Urban landslides triggered under similar rainfall conditions in cities globally

As growing cities expand into steeper terrain, urbanization activities like clearing vegetation, cutting and filling slopes, and building infrastructure can increase rainfall-triggered landslide hazard compared to unmodified slopes. Landslide early warning systems can help to reduce rainfall-triggered landslide risk in susceptible areas, but few cities worldwide have established dedicated systems. Such systems often rely on rainfall thresholds that identify landslide triggering conditions, but determining these thresholds requires landslide inventory data that is not available everywhere. Furthermore, the variability of thresholds between cities, and the applicability of previously estimated global thresholds to urban areas has yet to be assessed, such that cities with limited landslide inventory data have few options to learn from areas with more information. Here, we compiled 1216 rainfall-triggered urban landslide records from 26 cities worldwide to address two open questions: 1. how variable are rainfall thresholds between cities? and 2. how do global rainfall thresholds for urban landslides compare to previously estimated thresholds from multiple land use types?  Using hourly, station-based precipitation data from the Global Sub-daily Rainfall Dataset (Lewis et al., 2019), we applied Bayesian multi-level quantile regression to estimate intensity-duration thresholds for each city and a global threshold for urban landslides.  We found that landslides were triggered under surprisingly similar rainfall conditions in most cities despite widely varying climates, topographies, and income classes.  Median thresholds in cities with the highest and lowest mean annual precipitation were not credibly distinguishable, and in 77% of cities, the median threshold was indistinguishable from the global average. We show that urban landslides occurred at lower threshold intensities than previously reported for multiple land-use types, and that 31% of landslides occurred during moderate storms, not only during extreme rainfall. Our results suggest that urban hillslopes may be more adjusted to urbanization activities than to local environmental conditions, leading to similar thresholds between cities. Reports of urban landslides at relatively low intensity rainfall likely also reflects the role of anthropogenic hillslope modification and malfunctioning infrastructure in causing and triggering failures.  We offer a baseline for warning in cities with sparse landslide records and suggest that future updates to regional and global landslide early warning systems consider differing thresholds for urban and rural regions.