Puxi Li remembers the satellite images from East Asia’s catastrophic wet summer of 2020—swathes of deep red blooming across China, Japan, and South Korea, where torrential rains shattered records and flooded entire provinces. Now, for the first time, climate models can simulate those storms with such precision that individual thunderstorm cores emerge in near-realistic detail. These next-generation models use over 50 million pixels per layer—2.8 km of resolution—allowing them to capture storm dynamics without relying on crude approximations. But even at this unprecedented clarity, they still miss the mark in four consistent ways.

This leap in resolution matters because the world is hurtling into an era of more frequent and severe storms. Mesoscale convective systems (MCSs)—organized clusters of thunderstorms that drive flash floods, extreme rainfall, and damaging winds—now account for over 75% of the total precipitation increase across the East Asian summer monsoon region in the past two decades. As global temperatures rise, accurately predicting these systems isn’t just scientific progress—it’s a matter of public safety. That’s why a team led by the Chinese Academy of Meteorological Sciences, Beijing Normal University, ETH Zurich, and the CAS Institute of Atmospheric Physics put six of the world’s most advanced global models to the test: ECMWF (IFS), Max Planck Institute (ICON), UK Met Office (UM), U.S. Department of Energy (SCREAM), University of Tokyo (NICAM), and Chinese Academy of Sciences (CAS-ESM).

Using the extreme 2020 summer as a benchmark, the researchers compared simulations against actual satellite data. The results, published in Advances in Atmospheric Sciences, show that while the models correctly capture where MCS rain falls, how storms move, and their daily timing, they consistently generate too many short-lived systems, underestimate storm size by up to 30%, and over-intensify rainfall within storm cores. "They produce too many MCSs, which are too short-lived, too small in area, and too intense in rainfall rate," said Dr. Puxi Li, the study’s corresponding author. These biases persist across all six models, suggesting a systemic issue in how kilometer-scale models handle storm lifecycle and organization.

Despite these blind spots, the progress is historic. For the first time, models like ICON and IFS have completed multidecadal continuous simulations under European initiatives such as nextGEMS and Destination Earth—laying the groundwork for reliable decade-scale forecasts. As the KM-scale Global Modeling Summit 2026 in Hamburg approaches, scientists are focused on refining these models to close the gaps. With China already issuing its first national Red Alert for torrential rain in 2026, the need has never been more urgent. The models are sharper than ever—but to truly protect lives, they must learn not just to see storms, but to understand them.