UK-India workshop on downscaling and linking to applications, UEA, January 2009 Uncertainties in future projections of extreme precipitation in the Asian monsoon regions: a GCM perspective Andrew Turner 1 & Julia Slingo 2,1 1 NCAS-Climate, University of Reading, UK 2 Met Office, Exeter, UK
Introduction
Extreme rainfall can be damaging for a number of reasons: Agriculture – destruction of crops Infrastructure – damage to roads, buildings and utilities Disease – flooding exacerbates waterborne disease Little is known about what will happen to extreme rainfall events in monsoon regions.
Introduction #2 In this study we are concerned with two aspects of changes to monsoon rainfall extremes, with a focus on India. What will happen to the spatial distribution of the extreme rainfall events which happen each season? (e.g. few days per season with very heavy rainfall). What will happen to the intensity of the very heaviest rainfall events in different climate scenarios? (e.g. one-off localized convective storms)
Introduction #3 Both of these aspects were looked at in an earlier single GCM study (Turner & Slingo 2008): 100-year integrations of the HadCM3 coupled model under pre-industrial control and doubled CO2 forcing conditions. HadCM3 has relatively coarse resolution, even compared to other state-of-the-art GCMs: Atmosphere 3.75˚x2.5˚, 30 levels Ocean 1.25˚x1.25˚, 20 levels
Outline Introduction Uncertainties in the spatial pattern of changes to subseasonal monsoon rainfall extremes: Changes to the pattern of rainfall extremes in HadCM3 Comparison with the CMIP3 archive Uncertainty in predicted changes to intensity of the most extreme events over monsoon regions: Predictable changes in HadCM3 for India Case studies from the CMIP3 archive Summary and open questions relating to downscaling
Changes in subseasonal rainfall extremes Here we are specifically interested in those extreme rainfall days in each season, and how their patterns change between control and 2xCO 2 scenarios. 95 th and 99 th percentiles are calculated at each gridpoint over the Indian region, and for each season. These percentiles are averaged over the seasons available (100).
Changes in subseasonal rainfall extremes Extreme changes are large and statistically significant (95% level) especially in northern India, south/central China and South/East China Seas. From A.G. Turner & J.M. Slingo (2008). Submitted, QJRMS. HadCM3 1xCO 2 HadCM3 2xCO 2 2xCO 2 minus 1xCO 2 95th 99th
Mean precipitation / change (JJAS) HadCM3 1xCO 2 2xCO 2 minus 1xCO 2 Changes to 95 th and 99 th subseasonal percentiles in HadCM3 are larger than changes in the mean, but in strong qualitative agreement with them.
Spatial changes to extremes in a multi-model database #1 Even projected changes to the mean rainfall in monsoon regions are very uncertain. Late-21 st century results from the A1B emissions scenario in the CMIP3 / IPCC AR4 multi-model database. General positive trend in mean summer rainfall for India, however the inter-model standard deviation (noise) exceeds the ensemble mean change (signal).
Spatial changes to extremes in a multi-model database #2 How can we make assessments of changes to extremes of monsoon rainfall in future climate scenarios when even mean changes are uncertain? What the IPCC AR4 says about extremes: Wet extremes more severe in models where mean precipitation increases (Meehl et al. 2007). Projections concerning extreme events in the tropics remain uncertain (Christensen et al. 2007). We examine changes to the mean, 95 th & 99 th subseasonal rainfall percentiles in available daily data from the IPCC database: 15 models provide daily data at control and CO2 doubling (1pctto2x) conditions. Summer (JJA) only.
Uncertainty in spatial distribution of extremes change: the AR4 models – mean vs. extreme
Close examination of spatial changes to mean and extremes via pattern correlations for the monsoon regions at various upper percentiles: India (65-90˚E, 10-30˚N) Broad Asia-Pacific (40-180˚E, 25˚S-40˚N) Projections of changes to the local distribution of precipitation extremes are very uncertain across GCMs, however, like the mean change, they are generally positive Patterns of these subseasonal extreme change seem strongly linked to mean changes difficult to devise local strategies to deal with extreme change when there is such uncertainty in current GCM output.
