Forecasting of Precipitation-Induced Landslides Using Atmospheric Rivers: Opportunities and Challenges

Landslides are particularly costly disasters, causing about 4,500 fatalities and US$20 billion in damages worldwide each year. In Western North America, where intense and frequent precipitation events interact with complex topography and steep slopes, precipitation-induced landslides (PILs) are a serious geological hazard. Recently, it has been revealed that the majority of PILs in the region are triggered by precipitation from atmospheric rivers (ARs), transient channels of intense water vapor flux in the troposphere. However, the synoptic conditions differentiating landslide-triggering and non-triggering ARs remain unknown. In this study, we explore opportunities for improved landslide forecasting in Western North America using catalogs of land-falling ARs and PILs, along with ERA5 climatological data, from 1996 to 2018. First, we employ event synchronization, a non-linear measure specially tailored for event series analysis, to identify landslide-triggering ARs. Based on the AR-strength scale, which ranks ARs in levels from 1 to 5, we further characterize landslide-triggering ARs in terms of intensity and persistence. Subsequently, we spatially resolve the conditional probability of PIL occurrence given the detection of AR-attributed precipitation in the antecedent week, revealing the contribution of each AR level. Lastly, using hourly estimates of integrated water vapour transport, geopotential height, and precipitation at 0.25° spatial resolution, we differentiate the spatio-temporal evolution of synoptic conditions preceding landslide-triggering and non-landslide triggering ARs. Our results constitute a first, fundamental, and necessary step toward AR-based landslide forecasts, contributing crucial insights to improve forecasting accuracy at the short and early medium-range (1–7 days).