Probabilistic Predictions of Precipitation Using the ECMWF Ensemble Prediction System

Abstract

The forecast skill of the European Centre for Medium-Range Weather Forecasts Ensemble Prediction System (EPS) in predicting precipitation probabilities is discussed. Four seasons are analyzed in detail using signal detection theory and reliability diagrams to define objective measure of predictive skill. First, the EPS performance during summer 1997 is discussed. Attention is focused on Europe and two European local regions, one centered around the Alps and the other around Ireland. Results indicate that for Europe the EPS can give skillful prediction of low precipitation amounts [i.e., lower than 2 mm (12 h)?1] up to forecast day 6, and of high precipitation amounts [i.e., between 2 and 10 mm (12 h)?1] up to day 4. Lower levels of skill are achieved for smaller local areas. Then, the EPS performance during summer 1996 (i.e., prior to the enhancement introduced on 10 December 1996 from 33 to 51 members and to resolution increase from T63 L19 to TL159 L31) and summer 1997 are compared. Results show that the EPS has been remarkably more skillful during summer 1997 than summer 1996, with the gain in predictability up to 3 days for the highest [5 and 10 mm (12 h)?1] amounts of precipitation. Finally, the EPS performance during wintertime is analyzed. Two issues are investigated: the seasonal variability of the forecast skill of the new EPS, and the impact of the system upgrade on the wintertime performance. The comparison of the performance of the new EPS system during winter 1996/97 and during summer 1997 indicates that the EPS is more skillful during winter than during summer, with differences in predictive skill around 3 days for precipitation amounts larger than 2 mm (12 h)?1. The comparison of the EPS performance before and after the system upgrade on 10 December 1996 during winter confirms the summer conclusion that the upgraded system is more skillful than the old one.