Evaluation of humic acid removal by a flat submerged membrane photoreactor

A flat submerged membrane combined with a TiO2/UV photocatalytic reactor (FSMPR) was employed in batch mode to remove humic acid (HA). HA removal efficiency was characterized by UV254 absorbance, UV-vis spectra, dissolved organic carbon (DOC) concentration, specific UV absorbance (SUVA), and trihalomethane formation potential (THMFP). The FSMPR process was effective in removing more than 86% of DOC and nearly 100% of UV254 absorbance, while the THMFPs of samples were reduced to < 19 μg/L after 150 min of treatment. In addition, changes in transmembrane pressure (TMP) with and without UV were evaluated; TiO2/UV was effective at controlling membrane fouling by HA. Analysis of the molecular weight (MW) distributions and three-dimensional excitation-emission matrix (EEM) fluorescence spectra of HAs revealed that the effectiveness in membrane fouling control is a result of changes in HA molecular characteristics. The TiO2/UV photocatalytic reactor caused the degradation of high MW, hydrophobic humic-like molecules to low MW, hydrophilic protein-like molecules, although this fraction was not completely removed during 150 min of treatment and was less responsible for membrane fouling.

Humic acid (HA), an important component of natural organic matter (NOM), is derived from the decomposition of the plant and animal materials that are commonly found in surface and ground water [1]. The presence of humic substances can impart an undesirable taste and color to drinking water. Humic materials can compete for adsorption sites with targeted compounds in coagulation treatments and activated carbon adsorption. They can also react with chlorine during water disinfection processes to form disinfection byproducts such as trihalomethanes, haloacetic acids, and haloacetonitriles. Therefore, control of HA plays an important role in treating surface water. Previous studies have reported that NOM with HA is difficult to remove using conventional treatment processes, with TOC removal efficiencies of only about 10%-50% [2]. Microfiltration (MF) and ultrafiltration (UF) membranes are effective in removing aquatic substances such as turbidity and pathogens but they are not effective for NOM removal; MF/UF treatment processes can achieve only nominal (10%) removal of NOM [3] and can contribute to the fouling of membranes [4], which is usually considered to be the most challenging factor in membrane applications. Therefore, pretreatment of feed water to membranes has become more important.
One of the chemical methods for HA removal is heterogenous photocatalysis. TiO 2 is the most commonly used photocatalyst because of its considerable photocatalytic activity, high stability, non-environmental impact and low cost [5]. Generally, a suspension of catalyst is used for high efficiency photocatalytic degradation and the catalyst particles must be separated from the treated water after the reaction. The combination of photocatalysis and membranes was designed to solve the above requirements. The membrane could play the role of both a simple barrier for the photocatalyst and a selective barrier for the molecules to be degraded. Compared with conventional photoreactors, the combination of membranes with photocatalysts (membrane photocatalytic reactors, MPR) is advantageous in confining the photocatalyst within the reaction environment by the membrane, control of a residence time of molecules in the reactor and realization of a continuous process with simultaneous products separation from the reaction environment [6][7][8][9][10]. Therefore, the combination of two processes is commonly used for two major reasons: the enhancement of the removal of NOM, and the reduction of membrane fouling. Both reasons, in turn, influenced the cost and acceptability of membrane use [11].
There have been many studies on removal of NOM by MPR, most of which focused on the effect of treatment parameters, such as pH, TiO 2 concentration, airflow, irradiation intensity, the addition of CaCl 2 , catalyst loading, catalyst species and initial HA concentration on removal efficiency [6,[12][13][14][15][16]. In addition, these studies used bulk parameters, such as TOC analysis or UV spectrophotometry to assess the treatment performance and effect on membrane fouling. Although these parameters are sufficient to give a general assessment, they cannot elucidate the specific chemistry of the process.
In this study, a flat submerged membrane photocatalytic reactor was designed for the degradation of HA in batch mode at pH 7 using commercial TiO 2 . Several complementary analytical methods were used to further understand the degradation process. DOC levels, UV 254 absorbances, UV-vis spectra, SUVA values, and THMFPs of water samples were monitored over time. In addition, the MW distributions of organic matter in the samples were determined and EEM fluorescence spectra were obtained. Transmembrane pressure (TMP) was investigated to determine its affect on membrane separation performance. These measurements are important in determining the characteristics of intermediate treatment products and assessing how these products affect membrane separation processes.

