A proof-of-concept wildfire monitoring dashboard for Sentinel-2 satellite imagery. In a real deployment, imagery would be acquired on demand from SentinelHub or Copernicus Data Space over a configured zone of interest, and alerts would be issued in near real-time.
All imagery comes from the BC Wildfire Sentinel-2 Dataset — 516 paired RGB + SWIR preview tiles (256×256) from British Columbia, April–July 2025. Fire tiles overlap NASA FIRMS hotspot detections; clear tiles were randomly sampled within BC.
⚡ QuickFire CV — physics-based SWIR threshold algorithm (Pierre Markuse). Uses B11+B12 brightness plus a B12/B11 spectral ratio filter. Precision 0.95, recall 0.93, F1 0.94. Runs in under 10 ms with negligible power draw — ideal for onboard edge processing.
✨ AI Enhanced — SwinFuSR guided super-resolution applied to the SWIR bands before detection. Simulates the 2× resolution gap between Sentinel-2's 20 m SWIR and 10 m RGB bands. Uses the RGB image as a high-frequency spatial guide to recover detail lost at lower resolution. Switch between models by clicking a scene and using the toggle in the detail popup.
Marginal — faint SWIR signal, possible smouldering (B11+B12 > 1.0)
Moderate — established fire front (B11+B12 > 1.25)
High — intense fire activity (B11+B12 > 1.5)
Extreme — peak thermal emission, active crown fire (B11+B12 > 2.0)
The timeline slider replays detections in chronological order. The alert status updates based on how many active fire zones are present at each date.