Análise Bayesiana da Formação de Trihalometanos em Sistemas de Água Potável

18082023 0608 A multivariate Bayesian network analysis of water quality factors influencing trihalomethanes formation in drinking water dist httpswwwsciencedirectcomsciencearticleabspiiS0043135420312471frRR2refpdfdownloadrr7dcf21602cd10300 17 Pesquisa de água Volume 190 15 de fevereiro de 2021 116712 Uma análise de rede bayesiana multivariada de fatores de qualidade da água que influenciam a formação de trihalometanos em sistemas de distribuição de água potável Rebecca A Li James A McDonald Arumugam Sathasivan Stuart J Khan Mostre mais httpsdoiorg101016jwatres2020116712 Obtenha direitos e conteúdo Abstrato Controlar a formação de subprodutos da desinfecção e ao mesmo tempo garantir a desinfecção eficaz da água potável é importante para proteger a saúde pública No entanto compreender e prever a formação de subprodutos de desinfecção sob uma variedade de condições na água potávelos sistemas de distribuição permanecem desafiadores pois a formação de subprodutos de desinfecção é um fenômeno multifatorial Este estudo teve como objetivo avaliar a aplicação de modelos de Rede Bayesiana para prever a concentração de trialometanos a classe dominante de subprodutos de desinfecção halogenados usando vários parâmetros de qualidade da água Modelos bayesianos ingênuos e bayesianos semiingênuos foram construídos a partir de conjuntos de dados de Sydney e sudeste de Queensland em 15 sistemas de distribuição de água potável na Austrália A variável alvo concentração total de trihalometanos foi discretizada em 3 categorias 01 mg L 01 02 mg L e 02 mg L The Bayesian network structures were built using water quality parameters including concentrations of individual and total trihalomethanes disinfectant species free chlorine monochloramine dichloramine total chlorine nitrogen species free ammonia total ammonia nitrate nitrite and other physicalchemical parameters temperature pH dissolved organic carbon total dissolved solids conductivity and turbidity Seven performance parameters including predictive accuracy and the rates of true and false positive and negative results were used to assess the accuracy and precision of the Bayesian network models After evaluating the model performance the optimum models were selected to be Bayesian network augmented naïve models These were observed to have the highest predictive accuracies for Sydney 78 and South East Queensland 94 Although disinfectant residuals are among the key variables that lead to trihalomethanes formation potential concentrations of trihalomethanes in distribution systems can be more confidently predicted in terms of probability associated with a wider range of water quality variables using Bayesian networks The modelling procedure developed in this work can now be applied to develop systemspecific Bayesian network models for trihalomethanes prediction in other drinking water distribution systems Graphical abstract a a b e Compartilhar Citar 1 1 1 18082023 0608 A multivariate Bayesian network analysis of water quality factors influencing trihalomethanes formation in drinking water dist httpswwwsciencedirectcomsciencearticleabspiiS0043135420312471frRR2refpdfdownloadrr7dcf21602cd10300 27 Download Download highres image 262KB Download Download fullsize image Introduction Chemical disinfectants are widely used in municipal drinking water systems Primary disinfectants such as free chlorine or chlorine dioxide are commonly applied to achieve minimum disinfection doses at drinking water treatment plants WTPs Treated water is delivered to customers via drinking water distribution systems DWDSs comprising storage reservoirs pipes and pumping stations To maintain the quality of drinking water in distribution systems secondary disinfectants which may also include free chlorine but sometimes more stable residuals such as monochloramine are widely used An important aspect of chemical disinfection is the need to control the potential production of disinfection by products DBPs These are byproducts that are produced from reactions between disinfectants and other substances present in the water including both organic and inorganic chemicals Richardson and Ternes 2011 Xie 2003 Although most DBP formation is associated with disinfection processes at WTPs additional DBPs may also be produced during water transmission in DWDSs Epidemiological studies have indicated that longterm exposure to DBPs is associated with increased adverse health outcomes including bladder cancer Villanueva et al 2004 Although over 700 DBPs have been identified and more previously unidentified DBPs are continuing to emerge only a few selected groups are monitored routinely by water utilities Li and Mitch 2018 Richardson and Kimura 2020 Most prominently monitored are trihalomethanes THMs followed by haloacetic acids Li and Mitch 2018 THMs are known to be the dominant class by mass of halogenated Cl Br and I DBPs in drinking water which are formed by reactions