Membrane filtration of natural organic matter: factors and mechanisms affecting rejection and flux decline with charged ultrafiltration (UF) membrane

Membrane filtration of natural organic matter: factors and mechanisms affecting rejection and flux decline with charged ultrafiltration (UF) membrane

Journal of Membrane Science 164 (2000) 89–110 Membrane filtration of natural organic matter: factors and mechanisms affecting rejection and flux decl...

637KB Sizes 7 Downloads 100 Views

Journal of Membrane Science 164 (2000) 89–110

Membrane filtration of natural organic matter: factors and mechanisms affecting rejection and flux decline with charged ultrafiltration (UF) membrane Jaeweon Cho a,b , Gary Amy b , John Pellegrino c,∗ a

c

KJIST, Department of Environmental Science and Engineering, 1 Oryong-dong, Puk-gu, Kwangju 500-712, South Korea b Civil, Architectural, and Environmental Engineering, University of Colorado at Boulder, Boulder, CO 80309, USA National Institute of Standards and Technology, Physical and Chemical Properties Division, MS 838.01, Boulder, CO 80303, USA Received 14 September 1998; received in revised form 14 May 1999; accepted 18 May 1999

Abstract We studied natural organic matter (NOM) rejection and the membrane’s flux decline during natural water filtration using a charged ultrafiltration membrane based on thin-film-composite technology. NOM rejection mechanisms such as steric exclusion and aromatic/hydrophobic and charge interactions were considered. Water composition factors affecting NOM rejection and flux decline were investigated, including ionic strength, pH, and calcium ion concentration. The membrane’s effective relative molecular mass cutoff for the NOM in our study was between 1500 and 2300 (significantly lower than the manufacturer’s nominal value of 8000) and depended on the NOM characteristics in the source water. In particular the ratio of UV absorbance at 254 nm to dissolved organic carbon (related to the humic content) correlated with the rejection. Comparison of relative molecular mass distributions between fractionated NOM and recovered membrane foulants indicates that the foulants are the larger-sized neutral and/or basic NOM components, and not the humic substances that were efficiently rejected by this membrane. ©2000 Elsevier Science B.V. All rights reserved. Keywords: Flux decline; Fouling; Natural organic matter; NOM; Ultrafiltration: water treatment

1. Introduction This paper focuses on the filtration of natural organic matter (NOM) from natural waters (and NOM isolates from those waters) using a thin-film-composite (TFC) ultrafiltration (UF) membrane. The natural waters were prefiltered using a microfilter (with nominal 0.45 ␮m pore size) to minimize contributions to flux decline due to biofilm formation. ∗ Corresponding author. Tel.: +1-303-497-3416; fax: +1-303-497-5259. E-mail address: [email protected] (J. Pellegrino).

The emphasis is on describing the electrostatic and molecular-size interactions between NOM and the membrane that provide much of the basis for solute rejection and adsorptive flux decline. The measurements and analytical protocols we are studying may also be useful for the further development of predictive correlations for the filtration figures-of-merit. In addition, the NOM size-exclusion characteristics and transport properties for this particular membrane are more completely measured in the context of drinking water treatment. Flux decline and NOM rejection are two important issues in membrane filtration. Both are influenced by

0376-7388/00/$ – see front matter ©2000 Elsevier Science B.V. All rights reserved. PII: S 0 3 7 6 - 7 3 8 8 ( 9 9 ) 0 0 1 7 6 - 3

90

J. Cho et al. / Journal of Membrane Science 164 (2000) 89–110

water composition factors (bulk NOM concentration, humic/non-humic NOM fraction, molecular mass distribution, pH, Ca++ concentration and ionic strength); membrane physical and chemical properties (pore size, water permeability, charge, hydrophobicity/philicity); and filtration process conditions (the NOM concentration at the membrane interface, controlled by flux rate and mass transfer in the fluid boundary layer). NOM adsorption and gel-layer formation not only affects flux decline but will also affect the observed rejection characteristics. A rigorous characterization of NOM in raw and membrane-treated waters can provide guidance as to the applicability of various membrane processes. NOM characterization includes several analytical methods such as NOM fractionation by humic/nonhumic character, molecular size analysis by high-pressure size exclusion chromatography (HPSEC), aromaticity by UV absorbance (an index of the NOM structure), and acidic functional group content by potentiometric titration of NOM isolates. Membrane characterization includes contact angle as an index of relative hydrophobicity, zeta potential for indirect measurement of surface charge, and nominal relative molecular mass cutoff (MWCO) using polyethylene glycols (PEGs).

2. Background NOM properties, including structure (aromatic versus aliphatic, or hydrophobic versus hydrophilic), size (average relative molecular mass (RMM), and RMM distribution), and charge density, are important factors in the formation of disinfection by-products (DBPs) [1–3] and can also be influential factors in membrane processes due to hydrophobic and charge interactions [4–6]. UV absorbance at 254 nm (UVA254 ) has been used to monitor not only NOM concentration, but also the humic content or aromaticity of NOM based on specific UVA254 (SUVA = UVA254 /dissolved organic carbon (DOC)) [2,7]. NOM fractionation by XAD-8 (a nonionic adsorbent from acrylic ester polymer with nominal pore size 23.5 nm) and XAD-4 (a nonionic adsorbent from polystyrene with nominal pore size 5 nm) resins has been used to obtain hydrophobic DOC (XAD-8 isolate), transphilic DOC (XAD-4 isolate),

and hydrophilic DOC (effluent from XAD-8 followed by XAD-4) [8–10]. For example, XAD-8 was used to obtain hydrophobic NOM and hydrophilic NOM water (the effluent of XAD-8) for hollow-fiber membrane filtration by Nilson and DiGiano [6]. These groups of organic matter have different characteristics of hydrophobicity, size, shape, and charge density. The molecular size of NOM must always be an important factor influencing its rejection by membrane filtration. The relative molecular mass or size of NOM has been estimated by both UF membrane gel filtration and HPSEC methods. Ultrafiltration was performed with several different regenerated cellulose membranes (coded as YM series) and a stirred-cell dead-end unit to obtain RMM fractions between two nominal MWCOs [5,6,11]. In this case, using UF for RMM fractionation is not significantly influenced by pH due to the uncharged nature of YM membranes [11]. However, when NOM concentration is relatively high, the concentration polarization near the membrane surface is high and may influence the rejection (as measured by either DOC or UVA). When concentration polarization exists, a permeation coefficient model was suggested to calculate the percent rejection [12]. Chin et al. [13] used the HPSEC technique to analyze RMM distribution and to calculate weight-averaged RMM (Mw ), number-average RMM (Mn ), and polydispersivity (Mw /Mn ) (an index of NOM homogeneity). RMMs determined by HPSEC were smaller than those measured by the UF method, according to Chin et al. [13] and a comparison of results for fulvic acids from Jucker and Clark [5]. Charge density may also be an influential factor on NOM rejection and flux decline during filtration, because membranes are generally negatively charged [14,15]. The NOM charge density of bulk natural waters is not easy to measure by titration with base because of interference from inorganic solutes. Instead, just the hydrophobic acids (XAD-8 isolate) have been used in potentiometric titrations (pH between 3 and 8) to measure carboxylic acidity [16]. For the phenolic group content of NOM, twice the amount of NaOH required to titrate from pH 8–10 was used as an estimate [7]. It is desirable to measure the charge density of transphilic (XAD-4 isolate) and hydrophobic (XAD-8 isolate) acids in order to more completely elucidate electrostatic effects on rejection.

