Application of hybrid coagulation–ultrafiltration for decentralized drinking water treatment: impact on flux, water quality and costs

Inadequate access to safe drinking water imperils life, subdues opportunity and subverts human dignity (Watkins 2006). Yet 2.1 billion people – 29% of the global population – lack safely managed drinking water services (WHO/UNICEF 2017). Low-income countries (LIC) are disproportionately affected by the pernicious and persistent water crisis of today, underpinned by a shortage of fresh water resources due to climate change and escalating levels of water pollution. In many LIC, drinking water treatment is mainly focused on conventional technologies such as media filtration and chlorine disinfection. However, conventional technologies are designed for use in large centralized systems. Therefore, they are highly unlikely to be installed in rural communities due to their high investment costs.

Nowadays, decentralized membrane-based water treatment has emerged as a sustainable approach to safe water supply in LIC. However, membrane fouling remains an overriding obstacle. To alleviate fouling, raw water pretreatment using techniques such as coagulation, biofiltration, adsorption, and oxidation has been widely proposed. Among these techniques, coagulation remains the most successful method for controlling membrane fouling in full-scale water treatment (Arhin et al. 2016). Aluminum sulfate is the most commonly used coagulant and can suppress the hydraulically irreversible fouling rate of hollow fiber membranes by 75–100% (Hatt et al. 2011). However, currently, there is an upsurge in the use of pre-hydrolyzed coagulants, including polyaluminum chloride (PACl). PACl contains substantial proportions of highly charged tridecamer cationic species (Al₁₃), which are effective in neutralizing negatively charged colloids (Arhin et al. 2018). Much work has been aimed at expounding the antifouling mechanisms of coagulants and at identifying coagulation conditions most effective for fouling abatement (Howe & Clark 2006). Promising results seem to be contingent on the coagulant type, dosage, mixing conditions, flow configuration, membrane properties, and feed water characteristics. Although a plethora of coagulation–ultrafiltration (UF) studies have reported favorable results with regards to contaminant elimination and flux amelioration, studies on the impact of coagulation on treatment costs are relatively rare. Yet, information in this regard is essential for the widespread adoption of such systems in LIC.

The aim of this study was, therefore, to evaluate the influence of coagulation on the performance of a pilot-scale UF unit for decentralized water treatment in terms of flux, water quality, and costs. Although the hybrid coagulation–UF system is rarely applied in LIC, we presume it could represent a potentially cost-effective approach to alleviating the drinking water crisis in LIC. This hypothesis is premised on the diminishing cost of UF membranes, the relatively low cost of coagulants and the need to deal with freshwater quality problems in LIC, including eutrophication, cyanobacteria, and disinfection by-product precursors, which make conventional treatments rather costly.

The feed water used in this study was collected from Lake Victoria – at the inlet to Ggaba II Water Treatment Plant, Kampala, Uganda. The characteristics of the feed water (Table A1) and further information on the study area are presented in the Appendix (available with the online version of this paper).

Two sets of UF experiments were used to assess the influence of PACl on flux, permeate quality and costs. In the first, commercial liquid PACl solution (Zetafloc 553 L; Abby Laboratories, South Africa) was used to pretreat the feed water prior to UF (herein denoted as system A). Coagulation pretreatment was performed using the optimum conditions observed during bench-scale tests as described previously: PACl dose of 20 mg·L⁻¹, corresponding to 1.21 mg-Al·L⁻¹ and a hydraulic retention time of 14 min (Arhin et al. 2018). The pH of the raw water was not adjusted during coagulation in order to mimic full-scale conditions at the local water treatment plant. The setup for system A, therefore, was comprised of coagulation/flocculation followed by UF without sedimentation. In the second set of experiments (system B), the feed water was treated by UF without PACl coagulation (0 mg·L⁻¹ PACl). The setup for the two systems is schematically shown in Figure A2, Appendix (available online).

