An online air-sea exchange model framework for trace gases powered by machine- learning

Abstract

The ocean emits a wide range of trace gases, such as volatile organic compounds, or sulfur-,nitrogen-, and halogen-containing compounds. Many of these gases play critical roles in the atmosphere, including aerosol and cloud formation, tropospheric and stratospheric ozone budget, as well as the self-cleaning capacity of the atmosphere. Most chemistry-climate models use prescribed oceanic emissions (often derived from observations). These prescribed (offline) emissions generally do not respond to changes in local conditions. A process-level representation of the bi-directional oceanic emissions of trace gases remains challenging, mainly because the ocean biogeochemical<br />processes controlling the natural synthesis of these compounds in the seawater remain poorly understood. We present a new online air-sea exchange framework for the NCAR CESM2, with an observationally trained machine-learning emulator to couple the ocean biogeochemistry with the air-sea exchange. This machine-learning based approach so far has been tested for a number of important trace gases, including dimethyl sulfide (DMS), acetone, bromoform (CHBr 3 ), and dibromomethane (CH 2 Br 2 ), and the preliminary results are evaluated with observations around the globe. This new model framework is more skillful than the widely used top-down approaches for representing the seasonal/spatial variations and the annual means of atmospheric concentrations. The new approach improves the model predictability for the coupled earth system model, and can be used as a basis for investigating the future ocean emissions and feedbacks under climate change.