nextGEMS final paper marks a milestone for future European climate research

After four years of intensive model development and international collaboration, the nextGEMS project published its final results paper in October 2025. The project set out to achieve multidecadal climate simulations at kilometre-scale resolution, representing interactions between the atmosphere, ocean, and land in far greater detail than conventional climate models.

This ambitious goal was pursued through four coordinated “hackathon cycles”, each producing an improved configuration of the project’s models. Over time, these cycles progressively increased model stability, realism, and computational performance. The final paper reflects the contributions of a large research community spanning partner institutions in 14 countries. This blog post highlights some of the project’s main results. For a detailed overview, read “nextGEMS: entering the era of kilometer-scale Earth system modelling.”

Two complementary Earth system models

nextGEMS further developed two Earth system models capable of simulating the coupled climate system at around 10 km resolution. Both models adopted a common output grid based on the HEALPix sphere, which divides the Earth into equal-area cells and simplifies the comparison and analysis of model results. 

The ICOsahedral Nonhydrostatic (ICON) model represents the atmosphere, ocean, and land system and relies on few empirical simplified physical processes or parameterizations. Meanwhile, the Integrated Forecasting System – Finite-Element Sea Ice-Ocean Model (IFS-FESOM) model combines the ECMWF weather prediction model with a detailed ocean model and uses more parameterizations, but is designed for strong operational accuracy and stability.

The models were improved through bug fixes, updated radiation and cloud schemes, better land-surface processes, and stronger atmosphere-ocean coupling. Early simulations lasted about a year and showed physical imbalances in energy and water budgets. Through iterative testing and tuning, these issues were gradually reduced. By the final cycle, the models were able to run stable 30-year climate simulations at roughly 10 km resolution.

Computational performance also improved substantially through optimized data handling, parallelization, and model tuning. On the Levante supercomputer, simulations eventually reached about 500 simulated days per day – more than nextGEMS initial goal.

How realistic are nextGEMS km-scale simulations?

The simulations showed encouraging results in reproducing important climate features. Both models achieved a near-zero global energy imbalance and realistic seasonal cycles, largely due to improvements in cloud and radiation schemes. Additionally, large-scale patterns such as the tropical rainbelt over land were well represented. However, both models still show known challenges common to many climate models, including biases in the western Pacific warm pool and the double Intertropical Convergence Zone (ITCZ) pattern.

For regional processes, the results are promising. Phenomena such as soil-moisture precipitation feedbacks, stratocumulus cloud systems, and atmospheric blocking events were simulated with realistic variability. ICON performed particularly well in representing cloud-radiative effects, while IFS-FESOM better captured the frequency of atmospheric blockings.

Next steps in research and project impact

Future work will push model resolution even further, toward less than 10-km simulations on exascale supercomputers such as LUMI. Additional Earth system components, including ocean biogeochemistry, aerosols, and aerosol-cloud interactions need to be integrated to improve realism.

Beyond research, the models and datasets developed in nextGEMS contribute to European initiatives such as Destination Earth’s Digital Twin, which aims to support climate adaptation and operational climate services.