Coagulation/flocculation prior to low pressure membranes in drinking water treatment: a review

Microfiltration (MF) and ultrafiltration (UF) consistently remove suspended material and pathogens from drinking water; however, membrane fouling inhibits their application by increasing operation and maintenance costs. Coagulation/flocculation is a commonly used pretreatment method for the reduction of membrane fouling; in this review it has been grouped into three typical configuration types: Type 1: coagulation + no/incidental flocculation, Type 2: coagulation + flocculation, and Type 3: conventional coagulation, based on operational conditions. The impact of each configuration on floc properties, membrane fouling, and organics removal has been reviewed in detail. Due to relatively high membrane resistance and low NOM reductions, configuration Type 1 may not be optimal for fouling control and organics removal when compared to Types 2 and 3. Configuration Type 2 led to the lowest cake layer and specific cake layer resistance for both MF and UF, while there is evidence that Type 3 results in the greatest reduction in fouling rate by reducing mass flux towards the membrane surface. As expected, with no coagulant results indicate that UF achieves greater organics removal when compared to MF, but with the addition of coagulant performance is similar for all configuration types. By highlighting the connection between coagulation/flocculation configuration types and membrane performance, the review provides insight for the design and operation of pretreatment for low pressure membrane filtration. In addition, understanding the impact of configuration types on floc properties aids in revealing the fouling mechanisms that dictate membrane performance. Knowledge gaps have been identified for guidance on future research. Membrane fouling inhibits the application of low pressure membranes by increasing operation and maintenance costs. Coagulation/flocculation is a commonly used pretreatment method to reduce fouling, which may be grouped into three typical configuration types. This review provides insight for the design and operation of coagulation/flocculation by highlighting the impact of each configuration on floc properties, membrane fouling, and organics removal.


Introduction
Microfiltration (MF) and ultrafiltration (UF) membranes, commonly referred to as low pressure membranes, are widely employed due to their ability to consistently remove suspended material and pathogens from drinking water. 1 However, membrane fouling remains a challenge, causing increased transmembrane pressure (TMP), flux deterioration, and greater frequency of required backwashing and chemical cleaning. Thus fouling reduces operating efficiency and membrane life, and ultimately increases the operating and maintenance costs of membrane filtration. 2 Membrane fouling is typically characterized as reversible or irreversible based on the impacts of cleaning practices. Hydraulically reversible fouling can be addressed hydraulically (e.g. backwashing), while hydraulically irreversible fouling can be addressed chemically (e.g. chemical cleaning). Chemically irreversible fouling cannot be removed and its gradual increase contributes to membrane 'ageing', or irreversible changes to membrane performance and characteristics associated with long-term foulant and cleaning agent exposure. 3 Membrane fouling is highly impacted by natural organic matter (NOM). 4 Historically, hydrophobic humic substances (i.e. humic and fulvic acids) which constitute the majority of NOM present in surface waters 5 have been identified as the predominant NOM foulants. 6,7 However, there is increasing evidence that hydrophilic biopolymers (i.e. protein-and polysaccharide-like macromolecules) are the main contributor to membrane fouling. [8][9][10][11][12] As NOM is ubiquitous in source waters, identification of the NOM fractions responsible for membrane fouling is a primary concern for development of mitigation methods for drinking water treatment using low pressure membranes.
Coagulation/flocculation is commonly used prior to low pressure membrane filtration to reduce fouling, and has been reported to reduce pore blocking, decrease cake layer resistance, and increase backwash efficiency. 13 Previously published reviews regarding pretreatment, 1,14 as well as fouling and cleaning 15 for low pressure membranes, and an overview of coagulation/flocculation pretreatment for membrane treatment of drinking water and wastewater 16 can be found in the literature. Gao 1 summarized pretreatment methods (e.g. coagulation, adsorption, peroxidation, prefiltration) and operational conditions (e.g. running modes, rinsing modes, chemical cleaning, air scouring) for reducing fouling, while Huang 14 also reviewed pretreatment methods to address membrane fouling concluding that coagulation had been most successful. Shi 15 provided a summary of both conventional and non-conventional (e.g. electrical cleaning, ultrasonic) cleaning methods, as well as their impact on membrane materials though no results quantifying cleaning performance were included. Thus, the available review studies do not provide insight on the impact of coagulation/ flocculation prior to low pressure membranes on floc properties, membrane fouling, and NOM removal. In particular, no comprehensive review exists on the impact of coagulation/flocculation configuration types on membrane performance.
The present review classified coagulation/flocculation pretreatment configurations from all of the reviewed literature into three typical types: Type 1: coagulation + no/ incidental flocculation (i.e. coagulation with direct membrane filtration), Type 2: coagulation + flocculation (i.e. coagulation, flocculation, with direct membrane filtration), and Type 3: conventional coagulation (i.e. coagulation, flocculation, sedimentation, and membrane filtration). The impact of each configuration type on floc properties and membrane performance in terms of fouling and NOM removal was assessed. The number of studies considered in reviewing the impact of configuration type on floc properties, fouling performance, and DOC/TOC removal is summarized in Fig. 1.
Published studies that counted towards fouling performance include at least one of total resistance, cake layer/specific cake layer resistance, specific hydraulic resistance, mean rate of TMP increase/flux decline, and flux recovery after hydraulic and chemical cleaning. This review highlights the connection between coagulation/flocculation configuration types and membrane performance, providing insight for the design and operation of pretreatment for low pressure membrane filtration. Summarized results are used to understand the impact of configuration types on floc properties, which dictate fouling mechanisms and membrane performance. Knowledge gaps have also been identified to provide guidance for future research.