Uncertainty in spatial distribution of extremes change: the AR4 models – mean vs. extreme Broad consistency between models At the 90 th subseasonal percentile, around 80% of the variance in the spatial pattern of change to the extreme can be explained by the pattern of mean change. Still around 50% at the 99 th percentile. This implies such extremes are spatially quite predictable.
Predictability in changes to the heaviest events Upper extremes are assessed for all times and at all gridpoints in a given region (after the global assessment of Allen & Ingram, 2002). The upper two quartiles are measured in control and 2xCO 2 scenarios, and the ratio of their magnitudes is found. No area-averaging is performed. Allen & Ingram (2002) showed that the increase in magnitude of the heaviest rainfall (when all moisture in the atmospheric column precipitates out) was predictable based on thermodynamic arguments. Turner & Slingo (2008) showed that this was applicable to the Indian monsoon region in the HadCM3 model.
Predictability in changes to the heaviest events tropicsIndia Maximum precipitation intensity increases broadly inline with Clausius-Clapeyron and measured local surface warming.
Predictability in changes to the heaviest events Does this simple relationship apply in other coupled GCMs? We test the CMIP3 models again, under control conditions and at CO 2 doubling. Some modelling groups also provide output after CO 2 stabilization. Typical data input for India (65–90 E, 10–30 N, for JJA): 40yrs control 20yrs 1pctto2x at time of CO 2 doubling (red) 20yrs 1pctto2x from CO 2 stabilization (blue, 9 models) Surface air temperature used to measure climate sensitivity.
Predictability in changes to the heaviest events Median projected increase in heaviest events is higher than that predicted due to surface warming alone (box- whisker plots). Considerable uncertainty among the models. Time of CO 2 doubling CO 2 stabilization
Predictability in changes to the heaviest events: 3 case studies We examine 3 model case studies of different behaviour in predictions vs. measured changes: 1) Changes above those predicted: gfdl_cm2_1 and 5 others (6/15). 2) Changes below predictions: ipsl_cm4 and 2 others (3/15). 3) Changes inline with predictions: cnrm_cm3 and 5 others (6/15), as well as HadCM3. Results for some models suggest remarkable predictability of changes to maximum precipitation intensity based on purely thermodynamic arguments. Allen & Ingram (2002) argued that monsoon regions could undergo larger increases due to feedbacks between latent heat release and strengthening of the Somali Jet (moisture convergence).
Given there is greater predictability in mean temperature changes compared to mean precipitation, this result suggests there may be some skill in predicting change to the heaviest rainfall events in monsoon regions in future climate scenarios. The reasons for such different behaviour among the CMIP3 models (why some models feature a different dynamical component) remain to be understood. Predictability in changes to the heaviest events
Summary Mean rainfall changes over the Indian monsoon region predominantly range from zero to positive, despite uncertainty in their local distribution. Changes to subseasonal precipitation extremes occurring in each season are quite predictable based on the model-dependent pattern of mean change. Increase in the magnitude of the heaviest monsoon rainfall may be potentially predictable based on local surface warming, or this may provide a lower bound.
Open questions for the workshop How can the impact of these climatic changes to extremes (needed to assess impacts of flooding on infrastructure, crop damage etc) be assessed given the large uncertainty involved? When GCMs converge more strongly on a single mechanism / pattern of mean monsoon rainfall change, can we be confident in their predictions? Will GCMs always be at a scale where they cannot simulate the statistics of rainfall events (the pdf), such that downscaling methods will be required?
Thank you! See: Turner & Slingo (2008), submitted Q.J.R.M.S. Turner & Slingo (2009), submitted Atmos. Sci. Lett.
Rainfall distribution: model vs. obs. Model has a tendency to drizzle (common with many convection schemes). Cannot represent the upper tail correctly due to grid size. Clear tendency for increased frequency of heavy events at 2xCO 2, at the expense of moderate rainfall. This trend is also noted in the IMD dataset 1. 1 B.N. Goswami, V. Venugopal, D. Sengupta, M.S. Madhusoodanan, P.K. Xavier (2006). Science 314: (each gridpoint)
IPCC AR4 upper percentile intensity change (India)
IPCC AR4 upper percentile intensity change (India JJA)