Unified FLUXes (UFLUX) emulator

Process-based biosphere models, which formulate biophysical ecosystem processes, provide a mechanistic framework for understanding terrestrial carbon dynamics. In contrast, data-driven approaches—e.g., upscaling eddy covariance (EC) fluxes using satellite observations and machine learning—offer empirical estimates. These complementary methods often diverge significantly, particularly in estimating global photosynthetic uptake and ecosystem respiration, with discrepancies exceeding 50 PgC and highlighting persistent uncertainties.To bridge this gap, we adopt a hybrid strategy that embeds physiological understanding via semi-empirical models, refines it with EC fluxes constrained by machine learning, and integrates process-based allocation to resolve component fluxes. This process-informed hybrid approach links ecological knowledge with predictive models, enabling generalisation beyond flux tower sites and supporting the development of new insights.We assess global carbon dynamics over the past two decades, applying Bayesian inference to evaluate climate impacts on land carbon processes. Using the Unified FLUXes (UFLUX) emulating the carbon allocations of process-based models, our study delivers the first observational and process-informed hybrid assessment of global carbon flux and stock changes. Notably, while gross carbon uptake is consistent across methods (~130 PgC in 2022, increasing by 0.4 annually), respiration estimates diverge—from 6 PgC in process-based models to 26 in conventional upscaling, and 16 in our hybrid model—revealing structural limitations in respiration parameterisation.Improved representation of respiratory processes is essential to capture the competing roles of photosynthesis and respiration under climate change. Despite rising global carbon fluxes and biomass stocks, tipping point risks remain: in tropical regions, increased photosynthesis (0.1 PgC) is offset by rising respiration (0.05 PgC).