Understanding the uncertainty of estimating herbicide and nutrient mass loads in a flood event with guidance on estimator selection

Andrew Joseph Novic, Christoph Ort, Dominique S. O'Brien, Stephen E. Lewis, Aaron M. Davis, Jochen F. Mueller

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 r2 correlation between flux and flow increases (e.g. > 0.9); a stratified composite sampling approach, even with random samples, is a promising alternative.