Ultrafiltration membranes in managed aquifer recharge systems

The natural organic matter (NOM) and color content of surface waters are increasingly becoming an issue for artificial groundwater recharge plants. Water from Lake Bolmen, in southern Sweden, had in 2017 an annual average NOM and color content of 8.6 mg/L total organic carbon (TOC) and 57 mg Pt/L respectively, and values ranging from 7.8 to 9.6 mg/L TOC and 50–70 mg Pt/L. Since water from Lake Bolmen will be used at Vomb Water Works, an artificial groundwater recharge plant, the high NOM-content of Lake Bolmen must be reduced prior to aquifer recharge. From experiences of full-scale operations of chemical flocculation, lamella sedimentation and rapid sand filtration using ferric chloride, three different pre-treatment methods were proposed; conventional precipitation, stand-alone direct precipitation before ultrafiltration (UF), and conventional precipitation with ultrafiltration after lamella sedimentation. In this study, a hollow fiber membrane (MWCO of 150 kDa) was used in different configurations during a 15 months pilot trial. The results showed the possibility to reduce NOM equal to conventional precipitation when a stable net-flux of 40 and 70 L/(m· h) was used for direct precipitation before UF and conventional precipitation with UF, respectively. This paper presents these treatment methods and evaluates their viability as full-scale treatment steps. This is an Open Access article distributed under the terms of the Creative Commons Attribution Licence (CC BY 4.0), which permits copying, adaptation and redistribution, provided the original work is properly cited (http://creativecommons.org/licenses/by/4.0/). doi: 10.2166/ws.2020.082 om http://iwaponline.com/ws/article-pdf/20/4/1534/705604/ws020041534.pdf er 2021 K. Hägg (corresponding author) K. M. Persson Division of Water Resources Engineering, Faculty of Engineering LTH, Lund University, Lund 221 00, Sweden and Sweden Water Research AB, Ideon Science Park, Scheelevägen 15, Lund 223 70, Sweden E-mail: kristofer.hagg@tvrl.lth.se T. Persson O. Söderman K. M. Persson Sydvatten AB, Hyllie Stationstorg 21, Malmö 215 32, Sweden


INTRODUCTION
. Insufficient NOM removal rates, combined with growing populations and higher water consumption, have led artificial recharge plants to search for ways to pre-treat water before infiltration.
Common ways to accomplish pre-treatment in Sweden have been chemical flocculation in combination with either sedimentation or contact filtration, which are techniques also used by surface WTPs. Further, chemical flocculation synergizes well with infiltration due to the two different techniques' ability to remove different fractions of NOM (Eikebrokk et al. ). Other techniques include membrane technologies, which are growing in popularity within the water supply industry (Salehi et al. ) because of their ability to remove particles and microorganisms (Iannelli et al. ). The different types of membranes typically used come with their own set of advantages and disadvantages depending on the context. Nanofiltration (NF) membranes, which have been combined with aquifer recharge (Yangali-Quintanilla et al. ), have the ability to remove large molecular compounds, such as humic acid, as well as low molecular weight organic material (MWCO 300 Da) (Meylan et al. ). However, for the purpose of pre-treatment, this might seem excessive. Ultrafiltration membranes, with higher molecular weight cut-off, combined with coagulation could be a better alternative.
It has been shown that NF and coagulation combined with UF can reduce NOM in similar ways (Lidén & Persson ), although at different costs (Lidén & Persson ). In both studies, aluminum-based coagulant was used, and this coagulant has also been used successfully in several other studies (Choi & Dempsey ; Kabsch-Korbutowicz ; Keucken et al. ; Ding et al. ). Iron-based coagulants, such as FeCl 3 , in contrast are less commonly used in chemical flocculation techniques. One reason is that iron-assisted coagulation requires lower pH (around pH 5) and has a lower optimum pH range than aluminum-based coagulants (Gillberg et al. ). However, iron-assisted coagulation is successfully used at Ringsjö Water Works, a surface WTP in Southern Sweden, where this study was conducted. Konieczny et al. () shows that, when comparing ironand aluminum-based coagulants at similar pH (around pH 7) combined with UF membranes, the aluminum-based coagulant (aluminum sulphate) was more effective at reducing TOC, but it should be observed that the pH range was outside the optimum range for iron-based coagulation in this study. Iron-based coagulation and flocculation combined with microsieving has also been studied (Hägg et al. a); however, more research is needed on the efficiency of iron-based coagulation with UF membranes, and how this compares to iron-based coagulation, flocculation and sedimentation.
This study investigates FeCl 3 assisted coagulation combined with UF membranes in two different configurations, (1) combining direct precipitation with UF and (2) UF of full-scale pre-treated water (coagulation, flocculation and lamella sedimentation), and compares these results with (3) the full-scale treatment with iron-based coagulation, flocculation, lamella sedimentation and rapid sand filtration presently used at Ringsjö Water Works. At Ringsjö, the effort is to use the WTP's raw water source; that is, Lake Bolmen, as a new water source for Vomb Water Works, an artificial groundwater recharge plant 35 km southeast of Ringsjö WTP. Motivated by previous studies, indicating limitations in the infiltration fields' ability to produce groundwater with high quality from the NOM-rich waters of Lake Bolmen (Hägg et al. b), the water works are investigating ways to pre-treat the water before infiltration.
Thus, the objective of this study was to evaluate the three methods and assess the possibility to utilize the different techniques as pre-treatment steps in artificial recharge.

