Phosphate Complexation Model and Its Implications for Chemical Phosphorus Removal

July 2, 2008

By Smith, S Takacs, I; Murthy, S; Daigger, G T; Szabo, A

ABSTRACT: A phosphate complexation model is developed, in an attempt to understand the mechanistic basis of chemically mediated phosphate removal. The model presented here is based on geochemical reaction modeling techniques and uses known surface reactions possible on hydrous ferric oxide (HFO). The types of surface reactions and their reaction stoichiometry and binding energies (logK values) are taken from literature models of phosphate interactions with iron oxides. The most important modeling parameter is the proportionality of converting moles of precipitated HFO to reactive site density. For well-mixed systems and phosphate exposed to ferric chloride during HFO precipitation, there is a phosphate capacity of 1.18 phosphate ions per iron atom. In poorly mixed systems with phosphate exposed to iron after HFO formation, the capacity decreased to 25% of the well-mixed value. The same surface complexation model can describe multiple data sets, by varying only a single parameter proportional to the availability of reactive oxygen functional groups. This reflects the unavailability of reactive oxygen groups to bind phosphate. Electron microscope images and dye adsorption experiments demonstrate changes in reactive surface area with aging of HFO particles. Engineering implications of the model/mechanism are highlighted. Water Environ. Res., 80, 428 (2008).

KEYWORDS: adsorption, chemisorption, chemical phosphorus removal, chemical equilibrium, surface area, precipitation.



Achieving phosphate removal to very low levels recently refocused the attention of the wastewater engineering profession on the design principles and various mechanisms that can produce effluent orthophosphate concentrations below 10 [mu]g-P/L (0.01 mgP/L), with total phosphorus concentrations typically less than 100 [mu]g-P/L (0.10 mg-P/L). Current equilibrium models used for design (WEF, 1998) are based on dissociation and solubility principles and cannot account for variable precipitate stoichiometries and time dynamics. The District of Columbia Water and Sewer Authority (DCWASA) (Washington, D.C.) initiated a research project to investigate the potential chemical, physicochemical, and physical mechanisms and their consequences for the engineering design and operation of these systems.

Chemical phosphorus removal is a complex and poorly understood process. However, it is a widely applied technology in practice. Traditionally, phosphorus (P) residuals in wastewater treatment plant effluents with chemical treatment were in the range 0.5 to 1.0 mg-P/L; however, recently, effluent requirements are frequently lower than this range by one or two orders of magnitude. Design and operation of chemical dosing systems is either experience-based (i.e., rule-of-thumb, typical doses, and ratios) or uses empirical kinetic (Henze et al., 2000) or chemical equilibrium (WEF, 1998) models. The research that is the subject of this paper was initiated to:

* Investigate, in a comprehensive manner, the major factors affecting phosphorus removal; and

* Create a model-based engineering design and operations method that relies on fundamental physical and chemical principles as much as possible and is able to predict the most relevant design and operational parameters (i.e., dose and residual phosphorus) based on local conditions.

The first objective is addressed in the companion paper by Szabo et al. (2008). The second objective, achieved partially through a chemical surface complexation model, is discussed here. It was found that a phosphate complexation model predicts phosphate removal in the simple inorganic abiotic aqueous iron(III) phosphate (H+-PO^sub 4^^sup 3-^-Fe^sup 3+^) system. Conceptually, the model involves precipitation of hydrous ferric oxide (HFO), which can be simultaneous with the complexation of phosphate species; this results in phosphate being occluded into the HFO floe. By varying the single model parameter proportional to the reactive site density (referred to as the active site factor [ASF] in the text that follows), the same chemical equilibrium framework can be used to describe phosphate removal by preformed HFO floes. This highlights the importance of reactive site availability in predicting phosphate removal. Changes in particle characteristics with aging of HFO are demonstrated using electron microscopy and dye adsorption experiments, and implications of the proposed removal mechanism for wastewater treatment are highlighted.


Batch sorption assays for chemically mediated phosphate removal were performed. All chemicals used were reagent-grade or better. The general experimental protocol involved making a solution of 1 mg-P/ L from a stock tribasic sodium phosphate solution. Iron, as ferric chloride, was added to this solution to give the desired iron-to- phosphorus ratio. While this initial solution was stirred, sodium hydroxide was added to adjust the pH to the desired value; the initial pH was acidic because of the ferric chloride. This sample was stored in a polyethylene bottle and shaken on a rotary shaker for 24 hours, which is often taken as the operational definition of equilibrium for phosphate adsorption studies (Li and Stanford, 2000). After 24 hours, the samples were taken off the shaker, and the pH was tested again. The samples were closed to the atmosphere, so there was little drift in pH, but the final pH value was taken as the “true” value. The solid sample was filtered through a 0.45- [mu]m filter, and the filtrate was measured for soluble phosphate concentration using the method described below. The filter mass after filtration can be used to determine the mass of precipitate and, after digestion in nitric acid and flame atomic adsorption spectrophotometry (AAS), the iron concentration of the solid can be determined. The iron digestion and analysis was done according to U.S. Environmental Protection Agency (U.S. EPA) (Washington, D.C.) method 305OB using a Perkin Elmer model 3100 atomic adsorption spectrophotometer (Perkin Elmer, Wellesley, Massachusetts).

Sensitive Colorimetric Determination of Phosphate. Phosphate detection is based on the standard colorimetric technique (Riley and Murphy, 1962) and presented as the ascorbic acid technique (APHA et al., 1998). For samples above 0.01 mg-P/L, a Varian Gary 50 spectrometer (Varian, Ontario, Canada) was used, with a 1-cm path length. For more dilute samples, a 1-m light path was used, for a detection limit of approximately 0.0001 mg-P/L. The 1-m light path consists of a hollow fiber optic tube, 1 m long, from World Precision instruments (LWCC-2100) (WPI Incorporated, Florida), coupled to a xenon lamp light source (Ocean Optics HPX-2000, Ocean Optics, Dunedin, Florida) and a QE5000 detector from Ocean Optics. The analytical methods were verified by spike recovery tests.

