Calibration and Downscaling Methods for Quantitative Ensemble Precipitation Forecasts

Abstract

Abstract Two approaches for downscaling and calibrating error estimates from ensemble precipitation forecasts are evaluated; the two methods are intended to be used to produce flood forecasts based on global weather forecasts in ungauged river basins. The focus of this study is on the ability of the approaches to reproduce observed forecast errors when applied to daily precipitation forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF) Ensemble Prediction System (EPS) for a 10-day forecast period. The two approaches are bias correction with spatial disaggregation (BCSD) and an analog technique. Mean forecast errors and skills are evaluated with respect to Tropical Rainfall Monitoring Mission (TRMM) observations over the Ohio River basin for the period 2002–06 for daily and 5-day accumulations and for 0.25° and 1° spatial resolutions. The Ohio River basin was chosen so that a relatively dense gauge-based observed precipitation dataset could also be used in the evaluation of the two approaches. Neither the BCSD nor the analog approach is able to improve on the forecast prediction skill resulting from a simple spatial interpolation benchmark. However, both approaches improve the forecast reliability, although more so for the analog approach. The BCSD method improves the bias for all forecast amounts (but less so for large amounts), but the downscaled precipitation patterns are unrealistic. The analog approach reduces biases over a wider range of forecast amounts, and the precipitation patterns are more realistic.