“People were all sucking in this superthick smoke for a little over a week and kept asking, ‘When is it going to change?’ ” says Andrew Wineke, a communications representative with Ecology. On a typical day, the department’s air quality map sees 4,000 viewers. During the smoke storm, it saw about a million.
For smoke specialists like Ecology's Dr. Ranil Dhammapala, an atmospheric scientist, predicting that kind of change is complicated — and that complicates interactions with the public.
“At the beginning of that supermassive event, initially we thought the answer was ‘a little relief tomorrow,’ ” says Dhammapala. He and his colleagues had been working from home, still managing to provide smoke forecasts against the odds. But this particular forecast didn’t quite pan out: a weak front that they’d hope would start to clear out the smoke, instead petered out, in part due to how thick the smoke was. “the And wooooh, we got taken to task for that 'tomorrow' being like a week later, because something we had counted on didn't quite materialize,” he says.
While the team members felt proud for helping educate people through the unprecedented September smoke, as well as the rest of the season, they agreed that they could improve. The team collaborated with researchers from Washington State University last year to develop a two-day automated forecast that used machine learning to accurately predict the severity of smoke levels two-thirds of the time. This offered a substantial degree of certainty for the smoke forecasting field, but Dhammapala and the team wished for a longer-term smoke forecast to accompany it.
“The more lead time people have to be aware of what's likely to come their way, the more lead time — we think, we hope — they have to protect themselves and their health,” Dhammapala says. When smoke is incoming, Seattleites can plan to purchase equipment like box fans, filters and N95 masks; reschedule activities; or even plan to evacuate.
“People are hungry for that information, and it’s our job to give it to them. So we want to use every tool at our disposal, and in this case, a tool that was not at our disposal until Ranil and Washington State University created it, to help them stay forewarned and forearmed,” says Wineke.
Less than a year later and just in time for wildfire season, Ecology has a new tool in its toolbelt. The new five-day automated smoke forecast is accessible as part of Ecology’s online smoke map, with days three through five similar to the existing two-day forecast, but using slightly different streams of information.
Users can anticipate average daily air quality conditions in 53 separate zones across the state, each pegged to real-time air quality monitors (measuring things like health-affecting ozone and fine particulate matter) and air quality management regions. When it soft-launched with the five-day forecast Tuesday, the map became the first available automated five-day smoke forecast tailored to Washington.
While designing the five-day forecast was a feat, the team still has plans to further refine it and make it useful — and with the fire season we might face, that information could be essential to millions.
Unpredictable fires
To appreciate the magnitude of a mostly accurate two-day forecast, let alone any five-day forecast, it helps to understand what goes into a smoke forecast.
Dhammapala says meteorological forecasting is imperfect but has improved substantially in the past two decades. What makes smoke such a hard thing to forecast is how fires themselves develop over a given forecast period. Wildfire intensity and duration, how high their plumes rise, chemical and physical reactions that happen because of smoke, where fires start, and the success of firefighters battling them in the field can all drastically affect the amount of smoke.
“Forecasters use anything and everything that we can look at that might have a chance of telling us anything,” says Phil Swartzendruber, an air resources specialist with the Puget Sound Clean Air Agency, which uses Ecology data.
The two-day forecast makes projections by funneling the previous day’s air quality monitor levels through a machine learning program that considers the weather conditions and topography near those air quality monitors.
Extending the forecast to five days requires running at least a dozen “what if” scenarios depicting how the interplay of fires and meteorology might play out and averaging those simulations together, Dhammapala says. The state’s five-day smoke forecast relies on a number of models, including forecasts from regional clean air agencies, accessed via the national AirNow air quality database; WSU’s smoke forecast; at least four global forecasts; Ecology’s own smoke models; and the professional judgment of state and local forecasters.
“The thing that's really nice about the tool that Ecology developed is they actually pulled together a number of these larger global models, and that gives you a really good idea of the range of possibilities,” says Swartzendruber. Being able to provide that range, rather than an explicit projection, is extremely helpful, he says.