MATERIALS AND METHODS
This study investigated two possible ways of implementing UF membranes in full-scale drinking water production through a two-part membrane pilot trial and comparing the results with conventional precipitation with rapid sand filters (RSFs). The first part of the tests was direct precipitation before UF and the second part was full-scale conventional precipitation combined with UF. The feasibility of these three methods was compared based on NOM removal, scalability, cost, and operational flexibility and security. For all three methods, the water quality and NOM removal efficiencies were determined by measuring UVA 254 nm (m À1 ) using online sensors in the feed, filtrated and permeated water. To achieve successful flocculation pH, NaOH dosages during flocculation were controlled by online pH measurements. The capacity of the two treatment alternatives was compared with conventional precipitation using the actual net-flux and the temperature compensated permeability achieved during the pilot trials. Figure 1 shows the three different methods combined with artificial groundwater recharge.

Ringsjö Water Works
This study was conducted at Ringsjö Water Works, a surface WTP in southern Sweden operated by South Sweden Water Supply AB (Sydvatten AB). The current treatment process at the WTP includes coagulation with ferric chloride (FeCl 3 ), flocculation, lamella sedimentation, rapid and slow sand filtration, disinfection through UV-light and chlorination ( Figure S1, Supplementary material).
The raw water is taken from Lake Bolmen and is transported for one week through an 80 km long tunnel and a 25 km long pipeline before it reaches the WTP (Sydvatten AB ). The quality of the water varies seasonally and daily depending on previous conditions at the water source. Lake Bolmen is an oligotrophic lake with a catchment area dominated by forest, bogland and iron-rich soil which causes the water to have high NOM and color content (Persson ; Eikebrokk et al. ; SMHI ).

Pilot plant
To investigate the possibility to pre-treat water from Lake Bolmen, a membrane pilot study was conducted from Two different membrane pilot configurations were conducted during the trial period. The first part was direct precipitation before UF using two different coagulation configurations, inline and feed tank dosage. In the second part, pretreated water (after lamella sedimentation and before rapid sand filtration) from full-scale operations was combined with UF.
The cleaning of the membrane was conducted in three different ways, all using collected permeate from the permeate tank. Table 1 shows the hydraulic clean (backwash) and the two different chemical cleaning procedures used during the pilot trials.   hours. The first CIP was performed with citric acid but, because the permeability remained the same before and after, the method was abandoned and a combination of oxalic acid and ascorbic acid was used successfully instead.

Direct precipitation before UF
From 1 May to 15 September 2017, direct precipitation trials were conducted. Figure S2 shows the flow chart used during the pilot study. The two different coagulant dosage configurations used were: (1) inline coagulation and (2) feed tank coagulation.
The first part of the trial was inline coagulation, where FeCl 3 was added after the feed tank and before the feed pump to create adequate mixing. The 90 seconds dosage contact time was achieved by connecting a tube allowing for sufficient contact time for floc formation, and during the second part, FeCl 3 was added in a feed tank equipped with a stirrer, allowing for a 14 minute flocculation time.
To achieve the same NOM removal as conventional precipitation, the flux and coagulant dosage was 50-60 L/(m 2 · h) and 5-6.5 ppm (mg Fe 3þ /kg water) respectively. The corresponding recovery rate for both configurations was about 88%. The coagulant dosage was determined based on the resulting UVA and the NaOH dosage in the feed tank was determined based on the measured pH after flocculation.

