Ing-elvis-appiah/Accra_Flood_ML_Prediction

# Accra Flood Susceptibility Mapping Accra has been dealing with flooding for decades. Every rainy season, the same low-lying neighbourhoods go under — Korle Lagoon, the Odaw corridor, Accra New Town — and the same question gets asked: which parts of the city are actually at risk, and why? This project builds a supervised machine learning model that predicts flood susceptibility across the Greater Accra Metropolitan Area using satellite data extracted from Google Earth Engine. Flood labels were derived from Sentinel-1 SAR imagery captured during the June 2020 floods, and the model was trained on terrain, rainfall, vegetation, and land cover features at 30m resolution. --- ## Study Area Greater Accra Metropolitan Area (GAMA), Ghana Bounding box: 5.35°N – 5.85°N, 0.55°W – 0.05°W Reference flood event: June 2020 --- ## Data Sources | Dataset | Source | Use | |---|---|---| | Sentinel-1 SAR (GRD) | Copernicus / GEE | Flood label generation | | SRTM DEM 30m | USGS / GEE | Elevation, slope, aspect, TWI | | HydroSHEDS Flow Accumulation | WWF / GEE | Topographic Wetness Index | | JRC Global Surface Water | EC JRC / GEE | Distance to river | | CHIRPS Daily Rainfall | UCSB / GEE | Mean annual rainfall | | ESA WorldCover 2020 | ESA / GEE | Land use / land cover | | Sentinel-2 SR | Copernicus / GEE | NDVI | --- ## Features | Feature | Description | |---|---| | Elevation | Height above sea level (SRTM) | | Slope | Terrain steepness | | Aspect | Slope orientation | | TWI | Topographic Wetness Index | | Distance to river | Proximity to nearest water body (JRC GSW) | | Mean annual rainfall | 10-year average from CHIRPS (2010–2020) | | Land cover | ESA WorldCover 2020 class | | NDVI | Vegetation index from Sentinel-2 | --- ## Model Algorithm: Random Forest Classifier Training samples: 2,000 (1,000 flooded, 1,000 non-flooded) Validation: 80/20 train-test split + 5-fold stratified cross-validation | Metric | Result | |---|---| | AUC-ROC | 0.982 | | F1 Score | 0.927 | | A …