Overview of coagulation/flocculation theory
Coagulation and flocculation typically occur sequentially, governing the formation and properties of floc, and impacting the performance of downstream membranes. Initial aggregation of particles/NOM occurs by destabilization during coagulation, where mechanisms include charge neutralization, interparticle bridging, and sweep flocculation. 17 As particles in natural waters, including NOM, are negatively charged (pH 6.0 to 8.0), charge neutralization occurs when sufficient cationic metal hydroxides are adsorbed to reduce zeta potential to zero. 18,19 Sweep flocculation occurs when the concentration of a coagulant exceeds its solubility limit, precipitates, and enmeshes particles/NOM. 20 The preferred destabilization mechanism for membrane pretreatment is dependent on water quality, and may be different for each configuration type.
Following destabilization, particle-particle interactions (i.e. collisions) result in floc formation. 17,19 Fluid shear induced by mixing is the dominant flocculation mechanism when two colliding particles are >1 μm in diameter, while Brownian motion dominates when at least one particle is small (i.e. <1 μm in diameter) and differential sedimentation in all other cases. 21 Aggregation rates are highest when particles have been fully destabilized by coagulation, and are lower in the case of partial destabilization. The hydrodynamic conditions of coagulation/flocculation are typically described by mean velocity gradient, G _ , and contact time, t, and will be different for each configuration type. In some of the published studies impeller speed (rpm) rather than G _ value is used to describe hydrodynamic conditions; however, despite the limitations of G _ value, impeller speed is inadequate for quantifying the forces being applied to water during mixing, and hinders the reproducibility of study results. Finally, mass flux of floc towards the membrane surface results in fouling, where floc properties may be associated with the type of membrane fouling that predominates.

Typical configuration types
Coagulation/flocculation pretreatment for all of the reviewed literature could be classified into three typical configuration types: Type 1: coagulation + no/incidental flocculation; Type 2: coagulation + flocculation; and Type 3: conventional coagulation (Fig. 2). In this review, coagulation without flocculation/sedimentation is considered as coagulation + no/ incidental flocculation. Coagulation followed by flocculation in the absence of sedimentation is commonly referred to as direct filtration, 22,23 while here it is characterized as coagulation + flocculation. Conventional coagulation, which includes sedimentation, has been applied prior to membrane filtration to remove aquatic constituents that cause fouling. 14 However, it has been observed that this may not be effective in removing the NOM fractions which contribute to irreversible fouling, 24 and that similar fouling performance may be realized without sedimentation. 25 In coagulation pretreatment without sedimentation, floc size is only required to grow beyond that of membrane pores (i.e. submicron), thus reduced coagulant dosages may be applicable. 26 Coagulation pretreatment without flocculation/ sedimentation has been investigated for its potential to significantly reduce flocculation times and water treatment plant footprint. 27 A summary of coagulation/flocculation conditions for configuration Types 1, 2, and 3 from the literature is provided in Table 1. 2.2.1. Type 1: coagulation + no/incidental flocculation. For configuration Type 1, a coagulant is typically applied on a continuous basis followed by rapid or static mixing directly upstream of membrane filtration. This does not include a flocculation step, and contact time is typically <1 min, 9,23,27 though it may be slightly longer. 31 Two studies did not apply mixing after the addition of coagulant. 9,29 Incidental flocculation could result from the use of additional ancillary equipment, such as contact tanks 30 and recirculation lines, 32 or by extended contact time such as that which may be experienced in a membrane cell or tank. 33,41,43 No standardized method for the optimization of coagulant dosage for configuration Type 1 has been presented in the literature. Coagulant dosages that have been considered range from 0.59 to 5.68 mg L −1 Al and 0.34 to 10.00 mg L −1 Fe. Several studies have examined the impact of coagulant dose on the basis of turbidity and NOM removal by charge neutralization and sweep flocculation. 22,29,30,32 Coagulant dosage has also been optimized for reducing membrane fouling. Pronk 9 reported increased fouling at an FeCl 3 dosage of 5 ppm (1.70 mg L −1 Fe) compared to 1 and 2 ppm (0.34 and 0.68 mg L −1 Fe), while Judd and Hillis 27 observed that at coagulant dosages <0.035 mM Fe 3+ (1.96 mg L −1 Fe) fouling increased when compared to no coagulant addition. These results suggest there may be dose thresholds, above or below which fouling is exacerbated by the application of coagulant. Choi and Dempsey 32 examined the effect of alum and aluminum chlorohydrate (ACH) dosages ranging from 0.59 to 2.93 mg L −1 Al on membrane fouling. The authors suggested that low dose conditions (1.17 mg L −1 Al, pH 4.81), below those required for charge neutralization, could simultaneously reduce membrane fouling and coagulant costs. Konieczny 30 observed that FeCl 3 and Al 2 (SO 4 ) 3 doses which were 20% lower than those determined by jar testing (3.0 mg L −1 Fe and 3.6 mg L −1 Al, respectively) resulted in the lowest drop in permeate flux.
Where mechanical mixing or inline mixing were applied, mean velocity gradient (G _ ) and contact time (t) after coagulation for Type 1 have been reported to range from 0 to 5.25 s −1 , and 360 to 720 s, respectively. Studies that did not  include mixing did not report a G _ value but contact time ranged from 20 to 120 s. Hydrodynamic conditions for coagulation generally are not well described in studies where static mixing is used. 27 For incidental flocculation conditions, it is difficult to characterize mixing in terms of G _ and t, and where these values are provided justification for their selection is not always stated directly.
Coagulant dosages that have been considered range from 6.75 × 10 −4 to 39.96 mg L −1 Al and 22.00 mg L −1 Fe. As for Type 1, numerous studies considered coagulant dosages optimized for turbidity and NOM removal, 24,40,41,43,44 while others investigated a range of coagulant dosages 42 including those below what is considered optimum for NOM removal by sedimentation. Yao 33 observed that while a low coagulant dose led to higher total membrane resistance, when compared to an optimum dosage for humic acid and turbidity removal, irreversible fouling was lower. Wang 42   For configuration Type 2, G _ and t for coagulation have been reported to range from 100 to 184 s −1 , and 60 to 180 s, respectively. Flocculation G _ and t ranged from 14.85 to 80 s −1 , and 600 to 1800 s, respectively. Coagulation conditions have frequently been simulated using a jar test, where high mean velocity gradients (G _ ranging from 100 to 150 s −1 ) were applied for a short duration (t ranging from 60 to 180 s). 22,23,28,41,42 Compared to configuration Type 1, G _ values during flocculation are approximately 3 to 15 times greater and contact times generally longer. Howe and Clark 23 incorporated a shorter contact time (240 s) to promote the development of pin-floc, which are floc with relatively small size when compared to those typically formed to promote sedimentation.
2.2.3. Type 3: conventional coagulation. In addition to flocculation, configuration Type 3 also includes sedimentation, where conditions are optimized for floc growth and the reduction of turbidity or NOM during settling. Coagulant dosages that have been considered range from 0.04 to 9.23 mg L −1 Al. Several studies selected coagulant dosage using jar tests, with optimum alum dosages for turbidity and NOM removal ranging from 10 to 15 mg L −1 (0.92 to 1.22 mg L −1 Al) 23 and 30 to 70 mg L −1 (2.76 to 6.44 mg L −1 Al), 22,23 respectively. Optimization of coagulant dose for biopolymer removal has also been considered. 8,50,51 Carroll 24 and Kabsch-Korbutowicz 46 used coagulant dosages optimized for NOM removal of 3.20 and 3.59 mg L −1 Al, respectively. Howe and Clark 23 examined membrane fouling following Type 3 for five different source waters with alum doses ranging from 0 to 50 mg L −1 (0 to 4.50 mg L −1 Al) representing low dose, optimum turbidity removal, and enhanced coagulation conditions. Generally, low dose conditions exacerbated fouling, while enhanced coagulation resulted in reductions. At an optimum dosage for turbidity removal, reductions in membrane fouling were observed for two of the source waters, no improvement for one, and increased fouling for the remaining two. The latter three cases coincided with poor NOM removal (<10%) since coagulant dosages were not optimized for this purpose, which was suggested to have contributed to increased fouling. This emphasizes the importance of NOM as a membrane foulant, and also the importance of differences in water quality when considering coagulant dosages for membrane pretreatment.
For configuration Type 3, G _ and t after coagulation have been reported to range from 5.25 to 60 s −1 , and 360 to 1740 s, respectively. Mixing intensities for coagulation were not reported in terms of G _ , but impeller speeds ranged from 100 to 300 rpm, and t from 60 to 180 s. While the maximum flocculation time for Type 2 was 1800 s (30 min) that for Type 3 is slightly lower, but still expected to be relatively long to enhance floc development for subsequent settling. Settling times prior to membrane filtration ranged from 900 to 3600 s (15 to 60 min). 23,46,47 For Type 3, it was also demonstrated that tapered flocculation, or gradual reductions in mixing speed (65,40, and 25 rpm for 17 min each), could promote the formation of larger floc size and greater removal of turbidity prior to membranes. 23

