Hydrologic Impacts of Climate Change : Scale Issues and Uncertainties P P Mujumdar Dept. of Civil Engineering & Divecha Center for Climate Change, IISc. Organisation of the Talk Introduction – Hydrologic Processes Climate Change Impacts : Scope of Research Scale Issues & Uncertainties Meteorologic droughts River basins : water availability River Water Quality Urban flooding Summary 21 July 2011 Asian Climate Change 2 Hydrologic Processes in a Catchment Source: http://hydrogeology.glg.msu.edu/research/active/modeling-and-monitoring-hydrologicprocesses-in-large-watersheds 21 July 2011 Asian Climate Change 3 •Air Temperature •Net Radiation •Wind Speed •Vapour Pressure •Relative Humidity •Soil Moisture •Type of Vegetation/Crop •Season of Vegetation/Crop Growth 21 July 2011 Asian Climate Change Source for figure : http://eoedu.belspo.be/en/applications/evapcontexte.asp?section=4.1 Factors affecting Evapotranspiration 4 Physical characteristics affecting runoff 21 July 2011 Rainfall intensity Rainfall amount Rainfall duration Distribution of rainfall over the basin Antecedent moisture content Credit : USGS Asian Climate Change RUNOFF Land use Vegetation Soil type Drainage area Basin shape Elevation Topography Drainage network patterns Ponds, lakes, reservoirs etc. in the basin Hydro - meteorological factors affecting runoff 5 Increasing Temperatures Evapotranspiration Water Quality Change in Precipitation Patterns Streamflow; Water availability Intensity, Frequency and Magnitude of Floods and Droughts Groundwater Recharge Rise in Sea Levels Inundation of coastal areas Salinity Intrusion 21 July 2011 Asian Climate Change 6 Fig. Source: ww.engr.uconn.edulanboG229Lect111SWIntru.pdf Climate Change – Hydrologic Implications Research Issues of Interest •Water availability •How do water fluxes vary on catchment scale in response to global climate events? •Impacts on Water Quality •Change in Frequency and Magnitude of extreme events •Design storm intensities - Urban Flooding •Delays in onset of monsoon •Impact on agriculture •Over-year storage policies •Real-time adaptive decisions •Water Demands • Evapotranspiration • Municipal and Industrial Demands •Salinity Intrusions & Coastal flooding •Robust & Resilient water management policies Source for the map: to offset adverse impact due to climate change www.mapsofIndia.com 21 July 2011 Asian Climate Change 7 Hydrologic impact assessment Spatio-temporal scale mismatch Accuracy of tropospheric vs surface variables Source: Xu C.Y., Water Resources Management 13: 369–382, 1999. 21 July 2011 Asian Climate Change 8 Need for downscaling A schematic diagram describing the statistical downscaling approach. GCMs provide useful predictions for large-scale atmospheric patterns (lower part). Details contained within a grid box (upper part) are influenced by local features beyond the resolution of current global climate models.[Source: http://www.bom.gov.au/info/GreenhouseEffectAndClimateChange.pdf ] 21 July 2011 Asian Climate Change 9 Downscaling the GCM outputs to the river basin scales 21 July 2011 Asian Climate Change Downscale Global climate models (GCM: resolution coarser than 20) ; Size of grid box: Tens of thousands of square kilometers. GCM Grid (~2.50) Grid size of interest in hydrology (~0.20 – 0.5100) Projecting Climate Change Impacts on Hydrology Climate Change Projections (precipitation, temperature, radiation, humidity) Topography, Landuse Patterns; soil characteristics; Downscaling Hydrologic Model Possible Future Hydrologic Scenarios on Basin Scale 21 July 2011 (Streamflow, Evapotranspiration, Soil Moisture, Asian Infiltration, Groundwater Climate Change 11 Downscaling Downscaling: to model the hydrologic variables (e.g., precipitation) at a smaller scale based on large scale GCM outputs. Dynamic Downscaling: uses complex algorithms at a fine grid-scale (typically of the order of 50 Km × 50 Km) describing atmospheric process nested within the GCM outputs; commonly known as Limited Area Models (LAM) or Regional Climate Models (RCM). Statistical Downscaling: produces future scenarios based on statistical relationships between large scale climate features and hydrologic variables. Assumption- Statistical relationships hold good in future for changed climate scenario. Advantage- computationally simple. Climate Predictors : Must be reliably simulated by GCMs; readily available from archives of GCM outputs and strongly correlated with the surface variables of interest 21 July 2011 Asian Climate Change 12 Climate change effects in the Colorado river basin Source: Christensen et al. (2004), Climatic Change 62, 337–363 Naturalized: effects of water management removed Colorado River basin with 1/8degree VIC routing network and major system of reservoirs Drainage Area : 6,30,000 sq. km; Serves 7 states; 12 major reservoirs – water supply, hydropower and flood control ; 70% runoff from Snow pack; Average Annual Runoff : 18.6BCM Current demands in the basin are not much lower than the mean flow. A mere 10% reduction in mean annual flow has major implications for the reservoir system performance; Reliability of a reservoir system decreases rapidly as the demands approach the mean flow; 21 July 2011 Asian Climate Change 13 Climate change effects in the Colorado river basin Downscaled temperature and precipitation from Parallel Climate Model (PCM) – 105 year simulations 21 July 2011 Asian Climate Change Source: Christensen et al. (2004), Climatic Change 62, 337–363 HIST : Historical (observed) : 1950-1999 CTRL : Control Climate Simulation (1995 greenhouse gas levels) BAU: Business as usual scenario for periods 1–3: 2010–2039, 2040– 2069 and 2070–2098 14 Downscaling GCM Simulations to Precipitation : Orissa Meteorological Sub-division • Coastal Area GCM Grids Surrounding the Case Study Area • Increase of hydrologic extremes in recent past 21 July 2011 • Increase in temperature: 1.10C/century, whereas in average increase in India: 0.40C/century. Asian Climate Change 15 NCEP grid points surrounding the meteorological sub-division Orissa Statistical downscaling (Principal component analysis, fuzzy clustering, transfer functions) Climate Predictors: MSLP and 500 hPa geopotential height Data Used : Rainfall : 1950-2003 (source : IITM Pune) Climate Predictors : from NCEP Reanalysis Project 21 July 2011 Asian Climate Change 16 Projected Rainfall CCSR/NIES GCM with B2 Scenario Increase decrease 21 July 2011 Asian Climate Change 17 GCMs and Scenarios Used 21 July 2011 Asian Climate Change 18 Projections of SPI (Drought Indicator) 21 July 2011 Asian Climate Change 19 pdfs of Drought Indicator Water Resources Research , No. 43, (2007) All scenarios are equally possible Projections from all GCMs are equally likely to be realized. Time series generated by a downscaled GCM simulation with one scenario is considered as one realization. 21 July 2011 Asian Climate Change 20 Projections with A1B Scenario 21 July 2011 Asian Climate Change 21 Weights for A1B Scenario Assignment of weights : Reliability Ensemble Averaging Algorithm Two reliability criteria : (a) Performance of the model in reproducing the present-day climate (“model performance”) (b) Convergence of simulated changes across models (“model convergence”) 21 July 2011 Asian Climate Change 22 Projected Rainfall (Weighted Mean CDF; A1B scenario) Limitation Reducing the present knowledge about climate sensitivity to a single probability distribution would clearly mis-represent the scientific disagreement ( Hall et al., 2007). 