fire-spread

PyroWISE — open scientific documentation for the wildfire-spread simulation engine inside TerraWise. Methods open, engine commercial.

Attention-based spatiotemporal ConvLSTM models for wildfire spread prediction. Pairwise and patchwise self-attention mechanisms capture local fire front dynamics and long-range spotting effects. Validated on simulated percolation data and VIIRS satellite observations (California 2012–2020).

Wildfire spread forecasting that feeds the evacuation router: Rothermel cellular automata, hybrid bias correction, Dijkstra around roads predicted to burn, and submodular crew placement. Naive routing traps 27% of residents; forecast-aware traps none.

Anisotropic smoothing for change-point regression data

Fire-spread prediction in England’s non-domestic buildings, with temporal validation of Logistic Regression, Random Forest and XGBoost.

How do wildfires spread? An interactive, open-source explainer. Glassbox No. 109.

TFM de inteligencia artificial para predecir la severidad en incendios forestales utilizando datos históricos y modelos predictivos avanzados.

Predicted fire-spread for FIRESTORM — PyreCast/ELMFIRE 10/50/90 ensemble extent per active fire, UTM→WGS84 GeoJSON. MODELED forecast, agency public-safety use.