Ammonium removal of biological roughing filter for rural drinking water pretreatment

Access to safe drinking water is a big challenge in rural areas around the world, especially in developing areas. One major increasing concern is the pollution of source water, which could result in a low level of drinking water quality; ammonium is one of the predominant pollutants, resulting from environmental pollution, lack of regulation and protection of water sources, seasonal change and so on (Wilbers et al. 2014; Chen et al. 2016). It is toxic in high levels for aquatic system (Prenter et al. 2004); at a low level, although it is not directly related with human health, it might lead to the accumulation of toxic nitrite (Williams et al. 2010). Moreover, it can react quickly with chlorine, the most popular disinfectant, which might result in not only larger amount of chlorine dose and toxic disinfection by-products (Zhang et al. 2011; Bond et al. 2014), but also taste and odor problems (WHO 2011), thus presenting health risk and reducing the satisfaction degree of consumers. Therefore, ammonium is regulated by 81 countries and territories around the world, and over half of them restrict ammonium to a maximum concentration of 0.5 mg/L (WHO 2018). However, an ammonium concentration exceeding the regulated value in finished water is not rare in rural areas, which raises the public concerns (Sun et al. 2009; Zheng et al. 2011).

Improving drinking water treatment is necessary for solving this problem. It is noted that lots of rural water treatment plants only apply conventional water treatment or partial treatment (e.g. disinfection only, combined filtration and disinfection), which is limited in removing ammonium, and new treatment schemes are helpful. Conventional biological pretreatment technologies (e.g., biological aerated filters) are usually employed to deal with ammonium pollution. They are effective but need effective aeration as well as relatively frequent backwashing, leading to the problem of high energy consumption and difficulty of management and operation. Besides, breakpoint chlorination, a common disinfection method, could also remove ammonium. This method, however, might consume larger amount of chlorine dose and produce toxic disinfection by-products (Zhang et al. 2011; Bond et al. 2014), increasing cost and health risk. These characteristics impede their application in rural water supply. Roughing filtration is a traditional pretreatment technology, widely applied in rural areas (Nkwonta & Ochieng 2009). It is often designed for removing particles for its outstanding solids retention capability, while it also performs well biologically to remove iron, manganese, and organic pollutants as biofilm can grow on the surface of the filter medium (Lv et al. 1998; Pacini et al. 2005). Therefore, it can also serve as a biological technology to carry out nitrifying process for ammonium removal. Compared to conventional biological pretreatment technologies, roughing filtration is much cheaper without aeration and more efficient in turbidity control as well as much simpler in operation and management, which make it more advantageous in rural water supply. However, although it is an attractive technology to cope with ammonium pollution in rural areas, the majority of previous researches were focused on the function of removing particles (Nkwonta & Ochieng 2009). Our previous research tested the performance of roughing filters to cope with ammonium in a short-term experiment, which showed about 70% removal (Zeng et al. 2020), proving that it is a promising solution. However, more information, such as influence of flow rate and experiment time, are needed.

This paper was aimed at developing a roughing filter for rural water supply to handle ammonium pollution by investigating the performance of roughing filtration and its influence factor, and optimizing operation conditions. Unlike conventional filter media, such as gravel and broken brick, roughing filters in this study were filled with ceramic media to increase the surface area of the filter medium and improve the biological function. Ammonium removal as well as turbidity removal were explored. Furthermore, related microbial analysis (i.e., biomass, biological activity, microbial community structure) of roughing filters was also performed. This study could facilitate a better understanding of biological roughing filters, and provide reference for the development of on-site water treatment technologies suitable for rural water supply systems.

The structure of roughing filters is shown in Figure 1. Two acrylic columns (2 m in height and 34 cm in diameter) were established in parallel, packed with ceramic media made of shale (Jinyi Water Purification Material Co., Ltd, Gongyi, China) with a size ranging from 3 to 10 mm. The characteristics of ceramic media are demonstrated in Table 1. Filters were arranged in two layers, together with a supporting layer packed with 10 cm 8–16 mm gravel. In order to increase the solid retention capability of filters, raw water first went through the bottom layer filled with larger media (6–10 mm) and then flowed to the top layer packed with smaller media (3–6 mm). According to our previous study (Zeng et al. 2020), 40 cm of bottom layer filled with large size medium was appropriate, so in this study, the bottom layer was set to be 40 cm while its top counterpart was set to be 60 cm to obtain larger biomass for its larger surface area. Ceramic media was crudely washed using tap water before being packed in filter columns.

