Scholarlink

<p>The Western Ghats (Sahyadri) of India, a UNESCO World Heritage Site and globally recognised biodiversity hotspot, experience recurrent flooding driven by intense monsoonal rainfall interacting with complex hydrogeomorphic controls that regulate runoff generation and flow concentration. However, the role of geomorphology in regulating flood susceptibility (FS) remains underrepresented in conventional flood risk assessments, which primarily emphasise hydrometeorological forcing. Although hydrodynamic models routinely incorporate digital elevation models (DEMs), they typically ignore the geomorphic signatures embedded within them, such as flow convergence, valley confinement, and floodplain geometry, which are fundamental to how floods initiate, route, and persist. Here, we develop a geomorphology-driven framework to quantify FS using high-resolution DEM-derived geomorphic flood descriptors (GFDs) and machine learning (ML). Six ML classifiers were evaluated, including Random Forest, Gradient Boosting, XGBoost, LightGBM, CatBoost, and an XLC-Boost ensemble. The ensemble achieved the highest predictive performance (AUC = 0.945) and identified steep valleys and coastal lowlands as dominant flood-susceptible zones. Districts such as Palghar, Kozhikode, Alappuzha, and Virudhunagar exhibit high susceptibility due to strong flow convergence and drainage concentration. Multicollinearity screening (VIF &gt; 5) ensured a robust set of predictors, while explainable AI revealed that downslope indices, the geomorphic flood index, convexity, and horizontal overland flow distance are the primary controls on FS. These findings demonstrate that terrain-driven flow organisation, rather than rainfall magnitude alone, governs where floodwaters accumulate. The proposed framework provides a physically grounded, scalable approach for mapping intrinsic FS using only DEM data. By explicitly quantifying terrain predisposition, it complements hydrodynamic models and helps unravel a critical dimension of physical vulnerability for improving flood risk assessment in data-scarce mountainous regions.</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

Hydrogeomorphic Controls on Flood Susceptibility Revealed Through Machine Learning and Explainable AI in a Multi-hazard-prone Fragile Biodiversity Hotspot of India

  • 期刊论文

  • 发表日期：

    2026年5月20日

查看全文
  • Subhankar Karmakar,
  • Sabirul Sk

摘要

The Western Ghats (Sahyadri) of India, a UNESCO World Heritage Site and globally recognised biodiversity hotspot, experience recurrent flooding driven by intense monsoonal rainfall interacting with complex hydrogeomorphic controls that regulate runoff generation and flow concentration. However, the role of geomorphology in regulating flood susceptibility (FS) remains underrepresented in conventional flood risk assessments, which primarily emphasise hydrometeorological forcing. Although hydrodynamic models routinely incorporate digital elevation models (DEMs), they typically ignore the geomorphic signatures embedded within them, such as flow convergence, valley confinement, and floodplain geometry, which are fundamental to how floods initiate, route, and persist. Here, we develop a geomorphology-driven framework to quantify FS using high-resolution DEM-derived geomorphic flood descriptors (GFDs) and machine learning (ML). Six ML classifiers were evaluated, including Random Forest, Gradient Boosting, XGBoost, LightGBM, CatBoost, and an XLC-Boost ensemble. The ensemble achieved the highest predictive performance (AUC = 0.945) and identified steep valleys and coastal lowlands as dominant flood-susceptible zones. Districts such as Palghar, Kozhikode, Alappuzha, and Virudhunagar exhibit high susceptibility due to strong flow convergence and drainage concentration. Multicollinearity screening (VIF > 5) ensured a robust set of predictors, while explainable AI revealed that downslope indices, the geomorphic flood index, convexity, and horizontal overland flow distance are the primary controls on FS. These findings demonstrate that terrain-driven flow organisation, rather than rainfall magnitude alone, governs where floodwaters accumulate. The proposed framework provides a physically grounded, scalable approach for mapping intrinsic FS using only DEM data. By explicitly quantifying terrain predisposition, it complements hydrodynamic models and helps unravel a critical dimension of physical vulnerability for improving flood risk assessment in data-scarce mountainous regions.