Increasing landslide susceptibility in urbanized areas of Petrópolis identified through spatio-temporal analysis

Abstract

This study conducts a detailed spatio-temporal analysis of landslide susceptibility in Petrópolis, Rio de Janeiro, leveraging data from recorded landslide events between 1985 and 2022. Employing Logistic Regression (LR) and Multiscale Geographically Weighted Regression (MGWR), we generated susceptibility maps for years with the highest landslide occurrences, enabling an evaluation of spatial variability and localized influences of conditioning factors. Significant variables included lithology, land use and land cover (LULC), total precipitation (PRCPTOT), extreme rainfall events (R100), distance from rivers, soil type, slope aspect, elevation, and forest cover. Lithology and LULC showed the strongest positive correlations with landslide occurrences, while forest cover and elevation exhibited protective effects. The model's predictive performance was validated with AUC values ranging from 0.77 to 0.99 and accuracy between 0.77 and 0.97, demonstrating robustness across different temporal scales. Susceptibility maps revealed a temporal increase in areas classified as very high susceptibility, particularly in urbanized zones. These findings address critical gaps in the understanding of landslide dynamics, providing a robust framework for integrating susceptibility maps into disaster risk reduction strategies and urban planning. By identifying key conditioning factors and their spatial variability, this study contributes to more effective mitigation measures and the development of evidence-based policies to enhance resilience in high-risk areas.