Particulate contaminants and treatment decision-making: maximizing the value of raw water pathogen monitoring for drinking water safety Open Access

Over the past decades, contaminants of emerging concern in drinking water – such as per- and polyfluorinated substances, microplastics, and cyanotoxins – have received international focus, while revisiting regulatory frameworks for managing microbial contamination has received little attention. Nonetheless, management of acute risks caused by waterborne pathogens is still the paramount objective of drinking water treatment. Critically, little guidance exists to determine whether treatment requirements prescribed in drinking water regulations are sufficient or how to evaluate additional treatment needs. For example, growing evidence underscores that turbidity monitoring is not enough to ensure the sufficiency of treatment in all systems (Ballantyne et al. 2024). Moreover, risks from microbial and other particulate contaminants are exacerbated in a changing climate. Extreme rainfall events increase erosion and particulate contaminant runoff from the landscape (Emelko et al. 2024); they are recognized as a factor leading to fecal pollution, increased concentrations of microbial contaminants, and even outbreaks of waterborne disease (Charron et al. 2004; Wu et al. 2016; Graydon et al. 2022). Treatment disturbances caused by other climate-related events like wildfires can also affect pathogen and other particulate contaminant removal in drinking water treatment by affecting coagulation (Emelko et al. 2011). Additionally, climate change undermines assumptions of stationarity (i.e., the idea that natural systems fluctuate within fixed patterns of variability) that have historically facilitated the management of water demand, supply, and risk (Milly et al. 2008). Thus, landscape disturbances such as wildfires and floods (Emelko et al. 2024) as well as changes in water use such as widespread implementation of de facto potable reuse (Rice et al. 2015) also contribute to unpredictability in source water quality. Accordingly, the incorporation of recent contaminant concentration data into treatment decisions becomes more critical in a more variable and/or non-stationary context, which emphasizes the need to critically revisit common practices to enhance particulate contaminant risk management in the provision of safe drinking water. While health endpoints for particulate contaminants such as microplastics are still evolving and risk management approaches for treatment (such as the ‘Threshold Microplastics Concentration’) are limited by data availability (Chowdhury et al. 2024), approaches for management of acute risks from pathogens in drinking water are well established.

To control microbial risks in drinking water, the United States Environmental Protection Agency implemented a series of regulations starting in the late eighties (U.S. EPA 1989, 1998, 2002). These regulations culminated in the Long Term 2 Enhanced Surface Water Treatment Rule (LT2ESWTR) in 2006 (U.S. EPA 2006a), which inspired regulatory guidance in several other jurisdictions worldwide (e.g., NHMRC & NRMMC 2011; Brasil 2021). Filtration (or equivalent technology) and disinfection are the two treatment processes usually required to control risks from pathogens in systems depending on or influenced by surface water. In the LT2ESWTR and subsequent international analogues, a treatment-technique-based approach that awards treatment credits for ‘well-operated’ technologies is used to ensure sufficient treatment performance. For example, a minimum 3-log treatment credit for Cryptosporidium removal is given to conventional filtration systems whose 95th percentile of combined filter effluent turbidity is below 0.3 Nephelometric Turbidity Unit (NTU). Besides this treatment-based approach, monitoring of pathogen concentrations can be used to assess treatment needs more accurately than using generic indicators such as filter effluent turbidity. Nevertheless, monitoring is not a typical regulatory requirement for informing water treatment decisions.

To establish treatment requirements under the LT2ESWTR, the United States mandated two national monitoring rounds since 2006, and utilities were classified in bins requiring specific levels of treatment based on their Cryptosporidium oocyst concentrations. Notably, this implicitly assumes that risks from Giardia are appropriately controlled if sufficient treatment is provided for Cryptosporidium. In the Netherlands, water utilities must monitor source water and conduct a quantitative microbial risk assessment (QMRA) every three years for enterovirus, Campylobacter, Cryptosporidium, and Giardia to ensure sufficient microbial water quality (Schijven et al. 2011). In Canada, provinces and territories establish their own regulatory requirements, and these may be informed by national guidelines for drinking water quality. Health Canada guidelines recommend a minimum 3 log of removal and/or inactivation of Cryptosporidium and Giardia as well as use of QMRA to determine the level of treatment needed to meet health-based targets (Health Canada 2019); however, practical advice on how to collect and analyze monitoring data for this purpose is lacking. In general, little guidance exists to determine whether a 3-log minimum treatment requirement is sufficient or how to evaluate additional treatment needs.

