Modeling Metal Binding to Soils: The Role of Natural Organic Matter

To model metal binding to natural materials, one ap proach is to use the so-called ... in which the NICA-Donnan and CD-MUSIC models were used for the ...
0 downloads 0 Views 216KB Size
Environ. Sci. Technol. 2003, 37, 2767-2774

Modeling Metal Binding to Soils: The Role of Natural Organic Matter JON PETTER GUSTAFSSON* AND P A V L I N A P E C H O V AÄ Department of Land and Water Resources Engineering, KTH (Royal Institute of Technology), SE-100 44 Stockholm, Sweden DAN BERGGREN Department of Soil Sciences, Swedish University of Agricultural Sciences, Box 7014, SE-750 07 Uppsala, Sweden

The use of mechanistically based models to simulate the solution concentrations of heavy metals in soils is complicated by the presence of different sorbents that may bind metals. In this study, the binding of Zn, Pb, Cu, and Cd by 14 different Swedish soil samples was investigated. For 10 of the soils, it was found that the Stockholm Humic Model (SHM) was able to describe the acid-base characteristics, when using the concentrations of “active” humic substances and Al as fitting parameters. Two additional soils could be modeled when ion exchange to clay was also considered, using a component additivity approach. For dissolved Zn, Cd, Ca, and Mg reasonable model fits were produced when the metal-humic complexation parameters were identical for the 12 soils modeled. However, poor fits were obtained for Pb and Cu in Aquept B horizons. In two of the soil suspensions, the Lund A and Romfartuna Bhs, the calculated speciation agreed well with results obtained by using cation-exchange membranes. The results suggest that organic matter is an important sorbent for metals in many surface horizons of soils in temperate and boreal climates, and the necessity of properly accounting for the competition from Al in simulations of dissolved metal concentrations is stressed.

Introduction The understanding of the processes responsible for the binding of metal contaminants in soils is important in order to correctly assess the risk for plant uptake and leaching. Organic matter is an important metal sorbent in the surface horizon of soils (1-3). There are, however, also other soil colloids capable of scavenging metals. Oxide-type components (such as Fe oxide or allophane/imogolite) may sorb metals through surface complexation. Clay minerals (examples: montmorillonite, vermiculite) bind metals both through ion exchange and surface complexation. Finally, various precipitation and coprecipitation reactions may occur, mostly at large equilibrium metal concentrations and at high pH (1). To model metal binding to natural materials, one approach is to use the so-called component additivity (CA) approach in which metal sorption to different sorbing components (e.g. Fe oxide, humic substances, clay) is treated separately with different submodel descriptions; contributions from the different components are summed to give the * Corresponding author phone: +46-8-7908316; fax: +46-84110775; e-mail: [email protected]. 10.1021/es026249t CCC: $25.00 Published on Web 05/16/2003

 2003 American Chemical Society

overall metal sorption of the sample (4). Recent examples include an effort to simulate the speciation of Cu and Pb in the Humber rivers using the SCAMP model (5) and one study in which the free soil solution concentrations of various metals were simulated using the ECOSAT program, in which the NICA-Donnan and CD-MUSIC models were used for the humic and Fe oxide component, respectively (6). Moreover, Cd sorption could be described for an acidic B horizon using a combination of the NICA-Donnan model and the GainesThomas equation for cation exchange to clay minerals (7). There are, however, numerous difficulties with applying CAtype models for metals. One of the most critical ones is that it is difficult or even nearly impossible to verify experimentally the calculated sorption contributions from different components. Other modeling work has focused on the identification and subsequent metal sorption modeling of the most important sorbing component. Such a model may not be valid for all soils, but a clear advantage is that model calibration and verification is much more straightforward. The study of Hesterberg et al. (8) represents an early attempt to predict soil solution concentrations of Zn, Cd, and Cu in the soil solutions of Dutch sandy soils in which organic matter was assumed to be the main sorbent. Benedetti and coworkers (9) used the NICA-Donnan model to simulate Cu and Cd binding in a mountain lake and in two sandy A horizons. In recent work by our own group, it has been shown that the Stockholm Humic Model (SHM) can describe satisfactorily the dissolved concentrations of Zn, Pb, Cu, and Cd in mor layers of forest soils and that the competitive effect from Al on Zn sorption can be simulated (10). The latter model results indicated that organic matter may be the dominant sorbent in certain surface horizons, such as the O horizon of Spodosols and humus-rich A horizons in sandy soils. However, the variability between different soils in terms of metal binding is still poorly known; for example, it is still not known for what range of soils the consideration of organic matter only would suffice for an assessment of the metal sorption properties. In addition it is not known whether organic matter in different soil environments possesses different metal binding characteristics. For these reasons, previous modeling results cannot easily be generalized to other soils. The overall objective of this study was to examine whether the use of the SHM alone could describe proton binding and dissolved metal concentrations in a variety of soils. This would help to identify soils for which a “simpler” CA-type model, involving only organic matter and clay, would suffice and soils for which a more complete CA model, including oxide phases, would be necessary.

