The oceans mitigate climate change by absorbing approximately 25% of anthropogenic carbon emissions. Decadal variability in the ocean carbon sink, such as a weakening in the 1990s and a strengthening in the 2000s, has been suggested by pCO2-based reconstructions, but its causes remain poorly understood. This variability is also not well represented in climate models, raising concerns about our ability to accurately project future changes. To address potential biases from sparse observational data, machine learning methods have been applied to surface pCO2 and interior dissolved inorganic carbon (DIC), but global reconstructions of full-depth DIC remain lacking. We aim to determine whether ocean carbon sink variability is real and to understand the role of interior DIC inventory changes in the carbon budget. Using neural networks trained on GLODAPv2.2023 observations and predictors like atmospheric CO2, location, temperature, and salinity from EN4 analysis, we reconstruct full-depth global DIC distributions from the 1990s to the 2010s using a residual neural network (ResNet). Validation through prediction of independent datasets show an improvement over previous products. Validation with the ECCO-Darwin dataset results in an average RMSE of 15.1 µmol/kg and bias of -0.3 µmol/kg. The global average uncertainty is 16.85 µmol/kg. The global change in the DIC inventory exhibits pronounced peaks in decadal variability, especially in the early 2000s driven primarily by intermediate waters at depths of 300-1200 m, particularly in the Atlantic, Indian, and Southern Oceans, and to a lesser extent in the Pacific. The accumulation rate of DIC increases steadily from the mid-2000s.

The ocean is an important sink for anthropogenic carbon, its size providing a vital constraint on the global carbon budget. Climate projections also depend on accurate modeling of the ocean carbon sink. However, estimates of the sink from data products based on surface ocean observations differ significantly from the results of global biogeochemical models (GOBMs), with the former suggesting more decadal variability over the observational period and a steeper uptake trend since around the year 2002. We present an alternative method for diagnosing the ocean carbon sink and characterizing transport and mixing of carbon in the ocean interior, using observational data. Our method combines a machine learning reconstruction of ocean interior carbon with an optimal transformation method (OTM) that uses a water mass framework to simultaneously close budgets of heat, freshwater and carbon. Focusing on two decades, we find a global sink of 2.03±0.22PgCyr-1 for 1993-2002 and 2.86±0.25PgCyr-1 for 2003-2012. The trend of 0.83±0.5PgCyr-1 is 75% larger compared to the data products (although within uncertainties), and nearly quadruples that suggested by the GOBMs, and is driven mainly by the North Pacific (>10N) and Southern Ocean (>35S) basins. The transport and mixing diagnosed by the OTM suggests an intensification of carbon uptake during Southern Ocean mode and intermediate water formation and a reduction in uptake during North Atlantic Deep Water formation have contributed to a southwards redistribution of carbon in the Atlantic. Over the same period, ocean transport and mixing acted to redistribute carbon from south to north in the Indo-Pacific.