Scholarlink

<p>Gas hydrates have been accepted to be a potential triggering factor for submarine landslides; however, their quantitative impact on landslide susceptibility has not been thoroughly assessed. By integrating multi-sourced geological parameters (e.g., earthquakes, depth, slope, and faults) with Potential Hydrate Occurrence Depth (PHOD), this study proposes a cascaded machine learning framework for both Bottom Simulating Reflector (BSR) occurrence and landslide susceptibility assessment, using the North Atlantic as a case study. We compare the performance of 11 machine learning classifiers in assessing landslide susceptibility of the North Atlantic. The results show that ensemble learning methods, particularly eXtreme Gradient Boosting (XGBoost) and Light Gradient Boosting Machine (LightGBM), achieve the best prediction accuracy. Using explainable artificial intelligence (XAI) techniques, this study quantifies the nonlinear relationships between submarine landslides and key controlling factors, such as hydrate occurrence depth, earthquakes, water depth, faults, and slope. Notably, the impact of PHOD on landslide susceptibility exhibits a reversal with increasing depth. Finally, we generate a 2-km resolution landslide susceptibility map. This study provides both theoretical insights and practical tools for early warning of landslide susceptibility in high-latitude hydrate provinces, while also clarifying the mechanisms of slope instability under multi-factor interactions.</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

The interaction of gas hydrate with submarine landslides studied with cascading machine learning

  • 期刊论文

  • 发表日期：

    2026年1月5日

查看全文
  • Wei Wang,
  • Nengyou Wu,
  • Lizhong Wang,
  • Lilin Wang,
  • Zhi Liu,
  • Yanlong Li

摘要

Gas hydrates have been accepted to be a potential triggering factor for submarine landslides; however, their quantitative impact on landslide susceptibility has not been thoroughly assessed. By integrating multi-sourced geological parameters (e.g., earthquakes, depth, slope, and faults) with Potential Hydrate Occurrence Depth (PHOD), this study proposes a cascaded machine learning framework for both Bottom Simulating Reflector (BSR) occurrence and landslide susceptibility assessment, using the North Atlantic as a case study. We compare the performance of 11 machine learning classifiers in assessing landslide susceptibility of the North Atlantic. The results show that ensemble learning methods, particularly eXtreme Gradient Boosting (XGBoost) and Light Gradient Boosting Machine (LightGBM), achieve the best prediction accuracy. Using explainable artificial intelligence (XAI) techniques, this study quantifies the nonlinear relationships between submarine landslides and key controlling factors, such as hydrate occurrence depth, earthquakes, water depth, faults, and slope. Notably, the impact of PHOD on landslide susceptibility exhibits a reversal with increasing depth. Finally, we generate a 2-km resolution landslide susceptibility map. This study provides both theoretical insights and practical tools for early warning of landslide susceptibility in high-latitude hydrate provinces, while also clarifying the mechanisms of slope instability under multi-factor interactions.