Scholarlink

<p>Flooding in arid urban regions is an increasingly pressing concern due to the compounded effects of climate change, rapid urbanisation, and hydrologically underprepared infrastructure systems. These environments, often characterised by impervious surfaces and poor drainage, face heightened exposure to short-duration, high-intensity rainfall events. Makkah, a rapidly growing arid city in western Saudi Arabia with significant topographical variability, typifies such risk and necessitates a robust, infrastructure-specific flood risk modelling framework. This study aims to develop a scientifically rigorous and spatially detailed flood susceptibility and damage assessment framework tailored for critical infrastructure specifically, healthcare facilities and road networks. The methodological approach integrates advanced remote sensing, machine learning, and probabilistic simulation. A comprehensive flood inventory was prepared using historical satellite imagery, ground-truthing, and official flood records. Flood conditioning parameters including topographic (e.g. slope, curvature, elevation), hydrological (e.g. drainage density, rainfall), and anthropogenic (e.g. proximity to roads/rivers) were derived and validated. To address multicollinearity and ensure data integrity, correlation and variance inflation factor (VIF) analyses were conducted. Four machine learning models such as Random Forest, Support Vector Machine, Gradient Boosting Machine, and Categorical Boosting (CatBoost) were trained using optimised hyperparameters and validated through stratified k-fold cross-validation. Among these, CatBoost yielded the highest accuracy and reliability (AUC = 0.90), mapping approximately 282.02&#xa0;km<sup>2</sup> under ‘very high’ and 156.55&#xa0;km<sup>2</sup> under ‘high’ flood susceptibility zones. Sensitivity analysis further revealed Support Vector Machine to be the most robust against input perturbations. Infrastructure-specific exposure analysis, coupled with Monte Carlo-based probabilistic economic modelling, estimated potential damages at SAR 28.1&#xa0;billion for hospitals, SAR 15.9&#xa0;billion for buildings, and SAR 3.36&#xa0;billion for roads. Critical vulnerability clusters were identified in Aziziyah, Al-Haram, and Al Misfalah districts. This integrated framework offers a replicable model for infrastructure resilience planning in arid urban environments.</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

Machine learning-based flood risk prediction and asset damage estimation for critical infrastructure in Arid Makkah

  • 期刊论文

  • 发表日期：

    2026年1月9日

查看全文
  • Saeed Alqadhi,
  • Javed Mallick,
  • Abdullah Othman

摘要

Flooding in arid urban regions is an increasingly pressing concern due to the compounded effects of climate change, rapid urbanisation, and hydrologically underprepared infrastructure systems. These environments, often characterised by impervious surfaces and poor drainage, face heightened exposure to short-duration, high-intensity rainfall events. Makkah, a rapidly growing arid city in western Saudi Arabia with significant topographical variability, typifies such risk and necessitates a robust, infrastructure-specific flood risk modelling framework. This study aims to develop a scientifically rigorous and spatially detailed flood susceptibility and damage assessment framework tailored for critical infrastructure specifically, healthcare facilities and road networks. The methodological approach integrates advanced remote sensing, machine learning, and probabilistic simulation. A comprehensive flood inventory was prepared using historical satellite imagery, ground-truthing, and official flood records. Flood conditioning parameters including topographic (e.g. slope, curvature, elevation), hydrological (e.g. drainage density, rainfall), and anthropogenic (e.g. proximity to roads/rivers) were derived and validated. To address multicollinearity and ensure data integrity, correlation and variance inflation factor (VIF) analyses were conducted. Four machine learning models such as Random Forest, Support Vector Machine, Gradient Boosting Machine, and Categorical Boosting (CatBoost) were trained using optimised hyperparameters and validated through stratified k-fold cross-validation. Among these, CatBoost yielded the highest accuracy and reliability (AUC = 0.90), mapping approximately 282.02 km2 under ‘very high’ and 156.55 km2 under ‘high’ flood susceptibility zones. Sensitivity analysis further revealed Support Vector Machine to be the most robust against input perturbations. Infrastructure-specific exposure analysis, coupled with Monte Carlo-based probabilistic economic modelling, estimated potential damages at SAR 28.1 billion for hospitals, SAR 15.9 billion for buildings, and SAR 3.36 billion for roads. Critical vulnerability clusters were identified in Aziziyah, Al-Haram, and Al Misfalah districts. This integrated framework offers a replicable model for infrastructure resilience planning in arid urban environments.