Fire Forecaster

With a $600,000 NASA grant, Alaska's Natural Resources Department is using satellite imagery to improve firefighting and emergency response.'The thing about Alaska is, our maps are all old,' said Marc Lee, Fairbanks area forester for DNR's Forestry Division.The standard U.S. Geological Survey one-inch-to-the-mile quadrangle maps for the state date to the late 1950s and early 1960s with some updates from the 1970s.'We don't have any good current maps,' Lee said.Besides new maps, he wants new tools for fighting fires in the state's millions of forested acres. It's impractical to expect firefighters on the scene to calculate how fast a fire will spread with given types of fuel. So the Forestry Division is developing a detailed database of vegetation types for computer models predicting fire behavior.The division is tackling both jobs with one bird'the QuickBird satellite operated by DigitalGlobe Inc. of Longmont, Colo. QuickBird takes large-scale, high-resolution photos that show features as small as 60 centimeters across, or about 2 feet.The resolution is good enough that firefighters and emergency response teams can use geographic information system images in place of maps. If specific vegetation can be identified, the data can populate a database to fuel a wildfire model.Since May 2002, the Forestry Division has been receiving QuickBird images of the 30-million-acre Tanana Valley in central Alaska with a goal of imaging 10-mile squares around each of the valley's 15 villages.'We've probably collected about half the images,' Lee said. But acquiring them is just the first step. They must be processed to compensate for distortion. In about a year, Lee said, his team will be able to shift its attention to analyzing the vegetation.'We're not a huge staff,' Lee said. 'We've got about four people working on it.' They like pushing the boundaries of off-the-shelf image analysis, he said. 'I think in three to five years, we're going to see some good programs that will be able to pretty well project what a fire is going to do.'DigitalGlobe launched QuickBird in October 2001 and began offering commercial images the following April. The satellite circles the Earth 16 times a day in a 450-kilometer polar orbit, covering Alaska between 10:30 and 10:45 a.m. local time.'That's about the time we take all of our images anywhere in the world,' said Brett Thomassie, DigitalGlobe's director of civil government programs. It's known as high solar noon, giving the best light angle for gathering images.The satellite can take up to 57 images on the day side of each orbit, each covering an area of 16.5 kilometers, or about 10 miles, on a side. When QuickBird is not busy taking shots for paying customers, operators point it at areas of high interest for the company's archives, such as urban areas and locations in the news.How often the satellite is within range of a particular area depends on its location, Thomassie said. At higher latitudes such as Alaska, it typically revisits an area every two to four days. Closer to the equator, the revisit time is 10 to 12 days.Images are stored onboard in a 128G solid-state recorder. Each image takes up 1.2G to 1.4G, so the recorder can hold about two days' worth. The satellite downloads the data to one of two receiving terminals in Alaska and Norway, linked by dedicated fiber links to DigitalGlobe in Longmont, where images are decompressed and processed.Images can go to customers on CD-ROM, DVD or tape, or be downloaded from a File Transfer Protocol server.The 105-square-mile size and 2-foot resolution made the hard-copy images attractive to Alaska's firefighters because they could identify trails, structures and ponds that don't appear on standard maps.'The quality is so good you can map directly from it,' Lee said. And the images are accurate enough to integrate with GIS overlays. 'It makes your data sets very powerful,' he said.Drilling down to a visible image in a GIS shows information about the landscape, features, ownership and other details.'We also added power line coverage so we would know where to turn off electricity' in a crisis, Lee said.DNR uses an orthorectification software product from PCI Geomatics Enterprises Inc. of Richmond Hill, Ontario, to correct distortion in the raw images. The software sees embedded data about the lenses used, satellite location, path and angle. A second round of rectification corrects distortions caused by landscape elevation.Because of the high resolution, new analysis methods are needed to identify vegetation. Geological Survey Landsat images have a resolution of about 30 meters, and each pixel can be analyzed separately. At resolutions of about 0.6 meter, pixels must be grouped and examined as objects.'They're just starting to come out with software for high-resolution images,' Lee said. His team uses eCognition from Definiens Imaging GmbH of Germany.'It's very complex,' he said. 'We're among the first to use it for vegetation classification.'Analysis of 30-meter-resolution images is about 60 percent to 70 percent accurate. Higher-resolution images should improve the accuracy.'I'm optimistic,' Lee said. 'I think it's going to be uncanny, but we don't know yet.'The plan calls for loading vegetation data into ArcGIS from ESRI of Redlands, Calif. Then FarSight software developed by the U.S. Forest Service will work with ArcGIS to calculate a fire's behavior.If successful, the Alaska Tanana Valley program could change the way land is managed and fires are fought around the country.'People in a lot of state governments and federal agencies are watching to see how this goes,' Lee said.