Transition from positive to negative indirect CO2 effects on the vegetation carbon uptake

Terrestrial ecosystems absorb roughly 30% of anthropogenic CO2 emissions, and the terrestrial carbon sink has more than doubled over the past five decades. Elevated atmospheric CO2 (eCO2) affects vegetation carbon uptake through two mechanisms: a direct physiological stimulation of photosynthesis and water-use efficiency (eCO2(dir)), and an indirect pathway via radiative forcing that alters climate (temperature, water availability, atmospheric dryness) and related environmental conditions (eCO2(ind)). Recent evidence indicates a decline in eCO2(dir), raising the importance of understanding the sign, magnitude, and temporal evolution of eCO2(ind), which could increasingly control the future terrestrial carbon budget. However, eCO2(ind) dynamics remain poorly constrained because multiple climate drivers co-vary and interact nonlinearly with vegetation processes and with eCO2(dir). This study asks: How has the indirect effect of eCO2 on vegetation gross primary production (GPP) changed since the 1980s, what mechanisms underlie these changes, and how will they evolve under high-emission scenarios toward 2100?

Prior studies document weakening temperature–vegetation relationships in northern ecosystems, increasing negative impacts of vapor pressure deficit (VPD) on productivity in alpine grasslands, growing water constraints on vegetation, and heightened drought risks for the carbon cycle. Many assessments focused on single drivers (e.g., temperature, water availability, VPD), often neglecting covariation and interactions among climate factors, thus capturing only partial aspects of eCO2(ind). FACE experiments provide local-scale insights into CO2 fertilization but may not scale to regional-to-global levels. Recent work also suggests a global decline in direct CO2 fertilization effects, implying that indirect effects and their temporal variability will be critical for future carbon sink trajectories. Uncertainties in the representation of water limitation, disturbances, and vegetation mortality in Earth system models (ESMs) contribute to divergent estimates between observations and models.

- Models and experiments: The study uses seven CMIP6 Earth system models (ACCESS-ESM1-5, CanESM5, CNRM-ESM2-1, E3SM-1-1, MIROC-ES2L, MRI-ESM2-0, UKESM1-0-LL) participating in C4MIP. It analyzes historical (1982–2014) and SSP5-8.5 scenario (2015–2100) simulations under: fully coupled (historical, ssp585), biogeochemically coupled (hist-bgc, ssp585-bgc), and a CO2-only forcing run for the historical period where available (hist-CO2; only CanESM5 provides all three). Model outputs include GPP, NPP, Ra, Rh, Reco, NEP, ET, Tmax/Tmin/Tmean, precipitation (P), cloud cover (CL), relative humidity (RH), soil moisture (surface 0–10 cm and total), and derived VPD and PET-based aridity index (P/PET). All outputs are regridded to 0.5°. - Growing season definition: For each grid cell and period, growing season months are those with T > 0 °C, and for arid/semi-arid areas additionally constrained by cumulative P between 10% and 90% of annual P (alternatives tested for robustness). - Indirect CO2 effect from models: eCO2(ind) is derived via factorial simulations as the difference between trends in growing-season GPP in fully coupled and biogeochemically coupled experiments, normalized by the rate of CO2 increase: eCO2(ind) = (δGPP_FULL − δGPP_BGC)/δCO2. Analogous calculations are applied to NPP, Ra, Rh, Reco, and NEP to assess broader carbon flux responses. Trend significance is tested with the Mann–Kendall test. - Direct CO2 effect from models: eCO2(dir) is estimated using a multiple non-linear regression on annual anomalies from fully coupled runs: GPP = β(CO2) + C1(P) + C2(VPD) + C3(Tmin×VPD) + C4(P×CL) + C5 + ε. The coefficient β (g C m−2 ppm−1) represents eCO2(dir). Predictor collinearity is checked via VIF. An alternative factorial estimate for eCO2(dir) is computed for CanESM5 using hist-CO2, historical, and hist-bgc runs. - Observation-based datasets and approaches: Satellite GPP (NIRv-based) at monthly 0.05° resolution (1982–2014) is resampled to 0.5° and aggregated to the CMIP6-derived growing seasons. CRU v4.05 climate data provide T, P, CL, VP, PET; VPD is derived as SVP−VP. A climate analog framework identifies, within periods 1982–1996 and 2000–2014, pairs of years with statistically similar multivariate climate (via Mahalanobis distance in a PCA-transformed space) but different CO2, to isolate direct effects. From observations: eCO2(dir)_obs = ΔGPP_obs^CA/ΔCO2; eCO2(net)_obs = ΔGPP_obs/ΔCO2; and eCO2(ind)_obs = (ΔGPP_obs − ΔGPP_obs^CA)/ΔCO2. An alternative observational estimate of eCO2(dir) uses the same non-linear regression with CRU climate. - Aridity and mechanisms: The study examines relationships between changes in eCO2(ind) and changes in terrestrial water availability, using surface soil moisture (SM_surf), total soil moisture, and P/PET across historical and future periods. Binned analyses across mean SM_surf and ΔSM_surf quantify sensitivities and humid vs non-humid thresholds (SM_surf ≈ 0.26 m3 m−3 corresponding to P/PET ≈ 1). - Robustness checks: Sensitivity to data sources, temporal window lengths (12, 15, 16 years), growing-season definitions, and alternate GPP proxies (kNDVI) are conducted. Idealized 1% CO2 per year experiments further cross-check methods for direct/indirect decomposition.

