Three-Component Competitive Adsorption Model for Flow-Through

May 30, 2003 - Three-Component Competitive Adsorption Model for Flow-Through PAC Systems. 2. Model Application to a PAC/Membrane System. Qilin Li ... ...
0 downloads 6 Views 201KB Size
Environ. Sci. Technol. 2003, 37, 3005-3011

Three-Component Competitive Adsorption Model for Flow-Through PAC Systems. 2. Model Application to a PAC/Membrane System Q I L I N L I , †,‡ B E N I T O J . M A R I N ˜ A S , * ,† VERNON L. SNOEYINK,† AND CARLOS CAMPOS§ Department of Civil and Environmental Engineering, University of Illinois at Urbana-Champaign, 205 North Mathews Avenue, Urbana, Illinois 61801, and Suez Environment-CIRSEE, 38 Rue du President Wilson, 78230 Le Pecq, France

A three-component competitive adsorption kinetic model, developed and validated in part 1 of this study, was applied to a continuous-flow PAC/membrane system to study the effects of various system and operating parameters on organic removal. The model quantitatively describes the two competitive adsorption mechanisms that occur during adsorption of trace organic compounds by powdered activated carbon (PAC) in flow-through systems where the PAC is retained in the system: pore blockage and direct competition for adsorption sites. Model simulations were conducted to investigate the effects of influent water composition, membrane cleaning water quality, PAC pore size distribution, and system operation conditions such as hydraulic retention time, membrane cleaning interval, and PAC dosing method on treatment efficiency. Effects of these factors on adsorption capacity as well as surface diffusion rate and consequent removal of the trace organic compound were discussed. It was found that optimal operating conditions for maximum trace organic compound removal must be determined on the basis of the adsorption properties and concentrations of the competing compounds in the influent. For the conditions investigated in this study, the small strongly competing compound, p-DCB, had greater impact on atrazine removal than the large poreblocking compound, PSS-1.8k. Various process design and operating parameters had complex and interrelated effects on the impact of competitive adsorption and corresponding trace contaminant removal efficiency in hybrid PAC/membrane systems.

reverse osmosis and nanofiltration membranes can remove dissolved compounds, the more commonly used ultrafiltration (UF) and microfiltration (MF) membranes are generally unable to reject dissolved species. Application of powdered activated carbon (PAC) to membrane reactors is a simple and cost-effective way to improve the removal of organic compounds, including trace organic contaminants and natural organic matter (NOM) (1-9). Recent studies have focused on the development of adsorption models in hybrid PAC/membrane systems. Campos et al. (7) modeled singlesolute adsorption in a hybrid PAC/membrane system without taking the competitive effects of NOM into account. Matsui et al. (8) developed a competitive adsorption model to predict synthetic organic compound removal by the hybrid PAC/ membrane system using the Ideal Adsorbed Solution Theory (IAST) to describe competitive adsorption equilibrium but did not consider the competitive effect on adsorption kinetics. The general applicability of these models is limited because accumulation of NOM in carbon pores can cause blockage or constriction of the pore openings and severely impact the diffusion rate of trace organic compounds (10, 11). Therefore, neglecting the effect of pore-blocking NOM on adsorption kinetics may lead to overprediction of trace organic compound removal. In part 1 of this study (12), a three-component model was developed and verified to describe the time-dependent removal of the trace organic compound atrazine in continuous-flow PAC/membrane systems in the presence of a small strongly competing compound, 1,4-dichlorobenzene (pDCB), and a large pore-blocking compound, poly(styrene sulfonate) (PSS-1.8k), which has a nominal molecular weight of 1800 Da. These two model compounds were selected as surrogates for the strongly competing fraction and the poreblocking fraction of NOM, respectively. Both competitive mechanisms, direct competition for sites and pore blockage, were taken into account so that the model could properly describe the range of effects of NOM on both adsorption equilibrium and kinetics of the trace organic compound. Major assumptions made in the model include the following: (i) membrane reactor is completely mixed for both liquid and solid phases; (ii) constant membrane flux is maintained; and (iii) adsorption kinetics is described by the homogeneous surface diffusion model (13). The main objective of the part of the study reported herein was to apply the verified COMPSORB model to investigate the effects of influent and membrane cleaning water quality as well as various system and operating parameters on the treatment efficiency of the PAC/membrane system by model simulation. Two PACs with different pore size distributions were compared to provide important information for adsorbent selection.

