AI Insight
Researchers conducted the first global atmospheric simulation at 220-meter resolution using Japan's Fugaku supercomputer, finding that higher resolution reduces unrealistic "popcorn-like" precipitation patterns in storm models but does not improve cloud cover representation. The study tested resolutions from 3.5 kilometers down to 220 meters and found that while precipitation became better organized into smaller, distributed cells at finer scales, cloud cover decreased regardless of the turbulence scheme used. The findings suggest that improving turbulence parameterizations, particularly for boundary layer processes, is essential for better representing both precipitation and clouds in global storm-resolving models.
Why it matters
This research addresses critical limitations in global weather prediction models by identifying the optimal resolution needed for accurate storm forecasting and revealing that achieving realistic simulations requires not just higher resolution but also improved representation of atmospheric turbulence. The findings have direct implications for improving weather forecasts, particularly for extreme precipitation events and thunderstorms.
Understand the Science
Source: Geophysical Research Letters
Global storm-resolving models (GSRMs), which can simulate how moist air rises to form clouds, heavy rain, and thunderstorms, are an important tool for predicting weather. However, because their resolution is confined to a few kilometers, they have limitations. For instance, they can create “popcorn-like” precipitation patterns at coarser resolutions, showing rainfall as too intense and localized.
Increasing the model’s resolution could help resolve some of these inconsistencies by allowing researchers to better simulate the internal structures of a thunderstorm and the processes that lead to rainfall. Previous studies suggested resolutions around 200 meters are needed to resolve the inner workings of deep convection; however, at these finer scales, accurately representing atmospheric turbulence becomes another challenge to overcome.
Matsugishi et al. conducted the first global large-eddy simulation at 220-meter resolution—fine enough to capture some of the turbulent motions associated with deep convection processes that produce thunderstorms, clouds, and heavy rain.
Using the Nonhydrostatic Icosahedral Atmospheric Model (NICAM) on the supercomputer Fugaku, researchers examined the simulations from an 8-hour period on 5 August 2016. Researchers tested two turbulence schemes at resolutions ranging from 3.5 kilometers to 220 meters. As resolution increased, the popcorn-like precipitation predictions decreased, with precipitation becoming organized into smaller, more widely distributed cells. However, cloud cover decreased with increasing resolution regardless of the turbulence scheme, suggesting that though higher resolution can reduce unrealistic precipitation patterns in GSRMs, it does not necessarily reproduce realistic cloud distribution.
Ten-meter global large-eddy simulations could ultimately reduce uncertainties even further, but current supercomputers are limited to grid spacings of around 220 meters. At resolutions around 200 meters, improving turbulence parameterizations, particularly for boundary layer processes in the gray zone, is essential for better representing clouds, the authors say. The findings from this study are promising for better simulating precipitation, but more work is needed to improve how clouds are represented, they add. (Geophysical Research Letters, https://doi.org/10.1029/2026GL124411, 2026)
—Rebecca Owen (@beccapox.bsky.social), Science Writer
