Quanifying Fire Containment Patterns

The spread rate and containment of a wildfire shift as it burns across the landscape. The total area contained over time typicaly takes a sigmoid shape as the fire dies down or as fire crews actively suppress it.

We want to try and develop a better understanding for when, where, and how fast containment occurs for different fires. So we pulled a bunch of public data on containment and plotted 'em out. This is all done in the elmfire/containment/notebooks/PlotFitContainment.ipynb notebook.

Model fits

Shrubland fire	Forest Fire

Wildfires - especially Megafires - tend to spread rapidly at first and then slowly die down. These pattern are encoded when you plot just the percent contained over time.

Shrubland fire	Forest Fire

The scaled containment curves for all fires are shown in the hero image of this page.

For each fire with available data, we then fit a simple two-term logistic curve to estimate the containment curve midpoint (x_0) and the growth rate (k). Then, we plot how these values shift based on the size of each fire.

Curve Midpoint	Growth Rate

We haven't figured out, like, exactly what we want to do with these data yet. If you have any ideas - we're all ears!

Raw data

Temporally-explicit containment data is only accessible for a small number of fires via Wikipedia. Here's an example from the Tubbs Fire. There doesn't appear to be a standard process for collecting or reporting these data. Only a few fires have any available information, which tends to be reported as the fire is burning using information from the 10 AM reports provided by CALFIRE that some intrepid individual posts to the wiki for that fire.

For the fires that do have this information, we used this tool to export the tables to CSVs, cleaned them to ensure consistent naming and columns used, and stored the files in the ELMFIRE cloud storage bucket.

gs://elmfire/containment/*.csv