled climate model output. Geophys Res Lett, 27 (8), 1199. doi:10.1029/1999GL006078 Wilby, R. L., Hay, L. E., Leavesley, G. H. (1999). A comparison of downscaled and raw GCM output: implications for climate change scenarios in the San Juan River basin, Colorado. J Hydrol, 225 (1-2), 67-91. doi:10.1016/s0022-1694(99)00136-5

165

Wilby, R. L., Whitehead, P. G., Wade, A. J., Butterfield, D., Davis, R. J., Watts, G. (2006). Integrated modelling of climate change impacts on water resources and quality in a lowland catchment: River Kennet, UK. J Hydrol, 330 (1-2), 204-220. doi:10.1016/j.jhydrol.2006.04.033 Wilby, R. L. & Wigley, T. M. (1997). Downscaling general circulation model output: a review of methods and limitations. . Prog Phys Geog, 21, 530-548. doi:10.1177/030913339702100403 Wilks, D. S. & Wilby, R. L. (1999). The weather generation game: a review of stochastic weather models. Prog Phys Geog, 23 (3), 329-357. doi:citeulike-article-id:3067708 Xu, C.-y. (1999). Climate Change and Hydrologic Models: A Review of Existing Gaps and Recent Research Developments. Water Resour Manag, 13 (5), 369-382. doi:10.1023/a:1008190900459 Xu, Y., Xu, C., Gao, X., Luo, Y. (2009). Projected changes in temperature and precipitation extremes over the Yangtze River Basin of China in the 21st century. Quatern Int, 208 (1–2), 44-52. doi:http://dx.doi.org/10.1016/j.quaint.2008.12.020 Yang, W., Bárdossy, A., Caspary, H.-J. (2010). Downscaling daily precipitation time series using a combined circulation- and regression-based approach. Theor Appl Climatol 102 (3-4), 439-454. doi:10.1007/s00704-010-0272-0 Yaoming, L., Qiang, Z., Deliang, C. (2004). Stochastic modeling of daily precipitation in China. J Geogr Sci, 14 (4), 417-426. doi:10.1007/bf02837485 Yimer, G., Jonoski, A., Griensven, A. V. (2009). Hydrological response of a catchment to climate change in the Upper Beles River basin, Upper Blue Nile, Ethiopia Nile Basin Water Engineering Scientific Magazine, 2, 11 Yu, P.-S., Yang, T.-C., Wu, C.-K. (2002). Impact of climate change on water resources in southern Taiwan. J Hydrol, 260 (1–4), 161-175. doi:http://dx.doi.org/10.1016/S0022-1694(01)00614-X Zhang, X. C., Liu, W. Z., Li, Z., Chen, J. (2011). Trend and uncertainty analysis of simulated climate change impacts with multiple GCMs and emission scenarios. Agric Forest Meteor, 151 (10), 1297-1304. doi:10.1016/j.agrformet.2011.05.010 Zhao, F. F. & Xu, Z. X. (2008). Statistical downscaling of future temperature change in source of the Yellow River Basin. Plateau Meteorol 27(1), 153–161 (in Chinese with English abstract) Zhao, G., Hörmann, G., Fohrer, N., Zhang, Z., Zhai, J. (2010). Streamflow Trends and Climate Variability Impacts in Poyang Lake Basin, China. Water Resour Manag, 24 (4), 689-706. doi:10.1007/s11269-009-9465-7 Zorita, E. & Storch, H. (1999). The Analog Method as a Simple Statistical Downscaling Technique: Comparison with More Complicated Methods. Journal of Climate, 12, 2474-2489. doi:citeulike-article-id:5312291 166

APPENDICES Appendix A Characteristics of GCMs used in Third Assessment Report (TAR) of IPCC GCM Name ECHAM4/OPYC3 (Germany) HADCM3 (UK)

Spatial resolution (deg.) 2.8 × 2.8 2.5 × 3.75

167

CSIRO-Mk2 (Australia)

