ARTICLE pubs.acs.org/est
Competitive and Synergistic Effects in pH Dependent Phosphate Adsorption in Soils: LCD Modeling Liping Weng,* Flora Alonso Vega, and Willem H. Van Riemsdijk Department of Soil Quality, Wageningen University P.O. Box 47, 6700 AA, Wageningen, The Netherlands
bS Supporting Information ABSTRACT: The pH dependency of soluble phosphate in soil was measured for six agricultural soils over a pH range of 3 10. A mechanistic model, the LCD (ligand charge distribution) model, was used to simulate this change, which considers phosphate adsorption to metal (hydr)oxides in soils under the influence of natural organic matter (NOM) and polyvalent cations (Ca2+, Al3+, and Fe3+). For all soils except one, the description in the normal pH range 5 8 is good. For some soils at more extreme pH values (for low P-loading soils at low pH and for high P-loading soils at high pH), the model over predicts soluble P. The calculation shows that adsorption is the major mechanism controlling phosphate solubility in soils, except at high pH in high P-loading soils where precipitation of calcium phosphate may take place. NOM and polyvalent cations have a very strong effect on the concentration level of P. The pattern of pH dependency of soluble P in soils differs greatly from the pH effects on phosphate adsorption to synthetic metal (hydr)oxides in a monocomponent system. According to the LCD model, the pH dependency in soil is mainly caused by the synergistic effects of Ca2+ adsorption to oxides. Adsorption of Al3+ to NOM adsorbed plays an important role only at a pH < 4.5. Presence of NOM coating strongly competes with phosphate for the adsorption and is an important factor to consider in modeling phosphate adsorption in natural samples.
’ INTRODUCTION Behavior of phosphorus (P) in soils has been for a long time, and still is, an important topic of study. On one hand, P is an essential macronutrient for crops and is often a limiting factor in agricultural production. But on the other hand, over fertilization has resulted in accumulation of P in soils, which can lead to environmental problems such as eutrophication of surface waters.1,2 The most significant P compound in soil in terms of bioavailability is the orthophosphate anion. Phosphate ions can adsorb onto positively charged minerals such as Fe and Al (hydr)oxides and the edges of clay minerals. Phosphate containing minerals can also form. In acidic conditions, Fe and Al phosphates may be present, whereas in neutral to alkaline soils Ca phosphates are more likely to occur.3,4 In his review on P bioavailability,3 Hinsinger concluded that “In spite of the large attention that phosphorus has received over decades of intensive research in the 20th century, the mobility of inorganic phosphorus in most soils is still rather poorly understood and hardly predictable”. The common belief is that adsorption process controls solidsolution distribution of phosphate over most of the natural pH range.3,5 The mechanisms of phosphate adsorption to synthetic metal (hydr)oxides are very well understood.6 Sophisticated surface complexation models, such as the CD-MUSIC model,7,8 have been developed. Using consistent parameter sets, phosphate adsorption in both mono- and multi-inorganic component synthetic systems can be very well predicted by the model.6,9 r 2011 American Chemical Society
However, applying these surface complexation models to predict anion adsorption and speciation in natural samples remains a big challenge. Major difficulties that hamper this “model to field” step include: (1) Possible presence of different adsorbing materials, for example, iron oxides, aluminum oxides, and clay; (2) Difficulty to measure or estimate the amount and reactive surface area of adsorbing materials present in natural samples,10 which is an essential model input for the calculation; (3) Difficulty to account for the effects of natural organic matter (NOM) on anion adsorption, although NOM is almost always present in natural samples; (4) Presence of various ions that may have synergistic or competitive effects. (5) Difficulty to distinguish between adsorbed and precipitated phosphate. Only a few attempts have been made to use mechanistic models to describe adsorption of phosphate or other oxyanions like arsenate in soils.5,10 12 In the study of Gustafsson and study of Devau et al.,5,13 anion adsorption to soils was simulated with various oxides (ferrihydrite, goethite or gibbsite) and clay minerals (allophone, kaolinite or Illite). Effects of NOM on anion adsorption were not considered. However, it is known that NOM is present in all soils and adsorption of NOM to minerals has big effects on anion adsorption to minerals.14 16 In another Received: June 6, 2011 Accepted: August 24, 2011 Revised: July 29, 2011 Published: August 24, 2011 8420
dx.doi.org/10.1021/es201844d | Environ. Sci. Technol. 2011, 45, 8420–8428
Environmental Science & Technology
ARTICLE
Table 1. Properties of Soil Samples. (DOC and NOM Adsorbed Are from This Study. The Other Data Are from Hiemstra et al.10)a
soil
code in
10
pH
SOC
clay
%
%
Fe-DCB
Al-DCB
Fe-ox
mmol/kg
Al-ox
P-ox
Rev
A m2/g soil
mmol/kg
DOC mg/L
NOM mg/m2
9
3
5.6
3.7
8
852
27
342.2
18.6
38.2
50.4
32.7
16.2
0.98
11
4
5.6
3.3
11
125
34
92.7
32.7
31.5
55.4
34.0
13.7
1.26
16
7
5.7
0.9
11
135
27
50.3
19.1
14.9
14.3
6.6
8.7
0.79
18
9
4.6
4.9
21
242
36
211.3
27.5
34.9
41.2
27.3
37.0
1.35
30
15
6.4
4.1
28
200
30
117.1
38.9
24.0
18.7
9.8
17.7
0.96
40
18
5.6
2.1
12
116
36
67.8
28.8
19.3
21.9
13.6
17.5
1.05
a
Fe-DCB, Al-DCB: concentration of Fe and Al in dithionite-citrate-bicabonate extraction; P-ox, Fe-ox, Al-ox: concentration of P, Fe, or Al in ammonium oxalate extraction; Rev: total amount of reactively bound phosphate; A: effective reactive surface area of oxides in soil; DOC: DOC concentration measured in this study at a pH about 5 (Figure 1-b); NOM: amount of NOM adsorbed fitted.
