Machine Learning derived CCN concentrations provide better constraints on the first aerosol indirect effect than aerosol optical properties

The first indirect effect of aerosols on cloud reflectivity, primarily through changes in cloud droplet number concentration (N_d) or effective radius (R_eff), remains one of the most uncertain components of anthropogenic radiative forcing. The strength of the first aerosol indirect effect (AIE) is quantified using relationships between aerosol proxies and N_d/R_eff. For large-scale assessments, these relationships have historically been observed via satellites and serve as critical constraints for climate models calculating radiative forcing from aerosol-cloud interactions (ACIs). They have often relied on observations of aerosol optical depth or aerosol index, which are column-integrated proxies for Cloud Condensation Nuclei (CCN) concentration that may not be directly relevant for studying ACIs. Additionally, these proxies are influenced not only by particle concentration but also by size distribution, composition, and relative humidity. Since CCN represents only a fraction of the aerosol size distribution, there may not always be an obvious correlation between CCN and optical properties, introducing uncertainties in estimating indirect effects when using aerosol optical properties.

To address this issue, we developed a machine learning approach to estimate the vertical profile of CCN concentration at 0.4% supersaturation using airborne High Spectral Resolution Lidar Generation-2 (HSRL-2) data and collocated in situ CCN, the latter as truth to train a neural network model. Reanalysis data were used to enhance model performance. Our algorithm predicts vertically resolved CCN concentration within a mean relative uncertainty of 20% and is applicable to EarthCARE/ATLID measurements.  Utilizing this new CCN product derived from the full suite of HSRL-2 extinction and backscatter measurements and reanalysis data of relative humidity and temperature in ACTIVATE (Aerosol Cloud meTeorology Interactions oVer the western ATlantic Experiment) along with collocated cloud properties retrieved from Research Scanning Polarimeter data, we investigate the first AIE over the western North Atlantic Ocean. Our preliminary findings indicate that the new CCN product consistently constrains the relationships between CCN and N_d/R_eff, for a wide range of cloud liquid water paths. The separation of indirect effects for different aerosol types indicates the expected differences in aerosol properties relevant for ACI. Overall, our approach using ML-derived CCN yields tighter constraints and physically more plausible insights into ACIs than vertically-resolved aerosol extinction, vertically-resolved aerosol index (extinction multiplied by Angstrom exponent), or column-integrated aerosol optical depth. We will conclude our presentation by illustrating that the aerosol vertical distribution and hygroscopic growth characteristics are the primary reasons why aerosol optical properties are inadequate for directly constraining the first AIE in the western North Atlantic Ocean.