Evaluating soil maps for prediction of soil water properties

Conventional soil survey stratifies a region into mapping classes and characterizes each by a representative soil profile within it. The efficacy of the procedure for predicting particle-size fractions, bulk density, water retention, and available water capacity (AWC) of the soil at previously unvisited sites on the Plain of Languedoc in southern France is evaluated for three scales of survey (1/10 000, 1/25 000 and 1/100 000) and is compared to that of prediction from stratified random and simple random samples. Data from 85 soil profiles on a random transect were used for evaluation. Classification partitioned the variation of the measured properties, except for AWC, well at the 1/10 000 and 1/25 000 scales, whereas classification at the 1/100 000 scale was less effective. At the 1/10 000 and 1/25 000 scales both classification and stratified random sampling were better for prediction than simple random sampling for the same total sample. On average the representative profiles proved substantially better predictors than the stratified random samples, but in most situations where soil stratification performed well efficiencies of the two predictors were similar. In essence, the more successful the classification was the more difficult it was to improve prediction by selecting representatives instead of sampling randomly within classes. These results confirmed statistically that the soil surveyor can exercise intuition and judgement to classify and select representatives.