GCRF Blue Communities Research Programme -

16 December 2019

By Dr Olivia Langmead, University of Plymouth

In August 2019, training in the application of Bayesian belief networks to marine environmental issues was held at the 2nd Blue Communities Annual Meeting and Training Workshops in Plymouth, UK. The training comprised two parts; initially Dr Caroline Hattam introduced conceptual modelling to the participants.

Caroline outlined the rationale behind the construction of conceptual models and provided context to the exercise that followed. Participants were divided into groups to each develop a conceptual model for one of the key South East Asian benthic habitats: coral, seagrass, mangrove and sedimentary habitats. The emphasis was to map out causal linkages arising from fishing activities and to prioritise one or two commonly used fishing practices. Participants developed conceptual models that captured the pressures arising from the chosen fishing type(s) and how these could impact the ecological components, together with any critical thresholds (levels of pressure that cause structural/functional changes to benthic habitats). The relationships between habitats and ecosystem services were assessed using evidence gathered to support the scoring of the habitat-potential ecosystem service matrix as part of Projects 3 & 8.

The second part, led by Dr Olivia Langmead, introduced Bayesian belief networks (BBNs), a type of probabilistic graphical model. BBNs represent a set of variables and the relationships between them as statistical dependences. They are broadly characterised by a set of nodes representing the variables, connected by linkages or arrows that represent the causal relationships and a set of probabilities that represent the influence of the condition of the parents (upstream nodes) on the state of a node. Applications for BBNs in marine and coastal management were outlined, together with their pros and cons. More detail on how to go about developing a BBN was given with emphasis on model structure, indicators for variables, defining the states of nodes, constructing conditional probability tables, and making explicit our confidence in the relationships.

The final part comprised a practical and interactive session that involved participants returning to their previous groups according to benthic habitat. They then worked to refine their previously developed conceptual models into graphical models suitable for BBN modelling. Specifically each group was asked to consider and review the models structure, direction of influence, numbers of parent nodes, potential indicators and states, and assess the types of available data that could be used to populate conditional probability tables, which describe the relationship between variables. This BBN training will be put into practice as Project 8 progresses over the second half of Blue Communities.

Bayesian belief network training

16 December 2019

Other newsletter articles

Promoting alternative livelihoods to enhance sustainable use of marine resources

Great progress from partners in Vietnam

Metadata: What is it all about?