Traditional global climate models were like early digital cameras — they had only about ten thousand pixels to cover the entire planet. At that low resolution, big storm systems looked like blurry blobs. You couldn't see their true shape, how long they lasted, or where they dumped the heaviest rain.
That created serious problems. Because those coarse models couldn't directly simulate thunderstorms, they had to rely on rough approximations, called "convective parameterization schemes" — and those often failed. They produced too much light drizzle, missed extreme downpours, got the timing of rain wrong, and completely failed to capture the large, organized clusters of thunderstorms known as mesoscale convective systems (MCSs). Those MCSs are exactly what cause most flash floods, damaging winds, and record-breaking rain events.
Now, next‑generation "kilometer‑scale" models have jumped to over 50 million pixels per global layer — roughly 2.8 km (about 1.7 miles) of resolution. That's sharp enough to see individual thunderstorm updrafts and rainbands directly, without guesswork.
With climate change driving extreme weather toward "more frequent, more severe," scientists urgently need to know: do these sharp new models actually get storms right?
A new study in Advances in Atmospheric Sciences puts six of the world's leading global kilometer‑scale models to the test, using East Asia's record‑breaking wet summer of 2020 as a real‑world exam. That summer broke rainfall records across Asia : 10 Chinese provinces flooded, Japan saw 1,000 mm in three days, and South Korea's rainy season lasted 54 days — well above its usual 32. The models include six global kilometer-scale models participating in the World Climate Research Programme (WCRP) Global KM-Scale Modeling Hackathon — ECMWF (IFS), the Max Planck Institute (ICON), the UK Met Office (UM), the U.S. Department of Energy (SCREAM), the University of Tokyo (NICAM), and the Chinese Academy of Sciences (CAS-ESM) .
The international research team, led by the Chinese Academy of Meteorological Sciences (CAMS), Beijing Normal University, ETH Zurich, and the CAS Institute of Atmospheric Physics, compared model results against satellite observations.
"The good news: these models mostly get the big picture right — where MCS rain falls, how long storms last, how fast they move, and their daily timing. " said Xiaotong Huang, first author and M.Sc. candidate at CAMS.
The models successfully reproduced the overall spatial distribution of MCS precipitation during the extreme 2020 summer. They also captured fine‑scale details including system duration, propagation speed, areal extent, rainfall intensity, and the daily cycle of storm activity.
Despite their power, the models shared the same biases. "They produce too many MCSs, which are too short-lived, too small in area, and too intense in rainfall rate." said Dr. Puxi Li, corresponding author of the study.
Specifically, relative to satellite observations:
- Too many MCSs — with the excess being predominantly short‑lived systems
- Too short — simulated storms don't last as long as real ones
- Too small — the rain area is systematically underestimated
- Too strong — within the narrow core, deep convection and rainfall rates are too intense
These shared biases provide critical clues for improving the next generation of global storm‑resolving models.
The urgency is not academic. In late May 2026, an extreme rainfall event struck the middle and lower Yangtze River basin in eastern China, prompting China's Ministry of Water Resources and the China Meteorological Administration to jointly issue the first national‑level Red Alert for torrential rain and flash floods of the year.
A previous study by the same research group found that MCS‑associated rainfall has become more frequent and intense, and contribute more than 75% to the total precipitation increase over the East Asian summer monsoon rainband in the past two decades — a trend directly linked to a warming climate.
Kilometer-scale Earth system modeling is crossing a historic milestone. Under major European initiatives — nextGEMS, WarmWorld, and Destination Earth — models such as ICON and IFS have already successfully completed multi-decadal continuous integrations. These frontier achievements will be discussed in depth at the upcoming KM-scale Global Modelling Summit 2026 in Hamburg, Germany. Looking ahead over the next three to five years, the international research community must continue concerted efforts to refine cloud microphysical processes, boundary-layer turbulent mixing, and high-resolution atmosphere–ocean–land coupling.
According to the authors, their goal is straightforward: get the right process of high-impact weather in the right place, at the right time — and turn these 50‑megapixel simulations into real help for disaster preparedness and climate adaptation.
