MAGIC-CH: Machine Learning-based Advancement and Usability Assessment of GNSS Interferometric Reflectometry for Climatological Studies in Switzerland

The MAGIC-CH project aims to advance the application of environmental products for climatological monitoring derived from observations of Global Navigation Satellite Systems (GNSS) Interferometric Reflectometry (IR). The project explores the potential of the existing Swiss GNSS infrastructure for monitoring essential climate variables, including snow, soil moisture, and atmospheric water vapor.

In this contribution, we aim to apply machine learning techniques to directly retrieve soil moisture and snow heights from GNSS-IR observables. Time series of signal-to-noise ratio (SNR) from ground-reflected GNSS signals are utilized as features, while satellite-based soil moisture data and in-situ snow height observations serve as target variables. Additionally, azimuth and elevation angle, day of the year, and a digital elevation model are used as inputs in the machine learning framework.

Preliminary results for soil moisture retrieval are based on the XGBoost algorithm, using GNSS data from the Automated GNSS Network for Switzerland (AGNES) and the 1 km surface soil moisture product provided by the Copernicus Global Land Monitoring Service. For snow height, initial results are based on an artificial neural network, GNSS-IR measurements of the Plate Boundary Observatory, and snow height observations of SNOTEL sites. The performance of these machine learning models shows promising improvements, significantly reducing standard error measures compared to traditional retrieval methods.