Search

Rainfall forecasting and its spatio-temporal variability is important for many hydrological applications. It is critical to understand the uncertainty and verify the quality of rainfall forecasts provided by Numerical Weather Prediction (NWP) models. In the present study, the National Center for Environmental Prediction (NCEP) Global Forecast System (GFS) model performance is evaluated for day-1 to day-5 forecast with a threshold of 1 mm/day in the Nagavali and Vamsadhara river basins, India. From the results, the model predicted the rainfall with a correlation coefficient of >0.3 and probability of detection >0.6 for day-1 and day-3 forecasts. The bias in rainfall prediction shifted from overestimation to underestimation by 30% as forecast lead time increased. The total mean error is decomposed into hit, false, and missed bias. The main sources of total mean error are hit bias and false bias. However, missed bias influenced total mean error as lead time increased. Bias correction is applied for the rainfall events with a rainfall intensity >12 mm/day. RMSE improved by >18% for day-1 forecast in both the Nagavali and Vamsadhara basins, and the improvement ranged between 3% to 9% for other days. In the Nagavali basin, BIAS and ME improved and ranged from 44% to 65% for day-1 to day-5 forecast, whereas in the Vamsadhara basin, it ranged from 65% to 93%. Our findings are useful for early warning dissemination during the flood events, resource mobilization to protect communities, and sustainable water resources planning and management. • The skill of the NCEP GFS model to forecast precipitation was examined. • For day-1 to day-3 forecasts, a correlation of >0.3 was seen. • Bias correction for rainfall intensities >12 mm/day required. • GFS rainfall forecasts useful for flood and water management in the tropics. [ABSTRACT FROM AUTHOR]