Development of a Bayesian network for probabilistic risk assessment of pesticides

Abstract

Conventional environmental risk assessment of chemicals is based on a calculated risk quotient, representing the ratio of exposure to effects of the chemical, in combination with assessment factors to account for uncertainty. Probabilistic risk assessment approaches can offer more transparency, by using probability distributions for exposure and/or effects to account for variability and uncertainty. In this study, a probabilistic approach using Bayesian network (BN) modelling is explored as an alternative to traditional risk calculation. BNs can serve as meta-models that link information from several sources and offer a transparent way of incorporating the required characterization of uncertainty for environmental risk assessment. To this end, a BN has been developed and parameterised for the pesticides azoxystrobin, metribuzin, and imidacloprid. We illustrate the development from deterministic (traditional) risk calculation, via intermediate versions, to fully probabilistic risk characterisation using azoxystrobin as an example. We also demonstrate seasonal risk calculation for the three pesticides.

A critical review of effect modeling for ecological risk assessment of plant protection products

“…Fraser et al (2002) discussed of uncertainty in biomagnihcation factors and half-lives of métabolites, while Weijs et al (2013) used a Morris sensitivity analysis followed by the eFAST test to quantitatively test the influence of the most sensitive parameters on their model output. We also noticed an increas ing use of probabilistic methods, such as Bayesian inference (Weijs et al, 2013) or Bayesian Networks (BN) (Kaikkonen et al, 2020;Mentzel et al, 2021), which hâve proven their efficiency in quantifying uncertainties. And to go in the same direction, Rubach et al (2010) hâve even illustrated that a complementary use of least-squares htting with the Levenberg-Marquardt (LM) algorithm and Monte Carlo Markov Chain (MCMC) methods is much more useful than the use of LM alone.…”

“…The PNEC can be calculated from the HC$ (Tier-2 PNEC), accounting for uncertainty by dividing the HC$ by a certain coefficient. According to authors, the relationship between the HC$ and the PNEC may differ: it can be assumed equal to the médian HC$ (Brock et al, 2006), to its lower-limit (Daam et al, 2010), to the ratio of the HC$ by an uncertainty factor (Mentzel et al, 2021); in the regulatory context, either to the ratio of the HC$ by an appropriate Assessment Factor (AF, European Commission 2003) or also equal to the médian HCz estimate [e.g., EFSA PPR Panel (2015b)). Note that ratios based on SSD outputs are now preferred: for example the Tier-1 Regulatory Acceptable Concentration (RAC) is an EC^o/AF, while the Tier-2B RAC is an EfC^/AF as usually preferred in aquatic risk assessment (EFSA PPR Panel, 2013).…”

“…A probabilistic RQ is a more informative alternative to the traditional single-value RQ, which is often interpreted as a binary outcome. Indeed, it can be useful useful for ranking of different scénarios as well as prioritizing among alternative risk scénarios (in , cited in Mentzel et al (2021)). Among probabilistic approaches, the Bayesian Network (BN) models are increasingly being used to model environmental Systems, in order to: integrate multiple issues and System components; utilise information from different sources; and handle missing data and uncertainty (Chen and Pollino, 2012).…”

Probabilistic modelling of inherent field-level pesticide pollution risk in a small drinking water catchment using spatial Bayesian Belief Networks

“…The use of BBNs has gained increasing popularity in environmental modelling and risk assessment (Aguilera et al, 2011;Kaikkonen et al, 2021) with examples including pesticide risk management (Carriger and Newman, 2012;Henriksen et al, 2007) and probabilistic assessments of pesticide exposure and effects (Mentzel et al, 2021). While the integration of Bayesian networks with GIS in environmental risk assessment has also been growing steadily over recent years (Moe et al, 2021), to date spatial BBN has only been used for pesticide risk modelling on a single occasion to assess pesticide runoff risk at a basin scale across France (Piffady et al, 2020).…”

Probabilistic modelling of the inherent field-level pesticide pollution risk in a small drinking water catchment using spatial Bayesian belief networks

“…The use of BBNs has gained increasing popularity in environmental modelling and risk assessment (Aguilera et al, 2011;Kaikkonen et al, 2020), with examples including pesticide risk management (Carriger and Newman, 2012;Henriksen et al, 2007) and probabilistic assessments of pesticide exposure and effects (Mentzel et al, 2021). While the integration of Bayesian networks with GIS in environmental risk assessment has also been growing steadily over recent years (Moe et al, 2021), spatial BBN has currently only been used for pesticide risk modelling on a single occasion to assess pesticide runoff risk at a basin scale across France (Piffady et al, 2020).…”