Automatic Generation of Winter Storm Warnings

Abstract

It is a task of MeteoSwiss to warn the public and authorities of upcoming extreme weather. The current warning system is based on fixed geographical warning regions which are manually classified into levels of danger by weather forecasters. The aim of this master’s thesis is to automatize the generation of warnings of winter storms. An algorithm is designed and implemented which automatically shapes and classifies event-based warning polygons. The inputs to the algorithm are the wind speeds computed by the COSMO2-E numerical weather prediction model. The data is processed with the image processing technique Morphological Filtering to reduce the heterogeneity in the wind speed data and to obtain warning polygons. Evaluated on past winter storms, the algorithm’s probability of detection has improved by 0.11 to 0.48 while the false alarm ratio has increased by 0.1 to 0.85, compared to the preprocessed COSMO2-E wind speed output. Therefore, the implemented algorithm forms contiguous warning polygons while preserving the COSMO2-E model’s accuracy. In future work, the algorithm can be further developed to reduce heterogeneity and form warning polygons of other numerical weather predictions.