Disentangling Static and Dynamic Landslide Drivers: A Dual-Phase Bayesian Framework for a Paraglacial Tyrolean Basin

Abstract In high-relief Alpine environments, the spatial and temporal distribution of landslide hazards is governed by a non-linear interplay between long-term topographic evolution and short-term meteorological forcing. Traditional susceptibility models often conflate these distinct driver categories, relying on static climatological averages (e.g., Mean Annual Precipitation) that may lack deterministic power in rapidly changing paraglacial landscapes. Here, presented a dual-phase Bayesian analytical framework applied to a 712 km² Tyrolean basin to rigorously decouple these drivers. Phase 1 employs an Integrated Nested Laplace Approximation (INLA) with Stochastic Partial Differential Equations (SPDE) to model static susceptibility while filtering spatial confounding. Phase 2 utilizes Firth’s Penalized Logistic Regression to resolve the "small " problem inherent in analyzing rare triggering events. The results reveal a fundamental decoupling: susceptibility is governed by "Topographic Determinism" (Planar Curvature, Slope Position), while hazard is driven by "Meteorological Transience." A robust negative result is reported regarding Mean Annual Precipitation, which is statistically insignificant, suggesting the landscape operates under a transport-limited regime. Conversely, dynamic analysis identifies a "Convective Switch" mechanism, where rainfall intensity acts as a binary trigger. These findings provide the mechanistic link between recent observations of increasing hourly rainfall extremes and rising debris flow frequencies in the Alps.