Materials and Methods Soil Samples. We report results on solid-phase chemistry and batch experiments for 14 field-collected soil samples originating from 11 different locations in Sweden (10 locations were in central Sweden and the 11th, Ersna¨s, was from northeastern Sweden). Selected soil characteristics are listed in Table 1 (for additional characteristics see Table S1, Supporting Information). Two samples were O horizons from Spodosols (mor humus consisting of the Oa + Oe layers), 6 were A horizons from a range of different parent materials, 2 were B horizons from acid sulfate soils (Aquepts (11)), and the remaining 4 were B horizons from Spodosols. The soils were rather low in extractable Zn, Pb, Cu, and Cd (Table S1, Supporting Information). VOL. 37, NO. 12, 2003 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

2767

TABLE 1. Selected Properties of the Investigated Soil Samples soil

pH(H2O)

total C, %

BaCl2-Ca, mmol kg-1

BaCl2-Mg, mmol kg-1

Alp/Cp, mol mol-1

clay, %a

soil order

Risbergsho¨ jden O Tyresta O Igelba¨ cken A Kista E4 A Bjo¨ rkby sand A Tyresta A Lund A Bjo¨ rkby clay A Brantshammar B Ersna¨ s B Tyresta Bs Romfartuna Bhs Kloten Bs2 Risbergsho¨ jden Bs2

3.8 3.8 5.0 7.0 4.5 5.0 6.0 6.0 3.8 4.9 4.4 4.8 5.0 5.7

47.9 44.4 25.2 10.8 6.42 4.99 3.68 2.60 5.47 1.42 4.44 2.05 3.86 0.80

60 39 242 170 20.6 14.2 57 93 21.3 18.8 0.47 0.09 0.25 0.21

9.2 9.4 20.3 13.5 4.0 2.1 5.8 17.5 11.6 7.0 0.31 0.19 0.12 0.11

0.008 0.016 0.026 0.027 0.027 0.084 0.039 0.026 0.034 0.008 0.10 0.11 0.19 0.31

0 0 35 36 14 23 14 52 65 20 21 4 8 5

spodosol spodosol inceptisol inceptisol entisol inceptisol entisol inceptisol inceptisol inceptisol spodosol spodosol spodosol spodosol

a

Percent of total dry mass of soil.

