The UKESM1 CMIP6 DECK is now largely complete, with the pre-industrial control (piControl) having run ~900 years and continuing. Six historical simulations are complete and a further six are running. A number of the historical runs will be extended to 2100, following a range of CMIP6 scenarioMIP emission pathways (O’Neill et al. 2016). A first set of scenarioMIP projections are planned to begin before the end of 2018. We have also completed the two climate change simulations: (i) abrupt 4xCO2 increase (abrupt-4xCO2) and (ii) 1% transient CO2 increase (1pctCO2), which help define the Effective Climate Sensitivity (ECS) and Transient Climate Response (TCR) of models. We are in the process of drafting a series of papers documenting the performance of UKESM1. These will appear in the peer-reviewed literature in 2019. Here we give a brief overview of some headline results from the UKESM1 DECK and historical runs.
UKESM1 science configuration and couplings
As a reminder, UKESM1 is built on the physical model HadGEM3-GC3.1 N96ORCA1 (Kuhlbrodt et al. 2018), extending it through inclusion of; (i) Interactive stratosphere-troposphere chemistry coupled to the GLOMAP-mode aerosol scheme (Mulcahy et al. 2018). (ii) A global carbon cycle, including terrestrial carbon processes with nitrogen limitation on carbon uptake and dynamic vegetation. Marine carbon cycle processes are represented by the MEDUSA2 model, within NEMO-ORCA1. A further configuration of UKESM1 (UKESM1-is) that includes interactive treatment of the Greenland and Antarctic ice sheets is under development, details of which can be found in the article by Smith et al. in this newsletter.
UKESM1 includes a range of couplings, between ‘physical’ and ‘Earth system (ES)’ components, as well as across domains of the coupled model (i.e. between the land, ocean and atmosphere). These couplings increase the realism (and degrees of freedom) of the model, enabling an investigation of potential Earth system feedbacks arising from future anthropogenic CO2 emissions. The primary cross-domain model coupling is CO2, exchanged between the atmosphere, ocean and land and allowing UKESM1 to run either with prescribed atmospheric CO2 concentrations or with anthropogenic CO2 emissions. Other important couplings include; (i) Dust emissions that depend on predicted vegetation cover and climate and influence aerosols and radiation processes in the atmosphere and are a source of soluble iron for the ocean. (ii) Biogenic Volatile Organic Compounds (BVOCs), emitted by vegetation and influencing model cloud-aerosol formation. (iii) Marine dimethyl sulfide (DMS) and Primary Marine Organic Aerosol (PMOA) emissions, coupled to MEDUSA-predicted seawater DMS and chlorophyll and acting as cloud condensation nuclei in the model atmosphere. Finally, concentrations of O3, CH4 and N2O, simulated by the UKESM1 chemistry scheme are active in the model radiation parameterization. This degree of coupling likely makes UKESM1 the most “Earth system complete” model in CMIP6.
UKESM1 science performance
Pre-industrial control simulation (piControl)
As stressed in earlier newsletter articles, an important requirement of an Earth system model is a temporally stable (and realistic to the degree this can be established) simulation of the (unforced) pre-industrial climate. To this end, figure 1 shows global mean values of; left; top of atmosphere (TOA) net radiation (positive values indicating downward directed radiation), middle; global mean (for land points only) surface temperature and right; global mean surface temperature (for ocean points only). The dashed lines on the figure show values from the CMIP5 model, HadGEM2-ES. The UKESM1 net TOA radiation is centred on 0Wm-2, accompanied by long-term (i.e. on multi-century timescales) stable surface temperatures. Even though the piControl does not have any time varying external forcing, there is still evidence of variability in the simulated surface temperatures, particularly in the ocean, where an oscillation with a timescale of ~100 years is evident, particularly from 500 years into the piControl. This variability is most evident in surface temperatures over the Southern ocean around Antarctica and is linked to deep ocean overturning, periodic reduction in sea-ice extent and venting of deep ocean heat to the model atmosphere.
Figure 1: Global mean TOA net radiation (left), surface temperature, land points only (middle) and ocean point only (right) from ~900 years of the UKESM1 piControl (full black lines) and 500 years of the CMIP5 HadGEM2-ES piControl (dashed line).
Figure 2 illustrates the simulated global carbon cycle in the UKESM1 piControl. The left panel shows the global mean flux of carbon from the land to the atmosphere and the right panel the ocean to atmosphere flux. Thin black lines in the figure are a measure of inter-annual variability, while the thick black line is a 50-year running mean of the annual fluxes. Both land and ocean fluxes are essentially zero on long timescales, again with evidence of multi-decadal to centennial variability around the long term mean. The unforced nature of the piControl simulations requires, on sufficiently long averaging timescales, a zero global mean flux of carbon between the atmosphere and the land/ocean reservoirs.
