A Multivariate Approach for High-Resolution Soil Survey Development

First-order soil surveys (scales 91:12,000) are essential for detailed land use assessment. Recently developed technologies depicting landscape variability at high resolution are useful for first-order survey development. Our objective was to compare a first-order soil survey created using conventional techniques versus a multivariate firstorder survey developed using terrain attributes calculated from digital elevation models and electrical conductivity (EC) mapping. Two research sites (Macon [9 ha] and Dale [8 ha]) were located in the Coastal Plain physiographic region of Alabama, and first-order soil surveys (scale 91:12,000) were generated using conventional techniques. Soils are largely Aquic, Oxyaquic, and Typic Paleudults at the Macon site and Typic Kandiudults that differ in particle size family at the Dale site. Elevation data were collected using real-time kinematic global positioning system, terrain attributes were calculated, and field-scale EC data were collected. Three principal factors described 81% and 80% of the terrain and EC variability for the Macon and Dale sites, respectively, and fuzzy k-means clustering of principal factor scores was used to create multivariate zones. Random pedon sampling was used to compare techniques, and a rigid similar-dissimilar rule (one-class) was used for accuracy assessment. Probabilities of success (p) for observing the named soil within a map unit for the multivariate zone approach averaged 50% and 76% for the Macon and Dale sites, respectively, which was slightly less than the conventional approach. Estimated errors and confidence interval calculation indicate that for these Alabama Coastal Plain landscapes the overall accuracy of the two approaches was similar.