J. Cho et al. / Journal of Membrane Science 164 (2000) 89–110

Membrane properties, including hydrophobicity, pore size and surface charge, are also necessary for evaluation of NOM rejection, and associated fouling, during membrane filtration. Membrane hydrophobicity can be inferred by the material itself and by contact angle measurements. Contact angle is an index of hydrophobicity of the membrane surface, which can be measured by captive bubble and sessile drop methods. Zhang and co-workers [17,18], and Gekas et al. [19] used cellulose acetate (CA) and polysulfone (PSf) membranes to compare contact angles of relatively hydrophilic and hydrophobic membranes. There was no significant difference in contact angles between CA and PSf membranes in the measurements of Zhang and Hallstrom [18] and Gekas et al. [19]; however, CA exhibited higher contact angles than PSf in the results of Zhang et al. [17]. PSf membranes from different manufacturers showed very different contact angles [17,18]. Polyethersulfone (PES) membranes (coded as PM series) and regenerated cellulose membranes (the YM series) exhibited a large difference in contact angles based on the measurements of Jucker and Clark [5]. The contact angle of membrane surfaces is not easy to accurately measure because of the effects of surface texture, porosity and water wicking, but it is still regarded as a convenient semi-quantitative index of membrane hydrophobicity/hydrophilicity. Comparisons of contact angle measurements between material surfaces are more meaningful if the entire protocol (including elapsed times) is kept consistent for each sample. Membrane MWCO is the basic tool for predicting NOM rejection and other solute separations by membrane filtration because size exclusion is such an important mechanism. Manufacturers provide membrane MWCO information from the results of solute rejection tests using macromolecules such as proteins, dextrans and PEGs — often uncharged. As an alternative, recently, Ohya et al. [20] determined the MWCO of an asymmetric aromatic polyimide membrane using toluene solutions of aliphatic hydrocarbons with different RMMs. Direct measurement of the membrane surface charge is very difficult. Instead, the zeta potential (which is the potential at the shear plane in the diffuse layer) is used. It can be estimated using the streaming potential method. Causserand et al. [21] and Jucker and Clark [5] measured the zeta potentials of UF

91

membrane pores, while Elimelech and co-workers [14,15] measured the zeta potentials of the membrane surface (active layer). The Helmholtz–Smoluchowski equation was used to calculate the zeta potential from the streaming potential by all of these researchers. The zeta potentials of clean membranes were measured at different pH’s and ionic strengths, and membranes with either a protein [21] or humic acids [5] adsorbed were also measured. Also a Suwannee River humic acid solution itself was used as a background electrolyte for the zeta potential measurement of Childress and Elimelech [15]. The zeta potentials of membranes adsorbed with a protein or a humic acid were less than those of clean membranes before adsorption [5,21]. Zeta potentials became more negative as pH increased for PES and sulfonated PSf membranes (sulfonated PSf had the higher negative charge) [21], and for PA TFC membranes [15]. The influence of ionic strength on PEG rejection by a sulfonated PSf membrane was shown to provide higher PEG rejection with higher ionic strength, thus indicating that the pore radii of the membranes are decreased by higher ionic strength [22]. This is an indirect evidence of the surface electrostatic forces of the membrane material. NOM removal by nanofiltration (NF) and UF membranes is generally based on polyamide TFC, polysulfone, polyethersulfone, and sulfonated-polyethersulfone polymers. Beyond the membrane’s ability to reject the target solutes, flux decline and fouling of the membranes are important factors for economic reasons. Ideally, the membrane surface should have a negative charge to repel NOM, which also has a negative charge, thus inhibiting its adsorption and helping to maintain a high flux. The charge repulsion between NOM and the membrane surface, and NOM adsorption, will also be influenced by water-quality factors such as pH and ionic strength. Flux declines were measured for PSf and sulfonated PSf hollow-fiber membrane filtration of Suwannee River organic matter at different pH’s by Braghetta et al. [22], while Nilson and DiGiano [6] filtered a bulk NOM solution, a hydrophilic NOM solution (XAD-8 effluent), and a hydrophobic NOM solution (XAD-8 isolate). They observed that flux declines were relatively significant at pH 4 and 7, but were negligible at pH 10, probably due to the fact that the acidic components of NOM have a higher negative charge

92

J. Cho et al. / Journal of Membrane Science 164 (2000) 89–110

density at higher pH (from carboxylic and phenolic functional groups) thus inhibiting NOM adsorption and the resulting flux decline. Hydrophobic interactions are influential factors on flux decline, as indicated by results that a hydrophobic NOM solution (XAD-8 isolate) exhibited more flux decline than a hydrophilic NOM solution (effluent of XAD-8) [6]. A hydrophobic PSf membrane with a hydrophobic nonionic surfactant was also used to demonstrate more flux decline than a relatively hydrophilic membrane with the same surfactant [23]. This result substantiates the significance of hydrophobic interactions in the foulant-adsorption process. Further results along this vein were obtained with a stirred-cell, dead-end filtration apparatus used to measure flux declines with PES and RC UF membranes [4]. The PES membrane (PM series) exhibited more flux decline than the RC membrane (YM series). According to these studies, it seems likely that PSf and sulfonated PSf membranes can be fouled more easily by hydrophobic components of NOM due to the hydrophobicity of the membrane surface. Several NF membranes based on PA TFC (NF90 and TFCS) and sulfonated-PES (NTR7450) were used to measure flux declines with an NOM-containing natural (Ohio River) water [24]. The PA TFC membranes exhibited very little or almost no flux decline for 4–13 days even with a recycle line (which increased the bulk solute concentrations), while the sulfonated PES membrane showed greater flux decline over 4–7 days and required frequent membrane cleaning to maintain the specified flux. This review of prior work provides the basis for the measurements we have applied. The results in this paper will focus on filtration with a relatively ‘tight’ UF TFC that is negatively charged. The results provide both an improved qualitative understanding of the interaction of the heterogeneous NOM mixture and a negatively charged membrane, and further experimental measurements with which to advance our predictive capability.

ments. According to the manufacturer, the surface of this membrane is made of a combination of aromatic and aliphatic multifunctional monomers, and has ionizable functional groups such as carboxylic acids. A regenerated cellulose membrane (coded as YM3) which is highly hydrophilic but does not have ionizable functional groups was also used only for NOM rejection experiments. Thus, these filtration measurements provided a comparison of the solute rejection between charged and uncharged membranes. The surface zeta potentials for both membranes were estimated from streaming potential measurements using a KCl electrolyte solution (30 mS/m) and a commercial electrokinetic analyzer (EKA) measurement apparatus. The contact angle was measured between a water droplet, the membrane surface, and air, using a goniometer (sessile drop method). The membrane is first rinsed by floating it, skin side down, in a container of DI water (a deionized water prepared by filtration with two proprietary cation-exchange mixed beds, an anion-exchange bed, and a 0.2 ␮m filter) for 1 day, and changing the water three times. This procedure extracts water-soluble coating materials (for example, glycerine). The rinsed membranes are dried in a closed desiccator for a day and stored in a closed petri dish before measurements. Membrane samples are cut into small pieces and mounted on a support. An approximately 2.0 ␮l droplet of DI water is placed on the membrane specimen and the contact angle is measured with the goniometer immediately after the drop placement. Table 1 presents a listing of the characteristics of the membranes, all of which were provided by the manufacturers except for the contact angles, zeta potentials, and average clean-water permeance. The YM3 and GM membranes may be considered relatively hydrophilic and hydrophobic, respectively, based on our contact angle measurements.

3.2. Source waters 3. Methods and analyses 3.1. Membrane A membrane of the TFC (coded as GM) type was used for the NOM filtration and flux decline measure-

Two drinking water sources, Silver Lake surface water (SL-SW) and Orange County groundwater (OC-GW), representing relatively hydrophilic versus hydrophobic sources of NOM, respectively, were used to perform bench-scale membrane-filtration

J. Cho et al. / Journal of Membrane Science 164 (2000) 89–110

93

Table 1 Summary of membranes tested Code

Material

Contact angle

Manufacturer’s nominal MWCO

Zeta potential at pH 7 (mV)

Clean water permeance (l per day/m2 /kPa)

YM3 GM

Regenerated Cellulose Polyamide TFC

13.3◦ 54.7◦

3000 8000

−8.6 −17.0

4.51 8.00

Table 2 Characteristics of source watersa Source

DOC (mg/l)

UVA (/cm)

SUVA (/cm/mg l)

Baseflow SL-SW Runoff SL-SW OC-GW

2.00 (±0.10) 3.88 (±0.19) 9.80 (±0.49) 47.80 (±2.39)

0.048

0.024

0.172

Twitchell

Conductivity (␮S/cm)

pH

Alkalinity (ppm CaCO3 )

Humic content (% DOC)

21.4

6.2

0.045

29.9

6.4

11.7 (±0.1) –

0.480

0.049

477.0

8.8

1.770

0.037

1066.0

7.1

43.3 (±3.1) 56.9 (±3.4) 80.0 (±5.7) 60.6 (±3.9)

222.0 (±18.0) 79.0 (±6.4)

Ca (mg/l) 8.1 – 4.7 34.3

a

Note: In general, three replicate measurements were made and the coefficient of variation was less than 5%. Insignificant variations were observed in measurements of UVA, conductivity, pH, and Ca++ . The numbers in parenthesis are one standard deviation.

tests. SL-SW samples were collected during baseflow (normal) and runoff (spring snowmelt) periods to include seasonal variations in NOM. Unless otherwise indicated, all membrane filtration tests were performed on 0.45 ␮m pre-filtered water, corresponding to the definition of dissolved organic carbon (DOC). A third water source was also used: Twitchell water, representing an agricultural drainage feeding into the California State Water Project. Each source water was analyzed for DOC, UV absorbance at 254 nm (UVA254 ), specific UV absorbance at 254 nm (SUVA = UVA254 /DOC), conductivity, pH, alkalinity, Ca++ concentration, and humic (XAD-8 adsorbable) content of the NOM. The humic fractions of NOM source waters were determined by performing a DOC mass balance across an XAD-8 resin column, with the column effluent representing the non-humic fraction. These results are tabulated in Table 2. A typical natural water contains 40–60% humic matter as defined by isolation/adsorption onto XAD-8 resin. A water with a 40/60 split between humic and non-humic would be considered a ‘more non-humic water’ while a 60/40 split would infer the opposite. Our waters ranged from about 40% humic (fairly representative) to 80% (an ‘outlier’).

All the source waters used in these filtration measurements were stored under refrigeration prior to their use. The bulk water analysis was always repeated at the same time as the filtration measurements. We observed no significant change in composition over the period of this study and since the feed waters were prefiltered by nominal 0.45 ␮m microfilters, biological organisms were not considered to be a significant part of the membrane challenge. HPSEC was used to determine the RMM distribution of NOM. We used a modified silica column (separation range of 138–35 000 mass units) and a commercial UV spectrophotometric detector. Eluent for the HPSEC was composed of DI water buffered with phosphate (pH 6.8) and NaCl to provide ionic strength of 0.1 M [13]. Standard solutions for the RMM calibration curve were made with sodium polystyrene sulfonates (PSS) (1800, 4600, 8000, and 35000 mass units), and salicylic acid (138 mass units) was used to confirm the lower range of the calibration curve. The RMM distributions of baseflow and runoff SL-SW, OC-GW, Twitchell water, and several XAD-isolate solutions were determined. These RMM distributions and average RMMs are shown in Figs. 1 and 2and Table 3, respectively.

94

J. Cho et al. / Journal of Membrane Science 164 (2000) 89–110

Fig. 1. Relative molecular mass distribution (by SEC) of the NOM contained in the natural water sources: (䊐) baseflow Silver Lake surface water (SL-SW), (䊏) runoff Silver Lake surface water (RSL-SW ), (䊊) Orange County ground water (OC-GW), and (∇) Twitchell drainage.

Fig. 2. Relative molecular mass distribution (by SEC) of the NOM fractions contained in the (䊏) runoff Silver Lake surface water: (heavy solid line) hydrophobic DOC, (thin solid line) hydrophilic DOC, and (dashed line) transphilic DOC.

XAD-8 and XAD-4 resins were used to isolate the hydrophobic DOC (primarily polycyclic aromatic acids, the XAD-8 isolate), transphilic DOC (primarily the relatively hydrophilic aliphatic acids, the XAD-4 isolate), and hydrophilic DOC (the ef-

fluent from XAD-4 column whose feed had passed through an XAD-8 column) from the runoff SL-SW. The RMM distributions of the hydrophobic DOC, transphilic DOC, and hydrophilic DOC derived from the runoff SL-SW are presented in Fig. 2. We also ar-

J. Cho et al. / Journal of Membrane Science 164 (2000) 89–110

95

Table 3 Relative molecular mass determinationsa Source

Mn (number-averaged RMM)

Mw (weight-averaged RMM)

polydispersivity (Mw /Mn )

Runoff SL-SW

1270 (±15) 1330 785 804 984 965 1160 (±96) 1350 (±58)

1490 (±95) 1750 953 941 1200 1120 1590 (±41) 1700 (±45)

1.17

Hydrophobic DOC of runoff SL-SW Transphilic DOC of runoff SL-SW Hydrophilic DOC of runoff SL-SW Baseflow SL-SW Hydrophilic DOC of baseflow SL-SW OC-GW Twitchell water

1.31 1.21 1.17 1.22 1.16 1.37 1.26

a Note: three replicate measurements for the RMM distribution were used to calculate the M and M of the runoff SL-SW, OC-GW, n w and Twitchell feed waters. The calculated standard deviation for those results are in the parenthesis.

