A Bulk Single Category, Four-Moment Warm Rain Microphysics Scheme

Weather and climate models often struggle to represent low-level warm (liquid) clouds realistically. Previous studies have shown considerable sensitivity of warm cloud properties to changes in microphysics scheme parameters affecting, for example, the rate of precipitation formation. This can alter key climate features such as aerosol-cloud interactions, precipitation characteristics, and climate sensitivity. Current bulk cloud microphysics schemes in weather and climate models use separate categories for cloud (small drops) and rain (large falling drops). This two-category approach necessitates an ad-hoc “autoconversion” process to convert cloud droplets that collide and coalesce into rain drops. Autoconversion is usually formulated based on detailed process model output, but this typically assumes a drop size threshold to separate cloud and rain which contributes to large uncertainty in how rain formation is parameterized. In this study, we test a novel approach that does not have separate cloud and rain categories and instead represents small and large drops using a single liquid category. Drop collision-coalescence rates are calculated following the Arbitrary Moment Predictor (AMP) method from Igel (2019). This has an improved physical basis compared to current two-category bulk schemes and obviates the need for an ad-hoc autoconversion process to represent rain formation. In box model and 1D tests, the single liquid category approach gives much smaller error than current schemes over a wide range of cloud conditions when compared to a detailed reference model for rain formation (the Tel Aviv University bin microphysics scheme). Sedimentation of drops is also improved relative to the reference model. The single liquid category approach uses a mapping method to obtain drop size distribution parameters from the predicted state variables and numerical integration to obtain the collision-coalescence rates. These are both computationally expensive, and machine learning with neural networks is used to speed up the calculations and make the approach tractable.