Agroecosystems play a vital role in regional and global carbon cycles. Quantifying the carbon budget of agroecosystems remains a challenge mainly due to the complex impacts of human management on the carbon cycle of agroecosystems. Model-data fusion is a promising approach to accurately quantify the carbon budget of agroecosystems given more and more observations become available from satellite remote sensing. Here we present a model-data fusion system to quantify the carbon budget of the U.S. Midwestern agroecosystem. We first evaluated the performance of an advanced agroecosystem model, ecosys, in simulating carbon budget over the U.S. Midwestern. We conducted model simulations and evaluations at 7 cropland eddy-covariance sites in the U.S. Midwestern. The site-level simulations show that ecosys model captured both the magnitude and seasonal patterns of carbon fluxes (i.e. GPP, NEE, Reco), LAI, and dynamic of plant carbon allocation processes with high accuracy. We then scaled the simulations up to the 293 counties across three I-states (i.e. Illinois, Indiana, and Iowa). We constrained the model with a novel NIRv-based remotely sensed GPP product and crop yield data from USDA National Agricultural Statistics Service in the even years during 2001 and 2018, and evaluated the model performance in the odd years during the same period. The results show that the constrained ecosys model reproduced the spatial distribution and interannual variability of corn and soybean yield in the three I-states. The responses of the carbon cycle processes to the environmental variability obtained from the constrained model simulations were consistent with the observed ones, revealing the applicability of the constrained ecosys model in simulating the impacts of future climate change on the carbon cycle of the U.S. Midwestern agroecosystems. We finally quantified the carbon budget of the U.S. Midwestern agroecosystems at county scale using the constrained ecosys model under both historic and future climate conditions.