Improving seasonal predictions of German Bight storm activity

Abstract

Extratropical storms are one of the major coastal hazards along the coastline of the German Bight, the southeastern part of the North Sea, and a major driver of coastal protection efforts. However, the predictability of these regional extreme events on a seasonal scale is still limited. We therefore improve the seasonal prediction skill of the Max Planck Institute Earth System Model (MPI-ESM) large-ensemble decadal hindcast system for German Bight storm activity (GBSA) in winter. We define GBSA as the 95th percentiles of three-hourly geostrophic wind speeds in winter, which we derive from mean sea-level pressure (MSLP) data. The hindcast system consists of an ensemble of 64 members, which are initialized annually in November and cover the winters of 1960/61–2017/18. We consider both deterministic and probabilistic predictions of GBSA, for both of which the full ensemble produces poor predictions in the first winter. To improve the skill, we observe the state of two physical predictors of GBSA, namely 70 hPa temperature anomalies in September, as well as 500 hPa geopotential height anomalies in November, in areas where these two predictors are correlated with winter GBSA. We translate the state of these predictors into a first guess of GBSA and remove ensemble members with a GBSA prediction too far away from this first guess. The resulting subselected ensemble exhibits a significantly improved skill in both deterministic and probabilistic predictions of winter GBSA. We also show how this skill increase is associated with better predictability of large-scale atmospheric patterns.