Quantifying uncertainties in future climate change projections over India

QUANTIFYING UNCERTAINTIES IN FUTURE CLIMATE CHANGE PROJECTIONS OVER INDIA

RAM SINGH

CENTRE FOR ATMOSPHERIC SCIENCES INDIAN INSTITUTE OF TECHNOLOGY DELHI

APRIL 2019

©Indian Institute of Technology Delhi (IITD), New Delhi, 2019

QUANTIFYING UNCERTAINTIES IN FUTURE CLIMATE CHANGE PROJECTIONS OVER INDIA

by

Ram Singh

Centre for Atmospheric Sciences

Submitted

in fulfillment of the requirements of the degree of Doctor of Philosophy

to the

Indian Institute of Technology Delhi

April 2019

DEDICATED TO My Family

&

My Teachers

Certificate

This is to certify that the dissertation entitled “QUANTIFYING UNCERTAINTIES IN FUTURE CLIMATE CHANGE PROJECTIONS OVER INDIA” which is being submitted by Mr. Ram Singh to the Indian Institute of Technology Delhi for the award of the degree of Doctor of Philosophy is a record of bonafide work carried out by him under my sustained guidance and supervision. The dissertation has reached the standard fulfilling the requirements of the regulations relating to the degree. The results embodied in the dissertation have not been submitted to any other university or institute for the award of any degree or diploma.

Dr. Krishna AchutaRao New Delhi

Associate Professor April, 2019

Centre for Atmospheric Sciences Indian Institute of Technology Delhi New Delhi – 110016

Acknowledgements

The successful completion of any work that we take up in life is never an individual

effort. This thesis is no different. I would like to begin with thanking my supervisor, Dr. Krishna AchutaRao who has guided me throughout this work with not just valuable

scientific suggestions and moral encouragement, but being patient with me when I am stuck at some technical issue and unable to produce results. Without his support and encouragement, to think big and push my intellectual boundaries, this work would not have been possible. I express my sincere gratitude for his valuable guidance, constructive criticism, constant encouragement and affection throughout this research work.

I would like to take this opportunity to thank the successive Heads, Centre for Atmospheric Sciences, Indian Institute of Technology Delhi for providing me with a comfortable work place and all essential facilities in the Centre. I would also like to thank Dr.

S.B. Roy and my SRC (Student Research Committee) members - Prof. A.D Rao, Dr. Dilip Ganguly, and Prof. Ambuj Sagar - for their valuable suggestions during the course of this work. I am grateful to Prof. (Rtd) O. P. Sharma, Dr. H. C. Upadhyay (Rtd), Prof. (Rtd) S. K.

Dash, Prof. (Emeritus) Maithili Sharan, Prof. Manju Mohan, Dr. S. Dey, Dr. V. Pant, Dr. S.K Mishra, Dr. S. Sahany, Dr. Sandeep Sukumaran and Dr. Ravi Kumar Kunchala for their encouragement. I also thank the whole staff of Centre for Atmospheric Sciences for their help and support.

I would like to thank Dr. Anikender, Dr. Kanhu, Dr. Amit, Dhirendra, Himansu, Pushpraj, Shushant, Vijay, Abhishek Upadhayay, Abhishek Lodh, Ragi, Pawan, Puneet, Sarita, Piyush, Amit, Ankur, Gavendra, Saurabh, Sourangshu and colleagues for their active cooperation. I wish to express my thanks to Dileep, Sathiya, Rajeev, Roshni and Arulalan for their warm company as a part of Research Group, helpful suggestions, and advices offered at various stages of my Ph.D. I also convey special thanks to Dr. Aniket (NE-SAC), Dr.

Tarkeshwar (NCMRWF), Surendra (IMD), Dr. Hanuman (IIT Bombay), Chander Mohan IPS, Manish (BHEL), DK (PowerGrid), Vikas (ONGC), Rajesh kashnia, Sandip Karwasra, Krishna, Sandeep (MoLE), Anupriya, Himanshu, Jatin, Satish and Pawan with whom I shared my joy and sorrows during the long period of the research work.

I wish to express my gratefulness to Prof. James Hurrell (Colorado State University), Dr. Linda Mearns (NCAR) and Sean Santos (NCAR) for their help and suggestions. I would also acknowledge World Climate Research Programme’s (WCRP) Working Group on Climate Modelling for CMIP and Program for Climate Model Diagnosis and Intercomparison (PCMDI) for providing CMIP5 dataset and thanks to different modeling groups for producing and making available their model output. I also acknowledge the NCAR for provinding Community Earth System Model (various versions), NCAR’s Large Ensemble project community for CESM-LE and Swedish Meteorological and Hydrological Institute (SMHI) for CORDEX dataset. I also thank the CESM bulletin board for addressing the modeling issues faced during my PhD.

I gratefully acknowledge the financial support received from IIT Delhi, Ministry of Earth Sciences (GoI), CDKN and Climate Central. I sincerely acknowledge the High Pefromance Computing (PADUM) facility provided by IIT delhi and HPC help team for their support. I also aknowledge the CDAC – Pune for the HPC facility ParamYuva-II and their technical support.

I would like to express my love and gratitude to my family members who have supported and encouraged me through this journey. I would like to thank my parents, my brothers (Mange Ram, Satyawan and Prahlad Singh) and Sister-in-law (Pooja and Mamta) for their life-long support, everlasting love, and sacrifices, which sustained my interest in research and motivated me towards the successful completion of this study. I would be thankful to my wife Priyanka for her patience and immense support during this journey.

New Delhi Ram Singh

Abstract

Robust estimation of uncertainty in future climate change can help in improving climate models and designing appropriate policies and actions for mitigation and adaptation.

