1. Stable Machine-Learning Parameterization of Subgrid Processes with Real Geography and Full-physics Emulation
- Author
-
Hu, Zeyuan, Subramaniam, Akshay, Kuang, Zhiming, Lin, Jerry, Yu, Sungduk, Hannah, Walter M., Brenowitz, Noah D., Romero, Josh, and Pritchard, Michael S.
- Subjects
Physics - Atmospheric and Oceanic Physics - Abstract
Modern climate projections often suffer from inadequate spatial and temporal resolution due to computational limitations, resulting in inaccurate representations of sub-resolution processes. A promising technique to address this is the Multiscale Modeling Framework (MMF), which embeds a small-domain, kilometer-resolution cloud-resolving model within each atmospheric column of a host climate model to replace traditional convection and cloud parameterizations. Machine learning (ML) offers a unique opportunity to make MMF more accessible by emulating the embedded cloud-resolving model and thereby reducing its substantial computational cost. Although many studies have demonstrated proof-of-concept success of emulating the MMF model with stable hybrid simulations, it remains a challenge to achieve operational-level success with real geography and comprehensive variable emulation, such as explicit cloud condensate coupling. In this study, we present a stable hybrid model capable of integrating for at least 5 years with near operational-level complexity, including real geography, seasonality, explicit predictions of cloud condensate and wind tendencies, and land coupling. Our model demonstrates skillful online performance such as 5-year zonal mean biases when comparing to previous MMF emulation studies. Key factors contributing to this online performance include using an expressive U-Net architecture, leveraging input features that includes large-scale forcings and convection memory, and incorporating microphysical constraints. With microphysical constraints mitigating unrealistic cloud formation, our work is the first work that demonstrates realistic multi-year cloud condensate climatology under the multi-scale modeling framework. Our work showcases the potential of ML parameterization for operational-level climate simulations., Comment: 28 pages, 6 figures in the main text, 5 figures in appendix. This version is a minor editorial update from the initial version
- Published
- 2024