Tao Zhang, Cyril Morcrette, Meng Zhang, Wuyin Lin, Shaocheng Xie, Ye Liu, Kwinten Van Weverberg
Topic
‘Atmosphere', 'AI/ML'
Project
E3SM
Abstract
Parameterizations in Earth System Models (ESMs) are subject to uncertainties arising from subjective empirical assumptions and incomplete understanding of the underlying physical processes. Recently, the growing representational capability of machine learning (ML) in solving complex problems has spawned numerous interested in climate science applications. Specifically, the ML-based parameterizations have been developed to represent convection, radiation and microphysics processes in ESMs by learning from observations or high-resolution simulations, which can alleviate such uncertainties. Previous works have developed some surrogate models of these processes using ML, however, those surrogate models haven’t been coupled with the dynamics core of ESMs, making it difficult to investigate their performance in a coupled system. In this study, we develop a Fortran-C-Python deep learning bridge framework embedded in the E3SM to support online AI-based parameterizations. While there are other online frameworks available, they either rely on file-based variable passing between Fotran and Python, or just binding specific ML frameworks, such as Keras, to provide corresponding interfaces of layers in neural networks. Our framework is designed to be both efficient, allowing variable passing by memory, and flexible, supporting any Python package, especially Keras/Tensorflow, Scikit-learn and Pytorch. This framework has been successfully applied in E3SM, to incorporate the ML trigger function and closure in deep convection scheme, as well as wildfire models.