Canada is currently facing its worst wildfire season in recent memory, with blazes burning in nearly all provinces and territories.
Last week, we saw the air quality index in Ontario swell to unprecedented levels, blanketing skies in the province and the north-east region of the U.S. with a yellowish haze and heavy smog qualified as very risky to people’s health.
Prevention and early detection are key in extreme weather events, explained Rejean Bourgault, public sector lead for Amazon Web Services (AWS) Canada, as he outlined the numerous opportunities for the use of artificial intelligence (AI) to detect the locations most at risk for wildfires and to mitigate risks.
Earlier this year, the World Economic Forum (WEF) also released a report urging the increased use of AI and machine learning (ML) in predicting and directing responses to wildfires and managing forests.
“Wildfires represent a growing environmental problem,” said David Thogmartin, machine learning fellow at the WEF. “Intervention measures are more effective the earlier they are taken – saving lives, livelihoods, habitats, as well as the cost of the measures themselves. While a significant modelling challenge, better prediction of fire outbreak and spread – using technologies such as AI, cloud and digital twins – has the potential to upgrade resource planning and save firefighters valuable time.”
In May 2022, U.S. electricity company San Diego Gas & Electric deployed a fleet of drones to fly over about 75,000 of the utility’s 240,000 power poles – those in areas at highest risk of wildfire – to take images and identify equipment potentially in need of repair. With over 2.3 million drone shots, the company, alongside Accenture and Kaminsky, built machine-learning models running on AWS cloud to identify the most critical and vulnerable part of the system and then anticipate the most urgent fixes.
This technology saw the light of day after California recorded its largest wildfire season in the state’s history.
Australia also saw a devastating 2020-2021 bushfire season, a year after which, AWS lent its infrastructure to exci, another platform that employs deep machine learning to detect and predict wildfire locations by aggregating images collected via satellite imagery.
Governments and utilities in the western U.S. and Australia have also been using FireScout, a cloud-based machine vision AI that detects wildfires 24/7 in real-time, including during nighttime, with over a thousand fire cameras.
“FireScout, which boasts 99 per cent accuracy, employs its algorithm to identify wildfires within the early ‘golden time’ of 10 to 20 minutes, allowing for response before small ignitions become big wildfires,” the company said in a recent release. “AI continually learns smoke identification methods and improves its capabilities from photographs in the database and previously discovered fire images using deep learning.”
NASA’s Fire Information for Resource Management (FIRMS) is using low-orbit satellites to provide live fire data, reportedly with a lag of just 60 seconds between Earth observation and wildfire detection.
Whether the Canadian government and organizations will ramp up AI efforts following the grim wildfire season remains a question.
AWS confirmed that it continues to have conversations with organizations across the country about how AI technology can help combat wildfires, but said it cannot share customer names without their permission.
Bourgault however, maintains that Canada is not falling behind in its wildfire response and management. Deployment, he said, does not take very long. Taking critical data securely out of Ukraine following the early signs of invasion, for instance, took “only a few days.” He adds that AWS released a contact center in 48 hours to support citizens with the Covid-19 emergency fund in April, shortly after the pandemic hit.
Canada also touts its $170 million WildFireSat project that will enable wildfire managers on the ground to make near-real time decisions based on data captured by satellites.
“This is the very first time ever that a country is producing a dedicated fire monitoring satellite, an operational one,” Josh Johnston, primary investigator for the project, told the Toronto Star. “There have been scientific missions that have gone out, but they could only do a small amount of work. We’re talking about an operational system that is dedicated to supporting fire managers. It’s never been done.”
Bourgault says that the WildFireSat constellation, unlike existing satellites, will enable better coverage of the peak times when the fires are the most active, which is generally between about 3 pm and 6 pm. This project, however, is only slated to go live in 2029.
In the meantime, Natural Resources Canada is using satellite imagery stored on AWS Cloud, to power interactive maps like the Canadian Wildland Fire Information System (CWFIS) and the Canadian Forest Fire Danger Rating System (CFFDRS).
These systems seek to give forest managers a range of tools for assessing fire danger, predicting fires and responding as necessary.
Yet, WEF said, still more can be done to leverage AI in fighting wildfires. It outlined advances in the technology that could help, including:
- Developments in NLP to improve computers’ capacity to make predictions
- Research on autonomous vehicles has driven continuous refinements in computer vision AI, which can be applied to the science of wildfire prediction
- Sensors with inbuilt AI that can operate as smoke detectors and alert authorities within the first hour of a wildfire starting
