. Improving the modeling of collision-coalescence in a two-moment microphysics scheme with neural networks

Abstract
A bin-emulating two-moment microphysics scheme for warm cloud has been successfully applied to the simulation of marine stratocumulus and cumulus clouds. In this model, the mass and number concentration tendencies of cloud and rain modes due to collision-coalescence given mass and number concentrations of cloud and drizzle modes come from a lookup table which stores pre-calculated results using a bin model. The size of the lookup table becomes unmanageable if the mass and number concentrations of cloud and drizzle modes span a wide range. A neural network (NN) is trained to map the mass and number concentrations of cloud and drizzle modes to the tendencies calculated by the bin model. The neural network is implemented into the SAM model to simulate precipitating marine stratocumulus and cumulus clouds. We will discuss the advantages of using NN over both lookup table and the bin model.