Based on over 160,000 quality-controlled measurements of surface ocean carbon dioxide fugacity from 2000 to 2020, and employing machine learning methods, a satellite-based assessment model for sea-air carbon dioxide fugacity (fCO2) has been developed, aiming to reveal global changes in sea-air carbon dioxide fugacity over the past 20 years. Examining factors affecting fCO2, this study encompasses satellite data coordinates, basic seawater parameters such as salinity ,temperature, wind speed, seawater acidity and alkalinity, seawater velocity, surface geostrophic sea water velocity, surface partial pressure of carbon dioxide in sea water, surface downward mass flux of carbon dioxide expressed as carbon, as well as concentrations of dissolved inorganic carbon, phosphate, nitrate concentration, thickness of the marine mixed layer, seawater total alkalinity, silicate influencing seawater solubility, chlorophyll concentration indicating biological activity, and dissolved oxygen concentration. A comparative analysis was conducted on various machine learning methods, including XGBoost, Random Forest, Light Gradient Booster, Feedforward Neural Network, Convolutional Neural Network, and Backpropagation Neural Network. XGBoost machine learning algorithm was chosen for model construction based on the best performance. The results of independent field validation indicate that the model has a low root mean square error (RMES=18.08μatm) and mean absolute percentage error (MAPE=1.1%) and R-squared (R2=0.91). Finally, the global distribution of sea-air carbon dioxide fugacity at a resolution of 0.25°×0.25° from 2000 to 2020 has been reconstructed. The carbon dioxide fugacity in the global oceans has shown a slow upward trend, over the past 20 years, the carbon dioxide fugacity in global oceans has increased by 6.7%.