Climate modeling has become a crucial tool for research on climate and global change. Correct representation
of clouds in climate models is a question of great importance, given the potentially high impact of clouds upon climate change.

“The impact of clouds upon climate change is a very complex issue”, Järvinen explains. “On one hand, clouds trap emitted infrared radiation and thus tend to reduce the top-of-theatmosphere infrared radiation. On the other hand, they reflect solar radiation. The net effect, so-called ‘cloud radiative forcing’ is – at present – negative, i.e. in an undisturbed climate clouds tend to have a cooling effect on the planet.”

“Changes in the global climate will, however, undoubtedly also impact upon the ways in which clouds and temperature interact, known as cloud feedback” Järvinen adds; “whether the net effect of this impact is to change that feedback from negative to positive will depend, amongst other things, upon whether the cloud amount or water content, or their vertical distribution with altitude, change significantly.”

While an adequate representation of clouds in climate modelling is crucial, many important processes that occur in the atmosphere, such as the interaction of clouds with the radiation emitted by the Earth-atmosphere system, are still unresolved by current computer-driven models.

“Atmospheric General Circulation Models (GCMs) – which model the atmosphere – are the basic modelling tools used in climate research and climate simulation. However, the representation of clouds and their interaction with radiation remains a major issue, due to the coarse spatial resolution of this particular model: a grid cell typically covers an area of 200 km by 200 km in the horizontal. This implies that many cloud features important for radiation cannot be explicitly resolved. This is quite problematic as these interaction processes are essential to the Earth’s radiation budget which, in turn, largely determines the Earth’s climate.”

“There is, therefore, a strong need to improve the representation of clouds in climate models, and one could even say that a biased cloud scheme can ruin a climate simulation resulting from an otherwise accurate climate model.”

Stochastic cloud modelling in a coupled climate model

“In climate models, clouds are usually computed using grid box average quantities, such as grid box mean temperature and humidity. In our study, cloud formation is described as a stochastic process. It mimics the chain of events as they occur in nature but is inherently random. The need to preserve the grid-box mean quantities, however, represents a powerful constraint on the process.”

“Our study is based on several recent developments – from our own work, and that of other scientists – that enable us to address cloud subgrid-scale variability in climate models. The Tompkins cloud scheme (a statistical cloud scheme developed by Adrian Tompkins for data assimilation) allows us, for example, to derive the magnitude of subgrid-scale variations in cloud water amount in a physically consistent fashion. The Monte Carlo Independent Column Approximation (McICA) (a recently developed method for computing domain-average radiative fluxes) and the stochastic cloud generator enable a flexible description of subgrid-scale cloud structure in GCM radiation calculations.”

“The SSSC project represents the first test of these approaches in long integrations within a coupled atmosphere-ocean GCM,” Järvinen points out.

“The first tests with the stochastic subgrid-scale cloud modelling have been successful. In these tests, the atmospheric GCM used the observed sea surface temperature and sea-ice cover, which implies a forcing of the atmospheric simulation towards the observed evolution of the climate. In the SSSC project, the atmospheric GCM was coupled with an oceanic GCM and these were run in a coupled mode. In this case, there is no forcing towards the observed climate evolution, excepting perhaps the observed atmospheric CO2 concentration. A coupled atmosphere-ocean GCM was particularly required for our study, in order to avoid use of fixed sea surface temperature of an atmosphere- only GCM, which would have limited the validity of the simulations.”

“In a nutshell, our study aimed at exploring the use of an advanced treatment of subgrid-scale cloud radiation effects in climate models, by using a coupled atmospheric-ocean model. Its scientific objectives were twofold: first, to test and demonstrate the viability of the stochastic approach for the sub-grid-scale cloud and radiative transfer parametrizations; and second, to use this approach to improve the representation of cloud-radiation
interaction in GCMs.”

Computational resources

In order to carry out this study, a large amount of computational resources was required. Researchers used DEISA resources within the DECI framework and experiments were conducted in 2008 on the NEC-SX-8 at the High Performance Computing Center (HLRS) in Stuggart, Germany, where the Finnish team coupled the general atmospheric circulation model ECHAM5 to the general ocean circulation model MPIOM, two models that
were developed by the Max-Planck-Institute for Meteorology (MPI-M) in Hamburg, Germany.

“This model setup is very time-consuming for any computer and, in practice, it would have been impossible for us, using the resources available to us at our Institute in Finland alone, to conduct this experiment to the extent made possible by the DEISA quota,” Järvinen ackowledges.

“Three 240-year long experiments were carried out with three specific model configurations. The first experiment involved the standard version of the ECHAM5-MPIOM coupled atmosphere- ocean GCM; the second used a version employing the Tompkins cloud scheme; and the third a version employing the Tompkins cloud scheme together with McICA radiation calculations and the stochastic cloud generator.”

“On the NEC-SX-8 processors, each run took between 3 and 4 hours per simulated year Unexpected problems
regarding the parallelization of the coupled model across several 8-processor nodes precluded the use of a larger number of processors, but the experiments were completed successfully nevertheless,” says Järvinen.

Comparison of simulated time-mean near-surface air temperatures in Celsius. Figure EXP1 is computed with standard version of the coupled model. In figure EXP2 Tompkins cloud scheme is included. In figure EXP3 Tompkins cloud scheme included and subgrid-scale cloud information utilized in radiation calculations through the stochastic cloud generator and Monte Carlo Independent Column Approximation. The data used in these simulations is produced by the European Centre for Medium Range Weather Forecasts. © Heikki Järvinen

The results, however, proved to be quite unanticipated for the research team: “Perhaps surprisingly, the modifications tested here had overall a relatively small effect on simulated climate. Given that this new cloud modelling method represented quite a significant change compared to the standard model, we had anticipated that it would have a larger impact. Nothing major was detected, however, and this was rather unexpected.”

These unexpected results did not, however, mean that the experiment was worthless; quite to the contrary: “While this might seem like an unexciting result, it is actually an important indicator of the feasibility of the proposed stochastic approach in state-of-the-art climate models”, Järvinen points out. “It shows, in particular, that the new cloud modelling method can be safely included into the coupled climate models, and that its benefits can begin to be exploited. The new approach improves substantially the internal physical consistency of the model, and allows also us to produce a stable model climate. It is therefore a viable option for application in multi-century climate simulations.”

“We are now equipped with convincing evidence of the applicability of our approach within GCM implementations. Hopefully this work will contribute towards improving the models that will be used in the 5th assessment report of the Intergovernmental Panel on Climate Change (IPCC).”

Damien Lecarpentier
Photo of Heikki Järvinen: Antonin Halas/Studio Halas

More information on DEISA

The DEISA supercomputer environment is suitable for demanding projects that necessitate large computing capacity, ones that would be impossible to implement without the resources offered by DEISA. Research projects carried out using the DEISA resources are computationally amongst the most challenging ones. These projects represent nanosciences, materials sciences, and bioinformatics, among others. DEISA is the only body that offers HPC capacity for all European researchers against an application. Slightly less than ten percent of the supercomputers’ run time capacity is reserved for DEISA projects. In addition to CSC, supercomputer centers representing the Netherlands, Spain, UK, Italy, Germany, and France participate in DEISA.

http://www.deisa.eu/

http://www.csc.fi/english/research/Computing_services/grid_environments/deisa/index_html