Abstract
Greywater reuse is a strategy to address water scarcity, necessitating the selection of treatment processes that balance cost-efficiency and human health risks. A key aspect in evaluating these risks is understanding pathogen contamination levels in greywater, a complex task due to intermittent pathogen occurrences. To address this, faecal indicator organisms like E. coli are often monitored as proxies to evaluate faecal contamination levels and infer pathogen concentrations. However, the wide variability in faecal indicator concentrations poses challenges in their modelling for quantitative microbial risk assessment (QMRA). Our study critically assesses the adequacy of parametric models in predicting the variability in E. coli concentrations in greywater. We found that models that build on summary statistics, like medians and standard deviations, can substantially underestimate the variability in E. coli concentrations. More appropriate models may provide more accurate estimations of, and uncertainty around, peak E. coli concentrations. To demonstrate this, a Poisson lognormal distribution model is fit to a data set of E. coli concentrations measured in shower and laundry greywater sources. This model estimated arithmetic mean E. coli concentrations in laundry waters at approximately 1.0E + 06 MPN 100 mL(-1). These results are around 2.0 log(10) units higher than estimations from a previously used hierarchical lognormal model based on aggregated summary data from multiple studies. Such differences are considerable when assessing human health risks and setting pathogen reduction targets for greywater reuse. This research highlights the importance of making raw monitoring data available for more accurate statistical evaluations than those based on summary statistics. It also emphasizes the crucial role of model comparison, selection, and validation to inform policy-relevant outcomes.