Field demonstration of UV-LED disinfection at small and decentralized water facilities

Ultraviolet (UV) irradiation is a practical and efficient water disinfection method. UV disinfection is effective against diverse microbial species, including viruses, bacteria, and chlorine-resistant zoonotic parasites of Cryptosporidium and Giardia (United States Environmental Protection Agency 2006). UV light-emitting diodes (UV-LED) with germicidal emissions possess many unique properties, such as compactness, physical robustness, mercury-free components, quick start-up, and long lifetimes (Würtele et al. 2011). With these properties, UV-LED can be a good option for disinfecting water at small and decentralized water facilities in remote locations without easily accessible public water supplies, in which simple and small-footprint technologies are generally preferable. These water facilities include community-based water supply systems owned, operated, and maintained by the local residents. Such community-based systems are in practice worldwide (World Health Organization 2012), and the treatment installation is often based on a simple procedure of disinfection (Scheili et al. 2016; Deane & Mac Domhnaill 2021). Facilities in remote locations must source locally available water resources, which can be at high risk of fecal contamination from wild animals. Thus, appropriate disinfection practices are very important for community systems in both technical validity and operational adequacy.

At small facilities, chlorination is the most traditional and common practice to disinfect water. Many efforts have been made to promote chlorination in remote locations, including a challenge to generate sodium hypochlorite on-site with a solar-driven system (Chinello et al. 2019). However, chlorination may not be enough to secure public health if the source water is contaminated with chlorine-resistant zoonotic parasites. Also, chlorination has the concern of low public acceptance in some communities due to the concerns on the taste and odor as well as the need for refill and proper storage of the chemical. In such cases, an alternative could be UV disinfection offering high inactivation efficacy of chlorine-resistant parasites, no impact on odor, and no chemical requirement. Considering the drawback of UV providing no residual effects and thus being potentially vulnerable to bacterial regrowth afterward (Lund & Ormerod 1995; Sun et al. 2012), combining UV and chlorine as dual barriers against microbial contamination could be a better solution if local settings and resources allow. Potential repair of UV-inactivated microorganisms, namely photoreactivation and dark repair, would also be suppressed in such a dual barrier system (Quek et al. 2006).

Since the emergence of germicidal UV technology, many efforts have been made to characterize the UV sensitivity of various health-related microorganisms in water, using low- and medium-pressure mercury UV lamps (Hijnen et al. 2006) followed by the use of UV-LED with bench-top batch mode experimental setups (Würtele et al. 2011; Beck et al. 2017; Rattanakul & Oguma 2018; Pousty et al. 2021). To follow such bench-scale studies, UV-LED modules have been designed to treat water flow, and the performance of such flow-through modules has been reported with pure cultured challenge microorganisms at the laboratory scale (Keshavarzfathy et al. 2021; Wang et al. 2021; Romero-Martínez et al. 2022; Montazeri & Taghipour 2023). However, evaluating such flow-through modules in real fields is still very limited despite the necessity in understanding their performance against wild-type indigenous microorganisms and identifying the challenges that may arise in their use for certain periods of time.

A pioneering demonstration study in the United States evaluated a UV-LED module at a small water treatment plant in Colorado sourcing water from a creek (Hull et al. 2019). The treatment plant adopted slow sand filtration followed by chlorination before supply, and the UV-LED module was installed for demonstration at the outlet of a slow sand filter before chlorination at 0.5 L/min. The module's performance was monitored for one year, showing the equivalent disinfection efficacy to the chlorination system in practice at the treatment plant. Another field-testing project at a private water facility in Japan evaluated the effectiveness of two UV-LED modules of different designs at 2 and 10 L/min in treating a natural mountain stream without any pretreatment (Oguma & Watanabe 2020). The results showed the high efficacy and stability of both modules against indigenous indicator microorganisms throughout one year without the formation of a visible scale inside the module, supporting the usability of UV-LED modules in a decentralized manner at the point-of-use (POU) or point-of-entry (POE), as was envisioned for a new model for water distribution systems (Linden et al. 2019).

