Scholarlink

<p>Landslides frequently occur along mountainous highway corridors, posing substantial risks to infrastructure and reservoir operations. To enhance the structural reliability and spatial interpretability of landslide susceptibility mapping (LSM), this study proposes a point–line–area integrated framework that combines ensemble learning, slope-unit-based InSAR time-series analysis, and spatial clustering diagnostics within a unified GIS workflow. A stacking ensemble model integrating XGBoost, Random Forest, and Multilayer Perceptron was developed using a 70/30 train-test split and five-fold cross-validation. Hyperparameters were optimized via Bayesian search, achieving a strong predictive performance with an AUC of 0.9603 and an accuracy of 0.8925. A Consistency–Significance Index (CSI) was used to jointly evaluate cluster dominance and statistical significance and to enhance spatial coherence. Furthermore, Sentinel-1 InSAR data acquired between May 2021 and April 2024 were processed to classify deformation trends into accelerated, seasonal, and steady types, with − 5&#xa0;mm/year defined as the threshold for accelerated instability. A rule-based calibration strategy integrated these probabilistic, spatial, and dynamic indicators into a refined five-level susceptibility scheme. Quantitative validation via frequency ratio (FR) analysis and bootstrap testing (B = 5000) demonstrates that the refined map restores strict monotonic ordering, increasing the combined FR of the high- and very high-susceptibility zones from 1.062 to 1.717. This integrated approach provides a robust and interpretable framework for corridor-scale geohazard assessment and infrastructure risk management.</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 InSAR time-series and ensemble learning for corridor-scale landslide susceptibility assessment in the Three Gorges Reservoir Area, China

  • 期刊论文

  • 发表日期：

    2026年4月23日

查看全文
  • Ruijie Sun,
  • Wanli Liu,
  • Jiayu Ling,
  • Jiyun Wang,
  • Echuan Yan

摘要

Landslides frequently occur along mountainous highway corridors, posing substantial risks to infrastructure and reservoir operations. To enhance the structural reliability and spatial interpretability of landslide susceptibility mapping (LSM), this study proposes a point–line–area integrated framework that combines ensemble learning, slope-unit-based InSAR time-series analysis, and spatial clustering diagnostics within a unified GIS workflow. A stacking ensemble model integrating XGBoost, Random Forest, and Multilayer Perceptron was developed using a 70/30 train-test split and five-fold cross-validation. Hyperparameters were optimized via Bayesian search, achieving a strong predictive performance with an AUC of 0.9603 and an accuracy of 0.8925. A Consistency–Significance Index (CSI) was used to jointly evaluate cluster dominance and statistical significance and to enhance spatial coherence. Furthermore, Sentinel-1 InSAR data acquired between May 2021 and April 2024 were processed to classify deformation trends into accelerated, seasonal, and steady types, with − 5 mm/year defined as the threshold for accelerated instability. A rule-based calibration strategy integrated these probabilistic, spatial, and dynamic indicators into a refined five-level susceptibility scheme. Quantitative validation via frequency ratio (FR) analysis and bootstrap testing (B = 5000) demonstrates that the refined map restores strict monotonic ordering, increasing the combined FR of the high- and very high-susceptibility zones from 1.062 to 1.717. This integrated approach provides a robust and interpretable framework for corridor-scale geohazard assessment and infrastructure risk management.