A Holistic Investigation of Cloud Droplet Formation Representation in General Circulation Models

Atmospheric aerosol particles are essential to Earth’s climate, serving as the nuclei for cloud droplet formation. The process of cloud droplet formation directly links aerosols and clouds, thereby influencing cloud properties (e.g., albedo and lifetime). Model estimates of effective radiative forcing suggest a net cooling effect from aerosol-cloud interactions (–0.84 W m^‑2), but the wide range of values (–1.45 to –0.25 W m^‑2) has hindered future climate projections (Masson-Delmotte et al., 2021).

Aerosol activation in general circulation models (GCMs) is parameterised, with the Abdul-Razzak & Ghan (2000; ARG) scheme used in the Unified Model for its efficiency, while the Morales Betancourt & Nenes (2014; MN) scheme represents a recent addition. These parameterisations are based on adiabatic cloud parcel model theory and estimate the maximum supersaturation (S_max) from which the cloud droplet number concentration (N_d) is derived. We will demonstrate that an improved GCM representation of cloud droplet formation is vital for constraining estimates of the climatic effect of aerosol-cloud interactions using a three-step holistic framework:

• Direct comparison of parameterisations against an efficient cloud parcel model

In this study the DiffeRential Evolution Adaptive Metropolis (DREAM) Markov Chain Monte Carlo algorithm (Vrugt et al., 2009) is used to compare predictions of S_max and N_d from the ARG and MN schemes with those from the Pseudo-Adiabatic bin-micRophySics university of Exeter Cloud parcel model (PARSEC), using model input parameters selected near-randomly from predefined prior ranges that reflect those used in GCMs. These comparisons show that ARG underestimates both S_max and N_d relative to PARSEC (both by up to ~60%), while MN, to a lesser extent, overpredicts both values. Crucially for future climate projections, we identify differences in model sensitivity to input parameters (e.g., updraft velocity), and parameter combinations, between the parameterisations and PARSEC.

• Offline evaluation against in-situ observations

The DREAM algorithm is applied within an inverse modelling framework to perform N_d closure experiments, using data from three marine aircraft campaigns that span updraft- and number-limited regimes. The results show that, compared to PARSEC, both ARG and MN often fail to match observed N_d, with ARG significantly underestimating N_d. Our framework reveals parameter sensitivities and correlations, offering insights for refining models and guiding future measurement campaigns.

• Online evaluation – exploring the Southern Ocean albedo bias

GCM predictions of cloud albedo in the Southern Ocean are significantly underestimated relative to observations (Mulcahy et al., 2018). Here, PARSEC has been integrated into the UK Met Office climate model, allowing the first online evaluation of existing aerosol activation parameterisations. Importantly, we quantify the impact of cloud droplet formation representation on the observed Southern Ocean albedo bias.

Finally, we will discuss the implications of our findings from (1-3) for the effectiveness of current aerosol activation parameterisations for geoengineering via marine cloud brightening.

(1) Masson-Delmotte, et al., 2021, DOI:10.1017/9781009157896.001, (2) Abdul-Razzak, H. and Ghan, S. J., 2000, DOI: 10.1029/1999JD901161, (3) Morales Betancourt, R. and Nenes, A, 2014, DOI: 10.5194/gmd-7-2345-2014, (4) Vrugt, J. A., et al., 2009, DOI:10.1515/IJNSNS.2009.10.3.273, (5) Mulcahy, J, P., et al., 2018, DOI: 10.1029/2018MS001464.