The pilot-scale study was performed in the Institute of Urban Environment, Chinese Academy of Sciences, Xiamen, China. The man-made stream flowing through this Institute was employed as a water source. Characteristics of source water are shown in Table 2. In the start-up period, the flow rate was first set to 1 m/h (2–3 days) and then raised gradually to 2 m/h (2–3 days) and then remained at 4 m/h. The acclimation time included all the periods of 1 m/h, 2 m/h and 4 m/h. After acclimation, the hydraulic load was kept to 4 m/h. The filters were run from June 2018 to January 2019, including both summer and winter periods. Backwashing combining air and water was performed every 2–3 weeks using tap water. The cleaning procedure was: (1) 2 min with air; (2) 6–10 min with air and water; (3) 2–5 min with water. The intensity of air and water was 20–25 L/(m².s) and 8–9 L/(m².s), respectively.

Water samples were collected from the containers of raw water and effluent (Figure 1). Effluent mixture of the two columns was analyzed instead of individual filter columns. Turbidity was detected using a portable Turbidimeter (Orion AQ4500, Thermo, America). Ammonium concentration (calculated as NH₄⁺-N) was determined by Nessler's reagent spectrophotometry (Shimadzu, UV1800, Japan). N-(1-Naphthyl)ethylenediamine and ultraviolet spectrophotometry methods (Shimadzu, UV1800, Japan) were used for nitrite (calculated as NO₂⁻-N) and nitrate (calculated as NO₃⁻-N) analysis, respectively.

Ceramic media was collected from different heights of filter (20, 40 and 80 cm). The wet media was then weighted and put into 50 mL sterile centrifuge tubes. Next, sterile NaCl solution (0.9%) was added before shaking. Afterwards, the tubes were treated with ultrasonic (KQ-500D, Kunshan Ultrasonic Instruments Co., Ltd, Suzhou, China, 100% power) for 10 min with a 5 min interval for three times. During the interval, tubes were vortexed for about 30 s. The suspension then was used for biomass detection and DNA extraction. For biomass analysis, the Heterotrophic Plate Count (HPC) method was employed. The suspension was first diluted, and then three aliquots (100 μL) were spread onto nutrient agar plates, which were incubated for 48 h at 37 °C before counting.

Biological activity was determined by Oxygen Uptake Rate (OUR) method. Filter medium was taken from filters (height: 60 cm), which was separated into three parts. One remained untreated, one was slightly rinsed twice with sterile NaCl solution (0.9%), and the last part was rinsed twice and then treated with ultrasonic (KQ-500D, Kunshan Ultrasonic Instruments Co., Ltd, Suzhou, China, 100% power) for 10 min with a 5 min interval, for three times. During the interval, tubes were vortexed for about 30 s. After being pretreated, the filter medium was weighted and transferred into small sterile bottles. Raw water was aerated till saturated with oxygen after pasteurization at 80 °C for 30 min. Next, the bottles were filled with raw water with no bubbles and closed carefully for isolation of air. For better detection, the bottles were incubated at 37 °C, 180 rpm for 3–5 h to facilitate oxygen uptake. The value of OUR was calculated according to oxygen consumed, the weight of filter medium and the incubation time. Each experiment with different pretreated filter media was conducted in octuplicate, and more than three blank samples were prepared at the same time.

The suspension of ceramic media collected at 40 cm from the bottom (described above) was filtered using 0.22 μm membrane. The membrane was stored in −20 °C for DNA extraction. FastDNA SPIN kit (MP Biomedicals, America) was used in this study according to the instructions provided by the manufacturer. The DNA sample was then sent to a high-throughput sequencing company (Shanghai Meiji Biotechnology Co., Ltd, Shanghai, China) for quality testing and sequencing. The data were then analyzed in the I-Sanger platform (https://www.i-sanger.com) provided by this company.

One-way analysis of variance was performed by using IBM SPSS Statistics 20 and the differences were considered statistically significant if p < 0.05.

For a biological filter, an ammonium removal efficiency larger than 60% is usually recognized as a clue for developed biofilm (Hong 2013). In the start-up period, water temperature was high, between 28 °C and 33 °C, and roughing filters in this study only needed about half a month to achieve a stable ammonium reduction of more than 60% (Figure S1), much shorter than that reported by previous studies (Basu et al. 2016; Zeng et al. 2020), which have presented an acclimation period of about one month. This might be ascribed to the suitable water temperature and substrate concentration for bacteria growth.

