NASA GISS ModelE3: New Tuning Framework and Evaluation against Global Satellite Datasets

Abstract:

The NASA GISS ModelE3 general circulation model (GCM), recently completed for CMIP6, exhibits a number of improvements in the representation of moist turbulence, convective cloud systems, and stratiform cloud micro- and macro-physics. Coincident with the development and implementation of multiple new parameterizations is the introduction of numerous unconstrained GCM parameters. A new machine-learning tuning framework was developed to mine this multidimensional free-parameter state space, with the goal of determining to what extent multiple diverse parameter combinations lead to GCM-simulated energy and hydrologic cycle fields that agree with the large suite of satellite product climatologies available. A key new feature of the tuning framework is the accounting of observational uncertainties (or biases) in satellite products, an effort that is expected to become more important as climate models advance and errors approach the quantitative differences between two satellite products aiming to estimate the same parameter. The role of the Observations for Model Intercomparisons Project (Obs4MIPs) in this effort, the evaluation of ModelE3 against new state-of-the-art satellite products, and differences in ModelE3 ensemble emergent properties will be discussed.