Model Input Parameters Introduction In the past two decades, the use of membrane filtration processes in drinking water treatment has increased very rapidly. These processes can provide high removal efficiencies of particulate contaminants such as microbes from drinking water without producing hazardous byproducts. Although * Corresponding author e-mail: [email protected]; phone: (217)333-6961; fax: (217)333-6968. † University of Illinois at Urbana-Champaign. ‡ Present address: Department of Chemical Engineering, Yale University, P.O. Box 208286, New Haven, CT 06510. § Suez Environment-CIRSEE. 10.1021/es020990j CCC: $25.00 Published on Web 05/30/2003

 2003 American Chemical Society

Model simulation was conducted for two PACs, PAC A and PAC B. Important properties of these two PACs were presented in the accompanying paper (12). Single-solute adsorption equilibrium and kinetic parameters needed to run the model were obtained in batch adsorption tests and were given in part 1 of the study (12). Model calculations were performed with the simplifying assumption that p-DCB and atrazine molecules can access the same pores and thus that the single-solute adsorption parameters of both p-DCB and atrazine could be used in the IAST model. However, it should be realized that a fraction of p-DCB adsorbs in pores too small to be accessed by atrazine and that only a portion of the p-DCB competes with atrazine (11). The noncompeting VOL. 37, NO. 13, 2003 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

3005

FIGURE 1. Effect of hydraulic retention time on removal of (a) atrazine and (b) p-DCB by PAC A.

FIGURE 2. Effect of hydraulic retention time on removal of (a) atrazine and (b) p-DCB by PAC B.

fraction of p-DCB, which does not impact atrazine kinetics or equilibrium, was neglected to simplify the demonstration of the model. Consequently, the strongly competing compound used for simulation purposes has the adsorption equilibrium and kinetics of p-DCB but roughly the same molecular size of atrazine. The pore-blockage parameters β and qcr used in the simulations were presented in the accompanying paper (12). In all model simulations, the following conditions were used unless the effect of one of them was studied: hydraulic retention time (HRT) of 5 min; flow rate of 10 mL/min; pulse PAC doses (the amount of PAC used per liter of water treated) of 0.5, 1, 2, and 4 mg/L; influent atrazine concentration of 5 µg/L; influent p-DCB concentration of 200 µg/L; and influent PSS-1.8k concentration of 10 mg/L. PAC is only wasted intermittently when the membrane is backwashed or cleaned by hydraulic flushing. An additional base condition was that the membrane was flushed with raw water so that the initial concentration in the membrane reactor was equal to that of the influent water. The magnitude of the influent concentrations of p-DCB and PSS-1.8k was chosen to represent the concentrations of the small, strongly competing and the large, pore-blocking fractions of NOM generally found in natural waters. The average effluent concentrations were calculated using the following equation:

and 2 present average effluent concentrations of atrazine and p-DCB obtained with the two PACs when HRTs from 1 to 20 min are used. The average effluent concentration is the total mass of adsorbate in the effluent divided by the total volume of water processed during the time between membrane backwashes. The HRT was varied by changing the reactor size while keeping the influent water flow rate constant. When HRT increases, the contact time between the PAC and the water increases. On the other hand, the amount of raw water in the reactor at the beginning of the filtration cyclesthis water is used to flush the membrane at the end of the previous filtration cyclesis greater because of the larger reactor size. This water is not taken into account when the amount of PAC dosed as a pulse input is calculated on the basis of the milligrams per liter PAC dose and the product water volume. Therefore, the actual milligrams per liter PAC dose is lower when the HRT is longer; and the difference between the actual and the expected PAC doses is larger at higher PAC doses. For a longer MCI, the amount of the water unaccounted for is smaller relative to the total volume of water treated. Therefore, the actual PAC dose is closer to the expected dose calculated on the basis of product water volume. Figures 1 and 2 demonstrate how HRT affects atrazine and p-DCB removal. For a MCI of 360 min, HRT has only a slight effect on both atrazine and p-DCB removal due to the reasons explained above. For example, atrazine removal increases only by a maximum of 6 and 5% for PACs A and B, respectively, (PAC dose ) 1 mg/L) when the HRT increases from 1 to 20 min. On the other hand, when a short MCI of 30 min is used, the effect of HRT is substantial for both atrazine and p-DCB removal. In some of the cases shown, the average effluent concentrations increase with increasing HRT. This is a coupled effect of adsorption capacity and adsorption kinetics. With short HRTs, the size of the membrane reactor is smaller. For a given MCI and PAC dose, the amount of PAC added to the system in a pulse dose at the beginning of the filtration cycle does not change as HRT changes. Therefore, the PAC concentration in a small reactor is much higher than that in a large reactor, which leads to a quick initial drop of the effluent concentration, as shown in Figure 3. If the MCI is short, the average effluent concentrations obtained with a small reactor are actually lower than those obtained with a large reactor. The effect of HRT on PSS-1.8k removal is very small, corresponding to

