Activity Number:
|
185
- SPEED: Environmental Statistics
|
Type:
|
Contributed
|
Date/Time:
|
Monday, July 31, 2017 : 11:35 AM to 12:20 PM
|
Sponsor:
|
Section on Statistics and the Environment
|
Abstract #325121
|
|
Title:
|
Computer Model Calibration via the Ensemble Kalman Filter
|
Author(s):
|
Seiyon Lee* and Murali Haran and Klaus Keller
|
Companies:
|
Pennsylvania State University and Pennsylvania State University and Pennsylvania State University
|
Keywords:
|
Data Assimilation ;
Climate Models ;
Calibration ;
Ensemble Kalman Filter ;
Gaussian Process Emulators
|
Abstract:
|
Deterministic models are used to model complex systems in fields such as climate science and public health. Input parameters for these models are often uncertain. Computer model calibration involves inferring input parameters based on physical observations of the system. I investigate the use of ensemble Kalman filters (EnKF) for computer model calibration and compare the EnKF to the popular Gaussian process (GP) based approach for calibration. I will discuss how EnKF is computationally more efficient than GP methods even as the number of computer model runs (ensemble size) increases. I will also show, via simulated examples, that increasing the dimensions of the input or output can result in increased computational costs and poor calibration results. Finally, I will discuss new methods I am developing for EnKF with large input and output dimensions.
|
Authors who are presenting talks have a * after their name.