Biological ammonium transformation consumes a large amount of oxygen, but typically a roughing filter is often run without aeration and employs a small hydraulic load, usually below 1.5 m/h, to facilitate particle removal (Nkwonta & Ochieng 2009), which makes it very different from biological aerated filter. Although small flow rate can increase the empty bed contact time, as well as non-aeration operation it would affect aerobic biological processes, like nitrification, because of lack of oxygen and other nutrition. In this work, flow rates varying between 2 m/h and 5.3 m/h was considered for enhancing ammonium removal performance of roughing filters. Moreover, related water indexes (i.e., nitrite, nitrate) were also analyzed in this experiment. The results are illustrated in Figure 2.

The scenery at 2 m/h significantly differed from that in higher flow rates (p < 0.05), while 4 m/h and 5.3 m/h showed a close performance. In 2 m/h, only about 30% of NH₄⁺-N was eliminated, and when influent NH₄⁺-N concentration was about 1 mg/L, the effluent concentration was as high as 0.7 mg/L. This could be ascribed to the deficiency of oxygen at such low hydraulic load. Interestingly, our previous study, conducted in winter with temperatures of 10–25 °C, showed a good NH₄⁺-N removal at 2 m/h with turbidity and dissolved oxygen fluctuating from 1 to 3 NTU and 6 to 11 mg/L, respectively (Zeng et al. 2020). Possible explanations were that higher oxygen concentration and much lower particle concentration presented in raw water; experimental condition of winter which might lead to smaller biomass in the filter and thus lower oxygen consumption, compared to this experiment, performed in summer. By contrast, the conditions of 4 m/h and 5.3 m/h presented a similar removal efficiency of around 60%. These results indicated that small flow rate did affect filter performance and this undesirable situation could be improved by increasing the flow rate. Nevertheless, further increasing the flow rate from 4 m/h to 5.3 m/h could not enhance filter performance any more, which might be ascribed to insufficient substrate concentration. It can be seen that the effluent concentration at 4 m/h and 5.3 m/h was almost the same, about 0.3 mg/L, and it would be hard for bacteria to utilize substrate in such low concentrations (Rittmann & Snoeyink 1984; Tao et al. 2016). This result was in line with that of Fang et al. (Fang et al. 1998), which reported that ammonium removal efficiency at 7 m/h was almost the same as that at 4 m/h in bio-ceramic reactors. Bio-ceramic reactor could achieve high removal in about 5 min (Wang & Liu 1999), suggesting that moderately enhancing the hydraulic load would not affect ammonium reduction even with a shorter contact time.

The situation of nitrite and nitrate was similar to ammonium. It is noted that the higher the flow rate, the larger removal (accumulation) percentage of nitrite and obtained, which followed the regular pattern of ammonium transformation. For example, nitrite reduction at 4 m/h was 68%, close to that of 5.3 m/h (75%), which was about three times larger than its 2 m/h counterpart. This result indicated that transformation of nitrite into nitrate was effective in all flow rates tested in the experimental condition.

In summary, a flow rate of 2 m/h was too small for ammonium oxidizing in this study. At 5.3 m/h, in spite of good performance, particle removal capacity of filters as well as the length of backwash period would be influenced compared with lower flow rate. Since being the turbidity barrier of the whole drinking water treatment process is the prior, the final flow rate of roughing filters was set to 4 m/h according to the above results. If only considering the performance of ammonium elimination, flow rate could be moderately enhanced with the prerequisite of suitable contact time.

After the ripening period, ammonium removal was monitored for about 6 months, as demonstrated in Figure 3. In this experiment, we used the Chinese regulated value of NH₄⁺-N of 0.5 mg/L as criteria for better discussion, since China is a typical developing country facing the challenge of safe rural drinking water. It can be seen from Figure 3(a) that most water samples could achieve 0.5 mg/L in the outlet, other than the samples with NH₄⁺-N concentration in raw water higher than about 1 mg/L. This indicated that the largest NH₄⁺-N loading rate of roughing filters was about 0.1 kg/(m³·d), in order to meet the requirement of 0.5 mg/L in the effluent. Moreover, Figure 3(b) shows that removal efficiency was also influenced by initial NH₄⁺-N loading rate in raw water, with an average reduction of about 51%. Initial NH₄⁺-N concentration, ranging from about 0.4 mg/L to 1.7 mg/L, facilitated NH₄⁺-N removal with values varying between 44% to 86%. By contrast, under the condition that initial NH₄⁺-N concentration is less than 0.2 mg/L or larger than 1.7 mg/L, low removal efficiencies were obtained of between 18% to 40%. This might be ascribed to limitation of low substrate concentration (Rittmann & Snoeyink 1984; Tao et al. 2016) and oxygen, respectively. Lv et al. (Lv et al. 1998) employed a two-stage horizontal roughing filter to treat river water, and observed 90% ammonium removal from 0.75 mg/L to 0.075 mg/L. The higher removal might be attributed to the long contact time of 300 min and the supplement of oxygen by air and drop-aeration.

