At the CMIP Community Workshop 2026, the session “Climate sensitivity, clouds, and circulation” brought together a diverse set of perspectives on one of the central challenges in climate science: understanding how clouds and atmospheric processes shape Earth’s climate response.
Despite the variety of approaches, ranging from global storm-resolving models to observational constraints and emerging machine learning techniques, a common thread ran through the session: uncertainty in climate sensitivity remains closely tied to how we represent clouds, aerosols, and small-scale processes in models.
In my contribution, “Consistent Climate Feedbacks Across ICON Resolutions Suggest Model-Dependent Cloud Responses in Global Storm-Resolving Models,” I explored how climate feedbacks change with resolution in the ICON model. Interestingly, global climate feedbacks remained remarkably consistent across storm-resolving resolutions, pointing to a robust large-scale response within a given model framework. At the same time, it reinforces a broader issue: cloud radiative responses to warming appear to be strongly model-dependent in GSRMs, suggesting that differences between models, rather than resolution and convective parametrisations alone, continue to drive uncertainty in climate sensitivity.
This question of uncertainty was taken further by Brian Medeiros, who examined how much of it arises from the models themselves. Using perturbed parameter ensembles in the Community Earth System Model, his work explored how variations in model parameters translate into differences in climate projections. At its core is a fundamental question: how much of what we simulate is determined by physical understanding, and how much by tuning? By systematically sampling parameter space, these experiments help clarify the role of parametric uncertainty and provide a more structured way to interpret model spread.
From there, the focus shifted toward the physics of clouds. Masaki Satoh presented a compelling case for the importance of cloud microphysics, showing how something as specific as cloud-falling speed can influence climate sensitivity. Using observations from the EarthCARE satellite, particularly reflectivity and Doppler velocity, his work connected small-scale processes to large-scale radiative effects across different cloud regimes, including tropical convection and frontal systems. High clouds, in particular, emerged as highly sensitive to microphysical assumptions. A key takeaway was the growing potential to use observations such as top-of-atmosphere radiation, cloud fraction, and optical thickness to constrain models, alongside new initiatives like ECO-MIP aimed at improving cloud representation in high-resolution frameworks.
A different but complementary perspective came from Zhaoyi Shen, whose work pointed toward the integration of physics-based modelling and machine learning. Focusing on turbulence and unresolved processes, this approach aims to train models that better capture subgrid-scale dynamics while maintaining physical consistency. It reflects a broader shift in the field toward hybrid methods that combine data-driven tools with traditional modelling.
The role of aerosols added another layer of complexity. Kayla White explored how anthropogenic aerosols shape the historical evolution of radiative feedbacks, emphasising their strong spatial and temporal variability. Different aerosol types, such as sulfates and black carbon, interact with clouds in distinct ways, and their distribution can partially offset greenhouse gas forcing. This makes it challenging to disentangle causes of past climate change, and highlights how the location and composition of aerosols are critical for understanding feedbacks.
Finally, Yi Huang turned attention to the stratosphere, focusing on the role of water vapour. Both observations and models indicate that the stratosphere is gradually moistening, contributing a positive climate feedback. Using experiments with a climate model, including scenarios where stratospheric water vapour was doubled, his work quantified its radiative impact, showing that while smaller than cloud effects, it remains a non-negligible component of the climate system.
Taken together, the session painted a rich and interconnected picture. What stood out was not just the diversity of methods, but how they complement each other: high-resolution models revealing process-level behaviour, ensembles quantifying uncertainty, observations providing constraints, and new approaches expanding the modelling toolbox.
From my perspective, one message was particularly clear: progress in understanding climate sensitivity will not come from a single approach, but from bringing these perspectives together. At the same time, the persistence of model-dependent cloud responses reminds us that some of the most fundamental challenges remain unresolved.
Bridging the gap between processes and projections, between clouds and climate sensitivity, continues to be a central task for the climate modelling community.