Figure 2: Global mean flux of carbon from land to atmosphere (left) and ocean to atmosphere (right) from the UKESM1 piControl. Thin lines are inter-annually varying fluxes, thick black line is a 50 year running mean flux and the red full line indicates the long term mean flux.
Extending the analysis of the UKESM1 carbon cycle, figure 3 shows cumulative global mean land carbon uptake (top panel) in the UKESM1 CMIP6 historical simulations (covering the period 1850-2014) and the cumulative marine carbon uptake (right panel) for 1950 to 2014 of the same simulation. Three UKESM1 members are show by the black lines in figure 3. Observational estimates for both carbon uptakes are shown by the vertical blue lines on the figure, centred on year 2005. The pink plume shows the spread in the CMIP5 multi-model ensemble.
Figure 3. Top: Global mean cumulative terrestrial carbon uptake 1850-2014 from three UKESM1 historical simulations (black lines). The pink plume spans the CMIP5 multi-model ensemble and the blue vertical line is an observed range for 2005. Middle: As top panel but cumulative ocean carbon uptake for 1950-2014. Bottom: Diagnosed UKESM1 (black) carbon emissions compatible with the prescribed atmospheric CO2 concentrations and simulated carbon uptake from the atmosphere. Observed carbon emissions are shown by the white dashed line and the CMIP5 ensemble by the pink plume.
UKESM1 terrestrial uptake sits close to the centre of the observed range for 2005, with the land losing carbon to the atmosphere until ~1980 (the approximate time where the black lines reach a minimum value in the figure), before becoming a sink for atmospheric carbon. The loss of terrestrial carbon is due to land use change (primarily loss of forest cover) while the later terrestrial uptake of carbon is due to the increasing atmospheric concentrations driving an atmosphere to land gradient in carbon and a resulting flux greater in magnitude than the carbon loss from land use change. Simulated ocean uptake is at the lower end of the 2005 observational estimates, but still within the observed range. The ocean has acted as a strong sink for atmospheric carbon from at least 1950 through to present day. Combining these two uptake terms with the atmospheric CO2 concentrations prescribed in the CMIP6 historical runs, allows us to diagnose the historical (anthropogenic) carbon emissions compatible with the prescribed atmospheric CO2 concentrations. These are shown in the lower panel of figure 1 (black full lines) and compared with actual historical emissions (dashed line). The UKESM1 compatible emissions track the observed emissions closely, suggesting the global-scale carbon cycle is accurately simulated. This will be important when we run the model in emission-driven mode. Here we do not prescribe atmospheric CO2 concentrations but rather start the model from pre-industrial (PI) conditions and a PI estimate of atmospheric CO2 and prescribe anthropogenic historical emissions, allowing the models carbon cycle to determine where this carbon goes between the Earth’s reservoirs and on what timescales.
In figure 4 we show another important performance metric for UKESM1, namely the model’s ability to simulate the Antarctic ozone hole. We show the temporal evolution of total column ozone at the South Pole, simulated in two UKESM1 historical runs and from observations, the latter beginning in 1964. Monthly mean column ozone is shown for September, October, January and February through the entire historical simulation period. From the early 1970’s both the model and observations depict a decrease in column ozone at the South Pole (indicative of the onset of the stratospheric ozone hole). This decrease reaches a minimum, both in the model and observations, around 2005. The observed annual cycle of the ozone hole shows a rapid decrease through September and October (the Antarctic spring) and subsequent dissipation, as the polar vortex breaks up in the following January and February (the Antarctic summer). Both the annual cycle of the growth and decay of the ozone hole and its overall development, from initiation in the early 1970’s to minimum values around 2005, are well simulated. The overall magnitude of the ozone decrease also appears well captured by the model. For example, October column ozone decreases from ~310 Dobson units (DU) in the mid-1960s to a minimum of ~125 DU by ~2005, in line with observations.
Figure 4. Total column ozone at the South Pole in two UKESM1 CMIP6 historical simulations (green and blue lines) and observed at the South Pole (1964 to present-day, black crosses). Monthly mean ozone values are shown for September, October, January and February.
