Scholarlink

<p>Urban expansion in fast-growing cities often outpaces the ability of planners to anticipate its environmental and infrastructural consequences. Here, we integrated high-resolution Landsat imagery (2011–2021) with an ensemble of gradient-boosting algorithms, namely XGBoost, CatBoost, and LightGBM, to forecast the spatial likelihood of urban growth in Lucknow, India, by 2031. After rigorous preprocessing (variance-inflation screening, Synthetic Minority Oversampling Technique balancing, and randomized hyperparameter optimization), XGBoost achieved the best discrimination (accuracy = 92.4 %, AUC = 0.97). Explainable AI based on SHapley Additive exPlanations revealed that drive-time to the central business district, proximity to roads, elevation, and the projected population together accounted for &gt;70 % of the model influence, linking transport accessibility and topography to future land conversion. The resulting probability surface identified 205 km<sup>2</sup> (approximately 22 % of the study area) as high-risk for development – chiefly along planned expressways and peri-urban corridors – providing an actionable map for zoning, storm-run-off mitigation, and green-belt preservation. By coupling state-of-the-art machine learning with transparent attribution, our framework provides a transferable blueprint for evidence-based climate-responsive urban planning in the Global South.</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

Ensemble machine learning for predicting the urban expansion in Lucknow, India

  • 期刊论文

  • 发表日期：

    2025年10月11日

查看全文
  • Danish Khan,
  • Nizamuddin Khan,
  • Mohamed Yehia Abouleish,
  • Shruti Kanga,
  • Suraj Kumar Singh,
  • Gowhar Meraj

摘要

Urban expansion in fast-growing cities often outpaces the ability of planners to anticipate its environmental and infrastructural consequences. Here, we integrated high-resolution Landsat imagery (2011–2021) with an ensemble of gradient-boosting algorithms, namely XGBoost, CatBoost, and LightGBM, to forecast the spatial likelihood of urban growth in Lucknow, India, by 2031. After rigorous preprocessing (variance-inflation screening, Synthetic Minority Oversampling Technique balancing, and randomized hyperparameter optimization), XGBoost achieved the best discrimination (accuracy = 92.4 %, AUC = 0.97). Explainable AI based on SHapley Additive exPlanations revealed that drive-time to the central business district, proximity to roads, elevation, and the projected population together accounted for >70 % of the model influence, linking transport accessibility and topography to future land conversion. The resulting probability surface identified 205 km2 (approximately 22 % of the study area) as high-risk for development – chiefly along planned expressways and peri-urban corridors – providing an actionable map for zoning, storm-run-off mitigation, and green-belt preservation. By coupling state-of-the-art machine learning with transparent attribution, our framework provides a transferable blueprint for evidence-based climate-responsive urban planning in the Global South.