To comprehensively assess regional landslide hazards, we propose a geospatial approach that jointly evaluates both the probability of occurrence (susceptibility) and potential destructive power (intensity) within a single framework, overcoming the limitations of previous studies that treated these two disaster scenarios independently. Focusing on the largest landslide event triggered by the Wenchuan earthquake in China, we collected landslide occurrence and count data at the slope unit level, alongside 18 environmental factors, including seismic data. To enable this multi-hazard single-framework evaluation, we employed two Bayesian spatial joint regressions: the spatial shared component model (SSCM) and the spatial shared hyperparameter model (SSHM). This joint assessment focuses on three key components: identifying shared influencing factors, capturing shared spatial autocorrelated random effects, and jointly predicting susceptibility and intensity maps. Additionally, we enhanced the traditional absolute intensity index into the relative intensity by accounting for slope unit size. Both Bayesian SSCM and SSHM, incorporating multiple environmental drivers (seismic, topographical, geological, hydrological, and human activities), successfully evaluated landslide susceptibility and intensity under a single analytical frame. SSHM outperformed SSCM in terms of model fit and predictive accuracy, as revealed by cross-validation. While SSCM overfitted the landslide distribution of spatial autocorrelated random effect, SSHM provided a smoother, more spatially diverse representation. Both models consistently identified slope as the shared key factor influencing susceptibility and intensity, with the top four additional environmental factors varying slightly but all related to seismic activity. The concurrent susceptibility and absolute intensity maps produced by both models exhibited similar patterns, while relative intensity mapping identified new high-hazard areas within smaller slope units that were previously overlooked by susceptibility and absolute intensity. We established a Bayesian-based single modeling framework for joint hazard assessment and prediction of regional susceptibility and intensity, providing a cutting-edge geospatial paradigm for multi-objective hazard assessment in global environmental disaster management.
Read full abstract