| 10 Sep 2025
Remaining aerosol forcing uncertainty after observational constraint and the processes that cause it
Abstract. Aerosol radiative forcing remains a major source of climate model uncertainty, limiting climate model projection skill and slowing global action on addressing climate risks. Observations only modestly constrain the magnitude of aerosol radiative forcing despite advances in model fidelity, resolution and availability of observations. Our goals are to understand where aerosol-cloud forcing uncertainty resists efforts to reduce (or constrain) it and to identify the processes that cause the remaining uncertainty, to guide future observation campaigns and model constraint efforts. We map the aerosol forcing uncertainty in a global climate model perturbed parameter ensemble before and after constraint to satellite observations of several cloud, aerosol and radiative properties. Original uncertainty falls by more than 80 % in Northern Hemisphere marine regions and by 70 % for globally averaged aerosol forcing. However, the uncertainty remains large (more than 70 % of the original uncertainty) in Southern Hemisphere marine environments where stratocumulus clouds transition to cumulus, as well as in some highly populated industrialized areas. Regional clusters of shared causes of model uncertainty highlight common processes as targets for future observational constraint. Our findings highlight the value in re-evaluating the remaining causes of ΔFaci uncertainty during the constraint process and provide actionable information for prioritizing existing observations that should be included as constraints. Additionally, our results highlight targeted observations in persistent uncertainty hotspots where novel and process-specific data could further constrain aerosol forcing. This work provides a framework for model evaluation and development that prioritises aerosol forcing constraint to improve model skill at making climate projections.
Competing interests: At least one of the (co-)authors is a member of the editorial board of Atmospheric Chemistry and Physics.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.
Review for ACP of Regayre et al. manuscript entitled “Remaining aerosol forcing uncertainty after observational constraint and the processes that cause it.”
Summary and General Comments:
This analysis uses an existing climate model perturbed parameter ensemble (PPE) and after first discussing how the progressive/sequential addition of observations yields constraints (with decreasing return) on aerosol forcing, they analyze the role of parameters on the remaining aerosol forcing uncertainty. Interestingly, they find that parameters that initially seem unimportant contribute a larger role after the optimal constraint. They then take a different approach and use a K-means clustering algorithm to find the dominant clusters of parameters that contribute to the uncertainty, and they plot these clusters onto a latitude-longitude map, enabling visualization of spatial extent of clusters relative to the latitude-longitude resolved aerosol forcing uncertainty map. The paper has numerous quantitative details of parameter/cluster contributions to the uncertainty, and one has to pay close attention so as to not get lost in details, but the analysis is interesting, and the clustering approach toward understanding uncertainty contributions is an interesting approach that warrants publication. I do not have major comments, but I do have a few questions (below), mostly about the chosen clustering approach, that the authors may wish to consider in their revision.
Specific Comments/Questions:
I see that in the PPE being analyzed, less than 10 non-aerosol parameters were perturbed, which is small. But climate models have far more than 10 physics parameters that impact surface wind, surface-atmosphere fluxes, turbulence, convective mass flux, clouds, and by extension, aerosol-convection-cloud interactions. One cannot at this point go back to re-do this particular PPE, and I applaud that there are many aerosol parameters (the other 27 parameters), but is it overall possible that the non-aerosol portion of the parameter space really was not explored much and results are muddled by a lack of effort to explore the many other parameters that impact environments and clouds? Thus, when speaking of “remaining” uncertainty, would not the full parametric uncertainty be much larger and we’re really not at the level of looking at remaining parametric uncertainty?
Aside from the land domain of eastern Asia/China, it is obvious that the other geographic regions of large uncertainty in Fig. 2c – namely, the stratocumulus regimes off the west coasts of the continents – are covered by a diverse range of clusters (Fig. 4, with light/dark blues, oranges, reds, greens, purples, etc.). I did not anticipate this; in general, I would have thought that there would be a mixing of clusters in geographic domains where the uncertainty was smaller in absolute W/m2. Intuitively, for example, I would think that the oceanic stratocumulus regimes would have emerged as a consistent color that then transitioned to another color to the west as those cloud types transitioned into cumulus.
With the above paragraph in mind, how does one interpret this spatial mixing of clusters in Fig 4 coincident with the bullseyes of larger absolute W/m2 uncertainties in Fig. 2c? Might this imply that these clusters as defined are not as useful and could this have arisen due to arbitrary partitioning of a continuous distribution into 10-ish clusters? K-means clustering is most appropriate for data that does exhibit unique modes or clusters, but of course, one can impose k-means clustering on any continuous dataset, even a continuous one not suitable for clustering. Which brings up my second question – how were the number of clusters determined? Finally, instead of clusters connected to geographic domains, would it make more sense to define clusters according to cloud controlling factors (i.e., diagnostics related to stability, moisture, dynamic indices) and then check to see if similar clusters emerge no matter where you are on Earth, so long as you have a similar thermodynamic and/or dynamic environment?