Winds 30 Kilometres Up Steer Tropical High Clouds — and Climate Models Miss Half the Signal

High above the equator, some 30 kilometres up, the stratospheric winds reverse direction roughly every 28 months, swinging from easterlies to westerlies and back again in a rhythm known as the quasi-biennial oscillation, or QBO. For decades this oscillation was regarded as a curiosity of the upper atmosphere, a self-contained cycle driven by upward-propagating waves from the troposphere. But a growing body of evidence shows that the QBO reaches downward, nudging tropical convection, precipitation and even the Madden–Julian Oscillation, the great eastward-marching pulse of tropical weather. Now a new study published in Atmospheric Chemistry and Physics has quantified one of the most consequential of these downward influences: the QBO’s control over the thin, icy high clouds of the tropical tropopause layer, the band of atmosphere between roughly 14 and 17 kilometres that regulates how much heat Earth’s tropics radiate to space.

Aleena M. Jaison of Imperial College London, together with Paulo Ceppi and Sarah Wilson Kemsley, set out to answer a deceptively simple question: when the stratospheric winds blow from the west, what happens to the tropical high clouds, and can climate models reproduce it? The answer, drawn from a decade and a half of spaceborne lidar observations and a suite of state-of-the-art climate simulations, is that the clouds do respond — robustly, measurably and in a way that models capture only faintly. During the westerly phase of the QBO, high-cloud cover over the deep tropics between 10 degrees south and 10 degrees north shrinks. In the lidar observations, the tropical-mean cloud fraction in the 14-to-17-kilometre layer falls by about 0.54 percent for each standard deviation of the QBO wind index. The multi-model mean of seven CMIP6 climate models that simulate a realistic QBO captures the same sign of response, but at barely a fifth of the observed strength, around 0.12 percent.

The observational backbone of the study is the CALIPSO satellite, launched in 2006, whose Cloud-Aerosol Lidar with Orthogonal Polarization fires laser pulses downward and records the returning echoes, profiling clouds in vertical slices tens of metres thick. This active sensing is critical, because the clouds in question are often optically thin cirrus veils that passive imagers struggle to distinguish. To make the satellite data comparable with model output, the team used the GCM-oriented CALIPSO Cloud Product, a reprocessed dataset designed to match what a satellite simulator inside a climate model would see, and cross-checked the results against MODIS and ISCCP cloud records. Meteorological fields came from the ERA-5 reanalysis. Because El Niño–Southern Oscillation dominates tropical cloud variability on similar timescales, the Niño 3.4 index was regressed out of every dataset, isolating the QBO’s fingerprint.

That fingerprint is strikingly structured. When westerly wind anomalies occupy the lower stratosphere, a QBO-driven secondary circulation warms the layers immediately beneath, from about 70 hectopascals down past 100 hectopascals, near the tropopause. The warming dries the upper troposphere, lowering relative humidity across the equatorial band. And where the air warms and dries, clouds thin out: the observed cloud-fraction reduction extends from 100 to 200 hectopascals, with the strongest deficits over the Indian Ocean, the Maritime Continent and the western Pacific — precisely the regions where deep convection and tropopause-layer cloudiness climatologically peak. The QBO alone accounts for roughly a quarter of the variance in tropical-mean cloud anomalies in the 14-to-17-kilometre layer. Intriguingly, the response is not purely a thinning: at levels between 6 and 14 kilometres, cloud fraction shows a modest, if statistically fragile, increase during westerlies, suggesting the oscillation reshapes the vertical distribution of tropical cloudiness rather than simply switching it off.

To dissect the mechanism, the authors turned to a cloud-controlling factor analysis, a statistical framework that treats cloud fraction as a function of the meteorological environment. Six factors were chosen: upper-tropospheric temperature, static stability at 150 hectopascals, upper-tropospheric relative humidity, lower-tropospheric humidity at 700 hectopascals, vertical pressure velocity at 300 hectopascals, and wind shear between 300 hectopascals and the surface. The sensitivity of cloud fraction to each factor was estimated with ridge regression, a technique that penalises large coefficients to prevent overfitting when predictors are correlated, and computed over a wide 21-by-11 grid-box neighbourhood so that non-local influences on each grid cell are captured. Multiplying those sensitivities by the QBO-induced changes in each factor reconstructs the total QBO–cloud relationship — and the reconstruction works, achieving a spatial correlation of 0.87 with the observed pattern.

