Space-time explainable modelling of regional hillslope deformation, an example from the Tibetan Plateau

The future of InSAR applications will undoubtedly involve data-driven solutions to predict deformation across space and time. Recent advancements in subsidence research have already integrated such approaches, primarily in flat to near-flat landscapes. However, in mountainous terrains, space-time InSAR modelling has so far focused mainly on individual slopes or small catchments. Here, we propose a modelling protocol based on a deep learning architecture capable of predicting InSAR-derived hillslope deformation. This approach is developed primarily using morphometric and meteorological variables over extensive mountainous areas (∼15,000 km²) and extended time windows (∼7 years). By aggregating the deformation signal at the Slope Unit scale while maintaining 12-day temporal intervals consistent with Sentinel-1 acquisitions, we achieve high modelling performance (PCC = 0.7). If validated in other regions, this method could represent a crucial step towards a large-scale, consistent, and highly effective scenario-based warning system for hillslope deformation.