- Historical decline and sign change of eCO2(ind): CMIP6 model ensemble mean (CMIP6_SMA) shows a significant global decline in eCO2(ind) from 0.24 ± 0.32 gC m−2 ppm−1 (1982–1996) to −0.04 ± 0.24 gC m−2 ppm−1 (2000–2014) (p < 0.05). Observation-based estimates using the climate analog approach confirm a global weakening with an overall change of −0.38 gC m−2 ppm−1. - Spatial and climatic patterns: The decline is strongest in cold and dry zones, notably boreal regions (hotspots in eastern Canada, Scandinavia, south-central Siberia). Warm and wet zones show limited opposite tendencies. Model–observation agreement is strong in spatial patterns across latitudinal and climatological gradients. - Future projections (SSP5-8.5): eCO2(ind) is projected to continue declining and remain negative globally by late century. CMIP6_SMA projects a significant global decrease by 0.36 gC m−2 ppm−1 for 2086–2100 relative to 1982–1996 (p < 0.01). Declines are significant over 46.5% of vegetated land, especially in the Northern Hemisphere; increases occur mainly along the equatorial belt over smaller extents (32.7%). - Net ecosystem production: Historical global eCO2(ind)-NEP ≈ −0.02 gC m−2 ppm−1. It shifts from 0.05 ± 0.12 gC m−2 ppm−1 (1982–1996) to −0.05 ± 0.03 gC m−2 ppm−1 (2086–2100). The smaller decline in NEP’s indirect effect (−0.10 gC m−2 ppm−1) versus GPP’s (−0.36) suggests partial offset by respiration responses. - Direct effect also declines: Between 1982–1996 and 2000–2014, declines in eCO2(dir) are estimated as: CMIP6_SMA −0.44 gC m−2 ppm−1 (−22.8%); satellite analog obs −1.20 (−78.3%); satellite regression obs-RM −1.65 (−67.0%); CanESM5 factorial −1.38 (−69.2%). - Relative contributions and dominance: The relative contribution of eCO2(ind) to eCO2(net) decreases from 11.1% (1982–1996) to −22.6% (2086–2100), while eCO2(dir) accounts for a larger relative share due to faster decline of eCO2(ind). Negative eCO2(ind) is projected to outweigh positive eCO2(dir) over 30.3% of vegetated land by 2041–2055 and 48.0% by 2086–2100. - Joint changes: 66.9% of vegetated land exhibits the same direction of change in eCO2(ind) and eCO2(dir) between 1982–1996 and 2086–2100; the concurrent decrease (−/−) is most common (48.5%), particularly at northern latitudes. In these areas, changes in indirect and direct effects explain 47.9% and 52.1% of the reduction in eCO2(net) on growing-season GPP (−4.87 gC m−2 ppm−1), respectively. Northern lands show a sharper eCO2(net) decrease (−2.69 gC m−2 ppm−1; −82%) than the global mean (−1.65; −75.7%). - Mechanisms: Widespread land drying underlies the weakening indirect effect. By 2086–2100, surface soil moisture decreases by 7.3% globally (p < 0.01), with 82.6% of vegetated land drier than 1982–1996. In humid regions (SM_surf > 0.26 m3 m−3), eCO2(ind) declines with drying and improves with wetting. In water-limited regions (SM_surf < 0.26), the negative eCO2(ind) tends to weaken with drying, likely due to increased water-use efficiency under CO2 and drought conditions, though declines occur under both drying and wetting, pointing to roles of vegetation type and diversity.

