An accuracy assessment of forest disturbance mapping in the western Great Lakes

The increasing availability of satellite imagery has spurred the production of thematic land cover maps based on satellite data. These maps are more valuable to the scientific community and land managers when the accuracy of their classifications has been assessed. Here, we assessed the accuracy of a map of forest disturbance in the watersheds of Lake Superior and Lake Michigan based on an improved version of the Vegetation Change Tracker algorithm (VCTw). We constructed a probability-based sampling design using two stages of sampling with stratification at each stage. Results are presented for the portion of the map within the U.S. as well as separately for the U.S. portion of Lake Superiorʹs watershed and for Lake Michiganʹs watershed. We also present estimates and standard errors of the percent cover for each land cover class that incorporate both the mapʹs data and our sample data. The overall accuracy for the U.S. portion of the map is estimated to be 91% with a standard error of 0.8%. We discuss the relative strengths of the VCTw algorithm as well as the dependence of such an algorithmʹs success on the characteristics of the landscape being mapped.