As obvious in these demonstration studies, UV-LEDs have been considered mostly for POU and POE applications at low flow rates to date. However, deliberating the recent and projected progress in UV-LED technologies (Kneissl et al. 2019) and the emergence of commercial products designed for decentralized water systems, demonstration studies on UV-LED water treatment must be updated with currently available modules targeting higher flow rates that may serve for larger populations than POU and POE, such as community water supplies. Another missing component of UV-LED demonstration to date is the evaluation of the same module in multiple locations with different source water matrices to understand the consistency of disinfection performance and reveal the causes for the inconsistency, if any. Such a multi-location field test has never been reported, although it is inevitable to promote the implementation of UV-LED technology in many locations.

In this context, the objective of this study was to run three field-testing projects using the same UV-LED module model in remote locations in Japan. The three test sites were a community-based water supply system for residents, a water supply station for visitors at a quasi-national park in Japan, and a source water intake point for a water treatment plant. At each test site, indigenous microbial parameters, including Escherichia coli (E. coli), total coliforms, standard plate count (SPC) bacteria, and heterotrophic plate count (HPC) bacteria, were monitored before and after UV-LED disinfection for 1–2 years. The source water quality was also monitored in terms of physicochemical parameters, including UV transmittance (UVT) and turbidity, to understand the natural fluctuations of such parameters and their impacts on inactivation efficiency. Given the context, this study aims to evaluate the effectiveness of UV-LED technology as a reliable disinfection option for decentralized water systems and to support the establishment of a safe water supply in remote locations.

At each test site, samples were collected at two points, the inlet and outlet of the UV-LED module. A total of three samples were obtained, namely the source water at the inlet (hereafter referred to as ‘source’), treated water at the outlet with UV-LED turned on (‘UV-LED on’), and control sample at the outlet with UV-LED turned off (‘UV-LED off’). Sampling was conducted twice a month every 2 weeks, as far as the situation allowed.

As microbial parameters, indigenous E. coli, total coliforms, SPC bacteria, and HPC bacteria were monitored in all sets of three samples. Here, E. coli and SPC bacteria were targeted as the items in the Drinking Water Quality Standard in Japan (Ministry of Health, Labour and Welfare of Japan (MHLW) 2015), HPC bacteria were targeted as a complementary item of drinking water quality in Japan (MHLW 2015), and total coliforms were tested following the guidelines by the World Health Organization (2022). The concentrations of E. coli and total coliforms were determined using the dual-chromogen membrane filter technique (Standard Methods for the Examination of Water and Wastewater 2017) to obtain colony-forming units (CFU) per 100 mL of the sample. SPC and HPC bacteria were evaluated using plate count agar (24 h at 36 ± 1 °C) and R2A agar (7 days at 20 ± 1 °C), respectively, to determine CFU per 1 mL.

The statistical analysis software R version 4.3.0 was used for all statistical analyses. Spearman's rank correlation coefficient (ρ) was calculated to examine the correlation between two parameters. One-way analysis of variance (ANOVA) with the Bonferroni post hoc test was used to compare the disinfection efficiency among the three test sites and the changes in efficiency over time at each site.

Table 2 shows a summary of the source water quality at Sites S, A, and F. Overall, microbial parameters showed high fluctuations at all test sites, with F being the most contaminated site among the three. The catchment area of the source river for Site F runs through mountain valleys where wild animals such as deer, bears, and foxes are occasionally observed, implying a high possibility of microbial contamination from animal feces. Considering the compliance with the Drinking Water Quality Standards in Japan, which require no detection of E. coli in 100 mL and 100 CFU/mL of SPC bacteria (MHLW 2015), none of the three source waters was directly potable without treatment.

