Multivariate and geostatistical analyses to evaluate lowland soil levelling effects on physico-chemical properties

• The study shows the potential of multivariate analysis to evaluate land levelling.
• Geostatistical tools can be used to evaluate impacts caused by levelling.
• Levelling negatively affected lowland soil surface quality.
• Levelling affected spatial relationships among soil properties.
• Multivariate analysis can be used for the recuperation of degraded soils by levelling.

Levelling is used in rice fields to correct land irregularities for management practice improvement. Effects of levelling have been evaluated using univariate tools from classical statistics and geostatistics, which disregard possible effects on interrelations among soil properties. This study aimed to: (i) evaluate the effect of levelling on soil physico-chemical properties using classical statistics and geostatistical tools, (ii) group soil properties in homogeneous sets using multivariate cluster analysis, (iii) reduce the dimensionality of sets by principal component analysis, and (iv) characterize and map spatial distributions of principal components to compare them with individual maps to elaborate maps of homogeneous management zones. On a 1.0 ha field in Southern Brazil a grid of 10 × 10 m points was used to sample the 0–0.20 m soil layer, before and after levelling. All data sets were submitted to descriptive statistics and the Kolmogorov–Smirnov (K–S) test, as well as to geostatistical analysis. Maps were constructed using ordinary kriging or the inverse square distance interpolation. Groups were formed by cluster analysis. To reduce their dimensionality principal component analysis and geostatistics were used to evaluate spatial dependencies of the main principal components. Maps of the main principal components and of each soil property were visually compared to verify if they had similar patterns of spatial variability. Levelling caused an increase of the spatial range for the majority of the properties as well as an improvement of the goodness of fit of the semivariogram models. Variability structures of the majority of the properties were best modeled by Gaussian and spherical models. Levelling also caused changes in the variance and covariance structures of all properties as detected by cluster multivariate analysis. Soil bulk density and organic matter represented the sets that were most affected by land levelling, so that their maps are considered useful and accessible tools for farmers in future operations of land recuperation.