Seeing is Predicting: Water Clarity-Based Nowcast Models for E. coli Prediction in Surface Water

Abstract

Given the 24–48 h turn-around time of conventional surveillance approaches, methods are needed that improve the timeliness and accuracy of recreational water quality risk assessments. Although one useful approach is to combine existing monitoring programmes with predictive faecal indicator bacteria (FIB) models, these models are largely ‘top-down’ in their approach to safeguarding public health. Beyond being simply ‘advised when to avoid swimming’, there is an increasing awareness amongst the general public regarding the role they can play in water quality monitoring. Using quantile, maximum value and optimized incremental modelling approaches, this study reports on the possibility of developing intuitive, public-friendly models that are based on the physical appearance of water (clarity), to estimate 8103 nation-wide E. coli concentrations in rivers, and to assess whether water is safe to swim in. If swimmers were to avoid river waters with <1.1 m black disc visibility during autumn and summer, and river waters with values <0.5 m black disc visibility during spring and winter, they would also avoid microbial hazards that are associated with exceedances of the 540 CFU/100 mL single sample bathing water standard. Regardless of the climatic season, stream order classification, catchment land cover or geology of streams considered, the clarity-based E. coli models performed well as they presented with sensitivity, specificity and accuracy values of at least 72%. The developed models offer the benefit of providing a faster method for estimating E. coli concentration, potentially engaging the public in water monitoring, and allowing them to make informed decisions on whether it is safe to swim.

Interpretability Versus Accuracy: A Comparison of Machine Learning Models Built Using Different Algorithms, Performance Measures, and Features to Predict E. coli Levels in Agricultural Water

“…Each model was built using one of nine water quality or weather features (Polat et al, 2019). Studies conducted in nonagricultural, freshwater environments (e.g., swimming beaches) that focused on developing interpretable models used similar sets of physicochemical and weather features (Olyphant and Whitman, 2004;Francy and Darner, 2006;Efstratiou et al, 2009;Shiels and Guebert, 2010;Francy et al, 2013;Bradshaw et al, 2016;Dada, 2019). To ensure comparability with these previous studies, and provide baseline models that could be used to gauge full and nested model performance, we developed eight log-linear and a featureless regression model.…”

“…While past studies have shown that predictive models can be useful for assessing public health hazards in recreational water (Olyphant, 2005;Hou et al, 2006;Hamilton and Luffman, 2009;Francy et al, 2013;Francy et al, 2014;Dada and Hamilton, 2016;Dada, 2019;Rossi et al, 2020), no models, to the author's knowledge, have been developed to predict E. coli levels in surface water used for produce production (e.g., for irrigation, pesticide application, dust abatement, frost protection). Moreover, many of the recreational water quality studies only considered one algorithm during model development (e.g., (Olyphant, 2005;Hamilton and Luffman, 2009), including algorithms [e.g., regression, (Olyphant, 2005;Hamilton and Luffman, 2009)], which has more assumptions and may be less accurate than alternate algorithms (e.g., ensemble methods, support vector machines, (Kuhn and Johnson, 2016;Weller et al, 2020a)).…”

Feasibility of nowcasting SDG indicators: A comprehensive survey

“…Over half of the Goal 6 indicators come from 6.6.1's sub-indicators. In general, Goal 6 includes indicators that fall under ecological or environmental topics, for which modelling is frequently used [36][37][38][39][40][41]. Explanatory variables should be widely available for these indicators.…”