Inference for Deterministic Models in the Environmental Sciences

• Activity Number: 244
• Type: Invited
• Date/Time: Tuesday, August 9, 2005 : 10:30 AM to 12:20 PM
• Sponsor: IMS
• Author(s): Adrian E. Raftery*+
• Companies: University of Washington
• Keywords: probabilistic weather forecasting ; Bayesian model averaging ; Bayesian melding ; wildlife management ; environmental risk assessment ; air quality

Abstract

There are two main cultures of quantitative research: statistical modeling and mechanistic modeling, which often is deterministic, using systems of differential equations. Disciplines tend to rely mainly on one or the other. Both are often useful, however, and I will review efforts over the past decade to achieve a synthesis driven by environmental applications. I will mention applications in wildlife management and environmental risk assessment and review Bayesian melding, which allows one to take account of evidence and uncertainty about a mechanistic model's inputs and outputs when making inference about a quantity of policy or other interest. I will describe an extension to deal with situations where model outputs and the data relevant to them are on different scales. Bayesian melding is difficult when the model takes a long time to run. I will describe how one can use Bayesian model averaging to make calibrated inference using only a few model runs and apply it to probabilistic weather forecasting.