“This Is What We Don't Know”: Treating Epistemic Uncertainty in Bayesian Networks for Risk Assessment

Sahlin, Ullrika; Helle, Inari; Perepolkin, Dmytro

doi:10.1002/ieam.4367

Cited by 32 publications

(25 citation statements)

References 84 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Furthermore, the probabilistic assessment provided by the BBN methodology is more in line with the classical definitions of risk than the more commonly used single-value risk quotients (Moe et al, 2021). The model represents key processes to capture combined uncertainties stemming both from observational data and limited knowledge (Sahlin et al, 2021).…”

Section: Resultsmentioning

confidence: 99%

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

Troldborg

Gagkas

Vinten

et al. 2022

Hydrol. Earth Syst. Sci.

View full text Add to dashboard Cite

Abstract. Pesticides are contaminants of priority concern that continue to present a significant risk to drinking water quality. While pollution mitigation in catchment systems is considered a cost-effective alternative to costly drinking water treatment, the effectiveness of pollution mitigation measures is uncertain and needs to be able to consider local biophysical, agronomic, and social aspects. We developed a probabilistic decision support tool (DST) based on spatial Bayesian belief networks (BBNs) that simulates inherent pesticide leaching risk to ground- and surface water quality to inform field-level pesticide mitigation strategies in a small (3.1 km2) drinking water catchment with limited observational data. The DST accounts for the spatial heterogeneity in soil properties, topographic connectivity, and agronomic practices; the temporal variability of climatic and hydrological processes; and uncertainties related to pesticide properties and the effectiveness of management interventions. The rate of pesticide loss via overland flow and leaching to groundwater and the resulting risk of exceeding a regulatory threshold for drinking water was simulated for five active ingredients. Risk factors included climate and hydrology (e.g. temperature, rainfall, evapotranspiration, and overland and subsurface flow), soil properties (e.g. texture, organic matter content, and hydrological properties), topography (e.g. slope and distance to surface water/depth to groundwater), land cover and agronomic practices, and pesticide properties and usage. The effectiveness of mitigation measures such as the delayed timing of pesticide application; a 10 %, 25 %, or 50 % reduction in the application rate; field buffers; and the presence/absence of soil pan on risk reduction were evaluated. Sensitivity analysis identified the month of application, the land use, the presence of buffers, the field slope, and the distance as the most important risk factors, alongside several additional influential variables. The pesticide pollution risk from surface water runoff showed clear spatial variability across the study catchment, whereas the groundwater leaching risk was uniformly low, with the exception of prosulfocarb. Combined interventions of a 50 % reduced pesticide application rate, management of the plough pan, delayed application timing, and field buffer installation notably reduced the probability of a high risk of overland runoff and groundwater leaching, with individual measures having a smaller impact. The graphical nature of BBNs facilitated interactive model development and evaluation with stakeholders to build model credibility, while the ability to integrate diverse data sources allowed a dynamic field-scale assessment of “critical source areas” of pesticide pollution in time and space in a data-scarce catchment, with explicit representation of uncertainties.

show abstract

Section: Resultsmentioning

confidence: 99%

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

Troldborg

Gagkas

Vinten

et al. 2022

Hydrol. Earth Syst. Sci.

View full text Add to dashboard Cite

show abstract

“…As pesticide risk assessment is inherently uncertain, due to many complex and poorly characterised processes, the graphical BBN model helps to improve the transparency of the risk management process (Carriger and Newman, 2012). Furthermore, the probabilistic assessment provided by the BBN methodology is more in line with the classical definitions of risk than the more commonly used singlevalue risk quotients (Moe et al, 2021).The model represents key processes to capture combined uncertainties stemming both from observational data and limited knowledge (Sahlin et al, 2021).…”

Section: Resultsmentioning

confidence: 99%

“…4). Regardless of the absolute values, the relative difference in entropy between different management scenarios and risk status classes is informative for informing management interventions and safeguarding managers from putting too much or too little confidence in the final risk assessment (Sahlin et al, 2021).…”

