A Hybrid GeoAI–Remote Sensing Framework for Urban Flood Susceptibility Prediction in Coastal Megacities: A Case Study of Lagos, Nigeria

Abstract

Urban flooding poses a persistent and escalating threat to coastal megacities, particularly in rapidly urbanizing regions where climate variability and unregulated land use intensify vulnerability. This study develops a hybrid GeoAI–Remote Sensing framework for urban flood susceptibility prediction in Lagos, Nigeria. The research integrates nine multi-source geospatial datasets, which include Aspect, Curvature, Slope Length, Slope, Geology, Land Use/Land Cover, TWI, Altitude, and Proximity to River, within a machine learning-driven analytical environment. Advanced GeoAI techniques, specifically Random Forest (RF) and Logistic Regression (LR), were employed to model flood susceptibility by capturing relationships among hydrological and anthropogenic factors. A multicollinearity analysis indicated absence of problematic linear multicollinearity of the predictors (Average VIF = 1.41). Model performance was rigorously evaluated using AUC-ROC, R², and Mean Squared Error (MSE). The results revealed that the Random Forest (RF) model demonstrated higher predictive performance than the Logistic Regression model in predicting spatial flood susceptibility, achieving an excellent AUC of 0.976 and an R² score of 0.762, compared to the LR model’s AUC of 0.923 and R² of 0.444. Feature importance analysis identified Proximity to River (0.663) emerged as the strongest predictive factor governing spatial flood susceptibility across the study area. The resulting susceptibility maps reveal that approximately 43.85% of Lagos falls within "High" to "Very High" susceptibility zones, with respect to the RF model. This study provides a scalable decision-support tool for disaster risk management and offers practical implications for urban planners seeking to enhance climate resilience in coastal megacities.