Greywater recycling for diverse collection scales and appliances: Enteric pathogen log-removal targets and treatment trains

In light of increasingly diverse greywater reuse applications, this study proposes risk-based log-removal targets (LRTs) to aid the selection of treatment trains for greywater recycling at different collection scales, including appliance-scale reuse of individual greywater streams. An epidemiology-based model was used to simulate the concentrations of prevalent and treatment-resistant reference pathogens (protozoa: Giardia and Cryptosporidium spp. , bacteria: Salmonella and Campylobacter spp., viruses: rotavirus, norovirus, adenovirus, and Coxsackievirus B5) in the greywater streams for collection scales of 5, 100-, and a 1000-people. Using quantitative microbial risk assessment (QMRA), we calculated LRTs to meet a health benchmark of 10 – 4 infections per person per year over 10 ′ 000 Monte Carlo iterations. LRTs were highest for norovirus at the 5-people scale and for adenovirus at the 100-and 1000-people scales. Example treatment trains were designed to meet the 95 % quantiles of LRTs. Treatment trains consisted of an aerated membrane bioreactor, chlorination, and, if required, UV disinfection. In most cases, rotavirus, norovirus, adenovirus and Cryptosporidium spp . determined the overall treatment train requirements. Norovirus was most often critical to dimension the chlorination (concentration × time values) and adenovirus determined the required UV dose. Smaller collection scales did not generally allow for simpler treatment trains due to the high LRTs associated with viruses, with the exception of recirculating washing machines and handwashing stations. Similarly, treating greywater sources individually resulted in lower LRTs, but the lower required LRTs nevertheless did not generally allow for simpler treatment trains. For instance, LRTs for a recirculating washing machine were around 3-log units lower compared to LRTs for indoor reuse of combined greywater (1000-people scale), but both scenarios necessitated treatment with a membrane bioreactor, chlorination and UV disinfection. However, simpler treatment trains may be feasible for small-scale and application-scale reuse if: (i) less conservative health benchmarks are used for household-based systems, considering the reduced relative importance of treated greywater in pathogen transmission in households, and (ii) higher log-removal values (LRVs) can be validated for unit processes, enabling simpler treatment trains for a larger number of appliance-scale reuse systems.


Introduction
Water scarcity poses a significant threat to sustainable human development, as it increasingly impedes the provision of sufficient quantities of safe water for human use (Huang et al., 2021).In some regions, the practicability of centralized water treatment and distributions is being challenged by rapid urbanization, aging infrastructure, and poor institutional reliability (Larsen et al., 2016).On-site water reuse can offer flexible solutions to provide water and protect freshwater resources in water-scarce areas without the need for large-scale centralized infrastructure.Treatment of greywaterwater from sources such as showers, baths or washing machinesis increasingly regarded as a promising water source due to its high volume and relatively low pollutant concentrations (Gross et al., 2015).
Recent technological advancements have expanded the potential for greywater reuse, with the emergence of technologies that enable reusing water for residential on-site applications.These on-site technologies can be designed to serve varying population sizes, from as few as two people sharing a household-scale system to several hundred users in buildingor neighbourhood-scale reuse systems (De Gisi et al., 2016).Moreover, greywater reuse is also increasingly diverse with respect to reuse applications: while treated greywater was mostly used for irrigation and toilet flushing in the past (Jefferson et al., 2004;Oron et al., 2014), there are now expanding technological options for recycling specific greywater streams in closed-loop applications.Examples include technologies that recycle shower water (Gassie and Englehardt, 2017), hand washing water (Olupot et al., 2021;Reynaert et al., 2020), or laundry water (Manouchehri and Kargari, 2017).
Current legislation around (grey-)water reuse tends to take a onesize-fits all approach, ignoring the advantages of a fit-for-purpose approach for on-site water reuse, and potentially hindering the implementation of greywater reuse systems (Reynaert et al., 2021;Van de Walle et al., 2023).An important question is how the increasing diversity of greywater reuse applications influences the required level of treatment and whether differences in treatment requirements can allow for the use of simpler treatment systems for certain applications.
Pathogens are considered the primary threat to user health in greywater reuse and should be removed through appropriate treatment (Benami et al., 2016).Quantitative microbial risk assessment (QMRA) is a structured methodology that can be applied to determine the relationships between pathogen log-removal targets (LRTs) for combinations of wastewater sources, reuse applications, and likelihood of negative health consequences (i.e., infections, illnesses, or disability-adjusted life years).This approach can be used to find the critical LRTs to ensure that the risk to human health does not exceed a specified benchmark level (WHO, 2016).A particular challenge of QMRA for on-site greywater reuse, especially when studying small collection scales or individual greywater streams, is the limited availability of measured pathogen data for such systems (Jahne et al., 2017).Obtaining reliable estimates of pathogen concentrations is difficult due to the dynamic nature of pathogen concentrations with frequent non-detects resulting from sporadic infections in small populations (Jahne et al., 2017).The smaller the scale of reuse (e.g., single household), the higher the variability in pathogen concentrations (Gross et al., 2015).
Strategies to overcome data limitations related to pathogen concentrations in greywater QMRA include the use of (i) surrogate parameters such as E. coli instead of pathogen measurements (Busgang et al., 2015), (ii) the use of pathogen measurements from municipal wastewater, for which larger datasets are available, in combination with a factor estimating the fecal load in greywater compared to municipal wastewater (Pecson et al., 2022), or (iii) the use of epidemiology-based models to simulate pathogen concentrations and occurrences based on infection or incidence rates (Jahne et al., 2017).A notable recent strategy to determine LRTs for greywater reuse is the use of Monte Carlo methods in combination with epidemiology-based modelling of pathogen concentrations as implemented by several studies (Arden et al., 2020;Schoen et al., 2017Schoen et al., , 2018Schoen et al., , 2020aSchoen et al., , 2020b)).What these studies have in common is that they typically investigate LRTs required for the treatment of combined greywater sources that are reused for broad applications such as "indoor reuse" or irrigation.In this study, we extend this framework to investigate appliance-scale reuse scenarios, in which water is repeatedly recycled for the same application, thus increasing the resolution of reuse scenarios beyond the broader categories of prior work.
LRTs obtained through QMRA can then be used to guide the selection and design of appropriate treatment technologies that achieve the necessary level of pathogen removal.Typically, unit processes are combined in treatment trains in order to provide multiple barriers, ensuring the removal of different types of pathogens.If each barrier is assumed to operate independently, the pathogen log-removal value (LRV) for each process can be summed up to obtain the overall LRV (Slob, 1994).Reference pathogens are often selected based on high infection prevalence and/or excretion data (Sharvelle et al., 2017).
When using QMRA to determine appropriate treatment trains, it is also important to consider pathogen-specific treatment efficacy, as the most commonly occurring pathogens may not necessarily be those most critical for the dimensioning of treatment trains.
This study presents enteric pathogen LRTs for the recycling of individual or combined greywater streams for three different population sizes.For these pathogen LRTs, we propose greywater treatment trains consisting of a sequence of unit processes.It is not necessarily the same pathogens that determine the overall treatment train requirements and the dimensioning of unit processes.Hence, this study provides a comprehensive assessment of several protozoan, bacterial and viral reference pathogens to identify (i) relevant reference pathogens for the selection of overall treatment trains consisting of several unit processes, and (ii) the most critical reference pathogens for the dimensioning of individual unit processes.

