AbstractMost of China's carbon sink inversion research uses global atmospheric transport models to assimilate natural fluxes, which quantifies the biosphere and ocean carbon budget with a relative coarse spatial resolution and long timescale from a weekly or monthly perspective. Toward high‐resolution inversion of CO2 fluxes, a novel carbon flux forecast model was developed in this study, which was then further used to carry out carbon assimilation based on a regional chemical transport model (CMAQ) at higher spatial (64 × 64 km2) and temporal (1 hr) resolutions. An Ensemble Kalman Smoother was applied as the assimilation algorithm and further extended to assimilate surface CO2 observations. Concentrations and fluxes were simultaneously assimilated as state variables to help reduce the uncertainty in the initial CO2 fields with the joint data assimilation framework (JDAS). In general, the posterior fluxes reproduced the seasonal, daily and hourly variation effectively, demonstrating the ability to fully absorb and utilize observations. Moreover, the influence of the choice of assimilation window on the carbon flux inversion was assessed via sensitivity experiments, revealing 36 hr to be the optimum length. Evaluation of the prior and posterior flux simulations also indicated that JDAS offers reasonable improvements, making it suitable for fine‐scale flux optimization and estimation. In addition, the posterior biosphere estimation in mainland China (−682 TgC yr−1) tends to be the optimal mathematical solution under current sparse observation coverage with daytime photosynthetic uptake, which likely leads to the overestimation of the optimized CO2 sink. This study serves as a basis for future regional and urban assessment.