Assessment of post-fire riparian vegetation using hyperspectral imagery

S.K. Richardson and G.W. Minshall. Department of Biological Sciences, Idaho State University, Pocatello, ID, 83209

Remote sensing imagery is a cost-effective tool for studying large scale phenomena, such as wildfires, in remotely located regions. Hyperspectral imagery has the advantage of distinguishing among different species of vegetation due to the unique spectral reflectance of each species. High spatial resolution (5m) hyperspectral imagery was acquired to examine regrowth of riparian vegetation along several streams in the Big Creek drainage of the Frank Church River of No Return Wilderness that burned in summer 2000. Riparian vegetation endmembers were selected automatically in ENVI after running a Minimum Noise Fraction (MNF) transformation and a Pixel Purity Index, as well as manually with the aid of GPS points collected in the field. Vegetation was mapped using Mixture Tune Matched Filtering (MTMF) algorithm. We quantitatively mapped several riparian plant species and determined the relative species composition within 5m, 20m, and 30m riparian buffer zones. Using strictly the endmembers, Cliff Creek, which burned most severely in 2000, had the lowest success of classified vegetation (12%) while Goat Creek, which did not burn in 2000, had the highest success of classification (56%). Using spectral unmixing we were able to further quantify species abundance at the subpixel level.