Ultrafiltration fouling behavior of natural organic matter: a perspective of EEM and molecular weight distribution

NOM

natural organic matter

UF

ultrafiltration

RDA

redundancy analysis

PARAFAC

parallel factor analysis

AFM

atomic force microscope

SEM

scanning electron microscope

CM

combined mechanism

FEEM

fluorescence excitation emission matrix

EEM

excitation emission matrix

BP

biopolymers

SMP

soluble microbial product

HPSEC

high-pressure size exclusion chromatography

PLC

programmable logic controller

DOC

dissolved organic carbon

TOC

total organic carbon

UV₂₅₄

ultraviolet absorbance at 254 nm

AMWD

apparent molecular weight distribution

Subscripts

i

intermediate blocking

c

cake layer fouling

NOM is a complex matrix of organic materials, and its amount, character, and properties vary considerably according to the origins of the waters and depend on the biogeochemical cycles of their surrounding environments (Sillanpää 2015). Ultrafiltration (UF) is a promising technology to remove NOM with little chemical agent. However, the consequent membrane fouling has blocked the further application of UF to water treatment. Membrane fouling is the deposition of foulants on the membrane surface or in the pores, which results in the permeate flux decline or operational pressure increase. NOM is known as the most detrimental membrane foulant during water treatment (Shi et al. 2014; Wang et al. 2017b) and many studies have been focused on the contribution of NOM to membrane fouling till now (Kennedy et al. 2005).

Researchers have studied the influence of NOM characteristics (hydrophobicity/hydrophilicity, molecular weight, organic group, etc.) on membrane fouling. Chen Fei et al. conducted a pilot scale investigation of ultrafiltration for drinking water treatment between 2008–2010 and pointed out that the protein-like substances/biopolymers in NOM were major foulants during membrane fouling (Chen et al. 2014). Hiroshi Yamamura et al. put forward that the hydrophilic fraction resulted in severe fouling during ultrafiltration with the comparison of fouling difference between hydrophilic and hydrophobic fraction of NOM (Yamamura et al. 2014). In addition, contributions of fluorescent natural organic matter on membrane fouling were studied by Wang Hui et al. and fluorescent protein content played a key part in fouling layer formation (Wang et al. 2017a).

According to the mechanical sieving principle in low-pressure membranes, NOM fractions with greater molecular weight can block the membrane pores; while the fractions with relatively smaller molecular weight can enter the membrane pores interior. Thus, NOMs with different molecular weight fractions may have varied membrane fouling potential. Previous study showed that the high molecular weight fraction of NOM (>30 kDa) was responsible for the greater flux decline; for example, Kimura Katsuki et. al. pointed out that biopolymers (BP) with the molecular weight of >1 million Dalton had severe fouling potential for the large-size BP since they were not allowed to enter micro-pores. Lin et al. fractionated humic substances according to MW with gel permeation chromatography and hydrophobicity with resins. They found that the highest molecular weight (6.5–22.6 kDa) components for both hydrophobicity and hydrophilic fractions caused the greatest flux decline. Through examination of effects of molecular weight distribution of soluble microbial product (SMP) on membrane fouling, Teng Jiaheng et al. pointed out that the specific filtration resistance of SMP increased with molecular weight (Teng et al. 2020). However, through the filtration of different molecular fractions of HA, Li Chiwang et al. examined that the low molecular fraction (<3 kDa) caused the most severe membrane flux decline compared to other fractions, which they attributed to the interaction force between the molecules rather than the molecular weight (Li & Chen 2004).

These characteristics all have an impact on membrane fouling, but cannot give a good explanation of membrane fouling alone. Generally, membrane fouling is scarcely depicted by a single characteristic of NOM and combining the characteristics of NOM to give a comprehensive analysis is the feasible approach. The fouling behavior of NOM during UF was studied through advanced organic matter characterization methods including high-pressure size exclusion chromatography (HPSEC) and three-dimensional fluorescence spectroscopy coupled with parallel factor analysis (PARAFAC). In addition, the redundancy analysis (RDA) was also adopted to study the correlation between the removal of NOM fractions and membrane fouling resistance. Moreover, the combined mechanism (CM) model was used to evaluate the fouling mechanism of NOM fouling.

