Optimizing seismic hazard inputs for co-seismic landslide susceptibility mapping: a probabilistic analysis

Abstract

The significance of seismic hazard maps as inputs in co-seismic landslide susceptibility mapping is well-established. However, a research gap exists as no previous study has compared the effectiveness of various seismic hazard map inputs. The present research conducts a comprehensive comparative study, evaluating probabilistic seismic hazard assessment (PSHA)-based and specific scenario-based PGA maps as inputs for co-seismic landslide susceptibility mapping. In the study, the first step involved generating PSHA-based and scenario-based PGA maps, which served as seismic intensity inputs for the modified Newmark’s model. The modified model incorporates the rock joint shear strength parameters in displacement computations. To address uncertainties associated with the spatial variability of shear strength parameters of rock joints, Latin hypercube sampling along with Monte Carlo simulations were employed, resulting in a set of displacement values. The Latin hypercube sampling method ensured a more efficient and stratified sampling approach, enhancing the representation of uncertainty in the model. The simulations were conducted 10,000 times, generating 10,000 displacement values for each pixel. Subsequently, statistical calculations were performed to determine both the means and standard deviations of these displacement values, resulting in the creation of probability distributions. The predicted displacement probabilities surpassing 5 cm as threshold value were then displayed as landslide susceptibility maps. After generating the susceptibility maps, a comprehensive comparison was conducted based on various evaluation metrics, including confusion matrix, Kappa Coefficient, F1-score, and AUC-ROC values. The analysis revealed that the PSHA-based PGA input performed better than the scenario-based PGA input for co-seismic landslide susceptibility mapping. © The Author(s), under exclusive licence to Springer Nature B.V. 2024.