Understanding the variability and sources of atmospheric CO2 is essential for improving greenhouse gas monitoring and model performance. This study investigates temporal CO2 variability at the Xianghe site in China, which hosts both remote sensed (TCCON-affiliated) and in situ (PICARRO) observations. Using the Weather Research and Forecast model coupled with Chemistry, in its greenhouse gas option (WRF-GHG), we performed a one-year simulation of surface and column-averaged CO2 mole fractions, evaluated model performance and conducted sensitivity experiments to assess the influence of key model configuration choices. The model captured the temporal variability of column-averaged mole fraction of CO2 (XCO2) reasonably well (r=0.7), although a persistent bias in background values was found. A July 2019 heatwave case study further demonstrated the model’s ability to reproduce a synoptically driven anomaly. Near the surface, performance was good during afternoon hours (r=0.75, MBE =2.44 ppm), nighttime mole fractions were overestimated (MBE = 7.86 ppm), resulting in an exaggerated diurnal amplitude. Sensitivity tests revealed that land cover data, vertical emission profiles, and adjusted VPRM-parameters (Vegetation Photosynthesis and Respiration Model) can significantly influence modeled mole fractions, particularly at night. Tracer analysis identified industry and energy as dominant sources, while biospheric fluxes introduced seasonal variability – acting as a moderate sink in summer for XCO2 and a net source in most months near the surface. These findings demonstrate the utility of WRF-GHG for interpreting temporal patterns and sectoral contributions to CO2 variability at Xianghe, while emphasizing the importance of careful model configuration to ensure reliable simulations.