On the reliability of medium-range probabilistic rainfall predictions over river basins in India

Abstract

For hydrological applications especially for reservoir management and flood forecasting, it is required that the skill of the medium-range weather forecasts is evaluated and documented. In this study, rainfall forecasts up to 7 days for July and August 2018 have been examined at river basin scales. Reliability of the forecasts at basin scale was examined using reliability diagram for light and moderate rainfall categories. It is found that the model has reasonable skill in forecasting rainfall (ensemble mean) over the Indian regions up to 7 days in advance. The model could predict categorical rainfall (hits) for several rainfall events, however, the number of false alarms are larger than number of hits. The model also missed quite a few events in various categories. On many occasions, none of the ensemble members could forecast high amount of rainfall that was observed. The RMSE and ensemble spread remains almost the same as the forecast length increases over all the river basins considered. Therefore, forecast quality does not deteriorate as the forecast length increases. The ensemble spread is quite less and almost half the RMSE values for each of the river basin. The probabilistic forecasts are not reliable for any of the categories. The model forecasts overestimate the observed frequency over all the river basins. Forecasts with probability values of more than 70% do not have any skill.