Improved performance of multi-model ensemble through the bias correction based on ANN technique

The prediction of Indian summer monsoon (ISM) variability on extended range (3-4 pentads in advance, where pentad refers to 5-day average) is crucial due to its usefulness in agricultural and hydrological planning. Monsoon rainfall occurs in the form of active/break spells which causes large-scale flood/drought condition. Prediction before 3-4 pentad may not stop this condition but can help us to prepare for the same. Hence Climate Forecast System model version 2 (CFSv2) has been adopted to develop a dynamical prediction system for monsoon rainfall on different time scales. The Extended Range Prediction (ERP) group at Indian Institute of Tropical Meteorology has indigenously developed an Ensemble Prediction System based on CFSv2 for the real-time prediction of active/break spells of ISM and has been providing the experimental forecasts up to 4 pentad lead. A multi-model ensemble (MME) framework employing different variants of CFSv2 has shown better skill in the ERP of ISM, compared to the individual variants. Still CFSv2 model shows statistical biases which cause lack of skill in MME and causes false prediction. In this study, it is proposed to employ the artificial neural network based technique to reduce the inherent biases by post-processing the raw model output. The performance is evaluated with mean square error, correlation coordinate and Brier skill score. Result shows that the probabilistic skill of the model is improved with the ANN-based technique.