AI Drones Brave Smoke Plumes to Map Wildfires Faster
AI Drones Brave Smoke Plumes to Map Wildfires Faster
.jpg?width=850&height=478&ext=.jpg)
To combat the growing threat of wildfires and the far-reaching impact of smoke pollution, researchers are deploying AI-powered drones into smoke plumes to gather critical data.
Wildfires torch vast swaths of the United States and Canada every year, straining firefighting resources, damaging wildlife habitats, endangering property and people, which are a few reasons why researchers at the University of Minnesota Twin Cities are studying using AI and drones to infiltrate smoke plumes and collect data. The end goal is to better protect against the spread of fires and to prevent the harmful effects of smoke and other pollutants.
The U.S. experiences about 60,000 to 70,000 wildfires a year, scorching 7 million acres, according to the National Interagency Fire Center (NIFC). Canada has about 8,000 wildfires a year, burning an average of 2.1 million hectares, reported the Canadian Red Cross.
Prescribed burns are needed for a variety of reasons including consuming excess fuel, also present dangers. The U.S. Government Accountability Office estimated in 2024 that 43 wildfires were generated from 50,000 prescribed burns between 2012 and 2021, another reason for the need for better smoke management tools.
“A key step is understanding [how fires spread] is the composition of smoke particles and how they disperse,” said Jiarong Hong, a professor in the University of Minnesota’s Department of Mechanical Engineering. “Smaller particles can travel farther and stay suspended longer, impacting regions far from the original fire.”
More for You: Drones Could Aid First Responders, Researchers Say
Earlier simulation tools model fire and smoke particle behavior, however, they face limitations in data collection, modeling accuracy, and field observation of smoke plumes. In “3D characterization of smoke plume dispersion using multi-view drone swarm,” recently published in Science of the Total Environment, Hong and his colleagues tackled these challenges by improving methods for accurately modeling how smoke particles—some as small as a micron— moved and spread during wildfires and prescribed burns.
Researchers unleashed a fleet of AI-driven aerial robots designed to penetrate smoke plumes from all directions, creating detailed 3D models and mapping airflow dynamics. Unlike standard drones, these machines can sense smoke and steer straight into it, collecting data from within the haze.
“The most exciting part of this project was deploying drones during real-world prescribed burns to capture smoke particles,” said Srijan Kumar Pal, an electrical engineering doctoral candidate whose research focus is robotics. “My main role involved developing an autonomous single-drone system capable of capturing smoke particles without human input, especially in unpredictable wind conditions.”
“The computer runs with AI models,” Hong said. “The AI model then tells the computer to tell the drones to recognize a certain object. In this case, smoke plumes.”
With prescribed burns, the main smoke release lasts only minutes due to careful control.
“Our swarm system must be fully operational to collect smoke particle data within this narrow window, despite unpredictable winds and high temperatures,” Pal said.
An obstacle was building a system robust enough to work in the field, which spurred the team to also implement a digital twin strategy, Hong explained.
Discover the Benefits of ASME Membership
“If you pay attention to drones [you know] drones are likely to crash,” he said. “Therefore, we created a platform. We can create a virtual environment that allows us to test a lot of algorithms, drones and flights on our computer. And then once we get that mature, we can migrate into the real world.”
The team built their own drones, fabricating the frame and rotor blades, all 3D printed. The drones cost between $1,000 and $2,000, which will not be a big factor in eventually making them marketable, he added.
“There are still a lot of technical problems we must solve. It’s not like you have a fundamental theory and you just plug it in and run. We still have to make AI able to adapt to this specific application. And through this process, we probably have to invent new technology, new AI algorithms to be able to cope with new situations,” Hong said. “For example, one of the technical problems is how do you distribute a swarm of drones to better map the smoke field? Smoke is constantly changing because of wind changes, turbulence in the wind, and burning sites are changing constantly. But you have a limited number of drones with a limited number of sensors—so how do you have a drone flying in optimal configuration to extract the most information from the smoke plumes?”
