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

Chen, Zefeng; Wang, Weiguang; Forzieri, Giovanni; Cescatti, Alessandro

doi:10.1038/s41467-024-45957-x

Download PDF

Article
Open access
Published: 19 February 2024

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

Nature Communications volume 15, Article number: 1500 (2024) Cite this article

7529 Accesses
34 Altmetric
Metrics details

Subjects

Abstract

Although elevated atmospheric CO₂ concentration (eCO₂) has substantial indirect effects on vegetation carbon uptake via associated climate change, their dynamics remain unclear. Here we investigate how the impacts of eCO₂-driven climate change on growing-season gross primary production have changed globally during 1982–2014, using satellite observations and Earth system models, and evaluate their evolution until the year 2100. We show that the initial positive effect of eCO₂-induced climate change on vegetation carbon uptake has declined recently, shifting to negative in the early 21st century. Such emerging pattern appears prominent in high latitudes and occurs in combination with a decrease of direct CO₂ physiological effect, ultimately resulting in a sharp reduction of the current growth benefits induced by climate warming and CO₂ fertilization. Such weakening of the indirect CO₂ effect can be partially attributed to the widespread land drying, and it is expected to be further exacerbated under global warming.

Increasing impacts of extreme droughts on vegetation productivity under climate change

Article 25 November 2019

Increased control of vegetation on global terrestrial energy fluxes

Article 16 March 2020

Modeling suggests fossil fuel emissions have been driving increased land carbon uptake since the turn of the 20th Century

Article Open access 03 June 2020

Introduction

Terrestrial ecosystems absorb ~30% of anthropogenic carbon dioxide (CO₂) emissions and thus play a fundamental role in mitigating climate change^1,2. Over the past five decades, the terrestrial carbon sink has more than doubled at a pace that is consistent with the increase in anthropogenic CO₂ emissions^2,3. Current evidence demonstrates that the enhancement of the terrestrial carbon sink is partially attributable to the increased carbon uptake by vegetation under elevated atmospheric CO₂ concentration (eCO₂)^4,5,6,7. The eCO₂-induced changes in vegetation carbon uptake (represented by gross primary production (GPP)) are governed by two different mechanisms. The first is the direct effect of eCO₂ through the stimulation of photosynthetic carbon fixation and the enhancement of water-use efficiency (hereafter eCO₂(dir))^8,9. The second is the indirect effect of eCO₂ through its radiative forcing and the associated change in climate (e.g., temperature and water regime) and related environmental conditions (e.g., variation in nitrogen availability linked to temperature-driven changes in the mineralization rate of soil organic matter) (hereafter eCO₂(ind))^10,11,12. Recently, data-driven assessments based on in-situ and satellite observations have documented a declining trend in eCO₂(dir)^9,13. Given the dominant role of eCO₂ in the recent increase in GPP¹⁴, the sign and temporal variation in eCO₂(ind) is expected to increasingly control the future trajectory of the terrestrial carbon budget¹⁵. However, the dynamics of such indirect CO₂ effect on the terrestrial carbon budget remain largely elusive. The relative importance of future indirect versus direct effects of eCO₂ in regulating vegetation carbon uptake has not yet been quantified, and the underlying ecological mechanisms remain poorly understood. Such knowledge gaps are reflected in substantial uncertainties in the effectiveness of land-based climate mitigation policies.

eCO₂(ind) originates from the strong and non-linear effects of eCO₂-induced climate change on terrestrial GPP, which involve multiple pathways, including the plants’ response to changing temperature, water supply, atmospheric dryness (expressed by vapor pressure deficit, VPD) and their complex interactions¹⁶. In addition, these pathways via which climate influences GPP also interact with the eCO₂(dir)^17,18. For example, the rising VPD with eCO₂ generally causes a reduction in stomatal aperture, modulating the transpiration rate and—at the same time—the positive effect of CO₂ fertilization on photosynthesis¹⁹. In view of the variety of interacting feedbacks that regulate vegetation carbon uptake, it is challenging to quantitatively disentangle the total eCO₂(ind), particularly at regional-to-global scales, where local-scale findings of Free-air CO₂ enrichment (FACE) experiments may not be applicable^20,21. Recent studies based on satellite products and model simulations have reported the weakening of the temperature-vegetation relationship in northern ecosystems over the past 30 years²², the increasingly negative impact of VPD on alpine grassland productivity²³, and the increasing water constraint on vegetation growth in many regions across the globe^24,25 and the corresponding higher risk of droughts to the global carbon cycle²⁶. However, considering that changes in temperature, atmospheric dryness, precipitation, and soil moisture are single components of the climate response to the CO₂ radiative forcing, findings of the abovementioned studies can only partially reflect the temporal variations in eCO₂(ind). eCO₂ drives changes in various climatic factors, and their effects on vegetation carbon uptake are covariant, can be additive or offsetting, and may lead to nonlinearities due to different feedback mechanisms^27,28. Existing studies focusing on a single climate driver (e.g., temperature²²), generally assumed that the effects are independent by neglecting the covariation and the interaction between drivers. Therefore, the assessment of the variations in the total indirect effect of eCO₂ can only be partially represented.

