United States Environmental Protection Agency Office of Research and Development National Exposure Research Laboratory Research Abstract Government Performance Results Act (GPRA) Goal 4 Annual Performance Measure 277 Significant Research Findings: Accuracy assessment of the 1992 multi-resolution land characteristics (MRLC) national land cover data (NLCD) Scientific Many air and water-quality models use land cover as a primary data source. Problem and Accuracy of land-cover data are therefore important for understanding model Policy Issues results. A statistically rigorous sampling design was used to collect reference data to document the thematic accuracy of the 1992 NLCD, the most widely used land- cover data across EPA. Research The main objective of this task was to document the thematic accuracy of the 1992 Approach NLCD. We used a two-stage cluster sampling design to collect 18,000 reference samples across the conterminous United States. The land-cover labels for the photo-based reference samples were compared to the satellite derived map land- cover labels to determine the thematic accuracy. Results and Thematic accuracy results show that 1992 NLCD meets nominally established Impact standards (85%) for 8 of 10 EPA Regions and Anderson Level I. Level I thematic accuracy was only 74% for the south-central region (Region 6) and 70% for the mid-Atlantic region (Region 3). The lower overall Level I accuracy for Region 6 was due primarily to confusion between pasture and grassland, and the lower overall Level I accuracy for the mid-Atlantic was due primarily to confusion between forest and agriculture. Level II and I class-specific users accuracies and associated standard errors are reported in Stehman et al. (2002) and Wickham et al. (2004), and these results are also reported at htto://www .epa.gov/mrlc . The results also demonstrate that land-cover heterogeneity (i.e., many different classes within a small area of the map) and ambiguity in reference label assignment were the primary sources of error. The subset of samples for which no other land- cover classes were within 30 meters showed significantly higher accuracy. Most of the classification error is at the edges between land-cover classes. Research Collaboration and Research Products The research was done in collaboration with the U.S. Geological Survey and State University of New York (SUNY-ESF). The published research and results are listed below, and reported on the internet at http://www.epa.gov/mrlc/. Wickham JD, Stehman SV, Smith JH, and Yang L. 2004. Thematic accuracy of 1992 National Land-Cover Data for the western United States. Remote Sensing of Environment 91:452-468. Wickham JD, Stehman SV, Smith JH, Wade TG, and Yang L. 2004. A priori evaluation of two- stage cluster sampling for accuracy of large-scale land-cover maps. International Journal of Remote ------- Sensing 25:1235-1252. Stehman SV, Wickham JD, Smith JH, and Yang L. 2003. Thematic accuracy of the 1992 National Land-Cover Data (NLCD) for the eastern United States: statistical methodology and regional results. Remote Sensing of Environment 86:500-516. Smith JH, Stehman SV, Wickham ID, and Yang L. 2003. Effects of landscape characteristics on land-cover class accuracy. Remote Sensing of Environment 84:342-349. Smith JH, Wickham JD, Stehman SV, and Yang L. 2002. Impacts of Patch Size and Land Cover Heterogeneity on Thematic Image Classification Accuracy. Photogrammetric Engineering and Remote Sensing 68:65-70. Future Research Research on change detection accuracy assessment is underway. James D. Wickham U.S. EPA, Office of Research and Development National Exposure Research Laboratory www. epa.gov/mrlc/accuracy .html Phone: 919/541-3077 E-mail: wickham.james@epa.gov Funding for this project was through the U.S. EPA's Office of Research and Development, National Exposure Research Laboratory, and the work was conducted by the Environmental Science Division. Collaborators were funded through IAGs DW14937937-01, DW14938034-01, DW14939340-01, DW14939330-01, and DW14939339-01, and contracts 9V1031NAEE, 68-C5- 0065, 68-D-00-267, and 68-W60023. Contacts for Additional Information ------- |