What aerosol and cloud observations reveal about model robustness and aerosol forcing uncertainty

Aerosol radiative forcing remains one of the largest sources of uncertainty in climate projections, despite substantial advances in aerosol and cloud observations from ground-based networks and satellites. We quantify how effectively current aerosol, cloud, and radiative observations constrain aerosol forcing uncertainty in global climate models, and identify where and why significant uncertainty persists. Using large perturbed-parameter ensembles of an Earth system model evaluated against satellite-derived aerosol, cloud, and radiation products, we map the spatial distribution of aerosol forcing uncertainty before and after observational constraint.

We show that observational constraints reduce aerosol forcing uncertainty by more than 70–80% in Northern Hemisphere marine regions and substantially narrow the global mean forcing range. However, large uncertainties remain in key regions, notably Southern Hemisphere stratocumulus-to-cumulus transition zones and some industrialized continental areas. Analysis of uncertainty clusters reveals common controlling processes that resist constraint even when multiple observational datasets are applied.

A central outcome of this work is that applying observational constraints to large perturbed-parameter ensembles provides new insights into structural model behaviour. Simultaneous evaluation against multiple observed aerosol and cloud properties reveals where model tuning is by necessity a compromise. This approach exposes structural model deficiencies – model development priorities.

Our results provide actionable guidance for aerosol measurement communities, including ACTRIS and Harmonia, by identifying regions and processes where improved, harmonized aerosol and cloud observations could most effectively reduce aerosol-cloud radiative forcing uncertainty. The study underscores the need for coordinated observation-model development strategies to maximize the value of long-term aerosol datasets.