Mapping Shrub Abundance in Arctic Tundra from the Satellite High Resolution Record and Impacts on Albedo: Validation
Mark James Chopping, Montclair State University, choppingm@mail.montclair.edu (Presenter)
Darko Radakovic, Montclair State University, radakovicd1@montclair.edu
Angela Erb, University of Massachusetts Boston, angela.erb@umb.edu
Rocio Duchesne-Onoro, University of Wisconsin - Whitewater, duchesnr@uww.edu
Zhuosen Wang, NASA GSFC/University of Maryland, zhuosen.wang@nasa.gov
Crystal Schaaf, University of Massachusetts Boston, crystal.schaaf@umb.edu
High resolution 2.5 m shrub abundance maps were created for 127 2 km^2 sites in Alaskan Arctic tundra using QuickBird-2 (QB)/WorldView-2/3 (WV) panchromatic and NDVI image pairs over a 15- to 18-year period. The goal is to provide a dataset that can be used to assess the impact of changes in shrub abundance on summer surface albedo and to validate lower spatial resolution ABoVE remote sensing data products. The imagery was orthorectified to a 0.5 m grid using the Polar Geospatial Center code and conversion to 2.5 m shrub cover and abundance maps was achieved by calculation of a roughness metric (RM) in a 5 × 5-cell contiguous moving window, reflecting shrub density. Initial comparisons between the RM and Landsat-based shrub aboveground biomass (g/m^2; Berner et al. 2018) yielded R^2 = 0.7, though both RM- and Landsat-based approaches potentially suffer from inaccuracies over polygonal ground: image spatial variability is increased by presence of (linear) troughs; and Landsat's spectral measures are sensitive to green non-shrub vegetation in the troughs, where moisture can collect. In this work, validation of maps was performed using the 'Toolik' map from the ABoVE 'High-Resolution Vegetation Community Maps for the Toolik Lake Area, 2013-2015' (Greaves et al. 2019). The classes 0_No_Data, 5_Low dense shrub, 8_Shrubby tussock tundra, 10_Shrubby moist non-tussock tundra, 11_Low to tall moist shrub, and 12_Tall shrub were recoded 0_No_Data: 0; 5 & 8:1; 10 & 11: 2; and 12: 3. The 2009 QB and 2017 WV roughness maps were recoded to classes none, sparse, moderate, and tall, with <1.1, 1.1<1.5, 1.5<3, and >=3. A small region of invalid imagery was masked before calculation of confusion matrices and error metrics. The overall, user's, and producer's accuracies were 64%, 77%, and 82%, respectively, for the QB-derived map; and 61%, 78%, and 76%, respectively, for the WV-derived map. If classes are combined to none/sparse and moderate/tall, the respective accuracies are 82%, 87%, and 92% (QB); and 81%, 88%, and 89% (WV). Shrub cover was 11.4% and 14.5% in 2009 and 2017, respectively, though this may partly reflect the higher intrinsic resolution of the WV vs QB images rather than real change; if so, the rate of 0.39/year can be used to correct change estimates for the other 126 sites.
Poster: Poster_Chopping_2-19_186_35.pdf
Associated Project(s):
Poster Location ID: 2-19
Presentation Type: Poster
Session: Poster Session 2
Session Date: Wed (May 10) 5:15-7:15 PM
CCE Program: TE