Conventional precipitation with UF
From September 2017 to August 2018, water supply from full-scale operations after lamella sedimentation was connected to the pilot, as seen in Figure S3. In this pilot configuration, the UF membrane replaces the rapid sand filtration function of separating residual flocs from the water. The flux was varied between 60 and 90 L/(m 2 · h) with a recovery rate of about 95%.

Chemicals
The coagulant used was a 40% by weight FeCl 3 solution produced by Kemira (PIX-311, Helsingborg, Sweden), and the coagulant dosage was presented as mg Fe 3þ /L of water (ppm). A H 2 SO 4 (37%) and NaOH (25%) solution was used for pH adjustment and CEB to remove metal residues and organic matter respectively. A 12.5% sodium hypochlorite (NaClO) solution was used also for CEB. Citric, oxalic and ascorbic acid was used during CIP cleans.

Methodology
To be able to compare and determine the performance of the membranes in the different configurations, the results from full-scale precipitation at Ringsjö Water Works was used as a benchmark for water quality during the pilot study. Another important aspect is the targeted infiltration water quality when introducing pre-treatment of water before infiltration. Table 2 shows the yearly water quality averages of Lake Bolmen, the finished drinking water, the

RESULTS
In this section, the results from the pilot study and the economical evaluation are presented. As previously mentioned, the water quality, measured as UV-absorbance, from conventional precipitation was the target for the membrane pilot. The results from full-scale operation show that the water quality of the treated water remains around 5 m À1 despite fluctuations in raw water quality. However, the result depends on sufficient coagulant dosage, and when NOM content increases in the raw water, the coagulant dosage is increased. The results from full-scale operations can be seen in detail in Figure S4.

Hollow fiber UF membrane performance
This section describes the performance of the membrane in the different configurations. The first section presents the results from direct precipitation where the coagulant was applied at two different stages; inline and feed tank precipitation (see Figure S2). The second section shows the results from conventional precipitation combined with UF, where treated water from full-scale operation was used as feed water in the membrane pilot (see Figure S3). Figures 2 and 3 show the results from the first weeks of inline precipitation and feed tank precipitation, respectively.

Direct precipitation
As seen in Figure 2, the inline configuration resulted in an unstable performance due to fouling and rapid and great decreases in permeability. The initial flux used was 60 L/(m 2 · h) but was decreased to 50 L/(m 2 · h) after two days in an attempt to stabilize the permeability. After the first couple of days, CEB intervals were increased from once to twice a day. In the last part, pH was gradually increased, which resulted in a more stable performance regarding permeability. However, permeate quality, measured as UV absorption, deteriorated with increasing pH. The results from the first two weeks shows that the overall permeability decreases from around 500 to less than 200 L/(m 2 · h·bar).
During feed tank precipitation, the result showed a more stable performance in regard to permeability (Figure 3). The initial coagulant dosage was 6 ppm and was gradually increased to 7.5 ppm. As a result, the UV-absorption of the permeate decreased from around 7 to 5 m À1 . During the three-week trial, the overall permeability decreased from around 500 to 450 L/(m 2 ·h·bar).
During the last period of the trial, the capacity of the membrane was tested (Figure 4).  2 m 3 /s, this would require 180,000 m 2 membrane area.
Comparing the coagulant dosage needed for direct precipitation with conventional treatment, resulted in a potential chemical saving of up to 15% (from 6-7 ppm to 5-6 ppm).

