NextGEMS: The Start of a New Era in Climate Research

Climate researchers are addressing this challenge by developing climate models with a horizontal resolution of a few kilometers. The MPI-M has been working towards this goal for years with the ICON model. Developing kilometer-scale models is a challenging task, because it does not simply mean increasing the resolution of the old models. Rather, the new models must also represent processes such as storms, ocean eddies, or cracks in the ice.  In the past, models did not include these processes, but instead tried to account for their net effect through statistical corrections.  Storm-resolving models represent a new generation of Earth system models designed to account for these processes explicitly and physically.

Developing two such models was a major goal of the “next Generation Earth Model Systems” (nextGEMS) project, which started in 2021 and ended this year and which was funded by the European Union’s Horizon 2020 program. Led by Bjorn Stevens at MPI-M and Irina Sandu from the European Centre for Medium-Range Weather Forecasts (ECMWF), it involved 27 research institutions from 14 European countries and Senegal. At the start of the project, Sandu emphasized its relevance for society: “The nextGEMS models will provide information on scales—like a country or city—on which the impacts of severe weather events are felt, and on scales which are relevant for important applications and sectors like renewable energies or marine fisheries.”

The project was ambitious, and the challenges were significant at the beginning, because kilometer-scale models require enormous computing power and must use state-of-the-art supercomputer structures optimally. “They need to work on the largest supercomputers, which are also the most tricky ones to use, to produce data amounts we could have never imagined saving, let alone analyzing,” said Daniel Klocke, head of the MPI-M model development group, at the start of the project.

nextGEMS therefore pursued three specific goals: First, to develop two kilometer-scale Earth system models; second, to produce the first ever simulations with these models over a period of 30 years; and third, to provide new workflows to handle the enormous amounts of data they produced.

The strategy for achieving the goals involved several eight-month development cycles. At the end of each cycle, the participants tested and optimized the previous developments in hackathons. These interactive work meetings replaced the usual presentation-based project meetings to enable efficient collaboration among people from the participating countries. Scientists, technical experts, and potential users from the field of renewable energies were involved. The hackathons, organized by nextGEMS coordinator Heike Konow at MPI-M and others, became the hallmark of the project. While the first hackathon in Berlin in October 2021 had 80 participants, the organizers welcomed almost twice as many people two years later in Madrid. “In nextGEMS, hackathons became more than events. They became the heart of scientific progress,” says Eulàlia Baulenas from the Barcelona Supercomputing Center in a review of the modus operandi of nextGEMS. In addition, the project focused on engaging, inclusive communication, including an innovative video blog from which this text quotes extensively. The aim of this strategy was not only to inspire as many people as possible, but also to get them involved.

Model development did not start from scratch, but built on two existing models: ICON, which has been developed over the past 20 years by MPI-M, the German meteorological service (Deutscher Wetterdienst), and other partners, and the Integrated Forecasting System (IFS) of the ECMWF, coupled with the FESOM ocean model of the Alfred Wegener Institute Helmholtz Centre for Polar and Marine Research. The choice of these two models was not random—rather, they incorporate very different philosophies, and comparing them therefore becomes more interesting from a scientific perspective.

In the three development cycles, shorter simulations were run on the Levante supercomputer at the German Climate Computing Center (DKRZ) to test and improve the performance of the models. “It became clear that such short runs are often sufficient to see, understand, and evaluate the effects of model adjustments,” says MPI-M group leader Hans Segura.