The use of hard- and soft-modelling to predict radiostrontium solid–liquid distribution coefficients in soils

The solid–liquid distribution coefficient (Kd) is the parameter that governs the incorporation of contaminants in soils. Its estimation allows the prediction of the fate of contaminants in the short- and long-term after a contamination event. Here, the Kd of radiostrontium (Kd(Sr)), a radionuclide of significant environmental interest, was predicted by hard models, which are based on knowledge of the mechanisms governing its sorption, and by soft models based on Partial Least Squares (PLS), using a large data set with the main soil parameters. The two approaches were tested and compared for 30 soils in Spain. Correlations between the predicted and experimental values of Kd(Sr) obtained using hard- and soft-modelling showed slopes close to 1 and regression coefficients higher than 0.95, which confirms that both approaches are able to obtain satisfactory estimates for Kd(Sr) from soil parameters.

► Mechanistic and PLS-based models are compared to predict the Kd(Sr) in soils.
► A simple mechanistic model is proposed, based on CEC and Ca–Mg status in the soils.
► The PLS calibration model is constructed with 20 soil variables and 30 soils.
► The external validation of the PLS model confirms its robustness and validity.
► Both models succeed in describing 90% or more of the Kd(Sr) variance, with no bias.