Microscopy Methods. Samples were prepared for microscopy by taking the original aqueous suspension and centrifuging it to collect the solid at the bottom of the centrifuge tube. The supernatant was decanted and replaced with pure water (MilliPore Synthesis A10 system, with resistance > 18 Mohm cm^sup -1^ [MilliPore, Billerica, Massachusetts]). The solid was distributed back into suspension, and the process was repeated three times. The final aqueous suspension was dispersed onto a thin slide for microscopic analysis. The water was evaporated, and the solid was analyzed. This cleaning procedure is necessary to remove salts from the aqueous solution. Scanning Electron Microscopy (SEM) measurements were performed on a Hitachi S-5200 SEM (Hitachi, Itasca, Illinois), using the accelerating voltages indicated on the figures without using conductive coating.

Dye Adsorption Experiments. Dye adsorption can be used to determine the surface area of particles suspended in solution. The advantage of dye adsorption techniques is that the sample is not dried, as in the Brunauer, Emmet, and Teller (BET) technique (Brunauer et al, 1938). By avoiding drying the sample, the surface area should be more representative of what solutes will actually “see” in solution. Dye adsorption using Methylene Blue is a common technique to determine the surface area of clay minerals (Hang and Brindley, 1970). The method involved taking a suspension of HFO with a known particle density (milligrams per liter) and adding increasing amounts of Methylene Blue in batches. The pH of these batches is fixed by addition of dilute sodium hydroxide and nitric acid, as necessary. For the experiments reported here, the pH was fixed at 4.0, even though this value is not representative of wastewater, because that pH yielded the most reproducible results. It is suspected that samples closer to the zero point-of-charge of the solids had a greater degree of particle interactions and interfered with the dye adsorption technique. After 1 hour equilibration time, these batches were filtered through 0.45-[mu]m membrane filters, and the concentration of dye in solution was determined using fluorescence spectroscopy at 600 nm excitation and 690 nm emission. Fluorescence was measured using a Varian Gary Eclipse spectrofluorometer. The difference between the total added dye and the measured solution dye concentration was used to determine the amount of dye bound to the surface of the HFO. In this way, a sorption isotherm for the dye molecule was generated and fitted to a Langmuir adsorption model using nonlinear regression analysis. Assuming monolayer coverage, the area of a single dye molecule can be used to determine the surface area of the particles in solution (Hang and Brindley, 1970). Acid-Base Titrations. Experimental and modeling methods for titration curves are detailed elsewhere (Smith and Ferris, 200Ib); however, in brief, the method is based on measuring pH using a glass electrode during acid-base titration in a temperaturecontrolled environment at a solution ionic strength adjusted to 0.02 M, which is close to typical wastewater. Each titration point was allowed to reach steady-state before the next addition of titrant. The charge excess (b), calculated from the following equation, is the amount of negative charge required for electroneutrality:

b = [Cations^sup +^] – [Anions^sup -^] + [H+] – [OH-] (1)

Solutions of ferric chloride and sodium phosphate (80 mM) were titrated from pH 5 to 10. Titrations were performed using a Tanager 9501 automatic titrator (Tanager Scientific, Ontario, Canada) and separate glass electrode and reference electrode half-cell electrodes.

Mechanism and Surface Complexation Model. While the ferric iron solution added to the wastewater stream is acidic, sufficient alkalinity is generally present in the wastewater to neutralize it. The result is rapid precipitation of HFOs. These HFOs are probably most closely related to 2-line ferrihydrite and will occur alone or as coatings on particles (Small et al., 1999; Smith and Ferris, 2003). Simultaneously with the HFO precipitation, soluble phosphate is removed by either precipitation of iron phosphates, co- precipitation, or adsorption of phosphate onto existing HFO particles. The removal of phosphorus can occur via many different pathways, as follows:

(1) Adsorption of phosphate onto HFO;

(2) Co-precipitation of phosphate into the HFO structure;

(3) Precipitation of ferric phosphate; and

(4) Precipitation of mixed cation phosphates (i.e., calcium, magnesium, iron, or aluminum phosphates, or hydroxyphosphates).

To avoid confusion in terminology, the International Union of Pure and Applied Chemistry (Research Triangle Park, North Carolina) (IUPAC) definition of co-precipitation is adopted here, which is “the simultaneous precipitation of a normally soluble component with a macrocomponent from the same solution by the formation of mixed crystals, by adsorption, occlusion or mechanical entrapment” (McNaught and Wilkinson, 1997).

The work presented here will focus on the simple iron, hydrogen, phosphate system as a starting point toward quantitative modeling of chemically mediated phosphorus removal. Most of the geochemical literature on iron and phosphate systems is devoted to item 1 above- the adsorption of phosphate onto preformed iron oxides (crystalline oxides or amorphous oxides, such as HFO). For example, Dzombak and Morel (1990) summarize cation and anion adsorption onto hydrous ferric oxides. This work and extensions have been included in various geochemical modeling packages, such as MINEQL (Allison et al., 1991). This type of modeling is a starting point, because its use in wastewater treatment systems requires more mechanistic details to realistically predict chemical dose requirements and phosphorus residuals under engineering conditions. This starting point will be built upon in this paper and will also incorporate pathway 2.

Pathway 2 was investigated, and ferric phosphate precipitation in a simple iron-phosphate-hydrogen system was found to predominate only at acidic pH (4 or lower; see titration curve in Figure 1 and the SEM image in Figure 2). The influence of more complicated water chemistry and the possibility of mixed precipitates (pathway 4) has received less quantitative attention, but should be the emphasis of future work representing more realistic systems. Some work on item 4 has shown that the presence of phosphate (and silicate) can dramatically affect HFO particle size (Magnuson et al., 2001).

