ShallowLandslider: a physics-based component for predicting regional distributions of coseismic landslides

Suryodoy Ghoshal , Sarah J Boulton , T.C. Hales, Georgina Bennett , Amy Beswick, Josh N. Jones, Shaun Lewin, Zoe K Mildon , Martin Stokes , Michael R. Z. Whitworth , Benjamin Campforts 

Earthquakes can trigger thousands of shallow landslides across mountainous terrain, reshaping landscapes and posing severe hazards. Predicting their spatial distribution remains challenging because most existing models are empirical, event-specific, and lack physical interpretability. We introduce ShallowLandslider, a physics-based component within the open-source Landlab framework for regional coseismic landslide prediction. The model extends the classical Newmark sliding block approach to three dimensions, incorporating transient seismic accelerations, slope geometry, and variable soil properties on structured grids. Instability is assessed using critical acceleration thresholds, and a probabilistic selection scheme represents natural variability in failure occurrence. We validate ShallowLandslider against landslide inventories from two subregions affected by the 2015 Mw 7.8 Gorkha earthquake in Nepal. Model performance is evaluated using non-parametric distributional metrics (Kolmogorov–Smirnov, Kuiper, and Wasserstein distances) across landslide area, elevation, slope, and aspect. Results show that realistic soil-depth parameterisations and moderate cohesion values (10-15 kPa) are essential for reproducing observed topographic clustering and size distributions. While pixel-level prediction remains impractical, ShallowLandslider captures first-order spatial and statistical patterns of coseismic landsliding, offering a reproducible, physically grounded tool for regional hazard assessment. Its modular design enables coupling with other Earth-surface process models, providing a foundation for integrated simulations of landscape response to seismic forcing.