between chlorine and natural organic matter such as humic or fulvic acids Crittenden et al 2012 Krasner et al 2006 The presence of bromide or iodide ions in the water leads to brominated or iodinated THMs respectively Xie 2003 Four common THM species chloroform bromodichloromethane dibromochloromethane and bromoform collectively known as THM4 have been subjected to regulation in the US since 1979 Li and Mitch 2018 The World Health Organization WHO Guidelines for Drinkingwater Quality have set guideline values for individual THM4 in drinking water for chloroform 03 mg L bromodichloromethane 006 mg L dibromochloromethane 01 mg L and bromoform 01 mg L WHO 2017 The Australian Drinking Water Guidelines ADWG has recommended a similar guideline concentration for THMs but based on the sum of these four THMs which should not exceed 025 mg L NHMRC and NRMMC 2011 Finding a balance between disinfection efficiency and DBP formation in drinking water remains a challenge for water utilities To minimise THM formation during distribution monochloramine a weaker but more stable disinfectant than free chlorine has been applied as a secondary disinfectant in DWDSs Richardson and Postigo 2015 Understanding and predicting DBP formation under a variety of conditions in DWDSs is challenging since DBP formation is known to be a multifactorial phenomenon In addition to halogen ions and natural organic matter DBP formation can be influenced by other water quality and operational parameters in DWDSs including temperature pH water age turbidity and the concentrations of disinfectant residual and nutrients nitrogen phosphorus and sulphur Li et al 2019 Many studies have reported on the development of numerical predictive models and multivariate analysis of influencing factors for THMs formation Brown et al 2011 Chen and Westerhoff 2010 Le Roux et al 2012 Obolensky and Singer 2008 Although some regression models have achieved satisfactory prediction on DBPs formation with strong correlations to the selected predictors Golea et al 2017 these traditional regression models may not be the best approach to capture the complex interrelationships between water quality operational factors and DBPs formation Having examined this problem in detail we have previously developed the concept of applying 1 1 1 1 1 18082023 0608 A multivariate Bayesian network analysis of water quality factors influencing trihalomethanes formation in drinking water dist httpswwwsciencedirectcomsciencearticleabspiiS0043135420312471frRR2refpdfdownloadrr7dcf21602cd10300 37 Bayesian networks as an alternative to facilitate multifactorial modelling and prediction of DBP formation in DWDSs Li et al 2019 The advantages of Bayesian networks in comparison to regression models are described below Bayesian networks are probabilistic graphical models which can be used to describe causal relationships between model variables Marcot 2012 A Bayesian network structure consists of nodes representing variables and arcs representing directional connections Norsys 2017 An arc starts from a variable which is referred to as a parent node and ends at another variable which is called a child node showing their causal dependencies Each node is conditionally independent unless they are linked by arcs Holmes and Jain 2008 The target variable to be predicted or classified is called a class node if it is not a child node of any other variables which are also called attributes In most Bayesian network software continuous variables need to be discretised into a finite set of states Chen and Pollino 2012 A conditional probability table CPT attached to each node shows the quantified causal relationship between nodes The size of the CPT of a child node grows when the number of its parents and their states increase which increases the complexity associated with populating the CPT Chen and Pollino 2012 Bayesian networks can incorporate parameter uncertainty and variability Carvajal et al 2017 Based on other given observed data or conditional probability information Bayesian networks can also calculate the conditional probabilities of missing states when there are random missing variables Holmes and Jain 2008 The predictive values are presented as numerical values from regression models whereas in Bayesian networks they are presented in terms of probabilities in a range of values eg 90 01 mg L 9 in 01 02 mg L and 1 02 mg L While many researchers have used traditional regression numerical models to seek the relationship between known influencing variable and THM formation Bayesian network is able to capture the subtle interrelationships between variables Struhl 2017 When individual parameters are changed in response to observations or what if scenarios the full chain of influence can be observed from a Bayesian Network in which the interrelationships among predictive variables can be determined Similarly Bayesian Networks facilitate backcasting which enables users to determine the most likely system conditions required to achieve a specified outcome ie THM4 being below guideline values However these can be challenging to achieve in linear regression models which require known data inputs ie all required water parameters in the function to generate a target output ie THM4 concentration especially when there are missing data in the predictive variables of the function Thus Bayesian network modelling has been shown to offer greater functionality than regression when determining the effects of many variables on an outcome Struhl 2017 Therefore DBP formation under various conditions can be predicted more realistically in Bayesian networks with new conditional probabilities Based on Bayes Theorem developed in the 18 century Bayesian networks have been progressively used to determine conditional relationships among variables both attributes and target variables in water quality studies especially in the field of chemical and microbial risk assessment Carvajal et al 2015 2017 Graham et al 2019 Islam et al 2016 Zerouali and Zerouali 2020 Zhu et al 2010 Zhu et al 2014 There are only a limited number of THMs studies applying Bayesian networks among which a dynamic Bayesian network Zhu et al 2010 and a Bayesian belief network BBN Zhu et al 2014 have been successfully developed for predicting THM formation and associated health risks in drinking water with the source water quality at WTPs As it is a crossdiscipline technique no predictive model on DBP formation using Bayesian networks with water quality parameters in DWDSs has been reported to our knowledge In this study predictive Bayesian network models were developed to understand and predict the formation of four individual THMs and the composite THM4 in Australian chloraminated DWDSs It was hypothesised that routinely collected parameters could provide a strong predictive capability for THM formation following the model development and optimisation procedures Unlike the laboratorycontrolled studies routinely collected water quality parameters might not have significant variances in treated drinking water and often data for certain known influencing factors such as bromide concentration are unavailable In this work we demonstrate that a Bayesian network approach is particularly useful to determine the interrelationships among influencing factors relevant to those specific systems via backcasting and conditional scenario studies Water quality data routinely collected by water utilities can be investigated to identify useful predictive relationship in Bayesian network applications Section snippets 1 1 1 th 18082023 0608 A multivariate Bayesian network analysis of water quality factors influencing trihalomethanes formation in drinking water dist httpswwwsciencedirectcomsciencearticleabspiiS0043135420312471frRR2refpdfdownloadrr7dcf21602cd10300 47 Material and methods A procedure for Bayesian network modelling from the approach previously proposed by Carvajal et al 2015 with minor modifications was used This procedure incorporated the key steps of data collection and formatting structure development performance evaluation model comparison and selection and model exportation and prediction The focus of this study was to predict the formation of THM4 using other water quality variables Correlations of formation between the individual THM4 were also Results and discussion Sydney and South East Queensland regions represent two types of chloraminated DWDSs in which various recorded water quality parameters were observed Variations in THM concentrations and speciation were anticipated between the two regions especially at the end of pipes Sydney and South East Queensland are discussed individually and in comparison with each other in the following subsections Conclusion Bayesian network models for THM4 formation in DWDSs can be produced by following the procedures of model development and optimisation relying on the existing utility water quality data Potential THMs concentrations in DWDSs can be confidently predicted in Bayesian network in terms of probability using the concentrations of disinfectant residuals nitrogen species pH TDS DOC conductivity and other water quality parameters but in much less confidence with only disinfectant residuals and pH Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper Acknowledgement This research was funded by the Australian Research Council ARC LP1601009099 The authors are grateful to Sydney Water Corporation Seqwater and Urban Utilities for providing water quality data and reviewing this manuscript The authors also appreciate to Dr Hung Pham for MATLAB assistance and Dr Guido Carvajal and Dr Tom Beuzen for technical discussions on Netica RL is supported by Australian Government Research Training Program Scholarship RTP and UNSW Faculty of Engineering Topup References 46 MS Boyce et al Evaluating resource selection functions Ecological modelling 2002 G Carvajal et al Modelling pathogen log10 reduction values achieved by activated sludge treatment using naive and semi naive Bayes network models Water Res 2015 G Carvajal et al Bayesian belief network modelling of chlorine disinfection for human pathogenic viruses in municipal wastewater Water Res 2017 B Chen et al 18082023 0608 A multivariate Bayesian network analysis of water quality factors influencing trihalomethanes formation in drinking water dist httpswwwsciencedirectcomsciencearticleabspiiS0043135420312471frRR2refpdfdownloadrr7dcf21602cd10300 57 Predicting disinfection byproduct formation potential in water Water Research 2010 SH Chen et al Good practice in Bayesian network modelling Environmental Modelling Software 2012 M Deborde et al Reactions of chlorine with inorganic and organic compounds during water treatmentKinetics and mechanisms A critical review Water Research 2008 DM Golea et al THM and HAA formation from NOM in raw and treated surface waters Water Res 2017 S Kinani et al Analysis of inorganic chloramines in water TrAC Trends in Analytical Chemistry 2012 RA Li et al Disinfectant Residual Stability Leading to Disinfectant Decay and Byproduct Formation in Drinking Water Distribution Systems A Systematic Review Water Research 2019 BG Marcot Metrics for evaluating performance and uncertainty of Bayesian network models Ecological Modelling 2012 View more references Cited by 19 The mixedorder chlorine decay model with an analytical solution and corresponding trihalomethane generation model in drinking water 2023 Environmental Pollution Show abstract Predicting heterotrophic plate count exceedance in tap water A binary classification model supervised by cultureindependent data 2023 Water Research Show abstract Insight into mixed chlorinechloramines conversion and associated water quality variability in drinking water distribution systems 2023 Science of the Total Environment Show abstract Exposure and carcinogenic risk assessment of trihalomethanes THMs for water supply consumers in Addis Ababa Ethiopia 18082023 0608 A multivariate Bayesian network analysis of water quality factors influencing trihalomethanes formation in drinking water dist httpswwwsciencedirectcomsciencearticleabspiiS0043135420312471frRR2refpdfdownloadrr7dcf21602cd10300 67 2023 Toxicology Reports Show abstract Cumulative human health risk analysis of trihalomethanes exposure in drinking water systems 2022 Journal of Environmental Management Show abstract Multivariate experimental design provides insights for the optimisation of rechloramination conditions and water age to control disinfectant decay and disinfection byproduct formation in treated drinking water 2022 Science of the Total Environment Citation Excerpt Thus the chloramine dose and pH were selected as the first and second most significant variables in the Doehlert matrix Based on previous studies bromide Br concentration is also known to affect the speciation of DBPs and monochloramine decay Li et al 2021 Liu et al 2012 Sathasivan et al 2020 In addition as BrDBPs were reported to be more toxic than ClDBPs Richardson et al 2007 bromide concentration is an essential parameter to be investigated as the third parameter in the Doehlert matrix to determine the interactions of disinfectant dose and pH on disinfectant decay and DBP formation in bromidecontaining water Show abstract View all citing articles on Scopus Recommended articles 6 Research article The modelling of SurfaceWater photoreactions made easier introducing the concept of equivalent monochromatic wavelengths Water Research Volume 190 2021 Article 116675 Show abstract Research article Development of methods to estimate microcystins removal and water treatment resiliency using mechanistic risk modelling Water Research Volume 190 2021 Article 116763 Show abstract Research article A continuous variable Bayesian networks model for water quality modeling A case study of setting nitrogen criterion for small rivers and streams in Ohio USA Environmental Modelling Software Volume 69 2015 pp 1422 Show abstract Research article Bayesian belief network modelling of chlorine disinfection for human pathogenic viruses in municipal wastewater Water Research Volume 109 2017 pp 144154 Show abstract 18082023 0608 A multivariate Bayesian network analysis of water quality factors influencing trihalomethanes formation in drinking water dist httpswwwsciencedirectcomsciencearticleabspiiS0043135420312471frRR2refpdfdownloadrr7dcf21602cd10300 77 Research article Use of surrogate indicators for the evaluation of potential health risks due to poor urban water quality A Bayesian Network approach Environmental Pollution Volume 233 2018 pp 655661 Show abstract Research article Previsão de risco de acidentes de poluição com base em uma rede bayesiana integrada e modelo de qualidade da água para o projeto de transferência de água do sul para o norte Engenharia Ecológica Volume 96 2016 pp 109116 Mostrar resumo Ver texto completo 2020 Elsevier Ltd Todos os direitos reservados Copyright 2023 Elsevier BV ou seus licenciadores ou colaboradores ScienceDirect is a registered trademark of Elsevier BV