Dead-end filtration was adopted in the present study. The photos of the laboratory experimental setup are shown in Figure S1 (Supplementary Material). The experiments were carried out in a homemade membrane module at a temperature of 25°C ± 1°C, controlled by a water bath thermostat (DKB1915, JINGHONG, China). The feed water was driven through the membrane by a magnetic gear pump under a constant flux. The pressure and flow data were returned to a programmable logic controller (PLC) by a pressure transmitter (MIK-P300, Asmik, China) and flow transmitter (LSF38, OVAL, Japan) respectively every 1 minute.

The membrane module (Figure S1) was composed of a single membrane fiber (0.9 mm Multibore^®, inge, GmbH, Germany) sealed in a polymethyl methacrylate tube with a mixture of polyurethane and epoxy resin. The modules had an effective membrane surface area of 39.6 cm². Before the filtration tests, pristine membrane modules were operated with Milli-Q water for 24 h to remove the organic residuals and wetting agents on the membranes and reached a constant operation pressure.

The feed water for ultrafiltration experiments was collected from Tong Xin river in Tongji university. All samples were transported to the laboratory within 24 hours, then filtrated by 0.45 μm mixed cellulose ester filter (Xinya Co. Ltd, Shanghai, China) and stored in a refrigerator at 4°. The dissolved organic carbon (DOC), ultraviolet absorbance (UV₂₅₄) and pH were measured immediately.

The experiment procedure was controlled by a programmable logic controller (SIMATIC S7-1200, Siemens, Germany). The procedures and specific parameters were listed in Table 1.

The dissolve organic concentrations (DOC) were measured using a total organic carbon (TOC) analyzer (TOV-VCPH, Shimadzu, Japan). Ultraviolet light absorption at 254 nm was measured by a HACH UV-Vis spectrophotometer (DR5000, HACH, USA) and a 1 cm quartz cuvette. The molecular weight distribution of NOM was determined using a HPSEC method with a UV detector (UVA, Waters 2489) and on-line DOC detector (TOC, Sievers 900 Turbo TOC). Fluorescence excitation emission matrix spectra (FEEM) measurements were conducted using a Hitachi 7100 fluorescence spectrometer. The excitation and emission slits were set to 5.0 nm and excitation wavelengths were incrementally increased from 200 to 450 nm with a step of 10 nm. For each excitation wavelength, the emission wavelengths were detected at 2.0 nm gap from 275 to 575 nm.

PARAFAC was used to decompose the fluorescence components in the dataset of 36 EEMs of raw NOM samples. PARAFAC analysis was conducted using the DOMfluor toolbox (Stedmon et al. 2003) in Matlab^® (version 2016, MathWorks Inc., Natick, MA).

The morphology of the membrane surface was obtained by SEM (Zeiss Sigma 300, Carl Zeiss Ltd, UK) and the surface roughness were analyzed using an atomic force microscope (AFM, FM-NanoviewOp-AFM). Fouled membranes (approximately 1 cm²) were stuck to the magnetic stainless steel substrates and the surface was imaged at a scan size of 5 μm × 5 μm. AFM images were imported to the Gwyddion software to distinguish the height of samples by color scale, and then to calculate the surface roughness (Nečas & Klapetek 2012).

The normalized resistance variation during experiment was shown in Figure 1. The normalized total fouling resistances at the end of every cycle ranged from 2.5 to 4.0. The normalized hydraulic reversible resistance (R_r/R_m) and normalized irreversible resistance (R_ir/R_m) were calculated according to Equations (2) and (3), and results were shown in Figure 2, which is similar to the fouling behavior described by Maciej Szwast et al. (Szwast et al. 2013). Of note, R_r/R_m accounted for the main part of membrane fouling resistance. During the initial phase of filtration, R_ir/R_m experienced rapid growth for the mass organic colloidal adsorption. Subsequently, R_ir/R_m showed a dramatic decrease after initial adsorption, and gradually increased to around 0.1 R_m, which could be attributed to the pore blocking.

