Scholarlink

<p>Urban heat in fast-growing South Asian cities is driven by surface materials, urban form, vegetation, and weather conditions. In riverfront settings, these factors interact with riparian conditions and seasonal hydrology, yet most studies represent rivers as a static class with a fixed cooling effect. As a result, the strength and spatial reach of river influence on land surface temperature (LST) remain weakly quantified, especially in composite climates with high seasonal variation. This study examines Varanasi, India, over four seasons between 2024 and 2025, and integrates multi-sensor remote sensing with machine learning prediction models and spatial diagnostics to estimate seasonal LST and quantify the role of the river Ganga, relative to other drivers (R² ≈ 0.79–0.92). Landsat-9 and Sentinel-2 products (albedo and vegetation/built-up indices) are prepared at a common resolution and combined with morphology metrics and distance-to-river measures. Gradient-boosted models have been trained using spatial cross-validation, and performance is summarized across repeated folds with uncertainty intervals. Model interpretation uses Shapley Additive Explanations (SHAP) and Ordinary Least Squares (OLS) to assess variable importance, non-linear dependence, and key interactions and a distance-decay function is fitted to characterize the direction, magnitude, and half-distance of river effects under different seasonal backgrounds. Results show that vegetation is the most stable cooling determinant (up to ~ 3&#xa0;°C reduction in monsoon), compact, high built-up areas can be cooler than sparsely built, low-vegetation grids, and river proximity exerts a strongly seasonal, localized influence. These outcomes provide a consistent, season-aware basis to compare riparian influence with vegetation and built density, and deliberate on practical choices for riparian development and the impact of different factors on the LST. Keywords: Land Surface Temperature (LST), Riparian zone, Prediction model, Machine learning, Shapley additive explanations (SHAP).</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

Factors affecting urban land surface temperature in the Riparian zones of Varanasi: a SHAP-informed machine learning study

  • 期刊论文

  • 发表日期：

    2026年4月10日

查看全文
  • Amit Kumar Biswal,
  • Preeti Onkar

摘要

Urban heat in fast-growing South Asian cities is driven by surface materials, urban form, vegetation, and weather conditions. In riverfront settings, these factors interact with riparian conditions and seasonal hydrology, yet most studies represent rivers as a static class with a fixed cooling effect. As a result, the strength and spatial reach of river influence on land surface temperature (LST) remain weakly quantified, especially in composite climates with high seasonal variation. This study examines Varanasi, India, over four seasons between 2024 and 2025, and integrates multi-sensor remote sensing with machine learning prediction models and spatial diagnostics to estimate seasonal LST and quantify the role of the river Ganga, relative to other drivers (R² ≈ 0.79–0.92). Landsat-9 and Sentinel-2 products (albedo and vegetation/built-up indices) are prepared at a common resolution and combined with morphology metrics and distance-to-river measures. Gradient-boosted models have been trained using spatial cross-validation, and performance is summarized across repeated folds with uncertainty intervals. Model interpretation uses Shapley Additive Explanations (SHAP) and Ordinary Least Squares (OLS) to assess variable importance, non-linear dependence, and key interactions and a distance-decay function is fitted to characterize the direction, magnitude, and half-distance of river effects under different seasonal backgrounds. Results show that vegetation is the most stable cooling determinant (up to ~ 3 °C reduction in monsoon), compact, high built-up areas can be cooler than sparsely built, low-vegetation grids, and river proximity exerts a strongly seasonal, localized influence. These outcomes provide a consistent, season-aware basis to compare riparian influence with vegetation and built density, and deliberate on practical choices for riparian development and the impact of different factors on the LST. Keywords: Land Surface Temperature (LST), Riparian zone, Prediction model, Machine learning, Shapley additive explanations (SHAP).