Diagnostic analysis of Gulf area using Bayesian Decision Networks

[zenrabbit]

Author: Patricia Varela Topic: Environmental synthesis for net health of region

Objective

To identify activities developed in the Gulf Coast area and study their correspondent health, environmental, social and economic impact and combine them using Bayesian Decision Networks. Activities can range from oil & gas developments, constructions, farming, fisheries, hunting, tourism, shipping, among many others, based on the most important economies of the study area. In due course, the impact and intensity of these activities can be gathered in several locations that will allow to map them with a comprehensive, robust and systematic approach involving data mining, modeling, data integration and spatial statistics.

Method

The data should be classified in spatial and temporal domains and resolution, and be obtained, preferably, from official sources. The experts from the group are encouraged to collaborate on tasks such as: data identification and preparation, definition of cause-effect relationships between data, definition of variables required for the model, and ultimately, model design. This approach has been widely used for data integration, and allows scientists and engineers to include their professional beliefs, model predictions and monitoring data in the Bayesian Decision Network. The results of this type of analysis allows researchers and data analysists to have a better understanding of the consequences of a combination of selected parameters when a prognosis reasoning is performed (cause to effect). However, this methodology also allows to make management decisions by fixing a specific variable representing policy-making scenarios, and study its impact on the updated condition of the original variables, on what is known as a diagnostic reasoning (effect to causes). Improved decision making and risk management and reduction of uncertainty are among the most important contributions that can result from this proposed synthesis activity.

[vdtobias]

This is pretty exciting! I like the idea of using Bayesian Decision Networks to formalize decision-making. I think having a concrete way to look at trade-offs would be a very useful thing for the GOM. I don't know much about the specifics of Bayesian Decision Networks so I hope we'll get to learn about them through this proposal.

[pvarelag]

In my experience, the use of Bayesian Decision Networks encourages multidisciplinary collaborations that have been used not only for decision making but also for artificial intelligence, and uncertainty quantification. It's a very versatile tool capable of integrating data, model predictions and experts educated beliefs as sources of evidence. I hope I can contribute with this project to stimulate active collaboration between participants.

[ailich]

I don't have any experience with Bayesian Decision Networks, but being able to integrate data sets from various disciplines sounds interesting.

[kdorans]

I also would be interested to learn more about Bayesian Decision Networks and to see how they can be applied to policy decisions. I thought this article (focused on adaptive management during a restoration project), is a great example of how structured decision-making can be successfully used in practice: https://www.ncbi.nlm.nih.gov/pubmed/27623362. Did you have thoughts on what scale or level of complexity you might focus on for this work? For the adaptive management article, the framework got very complicated very quickly. I wondered if, for the synthesis project, it might work well to focus on one particular region or specific impacts of a few activities and then branch out from there?

[pvarelag]

Great question! Bayesian Networks also can get very complex very quickly, and there are recommended practices in the literature (e.g. Ref 1 & Ref 2) to keep them simple and meaningful. It is important to identify the simplifications made during the modeling process to draw the correct conclusions, and therefore to have a truly informative decision making tool.

The collaboration process between experts from different disciplines will always result in a comprehensive, elaborated network. My idea is to design what I call "The Big Picture" model first, and to separate teams to assess different aspects of the network. Each group would make decisions on the type of simplifications that need to be made based on data and time constrains. At the end of the three weeks, all team members will have a set of improvement methodologies that could enhance the capabilities of the model in their own research. Maybe will encourage further collaboration between experts.

I will take a look at the publication you shared. Thank you!

Ref1: Marcot, B. G., Steventon, J. D., Sutherland, G. D., & McCann, R. K. (2006). Guidelines for developing and updating Bayesian belief networks applied to ecological modeling and conservation. Canadian Journal of Forest Research, 36(12), 3063-3074.

Ref2: Chen, S. H., & Pollino, C. A. (2012). Good practice in Bayesian network modelling. Environmental Modelling & Software, 37, 134-145.

[kdorans]

@pvarelag, thanks so much! I will check these out. Austin.etal.2002.Bayeswatch.OverviewofBayesianStatistics.JEvalClinPract.pdf

Though this is not Bayesian network modeling specifically, I also wanted to pass along this reference, in case anyone is interested in learning about Bayesian compared with frequentist approaches for statistical analyses in the context clinical research: https://www.ncbi.nlm.nih.gov/pubmed/12060417. I think it provides a good overview of the Bayesian paradigm.