Remote Sensing based Sea Surface partial pressure of CO2 (pCO2) in China Seas (2003-2019)

This dataset is sea surface partial pressure of CO2 (pCO2) in China seas (0-42°N, 105-132°E) over 2003-2019 with a spatial resolution of 1km and temporal resolution of a month. This is our second version of pCO2 in China seas. The first version was published on the SatCO2 website (http://www.satco2.com/index.php?m=content&amp;c=index&amp;a=show&amp;catid=317&amp;id=188). We produce this dataset by creating a boost machine learning algorithm (XGBoost) based on the gradient boost decision tree (GBDT). The input parameters used in this are sea surface temperature (SST), chlorophyll-a concentration (Chl-a), remote sensing reflectance of three bands (Rrs412, 443, 488 nm), the temperature difference in longitude direction (SST_DIF), and the theoretical background <em>p</em>CO₂ (<em>p</em>CO₂_therm) under corresponding SST. SST_DIF is derived from SST by subtracting the mean value at the same latitude. To calculate <em>p</em>CO₂_therm, we first assume that the seawater background <em>p</em>CO₂ equals the annual average atmospheric <em>p</em>CO₂ (sea-air <em>p</em>CO₂ balanced under ideal conditions), and the equilibrium temperature is assumed to be the yearly average SST; Then, the background <em>p</em>CO₂ is corrected according to SST to obtain <em>p</em>CO₂_therm. Air pressure at sea level (SLP) and mole fraction of CO₂ in the air (xCO₂) are employed when calculating atmospheric <em>p</em>CO₂. The underway <em>p</em>CO₂ is first gridded to monthly 1 km and then divided into the training and validation sets with the volumes 151009 and 32584 samples. The validation set shows coefficients of determination (R²) between the XGBoost model-predicted and in-situ <em>p</em>CO₂ were 0.86, and the root means squared errors (RMSE) for the <em>p</em>CO₂ were 21.1 μatm.