Estimating Decadal Trends in the Southern Ocean Carbon Storage using a Deep-Learning Model

One of the major sinks for anthropogenic carbon dioxide is thought to be the ocean, in particular at the higher latitudes because the solubility of gases increases with decreasing temperature. It is estimated that the Southern Ocean, where persistent zonal (i.e., east-west) winds are strong and temperatures are relatively cold, is responsible for approximately 40% of the oceanic carbon sink of anthropogenic emissions. Understanding the changes in ocean carbon content is crucial not only for understanding the climate, but also marine chemistry. For example, an increased amount of dissolved carbon in the ocean has led to ocean acidification, which affects marine organisms. However, there are large uncertainties regarding the carbon storage potential of the ocean: studies have shown that after a period of decline during the 1990s, ocean carbon uptake may have increased in the 2000s. Because the shipboard measurements of carbon content in the ocean interior are sparse both spatially and temporally, estimating ocean carbon trends is difficult. Recent deployment of Argo floats (robotic instruments that drift around in the ocean interior) equipped with biogeochemical sensors has greatly enhanced the volume of available data; however, the timeseries of this data span a relatively short time period since 2014. We developed a deep-learning neural net model that takes as input ocean surface measurements, which are readily available from satellite observations at regular and high-resolution spatial and temporal intervals. Our model then outputs estimates of dissolved inorganic carbon concentrations from the ocean surface to 4 km depth. We applied this method to calculate trends in ocean carbon concentrations over the past three decades and found that while ocean carbon content has decreased in the 1990s and 2000s, it has been increasing at a faster rate since the 2010s. This means that the ocean’s potential to take up more atmospheric carbon has declined. However, the particular circulation dynamics that drove these changes may have differed across zonal sectors of the Southern Ocean. While the near-surface decrease in DIC concentrations would enhance atmospheric carbon dioxide uptake continuing the previously-found trends, weakened connectivity between surface and deep layers and build-up of DIC in deep waters could reduce the ocean’s carbon storage potential. Importantly, once trained, our model is fast and computationally inexpensive, which gives it an advantage over traditional ocean models. This makes our model a useful tool for monitoring future ocean carbon content from new satellite measurements as they become available, and similar models can be developed for other regions of the global ocean in order to better understand the ocean's capacity to take up additional atmospheric carbon.