The research team evaluated historical simulations from 41 climate models participating in the Coupled Model Intercomparison Project Phase 6, known as CMIP6, the international framework that underpins the assessments of the Intergovernmental Panel on Climate Change. Their benchmark was ERA5, the fifth-generation reanalysis produced by the European Centre for Medium-Range Weather Forecasts, which blends observations from satellites, weather balloons and surface stations into a physically consistent record of the recent climate. Reanalyses are not perfect, particularly over the data-sparse Arctic, but they represent the best available reconstruction of what has actually happened in the atmosphere, and they provide a rigorous yardstick against which model performance can be measured.

To compare models fairly, the researchers used a pattern projection metric that scores each model on how well it reproduces both the spatial pattern and the amplitude of lower-tropospheric warming across the Arctic, defined here as the region poleward of 75 degrees north. Rather than simply asking whether a model produces the right average temperature trend, this approach asks whether the geography of the warming matches reality: does the model place the strongest warming where observations do, and does it capture the right magnitude? Based on these scores, the 41 models were sorted into four groups, ranging from those that faithfully capture the observed warming to those that underestimate it, exaggerate it, or even reverse its sign in the lower troposphere.

The most eye-catching result of the analysis is a correlation coefficient of 0.91 between simulated wintertime warming and simulated moistening across the model ensemble. In climate science, correlations that high are rare, and they point to an intimate thermodynamic coupling: models that warm the Arctic more also moisten it more, and vice versa. The physical logic is straightforward. Water vapour is itself a powerful greenhouse gas, so additional moisture traps more outgoing longwave radiation and radiates extra energy back down to the surface, warming it further. Warmer air, in turn, can hold more moisture, closing a feedback loop. But the new study goes beyond documenting this coupling; it dissects the physical pathways through which moisture accumulates in the Arctic air column, and asks whether the models that get the warming right are also the ones that get the moisture pathways right.

The team identified two distinct moist-process feedbacks operating over the Arctic in winter. The first is a remote feedback. Warming patterns alter stationary atmospheric waves, the large-scale undulations in the circulation that steer weather systems, and these changes strengthen the transport of moist air poleward from lower latitudes. When humid air masses intrude into the Arctic, they carry with them both latent heat and the capacity to emit downward longwave radiation, warming the surface and the air above it. This mechanism has been implicated in previous studies of winter sea-ice decline in the Atlantic sector of the Arctic, where moist intrusions from the North Atlantic and North Pacific have been linked to episodes of rapid ice loss and dramatic temperature spikes.

The second pathway is a local feedback. As Arctic warming retreats the sea-ice edge, more open water is exposed to the cold winter atmosphere, and evaporation from the ocean surface increases. That locally generated moisture is added to the atmospheric column, enhancing the greenhouse effect close to where it is produced. Both feedbacks, remote and local, converge on the same signature: enhanced downward longwave radiation at the surface, surface warming, and increased near-surface diabatic heating, the release of heat into the lower atmosphere associated with condensation and other thermodynamic processes. The diagnostics used in the study trace these components across the Arctic, allowing the researchers to compare, region by region, how each model handles each link in the chain.

The comparison revealed a clear hierarchy. Models that more accurately reproduce the observed lower-tropospheric warming also better reproduce the directions, spatial patterns and magnitudes of the diagnosed components of both feedbacks. Models that underestimate, exaggerate or reverse the observed warming show corresponding biases in their moisture physics. In other words, getting the Arctic winter right is not a matter of luck or of tuning a single parameter; it depends on whether a model can represent the combined workings of poleward moisture transport and local evaporation-driven moistening. This provides climate scientists with a practical diagnostic: by checking how a model handles these two moist-process feedbacks, researchers can gauge, at least in part, how trustworthy its Arctic warming projection is likely to be.

Even the best-performing models, however, are not off the hook. Over the Barents-Kara Seas, the region between roughly 75 and 85 degrees north and 0 to 90 degrees east, which the study identifies as the principal Arctic warming hotspot, all models retain biases in the relative contributions of the two processes. There, models tend to overestimate local evaporation while underestimating the vertically integrated convergence of moisture fluxes, the net accumulation of moisture transported in from elsewhere. These two errors partially cancel each other out, which means a model can produce a plausible total moisture amount for the wrong reasons. Such compensating biases are notoriously difficult to detect in standard evaluations, and they matter because the balance between locally sourced and remotely transported moisture shapes where clouds form, where longwave radiation is enhanced, and ultimately where the strongest warming occurs.

