Identifying and quantifying uncertainties in process representations regulating aerosol-cloud-precipitation interactions and constraining them using deep learning techniques

Proper characterization of clouds and precipitation that resembles reality remains a major challenge for global atmosphere models. Here we show that significant improvements can be achieved through model retuning and better integration of parameterizations in the Energy Exascale Earth System Model (E3SM) Atmosphere Model version 1 (EAMv1). We find that the significant improved simulation of present-day cloud and precipitation climatology (including the marine stratocumulus, the stratocumulus-to-cumulus transition, the mid-latitude storm track clouds, the high-latitude clouds, and the cloud deck over the tropical warm pool) is accompanied with weaker aerosol-cloud-precipitation interactions (ACPI). The weaker ACPI is associated with a change of the simulated cloud regimes, which consists of different mechanisms (e.g., weaker convective mixing, weaker cloud top entrainment, less frequent decoupled boundary layer, and stronger inversion strength) and different cloud state (e.g., cloud liquid water path, droplet number, fraction, and depth). To constrain ACPI, we use deep learning techniques to derive new metrics to quantify the role of other significant aerosol, cloud, precipitation, or meteorological variables that have been conventionally overlooked in constructing ACPI metrics, providing a regime-independent ACPI metric. This study compares results from incremental parameter changes in the same model, identifying processes in parameterizations that affect model cloud and precipitation climatology, process rates, and their responses to aerosol perturbations. These efforts provide insights into processes affecting clouds, precipitation, circulation, and the climate, and highlight the importance of evaluating these process representations and their relationships with system responses against theories and observations in future model development.