Cavg )



MCI

0

Ceff(t) dt

MCI

(1)

where Cavg ) average effluent concentration over a filtration cycle; Ceff(t) ) effluent concentration at time t; and MCI ) membrane cleaning interval or the length of a filtration cycle.

Model Application Model simulations were performed with the COMPSORB model developed in part 1 (12) to demonstrate the effect of different operating conditions and characteristics of the influent water on the treatment efficiency of the PAC/ membrane system. The factors studied include HRT, MCI, PAC type and dose, PAC addition mode, influent water composition, and membrane cleaning water quality. Hydraulic Retention Time. Determination of HRT is a very important step in system design because it determines the size of the reactor given an influent flow rate. Figures 1 3006

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 37, NO. 13, 2003

FIGURE 3. Normalized atrazine effluent concentration during a filtration cycle (PAC B dose ) 2 mg/L, MCI ) 30 min). less than 1% change in average PSS-1.8k concentration for all the operating conditions studied (data not shown) because of its low adsorption capacity and slow adsorption kinetics. It is important to notice that there may be an optimal HRT depending on the MCI and PAC dose. Therefore, the size of the membrane reactor should be determined on the basis of the PAC dose required and the MCI that provides the best organic removal while maintaining a satisfactory product water flux. Membrane Cleaning Interval. The discussion presented in the preceding section revealed that MCI is also a very important parameter in membrane system design. A more detail assessment of its impact is presented in this section. Because membrane flux decreases with filtration time because of membrane fouling, MCI determines the average flux during a filtration cycle. When treated water is used for backwash, the membrane cleaning frequency determines the daily or monthly total amount of cleaning water needed and, therefore, the net water production rate. MCI also affects the extent of membrane fouling, which in turn determines the lifetime of membranes. Since PAC is only wasted when the membrane is cleaned, MCI is equal to the carbon retention time (CRT) when PAC is added as a pulse dose and twice the average CRT when a step input of PAC is used, as explained later. Longer MCIs provide better carbon utilization because they allow more contact time for the carbon. Consequently, for a PAC/membrane system, which must achieve both adsorption and filtration, an MCI should be chosen so that the CRT is long enough to obtain sufficient adsorption while maintaining a high flux and a low degree of fouling. Model simulations were conducted with MCI values ranging from 15 to 720 min for all four PAC doses. Figure 4 presents average effluent concentrations for atrazine, p-DCB, and PSS-1.8k (normalized by their influent concentrations) predicted using the COMPSORB model for both PACs. Unlike HRT, increasing MCI substantially improved the removal of both atrazine and p-DCB, and the most improvement was obtained with the highest PAC dose. For example, atrazine removal increased from 3% to 6% for 0.5 mg/L of PAC A and from 17% to 67% for 4 mg/L of PAC A over the MCI range studied. When a longer MCI is used, the PAC had more time to adsorb, so more adsorption capacity was used, which in turn resulted in lower average effluent concentrations. However, the increased adsorption benefit from increasing MCI decreases as MCI increases, a result of lower adsorption capacity and slower adsorption kinetics of atrazine due to the accumulation of p-DCB and PSS-1.8k as explained later. As shown in Figure 4a for a PAC A dose of 4 mg/L, 41% more atrazine removal is achieved when the MCI is increased from 15 to 360 min, while an increase in MCI by another 360 min (i.e., to MCI ) 720 min) only achieved an additional 8% atrazine removal. The results from the adsorption model should be coupled with those from a membrane filtration model that includes rate of fouling to determine the most

