Assessing the classification accuracy of digitized National Aerial Photography Program imagery: Comparison with botanical data

• Main Author: Alexander, Mark Arthur
• Format: Text
• Language: English
• Published: UNI ScholarWorks 2000
• Subjects: Photographic surveying; Botany > Wyoming > Remote sensing; Remote Sensing; Napp; Wind River; Tundra
• Online Access: https://scholarworks.uni.edu/etd/619; https://scholarworks.uni.edu/cgi/viewcontent.cgi?article=1886&context=etd

The ability to acquire and use remotely sensed data has revolutionized large-scale ecological studies by reducing dependence on difficult and expensive field survey techniques for acquisition of land use and land cover data. Multispectral satellite imagery, typically from the Landsat Multispectral Scanner (MSS), Thematic Mapper (TM) and the SPOT High Resolution Visible (HRV) instruments, has proven particularly valuable in surveying and classifying vegetation cover (Frank & Thorn, 1985). Further studies have concluded that alpine vegetation communities and habitats can be mapped successfully by including the geomorphic parameters of elevation, slope, and incidence when classifying remotely sensed data. This research analyzes the degree of classification accuracy that can be obtained by using digitized National Aerial Photography Program (NAPP) aerial photographs and topographic data derived from a digital elevation model (DEM) by comparing these to detailed tundra vegetation and topographic data for an alpine tundra area in the Wind River Range, Wyoming. Two main objectives are associated with this research. The first objective is to determine the extent to which alpine vegetation types are visible with the spatial and spectral characteristics of digitized NAPP imagery. The second objective is to determine if topographic information such as elevation, slope, and aspect data derived from manual field study and a USGS DEM will increase the overall classification accuracy when combined with the digital color channels of the digitized NAPP. Discriminant function analysis was used to conduct the statistical classifications of the data sets. Bivariate relationships among the field and digitized NAPP variables were evaluated to test for variable similarity and to aid in the prediction of what kind of results can be expected from discriminant function classification. Statistical classification using linear discriminant analysis produced overall classification accuracies of 76.17% for the spectral data (digitized NAPP RGB). The classification accuracy of the integrated digitized NAPP and DEM data sets was 84.38%. This is significantly greater than the classification of either data set alone. These results support the idea reported in the literature, it is necessary to integrated TM and DEM data in multispectral classifications of mountain environments.