Elevated atmospheric CO₂ concentration enhances global photosynthesis, a phenomenon known as the CO₂ fertilization effect, which substantially mitigates climate change. Despite its importance, significant discrepancies exist between satellite-based estimations and Terrestrial Biosphere Models (TBMs) simulations in quantifying CO₂ fertilization and photosynthesis. These uncertainties hinder a robust assessment of terrestrial carbon dynamics and future predictions.

Here, using machine learning, global eddy covariance measurements, and satellite remote sensing data, we evaluate CO₂ fertilization and develop predictive models to quantify Gross Primary Productivity (GPP), or ecosystem photosynthesis, from 1982 to 2020. Our results reveal a widespread positive effect of elevated CO₂ on photosynthesis, while standard satellite-based data products underestimate these impacts due to a neglect of the direct CO₂ effects on photosynthetic light use efficiency. By incorporating these direct CO₂ effects, our estimates demonstrate improved consistency with TBMs from the TRENDY ensemble regarding GPP long-term dynamics. Our work reconciles the discrepancy between satellite and TBM estimations of photosynthesis, providing a robust benchmark for Earth system models and important constraints to the global carbon cycle.