In recent history, the frequency and severity of wildfires has skyrocketed with the most notable occurrences happening in the West part of the United States in areas such as California and Colorado. While the West has the most severe wildfires, with some fires burning over 100,00 acres, the East has more wildfires per year than the West. This poses a problem for the Eastern United States because of the proximity of many residential areas regions that are susceptible to wildfires. Additionally, the severity of wildfires is growing at an alarming pace. In 2023, the United States reported 56,580 wildfires that burned 2,693,910 acres. In 2024, the number of reported wildfires rose to 64,897, with 8,924,884 acres burned—nearly three times the acreage lost per fire than in 2023. The rise in fire severity and the greater number of wildfires in the East motivates the need to research Eastern wildfire management.
One of the many labs that does Eastern Wildfire Research is the Laboratory for Autonomy in Data-Driven and Complex Systems in the Ohio State University Mechanical and Aerospace Engineering Department. One of the projects in this lab is that of PhD candidate Grant Mirka, who, under the direction of Dr. Mrinal Kumar, is using a Hyperspectral camera mounted on a drone to scan the prairie and extract relevant information for wildfire workers. A hyperspectral camera is different than a normal RGB camera in that instead of only taking in visible light, it can take in near infrared light along with hundreds of bands at each pixel.
This means that for every pixel, instead of having just three values, Red, Green, and Blue, it can take in an almost continuous signal of light. This allows for feature extraction that would otherwise be impossible with a normal RGB camera.
In Fall of 2025, a data campaign is being conducted to collect Hyperspectral images of the prairie and along with bagged field clippings of the vegetation in the prairie. One of the goals of this dataset is to unmix each pixel into its subsequent endmembers. An endmember is a pure spectral signature given off by a single material. In this case, an endmember might be goldenrod. By doing this, an abundance heatmap can be created showing where each type of vegetation is in the prairie. Another goal of this dataset is to determine the moisture content at every pixel in the image. To do this, the ground samples are weighed, dried, and then weighed again to get their moisture percentage and then each sample is mapped to their corresponding pixel in the image. Then each ground truthed pixel is used to train a machine learning model to understand how to map hyperspectral signals to moisture content. Once the model has been properly trained, it is then applied to the rest of the hyperspectral image to create a moisture abundance map where each pixel has a moisture value associated with it.
Both of these metrics are invaluable in making active wildfire decisions. They allow wildfire workers to focus on areas that will burn faster and hotter and dedicate less effort to areas that will naturally extinguish the fire.
Future work involves combining these metrics along with local topography and meteorology to create a ground accurate fire prediction model that wildfire workers can use during an active fire to help guide their efforts and predict fire behavior.