This study investigates the integration of QMRA and monitoring data into drinking water treatment decisions using the monitoring experience of a water utility in Canada as an example. The City of Calgary has conducted routine monitoring for protozoa following provincial requirements since 1999. In 2010, an agreement with the province of Alberta specified a set of rules for protozoa treatment requirements and associated monitoring as well as measuring and reporting of raw water protozoa concentration values. These rules have a bin classification approach (Alberta 2012) based on the LT2ESWTR for Cryptosporidium and apply a similar approach for Giardia using limits based on a study summarized in U.S. EPA (1991). A more contemporary risk-based approach to managing protozoan risks in drinking water, such as that reflected in the Guidelines for Canadian Drinking Water Quality for enteric protozoa (Health Canada 2019), became attractive for evaluation of treatment needs, especially during winter months when disinfection credits for Giardia declined due to lower water temperatures. Practices to monitor and control protozoan risk in Calgary were reviewed to develop an implementation plan that merges treatment criteria from the Health Canada guidelines into the existing regulatory approaches. The framework applied by the City of Calgary since 2021 is presented here to support the implementation of monitoring programs in drinking water treatment plants and offer additional recommendations to systems that apply QMRA as a risk management approach in water treatment. While protozoa were the focus of the Calgary framework, statistical tools reflecting their discrete nature (Emelko et al. 2010) are the same that would be applied to other particulate contaminants in water such as other pathogens and microplastics (Chowdhury et al. 2024; Zhu et al. 2024). Using the example of protozoa, this study aims to provide guidance for (1) monitoring particulate contaminants in the water sources of treatment plants to ensure the representativeness of the water entering the plant and (2) using monitoring data to assess the sufficiency of treatment to meet health-based targets. The need for periodic review of monitoring programs is also discussed along with the reassessment of the framework after 2 years of its implementation.

Critically, it is inappropriate to input a concentration value that is only representative of a single day or month into this model because the maximum acceptable risk is defined on an annual level – doing so determines what the required log-reduction would be if the concentration value were sustained year round. This misalignment between how model inputs and risk are evaluated directly contradicts the current philosophy of ‘acceptable risk’ in microbial risk management. The annual time frame allows for periods of higher and lower risk that average out over the course of a year to an ‘acceptable’ level. Although this metric may not be sufficient for the evaluation of short-term and high-magnitude risks (Signor & Ashbolt 2009), it is widely applied for the determination of treatment requirements (WHO 2022). In QMRA, variation in pathogen concentrations over time is often incorporated in the form of a distribution rather than just using a mean concentration value. Such variation does not need to be considered here because the mean annual risk is the same regardless of whether daily variation in concentration is incorporated or simply represented by the mean. This is a mathematical property of the type of dose–response model used, together with the assumed maximum acceptable risk as well as the constant water consumption and treatment level, as demonstrated in Supplementary Material S1. Thus, collecting sufficient data to obtain a suitably accurate estimate of the mean concentration is the principal monitoring objective from a risk assessment perspective. The effect of dependance between concentration and treatment efficiency on risk was explored by De Brito Cruz et al. (2024) and it was determined to be minimal in the present case. As such, mean concentration values can be used to estimate average treatment needs in a year.

Here, a mean concentration value is calculated every month and a 1-year rolling average concentration is used to update the required log-reduction. Using a rolling average rather than calendar years of data allows timelier treatment responses to changes in pathogen concentrations. A 1-year period is preferable to the two-year period suggested in the provincial guidelines for municipal waterworks (Alberta 2012) because it will be more rapidly responsive to increases in (oo)cyst concentrations and will be more conservative in that the peaks will be higher. It also does not allow a one-year period with high risk to be averaged out with a preceding year of lower risk.

In the risk-based approach proposed by Health Canada, the concentration of pathogens in the source water is an important model input to determine treatment needs. Hence, it is critical that the concentrations evaluated during the monitoring program accurately reflect the water entering the treatment plant to avoid introducing bias to the estimated treatment needs. Fitting how data are collected to the purpose of monitoring leads to more informative results for decision-making. Once monitoring data are collected, it is also important to ensure that data are appropriately used in the risk assessment to be consistent with model formulation and assumptions. To provide guidance for protozoa monitoring including planning, collection, and analysis of data, the framework resulting from the implementation plan is divided into two main parts: establishment of an updated protocol to collect and analyze raw water samples and use of the collected data to inform treatment decisions.