Uncertainty in climate change is different at the regional and sub-regional scales compared to global and continental scales. The work undertaken in this thesis is focused on understanding the uncertainties in climate change over India, home to a large population and considered to be one of the countries most vulnerable to climate change.

The thesis has three main objectives. The first is aimed at partitioning the uncertainty due to various sources over India and different homogenous regions within using coupled model output from CMIP5 with multiple realizations for each model. It is found that future temperature change over India is dominated by epistemic (model) uncertainty over aleatoric (internal variability) uncertainty that tends to increase with time although aleatoric uncertainty contributes significantly over sub-regional scales in the near term. Rainfall change for the JJA season has high uncertainty over various sub-regions due to more substantial aleatoric and epistemic uncertainties. It is also found that over certain regions aleatoric uncertainty remains larger during the entire 21st Century and dominates over epistemic uncertainty. A large ensemble of simulations from a single AOGCM (CESM-LE) is used to examine the role of internal variability, and it is found that aleatoric uncertainty is quite large and comparable to epistemic during the JJA and SON seasons. Aleatoric uncertainty dominates over epistemic in the DJF season rainfall during the much of 21st century with considerable decadal variability.

The estimation of aleatoric uncertainty using small number of ensemble members (as in the case of CMIP5 models) is studied using the CESM-LE ensemble. It is found that having a smaller number of ensembles results in an underestimate of the aleatoric uncertainty.

The second objective of the thesis is to explore the possibility of reducing epistemic/model uncertainty using performance-based model weighting strategies. It is found that some performance metrics (involving circulation parameters) do not add any value as the model weights are uniformly high across models. It is shown that the inclusion of both surface air temperature and precipitation variables is important as model weighting based on a single variable can be misleading. It is also found that observational uncertainty plays a crucial role in the model weights and its effect on the weighted change and uncertainty range is as much the different weighting strategies themselves.

The third objective of the thesis is to explore the impact of various dynamical downscaling techniques on the mean change and uncertainty over India. A set of 9 CMIP5 AOGCM projections are downscaled using an atmospheric general circulation model (CAM5.3; AGCM; Resolution 0.9°x1.25°) and compared against downloaded projections (of the same 9 AOGCMs) from a Regional Climate model (RCA4; CORDEX; 0.5°x0.5°

resolution). Results are compared against the driving AOGCM projections to examine the differences in change and uncertainty. Model simulations of present-day climate exhibit biases that carry the imprint of the model used for downscaling and not the driving AOGCM. The CAM5.3 exhibits comparatively lower spread across simulations as compared to the CMIP5 and CORDEX simulations. The CORDEX model projects a JJA season rainfall mean change and uncertainty pattern that is very similar to the driving CMIP5 models but the CAM5.3 exhibits a different spatial pattern with drastically reduced uncertainty. Precipitation change uncertainty shows a larger decadal scale variability during DJF, MAM and SON seasons as compared to the JJA season. It is concluded that dynamical downscaling using the CAM5.3 or CORDEX model lead to quite different regional climate change projections as well as different uncertainties as compared to the driving AOGCM and care needs to be exercised in interpreting and using the projections.

सार

भ%व'य के जलवायु प0रवत2न म5 अ7नि9चतता का मजबूत अनुमान जलवायु मॉडल को बेहतर बनाने

और शमन और अनुकूलन के Cलए उFचत नी7तयH और कायI को Jडजाइन करने म5 मदद कर सकता

है। जलवायु प0रवत2न म5 अ7नि9चतता वैि9वक और महाOवीपीय पैमानH कP तुलना म5 QेRीय और उप- QेRीय पैमानH पर अलग है। इस थीCसस म5 Vकए गए काय2 भारत म5 जलवायु प0रवत2न कP अ7नि9चतताओं

को समझने के Cलए है, जो एक बड़ी आबाद] के Cलए घर और जलवायु प0रवत2न के Cलए सबसे अFधक संवेदनशील देशH म5 से एक माना जाता है।

थीCसस के तीन मु`य उOदे9य है। पहला उOदे9य abयेक मॉडल के Cलए कई वाdत%वकताओं के साथ CMIP5 से युिjमत मॉडल आउटपुट का उपयोग करके भारत के भीतर %वCभlन समmप QेRH के Cलए अ7नि9चतता के %वCभlन nोतH का %वभाजन करना है। यह पाया गया है Vक भारत म5 भ%व'य के

तापमान प0रवत2न म5 oानाbमक (मॉडल) अ7नि9चतता, आंत0रक (आंत0रक प0रवत2नशीलता) अ7नि9चतता

पर हावी है, जो समय के साथ बढ़ता जाता है, हालांVक 7नकट अवFध म5 उप-QेRीय पैमानH पर aाकृ7तक अ7नि9चतता का काफP योगदान होता है। जेजेए सीजन के Cलए वषा2 प0रवत2न म5 %वCभlन उप-QेRH म5 उvच आंत0रक और oानाbमक संबंधी अ7नि9चतताओं के कारण उvच अ7नि9चतता है। यह भी पाया

जाता है Vक कुछ QेRH म5 संपूण2 21 वीं शता{द] के दौरान आंत0रक अ7नि9चतता अFधक बनी हुई है

और oानाbमक संबंधी अ7नि9चतता पर हावी है। एक एकल AOGCM (CESM-LE) से Cसमुलेशन का

एक बड़ा पहनावा आंत0रक प0रवत2नशीलता कP भूCमका कP जांच करने के Cलए उपयोग Vकया गया है, और यह पाया गया है Vक जेजेए और सओन सीजन के दौरान आंत0रक अ7नि9चतता काफP बड़ी और oानाbमक अ7नि9चतता के बराबर है। 21 वीं सद] के दौरान डीजेएफ सीज़न कP बा0रश म5 आंत0रक अ7नि9चतता दशकPय प0रवत2नशीलता के साथ oानाbमक अ7नि9चतता पर हावी है। CESM-LE का

