Feng Tao, Benjamin Z. Houlton, Yuanyuan Huang, Ying-Ping Wang, Stefano Manzoni, Bernhard Ahrens, Umakant Mishra, Lifen Jiang, Xiaomeng Huang, Yiqi Luo

The current generation of biogeochemical models produce large uncertainty in carbon-climate feedback projections. Structural differences in these models have been identified as a major source of inter-model uncertainties when simulating soil organic carbon (SOC) dynamics worldwide. However, parameterization could also play a role, particularly when common observational data are used to constrain model simulations. Here we demonstrate the critical role of observational data in reducing model-based uncertainty in global estimates of SOC. We applied the PROcess-guided deep learning and DAta-driven modeling (PRODA) approach to constrain both a microbial implicit model based on first-order kinetics (i.e., Community Land Model version 5, CLM5) and a microbial explicit model based on Michaelis-Menten kinetics (i.e., CarbOn cycle and Microbial PArtitioning Soil model, COMPAS) with >50,000 globally distributed SOC vertical profiles. Overall, the two constrained models predicted similar carbon transfer efficiency, baseline decomposition rate, and environmental effects on carbon fluxes. These converged model components contributed to similar SOC patterns simulated by the two structurally different biogeochemical models. Carbon input allocation and vertical transport were less constrained by SOC profile data and require other data sets to constrain. Moreover, after being constrained by SOC observations, the Michaelis constant in COMPAS tends to be much larger than its corresponding substrate concentration in SOC decomposition. Thus, the Michaelis-Menten kinetics in the COMPAS model can be approximated by multiplicative kinetics (i.e., first order with respect to both donor and received pool carbon) in these global scale simulations. Our results highlight the importance of observational data in informing model development and constraining model predictions.