Scholarlink

<p>Integrating&#xa0;dynamic factors such as rainfall and land use/cover (LULC) changes into landslide predictions is often overlooked. A combination of aforementioned dynamic factors, mountainous terrain and fragile geology increase risk of landslides in the Himalayan region. This study assesses the impact of both dynamic and static factors on landslide prediction. The XGBoost machine learning (ML) algorithm is employed for generating landslide susceptibility maps due to its superior performance and accuracy in the study area. Base map is prepared for the period from 1995 to 2020, taking into account significant changes in urbanization and climatic impacts observed in the study area. Results suggest that the ML algorithm performs well based on metrics such as accuracy (96.6%), precision (98.4%), recall (94.8%), Matthew’s correlation coefficient (93.2%), Cohen’s kappa coefficient (92%), F1 score (96.6%), and area under receiver-operating-characteristic (ROC) curve (99.3%). For future landslide susceptibility predictions, maps under different climate change scenarios are prepared using rainfall alone and both rainfall and LULC as dynamic factors. Results indicate an increase in high and very high susceptibility classes; the most significant increase (approximately 60% of the baseline) is observed in scenarios considering both the dynamic factors. It infers that including dynamic parameters in landslide prediction enhances the accuracy of landslide susceptibility analysis and improves reliability of disaster management 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

Integrating dynamic factors for predicting future landslide susceptibility

  • 期刊论文

  • 发表日期：

    2025年1月28日

查看全文
  • Suraj Lamichhane,
  • Arhat Ratna Kansakar,
  • Nirajan Devkota,
  • Bhim Kumar Dahal

摘要

Integrating dynamic factors such as rainfall and land use/cover (LULC) changes into landslide predictions is often overlooked. A combination of aforementioned dynamic factors, mountainous terrain and fragile geology increase risk of landslides in the Himalayan region. This study assesses the impact of both dynamic and static factors on landslide prediction. The XGBoost machine learning (ML) algorithm is employed for generating landslide susceptibility maps due to its superior performance and accuracy in the study area. Base map is prepared for the period from 1995 to 2020, taking into account significant changes in urbanization and climatic impacts observed in the study area. Results suggest that the ML algorithm performs well based on metrics such as accuracy (96.6%), precision (98.4%), recall (94.8%), Matthew’s correlation coefficient (93.2%), Cohen’s kappa coefficient (92%), F1 score (96.6%), and area under receiver-operating-characteristic (ROC) curve (99.3%). For future landslide susceptibility predictions, maps under different climate change scenarios are prepared using rainfall alone and both rainfall and LULC as dynamic factors. Results indicate an increase in high and very high susceptibility classes; the most significant increase (approximately 60% of the baseline) is observed in scenarios considering both the dynamic factors. It infers that including dynamic parameters in landslide prediction enhances the accuracy of landslide susceptibility analysis and improves reliability of disaster management strategies.