Home Artificial Intelligence A Recent Way: Wildlands App and Counting Sticks The state of the world Current fuel estimation processes when using Photoloading The Photoload App The chances The subsequent steps

A Recent Way: Wildlands App and Counting Sticks The state of the world Current fuel estimation processes when using Photoloading The Photoload App The chances The subsequent steps

A Recent Way: Wildlands App and Counting Sticks
The state of the world
Current fuel estimation processes when using Photoloading
The Photoload App
The chances
The subsequent steps

A photo of sticks and grass in a forest, marked up to indicate the amount of time each stick would burn for.
Microplot from Know Your Fuels, Know Your Fire Fall 2022 Workshop with illustration overlay of tools utilized by photoloading practitioners. Photo and illustration by Karen Goodfellow.

In at the least eight states across North America, the wildfire season of 2022 caused road closures, forced evacuations, and air quality that ranked worst on this planet resulting from the associated smoke. Six of the ten most destructive wildfires in California’s history have occurred throughout the last five years. Briefly — fires, caused each by humans and nature, can have devastating effects after they are unplanned or burn uncontrolled.

Historic wildland management practices often attempted fire exclusion, the prevention and rapid suppression of any fire. This practice inadvertently created wildlands vulnerable to extreme fires and interfered with the life cycles of some vegetation.

Restoring wildlands to their historical, fire-adapted state can reduce the likelihood of destructive, out-of-control fires; fuels reduction is vital to doing so. Surface fuels measurements are vital before and after fuels reduction treatment, whether through mechanical thinning or controlled burns. Practitioners can pick from several different techniques to estimate fuels. All of them demand consistent, rigorous fieldwork.

With this in mind, AI2’s latest Applied Sciences & Technologies team — the Wildlands team — got down to help fire managers, fuel specialists, and researchers by applying AI as an assistive tool for fuel load estimation. The team’s initial work is targeted on developing AI models that may perform the estimates required by the photoload sampling technique.

Today, some field practitioners use a way called photoloading to estimate surface fuels. This requires them to take a one-meter square PVC pipe frame, a size gauge, and a camera into the world they shall be measuring. They have to also bring an in depth book with reference images for visual comparison. Finally, practitioners must carry sampling data sheets, pens, and a calculator to record the info gathered. Other techniques–corresponding to transects–have similar recording and calculation requirements.

From left to right: In the sphere taking notes on fuel loadings. Photo by Jenna James // Traveling to photoloading site for Know Your Fuels, Know Your Fire Workshop led by Heather Heward. Photo by Karen Goodfellow // Compiling photoloading data right into a spreadsheet. Photo by Karen Goodfellow

Photoloading relies on following consistent sampling and estimation processes to amass useful data. The method for collecting plot sample data has been designed to remove human biases in where samples are collected, and is comparable for estimation techniques aside from photoloading as well.

Whatever the fuels estimation technique, practitioners must estimate the mass of several different surface fuel components: duff, litter, shrub, herb, and woody, where woody fuels are further subdivided based on their diameter into 1-, 10-, or 100- hour classes. In photoloading, the practitioners use the one-meter square frame to define several areas–known as a microplot that’s photographed–where the fuel components listed above have to be estimated. Taking a look at each fuel component in turn, the practitioner evaluates what they see throughout the plot frame and compares that to the reference images in an effort to derive mass estimates for the fuels. After making each estimate, the practitioner records them on a sample sheet. For some fuel classes, the practitioner must then perform a series of table lookups and adjustment calculations to derive and record a final value. When all other fieldwork is complete, the practitioner transcribes all data from their sample sheets right into a spreadsheet for durable preservation and further evaluation.

Fuels estimates are essential to observe forest health, predict wildfire risk and severity, and safely execute prescribed burns. On account of the high effort required, and the proven fact that there are few practitioners available to perform the work or teach these processes to latest people, the necessity for fuels estimates far exceeds the forestry community’s ability to assemble such data.

