Much of the agricultural region of the United States is still not mapped with a modern soil survey. A need exists to provide tools which will both expedite the soil survey and place the survey maps in a digital form for implementation in computer data banks. Computer-assisted interpretation of scanned aerial photography may provide an intermediate product which may be used by soil surveyors to more rapidly and accurately delineate boundaries between soil polypedons. Moreover, the resulting raster format map may be easily encoded into data banks for use in a general land records system. Output maps from such a system may be generalized as desired and produced at a scale convenient to the user. The purpose of this study is to investigate means of producing an interim product to be used by field soil surveyors as an additional tool in the mapping of soil individuals.
Color infrared photography at a scale of 1:24,000 was taken by the Department of Natural Resources on 23 April 1979 over a site located approximately 5 kilometers west of Neenah, Wisconsin. The soils of this are a are formed in calcareous clayey glacial till and glacial lakebed sediments thickly deposited over dolomite bedrock. The entire area has been blanketed to various depths by aeolian silt. Pre-settlement vegetation consisted of oak-hickory communities on the uplands and sugar maples and basswood communities transitioning to wetlands on the lowland areas. Relief is generally less than 20 meters, and topography is gently rolling. Most recent glaciation of the area ended approximately 10,000 years ago.
A 96 mm by 115 mm portion from the center of one of the color infrared transparencies was chosen for scanning and analysis. Pixel size chosen was 100 mµ. The transparency and its corresponding film wedge were scanned through narrow band (10 nm) interference filters centered at 450, 550, and 650 nm. The resulting digital data array was corrected to analytical densities by a transformation matrix and to relative log exposures by reference to the film wedge.
A computer-assisted supervised classification was performed on the corrected data using an elliptical table look-up algorithm. Input to the program consisted of statistics from 37 training sets and a user-selected number of standard deviations. Eigenfunctions described the location of the ellipses in spectral space, and the standard deviation input governed the size of the ellipse for each class. Pixels whose values were included in two or more overlapping classes were classified by means of a maximum likelihood classifier. Additionally, any pixel whose values placed it more than 4 standard deviations from the mean of the best class was designated as unclassified.
Comparison of the resulting colorcoded thematic map to the SCS detailed soils map shows general correspondence. However, the classification is much more detailed. Therefore, map evaluation is being conducted by ground study using soil pits and transects of soil auger borings. Preliminary conclusions indicate that the computer classification are accurately delineates soil series boundaries in bare soil areas than does the SCS detailed map.
A similar procedure was conducted on imagery over the same site acquired on 22 September 1978. Preliminary conclusions are that fall as well as spring imagery are useful data sources.
Thus, one may conclude that this technique may provide a useful tool for soil surveyors in regions where fall or spring plowing exposes relatively large areas of bare soil long enough for acquisition of photographic imagery.
Date of this Version