Recurrent Flood Mapping and Inundation Frequency Analysis in Bayelsa State, Nigeria, Using Sentinel-1 SAR Data and Machine Learning Prediction

Bayelsa State, located in Nigeria’s Niger Delta basin, is one of the nation's most flood-prone regions. Persistent inundation continually displaces populations, devastates local communities, and damages critical infrastructure. While the widespread construction of flood-break embankments underscores the severity of this crisis, the spatial extent, frequency, and future trajectory of these floods remain poorly quantified through systematic remote sensing. To address this gap, this study introduces an integrated framework that pairs multi-temporal Sentinel-1 Synthetic Aperture Radar (SAR) data with machine learning algorithms. Utilizing the Google Earth Engine platform, Sentinel-1 C-band SAR imagery was processed to leverage its all-weather, cloud-penetrating capabilities—a necessity for the cloud-heavy Niger Delta environment. A Random Forest classifier was applied to multi-temporal backscatter composites to separate flood and non-flood surfaces, producing inundation frequency maps that pinpoint chronically affected zones. Furthermore, a Long Short-Term Memory (LSTM) deep learning model was trained on historical SAR time series to project future flood extents by integrating seasonal rainfall and hydrological covariates. Validation was completed via desk-based cross-referencing against high-resolution Google Earth historical imagery, cloud-free Sentinel-2 optical data, and official disaster records from NEMA and SEMA. The results highlight persistent inundation hotspots along low-lying communities and major river corridors, where flooding spans multiple months annually. Predictive modeling indicates flood extents will intensify under current climatic trends. Ultimately, these findings deliver crucial geospatial intelligence to optimize flood-break placement, enhance early warning systems, and support disaster risk reduction planning across the Niger Delta.