The Barents-Kara hotspot deserves particular attention because it sits at the crossroads of Atlantic influence and Arctic climate. Warm, salty Atlantic waters flow into the Barents Sea, and the region has experienced some of the fastest winter sea-ice losses anywhere in the Arctic. The new findings suggest that the interplay between ocean-driven local evaporation and atmospheric moisture imported from lower latitudes is the crux of why models disagree about this region, and that resolving these biases is essential for narrowing uncertainty in projections of Arctic amplification. Given that Arctic warming influences mid-latitude weather patterns, permafrost carbon stores and global sea level, the stakes of getting these feedbacks right extend well beyond the polar night.

For the broader climate modelling community, the study offers both a warning and a roadmap. The warning is that a model’s overall global performance does not guarantee fidelity in the winter Arctic, where unique physics, including sea-ice thermodynamics, stable boundary layers and persistent temperature inversions, test model formulations in ways the tropics do not. The roadmap is the process-based evaluation framework itself: by scoring models not just on their warming answer but on the physical components that produce it, scientists can identify which links in the causal chain are broken and target model development accordingly. As the Arctic continues to warm at a pace that already approaches four times the global average, understanding which models capture the machinery of polar amplification, and which merely approximate its outcome, becomes a matter of considerable consequence for everyone who depends on credible climate projections.

Subject of Research: Evaluation of CMIP6 climate model performance in simulating accelerated wintertime Arctic warming and associated moist-process feedbacks

Article Title: Assessing the performance of CMIP6 models in simulating the accelerated wintertime Arctic warming

Article References: Assessing the performance of CMIP6 models in simulating the accelerated wintertime Arctic warming. (n.d.). https://doi.org/10.1007/s00382-026-08395-7

Image Credits: AI Generated

DOI: 10.1007/s00382-026-08395-7

Keywords: Arctic warming, CMIP6, climate models, polar amplification, moisture transport, sea ice, evaporation, downward longwave radiation, Barents-Kara Seas, ERA5, water vapour feedback, Climate Dynamics

The research, led by scientists from the National Institute of Polar Research in Japan and Nagoya University, draws on measurements collected during the 64th Japanese Antarctic Research Expedition, known as JARE64, aboard the research icebreaker R/V Shirase. Between December 2022 and March 2023, as the vessel threaded its way across the Southern Ocean toward Antarctica, instruments aboard the ship continuously recorded cloud properties, atmospheric temperature and humidity, surface radiation, and aerosol concentrations. This sustained campaign produced one of the most comprehensive observational benchmarks ever assembled for evaluating how well models reproduce clouds and radiation over a region where ground truth is extraordinarily scarce. Professor Jun Inoue of the research team notes that numerical models have long shown poor skill in reproducing clouds, particularly over the Southern Ocean and Antarctica, where cloud-related biases propagate into errors in the surface energy budget through distortions of the radiative budget.

To test how well existing tools perform against these new measurements, the team evaluated three widely used datasets: the ERA5 and MERRA-2 atmospheric reanalyses, which blend observations with numerical weather prediction models to create continuous global records, and the CAM-ATRAS climate model, which simulates the atmosphere from first physical principles. Reanalyses are the workhorses of climate research, providing researchers with gridded estimates of atmospheric conditions even in data-sparse regions, so any systematic flaws in their cloud representation can quietly contaminate a wide range of downstream studies. The comparison covered the full span of the expedition, allowing the scientists to assess cloud occurrence, cloud phase, and the downward longwave radiation that reaches the ocean or sea ice surface across a broad swath of the Southern Ocean in a single coherent framework.

The results revealed a pattern of errors that is both subtle and consequential. All three datasets broadly captured the large-scale patterns of cloudiness over the Southern Ocean, which might initially suggest that models are performing reasonably well. But the details told a different story. ERA5 and MERRA-2 consistently overestimated how often low-level clouds occurred, painting the boundary layer as cloudier than it really was. CAM-ATRAS, by contrast, came closest to the observations, particularly in reproducing both the frequency of clouds and their phase, that is, whether they were composed of liquid water droplets or ice crystals. Cloud phase matters enormously in this environment, because supercooled liquid clouds emit radiation differently than ice-containing clouds and persist under conditions that models often misjudge.

The most surprising finding, however, was that despite simulating abundant low-level clouds, all three datasets underestimated the amount of downward longwave radiation reaching the surface. Downward longwave radiation is the infrared heat emitted by clouds and atmospheric gases back toward the ground, and over the Southern Ocean it is a critical component of the surface energy budget, helping to determine sea surface temperatures, sea ice formation, and the exchange of heat between ocean and atmosphere. If models produce plenty of clouds yet still deliver too little heat to the surface, something fundamental is wrong with the physical character of those simulated clouds, not merely their quantity. This discrepancy, the researchers realized, pointed toward deeper problems in how the models represent cloud microphysics and the thermal structure of the atmosphere.