The study demonstrates a global transition of the indirect effect of elevated CO2 on vegetation—from positive in the late 20th century to near-zero or negative in the early 21st century—especially across boreal and other cold/dry regions. This aligns with observed weakening of temperature–vegetation relationships and increasing water limitations. Projections indicate that eCO2(ind) will remain negative and continue to decline under high emissions, with substantial implications for net ecosystem carbon uptake. The results indicate that increasing water limitation and associated aridity trends are key mechanisms, while interactions among multiple climate drivers (temperature, precipitation, VPD, radiation) and their covariation with direct CO2 effects shape nonlinear responses. Disturbance regimes (insects, pathogens, mortality) and rapid climate change further threaten productivity and resilience, potentially intensifying the negative indirect effects. Model–observation discrepancies in the magnitude of decline suggest that ESMs may underestimate water limitation, disturbances, and nonlinear mortality processes, leading to optimistic projections of the land carbon sink if indirect effects are misestimated.

This work provides multi-line evidence—from factorial ESM simulations and satellite-observation-based climate analog analyses—that the indirect effect of elevated CO2 on vegetation carbon uptake has weakened since the 1980s and has turned negative in the early 21st century, particularly in northern high latitudes. Both indirect and direct CO2 effects are projected to decline through the century under SSP5-8.5, with the negative indirect effect increasingly dominating regional dynamics and potentially driving eCO2(net) toward negative values over large areas. Widespread land drying and increasing aridity emerge as central mechanisms, with additional contributions from weakening temperature benefits and potential increases in climate-driven disturbances. These findings imply a reduced capacity of terrestrial ecosystems to offset anthropogenic emissions and highlight the risk of amplifying climate–carbon feedbacks. Future research should improve representation of water limitation, disturbance and mortality processes, and nonlinear interactions in ESMs; extend factorial experiments into future scenarios across more models; and continue integrating satellite constraints to refine projections and inform land-based mitigation strategies.

- Model limitations: Current ESMs may underrepresent water limitation, disturbances (insects, pathogens), and vegetation mortality, contributing to underestimation of the decline in eCO2(ind) and discrepancies with observations. Soil depth and hydrological representations vary across models. - Scenario and experiment coverage: Future CO2-only forcing experiments are unavailable; only CanESM5 provides the full set of historical factorial runs. Projections are limited to SSP5-8.5. - Observation-based assumptions: The climate analog approach assumes annual effects without legacy influences and that eCO2 dominates short-term climate change; satellite GPP proxies (NIRv, kNDVI) have uncertainties. Observational period ends in 2014. - Growing season and variable choices: Results depend on growing-season definitions and chosen hydrometeorological predictors, although multiple robustness checks were performed.