Materials
TiO 2 P25 (ca. 50 m 2 g 1 ; 25 nm) was purchased from Degussa (Hanau, Germany). Before use, 4.0 g of TiO 2 powder was weighed and mixed with a small amount of ultrapure water to prepare a TiO 2 slurry, which was added to the photocatalytic reactor. This resulted in a TiO 2 concentration of 0.5 g/L in the mixed liquor. HA was provided by Ju Feng Chemical Technology Co., Ltd., Shanghai, China. NaOH was used to improve the dissolution of HA in water. At pH 12, 1 g of HA powder was dissolved in 100 mL of a 0.1 mol/L NaOH solution. The resultant solution was diluted to 1000 mL with deionized water (MilliQ), filtered through a 0.45 μm cellulose acetate filter membrane (Xingya Purification Material Plant, Shanghai) to remove all suspended solids, and stored at 4°C. The pH values of the solutions were adjusted by adding HCl or NaOH. All reagents utilized were of analytical purity. The initial concentration of HA used in the experiment was about 10 mg L 1 (DOC). The membrane module was a polyvinylidene fluoride (PVDF) flat membrane with a mean pore size of 0.08 μm and a filtration area of 0.02 m 2 and was provided by Lantian Peier Membrane Co., Ltd., China. The membrane module was treated with an ethanol/water mixture (3/97 by volume) for 24 h and rinsed with purified water to remove residual dissolved organic matter. All membrane samples were stored in purified water at 4°C and the water was replaced regularly.

Photoreactor
The FSMPR process, shown in Figure 1, was designed and conducted on a laboratory scale. The reactor was made of stainless steel and had an effective volume of 8 L. A stainless steel light baffle separated the reactor into two parts: the photocatalytic oxidation (PCO) zone and the membrane separation zone. These were connected by a bottom flow channel and an upper overflow channel. Three low-pressure UV lamps (16 W, Philip) emitting at a wavelength of predominantly 253.7 nm were suspended vertically inside a quartz glass cylinder in the middle of the PCO zone. The intensity of radiation was 1.17 mW cm 2 . The mixture was circulated by a recycle pump and magnetically stirred during the irradiation to ensure a homogenous mixture. Air (4 L min 1 ) was supplied by a longitudinal air diffuser directly below the membrane module to provide good mixing and dissolved oxygen for photoreaction. The aeration also fluidized the TiO 2 particles and created sufficient turbulence along the membrane surface to avoid deposition of TiO 2 particles and to maintain the TiO 2 concentration.
During the experiments, the reaction temperature was maintained at about 25°C with the aid of recirculating cooling water. A constant flux of permeate (40.0 L m 2 h 1 ) was withdrawn by a peristaltic pump. The TMP in all experiments was measured by a precision pressure gauge on the permeate line. The penetrating fluid was returned to the photocatalytic reactor to keep the reactor volume constant (batch mode). Samples were taken from the permeate line at predetermined times and analyzed. To minimize the effect of adsorption, the suspension was left in the dark for 30 min before the UV light was switched on. The reaction rate constant of the photocatalytic oxidation was analyzed from 0 to 150 min.