tificially increased the DOC concentration of baseflow SL-SW from 2.0 to 9.2 mg/l by using the retentate from a nanofiltration step with a membrane having MWCO of 300 (NF70). From this ‘DOC-augmented’ baseflow SL-SW we produced an additional water sample: the effluent from an XAD-8 column providing a hydrophilic, nonhumic NOM solution (DOC = 5.2 mg/l). 3.3. Continuous crossflow, flat sheet membrane unit A commercial bench scale crossflow membrane cell was used to evaluate the filtration properties of the flat sheet specimens. The system is comprised of the membrane unit, the feed, permeate and retentate streams. The retentate line is also divided into recycle and waste lines (see Fig. 3). The system accommodates 60 cm2 flat sheet specimens under tangential feed flow with a channel height of 0.04 cm. Feed flowrates are variable between 100 and 450 ml/min. The crossflow velocity can be adjusted by varying the bulk feed flow, and this crossflow velocity was kept approximately constant for all measurements in this report at ∼8.6 cm/s by setting up a constant feed flowrate of 200 ml/min. The temperature was maintained at 298 K and the transmembrane pressure was kept constant at approximately 345 kPa (50 psi). At these conditions the Reynolds number is laminar, nominally 36. The system recovery R, which is defined as ratio of permeate to fresh feed mass flow, can be controlled by changing the amount of recycle from the retenate back

to the feed line (see Fig. 3). Higher system recovery ratios (for a given fresh feed concentration) will result in a higher NOM concentration at the membrane surface. For each filtration test, a new membrane specimen was soaked in DI water for 1 day to rinse away water soluble surface coatings. Clean water was filtered through the membrane until approximately constant flux was obtained, then the NOM solution (equilibrated to room temperature) was filtered. The flow rate, UVA, and DOC of the permeate were measured over time. The SUVA of the permeate was compared with that of the feed sample to provide a measure of the preferential rejection of the aromatic (hydrophobic) fraction of the NOM. System recovery R, as defined above, is the fraction of fresh feed flow that is recovered as permeate. In this case, retentate recycle is used to maintain the crossflow velocity at a consistent level. The purpose of this mode of operation is to obtain higher bulk concentrations of NOM. When retentate recycle was used to increase the system recovery, bulk rejection Rj (bulk) and feed rejection Rj (feed) was defined by R (bulk) = R(feed) =

Cb − Cp Cb CR (1 − R) . Cf

(1)

(2)

Cb is the bulk NOM concentration in the module, Cp is the NOM concentration in the permeate, CR is the

96

J. Cho et al. / Journal of Membrane Science 164 (2000) 89–110

Fig. 3. Schematic of crossflow membrane filtration apparatus.

NOM concentration of the retentate, and Cf is NOM concentration of fresh feed.

3.5. Flux decline and adsorption tests using crossflow filtration unit

3.4. Polyethylene glycol MWCO determinations of the membrane

Extending on the approaches of many prior investigations that used three-parameter models [25–28], a five-parameter membrane resistance-in-series model was used to quantify the influences on flux decline:

A range of PEGs (RMM = 200–10 000) were used to determine the membrane’s MWCO under the same crossflow filtration operating conditions as used in the NOM filtration measurements. A 20 mg/l concentration as DOC was used for each of the three PEG rejection measurements: in DI water; in DI water containing 10 mM NaCl; and in DI water containing 4 mM Ca++ . NaCl and Ca++ were added to determine the effects of increased ionic strength and calcium binding on the membrane’s apparent MWCO. The purpose of these measurements were not to ‘define’ the MWCO of the membrane but to provide a standard benchmark with a neutral, linear macromolecule with which to compare NOM rejection.

Jv =

1P µ(rm + rc + rg + ra1 + ra2 )

(3)

where Jv is flux through the membrane (cm/s), 1P is transmembrane pressure (Pa), µ is dynamic viscosity (Pa s or g s cm−1 ), rm is membrane hydraulic resistance, rc is concentration polarization resistance, rg is gel layer resistance, ra1 is weak adsorption resistance, ra2 is strong adsorption resistance (all resistances are in cm−1 ). For our study, the concentration polarization, gel layer, and weak adsorption resistances may be considered to be reversible by clean water and NaOH extraction, while strong adsorption resistance is not. Note that in this model the osmotic pressure

J. Cho et al. / Journal of Membrane Science 164 (2000) 89–110

97

Fig. 4. Schematic representation of the filtration protocol used to determine resistances in series.

term is subsumed into the concentration polarization resistance. DI water and NOM water were crossflow filtered using the following protocol (illustrated schematically in Fig. 4) to obtain all resistances. Clean water was first filtered through the membrane until a constant flux was obtained (Step 1); then NOM-containing water was introduced and the permeate rate was monitored over time (Step 2). After the permeate rate reached a constant value (that is, the permeate rate of the fouled membrane), DI water replaced the NOM-containing water, and the applied pressure was released to remove concentration polarization (Step 3). The fouled membrane was then rinsed (the flow rate was increased from 8.6 to 19.3 cm/s) with DI water so that the gel layer (highly concentrated NOM layer) was removed from the membrane surface and DI water filtration was again performed (Step 4). The membrane was then soaked in a 0.1 M NaOH solution for 1 day so that weakly adsorbed NOM on the membrane surface could be desorbed, then DI water was again filtered through it (Step 5). Using the flux values from steps 1–5, we could calculate rm , rc , rg , ra1 , and ra2 . The difference between concentration polarization and gel layer resistances is that the former is a thermodynamic modification of the pressure driving force, and the latter is the viscous resistance for flow through highly concentrated (precipitated or gelled) solutes. The weak adsorption can be defined as the NOM adsorption that can be removed by chemical cleaning with 0.1 M NaOH. Strong adsorp-

tion is attributed to the NOM that cannot be desorbed even with this chemical cleaning. The RMM distribution was determined (using HPSEC) for the NOM desorbed by 0.1 M NaOH extraction of a fouled membrane. The foulants were compared with the RMM distributions of hydrophobic, transphilic and hydrophilic DOC fractions from the runoff SL-SW feed water. 3.6. General discussion of measurement uncertainties Based on systematic uncertainties from resolution of the mass balance, volumetric standards, and timing devices, we estimate the uncertainties in the permeate flux measurements to be 0.2–1% of the reported values. The lower uncertainties are for the initial periods when the permeation rates are higher. Carrying this uncertainty into the calculation of the tabulated resistances-in-series, we estimate an uncertainty between 1% and 10% of the reported value, again depending on its magnitude. The expanded uncertainties (coverage factor of 2) due to random and systematic effects (based on 3 replicate analyses) in the reported NOM rejections are ±2–5% when based on DOC, and ±0.5–2% when based on UVA254 . The expanded coverage on the specific RMM values are estimated to be ±60 mass units based on the variance of peak times observed for replicate measurements with the RMM standards. 2–3 replicate measurements for the peak retention times on the mass standards were made such that standard deviation was within 0.01 min (the coefficient of variation was 0.1–0.2% over the range of

98

J. Cho et al. / Journal of Membrane Science 164 (2000) 89–110

Fig. 5. Mass % rejection of PEG by YM3 (solid line) and GM (dashed line) versus the PEG relative molecular mass under crossflow filtration and different solution conditions.

retention times). The same strategy was applied for the determination of RMM distribution of the water samples. 4. Results and discussion 4.1. PEG MWCO of membrane The rejection of PEG by the GM and YM3 membranes are presented against the average RMMs of the PEG fractions in Fig. 5. The MWCO of a membrane is typically defined as the RMM that is 90% rejected when the macromolecules used are non-charged model compounds [22,29]. As shown in Fig. 5, the GM membrane apparently has a MWCO value greater than 8000 for neutral linear macromolecules. As the RMM of PEG increases, its rejection by the GM increases almost linearly. Thus, the GM membrane exhibits a diffuse type of cut-off which implies a wide pore size distribution. With added NaCl or Ca++ in the DI water, PEG rejections (at a given RMM) increased. The YM3 membrane’s nominal MWCO exhibited no effect from the added ions and matched its manufacturer’s specification.