The decomposition reveals that three factors do nearly all the work: upper-tropospheric temperature, static stability at 150 hectopascals and upper-tropospheric relative humidity. The physics is coherent. During westerlies, warming near the tropopause increases static stability, which suppresses convection; the same warming dries the upper troposphere, starving cirrus formation of moisture; and temperature itself may act directly on ice-cloud microphysics and cloud-top radiative cooling. Notably, when the team swapped surface temperature in place of upper-tropospheric temperature, the surface term contributed almost nothing to the tropical-mean cloud response, confirming that the QBO acts on clouds through in-situ thermodynamic changes in the upper troposphere and lower stratosphere rather than through surface-based pathways. The remaining three factors, though individually important for clouds, barely budge under the QBO and so contribute little.

The models, by contrast, get the ingredients but misplace them. The CMIP6 multi-model mean reproduces the overall reduction in deep-tropical high clouds during westerlies, yet the spatial alignment of the three dominant contributions falls apart. In the models, the humidity-driven cloud reduction peaks over the central Pacific rather than the Indian Ocean and Maritime Continent, the temperature contribution actually points toward increased cloudiness, and the static-stability response to the QBO is not just weak but sign-reversed, showing a decrease in stability across most of the deep tropics where observations show a pronounced increase. When the researchers ran a diagnostic swap — combining the models’ QBO-driven meteorological anomalies with the observed cloud sensitivities — the reconstructed pattern snapped back toward the observed one, with strong cloud modulation near the Maritime Continent. The conclusion is uncomfortable but clarifying: the dominant model bias lies not in how the models simulate the QBO itself, though that is also flawed, but in how sensitively their clouds respond to upper-tropospheric temperature and stability.

The inter-model spread tells the same story from another angle. All seven models except HadGEM3 produce a cloud reduction during westerlies, but the magnitudes range widely, from the weakest response in CNRM-ESM2-1 to the strongest in MRI-ESM2-0 — an outlier whose clouds are exceptionally sensitive to upper-tropospheric temperature. Substituting a common multi-model set of cloud sensitivities into each model’s reconstruction collapses much of the spread, pinning the differences chiefly on how each model’s clouds react to temperature and stability, with secondary contributions from how strongly each model’s QBO modulates those same factors. This matters beyond model intercomparison: previous work has shown that models fail to capture the observed QBO–MJO connection even when their stratospheres are nudged toward reanalysis, and the strongest observed QBO–cloud signal sits squarely over the Maritime Continent where MJO activity peaks. Biased cloud sensitivities over that region could be one reason the connection goes missing.

The stakes extend to climate projection. Cloud feedback remains one of the largest uncertainties in estimates of future warming, and the upward migration of high clouds under greenhouse-gas forcing is a central mechanism. Because the QBO’s temperature anomalies are well observed and well understood, the QBO offers something rare: a natural, recurring experiment in how high clouds respond to changes in upper-tropospheric temperature, stability and humidity — the very factors that greenhouse-gas-driven stratospheric cooling will alter. Recent research even suggests the QBO itself may weaken or disappear under sustained global warming, making its cloud connections all the more urgent to understand. The authors propose their framework as a template for assessing how clouds will respond to that cooling. The message for model developers is equally concrete: fixing the QBO’s descent into the lower stratosphere is necessary but not sufficient. Until climate models represent the true sensitivity of tropical tropopause-layer clouds to upper-tropospheric thermodynamics, they will systematically understate one of the atmosphere’s most elegant teleconnections — and possibly part of the cloud response to a warming world.

Subject of Research: The influence of the quasi-biennial oscillation on tropical high-cloud variability in satellite observations and CMIP6 climate models

Article Title: The role of the QBO for tropical high-cloud variability in CMIP6 models and observations

Article References: Jaison, A. M., Ceppi, P., & Wilson Kemsley, S. (2026). The role of the QBO for tropical high-cloud variability in CMIP6 models and observations. Atmospheric Chemistry and Physics, 26(19), 14403-14416. https://doi.org/10.5194/acp-26-14403-2026

Image Credits: AI Generated

DOI: 10.5194/acp-26-14403-2026

Keywords: quasi-biennial oscillation, tropical high clouds, tropical tropopause layer, CMIP6, CALIPSO, cloud-controlling factors, cloud feedback, stratosphere-troposphere coupling, upper-tropospheric temperature, static stability, relative humidity, climate model bias