3.2 × 5.6

NCAR-CSM (USA) NCAR-PCM (USA)

2.8 × 2.8 7.5 ×4.5

R30 (USA)

3.75 × 2.25

CGCM2 (Canada)

3.75 × 3.75

CCSRNIES (Japan)

5.6 25×5.625

Scenario A2 B2 A2 A2b A2c B2 A1 A2 B1 B2 A2 A1B A2 B2 A2 B2 A2 B2 A2 B2

Temporal resolution Monthly/6hr Monthly/6hr Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly/Daily Monthly/Daily Monthly Monthly

Temporal coverage 1990-2100 1990-2100 1950-2099 1960-2099 1860-2099 1950-2100 1961-2100 1961-2100 1961-2100 same 2000-2099 1980-2099 1980-2099 2000-2099 1961-2100 1961-2100 1900-2100 1900-2100 1890-2100 1890-2100

Spatial coverage Lat / long -90o – 90o /0.0o-36 90o – 90o /0.0o-36 90o – 90o /0.0o-36 90o – 90o /0.0o-36 90o – 90o /0.0o-36 -90o – 90o /0.0o-36 -87.56o – 87.56o /0.0o-360 -87.56o – 87.56o /0.0o-360 -87.56o – 87.56o /0.0o-360 -87.56o – 87.56o /0.0o-360 -87.56o – 87.56o /0.0o-360 -87.56o – 87.56o /0.0o-360 -90o – 90o / 0.0o-360o -90o – 90o / 0.0o-360o -90o - 90o / 0.0o - 360o -90o - 90o / 0.0o - 360o -87o - 87o / 0.0o 360o -87o - 87o / 0.0o 360o -87o - 87o / 0.0o 360o -87o - 87o / 0.0o 360o

Download availability Yes (Y) Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y

Available variables 29 29 19 12 12 19 23 23 23 23 9 12 12 12 12 12 21 19 19 19

Appendix B Correlation coefficient among 26 NCEP predictors

168

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26

Predictor ncepmslp ncepp_f ncepp_u ncepp_v ncepp_z ncepp_th ncepp_zh ncepp5_f ncepp5_u ncepp5_v ncepp5_z ncepp500 ncepp5th noepp5zh ncepp8_f ncepp8_u ncepp8_v ncepp8_z ncepp850 ncepp8th ncepp8zh ncepr500 noepr850 nceprhum ncepshum nceptemp

1 1 0.199 -0.233 0.071 0.098 -0.006 0.328 0.133 0.104 -0.207 0.225 -0.307 0.045 0.199 0.182 -0.185 0.202 -0.088 0.954 0.172 -0.174 -0.318 -0.053 -0.53 -0.737 -0.719

2 0.199 1 -0.856 0.748 -0.259 0.164 -0.358 -0.456 -0.489 -0.367 0.042 0.366 0.36 0.353 0.653 -0.646 0.543 -0.585 0.248 0.44 -0.458 -0.465 -0.294 -0.297 0.117 0.141

3 -0.233 -0.856 1 -0.346 0.114 -0.173 0.062 0.44 0.474 0.374 -0.21 -0.244 -0.353 -0.363 -0.612 0.877 -0.245 0.464 -0.287 -0.566 0.202 0.551 0.269 0.269 0.003 -0.006

4 0.071 0.748 -0.346 1 -0.366 0.278 -0.645 -0.376 -0.375 -0.245 -0.22 0.432 0.224 0.23 0.297 -0.111 0.715 -0.538 0.112 0.171 -0.608 -0.196 -0.248 -0.248 0.253 0.291

5 0.098 -0.259 0.114 -0.366 1 -0.073 0.462 0.365 0.33 0.299 0.369 -0.695 -0.242 -0.278 0.049 -0.078 -0.081 0.715 0.106 0.037 0.041 0.28 0.337 0.337 -0.461 -0.521

