For each sample block, five dates of Landsat imagery were interpreted to map land use and land cover. The classification system consisted of 11 general land-use/land-cover classes (Water, Developed, Mechanically disturbed, Mining, Barren, Forest, Grassland/Shrubland, Agriculture, Wetland, Nonmechanically disturbed, Ice/snow). Land-use/land-cover delineation of each sample block began with the creation of a baseline reference land-use/land-cover dataset. The 1992 date was usually the starting point due to the availability of the 30-meter-resolution 1992 National Land Cover Data (NLCD). The more detailed NLCD classes were first aggregated to match the more general land-use/land-cover classification scheme used by the Land Cover Trends Project. The NLCD data were then manually edited on the computer screen using Erdas Imagine software, using on-screen interpretation methods while looking at 1992 Landsat TM data and also at various aerial photographs and geospatial data, which served as interpretation aids; a list of ancillary data is included in the accompanying document entitled "Land Cover Trends Dataset, 1973-2000" (available at <http://pubs.usgs.gov/ds/844/pdf/ds844.pdf>). This editing (that is, cleanup) procedure was done because NLCD data were created using automated image processing procedures at different scales (multistate blocks) and were not meant for use in local-scale assessments. A minimum mapping unit of 60 meters was used to guide delineations, and so features having ground footprints less than 60 meters wide were not mapped. Thus, the mapping excluded high-contrast features such as roads, which may have a distinct spectral signature but have ground dimensions of less than 60 meters. Land cover for the 1973, 1980, 1986, and 2000 periods was then backward- or forward-classified using the 1992 land-use/land-cover data as the template. Creation of the 2000 land-use/land-cover product, for example, began by making an exact copy of the 1992 land-use/land-cover product, which served as a baseline for the 2000 land-use/land-cover product. Changes between 1992 and 2000 were identified and then manually edited into the copied image by linking the 1992 Landsat imagery and the 2000 Landsat imagery, using the ERDAS Imagine geospatial-data authoring system. Each analyst panned through the entire sample block, examining the 1992 and 2000 Landsat imagery and any relevant aerial photography for valid land-use/land-cover changes between the two dates. Any land-use/land-cover changes identified were manually digitized on-screen, and the land cover was attributed on the 2000 land-use/land-cover map file. Upon completion of the 2000 land-use/land-cover product, the same procedures were used to create the 1986, 1980, and 1973 land-use/land-cover products, using 1992 to backward-classify change in 1986, and so forth. This manual process eliminated errors that can occur when change analyses are performed between independently created land-use/land-cover products. Because only manually identified, delineated, and coded land-use/land-cover changes were analyzed during this phase, classification errors were greatly reduced.
Sample blocks have dimensions of either 10 x 10 kilometers or 20 x 20 kilometers. Exceptions exist only where slivers were added along a coastline or in international boundary areas. Such sliver blocks have a file name that reflects adjacency to another block, and, therefore, the dimensions of the sliver sample blocks vary. Pixels were classified using a 60-meter minimum mapping unit. Interpreters classified objects that were discernible at the Landsat pixel resolution. All pixel values were assigned a value (between '1' and '11') that indicated the land-use/land-cover class.
Landsat imagery are not included as part of this release of the Land Cover Trends Dataset, but the image year is listed in the metadata to support approximate replication of the land-use/land-cover output.