Meeting Documents

Reconstructing the spatiotemporal evolution of the global interior ocean's anthropogenic carbon sink using deep learning

Abstract

The oceans play a mitigating role in climate change by absorbing approximately 25% of the anthropogenic carbon that is released. Previous pCO2 based reconstructions of air-sea CO2 flux have suggested that this carbon sink shows decadal variability, apparently weakening in the 1990s and strengthening in the 2000s. The reason for this is currently not well understood. Moreover, this variability is currently not well represented in climate models, and in the future climate predictions they produce.

At this point it is unclear if the estimated variability is a product of bias due to the sparsity of biogeochemical observations, especially in earlier decades or during wintertime in polar regions. To overcome the challenge of sparse observations, machine learning methods, particularly neural networks, have been applied, both to surface pCO2 and interior dissolved inorganic carbon (DIC). However, as yet reconstructions of the latter have not been produced to full depth over the global ocean.

Our aim is to assess whether ocean carbon sink variability is real and understand the interior inventory changes as a component of the carbon budget. To achieve this, we train a neural network to predict global spatiotemporal distributions of DIC and C* from the 1990s to the 2010s. C* is a quasi-conservative tracer using Redfield stoichiometric ratios to correct DIC for biological activity. ΔC*, for the change in the carbon budget between two times, has been used as a proxy for a change in added anthropogenic carbon.

We train a neural network with a multi-phase and multi-modal approach using data from GLODAPv2.2022, ARGO floats, the ECCO-Darwin biogeochemical ocean model, and satellite altimetry. We use time, location, temperature, and salinity from the UK Met Office monthly EN4 reanalysis product to make predictions. Here, we present preliminary results for the spatiotemporal evolution of full-depth interior carbon in the global ocean, quantifying the anthropogenic carbon sink and its variability over time. Furthermore, the results are being combined with a water mass based inverse method to investigate the drivers of variability.