The Northeast Pacific Ocean (NEP) is one of the important carbon sinks in the global ocean. The causes of carbon flux changes in this region have been widely studied, but the physical processes associated with large scale climate variability remain controversial primarily due to scarcity of spatially and temporally continuous observations. In this study, we constructed a high-resolution sea surface partial pressure of CO2 (pCO2) from satellite observations for the NEP from 2003 to 2020 using the machine learning based XGBoost model. By analyzing the interannual large-scale high-latitude atmospheric dynamics and ocean physical conditions over the NEP, we find that the CO2 flux density (FCO2) anomalies have a strong correlation with the Alaskan atmospheric blocking events. In the region north of 48°N, anomalous cyclones triggered by atmospheric blocking increased sea surface height (SSH), which reduced the replenishment of dissolved inorganic carbon (DIC) from deep seawater, leading to enhanced carbon uptake. By contrast, in the region south of 48°N, the increase in sea surface temperature (SST) triggered by atmospheric blocking reduced the solubility of CO2 in seawater, resulting in a decrease in regional carbon flux. These results provide new perspectives for better understanding and predicting the effects of high-latitude atmospheric dynamics on regional ocean carbon fluxes.