उपयोग, कम सं`या वाले मॉडल (जैसा Vक सीएमआईपी 5 मॉडल के मामले म5) उपयोग कP तुलना

करते हुए आंत0रक अ7नि9चतता के अनुमान का अ„ययन Vकया गया आवर यह पाया गया Vक कम सं`या म5 असे…बल होने से आंत0रक अ7नि9चतता का lयूनानुमान होता ह†।

थीCसस का दूसरा उOदे9य aदश2न-आधा0रत मॉडल भा0रत रणनी7तयH का उपयोग करके oानाbमक/मॉडल अ7नि9चतता को कम करने कP संभावना का पता लगाना है। यह पाया गया है Vक कुछ aदश2न मेˆ‰Šस

(संचलन मापदंडH को शाCमल करते हुए) कोई मू‹य नह]ं जोड़ते ह† ŠयHVक मॉडल वजन मॉडल भर म5 समान mप से उvच ह†। यह ˆदखाया गया है Vक सतह के वायु तापमान और वषा2 दोनH चर को शाCमल करना महbवपूण2 है ŠयHVक एक एकल चर के आधार पर मॉडल का भार •ामक हो सकता है। यह भी

पाया गया है Vक पय2वेQी अ7नि9चतता मॉडल भार म5 महbवपूण2 भूCमका 7नभाती है और भा0रत प0रवत2न और अ7नि9चतता सीमा पर इसका aभाव उतना ह] अलग होता है िजतना Vक dवयं कP रणनी7त।

थीCसस का तीसरा उOदे9य भारत म5 औसत प0रवत2न और अ7नि9चतता पर %वCभlन ग7तशील डाउनdकेल तकनीकH के aभाव का पता लगाना है। 9 CMIP5 AOGCM अनुमानH का एक सेट एक वायुमंडल]य सामाlय प0रसंचरण मॉडल (CAM5.3; AGCM; 0रज़ॉ‹यूशन 0.9 °x1.25 °) का उपयोग करके

डाउनdकेल Vकया गया है और QेRीय जलवायु मॉडल (RCA4) से डाउनdकेल (CORDEX; 0रज़ॉ‹यूशन 0.5 °x0.5 °) Vकए गए अनुमानH (उसी 9 AOGCMs) के साथ तुलना कP गयी है। प0रवत2न और अ7नि9चतता म5 अंतर कP जांच करने के Cलए šाइ%वंग एओजीसीएम अनुमानH के मुकाबले प0रणामH कP तुलना कP गयी है। वत2मान म5 चल रहे जलवायु aदश2न के मॉडल Cसमुलेशन म5 झुकाव šाइ%वंग एओजीसीएम के बजाय डाउनdकेCलंग के Cलए उपयोग Vकए जाने वाले मॉडल कP छाप है। CAM5.3 Cसमुलेशन म5 CMIP5 और कॉड›Šस Cसमुलेशन कP तुलना म5 तुलनाbमक mप से कम अ7नि9चता है।

CORDEX मॉडल म5 जेजेऐ सीज़न वषा2 म5 प0रवत2न और अ7नि9चतता पैटन2 जो šाइ%वंग CMIP5 मॉडल के समान है, लेVकन CAM5.3 एक अलग dथा7नक पैटन2 aदCश2त करता है िजसम5 काफP कम अ7नि9चतता है। डीजेऍफ़, एमएएम और सोन सीजन के प0रवत2न म5 जेएजेए सीजन कP तुलना म5 प0रवत2न अ7नि9चतता एक बड़े पैमाने पर दशकPय प0रवत2नशीलता ˆदखाती है। यह 7न'कष2 7नकाला

गया है Vक CAM5.3 या CORDEX मॉडल का उपयोग करके डायनेCमक डाउनdकCलंग काफP Cभlन QेRीय जलवायु प0रवत2न अनुमानH के साथ-साथ šाइ%वंग एओजीसीएम कP तुलना म5 %वCभlन अ7नि9चतताओं को जlम देती है और अनुमानH कP Ÿया`या और उपयोग करने म5 देखभाल कP आव9यकता होती है।

Contents

Certificate i

Acknowledgments ii

Abstract iv List of Figures xii

List of Tables xx

Abbreviations xxi

Chapter 1: Introduction 1

1.1 Introduction 2

1.1.1 Background 2

1.1.2 Global and Regional Aspects of Climate Change 4

1.1.3 Introduction to Study Domain: India 8

1.1.3.1 Climate Profile of India 10

1.1.3.2 Evidence of Climate Change Over India 10

1.1.3.3 Factors Affecting the Indian Climate 11

1.1.4 Motivation 12

1.1.5 Objectives 19

1.1.6 Structure of The Thesis 19

Chapter 2: Segregation of Uncertainty and role of Internal Variability 21

2.1 Introduction 22

2.2 Datasets Used 25

2.3 Methodology 28

2.4 Results and Discussion 28

2.4.1 Surface Air Temperature Change and Uncertainty 29

2.4.2 Precipitation Change and Uncertainty 35

2.4.3 Role of Aleatoric Uncertainty in Future Projections 44 2.4.4 Temporal Variation of Aleatoric and Epistemic Uncertainty 49 2.4.5 The Impact of Ensemble Size on Aleatoric Uncertainty 52 2.4.6 Co-evolution of Temperature and Precipitation Uncertainty 54 2.4.7 Land-sea Temperature Contrast and Monsoon rainfall 59