Ammonium oxidation needs a large amount of oxygen, and since it is run without aeration, it could be expected that a roughing filter, unlike other aerated treatment reactors, might be incapable of coping with high ammonium concentration. Besides, the large number of particles accumulated inside the filter and the long cleaning period would also affect the biological performance. As expected, roughing filters in this work could not perform well when fed with raw water containing ammonium greater than 1 mg/L. However, NH₄⁺-N effluent concentration could below 0.5 mg/L within a NH₄⁺-N loading rate of 0.1 kg/(m³·d). These results suggested that in general, a biological roughing filter could handle moderate ammonium pollution and might be expected to help rural drinking water plants adapt to the seasonal change and moderate deterioration of water quality with low cost and simple operation. Furthermore, if only considering ammonium removal, the loading rate could be enhanced to about 0.15 kg/(m³·d).

During operation, the media porosity was gradually clogged. Then backwashing using tap water was employed every 2–3 weeks, since hydraulic cleaning was limited (Zeng et al. 2020). For a roughing filter that employs a larger size medium, backwash intensity would be greater, which might impair biological performance. In this experiment, the impact on backwashing was investigated. As shown in Figure 4, ammonium removal in the first day after backwashing was slightly affected, while it gradually recovered in the next 2–3 days. This indicated that intense backwashing with tap water would not strikingly impact ammonium removal.

Nitrification is a biological process that depends on the amount and activity of biofilm developed on the filter medium. In this study, biomass of biofilm was tested by HPC method. As shown in Figure 5(a), about 10⁵ CFU cells were detected per gram wet ceramic media, which was obtained by HPC using Nutrient Agar. It is noted that Nutrient Agar is rich in nutrition and might lead to an underestimate of living cell numbers, as a drinking water system is an oligotrophic environment. Moreover, the biomass of the biofilm developed in various heights of the filters was at the same level, and only a marginal difference could be observed that the biomass in higher places was slightly larger, which was different from other types of up-flow biological filters (Sang et al. 2003). In general, the advantage of richer nutrition level in the bottom would facilitate a larger biomass than in the top part. However, in an up-flow roughing filter, the surface area of filter medium in the bottom layer was only 1.97 m²/g, while the top layer was 2.4 m²/g (Table 2), which could support larger biomass and contributed to this irregular phenomenon.

Since biofilm is attached on the filter medium, any forces that can detach biofilm, such as backwashing, will influence the biological activity. For a roughing filter that employs larger size medium, backwashing intensity would be greater, leading to more serious damage on the biofilm. Thus, when carrying out the OUR test, a filter medium with high- strength pretreatment was also employed. Results revealed that microbial activity was affected by different pretreatments to the ceramic media (Figure 5(b)). For example, original ceramic media without any treatment presented an OUR value of 7.04 μgO₂/(g·h), and similar to that reported by aerated bio-ceramic filter (Sang et al. 2003). However, rinsing carefully could markedly reduce biological activity of biofilm to 4.80 μgO₂/(g·h), mainly due to the removal of accumulated sludge and free bacteria. This result implied that the accumulation of sludge and free bacteria would consume lots of oxygen, which might act against ammonium removal. Nevertheless, it is noted that microbial metabolism still appeared with an OUR value of 3.88 μgO₂/(g·h) after being violently treated by combined rinse and sonication, indicating that after being fiercely pretreated, biofilm appeared with some damage in biological activity, but it could not be completely damaged and a large number of viable microorganisms were still living on the filter medium. These results suggested that biofilm could survive intense backwashing; since nitrifying bacteria exist in the inner part of biofilm, the performance of ammonium removal could recover soon in suitable conditions, which was in agreement with the above results (Figure 4).

In order to give a detailed insight into the microbial community and explore its relation with filter performance, the phylogenetic classification of bacterial sequences at phylum level (Figure 6) and genus level (Figure S2) were summarized. High diversity of bacteria was found (1128 OTUs), and the phyla of Proteobacteria and Chloroflexi dominated in the biofilm, which accounted for 27.04% and 22.15%, respectively. Unexpectedly, high abundance of Cyanobacteria (12.47%) was also observed. Moreover, the existence of nitrifying bacteria was found. It is noted that the relative abundance of Nitrospirae was as high as 6.59%, among which 4 OTUs were detected, all affiliated to the genus Nitrospira. The predominant OTU was Candidatus Nitrospira defluvii (OTU 762), occupying 5.02%, and the other three OTUs were only responsible for 1.57%. Furthermore, another 4 OTUs affiliated to the genus Nitrosomonas were detected with a relative abundance of 1.12%.

