Sea surface partial pressure of carbon dioxide (pCO₂) is an important parameter in the quantification of air-sea CO₂ flux, which further plays an important role in quantifying the global carbon budget. Marginal seas are a dynamic and significant part of the global carbon cycle. However, due to the large temporal and spatial variations of pCO₂ and the diversity of control factors in here, there are still large errors in remote sensing estimation of pCO₂. In this study, the best estimation model of pCO₂ based on different algorithms was established in the Yellow Sea (SYS) from 2011 to 2019, using in-situ data combined with the satellite sea surface temperature (SST), chlorophyll-a concentration (Chl-a), diffuse attenuation coefficient at 490 nm (Kd_{_490}) and salinity (Sal). Spatial and temporal variation of pCO₂ were further derived based on the constructed model for the years of 2003–2021. The results showed Random Forest (RF) based pCO₂ algorithm had best performance, with root mean square difference (RMSD) of 43.35 μatm, and coefficient of determination (R²) of 0.67, which pCO₂ ranged between 290 and 526 μatm. Sal was the most important variable effecting the pCO₂ variation in the SYS, followed by the variable of SST, Kd_{_490}, Chl-a. On temporal scale, the pCO₂ presented an obvious seasonal with high pCO₂ in summer and autumn and low pCO₂ in spring and winter. pCO₂ showed a rising tendency at a rate of 0.34 μatm per year (R­²=0.27, p<0.05). On spatial scale, pCO₂ varied from high value inshore waters to low value in offshore waters, which was consistent with the pCO₂ distribution derived with in-situ data.