The monitoring protocol is bound by constraints from the provincial approval signed in 2010 and the use of Method 1623.1 for the detection of Cryptosporidium and Giardia (U.S. EPA 2012). To ensure the representativeness of the raw water entering the treatment plant by the samples collected, the choice of sampling locations, number of samples per sampling location, sample frequency, and scheduling are important factors to consider. For example, when a treatment plant draws from two or more intakes, it is important to consider the contribution of each intake to the plant. Data from event-based monitoring should also be handled with care to avoid biasing treatment needs. Practical advice on general procedures for sample collection and analysis is also given to complement guidance from Method 1623.1 (Section 3.1.3).

The second part of the framework addresses the use of the monitoring data to inform treatment needs. This is also bound by constraints established in the provincial approval and Health Canada guidelines (Health Canada 2019). Results from the developed approach to estimate monthly and annual average pathogen concentrations from collected data and to evaluate treatment needs using these concentrations are presented. Finally, approaches to evaluate the sufficiency of treatment are discussed based on Calgary's regulatory framework that assigns credits based on treatment conditions (e.g., filter effluent turbidity). The main goal is to provide guidance on how to assess the sufficiency of treatment given the monitoring data and Health Canada guidelines rather than defining prescriptive rules of how to achieve a required treatment performance or criticizing pre-established treatment credits.

The framework to monitor protozoa and evaluate treatment sufficiency is composed of six main elements: the monitoring protocol includes (1) sampling location and representativeness of water entering the plant, (2) sample frequency and scheduling, and (3) general procedures for sample collection and analysis, while the analysis of treatment sufficiency includes (4) estimation of monthly and yearly average pathogen concentrations, (5) evaluation of treatment needs, and (6) use of treatment process monitoring data and credits to evaluate the sufficiency of treatment. Two years after the implementation of this framework in 2021, the monitoring program was reassessed to evaluate the impact of the proposed changes. Seven years of monitoring data are analyzed to discuss the main lessons learned from implementation of this framework.

When conducting a risk assessment to establish treatment requirements, the choice of sampling locations should ensure that the samples collected represent the raw water entering the plant well. For example, if a plant has multiple raw water intakes, monitoring blended water entering the treatment system can provide an uncomplicated representation of raw water quality. However, it is useful to collect samples from each of these intakes to inform risk management strategies such as the implementation of specific source water protection measures. At Bearspaw, there is no sampling location with blended water, so each raw water intake must be sampled separately. The LT2ESWTR's source water monitoring guidance (U.S. EPA 2006b) provides two options for water treatment plants that have multiple raw water intakes: a composite sample or a weighted average approach. In the composite sample option, the proportion of instantaneous flow contribution is reflected in the sample volume collected from each intake. The weighted average approach uses the instantaneous proportional flow contribution of each intake to weigh the averaging of concentrations estimated from samples analyzed independently. Two other options were considered to collect intake-specific information based on their proportional flow contributions: randomizing which intake to sample or adjusting the number of samples at each intake. The advantages and disadvantages of these five methods for accurately representing the raw water entering a treatment system are presented in Table 1.

Data from January 2016 to October 2020 were analyzed to compare the use of each raw water intake at Bearspaw. This analysis indicated that approximately 63% of Bearspaw water was drawn from the dam intake with almost all the monitoring data corresponding to this intake. It was decided to adjust the number of samples collected at each intake in the protozoa monitoring framework implemented in Calgary in 2021. To better represent the blend of water from the two intakes (approximately 2/3 from the dam and 1/3 from the river), the sampling approach for Bearspaw was modified to include at least two dam intake samples per month with a total analyzed volume of approximately 100 L in addition to one or more river intake samples per month with a total analyzed volume of approximately 50 L. If an intake is out of service in a particular month, then at least two samples with a total analyzed volume of approximately 100 L are drawn from whichever intake is in service. This method was chosen because it would lead to an uncomplicated and consistent monitoring protocol with a lower number of samples when compared to the weighted average approach. The contribution of each intake to the treatment plant and the impact of this change in the monitoring program were reassessed 2 years after implementation, and the results are also presented in Supplementary Material S2. At the Glenmore treatment plant, at least two samples with a total analyzed volume of 100 L are collected per month at the single raw water intake.