2.5 Summary and Conclusions 62

Chapter 3: Impact of Model Performance Weighting on Mean Change and

Uncertainty Range 65

3.1 Introduction 66

3.2 Observed and Modeled Data 71

3.3 Methodology 77

3.4 Results and Discussion 85

3.4.1 Model Performance Evaluation and Weights 85 3.4.2 Projected Mean Change and Uncertainty Range 99

3.4.3 Sensitivity of the Mean and Uncertainty to Extended

Weighting Methodologies 106

3.4.4 Impact of Observational Uncertainty 111

3.4.5 Reliability of Change and Effective Number of models 115

3.5 Summary and Conclusions 117

Chapter 4: Impact of Dynamical Downscaling on future Change and

uncertainty 120

4.1 Introduction 121

4.2. Datasets, Methodology and Simulation Details 127

4.2.1 Datasets 127

4.2.1.1. Regional Climate Model: Rossby Centre regional

Atmospheric (RCA4) 128

4.2.1.2. Atmosphere General Circulation Model (AGCM):

Community Atmosphere Model (CAM5.3) 129

4.2.2 Methodology 132

4.3. Results and Discussion 133

4.3.1 AOGCM, RCM and AGCM Performance Evaluation 133 4.3.1.1 Temperature and Precipitation Bias 133 4.3.1.2 Temperature and Precipitation Variability 140 4.3.1.3 Model Simulated Annual Cycle Climatology 144 4.3.2. Projected Future Climate Change and Uncertainties 150

4.3.2.1 Spatial Pattern of Temperature Change and

Uncertainty Range 151

4.3.2.2 Spatial Pattern of Precipitation Change and

Uncertainty Range 156

4.3.3 Projections of Change and Uncertainty Over Homogenous

Climatic Zones 164

4.3.3.1 Surface Air Temperature Change and Uncertainty 164

4.3.3.2 Precipitation Change and Uncertainty 166

4.3.4 Temporal Characteristics of Uncertainty 171

4.4 Summary and Conclusions 174

Chapter 5: Conclusions and Scope of Future Work 178

5.1 Conclusions 179

5.2 Scope of Future Work 184

References 185

Curriculum Vitae 205

List of Figures

1.1 Observed change in the surface temperature (1.1A) for the period 1901-2012 and precipitation (1.1B) for the period 1951-2010, derived from the temperature/precipitation trends calculated using the linear regression from one dataset grid box with greater than 70% of complete and more than 20% of availability during the first and last 10% of the time period. + sign indicate the grid points where trend is significant at 10% level (Source: IPCC 2013; Collins et al. 2013; Synthesis report: Fig 1.1 Panel (b) and (e))

5

1.2 Spatial pattern of the CMIP5 multi-model mean change and standard deviation across models in the DJF season surface air temperature (Upper row) and JJA season precipitation (lower row) for the 20 year time period 2080-2099 relative to the reference period 1986-2005 under the RCP8.5 scenario. A set of 27 CMIP5 models is regridded to the common resolution 0.5ºx0.5º (resolution of CRU observation)

6

1.3 Homogenous Rainfall Zones (after Parthasarathy et al. 1994) 9 1.4 CMIP5 model simulated (A) DJF season temperature change and (B) JJA season

precipitation change averaged over All India relative to the 1986-2005 reference period under historical (1860-2005) and RCP4.5 and RCP8.5 respectively scenarios. Colored dots represent the individual models and colored bands represent the 5-95 percentile envelope (Gray: historical, light blue for RCP4.5 and Orange for RCP8.5). The 5-95 percentile envelope is calculated as the difference between 95^th and 5^th percentile value of the model anomalies (relative to 1986-2005) at each year. The light gray, dark blue and red lines represent the corresponding multi-model ensemble means. The different thick black lines represent the available observations. The box-whiskers represent the 5, 25, 75, and 95^th percentile values along with median and mean averaged for near (2020- 39) and far (2080-99) future time periods

15

2.1 Surface air temperature change with respect to 1986-2005 and uncertainties for the JJA season in CMIP5 models under RCP4.5 scenario. The individual homogenous zones (from figure 1.3) are demarcated in black outline. Columns are for future time 2020-39, 2040-59, 2060-79 and 2080-99. Top row shows the multi-model ensemble mean change. Second and third rows are aleatoric (internal variability) and epistemic (model) uncertainty respectively. Bottom row shows the ratio of aleatoric and epistemic uncertainty.

30

2.2 As in figure 2.1, but for RCP8.5 scenario. 31

2.3 As in figure 2.1, but for DJF season under RCP4.5. 31

2.4 As in figure 2.1, but for DJF season under RCP8.5 scenario. 33 2.5 MME mean surface air temperature change over the All India region (black

dashed line) shown with aleatoric (blue shading) and epistemic (orange shading) uncertainty ranges (calculated as 2*1.64 σ) for RCP4.5 and RCP8.5 scenarios (columns) and all seasons (rows). Mean change is calculated with 20-year

35

reference to 1986-2005. Uncertainty range envelope shown is ±1.64 σ. Note that the two uncertainty envelopes are overlaid on top of each other.

2.6 Seasonal mean climatology of rainfall for MAM, JJA, SON and DJF (columns) over the Indian region from IMD observations (top row) and MME mean of CMIP5 Historical runs (bottom row) for the 1986-2005 period. The homogenous rainfall zones from figure 1.3 are demarcated in black outline.

36

2.7 Precipitation change with respect to the 1986-2005 reference period for the JJA season in CMIP5 models under the RCP8.5 scenario in mm/day (top row) and in

% (bottom row). Columns from left to right are the four 20-year time periods 2020-39, 2040-59, 2060-79 and 2080-99 respectively.

