|
Abstract:
|
Recent advances in airborne and spaceborne sensors have made high spatial
(≤1m/pixel) and spectral resolution images (e.g. IKONOS, SPOT 5, Quickbird 2) widely
available, raising questions regarding their utility for floristic identification and
classification. Additionally, the use of object-oriented software to perform automated
classification and mapping has increased throughout the past 20 years. Studies assessing the utility of these image and software options frequently center on large, homogeneous sites and do not address these applications to small, heterogenous areas typical of the Pacific Northwest. In this study, a high-density sampling grid was used (approximately 9.0 % sample), followed by agglomerative cluster analysis and ordination, to identify all vegetation alliances and associations on a 148-ha study site in Maple Creek, California. Supervised classification using object-oriented software was performed on three images of various high spatial resolutions (0.15 m 4-band aerial photo, 0.60 m 4-band satellite image, and 1 m 3-band satellite image). The resulting classifications were compared with the reference vegetation map (derived from plot and image data) to assess accuracy. Results show differences in classification accuracy between the 3 images with the 0.60m Quickbird image producing the highest overall accuracy (69%); followed by the 0.15m aerial photo (48%); and the 1m NAIP image (37%) when assessed at the alliance level. |