6 -0.006 0.164 -0.173 0.278 -0.073 1 -0.186 0.27 -0.231 -0.118 -0.075 0.183 0.036 0.111 -0.169 -0.111 0.154 -0.175 0.037 0.097 -0.116 -0.102 -0.121 -0.121 0.135 0.123

7 0.328 -0.358 0.062 -0.645 0.462 -0.186 1 0.357 0.343 0.065 0.416 -0.679 -0.139 -0.062 -0.167 0.003 -0.625 0.375 0.201 -0.103 0.582 0.005 0.266 0.266 -0.578 -0.652

8 0.133 -0.456 0.44 -0.376 0.365 -0.27 0.357 1 0.895 0.321 0.06 -0.577 -0.359 -0.298 -0.174 0.388 -0.206 0.501 0.025 -0.295 0.152 0.312 0.32 0.32 -0.424 -0.426

9 0.104 -0.489 0.474 -0.375 0.33 -0.231 0.343 0.895 1 0.209 -0.031 -0.506 -0.515 -0.193 -0.296 0.461 -0.256 0.454 0.013 -0.314 0.215 0.284 0.301 0.301 -0.366 -0.378

10 -0.207 -0.367 0.374 -0.245 0.299 -0.118 0.065 0.321 0.209 1 0.091 -0.29 -0.413 -0.954 -0.108 0.256 -0.064 0.41 -0.232 -0.184 -0.029 0.547 0.475 0.475 0.003 -0.045

11 0.225 0.042 -0.21 -0.228 0.369 -0.075 0.416 0.06 -0.031 0.091 1 -0.482 0.035 -0.068 0.249 -0.362 -0.153 0.299 0.169 0.173 0.11 -0.114 0.182 0.182 -0.428 -0.458

12 -0.307 0.366 -0.244 0.432 -0.695 0.183 -0.679 -0.577 -0.506 -0.29 -0.482 1 0.294 0.266 0.053 -0.116 0.205 -0.62 -0.125 0.116 -0.139 -0.259 -0.421 -0.421 0.758 0.831

169

13 0.045 0.36 -0.353 0.224 -0.242 0.036 -0.139 -0.359 -0.515 -0.413 0.035 0.294 1 0.397 0.268 0.31 0.141 -0.317 0.083 0.178 -0.091 -0.282 -0.233 -0.233 0.138 0.168

14 0.199 0.353 -0.363 0.23 -0.278 0.111 -0.062 -0.298 -0.193 -0.954 -0.068 0.266 0.397 1 0.122 -0.259 0.069 -0.385 0.221 0.178 0.023 -0.528 -0.459 -0.459 -0.013 0.036

15 0.182 0.653 -0.612 0.297 0.049 -0.169 -0.167 -0.174 -0.296 -0.108 0.249 0.053 0.268 0.122 1 0.64 0.47 -0.176 0.218 0.264 -0.439 -0.271 -0.107 -0.107 -0.079 -0.053

16 -0.185 -0.646 0.877 -0.111 -0.078 -0.111 0.003 0.388 0.461 0.256 -0.362 -0.116 -0.31 -0.259 -0.64 1 -0.132 0.155 -0.258 -0.641 0.128 0.501 0.246 0.246 0.097 0.074

17 0.202 0.543 -0.245 0.715 -0.081 0.154 -0.625 -0.206 -0.256 -0.064 -0.153 0.205 0.141 0.069 0.47 -0.132 1 -0.294 0.252 0.179 -0.919 -0.071 0.17 0.17 0.045 0.079

18 -0.088 -0.585 0.464 -0.538 0.715 -0.175 0.375 0.501 0.454 0.41 0.299 -0.62 -0.317 -0.385 -0.176 0.155 -0.294 1 -0.14 -0.116 0.229 0.396 0.32 0.32 -0.338 -0.365

19 0.954 0.248 -0.287 0.112 0.106 0.037 0.201 0.025 0.013 -0.232 0.169 -0.125 0.083 0.221 0.218 -0.258 0.252 -0.14 1 0.233 -0.218 -0.35 -0.104 -0.104 -0.629 -0.581