Ammonium removal of the roughing filter was inseparably interconnected with the high abundance of nitrifying bacteria. The predominant phylum Proteobacteria in the community, containing many species of ammonia-oxidizing bacteria, is associated with nitrification and denitrification. Actually, the two genera Nitrosomonas and Denitratisoma affiliated to this phylum, were found to make up 1.12% and 5.44% (Figure S2), respectively. It is worth noting that the high abundance of Denitratisoma associated with denitrification implied that the roughing filter system might facilitate the denitrification process. Except for the frequently-reported nitrifying bacteria Nitrosomonas, the biofilm was also abundant in Nitrospira, which was much more overrepresented than Nitrosomonas. Previous researches also showed high abundance of Nitrospira in other nitrifying system, such as submerged aerated biological filters (Yue et al. 2018) and rapid sand filters (Oh et al. 2018). In addition, it was also possible that some Nitrospira might be involved in complete ammonia oxidation (Palomo et al. 2016). Interestingly, among the Nitrospira detected, 76% were found to be Candidatus Nitrospira defluvii, which is very different from other nitrite oxidizers in enzymatic repertoire and metabolic pathways, and presents almost the same gene for nitrite oxidation as anaerobic ammonium-oxidizing organisms (LüCker et al. 2010).

Moreover, high abundance of Cyanobacteria was unexpectedly observed. This might be attributed to the accumulation of sludge containing algae and the light-permeability of columns, which facilitates growth of algae, since the cleaning period was as long as 2–3 weeks in this study.

In general, a roughing filter works as a pretreatment technology, and its primary function is to control the turbidity load of subsequent treatment processes, especially for slow sand filters. Thus, apart from ammonium removal, the turbidity removal capacity of biological roughing filters was also discussed in this study, as shown in Figure 7. In the first 9 days, turbidity removal of filters increased gradually, and meanwhile the effluent turbidity showed a declining trend (Figure 7(a)). This might be attributed to biofilm formation, which was beneficial for retaining particles through bio-flocculation and absorption. After this ripening period, turbidity removal efficiencies remained at a relatively stable level, ranging from 50% to 90% with an average removal of 71.5%. The effluent turbidity was less than 3 NTU when turbidity of raw water fluctuated between 2–8 NTU.

In general, a high flow rate could reduce the filter footprint and facilitate ammonium removal, but might affect filter performance in terms of turbidity removal. In our previous study (Zeng et al. 2020), the results showed that roughing filters run at 2 m/h could achieve a turbidity removal of about 70–75%. However, at 4 m/h, a similar performance (71.5%) was obtained in this study. It seemed that the change of flow rate did not strikingly influence filtrate turbidity, and synchronous high removal of both turbidity and ammonium could be achieved, which was very important in practice. Fang et al. (Fang et al. 1998) also reported that turbidity removal efficiency at 7 m/h was almost the same as that at 4 m/h. The deeper part of the filter bed (the top layer in this study) contributed to this result (Fang et al. 1998). At small flow rates, the majority of particles were retained in the bottom layer and the deep layer seldom contributed. When flow rate increased, more particles penetrated into the deeper part of the filter bed, and were then removed from water.

Figure 7(b) demonstrates the effect of backwashing on turbidity removal. After intense backwashing, effluent turbidity increased with decrease of removal efficiency. The impaired capability for particle retention might be mainly attributed to the increase in medium porosity after cleaning, resulting from discharge of sludge and expansion of the filter bed. Moreover, turbidity removal gradually recovered in the next 5–6 days, which was several days more than that of ammonium removal (Figure 4). It is noted that although being affected, roughing filters still presented strong capability for turbidity control with effluent less than 3 NTU and removal efficiency higher than 50%.

This study evaluated the performance of roughing filters in terms of ammonium and turbidity removal. Low flow rate did affect the nitrifying process while intense backwashing only showed a slight impact. At 4 m/h, the largest NH₄⁺-N loading rate of roughing filters was about 0.1 kg/(m³·d), in order to meet the requirement of 0.5 mg/L in the effluent, but highest NH₄⁺-N removal varying between 44% to 86% was obtained with initial NH₄⁺-N concentration ranging from about 0.4 mg/L to 1.7 mg/L. Turbidity removal efficiency was about 70%. High abundance of nitrifying bacteria (Nitrospira 6.59%, Nitrosomonas 1.12%) was found, contributing to ammonium removal.

This research work was financially supported by STS Program Supporting Project of Fujian Province (2018T3102).

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