After collection, the samples were stored in a cold room at +5 °C until further use. Before the experiments the samples were sieved ( 5.2. However, in three of the soils, Bjo¨rkby sand, Ersna¨s, and Igelba¨cken, the Alqr concentrations “plateaued” at 6, 3.5, and 3.7 µM, respectively, and did not decrease below this value at higher pH. Also for Tyresta Bs, the Alqr concentration at the highest observed pH was unexpectedly large, i.e., larger than the value predicted from Al(OH)3 solubility. As a consequence, the slope of the observed pH vs log {Al3+} at high pH for these four soils was clearly different not only from that of the other soils but also from the expected slope predicted by SHM (Figure 1). We are not able to explain why 2770

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 37, NO. 12, 2003

the pH - log {Al3+} slope at high pH would be much different for these four soils compared to the other ones. It seems likely that the Alqr concentrations at high pH in these four soils contained organically bound Al, i.e., that the Alqr method did not always work in the intended way. To filter out this effect from the model optimizations, observations leading to log {Al3+} < -5.5 were excluded from further treatment for these four soils. When the high-pH data for these four soils were disregarded, it was seen that the slopes of pH vs log{Al3+} were mimicked well with the bidentate reaction used in the model, which supports the hypothesis that two protons are released for every Al sorbed in the Al complexation reaction (26). Moreover, as most investigated suspensions were undersaturated with respect to Al hydroxide, Figure 1 clearly illustrates that dissolved Al is controlled by complexation to organic matter in a wide variety of soils. This is true not only for acid horizons of Spodosols (20, 27, 28) but also, apparently, for other sandy and silty A horizons, B horizons from Aquepts, and maybe even for many surface horizons from clay soils, even though the postglacial clay soil from Bjo¨rkby used in this study may not be representative for other clays. Simulations of Dissolved Metal Concentrations. Results from SHM fits are shown in Figures 2 and 3. These fits are predictions, based on the metal complexation constants optimized for Risbergsho¨jden O (10), and optimization of the HA, FA, and Al concentration for the individual soils. Examples of fits for individual soils are also shown in the Supporting Information, Figures S3, S4, S5, and S6. The RMSE value for Ca was as large as 0.16, despite the use of Ca in the optimization, see Table 4. The relatively poor fit may have been affected by the weathering of Ca that seems to have occurred in some soils at low pH, Lund A, Bjo¨rkby clay A, and Brantshammar B. If these three soils are disregarded, RMSE for Ca is decreased to below 0.1. Magnesium was better described by the model (RMSE ) 0.11). These results show that the BaCl2 extraction appears to be a good indicator for active Ca and Mg. For Zn and Cd, the SHM simulations in most cases agreed rather well with the observations, (Figures 2 and 3) and the overall RMSE values were 0.13 and 0.18, respectively (Table 4). However, the overall fit was improved if two of the A horizons, Kista E4 and Igelba¨cken, were excluded from the optimizations (resulting in RMSE ) 0.090 and 0.13 for Zn and Cd, respectively). For these A horizons, the model underestimated the concentration of Zn and Cd. The reason for this is unknown. One possibility is that the organic matter of the two soils contained more high-affinity sites for Zn and Cd complexation than the other soils. However, because the overall RMSE was < 0.2, it was concluded that the same set of metal complexation constants could be used to describe the dissolved Zn and Cd concentrations in all soils reasonably well. In other words, the organic matter in the different soil environments investigated appeared to behave similarly in terms of Zn and Cd binding. In addition, the model results re-emphasized the importance of Al competition for dissolved Zn and Cd, as was shown also in our preceding paper (10). Although there are many other factors that explain a particular soil’s relative affinity for Zn and Cd (such as organic matter content), it can be seen in Figures 1 and 2 that in general, soils with little soluble Al at a given pH also retain Zn or Cd more strongly. Although the model worked well for Cd at the relatively large Cd concentrations employed, it should be noted that some additional experiments were performed using variable initial Cd concentrations for the Lund and Romfartuna soils. These experiments showed that the SHM predicted larger equilibrium Cd concentrations than was observed at very small (but environmentally relevant) total Cd concentrations, whereas the model performed well at intermediate and large

FIGURE 1. The log {Al3+} plotted against pH for systems with added Cd (Lund, Romfartuna, and Kloten soils) or Zn (other soils). Observations are shown as points, whereas the SHM fits are shown as solid lines. The dashed line indicates the solubility of an Al(OH)3 phase with a log *Ks of 9.4 at 8 °C (19).

