Process?Based Climate Model Development Harnessing Machine Learning: III. The Representation of Cumulus Geometry and Their 3D Radiative Effects

Process-scale development, evaluation, and calibration of physically based parameterizations clouds radiation are powerful levers for improving weather climate models. In a series papers, we propose strategy process-based models that uses machine learning techniques. It relies on systematic comparisons single-column versions with explicit simulations boundary-layer dynamics (Large-Eddy Simulations [LES]). This paper focuses the cloud geometry parameters (vertical overlap, horizontal heterogeneity, size) appear in parameterization radiation. The solar component radiative transfer (RT) scheme includes 3D effects (SPARTACUS) is run offline mode an ensemble input profiles synthesized from LES outputs. space parameter values efficiently explored by sampling large number sets (configurations) which metrics computed using fast surrogate emulate SPARTACUS solver. sampled configurations evaluated comparing these to reference provided RT Monte Carlo model. best calibrated yield better predictions TOA surface fluxes than one fields: root-mean-square errors averaged over cumulus fields angles reduced ?10 Wm?2 LES-derived ?5 adjusted parameters. However, fails reduce absorption, remain around 2–4 Wm?2.