How hot will it get in the Sahara, how cold in Antarctica, how mild in Central Europe? Climate models calculate the temperature map of the Earth. Their results feed into the reports of the Intergovernmental Panel on Climate Change, underpin political agreements such as the Paris Climate Agreement and provide the basis for concrete, regional measures to adapt to climate change. But how good are these models, really? And how have they evolved since their beginnings?
To answer these questions, Lukas Brunner and colleagues from the University of Vienna, the ETH Zürich (among others) examined a total of 176 climate models spanning the past 30 years. His finding: on average, the temperature maps produced by climate models have become significantly more accurate and reliable. "But it is impressive what was already possible thirty years ago," says the physicist. "If you take a really good model from the 1990s, its performance is actually similar to that of a mediocre model today."
Kilometer-scale models
Particularly exciting is the look at the latest generation: so-called kilometer-scale models, which simulate the climate system on a grid of just five to ten kilometers. Previously, grids of around 100 kilometers were standard. This means that processes such as the formation of thunderstorm cells can now be directly simulated for the first time, rather than being approximated through statistical methods.
One of the most important of these models, the so-called ICON Earth System Model, was developed at the Max Planck Institute for Meteorology in Hamburg in collaboration with the CLICCS Cluster of Excellence. The Vienna group led by Aiko Voigt also works extensively with the ICON model, contributes to its development, and uses ICON in teaching; the model runs on the HPC computers of the ASC (Austrian Scientific Computing) in Austria and CINECA Leonardo HPC computers in Italy.
To put it pointedly: the modeling has caught up with our observational reality
Another model, the IFS, comes from the European Centre for Medium-Range Weather Forecasts, which works closely with Hamburg's climate research community. Some of the latest versions of the IFS model already come remarkably close to the observational data used to benchmark the calculations. "To put it pointedly: the modeling has caught up with our observational reality," says Brunner. The model is no longer the limiting factor – the accuracy of the observational data is.
Model scenarios compared with real-world data
As part of the study, which was published in the Nature journal *Communications Earth & Environment*, various scenarios were simulated using each model and then compared with actual measured values. Rather than relying on a single reference dataset, ten datasets were used comparatively to account for differences in data quality. The results revealed significant discrepancies, making clear that the assessment of model quality strongly depends on the available data.
"Simply doubling the resolution without changing anything else in the model will usually not improve the results," Brunner emphasizes. What matters is that model physics and fine-tuning are also adapted to the new resolution.
Higher resolution does not automatically improve the results
Aiko Voigt from the Department of Meteorology and Geophysics at the University of Vienna also emphasizes this: “In our study, we were able to show that drastically refining the model grid from 100 km to 5 km allows for improvements in model quality that were unattainable for low-resolution models in the past. At the same time, we were also able to show that a high-resolution model is not automatically better. This illustrates that simulation on the one hand and the physical understanding of the climate on the other move in step with one another and benefit from each other.”
In addition to the Cluster of Excellence CLICCS, the study involved the University of Vienna, the Max Planck Institute for Meteorology, the Alfred Wegener Institute and ETH Zurich.
Publication: Brunner, L., Ghosh, R., Haimberger, L. et al. Three decades of simulating global temperature patterns with coupled global climate models. Commun Earth Environ (2026). DOI: https://doi.org/10.1038/s43247-026-03497-w
