Calibration, Compositing, and Classification of Landsat Datasets and High-Resolution Imagery in Arctic Alaska

Abstract:

Providing calibrated, cloud-free, and phenologically consistent satellite basemaps at moderate (~30 m) and high resolution (≤2 m) is critical to the mapping of arctic tundra vegetation and landscape attributes across large study areas. We obtained ground cover (n = 107) and field spectra (n = 28) data for tundra vegetation across a network of field plots in a ~100,000 km² study area spanning the foothills and coastal plain ecoregions of Alaska’s North Slope. Calibration and atmospheric correction of Landsat TM, ETM+ and OLI, WorldView-2, and GeoEye-1 imagery were performed and we compared results across sensors and with ground spectra. For the Landsat imagery, we produced consistent basemaps using compositing approaches that focused on capturing central tendencies from multiple years of imagery within narrow phenological windows (e.g., green-up, peak growth, fall senescence). We leveraged information from the full 1985–2014 time series to optimize compositing year ranges to prevent bias due to directional changes (e.g., shrubification), and “step-changes” (e.g., tundra fire, lake drainage, ice-wedge degradation) in vegetation and landscape characteristics over the Landsat era. Finally, we explored automated classification of the calibrated Landsat and high-resolution imagery using the spectral rule-based classifier Satellite Image Automatic Mapper (SIAM).