Abstract

This study proposes an innovative approach forstatistical downscaling of rainfall based on scaling property ofmeteorological variables. The reanalysis data of five dominantmeteorological variables mean sea level pressure, relativehumidity, surface temperature, wind velocity (zonal andmeridional components) extracted from National Centre forEnvironmental Prediction (NCEP) are used as predictors toproject monthly rainfall of Kerala meteorological subdivision inIndia. The multiscale decomposition of predictor dataset of theregion and the monthly rainfall of a specific grid point isperformed simultaneously by employing the MultivariateEmpirical Mode Decomposition (MEMD) technique. Theindividual modes are predicted by fitting stepwise linearregression (SLR) by considering the potential predictors basedon p-value statistics. Subsequent addition of the predictedmodes gives the monthly rainfall. The method is demonstratedby a specific grid point of Chalakkudi river basin in Kerala,India. The method is found to be superior over the linearregression and M5 model tree based transfer functionapproaches. Further, the MEMD-SLR hybrid model is used forrainfall projections of the state of Kerala under threerepresentative concentration pathway scenarios (RCP2.6,RCP4.5 and RCP8.5) provided by Canadian Centre for ClimateModeling and Analysis (CCCMa).