21 July 2011 Asian Climate Change 23 Imprecise Probability Provides an envelope of probability distribution 21 July 2011 Asian Climate Change 24 Bounds for Probability of Drought Journal of Geophysical Research, 114 (2009) 21 July 2011 Asian Climate Change 25 Mahanadi River Basin - Streamflow Hirakud Dam Predictand: Predictors 2m Surface Temperature Geopotential Height at 500 hPa Specific Humidity Mean Sea Level Pressure 21 July 2011 Asian Climate Change Monsoon Streamflow of Mahanadi River at Hirakud Dam 26 Selection of Predictors Streamflow: result of rainfall, evaporation and infiltration. Monsoon: insignificant infiltration compared to streamflow. Rainfall: consequence of Mean Sea Level Pressure (MSLP) (Bardossy and Plate, 1991; Bardossy et al., 1995; Hughes and Guttrop, 1994), Geopotential Height (Stehlik and Bardossy, 2002) and Specific Humidity (Crane and Hewitson, 1998). Evaporation: mainly guided by temperature and humidity (Wilby and Harris, 2006) Predictors selected: 2m surface air temperature, MSLP, geopotential height at 500 hPa and surface specific humidity 21 July 2011 Asian Climate Change 27 Observed and Predicted Streamflow (from NCEP/NCAR reanalysis data) Observed Predicted 7332 Mm3 7384 Mm3 Std. Dev 5996 Mm3 4607 Mm3 Mean Nash-Sutcliffe coefficient, E = 0.67 21 July 2011 Asian Climate Change 28 Possibilistic Approach Assumptions of earlier models on GCM and scenario uncertainty: equal possibility and equi-probability of all the scenarios. With the passage of time, it is relevant to assess the effectiveness of the GCMs in best modeling climate change and also to judge which of the scenarios best represent the present situation under climate forcing. Scope of the study: assignment of possibility distribution to different GCMs and scenarios, measured in terms of their ability in modeling climate change based on their performance in the recent past (years 1991-2005) under climate forcing 21 July 2011 Asian Climate Change 29 Possibility Distribution of GCMs and Scenarios Water Resources Research , No. 44, (2008), AGU Model Uncertainty is greater than Scenario Uncertainty Possibility assigned to GCM : possibility with which the future hydrologic scenario is modeled best by the downscaled output of the GCM ; Possibility assigned to a scenario : possibility with which the scenario best represents the climate forcing in the study area 21 July 2011 Asian Climate Change 30 Projected Streamflow CDF 21 July 2011 Asian Climate Change 31 Projections for future monsoon inflows to Hirakud Reservoir Reduction in ‘normal’ (middle level) flows Range of projected future flow duration curves at Hirakud 21 July 2011 Asian Climate Change 32 Projected Irrigation Water Demand : CGCM2; A2 ; Source : Asokan and Dutta (2009) Projected Peak and Average Discharge; CGCM2; A2; Source : Asokan and Dutta (2009) Flood Storage Dam Live Storage Hydropower 21 July 2011 Asian Climate Change Irrigation 33 Hirakud Reservoir : Serves Flood Control, Hydropower and Irrigation Gross Storage : 5896 Mm3 Live Storage : 4823 Mm3 Installed Capacity : 347.5 MW Firm Power : 134 MW Adaptive Operating Policies : Derived with Stochastic Dynamic Programming, with tradeoffs between flood control, hydropower and irrigation, with an objective of maximising hydropower in future years. 2045-65 21 July 2011 Asian Climate Change 34 Impacts: Performance measures Adaptive Policy using SDP (Policy 1) Reliability-power 0.604 Current (1959-2005) A1B 0.5 2045-65 MIROC3.2 0.453 CGCM2 0.502 GISS A1B 2075-95 MIROC3.2 0.366 0.276 CGCM2 0.403 GISS Resiliency-power 0.229 Current (1959-2005) A1B 2045-65 MIROC3.2 0.214 0.218 CGCM2 0.215 GISS A1B 2075-95 MIROC3.2 0.155 0.123 CGCM2 0.178 GISS 21 July 2011 A2 0.484 0.523 0.515 A2 0.286 0.146 0.423 B1 0.462 0.471 0.514 B1 0.382 0.294 0.458 A2 0.215 0.224 0.221 A2 0.159 0.103 0.175 B1 0.206 0.202 0.213 B1 0.177 0.255 0.18 Reliability-irrigation 0.834 A1B 0.799 MIROC3.2 0.796 CGCM2 0.802 GISS A1B 0.592 MIROC3.2 0.544 CGCM2 0.599 GISS Vulnerability-power 0.688 A1B 0.824 MIROC3.2 0.956 CGCM2 0.911 GISS A1B 1 MIROC3.2 0.952 CGCM2 0.878 GISS Asian Climate Change A2 0.798 0.802 0.801 A2 0.497 0.447 0.614 B1 0.795 0.801 0.801 B1 0.601 0.535 0.634 A2 0.895 0.75 0.873 A2 0.933 1 0.883 B1 0.931 0.935 0.903 B1 0.966 1 0.865 Reliability-Flood Control 0.907 A1B A2 MIROC3.2 0.906 0.921 0.961 0.95 CGCM2 0.899 0.905 GISS A1B A2 0.93 MIROC3.2 0.944 0.95 0.988 CGCM2 0.916 0.902 GISS Deficit ratio-power 0.311 A1B A2 0.41 MIROC3.2 0.395 0.429 0.381 CGCM2 0.381 0.377 GISS A1B A2 0.65 MIROC3.2 0.558 0.677 0.8 CGCM2 0.525 0.501 GISS B1 