Abstract: Elevated atmospheric CO₂ (eCO₂) is a primary driver of the terrestrial carbon cycle, yet substantial uncertainties remain regarding the magnitude of vegetation productivity responses and their persistence over time across different ecosystems. Here, we conducted a global meta-analysis based on 373 pairwise observations from 70 experimental studies to quantify the responses of aboveground net primary productivity (ANPP) and belowground net primary productivity (BNPP) to eCO₂, and evaluated the predictive capability of Dynamic Global Vegetation Models (TRENDY v12). Our results show that eCO₂ significantly stimulated global ANPP and BNPP by 16.73% and 20.06%, respectively. This CO₂ fertilization effect exhibited a cumulative strengthening trend with experimental duration. We observed a distinct ecosystem-dependent pattern, where forests exhibited significantly stronger responses (ANPP: +31.72%; BNPP: +33.09%) compared to grasslands (ANPP: +9.94%; BNPP: +14.59%). Boosted Regression Tree analysis identified mean annual precipitation as the most dominant environmental driver regulating these responses, outweighing the influence of soil nutrient availability. However, comparisons with model simulations revealed that current Land Models fail to reproduce the observed divergent responses between forests and grasslands, despite capturing the general global positive trend. Specifically, the models overestimated the eCO₂ fertilization effect in grasslands while underestimating it in forests. In summary, these findings highlight the critical role of water availability and ecosystem-specific physiological traits in modulating the eCO₂ fertilization effect, indicating that Earth System Models will need to incorporate these differential mechanisms to improve the accuracy of future terrestrial carbon sink projections.

Key words: Vegetation productivity, Climate change, Elevated CO₂, Meta-analysis, Ecosystem types, Land Models