Monitoring the risk of Legionella infection using a general Bayesian network updated from temporal measurements in agricultural irrigation with reclaimed wastewater. Issue 1 (24th November 2022)

• Record Type: Journal Article
• Authors: Massiot, Gaspar; Courault, Dominique; Jacob, Pauline; Albert, Isabelle
• Is Part Of: Environmental science. Volume 9:Issue 1(2023)
• Page Start: 176
• Page End: 192
• Publication Date: 2022-11-24
• Subjects: Water-supply -- Periodicals; Water security -- Periodicals; Water resources development -- Periodicals; Water chemistry -- Periodicals; 553.705
• DOI: 10.1039/d2ew00311b
• Languages: English
• ISSNs: 2053-1400

Abstract

Abstract : This paper presents a model for the risk assessment of pathogen inhalation from sprinkler irrigation using treated wastewater. Abstract : Reuse of reclaimed wastewater for agricultural irrigation is an expanding practice worldwide. This practice needs to be monitored, partly because of pathogens that the water may contain after treatments. More particularly, sprinkler irrigation is known to generate aerosols which may lead to severe health risks to the population close to irrigated areas in case of the presence of Legionella bacteria in the water. A pilot experiment was conducted on two corn fields in South-Western France, irrigated with wastewater undergoing two different water treatments (ultra-filtration and UV). Water analyses have shown high levels of Legionella in the water even after a standard wastewater treatment plant (WWTP) cleaning process followed by the UV treatment (up to 10 6 GC per L in 2019). In this context, an updated general Bayesian network (GBN), using discrete and continuous random variables, in quantitative microbial risk assessment (QMRA) is proposed to monitor the risk of Legionella infection in the vicinity of the irrigated plots. The model's originality is based on i) a graphical probabilistic model that describes the exposure pathway of Legionella from the WWTP to the population using observed and non-observed variables and ii) the model inference updating at each new available measurement. Different scenarios are simulated according to the exposure time of the persons, taking into account various distances from the emission source and a large dataset of climatic data. From the learning process included in the Bayesian principle, quantities of interest (contaminations before and after water treatments, inhaled dose, probabilities of infection) can be quantified with their uncertainty before and after the inclusion of each new data collected in situ . This approach gives a rigorous tool that allows monitoring the risks, facilitates discussions with reuse experts and progressively reduces uncertainty quantification through field data accumulation. For the two pilot treatments analyzed in this study, the median annual risk of Legionella infection did not exceed the US EPA annual infection benchmark of 10 −4 for any of the population at risk during the past few months of the pilot experiment (DALYs are estimated up to 10 −5 ). The risk still bears watching with support from the method shown in this work.