study of Gustafsson on arsenate adsorption to soils, effects of NOM were simulated with a simplified approach, in which adsorption of NOM was represented by complexation between a surface group and a carboxylic group.12 The amount of adsorbed carboxylic group was derived by fitting. A similar approach was taken by Hiemstra et al. when they used mechanistic modeling to explain the change of soluble phosphate with the change of soil-solution ratio and to fit the reactive surface area of oxides in soils.10,17 However, in both approaches, the way in which the adsorbed NOM was treated was rather simplified, which does not allow for calculations of change of charge of adsorbed NOM with the change of solution composition (e.g., pH, Ca concentration). Recently, a mechanistic surface complexation model, the LCD (ligand charge distribution) model, has been developed for the assemblage of metal (hydr)oxides and NOM.18,19 In an earlier work, it has been demonstrated that the LCD model can successfully describe phosphate adsorption to goethite (αFeOOH) in the presence of humic acid (HA) and fulvic acid (FA).14 Compared to the simplified approaches mentioned above, the LCD model is more sophisticated in dealing with speciation calculation of adsorbed NOM. In stead of using fixed charge, the LCD model calculates surface complexation, protonation and cation adsorption of the reactive groups present on NOM. The site competition and electrostatic effects of NOM on anion adsorption depend on the speciation of NOM calculated. In this paper, the pH dependency of soluble phosphate concentration in six agricultural soils was measured. The LCD model was applied to simulate changes of soluble phosphate, taking into account the competitive effects of NOM and synergistic effects of major polyvalent cations (Ca2+, Al3+, and Fe3+). The objectives of this paper are 2-fold: (1) to test the feasibility of using LCD model to simulate anion speciation in natural samples (soils); (2) to get better understanding of the mechanisms that control the pH dependency of phosphate solubility in soils, based on the modeling results.
’ MATERIALS AND METHODS Soil Samples. Six soil samples were used in the experiment. These soils are chosen from a large collection of representative agricultural top soils of The Netherlands, known as the Copernicus Series.20 The chosen soils are also part of the samples used by Hiemstra et al. to develop a new method to estimate the effective reactive surface area of metal oxides present in soils.10,17 Basic soil properties and effective reactive surface area of oxides-like materials
of these samples are available from previous studies 10,17 (Table 1). The six soil samples cover a range of initial pH (4.6 6.4), soil organic carbon (SOC) content (0.9 4.9%), clay content (8 28%), and they vary also in the amount of iron and aluminum oxides, effective reactive surface area and phosphate loading (Table 1). Desorption Experiment. For each soil, 15 subsamples of 2.0 g each were put into Teflon bottles, to which 10 mL of 0.02 M CaCl2 solution was added. Appropriate amount of acid or base (0.1 M H2SO4 and 0.2 M KOH) was added to each subsample to adjust the pH to a range of 3 10. The amount of acid or base needed was estimated from a pretest using the same soil. Ultra pure water (UPW) was added to each bottle to a final total volume of 20 mL. The soil-solution ratio (SSR) is thus 1:10. The final concentration of CaCl2 added is 0.01 M. The soil suspensions were shaken end-over-end very gently (25 rpm) for10 days in a controlled temperature room at 20 °C, which was most of the time in the dark. pH was checked every day and readjusted if needed. The extra addition of acid and base leads to a final volume of 20.0 20.7 mL. The maximum amount of acid and base added was equivalent to a final concentration of respectively 14 mM H2SO4 (in soil 30) and 71 mM KOH (in soil 18). After 10 days shaking, pH in the suspension was measured. Thereafter, the suspensions were centrifuged at 3000 rpm for 15 min. The supernatant was then filtered through 0.45 μm filter before further analysis. Concentrations of inorganic and organic carbon were measured with a TOC analyzer (Shimadzu). After acidification to 0.1 M HNO3, concentration of colorimetric reactive P (P_PO4) was measured with an optimized molybdate blue method using a segmented flow analyzer (Skalar, The Netherlands). Concentration of soluble Fe and Al was measured with High Resolution ICP-MS (Thermo Scientific, Element2). Concentration of soluble Ca was measured with ICP-OES (Thermo Scientific, Iris Advantage). Modeling Approach. In the model, concentration of free Ca2+, 3+ Al , and Fe3+ was calculated from solution speciation using measured pH, concentration of soluble Ca, Al, Fe and (bi)carbonate as model input. In addition, 0.01 M Na and 0.02 M Cl were included. The concentration of Na was close to the sum of Na and K concentration measured in the 0.01 M CaCl2 extractions of these soils10 and the modeling results are not very sensitive to some changes in the Na concentration. The concentration of Cl was taken as that added in the soil extraction (0.01 M CaCl2). We assume that there are no inorganic particles present in the soil solution extracted. Formation of possible soluble species between various inorganic components was considered in the modeling. 8421
