There is no single silver-bullet solution for the early detection of wildfires, according to leading meteorological experts. A toolbox of technologies including drones, ground-based sensors, satellites and AI models needs to be used to predict and monitor fires to save lives and protect infrastructure.
Early wildfire detection is critical. “Extinguishing a wildfire in the first five minutes requires a bucket of water,” comments Grigoris Konstantellos, mayor of Vari-Voula- Vouliagmeni in Attica, Greece. “After 15 minutes it requires two fire trucks. After 30 minutes you need 10 airplanes to fight something uncontrollable.”
In five years Attica has lost 38% of its forests to wildfires aided by heat and drought and abetted by intense winds. The coastal municipality of Vari-Voula-Vouliagmeni is surrounded by pine forests parched to tinder by heat waves of increasing frequency.
“This is climate change,” adds Konstantellos. “We used to get maybe two days of 40°C heat. Now we have two weeks and see wildfires even in the winter. We’re slowly transforming green, livable areas north of the Mediterranean into desert.”
His administration has responded by developing drone-based early wildfire detection. The drones are equipped with digital cameras, thermal cameras with lasers to capture spot temperatures, transponders for identification and even 200m directional microphones so that warnings can be communicated to people in restricted forests during high-risk periods.
“In the summer, one drone patrols the city boundaries 24/7, with a second on standby for emergency situations,” Konstantellos explains. “Each drone is controlled remotely by a certified operator at our operations center and sends back real-time imagery. We obtained a license to fly remotely anywhere in the city, up to a 1,000ft [305m] operational ceiling.”
The system was conceived after a fire nearly engulfed the municipality in 2022. Fanned by severe winds, it quickly climbed a neighboring mountain and entered a ravine, which became a savage funnel of concentrated flame. Only when winds died down at sunset could firefighters master the flames.
“Initially they chased the fire rather than cutting in front,” Konstantellos recalls. “We fought the fire on the mountain. We lacked the situational awareness to make effective decisions. We ran behind the extinguishing forces to provide water, but they had moved elsewhere.”
The drone system was developed with Vanguard, a security company using drones to deter theft of copper power lines from railways. Previously two fixed cameras at panoramic vantage points had been used to detect smoke, but were little help in fighting a fire. Now Konstantellos can direct the drones and view real-time footage on his iPhone as an emergency situation unfolds.
In 2020, Californian wildfires burned 1,780,000ha, destroyed 10,000 structures and caused over US$12bn in damage. The California Department of Forestry and Fire Protection (Cal Fire) is evaluating detection technologies, but sees limited scope for drones.
“There is no determined need or immediately applicable solution for using unmanned aerial systems (UAS) in wildfire detection,” explains Marcus Hernandez, deputy chief of wildfire technology research and development. “Technologies are being developed to allow crewed and uncrewed aviation to work harmoniously. Cal Fire uses UAS for some operations, but only incorporates additional technologies that are proved safe and reliable.”
In Mendocino County, Cal Fire is testing networks of wireless ground-based sensors with infrared, gas and particulate detection capabilities. The trials will establish whether these sensors offer sufficiently reliable power and connectivity for dependable wildfire detection. Cal Fire sees promise in AlertCalifornia, a statewide network of more than 1,000 elevated pan-tilt-zoom cameras operated by the University of California, San Diego.
“The AlertCalifornia network has detected fires before they were reported via the 911 system,” says Hernandez. “The technology detects anomalies in the captured video that may indicate the presence of smoke. It enables a professional to verify what is detected, which reduces false detections. We’re assisting in training an AI model for automated detection of anomalies.”
NOAA provides actionable wildfire information from satellite imagery. GOES-R geostationary satellites 35,405km from Earth measure mid-wave infrared signals at moderate spatial resolution to capture diffuse heat from fires.
“Geostationary satellites are constant sentinels,” says Mike Pavolonis, wildland fire program manager at the NOAA Satellite Service (NESDIS). “They enable detection of house-sized fires within 5 to 30 minutes of ignition. AI is a force-multiplier that can mimic a human expert checking imagery. We have a prototype real-time AI system to detect fires and issue alerts.”
But geostationary satellites cannot see fires concealed by cloud, and their coverage deteriorates over northerly latitudes. To see fires in Alaska, NOAA relies on low-Earth-orbiting satellites that overpass only periodically. Ground-based technologies have a complementary role and must become interoperable with space-based detection.
