Skill of 2-m Temperature Seasonal Forecasts over Europe in ECMWF and RegCM Models

Abstract

arious measures of forecast quality are analyzed for 2-m temperature seasonal forecasts over Europe from global and regional model ensembles for winter and summer seasons during the period 1991 to 2001. The 50-km Regional Climate Model (RegCM3) is used to dynamically downscale nine-member ensembles of ECMWF global experimental seasonal forecasts. Three sets of RegCM3 experiments with different soil moisture initializations are performed: the RegCM3 default initial soil moisture, initial soil moisture taken from ECMWF seasonal forecasts, and initial soil moisture obtained from RegCM3 ECMWF interim Re-Analysis (ERA-Interim)-driven integrations (RegCM3 climatology). Both deterministic and probabilistic skill metrics are estimated.The better-resolved spatial scales in near-surface temperature by RegCM3 do not necessarily lead to the improved regional model skill in the regions where systematic errors are large. The impact of initial soil moisture on RegCM3 forecast skill is seen in summer in the southern part of the integration domain. When regional model soil moisture was initialized from ECMWF seasonal forecasts, systematic errors were reduced and deterministic skill was enhanced relative to the other RegCM3 experiments. The Brier skill score for rare cold anomalies in this experiment is comparable to that of the global model, whereas in other experiments it is significantly smaller than in global model.There is no major impact of soil moisture initialization on forecast skill in winter. However, some significant improvements in RegCM3 probabilistic skill scores for positive anomalies in winter are found in the central part of the domain where RegCM3 systematic errors are smaller than in global model.