We attribute our results to the supposition that the pore size of the GM membrane is affected by the charge interactions between its functional groups, such as the electrostatic repulsion between nearby carboxylates. Therefore, an increase in ionic strength and/or ion binding (such as with Ca++ ) would decrease the double layer and allow functional groups to approach closer to each other. Whether such a phenomenon would increase or decrease the apparent mean of the membrane’s pore size distribution depends on the specific morphology present. In the case of the GM membrane, it appears that the effective pore size distribution has a lower mean value when charge screening occurs. Our results and interpretations are consistent with the general observations of Braghetta et al. [22] for their measurements using sulfonated polysulfone nanofiltration membranes to filter solutions of varying pH and ionic strength. 4.2. NOM rejection mechanisms: size exclusion, hydrophobic interactions, and charge interactions When other factors are excluded, membrane filtration is a physical barrier process, rejecting macro-

J. Cho et al. / Journal of Membrane Science 164 (2000) 89–110

99

Fig. 6. Relative molecular mass distribution (by SEC) of the NOM contained in the (䊏) runoff Silver Lake surface water (RSL-SW ) feed, (thin solid line) permeate at 5% recovery, (medium dashed line) retentate at 5% recovery, and (heavy dashed line) retentate at 95% recovery.

Table 4 Size exclusion by GM membrane based on weight-average RMM Baseflow SL-SW OC-GW Runoff SL-SW Twitchell Mw Feed 1200 Permeate 827 NOM rejection (%) DOC 37.6 UVA 63.2

1590 925 84.2 94.0

1490 958 60.2 84.8

1700 1210 59.9 65.2

molecules larger than the membrane pore size. As UF membranes may have a wide pore size distribution (diffuse type), and NOM comprises a range of macromolecule structures and sizes, it is not straightforward to illustrate (or predict) macromolecule rejection by size exclusion effects. The Mw of the different feed waters and GM permeates are compared in Table 4. The Mw of the permeates are 20–24% less than those of feed waters for the various NOM sources. The RMM distributions of the GM permeate (recovery ratio = 5%) and GM retentate (recovery ratios = 5% and 95%), collected at steady state flux conditions, are compared to the RMM distribution of the feed (runoff

SL-SW) in Fig. 6. (Note: The retentate from a significantly more concentrated stream was required in order to discern any significant shift in the RMM distribution between fresh feed and retentate). A shift to smaller and larger RMMs, for the permeate and reject stream occurred, respectively. The vertical axis of Fig. 6 represents a normalized fraction percentage obtained by dividing each incremental height of the chromatogram with a sum of the heights when the chromatogram is divided into incremental mass intervals. As shown in Fig. 6, the 200–300 RMM fraction for the GM permeate is greater than that of the feed; meanwhile, the RMM fraction larger than 2000 is almost completely rejected. The RMM distribution shape of the GM retentate stream is generally similar to that of feed (runoff SL-SW), however, the distribution is shifted to a larger RMM by about 1000. Several source waters with different RMMs, SUVAs and humic contents (% DOC), were used for NOM filtration through the GM membrane. In general, when the source waters have a larger Mw , NOM based on DOC and UVA is more easily rejected by the GM membrane than source waters having a smaller Mw (see Fig. 7). But if we were to consider only av-

100

J. Cho et al. / Journal of Membrane Science 164 (2000) 89–110

Fig. 7. NOM% rejection based on (clear bars) DOC and (shaded bars) UVA during GM membrane filtration of the various feed waters. Crosshatched band between 10% and 20% represents expected DOC rejection based on size exclusion from the PEG measurements. Tabulated below each feed water are several NOM metrics.

erage molecular mass, then based on the PEG measurements all rejections should have been in the range of 10–20%. There is a generally monotonic relationship between NOM rejection and increasing SUVA content of the feed stream. In fact, even though the OC-GW NOM has a smaller Mw than Twitchell water, since its SUVA is higher, so is the rejection for both DOC and UVA. On the other hand, hydrophilic NOM (effluent of XAD-8 derived from the baseflow SL-SW) with both very low RMM and hydrophobicity (lowest SUVA), has virtually no measurable rejection of either DOC or UVA. Indeed, there is also a positive correlation between NOM rejection and humic content (based upon DOC). These results suggest that NOM aromaticity/hydrophobicity may become a quantitative predictor of NOM rejection. It appears that all the humic fraction of the NOM can be rejected by the GM membrane. This assertion is illustrated by the GM retentate stream from the runoff SL-SW having almost the same RMM distribution as

the hydrophobic DOC (XAD-8 isolate derived from the runoff SL-SW) (see Fig. 8). In fact, for all the waters (except the hydrophilic NOM solution) the hydrophobic fraction of NOM is preferentially excluded from the membrane, as indicated by the fact that the UVA rejection is higher than the DOC rejection. NOM rejections by the YM3 membrane are compared with those by the GM membrane (see Fig. 9) to evaluate charge interactions using several natural waters. The YM3 is presumably non-charged (actually less negatively charged based on zeta potential measurements) than the GM membrane (Table 1) and has a MWCO of 3000 (manufacturer’s rating and our measurements, Fig. 5). Based on a comparison of their nominal MWCOs, we anticipated that the YM3 membrane would exhibit higher NOM rejection than the GM membrane. However, the opposite was true. The GM membrane had higher NOM rejections than the YM3 membrane except for the case of the hydrophilic NOM-containing water. The higher NOM rejections

J. Cho et al. / Journal of Membrane Science 164 (2000) 89–110

101

Fig. 8. Relative molecular mass distribution (by SEC) of the NOM contained in the (䊐) hydrophobic fraction from runoff Silver Lake surface water and (heavy dashed line) retentate at 95% recovery from the same feed water.

Fig. 9. DOC% rejection during membrane filtration of the various feed waters using (clear bars) YM3 and (shaded bars) GM membranes filtration. Crosshatched band between 10% and 20% and 30% and 40% represents expected DOC rejection for GM and YM3, respectively, based on size exclusion from the PEG measurements and the average relative molecular mass of the DOC in the feed water.

102

J. Cho et al. / Journal of Membrane Science 164 (2000) 89–110

Fig. 10. pH versus acidity (milliequivalents per gram of C from DOC measurements) of the NOM isolates from the (䊉) XAD-8 and (䊊) XAD-4 fractionation of runoff Silver Lake surface water.

by the GM membrane are believed to be due to charge repulsions between the negative surface charge of the GM membrane and the negative charge density of NOM acidic fractions. The charge density of NOM fractions from runoff SL-SW were measured by potentiometric titrations (see Fig. 10). The charge density (per mass of DOC) of XAD-4 isolate (hydrophilic acids with aliphatic structure) is higher than that of the XAD-8 isolate (hydrophobic acids with aromatic structure), but both have significant acidity. The number of charged groups in the two types of isolates on an average molar basis would seem to be relatively equivalent when the average RMM is considered (see Table 3). The average solute molecular size (based on number average) of the OC-GW < runoff SL-SW < Twitchell (see Table 3) but they are all between 1150 and 1350. The YM3’s NOM rejection follows the order, OC-GW < Twitchell < runoff SL-SW, but the variation in measured rejection between the waters is small (between 47% and 55% based on DOC). Therefore, one can judge that the primary mechanism of NOM rejection for the YM3 membrane is a steric hindrance.

On the other hand, the GM’s NOM rejection follows OC-GW > Twitchell ≈ runoff SL-SW, which is the same order as the humic fraction. But it is instructive to note that it is only with the OC-GW that the NOM rejection of the GM significantly exceeds that of the YM3. The combination of the higher humic content and negative charge density of the OC-GW resulted in a solute that was significantly more excluded from the GM membrane than the YM3. This leads us to speculate that there is a strong synergy between the aromaticity (and therefore, shape instead of simply RMM) and charge density of the solutes that controls the partitioning of NOM constituents into membrane pores. This synergy may find its strongest expression when the membrane material has significant intrinsic charge density.

