1992 NLCD Accuracy
Accuracy of the 1992 NLCD land-cover map was conducted by EPA federal region (see figure) using a probability sampling design incorporating three levels of stratification and two stages of selection. Details of the methodology and results have been published. Although some regional variation in protocol and implementation exists, all of the regions shared a common general framework. The goals of this framework were to insure that: 1) satisfy protocols defining a probability sample; 2) sufficient sample sizes were acquired for each land-cover class; 3) reasonable cost controls were maintained and 4) a spatially well-distributed sample was acquired. Reference land-cover labels were acquired through the photo-interpretation of NAPP aerial photographs, or DOQQs. A correct classification was defined as occurring when the primary, or secondary reference label matched the mode class present in a 3x3 block centered on the sample point. Assessments were conducted at both the Anderson level I and II levels of classification.
In addition to the above accuracy reports, research has been conducted in evaluating different sampling schemes, as well as the impact of landscape characteristics on accuracy.
Error matrices for the EPA regions can be accessed at both classification levels by clicking on the appropriate region in the figure below. The cell values presented in the matrices were derived by weighting each sample pixel by the inverse of its inclusion probability resulting from the sampling design, thus calculating the estimated percentage of that class in that region. For example, a cell value of 8.523 would translate to 8.523% of the region's area. Included in the matrices are the errors of commission, denoted by the letter C, and errors of omission (O).