37

2.8 Precipitation change (%) with respect to 1986-2005 and uncertainties over India for the JJA season in CMIP5 models under the RCP4.5 scenario. Columns are for future time periods 2020-39, 2040-59, 2060-79 and 2080-99. Top row shows the multi-model ensemble mean change. Second and third rows are aleatoric (internal variability) uncertainty (1.64σa) and epistemic (model) uncertainty (1.64σe) respectively. Bottom row shows their ratio. The homogenous rainfall regions from figure 1.3 are outlined in each of the figures.

39

2.9 As in figure 2.8, but for the RCP8.5 scenario. 40

2.10 As in figure 2.8, but for the DJF season. 40

2.11 As in figure 2.10, but for RCP8.5. 42

2.12 MME mean precipitation change (%) in the All India region (dashed line) along with aleatoric (blue shading) and epistemic (orange shading) uncertainty ranges (calculated as 2*1.64 σ) for RCP4.5 and RCP8.5 scenarios (columns) and all seasons (rows). Mean change is calculated with reference to 1986-2005 using 20-year moving windows (overlapping by all but one year) for the period 2006- 2099. Uncertainty range envelopes shown are ±1.64 σ. Note that the two uncertainty envelopes are overlaid on top of each other.

44

2.13 Ensemble mean change in Annual surface air temperature under RCP8.5 for the four time periods (columns) in the CESM-LE (top row) and CMIP5 (second row).

Contour lines in the top two rows indicate fraction of models with same sign of change as the mean. Aleatoric or internal variability uncertainty (1.64σa) for the four time periods is shown from CESM-LE (third row) and CMIP5 (bottom row).

The homogenous regions(from figure 1.3) are demarcated in the bottom two rows.

45

2.14 Ensemble mean change in precipitation (%) during JJA season under RCP8.5 for the four time periods (columns) in the CESM-LE (top row) and CMIP5 (second row). Contour lines in the top two rows indicate fraction of models with same sign of change as the mean. Aleatoric (internal variability) uncertainty (1.64𝜎_&) for the four time periods is shown from CESM-LE (third row) and CMIP5 (bottom row). The homogenous regions (from figure 1.3) are demarcated in the bottom two rows.

47

2.15 As in figure 2.14, but for the DJF season. 48

2.16 As in figure 2.14, but for the MAM season. 48

2.17 As in figure 2.14, but for the SON season. 49

2.18 Epistemic (model) uncertainty and aleatoric (internal variability) uncertainty in surface temperature change (℃) for the six homogenous zones (shown in figure 1) for the four seasons (JJA, SON, DJF, and MAM) and annual mean. The epistemic uncertainty lines (dark blue and red line for RCP4.5 and RCP8.5 scenario respectively from the CMIP5 ensemble) represent the +1.64𝜎₎values calculated for 20-year periods with a moving window of twenty years that overlap by all but one year. The aleatoric (internal variability) uncertainty lines (light red for CMIP5 and grey for the CESM-LE ensemble) are the +1.64𝜎_& values calculated for 20-year periods. Note that the vertical scales are different for the 6 homogenous zones.

50

2.19 As in figure 2.18, but for precipitation change (%). Note that the vertical scales are different for the 6 homogenous zones.

51

2.20 The effect of ensemble size on aleatoric (internal variability) uncertainty calculated through bootstrapping procedure. The shaded area represents the 5^th- 95^th percentile range from 1000 samples of centered standard deviations (multiplied by 1.64) calculated for different ensemble numbers N (N ranges from 3-39) taken from the 40-member CESM-LE ensemble. The dark line represents the median of the distribution. The ranges are plotted for India and the 6 homogeneous zones within (rows) during the JJA season for the four 20-year time periods 2020-39, 2040-59, 2060-79 and 2080-99 (columns).

53

2.21 Scatter plot of model simulated precipitation change (%) in the Westcentral region for the JJA season against temperature change ratio (the model simulated temperature change DT over the region scaled by global mean temperature change DTG) for the CESM-LE ensemble (top row) and the CMIP5 ensemble (bottom row) for the four time periods 2020-2039, 2040-2059, 2060-2079, and 2080-2099 (columns). Individual ensemble members are shown in open colored circles and the model mean in filled circles.

55

2.22 As in figure 2.21, but for SON season. 57

2.23 As in figure 2.21, but for MAM season. Note change of scale on both axes. 58 2.24 As in Figure 2.21, but for the DJF season in the Peninsula region. Note change

of scale on both axes. 59

2.25 Scatter plot of model simulated precipitation change (%) in the Westcentral region for the JJA season against change in Land-Ocean temperature contrast - estimated as the difference in the values between the South Asian subcontinent (SAS; 70–85E, 10–30N) and western Indian Ocean (WIO; 50–65E, 5S-10N) as in Roxy et al., 2015. The CESM-LE ensemble is shown in the top row and the CMIP5 ensemble in the bottom row for the four time periods 2020-2039, 2040- 2059, 2060-2079, and 2080-2099 (columns). Individual ensemble members are shown in open colored circles and the model mean in filled circles.

60

2.26 As in Figure 2.25, but for the All India region. Note change of scale on both axes. 61

3.1 Schematic diagram of REA_E to calculate model reliability or weight using the

various performance criteria.. 81

3.2 Observed and modeled mean climatology plotted against natural variability (for 1986-2005 period) of temperature and precipitation over the Peninsula region.