Conventional precipitation with UF
During a two month period, pre-treated water from full scale operations was used as feed water for the membrane pilot.
The results can be seen in Figure 5. In this study, most of the flocs were already removed in the sedimentation basins which allowed for higher fluxes. During the first week, the flux was gradually increased from 50 to 80 L/(m 2 · h), which resulted in a rather stable permeability the following month.
However, there was a slight decrease in permeability on 15 October that coincided with a brief but strong turbidity increase in the feed water. The loss in permeability persisted after the event. The flux was further increased to 90 L/(m 2 ·h) during the last couple of weeks, which resulted in a permeability decrease to between 300 and 400 L/(m 2 ·h, bar). (direct precipitation) and 102,900 m 2 (conventional precipitation with UF) membrane surface area, respectively. An additional 12% membrane area would be required for direct precipitation for BW waste treatment, estimated from the observed recovery rate of 88%, which results in a total membrane area of 201,600 m 2 . This secondary membrane treatment step would not be required for conventional precipitation with UF, due to the possibility to return the back-wash water waste to the previous treatment step.
An additional cost that comes with UF facilities is replacing of the membrane modules after several years. The average lifetime of the membranes is typically 10 years if the CEB frequency is once per day and it is estimated to be 5-7 years if the CEB frequency is increased to twice per day, according to the manufacturers, after which the modules needs to be replaced.

Chemical consumption
One important aspect for water utilities and for sustainability is the usage of chemicals. Table 3 shows the amount of chemicals used to treat 63 M m 3 (2 m 3 /s) of raw water for each alternative in tons per year.
The chemical flocculation steps for all alternatives have the highest chemical demand. This includes the coagulant but also sodium hydroxide which is used to increase the pH of the raw water to ensure the right flocculation pH.
The results show the benefit of direct precipitation with the techniques ability to reduce the need for coagulant.
The reduction of coagulant dosage would also reduce the need for hydroxide, and in this case the reduction was 40%. However, the need for cleaning chemicals is high, which results in an overall hydroxide consumption of 996 tons/year. The total yearly CEB chemical amount for conventional precipitation with UF is around 215 tons and around 841 tons for direct precipitation.

Operational and investment costs
The feasibility to transform the alternatives to full-scale operations also depends on the operational and investment cost for each configuration. The operational costs for conventional precipitation, direct precipitation and conventional treatment with UF are summarized in Table 4. The data are taken from operational experience and estimated values from the pilot study, and only reflects costs associated with the pre-treatment of raw water before infiltration.
The results show that conventional precipitation is the cheapest option, $0.014/m 3 , where the coagulant cost contributes the most to the price followed by sodium hydroxide. Direct precipitation on UF membranes manage to reduce the cost of the coagulant and hydroxide used for pH adjustment significantly; however, the need for cleaning chemicals, the energy cost and replacement of the modules increases the price to 0.028 $/m 3 . The high replacement cost, compared to conventional precipitation with UF, comes from the increased number of modules that needs to be replaced and the more rapid deterioration associated with higher CEB frequencies. The price to pre-treat water with conventional precipitation with UF was 0.023 $/m 3 , 60% more than conventional precipitation. The cost for the cleaning chemicals was minor compared to the coagulant and hydroxide cost. Bisulphite cost, used to neutralize hypochlorite waste after cleaning cycles, was not included.
The investment costs for the membranes are based on a net-flux of 40 and 70 L/m 2 ·h for direct precipitation and conventional precipitation with UF, respectively. This translates to a required membrane surface of 201,600 m 2 and 102,900 m 2 to achieve a yearly production of 63 Mm 3 (2,000 L/s). As previously mentioned, it was expected that the investment cost per membrane surface would decrease with an increasing amount of membrane modules. This is due to the saving in other installments in the membrane facility, such as pumps and chemical facilities and so on.
The price per membrane surface area for a 201,600 m 2 membrane facility was estimated to be 10-15% less than a facility with 102,900 m 2 based on Huehmer (), which corresponds to approximately 290 $/m 2 (with 12.5% cost reduction) and 330 $/m 2 , respectively. The costs for the membrane facility are based on experiences from water utilities in Sweden and from Huehmer (). Table 5 shows the estimated costs for each alternative including shared costs such as building, sludge management and chemical facilities.
The results show that the alternative with the lowest and highest investment costs are conventional precipitation and direct precipitation, respectively. The reason why direct precipitation is the most costly is because the most expensive treatment step is being scaled up; that is, the membrane modules. For conventional precipitation with UF, the investment in flocculation and sedimentation steps is saving investment costs in the membrane modules; however, the alternative still costs more than conventional precipitation.
Both membrane alternatives have no need for rapid sand filters, which is one of the highest costs for conventional precipitation. In the end, conventional precipitation would cost $62,000,000; direct precipitation would cost $77,000,000; and conventional precipitation with UF would cost $69,000,000. The data were taken from operational experience at Ringsjö Water Works and data based on the pilot trials. The cost for replacing the membrane modules was estimated at $25/m 2 , based on costs taken from the retailer. The energy consumption in the UF stage used in the calculations were 0.035 kWh/m 3 based on experience from a full-scale plant in Lackarebäck in Sweden.
Other shared costs for each alternative, e.g. administrative costs, heating, lighting, etc., and costs associated with subsequent treatment steps, are expected to be the same for each alternative and thus not included in the calculation. The equivalent cost for HCl was used instead of sulphuric acid, which was used in the pilot trial. a Costs were based on an energy consumption of 290 kWh/day for flocculation and sedimentation basins, 200kWh/day for rapid sand filters, and 0.035 kWh/m 3 for the UF membranes. Energy cost was estimated at $0.11/kWh, based on the Swedish market. b Cost for cleaning chemical. c Calculated from a 7 year module lifetime and 201,600 m 2 membrane surface area. d Calculated from a 10 year module lifetime and 102,900 m 2 membrane surface area.

