Long and short-term perspectives on space–time landslide modelling

Abstract

Data-driven models applied to landslide prediction have historically been mostly confined to the pure spatial context, as per landslide susceptibility requirements. Its standard definition assumes that the occurrence probability is conditional on a broad set of static predictors and that in turn, it does not change with time. To find data-driven models where the probability is temporally dynamic, we need to explore early-warning systems. However, these models traditionally rely only upon rainfall (intensity-duration characteristics) and neglect influences from terrain, geological, and other thematic contributors. Space-time data-driven models can incorporate both static and dynamic predictors, allowing for a rich description of the landslide process and for the susceptibility to change both in space and time. In this work, we present an overview of potential variations of space–time landslide susceptibility models for an area in Chongqing, China. In doing so, we present space–time models suited for long-term (yearly or seasonal models) or short-term (monthly or daily) planning. Therefore, the manuscript presents elements of a review as well as elements of methodological innovation. The method of choice used across all the experiments corresponds to a Generalized Additive Model, whose structure will account for linear, nonlinear, spatial, and temporal effects.