Issues involved in the accuracy assessment of large scale land use/land cover mapping from remotely sensed data

The Multi-Resolution Land Characteristic (MRLC) Consortium, composed of several U.S. Government agencies, sponsored the creation of the National Land Cover Data (NLCD). This dataset provides a consistent land cover classification system for the lower forty-eight states. It is based on thirty meter spatial resolution Landsat Thematic Mapper (TM) satellite data. The NLCD, as a national land cover classification, is unprecedented in its coverage and resolution. The objective of this project was to assess the accuracy of a portion of the NLCD. An implementation scheme for the project was designed by the Center for Earth Observation group to assure the quality of the reference data collection when applied to such a large scale project. A number of issues were encountered in dealing with a project of this magnitude. These issues are of importance to researchers involved in large scale land cover mapping and change detection. Issues addressed include: Examining the impact of including alternate classes in the accuracy assessment, compensating for positional errors in the dataset, and examining reference data collection issues.