Evaluation of the predictive skill for warm spells in Germany at seasonal scale

Abstract This study evaluates the predictive skill of the downscaled German Climate Forecast System Version 2.1 (GCFS2.1) for warm spells in Germany on a seasonal scale for both warm (spring and summer) and cold seasons (autumn and winter). The analysis considers hindcast data from the 1991–2020 base period, statistically downscaled to a high spatial resolution over a region including Germany, along with five climate indices (Summer Days, Hot Days, Hot Spells Duration, Warm Days, and Warm Spells Duration). The skill is assessed using two metrics, the Mean Squared Error Skill Score (MSESS) and the Ranked Probability Skill Score (RPSS), with E-OBS data as the reference climatology. The main findings of this analysis are twofold: (1) The predictive skill for the warm seasons is highly heterogeneous, with simulations starting in March for indices like Summer Days, Hot Days, and Warm Days showing some skill in both MSESS and RPSS, whereas predictions for later starting months and the remaining indices exhibit limited skill. (2) High predictive skill is observed for the Warm Spell Duration index in MSESS and, to a lesser extent, in RPSS during cold seasons across most of Germany and simulation lead times, although skill slightly declines as lead time progresses. The degradation of the skill during warm seasons compared to that of the cold seasons is likely due to dynamical and methodological factors, such as a less accurate description of the large-scale circulation patterns like the North Atlantic Oscillation (NAO), East Atlantic (EA) or Scandinavian (SCAND) patterns and their link with extreme temperatures in Europe. This study shows that the downscaled GCFS2.1 model is a reliable tool for predicting warm spells in Germany during cold seasons.