FIGURE 2. Percentage sorbed Cd (Lund, Romfartuna, and Kloten soils) or Zn (other soils). Observations are shown as points, whereas the SHM fits are shown as lines. total Cd concentrations (Figure S7, Supporting Information). This indicates that the sorption of Cd to the high-affinity sites, which become important at small equilibrium concentrations, was not simulated very well in the present SHM application. Model tests were made using a larger ∆LK2 value for Cd, but this did not improve the fit considerably. For Cu and Pb, there was a pronounced difference between the Aquept B horizons and the other soils, as is seen in Figure 3. Whereas the simulations of Cu and Pb binding were rather successful in the latter group of soils (with RMSE

) 0.14 for both metals, see Table 4), Cu and Pb were predicted very poorly for the Aquepts. The SHM underestimated the dissolved Pb and Cu concentrations in these soils, for unknown reasons. Possibly, the organic matter of marine origin that was present in the Aquept B horizons may not bind Cu and Pb as strongly as the organic matter of the other soils. In the two soils with Al(OH)3 solubility control, SHM was not used to simulate metal sorption, as the fHS-tot could not be optimized. For these soils we made preliminary fits using VOL. 37, NO. 12, 2003 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

2771

FIGURE 3. Comparison of the simulated values of log [Zn]tot, log [Pb]tot, log [Cd]tot, and log [Cu]tot with measurements. The line indicates the perfect 1:1 fit. the average value of fHS-tot determined for the B horizons without Al(OH)3 solubility control (data not shown). However, metal binding was overestimated considerably. Possibly this may have been caused by the inactivation of humic functional groups following adsorption to oxide surfaces (29), which are present in large concentrations in the two soils in question. In addition, the importance of other metal sorption processes, such as oxide sorption or the formation of layered double hydroxides (30), cannot be assessed. It is of interest to note, however, that despite their large oxide content, the metal sorption at a given pH was much weaker than in the other soils (see results for Zn/Cd in Figure 2), showing the overall importance of organic matter for metal binding in soils. With the use of a more complete CA-type model, such as SCAMP (5) or the combined NICA-Donnan/CD-MUSIC (6), the importance of the oxides as sorbent phases may be evaluated from oxalate-extracted Fe and Al shown in Table S1, if certain rather crude assumptions are used. For example, using the assumption that the properties of Fe oxide can be approximated with those of 2-line ferrihydrite (31), it would probably be shown that Fe oxide is an important metal sorbent for several soils of this study, at least at high pH. However, the validity of such an assumption can certainly be questioned. It seems far from certain that all oxalateextracted Fe has the same large site concentration as 2-line ferrihydrite. In this perspective, we believe that it is of interest to see that a simpler CA-type model accounting for organic matter and clay appears to be able to simulate metal binding in some soils, without the need for crude assumptions as regards the properties of the oxide phases. Given the model results it seems possible, however, that oxide phases may be important also in these soils but at higher pH where metal sorption is close to 100%. Speciation of Cd and Pb with the Cation-Exchange Membrane Technique. Figures 4 and 5 show the results obtained, along with the SHM fits. To simulate the speciation in SHM/Visual MINTEQ, we used the total dissolved organic carbon concentration as an input parameter, assuming it to consist of 100% FA (as in the batch experiment). The major trends were clearly captured by the model, indicating that 2772

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 37, NO. 12, 2003

FIGURE 4. Solution speciation of Cd in the filtrates from the Lund A and Romfartuna Bhs soils (I ) 0.1 M). Points represent observations using the cation-exchange membrane technique, the solid line connects averaged concentrations, whereas the dotted line is the SHM fit for the Cd2+ concentration. the aqueous speciation in the batch experiment suspensions may have been well simulated. According to both methods, most of the dissolved Cd was Cd2+ over most of the pH range. Organic complexes were minor except at the highest pH. For Pb, both methods showed that half of the dissolved Pb was Pb2+ at pH < 5 (the Visual MINTEQ results showed that an additional 30-40% was PbNO3+ at the high ionic strength employed). At pH > 5, organic complexes became important and dominated completely above pH > 6. These results also demonstrate the robustness of the cation-exchange membrane method. If properly used, this method may offer a

be the same for all types of organic matter. To be able to use generic parameter values for metal-humic complexation, this aspect needs to be studied. Also, a possible future development of SHM is to include the option to simulate the partitioning of organic matter between solid and dissolved phases.

