Developments of a targeted ensemble prediction system

Abstract

AbstractBased on results presented in a previous paper, an Ensemble Prediction System, especially designed for the short‐range to early‐medium range for the European domain (TEPS), was further developed. Between 1 January 1999 and 27 February 2000 this system was run for two cases per week. Results were compared to the ensemble prediction system (EPS) that was operational at the European Centre for Medium‐Range Weather Forecasts at the time. The focus was on weather parameters. Several configurations were tested. Besides perturbations that optimize total error growth for the European domain at 48 h, perturbations relevant for shorter forecast times were included as well. Firstly, in accordance with EPS, evolved northern hemispheric singular vectors (ETEPS) and secondly, targeted singular vectors with an optimization time of 12 h (STEPS) were added. Compared to EPS, ensemble spread was increased in the short range, having a beneficial effect on statistical properties such as reliability and the frequency of outliers. Spread–skill correlations improved as well. Best performance was found for the STEPS configuration. The impact on Brier scores, relative operating characteristic curves and cost‐loss analyses was less clear. From the singular vector point of view, the positive impact of using targeted perturbations is evident. However, after integration (STEPS or ETEPS), it was difficult to find synoptic cases in which this clear impact is still present. Copyright © 2003 Royal Meteorological Society