The climate models agree on the outcome and disagree on how it happens

Clouds grow from the bottom up

The clouds that cool the planet are built from the bottom up, and a preprint posted this week says the models have been looking at the wrong controls. Marine stratocumulus clouds — the wide, flat decks that shade large stretches of ocean — grow by an upscale process: small vertical thermals feed energy upward into large horizontal structures, an inverse cascade of turbulent kinetic energy. The cells grow at roughly one boundary-layer depth per hour. What controls that growth, the researchers found, is the transfer of kinetic energy and the stratification of the air, rather than the interaction between humidity and radiation that modelers often reach for. For clouds that shade large stretches of ocean, the physics that sets their size is mechanical before it is radiative. m:…3472

The aerosol the models miss

A second preprint finds a forcing that the big climate models are mostly missing. The South East Asian Aerosol Plume — smoke and haze from biomass burning and gas flares — has a direct and statistically significant effect on the Indian Ocean Dipole, the pattern of sea-surface temperatures that shapes rainfall around the Indian Ocean. The plume cools the sea surface in the east and suppresses convection there, which warms the west. The gap in the models is large: the CMIP6 historical experiments underestimate aerosol levels in the plume region by 58 percent compared with NASA Terra MODIS satellite measurements. The preprint also finds that Indian Ocean Dipole events are intensified by the aerosols from biomass burning and oil industry gas flares. If the models are missing more than half the aerosol, they are missing part of what drives one of the most consequential ocean patterns on the planet. m:…0226

A bias that tilts the Pacific

A third preprint traces a different model error to a single ocean bias. In the equatorial Pacific, climate models tend to make the cold tongue — the band of cool water along the equator — too cold. That bias accounts for 28 percent of the difference between models in how much the Northwest Pacific anticyclone strengthens in the summers after an El Nino. Because of the bias, the CMIP5 and CMIP6 models overestimate that strengthening, and observation-based constraints point to a substantially weaker intensification than the raw model output suggests. The practical consequence is for anyone using model output to plan for post-El Nino summers around the Northwest Pacific: the raw projections overstate the change, and the correction is not a small tweak but a substantial downgrade. m:…3417

Storm models split in two

The last preprint looks at the newest storm-resolving models and finds they agree on the answer while disagreeing on the physics. All five kilometer-scale models evaluated underestimated both the total rainfall and the size of tropical oceanic mesoscale convective systems compared with satellite observations. But they split into two distinct regimes in how they get there: one group ties rainfall to both moisture and upward motion, the other is driven mainly by dynamics regardless of the moisture state. Two models can produce the same rainfall statistics for different physical reasons, which means the agreement is not evidence that the physics is right. The researchers point to the boundary layer and the lower free troposphere — the lowest few kilometers of the atmosphere — as the place where future model development has to focus. m:…3415

What this adds up to

Four preprints, one pattern: the models are good at the headline and shaky on the mechanism. Clouds grow by energy transfer, not by the humidity-radiation link the models emphasize. A major aerosol forcing is underestimated by more than half. A Pacific ocean bias inflates a post-El Nino signal by a wide margin. And the best storm models agree on rainfall totals while running on different physics. None of this says the models are useless — they reproduce the broad picture, and each preprint names the specific fix. But it does say that anyone using model output for a decision should ask which mechanism the model is using, not just what number it produces. All four results are preprints, and the fixes they propose still need to be built into the models and tested. The work ahead is not more resolution. It is getting the physics right at the scales the models already have. m:…3472 m:…0226 m:…3417 m:…3415