Optimising landslide initiation modelling with high-resolution saturation prediction based on soil moisture monitoring data

It has been widely recognised that the degree of soil wetness before precipitation events can be decisive for whether or not shallow rainfall-induced landslides occur. While there are methods to measure and/or model soil wetness in complex topography, they often exhibit limitations in spatial or temporal resolution, hindering their application in regional landside initiation modelling. In this study, we address the need for high-resolution predictions of initial saturation before rainfall events by employing data-driven linear regression models. The models were trained using in-situ soil moisture data collected from six measurement stations located in a landslide-prone region in Switzerland. Various topographic attributes, along with multiple antecedent rainfall and evapotranspiration variables were tested as input for the models. The final model consisted of five measurable variables, including cumulative antecedent rainfall, cumulative evapotranspiration, and the topographic wetness index (TWI). The model effectively reproduced the observed spatial and temporal variability of the in-situ measurements with a coefficient of determination R² = 0.62 and a root mean square error RMSE = 0.07. Subsequently, we applied the regression model to predict the spatial soil saturation at the onset of actual landslide triggering rainfall events and integrated these patterns into the hydromechanical model STEP-TRAMM. The results demonstrate improvements in predicting observed landslide occurrences compared to simulations assuming spatially uniform initial saturation conditions, highlighting the importance of in-situ measurements and a realistic extrapolation of such data in space and time for accurate modelling of shallow landslide initiation.