Quantitative microbial risk assessment for enteric pathogens
The goal of the QMRA was to systematically determine LRTs for combinations of greywater sources and reuse applications at different scales.LRTs were calculated to achieve a tolerable annual risk of infection P inf of 10 -4 infections per person using the standard QMRA equation from Schoen et al. (2017): with S the fraction of people in the exposed population susceptible to the reference pathogen, n exp the number of days of exposure over a year for a certain reuse application, DR(…) the dose-response model for the reference pathogen, V ing the daily volume of water ingested for a reuse application and C P,i the reference pathogen concentration in the greywater source randomly sampled from a yearly simulation (see Section 2.1.2).
Besides the probability of infection through routine ingestion (P inf, rout ), this study also accounted for the probability of infection due to accidental ingestion (P inf, acc ) and cross-connections of the reclaimed water with potable water (P inf, cc ).Assuming that only fractions of the population would be exposed to accidental ingestion and crossconnections, the combined probabilities of infection were then calculated based on Schoen et al. (2018): where f acc and f cc are the fractions of the population exposed to accidental ingestion (acc) and cross-connections (cc), respectively.

Reference pathogens
For each group of pathogens, we selected two reference pathogens linked to the highest number of illnesses as reported in a study by the United States Centre for Disease Control and Prevention (Scallan et al., 2011), and supported by the set of reference pathogens selected by Soller et al. (2017).For protozoa, this included Giardia spp.and Cryptosporidium spp., for bacteria Salmonella spp.and Campylobacter spp., and for viruses rotavirus and norovirus.For the design of treatment trains based on LRTs (see Section 2.2), it is also important that the reference pathogens are conservative in terms of treatment (Zhiteneva et al., 2020).In this study, unit processes included membrane bioreactor (MBR), chlorination, and UV disinfection.The determining reference pathogens for these unit processes are expected to be viruses (limited retention/inactivation in MBR and with UV disinfection) or protozoa (limited inactivation with chlorine).For protozoa, Cryptosporidium spp.
E. Reynaert et al. is a conservative reference pathogen because of its high resistance to chlorine (US EPA, 2003).For viruses, the most prevalent pathogenic viruses are not necessarily conservative in terms of treatment train design, as some less prevalent viruses are highly resistant to treatment.We thus additionally included Coxsackievirus B5, which is highly resistant to chlorination (Petterson and Stenström, 2015) and adenoviruses, which are highly resistant to UV disinfection (US EPA, 2020), as candidate reference viruses.

Pathogen concentrations in greywater
Due to the limited availability of measured pathogen concentrations for individual greywater streams, especially at small scales, we simulated pathogen concentrations following an epidemiology-based Monte Carlo approach as presented in Jahne et al. (2017).Daily pathogen concentrations were simulated over 10′000 years.The simulation consisted of three steps: 1. Simulation of total daily infections N for each reference pathogen and population size pop through random sampling from the distributions of incidence and infection duration.2. Simulation of feces concentrations in the GW stream based on random sampling from measured E. coli concentrations in individual GW streams C EC,GW (CFU⋅L -1 ) and the density of E. coli in feces C EC,F (CFU⋅wet g -1 ). 3. Simulation of pathogen concentrations in GW, C P,GW (#⋅L -1 ), using the following equation (with C P,F pathogen density in feces, in #⋅wet g -1 ), where dividing by the population size accounts for dilution effects by wastewater from non-infected individuals: The incidence, duration of excretion and the density of the reference pathogens in feces are presented in Supplementary Information (SI) Section 1.
Table 1 presents the modelled E. coli concentrations in the streams (C EC,GW ) along with the relative contributions of individual greywater sources to combined greywater.Following Jahne et al. (2017), E. coli concentrations were assumed to follow a lognormal distribution with a PERT-distribution of the mean of the lognormal distribution.These E. coli distributions were built on summary statistics from several studies.However, Sylvestre et al. (2024) showed that such parametric models built on summary statistics can significantly underestimate the variability and uncertainty in E. coli concentrations.For greywater from washing machines and showers, for which datasets of E. coli concentrations have been published (O'Toole et al., 2012), we therefore also included lognormal distributions of E. coli using the best-fit parameters presented in Sylvestre et al. (2024), which were shown to more accurately describe the distribution of E. coli.Source-separated toilet flush water represents flush water that is separated from the major part of urine and feces (urine-diverting flush toilet with solids separation).On-site systems that treat source-separate toilet flush water can for instance employ hydrocyclone separators (Reynaert et al., 2020), or rotating belts (Sahondo et al., 2020) to generate wastewaters with substantially lower fecal concentrations than blackwater.In terms of fecal contamination, source-separated toilet flush water was assumed to have a similar concentration as locally-collected mixed wastewater (see Supplementary Information (SI) Section 1).While source-separated toilet flush water is not generally considered to be greywater, it provides a boundary case with relatively high pathogen concentrations.Combined greywater was modeled as the sum of greywater provided from showering, washing machines and handwashing.

