🌿 Western Ghats — 2030 Frontier Risk Digital Twin

Random Forest · Landsat 5 / 8 / 9 · 2006–2026 · Predicting Forest Loss Probability by 2030
🛰️ Landsat Multi-Mission 🌍 Western Ghats, India 🤖 Random Forest ML 📅 21 Annual Frames
Loading Digital Twin map…

Study Area

~160,000 km²

Western Ghats UNESCO World Heritage Site. Runs 1,600 km along India's western coast.

Time Horizon

2006–2030

21 annual dry-season Landsat composites (2006–2026) used to predict 2030 deforestation risk.

Spatial Grid

~5 km

GEE export at 5 km grid (scale=5000). ~6,500 analysis pixels across the corridor.

Sensors

L5/8/9

Landsat 5 TM (2006–2012), Landsat 8 OLI (2013–2022), Landsat 9 OLI-2 (2023–2026). Dry season: Jan–Apr.

Risk Metric

RF Prob.

Random Forest probability of forest-to-non-forest transition by 2030. Magenta = extreme risk.

🛰️ Data Sources

Google Earth Engine collections LANDSAT/LT05/C02/T1_L2, LANDSAT/LC08/C02/T1_L2, LANDSAT/LC09/C02/T1_L2. Annual dry-season (Jan–Apr) median composites. Terrain from CGIAR/SRTM90_V4. Settlement proximity from JRC/GHSL/P2016/SMOD_POP_GLOBE_V1.

🌿 NDVI & Feature Engineering

Per-pixel annual NDVI computed for 2006–2026. Temporal features: OLS trend slope (NDVI/yr), standard deviation, minimum observed NDVI, count of loss-years (NDVI < 0.35). Terrain features: elevation, slope, distance to nearest settlement.

🤖 Random Forest Model

Binary classifier (200 trees, class_weight='balanced'). Training labels: pixels forested in 2006 that were deforested by 2015 (class=1) vs. still forested in 2015 (class=0). Output mode: predict_proba → continuous [0, 1] risk score.

🗺️ Visualisation

Interactive Folium map with two risk layers: (1) HeatMapWithTime showing 21 annual NDVI frames for historical context; (2) static RF probability heatmap using a yellow → orange → Digital Magenta palette to highlight extreme-risk frontier zones.

⚠️ Conservation Context

The Western Ghats is one of the world's eight biodiversity hotspots, harbouring over 5,000 plant species and 139 mammal species. With the UN 2026 Global Forest Goals Report reporting a global net loss of 40M+ hectares since 2015, predictive risk tools are critical for directing scarce conservation capital before ecological collapse.

🔧 Pipeline

GEE JS → WG_Landsat_Annual_NDVI_2006_2026.csv
→ wg_01_process_landsat.py (features + labels)
→ wg_02_rf_risk_map.py (RF model + Folium map)
→ wg_frontier_risk_2030.html (deploy to repo root)

Demo mode (no GEE required):
python python/wg_01_process_landsat.py --demo
python python/wg_02_rf_risk_map.py