Exploring the dynamic impact of urbanization on landslide susceptibility in Sichuan Province using an explainable XGBoost model

Abstract

Landslides pose significant threats to human life and are influenced by anthropogenic modifications associated with urbanization. Assessing the impact of urbanization on landslides remains a challenging task that has not yet been fully quantified. To address this issue, this study develops an interpretable machine learning model to quantify the variation in and driving mechanisms behind landslide susceptibility under urbanization from 2000 to 2020. In Sichuan Province, the study area experienced a 137 % increase in urban impervious surface area, covering an area of 2.4 × 10⁴ km². In this context, 18.56 % of the study area experienced an increase in landslide susceptibility, with an average increment of 14 %. The Shapley method was employed to identify the most influential factors on landslide susceptibility, including elevation, topographic relief, distance to roads, annual precipitation, and NDVI. In urban areas, road construction activities and rainfall were identified as the primary contributors to increased landslide susceptibility. In urbanizing areas, human activities, precipitation, and vegetation degradation emerged as key factors influencing changes in landslide susceptibility. The results confirm that urbanization increases landslide susceptibility and highlight the importance of using interpretable machine learning techniques to understand this phenomenon. These findings, along with the proposed analytical framework, offer new perspectives and insights for the in-depth study, prediction, and management of landslide risks.