Evaluating central tendency and variance of soil properties within map units

Users are demanding that soil surveys contain statistical information about soil and map unit properties. Numerous sampling and statistical methodologies are available, but statistical methods should be appropriate for both sampling methods and data distributions. Rigorous investigations of soil property distributions are rare, and statistical methodologies have received little attention for soil survey applications. Two hypotheses were tested: (i) frequency distributions of soil properties within map units are non-normally distributed, and (ii) variability estimates will depend on the statistical assumptions and methods. Twelve transects were randomly selected in a Boone County, Missouri soil survey map unit and 12 soil attributes were investigated from a population of 120 soil cores. Three methods were used to estimate variances: (i) simple random sampling, (ii) cluster sampling, and (ii) two-stage random sampling. Frequency distributions were significantly skewed for 11 of the soil properties. The median is an appropriate estimator of central tendency for these properties. The random sampling approach provided the smallest variances, but is conceptually questionable. Two-stage sampling is conceptually most correct, but requires dubious assumptions about population size. Non-parametric confidence intervals for variable medians were appropriate, but are wide for most properties. None of the applied assumptions and techniques are universally applicable. Statistical assumptions and methodologies should be tested across a spectrum of depositional, weathering, and land-use conditions to determine the most suitable approaches for particular user needs and soil conditions.