Desempenho de diferentes arranjos amostrais e modelos na espacialização de cátions trocáveis do solo

Abstract

Soil sampling for the mapping of its properties can be performed by a variety of arrangements with different levels of complexity. Considering the importance of exchangeable soil cations for crops, it is necessary to describe their distribution in space and time in the different agroecosystems. Among the variety of sampling arrangements used in the digital mapping of exchangeable soil cations, the Regular grid sampling (R), the Spatial Coverage sampling (S) and the Latin Conditioning Hypercube sampling (cLHS) stand out. Therefore, the objective of this research is to test whether the performance of the cLHS sampling arrangement that uses environmental covariates, when compared to the regular grid and spatial coverage arrangements, will increase the accuracy in the digital mapping of exchangeable soil cations, carried out by two methods of spatial modeling (geostatistics and mixed linear model), in an area of grain cultivation under central pivot irrigation. The study area covers 160 hectares. For the predictions of the spatial distribution of the exchangeable cations of the soil, the three sampling arrangements, regular grid, spatial coverage and conditioned Latin hypercube were used. The spatial predictions made in the ArcMap® software were used, using Krigagem, and also, mixed linear model in the software R. For the geostatistics the sampling arrangement that presented better accuracy in the prediction of Al and K was the mesh. R, the lowest predictive performance was presented in the S grid. In the case of Ca, the sampling arrangement that had the highest performance was that of the R mesh and the lowest was of the cLHS mesh. Finally, for Mg, the best performance was the S mesh, and the lowest performance was the R mesh. The sampling arrangement that showed the best accuracy in the prediction of exchangeable soil cations using the mixed linear model was the cLHS mesh model. It was observed that the mesh S obtained a good predictive performance, while, the lowest performance was of the R mesh sampling arrangement. Thus, the Hipercubo Latino Conditioned sampling arrangement proved to be superior to the other tested arrangements.