Addressing Challenges in Utilizing Rainfall Forecasts from Global NWP Models for Hydrological Forecasting in Major Indian River Basins

Quantitative Precipitation Forecast (QPF) obtained from a Numerical Weather Prediction (NWP) model is used for short-range streamflow forecasting. At a catchment level, the QPF needs to be compared with the observations to examine the forecast accuracy. National Centre for Medium Range Weather Forecasting (NCMRWF) regularly performs verification of forecasts (deterministic/ensemble) in large number of catchments across India. The comparison of QPF(s) with observation (satellite/gauge) is usually performed at the centroid of the sub-catchment. It is observed that the forecast performance depends on catchment characteristics such as size, topography, rainfall-generation-mechanism etc. However, a systematic investigation is yet to be performed to gain confidence in using forecasts in a given sub-catchment. Furthermore, a bias correction approach needs to be developed to improve the forecasts across all sub-catchments. To use a bias-correction for operational forecast services, its parameters need to be either known or follow some guideline. This study attempts to first compare various rainfall forecast products (deterministic/ensembles) from global NWP models, determine which observation data (satellite/gauge-based) to be used for comparison, relate the forecast accuracy with catchment characteristics, and subsequently suggest a suitable bias-correction approach. The forecast skill is evaluated for 177 subbasins in Ganga, Narmada, Mahanadi, Tapti and Godavari River Basins. The forecast accuracy is examined for monsoon season at multiple temporal resolution i.e. 6, 12, and 24 hours and for multiple lead times (up to 96 hours). Subsequently, we relate the forecast accuracy with subbasin characteristics and precipitation intensity. Proxy validation of the stream flow forecast simulated using QPF as input was performed using a hydrological model. Our results indicate that the forecast accuracy of NCMRWF is closely comparable with other forecast products for all lead time and temporal resolution. We concluded that selecting accurate observation data is very critical for the forecast performance evaluation. Furthermore, we observed that the forecast performance varies significantly with the sub basin characteristics and precipitation intensity.