The cLHS is considered as a robust sampling strategy for the selection of representative samples of the landscape, which uses environmental variables and their multivariate distributions. In this study, we propose a method based on cLHS for selecting soil sampling points in areas of difficult access, with the objective of minimizing inaccessibility problems in field campaigns, considering obtaining alternative samples with a lower cost and time demand. The study aims to analyze, above all, the practical operational performance of the method based on the potential and restrictions for application in digital soil mapping. For this, five predictive models (GBM, RF, SVM, kNN and C5.0) were initially used to select the most important variables to be inserted in the cLHS. The k-means method was applied to select the alternative points closest to the original points of the cLHS. Restrictions such as the euclidean distance from roads and the exclusion of urban and mining areas were incorporated. Approximately 30% of the original sample points of the cLHS could not be accessed in the field, the main operational restriction was due to the lack of access/routes to the selected points. However, the use of alternative sampling points allowed greater flexibility and accessibility in the field, where it was possible to collect 17% of the points and reduce the demand for time and cost. With this, the sampling strategy adopted made it possible to obtain an ideal minimum size of sampling points to be used in predictive models in digital soil mapping studies.