Natural organic matter and sulphate elimination from rainwater with nanofiltration technology and process optimisation using response surface methodology

Water resources are gradually decreasing all over the world due to population growth, climate change, and pollution. Nowadays, rainwater stands out as an alternative natural resource that can be easily collected to address limited water resources problems worldwide (UNICEF and WHO 2012; GhaffarianHoseini et al. 2016). It is possible to use rainwater as drinking water after a certain level of treatment (Silva et al. 2015; Oviedo-Ocaña et al. 2018). Additionally, the effective control of rainwater flow helps to close the hydrological cycle in urban areas. There are many examples throughout history of rainwater stored in cisterns being used as drinking water (Antoniou et al. 2014; Angelakis 2016; Yannopoulos et al. 2017). AbdelKhaleq and Alhaj Ahmed also presented many examples of rainwater harvesting and treatment in some facilities located in sanctuaries in both ancient and modern times in their paper (AbdelKhaleq & Alhaj 2007). The construction of a rainwater harvesting system and installation of a treatment system for the reuse of rainwater was common. However, rainwater may have serious contaminants, contrary to expectation (Jaiyeola 2017; Hamilton et al. 2019), as it is affected by airborne particles and pollutants on surfaces in the collection area. Sulphate and organic pollutant contents are especially important (Eriksson et al. 2007). In urban situations, the sulphate concentrations have increased because of the increasing fog (Penkett et al. 1979; Angrill et al. 2017). On the other hand, studies had indicated that the source of the organic matter in rainwater can be due to different causes (Tran et al. 2020). While biogenic organic compounds are dominant in rural areas, the anthropogenic compounds are more abundant in urban areas' rainwater samples (Kawamura & Kaplan 1986; Pantelaki et al. 2018). Health authorities generally do not recommend the use of untreated rainwater because of disease risks, and some previous studies have proven the relationship between rainwater consumption and water-borne illnesses (Lye 2002; Heyworth et al. 2006; Rodrigo et al. 2011). There is a limit for the daily intake of sulphate from water, air, and food. Generally, it is thought that food is the major source. However, in areas with water supply having high levels of sulphate, water may be the major source of intake (Fawell & Mascarenhas 2004). In the literature, it was found that there was a diarrhoea problem in piglets fed with drinking water including sulphate (Cocchetto & Levy 1981; Morris & Levy 1983). It is also reported that humans can face cathartic effects (US DHEW 1962; Chien et al. 1968), and these effects can be important for population groups like children and elders that may be more sensitive. On the other hand, organic matter in rainwater can be addressed with the parameter of Natural Organic Matter (NOM). NOM is a complex matrix of organic substances in aquatic ecosystems that are generated as a result of biological, geological, and hydrological cycles (Sillanpää et al. 2018). Intrinsically, NOM is not toxic but its presence in water can be hazardous. For instance, it can act as a carrier of toxic pollutants including pesticides and radionuclides (Knauer et al. 2017; Santschi et al. 2017) and increase their bioavailability through increased solubility (Reid et al. 2000). Moreover, Sillanää and colleagues also underlined that NOM tends to form some strong complexes with heavy metals (Matilainen et al. 2011; Tang et al. 2014) and it reacts with disinfectants, resulting in the formation of disinfection by-products (DBPs) (Golea et al. 2017; Goslan et al. 2017). This leads to the creation of carcinogenic compounds (Bond et al. 2012; Jiang et al. 2017). Besides, the presence of NOM causes the sensorial properties of rainwater to change. Water is naturally considered to be of poorer quality if taste or odour can be perceived (de França Doria 2010), and consumers are suspicious of drinking water with an unpleasant taste or odour. As a result, the concentration of sulphate and organic matter should be reduced, not only for health reasons but also for taste and odour reasons (Shen & Schäfer 2015). Taste disturbance varies according to the nature of the associated cation (WHO 2011) and a sulphate concentration of 25 mg/L is recommended by Turkish Standards as a threshold level for the minimal disturbance of taste (TSI 2005).

