Fine Spatial Resolution Satellite Sensor Imagery for National Land Cover Mapping

摘要

Three primary categories of land cover in the UK were identified (agricultural, semi-natural, urban). These categories had distinct characteristics and, therefore, different mapping requirements. 4 m spatial resolution multispectral imagery was used to classify three study areas, each of which corresponded to one of these land cover categories. 20 m spatial resolution imagery was also classified for comparison. In each case, per-pixel and per-field classification was performed, using either a hard or a fuzzy classifier. In addition, supplementary procedures were developed to increase classification accuracy (e.g., a potential error flag and a missing boundary flag). Overall, classification of 4 m spatial resolution imagery was more accurate than that of 20 m spatial resolution imagery, although this was not case for agricultural study area. Generally, per-field classification was more accurate than per-plxel classification and fuzzy classifiers (performed on the urban study area) were more accurate than hard classifiers.