Prediction of Indian summer monsoon rainfall: a weighted multi-model ensemble to enhance probabilistic forecast skills

India gets the maximum amount of rainfall during the months of June to September (JJAS) which is known as the summer monsoon season. The erratic nature of Indian summer monsoon rainfall (ISMR), in terms of both rainfall amount and distribution, is highly responsible for the interannual variability in agricultural production as well as occurrence of floods and droughts. Accurate seasonal predictions of ISMR are required for appropriate hydrological planning and disaster management systems. Studies have revealed that probabilistic prediction, based on the products of General Circulation Models (GCMs), can be generated in a parametric as well as non-parametric manner. The present paper discusses the enhancement of probabilistic prediction by improving the potential predictable signal obtained from these GCMs. A Singular-Value-Decomposition based multiple linear regression method (SVD-MLR) has been applied to improve the signal and a simple average of all GCMs (EM) has been used as the benchmark to examine the skill of the SVD-MLR method. The potential of the proposed method has been assessed through Brier Skill Score (BSS) and Rank Probability Skill Score (RPSS). A rigorous analysis has finally revealed that SVD-MLR method has better skill than EM in predicting the typical nature of observed monsoon rainfall in extreme years.