“Every technology has a detection latency, depending on conditions,” Pavolonis says. “No single technology will address early detection completely. California’s ground-based tower cameras have enabled some very early detections. Drones face complex logistical hurdles. It’s unclear where you’d position them, and the concept of operations is still evolving.”
“The challenge is how we routinely operate drones in a safe and automated manner,” adds NOAA senior scientist Dave Turner. “Aviation authorities are not just concerned with mid-air collisions. If a drone unexpectedly comes down on a highway it could cause an accident. The western US is vast, but drones only see a relatively small area, depending on altitude.”
NASA’s FireSense program aims to deliver operational technologies to partners tasked with managing wildfires. They address not just detection but also monitoring pre- and post-fire environments. FireSense will test novel sensors in flight campaigns with manned aircraft, typically over prescribed fires set to manage fire risk.
“We’re flying radars and lidars to map fuel, hyperspectral sensors to observe fuel composition and synthetic aperture radars and L-band microwave radiometers to detect fuel moisture,” explains FireSense program manager Michael Falkowski. “We’re flying sensors to look at post-fire impacts. We’re also pushing toward 24/7 detection and tracking with thermal sensors.” This means taking an instrument such as NASA’s spaceborne MASTER multispectral imager and reducing its form factor, weight and power requirements to enable more agile deployments.
“If we shrink those sensors we could potentially put them on high-altitude platform stations (HAPS), which hover in the stratosphere for 30 to 60 days, or on small satellites in low Earth orbit,” comments Falkowski. “In five years, privately owned constellations of 50 smallsats the size of mini refrigerators in low Earth orbit could provide 12-minute wildfire detection.” The discontinuous snapshots of low Earth orbiters could thus be improved by having many passing over in a continuous carousel.
If no single technology can solve early detection, University of California, Davis scientists are maturing ambitious plans for an integrated system using AI models, sensors, drones and cameras. It would predict areas of high wildfire risk to target drone-based detection.
“Factors that cause wildfire risk are temperature, humidity, and wind speed and direction,” says Prof. Anthony Wexler, director of the Air Quality Research Center at the University of California, Davis. “California’s complex topography means these conditions can vary wildly between valleys. Because conventional wind sensors are too expensive to put in every valley, we’ve developed US$25 prototype sensors with a 2D strain gauge to measure winds.”
A statewide network of palm-sized temperature, humidity and wind sensors could identify wildfire-conducive conditions and feed a predictive model envisaged by Wexler’s colleague, associate professor Zhaodan Kong. “We could fuse that sensor data, historic fire data and geographical understanding of where roads, campsites or power lines create a wildland-urban interface,” says Kong. “We detect fires reactively today, but a machine learning algorithm could proactively identify high-risk areas.”
“Fundamentally, two things cause wildfires – lightning and humans,” adds Wexler. “That’s it. Human infrastructure adjacent to forest automatically elevates the risk of wildfires that could damage human habitations.”
Drones equipped with chemical sensors would then be dispatched to at-risk locations to detect fires. Wexler argues that we often smell a fire before seeing it, and that wildfires producing visible smoke are already well advanced. Therefore, chemical sensors offer better odds than cameras of catching them in their infancy.
“Think of the drone with chemical sensors as a dog,” says Kong. “It sniffs out and tracks the fire to its source, then switches to a camera to monitor it. We would need a swarm to cover larger areas. Our current octocopter rotorcraft can only fly for 30 minutes, but a hybrid vehicle that takes off like a rotorcraft then flies in a fixed-wing configuration could offer greater endurance.”
There are some daunting obstacles, with little precedent for drones using their own sensors to decide where to fly. Currently Kong must obtain FAA approval for individual flights and each drone requires a dedicated pilot. This makes automated swarms a distant prospect, and a mission-tailored hybrid aircraft has yet to be built. “We have three drones equipped with the sensor package,” says Kong. “We’ve collected data around prescribed burns conducted by Cal Fire, to build a model of how chemicals from a fire propagate temporally and spatially. By next summer we aim to demonstrate that a drone can track a plume to its source.”
