As one of the major agricultural production areas in the world, the United States (U.S.) Midwest plays a vital role in the global food supply and agricultural ecosystem services. Although significant efforts have been made in modeling the carbon cycle dynamics over this area, large uncertainty still exists in the previous simulations in terms of reproducing individual components of the carbon cycle and their responses to environmental variability. Here we evaluated the performance of an advanced agroecosystem model, ecosys, in simulating carbon budgets over the U.S. Midwest, considering both the magnitude of carbon flux/yield and its response to environmental (climate and soil) variability. We conducted model simulations and evaluations at 7 cropland eddy-covariance sites as well as over 293 counties of Illinois, Indiana, and Iowa in the U.S. Midwest. The site-level simulations showed that ecosys captured both the magnitude and seasonal patterns of carbon fluxes (i.e., net ecosystem carbon exchange (NEE), ecosystem gross primary production (GPP), and ecosystem respiration (Reco)), leaf area index (LAI), and dynamic plant carbon allocation processes, with R2 equal to 0.92, 0.87, 0.87, and 0.78 for GPP, NEE, Reco, and LAI, respectively across all the sites compared with the observations. For regional scale simulations, ecosys reproduced the spatial distribution and interannual variability of corn and soybean yields with the constraints of observed yields and a new remotely sensed GPP product, with R2 of multi-year averaged simulated and observed yield equal 0.83 and 0.80 for corn and soybean, respectively. The simulated responses of carbon cycle dynamics to environmental variability were consistent with that from the empirical observations at both site and regional scales. Our results demonstrated the applicability of ecosys in simulating the carbon cycle and soil carbon dynamics of the U.S. Midwestern agroecosystems under different climate and soil conditions.

Accurate and cost-effective quantification of the carbon cycle for agroecosystems at decision-relevant scales is critical to mitigating climate change and ensuring sustainable food production. However, conventional process-based or data-driven modeling approaches alone have large prediction uncertainties due to the complex biogeochemical processes to model and the lack of observations to constrain many key state and flux variables. Here we propose a Knowledge-Guided Machine Learning (KGML) framework that addresses the above challenges by integrating knowledge embedded in a process-based model, high-resolution remote sensing observations, and machine learning (ML) techniques. Using the U.S. Corn Belt as a testbed, we demonstrate that KGML can outperform conventional process-based and black-box ML models in quantifying carbon cycle dynamics. Our high-resolution approach quantitatively reveals 86% more spatial detail of soil organic carbon changes than conventional coarse-resolution approaches. Moreover, we outline a protocol for improving KGML via various paths, which can be generalized to develop hybrid models to better predict complex earth system dynamics.