Machine learning-based shoreline change prediction and erosion analysis: a case study of Ogu/Bolo, Nigeria

Extensive shoreline retreat: Over 78% of transects showed erosion between 1984 and 2024, averaging − 1.49 m/year. Machine learning enhances prediction: Random Forest outperformed SVM and LSTM in predicting shoreline change with R² = 0.88. Key erosion drivers identified: Mangrove loss (–17 km²), low elevation, and flat slopes were major contributors. Future risks mapped: Projected shoreline changes to 2044 indicate retreat up to 20 m in erosion hotspots. Integrated framework proposed: Combines DSAS, geospatial analysis, and explainable AI for coastal risk assessment and management. This study examines shoreline change and erosion dynamics along the Ogu/Bolo coastline in Nigeria’s Niger Delta from 1984 to 2024, with predictive projections to 2044. The objective was to quantify past shoreline shifts, identify key environmental and anthropogenic drivers, and develop accurate predictive models for future shoreline positions. Multi-temporal Landsat imagery, combined with datasets on topography, land use/land cover, slope, elevation, and vegetation indices, was processed using the Digital Shoreline Analysis System (DSAS) across 94 transects. Shoreline displacement was evaluated using End Point Rate (EPR) and Weighted Linear Regression (WLR), while predictive modeling employed Support Vector Machine (SVM), Random Forest (RF), and Long Short-Term Memory (LSTM). SHapley Additive exPlanations (SHAP) analysis was applied to identify the most influential predictors, including distance to tidal inlets, Normalized Difference Vegetation Index (NDVI), and slope. Results showed that 78.72% of transects experienced net erosion, with a mean retreat of − 1.49 m/year. RF achieved the highest prediction accuracy (R² = 0.88; RMSE = 0.92 m). Morphological analysis indicated a cumulative land loss of 0.89 km², while land use assessments linked erosion hotspots to mangrove deforestation (–17 km²), low elevations (< 2 m), and flat slopes (< 1°), exacerbated by sand mining and dredging. Projections suggest erosion hotspots may retreat by over 20 m in some areas by 2044. The study demonstrates the value of integrating remote sensing, statistical analysis, and explainable artificial intelligence (AI) for shoreline prediction, providing actionable insights for coastal management, ecosystem restoration, and policy intervention.