The apparent molecular weight distribution (AMWD) in NOM samples with the same DOC concentration (5.0 ± 0.1 mg/L) are shown in Figure 3. AMWD in different NOM samples can be divided into three regions, i.e. low molecular region III (500 Da–8,000 Da), medium molecular region II (5,000 Da–20,000 Da) and high molecular region I (600,000 Da–2,000,000 Da). A similar molecular weight distribution of NOM can be found in previous studies (Kennedy et al. 2005; Yamamura et al. 2007). Their proportions are 0.73% ± 0.028%, 6.35% ± 0.626% and 92.92 ± 0.628%, separately. Of note, organics in region I show a rare UV₂₅₄ response, indicating the low aromatic structure of high molecular organics; organics in region II have the strongest UV₂₅₄ response along with the moderate DOC response, and organics in region III account for most DOC response with lower UV₂₅₄ response (Gao et al. 2021). It is thought that large molecules mainly consisted of polysaccharide or protein, whereas lower molecules mainly consisted of humic acid (Yamamura et al. 2007).

A total of 36 EEMs of NOM samples collected from Tong Xin river were used for PARAFAC analysis. According to the PARAFAC results, EEMs can be divided into three components as shown in Figure S2, which is consistent with NOMs from the previous studies (Table S1). Component 1 had an excitation peak at around 240 nm with an emission peak at around 430 nm. Furthermore, Component 1 was attributed to the humic acid component, indicating good matches with the published PARAFAC results (Chen et al. 2003; Senga et al. 2018). Component 2 had primary and secondary excitation peaks, occurring at around 200 nm–235 and 275 nm, respectively, and a single emission peak at around 300 nm. Component 2 was dominant in tryptophan-like substances, respectively, which differ in molecular weight and aromaticity. The prominent peak of Component 3 appeared at 225 nm/340 nm (E_x/E_m), linked to soluble microbial product (SMP) and aromatic protein fluorescence (Yamashita et al. 2010). Compared with Component 2, the blueshift of the EEM peak from 340 to 300 nm indicated a lower degree of condensation (Chen et al. 2003) and/or a lower molecular weight (Ishii & Boyer 2012).

The maximum fluorescence intensity (F_max) was used to track changes of each component during UF; however, comparison between different components depends on the magnitude of their quantum efficiencies and their responses to quenching effects (Baghoth et al. 2011). Figure 4 shows the removal rate of F_max of each component in every filtration cycle. The F_max is determined by selecting the point of maximum intensity for each component in every sample (Vera et al. 2017). F_max of components 1 and 3 has a clear decline as the filtration cycle increased, and the F_max of component 2 did not show an obvious tendency. This is mainly because the fluorescence of Component 2 is related to the medium and low molecular organics (shown in Figure 6), which showed little removal.

According to the AMWD of samples, the removal rate of different molecular weight (MW) fractions is shown in Figure 5. There was no significant difference between each filtration cycle in removal rates among each MW fraction besides fraction III. The removal rate of fraction III suffered a sharp decline and then slowly increased to a relatively stable value. That is mainly because of the rapid adsorption of low molecular organics in the initial phase and the retention capacity of the fouling layer in subsequent process.

The correlation between PARAFAC components and molecular weight regions is calculated by R (package stats v4.0.2) and the correlation heatmap is shown in Figure 6. The soluble microbial product and aromatic protein fluorescence (Component III) is significantly positively correlated with the high molecular region (Region I). That is, the soluble microbial product and aromatic protein in the feed water are mainly composed of high molecular organics (600,000 Da–2,000,000 Da). In addition, the humic acid component (Component 1) is significantly positively correlated with the low molecular region (Region III), which suggests that the humic acid in the feed water is mainly composed of low molecular organics (500 Da–8,000 Da).

