Remote Sensing based Sea Surface partial pressure of CO2 (pCO2) and air-sea CO2 flux (FCO2) in the East China Sea (2003-2019)

Based on in situ seawater pCO2 data collected on 51 cruises/legs over the past two decades, a satellite retrieval algorithm for seawater pCO2 was developed by combining the semi-mechanistic algorithm and machine learning method (MeSAA-ML). MeSAA-ML introduces semi-analytical parameters, including the temperature-dependent seawater pCO2 (pCO2,therm ) and upwelling index (UISST), to characterise the combined effect of atmospheric CO2 forcing, thermodynamic effects, and multiple mixing processes on seawater pCO2. Additionally, considering the biological effects and various sub-regional features, multiple ocean colour parameters were also used as inputs in XGBoost, the best-selected machine learning algorithm. Independent cruise-based data were used to validate the satellite-derived pCO2, which achieved excellent performance in this complicated marginal sea, with low root mean square error (RMSE=19.6 μatm) and mean absolute percentage deviation (APD=4.12%). Air-sea CO2 fluxes are calculated based on retrieved seawater pCO2.

本数据集基于近二十年来51个航次/航段获取的原位海水二氧化碳分压（pCO2）实测数据，结合半机制算法与机器学习方法，开发了一套海水pCO2卫星反演算法（MeSAA-ML）。MeSAA-ML引入了温度依赖性海水二氧化碳分压（pCO2,therm）与上升流指数（UISST）等半解析参数，以表征大气CO2强迫、热力学效应及多种混合过程对海水pCO2的联合调控作用。此外，考虑到生物效应与多样的区域子特征，本研究选取经筛选得到的最优机器学习算法XGBoost，将多种海洋水色参数作为其输入特征。利用独立航次实测数据对卫星反演得到的海水pCO2进行验证，在这片复杂的边缘海区域中，该算法取得了优异的反演性能，其均方根误差（root mean square error, RMSE=19.6 μatm）与平均绝对百分比偏差（mean absolute percentage deviation, APD=4.12%）均处于极低水平。基于反演得到的海水pCO2，可计算海-气CO2通量。