Error growth in poor ECMWF forecasts over the contiguous United States

Abstract

Successive improvements to the European Center for Medium-range Weather Forecasting model have resulted in improved forecast performance over the Contiguous United States (CONUS). While the overall performance of the model in this region was found to have improved during the period of the 1981-1990 winter seasons, the number of poor forecasts increased over this time. This study uses the Root Mean Square (RMS) error to measure the performance of 5-day 500 mb winter forecasts over the CONUS. Poor and good forecasts are defined in terms of the 10-year distribution of the RMS values between the 1981 and 1990 winter seasons. Subjective analysis of a subset of poor forecasts yielded no obvious patterns of error growth, location or propagation in the evolution of poor forecasts. A tendency is noted for in situ amplification of forecast errors. Additionally, successive forecasts verifying on the same day are found to have similar error patterns, with increased amplitudes at longer forecast lengths. This implies that the initial conditions are not a significant source of the error in poor forecasts. Empirical Orthogonal Function (EOF) analysis of error growth in time and space reveals significant differences between poor and good forecasts. Good forecasts are found to have the majority of RMS growth on day I while poor forecasts do not experience rapid error growth until days 3 and 4. For poor forecasts, the leading EOFs reveal a wave pattern down stream of the Rocky Mountains. This pattern evolves and propagates throughout the forecast period until it dominate the 5-day error field. No similar pattern is revealed in the error fields of good forecasts. This pattern suggests a dynamic link between the Rockies and the zonal wind, although no link with the 500 mb geostrophic wind could be established.