APPLICATION OF RIDGE REGRESSION TO QUANTIFY MARGINAL EFFECTS OF COLLINEAR SOIL PROPERTIES ON PHYTOACCUMULATION OF ARSENIC, CADMIUM, LEAD, AND ZINC

Soil properties that mitigate hazardous effects of environmental contaminants through soil chemical sequestration should be considered when evaluating ecological risk from terrestrial contamination. The objective of this research was to identify predominant soil chemical/physical properties that modify phytoaccumulation of As, Cd, Pb, and Zn to the non-hyperaccumulating higher plants: Alfalfa (Medicago sativa L.), perennial ryegrass (Lolium perenne L.), and Japanese millet (Echinochloa crusgalli L.). Transmission coefficients were estimated from a dose-response experiment with the use of aboveground tissue contaminant concentrations and correlated with selected soil property measurements to develop statistical prediction models for soil-specific adjustments to ecological risk assessments. Significant correlations between soil properties and transmission coefficients were observed for all four contaminants. Intercorrelation was also observed among soil properties, including cation exchange capacity (CEC) and soil pH (p = 0.035), CEC and total clay (p = 0.030), organic carbon (OC) and total clay content (p = 0.085), reactive iron oxides (FeOX) and OC (p = 0.078), and reactive Mn oxide (MnOX) and total clay content (p < 0.001). Ridge regression, a technique that suppresses the effects of multicollinearity and enables prediction, was used to assess the marginal contributions of soil properties found to mitigate phytoaccumulation. Prediction models were developed for all four contaminants. Significant variables were FeOX for As or pH, OC, CEC, clay content, or a combination of factors for cationic metal models. Ridge regression provides a powerful alternative to conventional multiple regression techniques for ecotoxicological studies when intercorrelated predictors are experimentally unavoidable, as with soil properties.