Reconstruction of global surface ocean pCO2 using region-specific predicators based on a stepwise FFNN regression algorithm

Various machine learning methods were attempted in the global mapping of surface ocean partial pressure of CO2 (pCO2) to reduce the uncertainty of global ocean CO2 sink estimate due to undersampling of pCO2. In previous researches the predicators of pCO2 were usually selected empirically based on theoretic drivers of surface ocean pCO2 and same combination of predictors were applied in all areas unless lack of coverage. However, the differences between the drivers of surface ocean pCO2 in different regions were not considered. In this work, we combined the stepwise regression algorithm and a Feed Forward Neural Network (FFNN) to selected predicators of pCO2 based on mean absolute error in each of the 11 biogeochemical provinces defined by Self-Organizing Map (SOM) method. Based on the predicators selected, a monthly global 1° × 1° surface ocean pCO2 product from January 1992 to August 2019 was constructed. Validation of different combination of predicators based on the SOCAT dataset version 2020 and independent observations from time series stations was carried out. The prediction of pCO2 based on region-specific predicators selected by the stepwise FFNN algorithm were more precise than that based on predicators from previous researches. Appling of a FFNN size improving algorithm in each province decreased the mean absolute error (MAE) of global estimate to 11.32 μatm and the root mean square error (RMSE) to 17.99 μatm. The script file of the stepwise FFNN algorithm and pCO2 product are distributed through the Institute of Oceanology of the Chinese Academy of Sciences Marine Science Data Center (IOCAS; http://dx.doi.org/10.12157/iocas.2021.0022, Zhong et al., 2021).