On the heels of one of the most destructive wildfire seasons in California’s history, the state is once again beset by massive blazes that have burned down homes in some communities and threatened others. The Ranch firepart of the Mendocino Complex fire currently burning in northern California—is the largest wildfire in state history, having consumed more than 300,000 acres. It bested the record set just eight months ago by the Thomas fire near Los Angeles.

In the face of such threats—exacerbated in part by climate change and development in fire-prone areas—experts want to improve predictions of wildfire behavior to make better use of limited firefighting resources. To do that, they need observations of how that behavior is shaped by fire weather. This includes the atmospheric conditions around the fire and the microscale weather the fire itself can generate, such as the huge, rotating plume unleashed by the Carr fire near Redding, Calif.

Craig Clements is at the forefront of efforts to gather information about fire weather to plug into models that simulate the fire–atmosphere interactions. In 2013 the San Jose State University meteorologist launched the Rapid Deployments to Wildfires Experiment (RaDFIRE) program (funded by the National Science Foundation), which deployed Doppler radars, LiDAR (which uses light pulses to generate three-dimensional images) and sensors on airplanes to gather information on winds and moisture around fires. Clements likens the effort to the dawn of tornado chasing in the 1980s. This year his team has been helping sample smoke plumes from fires around the state, and he hopes to soon launch RaDFIRE 2, a more ambitious successor to the original program, which ended in 2016.

Scientific American spoke with Clements about his efforts to capture fire weather to better understand wildfire behavior.

[An edited transcript of the interview follows.]

What are some the ways wildfires interact with the atmosphere?
What we call the actual fire front is the active burning zone. That’s the hot gases and the combustion and all the chemical reactions. The plume is actually the smoke, and that’s the hot gases that are rising. And this goes really, really high because the fire is really, really hot. The bigger the plume is, the bigger the fire is, generally. As the fire starts growing, it can start inducing its own wind. But more important is the ambient wind that is near the fire—so if it’s blowing 10 miles per hour, that’s not that big a deal; if it’s blowing 20 mph, that’s a big deal. When that happens, fires spread rapidly. And when they start spreading more rapidly, they start burning hotter and burning through more material, which puts more heat into the air, which affects the atmospheric boundary layer [the lowest layer of the atmosphere]. We don’t really know how this layer evolves around a fire. We know that the plume punches through the boundary layer. And that can actually have an effect on the fire’s dynamics because you can sometimes cause more acceleration of the fire once it breaks through. These things have been studied in numerical simulations and also in laboratory experiments, but we really want to capture these in the real world.

Why are real-world observations so important?
We have a tool that predicts fire spread and heat release, puts it into a model which generates a simulation of how the fire and atmosphere are coupled. How do we test that? How do we know if it’s right? We have a small experimental grass fire that we can simulate with these different models, and we can see that the model works for either the right or wrong reason. The next phase is—okay, let’s apply that model to the real environment: Let’s bring in the fuels data; let’s get all the terrain in there; and we need weather data. The idea is that now we can go to the fire and we can scan the plume as it’s occurring. My dream is that we would be able to get instantaneous wind data around the fire incident that would be fed through satellite into a fire behavior model and then that model would update with realistic wind profiles and maybe even a sounding [data that show how wind, temperature and other properties change with height in the atmosphere] at the site. Then you actually have something that can forecast a little bit better.

Why has this data been lacking to date?
Most of the research has been focused on fire danger on the larger scale: What are the fuel conditions? How is fire danger going to play out? When there is a large wildfire, they bring in the fire behavior analyst and the incident meteorologist—and that meteorologist is there to forecast the conditions.* But again, that’s kind of the larger-scale weather, not really the fire-induced air circulation. It doesn’t help us understand how fires introduce their own weather. But it’s really hard to observe that unless you have a network of remote-sensing tools that are scanning around the fire.

What are some examples of the observations you are getting when you go out to wildfires with these tools?
We were able to look at fire-induced air circulations at the base of the plume; we were able to look at surrounding air being pulled into plumes—which has never been done, and it’s really important for smoke modeling. And we were then able to detect the pyrocumulus cloud condensation level—where water condenses out in a wildfire plume—in multiple plumes and determine some likely explanations of why that occurs. For example: Where’s the moisture coming from? Is it moisture produced by the fire—which is one theory. Or is it just ambient entrainment of moisture? We kind of show that it’s most likely ambient moisture. But we only have a couple of samples, we just have a few observations. We have shown, though, that we can collect these observations on wildfires and what it takes to do that.

When will we have enough of these observations to validate these models and better understand wildfire weather and behavior?
I think in the next five years. But that is based on the fact that I believe that we are going to be able to get our RaDFIRE 2 campaign done. I believe our radar tool is really going to be groundbreaking, and we’re going to get on more fires and get these observations. But it’s also because the models are getting really good, and they have been validated. And I think once we validate a couple more model simulations on active wildfires, we’ll be able to use those models throughout a wildfire’s lifetime. It could be as easy as pulling up a forecast on your iPhone.

*Editor’s Note (8/17/18): This sentence was edited after posting to correct the terminology.