For the physicochemical parameters, the source waters at Sites S and A readily complied with the Drinking Water Quality Standards in Japan (MHLW 2015) regarding the tested parameters, except for the turbidity exceedance (>2°) at Site S three times. Unlike Sites S and A, Site F was not a treatment facility but the source water intake point for a municipal water treatment plant adopting rapid sand filtration followed by chlorination; thus, its relatively low source water quality is acceptable in practice. Site F is also unique out of the three sites in terms of climate, belonging to the subarctic and humid continental climate zone in Köppen's climate classification, and is registered as a special snowy area in Japan. This was notable at the lower water temperature at Site F.

The log inactivation of bacteria at all test sites is summarized in Table 3. For those entirely or mostly resulting in the estimated values due to negative detection in treated water, the statistical summary is inappropriate, and the maximum observed or estimated values are shown instead.

The Site S facility is located in a mountainous area where the public water supply is not accessible, and the local residents have been operating the community water supply under their responsibility and efforts with technical advice from local health authorities. At this site, the source water was frequently positive for E. coli (73%, 24 out of 33 samples), whereas the treated water was negative for E. coli throughout the test period, except for a single detection at 0.5 CFU/100 mL (limit of detection as a mean of duplicate measurements) in two years, which corresponded to the maximum observed E. coli inactivation of 2.7 log at Site S (Table 3).

From September to October 2020, total coliforms in the source water were notably high, while E. coli was low, probably because total coliforms can include non-fecal bacteria such as Klebsiella spp. which are ubiquitous in nature (Gerba 2015). However, total coliforms were similar to E. coli in the sense that the positive ratio and the concentration decreased notably after UV-LED treatment (94% positive in the source and 27% positive in treated water with a maximum concentration of 2 CFU/100 mL). The source was almost always positive for SPC and 100% positive for HPC at high concentrations, commonly beyond the interim target value of HPC as a complementary item of drinking water quality in Japan (2,000 CFU/mL) (MHLW 2015), which decreased to 685 CFU/mL or lower after the treatment.

In the source water, turbidity and color showed a significant positive correlation (ρ = 0.864, p < 0.001), while neither was correlated with UVT at Site S. In general, color and UVT can be negatively correlated, as was observed at Sites A and F, as detailed below, which seems reasonable because photon transmission in water is impaired in colored water due to absorption. Presumably, the major component causing the color of water might be unique at site S. For example, natural organic matter (NOM), including humic acid, is known as the typical dissolved component to increase water color and impair UVT (United States Environmental Protection Agency 2006), but NOM might not be the main cause of color at Site S. None of the physicochemical parameters showed a significant correlation or a similar trend with microbial parameters in the source, implying the difficulty of predicting the level of microbial contamination in the source water based on such easy-to-monitor parameters in field implementation. The relationship between UVT and disinfection efficiency is discussed below in the integrated discussion.

Simplicity in operation and maintenance is critical for residents, and the UV-LED module was proven effective without laborious procedures such as cleaning and LED replacement for the entire two years of this study. Such information and all monitoring data were shared with community residents and the local government, which served as the scientific evidence to support the effectiveness and stability of the module for the community water supply. With this and all other efforts by stakeholders, the module has been implemented in three community water supply systems in Shizuoka City, together with the chlorination for the residual chlorine, to ensure the microbial stability of water in the storage tank and supply pipes.

Site A is located in a deep forest in a quasi-national park, and the water has been supplied only for toilet flushing without disinfection. Recently, the local government decided to upgrade the water treatment system to supply potable water for visitors to the park to encourage tourism. Our study revealed that the source water at Site A readily met the Drinking Water Quality Standard in Japan (MHLW 2015) in the physicochemical parameters evaluated (Table 2), and the total organic carbon of 0.50 ± 0.28 mg/L was well below the standard value of 3 mg/L. This implies that the fundamental concern for potable use is microbial contamination at Site A. The UVT at 280 nm was consistently high throughout the one-year test period, suggesting that UV disinfection is suitable for this facility.

