StarGazer500/forest-lulc-ghana

Sentinel-1/2 land use land cover classification for Anwiaso East Forest Reserve, Ghana — Random Forest and deep learning (ResU-Net + MAnet) pipelines via Google Earth Engine. # Forest LULC Classification — Anwiaso East, Ghana A land use/land cover (LULC) classification pipeline for Anwiaso East Forest Reserve, Ghana, combining Sentinel-1 SAR and Sentinel-2 multispectral imagery with both classical and deep learning approaches. ## Overview This project classifies forest landscapes into 11 land cover types — including dense forest, degraded forest, cocoa farms, illegal mining (galamsey), and water bodies — using 10 m satellite imagery processed via Google Earth Engine. **Study Area:** Anwiaso East Forest Reserve, Ghana (~6.05–6.36°N, 2.13–2.26°W) **Resolution:** 10 m pixels (Sentinel-2 native) **CRS:** EPSG:32630 (UTM Zone 30N) **Season:** Dry season imagery (November–February) to minimize cloud cover ## Example Output *LULC classification result (ResNet50-based deep learning model) displayed in QGIS — Anwiaso East Forest Reserve, Ghana* ## Land Cover Classes | ID | Class | Description | |----|-------|-------------| | 1 | Cocoa | Cocoa plantations (~90.7% of study area) | | 2 | Degraded Forest | Partially cleared/disturbed forest | | 3 | Dense Forest | Intact closed-canopy forest | | 4 | Farms | Agricultural land | | 5 | Galamsey | Illegal artisanal mining sites | | 6 | Invasives | Invasive plant species | | 7 | Natural Forest | Secondary/mixed natural forest | | 8 | Open Areas | Bare land, cleared areas | | 9 | Road Buffer | Road corridors | | 10 | Swamp | Wetland/swamp areas | ## Feature Stack (26 bands) | Category | Features | |----------|----------| | Sentinel-2 Multispectral | B2, B3, B4, B5, B6, B7, B8, B8A, B11, B12 | | Spectral Indices | NDVI, NDWI, NDRE, EVI, SAVI, NBR, BSI | | Sentinel-1 SAR Statistics | VV/VH mean, min, max, std + VH/VV ratio | ## Notebooks Run notebooks in this order: ### 1. `data_prep.ipynb` Connects to Google Earth Engine, downloads and composites Sentinel-1 and Sentinel-2 imagery, computes spectral indices, samples training polygons, and exports the 26-band feature stack and training samples to …