Fire Characterization and Fire-Related Land Cover Classification Using Hyperion Data over Selected Alaskan Boreal Forest Fires

Abstract:

In this study, NIR and SWIR EO-1 Hyperion data acquired over two large Alaskan forest fires are used to detect active fires, map their immediate vicinity, and retrieve fire temperatures. The study sites are located in black spruce stands within the 2004 Boundary fire (215,000 ha total affected area) and the 2009 Wood River 1 fire (50,000 ha). Even though fires in the North American boreal forest ecosystem contribute greatly to global carbon cycling and large-scale air pollution, they have been less studied so far using satellite-borne imaging spectroscopy. We adapted the Hyperspectral Fire Detection Index (HFDI) so that it worked well for the high-latitude Hyperion data. This involved selecting suitable bands which best separated fire from non-fire pixels and averaging them to further improve the detection signal. Resulting fire detection maps compare favorably to uniform radiance thresholding of the Hyperion data and are consistent with fires detected on near-simultaneous Landsat 7 ETM+ data. Unsupervised classification of the vicinity of the active fire zones served to delineate 5 to 6 well separated classes: high- and low-intensity fire, various unburnt vegetation classes, recent fire scar, and a transitional zone ahead of the active fire front that shows evidence of fire impact but no emitted radiance component. Furthermore, MODTRAN5 was used for atmospheric correction to retrieve fire temperatures by modeling a mixture of emitted and reflected radiance signatures of the fire and background areas, respectively. As most of the carbon consumption and subsequent emissions in boreal forest fires stem from the combustion of dead plant material on the forest floor, estimates on fire intensities and high/low intensity burn areas provide valuable insight into the amount of carbon cycling in the system. Imaging spectroscopy can therefore contribute an important step forward in quantitative studies of boreal fires and their impacts. These techniques are set to advance further as future hyperspectral satellite missions start delivering a time series of Hyperion-like data at higher signal-to-noise levels.