Idealized climate change experiments (abrupt-4xCO2 and 1pctCO2)
In the abrupt-4xCO2 experiment, we start from a UKESM1 piControl model state and instantaneously quadruple the amount of atmospheric CO2. This results in a large (downward) net radiation imbalance at the TOA, which leads to a warming. After sufficient simulation years, the model will come back to zero TOA net radiation balance, with a warmer climate and a an increased emission of longwave radiation at TOA. As this adjustment to the model’s equilibrium climate response can potentially take thousands of simulation years, Gregory et al. (2004) developed a method to estimate the Effective Climate Sensitivity (ECS) of a model to an external forcing perturbation, such as a quadrupling of CO2. This method regresses the TOA radiative flux perturbation at the time of CO2 quadrupling against the global mean surface temperature change. Extrapolating the linear regression (best-fit line) to a zero TOA radiative perturbation provides an estimate of the model ECS to a quadrupling of CO2. Halving this value gives the more traditional ECS for a doubling of CO2.
Figure 5 shows the change in global mean surface temperature (dT) in UKESM1 as a function of simulation year (from a nominal start date of 1850) in the abrupt-4xCO2 experiment. Two UKESM1 realizations are shown (blue and red lines), compared to the HadGEM2-ES response and the range of CMIP5 models. The right panel shows the linear regression between the net TOA radiation imbalance (dN) and the surface temperature change (dT). This gives an estimate of the UKESM1 2xCO2 ECS of ~5.3K. The left panel indicates this response is larger than seen in the CMIP5 multi-model ensemble. There is a strong indication (not yet published) that a number of other CMIP6 models will have ECS values higher than the upper end of the CMIP5 range.
Figure 5: Left: Global mean surface temperature (GMST) response simulated by UKESM1 (red and blue lines) to an instantaneous quadrupling of atmospheric CO2. HadGEM2-ES is in black and the CMIP5 multi-model ensemble the light grey lines. Right: Linear regression between TOA net radiation perturbation (dN) and the change in GMST (dT) in two UKESM1 abrupt-4xCO2 experiments. ECS is estimated from 150 years of simulation.
Another important measure of a model’s sensitivity to increasing CO2 is the transient climate response (TCR). TCR is defined as the global mean temperature change, averaged over a twenty-year period centred on the time of CO2 doubling, in a transient simulation with CO2 increasing at 1% per year from PI values (Randall et al. 2007). Figure 6 shows two UKESM1 1pctCO2 experiments, again compared to the same simulation using HadGEM2-ES and the CMIP5 multi-model ensemble. The final UKESM1 TCR will be derived from a 4 member ensemble of 1pctCO2 experiments. From these two initial experiments the TCR lies in the approximate range 2.6-2.9K, slightly higher than HadGEM2-ES and at the upper edge of the CMIP5 range. Work is ongoing to understand the various feedbacks leading to the ECS and TCR in UKESM1 and will be reported in the peer-reviewed literature in 2019.
Figure 6: Global mean surface temperature (GMST) response in the UKESM1 1pctCO2 experiment (red and blue lines), compared to HadGEM2-ES (black) and the CMIP5 mutli-model ensemble (grey lines).
We are nearing completion of the UKESM1 CMIP6 DECK and historical simulation set. Output from these runs will be submitted to the UK Earth System Grid Federation (ESGF) node early in 2019. The UKESM team, along with collaborators at the Met Office, NERC centres and UK universities are actively analysing these simulations and papers documenting the performance of the model will appear in the scientific literature over the coming year.
- Gregory, J.M et al. (2004). A new method for diagnosing radiative forcing and climate sensitivity. Geophysical Res. Letts. 31 (3): L03205. doi:10.1029/2003GL018747
- Kuhlbrodt T. et al. (2018). The Low‐Resolution Version of HadGEM3 GC3.1: Development and Evaluation for Global Climate. J. Advances in Modeling Earth Systems. Published online: 29/10/18 https://doi.org/10.1029/2018MS001370
- Mulcahy J.P. et al. (2018). Improved Aerosol Processes and Effective Radiative Forcing in HadGEM3 and UKESM1. Published online 15/10/18: https://doi.org/10.1029/2018MS001464
- O’Neill, B. C. et al. (2016): The Scenario Model Intercomparison Project (ScenarioMIP) for CMIP6, Geosci. Model Dev., 9, 3461-3482, https://doi.org/10.5194/gmd-9-3461-2016.
- Randall, D.A.; et al. (2007). 8.6.2 Interpreting the Range of Climate Sensitivity Estimates Among General Circulation Models, In: Climate Models and Their Evaluation. In Solomon, S. et al. Climate Change 2007: Synthesis Report. Contribution of Working Groups I, II and III to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change.