Buying time: Inside the AI technology trained on B.C. wildfires

As frontline crews battled the worst wildfire season in B.C. history, a pilot project unfolded at a frantic pace at the BC Wildfire Service headquarters in Kamloops. 

Supt. of Predictive Services Neal McLoughlin and his team spent much of 2023 implementing two artificial intelligence programs to analyze the Kamloops and Coastal Fire Centres and do work typically carried out by human analysts: taking data inputs around geographical conditions, weather forecasts, drought conditions, even detailed descriptions of the type of vegetation where new fires were sparking. 

“(When done by people) a typical turn-around time would be anywhere from two to four hours for one fire simulation and you might get 14 in a day,” he said in an exclusive interview with CTV News. “Last year we were running probably 200 fires a day at minimum, and some of those fires would repeat and simulate a second time in a day when a new weather forecast came in.”

That means decision-makers don’t have to wait to assess what equipment and personnel should go where, which can make all the difference when responding to the kind of explosive wildfire growth B.C. saw last year.

As of March, the entire province has been under AI analysis at a time when the fire season has slowed due to cool wet weather in most areas. That’s allowed the team to pay close attention to a smaller number of fires in order to assess how well the parallel programs are predicting what happens when new fires start. 

The Canadian software, FireCast, and American Wildfire Analyst typically provide similar results, but sometimes they provide conflicting forecasts for where and how fast a fire can spread. For the time being they’ll run both systems to compare results.

Machine learning can save precious time

In recent years, persistent drought conditions and lightning storms have sparked hundreds of wildfires, some of which have destroyed infrastructure and homes. 

While computer modelling has been used for many years, the manpower required to forecast fire growth simply can’t keep up when there are hundreds of fires burning at the same time and limited resources to deploy. Having a digital helper that can quickly do the grunt work can lead to faster, more informed deployment.

“It takes away a lot of the busy work and it allows us to get ahead of fire,” said McLoughlin. “The whole impetus is to set the clock back so that we have more time to think about what a fire could do, think about how we're effectively going to use resources, and respond to a fire before anything's even happened.”

For the foreseeable future, BCWS will retain the machine-learning results within the organization, since they do not rely exclusively on modelling and take multiple factors into consideration before deciding how – and even if, when they’re in remote areas – to fight a wildfire. As with all modelling, the emphasis is that the projections represent what could happen, not what will certainly happen.

A case study

The AI programs crunched the numbers and conditions around Horsethief Creek wildfire in the Kootenays, which started burning in late July last year and prompted evacuation orders and restricted access to treacherous areas.

Their predictions saw the fire come dangerously close to Invermere, and that factored into BCWS’ decision to fight the fire from the eastern flank to avoid the worst-case scenario. After a robust attack over weeks, the fire was ultimately contained to a fraction of the area the modelling warned could burn without firefighting intervention.

The two systems work with very limited data sets, which is different from large language models like ChatGPT, providing more reliable results with very specific goals. It also means BCWS can begin customizing what they want to be part of the analysis, which they will slowly do in the coming years.

McLoughlin emphasized that humans are still making the call on what to do, and “this doesn't replace people but what it allows us to do is provide information to decision-makers in a more timely manner and to cover a lot more ground in the province when we have heavy fire seasons like 2023.”