Shallow landslide diagnostics for Southern California

Shallow landslides are significant natural hazards in the mountainous regions of Southern California during anomalously wet winter seasons, posing substantial risks to life and property. Better understanding and operational prediction of the atmospheric and land-surface conditions leading to the initiation of these impactful events are critical for mitigating life loss and reducing damage costs. This study aims to characterize and provide templates for operational prediction of the triggering conditions of shallow landslides in Southern California with the application of the MHEBC, a decade-long, high-resolution, prototype dataset of quantitative precipitation estimates. This dataset has region-specific calibration and is the result of methodologies tailored to capture the orographic signal of precipitation, which is a key element for the generation of the precipitation associated with landslide occurrence in this region. Using recorded landslide events from the California Geological Survey Reported Landslides Database, rainfall thresholds for landslide occurrence are derived on the basis of the MHEBC dataset. Significant quality control was necessary for both rainfall and landslide databases before high-quality data input could be provided for the analysis. Observed local landslide data availability is typically limited worldwide, and Southern California is no exception. For such conditions, estimates of landslide occurrence are obtained using manually-derived precipitation thresholds, following well defined operational procedures. In addition, a Random Forest (RF) machine-learning model and a few of its variants were applied to the same high quality but limited dataset to examine its utility for operational prediction under typically limited data. Under a variety of performance measures the automated RF does only marginally better than the well-established manual threshold procedure. Overall, the analysis demonstrates a valuable hydrological land surface application of the MHEBC dataset and provides insights towards skillful regional landslide forecasting in Southern California.