C5 - Dynamical feature-based ensemble postprocessing of wind gusts within European winter storms
Other researcher: Lea Eisenstein (PhD)
Wind gusts associated with cyclonic storms are among the most significant natural hazards in central Europe. Forecasting these with sufficiently high accuracy to issue effective warnings to the population remains a challenge. In Phase 1, research was organized around the idea of gust forecasting as a multi-scale problem. On the synoptic scale, an analysis of retrospective ensemble forecasts by the European Centre for Medium-Range Weather Forecasts (ECMWF) showed that a potential for early warning exists up to 10 days ahead but that satisfactory accuracy for cyclone position and intensity is limited to lead times of 2–4 days. On the mesoscale, distinct high-wind areas within cyclonic storms can be identified that are responsible for most damages. A novel 6-year dataset of 21-hour convection-permitting ensemble forecasts from the Deutscher Wetterdienst (DWD) was evaluated with respect to wind and strong gust forecasts showing that predictability is reduced in cases involving frontal convection. In some situations errors are largely inherited from the synoptic scale, leading to high forecast uncertainty even at short range. In collaboration with Project C7, ensemble postprocessing methods were applied showing overall performance improvement, but also deterioration for storms with uncharacteristic forecast errors. Finally, the role of boundary-layer processes in gust generation was investigated using high-resolution observations and modeling, including scanning Doppler lidar and tower measurements conducted at Karlsruhe in winter 2016-17 as well as large-domain, large-eddy simulations over Germany.
Building on these results, Phase 2 will concentrate on the development of ensemble postprocessing methods for individual dynamical features within winter storms over central Europe. The basic idea behind this is to objectively identify characteristic subregions of cyclonic storms such as cold, warm and sting jets as well as convective gusts, and to then conduct dynamical analysis, forecast evaluation, and ensemble postprocessing separately. We hypothesize that these features show distinguishable behavior with respect to their forecast performance and error characteristics. The data basis will be an extension of the high-resolution ensemble forecasts and observations over Germany (mostly surface stations) already used in Phase 1, complemented with coarser-resolution forecasts from the ICON (Icosahedral Nonhydrostatic) and ECMWF models. Necessary prerequisites are an improved understanding of mesoscale dynamics and based on this the development of an objective identification algorithm for the dynamic subregions, which will be achieved using suitable case studies and 3D visual analysis techniques in collaboration with Project C9. The feature-dependent postprocessing development will use modern machine learning methods and customized flow-dependent similarity-based estimation techniques, ensuring physical consistency. The new results will be compared to currently operational and alternative statistical methods in regular exchange with potential operational users such as DWD and ECMWF.