Implications of Using Robust Bayesian Analysis to Represent Diverse Sources of Uncertainty in Integrated AssessmentEPA Grant Number: R833666
Title: Implications of Using Robust Bayesian Analysis to Represent Diverse Sources of Uncertainty in Integrated Assessment
Investigators: Borsuk, Mark E. , Howarth, Richard B.
Institution: Dartmouth College
EPA Project Officer: Pascual, Pasky
Project Period: January 1, 2008 through December 31, 2009
Project Amount: $349,607
RFA: Uncertainty Analyses of Models in Integrated Environmental Assessments (2006) RFA Text | Recipients Lists
Research Category: Ecological Assessment , Economics and Decision Sciences , Ecosystems
We have the following objectives for the proposed project:
- To develop practical methods for using robust Bayesian analysis to separate the various sources of uncertainty in environmental modeling.
- To clarify the ramifications of environmental model uncertainty for linked economic analyses, particularly the problem of discounting under uncertainty.
- To establish how Bayesian networks can be used to integrate environmental and economic models in the robust Bayesian setting.
The objectives will be pursued in a multi-disciplinary context using a combination of environmental, economic, and statistical modeling. A particular focus of our project will be on the implications of epistemic uncertainty for integrated assessment of climate change. This provides a timely and policy-relevant problem context that builds on our previous research and is clearly transferable to other settings. For our climate system representation, we will use a global energy balance model, which we have previously modified for uncertainty analysis. To represent the world economy, we will use the Emissions Control & Optimal Taxation (ECOT) model. For the purposes of integrated uncertainty assessment, these models will be combined as a Bayesian network.
In our previous research, we showed that robust Bayesian methods can be used in environmental modeling to define a set of probability distributions for key parameters that captures the effects of expert disagreement, ambiguity, or ignorance. This entire set can then be updated against data using Bayes’ theorem to investigate the degree to which aleatory and/or epistemic uncertainty are reduced through additional observations. Further work is required to clarify the methods of selecting the appropriate set definitions in real-world applications. Such work addresses the first objective of the proposed project. In parallel research, we have demonstrated that economic analyses performed under conditions of uncertainty require specific, and previously unrecognized, methods and rates for discounting future benefits. We plan for our second objective to lead to a delineation of the exact consequences of this result for integrated assessment modeling. Finally, we hypothesize that the above two outcomes will have significant implications for decision support. This will be tested using the integrated robust Bayesian network to evaluate policies according to both conventional and alternative decision criteria.