Researchers at the University of Tokyo have developed a high resolution global atmospheric model capable of simulating weather patterns at a grid spacing of just 220 metres, helping eliminate unrealistic bursts of heavy rainfall known as “popcornlike” rain in computer simulations.
The breakthrough could improve scientists’ understanding of extreme weather events and help refine future weather forecasting and climate prediction models. The research was published in the journal Geophysical Research Letters.
Conventional global climate models typically divide the atmosphere into grid sections spanning tens to hundreds of kilometres. More advanced global storm resolving models use finer grids, generally ranging from one to 10 kilometres, to simulate clouds storms and atmospheric processes more accurately.
The new model developed by the University of Tokyo team takes this approach further allowing the atmosphere to be simulated at a horizontal resolution of 220 metres, roughly the length of two football fields.
According to the researchers the finer resolution enables the model to represent the internal structure of convective clouds including individual updrafts and downdrafts. These processes play an important role in the development of thunderstorms and intense rainfall.
The team described the approach as the first demonstration of a global large eddy simulation which can explicitly represent fine scale structures within deep convective clouds instead of relying primarily on simplified mathematical representations.
One of the major improvements was the elimination of “popcornlike” rain, a recurring error in some existing global storm resolving models. This occurs when simulations produce isolated intense bursts of rainfall that do not correspond to actual weather patterns.
In simulations of Typhoon Omais, the researchers found that the 220 metre model reproduced precipitation patterns more realistically than a model with a grid spacing of 3.5 kilometres.
However, achieving this level of detail required enormous computing resources. The simulation contained nearly one trillion three dimensional grid points. To simulate eight hours of atmospheric activity on August 5, 2016 the researchers used more than half of Japan’s Fugaku supercomputer.
The computational demand remains a major obstacle to practical use. Shuhei Matsugishi a researcher at the university’s Atmosphere and Ocean Research Institute said the model is currently too expensive to run as part of routine operational weather forecasting.
For now, the technology is better suited to research, including investigations into tropical convection, heavy rainfall and the detailed structure of storm clouds. It could also provide a high resolution reference for evaluating and improving less computationally demanding climate models.
Despite eliminating the popcornlike rain bias the model still showed inaccuracies in the distribution of cloud cover. Researchers say further work is needed to understand turbulence, cloud microphysics and other atmospheric processes.
The team plans to use the model to investigate convective clouds in greater detail and improve the representation of small-scale atmospheric processes in future simulations.
The findings could eventually contribute to more reliable predictions of extreme weather and reduce uncertainties in climate projections although significant computational and scientific challenges remain.
The study titled “Resolution Dependence in a Global Atmospheric Simulation from km to 220 m Grid Spacing,” was published in Geophysical Research Letters in 2026.