A high-resolution simulation of CO2 at 1×1 km horizontal resolution using the Weather Research and Forecasting Greenhouse gas (WRF-GHG) model was conducted, focusing on the Kanto region in Japan. The WRF-GHG simulations were performed using different anthropogenic emission inventories: EAGrid (Japan, 1 km), EDGAR (0.1o), and EDGAR-downscaled (0.01o). Our analysis showed that the simulations using EAGrid better captured the diurnal variability in observed CO2 compared to EDGAR and EDGAR-downscaled emissions at two continuous monitoring sites. The 1×1 km simulation performed better in simulating CO2 variability observed in surface sites (hourly) and aircraft observations, compared to the 27×27 km simulations. We compared the vertical profile distribution of CO2 and found that all the simulations performed similarly. During February (May), the anthropogenic (land biosphere) fluxes were the primary contributor to the vertical distribution of CO2 up to an altitude of 3200 m (4500 m), beyond which long-range transport influenced by lateral boundary conditions from Eurasia played a greater role. The sensitivity analysis of boundary conditions showed a systematic bias (~ 4 ppm) persisting above 3200 m altitude when fixed (a constant value) boundary conditions are applied, as compared to the simulation with boundary conditions from a global model. We also compared the WRF-GHG simulated column-averaged XCO2 from Orbiting Carbon Observatory-2 (OCO-2) satellite and found a statistically significant spatial correlation (r=0.47) in February. However, we found a weaker spatial correlation (0.17) in May, which could be caused due to under-representation of intense land biosphere activity in WRF-GHG.