Process evaluation suggests models misrepresent the precipitation-driven replenishment of cloud condensation nuclei

Accurately modeling the cloud condensation nuclei (CCN) budget is a key factor in reducing uncertainty in aerosol–cloud interactions in Earth system models. Wet deposition – the removal of particles by precipitation – is a major CCN sink, but rainfall can also trigger a replenishment phase via the formation and growth of new particles, partially offsetting losses. However, the ability of general circulation models (GCMs) to capture this precipitation-driven replenishment and size-dependent losses remains under-explored. Here, we evaluate three GCMs representation the size- and time-resolved effects of precipitation on the particle number size distribution (PNSD) and CCN budget, based on correlations between PNSD and precipitation rate along back trajectories from three long-term measurement stations. To better isolate the role of precipitation from confounding factors, we also apply a Machine Learning approach (XGBoost), training one regression model per site and source using a minimal set of physically relevant predictors. Our results show that at the two high-latitude stations, the models underestimate CCN replenishment following precipitation, with too weak new particle formation and growth. At ATTO, two of the models instead overestimate this effect, simulating an immediate CCN source after rainfall. Observations also suggest that CCN removal is weaker during colder conditions, a pattern that models struggle to capture – either overestimating or underestimating the precipitation effect, depending on the model. The XGBoost analysis confirms the key findings of the correlation analysis while helping to correct for likely confounding influences, showing promise for disentangling spurious correlations in model evaluation in process evaluation.