This page describes the first of our secondary experiments. In this experiment we investigated the effects of a slowdown of the North Atlantic meridional overturning circulation, also known as the thermohaline circulation (THC).

This sub-experiment was led by Nick Faull (climate science) and Tolu Aina (computing).

The experiment was not a prediction of the likelihood of a reduction in the strength of the THC. In this experiment we imposed a reduction of the THC consistent with earlier experiments with the Hadley Centre coupled model and studied the atmospheric response. We used the Hadley Centre atmospheric model in conjunction with a simplified thermodynamic (“slab” ocean) which comprised of a single layer ocean with prescribed heat and salinity transports. We imposed surface fields which reflected the fully-coupled model’s response to an imposed THC slowdown. The experiment was consistent with previous coupled model work with the same model at the same resolution. The essence of our THC experiment was to look at how the atmosphere would respond to such changes in the ocean, given a THC slowdown.

Motivation of the THC experiment

The motivation behind the THC expt was to look at the sensitivity of the THC to anthropogenic climate forcing. Rapid climate changes have occurred in the geological recent past, and their coincidence in time with the creation of the North Atlantic Deep Water formation, suggests an active role of the THC in these rapid climate changes. Modelling results show a wide range of responses of the THC to high latitude warming and freshening, characteristic of global warming scenarios. Many models indicate a modest weakening of the THC, but some models produce a substantial breakdown of the THC under sufficiently strong forcing.

The uncertainty regarding the response of the THC to anthropogenic forcing is crucial in identifying and understanding the physical processes which control the stability of the THC. By now it is generally accepted that past changes in the THC have been triggered by large amounts of freshwater discharge from continental ice sheets in the North Atlantic. However, in the future, enhancements in the hydrological cycle may change the freshwater budget in the North Atlantic sufficiently to induce entirely different dynamics of the THC. In this context, it is not unlikely that important components of the climate system are remotely controlled. [An example of this is the suggested impact of the SST pattern in the Pacific on the freshwater budget in the North Atlantic, and consequently on the THC.] Yet another motivation to study the stability of the THC is the evidence compiled by models of various complexity, that the THC may not be far away from an instability threshold and that we may be approaching it as a result of greenhouse warming.

Full Description of the THC experiment

A key determinant of possible changes in the THC over the coming century is the atmospheric hydrological cycle response both to externally driven (predominantly CO2-induced) temperature changes and to THC changes themselves. The hydrological cycle response to CO2, particularly the extent of freshwater forcing change in the North Atlantic, determines the size of what is likely to be the largest externally imposed forcing on the THC over the coming decades. The hydrological cycle response to surface heat fluxes accompanying any THC change (whether internal or externally generated) determines whether the atmosphere acts to reinforce or damp the development of ocean circulation anomalies.

This is an area of considerable uncertainty at present, for two main reasons.
Precipitation changes are extremely sensitive to the details of underdetermined parameterisations in models, particularly at the regional level, reflected in the much wider spread of precipitation responses to increasing CO2 relative to temperature responses in, for example, the CMIP-2 model inter-comparison project.

The mechanisms controlling the large-scale inter-annual hydrological cycle response to external forcing are such that most of the precipitation changes observed over the 20th century have, in contrast to recent temperature changes, been driven by natural (shortwave) forcing. Hence, the recent observational record provides a much less direct constraint on future precipitation changes than it does on future temperature trends. The balance between shortwave forcing, longwave forcing and thermal response will almost certainly change substantially over the next 30 years relative to the past 30 years, leading to much greater uncertainty in future precipitation changes relative to future temperature changes.

Previous model inter-comparison results make clear that any study of the large-scale hydrological cycle response to external forcing must account for model uncertainty. “Perfect-model” studies that are based solely on one specific representation of the atmosphere-ocean system miss, by construction, what is likely to be the most important source of uncertainty in the problem. Accounting for model uncertainty in climate prediction is still in its infancy, but recent years have seen considerable progress with the development of the first “perturbed physics” ensemble forecasting systems for the analysis of the response to anthropogenic (CO2) forcing. This project was the first to extend the perturbed-physics ensemble methodology to study the role of the hydrological cycle in possible rapid climate change.


