Filling the gaps in Southern Ocean carbon and heat: Machine learning-based products from sparse observations

The global ocean has taken up an estimated 89% of excess heat and 26% of annual CO₂ emissions resulting from anthropogenic activity in recent decades. Despite the crucial role of the ocean in the climate system, there remains significant uncertainty around ocean heat and carbon uptake. Processes of ocean carbon and heat uptake and redistribution are among the leading processes limiting scientists’ ability to predict the rate and magnitude of global climate change. Furthermore, despite being responsible for as much as 40–50% of global annual oceanic CO₂ uptake, the Southern Ocean remains the most controversial ocean basin, with large differences in both the magnitude and variability of CO₂ fluxes across various estimation methods. The disagreement regarding the magnitude and variability of Southern Ocean CO₂ fluxes largely arises due to the sparsity and uneven distribution of in-situ observations of carbon in the ocean, a problem which also constrains estimates of ocean heat uptake. However, machine learning has been extensively demonstrated in recent years to be an effective tool for overcoming such data limitations through gap-filling methods.

We leverage existing expertise in machine learning methods from generating the SOM-FFN and MOBO-DIC ocean carbon data products and apply this expertise to the development of a novel machine learning generated ocean heat uptake data product for the Southern Ocean. We integrate new Earth Observation data from Copernicus satellites with in-situ observation data from the Southern Ocean as input for our new data products. Here, we present a beta-version of our new machine-learning based products of carbon and heat uptake in the Southern Ocean and a first analysis of the variability and trends of the carbon and heat uptake of these data products. This work presents our first steps towards data products which will allow us to identify transport pathways of carbon and heat within the ocean interior. The novel machine learning-based products will also be used to support the advancement of Earth System Models and climate change predictions.