A novel approach for statistical downscaling of future precipitation over the Indo-Gangetic Basin

We propose a novel statistical downscaling method using Global Circulation Model (GCM) rainfall and satellite based precipitation estimate Tropical Rainfall Measurement Mission (TRMM; 3B43v7) to generate a high-resolution rainfall (0.25 degrees x 0.25 degrees) estimate over the Indo-Gangetic Basin (IGB) for 9 GCM and 4 Special Report on Emissions Scenarios (SAES) combinations. These precipitation values, along with the precipitation dataset from the APHRODITE's Water Resources project are then seasonally segregated (winter, pre-monsoon, monsoon and post-monsoon) and combined into a Bayesian framework to generate probability distribution of future precipitation change at regional scale. We considered present time as 2001-2010, and 3 non-overlapping time slices 2011-2040, 2041-2070, and 2071-2100 as future. The precipitation trends are heterogeneous in space and seasons, but there is an overall consistency in trends for different future time slices. The shapes of the final probability density functions given by the kernel density estimators show varying characteristics. Compared to traditional transfer function based statistical downscaling methods our framework allows downscaling to basin level gridded rainfall rather than station specific precipitation. It also allows an integrated estimate of uncertainties arising from different sources which is an essential diagnostic when datasets from various sources are considered. Furthermore, the Bayesian framework allows the analysis of means and precisions of precipitation, even when they reveal characteristics, such as multi-modality and long tails. (C) 2017 Elsevier B.V. All rights reserved.