FIGURE 4. Effect of membrane cleaning interval on the removal of (a) atrazine, (b) p-DCB, and (c) PSS-1.8k. cost-efficient MCI. The effect of MCI on the removal of PSS-1.8k was rather small as shown in Figure 4c. Less than 3% change in PSS1.8k effluent concentration was obtained for a 4 mg/L dose of PAC A when the MCI was increased from 15 to 720 min. This effect resulted again because of the low adsorption capacity and slow adsorption kinetics of PSS-1.8k. For the PAC dose range studied, PSS-1.8k removal did not change much with PAC dose. Therefore, only the results for the PAC dose of 4 mg/L are presented in Figure 4c. Figure 5 demonstrates how average atrazine equivalent Freundlich coefficient, K′ (i.e., the arithmetic average of the K′ values over a filtration cycle), changes with MCI. The general trend observed is that the average K′ value for atrazine decreases with increasing MCI because of the increasing accumulation of p-DCB on the PAC surface. When low PAC doses are used (e.g., 0.5 and 1 mg/L), atrazine K′ decreases continuously with increasing MCI because of the competition by p-DCB, which has a higher adsorption capacity and faster adsorption kinetics. At higher PAC doses (e.g., 2 and 4 mg/ L), atrazine K′ decreases initially and then reaches a plateau or increases slightly with increasing MCI. This trend is the result of the large amount of PAC added to the system at time zero as a pulse input and a corresponding higher atrazine uptake relative to p-DCB. Some of the variability depicted in Figure 5 is the result of atrazine K′ at any given time depending on the relative amount of atrazine and p-DCB in the system. This ratio changes with both time and carbon dose (12). Despite this variability, the lowest atrazine adsorption K′ values shown in Figure 5 correspond to the lowest PAC dose and the longest MCI. Figure 6 shows the average surface diffusion coefficient, Ds,avg, of atrazine during the adsorption process for doses of VOL. 37, NO. 13, 2003 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

3007

FIGURE 7. Effect of influent p-DCB concentration on atrazine removal (MCI ) 360 min).

FIGURE 5. Effect of membrane cleaning interval on atrazine equivalent Freundlich coefficient of (a) PAC A and (b) PAC B.

FIGURE 6. Effect of membrane cleaning interval on atrazine surface diffusion coefficients. 4 mg/L of PACs A and B. As depicted in Figure 6, Ds,avg consistently decreases as the MCI increases because of the pore blockage caused by accumulation of PSS-1.8k on the carbon surface. The average solid-phase concentration of PSS-1.8k increases with increasing MCI, leading to slower atrazine surface diffusion. Unlike the adsorption capacity of atrazine, the Ds,avg of atrazine is only slightly affected by PAC dose because the surface coverage with PSS-1.8k is not sensitive to PAC dose in the dose range studied. Previous studies (11, 14) have shown that, with the same solid-phase concentration of PSS-1.8k or NOM, pore blockage is more pronounced (i.e., the surface diffusion coefficient, Ds, of atrazine is lower) for PAC B than for PAC A. This difference in behavior is related to the difference in pore structure and adsorption properties of the two carbons. PAC B has a much greater PSS-1.8k Freundlich coefficient due to its larger mesopore surface area/volume, and its isotherm for this compound also has a much smaller 1/n value, indicating a much stronger energy of adsorption on this carbon (15). Thus, the average solid-phase PSS-1.8k concentration on PAC B can be much higher than that of PAC A when long MCIs are used. These factors cause the atrazine Ds,avg for PAC B to be smaller than that for PAC A for all MCIs except the shortest one of 15 min, as shown in Figure 6. Similar results (not shown) were obtained for the Ds,avg of p-DCB. 3008