Reuse applications
This study considered the recycling of all individual greywater sources (greywater source and reuse application are the same), the recycling of combined greywater for handwashing, laundry and showering, and the downcycling of combined greywater for irrigation (Fig. 1).Note that the recycling of showering water (SH) in this study is not representative of commercially available systems that recycle water for one single user/usage.Additionally, potable reuse of combined greywater was included as an aspirational boundary condition providing context for interpreting more feasible scenarios of non-potable reuse, as well as representing a situation where users would be exposed to large quantities of reclaimed water.
Table 2 presents the ingested volumes of water for each exposure route, along with the frequency and the fraction of the population exposed.Information on the frequency of accidental exposure and fraction of the population affected were not available for most reuse scenarios.This is why scenarios were used to assess the importance of both factors.We included weekly, monthly, and yearly frequencies with  10 %, 30 %, or 50 % of the population affected.Similarly, the duration of cross-connections with potable water is highly uncertain.Here, we tested the effect if cross-connections are detected after one day, one week, or one year.

Population-dependent parameters and dose-response models
Susceptibility of the population was conservatively set to 1 for all pathogens except of rotavirus, for which the susceptibility was set to 0.06 assuming that only young children are susceptible (Pott et al., 2012).We used the following dose-reponse models: Rose et al. (1991) (exponential) for Giardia spp.; Messner and Berger (2016)

Calculation of log-removal targets (LRTs)
LRTs were calculated by numerically solving Eq. ( 1) for each scenario over 10′000 Monte Carlo simulations, i.e., simulating the microbial risk over 10′000 years.The 95 % quantiles of the 10′000 simulations were used as a conservative point estimate for the recommended LRT.

Design of treatment trains
Model treatment trains were designed to meet the recommended LRTs.Basic treatment trains consisted of two main components: an aerated membrane bioreactor (MBR) and chlorination with sodium hypochlorite.In case LRTs could not be reached with these two technologies, UV disinfection was added before the chlorination.These three unit processes are well-established for on-site applications, and the inclusion of an MBR enables the fulfillment of additional water quality requirements besides pathogen control, including the reduction of organics, suspended solids, color, and odor (Pecson et al., 2022).
The maximum LRV attributed to each technology was either the maximum LRV of the reference pathogen reported in literature (but not more than 6.0), or the LRV of a more conservative indicator organism (Table 3).The maximum LRV of 6.0 was selected to ensure that monitoring of each barrier would be feasible: for barriers with rapid performance loss, monitoring frequencies in the order of 15 s are required in order to monitor treatment performance during operation (Sylvestre et al., 2023).For MBR treatment, we accounted for a fixed LRV (LRV MBR ) independent of LRT, while chlorination and UV were dimensioned based on required concentration × time (CT)-values and UV doses, respectively.CT values for chlorination and UV doses for each pathogen can be found in SI 3. Note that there are no widely-established CT values for rota-, noro-, and adenovirus, and results presented in this study should thus be treated as estimates.
In a first step, the CT value for each pathogen was determined taking into account the maximum removal reported in literature (LRV Cl2,max ).The maximum CT value (CT max ) was selected for the dimensioning of the chlorination unit, where CT max is the maximum of the CT values needed for each pathogen to reach either the remaining LRV requirements (LRT − LRV MBR ), or the maximum LRV achievable by chlorination (LRV Cl2,max ) (whichever is lower).: Similarly, for cases where LRT > LRV MBR + LRV Cl2 for at least one pathogen, the required UV dose dose max was calculated with the following equation, where dose max is the maximum of the UV doses needed for each pathogen to reach either the remaining LRV requirements (LRT − LRV MBR − LRV Cl2 (CT max )), or the maximum LRV achievable by UV disinfection (LRV UV,max ) (whichever is lower):

Table 2
The ingested volumes (V ing ) considered accounted for routine exposure, accidental exposure and exposure through cross-connections with potable water.Numbers in italic font were used as alternative scenarios due to a high level of uncertainty for these parameters.
Overall, the treatment trains are based on conservative estimates of enteric pathogen LRVs by unit processes, in line with previously published recommendations (Pecson et al., 2022;Branch et al., 2021;Salveson et al., 2021).Note that this paper represents a simplified approach to treatment train design, with the objective of evaluating the impact of differences in LRTs on suitable treatment trains.The dimensioning of the treatment units and optimization of treatment trains as a function of the raw water quality and reclaimed water quality may shift enteric pathogen LRTs; such variations in treatment unit performance may influence design of treatment trains.The choice between closed-loop recycling or once-through recirculation may influence treatment train design or performance.For pathogen removal, the concentrations recirculated into the system after treatment are negligible compared to the loads entering the water from infected individuals, so the difference between closed-loop and once-through is negligible.However, other contaminants not removed by treatment, such as contaminants that impact water fluency or chlorine demand, are expected to build up in closed-loop systems impacting pathogen removal and acceptability (color, odor) of the reclaimed water.In such cases, further treatment, removal of concentrates, or amendment with fresh water may be required.The treatment train design assumes that the LRVs of unit processes are always met during operation: this approach does not include process redundancy to account for treatment failures, as is typically implemented in large-scale direct potable reuse facilities (Pecson et al., 2017).