Complexation Model for Phosphate Removal. The model presented here for phosphate removal by ferric chloride addition is based on interactions between HFO and phosphate. The model is based on fundamental chemical binding principles and is designed to study the effects of aging and mixing. The basis of the model is that iron and phosphorus share an oxygen atom. This can be represented by the following symbolic reaction (charges omitted), in which the iron oxide surface is represented as FeOOH:

FeOOH + HOPO^sub 3^ = FeOOPO^sub 3^ + H2O (2)

The exact reaction depends on the nature of the oxygen at the HFO surface-in particular, the number of iron atoms sharing each oxygen atom. This aspect of the model is summarized by the MUltiSIte surface Complexation (MUSIC) model applied to phosphate/goethite interactions by Geelhoed et al. (1997). Goethite is a crystalline iron oxide, but the mechanism of interactions is expected to be similar for amorphous iron oxide (HFO), as demonstrated by Smith and Ferris (2001b). These surface complexation reactions will also be pH- dependent, as a result of proton competition for surface oxygen and phosphate oxygen. The typical term for this type of modeling is surface complexation modeling (SCM). In the model developed here, the SCM concept is generalized to include any set of molecules with similar short-range order similar to Goethitelike clusters.

The reactive oxygens are termed “surface sites”, the availability of which reflects mixing and aging conditions. Rapid mixing means that surface sites are readily available, whereas, with slower mixing, much of the HFO would form in the absence of phosphate and make “internal” oxygen atoms unavailable for binding. Fresh HFO has a very open structure (see Figure 3), and surface sites are available. However, as the mineral ages, the structure becomes more compact (Figures 4 and 5), and the surface oxygen atoms become less available for binding. Binding might still be possible, but it will become diffusion-limited (Makris et al., 2004), suggesting that kinetics may be important. Longer term kinetic data (Szabo et al., 2008) suggest that there is a two-step process-a fast “equilibrium” step and a slower “kinetic” step. In this paper, we only consider the equilibrium process. The majority of geochemical studies assume chemical equilibrium; however, on timescales relevant to the wastewater treatment plant, not all processes will reach equilibrium. There are competing sorption and aging limitations. The iron oxides change with time, and floes can aggregate and change with age This must be accounted for in an “engineering” model.

Chemical Equilibrium Model. The chemical equilibrium model is divided into two parts for numerical reasons. The first part solves for the solid species present, where Fe(OH)^sub 3^(s) and FePO^sub 4^(S) are allowed to precipitate, if thermodynamically favored. Once the solid species are determined, a second chemical equilibrium modeling step is performed, where phosphate is allowed to complex with active sites on the precipitated HFO. This second equilibration step is the phosphate complexation component of the overall model. Mechanistically, phosphate removal should be considered simultaneously with precipitation, if phosphate is available. Thus, the SCM formalism is used to calculate removal during the co- precipitation step, even though there are no defined surfaces present during the early stages of hydrolysis. The model describes this step through the use of the ASF, which will be kinetically linked to initial mixing.

The first part of the chemical equilibrium modeling process is represented in Table 1, in which the chemical reactions are represented in tableau notation (Morel and Hering, 1993). The reactants are given as the components across the top of the table, and the products are listed as the species in the last column. For example, the 9th row in Table 1 corresponds to the FeH^sup 2^PO^sub 4^^sup 2+^ formation reaction, where it is necessary to combine 2 H+, 1 Fe^sup 3+^, and 1 PO^sub 4^^sup 3-^, as reactants to yield the desired product. The log K value for this formation reaction is 22.11. The precipitation of potential solids is tested by comparing the reaction quotient (Q) with the solubility product, K^sub sp^. If Q is greater than K^sub sp^, precipitation is allowed to occur. In the case of multiple precipitates, the most insoluble (larger Q) is allowed to precipitate first, and iteration is subsequently performed to see if the second phase is still insoluble.

The numerical problem corresponding to solving for the equilibrium state of the system is to determine the value of X that minimizes the residuals in the mass balance. This calculation is performed as follows:

Minimize R as a function of X,

Where R = A’ x (10^sup logC^) – T and log C = K + AX’

Minimization of the residual vector is performed using the Newton- Raphson method implemented in Matlab (The Mathworks Incorporated, Natick, Massachusetts) following the method of Smith and Ferris (2001a). In the absence of precipitate, there are two elements in the X vector-the component concentrations [Fe^sup 3+^] and [PO^sub 4^^sup 3-^]. Here, pH is fixed, so [H+] is known. The stoichiometry matrix A is 9 x 2 for 9 species and two components. The T and K vectors contain the total concentrations of iron and phosphate and the log K values, respectively. Finally, C is a vector containing the concentration of each species. Concentrations in the X and T vectors are expressed as logarithms to the base 10 to ensure positive solution concentrations and a well-behaved numerical optimization.

When one solid phase is present, either [Fe^sup 3+^] or [PO^sub 4^^sup 3-^] is fixed, because one degree of freedom is lost, and the optimization proceeds with one entry in X. When two solids precipitate, there are no degrees of freedom, and the concentrations of all species can be solved without iteration. If the pH is unknown, the same solution method can be used with a corresponding increase in the matrix dimensions, as long as terms for the major cationic and anionic species are included, so that the total hydrogen concentration can be estimated. Once HFO is found to precipitate, an SCM calculation is run to determine how much phosphate is bound to the surface and other available oxygen binding sites as the precipitate is formed. The SCM model, shown in Table 2, still includes all aqueous reactions, but the amount of precipitate is fixed. The stoichiometry of the surface reactions was determined based on Geelhoed et al. (1997). These possible surface reactions are based on spectroscopic studies and theoretical surface structural considerations (i.e., MUSIC model). These possible reactions on known Goethite are taken as reasonable reactions that are possible on the HFO surface. Even though the iron oxide formed here is amorphous and not crystalline, it is reasonable that similar types of binding arrangements (iron to oxygen) will occur at their surface (Smith and Ferns, 2001b); these are the active phosphate binding sites. The values for the equilibrium constants in Table 2 were determined heuristically to describe the experimental data. The best description of the experimental data was taken as describing the lowest residual phosphorus concentrations. The idea is that these points are closest to equilibrium, which is expected to be the maximum removal of phosphate. While the exact values for the log K values are different from the Geelhoed et al. (1997) paper, the reaction stoichiometries are the same.