The removal of sulphate and NOM from drinking water supplies is becoming a challenging task requiring the application of reliable and highly efficient water treatment technologies. Many methods have been proposed to treat rainwater in the literature. These methods include physicochemical processes such as granular activated carbon (GAC) combined with ultraviolet (UV) disinfection or nanofiltration (NF) (Naddeo et al. 2013; Ding et al. 2018; Leong et al. 2018), chemical processes like chlorination and adsorption (Keithley 2012; Omar et al. 2017) and biological processes like biosand filter (Kassim & Hashim 2006). Membrane filtration technologies are gaining preference for rainwater treatment due to their low carbon footprint and low disinfectant consumption (Oosterom et al. 2000; Kus et al. 2013; Dobrowsky et al. 2015; Leong et al. 2017). According to the results of a study carried out with raw rainwater collected from a rainwater channel in Sydney and treated using GAC and ultrafiltration (UF) membrane, organic matter concentrations were significantly reduced (Kus et al. 2012). On the other hand, Kim et al. (2005) investigated the treatability of rainwater by preferring a metal membrane, and found it particularly suitable to treat rainwater due to its high purification efficiencies, but the expected delay in clogging could not be achieved (Kim et al. 2005). It is obvious that the treatability of rainwater with membrane processes is a research topic that deserves further examination. Besides, it can be also seen that the treatment studies in the literature using membranes were generally carried out using conventional and classical methods of experimentation. In this method, the researcher has to carry out many experiments in which one of the parameters is varied while the others remain constant. As this approach can be time-consuming and results in the lack of explanations with a comprehensive mathematical relationship, this ultimately causes poor optimisation of the process concerned. On the other hand, the response surface methodology (RSM), which is a methodology for the design of experiment (DoE), can be applied as an alternative. DoE is a statistical tool to reduce the number of experiments/runs and to ensure that the maximum data can be collected from the performed experiments, and RSM takes a statistical approach where all the variables are varied over a wide range of levels and in turn swiftly manages to eliminate the aforementioned drawbacks of the conventional and classical method (Myers et al. 2016; Montgomery 2017). RSM can be practically used by researchers to obtain mathematical models for making some predictions via simulating the membrane processes (Santafé-Moros et al. 2005; Xiarchos et al. 2008; Jadhav et al. 2016).

To sum up, urbanisation has increased the number of impermeable areas in rainwater collection basins, and the leakage and storage of rainwater are restricted (Zhang et al. 2015; Li et al. 2017). For this reason, it is important to design and construct rainwater collection and storage elements in newly built structures (Sürmeli 2016). Moreover, it must be highlighted that rainwater should go through proper treatment before using it as drinking water (Shen & Schäfer 2015) and researchers argue that the feasibility for this kind of treatment unit can be determined through a detailed analysis of design and operation parameters (Villarreal & Dixon 2005; Mun & Han 2012). Few investigations have addressed the application of the NF membrane, which is a type of pressure-driven membrane having properties in between those of UF and reverse osmosis membranes, in the treatment of rainwater. Some of these studies were exemplified above. Indeed, the NF process benefits from ease of operation, reliability, and comparatively low energy consumption, as well as high efficiency of pollutant removal, and it has given rise to worldwide interest. In this regard, the treatability of rainwater with NF was studied and optimised in this paper with samples taken from the cistern of the new mosque, which was built on campus in Istanbul. Sulphate and NOM removal in NF processes performed at different operating pressures with three commercial membranes (NF90, NP030, and NP010) have been evaluated in terms of potability of the permeate after disinfection. Finally, the technical feasibility of NF at low pressures with a membrane having bigger pore size and higher flux, as a result, was presented within the regard of the models from RSM and fouling analysis.