4.3. Solution conditions affecting NOM rejection The effects on NOM rejection by the GM membrane from changes in ionic strength, Ca++ concentration, and pH were evaluated with runoff SL-SW and Twitchell water (see Figs. 11 and 12). Only

J. Cho et al. / Journal of Membrane Science 164 (2000) 89–110

103

Fig. 11. NOM % rejection based on (clear bars) DOC and (shaded bars) UVA during GM membrane filtration of runoff Silver Lake surface water under various feed water modifications. Ambient means the unmodified R SL-SW.

Fig. 12. NOM% rejection based on (clear bars) DOC and (shaded bars) UVA during GM membrane filtration of Twitchell drainage water under various feed water modifications. Ambient means the unmodified Twitchell.

[Ca++ ] significantly influenced NOM rejection with either water. This is likely attributable to calcium binding between NOM molecules [30] and possibly also the negatively-charged membrane surface. Either

case would result in reduced surface charge repulsions between the NOM and the membrane. Hong and Elimelech [31] reported a decrease in water permeation through an aromatic-polyamide TFC

104

J. Cho et al. / Journal of Membrane Science 164 (2000) 89–110

nanofiltration (NF) membrane when increasing the Ca++ concentration. A thicker, more dense ‘fouling’ layer due to the rejected, coiled NOM was their reasonable explanation for those results. But they also reported almost no difference in the NOM rejection (based on total organic carbon — TOC) for their model humic substances with the increase in Ca++ concentration. The NF membrane they used had NOM rejection (under baseline conditions) typically over 90%, while the GM UF membrane has only 55–60% DOC rejection with the waters in these measurements. We consider that the intramolecular Ca++ binding decreases the electrostatic interaction, in one case between a molecule and a relatively inpenetrable surface, and in the other case, between the molecule and a pore mouth that it can actually ‘fit’ in! Diminishing the strength of the electrostatic repulsion in the former case allows the rejected molecules to more tightly aggregate as a layer on the surface, but in the latter case allows the formerly-rejected molecules to pass into and through the pore structure. One can also suppose that the intramolecular Ca++ binding decreases the NOM molar volume (as Hong and Elimelech [31] suggested by the descriptor ‘coiled’), then that fraction of the NOM size distribution in our measurements that was close to passing through the membrane (in the absence of Ca++ ) would then be small enough to permeate and lower the observed rejection. Slightly lowering the pH only has a minor effect on NOM rejection because a significant portion of the NOM acidity for both hydrophobic and hydrophilic acids is within a pH range of 3–4 (see Fig. 10). In the pH range between 6 and 8, there is significantly less acidity for those components. Clearly, our justification for a minor pH effect on rejection also requires no major change in the membrane surface charge characteristics in the same range of pH values. We did in fact do the streaming potential measurements with varying pH and ionic strength. The zeta potential varied smoothly from −15 to −17.7 mV as the pH was changed from 5.5 to 8.5. The increase in ionic strength also appeared to insignificantly influence NOM rejection in our measurements. It was anticipated that since increased ionic strength can cause a compacted double layer of the NOM macromolecules a significant reduction in rejection (due to the reduced negative charge density) would occur. However, our other measurements (see

Fig. 5) indicated an apparent reduction in GM membrane pore size was caused by a similar increase in ionic strength. Therefore, we infer that the effects on macromolecule charge and size and the membrane pore geometry were offsetting. When a recycle line was used to increase the recovery ratio up to 95%, NOM rejections can be calculated by two methods: based on bulk rejection by Eq. (1) and feed rejection by Eq. (2). The NOM bulk rejections at the 95% recovery was more than 90% based on both DOC and UVA with the increased bulk concentration (near the membrane surface) caused by the recycle stream (see Fig. 11). When NOM feed-based rejections were calculated, DOC and UVA rejections were 45.8% and 56.1%, respectively.

4.4. Solution conditions affecting flux decline and resistances in series Flux declines were monitored over time with runoff SL-SW as the bulk water and with that bulk water spiked with 10 mM NaCl; spiked with 4 mM Ca; pH adjusted to 4.32; and filtered at 95% recovery (see Fig. 13). The permeance, or water mass transfer coefficient (MTCw , l m−2 per day kPa−1 ), calculated at various times in the filtration process, is used as the general index of flux. When normalized by the initial flux, the decline (versus unmodified bulk water) was similar for all cases except that of the increased recovery ratio, that is, the increased bulk NOM concentration lead to a larger flux decline, as expected. The flux decline was also measured (Fig. 14) using each of the different source waters (runoff SL-SW, OC-GW, and Twitchell). Filtration of OC-GW has slightly less flux decline than runoff SL-SW, even though the OC-GW has a higher DOC concentration and hydrophobicity. More intuitively, though, the very high DOC concentration of the Twitchell water causes the largest flux decline of all three waters. These results indicate that the DOC fractions that are most effectively excluded (rejected) by the GM membrane are not the same fractions that contribute to increased flux decline. To evaluate the flux decline more quantitatively, the resistances in series defined in Eq. (3) were calculated. These are presented in Table 5, including for runoff SL-SW at the modified solution conditions

J. Cho et al. / Journal of Membrane Science 164 (2000) 89–110

105

Fig. 13. Permeance (l/m2 per day/kPa) versus time in crossflow filtration of runoff Silver Lake surface water under various feed water modifications. The unmodified feed water filtered at 5% recovery is the heavy solid line which connects the actual data points, whose symbols were omitted for visual clarity. (䊉) unmodified at 95% recovery, (䊊) with 10 mM NaCl at 5% recovery, (䊐) with 4 mM Ca2+ at 5% recovery, and () pH 4.32 at 5% recovery. The increase in permeance at end is due to flux recovery conditions.

(ionic strength, pH, Ca, recovery ratio). There are no significant differences except for the increased recovery ratio (95%) case and the Twitchell water. The concentration polarization and gel layer resistances are very small compared with weak adsorption resistance ra1 , suggesting that NOM-fouled membranes need to be chemically cleaned. Membrane flux recoveries (ratio of clean water flux after chemical cleaning to initial clean water flux) from fouled membranes are more than 90% for all cases, indicating effective cleaning for NOM-fouled membranes using a caustic solution. 4.5. Effective MWCO determination As previously mentioned, the charged GM membrane rejected NOM mass fractions are smaller than expected based on the membrane’s nominal MWCO (8000). It can be envisioned that there is a certain apparent MWCO for the charged membrane with the charged NOM. This was determined using the NOM

fractional rejection by Eq. (4) with the RMM distribution of the membrane permeate [29]. RMi =

WMi (feed) − WMi (perm)(1 − Roverall ) WMi (feed)

(4)

Here, RMi is the fractional rejection for a certain RMM, WMi is the mass fraction of that RMM, and Roverall is overall NOM rejection based on DOC. The NOM fractional rejections by the GM membrane with runoff SL-SW are shown in Fig. 15, along with the RMM distributions of runoff SL-SW, and the corresponding permeate after filtration at 5% recovery. The effective MWCO (based on a RMi value of 90%) was determined using Eq. (4) and are listed with SUVA values and Mw in Table 6 for the three source waters. The much smaller effective MWCOs for the GM membrane for the NOM contained in our test waters, compared to either the manufacturer’s nominal value or PEG-based measurements we made, are attributed to charge and hydrophobic interactions (note