Operational variations
While the conditions for each configuration described in section 2.2 provide an overview of the main alternatives, there are several operational variations that may be applied in each case. Yu 52 examined the addition of polyacrylamide (PAM) as a coagulant aid to reduce UF fouling, while Xu 53 investigated the use of titanium sulfate (Ti(SO 4 ) 2 ) and UF for treatment of waters containing algal organic matter. The application of natural coagulants in drinking water treatment, such as chitosan, 54,55 seed extract, 56 and starches, 57 has been investigated in order to address the sustainability of coagulation/flocculation. Encouraging results were obtained with respect to turbidity removal, as  well as in reducing membrane fouling when applied in combination with Al-based coagulants. It was suggested that pre-coating metal hydroxides prior to a permeation cycle may reduce fouling and enhance the removal of organic matter by forming a permeable, easily removable fouling layer. 29, 58 Pronk 9 suggested establishing a protective fouling layer would result in longer permeation cycle times, and examined the application of phased coagulation as an alternative. By applying 1 to 2 ppm FeCl 3 during the first 30 min of a 60 min permeation cycle, the authors observed similar total membrane fouling when compared to continuous coagulant addition. Phased coagulation would have the added benefit of significantly reducing coagulant costs.
3. Impact of configuration types on floc properties

Summary of results
Coagulation/flocculation, including coagulation conditions (i.e. coagulant dose, coagulant type, pH) and hydrodynamic conditions (i.e. G _ value, contact time) can impact floc properties such as floc size, growth rate, surface charge, structure, and strength 25,59 (Table 2). As the coagulation/ flocculation conditions of each configuration type differ, it is anticipated that floc properties will also vary. Potential relationships between coagulation/flocculation conditions and floc properties for Types 2 and 3 have been highlighted for results compiled from several studies (Table 8 and Fig. 7 and 8). Due to lack of data, similar figures were not included for Type 1.
Floc structure is commonly characterized by fractal dimension, D f , which may range from 1 to 3. 60 A D f value of one represents a linear aggregate having a mass proportional to length, whereas a D f value of three represents a uniform aggregate with mass approximately equal to the size cubed. 61 It has been suggested that as the value of D f increases the number of particle-particle bonds within the floc also increases along with strength. 62 However, while floc formed under sweep flocculation conditions are compact, they are less dense and may be more susceptible to compression under pressure. 44

Type 1: coagulation + no/incidental flocculation
Studies that have considered configuration Type 1 have investigated the impact of coagulant dose, type, and hydrodynamic conditions on floc size, surface charge, and D f ( Table 2). In general, floc sizes reported were smallest (∼1 to 8 μm) when compared to other configuration types. The size range reported by Cho 28 (49 ± 5 to 63 ± 5 μm) is an exception since contact time (20 min to 8 h) was extended in an effort to determine the effects of hydrodynamic conditions on floc properties. After 20 s of rapid mixing (230 rpm) and applying low (1.12 to 1.67 mg L −1 Fe) and high (4.02 mg L −1 Fe) coagulant doses, Judd and Hillis 27 reported particle aggregation into 2 to 5 μm floc of <10% and 65%, respectively. Similar observations were made by Cho 28 following 3 min of rapid mixing where zeta potential was near zero, suggesting that charge properties develop early in the coagulation process.
For Type 1 hydrodynamic conditions, Amjad 22 reported that floc formed at low G _ t (1890) were at least an order of magnitude smaller, and D f lower when compared to those at higher G _ t values (17 820 and 50 400). Cho 28 observed that D f decreased over time (2.30 ± 0.02 to 1.92 ± 0.01) as floc structure became less dense. While this was with rapid mixing only, the contact time was relatively long (3 min to 8 h) and it is unknown if similar results would be observed at shorter times. As it has been suggested that floc strength is related to D f , and since D f increases with increasing particle collisions, 62 it is expected that floc formed by configuration Type 1 will have relatively low D f values and may also have low strength.

