Rainfall induced shallow landslides: rainfall thresholds and antecedent conditions

Landslides are a natural hazard affecting alpine regions all over the world. They cause not only substantial economic damages, but also human casualties. The focus here is on rainfall induced shallow landslides, which happen following an increase in the pore water pressure in the soil. As the name suggests, this typically occurs after rainfall events, either prolonged in time or short but intense, and combining such rainfall data with landslide inventories allows the definition of landslide-triggering rainfall thresholds. Nevertheless, it is now widely accepted that antecedent conditions, i.e., the wetness of the soil prior to the (non) triggering rainfall, also plays an essential role. Not accounting for the soil condition prior to the rainfall event is the main limitation of rainfall thresholds, together with the fact that they do not consider spatial heterogeneities within the domain.

Here we take advantage of two long records of daily rainfall (MeteoSwiss) and landslides events (WSL) existing in Switzerland, as well as the hydrological estimates provided by two hydrological forecasting systems operational over Switzerland. We use these not only to confirm the importance of antecedent conditions, but also to explore how to best exploit them to improve upon classical rainfall thresholds to predict landslide occurrence.

We start by considering antecedent rainfall and demonstrate that it is helpful in reducing the misclassification associated with rainfall thresholds: missed landslide events are anticipated by high N-day antecedent rainfall, while false alarms by low N-day antecedent rainfall. Recognising the limit of this simple proxy of antecedent conditions, which cannot account for snowmelt or water redistribution, we proceed by considering the soil saturation provided by a) a European physically based hydrological forecasting system (TerrSysMP) and b) a Swiss conceptual hydrological model (PREVAH). The comparison between these two systems leads to the following main findings. First, the soil saturation estimates provided by PREVAH are more informative for landslides prediction, due to a much higher spatial resolution (Prevah 250m while TerrSysMP 12.5km). Second, if spatial heterogeneities in triggering conditions are considered by using the hydrological soil wetness estimates for the calculation of the Factor of Safety (infinite slope stability model), the separation between triggering and non-triggering conditions improves compared to just using saturation. Third, while the information content of antecedent conditions is evident, accounting for them in a regional warning system is not straightforward. In fact, we find a classical hydrometeorological threshold (with a measure of antecedent conditions on the x-axis and a measure of triggering rainfall on the y-axis) to be less successful than a pure rainfall threshold. Instead, we propose a sequential threshold, where first a soil saturation threshold is used to separate “wet” and “dry” conditions, and then 2 rainfall thresholds are utilised for the wet and dry antecedent conditions.