This story originally published in High Country News.

Over Labor Day weekend in the Pacific Northwest, high winds fanned wildfire ignitions in drought-ridden forests west of the Cascades. In a matter of hours, small fires erupted into about a dozen major blazes, destroying entire communities, displacing tens of thousands of residents and killing 10 people in Oregon and Washington.

The scale of the conflagrations, and the speed at which they grew, surprised even seasoned wildfire researchers. The scientific models used to predict and understand fires worked well in previous decades, but given current conditions across the West, trying to use them now “just sucks,” said David Saah, an environmental scientist at the University of San Francisco and a leader of the Pyregence Consortium, a team developing new wildfire models. “You know how we keep saying climate change is going to change everything? We’re there, we’re in it (and) we don’t know how to quantify it. We’re trying to figure it out.”

Fire has always been part of Western ecosystems; many animals and plants evolved to depend on periodic burns. And for thousands of years, Indigenous peoples have used fire to help keep forests healthy by reducing excess brush and encouraging new growth, a practice that continues today. But after a century of fire suppression — and with a rapidly changing climate that is drying out forests — Western wildfires are now much larger and more intense than before.

In the typically wet western Cascades, wildfires require certain conditions to grow: low humidity and powerful easterly winds. By early September, a 10-month drought had set the stage for dangerous blazes, and unusually dry and strong winds followed: Near Salem, Oregon, sensors logged the lowest combination of relative humidity and highest wind speed ever recorded at that location, said Larry O’Neill, the Oregon state climatologist. These conditions contributed to the “explosive” growth of the Santiam Fire, later renamed the Beachie Creek Fire, which has burned nearly 200,000 acres, destroyed thousands of homes and buildings in the towns of Detroit, Mill City, Gates and Santiam River, and killed five people, including Oregon environmentalist George Atiyeh. “(That) combination of conditions is essentially unheard of,” O’Neill said. But it might become more common in the future, thanks to climate change. Scientists are “very concerned” about the possibility that such rare wind events could become “more frequent or extreme,” O’Neill said.

When authorities are faced with major decisions — how to best protect homes and lives, and when to issue evacuation notices — they need to know how fast-moving, hot and severe a specific fire is likely to be, and where its perimeter might lie in the days ahead. For now, most state incident commanders and U.S. Forest Service firefighters rely on short-term wildfire models, computer-based calculations that forecast how a blaze might behave. The Rothermel surface fire spread model, developed in 1972, is the basis of many of the models used today. The basic inputs rely on knowing three main elements that drive wildfires: topography, weather and fuel flammability.

But the unusual easterly wind added an unexpected element to attempts to model the Labor Day fires, said Meg Krawchuk, a fire and landscape ecologist at Oregon State University. “Rare events are hard to model because you have so few cases to build and learn from,” she said. And wind in particular can stymie Rothermel-style models. Strong gusts can topple power lines, igniting new fires that build on the heat and vapor in the atmosphere, spawning an inferno large enough to create its own weather. When that happens, the original inputs are no longer accurate — and neither are the model’s results.

Understanding such complicated interactions requires a new kind of model. Coupled, or physics-based, models, for example, explicitly examine the interaction, or coupling, between fire behavior and the atmosphere. These models, which are expensive to run, are still being developed; right now they primarily live in “research land,” said Saah. They address the feedback cycle of what a fire consumes, how the heat released from that consumption impacts the atmosphere, how that in turn affects the weather, and how that weather then impacts the fire’s behavior. “The (Rothermel-based) models don’t capture that,” Saah said.

A third type of wildfire model, used for long-term planning and research, looks at general wildfire activity on a scale of decades rather than days. These statistical models look at long-term vegetation and climatological patterns and how they interact in a particular place, then project what wildfires will look like in 50 or 100 years. The inputs are based on historical data, and as climate change dries out the West and increases the frequency and intensity of extreme weather, wildfires will also change, making modeling them more difficult. “When you want to build large-scale systems models, you need lots and lots and lots of observations,” Saah said, something scientists don’t yet have, as climate change alters conditions in ways they haven’t seen before. “So that’s why people are freaking out.”