Type 2: coagulation + flocculation
Studies that have considered configuration Type 2 examined the effect of coagulant dose, type, pH, and hydrodynamic conditions on floc size, growth rate, D f , and strength (Table 2) (Fig. 7). The reported floc sizes (10 to 740 μm) generally increase with increasing coagulant dose. Coagulant doses below those required for charge neutralization may lead to insufficient hydrolytic coagulant species and poor floc formation. 33,42 Chakraborti 37 suggested that at a low alum dose (3.33 mg L −1 ) that there was insufficient coagulant present to cause significant floc growth within the first 10 min (600 s) of flocculation. It has also been reported that at coagulant doses below those optimized for NOM removal by sweep flocculation, D f values are also lower. 33,42 With respect to coagulant type, Feng 41 observed that monomeric AlCl 3 resulted in larger floc size at pH 6.0 than polymeric coagulants. Similarly, at pH 6.0 and optimum dosages for humic acid removal (7 to 13 mg L −1 ), Wang 64 reported that floc size was largest for aluminum chloride (AlCl 3 ) followed by polyaluminum chloride (PAC-1) and purified polyaluminum chloride (PAC-2), and that the corresponding growth rates were 0.59, 0.50, and 0.64 μm s −1 , respectively. The authors identified a lag time in floc formation where minimal growth occurred over the first 70 to 140 s of flocculation, followed by a growth period from 420 to 665 s. This lag is longer than that reported by Judd and Hillis, 27 where floc growth was observed within 20 s using an Fe-based coagulant. Wang 42 suggested that amorphous aluminum hydroxide (Al(OH) 3 ) produced by alum resulted in the formation of porous floc (i.e. low D f ), while aluminum polymers formed clusters and chains of small spheres with higher D f . At pH 6.0, Feng 41 reported that AlCl 3 produced floc with higher D f than PAC-1 and PAC-2; however, the variance in D f values was <5%. Dong 43 also reported that monomeric coagulant species form the highest density floc around neutral pH.
The predominant coagulation destabilization mechanism at various pH levels affects floc size, growth rate, and D f .   Results indicate that at acidic pH (4.0) larger steady-state floc size can generally be achieved when compared to more neutral or alkaline pH (6.0 to 8.0), while growth rates followed an opposite trend. 41,43 Low growth rates at acidic pH (4.0) were attributed to the predominance of charge neutralization by monomeric coagulant species, while at pH 6.0 and 8.0 monomeric coagulant species were rapidly transformed into polymers and solid precipitates. It was suggested that larger floc formed at lower pH because of charge neutralization and complexation, which involves stronger forces than sweep flocculation. Compared to more neutral and alkaline pH (>6.0), observations indicate that D f is typically lower at acidic pH (4.0). 36,41,43 The formation of more compact floc at higher pH was attributed to sweep flocculation. Yu 36 applied breakage tests as an indicator of floc strength, where floc were exposed to a high G _ value (184 s −1 ) and changes in size measured. It was observed that floc with low D f produced smaller particle sizes (∼38 to 58 μm, 41 to 60 μm) than floc with higher D f (∼76 to 102 μm, 130 to 133 μm), suggesting that D f has a positive correlation with strength. Feng 41 and Dong 43 also reported rapid breakage for floc formed at pH 4.0 when compared to those formed at alkaline pH suggesting lower strength. Results indicate that sweep flocculation produces floc with higher D f than charge neutralization, and that the D f of floc formed by monomeric coagulants is higher at acidic pH and lower at alkaline pH when compared to polymeric coagulants.
Previous studies have examined the impact of G _ value and contact time on floc size and D f . A positive correlation between contact time and floc size has been reported in multiple studies. 28,37 At a G _ value of 45 s −1 , Cho 28 observed that floc size increased (90 ± 9 to 447 ± 23 μm) as contact time increased from 20 min to 8 h, while D f decreased (2.29 ± 0.04 to 1.89 ± 0.01). The observation of a decrease in D f with contact time is unexpected, and while an explanation was not provided it may be related to the dramatic increase in floc size. Lower G _ values are expected to result in larger floc size, 35,64 while higher G _ values result in greater D f than floc formed by Type 1. 22,35,37 Floc formed at higher G _ values are expected to be more compact due to increased particle collisions, floc breakup, and restructuring. As floc grow, cluster-cluster interactions become more important and smaller compact clusters have the chance to penetrate the pores of larger flocs.

Type 3: conventional coagulation
Only one study has considered the impact of configuration Type 3 on floc properties, 22 investigating the effect of hydrodynamic conditions on floc size and D f ( Table 2). After flocculation, floc sizes (∼1 to 303 μm) are of similar magnitude as those of Type 2 conditions; however, neither floc size nor D f were reported after sedimentation. Therefore, the reported floc sizes are expected to be larger than those that would have been present during subsequent membrane filtration, as larger floc would have been removed by settling. Amjad 22 reported that floc with higher D f values were not preferentially removed by settling, and that the D f values of floc that were settled and those that   were not settled were not statistically different. While high D f values may indicate more compact floc as mentioned in section 3, it is not necessarily an indication of higher density or floc that are more readily settled.

