Classification of rain events using directional radio data of commercial microwave links

Abstract

Due to climate change, more and more extreme weather events are occurring. An accurate short-term forecast in terms of time and location represents a significant advantage for taking appropriate measures to prevent damage and to react and plan more efficiently. This requires a network of ground stations or remote sensing systems such as weather radar or satellites as dense as possible. In large parts of Austria, however, rough terrain limits the number of measuring stations and radar data are also only available to an insufficient extent in certain areas due to the topography. We aim to overcome these challenges by using physical data of directional radio links scattered across Austria to obtain information about the current precipitation situation. In this work, we introduce an approach for classifying rain events using a variety of different machine learning methods. The results can be used to improve numerical weather prediction models.