Analysing spatial patterns of soil properties in a landscape requires a sampling strategy that adequately covers soil toposequences. In this context, we developed a hybrid methodology that couples global weighted principal component analysis (GWPCA) and cost-constrained conditioned Latin hypercube algorithm (cLHC). This methodology produce an optimized sampling stratification by analysing the local variability of the soil property, and the influence of environmental factors. The methodology captures the maximum local variances in the global auxiliary dataset with the GWPCA, and optimizes the selection of representative sampling locations for sampling with the cLHC. The methodology also suppresses the subsampling of auxiliary datasets from areas that are less representative of the soil property of interest. Consequently, the method stratifies the geographical space of interest in order to adequately represent the soil property. We present results on the tested method (R2 = 0.90 and RMSE = 0.18 m) from the Guinea savannah zone of Ghana. •It defines the local structure and accounts for localized spatial autocorrelation in explaining variability.•It suppresses the occurrence of model-selected sampling locations in areas that are less representative of the soil property of interest.
Keywords: Auxiliary dataset; GWPCA; Localised spatial soil variability; Optimised soil sampling design; Sampling design to represent both the feature and the geographical space; cLHC.