A key uncertainty of wastewater-based epidemiology is the size of the population which contributed to a given wastewater sample. We previously developed and validated a Bayesian inference model to estimate population size based on 14 population markers which: (1) are easily measured and (2) have mass loads which correlate with population size. However, the potential uncertainty of the model prediction due to in-sewer degradation of these markers was not evaluated. In this study, we addressed this gap by testing their stability under sewer conditions and assessed whether degradation impacts the model estimates. Five markers, which formed the core of our model, were stable in the sewers while the others were not. Our evaluation showed that the presence of unstable population markers in the model did not decrease the precision of the population estimates providing that stable markers such as acesulfame remained in the model. However, to achieve the minimum uncertainty in population estimates, we propose that the core markers to be included in population models for other sites should meet two additional criteria: (3) negligible degradation in wastewater to ensure the stability of chemicals during collection; and (4) < 10% in-sewer degradation could occur during the mean residence time of the sewer network.

The suitability of passive samplers (Chemcatcher) as an alternative to grab sampling in estimating time-weighted average (TWA) concentrations and total loads of herbicides was assessed. Grab sampling complemented deployments of passive samplers in a tropical waterway in Queensland, Australia, before, during and after a flood event. Good agreement was observed between the two sampling modes in estimating TWA concentrations that was independent of herbicide concentrations ranging over 2 orders of magnitude. In a flood-specific deployment, passive sampler TWA concentrations underestimated mean grab sampler (n = 258) derived concentrations of atrazine, diuron, ametryn, and metolachlor by an average factor of 1.29. No clear trends were evident in the ratios of load estimates from passive samplers relative to grab samples that ranged between 0.3 and 1.8 for these analytes because of the limitations of using TWA concentrations to derive flow-weighted loads. Stratification of deployments by flow however generally resulted in noticeable improvements in passive sampler load estimates. By considering the magnitude of the uncertainty (interquartile range and the root-mean-squared error) of load estimates a modeling exercise showed that passive samplers were a viable alternative to grab sampling since between 3 and 17 grab samples were needed before grab sampling results had less uncertainty.

The aim of this study was to understand the uncertainty of estimating loads for observed herbicides and nutrients during a flood event and provide guidance on estimator selection. A high-resolution grab sampling campaign (258 samples over 100 h) was conducted during a flood event in a tropical waterway in Queensland, Australia. Ten herbicides and three nutrient compounds were detected at elevated concentrations. Each had a unique chemograph with differences in transport processes (e.g. dependence on flow, dilution processes and timing of concentration pulses). Resampling from the data set was used to assess uncertainty. Bias existed at lower sampling efforts but depended on estimator properties as sampling effort increased: the interpolation, ratio and regression estimators became unbiased. Large differences were observed in precision and the importance of sampling effort and estimator selection depended on the relationship between the chemograph and hydrograph. The variety of transport processes observed and the resultant variability in uncertainty suggest that useful load estimates can only be obtained with sufficient samples and appropriate estimator selection. We provide a rationale to show the latter can be guided across sampling periods by selecting an estimator where the sampling regime or the relationship between the chemograph and hydrograph meet its assumptions: interpolation becomes more correct as sampling effort increases and the ratio becomes more correct as the r correlation between flux and flow increases (e.g. > 0.9); a stratified composite sampling approach, even with random samples, is a promising alternative.

No abstract