Seminar 16 (August 27, 2025) of the CleanCloud series. 
Speaker is Prof. Harri Kokkola (Finnish Meteorological Institute; University of Eastern Finland)

Abstract: Understanding aerosol–cloud interactions is vital for accurate climate predictions, yet these interactions remain highly uncertain. Our research highlights significant challenges in both observations and modelling. Satellite retrievals of cloud droplet number concentration (CDNC) often show substantial positive biases, primarily due to spatial cloud variability and instrumental noise. Despite this, these studies suggest that changes in cloud liquid water path can be confidently determined from satellite data when atmospheric conditions are carefully selected. Conversely, long-term in situ observations reveal a higher susceptibility of cloud formation to aerosol changes than typically captured by satellite estimates, implying a stronger aerosol indirect radiative forcing at the upper end of current uncertainty ranges. A key issue is that traditional statistical methods applied to satellite data often yield unphysical relationships between cloud condensation nuclei (CCN) and CDNC, largely because they fail to account for the crucial influence of other factors e.g., activation updrafts. Machine learning approaches offer a more robust solution, providing physically plausible estimates by isolating the effect of CCN and considering other atmospheric variables. This underscores the critical need for improved model physics, especially regarding updraft velocities and aerosol size distributions, and more sophisticated analytical techniques to reduce uncertainties in aerosol–cloud interactions.