To address these knowledge gaps, here we investigate the dynamics in eCO₂(ind) at the global scale for the period 1982–2014 using both satellite retrievals and an ensemble of Earth system models (ESMs) participating in the Coupled Model Intercomparison Project Phase 6 (CMIP6)²⁹ (Table 1), and project potential changes in eCO₂(ind) up to the year 2100 under the SSP5-8.5 scenario. Factorial simulations derived from the fully coupled experiment and the biogeochemically coupled experiment are used to disentangle the eCO₂(ind) signal for the historical and scenario periods³⁰ (Table 2, details in Methods). To further evaluate the robustness of model-based results, we retrieve the eCO₂(ind) term from satellite observations (hereafter eCO₂(ind)_obs) through a statistical methodology within the climate analog framework (Methods). We complement the analyses by deriving eCO₂(dir) through multiple non-linear regression, incorporating CO₂ and climate drivers, and exploring its relationship with eCO₂(ind) across time and space. Finally, we investigate the sensitivity of eCO₂(ind) on land aridity to elucidate the underlying eco-hydrological mechanisms.

Table 1 Information of CMIP6 ESMs used in this study

Full size table

Table 2 Description of CMIP6 factorial simulations

Full size table

Results

Temporal change in the indirect effect of eCO₂

An ensemble of historical simulations from seven CMIP6 models (CMIP6_SMA, SMA: simple model averaging) shows that global eCO₂(ind) during the period 2000–2014 is significantly (p < 0.05, t test) lower than that during 1982–1996 (Fig. 1a, b). Averaged across the global vegetated areas, eCO₂(ind) simulated by CMIP6 models decreases from 0.24 ± 0.32 gC m⁻² ppm⁻¹ (mean ± s.e.) during 1982–1996 to −0.04 ± 0.24 gC m⁻² ppm⁻¹ during 2000–2014 (Fig. 1a, b). The emergence of negative eCO₂(ind) during 2000–2014 suggests the recent upsurge of climate stresses on the global vegetation carbon uptake, which is in agreement with the negative contribution of climate change on global GPP trend after 2000s reported in previous literature³¹. Remarkable differences in changes in eCO₂(ind) emerge across geographic areas and climatological gradients. Cold and dry climate zones experience a prominent decline in eCO₂(ind). The statistically significant decreasing signal is mostly in boreal regions (16.8% of global vegetated land with p < 0.05) with hot spots in eastern Canada, Scandinavia, and south-central Siberia (Fig. 1c, d). Warm and wet climate zones show an opposite tendency with more limited significant patterns (9.9% of global vegetated land with p < 0.05) (Fig. 1c, d).

**Fig. 1: Historical variations in the indirect effect of elevated atmospheric CO₂ concentration (eCO₂) on vegetation carbon uptake.**

In parallel, we used satellite retrievals of near-infrared reflectance of vegetation (NIRv) as a proxy of observed GPP³² to further verify the robustness of the signals derived from model simulations. The satellite-observed eCO₂(ind)_obs was disentangled from the other confounding effects through a climate analog approach³³, based on the identification of years with similar climate and distinct atmospheric CO₂ concentration (details in Methods). Observation-based results confirm a global weakening effect of eCO₂-driven climate change on GPP between the two periods (2000–2014 versus 1982–1996), with an overall change in eCO₂(ind)_obs of −0.38 gC m⁻² ppm⁻¹. We also found a good agreement between model-based and observation-based results in terms of spatial patterns emerging across climatological and latitudinal gradients (Fig. 1d–f).

A comprehensive set of experiments was additionally performed to test whether our model-based results were potentially affected by the data source, temporal window length, and the criteria used to define the growing season (Supplementary Text 1 and 2; Supplementary Figs. 1 and 2, and Table 1). Meanwhile, analyses replicated by using the kernel normalized difference vegetation index (kNDVI) as an alternative satellite GPP proxy were also performed to further verify the robustness of our results (Supplementary Text 3; Supplementary Fig. 3). Altogether, these results univocally show a substantial reduction of the indirect effect of eCO₂ at the global scale (Fig. 1a and Supplementary Figs. 1–3) and particularly in the Northern Hemisphere (Fig. 1d–f and Supplementary Fig. 3c). Such patterns agree with the weakening temperature-vegetation relationship in northern ecosystems documented in previous literature²², and appear plausibly influenced by the increasing water limitation (Supplementary Fig. 4).

eCO₂(ind) is expected to further decline in all investigated future temporal periods under the SSP5-8.5 scenario to the point that the global mean could persistently settle on negative values (Fig. 2a). Five out of seven individual ESMs agree that eCO₂-driven climate change will exert a negative role on the global vegetation carbon uptake for the period 2086–2100, albeit the inter-model spread is considerable (Supplementary Fig. 5a). For the period 2086–2100, the global eCO₂(ind)—as estimated by CMIP6_SMA—is projected to decrease significantly by 0.36 gC m⁻² ppm⁻¹ compared to the analogous estimate derived for the period 1982–1996 (p < 0.01, t test) (Fig. 2a). Such decreasing signal appears statistically significant (p < 0.05) over 46.5% of global vegetated land and prominently in the Northern Hemisphere (Fig. 2b,c). The global declining signal is partially dampened by opposite increasing patterns mainly occurring along the equatorial belt, which, however manifest statistically significant over a smaller extent (32.7%).