Panels A and B show values for temperature and precipitation respectively during the JJA season and panels C and D show corresponding values for the DJF season. The horizontal and vertical lines in different line styles are drawn denoting the location of observational datasets (two for temperature and three for precipitation) in each panel to enable easy comparison of observed and modeled values. Note differences in horizontal and vertical scales between panels.

85

3.3 As in figure 3.2, but for the Hilly region 86

3.4 The model temperature bias weights (f1) for the JJA season in the Peninsular region against reference datasets CRU and IMD (panels A and B respectively) and in the Hilly regions with reference dataset CRU and IMD (panels C and D respectively). The vertical lines in each panel indicate the value of natural variability in the observed dataset (values provided in the top right hand corner of each panel). Note differences in X-axis (temperature bias) between the two regions

87

3.5 The model precipitation bias weights (f3) for the JJA season for in the Peninsular region against reference datasets CRU, IMD, and APHRODITE (panels A, B, and C respectively) and in the Hilly regions with reference dataset CRU, IMD, and APHRODITE (panels D, E, and F respectively). The vertical lines in each panel indicate the value of natural variability in the observed dataset (values provided in the top right hand corner of each panel). Note differences in X-axis (precipitation bias) between the two regions

88

3.6 The model precipitation variability weights (f4) for the JJA season for in the Peninsular region against reference datasets CRU, IMD, and APHRODITE (panels A, B, and C respectively) and in the Hilly regions with reference dataset CRU, IMD, and APHRODITE (panels D, E, and F respectively). The vertical lines in each panel indicate the value of eCV in the observed dataset (values provided in the top right hand corner of each panel). Note differences in X-axis between the two regions

90

3.7 The model temperature bias weights (f1) in each season (rows) for the All India region against reference datasets CRU and IMD (columns). The vertical lines in each panel indicate the value of ε in the observed dataset (value of ε provided in the top right hand corner of each panel)

91

3.8 The model precipitation variability weights (f4) for each season (rows) in the All India region against reference datasets CRU, IMD, and APHRODITE (columns).

The vertical lines in each panel indicate the value of 𝜀_+, in the observed dataset (values provided at the top of each panel)

92

3.9 The observed (IMD, CRU and APHRODITE; first 3 panels, top row) and model simulated climatological (1986-2005 period) mean JJA precipitation (mm/day).

94

The 3 numbers in brackets next to the model names are the correlations with IMD, CRU and APHRODITE respectively

3.10 The observed (NCEP-NCAR Reanalysis; top left panel) and model simulated climatological (1986-2005 period) JJA mean sea level pressure (hPa). The numbers in brackets next to the model names are the correlations with NCEP- NCAR.

96

3.11 The observed (NCEP-NCAR Reanalysis; top left panel) and model simulated climatological (1986-2005 period) JJA mean zonal wind at 850hPa (ua850). The numbers in brackets next to the model names are the correlations with NCEP- NCAR.

97

3.12 Histogram showing number of models in each weight bin (0.1 bin width) for different weighting criteria during the JJA season over A) Peninsular region and B) Hilly regions. The four combinations of observational datasets (see Table 3.2) are shown in differently colored bars in each of the panels. Note that only weighting factors f1-f4 and f12 are different between the two regions.

98

3.13 The observed and model simulated climatological (1986-2005 period) annual cycle of rainfall over All India (the Indian land mass)

99

3.14 Weighted mean changes in tas (°C) for the four time periods from MME, REA_O, and REA_U methods using IMD&IMD observations under A) RCP4.5 scenario and B) RCP8.5 scenario respectively. Note that the MMEM does not require any observational data and the REA_O uses only the IMD observed temperature dataset to arrive at model reliability weights.

101

3.15 Unertainty range for surface air temperature change (°C) for the four time periods from MME, REA_O, and REA_U methods using IMD&IMD observations under the RCP8.5 scenario. Note that the MMEM does not require any observational data and the REA_O uses only the IMD observed temperature dataset to arrive at model reliability weights

102

3.16 Weighted mean changes in precipitation (%) for the four time periods from MME, REA_O, and REA_U methods using IMD&IMD observations under the A) RCP4.5 scenario and B) RCP8.5 scenario respectively. Note that the MMEM does not require any observational data and the REA_O uses only the IMD observed rainfall dataset to compute model weights

104

3.17 Uncertainty in precipitation change projections (%) for JJA season under the RCP8.5 scenario. Uncertainty range is calculated for 4 different time-slices (columns) from 2020 to 2099 and three approaches as spread in change around the mean (top row), REA_O and REA_U methods using IMD&IMD observations

105

3.18 Weighted change in JJA precipitation (%) under various weighting schemes using combinations of observations for four 20-year time slices from 2020-99 in A) Peninsular region and B) Hilly regions. Black filled circles represent the multi-model ensemble mean (MMEM) change and the bar represents the uncertainty range (±s). The red symbols show REA_O weighted change and associated uncertainty range for the three precipitation datasets. Symbols in blue represent REA_U and REA_E weighted change and bars the associated

107

uncertainty range. Note that for REA_U and REA_E, four observation combinations are shown

3.19 Weighted change in precipitation (%) and Temperature (℃) over Hilly region under various weighting schemes using combinations of observations for four 20- year time slices from 2020-99 in A) DJF precipitation change and B) JJA temperature change. Black filled circle represents the multi-model ensemble mean (MMEM) change and the bar represents the uncertainty range (±s). The red symbols show REA_O weighted change and associated uncertainty range for the three precipitation datasets. Symbols in blue represent REA_U and REA_E weighted change and bars the associated uncertainty range. Note that for REA_U and REA_E, four observation combinations are shown