The UF membranes used in this study were polysulfone (PS) hollow-fiber membranes (Frontec Environmental Co., Ltd, China) operated in an inside-out mode. A membrane module had an effective area of 4.5 m², a molecular weight cut-off (MWCO) of 100 kDa, a fiber outer diameter of 1.6 mm, and an inner diameter of 1.0 mm. Each experiment was conducted with a membrane in pristine condition. Prior to each experiment, the new membrane was soaked in clean water for 24 h and flushed copiously to remove any residual storage agent. The baseline membrane specific flux, J_o, was determined by 30 min filtration of clean water. Filtration experiments were conducted in the dead-end mode under a transmembrane pressure of 30 kPa. The filtration cycle in each system was set at 60 min, and backwashing was done after each cycle with permeate for 5 min (flux, 80 L·h⁻¹·m⁻²). The permeate flux, ⁠, was calculated from the permeate flow rate, Q (L·h⁻¹), and the membrane area, A (m²). The calculated flux was then adjusted to the equivalent flux at standard temperature (20 °C) using the expression ⁠, where J_s is the flux at standard temperature (L·h⁻¹ m⁻¹), T_s is the standard temperature (°C) and T is the measured temperature (°C) (Crittenden et al. 2012). The ratio of the equivalent flux at standard temperature, J_s, to the flux of clean water at standard temperature, J_os, was designated as the normalized specific flux, J_s/J_os, and was used for comparing the effect of each experiment. Chemical cleaning was done at the end of each experiment. The membranes were soaked in NaOCl (100 mg/L) for 3 h, followed by NaOH (0.02 N) for 3 h, then rinsed thoroughly with permeate.

The variations in the average normalized flux and UVA₂₅₄ removal rates for systems A and B are shown in Figure 1(a) and 1(b), respectively. As depicted in Figure 1(a), a two-stage flux decline mode consisting of a rapid flux decline during the initial filtration periods followed by a slow drop was observed in both systems, however, to varying extents. For system B the flux dropped rapidly to 58% on the 9th day. This rapid flux decline was followed by a gradual decrease over the next duration of the experiment. In contrast, the flux for system A declined rapidly to 82% on the 5th day and this was followed by a relatively stabilized flux in the subsequent days.

This two-stage flux decline mode exhibited by the normalized flux curves suggests potential changes in the fouling mechanisms, as per the pore blockage–cake filtration model described in previous studies (Ho & Zydney 2000). The rapid flux decline observed during the initial filtration periods suggests that pore blocking was the most dominant fouling mechanism at that stage, attributable to the adsorption of UVA₂₅₄ on/in the membrane via hydrophobic interaction. However, this was followed by cake layer formation in the subsequent days. After 25 days of operation, the flux of system A was 21% higher than system B, suggesting that PACl coagulation could ameliorate permeate flux.

Figure 1(c) presents the nature and extent of fouling in systems A and B. A higher UMFI_T value was observed in system B than in system A, suggesting that PACl was effective in reducing the total fouling rate. Furthermore, the value for UMFI_IR that indicates the residual fouling after hydraulic backwash was higher in system B than in system A, suggesting that hydraulically irreversible fouling was also suppressed by PACl coagulation. However, our results showed no distinction in the chemically irreversible fouling rate for the two systems. As shown by the matched-pair analysis in Figure 1(d), the line corresponding to y = 0 that shows whether the UMFI values for systems A and B are equal is within the mean ± 95% confidence interval (CI) range for UMFI_C, indicating that there was no statistical difference. This is probably due to the low levels of the UMFI_C values in both systems (<0.00005 m² L⁻¹), which were difficult to quantify as previously observed (Huang et al. 2012).

The quality of permeate produced from systems A and B in comparison with the local drinking water standards is presented in Table A3 (Appendix, available with the online version of this paper). As expected, UF was found to be very effective in removing pathogenic indicators (E. coli and total coliforms), with zero indicator bacteria detectable in the permeate in both systems. However, besides bacteria removals, PACl coagulation enhanced the removal of all other aquatic contaminants monitored.

Turbidity and color were two key quality parameters monitored. From an operational perspective, these two parameters are usually indicative of the performance of the treatment processes and their ability to conform to appropriate water quality standards (Ho et al. 2012). Consistent with the results of previous coagulation–UF studies (Bergamasco et al. 2011), the hybrid PACl-UF system was very effective in removing compounds responsible for turbidity and color. As shown in Figure 2(a), higher turbidity (99.2%) and color (100%) removal rates were attained in system A compared with system B (98.1% and 97.5%, respectively).

