Using Drones to Combat Wildfires
The Canadian province of British Columbia has experienced the worst wildfire season in recorded history. Over 2 million acres have been burned, and at its peak, 45,000 people were displaced.1 The British Columbia Wildfire Service (BC Fire) deployed every available resource, including four types of fire crews, 24 fixed-wing aircraft, 6 helicopters, and out-of-province resources.2 This year, BC Fire tested a new tool: drones.
Drones equipped with thermal infrared cameras can see hot spots in the forest that are primed for ignition, as well as areas that may have escaped containment. In some cases, roots may smolder for several days beneath the ground surface. If hot spot locations are provided to the fire crew in a timely fashion, they can cool or extinguish them to prevent further spreading. It is also critical to “clear” the area before civilians are allowed to return into an evacuation zone. Drones excel at mapping and data collection, which frees up larger wildfire aviation assets to deploy crews, move equipment, and drop flame retardants and water onto the wildfire.
Source: British Columbia Wildfire Service via Facebook
Creating a Hot Spot Identification Workflow
Raptor Maps was approached by Sarvus Unmanned Systems to design its flight operations and data processing pipeline to provide an actionable deliverable for the BC Fire crews. Sarvus is a drone service provider based in 100 Mile House, BC. They are one of only a handful of providers in the region authorized by Transport Canada and BC Fire to provide drone services for wildfire support, and their parent organization, Sarvair Helicopters, provides manned aircraft support. Sarvus was tasked by BC Fire to identify hot spots and provide the information to fire crews.
|Drone||DJI Inspire I|
|Thermal Camera||Zenmuse XT Radiometric from RMUS|
|Resolution||640 x 512|
Requirements for a Hot Spot Map Deliverable
Fast turnaround. Sarvus has a night waiver to fly and collect thermal imagery after dark. The results needed to be available to the fire crews by 7:00am the next morning to be actionable for the crews.
Maximize area covered in given time. Wildfires occur in remote locations, and travel time can often be greater than one hour. Given the limited time window, Sarvus wanted to maximize the covered area to increase the situational awareness for crews on the ground. Raptor Maps determined it could reduce the needed overlap and sidelap to 67%. This means they can cover 1.7x more area as compared to 80% overlap and sidelap, and 3.4x more area compared to 90% overlap and sidelap in a single flight.
Accuracy in identification. Reaching suspected hot spots isn’t easy. Fire crews have to make their way through thickly wooded areas while carrying heavy equipment. False positives (identifying hot spots that are not there) mean extra work for the crew, while false negatives (missing a hotspot) can lead to ignition and further spreading of the wildfire. RMUS provided Sarvus with a camera for an initial survey of hot spots in the area. Sarvus worked with the fire crews to identify examples of hotspots from the air. This was provided to Raptor Maps to tune the sensitivity of our machine learning algorithms specifically for Sarvus.
Accuracy in location. Crews need to get close enough to the hotspot to localize it. This is done with a combination of vision (smoke) and sound (crackling noises). For wildfire detection, we aimed for an accuracy of 25 meters. This accuracy can be significantly improved over flat terrain or with the addition of an accurate terrain map.
Accessible from handheld device. One of the challenges Sarvus worked through was delivering the hot spot data to the fire crews. Due to the scale of the wildfires, the provincial government brought in outside crews to assist, so not everyone was using the same handheld devices. Some of the crew used handheld GPS (e.g., Garmin), while others used their smartphones. Some had access to local base maps (to load into Avenza Maps), while others used default satellite imagery within Google Maps on their smartphones. We ultimately decided on providing the hotspot deliverables in 2 ways:
1) A .KML file containing the hot spots locations. This small file can be loaded into most GIS software, such as ArcGIS, Avenza Maps, Google Earth, and handheld GPS units. Many BC Fire personnel used the Avenza Maps app due to the ability to upload a custom basemap.
2) A Google My Maps link to the hot spot locations. This link can be emailed to any interested party and opened in a web browser (or Google Maps on a smartphone) without the end user needing to install any software.
Workflow and Results
Import Thermal Images into Raptor App
Receive .KML file and share link (< 1 MB total)
Deliverable provided by Raptor Maps. Available as .KML file and shared link to support various devices, applications, and levels of expertise.
A Successful Outcome
“Each pin took him [the fire boss] directly to a hot spot, which is great,” says Landon Sarver, President of Sarvus Unmanned Systems. For the next fire season, Landon wants to lower the threshold for hot spot identification even further to see tennis-ball sized hot spots. His conclusions from this year: “I’m happy we were able to provide them with some meaningful data.”
Remember, drone service providers flying near wildfires must do so in coordination with the fire fighting crews. Do not fly your drone near a wildfire unless you are authorized. Please see this post from the National Interagency Fire Center to learn more.