Assessing soil quality data by positive matrix factorization

Multivariate data analysis techniques such as factor analysis are useful tools in soil quality assessment. In this study, the applicability of a new factor analysis model, positive matrix factorization (PMF), which produces strict non-negative factor loadings and scores, was evaluated. The dataset used was obtained from soil samples taken from a long-term reclaimed wastewater-irrigated cropland and a canal water-irrigated control field. The dataset included 24 measured soil properties (8 physical attributes, 13 chemical attributes, and 3 biological attributes) for each field. The dataset was best modeled by a two-factor model. The first factor was interpreted as the soil component that was coarse-textured, slightly acidic, and physically loose, and the second factor was interpreted as the soil component that was fine-textured, slightly basic, and physically compacted. The factor loadings reflected the characteristic values of the soil attributes for these two components. Further, the new model was able to provide additional information such as cycling patterns of the field variability. Compared to traditional multivariate models, results by PMF model were more mathematically sound and more physically explainable, due to the non-negativity constraints and an individual weighting algorithm devised in the model. Overall, the PMF model proved to be a new multivariate approach that can be useful as a soil quality assessment tool.