Can we detect hail from GNSS observations? Case studies from severe weather events in Switzerland 2021

Tropospheric products from Global Navigation Satellite Systems (GNSS) have become a vital data source for Numerical Weather Prediction (NWP) and meteorology in general over the last decades. Commonly derived parameters like path delays and integrated water vapor are utilized for data assimilation and have proven to be beneficial for precipitation forecasts. In recent years, also other meteorological phenomena and parameters, such as soil moisture variations, snow properties or foehn winds, have been studied by means of GNSS remote sensing. However, typical troposphere products only consider the gaseous constituents of water to influence GNSS observations. Therefore, GNSS troposphere products neglect the effect of hydrometeors (liquid and solid particles of water) present in the air. Typically, their contribution to tropospheric delays is small, but becomes significant for severe weather events.

This study investigates the signature of hydrometeors, in particular hail, in time series of GNSS data, aiming to detect precursors of hail formation in the vicinity of GNSS stations. Hail represents one of the most treacherous types of severe weather and, at the same time, can only be detected from radar images. As radar observing systems are cost-intensive and therefore very sparsely distributed, the possibility of hail detection from GNSS observations would be very beneficial, especially for nowcasting applications.

The study presents an analysis of different hail events, e.g. the severe thunderstorm happening over the city of Zurich in the night of 12/13.07.2021, which caused tremendous damages on infrastructure. Therefore, a combination of different GNSS observations (tropospheric delays, signal-to-noise ratio, carrier phase residuals) will be used to investigate potential signs of the hail formation and compared to operational radar observations and event reports. Finally, an outlook on the usability of data-driven methods such as machine-learning-based detection methods for this use case will be given.