To investigate the role of aerosols, the tiny airborne particles on which cloud droplets and ice crystals form, the team first compared aerosol concentrations in the reanalysis datasets against the shipboard measurements. The reanalyses, it turned out, contained higher aerosol concentrations than were actually observed over the Southern Ocean, a region famously remote from industrial pollution sources and often used as a natural laboratory for studying pristine marine clouds. The researchers then ran sensitivity experiments with CAM-ATRAS, artificially increasing aerosol emissions across the Southern Hemisphere to see how the added particles would influence cloud formation and surface radiation. The experiments showed that higher aerosol concentrations did indeed produce more low-level clouds, but the effect on surface radiation was limited. Aerosol loading alone, in other words, could not explain the radiation deficit.

The true culprit lay in the physical properties of the simulated clouds rather than in their abundance. In the models, the clouds contained excessive ice, and ice-rich clouds emit less infrared heat downward than the warmer, liquid-dominated clouds actually observed over the Southern Ocean. This microphysical flaw reduced the longwave radiation reaching the surface even as the models churned out plenty of cloud cover. Yet even correcting for this did not close the gap entirely. The study demonstrated that biases in cloud representation alone cannot account for the underestimated downward longwave radiation; the numerical models also carry an inherent cold temperature bias, systematically simulating atmospheric temperatures that are too low, which further suppresses the infrared emission reaching the surface. Both problems, the excessive ice and the cold bias, conspire to starve the modeled surface of heat.

The implications of this diagnosis reach well beyond the Southern Ocean. For decades, climate models have exhibited persistent biases in this region, including the notorious double bias in which too little sunlight is absorbed and the Southern Ocean remains too cold in simulations, with knock-on effects for global heat uptake and projections of future warming. The new findings indicate that accurately representing cloud phase and atmospheric temperature is more important than simply reproducing how often clouds appear. A model can generate the right amount of cloudiness and still get the energy budget badly wrong if those clouds are made of the wrong stuff and sit in an atmosphere that is too cold. This reframing of the problem gives model developers a concrete target: improving cloud microphysics schemes, refining aerosol-cloud interactions, and correcting background temperature biases will do more for simulation fidelity than tuning cloud frequency alone.

Progress, the researchers emphasize, will depend on both better observations and better use of the observations that already exist. Because measurements over Antarctica and the surrounding ocean remain sparse, numerical models still carry substantial uncertainties in their representation of the polar atmosphere. Assistant Professor Kazutoshi Sato, who is affiliated with the National Institute of Polar Research, points out that incorporating existing but currently underutilized observations into numerical models may offer an effective solution. He highlights the example of the PANSY radar at Japan’s Syowa Station, a sophisticated atmospheric radar whose measurements are not yet routinely assimilated into numerical weather prediction systems. Assimilating such data, he suggests, could help reduce model biases and improve forecast accuracy across the region. Expanded observational campaigns over the Southern Ocean and Antarctica, particularly of fundamental variables such as temperature, will be equally essential for rooting out the cold bias identified in this study.

By tracing persistent cloud biases to their physical roots, the JARE64 analysis marks an important step toward more trustworthy climate projections. The Southern Ocean will remain a linchpin of Earth’s climate system, absorbing heat and carbon while moderating the pace of global warming, and the clouds above it will continue to set the terms of that bargain. Models that misrepresent the phase of those clouds and the temperature of the air around them will misjudge how much heat the surface receives, with consequences for sea ice, ocean circulation, and the trajectory of climate change itself. The voyage of the R/V Shirase has shown that the path to better predictions runs through the details: the ice crystals in a cloud, the droplets on an aerosol particle, and the degrees of warmth in a column of air above one of the most inhospitable oceans on Earth.

Subject of Research: Evaluation of low-level cloud, temperature, and radiation simulations over the Southern Ocean using ship-based Antarctic expedition observations, reanalysis datasets, and a climate model

Article Title: Antarctic expedition reveals why Southern Ocean clouds remain one of climate science's greatest challenges

Article References: Antarctic expedition reveals why Southern Ocean clouds remain one of climate science's greatest challenges. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: Southern Ocean, clouds, climate models, Antarctica, downward longwave radiation, cloud phase, aerosols, reanalysis, JARE64, surface energy budget, cloud microphysics, cold bias