The source water was 96% positive for E. coli, with one negative out of 26 samples, whereas the positive ratio decreased to 42% at a maximum concentration of 4 CFU/100 mL in the treated water. Total coliforms decreased, but the positive ratio remained high at 62% after treatment, with the maximum concentration in treated water being 120 CFU/100 mL. Among the physicochemical parameters in the source water, turbidity and color were not correlated at Site A, nor between turbidity and UVT, probably because turbidity was below the limit of detection (<0.2°) on occasion, whereas color and UVT showed a significant negative correlation (ρ = –0.792, p < 0.001). Physicochemical and microbial parameters at the source were not significantly correlated for any combination.

All monitoring data were shared with the local government, which eventually decided to implement the UV-LED system as the primary disinfection method at the facility, together with residual chlorine injection as the secondary disinfectant. This dual barrier approach will allow the facility to supply potable water to visitors at the park, as was initially intended by the local government. As such, a combination of UV and chlorine is a feasible strategy for ensuring the microbial safety of water in remote locations.

Site F is located in nature with a considerable influence by wild animal activities. The source water was 100% positive for E. coli, with the positivity ratio decreasing to 13% with a maximum concentration of 3 CFU/100 mL in treated water. Source water contained high total coliforms, roughly one order of magnitude higher than the other two sites, which were effectively inactivated with the maximum observed inactivation of 3.6 log (Table 3) when the turbidity, color, and UVT of the source water were within the typical ranges of 1.5°, 3°, and 81.1%, respectively. At Site F, turbidity and color showed a significant positive correlation (ρ = 0.893, p < 0.001) while a significant negative correlation was found between turbidity and UVT (ρ = 0.864, p < 0.001) as well as color and UVT (ρ = –0.929, p < 0.001). In contrast, physicochemical and microbial parameters in the source did not show a significant correlation in any combination, as was the case at Sites S and A.

Considering the potential risk of zoonotic disease through water, monthly measurements of pathogenic protozoa, including Cryptosporidium and Giardia, were conducted in 10 L of source water for the entire test period at Site F. The samples were filtered through a hydrophilic polytetrafluoroethylene (PTFE) membrane to obtain the residues, which were then collected, placed on a well slide, and stained with FITC-labelled agents to count the number of (oo)cysts under fluorescent microscopy (Standard Methods for the Examination of Water and Wastewater 2017). The protozoan species was negative from the beginning of the demonstration for 11 months (March 2022–January 2023), whereas one cyst in 10 L of Giardia was detected continuously in February and March 2023. This might be partially due to snowmelt, which may have caused the flushing out of fecal deposits of wild animals into the river, although E. coli and total coliforms were not necessarily high in those Giardia-positive samples. The seasonality of protozoan contamination should be examined further in future studies.

UV disinfection is recommended as one of the effective measures against Cryptosporidium and Giardia in Japan (MHLW 2019), based on the scientific evidence supporting UV as a promising disinfection method against those chlorine-resistant zoonotic parasites (United States Environmental Protection Agency 2006). In this context, Site F would benefit from adding UV disinfection to the treatment process as a secondary barrier against parasites, which so far has been well ensured by the rapid sand filtration system at the treatment plant with stringent monitoring of turbidity at the outlet of the filtration pond.

One-way ANOVA with the Bonferroni post hoc test indicated that the HPC inactivation efficiency was statistically similar (p > 0.01) at all three test sites. This may be partially due to the high fluctuation in the inactivation data at each site, which overwhelmed the site-specific inactivation trends. Since HPC bacteria include diverse bacterial species that can grow under nutrient-poor conditions at moderate temperatures, their members should differ on different sampling days, even at the same test site. This may have resulted in high fluctuations in HPC responses to the UV-LED treatment.

