Abstract

For anarea south of Wager Bay, Nunavut (NTS map sheets 046D, E, 055P, 056A, H), a map detailing 22 surface material classes was produced using a non-parametric classifier, Random Forests, applied to a combination of RADARSAT-2 C-band dual- polarized (horizontal transmitted and horizontal received (HH) and horizontal transmitted and vertical received (HV)) and Landsat-8 OLI images with a digital elevation model and slope data. We show that the addition of RADARSAT-2 C-HH and C-HV images to the optical Landsat-8 OLI image in the classification process increases the overall classification accuracy from 96.7% to 99.3%. Similarly, the accuracy determined by comparing the resulting maps with georeferenced field data (i.e., mapping accuracy) increases from 72.1% to 78.0% when RADARSAT-2 C-HH and C-HV imagesare added to the classification. The material classes with the highest mapping accuracies were flooded alluvium and boulders, both with 100%. The class with the lowest mapping accuracy was thin sand and gravel over bedrock (11.1%), commonly confused with sand and gravel with vegetation and bedrock. Adding RADARSAT-2 data in the classification increases also the mapping accuracy that was established by comparing to georeferenced point observations.