Detection at the spectral level, not the pixel level
NIR reflectance drops sharply when forest canopy is cleared. SWIR reflectance rises as bare soil is exposed. The Biodivex pipeline reads those band-delta signatures against a 6-week rolling baseline — and flags the polygon within 48 hours of the Sentinel-2 overpass.
Surface reflectance, not raw radiance
Biodivex ingests Sentinel-2 Level-2A surface reflectance data — atmospherically corrected to remove cloud and aerosol interference. This is the only input format that enables reliable spectral differencing across different acquisition dates.
- Sentinel-2 MSI at 10m/20m resolution, 5-day revisit at equator
- Sentinel-1 GRD C-band SAR for cloud-penetrating corroboration
- Landsat 8/9 OLI/TIRS for extended 30m baseline (16-day cycle)
- Planet SuperDove where licensed, for higher temporal density
Band differencing on NIR, SWIR1, SWIR2 against a 6-week baseline
Spectral change detection compares current acquisition values against a rolling 6-week cloud-free baseline for each pixel. Deforestation produces characteristic signatures: NIR drops sharply (healthy canopy absorbs NIR strongly), while SWIR bands increase (exposed bare soil reflects SWIR). Change polygons are generated via watershed segmentation from the difference image.
- Threshold calibrated per biome class (tropical moist, dry forest, savanna, mangrove)
- Seasonal NDVI variance factored into baseline to reduce false positives from deciduous behavior
- Minimum polygon area: 0.3 ha — filters noise while capturing small-scale clearing events
Each alert is scored, not just flagged
A raw spectral change polygon is not the same as a verified deforestation event. Biodivex attaches a confidence score to each alert based on three independent factors — reducing alert fatigue by distinguishing high-confidence events from uncertain detections.
- Cloud-free observation count in the 6-week baseline window (more observations = higher confidence)
- Seasonal NDVI variance (high seasonal variance = lower confidence, flagged for review)
- SAR corroboration status: Sentinel-1 C-band SAR confirmation adds ~15 percentage points to score
Three ways to receive detection data
Dashboard
Web-based map interface for non-technical teams. Filter by area, date range, and confidence threshold. Export polygons as Shapefile or GeoJSON.
REST API
GeoJSON polygon feed queryable by AOI bounding box, date range, minimum confidence, and biome class. Authenticated with API key. Rate limit: 1,000 requests/day.
Webhooks
POST to your system within 2 hours of detection. Configure per-AOI thresholds to reduce noise. Retry on failure with exponential backoff.
Configure a pilot AOI — we'll confirm sensor coverage and expected detection performance
Share the geographic coordinates or country/region and biome type. We'll describe revisit frequency, minimum detectable patch size, and typical confidence score distribution for your area.