Impact of configuration types on membrane fouling
4.1. Overview of membrane fouling theory As discussed in section 3, coagulation/flocculation configuration types dictate floc characteristics, which combined with membrane operating conditions, including flux, transmembrane pressure (TMP), and dead-end vs. crossflow modes, influence membrane fouling. 33,65,66 Membrane fouling depends on the mass flux of floc to the membrane surface as well as particle adsorption on the membrane surface and in membrane pores. Particlemembrane interactions initially dictate cake layer formation, followed by particle-particle interactions that may become more prominent once a cake layer has formed. Total membrane resistance, R (t) , which increases with time, can be quantified using the relationships in eqn (1) and (2): 17 where R m is the intrinsic membrane resistance to flow, and R c(t) is cake layer resistance at time t, ΔP is the TMP, μ is the dynamic viscosity, J is the membrane flux, R rev(t) is the hydraulically reversible fouling resistance at time t, and R irr(t) is the hydraulically irreversible fouling resistance at time t. R c(t) is determined by eqn (3), with the empirical relationship for specific cake layer resistance, α, described by Carmen's equation (eqn (4)): 67 where M is the mass flux of foulant deposited onto the membrane surface, C b is the bulk concentration of particles, ε is the cake porosity, ρ p is the particle density, and d p is the particle diameter. According to eqn (3), R c(t) is the resistance due to mass of foulants accumulated on the membrane surface per unit area at time t, while eqn (4) indicates that α is the resistance per unit mass of foulant. Membrane fouling may also be quantified as specific hydraulic resistance, R′ c , which is the resistance per unit depth of foulant as expressed in eqn (5). 68 where ΔP (t) is the TMP at time t, Ø s is the volume fraction of solids in the feedwater, and Ø c is the volume fraction deposited in the cake layer. Eqn (5) was applied by Judd and Hillis 27 to quantify membrane fouling for a constant fluxvariable pressure system, while Pikkarainen 38 applied it to a constant pressure-variable flux system.

Summary of results
This section discusses the impact of coagulation/flocculation configuration types on membrane fouling in terms of resistance (i.e. resistance to flow), fouling rate, as well as hydraulically and chemically reversible/irreversible fouling. For comparison with configuration types, results for membrane filtration with no coagulant addition have also been included. Published results on the impact of coagulation/flocculation configuration types on membrane resistance (i.e. total resistance, cake layer resistance, specific cake layer resistance, specific hydraulic resistance) for MF and UF are summarized in Tables 3 and 4 Resistance. Several authors reported a decrease in both total membrane resistance and specific hydraulic resistance with increasing coagulant dose, whereby thresholds were observed above or below which resistance increased 9,27,32 (Tables 3 and 4) (Fig. 9). Judd and Hillis 27 observed that with increasing coagulant dose the percentage of particles in the size range of 2 to 5 μm increased, which would be expected to reduce internal fouling and specific hydraulic resistance. At dosages <0.035 mM Fe (<1.95 mg L −1 Fe) formation of floc with diameter less than the membrane pore size resulted in higher specific hydraulic resistance than with no coagulant as well as increased internal fouling. Similar observations were reported by Choi and Dempsey 32 for total membrane resistance at a low coagulant dosage (0.59 mg L −1 Al) and alkaline pH (7.47 and 8.73). Applying increasing doses of FeCl 3 , Pronk 9 reported approximately equal total membrane resistances (∼7.4 × 10 11 m −1 ) during the first permeation cycle, then increasing resistances over the next three cycles due to the formation of a thicker, less permeable cake layer. The same authors observed similar membrane resistance for continuous vs. phased coagulation, where 2 ppm (0.68 mg L −1 Fe)   This journal is © The Royal Society of Chemistry 2020     This journal is © The Royal Society of Chemistry 2020  coagulant was added for only the initial 50% of the duration of the first permeation cycle, followed by 1 ppm (0.34 mg L −1 Fe) for the initial 50% of the duration of subsequent cycles. The ability to achieve similar or lower total membrane resistance with phased coagulation has the potential to significantly reduce coagulant requirements. There are no apparent trends in the results for total membrane resistance (Fig. 3a); however, both Pronk 9 and Choi and Dempsey 32 reported lower fouling for configuration Type 1 when compared to no coagulant addition. The limited results presented in Fig. 3b suggest that cake layer resistance for UF may be lower for configuration Type 1 when compared to no coagulant and Type 3, but higher when compared to Type 2. It was suggested that greater specific cake layer resistance for Type 1 could be due to the formation of smaller floc, which may form a less permeable cake layer when compared to the larger floc formed by configuration Type 2 (ref. 22 and 29) (Fig. 3c). As noted by Amjad, 22 specific cake layer resistance for Type 1 is likely to be lower when compared to that of Type 3 because of smaller floc that remain following settling (Fig. 3c). Specific cake layer resistance has been reported to decrease as rapid mixing time increased (20 min to 8 h) due to a decrease in the fractal dimension of floc. 28 Lower fractal values may not reduce resistance where compression occurs, although this may be negligible when operating under low or moderate pressures (e.g. ≤40 kPa). Amjad 22 observed that while cake layer thickness and resistance increased over the duration of a permeation cycle, porosity also increased.

