Benchmarking inference methods for water quality monitoring and status classification

Abstract

River water quality monitoring at limited temporal resolution can lead to imprecise and inaccurate classification of physicochemical status due to sampling error. Bayesian inference allows for the quantification of this uncertainty, which can assist decision-making. However, implicit assumptions of Bayesian methods can cause further uncertainty in the uncertainty quantification, so-called second-order uncertainty. In this study, and for the first time, we rigorously assessed this second-order uncertainty for inference of common water quality statistics (mean and 95th percentile) based on sub-sampling high-frequency (hourly) total reactive phosphorus (TRP) concentration data from three watersheds. The statistics were inferred with the low-resolution sub-samples using the Bayesian lognormal distribution and bootstrap, frequentist t test, and face-value approach and were compared with those of the high-frequency data as benchmarks. The t test exhibited a high risk of bias in estimating the water quality statistics of interest and corresponding physicochemical status (up to 99% of sub-samples). The Bayesian lognormal model provided a good fit to the high-frequency TRP concentration data and the least biased classification of physicochemical status (< 5% of sub-samples). Our results suggest wide applicability of Bayesian inference for water quality status classification, a new approach for regulatory practice that provides uncertainty information about water quality monitoring and regulatory classification with reduced bias compared to frequentist approaches. Furthermore, the study elucidates sizeable second-order uncertainty due to the choice of statistical model, which could be quantified based on the high-frequency data.

An investigation into the impact of nine catchment characteristics on the accuracy of two phosphorus load apportionment models

“…The impact of 'flashy' hydrographs and low sampling frequency on nutrient load estimation uncertainty has long been proposed (Johnes, 2007), with it still being highlighted as a barrier to robust models and reliable outputs in contemporary studies (Hollaway et al, 2018;Jung et al, 2020).This reduction in high Q data will be a further likely source of model uncertainty as true levels of diffuse contributions are masked (Johnes, 2007;Bowes et al, 2008).…”

“…Although the GM did not, holistically, provide an accurate representation of observed data points, the BM yielded results which demonstrate the algorithm generally performs well on datasets of the type analysed within this study. However, a challenge remains that these datasets are unlikely, given sampling frequency, to capture the full range of Q-P variation that occur within watercourses as recently shown by Jung et al (2020). Only by using high frequency Q-P data can true patterns be identified (Bieroza and Heathwaite, 2015;Williams et al, 2015;Elwan et al, 2018) and thereby increase the accuracy of BM P apportionment.…”

Benchmarking a decade of holistic agro-environmental studies within the Agricultural Catchments Programme

“…High-frequency monitoring of streamflow and stream water chemistry was recognised as a requirement to target the knowledge gaps for the underlying mechanisms to nutrient loss to water (Bol et al, 2018). The approach allows the capture of the full dynamics of nutrient loss to water over a year, without being skewed to specific sampling events or periods, and provides insights into water quality during both low-flow and high-flow conditions (Cassidy & Jordan 2011;Jordan et al, 2012a;Jung et al, 2020). Synchronous highfrequency water quality and discharge data provided new possibilities to analyse nutrient transfer pathways.…”

Modeling Water Quality in Watersheds: From Here to the Next Generation

“…Both statistical and machine learning methods provide ways of estimating temporal dynamics and patterns of variability in constituent concentrations and loads, assisting the identification of system stressors ( Glendell et al, 2019 ), setting of regulatory targets ( Jung et al, 2020 ), and model simplification ( Jackson-Blake et al, 2017 ). However, extrapolation of fitted behavior beyond the ranges and environmental conditions of measured data requires extreme caution.…”