Pathogen concentrations in greywater
As a first step to calculate required LRTs, we simulated reference pathogen occurrence and concentrations in the individual greywater sources and in combined greywater (Fig. 2).
Both the occurrence and the mean log 10 -concentrations of pathogens varied as a function of the population size: the smaller the population size, the lower the occurrence (due to a lower number of total infections), but the higher the mean concentration when occurring (due to less dilution by the non-infected part of the population).For all pathogens except norovirus, the occurrence in the 1000-people scenario was more than 100-fold higher than in the 5-people scenario.The difference in norovirus occurrence between the population sizes was lower due to the comparatively high incidence of norovirus compared to other pathogens (see SI 2), leading to a relatively high occurrence of 3 % even for a 5-people scenario.While the occurrence was highest for norovirus among all population sizes, mean concentrations when occurring were highest for adenovirus and rotavirus.
For all pathogens, concentrations when occurring where highest in source-separated toilet flush water [TF] and lowest in laundry greywater [WM], with differences in the mean log 10 -concentration of around 5. Combined greywater [GW], handwashing greywater [HW] and shower greywater [SH] had intermediate pathogen concentrations (in decreasing order).

Log-removal targets for recycling of greywater
LRTs were calculated for the recycling of combined greywater for all considered reuse applications, and for the recycling of individual   1).
greywater sources.Two additional scenarios were included: the reuse of combined greywater for potable consumption (as a boundary condition, not necessarily a feasible scenario), and the downcycling of combined greywater for irrigation.Three ingestion scenarios were investigated: routine ingestion, routine ingestion combined with accidental ingestion, and routine ingestion combined with ingestion due to cross-connections with potable water supplies.

Routine ingestion
The 95 % quantiles of LRTs for routine ingestion did generally not vary by more than 0.1 units between different runs of the QMRA simulation (Fig. 3).Exceptions were (1) rotavirus and adenovirus, for which 95 % quantiles of LRTs were 0 for some runs in the 5-people scenario, and (2) adenovirus for which the difference in 95 % quantiles of LRTs between different runs was up to 0.5 in the 5-people scenario.SI 4 presents the full distribution of LRTs for all scenarios.
Generally, there were only minor differences in LRTs for the 100people and 1000-people scenarios.For the 5-people scenario, the 95 % quantiles of LRTs were 0 for all pathogens except of norovirus and adenovirus, and (for some simulations) rotavirus.LRTs for these viral pathogens are attributable to their relatively high occurrence, high shedding, and high infectivity.For norovirus, the 95 % quantile of LRTs was as high for the 5-people scenario due to the high occurrence of around 2.9 % in combination with high concentrations when occurring.In contrast, the 95 % quantile of LRTs for adenovirus was lower for the 5people scenario compared to the 100-people and 1000-people scenarios.
Notably, LRTs can be impacted by the underlying assumption on the distribution of E. coli.The LRTs to simulate pathogen concentrations in laundry and shower greywater are based on the lognormal E. coli distributions from Sylvestre et al. (2024) (Table 1).LRTs using the original E. coli distribution from Jahne et al. (2017) are included in SI 5: while the LRTs for recycling of shower water are minimally impacted, recycling of laundry water required up to ~1-log higher removals when using lognormal distributions from Sylvestre et al. (2024) compared to the original distributions (Table 1).
The degree of source-separation and reuse application had a substantial influence on the LRT, with the lowest 95 % quantiles of LRTs (laundry and handwashing greywater) being around 3-log lower than Fig. 3. 95 % quantiles of log-removal targets (LRTs) and changes in LRT for the recycling of greywater at different scales for three ingestion scenarios: routine ingestion, routine ingestion with 30 % of the population accidentally ingesting water during 12 days/year, and routine ingestion with 0.1 % of the population exposed to cross-connections with potable water during 7 days/year.Reuse applications: Pot: potable reuse (boundary condition); GW: recycling of greywater for all considered non-potable applications; WM: washing machine; HW: handwashing; SH: showering; TF: toilet flushing; Irrig: downcycling of combined greywater for irrigation.Reference pathogens: Giardia: Giardia spp.; Crypto: Cryptosporidium spp.; Salm: Salmonella spp.; Campylo: Campylobacter spp.; RoV: rotavirus; NoV: norovirus; AdV: adenovirus; CVB5: Coxsackievirus B5.For WM and SH, pathogen concentrations were calculated using the E. coli distributions from Sylvestre et al. (2024).
E. Reynaert et al. the one for combined greywater.