Fouling rate.
Results from multiple studies indicate that the duration of membrane permeation cycles (i.e. the time between required hydraulic and chemical cleaning) can be increased using coagulation/flocculation pretreatment by producing floc that is larger than the membrane pores 27,29 (Tables 5 and 6) ( Fig. 4a and b). Low coagulant dosages may either exacerbate or reduce fouling rates (i.e. rate of TMP increase), impacting permeation cycle duration. Where partial charge neutralization occurs, incomplete aggregation may result in floc that are smaller than membrane pores thus increasing internal pore blocking. 27 However, where floc are sufficiently small, and foulant-membrane interactions limited, particles may also pass through the membrane filter, which would reduce fouling. Amjad 22 reported lower flux decline (∼25.0% L −1 ) over a permeation cycle for flocculation with low G _ t when compared to flux declines for medium  This journal is © The Royal Society of Chemistry 2020 (∼36.7% L −1 ) and high G _ t (∼36.7% L −1 , ∼30.0% L −1 ) ( Table 6). Low flux decline with low G _ t coincided with the smallest floc size and lowest solids removal (60.3%).
4.3.3. Reversible/irreversible fouling. Membrane flux can be recovered to varying degrees by hydraulic and chemical cleaning. For configuration Type 1, hydraulically and chemically reversible fouling ranged from 9 to 99% and 0 to 100%, respectively ( Table 7). The results in Fig. 5a and b indicate that for UF the mean value of hydraulically reversible fouling increased and that the mean value of chemically irreversible fouling decreased for Type 1 when compared to no coagulant. Regarding hydraulically reversible/irreversible fouling for UF, the mean values for Type 1 are similar to those reported for Type 3. The higher percentages of hydraulically reversible fouling reported by Judd and Hillis 27 (88 to 96%) when compared to those reported by Kimura et al. 70 (9 to 24%) may  be partially due to greater backwash frequency and intensity. Backwashes were conducted once every 10 min at 200 LMH in the former case, and once every 30 min at 94 LMH for the latter, which suggests that greater backwash frequency and intensity could increase the degree of hydraulically reversible fouling. Several studies investigated the impact of coagulant dose on reversible/irreversible fouling. Applying a range of coagulant dosages at various pH levels, Choi and Dempsey 32 observed comparatively high hydraulically and chemically reversible fouling for a low coagulant dose (0.59 mg L −1 ) (99% and 97%). This may have been due to partial charge neutralization, and formation of floc with a slightly negative charge that are more readily removed from the negatively charged membrane surface. Depending on coagulation  Resistance. Typically, both cake layer and specific cake layer resistance for configuration Type 2 are expected to be lower than those for Type 1 due to the formation of larger floc 28,29,63 (Tables 3 and 4) (Fig. 3b and c). However, the impact of intra-and inter-aggregate porosities has been reported to be a contributing factor to cake layer resistance as well. 22,63 Amjad 22 reported similar resistance for configuration Types 1 and 2 despite the formation of larger floc, which would be expected to result in greater inter-aggregate porosity. This could be due to floc formed by Type 1 conditions being associated with lower D f , which would result in higher intraparticle porosity despite smaller size. In addition, the authors suggested that while Type 2 results in the formation of larger floc with higher D f when compared to Type 1, cake layer restructuring caused the specific cake layer resistance for Type 2 to decrease over time, eventually converging with that of Type 1. Wang 42 reported that smaller floc formed using monomeric alum coagulant resulted in the formation of a more porous cake layer than that of polynuclear PACl and aluminum chloro-hydrate (ACH), which may form gels and a tight cake layer structure. This could be attributed to the alum flocs having lower D f , and greater intra-aggregate porosity. Where D f values are similar, larger floc would still be anticipated to result in lower cake layer resistance. 28 For cake layers formed by particles with diameters of 400 and 40 μm, at a pressure of 40 kPa the reduction of inter-and intra-particle porosities were 67% and 88%, and 14%, 29%, respectively. 63 The significant reduction in inter-particle porosity indicates that the inter-particle porosity of particles has a greater effect on the overall cake layer porosity, which may be anticipated since inter-particle voids are expected to be larger when compared to intra-particle voids. Lee 72 suggested that the effect of pressure on intra-particle permeability is significant for small floc (e.g. 8.1 to 17 μm) while it has little influence on large floc (e.g. 40.6 to 48.5 μm).
Operating MF membranes in dead-end mode, Lee 44 reported that relative specific cake layer resistance (α coagulation /α raw ) was lower when charge neutralization was the destabilization mechanism (<0.7) when compared to sweep flocculation (>1.0). Floc size distributions were similar for both mechanisms, thus it was suggested that the difference was due to floc compressibility, where floc formed under sweep flocculation were three times more compressible than floc formed by charge neutralization. That floc formed by sweep flocculation are more compressible appears counter-intuitive given that several studies have  This journal is © The Royal Society of Chemistry 2020   reported that floc formed by sweep flocculation have higher D f than those formed by charge neutralization. However, the higher compressibility of floc formed by sweep flocculation may be due to their higher water content, while at the same time they are gelated, more compact, and less porous because of being predominantly made up of aluminum hydroxide precipitates. 44 Lee 44 suggested that floc formed by charge neutralization consist of aluminum cation and inorganic/organic complexes that are less compressible. 4.4.2. Fouling rate. Similar to configuration Type 1, several studies have reported that the rate of membrane flux decline can be reduced by configuration Type 2 (ref. 24 and 33) ( Table 6). Evidence suggests that the fouling rate for Type 2 can be lower than that for Type 1, as longer permeation durations have been observed with both crossflow (1 m s −1 crossflow velocity) and dead-end modes for the same coagulation conditions 29 (Table 5). This is expected given the observation that configuration Type 2 produces larger floc, which are expected to result in a more permeable cake layer. 22,28,29 As such, it follows that as mixing time increases during both coagulation and flocculation, fouling rate decreases. 28 When compared to configuration Type 3, for medium and high G _ t followed by direct filtration Amjad 22 observed lower specific cake layer resistances for Type 2, but a greater rate of flux decline. This was attributed to a greater mass flux towards the membrane surface without settling.
Again, it was observed that a threshold coagulant dose exists below which fouling rate may increase. However, it may also be the case that low coagulant doses result in lower flux decline due to insufficient floc development and fewer particles being retained on the membrane. 42 Ma 26 observed critical doses of Al that resulted in dramatic flux reduction for water matrices containing humic acid (HA), bovine serum albumin (BSA), and a 1 : 1 mass ratio of HA/BSA, which were mainly induced by particle size. Flux decline varied slightly with pH for HA and significantly for BSA solutions, indicating that NOM type plays an important role in fouling. Dong 43 reported that at pH ranging from 7.0 to 9.0 flux declines were more dramatic than at pH ranging from 4.0 to 6.0. It was suggested that lower fouling rates at lower pH were due to either greater floc size or lower D f . In addition, the better performance of FeCl 3 and polyferric chloride with basicity of 1.0 (PFC 10 ) at pH 6.0 and 7.0 was attributed to the predominance of monomeric and polymeric species. 43 4.4.3. Reversible/irreversible fouling. Yao 33 observed lower overall fouling at a coagulant dosage optimized for turbidity and humic acid removal (0.1 mM Al), compared to fouling at a low dose (0.025 mM Al) (Table 7). However, at the low dose a greater proportion of fouling was hydraulically reversible, while hydraulically irreversible fouling was only slightly lower. Thus, it was suggested that the application of low dose coagulation be investigated further. As discussed in section 4.4.2., Amjad 22 observed lower specific cake layer resistances for configuration Type 2 than for other configurations, but greater rates of flux decline. Since porous cake layers are likely easier to remove by hydraulic backwashing, the combination with greater flux decline indicates that there may be a trade-off between greater hydraulic and chemical cleaning efficiency and higher cleaning frequency because of more rapid flux decline.

