Variations in land cover area estimated from remotely sensed data

Abstract

Land use and land cover (LULC) information is required for a host of environmental applications. Such information is needed at a variety of spatial resolutions ranging from meters to, on a global scale, hundreds of kilometers. This research explored impacts of selected upscaling procedures and landscape structure on LULC representation at multiple spatial resolutions. A LULC dataset having various spatial resolutions was derived from a Landsat Thematic Mapper image. Comparisons of class areas among different resolutions and between maps obtained from two different upscaling procedures were conducted. Effects of landscape structure on class area estimation were also examined. Results indicated that class areas changed substantially through resolutions, but that patterns of change were different for different classes and upscaling procedures. While class areas at coarse resolutions were predominantly determined by initial areas on the full resolution map, they were also significantly related to several metrics that characterize landscape structure (e.g., patch size, shape and interspersion indices). Considerable differences were found between coarse resolution LULC maps aggregated from the full resolution map and those classified from degraded images, indicating the difficulty of comparing a LULC map classified from coarse resolution images with one aggregated from a fine resolution classification.