Comparison of different approaches for modelling heavy metal transport in acidic soils.
MICHEL, K., ROOSE, M. and LUDWIG, B. (2007)
Modelling can be valuable in evaluating potential risks that heavy metals may pose to the environment, e.g. groundwater pollution, but there is no general agreement about the usefulness and limitations of different sorption approaches. Our objective was to test and compare different empirical and mechanistic models with respect to their suitability to describe and predict the transport of heavy metals in soils. Undisturbed columns of the acid subsoils of a sandy Podzol and a silty Cambisol were irrigated for two years with 4 mm day-1 at 8 oC with either relatively small (HM1, 0.1 to 1 mg L-1 corresponding to 0.9 to 19.2 µmol L-1) or markedly increased concentrations (HM2, 0.3 to 3 mg L-1 corresponding to 2.7 to 57.7 µmol L-1) of the heavy metals Pb, Cr(III), Cu, Ni and Cd. PHREEQC2 was used to calculate one-dimensional transport and homogenous and heterogenous reactions. The following models were tested: Sorption isotherms considering concentrations (Freundlich-1, Langmuir-1) or ion activities (Freundlich-2, Langmuir-2), cation exchange (CE) and competitive sorption (CSM)). Furthermore, two different calibration approaches were evaluated: a calibration procedure using the data of experiment HM2 followed by an independent validation (HM1) (approach A) was compared to a combined calibration using both data sets for parameter optimisation (approach B). Parameters were optimised for all models using the parameter optimization program PEST. To evaluate the performance of the different models the modelling efficiency (EF) which provides a comparison of the efficiency of the respective model to the efficiency of describing the data as the mean of the measured data was used. Copper and Pb were retained in all columns to more than 99 %, whereas Cd and Ni reached concentrations up to 27 µmol L-1 (HM2) in solution. The modelling efficiency values for prediction (EFpredA) did not simply follow the order according to the number of adjustable parameters in the different models [CSM (3) > Freundlich, Langmuir (2) > CE (1)]. Model CSM predicted transport behaviour of Ni and Cd in the Podzol better than the best empirical model approach as reflected by higher EFpredA values (CSM: 0.85 (Ni) and 0.76 (Cd) vs. 0.60 (Ni; Freundlich-2) and 0.74 (Cd; Freundlich-2)). In the Cambisol, Cd transport dynamic was better predicted than Ni transport dynamic by all models tested. The best prediction was obtained by using Freundlich isotherms (EFpredA = 0.72 and 0.73 for Freundlich-1 and Freundlich-2, respectively). As expected, independently of the model used, combined calibration of the HM1 and HM2 data generally improved the agreement between modelled and measured Ni and Cd concentrations (HM1) in both soils, where the prediction in approach A was only satisfactorily (EFpredA < 0.60). Our results indicated that neither the application of mechanistic nor empirical models can be generally recommended for all soils. Thus, calibration experiments appear to be generally required for a subsequent accurate prediction of independent experiments. PEST was well-suited for the parameter optimization of all calibration models.