With flow-through UV modules, the flow rate affects the inactivation efficiency as it governs the hydraulic retention time inside the module, thus altering the time for UV exposure. A higher flow rate gives shorter exposure time, thus delivering lower UV fluence (dose) in general. However, flow rates were inconsistent in this study at three test sites, while the inactivation efficiency was statistically similar. One possible reason for this discrepancy is the diversity of member species in HPC bacteria, as the nature of this bacterial indicator. Considering that sensitivity to UV differs significantly among different species of bacteria (Hijnen et al. 2006; Masjoudi et al. 2021), HPC was not sensitive enough to highlight the impact of flow rates on the inactivation efficiencies. Another possible reason is specifically for Site F, where the flow rate fluctuated much lower than Sites S and A, while UVT was rather lower than the others. Those positive (low flow rate) and negative (low UVT) factors for UV inactivation might have canceled with each other.

Changes in disinfection efficiency with operational time are of concern for practical applications. For this purpose, the full test period at each test site was divided into three defined terms (shown by arrows in Figure 6), as interrupted by two shutdowns at Site S, and as three equal parts for the full periods at Sites A and F. Namely, the three terms were August 2020–January 2021 (n = 10), June 2021–February 2022 (n = 18), and June–July 2022 (n = 5) at Site S; July–October 2021 (n = 8), November 2021–February 2022 (n = 8), and March–June 2022 (n = 8) at Site A; and March–July 2022 (n = 8), August–November 2022 (n = 8), and December 2022–March 2023 (n = 8) at Site F. A comparison of the three terms at each test site indicated that the disinfection efficiency of HPC bacteria was not statistically different for the three continuous terms at Sites A (p > 0.01) and F (p > 0.01). That is, a significant performance decline over time was not observed at Sites A and F.

In contrast, at Site S, HPC inactivation in the first term was significantly higher than that in the subsequent second and third terms (p < 0.01), implying a decrease in disinfection efficiency after the first shutdown from February to May 2021. The difference between the second and third terms was insignificant at Site S. A simple interpretation of the observed deterioration is the expected output power decline of the UV-LED over time. It is to be reminded that UV-LED remained on even during shutdown periods while water flow was stopped. The nominal lifetime of the LED package in the tested module was 35,000 h to show a 30% decrease in the output power; assuming a linear decline with time, the output power would decrease by approximately 6.2% in 10 months, which is roughly the beginning of the second term at Site S after the first shutdown (June 2021). For comparison, Hull et al. (2019) reported a 27% decrease in output after one year of continuous operation in the field and some laboratory testing periods regarding their UV-LED module. Such a time-dependent output decrease may have partially contributed to the observed decline in disinfection efficiency, but cannot explain why this was only observed at Site S. Therefore, a more rational explanation would be the impact of a shutdown that occurred exclusively at Site S. As mentioned earlier, the UV-LEDs remained on during the shutdowns to avoid microbial regrowth and biofilm formation inside the module, but this measure was likely insufficient in fully sustaining the module integrity with stagnant water. The second shutdown was shorter than the first one and the subsequent sampling number was limited (n = 5), which might be the reasons for not observing a significant decline after the second shutdown. Alternatively, or possibly together with the reasons mentioned above, any unknown site-specific factors, such as the unique composition of source water, might have caused the performance decline exclusively at Site S. Thinking of the no correlation between color and UVT only at Site S, more detailed analysis and characterization of source water quality may reveal the uniqueness of Site S. Future studies should further investigate the mechanisms of the performance deterioration in use.

