Land cover classification of tundra environments from Landsat, 2000-2001, Lena Delta, Laptev Sea Region (RU)

The study was based on land cover classification of three almost cloud free Landsat-7 ETM+ satellite images. The acquisition dates are 27 July 2000 (path 131, rows 8 and 9) and 26 July 2001 (path 135, row 8). Both were taken approximately at the peak of the vegetation period. ERDAS Imagine software was used to carry out all image processing tasks. In addition to the ETM+ satellite imagery, we acquired and utilized numerous other ancillary data for determination of typical land cover classes and field training sites: vegetation field data, soil information, field and aerial photography. To minimize radiometric differences between the three scenes due to different atmospheric conditions, a basic radiometric and image-based atmospheric correction according to Chavez (1996) was applied. Finally, the three scenes were projected to UTM Zone 52 with the geodetic datum WGS 1984 and a mosaic of the Lena Delta was composed. Supervised classification was carried out using the spectral Landsat bands 1-5 and 7 (VIS, NIR, SWIR) with 34 training areas for ten land cover classes. The training areas were distributed on the active floodplain and first terrace (21 sites), on the second terrace (8 sites), and the third terrace (5 sites). For the supervised classification, the minimum distance algorithm was used. After evaluation of the classes regarding their methane emission two classes were merged to the final number of nine. The accuracy assessment for our classification was based on 36 validation sites based on an image mosaic of Hexagon (synonymous with 'Keyhole-9') providing a dataset independent from the Landsat-7. Our accuracy assessment of the Landsat-7 supervised classification indicates a reasonable well overall accuracy of 77.8% (Kappa=0.74) for such a large and remote study area.

Detailed information about the methods can be found in the publication to which this dataset is a supplement.