0.939 0.955 0.897 B1 0.916 0.984 0.894 B1 0.43 0.395 0.371 B1 0.571 0.673 0.466 35 35 Rule curve for adaptive policies 194 192 Reservoir level (m) Curr rule curve min 190 Curr rule curve max 188 SDP 2045-65 186 Adaptive policy 1 Adaptive policy 2 184 Adaptive policy 3 182 SDP 1959-2005 180 178 1-Jul 1-Aug 1-Sep 2-Oct Date 21 July 2011 Asian Climate Change 36 Advances in Water Resources (2010) Rule curve at Hirakud for adaptive policies Impacts on River Water Quality Non-point source of pollution 21 July 2011 Asian Climate Change 37 Schematic Diagram of Tunga-Bhadra River Shimoga City Sewage Tunga Shimoga 4 4 MPML VISL 3 3 Bhadra Lakavalli 1 1 Tunga - Bhadra River 5 5 6 Honnali City Sewage 2 Harlahalli 6 2 Kumudavathi 7 7 Honnali 16 8 BhadravathiCity Head Water Flow Point Load 8 9 Kuppelur Reach Reach End point Check point MPM VISL HPF - 21 July 2011 15 12 9 14 Harihar City Sewage 10 13 11 10 11 12 13 HP 14 Dhavangere City Sewage Byladahalli Haridra Mysore Paper Mill Vishveshwaraya Iron and Steel Limited Harihara Poly Fibers Asian Climate Change 38 Temperature Station Variable Air Temperature Shimoga Water Temperature Air Temperature Honnali Water Temperature Air Temperature Kuppelur Water Temperature 21 July 2011 Period 1988 – 1999 To 2000 - 2006 1988 – 1999 To 2000 - 2006 1988 – 1999 To 2000 - 2006 1988 – 1999 To 2000 -2006 1991 – 2001 To 2002 - 2006 1991 – 2000 To 2002 - 2006 Annual Mean Change Streamflow Increase by 0.215 oC Increase by 0.599 oC Increase by 0.315 oC Increase by 3.34 oC Increase by 1.39 oC Station Period 1971 – 1991 To 1992 2006 1980 – 1990 To 1991 Honnali 2006 1991-1999 To 2000Kuppelur 2006 Byladahal 1985 – 1995 To 1996 li 2005 Shimoga % Reduction in Annual Mean Flow 3.1 12.26 16.8 24.16 Increase by 1.79 oC Asian Climate Change 39 Climate Change Impact Assessment ; Adaptive Policies Climate Change Projections Statistical Downscaling River Water Quality Simulation Model Water Quality Responses 21 July 2011 Optimal Effluent Treatment Levels (Fuzzy Effluent Load Allocation AsianModel) Climate Change 40 40 Observed and CCA Predicted Projections from MIROC 3.2 GCM (A1B) at Shimoga along Tunga River Average Air Temperature Maximum Air Temperature Observed Predicted (A) 30 25 35 20 30 25 Observed Predicted (A) 40 35 26 2010-2040 2040-2070 (B) 60 Observed (A) Predicted 80 75 70 65 Monthly Wind Speed in kmph Monthly Relative Humidity 80 40 24 22 2040-2070 (B) 2010-2040 2070-2100 Wind Speed Relative Humidity 100 Predicted (A) 18 2010-2040 2070-2100 Observed 20 30 22 15 8 35 6 30 4 2 Observed 2040-2070 (B) 2070-2100 River Water Temperature Predicted (A) 6 4 Water Temperature in degree C 20 40 Monthly Minimum Air Temperature in degree C 25 Monthly Maximum Air Temperature in degree C Monthly Average Air Temperature in deg C 30 Minimum Air Temperature 25 20 Observed Predicted (A) 35 30 25 2 2010-2040 2040-2070 (B) 21 July 2011 2070-2100 20 2010-2040 2040-2070 (B) 2070-2100 2010-2040 2040-2070 (B) 2070-2100 41 Asian Climate Change 41 21 July 2011 Asian Climate Change 42 Present and Future Estimates of DO Levels at Various Check Points along Tunga-Bhadra River Check Point 11 7.00 6.50 6.00 Dissolved Oxygen mg/L Dissolved Oxygen mg/L Check Point 1 6.72 6.06 5.50 5.55 5.00 4.50 4.89 4.00 3.50 present 7.00 6.50 6.00 5.50 5.82 5.00 4.50 5.24 4.96 4.70 4.00 3.50 present 2010-2040 2040-2070 2070-2100 2010-2040 2040-2070 2070-2100 Dissolved Oxygen mg/L Check Point 14 21 July 2011 6.50 6.00 5.50 5.00 5.41 4.85 4.50 4.44 4.00 4.30 3.50 present 2010-2040 2040-2070 2070-2100 Asian Climate Change 43 Current and Projected Treatment Policy Fractional Removal Levels 1 0.9 Current 0.8 0.7 2010-2040 0.6 2040-2070 0.5 2070-2100 0.4 0.3 1 2 3 4 5 6 7 8 Discharger 21 July 2011 Asian Climate Change 44 Urban Flooding Urbanisation alters the hydrology of a region; rainfall – runoff relationships get affected; quicker and higher peak flows ; more runoff After Urbanization Before Urbanization Q 21 July 2011 Asian Climate Change t 45 Bangalore Floods How do the short term intensities of rainfall respond to the climate change? 