Interactions between perturbations to different Earth system components simulated by a fully-coupled climate model

Abstract

We examine the global mean surface temperature and carbon cycle responses to the A1B emissions scenario for a new 57 member perturbed-parameter ensemble of simulations generated using the fully coupled atmosphere-ocean-carbon cycle climate model HadCM3C. The model variants feature simultaneous perturbation to parameters that control atmosphere, ocean, land carbon cycle and sulphur cycle processes in this Earth system model, and is the first experiment of its kind. The experimental design, based on four earlier ensembles with parameters varied within each individual Earth system component, allows the effects of interactions between uncertainties in the different components to be explored. A large spread in response is obtained, with atmospheric CO₂ at the end of the twenty-first century ranging from 615 to 1,100 ppm. On average though, the mean effect of the parameter perturbations is to significantly reduce the amount of atmospheric CO₂ compared to that seen in the standard HadCM3C model. Global temperature change for 2090–2099 relative to the pre-industrial period ranges from 2.2 to 7.5 °C, with large temperature responses occurring when atmospheric model versions with high climate sensitivities are combined with carbon cycle components that emit large amounts of CO₂ to the atmosphere under warming. A simple climate model, tuned to reproduce the responses of the separate Earth system component ensembles, is used to demonstrate that interactions between uncertainties in the different components play a significant role in determining the spread of responses in global mean surface temperature. This ensemble explores a wide range of interactions and response, and therefore provides a useful resource for the provision of regional climate projections and associated uncertainties.

Appendix: Spinning up the ESPPE simulations

ESPPE runs were started from a stable flux-adjusted control state of HadCM3C with standard parameters. In the initial “Haney” phase, sea surface temperatures (SSTs) and salinities were strongly relaxed towards pre-industrial conditions of 1860, using a relaxation coefficient for surface forcing of temperature H ₀ = 81.88  Wm⁻²K⁻¹ (Haney 1971; Collins et al. 2011a). Atmospheric CO₂ concentration was held constant at an estimate for 1860 of 286.56 ppm, allowing the land and ocean models to take up or release the CO₂ necessary to come into equilibrium with pre-industrial atmospheric conditions. We set equilibrium criteria for both the land and ocean of carbon uptake or loss of less than 0.4 GtCyr⁻¹ (equivalent to a change of about 0.2 ppmyr⁻¹ in atmospheric CO₂ concentration), and for the vegetative and soil fractions of land carbon storage to be stable. The uptake and loss criterion for the land and ocean was chosen to ensure that any residual flux was smaller than 10 % of anthropogenic emission during the 1970’s (Ian Totterdell, pers. comm., 2008). It was found that most ocean models needed to expel a large amount of CO₂ to approach equilibrium, requiring hundreds or even thousands of years of model integration. Fortunately, it turned out that intra-ensemble differences in ocean carbon cycle behaviour were largely controlled by two parameters that control horizontal and vertical mixing in the ocean. The first allows the sub-grid scale eddy driven ocean mixing resulting from baroclinic instability to depend on the local Richardson Number (Visbeck et al. 1997). The second affects the vertical mixing in the mixed layer and its interaction with the deeper ocean, where the parameter determines whether the LARGE scheme (Large et al. 1994) is implemented using a quadratic or cubic function for the shape function used to determine how the vertical diffusivity in the mixed layer varies with depth. By running long spin-up simulations sampling each of the four possible permutations (several thousand years) the resulting ocean carbon states could be substituted into all of the other ensemble members that shared those permutations. Subsequent spin-up of the larger ensemble subsequently only required 300–400 years for the ocean carbon stores to adjust to the other parameter perturbations and allowed ensemble members to quickly approach our ocean carbon criterion for the historical simulation. The land carbon cycle spins up rapidly, as the TRIFFID vegetation model (Cox et al. 1998; Cox. 2001) is able to run thousands of years very rapidly by coupling to the atmosphere only periodically.

Having achieved a stable surface temperature, ocean carbon state, and land carbon and vegetation states, the ocean heat and salinity fluxes that maintain the control climate during the last 50 years of the Haney phase were averaged to produce seasonally varying adjustment fields, updated every 5 days. The model was then run in “flux-adjusted” mode, where the correction fields were applied at the ocean surface rather than using the Haney term to relax to observed climatology. Use of flux-adjustments rather than relaxation allows the model to show more unforced variability, while remaining close to 1860 conditions on decadal timescales. More importantly, it allows us to run forced climate change experiments for different model parameters starting from realistic, stable control climates. Typically, on application of the flux adjustments there is a small adjustment in global mean surface temperature, and a small effect on ocean carbon uptake that quickly stabilizes.

Finally, the runs were put into “free-atmosphere” mode, where atmospheric CO₂ is allowed to respond to uptake and loss by the ocean and land surface. Further small adjustments to surface temperature and oceanic and land surface CO₂ uptake followed. CO₂ concentration stabilized at values ranging from 282–291 ppm, similar to ice core observations for 1860 (Etheridge et al. 1996). Final global mean control temperatures ranged between 286–288 K. In all but four cases, ocean heat uptake stabilizes at less than 0.3 Wm⁻². These simulations provided starting conditions for the ESPPE simulations of forced climate change.