Development of the United States National Land Cover Database: Experience from 1992 and 2001 Implementation

The Multi-Resolution Land Characteristics (MRLC) Consortium is a group formed by the U.S. federal agencies in 1990 to purchase Landsat 5 Thematic Mapper imagery for the conterminous U.S. and to develop a National Land Cover Dataset (NLCD 1992). The NLCD 1992 was completed in 2000. In 1999, a second-generation MRLC (MRLC 2001) consortium was formed to purchase three dates of Landsat 7 Thematic Mapper Plus (and some Landsat 5) imagery for the entire United States and to coordinate the production of a comprehensive National Land Cover Database (NLCD 2001). The NLCD 2001 was completed in April 2008.The NLCD 1992 and NLCD 2001 are specifically designed to meet the needs of the U.S. federal agencies for nationally consistent satellite remote sensing and land cover database. In addition, the MRLC consortium provides the database as public domain information. This paper summarizes the philosophy and scientific and technical issues with regard to design and implementation of the NLCD 1992 and NLCD 2001. Experiences and lessons learned from the intensive efforts to develop the two national land cover datasets are documented, including project design, technical approaches, accuracy assessment strategy, and project implementation. ∗ Information presented in this paper represents the views of the author only. It does not necessarily reflect the views of USGS or any other participating organizations and no official endorsement should be inferred. I. Background The late 1990s and early 2000s has witnessed several major and formative developments in globaland continental-scale land cover dataset using remotely sensed data (e.g. Townshend et al. 1997; DeFries et al. 1999; Loveland et al. 2000; Vogelmann et al. 2001; Friedl et al. 2002; Hansen et al. 2002; Bartholome et al. 2002; Homer et al. 2004; Liu et al. 2002; Homer et al. 2007). The endeavour is primarily driven by an increasing demand for up-to-date and accurate land cover dataset for study of global change at regional to global scale. The successful completion of those large-area land cover characterization activities are attributable primarily to the improved technological and algorithm development for land cover characterization, and the availability of the high quality remote sensing and geospatial datasets. Along with development of the continental and global land cover datasets, significant progress has also been made in designing and implementing accuracy assessment for those land cover datasets based on rigorous and statistically defensible design (e.g. Scepan 1999; Yang et al. 2001; Zhu et al. 2001; Stehman et al. 2003). A closer dialogue between the land remote sensing communities and data users is emerging. The continued improvement and availability of remote sensing and geospatial data has provided an excellent opportunity to further improve large-area land cover database development. One of the major efforts in large area land cover mapping is the development of two-time National Land Cover Datasets in late 1990s and early 2000 under auspices of the Multi-Resolution Land Characteristics Consortium (MRLC) (Loveland and Shaw, 1996). The MRLC was originally formed in early 1990 and consisted of several US federal agencies including: United States Geological Survey (USGS), Environmental Protection Agency (EPA), National Oceanic and Atmospheric Administration (NOAA), and United States Forest Service (USFS) (Loveland and Shaw 1996). Under the coordination of MRLC 1990, a National Land Cover Data 1992 (NLCD 1992) was developed (Vogelmann et al. 2001). Entering 2000, the need for updated national land-cover data has led to the development of a new initiative under MRLC 2000 to develop a National Land Cover Database 2001 (NLCD 2001) (Homer et al. 2007). The goals of this paper are two fold: 1) to summarize briefly major experiences learned from the development of NLCD 1992 activities, and 2) to present a new framework under which the NLCD 2001 has been developed. We illustrate the merits of the new framework through discussion of the NLCD 2001 design and implementation. The quality assessment of NLCD 2001 is also discussed. II. Review of NLCD 1992 project In early 1990 several federal agencies recognized the need for an up-to-date national land cover dataset of medium spatial resolution to support their national and regional environmental programs. This had let to the formation of the MRLC to pool resources for purchasing and processing Landsat data, and to develop a seamless national land-cover map of the conterminous U.S. The primary satellite data used for NLCD 1992 land cover mapping project was the Landsat 5 TM data of circa 1992, acquired from both leaf-on and leaf-off seasons. In addition, ancillary geospatial data were compiled for support land cover