Landslides can cause serious damage, but early warning systems are a way to help prevent it. Recent research shows that knowing how wet the soil is before it rains can improve landslide forecasts. In this study, we use only on-the-ground (in-situ) measurements of soil moisture and weather data to create models that predict where and when landslides might happen in Switzerland.

We combined information from 86 soil stations that measure how much water is in the ground along with weather data from nearby stations. Using modern machine learning—specifically neural networks—we analyze this data and estimate the chance of a landslide.

Our results show that this method works better than traditional techniques. These new models can recognize complex patterns between soil wetness and rainfall that simpler methods miss. We also found that including soil moisture helps predict landslides caused not just by rain, but also by other factors like melting snow.