Uncertainty Quantification and Global Sensitivity Analysis of the Los Alamos Sea Ice Model

TitleUncertainty Quantification and Global Sensitivity Analysis of the Los Alamos Sea Ice Model
Publication TypeJournal Article
Year of Publication2016
AuthorsUrrego-Blanco, Jorge R., Urban Nathan M., Hunke Elizabeth C., Turner Adrian K., and Jeffery Nicole
JournalJournal of Geophysical Research: Oceans
Volume121
Number4
Pages2709-2732
Date Published03/2016
Abstract / Summary

Changes in the high-latitude climate system have the potential to affect global climate through
feedbacks with the atmosphere and connections with midlatitudes. Sea ice and climate models used to understand these changes have uncertainties that need to be characterized and quantified. We present a quantitative way to assess uncertainty in complex computer models, which is a new approach in the analysis of sea ice models. We characterize parametric uncertainty in the Los Alamos sea ice model (CICE) in a standalone configuration and quantify the sensitivity of sea ice area, extent, and volume with respect to uncertainty in 39 individual model parameters. Unlike common sensitivity analyses conducted in previous studies where parameters are varied one at a time, this study uses a global variance-based approach in which Sobol’ sequences are used to efficiently sample the full 39-dimensional parameter space. We implement a fast emulator of the sea ice model whose predictions of sea ice extent, area, and volume are used to compute the Sobol’ sensitivity indices of the 39 parameters. Main effects and interactions among the most influential parameters are also estimated by a nonparametric regression technique based on generalized additive models. A ranking based on the sensitivity indices indicates that model predictions are most sensitive to snow parameters such as snow conductivity and grain size, and the drainage of melt ponds. It is recommended that research be prioritized toward more accurately determining these most influential parameter values by observational studies or by improving parameterizations in the sea ice model.

URLhttps://doi.org/10.1002/2015JC011558
DOI10.1002/2015JC011558
Journal: Journal of Geophysical Research: Oceans
Year of Publication: 2016
Volume: 121
Number: 4
Pages: 2709-2732
Date Published: 03/2016

Changes in the high-latitude climate system have the potential to affect global climate through
feedbacks with the atmosphere and connections with midlatitudes. Sea ice and climate models used to understand these changes have uncertainties that need to be characterized and quantified. We present a quantitative way to assess uncertainty in complex computer models, which is a new approach in the analysis of sea ice models. We characterize parametric uncertainty in the Los Alamos sea ice model (CICE) in a standalone configuration and quantify the sensitivity of sea ice area, extent, and volume with respect to uncertainty in 39 individual model parameters. Unlike common sensitivity analyses conducted in previous studies where parameters are varied one at a time, this study uses a global variance-based approach in which Sobol’ sequences are used to efficiently sample the full 39-dimensional parameter space. We implement a fast emulator of the sea ice model whose predictions of sea ice extent, area, and volume are used to compute the Sobol’ sensitivity indices of the 39 parameters. Main effects and interactions among the most influential parameters are also estimated by a nonparametric regression technique based on generalized additive models. A ranking based on the sensitivity indices indicates that model predictions are most sensitive to snow parameters such as snow conductivity and grain size, and the drainage of melt ponds. It is recommended that research be prioritized toward more accurately determining these most influential parameter values by observational studies or by improving parameterizations in the sea ice model.

DOI: 10.1002/2015JC011558
Citation:
Urrego-Blanco, JR, NM Urban, EC Hunke, AK Turner, and N Jeffery.  2016.  "Uncertainty Quantification and Global Sensitivity Analysis of the Los Alamos Sea Ice Model."  Journal of Geophysical Research: Oceans 121(4): 2709-2732.  https://doi.org/10.1002/2015JC011558.