When determining the frequency of sampling, it is important to consider that risk is generally assessed on an annual basis in drinking water QMRA. To characterize annual risk, factors influencing risk (e.g., pathogen concentrations and drinking water consumption) must be well characterized throughout a year. Consequently, monitoring should be strategically planned so that the average pathogen concentration through a year is well represented by the samples collected. If water consumption (i.e., ingestion) patterns also change substantially throughout a year, it is recommended to sample more frequently when the consumption is higher. Collecting many samples is not always feasible, so it is necessary to balance the better representation of temporal variations in microbial water quality of high-frequency sampling with the drawback of expending more human and financial resources.

To determine a suitable sampling frequency in Calgary, historical pathogen concentration data from January 2016 to June 2020 were analyzed assuming no temporal autocorrelation between sampling events and that these data could be informative of future pathogen concentrations. Five different numbers of samples collected during a year were considered and 10,000 simulations were conducted for each to compare the precision of estimated yearly treatment requirements. Because variability of the mean concentration is not practically relevant when the 3-log minimum treatment is not exceeded, precision was assessed in terms of treatment requirements rather than average concentrations. Precision can be described using 95% probability intervals so that a relatively wide interval indicates poor precision. However, the length of an interval does not provide sufficient information about over- and underestimation of required treatment if it is asymmetrical, and these have different practical implications for safety. Considering the treatment required using all historical data as the true required treatment, overestimation is measured by the difference between the 97.5th percentile of the 10,000 simulated treatment values and the true treatment required, and underestimation is represented by the difference between the true treatment required and 2.5th percentile of simulated values. Equations (2) and (3) were used to calculate the log-reduction required, in which represented the average of n samples. The results (Supplementary Material S3, Table S1) show that the sampling frequency of twice a month previously implemented in Calgary is sufficient to estimate treatment requirements with an error of 0.16 log (i.e., neither the 97.5th nor the 2.5th percentile of log-reduction required is more than 0.16 log away from the true required treatment value). This frequency was maintained in Calgary for continuity, but a lower frequency could be justified, especially if those concentrations were observed in systems that cannot afford this frequency of sampling. For example, sampling once a month would be acceptable in Calgary so treatment is not under- or overestimated by more than 0.25 log.

While both the LT2ESWTR and monitoring requirements established in the Netherlands (Schijven et al. 2011) use the size of a system to determine the frequency of sampling, here it is recommended to consider the variability of previous pathogen concentrations observed in the source water. Although this approach may be limited when past concentrations cannot inform about the future, the site-specific variability in concentration is useful to determine sampling frequency because it is more directly linked to source water quality. For example, sampling costs can be saved if the variability of pathogen concentrations is low. It is a challenge to balance the benefits of a one-size-fits-all approach – which is easier to implement and possibly less expensive – with the advantages of tailoring monitoring practices (which can provide more information and reveal the specific needs of a particular system). Given a specified frequency of sampling that is deemed acceptable for a particular system, it is also important to consider when these samples are collected. To prevent bias in concentration data due to any possible systematic patterns in source water quality (e.g., greater wastewater discharges during a specific period of the day or day of the week, autocorrelation in pathogen concentrations), a random approach is suggested to choose dates and times for sample collection.

In addition to routine monitoring, it is also important to develop protocols for an event-based monitoring program given sudden water quality variations that can occur following extreme events. It is recommended, however, that such event-based monitoring and discretionary forms of exploratory monitoring should be excluded from routine analyses to determine the log-reduction required unless they are deemed to have an acceptable effect upon the result (e.g., not reducing the estimated average concentration or biasing it with substantial overrepresentation of short periods of time or rare conditions).

Overrepresentation of specific events is more impactful with lower numbers of samples contributing to the yearly average concentration. For example, a yearly average concentration of 8.9 oocysts/100 L requires 3-log reduction of Cryptosporidium in Calgary (Equation (3)). When calculating a 1-year rolling average concentration, replacing one routine sample from an average value of 8.9 oocysts/100 L with a concentration of 1,000 oocysts/100 L measured after an extreme event increases the treatment requirement to 3.8 log if 24 samples compose the yearly average. On the other hand, if 6 samples compose the yearly average, the treatment required would be 4.3 log. In this example, the new concentration of 1,000 oocysts/100 L practically represents 0.5 and 2 months given 24 or 6 samples per year, respectively. Thus, if a high concentration after an event does not last for this long, the treatment required is overestimated. Similarly, treatment needs might be underestimated if presumed events correspond to low concentrations or fail to capture periods with the highest concentrations. For example, peak Giardia concentrations at the Glenmore intake in Calgary have previously been associated with snowmelt run-off caused by chinooks rather than rain events (Sokurenko 2014). If, in a hypothetical situation, all sampling after extreme events results in non-detects, incorporating these data in the routine monitoring program could decrease treatment requirements. It is important for sampling used in QMRA to be performed at random without biasing results by targeting specific events. Extremely high or low concentrations should not be excluded from data analysis if they were randomly observed (e.g., occurring in proportion to their relative frequency and not especially sought out).