Soluble phosphorus is the sum of all phosphorus species not bound to the iron oxide surface or precipitated. To quantify a value for this, it is necessary to determine how much phosphate capacity exists. In the tableau presented in Table 2, the total binding site capacity for sites 1 and 2 can be referred to as SlT and S2T. Using the same assumptions as Geelhoed et al. (1997), the concentration of these binding sites are taken to have the same value. These need to be related to the total iron concentration (FeT) in the precipitate, as follows:

S1T = S2T = ASF 7times; (FeT in HFO) (4)

The crucial factor in modeling is the ASF. This is related to available binding sites before, after, and during precipitation and represents the fraction of reactive surface oxygen atoms per bulk iron in the HFO. Best-fit values for ASF, given in the Fitting of the Active Site Factor Parameter section, represent the only model parameter that needs to be varied to describe the data presented in Figures 6 and 7.

The value of the area factor parameter is linked to the mixing conditions (C value) and age of floe (solids retention time [SRT]) through a kinetic function, as follows:

ASF = f (G, SRT) (5)

This function, together with expressions describing diffusion limitations within the floe, forms the kinetic part of the model. The exact form of the functions and their parameters have not yet been determined and will be the subject of a future paper.

Results and Discussion

Characterization of Iron Species Formed in the Presence of Phosphate. Hydrous ferric oxide forms when acidic ferric chloride solutions are neutralized. The products of iron oxyhydroxide formation are variable, depending on conditions. In particular, the rate of base addition and the aging of the sample are known to create products of variable structure (Schwertmann and Cornell, 2000). Because the conditions of HFO synthesis in this current work were variable in that different pH values were selected as the endpoint, it is necessary to determine if the same HFO product was formed independent of the final pH. To accomplish this, the iron content of the HFO was determined by AAS. There were no trends versus pH in the percent iron by mass, although the value in the solid was quite variable (most of the values were between 45 and 60%). The percent iron value for 6 observations on HFO formed between pH 7 and 10 was 52.7 +- 9.7, at a 95% confidence level. These results suggest that the HFO prepared had a variable composition and that this variation was random and not dependent on the formation pH. For comparison, the percent iron in Fe(OH)^sub 3^ without water for hydration is 52 and for geothite (FeOOH) is 62.6. One proposed stoichiometry for HFO is Fe^sub 5^(HO)^sub 8^ . 4H^sub 2^O (Jambor and Dutrizac, 1998), which would have a iron composition of 57.9% by mass. The percent iron of FePO^sub 4^(S) is 37.1, so it is extremely unlikely that the precipitate has the stoichiometry of ferric phosphate.

A further test for ferric phosphate precipitation was performed by titration of a ferric chloride and sodium phosphate solution and charge modeling. The measured b values for two titrations are shown in Figure 1. The total iron and total phosphate concentrations were 80 mM for both data sets. Calculated charge excess is modeled with K^sub sp^ for FePO^sub 4^(S) as the only unknown. Parameters for soluble ferric phosphate complexes FeH^sub 2^PO^sub 4^^sup 2+^ and FeHPO^sub 4^^sup +^ are taken as National Institute of Standards and Technology (Gaithersburg, Maryland) (NIST, 2001 ) values (see Table 1) and corrected for ionic strength using the Davies equation. The charging model for the HFO surface is taken from Smith and Ferris (2002a). The best-fit value for log K^sub sp^ for Fe(OH)^sub 3^(S) is taken as -38.25, determined from previous experiments in the absence of phosphate (data not shown).

Based on these experiments, the best-fit value of log K^sub sp^ for FePO^sub 4^(s) is greater than -23. This model calculation is shown as the solid line on Figure 1 and implies that no pure ferric phosphate precipitation occurs above pH 5. As an illustration, the dashed line, using log K^sub sp^ of FePO^sub 4^(s) = -26, shows how charge excess would look if FePO^sub 4^(S) started to form around pH 7.2. The NIST (2001) contains a range of log K^sub sp^ values for FePO^sub 4^(S), depending on particle size and form, of between – 21.8 (amorphous) and -26.8 (crystalline).

Evidence of ferric phosphate formation at low pH is shown in Figure 2, where FePO^sub 4^(S) is formed at approximately pH 3.5. Several samples were tested for FePO^sub 4^(S) at higher pH, but no evidence of FePO^sub 4^(S) was found at pH values above 5. This observation is consistent with the titrimetric studies shown in Figure 1.

From the literature (de Haas et al., 2000; WEF, 1998) and the data in Figure 6, it is clear that phosphate is removed from solution up to pH 8 and higher. It is proposed that co- precipitation and adsorption of phosphate ions on ferric hydroxide floes is occurring around neutral pH values (Pierri et al., 2000). In this context, coprecipitation refers to phosphate removed during the precipitation of HFO. It is possible that the mechanism of removal is the trapping of surface-bound phosphate on tiny HFO floes, as the floes accumulate and grow, and does not involve pure ferric phosphate (FePO^sub 4^) formation.

Model Description of the Effect of pH and Mixing. Figure 6 shows experimental data compared with the model calculation for more than 100 data points. Batch removal tests were performed, with an initial concentration of 1 mg-P/L dosed with 10 mg-Fe/L, where a ferric chloride stock solution was the source of iron, followed by pH adjustment and 24-hour equilibration time. This corresponds to a 5.5 molar ratio of iron to phosphorus. The model qualitatively describes the data and captures the trend of greatest removal at low pH, with gradual decrease in efficiency at a more basic pH. Essentially no phosphate removal occurs at pH values above 9.5. With this high dose of iron, it was possible to achieve residual phosphate concentrations of 10 [mu]g/L for samples at pH values near 5.5. There is approximately one-half an order of magnitude scatter in the data, however, and 10 [mu]g/L should be viewed as a best value. For pH values near neutrality, the residual phosphate values were closer to 0.1 mg-P/L.

Much experimental uncertainty exists in the data, likely reflecting the sensitivity of phosphate removal to the conditions of HFO formation. If the solution is well-mixed, then phosphate is removed as the HFO is formed; phosphate complexation occurs simultaneously with precipitation and can be termed co- precipitation. The experimental uncertainty is possibly a result of the difficulty in mixing reproducibly during base addition and HFO formation.