Three commercial flat sheet polymeric membranes were used in this study. These membranes were subjected to the rainwater treatment after their characterisations. The technical properties of these membranes are given in Table 1. Deionised water (DI water) was preferred for all experiments, and DI water was supplied by using a Milli-Q^® IQ 7003/05/10/15 system (Germany). The Dionex™ Combined Seven Anion Standard II was used for the calibration curve preparation to determine the anion concentrations in the samples (Thermo Fisher Scientific, USA) using Ion Chromatography (IC). Metal concentrations in the samples were determined with the calibration curve created by single-element standards manufactured for Inductively Coupled Plasma Optical Emission Spectrometry (ICP-OES) by Inorganic Ventures (USA). The rainwater was taken directly from the cistern, of which details are explained in the following sections. The characteristics of rainwater are shown in Table 2.

The rainwater was supplied from the cistern located in the garden of a mosque, newly constructed on the university campus. While the volume of the cistern was 278 m³, the rainwater collection area was 2,000 m². There were four channels for the rainwater collection in the tetrastoon and a weir in the cistern to remove the excess amount of rainwater during the heavy storm days. The water sample was taken using clear plastic bottles. The bottle was filled and sealed. The temperature, pH, and conductivity were measured on-site. All on-site measurements were triplicated. Following this, rainwater was characterised prior to the NF experiments to determine the concentration of some metals, anions, and cations (Step-1 in Figure 1). Commercial membranes were also subjected to the characterisation prior to the NF studies (Step-2). NF experiments had been carried out by following the experimental combinations given by RSM (Step-3). The optimum membrane and operating pressure were determined for a feasible rainwater treatment. Finally, to determine the reuse potential of commercial membranes, the four filtration cycles were also carried out using the rainwater at optimum operating pressure (Step-4). The experimental plan is schematically summarised in Figure 1.

All filtration experiments including membrane characterization and rainwater treatment were performed with Sterlitech Brand High-Pressured Filtration Cell (HPFC) (HP4750, Sterlitech Coop, USA). The system consists of a filtration cell, a nitrogen gas tank with a regulator, a permeate collection cell, and a computer that records data for the calculation of flux. Schematic representation of the experimental setup and technical properties of the high-pressure filtration cell are given in Figure 2 and Table 4, respectively.

After the determination of hydraulic permeability, membranes were subjected to additional characterisation with zeta potential (ZP) and contact angle (CA), and scanning electron microscopy (SEM) analysis, because the surface characteristic of the membrane plays an important role in the permeation and rejection. The top surface morphology of the membranes was observed by using the FEI scanning electron microscope at standard high-vacuum conditions. Polaron sputter coater was used to coat the outer surfaces of membrane samples with gold. The hydrophilicity of the membrane was determined using contact angle measurements. ZPs of clean membranes were determined using an electrokinetic analyser (Anton Paar Surpass, Austria). CAs of clean NF membranes were measured by the KSV Attension Theta Brand contact angle measure instrument (Sweden) using scissile drop mode on dried membranes. A small droplet of DI water was delivered onto the membrane surface and after waiting 1.5 s, an image of the droplet on the membrane surface was taken. Each ZP and CA of the membrane was measured at a minimum of five different points on the samples in order to calculate the average value.

Raw rainwater and permeate samples were collected for analysis during the NF experiments. Elemental analysis was performed using ICP-OES (3000 DV, Perkin Elmer, USA) and IC (Dionex ICS-3000, USA) for the raw rainwater characterisation. Also, a total organic carbon (TOC) analyser (Shimadzu V-CPN, USA) was used for the determination of NOM concentration in raw and treated rainwater as described in the paper of Pantelaki et al. (2018). The sulphate concentration in raw and treated rainwater was determined using IC. At the end of the NF processes, the morphology of the membranes was observed again using SEM (FEI Quanta FEG 250, USA). Before scanning, the surface of the membranes was coated with 3–4 nm gold-palladium (Pd–Au) using the Quorum SC7620 model sputter coating machine (UK) as carried out with clean membranes. All pH determinations were realised with the Fisher Scientific probe (XL Series, USA). All analytical measurements were conducted in triplicate, and the average values with standard deviations are presented in the paper.