Besides turbidity and color, DOC removal also increased with PACl coagulation. However, DOC is only a measure of the dissolved NOM concentration and does not reflect the characteristics of the NOM. Thus, in a further step, the degree of conjugation and aromaticity in the treated water was determined from the SUVA calculations. As shown in Figure 2(a), SUVA removal increased with PACl coagulation. The average percent removals in systems A and B were 57.1% and 21.9%, respectively. A matched-pair analysis was used to statistically evaluate the impact of PACl on the water quality parameters monitored. As shown in Figure 2(b), the line corresponding to y = 0 is above the mean ± 95% CI range for all the quality parameters assessed, indicating that the differences between system A and B values were statistically different from zero.

As depicted in Figure 3(a), the principle of economy of scale was found applicable for the impact of PACl dosing on the total investment costs, with a break-even point at 10 m³/d plant capacity. Thus, for a small community-scale plant with a capacity <10 m³/d, PACl dosing at 1.21 mg-Al·L⁻¹ impacts negatively on the total investment costs. Yet, for larger plants, PACl dosing and higher-flux operation result in an overall reduction in the total investment costs. CAPEX, however, is computed by depreciating investment costs over a period of time and expressed as costs per unit volume of water produced. The results of this conversion are shown in Figure 3(b). It should be noted that UF membranes are packed in modules and so the investment costs increase almost linearly with plant capacity as depicted in Figure 3(c). However, the auxiliary costs associated with coagulation follow the principle of economy of scale (Guo et al. 2014), which implies that the costs increase less than linearly with scale as depicted in Figure 3(d). Consequently, as plant capacity increases, CAPEX increments associated with PACl dosing diminish beyond CAPEX reductions resulting from PACl dosing and higher-flux operation, leading to overall savings in CAPEX as shown in Figure 3(b).

It is noteworthy that OPEX analysis based on the assumptions made focuses only on fouling control and does not reflect any added benefits or costs associated with contaminant removal. Considering that PACl dosing at 1.21 mg-Al·L⁻¹ provides additional benefits, including enhanced removal of NOM, these additional benefits could offset the presumed OPEX penalty in the worst-case scenario. For instance, higher removal of DOC after coagulation would be expected to reduce chlorine demand and chlorine dose (Xie 2004), which in turn would lower OPEX. However, this justification is probably site-specific and dependent upon the specific characteristics of the source water and the overall treatment goals. For source waters with low aromaticity and no need for enhanced coagulation, increased OPEX due to coagulation cannot be offset by savings in chlorine dosing. Nevertheless, in such scenarios, effective fouling control using low coagulant doses for charge-neutralization, rather than doses for sweep-flocculation, has been shown to provide OPEX benefits (Hatt et al. 2011).

PACl coagulation ameliorated the permeate flux in the pilot-scale UF unit by significantly suppressing both the total fouling and hydraulically irreversible fouling. Additionally, PACl improved the permeate quality by significantly enhancing turbidity, color, DOC, UVA₂₅₄, and SUVA removal. In comparison with the conventional UF system, the CAPEX of the hybrid coagulation–UF system shows a strong correlation with production capacity. Hence, the hybrid system is more favorable than the conventional UF system at larger capacities. OPEX, on the other hand, appears to be highly correlated to factors such as coagulant and membrane replacement costs, which are also dependent on local variables, including inflation. However, even in the worst-case scenario, OPEX penalties resulting from coagulation could be offset by savings in chlorine dosing. Overall, the results strongly suggest that the hybrid coagulation–UF system is potentially a cost-effective model for decentralized water supply.

This work was supported in part by the Swiss Federal Institute of Aquatic Science and Technology (Eawag) through the Eawag Partnership Program for Developing Countries and the Mobility to Enhance Training of Engineering Graduates in Africa (METEGA) Project. Access to research facilities of the National Water and Sewerage Corporation (NWSC), Uganda, is acknowledged.