Green Shift Hacks 2026 · Ecosystems & climate resilience
Forest loss, measured from orbit.
Sylvan Watch compares real Sentinel-2 imagery across years, finds where canopy disappeared, measures how many hectares, and evaluates model predictions against Hansen reference labels.
Loading model evaluation…
From pixels to polygons
Every stage runs on real data: cloud masking, seasonal composites, 27 spectral and temporal features, a trained detector, and georeferenced vectorisation.
Real imagery, no mock-ups
Sentinel-2 surface reflectance fetched live from Earth Engine for the area you draw.
Three detectors, one test set
A spectral rule, a Random Forest and a Siamese U-Net are scored on the same held-out sites.
Hectares and coordinates
Each loss patch becomes a polygon with area, centroid, confidence and land-cover context.
What it does not claim
- It cannot tell whether clearing was legal; imagery shows loss, not permits.
- Hansen labels are a reference product, not field truth, and every metric says so.
- Selective logging of single trees is below 10 m resolution.
- Accuracy is reported for the evaluated sites only, never worldwide.