Because of the very complex interconnected nature of climate systems, much of the work of studying what happens to them and why is done in supercomputers, although limitations in computing power have hitherto limited the possibilities for such simulations.
The situation has now changed. And a team of climatologists from the Swiss Higher Technical School of Zurich (ETH Zurich) was able to use supercomputer power to create ultra-high-resolution climate models covering all of Europe and the central Atlantic Ocean.
When analyzing climate and weather, low resolution can pose serious problems, inhibiting the modeling of critical arrays such as clouds and storms. In turn, this generates unevenness in modeling, which leads to great uncertainty. For example, estimates of the future temperature rise on Earth if the CO2 in the atmosphere will double, ranging from 1.5 ° C to 4.5 ° C. This is a huge range, which, according to Christoph Schär, professor of climatology at ETH Zurich, “mainly associated with low resolution models of the current climate“, The best of which often still have a wide range from 12 to 50 km.
Cher and his colleagues decided to change this paradigm. They are collaborating with the Swiss National Supercomputing Center (CSCS) and MeteoSwiss (Switzerland’s national meteorological office) to adapt the popular COSMO non-hydrostatic atmospheric model, previously only for CPU computing, for use on GPUs – a move that Scher says, “makes computing more efficient, faster and cheaper“.
The Partnership for Advanced Computing in Europe (PRACE) has provided scientists with the ability to operate the Piz Daint supercomputer at the Swiss National Supercomputing Center (CSCS). Piz Daint contains 5704 Cray XC50 nodes (each with Intel Xeon E5-2690 and NVIDIA Tesla P100) in addition to 1813 XC40 nodes (each with two Intel Xeon E5-2695). Piz Daint delivers 21.2 Pflops of Linpack performance, placing it among the top ten most powerful supercomputers in the world in the latest Top500 ranking.
Using COSMO and Piz Daint, the researchers created climate projections for much of Europe (including Scandinavia, the Mediterranean, and Africa) at an ultra-precise 2.2 km resolution. The higher resolution allowed researchers to make new predictions, such as that hourly precipitation would increase in intensity by 7% for every degree of temperature rise. “In Europe, we are primarily interested in short-term torrential rains, as they often occur in summer“, Said Cher. Based on the simulation results, he reported that “water infrastructure must be adapted to more frequent and heavy rainfall“.
Another important element of the new modeling was a more accurate estimate of albedo, a measure of the reflectivity of an object’s surface. A more accurate representation of clouds (which have relatively high albedo) through new modeling has helped researchers measure their overall albedo more accurately and demonstrate that even small differences in these values can have colossal consequences.
As Cher noted, scientists “there is still a long way to go“. Even with the dramatic increase in resolution, climate modeling was limited to Europe. For a truly accurate global climate model, it must cover the entire globe.