Mapping soil properties in a poorly-accessible area

Soil maps are important to evaluate soil functions and support decision-making process, particularly for soil properties such as pH, carbon content (C), and cation exchange capacity (CEC), but the spatial resolution and soil depth should meet the needs of users. On another hand, the efficiency of statistical models to create soil maps, with an acceptable level of accuracy, often require a large number of samples with an appropriate distribution across the area of interest. However, accessibility for sampling can be a trouble in remote areas, such as the kabala National Park (INP). The hypothesis of this work is that it is possible to obtain a viable result in soil mapping of areas with limited access by using DSM tools. The general objective of this paper was to create 2- and 3-D maps of the soil properties pH, carbon content, and CEC, with the correspondent spatial uncertainty, in the INP plateau. The sampling strategy was designed using conditioned Latin Hypercube Sample (cLHS), and different methods were tested to produce the soil properties maps. For calibration of the models: linear (MLR, multiple linear regression) and nonlinear (GAM, Generalised Additive Models). The results showed differences in predictive performance for all statistical methods and covariate selection approaches. The GAM, with covariates selection based on soil formation factors, was the best method for the limited number of soil samples. The greatest uncertainty was associated with areas with the lowest accessibility and, consequently, with low sampling density and/or noises in covariates. Even though the 2- and 3-D maps of soil properties, each associated with explicit uncertainty, can contribute to the INP decision makers/managers by providing information not available before.