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 37, NO. 13, 2003

FIGURE 8. Effect of influent p-DCB concentration on the average equivalent Freundlich coefficient of atrazine (MCI ) 360 min). Despite the lower surface diffusion coefficients of atrazine and p-DCB for PAC B, this carbon achieves more atrazine and p-DCB removal than PAC A, as shown in Figure 4. There are two primary reasons for the better performance of PAC B. First, PAC B has higher adsorption capacity for both atrazine and p-DCB. Second, its smaller particle size, 6 µm in diameter versus 10 µm for PAC A, yields faster adsorption kinetics even though the surface diffusion coefficient is lower. Influent p-DCB Concentration. Natural waters of different sources contain different amounts of strongly competing NOM. The effect of the strongly competing NOM fraction can be demonstrated by varying the influent concentration of p-DCB in the simplified three-component system. Figure 7 presents the average effluent concentration of atrazine obtained with an MCI of 360 min for influent waters containing different concentrations of p-DCB. As shown in Figure 7, the average effluent concentration of atrazine consistently increases as the influent p-DCB concentration increases for both PACs. This is because the adsorption capacity of atrazine is lower for higher influent concentrations of p-DCB, which competes strongly with atrazine for adsorption sites. Figure 8 shows how the average atrazine equivalent Freundlich coefficient, K′, of PACs A and B, calculated over a membrane filtration cycle of 360 min, changes with the influent p-DCB concentration. The average atrazine K′ of both PACs decreases rapidly with increasing influent p-DCB concentration, especially in the low p-DCB influent concentration range. An increase in p-DCB influent concentration from 50 to 500 µg/L results in a decrease up to 79 and 82% in atrazine K′ for PAC A and PAC B, respectively. The corresponding increases in average atrazine effluent concentration range from 12 to 490% for PAC A and from 17 to 500% for PAC B depending on the PAC dose used. PAC dose has a small effect on atrazine K′ possibly because of slightly different relative uptake rate of atrazine and p-DCB taking place at different PAC doses. Influent PSS-1.8k Concentration. The effect of PSS-1.8k influent concentration was also studied in order to determine the effect of the concentration of pore-blocking substances. Model simulations were conducted using an MCI of 360 min

FIGURE 9. Effect of influent PSS-1.8k concentration on the removal of (a) atrazine and (b) p-DCB (MCI ) 360 min). and four PAC doses for influent water containing PSS-1.8k concentrations ranging from 0 to 20 mg/L. Figure 9 presents average effluent concentrations of atrazine and p-DCB normalized by the influent concentrations obtained under these conditions. For PAC A, an increase of influent PSS-1.8k concentration in the concentration range from 0 to 20 mg/L causes a gradual increase in the average effluent concentration of both atrazine and p-DCB. However, for PAC B, the effect of increasing influent PSS-1.8k concentration is more pronounced in the lower PSS-1.8k concentration range. Relatively low PSS-1.8k content, 0.5 mg/L for p-DCB and 2.5 mg/L for atrazine, causes a noticeable decrease in the removal of atrazine and p-DCB, while further increases in the influent PSS-1.8k concentration result in relatively smaller increases in average effluent atrazine and p-DCB concentrations. The difference between the responses of the two PACs could be explained by their different adsorption capacity for PSS-1.8k. The PSS-1.8k adsorption isotherm of PAC B has a much higher K value, 116.3 (µg/mg)(L/µg)0.0662 and a much lower 1/n value (0.0662), consistent with its relatively high equilibrium solid-phase concentration that does not change much with the equilibrium liquid-phase concentration (15). Because atrazine adsorption does not affect that of p-DCB because of its relatively low concentrations, as demonstrated in the companion paper (12), the adsorption of p-DCB is only affected by a reduction in its surface diffusion coefficient as a result of the pore-blockage effect caused by PSS-1.8k. As shown in Figure 9b, the average p-DCB effluent concentration is more sensitive to PSS-1.8k influent concentration at lower PAC doses because of the corresponding higher PSS1.8k solid-phase concentration. In contrast to the observation for p-DCB, the adsorption of atrazine is affected by both pore blockage by PSS-1.8k and direct competition by p-DCB. Although the surface diffusion of atrazine is lower at higher PSS-1.8k concentrations because of the greater extent of pore blockage, the solid-phase concentration of p-DCB is also lower because of the pore-blockage effect, leading to less competition with atrazine. The net effect is that atrazine removal is more sensitive to influent PSS-1.8k concentration at higher PAC doses at which p-DCB effluent concentration does not change much with influent PSS-1.8k concentration (see Figure 9). A comparison of the predictions shown in Figures 7 and 9 reveals that p-DCB has a greater effect on

