A database approach to a global land cover characterization: Foundations, challenges, and lessons.

摘要

There is a significant need for improved global land cover data due to limitations in existing data sets and because of the rapidly increasing sophistication of applications using land cover. Scientific organizations, including the International Geosphere Biosphere Programme, have called for the generation of improved global land cover data. A review of applications of land cover data for large-area environmental modeling shows that land cover data needs vary between different types of applications, and even within common applications. To meet the need for improved data, a land cover characterization methodology was implemented at the global scale. The approach employed 1992-1993 1-km resolution Advanced Very High Resolution Radiometer remotely sensed data and other spatial data to produce a global land cover characteristics database. The methodology, which used unsupervised classification of multi-temporal satellite data followed by extensive post-classification refinement, produced a multi-layer database consisting of eight derived land cover data sets, descriptive attributes, and corresponding source data. The most detailed portrayal of global land cover is the 961 seasonal land cover regions, which represent lands with common characteristics of land cover, seasonality, and relative primary productivity. These were aggregated to produce seven land cover data sets commonly used for environmental applications. This flexible database strategy permits tailoring land cover data on a case-by-case basis to meet the needs of individual applications. Classification results show that forest cover has the largest area on Earth, followed by agriculture, grassland and savannas, barren land, snow and ice, shrubs, tundra, wetlands, and urban land. Generally, the relative primary productivity of land cover decreases as latitude increases, while phenological properties become more predictable as latitude increases. Satellite and reference data quality, methodology, interpreter performance, and logistical constraints likely affected the quality of the classification results. These results, however, are unvalidated. This work demonstrates a new, flexible method for approaching the development of land cover data. Hopefully, this work will focus the attention of organizations with mandates to produce such data on the applicability of the land characteristics database model.