Type 3: conventional coagulation
4.5.1. Resistance. Floc that are removed by settling during Type 3 do not directly contribute to membrane fouling, instead it is the properties of the particles that remain after settling that contribute to membrane fouling. As discussed in section 3.4, there may be cases where the D f of particles remaining after settling is similar to that of the settled floc. Amjad 22 observed similar cake layer fractal dimension, D c , with (2.943 ± 0.007 to 2.987 ± 0.005) and without settling (2.931 ± 0.001 to 2.977 ± 0.003). Despite similar D c , specific cake resistance was higher when floc was removed by settling than without floc removal (Fig. 3c). The authors suggested that this was due to the remaining particles being smaller and packing into less porous cake. The same authors also observed effective cake porosity to typically increase with permeation time, a phenomenon attributed to breakup and restructuring of floc within the cake layer. [72][73][74] 4.5.2. Fouling rate. Carroll 24 reported no change in the rate of flux decline when examining configuration types with or without settling (Table 6). Comparing both configuration types, the portion of the fouling rate associated with floc could be separated from that associated with dissolved NOM. The authors suggested that dissolved NOM, which remained following settling, was the main contributor to membrane fouling. When comparing low, medium, and high G _ t conditions followed by settling, Amjad 22 reported lower rates of flux decline with settling (ranging from 17.3 to 24.0% L −1 ) than without settling (ranging from 25.0 to 36.7% L −1 ), likely due to the reduction of solids in the feedwater when measured gravimetrically. Multiple studies have reported similar flux declines when evaluating raw water and water following coagulation with an optimum coagulant dosage for turbidity removal, 23,24 suggesting NOM as being the main contributor to flux decline. At higher coagulant dosages more suitable for NOM removal (3.2 to 4.0 mg L −1 Al 3+ ), reported flux declines were much lower. 4.5.3. Reversible/irreversible fouling. Limited information exists regarding both hydraulically and chemically reversible/ irreversible fouling for configuration Type 3. However, since the particles remaining in feedwater following settling are expected to be small, it has been suggested that they will form a less porous cake layer that may be more difficult to remove during hydraulic and chemical cleaning. 22 Thus, while the rate of flux decline may be slower with Type 3, the resulting cake layer properties may have important implications for the reversibility of fouling. Additionally, as NOM has been identified as an important contributor to membrane fouling, [8][9][10][11] the potential to achieve greater removals of NOM prior to membrane treatment by enhanced coagulation and settling could reduce irreversible fouling.
Wray and Andrews 8 reported different results for the impact of coagulant dose on hydraulically reversible/irreversible   fouling for various source waters. It was reported that for Lake Ontario, the addition of 15 mg L −1 alum increased reversible fouling, and greater variability in the hydraulic reversibility of fouling was observed at this higher coagulant dose. In contrast, for Lake Simcoe and Otonabee River waters the addition of alum reduced reversible fouling at dosages of both 0.5 and 15 mg L −1 . As mentioned in section 2.2.3., this emphasizes the importance of differences in water quality when considering coagulant dosages for membrane pretreatment.
5. Performance of configuration types related to removal of organic matter

Summary of results
Removals of dissolved and total organic carbon (DOC/TOC) for coagulation/flocculation configuration types are summarized in Fig. 6 ( Table 9). For comparison with configuration types, results for membrane filtration with no coagulant addition have also been included. While less results are available for configuration Type 1 regarding floc properties, when compared to Types 2 and 3 there are more results for Type 1 regarding DOC/TOC removal. This may be because the analysis of floc properties for configuration Type 1 is more difficult due to smaller floc size.

Type 1: coagulation + no/incidental flocculation
A number of studies have investigated the impact of coagulant dosage, coagulant type, and hydrodynamic conditions on the removal of organics and particulates for configuration Type 1. Reported DOC/TOC removal ranges from 42% to 75%, with a mean of 56% for MF, and 16% to 62%, with a mean of 40% for UF (Fig. 6). The mean removal for MF is greater than that reported for no coagulant (9%), below the mean value for configuration Type 2 (61%), and above the mean for configuration Type 3 (38%). The mean removal for UF is approximately the same as that reported for no coagulant (40%), and below the mean values for both configuration Types 2 (47%) and 3 (47%). With no coagulant, results indicate that UF achieves greater DOC/TOC removal when compared to MF, but with the addition of coagulant performance is similar for all configuration types. In general, results suggest that the removal of DOC/TOC increases with increasing coagulant doses. 25,27 There is evidence that charge neutralization, sweep flocculation, and low dose conditions at pH <5.0 are favorable for the removal of DOC/TOC, while low dose conditions at pH 7.0 to 9.0 lead to poor removal. 30,32 Fe-Based coagulants may perform better than Al-based coagulants for biopolymer removal, possibly due to faster generation rates of hydrolysis products and precipitates. 75 Guigui 25 noted that at approximately neutral pH (5.5 to 7.5), there was a greater variance in DOC removal for high (16 to 42%) compared to low (27 to 33%) coagulant doses. Amjad 22 reported poor solids removal (60.3%, as measured gravimetrically) at low G _ t (1890) for configuration Type 1 when compared to higher G _ t for Types 2 and 3, likely because of incomplete coagulation and dissolved solids not being retained on the membrane surface. Similarly, Howe and Clark 23 reported turbidity reductions after MF of 57% for rapid mixing alone compared to ≥89% for configuration Type 3. Again, this may be expected due to the formation of smaller floc, resulting in lower retention on the membrane surface.   22 observed nearly 100% solids removal following UF. This solids removal was greater than the removal reported for configuration Type 1, and indicated effective aggregation of humic acid. Howe and Clark 23 observed similar turbidity reductions (57%) after rapid mixing with 4 min flocculation (65 rpm) followed by MF or UF. Despite the addition of flocculation, turbidity removal was the same as that for 30 s rapid mixing only.

