Statistical Downscaling of Multisite Daily Precipitation for Tapi Basin Using Kernel Regression Model

The study presents fine resolution multisite daily precipitation projection for the Tapi basin using the kernel-regression (KR) based statistical downscaling methodology developed by Kannan and Ghosh with and without conditioned on the estimated rainfall state. The models are applied in downscaling of daily monsoon precipitation at a fine resolution of 0.25 comprising 351 grid points in and around the basin. The air temperature, specific humidity, zonal and meridional wind (at surface, 250, 500 and 850 hPa); mean sea level pressure and geopotential height at surface are utilized as predictors from five GCMs under CMIP-5 for two future scenarios, viz. RCP4.5 and RCP8.5. The performance of the downscaling model examined with respect to reproduction of various statistics for training period and indicated the better performance of KR model conditioned on the rainfall state than KR model without conditioned on the rainfall state of the basin. The KR model conditioned on the rainfall state is employed for future projections from GCMs outputs. The statistically downscaled daily precipitation from GCM (MPI-M) and CORDEX (COSMO-CLM) data is compared to quantify uncertainty. The statistically downscaled daily precipitation performs better than corresponding CORDEX data for the present study area. The study also revealed a possibility of decrease in the occurrences of extreme events with an increase in the medium rainfall events in the basin for future.