Analytical methods
The DOC was analyzed with TOC-V CPH (Shimadzu, Japan). UV 254 absorbance and UV-vis spectra were measured by a UV-spectrophotometer (Shimadzu UV-2550, Japan). The THMFPs of samples was tested according to USEPA Method 551.1 with gas chromatography-electron capture detector (GC-ECD) (Agilent 6890N, USA) equipped with a capillary column (30.0 m × 0.32 mm × 0.25 μm, HP-5, Agilent J&W, USA). The MW distribution was measured with the gel permeation chromatography (GPC) method on a high performance liquid chromatography (LC-10AD, Shimadzu, Japan) system coupled with an SPD-20A UV detector and a TSK-GEL G3000PWXL column (7.8 mm × 300 mm). Anhydrous sodium sulfate (0.05 mol L  ) was used as the isocratic mobile phase. The separated compounds were detected by UV absorbance at 254 nm. The MW distribution pattern was derived by calibration with poly-styrene sulfonate MW standards of 14, 7.5, 4.3, 1.4, 0.7, 0.5 and 0.21 kDa. The weight and number average molecular weights (MW and MN, respectively) were calculated from the GPC results. The ratio of MW/MN or polydispersity index can be used as a measure of the width of the MW distribution.
The three-dimensional EEM spectra were measured by a fluorescence spectrophotometer (F-4600, Hitachi, Japan). In this study, the EEM spectra were collected with corresponding scanning emission spectra from 200 to 550 nm at 5 nm increments by varying the excitation wavelength. The excitation and emission slits were maintained at 10 nm and the scanning speed was set at 1200 nm/min. The software Origin 7.5 and Surfer 8.0 were employed to process the EEM data. result of the oxidation reaction, did not adsorb as strongly on the TiO 2 surface as the original compounds. Because the mineralization rate of the HA was less than the rate of photocatalytic degradation, release of these intermediate products from the TiO 2 surface led to an increase in DOC concentration in the solution. The decrease in DOC concentration afterwards is attributed to the mineralization of HA through photocatalytic oxidation. Figure 2(b) shows the changes in the UV-vis spectra during the degradation of HA at pH 7 in the FSMPR process. UV-vis absorbance at 190-600 nm of the HA solution continuously decreased and nearly disappeared with increased duration of photocatalysis. The initial drop at 0 min is ascribed to TiO 2 surface adsorption; after this point, any decrease in absorbance can be attributed to TiO 2 photocatalysis. TiO 2 nanoparticles can catalyze the decomposition of HA upon exposure to UV radiation. The decrease in UV-vis absorbance indicates that the intermediate decomposition products of the photocatalytic oxidation have no UV absorbance, especially at 254 nm, which confirms the above-mentioned conclusion.
The removal efficiency of UV 254 was higher than that of DOC, with nearly 100% removal of UV 254 and only 86% removal of DOC after 150 min treatment. The reason for this is that HA is first degraded into intermediates such as low MW organic carboxylic acids and further degraded into carbon dioxide and water [17]. The removal of DOC reflects the actual degree of mineralization of organics after the breakup of large aromatic structures. The different removal efficiencies indicate that the mineralization rate of HA was less than the rate of photocatalytic degradation. The degradation kinetics of DOC and UV 254 during FSMPR both approximately followed a pseudo-first-order rate law, as shown in Figure 2(c). The degradation rate constants of UV 254 and DOC were 0.021 and 0.010 min -1 , respectively. The apparent reaction rate constant of UV 254 is 2.1 times that of DOC. This indicates that TiO 2 exhibited excellent photocatalytic activity for the removal of UV 254 compared to the removal of DOC. After 150 min of treatment, approximately 1.5 mg L  residual DOC remained in the water sample, indicating that complete mineralization was not achieved. This may be due to the presence of refractory compounds in the original water sample or byproducts of oxidation. To achieve high mineralization efficiency, the durations of these photocatalytic oxidations were increased.
Specific UV absorbance (SUVA) is defined as a sample's UV absorbance at 254 nm divided by the DOC concentration of the sample. High-SUVA waters are generally rich in hydrophobic NOM, such as humic substances. SUVA values have also been found to correlate well with both the MW and aromaticity of aquatic NOM [18]. Therefore, SUVA values indicate aromatic compounds in DOC and can be used to estimate the chemical nature of the DOC at a given location. SUVA data are presented in Figure 2(d). The initial reduction in SUVA at 0 min indicates the preferred adsorption of hydrophobic HA on TiO 2 . The sharp decrease in SUVA at 30 min suggests that the intermediate oxidation products released from the TiO 2 surface do not absorb significantly at 254 nm, as discussed above. After this sharp decrease, the SUVA values of the HA solution continuously decreased with reaction time. It has been reported that membrane fouling and DBP formation are closely related with SUVA [19][20][21]. Therefore, the reduction in SUVA suggests that the membrane fouling and DBP formation potential of the treated water should be significantly less than that of the untreated water.
In contrast, the control experiment confirmed that there was no catalytic degradation activity in the absence of light even after an extended reaction time, which is consistent with other results [13]. All changes in DOC, UV 254 , UV-vis spectra, and SUVA values were ascribed to TiO 2 surface adsorption.

