Reconstruction of global surface ocean <i>p</i>CO<sub>2</sub> using region-specific predictors based on a stepwise FFNN regression algorithm

Abstract. Various machine learning methods were attempted in the global mapping of surface ocean partial pressure CO2 (pCO2) to reduce uncertainty sink estimate due undersampling pCO2. In previous research, predictors pCO2 usually selected empirically based on theoretic drivers pCO2, and same combination was applied all areas except where there a lack coverage. However, differences between different regions not considered. this work, we combined stepwise regression algorithm feed-forward neural network (FFNN) select mean absolute error each 11 biogeochemical provinces defined by self-organizing map (SOM) method. Based selected, monthly 1∘ × product from January 1992 August 2019 constructed. Validation combinations Surface Ocean Atlas (SOCAT) dataset version 2020 independent observations time series stations carried out. The prediction region-specific FFNN more precise than that research. Applying size-improving province decreased (MAE) 11.32 µatm root square (RMSE) 17.99 µatm. script file are distributed through Institute Oceanology Chinese Academy Sciences Marine Science Data Center (IOCAS, https://doi.org/10.12157/iocas.2021.0022, Zhong, 2021.