A raster version of the Circumpolar Arctic Vegetation Map (CAVM)
New publication by Martha K. Raynolds, Donald A.Walker, Andrew Balser, Christian Bay et al.
Land cover maps are the basic data layer required for understanding and modeling ecological patterns and processes. The Circumpolar Arctic Vegetation Map (CAVM), produced in 2003, has been widely used as a base map for studies in the arctic tundra biome. However, the relatively coarse resolution and vector format of the map were not compatible with many other data sets. We present a new version of the CAVM, building on the strengths of the original map, while providing a finer spatial resolution, raster format, and improved mapping. The Raster CAVM uses the legend, extent and projection of the original CAVM. The legend has 16 vegetation types, glacier, saline water, freshwater, and non-arctic land. The Raster CAVM divides the original rock-water-vegetation complex map unit that mapped the Canadian Shield into two map units, distinguishing between areas with lichen- and shrub-dominated vegetation. In contrast to the original hand-drawn CAVM, the new map is based on unsupervised classifications of seventeen geographic/floristic sub-sections of the Arctic, using AVHRR and MODIS data (reflectance and NDVI) and elevation data. The units resulting from the classification were modeled to the CAVM types using a wide variety of ancillary data. The map was reviewed by experts familiar with their particular region, including many of the original authors of the CAVM from Canada, Greenland (Denmark), Iceland, Norway (including Svalbard), Russia, and the U.S. The analysis presented here summarizes the area, geographical distribution, elevation, summer temperatures, and NDVI of the map units. The greater spatial resolution of the Raster CAVM allowed more detailed mapping of water-bodies and mountainous areas. It portrays coastal-inland gradients, and better reflects the heterogeneity of vegetation type distribution than the original CAVM. Accuracy assessment of random 1-km pixels interpreted from 6 Landsat scenes showed an average of 70% accuracy, up from 39% for the original CAVM. The distribution of shrub-dominated types changed the most, with more prostrate shrub tundra mapped in mountainous areas, and less low shrub tundra in lowland areas. This improved mapping is important for quantifying existing and potential changes to land cover, a key environmental indicator for modeling and monitoring ecosystems. The final product is publicly available at www.geobotany.uaf.edu and at Mendeley Data, DOI: 10.17632/c4xj5rv6kv.1.