Quantifying the Accuracy, Uncertainty, and Sensitivity of Soil Geochemical Multisurface Models

Abstract

Geochemical multisurface models and their generic parameters for the solid-solution partitioning and speciation of metals have been used for decades. For soils the collective uncertainty and sensitivity of model parameters and soil-specific reactive surface properties has been insufficiently evaluated. We used statistical tools and data of diverse soils to quantify for Cd, Cu and Zn the uncertainty of model parameters and input values of the nonideal competitive adsorption (NICA)-Donnan model for organic matter (OM) coupled with the generalized two-layer model for metal-oxides. Subsequently, we quantified the uncertainty of speciation predictions and the sensitivity to model parameters and input values. Importantly, we established new generic NICA-Donnan parameters that substantially improved model accuracy, especially for Zn. Uncertainties generally followed Cu < Cd < Zn. With OM being the major binding surface across most soils, the affinity parameters (log K_i) were most influential. Compared to a “best-case” scenario with all relevant soil properties measured, a “simplified” scenario with assumptions about OM fractionation and metal-oxide specific surface area could be employed with a negligible effect on model accuracy and uncertainty. Our study provides a reference work with quantitative measures of model performance, which facilitates broader adoption of mechanistic multisurface models in addressing environmental challenges.