DISCUSSION
The results from the pilot study shows that it is possible to achieve the same NOM removal with both membrane configurations to the same degree as conventional precipitation.
In this case, exceeding the minimum estimated water quality for infiltration water and reaching drinking water quality.
This leads to other considerations when evaluating the different treatment methods, the first being the amount of chemicals needed to treat the water. During the direct precipitation trials, it was possible to reduce the coagulant dosage by a significant amount (15%) and therefore also the need for hydroxide (40% The benefits of conventional precipitation come from over 30 years of operational experience (Sydvatten AB ) and as seen in Figure S4, the consistent treatment result despite peaks in organic matter from the raw water source. Based on the results from direct precipitation, the membrane had issues with fouling and operational stability, and it seems unlikely that this treatment option will be able to handle strong variations in NOM, which are common after storms, let alone Cost commonly set at 20% of the total cost of the facility. In this case, the average cost for each alternative was used is are therefore the same.
long-term trends of brownification of Lake Bolmen. Also, based on the organic and inorganic load the membrane modules need to withstand, the average lifetime will likely be significantly less. Direct precipitation is most likely not suitable for high production facilities combined with NOM-rich source waters. The treatment method lacks operational stability which is paramount for drinking water producers. Conventional precipitation with UF showed a more stable performance than direct precipitation, and during short periods of time it was possible to increase the gross-flux to 90 L/m 2 ·h. This could provide more flexibility and allow for higher water production when peaks of water consumption occur, for example during warm periods in summer.
Taking everything into consideration, direct precipitation is not a viable option for treating water from Lake Bolmen.
Based on production capacity and NOM removal, both conventional precipitation and conventional precipitation with UF were sufficient. Both techniques are also expected to be better equipped to handle increases in NOM content, which is predicted to occur in Lake Bolmen in the future (Persson ; de Wit et al. ). Conventional precipitation was the cheapest option, however, UF membranes comes with other benefits. One benefit that a membrane facility provides is the extra microbial barrier, which would protect the groundwater supply in Vomb infiltration fields from outbreaks originating upstream. Another safety benefit comes from the operational security of being able to bypass artificial recharge with a sufficient microbial barrier during a crisis, if the infiltration field is compromised because of contaminates, e.g. chemical and oil spills, etc.

CONCLUSIONS
This study showed different ways of applying chemical flocculation and UF membranes in managed aquifer recharge.
The main findings can be summarized as follows: • Conventional precipitation combined with UF was deemed the most suitable choice for full-scale operations due to the increased flexibility and security provided by UF membranes.
• Sufficient water quality can be achieved with all three methods (UVA 254 nm below 5 m À1 ).
• Conventional precipitation had the lowest installation cost followed by conventional precipitation with UF and lastly direct precipitation.
• Treating water using direct precipitation and conventional precipitation with UF increased the production cost of the drinking water ($/m 3 ) by 100 and 60%, respectively, compared with conventional precipitation.
• Direct precipitation resulted in minor coagulant savings (around 15 percent) but with higher consumption of cleaning chemicals.
• A net flux of 40 and 70 L/m 2 ·h was achieved with direct precipitation and conventional precipitation with UF, respectively.
• The CEB intervals were decreased from 12 to 24 hours when using treated feed water in the membrane pilot.