High-resolution precipitation forecasts for Hurricane Helene improve landslide predictions

Hurricane Helene poured rain on the southern Appalachian Mountains, triggering thousands of landslides. The day before, a global landslide forecast that relied on a deterministic coarse-resolution precipitation forecast identified the potential for most (57%) of the reported landslides. This raised the question: could a high-resolution ensemble forecast improve the outcome? In this case study, observations from rain gauges and gridded quantitative precipitation estimates were compared to precipitation from the Goddard Earth Observing System (GEOS) forward processing (FP) model, an experimental version of GEOS, the High-Resolution Rapid Refresh (HRRR) model, and the North American Mesoscale (NAM) model. Relative to reference precipitation, HRRR performed the best at hourly, daily, and two-day resolutions. HRRR also captured the spatial structure of landslide-triggering precipitation better than the other forecasts, while neither GEOS-FP nor NAM captured the heaviest precipitation. Next, each precipitation forecast was ingested into the global Landslide Hazard Assessment for Situational Awareness system. Then, the hindcasts were combined into two ensembles: one with all four hindcasts and one with the high-resolution hindcasts. All were assessed with reference to an inventory of landslides triggered by Hurricane Helene. With the lowest false negative rate (0.16) and highest area under the receiver operating characteristic curve (0.88), the HRRR- based landslide hindcast outperformed the others, including both ensemble means. In conclusion, the best landslide forecast was not an ensemble, but the landslide hindcast derived from the currently operational HRRR, which should encourage development of better landslide forecasts for the southeastern United States.