Statistical Issues in Assessing the Accuracy of Multi-Category Raster Maps

• Activity Number: 236
• Type: Invited
• Date/Time: Tuesday, August 13, 2002 : 2:00 PM to 3:50 PM
• Sponsor: Environmental and Ecological Statistics
• Author(s): Stephen Stehman*+ and Ray Czaplewski
• Affiliation(s): State University of New York, Syracuse and U.S. Department of Agriculture
• Keywords: probability sampling ; contingency table analysis ; agreement

Abstract

The combination of remote sensing and geographic information systems technology has resulted in diverse science and policy applications incorporating large-area maps of environmental attributes (e.g., land cover, land-cover change, forest canopy, and impervious surface). The widespread use of these maps increases the importance of assessing their accuracy. Accuracy assessments are based on comparing the map label to the "truth" at a sample of locations. The process may be viewed as consisting of three components: 1) the sampling design; 2) the response design, which is the measurement protocol for determining the true label; and 3) the analysis protocols for describing accuracy. Several statistical issues confront the sampling design and analysis components. Sampling design faces the historical survey sampling problems related to limited budgets, multiple objectives, and large populations. Analysis problems in accuracy assessment include describing and modeling agreement, and spatial analyses of classification error. Current practice will be reviewed, followed by discussion of potential directions to advance the statistical rigor and quality of these assessments.