21 July 2011 Urban Flooding Asian Climate Change Likely changes in IDF (Intensity-DurationFrequency) relationships due to climate change 46 Toronto Source : Simonovic, 2005 Precipitation Intensity (mm/hour) 95 2090 90 85 2050 80 75 1985 70 65 60 55 50 21 July 2011 10 20 30 40 50 60 70 Extreme precipitation recurrence time (Years) Asian Climate Change 80 47 Bangalore City – Change in the IDF Relationships Comparison of IDF for return period of 10 years 100 Rainfall Intensity (mm/h) 90 80 90.174 1969-2003 76.789 1969-1986 1987-2003 70 60 62.672 59.653 53.898 50 43.471 40 33.651 30 26.968 20 19.124 15.25 9.5709 17.45 10 0 1 2 6 12 24 Duration (hours) Results not conclusive, because of the small sample of data available 21 July 2011 Asian Climate Change 48 Summary Climate change is likely to impact most hydrologic processes Impacts need to be assessed at regional/riverbasin and smaller scales GCMs are the most credible tools available today for impact assessment Scale issues and uncertainties are addressed in recent studies Results from the studies are useful in developing adaptive responses (e.g., long term reservoir operating policies; modifications in hydrologic designs; change in cropping patterns; water use adjustments etc.) Similar results may be used in developing IntenstyDuration-Frequency (IDF) relationships and FlowDuration curves, accounting for Climate Change. 21 July 2011 Asian Climate Change 49 THANK YOU 21 July 2011 Asian Climate Change 50 Kernel Density Estimation • Basic Equation fˆ x nh 1 fˆ x n h n K x X h l l 1 - kernel density estimator of a pdf at x - number of observations - smoothing parameter known as bandwidth Selection of bandwidth - an important step in kernel estimation method. Conventional Method (Silverman, 1986): 21 July 2011 h0 1.587 n 1 3 IQR min S , 1.349 Asian Climate Change 51 Kernel Density Estimation: Drawbacks A large sample can give a better estimate of kernel density estimator. In the present analysis, the sample size is small only having the downscaled SPI of the available GCM output, which may not lead to accurate results The bandwidth is estimated by assuming the actual density as normal, which may not be the actual case. In such cases the estimate may be inaccurate. 21 July 2011 Asian Climate Change 52 Orthonormal Series Method Orthonormal series: series of functions of an orthonormal system Properties: x x dx 0 s j x dx 1 s Used to determine the nonparametric PDF of a small sample s j 2 s (Effromovich, 1999). f J x j j x J or j 0 f J x j j x j 0 Series Used: Cosine System 0 x 1 j x 2 cosjx 21 July 2011 Asian Climate Change j0 53 Determination of Coefficients Estimation of j j f ( x) j x dx From probability methods: E j x f x x dx j It can be said j E j x or 21 July 2011 1 n j j xl n l 1 Asian Climate Change 54 Algorithm for pdf Estimation with Orthonormal Series Determination of bounds/support of the data set Scaling of data set Determination of functions of orthonormal series Determination of smoothing parameter Determination of cut-off Determination of Fourier coefficients Modifications for smoothness, area, negative values Determination of pdf of scaled data Determination of final pdf of unscaled data 21 July 2011 Asian Climate Change 55 Objectives of the Possibilistic Approach Assumptions of earlier models on GCM and scenario uncertainty: equal possibility and equi-probability to all the scenarios. With the passage of time, it is relevant to assess the effectiveness of the GCMs in best modeling climate change and also to judge which of the scenarios best represent the present situation under climate forcing. Scope of the study: assignment of possibility distribution to different GCMs and scenarios, measured in terms of their ability in modeling climate change based on their performance in the recent past (years 1991-2005) under climate forcing 21 July 2011 Asian Climate Change 56 Theory of Possibility Distribution X: variable in the universe , and is not possible to measure precisely Possibility that X can take a value x: X(x): [0,1] X(x)=0: Denotes X=x is impossible X(x)=1: Denotes X=x is possible without any restriction X x 1, x : Interpreted as complete ignorance about X 21 July 2011 Asian Climate Change 57 Modeling GCM and Scenario Uncertainty with Possibility Theory Assignment of possibility to GCMs and scenarios: based on system performance in recent past (1991-2005) when climate forcings are visible. System performance measure: Deviation of the predicted CDF from the observed CDF. System performance measure (C): similar to Nash-Sutcliffe coefficient. 