Can routine monitoring of E. coli fully account for peak event concentrations at drinking water intakes in agricultural and urban rivers?

Highlights

• Variability and uncertainty in Escherichia coli concentrations were modeled.

• Critical periods of microbial contamination of raw water were identified.

• Parametric models could predict peak Escherichia coli concentrations.

• Long-term routine monitoring accounted for peak events at drinking water intakes.

Abstract

In several jurisdictions, the arithmetic mean of Escherichia coli concentrations in raw water serves as the metric to set minimal treatment requirements by drinking water treatment plants (DWTPs). An accurate and precise estimation of this mean is therefore critical to define adequate requirements. Distributions of E. coli concentrations in surface water can be heavily skewed and require statistical methods capable of characterizing uncertainty. We present four simple parametric models with different upper tail behaviors (gamma, log-normal, Lomax, mixture of two log-normal distributions) to explicitly account for the influence of peak events on the mean concentration. The performance of these models was tested using large E. coli data sets (200–1800 samples) from raw water regulatory monitoring at six DWTPs located in urban and agricultural catchments. Critical seasons of contamination and hydrometeorological factors leading to peak events were identified. Event-based samples were collected at an urban DWTP intake during two hydrometeorological events using online β-d-glucuronidase activity monitoring as a trigger. Results from event-based sampling were used to verify whether selected parametric distributions predicted targeted peak events. We found that the upper tail of the log-normal and the Lomax distributions better predicted large concentrations than the upper tail of the gamma distribution. Weekly sampling for two years in urban catchments and for four years in agricultural catchments generated reasonable estimates of the average raw water E. coli concentrations. The proposed methodology can be easily used to inform the development of sampling strategies and statistical indices to set site-specific treatment requirements.

Introduction

The World Health Organization’s (WHO) water quality guidelines recommend a preventive and risk-based approach for drinking water quality management. For this purpose, a spectrum of microbial risk assessment approaches is available, from simple sanitary inspections and risk matrices to more complex ones such as quantitative microbial risk assessment (QMRA) (WHO, 2016b). The QMRA approach can provide relative estimates of microbial risks at drinking water treatment plants (DWTPs), which may be particularly useful to prioritize investments in improving water treatment or in implementing source water protection measures. However, in many situations, data on pathogen occurrence and concentrations in raw water are not available at DWTPs, and only faecal indicator bacteria (FIB) are measured to characterize source water quality. Therefore, to support the implementation of a source-to-tap approach, simplified classification methods, known as “bin classification” were developed to determine minimum treatment requirements according to a specific level of FIB concentrations in raw water. Different summary statistics (e.g., mean concentration, maximum concentration) and sampling strategies (e.g., weekly sampling, monthly sampling, event-based sampling) are specified in regulatory requirements worldwide for bin classification (Table A1).

The arithmetic mean is a valid metric for characterizing microbial concentrations in order to characterize the risk of multiple exposures to low doses of pathogens (Haas, 1996). The annual mean is usually considered in QMRA because annual health-based targets are recommended in guidelines and regulations (Sinclair et al., 2015). Precise estimation of the mean is challenging for surface water because microbial concentrations can vary over several orders of magnitude within hours or days (Burnet et al., 2019a). This metric relies on the law of large numbers; as the sample size grows, its sample mean gets closer to the true mean. However, the meaning of “large” depends on the distribution of the data. The convergence is much faster for normal or thin-tailed distribution than for heavy-tailed distributions. If the variance is very large, any new observation can be large enough to overwhelm all previous observations, regardless of the number of accumulated observations.

Numerous studies have shown that heavy rainfall can rapidly increase microbial contamination loads in water (Atherholt et al., 1998, Burnet et al., 2014, 2019a; Kistemann et al., 2002; Signor et al., 2005). In urban areas, combined sewer overflow (CSO) discharges induced by heavy rainfall or snowmelt events can cause recurring microbial peaks in raw water at DWTPs (Jalliffier-Verne et al., 2016; Madoux-Humery et al., 2016). In agricultural areas, similar heavy rainfall episodes can increase microbial contamination of surface waters as a result of overland transport, tile drainage systems and resuspension from stream sediments (e.g., Dorner et al., 2006). A statistical approach was proposed to incorporate such peak events in the risk assessment for a hypothetical frequency of occurrence (Petterson et al., 2006). However, methods for the estimation of peak event frequency have not been proposed yet. Stochastic models are used in other fields to evaluate the frequency of extreme precipitation events (Katz et al., 2002), streamflow peaks (Katz et al., 2002), and extreme pollution from runoff (Harremoës, 1988). These models have implications for quantifying the frequency of extreme events at DWTPs but have not been utilized in the context of microbial safety of drinking water.

In Quebec, Canada, raw water E. coli concentrations are measured since 2013 at least weekly for large DWTPs (>10,000 inhabitants). These extensive datasets provide a unique opportunity to study temporal variations in different catchments. The objective of the study was to first develop a methodology to correctly estimate the mean E. coli concentrations in surface drinking water sources by considering peak events. Large routine monitoring datasets from six DWTPs were fitted with parametric distributions having different upper tail behaviors. For the best-fit distributions, we then evaluated the required minimum sample size to estimate the mean concentration for different ranges of uncertainty. Secondly, key contributors to the mean concentration level were identified by examining the influence of seasonality and hydrometeorological factors on temporal variations. Finally, we conducted event-based sampling during two hydrometeorological events at an urban DWTP to evaluate whether the selected parametric distributions predicted these targeted peak events. Implications for the development of sampling strategies and probabilistic models are discussed for setting health-based drinking water treatment requirements.