Fine Scale Projections of Indian Monsoonal Rainfall Using Statistical Models

General Circulation models (GCMs) simulate climate variables globally accounting for the effects of green house emission; however, they mostly work in coarse resolutions and hence their performances for simulations of precipitation are not always reliable. To overcome this limitation we are using statistical techniques as downscaling methods for projecting precipitation as finer resolution (25 km grid approximately, 0.22° latitude x 0.22° longitude). Here we use conventional statistical downscaling where the relationship between predictor climate variables (other than precipitation) and precipitation are determined and then applied to GCM output for projections of precipitation. Kernel regression is used for developing the statistical relationship. The results are compared with interpolated, quantile based bias corrected GCM simulated precipitation output. Both the methodologies are applied to CMIP3 and CMIP5 simulations and the multi-model averaged (MMA) results are compared. The GCMs used are MRI, MIROC, BCCR, MPI, and CCCMA. We first evaluate the 20C3M simulations with the observed data, and we find that conventionally downscaled MMA simulations of CMIP5 do not show significant improvements over those of CMIP3, which suggests that there is no significant change in predictor simulations by the CMIP5 GCMs over those of CMIP3. However, when we do the same exercise with bias correction, we find bias corrected CMIP5 simulations are significantly improved. This shows that simulation of precipitation in Indian region for observed period has been improved with CMIP5 models. After validations, both the models are applied for future projections. It is observed that, though bias corrected models perform well for observed period, they simulate spatially uniform changes of precipitation in the entire country. The conventional downscaling method, involving predictors other than precipitation, simulates non uniform changes for future, which is similar to the trend of last 50 years of Indian precipitation pattern. The reason behind the failure of bias corrected model in projecting spatially non-uniform precipitation is the inability of the GCMs in modeling finer scale geophysical processes in changed condition. The results highlight the need to revisit the bias correction methods for future projections, to incorporate of finer scale processes.