From automated inventory enrichment to interpretable multiclass modeling of extreme rainfall-induced mass wasting

Abstract Climate change is amplifying rainfall-induced mass-wasting risks, while limited detail on failure types and source areas in current inventories hinders mechanistic insight. Here we present a spatial algorithm-based automatic labeling method for enhancing mass-wasting inventories. Events are distinguished into four types based on movement behavior. A visually interpreted inventory was created from the July 2023 extreme precipitation event in Beijing, China, to validate the proposed method. Results show an overall consistency of 82% to 85% between manual and automatic labeling, with lower consistency for debris floods (59%). The method was then applied to a larger inventory to automatically identify failure types and delineate source areas. Using Shapley Additive Explanations, we quantified factor contributions by mass-wasting types based on the automatic labels in a downstream analysis. This method enhances inventories without large labeled datasets and provides type-specific insights into controls on mass wasting.