Explore the past. Understand what comes next.
Loading wildfire observations…
A browser-based workspace for the original FireCast wildfire research project. Explore 624 historical fires, daily boundaries, terrain and weather from 2018–2019.
The five-state Gaussian HMM uses 24 standardized features. Viterbi decoding, forward–backward inference and transition-based forecasts run locally using the exported trained model and scaler.
Forecasts estimate future hidden-state probabilities from observations through the selected cutoff. They do not predict geographic boundaries, damage, or containment. This research prototype is not an operational emergency tool.
Fire boundaries are simplified at 0.0001° and rounded to five decimals for display. Terrain rasters are sampled to 48 × 48 cells. Weather and model features retain source values. Satellite tiles: Esri and contributors. Dark basemap: CARTO / OpenStreetMap.