Another advantage of using an automated system compiling all these models is that it provides some consistency for human forecasters. “It puts a number of good [models] in one place and pulls them together in a way that, at this point, I don't think there's any other ensemble of wildfire smoke models like this. … It’s an improved tool over what we had before and, as we use it, we’ll learn where its strengths and weaknesses are,” says Swartzendruber.
Ecology’s forecasters can override the five-day forecast’s machine learning prediction if they feel it’s necessary, Wineke says, and forecasters from the state’s seven regional clean air agencies can also enter their own forecasts—but overrides are rare. Usually, Dhammapala says, the program’s predictions are reasonable.
Still, most experts say the public should expect that there are limits to the accuracy of a five-day forecast.
“Having a longer forecast allows more time to inform the public of the coming smoke and gives them more time to decide how to react,” says Seth Preston of the Northwest Clean Air Agency. “But the information may not be quite as accurate due to the decreased accuracy of longer-term meteorological forecasts and uncertainty about new fire starts/fire behavior.”
Room to grow
Dhammapala says we won’t truly be able to gauge the tool’s accuracy until it can be tested and refined against real-world smoke events.
“Don't ask us how accurate it is — we don't know,” he says. “You need a good smoke event to tell that, unfortunately. And smoke is bad for everyone, so we don't really want the public breathing the smoke just to evaluate some geeky construct of ours.”
He hopes the tool will provide “reasonably reliable” five-day forecasts by late July, when wildfire season hits its stride. “I'm sure we're going to discover bugs, and I’m sure we’re going to find ways of improving it” in the interim, he says. “But we didn't want to have to test it for a year before we launched it.”
In addition to simply becoming more accurate, the tool has room to expand coverage. Right now it doesn’t cover the entire state. Areas without air quality monitors, for instance, aren’t included in the map. Often, these include wilderness areas, national parks and national forests with extreme topography. It also doesn’t take into account newly installed or temporary air quality monitors, which often don’t have the three-year minimum amount of data from which these models are built.
But even places without air monitors can still benefit from this map, says Northwest Clean Air Agency’s Preston, if they know how to access it and interpret the data.
“Smoke is pervasive — it isn’t selective. Chances are good that if there’s wildfire smoke in an area with a monitor, there’s wildfire smoke in areas without monitors,” Preston says. “So if there’s smoke at our monitor in downtown Mount Vernon, there’s a very good chance there’s going to be smoke in Burlington and Sedro-Woolley. Everyone benefits from this tool, but they have to know that it exists, where they can find it, and how to use it.”
What’s most important is how the public interprets information from a longer-term forecast. A forecast is never a guarantee: It can be a warning sign that should warrant attention, especially from those in vulnerable populations.
“This is a tool that [shows] us that the potential for something bad might be coming,” Swartzendruber says. “You have to frame this as a warning.”
“Something is better than nothing,” adds Dr. Hasan Tahat, the Yakima Regional Clean Air Agency’s compliance and engineering division supervisor. “If it helps, then we will use it. Why not?”
The summer ahead
Anticipating the summer season induces anxiety among air quality specialists. “The entire air quality program is trying to walk and chew gum at the same time,” Wineke says. “And it's a really big piece of gum, so that is stressing me out.”
“I use my lawn as a forecasting tool and it's not looking good. It's half dead already and it's not even June,” Swartzendruber says.
That’s a drastic shift from forecasting of past years, when the biggest thing that might happen from a smoke event was a pretty sunset, Swartzendruber says.
Forecasters say no long-forecast tool can really change the basic fact that, for Washingtonians, smoke in the summer is a fact of life now: Being prepared for it should be the default, not something to start thinking about a few days out.
“It's summer in the Northwest — you should really use this opportunity, when there's no panic, to be prepared,” Swartzendruber says. He compares smoke preparation to fixing a leaky roof. “When it's dry, there's no problem to fix,” he says. “But when it's raining and the roof is leaking, you can't go up on the roof and fix it because it's raining, so take advantage of when there's not a problem to make sure you're ready for when there might be.”
