NLCD Change (NLCD 1992 versus NLCD 2001)
Direct, pixel-to-pixel comparison of NLCD 1992 and NLCD 2001 land cover is not recommended for several reasons: 1) NLCD 1992 was based on an unsupervised classification algorithm, whereas NLCD 2001 was based on a supervised classification and regression tree algorithm; 2) terrain corrections were based digital elevation models (DEM) with a 90-meter spatial resolution for NLCD 1992, whereas terrain correction for NLCD 2001 used 30-meter DEMs; 3) the impervious surface mapping that is part of NLCD 2001 resulted in the identification of many more roads than could be identified in NLCD 1992, however, most of these roads were present in 1992; 4) NLCD 2001 imagery was corrected for atmospheric effects prior to classification, whereas NLCD 1992 imagery was not, and; 5) there are subtle differences between the NLCD 1992 and NLCD 2001 land-cover legends. These factors result in substantially different pixel-by-pixel labeling in the two dataset, much of which is probably not genuine land-cover change, as shown in the pictures below.
![]() original 1992 NLCD land cover Link to larger image. |
![]() 2001 NLCD land cover Link to larger image. |
![]() 1992 change product land cover Link to larger image. |
![]() Legend |
MRLC is correcting this problem by re-classifying MRLC 1992 era imagery, using NLCD 2001 as the template. The result will be a product that identifies change and no change on a pixel-by-pixel basis at the more generalized Anderson Level I classification scheme (e.g., water, urban, barren , forest, shrub/grass, agriculture, and wetland). Users should note that wetlands include woody wetlands (e.g., riparian forest) in the generalized scheme.