Object-based urban land cover mapping using high-resolution airborne imagery and LiDAR data

Urban land cover information is important for a number of applications. In recent years, the availability of airborne light detection and ranging (LiDAR) and high spatial resolution (HSR) imagery makes it possible to generate land cover information at fine scales. In this study, we proposed an object-based image analysis (OBIA) method to derive 1m resolution land cover classification from airborne LiDAR and multi-spectral image data. A series of rules were developed for identifying 7 land cover features (low impervious cover, buildings, shrub/tree, grass, soil/rock, rivers/lakes, and swimming pool). Experiments were performed in two sites in Richland County, South Carolina, USA. The classification results yielded an overall accuracy of 92.23% and a kappa coefficient of 0.8996. Confusion occurs between soil/rock and grass land and low impervious surface due to their spectral similarity. The algorithm shows promise for large-area classification in forested urban landscapes with similar datasets.