In essence, our methodology extended the project, using distributed computing to build a perturbed-physics ensemble simulation of the range of responses to increase CO2, to recent studies by Dong and Sutton (2002) and Palmer (2002) examining the response of a particular atmospheric model to changes in THC. Under, a current climate-resolution atmospheric general circulation model (the standard version of the HadAM3 model, with longitude/latitude resolution of 3.75 by 2.5 degrees and 19 levels in the vertical) was coupled to a simple thermodynamic “slab” ocean model. The properties of the model atmosphere were then changed using some combination of perturbations to over 20 parameters (including switches that allowed entire model sub-systems to be activated or deactivated) and the resulting perturbed model was distributed to a volunteer participant in the project to be ran on a personal computer.

The standard experiment comprised a 15-year integration during which sea-surface temperatures were strongly relaxed back to observed present-day conditions and the Ocean Heat Flux Convergence (OHFC) field was calculated to balance the resulting down-welling surface heat flux from the atmosphere. This was followed by a 15-year “control” phase with pre-industrial CO2 levels, and a 15-year “doubled-CO2” phase in which CO2 levels were increased to 560ppmv. A standard diagnostic set comprising approximately 0.5Gb of output was archived on the participant’s personal computer while a set of summary diagnostics (~5Mb per experiment) was automatically uploaded to one of the projects’ servers at the end of the experiment. These summary diagnostics provided an indication of the range of behaviour across the ensemble as well as highlighting particularly interesting parameter combinations for which we could either request the full archive to be uploaded (if the participant’s internet connection allowed) or repeat the run in-house.

In this revised experiment, an additional OHFC perturbation was applied at the end of the doubled-CO2 phase of the standard experiment to represent the change in net ocean-atmosphere surface heat flux that might result from a substantial reduction in the THC and the model integrated for a further 15 years. Although most models would not all have reached equilibrium by the end of this period, feedbacks in a slab-ocean set-up were sufficiently close to linear in the temperature change for the degree of disequilibrium to be inferred from the time-history and energy budget of the response. As an additional check, however, the OHFC perturbation phase was started from the end of the 14th year of the doubled-CO2 phase thereby ensuring a 1 year overlap between the end of one phase and the beginning of the next. With an initial condition ensemble this overlap provided information about linearity of the temperature response when the OHFC perturbation was applied.

This heat flux perturbation was derived directly from the work of Palmer (2002), who used a multiple regression analysis to separate out the OHFC changes associated with wind-stress curl anomalies (which affect any index of THC but are not directly relevant to the thermo-haline signal) from the OHFC anomaly associated with THC fluctuations in the HadCM3 coupled model control variability.

Palmer (2002) performed an experiment in which this OHFC response-pattern was scaled to correspond to a ~50% reduction in THC magnitude from HadCM3 control values and obtained a ~2K global mean cooling in response, as compared to Vellinger and Wood (2002) who induced a near-complete THC shut-down associated with a ~4K global mean cooling. We proposed initially to use double Palmer (2002)’s perturbation, although the perturbed-physics ensemble set-up made imposing a variety of OHFC perturbations very straightforward, and sensitivity of response to the imposed pattern and amplitude was explored in the course of this project. Following Palmer (2002), the imposed OHFC perturbation was seasonally varying.

Comparison of the mean precipitation response in the region of the North Atlantic where Vellinga and Wood (2002) imposed a freshwater forcing to induce an THC shutdown revealed that the magnitude (and possibly even the sign) of the atmospheric hydrological-cycle feedback on THC changes is highly sensitive to underdetermined aspects of atmosphere model formulation. Under this project, we used diagnostics returned from the perturbed-physics ensemble to identify parameter combinations which result in atmospheres that are (1) sensitive to CO2 forcing; (2) sensitive to THC changes with the precipitation response giving a positive feedback and (3) relatively realistic in terms of their base climate. We assessed to what extent current observations allowed us to constrain the ensemble in these “pathological” parameter-space directions and provide a range of representative statistical models of possible atmospheric responses to internal ocean variability, CO2 and THC changes for use as upper boundary conditions in ocean THC predictability studies.