FIGURE 10. Effect of membrane cleaning method on the removal of (a) atrazine, (b) p-DCB, and (c) PSS-1.8k by PAC A. atrazine adsorption than PSS-1.8k because it strongly competes with atrazine for adsorption sites. Membrane Cleaning Water Quality. Low-pressure membranes are usually cleaned by pressurized backwash with treated water or hydraulic flushing using raw water depending on the configuration of the membrane reactor. The quality of the water used in membrane cleaning determines the initial concentration of organics in the reactor for each filtration cycle. Figure 10 compares the effluent concentrations obtained with PAC A when the membrane is cleaned by clean water backwash and raw water hydraulic flushing. To compare the most extreme conditions, the backwash water is assumed to contain zero organic concentration. Therefore, the reactor has an initial concentration of zero after the membrane is backwashed. The dilution effect of the clean water leads to lower effluent concentrations as compared to when raw water hydraulic flushing is used, especially for short MCIs. When short MCIs are used, the total volume of water treated in one filtration cycle is relatively small. Consequently, the dilution effect of clean backwash water in the reactor at the beginning of a filtration cycle is substantial because the volume of the clean water is a relatively large percentage of the total product water. When longer MCIs or higher PAC doses are used, the effect of dilution on the effluent concentration is not significant as compared to that of the adsorption process. Therefore, the effluent concentration curves obtained with the two membrane cleaning methods converge. Similar results are obtained with PAC B. PAC Dosing Scenario. PAC can be added to the membrane reactor either as a pulse (referred to as pulse input) or as a continuous feed (referred to as step input). The two PAC dosing scenarios provide different carbon retention time as well as carbon concentration in the reactor. For a pulse input VOL. 37, NO. 13, 2003 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

3009

FIGURE 11. Effect of PAC dosing method on the removal of (a) atrazine, (b) p-DCB, and (c) PSS-1.8k by PAC B. of PAC, all the carbon is added to the membrane reactor at the beginning of each filtration cycle and the PAC is wasted when the membrane is cleaned. Therefore, the CRT of every particle is equal to the MCI. In contrast, for a step input of PAC, the carbon particles have a linear CRT distribution, and the average CRT is equal to half of the MCI. Therefore, for the same MCI, a pulse input provides twice as much CRT as a step input. Moreover, when PAC is dosed as a pulse input, the carbon concentration in the reactor remains the maximum through the whole filtration cycle, while it increases linearly from zero at the beginning of a filtration cycle to the maximum concentration at the end of the filtration cycle in the case of step input. The higher carbon concentration provided by the pulse input results in lower solid-phase concentration of p-DCB and PSS-1.8k, leading to less competition. However, the longer CRT also results in more adsorption of PSS-1.8k and p-DCB, which compromises atrazine removal with their respective pore-blockage and direct competition effects. The benefit of longer CRT for pulse input versus step input is therefore somewhat diminished by a relatively lower adsorption capacity and surface diffusion coefficient of atrazine. Figure 11 compares the removal of all the three compounds obtained with pulse and step inputs of PAC B. A pulse input results in greater removal of all the three compounds for all the operating conditions studied, indicating that the effects of longer carbon residence time and higher carbon concentration in the reactor is more pronounced than the countering effects of pore blockage and direct competition, especially at higher PAC doses, for which the total adsorption capacity available is higher. PAC A shows similar behavior. 3010

