FireCon targets regional wildfire containment forecasts with distributed compute

Researchers added hourly weather variability to a historically based containment model, producing FireCon’s forecasts of containment potential up to 48 hours ahead. A partnership with Wherobots targets the computing and data-preparation constraints that prevented timely regional runs.

For local governments coordinating wildfire response, a regional view could help identify changing containment opportunities and inform resource discussions across jurisdictions. Forecasts would still need operational interpretation: personnel deployment and microclimate conditions can affect containment outcomes in ways the model does not fully capture.

FireCon shifts wildfire containment modeling from annual maps of potential control locations toward forecasts that incorporate changing weather. Built on the Forest Service’s Potential Control Locations model, it uses a neural network to predict where containment features may hold or weaken up to 48 hours ahead. Individual-fire testing began in 2025, but demand for regional coverage exposed a computing bottleneck: those runs would have taken weeks on the existing infrastructure. Wherobots’ Rasterflow is intended to address that constraint through distributed processing and preparation of geospatial inputs, with outputs at 30-meter resolution. For counties and municipalities coordinating with fire-management agencies, regional forecasts could support discussions about resource availability and changing containment opportunities. Adoption would require agencies to interpret predictions alongside field observations and staffing constraints. The article reports validation methods but no accuracy metrics or achieved regional runtimes; deployment levels and operational benefits remain unclear.

Evidence supplied by Geo Week.