FRR values of commercial membranes were evaluated regarding the analysis of the foulants observed on the membrane surface after the rainwater filtration. SEM images of used membranes were obtained with this aim. Finally, the four filtration cycles were carried out with a high volume of rainwater to determine the reuse potential of commercial NF membranes. At the end of the first cycle, the membrane was physically cleaned with DI water and the second filtration cycle was started with the same membrane. These applications were complete after four runs and a chemical cleaning was applied at the end. NaOH was preferred as a model chemical cleaning solution and prepared as described in the paper of Li & Elimelech (2004). The solution (3% v/v) was freshly prepared with DI water right before the experiment.

The characterisation of rainwater used as the feed solution during the NF process is presented in the Materials and Methods Section as Table 2. Rainwater was completely colourless and odourless. The conductivity of rainwater was 320.0 μS/cm and lower than the conductivity of conventional tap water and the maximum conductivity value (1,017.0 μS/cm) reported in the study of Leong et al. (2017). On the other hand, the NOM concentration (3.30 mg/L) was slightly higher than for a rainwater sample presented in the literature (Leong et al. 2017). The concentrations of fluoride, chloride, and manganese in the rainwater were as expected. Besides, the only anion that had a high concentration when compared to others was sulphate. The sulphate concentration was two times higher than the concentration that was originated from asphalt surfaces in the study of Angrill et al. (2017), and this can be related to combustion emissions from transportation on the campus. The metal concentrations in the rainwater were negligible.

Clean NF membranes were also characterised before the experiments. The DI water permeability, ZPs, and CAs of three commercial membranes are presented in Table 5. While the highest DI water permeability was obtained with NP010, the lowest value was obtained with NF90. This means that DI water permeability was directly proportionate to the pore size, as expected. Then, the surface properties of the three NF membranes were assessed by measuring the streaming potential, expressed as ZP. While the ZP of NP010 was positive at the pH of rainwater, it was negative for NP030 and NF90. This means that NF90 has more potential to reject anions like sulphate, since the membrane surface and anions were both negatively charged. Finally, the CA values of the clean NF membranes were also investigated to evaluate their hydrophilicity. The highest hydrophilicity can be exhibited by NP030 and NF90 with their lower contact angles when compared to NP010 with a contact angle of 72°.

The characterised membranes were then subjected to rainwater filtration as explained in detail in the Materials and Methods section (Step-3 in this study). The average flux values for rainwater treatment are presented in Figure 3. The rainwater fluxes during NF were lower than DI water flux for all three NF membranes. Rainwater flux increased with the increasing operating pressure, but the pore size was another factor in this study. To understand the individual and combined effect of the pore size and operating pressure on the flux, a unique RSM study was carried out.

The effect of pressure on sulphate removal during the NF of rainwater was not significant when compared to the effect of the pore size in this study. Al-Zoubi et al. (2007) presented similar results from their NF studies carried out with MgSO₄ and Na₂SO₄ salts. The rejections did not change with increasing pressure, especially at operating pressures higher than 6 bar (Al-Zoubi et al. 2007). Stable removal efficiencies with changing operating pressure generally prove that the concentration polarization on the membrane surface was slightly abundant. The effect of concentration polarization is to reduce actual product water flow rate and salt rejection versus theoretical estimates. Theoretical estimation was an increasing rejection with increasing operating pressure. Jadhav and colleagues confirmed that the key mechanism of ion rejection was the charge exclusion (2016) and found also that pressure was the most ineffective parameter when compared to characteristics of membranes and feed solution.

As it is mentioned in the literature that the insoluble organic matter in the rainwater has an average pore size of 0.3–1 μm (Holecek et al. 2007), all NF membranes could successfully reject the organic matter. The NOM removals were not significantly affected by the pressure and the pore size. The average NOM removal efficiencies of NP010, NP030, and NF90 were 97.27%, 98.06%, and 99.20%, respectively. The results fit the NOM rejection efficiencies obtained in the study of Shen & Schäfer (2015). They had found that the efficiencies with NF90 were higher than the efficiencies with other NF membranes. When the NOM rejection by membranes was evaluated, it can be said that NOM rejection is independent of MWCO, as the molecular size of NOM is significantly larger compared to the membrane pore size (Shen & Schäfer 2015). Only small amounts of NOM could pass through the membrane.

