Perturbed parameters ensembles of idealized experiments to understand the impact of parametric uncertainty on model performance and response to climate change
Abstract
We use computationally efficient, idealized set-up to construct five sets of perturbed parameter ensembles (PPEs) based on HadGEM3-GA4.0 to explore the uncertainty in its historical performance and climate change response. The PPEs, constructed by perturbing 21 parameters simultaneously, include NWP and AMIP-style experiments to understand model performance on weather and climate timescales respectively, and 3 idealized experiments to determine the effect of parameter perturbations on forcing and feedback components that drive model responses to climate change. We demonstrate that there is a strong relationship between model errors at weather and climate timescales for a variety of key variables. These relationships indicate that the inexpensive NWP hindcasts can be used to efficiently pre-screen the parameter space to exclude parts that give rise to physically unrealistic model behavior, before investing in longer climate simulations either in atmospheric or coupled mode. In terms of model response, the effect of perturbing parameters is to produce wide ranges in most forcing and feedback components that are similar in magnitude to those seen in the CMIP5 MME. We us statistical emulation to explore the parameter space thoroughly, and filter out parts of parameter space that do not give credible simulations of historical climate without affective diversity in global-scale climate change response. This provides a computationally efficient framework to identify a small set of plausible realizations of an expensive model for projection studies.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2018
- Bibcode:
- 2018AGUFMGC43J1668K
- Keywords:
-
- 1626 Global climate models;
- GLOBAL CHANGEDE: 1968 Scientific reasoning/inference;
- INFORMATICSDE: 1990 Uncertainty;
- INFORMATICSDE: 3275 Uncertainty quantification;
- MATHEMATICAL GEOPHYSICS