Contribution of chemical components to cloud condensation nuclei prediction across multiple continental sites

Aerosol–cloud interactions represent a major uncertainty in climate projections. Cloud condensation nuclei (CCN) are particles upon which cloud droplets form, yet in situ CCN measurements remain scarce. Aerosol chemical component exerts important control over CCN activity. It remains unclear how reliably aerosol chemical composition predicts CCN number concentration (N CCN ). This study compiled a dataset of N CCN (supersaturation of 0.1 to 1%) collocated with aerosol chemical components at 20 global continental sites, spanning from remote natural to heavily polluted regions. This dataset was applied to evaluate global N CCN retrieval products from spaceborne Cloud Aerosol Lidar with Orthogonal Polarization (CALIOP) and Copernicus Atmosphere Monitoring Service (CAMS), and to develop a chemical component-based random forest (RF) framework for N CCN prediction and key predictor identification at single-site, site-type, and unified spatial scales. CALIOP showed total normalized mean bias (NMB) of 33.6 ± 79.4% at 0.2% supersaturation (Pearson’s r of 0.72), and CAMS showed –33.6 ± 61.5% to 48.6 ± 79.7% across 0.1 − 1.0% supersaturation (Pearson’s r of 0.29 to 0.60). Three RF modeling scales produced an overall Pearson’s r of 0.80–0.85 and total NMB of −17.3 ± 20.9% to 0.1 ± 49.6% across 0.1–1% supersaturation over the compiled dataset, confirming the feasibility of this framework for N CCN prediction. Interpretability analysis identified inorganic species (ammonium at anthropogenic sites and sulfate at natural sites) as leading predictors at 0.1% supersaturation, whereas organic matter becomes dominant for higher supersaturation. Future work incorporating more observational datasets across diverse environments is expected to further improve the accuracy and generalizability of N CCN predictions.