Towards an improved understanding of cloud microphysics via data-driven process-rate diagnostics

Small-scale microphysical processes describe the interactions of cloud particles and the phase transitions of condensed water in the atmosphere. In numerical weather prediction and climate models, they are represented empirically by a parametrization scheme, which describes their impact on and coupling to the resolved scale. Incomplete process-level understanding of cloud microphysics contributes to the significant model uncertainties linked to the parameterization of sub-grid scale processes. However, progress in reducing these uncertainties is hindered by the lack of microphysical process rate data. Within the parameterization, microphysical process rates are computed at interim steps to update the prognostic cloud variables. Yet, despite their informative value, they are usually not included in the output of km-scale simulations due to resource limitations.For this purpose, we developed PRecover (microphysical Process Rate recovery), a data-driven post-processing method to recover microphysical process rates in a two-moment microphysics scheme from high-resolution simulation output of the ICOsahedral Nonhydrostatic (ICON) model. Based on machine learning, PRecover emulates the computation of multiple warm-rain and ice microphysical process rates efficiently and flexibly, using a two-step classification-regression approach. Here, we use PRecover for a systematic evaluation of cloud microphysical processes. With a focus on instantaneous process rates, we demonstrate the functionality of PRecover. Additionally, we study the relevance of different microphysical processes and quantify their relative contribution to pathways of precipitation formation, e.g. the relative contributions of autoconversion and accretion to warm rain formation in different cloud regimes. In contrast to previous studies, which were often limited to idealized simulations, we are able to analyze the output of extensive high-resolution simulations in a regional and global configuration.