Towards the Emulation of Aerosol Effects in Convective Precipitation in a General Circulation Model

While aerosol-cloud interactions (ACI) are represented with a high degree of detail in small-scale high-resolution models (high-res), they are not taken into account in the parameterization of convection, which means that there is no representation of the aerosol load in e.g. the Tiedtke-Bechtold convection scheme. The aim of this study is to use the results of high-res experiments as input data for the training of an emulator predicting the ACI under different environmental conditions. This emulator will be embedded in the convection parameterization of the ICON model configured as general circulation model (GCM).
We will present the creation of the training data set as well as the emulator itself. The training data set consists of 206 experiments in the high-res setting of ICON (300 m grid spacing, torus grid, two-moment cloud scheme, 3D turbulence). An adaptation of idealized Weisman-Klemp profiles representing the environmental conditions of convective cells in several GCM experiments was used as input data, combined with perturbations in various input variables. The emulator is trained on the CCN-sensitivity for precipitation. In order to calculate these sensitivities, every experiment is run twice: once with a specific amount of aerosol and a second time with half the amount of aerosols. The training inputs are derived features like CAPE, relative humidity and mean updraft speed, which can also be extracted from the GCM setting later on. As validation we will present R² scores, RMSE and SHAP values. Due to the high variability of the investigated convective systems, the sensitivities to CCN, which contributes only to a small part of the total variability, is very hard to predict and varies a lot – even under very similar environmental conditions. Therefore, a positive correlation of R² ~ 0.4 (depending on the configuration) is seen as a success.
A first version of the emulator embedded in the convective scheme of the GCM will also be presented.