Reconstruction of Pco2 Data in the Southern Ocean Based on Feedforward Neural Network

The Southern Ocean accounts for 20% of the world’s ocean and nearly 40% of the global ocean’s total carbon sink, effectively reducing the impacts of anthropogenic carbon dioxide emissions. Due to the scarcity of observation data, the changing trends of carbon sinks in the Southern Ocean and its fluctuation reasons are still uncertain. In this chapter, we determine covariates through data association analysis, and build a feedforward neural network based on these parameters to improve the accuracy of carbon flux estimations in the Southern Ocean. Based on SOCAT observation data, we reconstructed the pCO2 gridded data of the Southern Ocean during 1998-2018. The root-mean-square error obtained by fitting the observation data was 8.86 $$\mu $$ atm, indicating that the results were better than other feedforward neural network model in the Surface Ocean pCO2 Mapping Intercomparison. The research results also showed that since 2000, the capacity of the Southern Ocean as a carbon sink has been gradually enhanced. During 2010-2013, it decreased, but then increased significantly. Due to the increase of upwelling, the seasonal feature of the Southern Ocean is the lowest carbon absorption in winter; then, the maximum absorption increased rapidly in summer, which is mainly driven by biology. As other studies have pointed out, there is an obvious double ring structure in the Southern Ocean. This study confirms that the inner ring ( $$50-70\,^{\circ }$$ S) is a carbon source area gradually transforming into a carbon sink, while the outer ring ( $$35-50\,^{\circ }$$ S) continues to serve as a carbon sink.