220-metre climate simulation breaks global resolution record

A global climate simulation has reproduced Earth's atmospheric behaviour at a horizontal resolution of 220 metres — a record for worldwide models, and a scale fine enough to resolve the internal structure of individual convective clouds. Shuhei Matsugishi of the Atmosphere and Ocean Research Institute at the University of Tokyo led the work, which the team reports in Geophysical Research Letters.

The result matters because conventional global climate models divide the atmosphere into grid cells spanning tens to hundreds of kilometres — far too coarse to capture localised, dangerous storms. A newer generation of Global Storm-Resolving Models (GSRMs), enabled by advances in supercomputing, has pushed grid sizes down to 10 km or even 1 km. Matsugishi's team went beyond that with a large eddy simulation built on the Nonhydrostatic Icosahedral Atmospheric Model, run on Japan's Fugaku supercomputer, one of the world's fastest machines.

What does the finer grid change?

The clearest measured effect concerns rainfall. At higher resolution, the model produces fewer of the very intense rain bursts over small, localised areas that plague current-generation GSRMs. These inaccurate, isolated cloudbursts are known in the field as "popcorn-like" rainfall, and they have cropped up across many simulations.

"This is fine enough to explicitly represent the detailed internal structure of convective clouds, including individual updrafts and downdrafts," Matsugishi said. "That is not possible in conventional global atmospheric models."

The experimental design was deliberate. Because previous work indicated that horizontal resolution should affect cloud optical properties more than vertical resolution, the researchers fixed the vertical grid at 78 layers — a configuration similar to an earlier study — and varied only the horizontal spacing, testing 3.5, 1.7, 0.87 and 0.44 km before the record-setting 220-metre run. The model's physics included light radiation, cloud microphysics, surface air fluxes and sub-grid-scale turbulence.

The team also found that how rainfall and clouds are represented depends on how atmospheric turbulence is modelled. "This is exciting because it allows us to study how individual convective clouds interact with each other and with the large-scale atmospheric circulation," Matsugishi said. "Such interactions are important for understanding extreme rainfall, tropical weather systems and the role of clouds in the climate system."

What are the computational limits?

The cost is the catch. A single global simulation at 220-metre resolution requires nearly one trillion grid points and what Matsugishi calls an unprecedented amount of memory and data storage. The team optimised the model to run across a large number of processors simultaneously, but Matsugishi concedes the approach is far too expensive for routine weather forecasting or long-term climate projection.

Its realistic role, he says, is as a reference: a high-fidelity benchmark against which researchers can evaluate and improve the coarser models that do run operationally. For R&D managers weighing compute budgets, the message is that this is a research instrument, not a production forecasting system.

The context sharpens the point. As climate change makes strong storms more common, prediction helps reduce risks to life and infrastructure, and physics-based models such as this one still outperform AI approaches for predicting extreme weather events.

What comes next?

The researchers plan to use the model to study convective clouds in greater detail. Specific tasks include examining how cloud size, organisation and vertical air motion vary with increasing resolution, and then assessing how realistically the model represents those characteristics. A further goal is to work out how turbulence, cloud microphysics and other small-scale processes should be represented as global atmospheric models move from kilometre-scale grids to resolutions of several hundred metres.

"Through these investigations, we hope to identify the improvements needed for the next generation of global weather and climate models," Matsugishi said.