Scholarlink

This chapter aimed to construct susceptibility maps for landslides along parts of National Highway 110 (NH-110) connecting hilly areas of Darjeeling district to Siliguri city (plains within Darjeeling district), West Bengal, India. Through the Bhukosh open source portal (Geological Survey of India), global landslide catalog (NASA), and Google Earth satellite imagery, 220 landslide points were found for the region of interest. The identified landslide causative factors were slope gradient, elevation, convergence index, distance to roads and rivers, drainage density, lineament density, land use/land cover, geomorphology, annual average rainfall, normalized differential vegetation index, normalized differential build-up index, and the same were used to construct the geodatabase. Multicollinearity diagnostics indicated no issue of multicollinearity within the identified factors, and thus all the factors were considered for the chapter. Geospatial multi-criteria decision making, viz., fuzzy analytic hierarchy process (FAHP) and four contemporary machine learning techniques, viz., boosted tree (BTree), boosted logistic regression (BLR), k-nearest neighbor (KNN), gradient boosting machine (GBM) were utilized for constructing four hybrid ensembles to delineate land-slide susceptibility maps (LSMs) for the intended study region. Further, the area under the receiver operating characteristics (AUC-ROC) curve was employed to establish the scientific significance of the LSMs. The yields from the ROC curves were 0.8310 (FAHP-BTree), 0.9233 (FAHP-BLR), 0.9281 (FAHP-KNN), and 0.9667 (FAHP-GBM) which demonstrated excellent accuracy in assessing landslide susceptibility. FAHP-GBM was found to be the best-suited model for the study. This chapter aimed to combine GIS and machine learning-based techniques for susceptibility zonation along parts of NH-110 transport arteries.

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

Enhanced Landslide Susceptibility Assessment Along Transport Infrastructures Through Hybrid Ensemble Learning Techniques

  • 章节

  • 发表日期：

    2025年

查看全文
  • Sumon Dey,
  • Deepanjan Sen,
  • Swarup Das

摘要

This chapter aimed to construct susceptibility maps for landslides along parts of National Highway 110 (NH-110) connecting hilly areas of Darjeeling district to Siliguri city (plains within Darjeeling district), West Bengal, India. Through the Bhukosh open source portal (Geological Survey of India), global landslide catalog (NASA), and Google Earth satellite imagery, 220 landslide points were found for the region of interest. The identified landslide causative factors were slope gradient, elevation, convergence index, distance to roads and rivers, drainage density, lineament density, land use/land cover, geomorphology, annual average rainfall, normalized differential vegetation index, normalized differential build-up index, and the same were used to construct the geodatabase. Multicollinearity diagnostics indicated no issue of multicollinearity within the identified factors, and thus all the factors were considered for the chapter. Geospatial multi-criteria decision making, viz., fuzzy analytic hierarchy process (FAHP) and four contemporary machine learning techniques, viz., boosted tree (BTree), boosted logistic regression (BLR), k-nearest neighbor (KNN), gradient boosting machine (GBM) were utilized for constructing four hybrid ensembles to delineate land-slide susceptibility maps (LSMs) for the intended study region. Further, the area under the receiver operating characteristics (AUC-ROC) curve was employed to establish the scientific significance of the LSMs. The yields from the ROC curves were 0.8310 (FAHP-BTree), 0.9233 (FAHP-BLR), 0.9281 (FAHP-KNN), and 0.9667 (FAHP-GBM) which demonstrated excellent accuracy in assessing landslide susceptibility. FAHP-GBM was found to be the best-suited model for the study. This chapter aimed to combine GIS and machine learning-based techniques for susceptibility zonation along parts of NH-110 transport arteries.