In summary, the NF process, which is a suitable process for obtaining treated rainwater without causing any problem in taste, could be optimised using model results to find the optimum point that meets the required permeate quality and lowers energy consumption. Silva et al. (2016) also reported in their studies that the model parameters obtained by independent methods can be used as a predictive tool. The process was optimised by considering the minimum sulphate removal efficiency. According to Model-S, the only important and effective factor was the pore size of the membrane. When Model-S was solved, it was seen that the biggest pore size of the NF membrane should be approximately 800 Da (62.8% removal). Since the operating pressure was not a significant factor when compared to the pore size of a membrane, NF processes can be carried out at a low operating pressure like 6 bar. When Model-N was used, the NOM removal was found as approximately 99%. As the final step, the flux was estimated as 36.4 L/m²/h utilising Model-F. This flux is higher when compared to the permeate flux (∼12.5 L/m²/h) presented in Shen and Schafer's study carried out at 6 bar with natural waters (Shen & Schäfer 2015). As a result, it can be said that a package NF unit, including two spiral wound 8-inch elements, can be installed on the basement floor of the building and 13 tons of treated water can be supplied in a day with only a 6 hours-long operation.

To understand the fouling tendencies of used commercial membranes, the flux reduction ratio values (FRR) were calculated as described in Section 2.5 and are presented in Figure 6. The FRR values of membranes on rainwater filtration varied between 8 and 15.7, 4.4 and 27, and 26.2 and 35.3% for NF90, NP030, and NP010, respectively. Concerning these values, it can be said that FRR values were directly proportional to pore size characteristics of commercial membranes. This result fits the results obtained by Koseoglu-Imer (2013). She also mentioned in her paper that the change of FRR in five different UF membranes manufactured at the laboratory was proportional to the porosity (Koseoglu-Imer 2013). The increase of pore size increased the adsorptive fouling except for the filtration at 9 bar. The effect of operating pressure was negligible. Besides, all membranes were subjected to microscopic investigation after the rainwater treatment to visualise the fouling effects. The SEM images can be seen in Figure 7. The virgin NF90 membrane exhibits the network-like structure typical of a membrane polyamide layer. The SEM image of fouled membrane clearly shows the accumulation of a foulant layer onto the membrane surface. The visual observation of the SEM image indicates that surface adsorption may be the main mechanism for NF fouling. When the shape and surface morphology with minor colloids on the used membranes was evaluated, it can be said that these are indicative of organic fouling. It is also observed that some white crystal substances appear in the fouling layer, which should be sulphate. NP010 membrane, which has the biggest pore size in the membranes used in this study, had less foulant on its surface, in line with its lower treatment efficiencies. NP030 membranes had more foulants on the surface than the foulants on NP010. The highest amount of foulant was seen on the surface of the NF90 membrane. In the case of NF90, the colloids were distributed over the entire membrane surface due to hydrophobic interactions between the colloids and the membrane surface. As a result, valley clogging did not occur.

It is important to prove that the selected membrane will have flux recovery property and a long lifetime during the operation. As explained in Section 2.6, commercial membranes were used four times to see their flux reduction profiles. The flux profiles can be seen in Figure 8. At the beginning of the first cycle, the membrane was conditioned and each cycle included 1.5 h rainwater filtration. At the end of the first cycle, the membranes were physically cleaned with DI water and the second filtration cycle was started with the same membranes. After four runs of rainwater filtration, which equals 6 h long NF in a day, the membrane was subjected to chemical cleaning using sodium hydroxide. The aim was to simulate a non-continuous operated plant described in the previous part. It is expected that flux will decline in time because of fouling. This decline could be seen in each cycle continuing 90 minutes. The physical cleaning applied at the end of each 90 minutes helped to increase the flux; however, the initial flux in the second, third, and the fourth cycle was lower than the initial flux value observed in the previous cycle. This means that physical cleaning could not be sufficient to remove all foulants on the membrane surface.

