Morocco, Spain and France are currently experiencing a very significant heat wave that surprises meteorologists because it arrives early. With climate change, this phenomenon is likely to recur every year, and perhaps even earlier in the season. If this year, it raises many concerns, it is because the spring has been particularly dry, vegetation is suffering from water stress, and the risk of forest fires is very high. Preventing these fires or containing them very quickly is essential and AI can play a major role.
The year 2021 has seen major forest fires all over the world: North America, especially in California, Canada, all around the Mediterranean basin (Iberian Peninsula, Algeria, Morocco, Italy, Greece, Turkey, France) but also in Russia. Siberia was the scene of deadly fires last May.
The hydric stress that trees are currently undergoing, to which is added for the most part violent winds, explains the extent of the fires we are experiencing.
Thus, if last year, fires destroyed 15,007 hectares of forest in France, almost half of this surface (8,000 ha) was burned by the fire that started on August 16 from a freeway service area in Gonfaron, in the Var department, which, pushed by a strong mistral wind, devastated the national nature reserve of the Maures plain.
Preventing forest fires with AI
Natural means of fire prevention exist: clearing brush, installing firebreaks, transplanting less flammable species… But if it is impossible to eradicate all these fires, which are often caused by human negligence (poorly extinguished cigarette butts, sparks from construction work, burning…), detecting them, monitoring their progress to contain them more quickly is possible thanks to AI, and many countries, such as Morocco, have resorted to it.
AI provides a processing of data collected by sensors, cameras, drones, satellite images, among others, faster and more accurate, which can save the flora but also human lives, wildlife …
Several works have been completed or are underway in this area:
ALERTWildfire
ALERTWildfire is an extension of the ALERTTahoe network, which was a pilot program deploying PTZ cameras and microwave networks in the area surrounding Lake Tahoe, located in the mountains between Nevada and California.
This consortium of the University of Nevada, Reno, the University of California, San Diego, and the University of Oregon provides fire cameras and tools to help firefighters and first responders to :
- discover, locate and confirm fire ignition
- rapidly increase or decrease fire resources,
- monitor fire behavior during containment,
- assist evacuations through improved situational awareness,
- observe contained fires for outbreaks.
A different strategy was adopted in early 2018 that involves installing the cameras on existing third-party wireless networks, to build larger virtual networks, produce regional coverage, and thus do it much faster. In this model, “opportunity towers” (those of wireless internet service providers, state and county departments, and other private point-to-point communications infrastructure) are equipped with fire cameras and associated equipment. Data from these confederated networks is seamlessly integrated into NSL’s back-end acquisition systems and presented on the Amazon Web Services website.
The network currently has more than 1,000 cameras, mostly in California, many of which incorporate AI as an additional analysis tool. They provide invaluable information for differentiating fog from smoke in wildfires, especially in hot, dry, windy weather. AI can scan images and detect differences in images very quickly, much faster than humans can.
FireScout
The AI-based wildfire detection SaaS solution, FireScout, was developed by Alchera, a leading AI technology provider. It detects smoke from a fire using a network of surveillance cameras that operates 24 hours a day, with nighttime detection possible using infrared cameras.
In 2021, Pacific Gas and Electric Company collaborated with ALERTWildfire to install 138 new HD cameras in high fire risk districts, 46 of which, WildfireLIVE cameras, designed to protect PG&E’s electrical transmission assets, are part of an AI testing program in partnership with Alchera and ALERTWildfire.
Syntecsys
Syntecsys, a Brazilian startup and commercial agricultural technology company, and Omdena, a platform that aims to bring the AI ecosystem together around current problems, have partnered to build a system that identifies smoke and flames with more than 95 percent accuracy. For this machine learning-based project, Omdena assembled a diverse team of 47 data scientists from 22 countries who collaborated for eight weeks with Sintecsys’ AI experts.
Syntecsys’ commercialized solution, because of its accuracy, significantly reduces false positives and fire rescue arrival time. Sintecsys and Omdena are working on a second project that will enable night-time smoke and fire detection. Currently, cameras equipped with this solution are monitoring, from towers, 4 million ha mainly planted with sugarcane and eucalyptus in Brazil, where fires are a daily occurrence.
Bee2FireDetection
In 2019, The Weather Company, a subsidiary of IBM’s Watson & Cloud Platform business unit partnered with Compta Emerging Business to design Bee2FireDetection, a solution that aims to better predict wildfires using IBM Watson AI, IoT data and The Weather Company’s humidity, wind speed and local weather data. This allows firefighters to be on alert during high-risk periods.
The system can be integrated with HD cameras, including existing surveillance systems, to automatically detect smoke or other signs of ignition with IBM Watson Visual Recognition technology. In late 2021, Bee2FireDetection was selected by the rePLANT project to protect the Portuguese forest.
The NVIDIA-Lockheed Martin labcom
In November 2021, NVIDIA and Lockheed Martin partnered with the U.S. Department of Agriculture Forest Service and the Colorado Division of Fire Prevention and Control (DFPC) to fight wildfires using AI and digital twin simulation.
The two companies are building the world’s first AI-centric lab dedicated to wildfire prediction and response. It will use NVIDIA AI infrastructure and the NVIDIA Omniverse advanced visualization and virtual world simulation platform to process the size of a fire and predict its progression. By recreating the fire in a physically accurate digital twin, the system will be able to suggest the most appropriate actions to deal with it.
Lockheed Martin’s Mission Managere System (CMM) platform will combine real-time fire sensor data with other data sources (vegetation type, topography, wind direction and strength…) to predict fire spread.
Translated from Quel rôle l’intelligence artificielle joue-t-elle dans la lutte contre les incendies de forêt ?
