Assessing 32-Day Hydrological Ensemble Forecasts in the Lake Champlain–Richelieu River Watershed

Abstract

This paper explored various configurations of the ensemble Kalman filter, the GR4J hydrological model, and the Global Environmental Multiscale (GEM) atmospheric model in order to maximize the skill of ensemble hydrological forecasts for the Lake Champlain–Richelieu River watershed. In open-loop mode, the hydrological model represented very well the observed streamflow (Nash–Sutcliffe value above 90%). It sufficed to assimilate hydrological data to obtain a reliable and skillful analysis of streamflow; assimilation of snow water equivalent (SWE) information did not bring additional benefits. In forecast mode, the opposite was true: hydrological assimilation alone did not improve forecast performance, but assimilating SWE data improved reliability and skill of forecasts with lead times of 15 days to 1 month. The impact of SWE assimilation also depended on the quality of the precipitation analysis. It therefore is recommended to use SWE assimilation for monthly forecasting, especially if the precipitation data used to drive the hydrological model are biased.