Logo Lanfrica

monte-kluemper/qgis-rwanda

Domaine:

environment and energygeospatial

Type de record:

software
Créateur:
mon
Hôte:
# Land Degradation Analysis (Sentinel-2) This package provides tools to analyze **land degradation** using **Sentinel-2 imagery** following UN-aligned indicators. It includes Earth Engine (GEE) scripts to export annual composites and Python scripts to calculate degradation indices and change maps. --- ## 1. Export Sentinel-2 Data from GEE ### a) Load Rwanda districts In the Google Earth Engine Code Editor, use the export_s2.js script in the gee_scripts directory: ### b) Download - Run exports from the **Tasks** tab in GEE. - Files will appear in **Google Drive → EarthEngine/** as GeoTIFFs (one per district/year). - Download them to a local folder, e.g. `C:/data/sentinel2_rwanda/`. --- ## 2. Python Setup ### Requirements - Python 3.9+ - Install packages: ```bash pip install rasterio numpy scikit-learn ``` ### Configure - Set your data folder: ```powershell $env:S2_DATA_DIR="C:/data/sentinel2_rwanda/" ``` - Place all `District_YYYY_S2.tif` files in that folder. --- ## 3. Scripts ### a) Long-term Degradation Trend Script: `compute_degradation_trend.py` - Uses **all years of Sentinel-2 data** for each district. - Computes the **NDVI trend (linear regression)** plus soil/vegetation condition indices. - Outputs: - `District_degradation_index.tif` → continuous raster (0–1). - `District_degradation_class.tif` → categorical raster (1–5 classes). Run: ```bash python compute_degradation_trend.py ``` --- ### b) Year-over-Year Change Script: `compute_degradation_yoy.py` - Compares each year’s indices to the **previous year**. - Highlights **areas of degradation (positive values)** and **improvement (negative values)**. - Outputs one raster per year pair: - `District_degradation_2020_vs_2019.tif` - `District_degradation_2021_vs_2020.tif` - etc. Run: ```bash python compute_degradation_yoy.py ``` --- ## 4. Visualization in QGIS - Load the output rasters. - For **classification rasters (1–5)** → use *Singleband pseudocolor* with discrete classes (green → yellow …