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.
- Drinking water
- Escherichia coli
- Event-based sampling
- Exposure assessment
- Quantitative microbial risk assessment