20 0.172 0.44 -0.566 0.171 0.037 0.097 -0.103 -0.295 -0.314 -0.184 0.173 0.116 0.178 0.178 0.264 -0.641 0.179 -0.116 0.233 1 -0.162 -0.361 -0.249 -0.249 -0.095 -0.064

21 -0.174 -0.458 0.202 -0.608 0.041 -0.116 0.582 0.152 0.215 -0.029 0.11 -0.139 -0.091 0.023 -0.439 0.128 -0.919 0.229 -0.218 -0.162 1 0.015 0.106 0.106 -0.016 -0.044

22 -0.318 -0.465 0.551 -0.196 0.28 -0.102 0.005 0.312 0.284 0.547 -0.114 -0.259 -0.282 -0.528 -0.271 0.501 -0.071 0.396 -0.35 -0.361 0.015 1 0.519 0.519 0.15 0.057

23 -0.053 -0.294 0.269 -0.248 0.337 -0.121 0.266 0.32 0.301 0.475 0.182 -0.421 -0.233 -0.459 -0.107 0.246 -0.17 0.32 -0.104 -0.249 0.106 0.519 1 1 -0.068 -0.263

24 -0.53 -0.297 0.269 -0.248 0.337 -0.121 0.266 0.32 0.301 0.475 0.182 -0.421 -0.233 -0.459 -0.107 0.246 -0.17 0.32 -0.104 -0.249 0.106 0.519 1 1 -0.068 -0.263

25 -0.737 0.117 0.003 0.253 -0.461 0.135 -0.578 -0.424 -0.366 0.003 -0.428 0.758 0.138 -0.013 -0.079 0.097 0.045 -0.338 -0.629 -0.095 -0.016 0.15 -0.068 -0.068 1 0.931

26 -0.719 0.141 -0.006 0.291 -0.521 0.123 -0.652 -0.426 -0.378 -0.045 -0.458 0.831 0.168 0.036 -0.053 0.074 0.079 -0.365 -0.581 -0.064 -0.044 0.057 -0.263 -0.263 0.931 1

Appendix C Characteristics of sub-basins of Jhelum River basin Sub-basin

170

1 2 3 4 5 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Total Area

Shape Length (m) 98712 385366 283626 31590 230671 393135 272544 231243 360974 629748 207764 273001 160636 46785 664537 143384 273287 468940 296936 212391

Area (km2) 200 2,394 1,874 28 897 1,442 1,201 1,278 1,927 6,672 665 1,283 361 46 5,022 405 1,539 3,739 1,324 1,011 33,308

Basin Slope (%) 49.95 63.02 33.23 37.71 22.20 55.24 20.32 34.82 53.45 30.92 19.60 5.42 12.77 12.49 57.92 45.77 41.73 40.74 60.19 54.76

Imp. (%) 4.99 10.93 5.12 5.00 5.07 23.73 6.87 10.55 5.05 9.02 10.14 6.34 27.47 64.55 25.24 4.98 4.99 5.89 13.13 37.92

Basin CN 70.94 72.10 74.91 69.36 83.63 78.63 78.68 78.77 74.46 77.93 84.51 80.99 82.17 91.36 75.97 77.02 81.12 77.79 72.52 80.47

Elevation (m ASL) Mean Max 2,708 636 4,978 665 4,311 1,530 1,441 630 1,694 316 5,385 1,556 4,570 1,500 5,056 1,469 4,337 671 5,191 1,392 1,301 316 854 330 1,304 214 562 316 6,266 1,766 2,911 573 3,396 421 4,697 464 4,994 803 5,195 2,418

Mean 1,672 2,762 2,873 1,036 1,005 3,420 2,987 3,174 2,469 3,220 809 592 759 439 3,990 1,742 1,909 2,500 2,858 3,704

Appendix D Elevation-storage relationship for the Mangla reservoir Elevation (m)

Storage (106 m3)

Elevation (m)

