High-Resolution Soil Moisture Maps Over Landslide Regions in Northern California Grassland Derived from SAR Backscattering Coefficients

Slow-moving landslides are destabilized by accumulated precipitation and consequent soil moisture. Yet the continuous high-resolution soil moisture measurements needed to aid understanding of landslide processes are generally absent in steep terrain. Here we produce soil moisture time-series maps for a seasonally active grassland landslide in the northern California Coast Ranges, USA using backscattering coefficients from NASAs Uninhabited Aerial Vehicle Synthetic Aperture Radar (UAVSAR) at 6-m resolution. A physically based radar scattering model is used to retrieve the near-surface (5-cm depth) soil moisture for the landslide. Both forward modelling (backscattering estimation) and the retrieval (soil moisture validation) show good agreement. The root mean square errors (RMSE) for VV and HH polarizations in forward model comparison are 1.93 dB and 1.88 dB, respectively. The soil moisture retrieval shows unbiased RMSE (ubRMSE) of 0.054 m3/m3. Our successful retrieval benefits from surface and double-bounce scattering, which is common in grasslands. The retrieved maps show saturated wetness conditions within the active landslide boundaries. We also performed sensitivity tests for incidence angle and found that the retrieval is weakly dependent on the angle especially while using HH and VV together. Using the two co-pol inputs, the retrieval is also not sensitive to the change of orientation angles of grass cylinders. The physical model inversion presented here can be generally applied for soil moisture retrieval in areas with the same vegetation cover types in California.