METHODS
All imagery was loaded into Erdas Imagine 8.3 and prepared for georeferencing. The digital-ortho photo was used as the reference image or "truth" and both the Landsat MSS scene and the Airborne Multi-Spectral Digital Camera (AMDC) image were georeferenced to it by using ground control points.
The scenes were all subsetted to a uniform geographic size by using a standard template definied by the interpreter. The extent and local of the subset was chosen based on the heterogeneity of landcover types within the subset and the relative proximity of the scene to the interpreter's place of employment. This second consideration was minor, but proved convenient for ground truthing of unknown objects and relative distances.
The interpretation of the digital-ortho photo was done by traditional photo interpretation methods. Areas identified as "impervious surfaces" by the interpreter were designated by enclosing them in a polygon and adding them to an appropriate vector coverage. Three major classes of impervious surfaces were identified: buildings, parking lots, and roads. The areas of all the polygons in each of these three vector coverage's were summed and then multiplied by the appropriate pixel size. Percent impervious cover and approximate area were calculated from these data.
AMDC data was interpreted by using a basic unsupervised classification technique as described in the Erdas Imagine version 8.3 Tour Guides Manual. The program ran an algorithm which divided the spectral signatures / intensities of the scene up into 10 gray scale classes and arranged them in an attribute table. The interpreter then illuminated each class individually and made a determination as to the true nomenclature (cover type) of the features identified in that class. After all ten classes were assigned appropriate names, they were grouped into one of two major categories -vegetation or impervious surface. The area and percent impervious cover were determined from the data amassed in the attribute table by summing the impervious values and multiplying them by the appropriate pixel value.
The Landsat MSS data was previously processed at ERIM International by a process know as "Tassel Capping." This algorithm helps to identify areas that are not "wet" and reflect energy rather than absorb it. The processing of the Landsat data into the Tassel Capped format was done previous to this study and was not part of the interpreters analytical technique. The Landsat data was interpreted by a level slice technique. A simple attribute table was created in Imagine and the distribution of the the data's radiated values (from 0 to 255) were displayed in tabular form.
The viewer containing the Tassel Capped data was linked with one containing the unclassified AMDC scene. An inquire cursor was used to take "sample readings" from know impervious surfaces in the AMDC scene so the corresponding pixel values in the Tassel Capped scene could be noted. Approximately 10 samples were taken from each of three major impervious classes (buildings, roads, and parking lots) and a simple mean was calculated for each of these samples. The means were used to determine a range of values (from 0 to 255) that were determined by the interpreter to be most indicative of the impervious surfaces in this Landsat scene. The values in this range were highlighted and again the values in this impervious range were summed and multiplied by the appropriate pixel size to determine percent and total area of impervious cover.
Percentage of each scene determined to be impervious was compared against each other, with the digital-ortho photo used as a true measure of impervious surface. Cost of data, accuracy of identification, and time required to interpret data were used to determine the effectiveness of each technique.