Cloud condensation nuclei (CCN) participate in controlling the climate, and a better understading of their number concentrations is needed to constrain the current uncertainties in Earth’s energy budget. However, estimating the global CCN concentrations is difficult using only localised in-situ measurements. To overcome this, different proxies and parametrisations for CCN have been developed. In this thesis, accumulation mode particles were used as a substitute for CCN, and continental proxy for number concentration of N100 was developed with CO and temperature as tracers for anthropogenic and biogenic emissions. The data utilised in the analysis contained N100 measurements from 22 sites from 5 different continents as well as CO and temperature from CAMS reanalysis dataset.
The thesis aimed to construct a global continental proxy. In addition to this, individual proxies for each site (the site proxy) and proxies trained with other sites’ data (the site excluded proxy) were developed. The performance of these proxies was evaluated using a modified version of K-fold cross-validation, which allowed estimating the effect of dataset selection on the results. Additionally, time series, seasonal variation, and parameter distributions for developed proxies were analysed and findings compared against known characteristics of the sites.
Global proxy was developed, but no single set of parameters, that would achieve the best performance at all sites, was found. Therefore, two versions of global proxy were selected and their results analysed. For most of the sites, the site proxy performed better than the global proxies. Additionally, based on the analysis from the site excluded proxy, extrapolating the global proxy to new locations produced results with varying accuracy. Best results came from sites with low concentrations and occasional anthropogenic transport episodes. Additionally, some European rural sites performed well, whereas in mountainous sites the proxy struggled. Comparing the proxy to literature, it performed generally less well or similarly as proxies from other studies. Longer datasets and additional measurement sites could improve the proxy performance.