Type 3: conventional coagulation
Several studies have considered DOC/TOC removal following configuration Type 3; however, the majority of results reported in the literature are for removal following sedimentation only. DOC/TOC removal after sedimentation ranged from 22% to 51% with a mean of approximately 37% (Table 9). Reported DOC/TOC removal ranges from 25.3% to 50% for MF, with a mean of 38%, while reported DOC/TOC removal ranges from 20.5% to 66.2% for UF, with a mean of 47% (Fig. 6). The mean removal for MF is greater than that reported for no coagulant addition, and lower than that reported for configuration Types 1 and 2, while the mean removal reported for UF is greater than that reported for both no coagulant addition and configuration Type 1, and approximately the same as Type 2. Using alum and PAC, Kabsch-Korbutowicz 46 observed TOC removals to increase by 24.1% and 14.8% for configuration Type 3-UF compared to conventional coagulation with settling alone. Dixon 47 used Type 3-UF to treat river water with PACl dosages optimized for UV 254 removal and enhanced coagulation, and reported similar DOC and UV 254 removals after conventional coagulation with settling and Type 3-UF for both coagulant doses. If the removal of organic matter following settling does not increase it could be attributed to dissolved organics and colloids passing through the membrane. 24 Amjad 22 observed that for hydrodynamic conditions of low (1790), medium (17 820), and high (50 400) G _ t, solids removals following settling were approximately 19%, 76%, and 82%, respectively. Subsequent removals by UF were 41.0%, 22.2%, and 15.5% totaling approximately 60% (similar to low G _ t without settling), and nearly 100% for the latter two conditions. Solids removals indicate that medium and high G _ t result in the formation of larger, more settlable floc than low G _ t that are more readily removed by UF.

Discussion and conclusion
Results indicate that configuration Type 1 (coagulation + no/ incidental flocculation) leads to the formation of small floc with low D f . There is evidence that this causes the formation of less porous cake layers with greater resistance than those formed by configuration Type 2 (coagulation + flocculation) (Fig. 3), and results in lower removal of solids and turbidity (section 5.2.). As a result, configuration Type 1 may not be optimal for fouling control and DOC/TOC removal when compared to Types 2 and 3; however, more evidence is required. When considering fouling rate, reversibility, and performance in terms of DOC/TOC removal, floc size has been identified as an important factor (sections 4.2. and 5.2.). While a wide range of coagulant doses have been examined for configuration Type 1, it may be beneficial to achieve a balance between thresholds where internal fouling is minimized and DOC/TOC removal targets are achieved, while limiting additional fouling from the accumulation of hydrolytic coagulant products on the membrane surface (section 2.2.1.). Results of several studies indicate that such a balance may be realized at low dose conditions. While characteristically short contact times (≤2 min) appear adequate for developing floc sizes greater than those of membrane pores, the effects of hydrodynamic conditions on floc properties are not clear, perhaps because of difficulty in accurately simulating these conditions at bench-scale. In addition, more information is needed comparing the impact of NOM type on floc properties and subsequent membrane fouling. It is expected that small floc with low D f would form a cake layer with higher resistance. However, there is also evidence that floc formed at low dosages and acidic pH conditions can result in permeable cake layers with high hydraulic and chemical recoveries (section 3.3), though it is not clear if this would occur for configuration Type 1 without flocculation. Phased coagulation, whereby coagulant is applied during only a portion of the permeation cycle (e.g. This journal is © The Royal Society of Chemistry 2020 50%), is an operational variation that should be investigated further, as it has been reported that fouling and coagulant requirements may be reduced simultaneously (section 4.2.1.). Application of configuration Type 2 may provide greater control of floc properties. As for configuration Type 1, a range of coagulation doses and destabilization mechanisms have been examined, and as expected for increased particle collisions and contact time, reported floc sizes were generally larger and D f higher (section 3.3.). Floc formed at acidic pH (≤5.0) have a slower growth rate, but reach larger steady-state size, which could lead to the formation of a more porous cake layer. Results also suggest that floc formed at acidic pH have lower D f . During mixing floc with lower D f experienced greater breakage than those formed by the sweep flocculation mechanism with higher D f . The apparent correlation between D f and the degree of floc breakage indicates that a correlation also exists between D f and floc strength. However, it has been reported that floc formed by sweep flocculation are more compressible, which can reduce cake layer permeability due to hydraulic pressure during membrane operation. Future research efforts should be directed towards more clearly distinguishing shear strength and compressibility of floc. While some studies reported that larger floc resulted in lower cake layer resistance, others reported similar resistances to those observed without flocculation (section 4.3.1.). Amjad 22 explained that despite forming cake with higher intra-particle permeability, the higher D f of floc could compensate for this with lower interparticle permeability. Finally, without settling, mass flux of floc towards the membrane surface may be comparatively high. While the larger floc may form a more porous and easily removed cake layer, rapid accumulation of material could increase cake layer thickness causing rapid flux decline (sections 4.4.2. and 4.5.2.). A trade-off may exist between cleaning efficiency and frequency. There is evidence that Type 2 conditions result in greater NOM removal when compared to Type 1 (section 5.3.).
For configuration Type 3 (conventional coagulation), settling prior to membrane filtration results in lower mass accumulation on the membrane surface, thus lower fouling rate, but higher specific cake layer resistance (section 4.5.). While the summarized results for DOC/TOC removal (Fig. 6) do not suggest a clear trend when comparing configuration types, Amjad 22 reported greater solids removal for configuration Type 3 when compared to Type 1 and similar solids removal when compared to Type 2. In some cases, it was reported that the rate of flux decline was similar with and without settling, which was attributed to poor removal of NOM despite effective reduction of turbidity (section 4.5.2.). In order to reduce membrane fouling using configuration Type 3, the application of enhanced coagulation for increased NOM removal may be required. Limited information exists regarding the cleaning efficiency of membranes incorporating conventional coagulation (section 4.5.3.). While cake layers are anticipated to be thin, they may also be more compact. In addition, the accumulated foulants may include a lower concentration of hydrolytic coagulant products, which could result in greater irreversible fouling.
This review compiled the results of 36 studies on the impact of coagulation/flocculation pretreatment on floc properties and membrane performance. Despite the significant number of published studies, no clear guidance can yet be obtained to optimally design coagulation/flocculation pretreatment for membrane filtration. Table 8 Potential relationships between coagulation/hydrodynamic conditions, floc properties, and resistance for typical coagulation/flocculation configurations    This journal is © The Royal Society of Chemistry 2020   This journal is © The Royal Society of Chemistry 2020

Conflicts of interest
There are no conflicts of interest to declare.