Contrary to expectations, no correlation was observed between UVT and HPC inactivation at any site. For example, at Site A, UVT was consistently high, with a median value of 97.6% and a narrow standard deviation of 1.1%, while the HPC inactivation varied widely in the range of 0.08–2.9 log. At Site F, the minimum UVT was as low as 49.8%, but this sample resulted in 0.9 log inactivation of HPC, which was the median value of observed inactivation at this site. UVT alone would not be a useful benchmark for predicting UV disinfection efficiency under the conditions adopted in this study. These results may be because, as previously noted, HPC bacteria are a mixture of heterotrophic bacteria and the member species can differ on different sampling days. By targeting indigenous microbial species, as was strategically designed in this study, determining the disinfection efficiency with a specific species (e.g., E. coli) is almost impossible, as its concentration is generally low in drinking water sources to demonstrate log-based inactivation. If a spike test with pure cultures is accepted and approved, a more scientific comparison of the performance at different test sites would be possible. In the study by Hull et al. (2019), coliphage MS2 was spiked in the system to conduct quarterly challenge tests in the field at various UVTs. The results showed the trend of lower MS2 inactivation at lower UVT, which seems theoretically reasonable, while the range of UVT challenged (86.7–95.5%) was narrower than that observed in this study.

Coliphages were not examined in this study, while viral contamination of drinking water is of concern. A PCR-based microbial source tracking revealed contamination by animal-specific viruses even very upstream of a river in the forest without apparent human activities (Malla et al. 2019), implying the importance of testing viruses in such remote locations. Viruses are to be tested in future demonstration studies. Other challenges for future studies include deriving a mechanistic understanding of performance deterioration in long-term operation, identifying key factors causing fluctuations in disinfection efficiencies, and developing a rational or empirical approach to predict disinfection efficiency based on water quality. Such knowledge and expertise would support the strategic implementation, operation, and maintenance of UV-LED disinfection in small and decentralized water systems, eventually delivering benefits to people relying on such water systems.

A flow-through UV-LED disinfection module was evaluated at three water treatment facilities in remote locations (Sites S, A, and F). The disinfection efficiency of the module was monitored for 1–2 years using indigenous E. coli, total coliforms, SPC bacteria, and HPC bacteria. Physicochemical parameters, including UVT, were also monitored in the source water.

The UV-LED module showed efficient disinfection at all locations, achieving E. coli inactivation at 2.7 log, over 2.3 log, and over 1.9 log as the maximum value during test periods at Sites S, A, and F, respectively. Statistical analysis revealed that disinfection efficiency, as evaluated by HPC inactivation, did not significantly differ among the three test sites (p > 0.01). For the changes in efficiency over time, Site S experienced a significant decrease in disinfection efficiency after 10 months of operation (p < 0.01), whereas the other two sites did not show performance deterioration after one year. The observed HPC inactivation was not significantly correlated with the source water UVT at any location, implying the difficulty in using UVT as a single parameter to predict disinfection efficiency in practical applications. To the best of my knowledge, this is the first study reporting the long-term field demonstration of the same model of UV-LED module in multiple locations in a comparable manner, which enabled the study to confirm the consistency of the module's performance among three test sites with regard to the HPC inactivation. Meanwhile, an inconsistency was observed in the performance deterioration behaviors over time, and the mechanisms of the deterioration observed exclusively at Site S need to be investigated in future studies. These findings would be relevant to the stakeholders of decentralized water systems including community residents, local government, and health authorities, and also to water treatment industries including water system engineers and UV-LED module manufacturers. To conclude, this study revealed that UV-LED technology is a practical and effective option for disinfecting water in small and decentralized water supply systems to improve access to safe water in remote locations.

This study was supported by the Japan Society for the Promotion of Science (JSPS) Grants-in-Aid for Scientific Research (A) (20H00257), the Japan Science and Technology Agency (JST) in the EIG CONCERT-Japan program, the Ministry of Health, Labour and Welfare (MHLW) Scientific Research Program, and the Ministry of the Environment (MOE) in the program for the implementation of innovative infection control. The author profoundly appreciates the local governments of Shizuoka City, Aya Town and Furano City as well as the community residents in Shizuoka City for their commitment and continuous support of the field-test projects. The author is also grateful to Dr Shinya Watanabe and his team at Nikkiso Co. Ltd for their technical support.

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

The authors declare there is no conflict.