Routine and accidental ingestion
Differences in LRTs resulting from accidental exposure were more pronounced for pathogens with higher occurrence, and thus also for larger population sizes (Fig. 3, 30 % of the population affected for 12 days/year, and SI 6 for absolute values of LRTs).In terms of reuse scenarios, differences in 95 % quantiles of LRTs were strongest for reuse for handwashing because of a low routine ingestion volume but a high volume for accidental ingestion.In contrast, there were only minor differences in LRT for shower water reuse, for which the routine ingestion volume was already comparatively high.
Despite to the high uncertainty on the fraction of the population affected by accidental exposure, the modeled percentage of the population affected did not appreciably change the required LRTs (less than 1-log difference for 10 % or 50 % of the population affected, see SI 7).Generally, the change in LRT was smaller for the 5-people scenario, due to the low frequency of accidental ingestion (compared to routine ingestion) in combination with low pathogen occurrence.Besides of the percentage of population, the frequency of the exposure is an important factor linked to high uncertainty.SI 7 presents results for alternative scenarios with weekly or yearly exposure frequencies.In contrast to the percentage of the population affected, the exposure frequency had a major impact for small population sizes.Increasing the frequency from monthly to weekly further increased the LRT by up to 2-log for all population sizes because of the higher likelihood of pathogens being present in the water.

Routine ingestion and cross-connection with potable water
The maximum difference in LRT due to cross connections was 0.9-log (Fig. 3, 0.1 % of the population affected for 7 days/year, and SI 6 for absolute values of LRTs).The change in LRT was lower for the smallerscale scenarios due to the lower occurrence of pathogens in these scenarios.Even for a highly conservative scenario in which the crossconnections are not detected during an entire year, the change in LRT was below or equal to 2, with the exceptions of irrigation and laundry washing.For these two scenarios, the increased frequency of exposure compared to routine exposure alone resulted in increased changes in LRT, with a maximum difference of 7.3 for irrigation in the 5-people scenario, where the 95 % quantile of LRTs for adenovirus was 0 without accounting cross-connections but 7.3 for cross-connections lasting 365 days/year (SI 8).
SI 6 additionally presents LRTs for combined accidental ingestion and cross-connections, highlighting that changes in LRTs compared to routine ingestion alone are primarily driven by accidental ingestion.

Comparison with log-removal targets from literature
The current study follows an epidemiology-based approach that has been used in a previous study that proposes LRTs for non-potable greywater reuse for 5-people and 1000-people scenarios (Schoen et al., 2017).Two reuse applications, namely unrestricted irrigation and indoor use, are similar to the scenarios for irrigation reuse and recycling of combined greywater modelled in the present study.Similarly, Pecson et al. (2022) presents LRTs for the same reuse applications for a 1000-people scenario.Unlike this study, Pecson et al. (2022) use pathogen measurements from municipal wastewater and assume pathogen concentrations in greywater are 1 % of wastewater.With this assumption, pathogen concentrations no longer depend on scale.Table 4 presents a comparison of LRTs from the current study and the two reference studies.The highest LRT is reported whenever multiple dose-response models were tested.SI 9 additionally presents LRVs for reuse of mixed greywater when fulling aligning the ingestion scenario with assumptions from Schoen et al. (2017).In spite of differences in the modelled greywater composition and volumes of reclaimed water ingested for the irrigation scenario, LRTs from Schoen et al. (2017) are similar to those reported in this study.In contrast, LRTs in this study are generally higher for indoor reuse due to the inclusion of showering as a reuse application.LRTs for Campylobacter spp.are considerably higher in this study, due to the use of a hypergeometric dose-response model by Teunis et al. (2018), which was published after the study by Schoen et al. (2017).The same observation of similar (irrigation) and higher (indoor reuse) LRTs also applies to the comparison with results from Pecson et al. (2022).Again, differences can likely be attributed to higher ingestion volumes for indoor reuse in this study.

Critical reference pathogens and selected treatment trains
Fig. 4 presents the technology treatment trains required for each pathogen and each reuse scenario for routine ingestion.Technology trains for the alternative ingestion scenarios are included in SI 10.Fig. 4 can be used to investigate three aspects: (1) which reference pathogens,

Table 4
Comparison of 95 % quantiles of log-removal targets (LRTs) to meet an infection benchmark of 10 -4 per person per year.The "GW" (recycling all greywater for all applications) and "Irrig" (greywater reuse for irrigation) scenarios are compared to the "indoor use" and "irrigation" scenarios from other studies.When two LRTs are presented, these correspond to the 5-people and 1000-people scenarios, respectively.n/a: not available.in combination with different dose-response models, determine treatment trains, (2) the effect of scale on the treatment trains, and (3) the effect of separating greywater sources.The primary objective of assembling treatment trains was to explore the option space, however, not all treatment trains are necessarily practical (e.g., MBR-based treatment for appliance-scale reuse for 5 people).
In most cases, the overall treatment trains were determined by several pathogens, mostly adenovirus, norovirus, rotavirus and Cryptosporidium spp.Exceptions were the reuse for irrigation scenario at a 5people scale and the reuse for handwashing scenario at 100-and 1000-people scales, where a single pathogen and dose-response model determined the overall treatment train.
Scaling down greywater reuse systems did not lead to simpler treatment trains in most scenarios.Due to the high occurrence of noroand adenovirus in all scenarios, the resulting treatment trains were similar across scales.The only exceptions were the handwashing scenario, for which a simple MBR+Cl 2 system met the LRT in the 5-people scenario for some of the model simulations (LRTs for adenovirus varied by up to 0.5-log), but not in the 100-people and 1000-people scenarios, and the recirculating washing machine, for which MBR+Cl 2 were sufficient at the 5-and 100-people scales.Notably, for handwashing, the need for UV disinfection for handwashing was driven only by risks for adenovirus at larger scales, for all other pathogens MBR+Cl 2 would have been sufficient.
Similarly, there was only limited advantage of separating greywater sources on the design of treatment trains.The separation of greywater sources allowed for simpler treatment trains only for the recycling of handwashing water designed for 5 people, and for recirculating washing machines at the 5-and 100-people scales.

