Scholarlink

<p>Landslides are among the most hazardous geological events in mountainous regions, causing sudden and widespread destruction. They threaten human lives and destroy houses and infrastructure. Landslide susceptibility maps can provide crucial spatial information about areas at risk in the future. Meanwhile, proactively identifying the number of houses potentially exposed to landslide hazards helps to plan prevention and minimize unnecessary damage. Thus, this study used ML models and geospatial analysis to present a quantifying method for assessing the number of houses potentially exposed to landslide hazards. Three advanced hybrid ML models, including Decorate UBoost (DCR-UBoost), Best First Decision Tree UBoost (BFT-UBoost), and Cost Sensitive Forest UBoost (CSF-UBoost) with UltraBoost (UBoost) as a base classifier were developed in the WEKA software to create landslide susceptibility maps using the landslide inventory data and 15 landslide causative factors. Receiver Operating Characteristic (ROC) curve analysis was used to compare hybrid ML models, identifying DCR-UBoost as the most accurate for landslide susceptibility mapping. The final map was overlaid with housing data to quantify exposure, revealing the highest-risk districts: Bac Yen (91.10%), Sop Cop (80.17%), and Muong La (75.21%). The proposed approach helps capture the spatial distribution and scale of landslide impacts on houses, adding referenced information for local authorities in disaster management activities to protect the community and resettlement strategies.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Assessing housing exposure to landslide hazards using hybrid machine learning and spatial modeling

  • 期刊论文

  • 发表日期：

    2025年5月17日

查看全文
  • Hang Ha,
  • Quynh Duy Bui,
  • Viet-Phuong Nguyen,
  • Xuan Thong Tran,
  • Dinh Quoc Nguyen,
  • Chinh Luu

摘要

Landslides are among the most hazardous geological events in mountainous regions, causing sudden and widespread destruction. They threaten human lives and destroy houses and infrastructure. Landslide susceptibility maps can provide crucial spatial information about areas at risk in the future. Meanwhile, proactively identifying the number of houses potentially exposed to landslide hazards helps to plan prevention and minimize unnecessary damage. Thus, this study used ML models and geospatial analysis to present a quantifying method for assessing the number of houses potentially exposed to landslide hazards. Three advanced hybrid ML models, including Decorate UBoost (DCR-UBoost), Best First Decision Tree UBoost (BFT-UBoost), and Cost Sensitive Forest UBoost (CSF-UBoost) with UltraBoost (UBoost) as a base classifier were developed in the WEKA software to create landslide susceptibility maps using the landslide inventory data and 15 landslide causative factors. Receiver Operating Characteristic (ROC) curve analysis was used to compare hybrid ML models, identifying DCR-UBoost as the most accurate for landslide susceptibility mapping. The final map was overlaid with housing data to quantify exposure, revealing the highest-risk districts: Bac Yen (91.10%), Sop Cop (80.17%), and Muong La (75.21%). The proposed approach helps capture the spatial distribution and scale of landslide impacts on houses, adding referenced information for local authorities in disaster management activities to protect the community and resettlement strategies.