Evaluation of the performance and limitations of empirical partition-relations and process based multisurface models to predict trace element solubility in soils

Abstract

Here we evaluate the performance and limitations of two frequently used model-types to predict trace element solubility in soils: regression based "partition-relations" and thermodynamically based "multisurface models", for a large set of elements. For this purpose partition-relations were derived for As, Ba, Cd, Co, Cr, Cu, Mo, Ni, Pb, Sb, Se, V, Zn. The multi-surface model included aqueous speciation, mineral equilibria, sorption to organic matter, Fe/Al-(hydr)oxides and clay. Both approaches were evaluated by their application to independent data for a wide variety of conditions. We conclude that Freundlich-based partition-relations are robust predictors for most cations and can be used for independent soils, but within the environmental conditions of the data used for their derivation. The multisurface model is shown to be able to successfully predict solution concentrations over a wide range of conditions. Predicted trends for oxy-anions agree well for both approaches but with larger (random) deviations than for cations.

Assessing the variability of the bioavailable fraction of zinc at the global scale using geochemical modeling and soil archetypes

“…Some authors found that these types of regressions could be applied to a wide range of soil conditions (Groenenberg et al 2010;Rodrigues et al 2010a). When comparing the two approaches, Groenenberg et al concluded that, although empirical regressions can be robust speciation predictors for most cations, they depend on the choice and number of soil properties used and are generally applicable only to the soils that were used to obtain them (Groenenberg et al 2010(Groenenberg et al , 2012. Groenenberg et al showed that models yield satisfactory results for a wide range of conditions and are better predictors of extreme conditions, but the results may vary according to model setup and parameterization (Groenenberg et al 2012).…”

“…Alternatively, Owsianiak et al used empirical regressions instead of speciation models to determine bioavailable fractions through multiple linear regressions performed on field data relating metal speciation (soluble metal concentration or free ion activity) to soil properties (total metal burden, soil pH, and soil organic matter (OM) content) (de Vries and Groenenberg 2009;Groenenberg et al 2010Groenenberg et al , 2012Rodrigues et al 2010a, b;Römkens et al 2004). Some authors found that these types of regressions could be applied to a wide range of soil conditions (Groenenberg et al 2010;Rodrigues et al 2010a).…”

Dynamic Geochemical Models to Assess Deposition Impacts of Metals for Soils and Surface Waters

“…Transfer functions are used to relate the Freundlich adsorption constant to soil parameters like the organic matter content, clay content, content of Fe-oxides and soil solution parameters like the pH and the concentrations of DOM. Groenenberg et al (2012) showed that such an empirical approach is justified for soils in the pH range 4-7.…”

Impact of cadmium levels in fertilisers on cadmium accumulation in soil and uptake by food crops

“…Clearly reliable estimates of leaching losses based on model predictions are prone to model uncertainty but during recent years various robust Cd partition models have been developed and calibrated that allow the user to calculate soil solution concentration based on a combination of soil properties including pH, organic matter and clay content (a.o. Groenenberg et al, 2012;Six and Smolders, 2014). Such models appeared to give reliable results over a wide range of soils when applied to independent data.…”

“…12 using a value of 1 for n. In the current version of Integrator used in this study, the value of n was not a priori assumed to be 1 but instead a non-linear version of the Freundlich model (n ≠ 1) was used as described by Römkens et al (2004) and later on used by a.o. Groenenberg et al (2012) for a number of metals. In this version the variation in the Cd concentration in solution is related directly to that of the soil Cd content and soil properties: 10 logCd solution = Constant + α 2 ·pH + β 2 ·log(Soil Organic Carbon) + γ· 10 log(clay) + δ· 10 log(Cd soil ) [14] The critical difference between both approaches is the consideration of non-linearity in the model used in integrator as represented by the coefficient δ in equation [14] whereas the model used by Smolders implicitly assumes that the value of δ equals 1 which then would render the same model in both cases, that is, if the same soil properties are being used to explain the variation in Cd in solution.…”