Delivering AI-backed tools that improve the health and resilience of our wildlands by supporting fire management

All over the world, the health of our forests impacts the quality of the air we breathe and the water we drink. Effective management of burnable material in forests is essential to support healthy, resilient forests that are less susceptible to catastrophic wildfires.

Smoke filled Seattle skyline.

Our Approach

We are a team of machine learning researchers, engineers, and designers working together to apply machine learning to support wildland fire management and research workflows. We are focusing our work on ground-level fuels that are not measurable with current satellite technology. Surface fuels are a neglected research area in wildfire research and management – we are taking current low-tech methodology and applying computer vision to greatly amplify the impact.

Diagram identifying the different fuels of a forest (canopy, above duff, ground, etc).

What is Fuel Management?

Anything burnable is considered fuel in a forest ecosystem. When fuels are in excess and conditions are right, fires can burn hotter, longer, and faster than expected, making them more challenging to manage.

Fire plays an important role in minimizing the negative impacts on forests, and not all fire is bad. Prescribed burning can help reduce fuels to help improve habitats and maintain ecosystems. By managing fuels, we can prevent fires from growing out of control, helping to save lives and property.

Fuel loading estimates for fire-prone ecosystems are vital for accurately predicting fire behavior and effects, including flame length and intensity, as well as smoke emissions, soil heating, nutrient dynamics, and water filtration rates.

Improving Current Techniques

Photoloading is a technique for estimating the fuel loading of a forest floor by visually comparing conditions captured in the field against a set of standards, or sequences. Photoload sequences are downward-looking, close-up photographs showing graduate fuel loadings of synthetic fuel beds.

While the photoload technique is relatively accurate and inexpensive, it is still a time-consuming process that is not widely adopted. With the help of machine learning, we hope to make this process faster and easier, greatly increasing the number of measurements available to help support fuel analysis and fire management.

Forest floor with sticks and debris highlighted by a computer vision model.Forest floor with random sticks and debris.

Our Project

The Wildlands team is working toward a proof-of-concept project to use machine learning to estimate the loading for 1, 10, and 100-hour fuels in Kg/m2 using photos taken with a digital camera or cell phone. This project will enable any forester with a camera to report quick, accurate fuel loading estimates, allowing fire managers, fuel specialists, and researchers to make more accurate planning predictions. This is part of a longer-term vision to apply machine learning to support many aspects of wildland fire management.

1 meter by 1 meter testing frame on forest floor.
Four hands working together
Collaborate With Us
Have ideas about how computer vision, machine learning, or other technology could be used to improve forest health? We are actively looking for project collaborators – please get in touch with us at wildland-fire-team@allenai.org.