Scientists create highly detailed digital twin of Earth to model climate

Scientists have created a highly detailed digital twin of Earth capable of simulating the planet's weather and climate systems with near-kilometer accuracy, a development researchers say could transform climate modeling.

Developed by a team led by Daniel Klocke at the Max Planck Institute for Meteorology in Germany, the model reaches a spatial resolution of 1.25 kilometers -- a level many atmospheric scientists have described as a "holy grail" for climate research.

The digital twin divides the globe into around 336 million land and ocean cells, each paired with a corresponding atmospheric layer. That amounts to 672 million points of calculation, allowing scientists to examine changes in both the atmosphere and the planet's surface in unprecedented detail.

The model integrates both "fast" and "slow" natural processes. Fast systems, such as energy and water cycles, reflect short-term weather events like storms and rainfall. Slow systems, including the carbon cycle and ocean geochemistry, evolve over years or decades. Linking the two allows researchers to simulate daily weather patterns and long-term climate shifts within the same framework -- something previously thought unattainable.

To run the model, the team used the ICOsahedral Nonhydrostatic (ICON) model, developed with the German Weather Service, and optimized it through Data-Centric Parallel Programming to enable efficient processing on modern hardware.

The simulation was executed on two European supercomputers -- JUPITER in Germany and Alps in Switzerland -- using 20,480 Nvidia GH200 Grace Hopper chips that combine GPU and CPU capabilities for massive parallel computation. This allowed scientists to simulate nearly five months of Earth's climate -- 145.7 days -- in just 24 hours, generating nearly a trillion data points.

However, researchers acknowledge that such models remain inaccessible for everyday forecasting due to immense hardware demands.