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

Abstract

Abstract. Various machine learning methods were attempted in the global mapping of surface ocean partial pressure of CO2 (pCO2) to reduce the uncertainty of the global ocean CO2 sink estimate due to undersampling of pCO2. In previous research, the predictors of pCO2 were usually selected empirically based on theoretic drivers of surface ocean pCO2, and the same combination of predictors was applied in all areas except where there was a 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 select predictors of pCO2 based on the mean absolute error in each of the 11 biogeochemical provinces defined by the self-organizing map (SOM) method. Based on the predictors selected, a monthly global 1∘ × 1∘ surface ocean pCO2 product from January 1992 to August 2019 was constructed. Validation of different combinations of predictors based on the Surface Ocean CO2 Atlas (SOCAT) dataset version 2020 and independent observations from time series stations was carried out. The prediction of pCO2 based on region-specific predictors selected by the stepwise FFNN algorithm was more precise than that based on predictors from previous research. Applying the FFNN size-improving algorithm in each province decreased the mean absolute error (MAE) of the 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, https://doi.org/10.12157/iocas.2021.0022, Zhong, 2021.

Global ocean CO2 uptake dominated by the ENSO in the last thirty years

“…In this paper the surface ocean pCO 2 was constructed using the Stepwise FFNN method in each province divided by the SOM (Zhong et al, 2022). The Surface Ocean CO 2 Atlas version 2021 (SOCATv2021) dataset was used for neural network training.…”

“…The Surface Ocean CO 2 Atlas version 2021 (SOCATv2021) dataset was used for neural network training. Compared with previous work (Zhong et al, 2021;Zhong et al, 2022), we used the average outputs of serval networks as nal pCO 2 prediction value to remove the in uence of random assigned initial state of FFNN and the training data sort order. The updated pCO 2 predictors in each province were listed in the table S1 in the supplements.…”

“…During the strong El Nino events of 1997-1998 and 2015-2016, the equatorial Paci c source weakened near 0.3 Pg C yr -1 in a few years and restrengthened in rapid sequence when the El Nino events nished (Figure 6). The predictor salinity, temperature, xCO 2 of dry air, chlorophyll concentration, and mixed layer depth contributed most in decreasing the predicting error of the stepwise FFNN product in the equatorial Paci c (Zhong et al, 2022). As the most import carbon source area, the equatorial Paci c Ocean transport deep waters toward surface and transport warm surface waters toward pole regions through the ocean overturing circulation (Resplandy et al, 2018).…”

“…Similarly, a few research suggested a weakening global ocean carbon sink since 2018 (Watson et al, 2020), while a more rapidly strengthening was found in other research (Chau et al, 2021;Iida et al, 2021). To better constrain the interannual-to-decadal variability of the global ocean carbon sink, the pCO 2 product constructed from a novel method based on the stepwise FFNN algorithm was used for the estimate of CO 2 ux (Zhong et al, 2022). Then the variation of the global ocean carbon sink in the last three decades were discussed based on the data.…”

The increasing big gap of carbon sink between the western and eastern Pacific in the last three decades

“…While in the temperate Pacific Ocean, the west-east difference of the carbon sink also existed for a long time (Takahashi et al, 2009), with still unclear influences and dynamic changes. Thus, based on the drivers selected by the Stepwise FFNN algorithm (Zhong et al, 2022), we reestimated the Pacific Ocean carbon sink in the last three decades and then explored the mechanism and long-term influence of the carbon sink differences between the western Pacific and the east.…”