Scholarlink

<p>Heavy rainfall and occasional powerful earthquakes are common occurrences in rocky and steep mountainous terranes. This apart, the region has heightened risk of landslides and debris flows poses a continual threat to downstream communities, creating a continual sequence of geological hazards. This study aims to bright up a potential framework of Artificial Intelligence (AI) scenario for meticulously assessing landslide risks in the Sikkim Himalayas of North-Eastern India. By leveraging regional geological features, topographical data, hydro-geomorphological factors and anthropogenic factors as hazard indicators, we initially crafted a comprehensive data processing in Geographic Information System (GIS) platform. Simultaneously, within a GIS framework, we developed an Earthquake Hazard Zoning (EHZ) map based on earthquake event magnitude and frequency data. Our analysis of the EHZ map revealed pronounced vulnerability to high-magnitude earthquakes in the northern Sikkim region, characterized by the South-Tibetan Detachment System (STDS). Conversely, the mid to south-western areas bounded by the Main Central Thrust (MCT) exhibited a heightened susceptibility to earthquakes, potentially triggering landslides. Notably, the risk of earthquakes in south-eastern Sikkim appeared relatively lower. Leveraging the EHZ map as a pivotal input, we incorporated it into the landslide susceptibility assessment process, yielding valuable insights for identifying landslide-prone areas. In constructing our AI framework for Landslide Susceptibility Map (LSM) modeling, we meticulously selected machine learning (ML) methods with the highest accuracy. Following rigorous evaluation, the ‘Random Forest’ method emerged as the most suitable, boasting 0.910 area under the Receiver Operating Characteristic curve, and an impressive model accuracy of 84.122%, surpassing alternative methods including Bayesian Network, Decision Tree, Logistic Regression, and Multi-Layer Perceptron. Moreover, we classified future landslide occurrence probabilities into distinct categories, ranging from very low to very high susceptibility indices. This approach provides stakeholders with actionable insights for targeted risk mitigation strategies. In summary, our integrated framework offers a robust methodology for advancing risk management and mitigation efforts in the Sikkim Himalayas, empowering decision-makers with the necessary tools to safeguard communities against geological surface processes and hazards.</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

Towards synergistic AI-driven ensemble framework for earthquake and rainfall induced landslide risks in Sikkim Himalayas

  • 期刊论文

  • 发表日期：

    2025年2月13日

查看全文
  • Litan Dutta,
  • Niptika Jana,
  • Yunus Ali Pulpadan

摘要

Heavy rainfall and occasional powerful earthquakes are common occurrences in rocky and steep mountainous terranes. This apart, the region has heightened risk of landslides and debris flows poses a continual threat to downstream communities, creating a continual sequence of geological hazards. This study aims to bright up a potential framework of Artificial Intelligence (AI) scenario for meticulously assessing landslide risks in the Sikkim Himalayas of North-Eastern India. By leveraging regional geological features, topographical data, hydro-geomorphological factors and anthropogenic factors as hazard indicators, we initially crafted a comprehensive data processing in Geographic Information System (GIS) platform. Simultaneously, within a GIS framework, we developed an Earthquake Hazard Zoning (EHZ) map based on earthquake event magnitude and frequency data. Our analysis of the EHZ map revealed pronounced vulnerability to high-magnitude earthquakes in the northern Sikkim region, characterized by the South-Tibetan Detachment System (STDS). Conversely, the mid to south-western areas bounded by the Main Central Thrust (MCT) exhibited a heightened susceptibility to earthquakes, potentially triggering landslides. Notably, the risk of earthquakes in south-eastern Sikkim appeared relatively lower. Leveraging the EHZ map as a pivotal input, we incorporated it into the landslide susceptibility assessment process, yielding valuable insights for identifying landslide-prone areas. In constructing our AI framework for Landslide Susceptibility Map (LSM) modeling, we meticulously selected machine learning (ML) methods with the highest accuracy. Following rigorous evaluation, the ‘Random Forest’ method emerged as the most suitable, boasting 0.910 area under the Receiver Operating Characteristic curve, and an impressive model accuracy of 84.122%, surpassing alternative methods including Bayesian Network, Decision Tree, Logistic Regression, and Multi-Layer Perceptron. Moreover, we classified future landslide occurrence probabilities into distinct categories, ranging from very low to very high susceptibility indices. This approach provides stakeholders with actionable insights for targeted risk mitigation strategies. In summary, our integrated framework offers a robust methodology for advancing risk management and mitigation efforts in the Sikkim Himalayas, empowering decision-makers with the necessary tools to safeguard communities against geological surface processes and hazards.