Abstract:
Hydrological triggering of landslides is strongly connected to the water content of the soil. Existing models that predict landslides mostly rely on antecedent rainfall indices as a proxy for soil moisture conditions, as precipitation data has been perceived to be more readily available than soil moisture data. Our research question is whether landslide prediction can benefit from using more precise information of soil moisture conditions instead of relying on proxies.

To tackle this question we examined soil hydrological conditions at times and locations of known landslide occurrences (Global Landslide Catalog, Kirschbaum et al. 2015). More specifically, we investigated surface and root zone soil moisture, as well as total water storage estimates simulated by the Catchment Land Surface Model (CLSM, Koster et al. 2000), or resulting from assimilation of surface soil moisture retrievals from the Soil Moisture Ocean Salinity (SMOS) mission or terrestrial water storage retrievals from the Gravity Recovery and Climate Experiment (GRACE) mission.

A first coarse-scale global analysis for the years 2011 through 2018 indicates that soil moisture and total water storage estimates are adequate alternatives to antecedent rainfall indices for the prediction of landslides. About 50% of the landslides occurred at soil moisture or total water storage values above the local 90th percentile. The same conclusion is found for results based on CLSM only and SMOS or GRACE data assimilation, because the modeled water storage climatology is roughly maintained in all three experiments by design. In contrast, SMOS or GRACE satellite observations by themselves are too sparse and noisy to clearly distinguish the different hydrological conditions at landslide and non-landslide events. The data assimilation schemes did not yet introduce any significant improvement in our ability to predict landslides, but local improvements and degradations are found and further investigated.