Fitting of the Active Site Factor Parameter. The chemical equilibrium parameters for the model plotted in Figure 6 are presented in Tables 1 and 2. The ASF value was the only parameter adjusted to allow the model to fit the data. The ASF value for HFO formed in the presence of phosphate was determined to be 1.18, which means that, for every 100 iron atoms in the HFO, there are 118 available oxygen sites for binding. Note that, in SCM, these oxygen sites are just potentially available for binding; however, as can be seen from the data, actual binding is pH-dependent. The model takes this into account using protonation/deprotonation reactions for the reactive sites and the phosphate anion. To assess if this model parameter (ASF) value is reasonable, some simple scenarios are considered. For a hydrated iron atom [Fe(H^sub 2^O)^sub 6^^sup 3+^] in solution, the ASF value would be 6. This is not a chemically realistic value; for energetic and steric reasons, 6 phosphate ions would not bind to a single iron cation. This theoretical values for a hexahydrated ferric cation does put an upper constraint on the value for ASF. For bernalite, a mineral with stoichiometry Fe(OH)^sub 3^(S), the unit cell is 0.7 nm x 0.7 nm x 0.7 nm. Using the crystal structure reported by Birch et al. (1993) and the visualization software Mercury (Cambridge Crystallographic Data Centre, Cambridge, United Kingdom), it is possible to estimate (count) reactive surface oxygen atoms per iron (total bulk iron). For a 1-nm cube, there are 4.5 reactive oxygen atoms per iron; for a 2 nm x 2 nm x 2 nm cube, this value decreases to 2.8, and the ratio is 1.3 for a 50-nm cube. Particles on the order of 50 nm in size are present in these samples (see Figure 3), so the maximum ASF value of 1.18 is reasonable. This ASF value implies that 1.18 phosphate ions are bound per 1 iron atom. This ratio is very similar to the 1:1 stoichiometry for formation of the pure component FePO4, which is potentially the source of the statements in applied wastewater treatment literature, that FePO^sub 4^ is precipitating at a pH range 6 to 8. Indeed, FePO^sub 4^ is likely not precipitating, as described in the previous subsection. Figure 7 shows experimental data and model fit for preformed iron hydroxide floes prepared by base addition before the addition of phosphate. These samples have the same total iron and total phosphate as the data shown in Figure 6, but the reagents were added in a different order. The phosphate removed in this system has a minimum value of two orders of magnitude higher than in the system where HFO is generated in the presence of phosphate. This data directly shows the effect of the order of reagent addition, but also can be interpreted as demonstrating the importance of mixing in phosphate removal. Co- precipitation (the removal of phosphate via SCM-like mechanism as HFO precipitates) is of less importance if the floes are forming in the absence of solution phosphate, which is reflected in the ASF value of 0.31 used to describe the data in Figure 7. This is approximately 25% of the value of the ASF value used in Figure 6 and results because the phosphate was not exposed to the particles as they were forming.

Implications of pH and Mixing for Design and Operations. The data and modeling results in Figures 6 and 7 have implications for engineering design and practice. It can be inferred that mixing should be done at the site of addition of acidic iron solution. The presence of dead zones (incomplete mixing) will result in a lower ASF value, thus affecting the efficiency of chemical dosage, because a low sorptive capacity will result in low removal. As Szabo et al. (2008) discussed, in addition to complete mixing, the amount of mixing energy at the site of acidic iron addition is also important. Szabo et al. (2008) also showed that pH has little effect between pH 5 and 7, if metal doses are large, resulting in very low residuals (less than 0.02 mg-P/L). For solutions above pH 7 and for higher than 0.05 mg-P/L residuals, pH does effect removal, as shown in Figures 6 and 7. Wastewater treatment plants operate in the pH range 6 to 8. Thus, mixing energy and pH in certain conditions are two factors that strongly influence removal efficiency, and it would be potentially beneficial to consider these factors in engineering practice.

Model Description of Iron Dosing Effects. A limited number of experiments were performed at higher and lower doses of iron with phosphate, in the presence of HFO, as it is forming. These experimental and model results are shown in Figure 8, with an ASF value of 1.18 for comparison purposes. High doses of iron do tend to remove more phosphate; based on model calculations, levels as low as 0.1 [mu]g/L are possible; however, in practice, the lowest values measured were at the 5-[mu]g/L level. It should be noted that the 0.1-[mu]g/L value is a theoretical value calculated using the model presented in this paper and was not measured. Also, to achieve such a low value, an extreme molar dose of 55 times more iron then phosphorus would be necessary. As can be seen in Figure 8, obtaining reproducible data for such high iron doses is very difficult. Part of the reason for such scattered data is that the formation of HFO particles in the presence of high iron concentrations is likely different from the HFO formed in dilute solution. In practice, particles will associate with each other, and this association can be the predominant interaction in highly concentrated solutions (Anderson and Benjamin, 1990). The modeling presented in this paper assumes no particle-particle interactions; the only surface reactions are between HFO and phosphate.

Optimal Iron Dose. To test for the particle-particle interactions mentioned above, iron overdose experiments were performed at fixed final pH of 7; HFO particles were formed by base addition in the presence of phosphate. The results of these experiments are shown in Figure 9. The right panel of Figure 9 shows that the residual phosphate decreases with an increasing molar iron dose of up to 100 moles Fe/mole P. At higher doses, the removal efficiency starts to decrease, likely as a result of particleparticle interactions (Anderson and Benjamin, 1990). The solid line in Figure 9 is data interpolation (not a model calculation) to emphasize the trend. The left panel of Figure 9 is a subset of the total data shown in the right panel. Phosphate removal at “lower” doses is more efficient, in terms of concentration of phosphorus removed. The solid line on the left panel of Figure 9 represents an ASF value of 0.91, which reasonably represents the data at lower molar ratios. At higher molar ratios, the 0.91 ASF value does not represent the data; a value of 0.2 is necessary to describe the higher dose data. The model calculations corresponding to an ASF value of 0.2 are shown as a dashed line in the left panel of Figure 9. The change in ASF can be interpreted as iron atoms predominantly being exposed to other iron atoms during the co-precipitation step, and most of the HFO formation occurs in the absence of available phosphate. Thus, less reactive oxygen atoms are available for phosphorus sharing with iron as HFO is formed.

