Generalized linear models and multivariate analysis applied to predict soil spatial distribution in south Brazil

摘要

Digital Soil Mapping (DSM) is an interdisciplinary science involving soil, statistics, mathematics, and geomatics knowledge applied to generate soil spatial information. This study aimed to use Principal Component (PC) as covariates in logistic modelsfor the prediction of soil classes in the south of Brazil. Principal Component Analysis (PCA) was applied to nine terrain attributes: elevation, slope, distance to the nearest stream; planar curvature, profile curvature, radiation index, natural logarithm of contributing area, topographic wetness index and sediment transport capacity. The retained PC was used as explanatory covariates in Multiple Logistic Regressions (MLR), which were trained with soil information provided by an available soil map on 1