Scientists have developed a new technique to predict the behavior of wildfires, using high-resolution satellite imagery to periodically check and revise computer simulations.

The work, which was carried out in collaboration by scientists at the National Center for Atmospheric Research (NCAR) in Boulder, Colo., and the University of Maryland, appeared yesterday in an online issue of the Geophysical Research Letters.

Wildfire behavior is, as a general rule, notoriously unpredictable. A small blaze can smolder for days before flaring up in force, and once going, it tends to travel wherever the wind blows. Factor in variable terrain, changing fuel loads and fire's aggravating tendency to create its own weather systems, and you have a moving picture whose final form is maddeningly difficult to forecast.

While scientists have labored over the past several decades to model interactions between fire, weather and on-the-ground conditions, their best efforts tended to unravel after the first day or so of predictions.

"When you're trying to look forward, everything is a forecast -- our weather models, our fire behavior models, how they work together," said Janice Coen, a scientist with NCAR and lead developer of the model. "When you run your models, you may get good results for 12 hours or so, but after a while, the inaccuracy builds up."

"The question was, how do you model an event when your model's accuracy only lasts for a day or two?"

Coen's answer, it turns out, was traveling in a sun-synchronous orbit 512 miles above the Earth.

Getting help from an eye in the sky
The Suomi NPP, a weather satellite operated by the National Oceanic and Atmospheric Administration, carries a special infrared sensor capable of capturing high-resolution images anywhere on the globe. Called the Visible Infrared Imaging Radiometer Suite (VIIRS), this small device can observe the Earth's surface at a much finer scale than the average satellite sensor.

Shifting through the troves of data sent back from the VIIRS, University of Maryland professor of geography Wilfred Schroeder identified a signature pattern that denotes wildfire activity. By feeding the resulting fire image data into Coen's model every 12 hours, the two scientists were able to essentially "reboot" the system, depicting the outlines of a fire with a high degree of accuracy, even when it lasted for several weeks.

"The satellite passes over the same spot on the Earth" -- say, the location of a particular fire -- "about twice a day," Schroeder said. "So every 12 hours, we initiate the model and assimilate the new satellite data. If the data matches the model, [Coen] verifies its accuracy; if the model deviates from the data, she corrects the model to correspond."

That kind of real-time updating may be critically important to firefighters, who are constantly on guard for changing conditions.

"Fire has always been the domain of forestry, and forest managers have always [monitored] fuel impacts and [the effect of] terrain on fire. Weather has been more of an unknown, and that's the perspective we bring," Coen said. "Most people, when they look at a fire, see moving flames. But we as meteorologists see air as a fluid, and as a fluid, there are rules for how it has to behave."

Those rules, however, can be confoundingly complex. In addition to the often fickle nature of weather systems, fires -- particularly large fires -- can alter the atmospheric conditions in which they operate. Fires often drive air upward, creating a vacuum that sucks more air toward the fire front, said Coen. In extreme cases, this can cause the fire to blow back on itself, or even reverse direction.

Coen and Schroeder are currently working with national firefighting agencies to adapt their technique to the on-the-ground needs of firefighters. They hope to begin introducing a practicable model next year.

Reprinted from Climatewire with permission from Environment & Energy Publishing, LLC., 202-628-6500