INFLUENCE OF THE SPATIAL DISTRIBUTION OF SAND CONTENT ON SAMPLING PATTERNS

Abstract

The spatial variability of sand content (0–15 cm) was examined on a crop rotation study site located on a Brown Chernozemic loam soil at Swift Current, Saskatchewan. The autocorrelation for surface sand content indicated a significant cyclic pattern, with a wavelength of approximately 56–65 m, which was considerably longer than the experimental plots (40 m). Since the sand content was not randomly distributed in space (and therefore all soil properties correlated to sand would also be nonrandomly distributed in space), an alternative sampling pattern had to be devised which would allow for an adequate measurement of the soil properties in each plot. Based on the autocorrelation function, a sampling interval of approximately 10 m was selected. The bootstrap method was used to compare a random versus a fixed interval-sampling pattern for these plots. Both patterns adequately estimated the mean of the population, but the random sampling pattern gave a greater amount of variation in the mean values. In addition, the random sampling pattern consistently underestimated the variance by as much as 30%. The fixed interval-sampling pattern gave a much better estimate of the variance (usually within 10%). The techniques used in this study would allow the development of sampling patterns to obtain unbiased estimates of soil properties, within experimental plots, which may not be randomly distributed in space and/or normally distributed. Key words: Soil variability, spatial distribution, sampling patterns, sand content