Critical reference pathogens for dimensioning of chlorination and UV disinfection
For chlorination, different reference pathogens were critical for the determination of the CT value as compared to the overall treatment trains (Table 5).In the 5-people scenario, norovirus and, in some cases rotavirus, 1 were the most critical reference pathogen due to the high associated LRTs.For the 100-people and 1000-people scenarios, Giardia spp.and norovirus or rotavirus determined the required CT value.Giardia spp.determined the CT value for the applications requiring the highest LRTs (potable, reuse of greywater for all indoor applications, showering, toilet flushing).The reason for this is the high resistance of Giardia spp. to chlorine, and thus high associated CT values.Norovirus and rotavirus, which have very similar CT values, determined the CT values for the cases where Giardia spp. was completely retained by the MBR.In comparison to norovirus and rotavirus, adenovirus requires slightly lower CT values.Because of the very low concentrations in greywater, Coxsackievirus B5 never determined CT values in spite of its high resistance to chlorine.As mentioned in Section 2.2, there are no widely-established CT values for rota-, noro-and adenovirus, and these results should be interpreted as estimations (see SI 3 for values used).

Treatment trains and unit processes: norovirus and adenovirus are most critical reference pathogens
Reference pathogens used for the design of treatment trains should provide a conservative model for risk assessment in order to ensure that health targets are met.An ideal reference pathogen would be characterized by high concentrations in wastewater, high pathogenicity, and/ or poor removal during treatment (Zhiteneva et al., 2020).To determine which pathogens are most critical for the determination of treatment trains, the present study incorporated two protozoa, two bacteria and four viruses.
In most modelled reuse scenarios, there were typically several reference pathogensnoro-, and adenovirus at all scales and additionally rotavirus and Cryptosporidium spp. in the larger-scale scenariosthat defined the treatment train requirements.Consequently, the modeling of each pathogen and selection of dose response-models was of minor importance for the determination of overall treatment trains, and the modeling of conservative (i.e., treatment-resistant) reference pathogens was mostly required for the dimensioning of unit processes.There were, however, exceptions for reuse scenarios with low levels of contamination (washing machine) or low frequency of reuse (irrigation), where a single reference pathogennorovirus for the 5-people scenario or adenovirus for the 100-and 1000-people scenariosnecessitated more advanced treatment.Better understanding of health risks associated with adenovirus in larger-scale systems and norovirus in household-scale systems is therefore important.
Current concerns around the QMRA model include the lack of data on infectivity, secondary transmission, and, for adenovirus, the lack of a widely accepted dose-response model.For adenovirus, the generalized dose-response model from Teunis et al. (2016) combines information from inhalation, oral ingestion and droplet inoculation.Due to the limited data for oral ingestion, the model parameters are primarily driven by the inhalation route.We have adopted this model because it currently represents the best available information, however, LRTs may need to be updated once dose-response models specific to oral ingestion of adenoviruses become available.For the determination of treatment trains, an additional challenge is the uncertainty around (maximum) pathogen LRVs attributed to the unit processes, as these are often determined by the detection limit of the quantification method (Zhiteneva et al., 2020).Some of the technology LRVs selected in this study might therefore be on the conservative end, as actual LRVs might be higher than the maximum detectable LRV, e.g., for the removal of protozoa in MBRs (Branch et al., 2020) or maximum LRVs for viruses through chlorination.In this study, this was for instance the case for the maximum removal of noro-and adenovirus through chlorination, that were often limited to 4 or 5 LRVs due to relatively low concentrations in wastewater (e.g., Petterson and Stenström, 2015).Improved quantification methods with higher measurement sensitivity could help with more accurate selection of treatment trains.Notably, required treatment trains could be reduced to only MBR with chlorination for handwashing at larger scale if validated LRVs for rota-, noro-and adenovirus for MBR treatment or chlorination can be increased by 1-log unit, e.g., from an LRV of 5 to an LRV of 6 for chlorination.
In contrast to the definition of treatment trains, where several reference pathogens determined the overall trains, it is single reference pathogens that determine the dimensioning of chlorination and UV disinfection.The importance of more accurate norovirus and adenovirus modelling in microbial risk assessment is also reflected in the dimensioning of the unit processes, where adenovirus mostly defined UV dose requirements and norovirusalong with rotavirus, which requires a similar CT-value for 5 LRVsoften determined chlorine CT-values.In some cases, Giardia spp. was also critical for the determination of CTvalues, due to its high resistance to chlorination.However, this paper presents a simplified approach to treatment train design, where removal with chlorination is maximized before a UV disinfection unit is added.For all cases where Giardia spp. was found to determine the CT-value, a UV disinfection unit was required as part of the treatment train.An optimized treatment train design approach would allow to reduce CTvalues, as the required UV dose (determined through adenovirus) provides sufficient inactivation of Giardia spp.In this case, rota-and norovirus would again determine chlorine CT-values.While established UV dose requirements exist for adenovirus, reported CT-values for rota-and norovirus are only available for up to 4 or 5 LRVs due to limitations by their concentrations in the untreated water.As for the determination of treatment trains, improved detection methods will allow to determine more precise CT-values for higher LRVs, thereby allowing for the accurate determination of (minimum) requirements for chlorination.
This study assembles treatment trains with a primary focus on enteric pathogens, but does not further examine human health risks related to the (re)growth of engineered water system-associated pathogens such as Legionella pneumophila and Mycobacterium avium.For safe on-site water reuse, treatment trains must also be able to reliably control the (re) growth of opportunistic pathogens, especially for applications associated with aerosol generation such as showering.

