Applications and future developments of Bayesian Networks in risk and impact assessments and environmental decision making

Open seminar and workshop

When: Wednesday March 29th 2017. From 10 to 14

Where: The Blue Hall and the Red Room in the Ecology Building, Lund University, Sölvegatan 37, Lund, Sweden

Scroll down to view the list of talks

Flyer BN Open Seminar


Bayesian Networks is a type of probabilistic modelling with wide applications in science and decision making. BNs is a modelling framework that enable us to integrate evidence to inform decisions, based on causal relations between decision, states and impact variables. BNs allows integration of data and expert knowledge.

This workshop will demonstrate the application of BNs in risk and impact assessment and environmental management, and, discuss and critically reflect on the developments and applications of BNs in research and decision making, through the use of case studies.


Blue Hall

10h00 -11h00  Adaptive management using Bayesian tools to meet the challenges of uncertainty, climate change and the challenge of making decisions – Wayne Landis

11h00 -11h45   Learning chains in oil spill risk analysis – Sakari Kuikka

11h45 – 12h30  Lunch

Red Room

12h30 – 14h00

Assessing multiple climate change impacts on water quality: a Bayesian Networks approach – Anna Sperotto

A Bayesian approach for safety barrier portfolio optimization – Alessandro Mancuso

Application of Bayesian Networks in integrated water resource management – Hazel Indrani Govender

What is needed to get Bayesian Networks robust to weaknesses in knowledge? – Ullrika Sahlin

Invited speakers and Abstracts

Invited speakers

Wayne Landis, Institute of Environmental Toxicology, Huxley College of the Environment, Western Washington University, United States. Wayne is a Professor in Environmental Science and a Director of the Institute of Environmental Toxicology at Western Washington University. He has over 20 years of experience in ecological risk assessment research, using Bayesian networks to guide decision making.

Sakari Kuikka, University of Helsinki, Finland. Sakari is a Professor in fisheries science and a head of the Fisheries and Environmental Management Group research group.


Adaptive management using Bayesian tools to meet the challenges of uncertainty, climate change and the challenge of making decisions – Wayne Landis

Environmental contamination, invasive species, changes in technology, and climate change are individually management challenges, but in reality, each intersects with the other.  This interaction is coupled with dramatic social changes in Europe, North America and across the world.  The question becomes how to effectively manage this mélange of factors within the cacophony of social norms.  Terms like “risk” and “uncertainty” also have multiple meanings, often with emotional undertones.  This talk will introduce the approach of using a Bayesian net based risk assessment framework coupled with an adaptive management framework.  Bayesian networks can be built to include multiple stressors affecting endpoints that represent ecological status and human well-being.  The process of adaptive management includes an initial risk assessment coupled to specific predictions regarding management options (hypotheses), followed by observation of the environment, and reconstruction of the Bayesian network to update the model.  The approach is closely coupled to decision making and is intended to be adaptable.  One of the goals in this approach is to accept uncertainty as a normal situation both in the understanding of the managed system, the efficacy of the management solutions, and in the societal norms.

Learning chains in oil spill risk analysis – Sakari Kuikka

We review the experience obtained in developing integrative Bayesian models in interdisciplinary risk analysis focusing on oil spill in the Gulf of Finland. Moreover, we also discuss the future challenges in this demanding modeling task. We have applied Bayesian models to the oil spill risk analysis in interdisciplinary questions. Bayesian belief networks are flexible tools that can take into account the different research traditions and the various types of information sources.

One of the advantages of using Bayesian decision analysis for management is that the uncertainty estimates are scientifically justified. Moreover, the Bayesian inference offers and important possibility to learn effectively from many sources of information, and the results of one integrative model can, and we argue that they should, be used as priors for next accidents so that the learning component from previous spills is as high as possible. Especially in cases where society is assumed to be risk averse, the uncertainty estimates have a crucial role.

Assessing multiple climate change impacts on water quality: a Bayesian Networks approach – Anna Sperotto, University Ca’ Foscari Venice, Italy

Bayesian Networks are employed for the implementation of a multi-risk model to assess cascading impacts induced by multiple stressors on water quality taking into account multiple climate and land use scenarios. Specifically, Bayesian Networks are applied as a meta-modelling tool for structuring and combining the information coming from existing hydrological models simulations, climate change and land use scenarios and to prioritize the contribute of different stressors on water quality status.

A Bayesian approach for safety barrier portfolio optimization – Alessandro Mancuso, Aalto University, Finland

In the framework of Probabilistic Risk Assessment (PRA), we develop a method to support the selection of cost-effective portfolios of safety measures. This method provides a systemic approach to determining the optimal portfolio of safety measures that minimizes the risk of the system and thus provides an alternative to using risk importance measures for guiding the selection of safety measures. We represent combinations of events leading to system failure with Bayesian Belief Networks (BBNs) which can be derived from traditional Fault Trees (FTs) and are capable of encoding event dependencies and multi-state failure behaviors. We also develop a computationally efficient enumeration algorithm to identify which combinations (portfolios) of safety measures minimize the risk of failure at different costs of implementing the safety measures. The method is illustrated by revisiting an earlier case study concerning the airlock system of a CANDU Nuclear Power Plant (NPP). The comparison of results with those of choosing safety measures based on risk importance measures shows that our approach can lead to considerably lower residuals risks at different cost levels.

Application of Bayesian Networks in integrated water resource management  – Hazel Indrani Govender

Water catchments are complex, with water resources receiving impacts from a vast range of land-use activities. Integrated Water Resource Management (IWRM) is a holistic approach that attempts to integrate the sustainable management of the water, land and related resources within the broader socio-economic and political context.  Risk assessment at a catchment scale, requires the consideration of multiple stressors and many causal relationships that result from the interactions between the components of ecosystems.  The Relative Risk Model, using Bayesian Networks (BNs), is used to assess the risks in a water stressed, economically critical water catchment in South Africa.  The focus on the study area by the authorities and government, has facilitated a number of research efforts and collaborations.  This is bringing together experts and a range of stakeholders in working towards protection of the water resources in the catchment.  This is beneficial to the application of Bayesian Networks as the information and data resulting from these research efforts can contribute to knowledge gaps and missing data.  This can facilitate updating of the BNs and contribute to an adaptive management approach to protecting water resources in the catchment.

What is needed to get Bayesian Networks robust to weaknesses in knowledge? – Ullrika Sahlin, Lund University

Also the sun has its spots. Bayesian Networks are useful, but has its limitations. I will mention some problems with BNs coming from weaknesses in knowledge. Instead of leaving you in total misery, I will end with some suggestions on how to deal with these issues without totally abandoning Bayesian Networks.

This workshop is funded by the research school ClimBEco 


This entry was posted in meetings, risk research. Bookmark the permalink.

Write a Comment

* Required