Implications of Iron Dosage for Design and Operations. If the mechanism proposed in this paper is correct, as the molar ratio of iron to orthophosphate is increased based on random collisions, the iron particles are more likely to associate and interact with other iron particles rather than phosphate molecules, thus decreasing the efficiency of the process. At extreme iron levels, the ASF factor is not only dependent on mixing and time; it is also dependent on iron dose. It should be noted that the iron doses in Figure 9 are, for the most part, well outside the range of engineering practice. These experiments were performed to see if there was a minimum measurable phosphate concentration. If phosphate is being removed by reaction with iron oxide surfaces, then increasing the iron concentration should decrease the residual phosphorus. The question to address was as follows: is there a lower limit on residual phosphorus via this removal mechanism? The minimum measured value was approximately 5 [mu]g-P/L (with large experimental uncertainty). Lower levels were not possible, even with higher doses. In fact, higher doses led to less efficient removal, which is likely the result of particle- particle interactions, and these proposed particle-particle interactions limit the practical minimum residual phosphorus (Figure 9).

Aging of Hydrous Ferric Oxide: Dye Adsorption Experiments. The batch phosphate removal tests demonstrate the importance of surface area in controlling phosphate removal via ferric chloride addition. There are several surface area determination techniques available. The most commonly used technique is BET analysis, which requires sample drying, which will inevitably change the nature of the particles. To avoid this, solution speciation techniques are used, such as dye adsorption (Hang and Brindley, 1970). In the dye adsorption method, a known concentration of dye is added to the aqueous sample of particles, where it interacts with the surface and is removed from solution. Thus, a measure of soluble dye can be used to determine how much dye is adsorbed to the surface, by finding the difference between the known amount of dye added and the amount measured in solution. The result is an adsorption isotherm for the dye molecule onto the reactive surface. The isotherm can be mathematically fit to the Langmuir equation and the total sorptive capacity determined (Hang and Brindley, 1970). This total capacity can be converted to surface area, if the size of the dye molecule is known and monolayer coverage is assumed. Figure 10 shows data and corresponding best-fit isotherms for fresh and old HFO. The fresh HFO was prepared according to the method detailed in the Methodology section (i.e., neutralize a 10 mg-Fe/L solution with base) and used immediately. The old HFO material was prepared in the same way, but stored at 4[degrees]C for 2 years.

For fresh HFO, the surface area was determined to be 5500 +- 170 m^sup 2^/g. The old HFO (2 years) was determined to have a surface area of 920 +- 170 m^sup 2^/g. For comparison, the typical value for HFO surface area based on BET analysis is taken as 600 m^sup 2^/g, with a reported range of 100 to 1200 m^sup 2^/g and 600 m^sup 2^/g somewhat arbitrarily selected as a representative value (Dzombak and Morel, 1990). It is reasonable that solution dye adsorption should give higher surface areas than BET measurement, because the samples are not exposed to the compaction and reorganization associated with drying. The surface areas determined here show the expected qualitative trend-much higher surface area for the fresh material and much lower surface area for the old material. In terms of proportional decrease, the aged surface area has decreased to 17% of the fresh HFO value. For comparison, in the preformed floe experiments, the ASF value decreases to 25% of the fresh value. The implication is that the ASF parameter can be related to available surface area, and it might be possible to model phosphate interactions during aging of floes by varying this single parameter. Implications of Aging for Design and Operations. As the HFO molecules age, they become denser (Dzombak and Morel, 1990) and should limit the ability of orthophosphate diffusion within the molecular structure (Makris et al., 2004). In kinetic modeling of dynamic treatment processes, the surface area changes of HFO must be taken into account. The age of floes may be significant to practice. For example, whether to recycle iron or aluminum containing water treatment plant residuals for phosphorus removal may significantly dependon the floe age. Takacs et al. (2006) describe a plant that recirculates its relatively “fresh” tertiary sludge to the primary process, resulting in substantial additional phosphorus removal in the primary process. Georgantas and Grigoropoulou (2005) present comparison data for phosphate removal using either fresh alum or spent alum sludge. It was found that the fresh material removed phosphate 5 times more efficiently, in terms of residual concentrations, than the old spent material.

Aging of Hydrous Ferric Oxide: Microscopy. The batch sorption data (Figures 6 and 7) allows a model to be developed to quantitatively describe the removal of phosphate. The important parameter in this model is the proportionality constant (ASF), to convert total iron into binding site density. An obvious question is the following: how does binding site density change as the particles age? To address this issue, SEM and transmission electron microscopy (TEM) images were obtained for HFO particles of different ages. The freshest HFO that could be obtained was prepared in the laboratory from 10 mg-Fe(III)/L and immediately “quenched” by rinsing with pure water. The image was obtained within 24 hours of particle formation. A TEM image of this sample is shown in Figure 3. The TEM images can show size (approximately 50 nm short axis and 100 nm long axis) and density. The dark regions in the image are denser than the lighter regions of the image. This fresh HFO particle is much less dense, suggesting that significant potential exists for diffusion into the particle. The total particle size is consistent with the estimate of ASF of 1.18 for phosphate removal when HFO is formed in situ. For 50 nm particles, the estimate of ASF is 1.3.

The sample shown in Figure 3 is a typical TEM image of HFO formed at circumneutral pH. Similar images are obtained when the HFO is formed in the presence of phosphate, except at an acidic pH. At acidic pH, FePO^sub 4^(s) was observed (see Figure 2).

