Beyond Coarse Data: Soil Thickness and Rainfall forLandslide Hazard Modelling

An increasing population in mountainous regions, where gentle and stable topography is scarce, drives residents to settle on steep slopes. These slopes are particularly prone to shallow landslides, which involve the displacement of the upper soil layers and are more easily triggered by rainfall. Therefore, accurate landslide hazard models are needed to safeguard populations.

These models typically include spatial data, such as soil thickness and rainfall. However, the lack of detailed inputs often means that models operate at coarse scales, which can mask local variability and potentially underestimate hazard levels. To address this gap, our research question is whether simulations of shallow landslides can be improved by enhancing the spatial resolution of two critical variables derived from coarse satellite data: (i) soil thickness determining the volume of material available for sliding, and (ii) rainfall controlling soil saturation and pore-water pressure dynamics.  

To demonstrate the scalability and applicability of the method to other regions prone to landslides, we tested this approach in La Estrella, Colombia, a municipality with a long history of landslides and rapid population growth on steep slopes. For soil thickness, we applied a geomorphological model that relates soil depth to slope angle and distance to the drainage network. We validated the estimates against borehole measurements, finding strong agreement at three of five test sites. For rainfall, we integrated CHIRPS with local rain-gauge data, using spatial interpolation and regression-based downscaling to produce high-resolution rainfall fields. The downscaling model was then evaluated using statistical metrics, including the Pearson correlation coefficient (r), bias, and Nash–Sutcliffe efficiency (NSE).

In the next step, we will feed these two outputs into a Landlab shallow landslide probability model that couples hydrological response with soil mechanical stability. This will allow us to quantify the influence of input resolution on predicted landslide probability patterns.