Bias correction of ensemble precipitation forecasts in the improvement of summer streamflow prediction skill

Abstract

In recent years, ensemble precipitation forecasting has been able to provide more uncertainty information and plays an increasingly important role in basin-scale hydrologic predictions. The Global Ensemble Forecast System (GEFS) reforecast data released by the National Centers for Environmental Prediction (NCEP) has a long-term data archive and relatively stable systematic error, which shows great potential in hydrologic applications. This paper aims to bias correct the GEFS reforecast precipitation forecasts and analyze the impact of bias-corrected ensemble precipitation forecasts on the improvement of summer streamflow prediction skill. Driven by the GEFS reforecast ensemble precipitation forecasts, the Variable Infiltration Capacity (VIC) distributed hydrological model is applied to simulate the 2000–2010 summer streamflow over the Huaihe River basin.The results show that the GEFS reforecast data generally underestimates the precipitation amount with the increasing leadtime and the streamflow simulation tends to underestimate the peak due to the spatial precipitation error. In this study, both the frequency matching method (FMM) and the analog method are applied to calibrate the GEFS reforecast ensemble precipitation forecasts. The searching criteria of the analog method has been improved based on the basin-scale areal precipitation.After bias correction, the FMM method mainly improves the forecast skill of light rain and moderate rain for the 1–8 d leadtime. By contrast, the analog method improves probabilistic precipitation forecasts and increases ensemble spread to alleviate the underdispersion of raw ensemble forecasts for the 1–8 d leadtime. The analog method can also improve spatial distributions with the downscaling information from the observed precipitation, producing better probabilistic precipitation forecasts. The streamflow prediction using bias-corrected precipitation forecasts better resembles the observed streamflow. The analog-corrected precipitation forecasts can effectively improve the streamflow simulations for the 1–5 d leadtime, with the maximum improvements at the 2–3 d leadtime. The FMM method only improves the streamflow predictions for the 2 d leadtime.For the heavy flood events caused by extreme rainstorms in the summer of 2003, the ensemble mean forecasts have limited forecast skill, but probabilistic precipitation forecasts show great potential in predicting the magnitude of extreme rainfall events. Meanwhile, the ensemble streamflow predictions using the analog-corrected precipitation forecasts have larger ensemble spread and a higher Nash coefficient of extreme streamflow.The encouraging results suggest that the reforecast ensemble dataset exhibits a great value to improve hydrometeorological predictions for operational applications.