DYAMOND – an initiative which pushes the limits

Since 2017, ESiWACE has been co-organising and substantially supporting DYAMOND, an intercomparison initiative of global storm-resolving models, proposed in October 2017 by Bjorn Stevens (MPI-M) and Masaki Satoh (University of Tokyo).

The DYAMOND initiative is the first global high-resolution simulation intercomparison of its kind. Never before had there been storm-resolving simulations that performed the same experiment.

Why are global storm-resolving simulations of importance?

In 2017 and still today, typical climate models, simulating the whole globe, have a horizontal resolution of 40 to 80 km. But this coarse resolution does not allow for an explicit simulation of the crucial vertical energy transfer which determines the deep convection mechanism in the atmosphere and especially in the tropics. Instead of explicit simulation parametrization of this mechanism has to be used.
The global storm-resolving DYAMOND simulations, with a horizontal resolution of 9 to 1.5 km, for the first time allowed scientists to study the interaction of the vertical energy transport in convective storm systems with the global-scale circulation (see Fig.1). Thus, these simulations are providing an unprecedented opportunity for the study of climatic extreme events and their future evolution in a changing climate.
Another advantage of this high horizontal resolution is: comparison with observational data becomes possible. As no analytical solutions for the whole climate system exist, comparisons with observational data as well as model intercomparisons are essential for assessing climate models.

Visualisation of convective processes based on the output of the ICON simulations with 156 m resolution (top) and 40 km resolution. The colours denote ice (pink), liquid cloud water (grey) and precipitation (blue). [Stevens et al. 2020].

Therefore, DYAMOND brings the participating climate models a further big step towards proven climate simulations.

New technical challenges

Surely, the increased horizontal resolution of a global simulation leads to an increased amount of compute and memory resources. This new kind of simulations became only possible by the growth of supercomputer capacities and an emerging class of atmospheric circulation models, which can make efficient use of these capacities.
Also the increasing amount of output data, up to 200 Terabytes per DYAMOND simulation, brings up new challenges. They range from fast and efficient output writing during the simulation, via data provision in a well-arranged manner for, a broad community, to new tools for fast data loading, efficient calculation and plotting mechanisms.

This ambitious scientific initiative perfectly fits to the ESiWACE2 objectives and pushes the boundaries of the models, the skills of the scientists involved and the data-providing institutions.

Two successful phases

Up to now, two experiment phases have been implemented: DYAMOND Summer and DYAMOND Winter. Both experiments cover 40 days, either during the boreal summer or the boreal winter period. In total more than 50 simulations simulated by 16 different models, which are being developed in Europe, Japan, USA and China, have been contributed, and up to now 2.7 Petabytes of output data have been made available by ESiWACE.
In the first phase, DYAMOND Summer, modelers overcame the challenge of reaching this high horizontal resolution for the first time. And at the end, all simulations showed a good agreement of mean precipitation with observational data without tuning simulation parameters [Stevens et al. 2019].
To bring atmospheric climate models another step forward, the second phase, DYAMOND Winter asked for coupled atmosphere-ocean simulations. Aim is to get closer to storm- and ocean eddy resolving-climate models, allowing scientists to investigate the atmosphere-ocean interaction on the convective scale. DYAMOND Winter was aligned with the EUREC4A field study, one of the most important campaigns to study the interplay between clouds, convection and circulation and their role in climate change, and provided a good opportunity to validate the DYAMOND climate simulations.
In two Hackathons and so far more than 20 scientific publications, including a special edition (2019-2021) in the Journal of the Meteorological Society of Japan (JMSJ), the analysis of this kind of data set has been shown as feasible. DYAMOND data is still being analysed, although first contributions are now already 4 years old.

So far and beyond

As DYAMOND experiments ask for pushing the limits of atmospheric climate models, the simulation results have shown that global storm-resolving models are becoming possible and increasingly feasible. ESiWACE, as co-organiser and supporter of this ambitious initiative, enables access, compute resources for data analysis, tutorials for data handling and training of scientists in the field of model optimisation.
Furthermore, DYAMOND simulations have served as a starting point for the even more ambitious project nextGEMS. NextGEMS aims for global decadal simulations at storm-resolving resolution, which will further reduce the need for parametrization by direct simulation of the physics process and therefore increase the model validity. These global storm-resolving simulations will help us to gain new insights and address long-standing questions with new tools.