106

J. Cho et al. / Journal of Membrane Science 164 (2000) 89–110

Fig. 14. Permeance (l/m2 per day/kPa) versus time in crossflow filtration of: (solid line) runoff Silver Lake surface water, (short dashes line) Orange County ground water, and (long dashes line) Twitchell drainage water. The lines connect the actual data points whose symbols have been eliminated to increase visual clarity. All filtrations were at 5% recovery with GM membrane. Increase in permeance at end is due to flux recovery conditions. Table 5 Resistances in series Feed waters

rm (cm−1 )

rc (cm−1 )

rg (cm−1 )

ra1 (cm−1 )

ra2 (cm−1 )

Flux recovered (%)

runoff SL-SW runoff SL-SW and NaCl (10 mM) runoff SL-SW and Ca++ (4 mM) runoff SL-SW @ pH = 4.3 runoff SL-SW @ R = 95% OC-GW Twitchell water

213.4 187.66 209.2 180.8 213.4 210.6 209.4

0.0 0.88 3.2 0.44 0.0 0.50 9.06

6.26 11.16 6.18 4.68 26.98 4.98 45.82

61.68 26.94 62.78 46.44 245.16 34.02 135.22

0.0 6.16 1.42 0.0 0.0 1.08 5.08

100 96.8 99.3 100 92.5 99.4 97.6

the inverse correlation between SUVA and MWCO) between the membrane surface and NOM. Thus, we see once again that the effective or observed MWCO is not an absolute membrane quantity, but is dependent on the specific solute chemistry that is presented to it. 4.6. Analysis of GM membrane foulants The results presented in Figs. 13 and 14 suggest that hydrophobic fractions of NOM did not cause significant flux decline. Even after pH was decreased and

ionic strength was increased to screen the NOM’s negative charge and enhance the hydrophobic interactions, flux decline was not significantly increased (see Fig. 13). This fact suggests that other NOM fractions such as transphilic DOC (XAD-4 isolate) and hydrophilic DOC (effluent of XAD-8 and 4, the neutrals and bases) may be foulants of the GM membrane. RMM distribution of NOM desorbed from the fouled GM membrane by a NaOH solution was compared to that of the hydrophilic DOC derived from runoff SL-SW (see Fig. 16). These SEC chromatograms reveal that the

J. Cho et al. / Journal of Membrane Science 164 (2000) 89–110

107

Fig. 15. Relative molecular mass distribution of NOM in runoff Silver Lake surface water (䊏) feed and (solid line) permeate (filtration at 5% recovery with GM membrane). Also plotted is the (dashed line) calculated % rejection of specific relative molecular masses. The 90% rejection point is ∼1580. Table 6 Effective MWCOs of the GM membrane (manufacturer’s nominal MWCO = 8000) NOM source

SUVA (/cm/mg l)

Mw of NOM

Effective MWCO

Twitchell water Runoff SL-SW OC-GW

0.037 0.045 0.049

1670 1490 1590

2220 1580 1500

hydrophilic DOC has two main peaks in the RMM distribution, and the larger RMM peak (between 700 and 2000) is very similar to the larger peak of desorbed NOM from the fouled GM membrane. The Mw and polydispersivity of the larger RMM peak of hydrophilic DOC are 937 and 1.04, respectively. These are very close to those of the desorbed NOM from the fouled GM membrane, 1,150 and 1.04, respectively. If we compare this larger RMM peak of desorbed NOM with the four different RMM distributions of XAD isolates from runoff SL-SW, there are no matching peaks except for the larger RMM peak of hydrophilic DOC (see Figs. 2 and 16). This supports the assertion that the fraction described as hydrophilic DOC is one of the membrane foulants responsible for flux decline. The smaller RMM region of the hydrophilic DOC seems to correlate with a similar region in the GM per-

meate when comparing the RMM distributions of hydrophilic DOC and GM permeate (see Fig. 17). There is also good agreement between RMM distributions of the GM permeate and transphilic DOC associated with runoff SL-SW. This suggests that the GM permeate is composed of a combination of transphilic DOC (hydrophilic acids) and the smaller RMM of hydrophilic DOC (neutrals or bases). The hydrophobic DOC (hydrophobic acids) is rejected by the GM membrane due to its negative charge (electrostatic repulsion) and high RMM. Thus, in the case of the GM membrane, we hypothesize that the larger hydrophilic fractions of the NOM (without significant ionizable functionality) are major components of the adsorbed foulants that lead to significant, long-term flux decline. The specific mechanisms may include pore mouth adsorption and subsequent narrowing of the pores, since these species are

108

J. Cho et al. / Journal of Membrane Science 164 (2000) 89–110

Fig. 16. Relative molecular mass distribution of (䊊) hydrophilic DOC fraction and (heavy solid line) NaOH desorbed DOC from filtration (at 5% recovery with GM membrane). Both DOCs are from runoff Silver Lake surface water.

Fig. 17. Relative molecular mass distribution of (䊊) hydrophilic DOC fraction, (䊏) transphilic DOC fraction, and (heavy solid line) permeate from filtration (at 5% recovery with GM membrane). All DOCs are from runoff Silver Lake surface water.

small enough not to be excluded from the membrane on strictly steric considerations. Initial adsorption by these species would screen electrostatic repulsions, thus allowing further mass to accumulate including the larger aromatic NOM components.

5. Conclusions NOM rejection by a (negatively) charged UF membrane is higher than expected based solely on steric considerations, because of the combination of the size,

J. Cho et al. / Journal of Membrane Science 164 (2000) 89–110

shape, aromaticity and the negative charge density of this NOM. The NOM’s aromaticity (with larger RMM) and charge are contained in the hydrophobic acids (i.e., XAD-8 resin column isolate) which have a greater negative charge due to carboxylic and phenolic moieties than neutrals and bases (effluent of XAD-8/4). The hydrophilic acids, even with a high negative charge density, could not be rejected completely by this negatively charged UF membrane. The lower RMM fractions permeate the membrane presumably due to the combinations of overall size and shape. NOM rejection by the charged UF membrane is decreased when calcium ions are in solution. This may be due to charge neutralization of both the NOM and the membrane surface. Small changes in ionic strength and pH had minor effects on NOM rejection by this charged UF membrane. There is preferential rejection of the aromatic/hydrophobic NOM fraction, which is indicated by the higher rejection based on UVA versus DOC. This fraction also has a larger RMM and negative charge/molecule than the other fractions. With the same source water, the recovery ratio (based on the fresh feed rate) can influence flux decline by increasing the bulk NOM concentration. Again, small changes in ionic strength and pH are minor factors affecting flux decline by the charged UF membrane, but, in this case, added Ca++ concentration is not a significant factor. This latter observation is presumably due to the small role of humic substances in causing flux decline for this membrane and natural waters. This presumption was supported by further measurements with different source waters where variations in NOM aromaticity/hydrophobicity apparently did not influence the flux decline of this charged UF membrane. We think the neutral and base fraction of the NOM to be a factor in this membrane’s fouling based on RMM distributions determined using HPSEC. Our flux recovery measurements indicate that the fouled membrane can be cleaned effectively only by chemical cleaning. In our case, we used 0.1 N NaOH but other cleaning formulations may be even more effective. Effective relative molecular mass cutoff (MWCO) was confirmed to be dependent on the specific source waters, and SUVA was a consistent indicator of the trend. Increasing SUVA content correlated with lower apparent effective MWCO.

