Engineering Soils Mapping from Multispectral Remote Sensing Data Using Computer Assisted Analysis

Research objectives included 1) determine relationships if any between spectral response of soil and soil textural groups, Unified Soil Classification subgroups and selected landforms and 2) develop computer assisted techniques for engineering soils mapping of remotely sensed data. The sire in northeastern Kansas, 1.6 Km by 43 Km (1.0 mi by 27 mi), contained glacial and alluvial soils. Multispectral scanner data (0.40 to 1.80 pm) collected at 915 meters altitude (3,000 feet) were analyzed using two distinct computer techniques: the nonsupervised approach involving delineation of soil training areas directly by computer and the supervised approach entailing training area selection manually by the researcher. Computer classification of surface materials follows both methods. Non supervised analysis results failed to show a consistant relation- ship between landform type and soil spectral class. Landforms correlating well displayed either unusually bright or unusually dark spectral signatures. Supervised analysis results showed a strong relationship between soil texture and soil spectral class. A moderately strong tie was shown between engineering soils groups (ML, CL, CH and OH) and their respective spectral classes, and a weak association between landform type and soil spectral class. An analysis procedure for engineering soils mapping by computer of remotely sensed data was developed. 1. Locate cultural features by visual examination of imagery. 2. Produce generalized bare soil-vegetation- water map using nonsupervised technique. 3. Outline significant soil fields for computer training using supervised approach. 4. Computer-classify entire area based on these soil fields.