Acknowledgments This study was funded by the Swedish Research Council (VR) and by the Foundation for Strategic Environmental Research (MISTRA). Gunilla Lundberg, Lise Gustafsson, Gunilla Hallberg, and Kenth Andersson made some of the laboratory analyses. Gustav Sohlenius is acknowledged for providing the Ersna¨s and Brantshammar samples. Steve Lofts and Ingrid O ¨ born are thanked for many valuable discussions during the course of this project.

Supporting Information Available

FIGURE 5. Solution speciation of Pb in the filtrates from the Lund A and Romfartuna Bhs soils (I ) 0.1 M). Points represent observations using the cation-exchange membrane technique, the solid line connects averaged concentrations, whereas the dotted line is the SHM fit for the Cd2+ concentration. cheap and attractive alternative for metal speciation in natural waters. Studies are underway to develop this technique further, i.e., to improve the relationships to take account for competing ions and to investigate if the membranes may be used successfully in solutions of low ionic strength (5% organic matter, and at rather low pH where metal sorption is lower than 100%. The results also demonstrate that the competitive effect of Al on the sorption of other metals has to be considered. Because organic complexation determines dissolved Al in many soils, the use of a too simplified approach to estimate Al solubility in simulations (such as the gibbsite solubility model) may lead to irrelevant or erroneous results concerning the dissolved concentrations of heavy metals. A number of problems remain before our model approach can be used on a routine basis. For example, methods have to be developed to estimate active HA, FA, Al, and heavy metals from available extraction procedures. The optimized values of HA, FA, and Al do not show any clear relationship with the extraction methods used in the present study. If an extraction method was found that more closely matches the active HA, FA, and Al, it would enable better generic model descriptions, requiring less input data in the model for the user. The exact role of Fe(III) needs to be elucidated in more detail, to allow simulations in which all important competing ions are considered. The poor fits for Pb and Cu in two of the soils, and the less satisfactory fits for Cd and Zn in two other soils, suggest that the affinity of these metals may not

Observed dissolved organic carbon concentrations (Figure S1); SHM fits to the acid-base characteristics (Figure S2); detailed results for Ca, Zn, and Pb binding in the Tyresta A, Bjo¨rkby clay A, and Ersna¨s B soils (Figures S3-S5); comparison of the simulated values of [Ca]tot and [Mg]tot with measurements (Figure S6); SHM fit to the Cd sorption isotherms for Romfartuna Bhs and Lund A (Figure S7); additional soil properties (Table S1); number of data points and pH ranges in the batch experiments (Table S2); SHM parameter values for proton binding (Table S3). This material is available free of charge via the Internet at http:// pubs.acs.org.

