Rainfall Classification in Genoa: Machine Learning Versus Adaptive Statistical Models Using Satellite Microwave Links

Monitoring rainfall is becoming increasingly important due to the impacts of climate change. In this context, opportunistic sensing based on satellite microwave links (SMLs) is gaining significant interest for its potential to provide real-time, low-cost, and valuable information. In this work, we deployed a network of 26 SML sensors over a survey territory corresponding to a catchment area of 140 km2, and the data collected from it were used to train two machine learning (ML) algorithms and a statistical method to predict wet/dry conditions. Additionally, a voting mechanism based on the majority vote of the three classifiers enhances the performance of the individual methods. The results show that ML algorithms are valuable solutions for real-time monitoring of rainfall, even in an extended sensor network. The voting mechanism improves the prediction of low precipitation at the expense of a higher false positive rate.