Spatiotemporal Landslide Susceptibility Mapping using Hybrid Machine Learning (BayesOpt-XGBoost) and GIS

This study is based on the hypothesis that landslide susceptibility on Penang Island changes over time in response to variations in topography, vegetation cover, and land use, and that a BayesOpt-XGBoost model can effectively capture these spatiotemporal dynamics with high predictive accuracy. Landslide and non-landslide grid units were extracted using an inventory of 443 landslide and 443 non-landslide points. Four primary conditioning factors, namely Digital Elevation Model (DEM), slope, land cover (LC), and Normalized Difference Vegetation Index (NDVI), were derived for the years 2014, 2017, and 2025. DEM and slope were generated from elevation data, NDVI was calculated from satellite imagery, and LC was obtained from classified land use maps. T

本研究基于如下假设：槟城岛的滑坡敏感性随时间动态变化，以响应地形、植被覆盖与土地利用的差异；且贝叶斯优化-极限梯度提升树（BayesOpt-XGBoost）模型可有效捕捉此类时空动态特征，并具备较高的预测精度。研究利用包含443个滑坡点与443个非滑坡点的编目数据集，提取得到滑坡与非滑坡网格单元。本次研究针对2014、2017及2025三个年份，提取了四类核心滑坡敏感性条件因子，即数字高程模型（Digital Elevation Model，DEM）、坡度、土地覆盖（Land Cover，LC）以及归一化差分植被指数（Normalized Difference Vegetation Index，NDVI）。其中，数字高程模型与坡度由高程数据生成，归一化差分植被指数由卫星影像计算得到，土地覆盖数据则源自分类后的土地利用地图。T