“There may be rarely enough time to coach people to develop into proficient enough within the system that they feel independently capable and motivated to gather fuel-loading data,” says Heather Heward, senior Fire Ecology & Management instructor on the University of Idaho.

“There may be also an ideal deal of decision-making in fuels sampling, which could be very tiring and is less appealing than using visual estimates, or nothing in any respect to measure fuels.”

“This could be very much a flossing your teeth activity,” said Paul Albee, a lead research engineer on the AI2 Wildlands team. “You realize you must do it, but no one desires to.”

The Wildlands team joined practitioners in the sphere to perform photoloading and saw opportunities to enhance the method’s efficiency and accuracy. Inspired, they began developing an application to make it easier to gather surface fuels data.

This application mechanically performs the calculations and table lookups that practitioners manually perform today. It also keeps photos and data together and linked, eliminating the necessity for practitioners to diligently sequence their photos to align with their data. Each of those features reduce the danger of mistakes in transcribing and reviewing collected data, a crucial consideration for practitioners who could also be physically and mentally drained given the difficult nature of their fieldwork.

Screenshots from the Photoload App. Left: Users can enter data for their microplot. Right: Photos can be associated for each microplot.
Screenshots from the Photoload App. Left: Users can enter data for his or her microplot. Right: Photos could be associated for every microplot.

The team hopes to deliver an application that’s intuitive for those trained in surface fuels measurement and simple to make use of in the sphere. The app will work offline during data collection, and can offer several ways for practitioners and land managers to eat the info when the app is back in cellphone range. Finally, the AI2 team will use the collected data for developing assistive AI.

“This app will streamline the info collection process, make training easier, and make reporting easier. It should throughout help increase the likelihood that individuals will collect data and that the info they collect is consistent between sampling events and between individuals who sample,” says Heather.

The Wildlands team is already establishing partnerships with the University of Idaho, the US Forestry Service, and the Washington State Department of Natural Resources to implement this app out in the sphere. As practitioners adopt the applying, their feedback may also help to shape improvements that may make it as useful in the sphere as possible.

In talking with Heather, she identified three essential challenges that the photoload application could assist with.

“First, higher information will help us know once we are and are usually not meeting the intended objectives for our management actions,” she says. “We also need more stories to inform stakeholders in our lands in regards to the positive work that we’re doing on the bottom. And eventually, a significant application of fuel loading is smoke estimates. If we’ve higher fuel loading measurements, we’ve higher smoke estimates.”

Moreover, AI could be brought into the conversation. With enough photos (Albee’s goal is 1,000 to start out), the Wildlands team can use the info collected from the sphere within the app to implement machine learning, and create a model that may begin to more accurately predict the consequence of surface fuels within the environment.

Within the last yr, Albee and his team were in a position to develop an AI model that may discover 1, 10, and 100 hours fuels and generate a mass estimate.

The initial model was constructed using 60 sample microplot images and the mass estimation reference images. By allowing Albee’s team to access additional real data collected by practitioners, the AI model could be improved, and incorporated into the info collection application to supply in-the-field estimates of fuel loadings. The team’s initial goal is to be inside 20% of actual lab-measured fuel loading for a microplot.

Albee and the Wildlands team want to recruit more fire managers and people excited by collecting surface fuels data to make use of this app of their surface fuels management. Their hope is that the app makes this process easier and more accurate, which could allow the practice to be implemented more continuously. Moreover, with higher adoption, more data shall be collected to feed the AI model, and that may result in greater wildfire prevention and management in the long run.

“No person knows about fires that don’t occur,” Albee said. “But prevention is vital. That is about long-term vision.”

We’re excited by feedback to assist shape this application into a very useful and time-saving tool. Please get in contact to tell us your thoughts or join to be a part of testing: wildland-fire-team@allenai.org


Please enter your comment!
Please enter your name here