🌊 Flood Detection & Mapping System

A satellite-based flood detection system using Sentinel-1 SAR (Synthetic Aperture Radar) imagery and change detection algorithms. Detects flooded areas by comparing pre- and post-flood radar backscatter, maps flood extent, calculates affected area in km², and classifies severity — deployed as an interactive Streamlit web app.

Select a real flood event → AI analyses SAR backscatter change → Flood extent map with area statistics and severity rating.

• 5 real flood events — Pakistan 2022, Libya 2023, Nigeria 2022, Bangladesh 2022, Australia 2022
• Lee speckle filter — removes SAR multiplicative noise before analysis
• Log-ratio change detection — post_dB − pre_dB highlights backscatter decrease
• Otsu thresholding — automatic optimal threshold separating flood from land
• Morphological clean-up — binary opening + closing to remove noise and fill holes
• Flood extent map with blue overlay on SAR imagery
• Area calculation in km² using geographic pixel dimensions
• Severity classification: Minor (<200 km²) / Moderate (200–2,000 km²) / Severe (>2,000 km²)
• Upload your own GeoTIFF — works on any Sentinel-1 SAR file
• OpenEO integration — download real satellite data with Copernicus credentials
• Multi-event comparison chart

Floods happen during storms — clouds are always present. SAR is the correct sensor choice.

1. Clone the repo

git clone https://github.com/manny2341/flood-detection.git
cd flood-detection

2. Install dependencies

pip install -r requirements.txt

3. Start the app

streamlit run flood_app.py

Open http://localhost:8501

1. Pre-flood SAR scene loaded (VV band, linear scale)
2. Post-flood SAR scene loaded for same geographic area
3. Lee speckle filter applied to both scenes — reduces multiplicative SAR noise
4. Both scenes converted to dB scale: 10 × log₁₀(linear)
5. Log-ratio computed: change = post_dB − pre_dB
   • Flooded land has lower backscatter than dry land → strong negative change
   • Permanent water unchanged → near-zero change
6. Otsu threshold applied to -change — automatically finds optimal cutoff
7. Morphological operations clean the binary mask (opening removes noise, closing fills holes)
8. Area calculated from pixel count × geographic pixel size
9. Severity classified by flooded area threshold

• Synthetic mode: Realistic SAR scenes generated from statistical models (speckle + terrain + flood simulation) for immediate demo without credentials
• OpenEO mode: Real Sentinel-1 GRD data from Copernicus Data Space Ecosystem — free account required
• Upload mode: Your own Sentinel-1 GeoTIFF files from Copernicus Data Space Ecosystem (replaced scihub in 2023)

flood-detection/
├── flood_app.py         # Streamlit web application
├── flood_engine.py      # Core pipeline: SAR generation, Lee filter, log-ratio, Otsu, visualisation
├── outputs/             # Generated flood maps and comparison charts
├── data/                # Downloaded GeoTIFF files (auto-created)
├── requirements.txt
└── README.md

@manny2341