Fixing a frequency for event-based sampling is not possible, but there should be protocols to determine when it should occur. For example, the baseline conditions of precipitation and flow can be characterized so that thresholds can be established to trigger sampling. Knowing the characteristics of a system can also guide assessment of the need for event-based sampling. For example, reservoirs may have a role in controlling pathogen loading after rainfall events (Signor et al. 2007).

Method 1623.1 (U.S. EPA 2012) is a common protocol to detect Cryptosporidium oocysts and Giardia cysts in water matrices. Adaptations of this method as well as other laboratory techniques such as membrane filtration, flocculation, and molecular methods have also been applied worldwide (Rosado-García et al. 2017). Despite being expensive and requiring both adequate infrastructure and technical expertise of laboratory staff, Method 1623.1 and previous similar versions have been applied since 1999 in Calgary. Specific details of how this method is implemented in Calgary are critically reviewed and some lessons are discussed to provide general guidance, particularly related to the evaluation of analytical recovery and (oo)cyst enumeration practices.

Regardless of the analytical method used to assess concentrations of protozoa in water, it is essential to incorporate knowledge of the losses associated with imperfect analytical recovery into description of raw water concentrations – ignoring the effect of these losses upon observed counts leads to underestimated concentrations. Calgary's monitoring program typically follows the national guideline recommendation to measure the recovery efficiency of each sample (Health Canada 2019). Sample-specific analytical recovery is quantified using internal standards in every sample (or at least on every sampling occasion for a given source). Calgary's practice is to reject samples in which the internal standard recovery estimate is less than 33% for Cryptosporidium oocysts or less than 22% for Giardia cysts and to resample the associated source to ensure a minimum number of reliable samples that are not affected by significant matrix effects or other problems leading to unusually low recovery. These thresholds are based on the ongoing precision and recovery quality control acceptance criteria from Tables 3 and 4 of Method 1623.1. Although rejecting samples with low recovery is a good practice for quality control, it is worth mentioning that these acceptance criteria are for method validation of mean recovery in reagent water and are not a requirement for every sample analyzed using Method 1623.1.

Using an internal standard in every sample supports laboratory quality control (including resampling when achieved analytical recovery is low), enables evaluation of matrix effects or other shifts or trends in analytical recovery patterns, permits minor variations in methodology after seeding by ensuring corresponding evaluation of recovery, and improves the accuracy of individual concentration estimates (provided that the estimated recovery is not near zero). Despite being a good practice, using spiking suspensions for every sample can also be costly. Sporadic recovery analysis may be a feasible alternative in small systems.

Regarding sample volumes, the objective is to collect approximately 50 L per sampling occasion in Calgary. When multiple filters are needed to process the desired volume of water, the (oo)cyst counts and volumes of all filters are added and the sample-specific recovery information of one filter is applied to these cumulated data. To minimize variations of analytical recovery that might be caused by different sample volumes, similar volumes should be run through each filter in such circumstances. In some samples, the packed pellet may be too large for a single immunomagnetic separation (IMS). In Calgary, this sometimes happens during spring freshets when the full pellet is analyzed in several parts. Subjecting only a portion of the resuspended pellet to IMS reduces the effective analyzed volume of the sample and the precision of the analytical recovery estimate compared to analysis of the full pellet as not all seeded (oo)cysts are in the analyzed portion. To meet a monthly analyzed volume target and avoid compromising the value of internal standards added to every sample, the entire packed pellet of such samples should be analyzed in several portions. Statistically, the counts from the various portions may be added as though they were processed in a single IMS procedure. Referring to Equation (4), and would be the sums of the counts and volumes, respectively, for all filters and/or IMS tubes utilized.