2 Q C 1 Q F F OF OF QPF QO 2 Normalization of C values to obtain possibility distribution 21 July 2011 Asian Climate Change 58 Graphical models Combine probability theory and graph theory Family of probability distributions that factorize according to an underlying graph G=(V,E). Vertices ~ random variables, edges ~ statistical dependencies Directed graph (Bayesian Network) Variable is conditionally independent of all other variables given its parents Undirected graph (Markov Random Field / Markov network) Variable is conditionally independent of all other variables given its neighbors Asian Climate Change 21 July Do2011 not impose acyclicity constraint / constraints on causality 59 Graphical models (contd.) Factor graph representation Every undirected model can be represented as a factor graph Distribution over a large number of random variables represented as product of local functions Where is a global normalizing constant. are clique potentials, functions from sets of nodes to nonnegative reals. 60 Asian Climate Change 21 July 2011 Generative vs discriminative models Generative model : Based on joint distribution p(y,x) Simplifying independence assumptions, else has to account for correlated features of input Eg: Naïve Bayes classifier Task: Predict class variable y (say, whether an email is spam / not spam) given a vector of features x = (x1,x2,...xk) (say, from address in predetermined list, contents more than certain size, subject contains word from predetermined list, etc). Model of joint distribution is K p( y, x) p( y) p( xk | y) k 1 Discriminative model : Based on conditional distribution p(y|x) Does not need model for p(x) Sufficient for classification tasks Model does not need to account for complex dependencies among input variables. Better suited to include rich, overlapping features Eg: Logistic regression (maximum entropy) 21 July 2011 K 1 p( y | x) exp{ AsianClimate Change y y, j x j } Z ( x) j 1 61 HMMs vs linear chain CRFs HMMs Linear chain CRFs Type of Bayesian Network Markov random field globally conditioned on input (observation) variables x Generative model Discriminative model Assigns a joint probability to paired observation and label sequences Assigns a conditional probability to label sequences given observation sequence Parameters trained to maximize joint likelihood Parameters trained to maximize conditional log likelihood Joint distribution may have many forms, one of which is an HMM Conditional distribution is a linear chain CRF which includes features only for the current input variable xt (at time t) Graphical model representation of an HMM 21 July 2011 One of the many graphical model representations of a linear chain CRF Asian Climate Change 62 IPCC SRES (2001) Scenarios(40) A1 A1F Fossil fuel Scenario family World order Ecologically friendly Population Economic growth Technology growth 21 July 2011 A2 A1B B1 B2 (family) (group) A1T Balanced Non fossil fuel A1 A2 B1 B2 Integrated Divided Integrated Divided No No Yes Yes Increases till 2050 and then declines Rapid Continuously increasing Same as A1 Increasing but lower than A2 Regionally oriented Rapid (serviceoriented) Intermediate Rapid Slower fragmented Rapid Slower fragmented Asian Climate Change 63 IPCC Scenarios 21 July 2011 Asian Climate Change 64 Climate change projections Source:Meehl et al., Climate Change 2007: The Physical Science Basis, WG I, Asian Climate Change AR4, IPCC 21 July 2011 65