Spatial Characterization of Cone Index and Some Nutrients in a Sandy Loam Soil (Eutric Leptosol) Using the Multivariate Analysis

A multivariate analysis was performed on some soil nutrient and Cone Index (CI) data from the research and demonstration farm of the Dept. of Agricultural Sciences, University of Juba in South Sudan. The main objective of the study was to characterize the spatial distribution of the soil nutrients: N, P, K, Fe and Mn as well as soil penetration resistance CI. A Principal Component Analysis (PCA), Gaussian Mixture Model (GMM) and Hierarchical Cluster Analysis (HCA) were performed on the analyzed samples. The Bayesian Information Criterion (BIC) was used for model selection between the Equal size, Equal shape and Equal orientation (EEE) and Equal size, Equal shape and Variable orientation (EEV) models which defined the size, shape and orientation of the ellipsoid with full covariance matrices. Eigenvalues of the three major principal components F1, F2 and F3 accounted for 75.67% of the total variance of the data. From hierarchical clustering, P was observed to cluster with Fe, Mn with N which at second level clustered with K then with CI. The results of the PCA showed that Nitrate-N, Mn and Kwere strongly influenced by CI and so determining their spatial distribution. This could be associated mainly to earlier anthropogenic activities on the soil. The results of this study also showed spatial relationships between individual soil nutrients with both K and P mutually antagonistic with Nitrate-N, whereas between K and P where mutually synergistic. While P was strongly adsorbed to Fe, this was associated to lithogenic soil materials and therefore interpreted as derived from natural sources of the Eutric Leptosol. The goodness-of-fit test using the Kolmogorov-Smirnov (KS) showed that the values of the variables: CI, K, P and Fe were significant at p ≤0.05 and that the data followed normal distribution, whereas Mn and Nitrate-N were not. The KS test also corroborated the results of strong spatial dependency of each variable at less than 25%. The multivariate GMM adequately described the spatial distribution of all measured variables than the unimodal Gaussian.