THMFP analysis
The total THMFP (TTHMFP) as a function of irradiation time was measured to assess the reactivity of degradation products with chlorine at different stages of the photocatalytic treatment. THMFP levels of HA samples subjected to different treatment times in the FSMPR process are shown in Figure 3. Only chloroform (CHCl 3 ) contributed to the total THMFP because the original HA sample had a low level of bromine. The HA sample treated for 30 min had a THMFP of 570 μg L  . In aqueous solution, chlorine is more likely to react with electron-rich sites in organic molecules such as activated aromatics (aromatic rings bearing OH, NH 2 , and heterocyclic nitrogen atoms) and 1,3-dicarbonyl aliphatics [22]. Therefore, the organic structure of NOM will strongly influence the extent of chlorine consumption and THM formation. The THMFP dropped to 19 µg L  after 150 min of treatment. The effluent THMFP levels of samples subjected to 90 min of treatment met the requirement of the Chinese National Standards for Drinking Water Quality (GB5749-2006, CHCl 3 < 60 µg L  ). This demonstrates an efficient removal of THM precursors. The low total THMFP concentration after photocatalytic treatment is consistent with the significantly reduced DOC concentration and UV 254 absorbance after oxidation. The percentage of THMFP removal exceeds that of DOC removal, as indicated by a decrease in the specific TTHMFP (TTHMFP/DOC) from 57 to 14 μg mg  . The fraction of organic matter that forms trihalomethanes is susceptible to photocatalytic treatment. Although complete mineralization was not achieved in these experiments, photocatalytic oxidation does induce changes in the chemical structure of THM pre cursors, rendering them less reactive to chlorine [23].

Membrane performance analysis
Changes in TMP were monitored to evaluate membrane fouling. Under normal circumstances, when a certain membrane flux has been guaranteed, TMP should ideally be as low as possible. When TMP increases, this indicates a rise in membrane fouling resistance, which shortens the membrane wash cycle and reduces the membrane life [24]. Membrane fouling during the filtration of different solutions occurs because of the formation of a cake layer and membrane pore plugging, which leads to an increase in TMP. The resistance-in-series model was applied to evaluate fouling characteristics during a 240 min filtration. The resistances of membrane, cake layer and pore plugging were measured (Table 1). Membrane resistance is constant and depends on physical and chemical properties such as membrane thickness and morphological features. Cake resistance is mainly determined by porosity, thickness and compressibility of the cake. Pore blockage resistance depends on the amount of intruding photocatalysts or HA inside the membrane pore. According to this model, the membrane flux, J, can be expressed as follows [25][26][27][28][29]: where J is the membrane flux (L m 2 h 1 ), P  is the TMP (Pa), µ is the viscosity of permeate (Pa s), R t is the total resistance (m 1 ), R m is the intrinsic membrane resistance, R c is the cake resistance, and R p is the fouling resistance due to pore plugging. The TMP during filtration of the TiO 2 solution (0.5 g L 1 ) only slightly increased initially. The resistance was caused by minor TiO 2 pore blockage resistance (0.2×10 11 m 1 ) and minor cake resistance (0.3×10 11 m 1 ). The aeration below the module may provide shearing force along the membrane surface, preventing TiO 2 particle deposition and reducing the TMP. Humic substances contained in natural waters have been demonstrated to be the foulant that causes membrane fouling in a number of studies [4,30,31]. The TMP increment rate during filtration of the HA solution (DOC 0 , 10 mg L 1 ) was 15 Pa/min. The resistance was caused by major HA pore blockage resistance (3.3×10 11 m 1 ) and minor cake resistance (0.2×10 11 m 1 ). The TMP increment rate during filtration of the HA (DOC 0 , 10 mg L 1 ) and TiO 2 (0.5 g L 1 ) mixed solution without UV was 23 Pa/min, almost 1.54 times the rate during the filtration of the HA-only solution. The resistance was caused by major HA pore blockage resistance (2.9×10 11 m 1 ) and major cake resistance (2.7×10 11 m 1 ). Lee et al. [12] investigated the hybrid TiO 2 photocatalytic membrane reactor for HA degradation with and without ultraviolet light and illustrated the possible mechanisms of membrane fouling by TiO 2 and HA. Humic acids can occupy the vacancies between TiO 2 particles and a portion of the deposition layer resistance actually increased when the TiO 2 particles and humic acids were mixed together. Thus, humic acid-laden TiO 2 particles can be expected to give rise to a greater resistance to permeation. The cake resistance of the mixture (2.7×10 11 m 1 ) was more than nine times higher than the resistance due to TiO 2 alone (0.3×10 11 m 1 ) and more than thirteen times higher than that of HA alone (0.2×10 11 m 1 ). The degradation of humic acids present in suspension or bound to the TiO 2 particle surfaces played a central role in the decline of TMP when UV irradiation was provided. The TMP during filtration of the HA (DOC 0 , 10 mg L 1 ) and TiO 2 (0.5 g L 1 ) mixed solutions with UV slightly increased initially, then remained constant throughout the rest of the filtration. The resistance was caused by minor TiO 2 pore blockage resistance (0.4×10 11 m 1 ) and minor cake resistance (0.6×10 11 m 1 ). The stabilized TMP indicates that membrane fouling did not occur and that TiO 2 /UV photocatalysis was effective in controlling membrane fouling by HA.

