CCN estimations at a high-altitude remote site: role of organic aerosol variability and hygroscopicity

Abstract

High-altitude remote sites are unique places to study aerosol–cloud interactions, since they are located at the altitude where clouds may form. At these remote sites, organic aerosols (OAs) are the main constituents of the overall aerosol population, playing a crucial role in defining aerosol hygroscopicity (κ). To estimate the cloud condensation nuclei (CCN) budget at OA-dominated sites, it is crucial to accurately characterize OA hygroscopicity (κOA) and how its temporal variability affects the CCN activity of the aerosol population, since κOA is not well established due to the complex nature of ambient OA. In this study, we performed CCN closures at a high-altitude remote site during summer to investigate the role of κOA in predicting CCN concentrations under different atmospheric conditions. In addition, we performed an OA source apportionment using positive matrix factorization (PMF). Three OA factors were identified from the PMF analysis: hydrocarbon-like OA (HOA), less-oxidized oxygenated OA (LO-OOA), and more-oxidized oxygenated OA (MO-OOA), with average contributions of 5 %, 36 %, and 59 % of the total OA, respectively. This result highlights the predominance of secondary organic aerosol (SOA) with a high degree of oxidation at this high-altitude site. To understand the impact of each OA factor on the overall OA hygroscopicity, we defined three κOA schemes that assume different hygroscopicity values for each OA factor. Our results show that the different κOA schemes lead to similar CCN closure results between observations and predictions (slope and correlation ranging between 1.08–1.40 and 0.89–0.94, respectively). However, the predictions were not equally accurate across the day. During the night, CCN predictions underestimated observations by 6 %–16 %, while, during morning and midday hours, when the aerosol was influenced by vertical transport of particles and/or new particle formation events, CCN concentrations were overestimated by 0 %–20 %. To further evaluate the role of κOA in CCN predictions, we established a new OA scheme that uses the OA oxidation level (parameterized by the f44 factor) to calculate κOA and predict CCN. This method also shows a large bias, especially during midday hours (up to 40 %), indicating that diurnal information about the oxygenation degree does not improve CCN predictions. Finally, we used a neural network model with four inputs to predict CCN: N80 (number concentration of particles with diameter > 80 nm), OA fraction, f44, and solar global irradiance. This model matched the observations better than the previous approaches, with a bias within ± 10 % and with no daily variation, reproducing the CCN variability throughout the day. Therefore, neural network models seem to be an appropriate tool to estimate CCN concentrations using ancillary parameters accordingly.