On the utility of Ensemble Rainfall Forecasts over River Basins in India

Abstract Rivers form a lifeline for the agriculture based economy in India, but recent heavy rainfall events have caused major floods in the rivers resulting in loss of life and property. In order to accurately forecast the stream flow from the rivers firstly, an accurate forecast of rainfall over the river basins (RB) is required. Until recently, for operational flood forecasting in India, rainfall forecasts from deterministic models were used. Deterministic models often result in incorrect forecasts as they do not contain the uncertainty information. Ensemble prediction systems (EPS) sample this uncertainty and can add value to the deterministic forecasts. This study seeks to address the question ‘ whether the ensemble rainfall forecasts over RBs in India are ready for hydrological applications? ’ In order to answer this and generate more confidence in using probabilistic rainfall forecasts from an EPS for hydrological purposes the accuracy of the forecasts has to be established. For this purpose, we have carried out an in-depth verification of the probabilistic rainfall forecasts obtained from the NCMRWF EPS (NEPS) over 8 major RBs of India during the southwest monsoon (SWM) seasons of 2018 to 2021. The basin averaged rainfall forecasts from NEPS and observations from the Integrated Multi-satellitE Retrievals for GPM (IMERG) are used in this study. It was seen from the study that the model possesses good skill in predicting low to moderate rainfall over Himalayan rivers like Ganga and peninsular rivers like Tapi, Narmada, Cauvery, and Krishna. This is seen in terms of a low Brier Score (BS), high Brier Skill Score (BSS) and low Continuous Ranked Probability Score (CRPS), as well as lower RMSE in the ensemble mean. The skill of the model is further confirmed by comparing the RMSE in the mean with the spread in the members. The best match between the RMSE in ensemble mean and spread is seen for Ganga RB. The Relative Economic Value (REV) determines the economic value of forecasts and it shows that over Ganga, Mahanadi, and Narmada the rainfall forecasts show the maximum economic value. However, the model shows relatively poorer skill in predicting rainfall over the Brahmaputra RB located in northeastern India. From this study it can be concluded that NEPS model has reasonably good skill in predicting rainfall over RBs in northern and peninsular parts of India and it would be beneficial to use these forecasts for forecasting floods.