109

The emergent physical picture is that the molecular shape and molar charge density of the NOM determines whether it will be excluded from the membrane’s pore structure. RMM (by SEC) is one indirect indicator of these characteristics of the NOM. Other chromatographic measures like XAD-8 and XAD-4 resin fractionation are further indicators. NOM molecules that are not completely excluded from the membrane’s pores, but that are large enough and can adsorb near the pore openings do so, and are the primary species leading to flux decline. These species are removed with NaOH cleaning, an observation that supports the hypothesis that they are adsorbing near the pore mouth. NOM molecules whose combination of size and charge are low enough not to be influenced by the membrane’s pore structure are found in the permeate. Further measurements to determine chromatographic indicators of molecular size, shape, functionality, and charge will be helpful in providing improved quantitative correlations for the membrane filtration behavior of complex mixtures.

Acknowledgements This work was supported by the American Water Works Association Research Foundation - Traci Case, project manager.

References [1] G.L. Amy, P.A. Chadik, Z. Chowdhury, Developing models for predicting THM formation potential and kinetics, J. Am. Water Works Assoc. 79 (1987) 89–97. [2] J.K. Edzwald, W.C. Becker, K.L. Wattier, Surrogate parameters for monitoring organic matter and THM precursors, J. Am. Water Works Assoc. 77 (1985) 122–132. [3] G.V. Korshin, C.-W. Li, M.M. Benjamin, Monitoring the properties of natural organic matter through UV spectroscopy: a consistent theory, Water Res. 31(7) (1997) 1787–1795. [4] J.M. Laine, J.P. Hagstrom, M.M. Clark, J. Mallevialle, Effects of ultrafiltration membrane composition, J. Am. Water Works Assoc. 11 (1989) 61–67. [5] C. Jucker, M.M. Clark, Adsorption of aquatic humic substances on hydrophobic ultrafiltration membranes, J. Membr. Sci. 97 (1994) 37–52. [6] J. Nilson, F.A. DiGiano, Influence of NOM composition on nanofiltration, J. Am. Water Works Assoc. 88 (1996) 53–66.

110

J. Cho et al. / Journal of Membrane Science 164 (2000) 89–110

[7] D.M. Owen, G.L. Amy, Z.K. Chowdhury, Characterization of natural organic matter and its relationship to treatability, AWWA Research Foundation, 1993. [8] E. Thurman, R. Malcolm, Preparative isolation of aquatic humic substances, Environ. Sci. Technol. 15 (1981) 463–466. [9] J.A. Leenheer, T.I. Noyes, A filtration and column adsorption system for on site concentration and fractionation of organic substances from large volumes of water. U.S. Geological Survey Water-Supply Paper 2230, 1984. [10] G.R. Aiken, D.M. McKnight, K.A. Thorn, E.M. Thurman, Isolation of hydrophilic organic acids from water using nonionic macroporous resins, Org. Geochem. 18(4) (1992) 567–573. [11] G.L. Amy, M.R. Collins, C.J. Kuo, P.H. King, Comparing gel permeation chromatography and ultrafiltration for the molecular mass characterization of aquatic organic matter, J. Am. Water Works Assoc. 79 (1987) 43–49. [12] B.E. Logan, Q. Jiang, Molecular size distribution of dissolved organic matter, J. Environ. Eng. ASCE 116 (1990) 1046– 1062. [13] Y. Chin, G. Aiken, E. O’Loughlin, Molecular weight, polydispersity, and spectroscopic properties of aquatic humic substances, Environ. Sci. Technol. 28 (1994) 1853–1858. [14] M. Elimelech, W.H. Chen, J.J. Waypa, Measuring the zeta (electrokinetic) potential of reverse osmosis membranes by a streaming potential analyzer, Desalination 95 (1994) 269–286. [15] A.E. Childress, M. Elimelech, Effect of solution chemistry on the surface charge of polymeric reverse osmosis and nanofiltration membranes, J. Membr. Sci. 119 (1996) 253– 268. [16] M.R. Collins, G.L. Amy, C. Steelink, Molecular weight distribution, carboxylic acidity, and humic substances content of aquatic organic matter: implications for removal during water treatment, Environ. Sci. Technol. 20 (1986) 1028–1032. [17] W. Zhang, M. Wahgren, B. Sivik, Membrane characterization by the contact angle technique II. characterization of UF-membranes and comparison between the captive bubble and sessile drop as methods to obtain water contact angles, Desalination 72 (1989) 263–273. [18] W. Zhang, B. Hallstrom, Membrane characterization using the contact angle technique I. methodology of the captive bubble technique, Desalination 79 (1990) 1–12.

[19] V. Gekas, K.M. Persson, M. Wahlgren, B. Sivik, Contact angles of ultrafiltration membranes and their possible correlation to membrane performance, J. Membr. Sci. 72 (1992) 293–302. [20] H. Ohya, I. Okazaki, M. Aihara, S. Tanisho, Y. Negishi, Study on molecular weight cut-off performance of asymmetric aromatic polyimide membrane, J. Membr. Sci. 123 (1997) 143–147. [21] C. Causserand, M. Nystrom, P. Aimar, Study of streaming potentials of clean and fouled ultrafiltration membranes, J. Membr. Sci. 88 (1994) 211–222. [22] A. Braghetta, F.A. DiGiano, W.P. Ball, Nanofiltration of natural organic matter: pH and ionic strength effects, J. Environ. Eng. ASCE 123(7) (1997) 628–641. [23] K. Yamagiwa, H. Kobayashi, A. Ohkawa, M. Onodera, Membrane fouling in ultrafiltration of hydrophobic nonionic surfactant, J. Chem. Eng. Jpn. 26(1) (1993) 13–18. [24] S.C. Allgeier, A.M. Gusses, T.F. Speth, J.J. Westrick, R.S. Summer, Verification of the rapid bench-scale membrane test and ICR requirements, Proc. 1996 AWWA Workshop on GAC, Membranes, and ICR, 3–5 March 1996, Cincinnati, Ohio. [25] A. Suki, A.G. Fane, C.J.D. Fell, Flux decline in protein ultrafiltration, J. Membr. Sci. 21 (1984) 269. [26] H. Nabetani, M. Nakajima, A. Watanabe, S. Nakao, S. Kimura, Effects of osmotic pressure and adsorption on ultrafiltration of ovalbumin, AIChE J. 36 (1990) 907. [27] A.P. Peskin, M.K. Ko, J. Pellegrino, Three layer membrane model for characterizing ultrafiltration membranes, J. Membr. Sci. 60 (1991) 195–206. [28] M.K. Ko, J. Pellegrino, Determination of osmotic pressure and fouling resistances and their effects on performance of ultrafiltration membranes, J. Membr. Sci. 74 (1992) 141–157. [29] M. Mulder, Basic Principles of Membrane Technology, 2nd ed., Kluwer Academic Publishers, Dordrecht, 1996. [30] J.G. Hering, F.M.M. Morel, Principles and applications of aquatic chemistry, Wiley, New York, 1992. [31] S. Hong, M. Elimelech, Chemical and physical aspects of natural organic matter (NOM) fouling of nanofiltration membranes, J. Membr. Sci. 132 (1997) 159–181.