108

3.20 Mean change in JJA precipitation (%) for the 2080-2099 time-period from the MMEM, REA_O, REAU, and REA_E_01 – REA_E_17 weighting schemes using the IMD&IMD observational data combination. Note that the MMEM does not require any observational data and the REA_O uses only the IMD observed rainfall dataset to arrive at model reliability weights

110

3.21 Uncertainty range in JJA precipitation (%) for the 2080-2099 time period from the MMEM, REA_O, REAU, and REA_E_01 – REA_E_17 weighting schemes using the IMD&IMD observational data combination

112

3.22 Weighted mean change in precipitation (%) for the 2080-99 period during A) JJA and B) DJF season under selected weighting schemes (top row MMEM; second row REA_O; third row REA_U; fourth row REA_E_08; bottom row REA_E_15) using four combinations of observational datasets (columns). Note that for MMEM there is no influence of observational datasets and the REA_O uses only the precipitation observations. The spatial mean change is shown at the top right hand corner of each of the panels

114

3.23 Collective reliability (as defined in eqn. 3.10) for the five seasons in the individual homogeneous regions shown for weighting combinations under REA_E_01 to REA_E_17 as well as REA_U (shown here as REA_E_00). The four observational combinations (as in Table 3.2) are each shown in separately

116

3.24 Effective number of models (as defined in eqn. 3.8) for the five seasons in the individual homogeneous regions shown for weighting combinations under REA_E_01 to REA_E_17 as well as REA_U (shown here as REA_E_00). The four observational combinations (as in Table 3.2) are each shown in separately

117

4.1 Model bias in DJF seasonal mean surface air temperature (℃) for the period 1986-2005 with reference to IMD observations over the Indian region for CMIP5 (top two rows), CORDEX (middle two rows) and CAM (bottom two rows).

Homogeneous zones are demarcated with thick black line.

135

4.2 Model bias in DJF season mean precipitation (mm/day) for the reference period 1986-2005 with respect to IMD observations over the Indian region for CMIP5 (top two rows), CORDEX (middle two rows) and CAM (bottom two rows).

Homogeneous zones are demarcated with thick black line.

136

4.3 Model bias in the JJA seasonal mean surface air temperature (℃) for the

reference period 1986-2005 with respect to IMD observations over Indian region 137

for CMIP5 (top two rows), CORDEX (middle two rows) and CAM (bottom two rows). Homogeneous zones are demarcated with thick black line.

4.4 Model bias in JJA season mean precipitation (mm/day) for the reference period 1986-2005 with respect to IMD observations over the Indian region for CMIP5 (top two rows), CORDEX (middle two rows) and CAM (bottom two rows).

Homogeneous zones are demarcated with thick black line.

138

4.5 Scatter plot of the observed and modeled variability against the climatological mean during 1986-2005 for surface air temperature (ºC; upper row) and precipitation (in mm/day; lower row) over the Hilly Regions in different seasons (columns).

140

4.6 As in fig. 4.5, but for the Northwest Indian region. 141

4.7 As in fig. 4.5, but for Peninsula Region. 143

4.8 Taylor diagram for spatial pattern of seasonal climatology of surface air temperature for reference period 1986-2005 over All India region for CMIP5, CORDEX and CAM simulations and CRU observations with reference to IMD observations for A) DJF, B) JJA, C) SON, D)MAM and E) Annual climatology.

145

4.9 Taylor diagram for spatial pattern of seasonal climatology of precipitation for reference period 1986-2005 over All India region for CMIP5, CORDEX and CAM simulations and observations from CRU and APHRODITE with reference to IMD observations for A) DJF, B) JJA, C) SON, D)MAM and E) Annual climatology.

146

4.10 Multi-model mean and spread as the difference between maximum and minimum for each month of climatological annual cycle of monthly rainfall (mm/day) simulated by CMIP5 (orange shading), CORDEX (blue shading) and CAM (grey shading) for the reference time period 1986-2005 over a) All India region, b) Peninsular region, c) West-central region, d) Northeast region, e) Central Northeast, f) Northwest region and g) Hilly region.

147

4.11 As in fig. 4.10, but for surface air temperature (℃). 149 4.12 Mean change and uncertainty range of surface air temperature (tas) relative to

1986-2005 during the DJF season under the RCP4.5 scenario. Columns represent the future time periods 2020-39, 2040-59, 2060-79 and 2080-99. The top 3 rows represent the mean change from CMIP5, CORDEX and CAM models respectively and lower 3 rows represent the uncertainty range.

150

4.13 As in fig. 4.12, but for RCP8.5. 151

4.14 As in fig. 4.12, but for JJA season and the RCP4.5 scenario. 153 4.15 As in fig. 4.12, but for JJA season under the RCP8.5 scenario. 154

4.16 As in fig. 4.12, but for MAM season under RCP8.5. 155

4.17 Mean change and uncertainty range of surface air temperature (tas) relative to 1986-2005 as described in fig. 4.12. but, for the SON season under the RCP8.5 scenario.

155

4.18 Mean precipitation change (%) relative to 1986-2005 and associated uncertainty range during the winter (DJF) season under the RCP4.5 scenario. Columns represent the time periods 2020-39, 2040-59, 2060-79 and 2080-99. The top 3 rows represent the mean change from CMIP5, CORDEX and CAM models respectively and lower 3 rows represent the uncertainty range.

