Fuzzy classification and post-processing of satellite imagery to derive watershed model parameter values
Fuller, Robert Clark
MetadataShow full item record
Nonpoint source water pollution, because it is diffuse in its origin, is more difficult to locate and control than point source water pollution. Modeling of the hydrologic and water quality characteristics of watersheds is a valuable tool in the effort to control nonpoint source pollution. Remotely sensed imagery, including satellite imagery and aerial photography, are increasingly being used as sources of data in the formulation of watershed models. Image processing software and geographic information systems (GIS) have been used to classify remotely sensed data, organize those data, and prepare them for use in watershed models. In prior studies of this type, remotely sensed data have been classified into land use/land cover classes, then linked to tables containing watershed model parameter values associated with the classes. In this study, field estimates of selected watershed model parameter values were used as training set data for fuzzy classification of Landsat Enhanced Thematic Mapper Plus (ETM+) satellite imagery into three sets (image layers) of possible values of the parameters. The associated spectral distance files, based on Mahalanobis distances, were used in a postclassification model to assign weighting values to the three fuzzy layers and calculate composite parameter values for subsequent use in the AGNPS watershed model. A second spatial model was developed to perform accuracy assessment on the classification and modeling used to generate watershed model parameter values. The technique was tested on three watersheds in Georgia, one in the Coastal Plain and two in the Piedmont. The technique produced reasonable results but it was not possible to judge its merits relative to existing techniques based on the specific application that was tested. Several other, more promising applications for the technique are suggested.