August 06, 2021 - by Florian Meyer, ETH Zurich

Torrential rain, hailstorms and floods in the Alpine region and northwest Europe: the past few weeks have highlighted the impacts of severe thunderstorms. But how exactly are extreme weather events connected to global warming? This is one of the central questions for researchers studying and modelling the interaction between weather and climate.

By representing the underlying fundamental physical processes, models are a very powerful tool to understand these interactions. But current models and the required computer infrastructure have reached a wall, limiting the extent to which researchers can draw conclusions about how, for example, climate change affects extreme weather. To overcome this issue, ETH Zurich has teamed up with partners to launch the EXCLAIM research initiative. This project aims to dramatically increase the spatial resolution of the models, thereby enhancing their accuracy in simulating the weather on a global scale in a future, warmer world.

Seamless weather simulations in climate models

“Thanks to their high resolution, the new, global models will simulate key processes such as storms and weather systems in much more detail than before, allowing us to study the interaction of climate change and weather events much more accurately,” says Nicolas Gruber, EXCLAIM lead PI and Professor of Environmental Physics.

EXCLAIM is interdisciplinary: along with the climate researchers from the ETH Center for Climate Systems Modeling (C2SM), ETH computer scientists, the Swiss National Supercomputing Centre (CSCS), the Swiss Data Science Center (SDSC), the Swiss Federal Laboratories for Materials Science and Technology (Empa) and MeteoSwiss, the Federal Office of Meteorology and Climatology, are all involved in the project. Not only will this collaboration improve the modelling of climate, it will also make the weather forecasts provided by MeteoSwiss more reliable. International project partners include Germany’s National Meteorological Service, Deutscher Wetterdienst (DWD), and the Max Planck Institute for Meteorology (MPI-M), which together developed the ICON (Icosahedral Nonhydrostatic) Model – the basis of EXCLAIM – as well as the European Centre for Medium-Range Weather Forecasts (ECMWF), of which Switzerland is a full member.

With EXCLAIM, researchers are aiming to radically scale up the spatial resolution of the weather and climate models. To simulate global weather and climate with all its regional detail, such models place a virtual, three-dimensional grid over the Earth. Researchers then use the laws of physics to calculate the respective climate conditions for each cell in their models. Current global climate models typically have grid cells with a width of 50 to 100 kilometres. In the long run, EXCLAIM researchers aim to increase the resolution to just one kilometre.

In the past, given the limited computing power of modern supercomputers, only regional weather could be simulated with such a fine grid – and for relatively short periods of time at most. With the new models, the researchers now hope to attain this fine resolution worldwide, enabling them to simulate weather patterns from a global climate perspective and with a much sharper focus. This is like giving global climate models an additional zoom function for small-scale events. “What’s more, the new models will pave the way for ‘forecasting’ weather in the future climate, providing the answers as to how extreme weather events like the torrential rain we experienced this summer might look in the future,” says Christof Appenzeller, Head of Analysis and Forecasting at MeteoSwiss.

Powerful infrastructure for climate simulations

Customised computer infrastructure is essential to get the best out of the new models. Weather and climate models are some of the most complex, most data-intensive computational problems there are, which is why the EXCLAIM models are being developed in parallel with the hardware and software for supercomputers. “The computing and data infrastructure is being tailored to the exact requirements of the weather and climate models,” says Thomas Schulthess, Director of the Swiss National Supercomputing Centre (CSCS) in Lugano. For example, the new “Alps” supercomputing system is configured to allow the high-resolution climate models to properly resolve convective systems, such as thunderstorms.

To effectively simulate weather and climate on a global scale over several decades with a grid width of just a few kilometres, the model will have to run approximately 100 times faster than is currently possible. The first option for achieving this goal is to deploy faster, more powerful computers. Switching from the current supercomputer at CSCS to the “Alps” system will be instrumental in this regard.

One challenge is the end of “Moore’s law”, which holds that processor performance doubles approximately every 20 months. “As processor haven’t increased in serial performance for about 15 years, the only way of improving supercomputer performance is to improve their parallel processing architecture,” Schulthess says. “Furthermore, it’s worth setting up the supercomputer architecture specifically to allow it to solve classes of research problems in an optimal manner.” The key to providing the requisite computing power here lies in a hybrid computer architecture in which conventional CPUs (central processing units), responsible for performing calculations and sharing data between the memory and components, are deployed in conjunction with GPUs (graphical processing units).

The second option concerns the software, namely the optimisation of the model code to ensure it fully benefits from the hybrid computer architecture. EXCLAIM is taking a revolutionary approach by splitting the source code into two parts: a first part that represents the interface to the model developers and users; and an underlying software infrastructure part in which the model’s central algorithms are implemented with a high degree of efficiency for the respective hardware. CSCS, MeteoSwiss and C2SM have already used this approach in the current MeteoSwiss weather model with great success. This approach is now being applied to the ICON weather and climate model. “We were able to accelerate the MeteoSwiss weather model by a factor of ten, improving the reliability of the MeteoSwiss forecasts as a result,” Schulthess says.

Managing the flood of data

Computing speed alone is not the decisive factor. Increasing the resolution of the models also leads to a data explosion. Furthermore, weather and climate research require and produce a high diversity of data. To ensure effective throughput, it is equally crucial that the computers are able both to access the data and to write the results to storage media as quickly as possible. The computing processes have to be organised accordingly, while memory bandwidth is maximised and costly data transfers avoided. “For the new weather and climate models to produce useful results, we have to optimise the entire infrastructure. To this end, we’re leveraging the expertise gained from many years of working with MeteoSwiss and the ETH domain,” Schulthess says.

Further information: