Integrating Multi-Temporal Earth Observation Data, Machine Learning, and CA–Markov Modelling for Historical and Future Flood Susceptibility Mapping in Lokoja, Nigeria

Flooding remains one of the most critical environmental challenges affecting urban and riverine communities in Nigeria, with the Lokoja Local Government Area (LGA) in Kogi State particularly vulnerable due to its location at the confluence of the Niger and Benue Rivers. This study integrates multi-temporal Earth Observation data, Geographic Information Systems (GIS), and Machine Learning (ML) techniques to evaluate historical flood susceptibility and predict future flood risk patterns in Lokoja. Multi-source geospatial datasets, including multi-temporal Landsat satellite imagery (Landsat 7 ETM+ and Landsat 9 OLI/TIRS) for 2004, 2014, and 2024, Shuttle Radar Topography Mission (SRTM) Digital Elevation Model (DEM), Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS), river and road network data, land use/land cover information, and historical flood records were integrated within a GIS environment. Random Forest (RF) and Support Vector Machine (SVM) algorithms were employed to generate and compare flood susceptibility maps based on environmental and anthropogenic conditioning factors. Future flood susceptibility scenarios were simulated for 2044 using Cellular Automata–Markov (CA–Markov) modelling implemented in IDRISI TerrSet. The findings reveal significant temporal changes in flood susceptibility between 2004 and 2024, with increasing high-risk zones associated with urban expansion, vegetation decline, terrain characteristics, rainfall variability, and proximity to major river systems. The Random Forest model demonstrated superior predictive performance compared with the Support Vector Machine, highlighting its ability to capture complex relationships among flood-driving variables. Future projections indicate potential expansion of severe flood-prone areas by 2044, emphasizing the influence of ongoing land transformation and climatic variability. This study demonstrates the value of integrating multi-source remote sensing data with machine learning and predictive modelling approaches for flood risk assessment. The findings provide important scientific evidence to support sustainable urban planning, disaster risk reduction, and climate adaptation strategies in vulnerable tropical environments.