Evaluating the Impact of Geoid Model Variations on Landslide Susceptibility: A Case Study in Taiwan's Mountainous Regions

Taiwan’s terrain is predominantly mountainous, with approximately 73% of its total land area classified as sloped terrain. The region is characterized by complex geological conditions, short yet steep and fast-flowing rivers, and a high susceptibility to natural disasters. During the summer and autumn seasons, typhoons and heavy rainfall frequently trigger landslides in mountainous areas, posing significant risks to infrastructure and communities. While numerous studies have investigated landslide susceptibility in this region, few have examined the impact of different geoid models on these analyses. In Taiwan, geoid models are periodically updated, and these changes can influence key analytical factors in landslide susceptibility assessments, potentially affecting the outcomes. This study utilizes high-resolution digital elevation models (DEMs) based on various global geoid models, such as EGM96 and EGM2008, as well as regional geoid models like TWGEOID2014, TWGEOID2023, and TWGEOID2024, to assess their influence on landslide susceptibility. This study focuses on the mountainous areas of central Taiwan, which also exhibit the largest differences in geoid models. Logistic regression analysis is performed using IBM SPSS statistics software, incorporating terrain factors such as aspect, slope, curvature, relief, and roughness to evaluate landslide susceptibility. Landslide susceptibility maps and receiver operating characteristic (ROC) curves are generated for each geoid model and compared to assess their differences. The findings of this research aim to improve the precision of disaster prediction and provide valuable insights for disaster prevention efforts, soil and water conservation, and integrated risk management strategies. Additionally, this study highlights the importance of geoid model selection in geospatial analyses and its broader implications for environmental and engineering applications.