Mapeamento digital de atributos do solo no complexo intrusivo de Santa Angélica-ES: arranjos amostrais e distribuição espacial

Abstract

The dissertation was divided into two chapters in which the performance of strategies in the sample design was analyzed through some environmental covariates, analysis and geostatistical modeling of chemical and physical parameters at a mafic rock concentration site in the south of Espírito Santo state, specifically, in the district of Santa Angélica, belonging to the municipality of Alegre-ES. The first chapter was designed to deal with sampling planning in order to acquire statistical representativeness of soil formation factors, for this some environmental covariates (digital elevation map, slope, profile and curvature plane, wetness index, and geology), the Latin Hypercube (LHS), Conditional Latin Hypercube (cLHS), regular and random mesh with two repetitions were tested and analyzed using descriptive statistics and statistical tests of variance (F test), mean (Student's t test), frequency (Kolmogorov-Smirnov; KS) and histogram analysis. In the results was diagnosed a superiority of the LHS in all verifications to the detriment of the others, the random and regular mesh presented reasonable results, possibly associated with a number of points that favored its distribution, and the cLHS that did not generate an expected result, however the literature sustains the inadequacy of this method in rough terrain. In the second chapter, based on the analyzes obtained in the laboratory, soil attribute predictor methods for the Santa Angélica Intrusive Complex (Cisa) were evaluated, including Kriging Regression (RK), Linear Multiple Regression (RLM), Co-kriging and Ordinary Kriging (KO). The covariates used were the same as those of the first chapter, plus geophysical data such as gamma radiation in the uranium, thorium and potassium channels. The criteria that allowed the choice of variables and methods were statistical correlations at a significance level of 0.05 and experimental semivariogram analysis. As a result, it was noted by the root mean square error (RMSE) that for KO sand was superior in all predictions, followed by RK, RLM and co-krigaging, whereas clay followed by KO, RLM, RK and cokrigagem, Al3 + only to KO and RLM, as it had significance in multiple regression and a poor fit on semivariogram, however, it did not present desired spatial covariance. It was generally concluded that digital soil mapping allows satisfactory and important results for better scale detailing. Some interpolators such as KO are simple and effective, however they rely on a sufficient number of samples, which can become costly. On the other hand, RLM and RK depend on a smaller number of points and achieve satisfactory results in the construction of digital attribute maps. The reference values for the local soil in general were framed as average, with higher Mg2+ due to the influence of ferromagnetic rocks.