SCALE CHANGE EVALUATIONS FOR ARID LAND IMAGE INTERPRETATION CASE STUDY AT CAMP WILLIAMS, UTAH

摘要

The use of different scales of aerial and satellite digital imagery in mapping vegetation species/community and monitoring land degradation at three sites at Camp Williams, US Army National Guard Base, Utah, was examined. Our purpose was to assess the changing information content of varying resolution and to study the transfer of information from large-scale to small-scale data, also known as 'Upscahng'. Color-space conversions and filtering in frequency space were employed as pre-processing techniques to remove color band autocorrelation and high frequency noise. The large-scale aerial Kodak imageries were classified using an unsupervised classification algorithm to extract ecologically significant vegetation classes (both at species and community level). Low-resolution IKONOS, IRS and Landsat TM images for the entire scene were also processed and classified. Kodak 4mpp and 5mpp merged classifications were compared with IKONOS and IRS cookie-cut (clipped) image classifications. Two resampling methods -Nearest Neighbor and Bihnear Interpolation were used during merging and tested against each other. Large-scale images are useful for mapping and monitoring small, local areas in detail, but their usefulness in regional and global studies is restricted by their limited coverage. Therefore a method to scale up vegetation cover estimates to larger geographic extents was developed. The classified Kodak images were resampled (degraded) and used as the training data for the supervised classifications of the entire scenes and tested for accuracy using error matrices. Comparison of classification accuracies indicates an increase in overall accuracy and Kappa statistic with the upscahng of images.