Response Model Based Analysis of Climate Model Sensitivities and Uncertainties using the LLNL UQ Pipeline
Abstract
A Lawrence Livermore National Laboratory (LLNL) multi-directorate strategic initiative is developing uncertainty quantification (UQ) tools and techniques that are being applied to climate research. The LLNL UQ Pipeline and corresponding computational tools support the ensemble-of-models approach to UQ, and these tools have enabled the production of a comprehensive set of present-day climate calculations using the Community Atmospheric Model (CAM) and, more recently, the Community Earth System Model (CESM) codes. Statistical analysis of the ensemble is made possible by fitting a response surface, or surrogate model, to the ensemble-of-models data. We describe the LLNL UQ Pipeline and techniques that enable the execution and analysis of climate UQ and sensitivities studies on LLNL's high performance computing (HPC) resources. The analysis techniques are applied to an ensemble consisting of 1,000 CAM4 simulations. We also present two methods, direct sampling and bootstrapping, that quantify the errors in the ability of the response function to model the CAM4 ensemble. This work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344 and was funded by the Uncertainty Quantification Strategic Initiative Laboratory Directed Research and Development Project at LLNL under project tracking code 10-SI-013.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2011
- Bibcode:
- 2011AGUFMGC11A0893B
- Keywords:
-
- 3333 ATMOSPHERIC PROCESSES / Model calibration