Finally, when using microscopy to enumerate (oo)cysts, care should be taken to count everything that exhibits morphological features and staining attributes like Cryptosporidium oocysts and Giardia cysts (e.g., without discrimination based on subtle variations in size or shape) rather than rigidly limiting counts to respective size ranges (e.g., 4–6 μm diameter for Cryptosporidium oocysts, 5–15 μm width and 8–18 μm length for Giardia cysts). Microscopy-based methods have the limitation of not accounting for pathogen viability (i.e., determining if the pathogen is alive or not) and infectivity (i.e., differentiating human pathogenic species from others), which can lead to overestimation of epidemiologically relevant pathogen concentrations in water. In the absence of practical methods to evaluate the viability of (oo)cysts and their potential to cause illness in humans for each monitoring sample, presuming that all observed (oo)cysts are infectious provides a substantial safety factor by overestimating risk or treatment required (Lapen et al. 2016).

The methodology presented herein to compute required log-reduction makes many assumptions due to limitations in the best available science (e.g., representativeness of dose–response models) or to facilitate uncomplicated data analysis approaches. Supplementary Material S5 provides a summary of these assumptions and their likely effects on the computation of required log-reduction.

Treatment sufficiency is not usually determined empirically due to the difficulties of quantifying pathogen concentrations in treated water. Instead, treatment credits are often established in drinking water regulations according to design and operational conditions. This section discusses metrics to compare levels of treatment credited and required based on the Calgary monitoring and reporting experience. The accuracy of these credits with respect to actual treatment efficacy is beyond the scope of this study.

Log-reduction credits have been granted to the treatment system in accordance with the operating approval from the province of Alberta, which includes a 3-log basic credit for conventional filtration, a 0.5-log or 1.0-log additional filtration credit assigned monthly based on filter effluent turbidity, and a further credit for chlorination (for Giardia only) assigned based on each day's minimum achieved CT (the product of residual free chlorine concentration and contact time). The turbidity-based filtration credit is awarded based on a calendar month of turbidity data. The 0.5-log credit is granted when at least 99% of combined effluent turbidity data are below 0.15 NTU, whereas the 1.0-log credit is given when 99% of individual filter effluent turbidity data are less than 0.1 NTU. Calgary computes the credit daily allowing a cumulative maximum of 15 min above the turbidity threshold. This daily time span is generally more stringent than the monthly criteria described above because it does not allow days with more frequent turbidity exceedances to be averaged out with days with less frequent turbidity exceedances over a month. The disinfection credit is evaluated once per minute based on CT data that depend on relatively variable parameters such as temperature and flow. The current approval specifies reporting and use of a minimum daily CT value, and the resulting credit is limited to a maximum 3 log.

Figure 3 presents the log-reduction required and credited for Cryptosporidium oocysts and Giardia cysts at both treatment plants in Calgary according to the rules established in the provincial approval. Treatment credits are equal to or higher than the required treatment during the whole period analyzed, complying with provincial requirements. In Bearspaw, an effect of seasonality on the treatment credit for Giardia cysts can be observed. This is explained by the reduction in chlorination efficiency due to low temperatures during the winter. Unlike at Glenmore, the size of the chlorination tanks and the water flow rate at the Bearspaw treatment plant do not allow for an increase in contact time, which results in lower CT values. Risk can increase if these periods of lower treatment are associated with high pathogen concentrations in water. De Brito Cruz et al. (2024) investigated the impact of the correlation between treatment and concentration data in Bearspaw, and no substantial impact on risk estimates was found.

In Calgary, achieved treatment is only allowed to fall below the required treatment once per month to a minimum of 90% of the required treatment. However, this metric is problematic because it gets more lenient when there is a greater need for treatment (90% of 3.0-log is 2.7-log, 90% of 4.0-log is 3.6-log, and so on). Rather than relying on unquantified conservatism to assure safety (which may be overly cautious in some cases and not safe enough in others), it is better to apply a well-reasoned and equitable safety factor. It is suggested to use a fixed 0.3-log difference (10% of the 3-log minimum) so that the achieved treatment is expected to exceed the required treatment daily with at most one day per month where the achieved treatment falls no more than 0.3-log short of the required treatment. The difference between achieved and required log-reduction and how it varies over time is a useful metric for operational targets and it is more meaningful in the drinking water industry than log-reduction ratios. Several levels could be set for different purposes to trigger responses, such as an alert level (e.g., requiring action by operators and utility managers), a notification level (e.g., requiring discussion with regulators), and a warning level (e.g., declaring that existing treatment may not have assured adequate safety of the treated water over a 1-year period).

