Estimation of map unit composition from transect data

We are concerned with the documentation of taxonomic variability within soil survey map units, and how that variability is expressed statistically. Twelve randomly selected transects were used to sample soil properties and classify pedons within a fluvial map unit of the Boone County, Missouri, soil survey. Map unit taxonomic variability was high, with 28 taxonomic family classes. Twenty-five percent of the sampled soils were in the named taxonomic class of the map unit, but most were similar inclusions. Interpretive purity of the sample was 83%. Variances about this sample proportion were calculated using three methods: (i) simple random sampling, (ii) cluster sampling, and (iii) two-stage random sampling. The simple random sampling formula provided the smallest variance, compared with the cluster formula (55% larger) or the two-stage formula (72% larger). However, the interpretation of transect sampling as simple random sampling is questioned. The two-stage random sampling method seemed most conceptually correct, but is computationally intensive and requires questionable assumptions about the population size. Confidence intervals based on small sample sizes drawn from binomial populations are highly suspect, and perhaps should be referred to as approximate bounds. It is important to carefully consider the assumptions and effects of any statistical technique chosen to analyze map unit composition.