Redundancy analysis (RDA) has been proved to be an effective multivariate statistical analysis method on the influence affecting membrane fouling (i.e., DOC, Zeta potential, metal ion concentration and so on) (Liu et al. 2019; Liu et al. 2020). Liu et al. analyzed the quantitative correlation between membrane fouling resistances and the influent characteristics by RDA during wastewater reclamation, and found that DOC, UV₂₅₄ and protein were the most important positive factor in membrane resistance, and they verified the fouling alleviation effect of coupling ferrate pretreatment and in-situ ozonation (Liu et al. 2019). In the current study, RDA combined with the results of AMWD and PARAFAC was applied to further explore the specific correlation between the removal of different NOM fractions and membrane fouling resistances. As shown in Figure 7, axes 1 and 2 represent 71.93% and 19.77% of the total variance, respectively. Clearly, the removal of Components 1, I and III has positive correlation with normalized hydraulically reversible fouling resistance (R_r/R_m) and total fouling resistance (R_t/R_m). The contribution to R_r/R_m followed the sequence of the removal rate of Component 1 > I > III > Component 2–II, while the contribution to R_ir/R_m followed the sequence of the removal rate of II > Component 2 > III > I > Component 1. This can be due to the fouling resistance by Component 1 (p = 0.05**) and I (p = 0.074*) being 47.7% and 19.3%, which indicated they were the dominant factors in membrane fouling. In combination with the PARAFAC and AMWD results, humic acid-like, SMP and aromatic protein had highly positive correlation with R_ir/R_m, R_r/R_m and R_t/R_m.

The combined mechanism model was used to analyze the fouling data. The model was described in detail in our previous study (Gao et al. 2021) and the fit result is presented in Figure 8 and Table 2. The cake-intermediate model exhibited higher R² than other blocking models. Since the terms and have similar units (m⁻¹), the contributions of individual fouling mechanisms could be evaluated by comparison of their obtained fitted parameter values (Cerón-Vivas et al. 2018). Notably, the value of the ratio was 0.03–0.04 indicating that the cake layer fouling mechanism is dominant during the filtration and the morphology of the membrane surface also shows the existence of a dense cake layer (Figure S4).

The morphologies of the UF membranes used in the experiments are shown in Figure S3. There was a considerable difference between the pristine membrane and fouled membrane and the fouling layer can be easily observed on the membrane surface.

The surface morphology of fouled membranes was further detected by AFM. The three dimensional map visually displays undulation on the surface of the sample (Figure S4) and the surface parameters, the mean roughness (Sa) and the height of the foulant layer (h) are obtained by the Gwyddion software (Table 3). The surface of the pristine membrane showed a typical peak-to-valley structure, with Sa being 5.30 ± 0.12 nm. After being fouled by NOM, the membrane surface morphology varied significantly, attributed to the deposition of foulants. The Sa of the sample was 87.6 ± 3.89, and the h of samples increased to 37.5 ± 5.78 nm.

In this study, the fouling behavior of NOM was investigated by the fluorescent component and MW distribution. The primary conclusions were summarized as follows:

• (1)

  Combined with the previous studies, PARAFAC results indicated that fluorescent NOM in feed water was mainly composed of humic acid, tryptophan-like substance, SMP and aromatic protein. The correlation analysis between PARAFAC components and molecular weight regions showed that the SMP and aromatic protein in feed water were mainly composed of high molecular organics (600,000 Da–2,000,000 Da) and the humic acid in the feed water was mainly composed of low molecular organics (500 Da–8,000 Da).

• (2)

  The RDA results indicated that humic acid-like, SMP and aromatic protein had highly positive correlation with R_ir/R_m, R_r/R_m and R_t/R_m. That is, they were the dominant factors of membrane fouling.

• (3)

  The fouling mechanism analysis indicated that the cake-intermediate model made a good description on the fouling behavior and the cake layer fouling had a larger predominance over the intermediate blocking on NOM fouling. Subsequently, the SEM and AFM images proved the existence of fouling layer.

The authors would greatly appreciate the financial support from the Major Science and Technology Program for Water Pollution Control and Treatment (No. 2017ZX07201001).

The authors have no conflicts to declare.

All relevant data are included in the paper or its Supplementary Information.