Influence of river incision on landslides triggered in Nepal by the Gorkha earthquake: Results from a pixel-based susceptibility model using inlabru

TL;DR

This paper introduces a novel, physically meaningful covariate, the normalized channel steepness index, into a Bayesian spatial model to improve landslide susceptibility mapping after earthquakes, with a focus on transferability and uncertainty quantification.

Contribution

The study develops a new framework integrating the $k_{sn}$ covariate into a Bayesian model using inlabru, enhancing landslide prediction accuracy and transferability across regions.

Findings

Elevated $k_{sn}$ correlates with higher landslide susceptibility.

The model effectively predicts landslide locations and sizes.

Uncertainty quantification improves model reliability.

Abstract

This study presents a comprehensive framework for modelling earthquake-induced landslides (EQILs) through a channel-based analysis of landslide centroid distributions. A key innovation is the incorporation of the normalised channel steepness index ($k_{sn}$) as a physically meaningful and novel covariate, inferring hillslope erosion and fluvial incision processes. Used within spatial point process models, $k_{sn}$ supports the generation of landslide susceptibility maps with quantified uncertainty. To address spatial data misalignment between covariates and landslide observations, we leverage the inlabru framework, which enables coherent integration through mesh-based disaggregation, thereby overcoming challenges associated with spatially misaligned data integration. Our modelling strategy explicitly prioritises prospective transferability to unseen geographical regions, provided that explanatory variable data are available. By modelling both landslide locations and sizes, we find that elevated $k_{sn}$ is strongly associated with increased landslide susceptibility but not with landslide magnitude. The best-fitting Bayesian model, validated through cross-validation, offers a scalable and interpretable solution for predicting earthquake-induced landslides in complex terrain.