A second TEM image is shown in Figure 4 for HFO that was prepared in the same way as the fresh HFO, but allowed to age for 4 days before centrifugation and rinsing. The TEM image shows that the density increased dramatically, and the low-density fresh HFO is more like dense hard spheres. The total size of the particle has also increased, with a long axis greater than 200 nm and a short axis of approximately 100 nm. An increase in particle size is consistent with a decrease in surface area and a decrease in the ASF parameter. Smaller aggregates are apparent as constituents of this total particle, however. Based on these images, it seems likely that, as HFO particles age, the capacity for phosphate removal will decrease.

Further emphasis of the proposed decrease in removal capacity with age is provided by an image (Figure 5) of 2-year-old HFO (actually crystallized goethite), which was prepared by rapid neutralization of ferric chloride and stored at 4[degrees]C. This sample (Figure 5) shows very large particle size with a “mat” greater than 400 nm. In addition to the large particle size, there is evidence of crystallization. This demonstrates that, not only does particle size change, but the actual chemical structure of HFO evolves with time.


Experiments were conducted to develop a surface complexation model for chemical phosphorus removal using HFO. The results indicate that phosphorus removal occurs by two fundamental mechanisms-(1) adsorption of phosphate onto preformed HFO and (2) co- precipitation of phosphate into the HFO structure. The direct precipitation of pure ferric phosphate does not appear to occur to any significant extent at pH values above 5. The second mechanism requires that phosphate be present when HFO is being formed, and it appears to occur very rapidly and is, therefore, significantly influenced by the specific mixing conditions occurring during the formation of HFO. Orthophosphate concentrations can be reduced to quite low values (see Figure 6) when this mechanism is functioning effectively. The first mechanism is significantly affected by the structure of the HFO, which changes with time as it “ages” and transitions from a more amorphous to a more crystalline structure. It should be noted that, on the timescale relevant to a wastewater treatment plant, the HFO should increase in density, but is not likely to crystallize. The more open amorphous structure appears to allow more of the oxygen atoms in the HFO to be available for the diffusion of phosphate to active sites where oxygen atoms in the HFO and phosphate molecule can be shared via covalent bonds, thereby allowing phosphate to be removed from solution. As the HFO increases in density, diffusion is retarded, which makes fewer active sites available for this reaction. The conceptual model presented here is supported qualitatively by the electron microscopic and dye adsorption experiments.

To summarize the conceptual model developed in this research for phosphate removal at a molecular level, the mechanism is based on iron and phosphorus sharing an oxygen atom. The empirical and modeling data presented in this paper is consistent with this mechanism. The implications of this proposed molecular-level mechanism are that, in a well-mixed system of ferric ions and phosphate, this oxygen sharing can occur as the initial iron hydroxide oligomer precursors of the solid phase are forming; that is, initially, for a very short time, there is molecular complexation of the phosphate by soluble iron hydroxide species. In well-mixed systems, this part of the process was shown to bind most of the available phosphate. This process is consistent with the definition of co-precipitation.

Continuing with this molecular-level description of the proposed mechanism of phosphate removal, as the process progresses in the next few seconds and minutes, the oligomers accumulate and polymerize to form a network of solid HFO. If the system was wellmixed during polymerization, the HFO will contain associated phosphate occluded into the solid. In the case of typical, poorly mixed plant dosing conditions, the HFO forms with much less phosphate occluded. Once this solid phase forms, the available sites for phosphate binding are drastically reduced on the surface. Deeper sites are expected to be kinetically limited through diffusion (Makris et al., 2004).

The ASF refers to all the available sites as the HFO forms, not just terminal sites on the surface of the mineral at the end of precipitation. The ASF value of 1.18 corresponds to a phosphate capacity of essentially 1 mole phosphate/mole iron. Thus, the “coprecipitate” would have a stoichiometry similar to iron phosphate, but a different chemical structure (i.e., not a salt structure), as revealed by SEM images in this manuscript. A similar surface complexation method for phosphate removal has been proposed by other researchers (Moller, 2006), based on observations from a treatment plant using iron-oxide-coated sand to remove phosphate. The work by Moller (2006) shows similar residual phosphate values (down to 0.01 mg-P/L as orthophosphate) and also a solidphase stoichiometry of 1:1 iron to phosphorus.

The equilibrium part of the mathematical model does not differentiate between “soluble” and “solid” surface oxygen atoms and offers a unified description for the oxygen-sharing mechanism itself. The initial fast removal in well-mixed systems and the slow removal experienced later are described by varying ASF as a function of time and initial mixing.


Funding for the project from DCWASA (Washington, D.C.) and Natural Sciences and Engineering Research Council of Canada (Ottawa, Ontario, Canada) is gratefully acknowledged. The authors thank Vladimir Kitaev for the SEM and TEM analysis and the Wilfrid Laurier University (Waterloo, Ontario, Canada) undergraduate students (Kelly Fischer, Ashley St. Pierre, and Lisa Rabson) who performed the measurements presented here. The authors thank three anonymous reviewers for their comments; this paper has benefited from their input.

Submitted for publication February 13, 2007; revised manuscript submitted October 31, 2007; accepted for publication January 8, 2008.

The deadline to submit Discussions of this paper is August 15, 2008.


Allison, J.; Brown, D. S.; Novo-Gradac, K. (1991) MINTEQA2/ PROEFA2. A Geochemical Assessment Model for Environmental Systems. U.S. Environmental Protection Agency, Environmental Research Laboratory: Athens, Georgia.

American Public Health Association; American Water Works Association; Water Environment Federation (1998) Standard Methods for the Examination of Water and Wastewater, 20th ed.; American Public Health Association: Washington, D.C.

Anderson, P. R.; Benjamin, M. M. (1990) Modeling Adsorption in Aluminum-Iron Binary Oxide Suspensions. Environ. Sci. Technol., 24, 1586-1592.

Birch, W. D.; Pring, A.; Relier A.; Schmalle, W. H. (1993) Bernalite, Fe(OH)^sub 3^, a New Mineral from Broken Hill, New South Wales: Description and Structure. Am. Mineral., 78, 827-834.

