During the summer of 2016, I had the opportunity to serve as a wildland firefighter for the United States Forest Service. While in this position I noted many potential applications for the use of information technology to vastly improve prediction of forest fire starts, which could, in turn, lead to better protection of our forests and citizens.
Starting with the problem
Last year alone the U.S. Forest Service spent over $2 billion suppressing forest fires and yet nine million acres of forest and habitat were still lost. The economic, environmental, and social impacts that forest fires have continue to worsen each year due to our warming planet and poor forestry management practices. One of the main reasons that fires get to a point where they can decimate habitats and communities is that they are not caught early enough. Almost all naturally occurring forest fires are started by lighting and these starts can often lay dormant and small for days, but despite being easy to suppress at this time are often not found until they are much larger and more difficult to control.
With these issues in mind I set out to find a way to improve the situation at hand. As I always do, I begin my projects by talking to others. In this case, I particularly wanted to make sure that I had a firm understanding of the problem. As I have talked with experts from the Forest Service, NOAA, Bureau of Land Management, and NASA, I have become increasingly more convinced that there is much room for improvement in this field.
Currently, forest fires are located via one of four methods: lookout sighting, civilian sighting, aircraft sighting, or infrared satellite detection. There are several key issues with this current methodology. First, sightings require physical people to be in the area–either on the ground or in the air. Monitoring our millions of acres of forest in such a way is impossible.
This leaves us with infrared satellite data. While this method sounds good in theory, its veracity is highly dependent on cloud cover, which can block infrared heat signatures. As almost all natural fire ignitions are caused by lighting this method is not very robust, since lightning is always accompanied by cloud cover. Furthermore, only two satellites, NASA’s MODIS and VIIRS, are capable of fire detection. These satellites only pass over the US once every 6 hours, which is often too long for the fire to be detected in its early stages.
Building a better solution
I am currently working on utilizing a machine learning algorithm which could predict fire ignition locations based on instantaneous lightning strike data collected by electromagnetic sensors stationed across the United States. This prediction method would not be dependent on cloud cover, or human resources and as a result, could significantly improve incident response times.
Check in later for more updates!