Improving precipitation forecasts skill over India using a multi-model ensemble technique

Abstract

In this paper a Multi-Model Ensemble (MME) technique is experimented for improving day to day rainfall forecast over India in short to medium range time scale during summer monsoon of 2010. Four operational global Numerical Weather Prediction (NWP) models namely, ECMWF, JMA, NCEP GFS and UKMO available on real time basis at India Meteorological Department (IMD), New Delhi are used simultaneously with appropriate weights to obtain the MME Technique. In this technique, weights for each NWP model at each grid point is assigned on the basis of the correlation coefficient (CC) between model forecasts and observed daily rainfall time series of south west monsoon (JJAS) season. Apart from MME, a simple ensemble mean (ENSM) forecast are also generated and experimented. The rainfall prediction skill of the weighted MME is examined against ENSM and member models. The inter-comparison reveals that the weighted MME is able to provide more accurate forecast of rainfall over Indian monsoon region by taking the strength of each constituent member model. It has been further found that the rainfall prediction skill of MME is higher as compared to ENSM and member models in the short range time scale. The rainfall prediction skill of weighted MME technique improved significantly over India.