Monitoring in an ongoing fashion rather than, for example, over a single one-year period accounts for possible changes in the watershed that could lead to non-stationarity (i.e., substantial shifts relative to historical patterns) and allows more timely treatment responses to high pathogen concentrations. However, ongoing monitoring requires revisions, especially if there are major changes in the watershed or operation of the treatment plant, as well as when there is a sustained shift in pathogen occurrence patterns. In the absence of those, periodic revisions are also suggested to reflect current practices and/or incorporate updates of the best available science. Two years after implementation, the monitoring program in Calgary was reassessed, particularly to consider new data from the river intake at Bearspaw. Moreover, the relative sampling frequency at the two intakes depends on their proportional contribution, which should be reassessed periodically. The impact of including an additional sampling point representing the river intake at the Bearspaw treatment plant is discussed in Supplementary Material S2. In Calgary, the water quality of the two intakes is not substantially different (the river intake is 4 km downstream of the dam intake) which allows less frequent analysis.

To reassess the frequency of sampling, the same type of analysis described in Section 3.1.2 was conducted with two different datasets: the first considers 2 years of data after the implementation of the framework (from May 2021 to April 2023) and the second uses the whole dataset available since January 2016. The results are presented in Supplementary Material S3 (in Tables S2 and S3, respectively). Similar to the results found in 2020, a sampling frequency of twice a month allows estimation of treatment needs with an error of no more than 0.14 log in Glenmore and 0.11 log in Bearspaw when the last 2 years of data are used. When the whole dataset is used, the same frequency results in a maximum error of 0.27 log in Glenmore and 0.15 log in Bearspaw. These two analyses were conducted to compare the effect of the two considered distributions of pathogen concentrations. When using all available data, there is an underlying assumption that they all represent samples from the same distribution. If it is believed that the distribution of pathogen concentrations is changing over time (i.e., the dataset shows non-stationarity), then it is important to use a timeframe that better describes the current pattern of concentrations observed to tailor monitoring practices and treatment needs accordingly. On the other hand, it is also critical to not respond to random noise as though it were a change in pattern. Because the frequency of sampling is influenced by the variability of concentration in these analyses, it is important that a few very extreme data points do not lead to an impractical sampling frequency.

Managing microbial risks in drinking water involves the challenge of balancing generic assumptions taken to simplify modelling and regulations with site-specific considerations necessary to accurately represent the context studied. Historically, regulatory policies have focused on prescriptive rules that can cover as many situations as possible such as the designation of ‘well-operated’ filtration systems or the use of system sizes when determining sampling frequencies. For simplification, one-size-fits-all approaches are frequently needed to develop global guidelines or national policies in populous or low-income countries. While this approach may work in most systems under stable conditions, it is important to acknowledge that water utilities are required more and more to operate under challenging conditions, which calls for more specific guidance. For example, changes in the magnitude and variability of microbial contamination caused by extreme events can rapidly disrupt water treatment operations, compromising drinking water safety. In a changing climate, it is important to revisit knowledge about source water quality more frequently to meaningfully inform treatment decisions. Widespread implementation of de facto reuse also necessitates more guidance to ensure that water reaching drinking water treatment plants can be sufficiently treated. Moreover, the criteria to define good operation for one system might not be the same for another. Research has demonstrated how the current regulatory turbidity goal of 0.3 NTU might not ensure sufficient Cryptosporidium removal in systems relying on high-quality source water (Ballantyne et al. 2024). Little guidance is given for systems that need more than the 3-log minimum treatment requirement for Cryptosporidium. Furthermore, relying only on Cryptosporidium monitoring may not be sufficient to determine treatment requirements if Giardia cysts are more abundant in the source water. Hence, a one-size-fits-all approach can be under-protective in some cases if system-specific details are not considered.