dx.doi.org/10.1021/es201844d |Environ. Sci. Technol. 2011, 45, 8420–8428
Environmental Science & Technology
ARTICLE
Figure 1. Concentration of P_PO4, Ca, Al, Fe, and DOC measured in the soil solutions.
In addition, adsorption of Ca, Al, and Fe to dissolved organic matter (DOM) was calculated using the NICA-Donnan model.21 DOM was calculated as two times of DOC measured (i.e., C content in DOM is 50%), and all DOM was assumed to behave as generic fulvic acid (FA) in terms of cation adsorption. The model
parameters (Supporting Information SI-Table 1) were taken from the generic parameters for fulvic acid,22,23 except those for Fe3+, which were adopted from Hiemstra et al.24 The LCD model18,19 was used to calculate phosphate adsorption to the assemblage of oxides and NOM, which integrates the 8422
dx.doi.org/10.1021/es201844d |Environ. Sci. Technol. 2011, 45, 8420–8428
Environmental Science & Technology NICA model and the CD-MUSIC model. Total amount of reactively adsorbed phosphate (Rev) and amount of effective reactive surface area of oxides in soils (A) were taken from those estimated by Hiemstra et al.11 (Table 1). The estimated effective reactive surface area of oxides-like materials in soils used in the modeling is a collective amount that may include various phases (iron oxides, aluminum oxides and clay edges). Goethite was used in the modeling as the representative material for oxides-like materials in soils. This is because first goethite is often an important component of the oxide fraction in soils. Second, phosphate binding capacity of clay minerals is more than 100 times lower than that of iron oxides.25 It is difficult to prepare clay materials that are free of oxide impurities;26 therefore the adsorption data collected on clay materials can have contributions from oxides present on clay. Third, we have the best quantitative insight in the multi component behavior of goethite, whereas understanding of adsorption to clay edges is not at the same level. Forth, there are lots of similarities in the adsorption behavior of various phases in soils, which allows this simplification used in our modeling. The CD-MUSIC model10,17 was used in the LCD to describe the reactivity and electrostatics at the goethite surface, while the Extended Stern model was used for the compact part of the electric double layer (EDL). In the extended stern model, there is one stern layer between the surface plane (0-plane) and first outer-electrostatic plane (1-plane) and a second stern layer between 1-pane and 2-plane (second outer-electrostatic plane).27 CD-MUSIC model parameters for ion (H+, Na+, Cl , Ca2+, PO43 ) adsorption to oxides were taken from Hiemstra et al.10 (Supporting Information SI-Table 2). It is assumed that the interaction between NOM and phosphate at the surface of oxides can be simulated in the LCD model using a layer of adsorbed fulvic acid (FA), which is present and equally distributed between the first and second Stern layer.14 Previous study has shown that the FA particles can fit into the compact part of the EDL, whereas a large part of the adsorbed humic acid (HA) is located beyond the EDL, in the diffuse double layer.14 Although it is very possible that the adsorbed NOM particles are larger than FA and stick out of the compact Stern layers, it has been shown that the fraction of NOM that is located by the model in the compact layer is the key factor for the correct prediction of the effect of NOM on phosphate adsorption. It was assumed that the carboxylic groups (RCOO ) of NOM in the first compact layer can form innersphere complexes (tFe1OOCR 0.5) with the singly coordinated surface sites on goethite.19 The corresponding charge is distributed between the 0-plane and 1-plane (Supporting Information SI-Table 2). The other carboxylic and phenolic groups on the NOM can bind H+, Ca2+, Al3+, and Fe3+. The reactions between the NOM ligands and surface sites, protons and other cations were calculated in the LCD with the NICA model. It was assumed that the NICA model parameters remain the same as for FA in the solution (Supporting Information SI-Table 1). The NICA model calculates the fractions of NOM ligands that are protonated, complexed with Ca2+, Al3+, and Fe3+ or bound to surface sites. Together with the amount of NOM adsorbed, the total charge carried by NOM at the mineral surface can be derived. The charge balance and electrostatic potential at each electrostatic plane at the oxide surface was calculated with the CD-MUSIC model by taking into account the charge contribution from NOM. The mean-field approximation was used to link the charge density and the potential in the various electrostatic planes. In the
