|
Activity Number:
|
532
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Thursday, August 2, 2007 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #308951 |
|
Title:
|
Bayesian Kalman Filter for Emulation of Complex Computer Models
|
|
Author(s):
|
Gentry White*+ and Peter Reichert
|
|
Companies:
|
North Carolina State University and Swiss Federal Institute of Aquatic Science and Technology
|
|
Address:
|
2501 Founders Drive, Raleigh, NC, 27695-8203,
|
|
Keywords:
|
Bayesian ; Emulator ; State-space Model ; Kalman FIlter ; Dynamic Systems
|
|
Abstract:
|
Non-linear computer models can be used to simulate complex dynamic systems. The expense of running these models limits exploration of overall system behavior. In this cases a simpler linearized computer model called an emulator is desirable to explore system behavior under a variety of inputs. A good emulator should include the relevant physics for the system, perfectly interpolate the observed data and provide reasonable estimates of the complex model output when there are no observations. In the case of modeling dynamic systems we apply a state-space model, using a Kalman filter in the context of Bayesian estimation in order to construct an emulator. This model is a Gaussian Process and thus has good emulator properties. The model developed here is shown to provide a reasonable emulator for a real world dataset concerning a hydrological ground water runoff model.
|