Section: Can We Characterise the Spatial And Temporal Variability Of Pesticide Pollution Risk From Groundwater Leaching And Overland Flowmentioning

confidence: 99%

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

Troldborg

Gagkas

Vinten

et al. 2021

Preprint

View full text Add to dashboard Cite

Abstract. Pesticides are contaminants of priority concern that continue to present a significant risk to drinking water quality. While pollution mitigation in catchment systems is considered a cost-effective alternative to costly drinking water treatment, the effectiveness of pollution mitigation measures is uncertain and needs to be able to consider local biophysical, agronomic, and social aspects. We developed a probabilistic decision support tool (DST) based on spatial Bayesian Belief Networks (BBN) that simulates inherent pesticide leaching risk to ground- and surface water quality to inform field-level pesticide mitigation strategies in a small drinking water catchment (3.1 km2) with limited observational data. The DST accounts for the spatial heterogeneity in soil properties, topographic connectivity, and agronomic practices; temporal variability of climatic and hydrological processes as well as uncertainties related to pesticide properties and the effectiveness of management interventions. The rate of pesticide loss via overland flow and leaching to groundwater and the resulting risk of exceeding a regulatory threshold for drinking water was simulated for five active ingredients. Risk factors included climate and hydrology (temperature, rainfall, evapotranspiration, overland and subsurface flow), soil properties (texture, organic matter content, hydrological properties), topography (slope, distance to surface water/depth to groundwater), land cover and agronomic practices, pesticide properties and usage. The effectiveness of mitigation measures such as delayed timing of pesticide application; 10 %, 25 % and 50 % reduction in application rate; field buffers; and presence/absence of soil pan on risk reduction were evaluated. Sensitivity analysis identified the month of application, land use, presence of buffers, field slope and distance as the most important risk factors, alongside several additional influential variables. Pesticide pollution risk from surface water runoff showed clear spatial variability across the study catchment, while groundwater leaching risk was uniformly low, with the exception of prosulfocarb. Combined interventions of 50 % reduced pesticide application rate, management of plough pan, delayed application timing and field buffer installation notably reduced the probability of high-risk from overland runoff and groundwater leaching, with individual measures having a smaller impact. The graphical nature of the BBN facilitated interactive model development and evaluation with stakeholders to build model credibility, while the ability to integrate diverse data sources allowed a dynamic field-scale assessment of ‘critical source areas’ of pesticide pollution in time and space in a data scarce catchment, with explicit representation of uncertainties.

show abstract

“…Well-suited for decision-support: ability to reveal hidden dependencies and indirect impacts (Marcot et al, 2006;Chen and Pollino, 2012;Rachid et al, 2021;Sahlin et al, 2021;Zhou et al, 2021) What-if scenario for the dependence on flood-sensitive crops was illustrated based on exemplary BN. The example suggested that alterations in the dependence on flood-sensitive crops (high vs. low) lead to changes of risk drivers pertaining, vulnerability and the final risk, including drivers with considerable causal distance to the observed driver (dependence on flood-sensitive crops).…”

Section: Benefitsmentioning

confidence: 99%

Assessing Flood Risk Dynamics in Data-Scarce Environments—Experiences From Combining Impact Chains With Bayesian Network Analysis in the Lower Mono River Basin, Benin

et al. 2022

View full text Add to dashboard Cite

River floods are a common environmental hazard, often causing severe damages, loss of lives and livelihood impacts around the globe. The transboundary Lower Mono River Basin of Togo and Benin is no exception in this regard, as it is frequently affected by river flooding. To enable adequate decision-making in the context of flood risk management, it is crucial to understand the drivers of risk, their interconnections and how they co-produce flood risks as well as associated uncertainties. However, methodological advances to better account for these necessities in risk assessments, in data-scarce environments, are needed. Addressing the above, we developed an impact chain via desk study and expert consultation to reveal key drivers of flood risk for agricultural livelihoods and their interlinkages in the Lower Mono River Basin of Benin. Particularly, the dynamic formation of vulnerability and its interaction with hazard and exposure is highlighted. To further explore these interactions, an alpha-level Bayesian Network was created based on the impact chain and applied to an exemplary what-if scenario to simulate changes in risk if certain risk drivers change. Based on the above, this article critically evaluates the benefits and limitations of integrating the two methodological approaches to understand and simulate risk dynamics in data-scarce environments. The study finds that impact chains are a useful model approach to conceptualize interactions of risk drivers. Particularly in combination with a Bayesian Network approach, the method enables an improved understanding of how different risk drivers interact within the system and allows for dynamic simulations of what-if scenarios, for example, to support adaptation planning.

show abstract

“This Is What We Don't Know”: Treating Epistemic Uncertainty in Bayesian Networks for Risk Assessment

Abstract: This is what we don't know'-Treating epistemic uncertainty in Bayesian networks for risk assessment

Cited by 32 publications

References 84 publications

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

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

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

Assessing Flood Risk Dynamics in Data-Scarce Environments—Experiences From Combining Impact Chains With Bayesian Network Analysis in the Lower Mono River Basin, Benin

Contact Info

Product

Resources

About