Branauer, S.; Emmett, P. H.; Teller, E. (1938) Adsorption of Gas in Multimolecular Layers. J. Am. Chem. Soc., 60, 309-319.

de Haas, D. W.; Wentzel, M. C.; Ekama, G. A. (2000) The Use of Simultaneous Chemical Precipitation in Modified Activated Sludge Systems Exhibiting Biological Excess Phosphate Removal Part I: Literature Review. Water SA, 26, 439-452. Dzombak, D. A.; Morel, F. M. M. M. (1990) Surface Completion Modeling, Hydrous Ferric Oxide; Wiley-Interscience: New York.

Geelhoed, J. S.; Hiemstra, T.; Van Riemsdijk, W. H. (1997) Phosphate and Sulfate Adsorption on Geothite: Single Anion and Competitive Adsorption. Geochim. Cosmochim. Acta, 61, 2389-2396.

Georgantas, D. A.; Grigoropoulou, H. P. (2005) Phosphorus Removal from Synthetic and Municipal Wastewater Using Spent Alum Sludge. Water Sci. Technol., 52, 525-532.

Hang, P. T.; Brindley, G. W. (1970) Methylene Blue Adsorption by Clay Minerals: Determination of Surface Areas and Cation Exchange Capacities. Clays Clay Min., 18, 203-212.

Henze, M.; Gujer, W.; Mino, T.; van Loosdrecht, M. C. M. (2000) Activated Sludge Models ASM1, ASM2, ASM2d and ASMS, IWA Scientific and Technical Report No. 9; IWA Publishing: London, United Kingdom.

Jambor, J. L.; Dutrizac, J. E. (1998) Occurrence and Constitution of Natural and Synthetic Ferrihydrite, a Widespread Iron Oxyhydroxide. Chem. Rev., 98, 2549-2585.

Li, L.; Stanforth, R. (2000) Distinguishing Adsorption and Surface Precipitation on Phosphate and Goethite (alpha-FeOOH). J. Coll. Interf. Sci., 230, 12-21.

Magnuson, M. L.; Lytle, D. A.; Frietch, C. M.; Kelty, C. A. (2001) Characterization of Submicrometer Aqueous Iron (III) Colloids in the Prescence of Phosphate by Sedimentation Field Flow Fractionation with Multiangle Laser Light Scattering Detection. Anal. Chem., 73, 4815-4820.

Makris, K. C.; El-Shall, H.; Harris, W. G.; O’Conner, G. A.; Obreza, T. A. (2004) Intraparticle Phosphorus Diffusion in a Drinking Water Treatment Residual at Room Temperature. J Coll. Interf. Sci., 277, 417-423.

McNaught, A. D.; Wilkinson, A. (1997) IUPAC Compendium of Chemical Technology; Blackwell Science: Oxford, United Kingdom.

Moller, G. (2006) Nutrient Removal. Water Environment Research Foundation Workshop 05-CTS-1W, Washington, D.C., March 9-11; Water Environment Research Foundation: Alexandria, Virginia.

Morel, F. M. M.; Hering, J. G. (1993) Principles and Applications of Aquatic Chemistry; John Wiley and Sons: New York.

National Institute of Standards and Technology (2001) NIST Standard Reference Database 46. National Institute of Standards and Technology: Gaithersburg, Maryland.

Pierri, E.; Tsamouras, D.; Dalas, E. (2000) Ferric Phosphate Precipitation in Aqueous Media. J. Crystal Growth, 213, 93-98.

Riley, J. P.; Murphy, M. J. (1962) A Modified Single Solution Method for Phosphate in Natural Waters. Anal. Chim. Acta, 12, 31- 36.

Schwertmann, U.; Cornell, R. M. (2000) Iron Oxides in the Laboratory: Preparation and Characterization; Wiley: New York.

Small, T. D.; Warren, L. A.; Roden, E. E.; Ferris, F. G. (1999) Sorption of Strontium on Bacteria, Fe(III) Oxide, and Bacteria- Fe(III) Oxide Composites. Environ. Sci. Technol., 33, 4465-4470.

Smith, D. S.; Ferris, F. G. (2001a) Methods in Enzymology Volume 337: Microbial Growth in Biofilms, Doyle, R. (Ed.); Academic Press: San Diego, California, 225-242.

Smith, D. S.; Ferris, F. G. (2001b) Proton Binding by Hydrous Ferric Oxide and Aluminum Oxide Surfaces Interpreted Using Fully Optimized Continuous pKa Spectra. Environ. Sci. Technol., 35, 4637- 4642.

Smith, D. S.; Ferris, F. G. (2003) Specific Surface Chemical Interactions Between Hydrous Ferric Oxide and Iron Reducing Bacteria Determined Using pKa Spectra. J. Coll. Interf. Sci., 266, 60-67.

Szabo, A.; Takacs, I.; Murthy, S.; Daigger, G. T.; Licsko, I.; Smith, S. (2008) The Significance of Design and Operational Variables in Chemical Phosphorus Removal. Water Environ. Res. (in press).

Takacs, I.; Murthy, S.; Smith, S.; McGrath, M. (2006) Chemical Phosphorous Removal to Extremely Low Levels: Experience of Two Plants in the Washington D.C. Area. Water Sd. Technol., 53, 21-28.

Water Environment Federation (1998) Biological and Chemical Systems for Nutrient Removal; Water Environment Federation: Alexandria, Virginia.

S. Smith1, I. Takacs2*, S. Murthy3, G. T. Daigger4, A. Szabo5

1 Department of Chemistry, Wilfrid Laurier University, Waterloo, Ontario, Canada.

2 EnviroSim Associates Ltd., Flamborough, Ontario, Canada.

3 DCWASA, SW Washington, D.C.

4 CH2M HILL, Englewood, Colorado.

5 Budapest University of Technology and Economics, Budapest, Hungary.

* EnviroSim Associates Ltd., 7 Innovation Dr., Suite 205, Flamborough Ontario, Canada, L9H 7H9; e-mail: imre@envirosim.com.

Copyright Water Environment Federation May 2008

(c) 2008 Water Environment Research. Provided by ProQuest Information and Learning. All rights Reserved.

comments powered by Disqus