laureboudinaud/s2-village-burn-assessment

Scalable open-source pipeline for detecting and characterising burn damage at settlement level using Sentinel-2 dNBR and BAI. Outputs monthly burn chronology, burned area estimates, and building-level exposure statistics per settlement. Example application: El Fasher area, North Darfur, Sudan for the period Dec 2024 to Apr 2025. # Village Burn Assessment **Satellite-based burn detection at settlement level in the surroundings of El Fasher (North Darfur)** Dec 2024 – Apr 2025 --- ## Overview This repository documents a satellite-based pipeline for detecting and characterising burning events at the village level in the area of El Fasher (North Darfur, Sudan). The pipeline combines Sentinel-2 spectral indices processed in Google Earth Engine with OpenStreetMap settlement extents and VIDA building footprints to produce monthly, village-level burn exposure statistics. --- ## Repository structure ``` village-burn-assessment/ ├── README.md ├── requirements.txt ├── .gitignore ├── 01_burn_score_compute.ipynb # GEE computation and monthly burn score rasters export ├── 02_settlements_process.py # Settlement extraction, burn statistics, confidence scoring ├── 03_figures_plot.py # Figures (spatial maps, monthly and locality aggregations) ├── run.py # Entry point for 02_settlements_process.py and 03_figures_plot.py ├── s2_3pts_inspect.js # GEE visual inspection tool (3-period NDVI timescan, Dec–Apr) ├── data/ # Input data └── outputs/ # Generated CSVs, geopackages and figures ``` --- ## Pipeline The full workflow is structured as five sequential steps: **(1) Settlement extraction** : merge HOT OSM polygons with manually digitised extents; join name attributes from the OSM point layer using two-pass spatial matching (containment, then nearest-neighbour with a 3 km distance cap). **(2) Monthly burn detection** : apply a burn score threshold to monthly Sentinel-2 composite rasters derived from dNBR and BAI indices with weighted confidence scoring and morphological denoising; classify each settlement as burned / not burned per month. **(3) Settlement statistics** : intersect burn masks with settlement polygons to compute burned area (ha) and percentage of total settlement extent; intersect burn masks with building footprints to count burned buildings per settlement and th …