Storage (106 m3)

Elevation (m)

Storage (106 m3)

Elevation (m)

Storage (106 m3)

290 290 290 291 291 291 292 292 292 293 293 293 294 294 294 294 295 295 295 296 296 296 297 297 297 297 298 298 298 299 299 299 300 300 300 301 301 301 301 302

1 2 3 5 6 7 8 10 11 12 14 15 17 18 20 22 23 25 26 28 28 28 28 28 28 28 29 29 30 32 33 34 36 37 39 41 42 44 45 47

302 302 303 303 303 304 304 304 304 305 305 305 306 306 306 307 307 307 308 308 308 308 309 309 309 310 310 310 311 311 311 312 312 312 312 313 313 313 314 314

48 50 52 53 55 57 59 60 62 64 66 68 70 71 73 75 77 79 81 83 85 87 90 92 94 96 99 101 103 106 108 111 113 116 119 121 124 127 130 132

314 315 315 315 315 316 316 316 317 317 317 318 318 318 319 319 319 319 320 320 320 321 321 321 322 322 322 322 323 323 323 324 324 324 325 325 325 326 326 326

135 138 141 144 146 149 152 156 159 163 170 175 181 187 194 200 207 213 220 227 234 241 249 257 265 273 282 291 300 309 318 328 338 348 358 369 380 392 404 417

326 327 327 327 328 328 328 329 329 329 329 330 330 330 331 331 331 332 332 332 333 333 333 333 334 334 334 335 335 335 336 336 336 336 337 337 337 338 338 338

430 444 458 472 487 501 517 532 547 567 584 601 618 636 655 673 692 712 732 752 772 793 814 835 857 879 902 925 948 972 996 1020 1044 1069 1094 1120 1146 1172 1199 1226

171

Elevation (m) 339 339 339 340 340 340 340 341 341 341 342 342 342 343 343 343 344 344 344 344 345 345 345 346 346 346 347 347 347 347 348 348 348 349 349 349 350 350 350 351

Storage (106 m3) 1253 1282 1310 1340 1369 1400 1430 1461 1493 1525 1557 1590 1624 1658 1693 1728 1764 1800 1836 1873 1911 1949 1988 2028 2067 2108 2149 2190 2232 2275 2319 2363 2407 2451 2496 2542 2588 2634 2681 2728

Elevation (m) 351 351 351 352 352 352 353 353 353 354 354 354 354 355 355 355 356 356 356 357 357 357 358 358 358 358 359 359 359 360 360 360 361 361 361 361 362 362 362 363

Storage (106 m3) 2776 2824 2873 2922 2972 3022 3072 3123 3174 3226 3278 3330 3383 3437 3491 3546 3601 3656 3712 3768 3825 3882 3940 3998 4056 4114 4173 4232 4292 4351 4412 4472 4533 4594 4655 4717 4779 4842 4904 4968

Elevation (m) 363 363 364 364 364 365 365 365 365 366 366 366 367 367 367 368 368 368 369 369 369 369 370 370 370 371 371 371 372 372 372 372 373 373 373 374 374 374 375 375

172

Storage (106 m3) 5031 5095 5160 5225 5291 5356 5423 5489 5556 5623 5690 5765 5846 5927 6008 6088 6169 6250 6331 6412 6498 6584 6670 6756 6842 6927 7013 7099 7185 7271 7362 7453 7544 7635 7726 7817 7908 7999 8089 8180

Elevation (m) 375 376 376 376 376 377 377 377 378 378 378 379 379 379 379 380 380 380 381 381 381 382 382 382 383 383 383 383 384 384 384 385 385 385 386 386 386 386 387 387

Storage (106 m3) 8276 8372 8467 8563 8659 8754 8850 8946 9042 9137 9238 9338 9438 9539 9639 9740 9840 9940 10041 10141 10247 10353 10458 10564 10670 10776 10881 10987 11093 11199 11309 11420 11530 11641 11751 11862 11973 12083 12194 12304