Evaluating the precipitation impact on particle number size distribution in climate models based on correlation analysis

For models to reliably predict future climate and air-quality scenarios, an accurate representation of the cloud condensation nuclei (CCN) budget is key. In this regard, the effect of precipitation on the particle number size distribution (PNSD) is important in at least two ways: 1) wet deposition, generally considered a dominant sink of CCN, and 2) CCN replenishing, which has been shown to frequently follow precipitation via the process of formation and growth of new particles, thereby buffering the loss process. Together, these effects illustrate the complexity of precipitation–PNSD interactions.

In this study, we use correlations between measured PNSD at three stations and precipitation rates along back trajectories to evaluate precipitation-PNSD interactions in three general circulation models (GCMs; NorESM, EC-Earth and ECHAM-SALSA). This approach allows us to focus on the size- and time-resolved effects of precipitation on the CCN budget. The long-term measurement sites used in the study are Zeppelin (Arctic), Hyytiälä. (boreal forest), and ATTO (Amazon rainforest). To investigate potential confounding factors, we further apply eXtreme Gradient Boosting (XGBoost) and build a separate regression model for each site and data source using a minimal set of physically relevant predictors.

For CCN replenishment following precipitation, the models tend to underestimate new particle formation (NPF) and particle growth to CCN sizes at the two high-latitude stations. In the Amazon (ATTO), by contrast, two models simulate an immediate CCN source after rainfall, whereas observations show a weaker response that takes time to grow to CCN sizes, indicating overly rapid aerosol growth in the models. Finally, observations suggest weaker wet deposition during cold periods than warm periods, likely due to phase dependency. The models are in general better at reproducing patterns during warm periods, while in cold periods one model (EC-Earth) has too strong positive correlations with precipitation, while another has strongly negative correlations (ECHAM-SALSA).

The XGBoost analysis largely confirms the key findings from the correlation evaluation, but also uncovers likely confounding influences, such as the correlation between emission regions and regions with strong precipitation. For example, a feature that appears as a precipitation-driven source of large particles in correlation analyses is instead attributed by the machine-learning model to shifts in air-mass origin. This approach shows potential for disentangling spurious correlations and controlling for confounding factors in model evaluation.

Overall, evaluating the size-resolved impacts of precipitation on particle number highlights model shortcomings in new particle formation and growth, and underscores the importance of disentangling these processes from the direct deposition effect of precipitation when improving models.