Modeling Fire Behavior for NW California Fires

We use many different tools and models to evaluate upcoming threats of major fire spread or conditions which could allow rapid growth on new starts.

I’m working as an evaluator on a project which is building new fire behavior forecasting tools for the California Energy Commission. This work is being supervised by the Spatial Informatics Group (SIG) under an umbrella project called Pyregence.  The fire spread model is called Pyrecast.  This project open source, so check it out if you are interested (it runs best in Google Chrome).

How Pyrecast Works
The Pyrecast model uses MODIS heat satellite data to map existing fire areas (ignition sources for future fire spread), and models the potential spread of the current fire using existing 30m-resolution fuel loading/vegetation mapping, terrain models, along with current fuel moisture conditions and forecast weather (GFS, NAM, and HRRP models). It can be run over an 8 day forecast period. It is a probabilistic model, meaning that it looks at historic weather to infer the likelihood of a given wind speed or direction across a certain time period. So if you run the model with the 90th percentile conditions, it will find the 90% hottest, driest days from the period of record and use those along with the forecast winds to estimate potential fire spread and intensity.

My Experience With the Model
My job on the Pyregence project is to test the model’s performace against observed conditions on major wildfires. In the past 3 seasons, it has accurately predicted several major blowups on fires including the 2020 Bear Fire, 2021 Dixie Fire, and 2022 Mosquito Fire. On the 2021 Dixie Fire, the model correctly called a major run up the North Fork Feather River Canyon, but underestimated the time it would take –  it thought it would take 4 days to do what the fire actually did in 2). It did a good job predicting the major SE run on the Mosquito Fire on Eldorado National Forest last year that pushed hard on Blodgett Forest. During 2020 Bear Fire, the model predicted a 15 mile westward run (3 days before it happened). The fire actually ran over 30 miles on the predicted day.

How is the Model Performing?
Like all of our current models, Pyrecast falls short in dealing with effects of inversions socking in canyons and suppressing fire behavior, so it does best when we have conditions which clear out the inversion for an extended period of time so the fire can just run. It doesn’t seem to do very well with backing fire, either. What the model DOES do well is show places where forecast weather, topography, fuel loading, and existing fires align. It’s not a precision tool, but it is useful is showing potentially dangerous places in the near future.

Applying Pyrecast to Current NW California Fires
Running higher percentile runs, Pyrecast has some pretty grim predictions over the coming 4 days for fires in NW California. The major weather models are forecasting NW winds starting tomorrow, transitioning to N winds Weds morning over the Smith River Complex. Many of the areas to S and SE of Gasquet (and also between Somes Bar and Mosquito/Bluff Fires) haven’t had major fires for a long time, it looks like there is some potential for major growth on these fires to the south and east.

Interpreting The Outputs
In the runs below, the ‘time slider’ on the bottom center shows the time and date for the predicted fire shapes shown in shades of gray in the main map. Pink areas are already burned as of the time the model was run, and dark blue areas are the most recent satellite heat detections at time model was run. The ‘Predicted Fire Size’ option on the left menu shows which percentile (probability) was used to generate the weather inputs against the forecast data – basically, the model is using historic weather to assign a likelihood to how strong the winds actually might end up being. 90th percentile is the highest option, and generates the worst-case scenario. 70th percentile is more toned-down, and 50th generally doesn’t usually move the fire in any major/dramatic fashion. You can play with all of these options yourself at

The model below shows 90th percentile (most aggressive) forecasts for fire spread on the Smith River Complex through Thursday 8/31/2023.

Screen Shot 2023-08-27 at 12.11.24 PM

The Pearch Fire shows potential for a strong eastward push this coming week on the same model.

Pearch Fire – 70th percentile run.

A lot of good firefighting has been happening on the Mosquito and Bluff Fires, north of Weitchpec, especially on the eastern flanks, and they’ve been cooling down quite a bit past few days, but model shows north end of Mosquito blowing out and running toward Somes Bar over the next week.

Mosquito/Bluff Fires

The model shows potential for the Blue 2 Fire to push west, downhill, almost to the Elliot Fire. In the past, I’ve found the model more inaccurate on fires moving against the topography, and as with all of these runs, you’ve got to take them for what they’re worth.