A Caveat Note on Tuning in the Development of Coupled Climate Models. (13th January 2018)

• Record Type: Journal Article
• Authors: Dommenget, Dietmar; Rezny, Michael
• Is Part Of: Journal of advances in modeling earth systems. Volume 10:Number 1(2018)
• Page Start: 78
• Page End: 97
• Publication Date: 2018-01-13
• Subjects: optimization -- coupled general circulation model -- model bias -- model errors -- climate sensitivity uncertainty; Geological modeling -- Periodicals; Climatology -- Periodicals; Geochemical modeling -- Periodicals; 551.5011
• DOI: 10.1002/2017MS000947
• Languages: English
• ISSNs: 1942-2466

Abstract

Abstract: State‐of‐the‐art coupled general circulation models (CGCMs) have substantial errors in their simulations of climate. In particular, these errors can lead to large uncertainties in the simulated climate response (both globally and regionally) to a doubling of CO2 . Currently, tuning of the parameterization schemes in CGCMs is a significant part of the developed. It is not clear whether such tuning actually improves models. The tuning process is (in general) neither documented, nor reproducible. Alternative methods such as flux correcting are not used nor is it clear if such methods would perform better. In this study, ensembles of perturbed physics experiments are performed with the Globally Resolved Energy Balance (GREB) model to test the impact of tuning. The work illustrates that tuning has, in average, limited skill given the complexity of the system, the limited computing resources, and the limited observations to optimize parameters. While tuning may improve model performance (such as reproducing observed past climate), it will not get closer to the "true" physics nor will it significantly improve future climate change projections. Tuning will introduce artificial compensating error interactions between submodels that will hamper further model development. In turn, flux corrections do perform well in most, but not all aspects. A main advantage of flux correction is that it is much cheaper, simpler, more transparent, and it does not introduce artificial error interactions between submodels. These GREB model experiments should be considered as a pilot study to motivate further CGCM studies that address the issues of model tuning. Plain Language Summary: State‐of‐the‐art climate models are highly complex models of the earth's climate. To achieve optimal simulations of the present climate these climate models are tuned. The tuning process as such is not well understood and it is unclear how good it performs. Alternative approaches, such as not tuning, but introducing correction terms, are less common. The study presented here illustrates the problems that tuning of a climate model introduce and shows that the alternative, of not tuning, but introducing corrections terms, is likely to be the better strategy. Key Points: Climate model tuning concepts are tested and compared versus flux corrections Climate model tuning is unlikely to improve the model significantly, but does introduce artificial errors Flux corrections are not perfect, but perform better than model tuning