Scholarlink

<p>The southeastern Tibetan Plateau is characterized by complex geological structures and pronounced topographic relief, making it one of the most landslide-prone regions worldwide and posing significant threats to regional safety and socio-economic development. To address the highly complex landslide environment, this study employs machine learning methods to conduct a high-precision landslide susceptibility assessment. A total of 14 key environmental factors, including elevation, slope, terrain position index (TPI), distance to roads, and annual rainfall, were selected. Based on 20,514 historical landslide samples, three models including Random Forest (RF), Extreme Trees (ET), and Support Vector Machine (SVM) were developed and comparatively evaluated. The results demonstrate that the RF model outperforms the other models, achieving an AUC of 0.897 and an accuracy of 0.819. Feature importance analysis indicates that TPI is the dominant controlling factor for landslide occurrence in the study area. Using the optimal RF model, a landslide susceptibility map with a spatial resolution of 30&#xa0;m was produced, classifying the area into five susceptibility levels. High and very high susceptibility zones account for 12.8% and 2.2% of the total area, respectively, and collectively contain over 59% of historical landslides. These findings confirm the model’s reliable predictive capability and reasonable spatial consistency, providing valuable scientific support for landslide risk identification and disaster prevention in the southeastern Tibetan Plateau.</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

Evaluation of landslide susceptibility in alpine canyon area based on random forest method: a case study of southeast Tibetan Plateau

  • 期刊论文

  • 发表日期：

    2026年6月16日

查看全文
  • Haoran Dong,
  • Qing Yang,
  • Chong Xu,
  • Huiran Gao,
  • Shuzhong Sheng,
  • Yulong Cui,
  • Saier Wu,
  • Lei Li,
  • Peng Wang,
  • Jingjing Sun

摘要

The southeastern Tibetan Plateau is characterized by complex geological structures and pronounced topographic relief, making it one of the most landslide-prone regions worldwide and posing significant threats to regional safety and socio-economic development. To address the highly complex landslide environment, this study employs machine learning methods to conduct a high-precision landslide susceptibility assessment. A total of 14 key environmental factors, including elevation, slope, terrain position index (TPI), distance to roads, and annual rainfall, were selected. Based on 20,514 historical landslide samples, three models including Random Forest (RF), Extreme Trees (ET), and Support Vector Machine (SVM) were developed and comparatively evaluated. The results demonstrate that the RF model outperforms the other models, achieving an AUC of 0.897 and an accuracy of 0.819. Feature importance analysis indicates that TPI is the dominant controlling factor for landslide occurrence in the study area. Using the optimal RF model, a landslide susceptibility map with a spatial resolution of 30 m was produced, classifying the area into five susceptibility levels. High and very high susceptibility zones account for 12.8% and 2.2% of the total area, respectively, and collectively contain over 59% of historical landslides. These findings confirm the model’s reliable predictive capability and reasonable spatial consistency, providing valuable scientific support for landslide risk identification and disaster prevention in the southeastern Tibetan Plateau.