Enhancing flood susceptibility mapping in Sana’a, Yemen with random forest and eXtreme gradient boosting algorithms

Floods pose a significant risk to urban areas worldwide, causing extensive damage to infrastructure, property, and human lives. The goal of this work is to improve Sana’a City, Yemen’s flood susceptibility mapping by utilizing two cutting-edge machine learning RF and XGBoost. The RF and XGBoost algorithms were selected for their robust performance in handling complex datasets and producing accurate predictions. The study’s methodology involved optimizing the algorithms through grid search and cross-validation techniques, followed by validation using historical flood data. Both models demonstrated high accuracy in predicting flood-prone areas, with RF achieving an accuracy of 92% and XGBoost slightly outperforming it with an accuracy of 94%. Sana’a City’s identified flood-prone areas were precisely highlighted on the flood susceptibility maps, which offered insightful information about the spatial distribution of flood hazards. The results offer practical implications for urban planners and policymakers, facilitating targeted interventions to mitigate flood risks and improve urban resilience.