Parameterization of Shallow Cumulus Entrainment and Detrainment Rates Using Global Satellite Observations: Physical and Machine Learning Approaches

Abstract Accurate parameterizations of entrainment and detrainment rates ( ε and δ ) of shallow cumulus are crucial for improving simulations of atmospheric energy and water cycles. Existing ε and δ parameterizations are often derived from theoretical derivations, limited observations, or numerical simulations, which lack comprehensive global observational support. To address this limitation, parameterizations of shallow cumulus ε and δ are developed and evaluated using a global satellite‐derived data set. For parameterizations based on the physical approach, the recommended scheme incorporates environmental relative humidity (RH e ) and vertical velocity to parameterize ε , while δ is parameterized using ε and RH e . Furthermore, the machine learning (ML) approach trained on thermodynamic, dynamic, and cloud microphysical properties can accurately predict ε and δ . Comparative analysis reveals that ML performs better than the physical approach. These findings provide valuable insights for refining cumulus parameterizations and enhancing the accuracy of climate model simulations.