Improving epidemiological models for small collection scales and individual greywater sources
The lack of measured pathogen data, especially at small collection scales, motivated the use of simulated pathogen concentrations using an epidemiology-based model.A recent monitoring campaign quantifying norovirus and adenovirus in two on-site greywater treatment systems serving 500 to 800 people allowed to validate the epidemiology-based approach for these population sizes (Jahne et al., 2020).Comparison of LRTs from the present study with results from Pecson et al. (2022), who modelled similar reuse scenarios based on measured pathogen data, shows that the 95 % quantiles of LRTs are relatively similar despite some differences in model assumptions.This comparison provides further evidence supporting the interchangeability of epidemiology-based models and measurement-based models in determining LRTs for greywater reuse at larger collection scales.
For very small systems, such as household-based systems, however, the lack of pathogen data will remain an inherent problem for QMRA: due to the low occurrence of most pathogens, it is challenging to design adequate sampling schemes that are able to capture events with occurrence of pathogens.In greywater streams with low levels of fecal contamination, such as handwashing or laundry greywater, the low concentrations of pathogens, even when occurring, are an additional challenge, as concentrations can be below measurement sensitivity limits (see also Section 4.1).For this reason, epidemiology-based models will continue to play a critical role in informing public health policies in the foreseeable future.
Future research efforts should thus prioritize improving the accuracy of underlying model assumptions, particularly those related to pathogen shedding, population infection dynamics, and spatial/temporal variability (Jahne et al., 2017).Two assumptions seem particularly important for small collection scales and individual greywater sources.First, E. coli concentrations compiled by Jahne et al. (2017), which were used to estimate the concentration of feces in greywater, are highly variable.The importance of E. coli distributions on the LRTs was illustrated through the use of alternative distributions from Sylvestre et al. (2024) for greywater from washing machines and showers, which increased LRTs by up to 1-log and had implications on appropriate treatment trains.Improved characterization of these distributions and understanding the factors contributing to the variability would help reduce the uncertainty of epidemiology-based predictions.Uncertainty in distribution parameters could also be explicitly included in the QMRA model by using second-order Monte Carlo methods.Including variability and uncertainty of pathogen concentrations is especially critical for appliance-scale reuse, where only limited data on fecal contamination generally, and on pathogen concentrations specifically, is available for individual greywater sources.It should also be noted that the implemented epidemiology-based model does not include secondary transmission from an infected individual to other people in close contact, which may limit the ability to realistically represent household-scale systems.Due to the high exposure intensities in household contexts, including dynamic modelling of secondary transmission in the epidemiology-based model could improve the accuracy of microbial risk assessment at small collection scales (Brouwer et al., 2018;Soller and Eisenberg, 2008).

Impact of non-routine exposures on the design of greywater treatment trains
In this study, we compared LRTs required if only routine ingestion is considered with LRTs including safety factors to account for accidental ingestion or cross-connections with potable water.The study shows that cross-connections, if detected within a week, had no substantial influence on the LRTs due to the low occurrence of most pathogens, and the low likelihood that contamination occurs during the period of the crossconnection.This is in line with results from Schoen et al. (2018), who found that short contamination events of potable water with reclaimed water did not increase human health risk above the selected benchmark.It can be assumed that the risk of cross-connections is further reduced in appliance-scale reuse (e.g., a recycling shower or recycling washing machine), as such systems are self-contained.
In contrast to cross-connections, including accidental ingestion in the QMRA model increased the LRTs by around 2-log, which can also impact the selection of appropriate treatment trains.Our results demonstrate that the frequency of accidental ingestion is important for the determination of LRTs.In most cases, the increase in LRTs from accidental ingestion was not linked to more complex technology trains, though the dimensioning of unit processes would be impacted.An exception was the recycling of laundry greywater at a 100-people scale, where including accidental ingestion required the implementation of UV disinfection.However, the limited effect of LRT differences on technology trains is also due to the use of conservative technology LRV estimates (see Section 4.1), which necessitated UV disinfection for most modelled scenarios.With less conservative LRV estimates, there would be more scenarios where MBR with chlorination would meet the routine ingestion LRTs, whereas UV disinfection would be required in the same scenarios when incorporating accidental ingestion into the LRT calculation.
The importance of non-routine exposure for the determination of LRTs is consistent with findings from another study: Schoen et al. (2020b) showed that refinement of assumptions for routine exposure is of limited value for uses with small routine exposure volumes, as non-routine exposure drives the human health risk in such cases.The relevance of accidental ingestion in selecting suitable treatment trains is problematic due to a scarcity of research on accidental ingestion of reclaimed water for various reuse purposes, and the generalizability of findings from such studies to other contexts is unclear.A particular challenge of assessing the effect of accidental exposure is that it will likely be context-specific, as accidental exposure will depend on user behavior and technology design.When it comes to technological advancement, it may be advisable to prioritize the implementation of preventive measures that restrict accidental ingestion, rather than incorporating safety factors accounting for accidental ingestion in the treatment trains.This may be particularly promising for small-scale systems, where users may be better aware of the source and treatment of the water and may take greater personal responsibility (Gross et al., 2015).An example of a water reuse technology that employs constructive methods to restrict inadvertent ingestion is a recycling handwashing station, in which the tap and the sink are positioned sufficiently close together to limit filling of containers or direct drinking, limiting the likelihood of misuse of the system (Sutherland et al., 2021a).Such approaches would allow the design of treatment trains with broader applicability, reducing the requirement to consider user behavior specific to each context.

