WSP11837 - Laserfiche WebLink

Home Browse Search

The fast data-processing and analysis functions of the Image-100 System were used advantageously for a number of tasks. The entire Landsat image was examined and parts of the image were selected for detailed study (fig. 6). The selected parts were enhanced to improve interpretability by contrast stretching; that is, the contrast between features and their surroundings was enhanced. A color-composite image of data from three spectral bands was displayed. Filter colors and gain-and-bias controls on the Image 100 were adjusted to en~ance selected features of interest. Classification was accomplished using the parallelepiped supervised technique (p. 46). For this approach, a data group which was considered to be representative of a land-use category was selected as a training set. Several areas outlined as training sets for irrigated hay meadows that are located north and west of the Hayden Powerplant (fig. 1) are illustrated on f i gu re 7. Using the Image-IOO System, the analyst must decide whether or not a spectral signature from a training set is a valid representation of the land- use category of interest. The training set should include only the area occupied by the desired land-cover type. This is complicated when individual pixels contain parts of adjacent land-use types. Attempting to broaden the spectral signature of one class to include the boundary pixels generally produces overlap, so that some pixels are classified as two or more land- cover types (p. 46). If an effort is made to avoid overlap, some pixels remain unclassified. A classification of 85 to 99 percent of the pixels generally is considered adequate for analysis. A decision was made as to the approximate number of spectrally different land-use classes in the selected part of the image. The results were tested on other parts of the image by applying and evaluating the preliminary spectral signatures (see Glossary). The objective was to check the results that would be obtained in other areas having a different mix of land use. Analysis of land-use types on the August 24, 1975, image was not completely successful using the Image-1DO System. This was because the paralleiepiped technique was not always able to differentiate the complex variety of land-use classes with a wide range of I ighting conditions (direct sunlight to almost complete shade) in the mountainous terrain characterizing the region. For example, it was determined that several classes of range and dryland agriculture had distinctive spectral signatures, but that riparian vegetation along the streams did not have enough areal extent or density to produce a good training area or a unique signature. Also, the recently worked areas of surface mines proved to have the same signature as other areas of bare soil and rock (that is, sparse vegetative cover), and recently reclaimed surface-mine areas had the same spectral signature as some other areas of agricultural land. In general, most boundary areas between different land-use categories either were classified as more than one category or were left unclassified. To demonstrate land-cover classification, broad categories were obtained by combining signatures on this Landsat image. of land use The following 14