Physically regularized machine learning emulators of aerosol activation

Abstract. The activation of aerosol into cloud droplets is an important step in the formation clouds and strongly influences radiative budget Earth. Explicitly simulating Earth system models challenging due to computational complexity required resolve necessary chemical physical processes their interactions. As such, various parameterizations have been developed approximate these details at reduced cost accuracy. Here, we explore how machine learning emulators can be used bridge this gap parameterization We evaluate a set detailed parcel model using physically regularized regression techniques. find that reproduce higher accuracy than many existing parameterizations. Furthermore, regularization tends improve emulator accuracy, most significantly when emulating very low fractions. This work demonstrates value constraints development enables implementation improved hybrid next-generation models.