We are nearing the completion of UKESM1 development, with the planned science capabilities now included and all components coupled together. What remains before finalising the model is to calibrate the fully coupled system to achieve good performance across a range of observed climate indicators, such as radiative fluxes, temperature and rainfall. This calibration step is usually referred to as “tuning” in the climate model development community. There are parameters in all climate models which are only weakly constrained by observations, and these are often adjusted to improve the overall performance of the model. Tuning is an essential final step in the development of all climate models; for more background see for example Schmidt et al (2017) and Hourdin et al (2016).
The individual component models of UKESM (such as atmospheric chemistry, and terrestrial and ocean biogeochemistry) have been developed such that they perform well in isolation, but within the coupled Earth system model they interact with other components and the performance can change markedly. In some cases small biases are amplified by feedbacks between components, while in other cases a bias in one component (deemed acceptable in its own right) has a disproportionate impact on the components, when coupled. A notable example for HadGEM2-ES was the impact of tropical precipitation biases upon dynamic vegetation and hence surface albedo and dust emissions (Martin and Levine, 2012). The tuning phase then involves work to understand and reduce these/such biases, and in some cases tuning the impacted component to compensate for the bias it is sensitive to. The latter form of tuning has to be treated with care, as it has the potential “get it right for the wrong reasons” and to compromise the model’s response to climate change, though this depends greatly on the nature of the parameters concerned. In choosing parameters to tune, we are guided by deep understanding of the formulation of the model, taking advice from component model experts both within and outside of the UKESM team. Where possible we also make use of the extensive perturbed-parameter ensembles performed by the Met Office and NCAS.
An example of a bias we are tackling in the final stages of UKESM1 development relates to surface albedo in northern mid-latitudes. Figure 1 shows differences between HadGEM3-GC3.1 and UKESM1 for top of atmosphere clear-sky outgoing shortwave radiation. The large differences in N America and Asia in the top-right plot arise from large differences in surface albedo, which has a detrimental impact on near-surface temperatures in these regions. The underlying cause is a bias in the vegetation distribution in the prototype model. UKESM includes dynamic vegetation, which responds to climate change, CO2 fertilisation and land-use change, but the inclusion of additional interactive model components invariably introduces biases. In this case there is too widespread cover by trees and shrubs where there should be grass. These vegetation types are less readily covered by snow in winter, leading to lower wintertime surface albedo on average in these regions.
Our approach to reducing the albedo bias has two strands. Firstly we are tuning vegetation dynamics parameters, such as mortality rate, to improve the vegetation distribution. Secondly, we will tune parameters which govern the interaction of snow cover and the vegetation canopy, such as the proportion of vegetation visible above a certain snow depth, and the rate at which accumulated snow unloads from needleleaf trees. The first approach can be viewed as addressing the underlying cause of the bias, while the second aims to mitigate its impact on radiation fluxes. In both cases we will constrain the tuning parameters within plausible physical limits. Sensitivity tests indicate that we can make significant improvements through this approach, and we are optimistic of good radiation performance in UKESM1.
Figure 1. Top of atmosphere clear-sky outgoing shortwave radiation (December-January-February average) for the prototype model version UKESM0.6 and the physical model configuration on which it is based, HadGEM3-GC3.1. a) UKESM0.6. b) difference between UKESM0.6 and HadGEM3-GC3.1. c) HadGEM3-GC3.1 errors against CERES-EBAF satellite observations. d) UKESM0.6 errors against the same observations.
Hourdin, F., T. Mauritsen, A. Gettelman, J.-C. Golaz, V. Balaji, Q. Duan, D. Folini, D. Ji, D. Klocke, Y. Qian, F. Rauser, C. Rio, L. Tomassini, M. Watanabe, and D. Williamson (2016, July). The art and science of climate model tuning. Bull. Amer. Meteor. Soc. 98 (3), 589-602.
Martin, G. M. and Levine, R. C.: The influence of dynamic vegetation on the present-day simulation and future projections of the South Asian summer monsoon in the HadGEM2 family, Earth Syst. Dynam., 3, 245-261, doi:10.5194/esd-3-245-2012, 2012.
Schmidt, G. A., Bader, D., Donner, L. J., Elsaesser, G. S., Golaz, J.-C., Hannay, C., Molod, A., Neale, R., and Saha, S.: Practice and philosophy of climate model tuning across six U.S. modeling centers, Geosci. Model Dev. Discuss., doi:10.5194/gmd-2017-30, in review, 2017.
Williams et al, in prep. JAMES special issue on HadGEM3 and UKESM1 contributions to CMIP6.