Deriving rainfall thresholds with XAI and GNSS measurements for a large landslide

Abstract Landslides represent one of the most serious natural hazards in Italy, frequently resulting in fatalities and several damage to infrastructure. Recent advancements in Global Navigation Satellite Systems (GNSS) have enabled the development of monitoring systems that can monitor slope displacements continuously, accurately, and at relatively low cost. In this study, we present a monitoring network based on single-frequency GNSS sensors, integrated with rainfall data, applied to a large landslide affecting a village in the Carnic Alps (northern Italy). The objective is to showcase a practical example of continuous landslide monitoring using GNSS technology. By implementing a specific velocity threshold for GNSS measurements, we created an inventory of landslide reactivation events, which was then used to define rainfall thresholds based on data from two pluviometer stations. To enhance the interpretability of machine learning (ML) models used in the analysis, we adopted an innovative approach employing Partial Dependence Plots (PDPs). This technique allowed for a straightforward assessment of the influence of cumulative rainfall across various time windows, facilitating the identification of the most critical accumulation periods.