Literature Cited (1) Brown, G. E.; Parks, G. A. Int. Geol. Rev. 2001, 43, 963-1073. (2) Nakhone, L. N. Discov. Innovat. 1997, 9, 53-71. (3) Strobel, B. W.; Hansen, H. C. B.; Borggaard, O. K.; Andersen, M. K.; Raulund-Rasmussen, K. Geochim. Cosmochim. Acta 2001, 65, 1233-1242. (4) Davis, J. A.; Coston, J. A.; Kent, D. B.; Fuller, C. C. Environ. Sci. Technol. 1998, 32, 2820-2828. (5) Lofts, S.; Tipping, E. Sci. Total Environ. 2000, 251, 381-399. (6) Weng, L.; Temminghoff, E. J. M.; Van Riemsdijk, W. H. Environ. Sci. Technol. 2002, 35, 4436-4443. (7) Voegelin, A.; Vulava, V. M.; Kretzschmar, R. Environ. Sci. Technol. 2001, 35, 1651-1657. (8) Hesterberg, D.; Bril, J.; del Castilho, P. J. Environ. Qual. 1993, 22, 681-688. (9) Benedetti, M. F.; Van Riemsdijk, W. H.; Koopal, L. K.; Kinniburgh, D. G.; Gooddy, D. C.; Milne, C. J. Geochim. Cosmochim. Acta 1996, 60, 2503-2513. (10) Gustafsson, J. P.; Van Schaik, J. W. J. Eur. J. Soil Sci. 2003, 54, in press. (11) O ¨ born, I. Geoderma 1989, 45, 197-219. (12) Jou, A. S. R.; Kamprath, E. J. Soil Sci. Soc. Am. J. 1979, 43, 35-38. (13) Van Reeuwijk, L. P. Procedures for soil analyses, 5th ed.; International Soil Reference and Information Centre: Wageningen, Netherlands, 1995. (14) Clarke, N.; Danielsson, L.-G.; Spare´n, A. Int. J. Environ. Anal. Chem. 1992, 48, 77-100. (15) Berggren, D.; Spare´n, A. Int. J. Environ. Anal. Chem. 1996, 62, 115-128. (16) Berggren, D. Int. J. Environ. Anal. Chem. 1990, 41, 133-148. (17) Liang, J.; Schoenau, J. J. Commun. Soil Sci. Plant Anal. 1996, 27, 3013-3026. (18) Gustafsson, J. P. http://www.lwr.kth.se/english/OurSoftware/ vminteq/index.htm. (19) HydroGeoLogic Inc.; Allison Geoscience Consultants Inc. MINTEQA2/PRODEFA2, a geochemical assessment model for Environmental systems: User Manual Supplement for Version 4.0; U.S. Environmental Protection Agency: Athens, GA, 1999. (20) Gustafsson, J. P.; Berggren, D.; Simonsson, M.; Zysset, M.; Mulder, J. Eur. J. Soil Sci. 2001, 52, 655-665. (21) Gustafsson, J. P. J. Colloid Interface Sci. 2001, 244, 102-112. (22) Lofts, S.; Woof, C.; Tipping, E.; Clarke, N.; Mulder, J. Eur. J. Soil Sci. 2001, 51, 189-204. (23) Peters, A. J.; Hamilton-Taylor, J.; Tipping, E. Environ. Sci. Technol. 2001, 35, 3495-3500. VOL. 37, NO. 12, 2003 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

2773

(24) Tipping, E.; Rey-Castro, C.; Bryan, S. E.; Hamilton-Taylor, J. Geochim. Cosmochim. Acta 2002, 66, 3211-3224. (25) Gustafsson, J. P. J. Environ. Qual. 1995, 24, 882-888. (26) Simonsson, M. Eur. J. Soil Sci. 2001, 51, 655-666. (27) Berggren, D.; Mulder, J. Geochim. Cosmochim. Acta 1995, 59, 4167-4180. (28) Simonsson, M.; Berggren, D. Eur. J. Soil Sci. 1998, 49, 317-326. (29) Filius, J. D.; Lumsdon, D. G.; Meeussen, J. C. L.; Hiemstra, T.; Van Riemsdijk, W. H. Geochim. Cosmochim. Acta 2000, 64, 5160.

2774

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 37, NO. 12, 2003

(30) Voegelin, A.; Scheinost, A. C.; Bu ¨ hlmann, K.; Barmettler, K.; Kretzschmar, R. Environ. Sci. Technol. 2002, 36, 3749-3754. (31) Dzombak, D. A.; Morel, F. M. M. Surface complexation modeling. Hydrous ferric oxide; Wiley & Sons: New York, 1990.

Received for review October 18, 2002. Revised manuscript received April 1, 2003. Accepted April 17, 2003. ES026249T