The team is now looking for state and federal funding to take the project to the next level. And Hong envisions the technology being used for multiple applications, from sandstorms to volcanic eruptions or even chemical dispersal after a crash or explosion.
“The purpose of this project is to build something everyone can use and benefit from,” Hong said.
Annemarie Mannion is a technology writer in Chicago.
The U.S. experiences about 60,000 to 70,000 wildfires a year, scorching 7 million acres, according to the National Interagency Fire Center (NIFC). Canada has about 8,000 wildfires a year, burning an average of 2.1 million hectares, reported the Canadian Red Cross.
Prescribed burns are needed for a variety of reasons including consuming excess fuel, also present dangers. The U.S. Government Accountability Office estimated in 2024 that 43 wildfires were generated from 50,000 prescribed burns between 2012 and 2021, another reason for the need for better smoke management tools.
“A key step is understanding [how fires spread] is the composition of smoke particles and how they disperse,” said Jiarong Hong, a professor in the University of Minnesota’s Department of Mechanical Engineering. “Smaller particles can travel farther and stay suspended longer, impacting regions far from the original fire.”
More for You: Drones Could Aid First Responders, Researchers Say
Earlier simulation tools model fire and smoke particle behavior, however, they face limitations in data collection, modeling accuracy, and field observation of smoke plumes. In “3D characterization of smoke plume dispersion using multi-view drone swarm,” recently published in Science of the Total Environment, Hong and his colleagues tackled these challenges by improving methods for accurately modeling how smoke particles—some as small as a micron— moved and spread during wildfires and prescribed burns.
Researchers unleashed a fleet of AI-driven aerial robots designed to penetrate smoke plumes from all directions, creating detailed 3D models and mapping airflow dynamics. Unlike standard drones, these machines can sense smoke and steer straight into it, collecting data from within the haze.
“The most exciting part of this project was deploying drones during real-world prescribed burns to capture smoke particles,” said Srijan Kumar Pal, an electrical engineering doctoral candidate whose research focus is robotics. “My main role involved developing an autonomous single-drone system capable of capturing smoke particles without human input, especially in unpredictable wind conditions.”
“The computer runs with AI models,” Hong said. “The AI model then tells the computer to tell the drones to recognize a certain object. In this case, smoke plumes.”
With prescribed burns, the main smoke release lasts only minutes due to careful control.
“Our swarm system must be fully operational to collect smoke particle data within this narrow window, despite unpredictable winds and high temperatures,” Pal said.
An obstacle was building a system robust enough to work in the field, which spurred the team to also implement a digital twin strategy, Hong explained.
Discover the Benefits of ASME Membership
“If you pay attention to drones [you know] drones are likely to crash,” he said. “Therefore, we created a platform. We can create a virtual environment that allows us to test a lot of algorithms, drones and flights on our computer. And then once we get that mature, we can migrate into the real world.”
The team built their own drones, fabricating the frame and rotor blades, all 3D printed. The drones cost between $1,000 and $2,000, which will not be a big factor in eventually making them marketable, he added.
“There are still a lot of technical problems we must solve. It’s not like you have a fundamental theory and you just plug it in and run. We still have to make AI able to adapt to this specific application. And through this process, we probably have to invent new technology, new AI algorithms to be able to cope with new situations,” Hong said. “For example, one of the technical problems is how do you distribute a swarm of drones to better map the smoke field? Smoke is constantly changing because of wind changes, turbulence in the wind, and burning sites are changing constantly. But you have a limited number of drones with a limited number of sensors—so how do you have a drone flying in optimal configuration to extract the most information from the smoke plumes?”
The team is now looking for state and federal funding to take the project to the next level. And Hong envisions the technology being used for multiple applications, from sandstorms to volcanic eruptions or even chemical dispersal after a crash or explosion.
“The purpose of this project is to build something everyone can use and benefit from,” Hong said.
Annemarie Mannion is a technology writer in Chicago.
