From outside, wildfire smoke may look like a grey cloud wafting. However, for scientists, these plumes are dynamic, complex and potentially dangerous. They can grow hundreds of miles and affect air quality, visibility and public health. Until now, capturing accurate data on how these smoke particles move and behave has been one of the most challenging tasks in atmospheric science.
Researchers at the University of Minnesota Twin City have developed a groundbreaking method for observing and analyzing wildfire smoke. This is a swarm of AI-powered aviation robots that can detect, track and build 3D models of smoke drops.
Unlike traditional drones, these small flying machines act as teams. They recognize smoke and fly directly there to take high-resolution images from multiple angles. Their mission is to help them understand better how smoke moves. This is an understanding that can reconstruct ways of predicting air pollution and responding to environmental hazards.
Published in the peer-reviewed journal Science of the Total Environment, this new study opens the door to more accurate fire behavior models and better air quality predictions, not just wildfires, but also prescription burns, volcanic eruptions, sandstorms and other particle-driven events.
The growing crisis meets high-tech tools
Between 2012 and 2021, approximately 50,000 prescription burns were committed in the United States. This is an intentional fire set under controlled conditions to improve forest health and reduce the risk of wildfires. But even controlled burns carry risk. According to a 2024 report by the Associated Press, 43 of these burns went out of control and became wildfires.
These numbers are profoundly problematic, while still small in proportion. This is because smoke particles, especially small, can stay in the air for days and travel far from the source. “The key step is to understand the composition of smoke particles and how they are dispersed,” explained jiarong Hong, professor of mechanical engineering at the University of Minnesota and a senior author of the study. “Small particles move farther away and remain suspended longer, affecting areas farther away from the original fire.”
Understanding how these plumes evolve over time is essential for early hazard detection, public health responses, and emergency planning. However, traditional tools for studying smoke, such as satellites, remote sensing, and riders, are shortened. These tools do not have the details and flexibility needed to capture the rapidly changing flow of smoke, especially in rough terrain and remote areas.
Related Stories
That's where a new pack of drones intervene. These AI-enabled robots are designed to adapt to the size and shape of the smoke. They collect rich data in real time. Existing technologies are not affordable or efficient.
The science behind the flock
The team's drone system includes one manager drone and four work card loans. These drones are not just fantasy cameras, but also mini-laborators in the sky.
Each drone is equipped with a 12 megapixel camera mounted on a three-axis gimbal to catch the smoke of movement. They are powered by a long-term 6000 MAH battery, leading to sophisticated flight controllers and Nvidia Jetson processors. These processors allow drones to recognize smoke in real time, adjust paths, and capture the best angles for imaging.
When launched, the drone works together to fly around the smoke plume, snapping high-resolution images from multiple directions. These images are grouped by time intervals and fed to the computer model using what is called a neural radiance field (NERF). This advanced AI model helps to transform 2D images into realistic and detailed 3D reconstructions of smoke plumes.
This step is important. 3D models allow researchers to analyze smoke shape, direction, and flow over time. It provides important data such as volume, angle of motion, and distribution speed. This is important to improve fire and smoke simulation tools.
Other cutting edge AI technologies such as Dynamic NERF (D-NERF) and RodynRF have been taken into consideration. However, these systems wrestlew with smokeless subjects and require long training times. The drone herd approach avoids these problems by directly capturing data within the field.
“This approach allows for high-resolution data collection in a wider area where costs less than satellite-based tools,” said Nikil Nrishnakumar, the study's first author and graduate researcher at the Minnesotaro Robotics Institute.
From research to real world impact
The drone herd has already been tested in field deployments, showing promising results. This system allows teams to generate multiple 3D reconstructions over time, creating a time-lapse view of how the smoke plume changes in real time. It's like watching plumes evolve in 3D. It is a powerful tool for scientists and emergency responders.
However, the advantages of this technology go far beyond wildfire science.
“Early identification is important,” Hon emphasized. “The sooner you can see the fire, the faster you can respond.”
Drones can also be used in other dangerous scenarios, such as volcanic eruptions, dust storms, and even urban pollution events. The system is modular and cost-effective, allowing it to scale up and down based on the size of the area being investigated. This flexibility makes it a powerful candidate for use by government agencies, environmental researchers, and emergency crews.
The next step for the team will involve making the system more autonomous and scalable. They are now integrating fixed wing drones with vertical takeoff and landing (VTOL) capabilities. These new drones can fly longer distances for more than an hour at a time, and do not require a runway to take off. This opens doors to monitor vast forests and hard-to-reach areas.
Additionally, the team will explore digital inline holography to improve particle characterization. This method can provide even more insight into what particles exist in the smoke plume and how they interact with the environment.
“We're not just building tools,” says Nrishnakumar. “We lay the foundation for a smarter, faster, and safer response to environmental hazards.”
A new era of smoke science
Many modern simulation tools, such as Firetec and QUIC fires, already exist to model how fires spread and how smoke particles behave. These systems use complex inputs ranging from fuel type and humidity to wind speed and terrain. But even the best models have one major limitation. Actual data is required to verify the forecast.
That's why a bunch of drones is extremely important. Provide missing pieces. These simulations are real, time-sensitive, high-resolution data that can make them more accurate and convenient.
Until now, simulation models have struggled to work in areas without detailed 3D maps of vegetation and topography. Furthermore, predictions could not be compared to actual smoke movements, especially in complex or rapidly changing environments. Drone herds change by creating accurate 3D ground truth models that can be used for comparison and improvement.
As the climate warms and wildfires rise, these tools may become essential to protecting ecosystems and human health. The study did not come at a better time as more than 40% of the US population live in areas that are prone to wildfires.
The project was supported by the National Science Foundation's leading research instrument programme and was carried out with the help of the St. Anthony Falls Institute. In addition to Hong and Nrishnakumar, the research team included Shashank Sharma and Srijan Kumar Pal from the Minnesota Robotics Institute.
