Ignitions are placed at controlled intervals along your infrastructure so every span is tested.
Methodology
How Flarecast turns weather, fuels, terrain, and infrastructure into a defensible risk score.
Thousands of fires are modeled along your infrastructure. Each ignition is spread under historical, current, and forecasted conditions, and the structures it reaches are counted toward the segments that produced it.
How the risk model works
The model runs a probabilistic ensemble: place an ignition, sample a weather stream, grow the fire, and measure the consequence. The results roll up to a score for each segment.
Ignitions are placed at controlled intervals along your infrastructure so every span is tested.
Model inputs
Five inputs feed the model. Dataset names and providers are published in the technical documentation. Forecast weather runs on the US operational models by default. Where a team already has a commercial or in-house feed under contract, such as DTN, Vaisala, or another provider, Flarecast can ingest that instead.
Hover an input to see what it contributes.
- Weather: Historical trends, current observations, and forecast conditions, including wind and fire-weather indices such as ERC.
- Fuels: Fuel type and current status across the landscape each modeled fire moves through.
- Terrain: Slope and aspect, which steer fire growth as strongly as wind.
- Values: Buildings and values a modeled fire can reach, used to quantify consequence.
- Infrastructure: Lines and assets from your GIS data, segmented for scoring.
Sensitivity in practice
Ignition modeling along infrastructure
Rather than rating a corridor as one object, the model tests it span by span. Simulated ignitions are placed at controlled intervals along the line. Each one grows into a fire under the conditions of the day, and the structures those fires reach are attributed back to the segments that produced them.
The result reads directly on the map: exposure clusters where consequence concentrates, and the segments driving it are obvious. That is the evidence behind operational calls like Public Safety Power Shutoffs (PSPS).
Ignition spacing is derived for each deployment from corridor length and asset density, then tightened until segment scores stop changing. The interval used for your system is stated in the technical documentation.
Validation
Fire growth in Flarecast is simulated by a peer-reviewed spread model with a public operational record. For more details including full citations and validation results, contact us for the technical documentation.
Peer reviewed
updated2026
The spread model and its automated calibration method were published in a peer-reviewed journal and have been refined continuously since, with the latest engine update this year.*1
Operational since 2020
6seasons
Fire seasons of real-time forecasting on most large fires in the continental US, with spread forecasts out to 14 days on the public platform the same engine drives.*2
Graded against real fires
214forecasts
Seven-day spread forecasts from the 2022 season, compared after the fact with what burned; 96% rated acceptable or better. Independent UC Berkeley tests on the 2017 Tubbs and Thomas fires found burned-area overlap above 85%.*3
- *1 Published in Fire Safety Journal, 2013. Most recent engine update: March 2026.
- *2 Real-time forecasting from 2020 onward, per the engine's validation documentation.
- *3 Retrospective review of 214 seven-day forecasts from the 2022 fire season in the engine's validation documentation: 205 rated acceptable, good, or excellent. Perimeter overlap from UC Berkeley research published in Proceedings of the Combustion Institute, 2024.
- *4 Perimeter comparison is an illustrative reconstruction on a synthetic corridor, not the study fires.
Cadence and horizon
Timing follows the US operational weather models that drive the forecast by default, all of them run by NOAA, the National Oceanic and Atmospheric Administration, within the US Department of Commerce. A commercial or in-house forecast feed can be ingested in their place, in which case cadence and horizon follow that feed. Ignition spacing is not listed here because it is derived for each deployment from corridor length and asset density.
- Four runs a day, aligned to the 00, 06, 12, and 18 UTC weather cycles.
- Those four cycles are where the driving models issue their long-range output. HRRR, the High-Resolution Rapid Refresh, extends from 18 to 48 hours, the 3 km regional tier to 84 hours, and the global tier to 16 days.
- Model runs on intermediate hours refresh the near term at a shorter horizon, so they revise the next few hours without moving the multi-day outlook a shift is planned against.
- Set by the same 6-hourly cycle, so the ceiling moves with the tier of weather data in use.
- Where a client supplies its own forecast weather, the horizon follows that feed rather than the figures below.
- All three models are produced by NOAA, the National Oceanic and Atmospheric Administration, an agency of the US Department of Commerce.
- HRRR, the High-Resolution Rapid Refresh, reaches 48 hours over the continental US.
- RRFS, the Rapid Refresh Forecast System and the 3 km regional tier, reaches 84 hours.
- GFS, the Global Forecast System, reaches 384 hours, roughly 16 days, at coarser detail.
Technical documentation is available upon request.
Dataset providers, model configuration, and validation results are shared with qualified teams during evaluation.