As a result, it seemed that physical cleaning was sufficient to complete the day with a 33% loss of flux averagely for all membranes used during 6 hours. At the 360th minute, membranes were subjected to chemical cleaning. The flux at the 361st minute of the operation was almost equal to the initial flux obtained at the very first minute of the operation day. So, results showed that caustic cleaning was effective at controlling membrane fouling and chemical cleaning could efficiently recover the flux at the end of the day/operation. In good agreement with the literature, high membrane permeability recovery was observed here. However, alkaline cleaning may lead to temporary enlargement of the membrane pores, and it can be said that this temporary enlargement of the membrane pores due to alkaline cleaning subsequently may result in some notable changes in the rejection of contaminants. For instance, it was reported in the literature that sequential alkaline cleaning caused a decrease in rejection (Liikanen et al. 2002; Simon et al. 2013). To prefer the use of more dilute alkaline cleaning agent as in this study is suggested.

This study proposed a promising approach for rainwater recycling by the NF treatment process. Elimination of natural organic matter and sulphate from rainwater using NF membrane technology was discussed in this regard. It was seen in the literature that NF can be a suitable process for obtaining treated rainwater without causing any problem in taste. In addition to treatment performances, the membrane fouling was also identified.

RSM was successfully applied for evaluating the importance of operating parameters and approaching the optimal NF process conditions for rainwater treatment. RSM based on CCF design was applied to study the effects of operating parameters including operating pressure and membrane pore size on the NF of rainwater containing sulphate and NOM. From the results of the experiments, the following conclusions can be drawn:

• (1)

  The more effective factor was the pore size of the membrane, and it was seen that the biggest pore size of the NF membrane should be a minimum of 800 Da. This means that NP010 should not be preferred for rainwater treatment when it is required to meet the health standards. On the other hand, it is not an obligation to prefer a tight NF membrane to prevent the taste problem while rainwater drinking. This means that the feasibility can be enhanced with reasonable decisions in the planning and design steps.

• (2)

  By considering the removal of natural organic matter and sulphate with maximum allowable concentrations, it could be found that NF processes can be carried out at low operating pressure (6 bar in this study). From the resultant models, it is expected to achieve a NOM removal with a percentage of 99% when the concentration of sulphate is below the limit. In an NF process carried out at an operating pressure of 6 bar with the proper membrane, the flux would be 36.4 L/m²/h, which is a relatively high flux compared to the flux values presented in the literature. So, it is possible to operate a low-risk system concerning occupational health and safety and operational costs by making no concessions to have a sufficient amount of treated water.

• (3)

  Membrane fouling was further examined based on SEM analyses and membrane cleaning procedures. Considering SEM measurements, it was determined that surface fouling was effective on the flux decline caused by fouling. The amount of the foulants on the membrane surface were not at high levels, and commercial membranes presented high cleanability on long-term rainwater filtration. A diluted alkaline cleaning agent was suggested by the author to prevent commercial NF membranes from pore extensions.

To conclude, rainwater recycling systems can save a large quantity of relatively high-quality water at a reasonable cost although there had been some published results based on the misrepresented operational costs of membrane systems. There is a growing awareness of water-sensitive urban design with water recycling. It seemed important to consider the costs and environmental impacts before choosing a treatment method for that recycling. Membrane processes have better drinking water quality, safe drinking water, and lower environmental impact compared to the other treatment alternatives. Especially regarding the energy demand, NF is known to have a large demand due to the required pressures; however, it was proven in this study that lower operating pressures can be also applied. Generally speaking, NF was found to be a proper method for NOM and sulphate removal from rainwater, and a package NF unit can be easily installed on a university campus for the reuse of water.

There was no funding provided for this study. The author would like to thank the National Research Center on Membrane Technologies (MEM-TEK) for SEM analysis. Also, the author is thankful to the campus administration for the rainwater supply. The graphical abstract was prepared with the software of Microsoft PowerPoint.

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