Saving lives in Greece
At 5:00am on August 26, 2023, Konstantellos was woken by a call from the operations center advising him to check his camera feed. Beset by a rare and deadly dry thunderstorm, Vari-Voula- Vouliagmeni recorded 270 lightning strikes exceeding 1.5MV and six simultaneous wildfires. One drone provided situational awareness while a second patrolled in case new fires broke out.
“We extinguished six fires in 40 minutes,” explains Konstantellos. “It was a textbook operation, wholly coordinated through the drones. We could see all six fires and the system directed the fire service where to proceed. It helped in critical decisions that could have affected lives. Sometimes those decisions are dressed with a lot of luck.”
The system proved its worth and continues to evolve. Satellite guidance may soon allow drones to fly remotely and transmit imagery even behind mountains. New batteries could add 30 minutes of endurance. Vanguard has partnered with Intercom in developing an AI application to screen drone imagery and issue automatic alerts.
“We plan to have three drones flying continuously at 200m, tethered by a cable that provides continuous power and data transmission,” says Konstantellos. “They can also be unhooked, fly and check, and then return.”
Last year drones were twice scrambled to assist neighboring municipalities in combating wildfires and Konstantellos believes the system could be rolled out nationwide. Europe has faced crises before, but whereas wars eventually end, climate change is forever. “Everyone on the scale of authority must do something proactive,” he says. “In Greece we saw a break in the bond of trust between citizens and the state. If you don’t feel secure for yourself, your family and your property, we move to another scenario – one of criminality and social unrest.”
Predicting short-term wildfire evolution at high resolution
The National Oceanic and Atmospheric Administration is deploying radar, lidar and infrared instruments across the western US to make long-term measurements of fire weather and determine how a wildfire interacts with the boundary layer.
“Thermal conditions, humidity and wind determine a fire’s growth,” says NOAA senior scientist Dave Turner. “If temperatures near the ground are colder than aloft, smoke becomes trapped in valleys, creating negative health effects. The fire itself modifies the boundary layer, so conditions upwind and downwind can look quite different.”
NOAA will deploy sensors at sites in Arizona, California, Colorado and Idaho to capture these interactions close to active fires in four distinct climatological regimes. The sensors include radar wind profilers, ceilometers and infrared spectrometers. “We’re using both active and passive sensors,” says Turner. “We use instruments similar to the infrared atmospheric sounding interferometer (IASI) on European Met-Op satellites to measure temperature and humidity. It’s like one pixel from IASI on the ground looking upward.”
Data collection will begin in spring 2025 and will support ambitions for an on-demand, operational National Weather Service model to aid decision makers by predicting short-term wildfire evolution. Since fire behavior depends on complex and localized terrain, this may require 100m spatial resolution. “Doubling resolution makes a model eight times more computationally expensive,” says Turner. “Going from a 3km to a 100m grid-spacing costs thousands more, so we make it affordable by using a small, perhaps a 50km by 50km, domain. As winds quickly move through that, so a second model must provide boundary conditions.”
The 3km High-Resolution Rapid Refresh model, which NOAA restarts with new observations every hour to produce 48-hour forecasts, would provide boundary conditions to drive the 100m wildfire mode. Improving its accuracy is a central aim of Turner’s measurements, he notes.
“We will collect data for up to 10 years,” adds Turner. “Of course that depends on instrument health and continued funding, but it’s hard to separate weather from climate events like El Niño and La Niña without a long data record.”
NCAR study on simultaneous wildfire frequency
The National Center for Atmospheric Research (NCAR) has found that simultaneous outbreaks of large wildfires will become more frequent in the western USA this century as the climate warms – putting major strains on efforts to fight fires.
The new study focused on wildfires of 404ha or larger. It found that wildfire seasons in which several such blazes burn concurrently will become more common, with the most severe seasons becoming at least twice as frequent by the end of this century.
“Higher temperatures and drier conditions will greatly increase the risk of simultaneous wildfires throughout the west,” says Seth McGinnis, NCAR scientist and the lead author of the study. “The worst seasons for simultaneous fires are the ones that are going to increase the most in the future.”
McGinnis says that decision makers can take steps to manage the future risk and impact of simultaneous fires. These range from thinning forests and conducting prescribed burns to increasing firefighting crews and equipment.
This article originally appeared in the January 2024 issue of Meteorological Technology International. To view the magazine in full, click here.