Using Discriminant Analysis to Examine Spectral Differences Among Four Tundra Vegetation Communities at Ivotuk, Alaska

Thursday, 18 December 2014
Sara Bratsch, University of Virginia Main Campus, Charlottesville, VA, United States and Howard E Epstein, University of Virginia Main Campus, Environmental Sciences, Charlottesville, VA, United States
Warming in the Arctic has resulted in changes in the distribution and composition of tundra vegetation in addition to a lengthening of the growing season. Past studies have mapped tundra vegetation at relatively coarse spatial scales; however, vegetation changes in the Arctic are occurring at spatial scales within a few meters. This research uses hyperspectral remote sensing data to differentiate among four vegetation communities at Ivotuk, Alaska (68.49°N, 155.74°W). Ivotuk is located on the North Slope, and is dominated by four plant communities including moist acidic tundra (MAT), moist nonacidic tundra (MNT), mossy tussock tundra (MT), and shrub tundra (ST). Hand-held hyperspectral data were collected during the 1999 growing season (5 June-27 August) at biweekly intervals using narrow, ~1.42 nm wavebands. Only wavebands within 400-1060 nm were used in analysis. Two sets of comparisons were conducted using stepwise discriminant analysis: 1) MNT and ST, and 2) all four tundra plant communities. MNT and ST classification accuracy ranged from 91.3-100%, with 100% classification and cross-validated accuracy occurring on 27 July. Classification accuracy for the overall growing season was 97.9% for MNT and 98.1% for ST. The stepwise function indicated 18 significant bands including 8 near infrared (NIR) and 4 blue bands. MAT, MNT, MT, and ST classification accuracy ranged from 58-100%, with greatest classification accuracy (100%) also occurring during peak growing season on 27 July. Overall classification accuracy for the growing season was 97.6% for MNT, 92% for MT, 84% for ST, and 70.8% for MAT. There were 14 significant bands including 6 NIR and 3 blue bands. The results presented here demonstrate that discriminant analysis can be useful in distinguishing among the dominant tundra vegetation communities in the Arctic, and can potentially help us to better understand and monitor arctic vegetation and ecosystem responses to environmental changes.