Characteristics of MW distribution
GPC chromatograms are presented in Figure 5. The chromatogram of the initial sample at 30 min has a broad molecular weight distribution and consists mostly of large molecules (1-10 kDa), which exhibit a high response. These are typical characteristics of hydrophobic aromatic and long-chain aliphatic molecules. The MW distribution of   To better evaluate and understand MW distributions, the analysis of MN, MW and the coefficient of MW distribution(MW/MN) were investigated in this study. The MW, MN and MW/MN ratio of the initial sample were 2403 Da, 198 kDa, and 12.1, respectively, while those of samples subjected to 150 min of treatment were 1246 Da, 296 Da, and 4.2, respectively. These results indicate that the FSMPR effluent had much narrower MW distribution after treatment.

Three-dimensional EEM fluorescence spectra
Fluorescence EEMs are very useful for distinguishing between different types of organic matter [32]. In this study, photodegradation of humic substances caused great changes in the EEM fluorescence properties of humic acids and the peak intensities and positions of peaks in fluorescence EEMs obtained from treated HA samples both changed with increasing photocatalytic irradiation time. As shown in Figure 6, there were different peak positions and intensities in the fluorescence EEMs of HA solutions that had undergone different durations of treatment in the FSMPR process. The intensities of two dominant peaks, Peak A and Peak T, decreased with time in the FSMPR system, indicating that the fluorescence characteristics of HA were gradually changed with photooxidation.
The dominant excitation and emission wavelengths (Ex/Em) of 275 and 445 nm (Peak A intensity, 970.5) of the initial sample treated for 30 min (Figure 6(a)) match the excitation and emission wavelengths reported for humic  acid-like fluorescence [18,[33][34][35]. After 30 min of dark adsorption ( Figure 6(b)), the fluorescence EEMs decreased in peak intensity (from 970.5 to 639.1) but the location of Peak A (Ex/Em, 275/445) was unchanged. This indicates that there was no new organic matter appearance and that TiO 2 adsorption did not change the chemical structure of HA. As shown in Figure 6(c), the Peak A intensity (Ex/Em, 265/435; intensity, 354.1) decreased after 30 min of photocatalysis, while a new peak, Peak T (Ex/Em, 215/330; intensity, 491.7), appeared. Peak T has been reported as a protein-like peak [36,37]. Both the red and blue shifts of Ex/Em were 10 nm. A red shift is related to the presence of carbonyl-containing substituent, hydroxyl, alkoxyl, and carboxyl groups as well as amino groups [38]. A blue shift is associated with the decomposition of condensed aromatic moieties and the breakup of large molecules into smaller fragments. This breakup may include a reduction in the degree of the -electron system, a decrease in the number of aromatic rings, a reduction of conjugated bonds in a chainstructure, a conversion of a linear ring system to a nonlinear system, or an elimination of particular functional groups including carbonyl, hydroxyl and amine groups [39][40][41]. The intensities of Peak A (Ex/Em, 260/420; intensity, 139.7) and Peak T (Ex/Em, 230/350; intensity, 247.7) decreased after 90 min of treatment ( Figure 6(d)). Peak A disappeared and Peak T (Ex/Em, 215/330; intensity, 296.3) slightly increased after 150 min of treatment ( Figure 6(e)), indicating that the protein-like peak fraction was due to byproducts of humic acid oxidation. Based on the analysis of GPC, the presence of this fraction represented by the main peak at 1565 Da had not caused membrane fouling, as discussed in Section 2.3.

Conclusions
The removal of HA at pH 7 by FSMPR was investigated in batch mode using multiple analysis methods. The following conclusions can be drawn: (1) The FSMPR process could remove more than 86% of DOC and nearly 100% of the UV 254 absorbance of HA after 150 min treatment, as well as the effective decrease of THMFP.
(2) The TiO 2 /UV pretreatment is very effective in controlling membrane fouling by HA. No remarkable increase of TMP was observed as the filtration process with the photocatalytic reactions took place.
(3) The hydrophobic humic-like molecules with high MW were degraded to hydrophilic protein-like molecules with low MW as a result of photocatalysis reaction, which were less responsible for membrane fouling.