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 37, NO. 13, 2003

PAC Dose. As explained above, PAC dose determines how HRT, MCI, and influent water composition affect atrazine removal efficiency of the PAC/membrane system. It has more meaning than the total adsorption capacity provided for atrazine. A low PAC dose may result in a continuous decrease in both atrazine adsorption capacity and surface diffusion coefficient because of the continuous accumulation of p-DCB and high surface concentration of PSS-1.8k. Therefore, a short MCI should be used when the PAC dose is low. PAC Type. The model simulations shown above reveal that p-DCB has similar effects on atrazine adsorption by the two PACs investigated (see Figure 7). This finding is consistent with the two PACs having similar surface area and volume of micropores, where both atrazine and p-DCB preferably adsorb. Because of the larger mesopore surface area and volume of PAC B, the pore-blockage effect is less pronounced for PAC B than for PAC A when the same amount of PSS-1.8k is adsorbed, although the much higher PSS-1.8k adsorption capacity of PAC B may result in a lower atrazine surface diffusion rate (see Figure 6). On the other hand, PAC B provides overall greater atrazine removal than PAC A, especially at very high influent concentrations of PSS-1.8k. As shown in Figure 9a, the surface diffusion coefficient of atrazine for PAC B becomes independent of the influent PSS1.8k concentration in the high concentration range while that for PAC A continuously decreases with increasing influent PSS-1.8k concentration. Implications for Treatment of Natural Waters. The results obtained using the model can be used as a guideline for water utilities to determine the optimal operating conditions according to the influent water composition and the characteristics of the adsorbent. For the influent concentrations and PAC doses used in the model simulation, the strongly competing compound p-DCB is found to play a more important role as compared to that of the poreblocking compound PSS-1.8k in determining atrazine removal efficiency by the PAC/membrane system because of its relatively strong effect on atrazine adsorption capacity. However, it is important to realize that p-DCB and PSS-1.8k might not completely represent the corresponding NOM fractions. In real natural water, the relative importance of the strongly competing NOM fraction and the pore-blocking NOM fraction may be different depending on the adsorption properties and molecular sizes relative to the target compound. In order for the model to be applicable to real natural waters, adsorption properties of NOM in natural water must be characterized in order to define the small, strongly competing fraction and the large, pore-blocking fraction.

Acknowledgments The authors thank L. Ding for her assistance with model computation and L. C. Schideman for helpful discussions concerning this work. This research was supported by the University of Illinois, the National Science Foundation under Grant 0123281, and Suez Environment-CIRSEE (Paris, France). The opinions in this paper are not necessarily those of the sponsors.

Literature Cited (1) Adham, S. S.; Snoeyink, V. L.; Clark, M. M.; Anselme, C. J. Am. Water Works Assoc. 1993, 85, 58-68. (2) Adham, S. S.; Snoeyink, V. L.; Clark, M. M.; Bersillon, J. L. J. Am. Water Works Assoc. 1991, 83, 81-91. (3) Pirbazari, M.; Badriyha, B. N.; Ravindran, V. J. Am. Water Works Assoc. 1992, 84, 95-103. (4) Schimmoller, L. J.; Snoeyink, V. L.; Anselme, C.; Baudin, I. Proceedings of the American Water Works Association Membrane Conference, Reno, NV, 1995; pp 295-309. (5) Clark, M. M.; Baudin, I.; Anselme, C. In Water Treatment Membrane Processes; McGraw-Hill: New York, 1996; pp 151-152.

(6) Anselme, C.; Laine´, J. M.; Baudin, I.; Chevalier, M. R. Proceedings of the American Water Works Association Membrane Conference, New Orleans, LA, 1997; pp 783-804. (7) Campos, C.; Marin ˜ as, B. J.; Snoeyink, V. L.; Baudin, I.; Laine´, J. M. J. Environ. Eng. (Reston, Va) 2000, 126, 104-111. (8) Matsui, Y.; Yuasa, A.; Ariga, K. Water Res. 2001, 35, 455-463. (9) Matsui, Y.; Colas, F.; Yuasa, A. Water Res. 2001, 35, 464-470. (10) Lebeau, T.; Lelie`vre, C.; Wolbert, D.; Laplanche, A.; Prados, M.; Coˆte´, P. Water Res. 1999, 33, 1695-1705. (11) Li, Q.; Snoeyink, V. L.; Marin ˜ as, B. J.; Campos, C. Water Res. 2003, 37, 773-784. (12) Li, Q.; Marin ˜ as, B. J.; Snoeyink, V. L.; Campos, C. Environ. Sci. Technol. 2003, 37, 2995-3002.

(13) Hand, D. W.; Crittenden, J. C.; Thacker, W. E. J. Environ. Eng. (Reston, Va) 1983, 109, 82-101. (14) Li, Q.; Snoeyink, V. L.; Marin ˜ as, B. J.; Campos, C. Submitted for publication in Water Res. (15) Li, Q.; Marin ˜ as, B. J.; Snoeyink, V. L.; Campos, C. Submitted for publication in J. Environ. Eng. (Reston, Va).

Received for review October 16, 2002. Revised manuscript received March 29, 2003. Accepted April 23, 2003. ES020990J

VOL. 37, NO. 13, 2003 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

3011