157

4.19 As in fig 4.18, but for the RCP8.5 scenario. 158

4.20 As in fig 4.18, but for the JJA season. 159

4.21 As in fig 4.20, but under RCP8.5 scenario. 160

4.22 As in fig 4.20, but for the MAM season under RCP8.5 scenario. 161 4.23 As in fig. 4.20, but for the SON season under the RCP8.5 scenario. 163 4.24 Surface air temperature change and uncertainty range (ºC) calculated as 1.64*s

during the DJF season for time periods 2020-39, 2040-59, 2060-79 and 2080-99 (rows) as simulated by CMIP5 (red symbols) and dynamical downscaled models using RCA4 (blue symbols) and CAM (black symbols) averaged over various homogenous rainfall zones and All India. Open circles denote the RCP4.5 scenario and filled circles denote the RCP8.5 scenario.

165

4.25 As in fig 4.24, but for the JJA season. 166

4.26 Precipitation change and uncertainty range (%) calculated as 1.64*s during the DJF season for time periods 2020-39, 2040-59, 2060-79 and 2080-99 (rows) as simulated by CMIP5 (red symbols) and dynamical downscaled models using RCA4 (blue symbols) and CAM (black symbols) averaged over various homogenous rainfall zones and All India. Open circles denote the RCP4.5 scenario and filled circles denote the RCP8.5 scenario.

167

4.27 As in fig. 4.26, but for the JJA season. 169

4.28 As in fig. 4.26, but for the SON season. 171

4.29 Uncertainty upper bound (1.64s) for surface temperature change (℃) over the six homogeneous zones and All India for the four seasons (MAM, JJA, DJF and SON) and Annual mean from the CMIP5 (red line), CORDEX (blue line) and CAM (black lime) datasets. Darker shades represent the RCP8.5 and lighter shape RCP4.5. The uncertainty upper bounds (Equation. 2.2 in the Chapter 2) are calculated using a 20-year moving window (overlapping by all but 1 year) between 2006-2099. The uncertainty upper bound value is plotted at the mid- point of the 20-year period. Note vertical scales are different for each region.

172

4.30 As in fig. 4.29, but for precipitation change (%). 173

List of Tables

2.1 Details of CMIP5 models and model realizations used for the two variables (Surface Air Temperature and Precipitation) analysed.

27 3.1 Details of CMIP5 model experiments used shown with the realization

numbers for different variables analyzed.

76

3.2 The combinations of observations used for surface air temperature (tas) and precipitation (pr) are used to create set of observation for model performance evaluation. The IMD gridded surface air temperature data is from Srivastava et al., 2009 (a) and the IMD rainfall dataset is from Rajeevan et al., 2008 (b). Rainfall and Surface air temperature data CRU TS3.21 is a monthly gridded (0.5⁰ x 0.5⁰ horizontal resolution) dataset from Climate Research Unit (CRU) version 3.21 is from Harris et al., 2015(c).

The Aphrodite precipitation data is from Yatagai et al., 2012 (d). The NCEP—NCAR reanalysis from Kalnay et al., 1996 (e) is common across all combinations.

83

3.3 The weighting factors used in REA_U and different REA_E combinations.

The value of the exponent mi determines whether a weight is active (mi = 1) or inactive (mi = 0). The ^* symbol indicates that when individual homogenous zones weights are calculated, the annual cycle of the homogeneous zone rainfall (HZR) is used and when grid point weights are being computed, the annual cycle of All India Rainfall (AIR) is used.

84

4.1 List of AOGCMs analysed and source of downscaled model data. 128

Abbreviations

AGCM Atmospheric General Circulation Models AOGCM Atmosphere-Ocean General Circulation Models AMWG Atmospheric Model Working Group

APHRODITE Asian Precipitation - Highly-Resolved Observational Data Integration Towards Evaluation

AR4 Fourth Assessment Report AR5 Fifth Assessment Report CAM Community Atmosphere Model CESM Community Earth System Model

CESM-LE Community Earth System Model Large Ensemble CNEIndia Central Northeast India

CMIP Coupled Model Inter-comparison Project CORDEX COrdinated Regional Downscaling EXperiment CRU Climate Research Unit

CV Coefficient of Variation DOCN Data Ocean

ENSO El-Niño Southern Oscillation GCM General Circulation Models GHG Green House Gas

HHPC Hybrid High-Performance Computing HIRLAM High Resolution Limited Area Model

HRIndia Hilly Region India

IITD Indian Institute of Technology Delhi IITM Indian Institute of Tropical Meteorology IMD Indian Meteorological Department IOD Indian Ocean dipole

IPCC Intergovernmental Panel on Climate Change ISMR India Summer Monsoon Rainfall

IS92 IPCC scenarios 1992

ITCZ Inter Tropical Convergence Zone IV Inter-annual Variability

MME Multi Model Ensembles MMEM Multi Model Ensemble Mean

NCEP National Centers for Environmental Predictions NCAR National Center for Atmospheric Research NEIndia Northeast India

NOAA National Oceanic and Atmospheric Administration NV Natural Variability

NWIndia Northwest India

PDF Probability Density Function PIndia Peninsular India

PRECIS Providing REgional Climates for Impacts Studies RCA4 Rossby Centre regional atmosphere model version 4 RCD Regional Climate Downscaling

RCM Regional Climate Model

RCPs Representative Concentration Pathways REA Reliability Ensemble Averaging

REA_O Original Reliability Ensemble Averaging REA_U Upgraded Reliability Ensemble Averaging REA_E Extended Reliability Ensemble Averaging RRTMG Rapid Radiative Transfer Method for GCMs SA90 1990 Scientific Assessment of the IPCC SAS South Asian Subcontinent

SMHI Swedish Meteorological and Hydrological Institute SPM Summary for Policy Makers

SRES Special Report on Emission Scenarios SSPs Shared Socio-economic Pathways SST Sea Surface Temperature

STD Standard Deviation WCIndia West Central India

WCRP World Climate Research Program WGCM Working Group on Coupled Modelling WIO Western Indian Ocean