Soil sampling is one of the most important stages in soil surveys. However, in conventional soil surveys, sampling schemes have proven to be inadequate for digital soil mapping because they may affect the results obtained and, in addition, they do not allow statistical estimations to be performed. This study attempts to evaluate the conditioned Latin Hypercube Sampling (cLHS) method with four environmental covariates (elevation, slope, curvature and land cover/land use map), in comparison to random sampling, in allocating 100 sample points, seeking a more representative description of the environmental characteristics of the Macacu watershed area. The performance of the methods was evaluated by qualitative assessment of histograms of distribution, and statistical analyses by the F test, Student T Test and Kolmogorov-Smirnov test (K-S) for each covariable. The sampling points selected by the cLHS method showed better geographic distribution than those obtained by random sampling. Furthermore, the cLHS method better maintained the frequency distribution of the continuous covariables than random sampling did, and the categorical covariable of land cover/land use showed no significance differences between the cLHS and random methods. Statistical analysis confirmed the better performance of the cLHS method compared to the random method, and the cLHS samples showed no significance differences from the watershed data. In contrast, the random samples showed significance differences from the watershed data for all the continuous covariables for at least one of the statistical tests used. Thus, the cLHS method is a satisfactory method for selecting points for soil sampling in heterogeneous areas, like those of the present study, for the purpose of use in digital soil mapping.
sampling design; soil survey; digital elevation model