Complementing general guidance provided in drinking water policies, the development of fit-for-purpose methods and models can contribute to decision-making regarding water safety. ‘Know your system’ is an expression emphasized more and more in the water industry, which underscores the need for practical guidance. This work provides specific guidance for water utilities interested in implementing a monitoring protocol for protozoa with the purpose of informing drinking water treatment decisions. When water contaminants – which can be protozoa, other pathogens, or even microplastics – are the proverbial ‘needle in a haystack’ (i.e., discrete and found in low concentrations in water), it is especially important that monitoring adequately represents the system being studied by designing protocols and analyzing data in a way that is fit-for-purpose for how risk is characterized. While it is common practice to implement monitoring as a one-size-fits-all approach, it is important to understand that monitoring the watershed to inform source water protection measures is different from monitoring the water entering the treatment plant to inform treatment decisions. A monitoring program that is designed to characterize average source water quality may not be informative enough to determine treatment decisions about the water entering the plant and to ensure the provision of safe drinking water not just on average but all the time. Peak raw water contaminant concentrations may not be captured by watershed monitoring or routine sampling at treatment plants. Although not addressed in this study, event-based monitoring can also help improve resilience during these episodic periods of deteriorated source water quality. With the exception of a few daily risk references in QMRA, guidance for short-term risk management in drinking water is still scant. In any case, it is critical to know when and where to collect data to serve the intended purpose and improve the return on investment of these costly monitoring programs. Moreover, this paper creates a bridge between collection of data and their subsequent use in a QMRA model. Typically, QMRA has been conducted with whatever data are available (or assumed values from the literature) to respond to risk assessment questions. Here, a framework was developed detailing how to collect data to better inform a risk assessment. There is a difference between making the most of available data and using data that are strategically gathered to be informative. Future research can develop frameworks to link the collection of data to other QMRA applications.

While water utilities can benefit from information provided by monitoring programs, it is important to acknowledge that routine monitoring of pathogen concentrations is not feasible and/or affordable in many places. Although this study focuses on a large system, some options to reduce sampling costs were noted. Even when it is not always possible to characterize pathogen concentrations in water, understanding how treatment requirements are influenced by different levels of concentration helps to inform decision-making. This work is not a defense for ‘monitoring at all costs’. Instead, it is argued that monitoring should be conducted strategically to better support decision-making in water treatment and effectively protect public health.

This paper provides guidance on the development of protocols for monitoring the raw water of treatment plants and the use of monitoring data in QMRA to evaluate treatment needs. The developed framework serves water utilities interested in maximizing the value of monitoring for drinking water treatment decisions and regulators that use monitoring as one of the strategies to control microbial risks in drinking water. The main conclusions are as follows:

• • It is important to know the quality of raw water to efficiently treat it (i.e., ensuring public health protection while avoiding unnecessary treatment costs). Tailored monitoring programs are needed to adequately characterize the raw water of a system and meaningfully support water treatment decisions, especially for increased water quality variability in a changing climate.

• • Data collection should be planned strategically to reflect the purpose of the monitoring program (e.g., to inform treatment decisions), extract maximum value from the allocated resources, and ensure the studied system is accurately represented.

• • o In systems with multiple raw water intakes, it is important for the contribution of each intake to be reflected in the sampling approach. There are several methods to choose from depending on the needs and capabilities of each water utility: (1) sampling blended water before treatment, (2) compositing samples of different volumes from each intake, (3) calculating a weighted average concentration from independent samples collected at each intake based on proportional flow contribution, (4) randomly selecting the intake given proportional flow contribution to the total water supply, or (5) representing the flow contribution by adjusting the number of samples per intake. Monitoring for treatment decisions needs to be representative of the water entering the treatment plant because this is the water that will ultimately be consumed.

  • o It is important to establish a frequency of sampling that can represent a system's needs, which can be informed by observed pathogen concentrations. This frequency can be adjusted to achieve desired precision in estimated treatment requirements given past data. Precisely determining treatment needs contributes to a better balance between water safety and sampling costs (e.g., a fit-for-purpose sampling protocol can save costs if the variability of pathogen concentrations is low).

• • When using risk assessment to determine treatment requirements, concentrations used in the model need to be consistent with how risk is measured. Computing annual risk using an individual concentration value assumes that it is sustained all year round, which can substantially overestimate treatment requirements and misinform decisions.

• • When compliance rules are established, it is important to set fixed-magnitude safety factors for consistently evaluating the sufficiency of treatment across systems. Instead of log-reduction ratios, the difference between treatment credits and requirements is recommended. Several thresholds could be set to trigger different risk management strategies (e.g., operational changes, regulator notification).

• • Treatment requirements may be driven by Giardia instead of Cryptosporidium if Giardia cysts are more abundant, which emphasizes the importance of also including this pathogen in protozoa monitoring programs and/or in regulatory requirements.

This work was supported by the Canada Research Chairs Program (ME – CRC in Water Science, Technology and Policy) (950-232469); the City of Calgary and the Alberta Innovates Water Innovation Program (AI2385B); and the forWater NSERC Strategic Partnership Grant for Networks (NETGP 494312-16). The City of Calgary contributed to the collection of data presented in this work. The authors also would like to thank Alice Stephanie Gomes for her help with Figure 1.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.