Scholarlink

<p>Landslides are a major geohazard in mountainous regions such as the Western Ghats of India. This study evaluates and compares two widely used Landslide Susceptibility Mapping (LSM) approaches, Frequency Ratio Analysis (FRA) and Random Forest Classifier (RFC), for the Kodagu district. Eleven conditioning factors were derived from remote sensing and GIS datasets, and models were developed using an 80:20 training-validation split. Both models demonstrated strong predictive performance, with AUC values of 0.876 for FRA and 0.921 for RFC. Beyond accuracy assessment, the study integrates sensitivity analysis (Leave-One-Out and weight variation) and Monte Carlo simulation to quantify uncertainty and evaluate model performance. Results indicate that RFC provides improved spatial consistency and lower uncertainty, while FRA offers greater interpretability but higher sensitivity to input perturbations. Key influencing factors identified include road proximity, elevation, and drainage density. The findings highlight that while RFC offers improved predictive stability, FRA remains a reliable and transparent alternative. The study provides a comprehensive framework for evaluating LSM models in complex terrains.</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

Sensitivity and Uncertainty Analysis of Statistical and Machine Learning based Landslide Susceptibility Mapping in a Part of Western Ghats, India

  • 期刊论文

  • 发表日期：

    2026年6月23日

查看全文
  • Priyajit Kundu,
  • Sreevalsa Kolathayar,
  • Pruthviraj Umesh

摘要

Landslides are a major geohazard in mountainous regions such as the Western Ghats of India. This study evaluates and compares two widely used Landslide Susceptibility Mapping (LSM) approaches, Frequency Ratio Analysis (FRA) and Random Forest Classifier (RFC), for the Kodagu district. Eleven conditioning factors were derived from remote sensing and GIS datasets, and models were developed using an 80:20 training-validation split. Both models demonstrated strong predictive performance, with AUC values of 0.876 for FRA and 0.921 for RFC. Beyond accuracy assessment, the study integrates sensitivity analysis (Leave-One-Out and weight variation) and Monte Carlo simulation to quantify uncertainty and evaluate model performance. Results indicate that RFC provides improved spatial consistency and lower uncertainty, while FRA offers greater interpretability but higher sensitivity to input perturbations. Key influencing factors identified include road proximity, elevation, and drainage density. The findings highlight that while RFC offers improved predictive stability, FRA remains a reliable and transparent alternative. The study provides a comprehensive framework for evaluating LSM models in complex terrains.