Benefits and disadvantages of small-scale and appliance-scale reuse
In household-scale systems, the occurrence of pathogens is sporadic but concentrations can be high, whereas the occurrence of pathogens becomes more frequent but concentrations decrease in larger-scale systems.These differences in pathogen dynamics result in different LRTs depending on the population size: scaling down greywater reuse to household scale reduced the 95 % quantiles of LRTs to 0 for all pathogens except of noro-and adenovirus, implying that no removal of these pathogens is necessary to meet the health benchmark.For practical applications, an important question is how these differences in LRTs caused by different collection scales affect the required treatment trains.Due to the high virus LRTs, smaller systems mostly required the same treatment trains as larger systems.The only exception where reducing the collection scale allowed for simpler treatment was for recycling of handwashing water.
From a health-risk perspective, small-scale systems may still offer advantages despite the need for similar treatment trains.One advantage is a reduced relative importance of greywater reuse for the transmission of pathogens in small-scale systems: In a household context, there are many routes of disease transmission besides use of recycled water (e.g., hands, fomites, food preparation).In contrast, the likelihood of the recycled water being the vehicle for pathogen transmission increases with increasing system size (Gross et al., 2015).The selected infection benchmark might thus be unnecessarily conservative for household-scale systems, and the use of less strict benchmarks for single-household systems could be considered (LRTs and treatment trains for a health benchmark of 10 -2 pppy included in SI 11).
In contrast to reducing the collection scale, the separation of greywater sources led to substantial differences in LRTs for rotavirus, norovirus and adenovirus.However, these differences in LRTs did often not result in simpler treatment trains, with the exception of recycling of laundry and handwashing water at small scales.As discussed in Sections 4.1 and 4.3, if higher LRVs could be validated for the unit processes, there would be more cases where simpler technologies can be used for systems treating greywater sources with low fecal contamination.
Though the most critical for the protection of human health, pathogen removal is only one of the considerations for the design of treatment trains and the assessment of scale effects.User acceptance and support, which are critical for the success of on-site water reuse systems, will depend on multiple contextual factors, including the technological configurations and implementation characteristics, such as the scale (Contzen et al., 2023).Field testing of on-site reuse technologies particularly underscored the need for users to understand how these technologies function (Sutherland et al., 2021b) and highlighted the role of real-time monitoring strategies in fostering user trust in the quality of the reclaimed water (Sutherland et al., 2021a).Treatment costs and environmental impacts are two other important aspects that are not covered in this study.Both aspects are highly dependent on the specific scenario, e.g., whether the larger-scale greywater reuse systems would collect water from individual households or from single buildings.Generally, larger-scale decentralized greywater reuse systems have lower environmental impacts compared to household-scale systems (Opher et al., 2019).The need for larger greywater collection systems are probably not sufficient to cancel the advantages of economies of scale in larger-scale systems (Kobayashi et al., 2020).But appliance-scale systems might become more competitive if simpler treatment trains can be used.

Conclusions
• In light of increasingly diverse greywater reuse applications, this study proposes LRTs to aid selection of treatment trains for recycling of greywater at different scales, including appliance-scale reuse of individual greywater streams.• The study reveals that, generally, several reference pathogens determine the overall treatment train requirements, suggesting that the specific selection of reference pathogen and dose-response models is of less importance.In contrast, the study highlights the importance of including treatment-resistant reference pathogens, such as adenovirus for UV disinfection, for the dimensioning of unit treatment processes.
• Separating greywater streams down to appliance-scale reuse significantly reduced required LRTs for applications associated with low levels of contamination, but mostly did not allow for simpler treatment trains considering current validated LRVs for unit treatment processes.• Similarly, reducing the collection scale (from 1000 to 100 or 5 people) generally did not allow for simpler treatment trains due to high LRTs associated with noro-and adenovirus.• Small-scale and application-scale reuse may still allow for simpler treatment trains if (i) less conservative health benchmarks are used for household-based systems due to a reduced relative importance of treated greywater in pathogen transmission, and (ii) higher LRVs can be validated for unit processes, allowing for simpler treatment trains for a larger number of appliance-scale reuse systems.

Declaration of competing interest
The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Timothy R. Julian reports financial support was provided by Eawag Swiss Federal Institute of Aquatic Science and Technology.Eberhard Morgenroth reports financial support was provided by ETH Zurich.Eberhard Morgenroth reports financial support was provided by Eawag Swiss Federal Institute of Aquatic Science and Technology.Eberhard Morgenroth is Editor in Chief of Water Research.If there are other authors, they declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
of recycled water with drinking water.E.Reynaert et al.

Fig. 2 .
Fig. 2. Boxplot distribution of simulated pathogen concentrations in greywater from different sources at three different scales (all concentrations in log 10 #/L).Number above each boxplot: occurrence.Number below: median concentration when occurring.GW: combined greywater.WM: washing machine.HW: handwashing.SH: showering.TF: toilet flushing (source-separated).CFU: colony-forming units.TCD 50 : 50 % tissue culture infective dose.Whiskers represent 1.5 times the interquartile range and black dots are simulated concentrations beyond that range.For WM and SH, pathogen concentrations were calculated using the E. coli distributions from Sylvestre et al. (2024) (Table1).

Table 1
Modelled E. coli concentrations in individual greywater and relative contribution to combined greywater.Greywater reuse scenarios as presented in Fig. 1.
c Based on per person use from DeOreo et al. (2016) (40 L per person per day, data from the US).

Table 3
Log-removal/inactivation of reference pathogens (when meaningful data available) or more conservative indicator organisms in membrane bioreactor, UV disinfection and chlorination.
a Reported LRVs from full-scale MBR systems, based on membranes with pore sizes ≤ 0.4 µm.bLRVs are limited by pathogen concentrations in the wastewater, and so may be conservative.cCampylobacterspp.more sensitive to chlorine than E. coli.E.Reynaert et al.