Multivariate analysis applied to reduce the number of predictors in digital soil mapping

Abstract

The objective of this work was to assess the possibility of generating a smaller set of uncorrelated predictors, potentially applicable to digital soil mapping, by multivariate statistical analysis. The terrain attributes, elevation, slope, stream distance, planar curvature, profile curvature, relative available radiation, natural logarithm of the contributing area, topographic wetness index, and sediment transport capacity, were transformed by the Varimax method into the variables: altimetry, hydrology, and curvature. This transformation represented a variability concentration of 65.57% of the original data in the three new components. The new variables enable the use of a smaller amount of data set in prediction models, besides the fact that they are uncorrelated. Varimax rotation allows the relationship between environment and soil formation to be explicitly included in the prediction models.