ARTICLE
modeling, the amount of NOM adsorbed was kept constant for one soil sample and was derived as a fitting parameter. The model calculations were carried out using the computer program ORCHESTRA.28 In summary, processes that are included in the modeling are (1) formation of inorganic soluble species; (2) cation adsorption to DOM in soil solution; (3) pH effects on the protonation of the surface sites of oxides; (4) formation of ion pairs between the surface sites and Na, Cl; (5) adsorption of Ca to surface sites; (6) presence of NOM at the surface; (7) adsorption of H, Ca, Al, and Fe to NOM. Direct adsorption of Fe and Al to iron oxides is not considered. Adsorption of other inorganic anions such as sulfate, (bi)carbonate and silicate is also ignored. Soluble inorganic C concentration was measured (data not shown) and it is at the maximum 230 mg/L and is lower than DOC. We did not measure Si concentrations, but in the work of Hiemstra et al.,11 Si concentrations for these soils were measured and it is in general between 0.2 and 2 mg/L, 2 10 times lower than the phosphate concentrations in the same samples. Due to these relative low concentrations and due to the strong interaction with NOM, competition from other inorganic anions on phosphate adsorption is expected to be small.
’ RESULTS AND DISCUSSION Experimental Results. The measured soluble phosphate concentration (P_PO4) is given in Figure 1-a, which ranges from 7; For the high P-loading soils (soil 16 and 30), a local maximum of P_PO4 is seen around pH 4, and P_PO4 concentration stays at a relatively low level around pH 6 8 and decreases further from pH 8 to 10. The DOC concentration ranges from 20 to 800 mg/L (Figure 1-b). In general, DOC concentration remains relatively flat in pH 3 6, with mostly a small decrease with the increase of pH. Above pH 6, DOC concentration increases strongly with pH. Figure 1-c shows the soluble Ca concentration measured in the soil extracts. At a pH below the original soil pH, the Ca concentration measured is in general equal or higher than the 0.01 M concentration used in the extraction. Soil 30 that has the highest initial pH showed the highest Ca concentration at a low pH. Soluble Ca concentration decreases with the increase of pH, and reaches a level that is much lower than Ca added at high pHs. Concentration of Al and Fe in solution is in the range of respectively 1.8 853 μM (0.05 23 mg/L) (Figure 1-d), and 17 234 μM (0.01 13 mg/L) (Figure 1-e), and shows a minimum around pH 6 8, except in soil 18. Part of Al and Fe in solution can be present as colloidal particles, especially at high pH. But in the modeling of this study, presence of colloidal Al and Fe in the solution was not considered. Modeling Results. The calculated change of P_PO4 (lines) is compared to the data (symbols) in Figure 2. Please note that the unit used in Figure 2 is in logarithm scale, whereas in Figure 1-a it is in linear scale. The comparison shows that for the pH range of 5 8, which is the normal pH range for these agricultural soils, the pH dependency is very well described by the model except for soil 18. For low P-loading soils (soil 9 and 18) at low pH (8), the adsorption model overestimated soluble phosphate. The pH dependency of phosphate solubility in the pH range of 4 7 observed in soil samples is opposite to that found in batch experiments using synthetic iron oxides in a monocomponent system. In experiments using synthetic iron oxides, phosphate adsorption decreases with increase of pH and correspondingly P_PO4 increases with pH.6 This apparent contradiction can be reproduced by the adsorption model rather well. The decrease of P_PO4 with increase of pH from pH 4 to 7 in soils can be attributed largely, according to the model, to the synergetic effects of adsorption of Ca. Adsorption of Ca to oxides and to NOM increases with the increase of pH (Figure 3). Direct adsorption of Ca to oxides is much more important than Ca adsorbed to NOM. Amount of Fe adsorbed to NOM is relatively small due to a very low concentration of free Fe3+ in solution. Amount of Al adsorbed to NOM is higher than Ca adsorbed at a low pH (