**Fig. 2: Projection of future variations in indirect effect of elevated atmospheric CO₂ concentration (eCO₂) on vegetation carbon uptake.**

To derive a more comprehensive picture of the terrestrial ecosystem response to eCO₂-driven climate change, we explored the temporal change in the strength of indirect CO₂ effect on carbon release by respiration (Supplementary Fig. 6), and on the net ecosystem carbon uptake (eCO₂(ind)-NEP, Fig. 2d, e) through factorial experiments of CMIP6 ESMs. We estimated a global eCO₂(ind)-NEP of −0.02 gC m⁻² ppm⁻¹ during the whole historical period (1982–2014), which is consistent with previous findings about the negative carbon-climate feedback from the land’s perspective (i.e., positive from the atmosphere’s perspective)³⁴. In pace with the attenuation of the indirect effect on total vegetation carbon uptake (Fig. 2a), global eCO₂(ind)-NEP is projected to decrease from 0.05 ± 0.12 gC m⁻² ppm⁻¹ during 1982–1996 to −0.05 ± 0.03 gC m⁻² ppm⁻¹ during 2086–2100 (inset box in Fig. 2d). The smaller decline in eCO₂(ind)-NEP compared to that in eCO₂(ind) (−0.1 versus −0.36 gC m⁻² ppm⁻¹) suggests the concurrently reduced influence on ecosystem respiration and its consequent offsetting effect. The latitudinal gradient of the changes in eCO₂(ind)-NEP between the two periods is largely concordant with the one derived from the changes in eCO₂(ind), thus reflecting similar spatial dependences on environmental factors (Fig. 2b–e).

Relationship between the indirect and direct effects of eCO₂

To quantify the relative importance of indirect versus direct effects of eCO₂ in regulating vegetation carbon uptake, simulated and observed eCO₂(dir) was derived based on a multiple non-linear regression (Methods) (i.e., CMIP6_SMA and obs-RM in Fig. 3a). The analyses were complemented by two additional independent estimates of eCO₂(dir) based on factorial experiments of CanESM5 (i.e., CanESM5-FE in Fig. 3a), and on the climate analog approach applied to observational datasets (i.e., obs in Fig. 3a) (details in Methods). We found that along with the decrease in eCO₂(ind), global eCO₂(dir) has dropped as well in recent years and is expected to further decline in the coming decades (Fig. 3a). Model results based on factorial experiments and non-linear regression show a strong reduction in global eCO₂(dir) between the periods 2000–2014 and 1982–1996, largely in agreement with satellite-derived estimates (Fig. 3a). Nevertheless, the magnitude of the decline simulated by CMIP6_SMA (−0.44 gC m⁻² ppm⁻¹, or −22.8%) is clearly lower than the analogous estimate derived from satellite product (obs: −1.20 gC m⁻² ppm⁻¹ or −78.3%; obs-RM: −1.65 gC m⁻² ppm⁻¹ or −67.0%) and from dedicated factorial experiments (CanESM5-FE: −1.38 gC m⁻² ppm⁻¹ or −69.2%) (Fig. 3a). While we recognized the intrinsic difficulties of disentangling drivers and producing robust causal attribution in observation-based analysis, we argued the emerging differences between models and observations could be partially attributable to the simplifying assumptions of CMIP6 models.

**Fig. 3: Relationship between direct and indirect effects of elevated atmospheric CO₂ concentration (eCO₂) on vegetation carbon uptake.**

Under the investigated SSP5-8.5 scenario, the relative importance of eCO₂(ind) and eCO₂(dir) for the terrestrial carbon cycle is expected to vary greatly. The relative contribution of eCO₂(ind) to the net effect of eCO₂ (i.e., eCO₂(net), the sum of eCO₂(dir) and eCO₂(ind)) will likely decrease from 11.1% (1982–11996) to −22.6% (2086–12100) (Fig. 3b). On the contrary, the relative contribution of eCO2(dir) is projected to increase, mainly due to the higher relative decreasing rate of eCO₂(ind). However, in view of the expected progressive decline in both eCO₂(ind) and eCO₂(dir), eCO₂(net) could become negative, and eCO₂(ind) could emerge as the dominant driver of the future temporal dynamic of GPP. Some regions of the globe, such as central Canada, northern Amazon, and western and southern Africa, could exhibit a dominant role of eCO₂(ind) by the end of 21st century (Supplementary Fig. 7). A detailed analysis suggests that the negative eCO₂(ind) will overcome the positive eCO₂(dir) over 30.3% of global vegetated land by 2041–12055, and over 48.0% by 2086–12100 (Supplementary Fig. 8). These results agree with previous studies which have emphasized the expected net negative role of eCO₂ on the terrestrial carbon uptake as a result of the increasing detrimental impacts of climate change on vegetation and a saturating CO₂ fertilization¹³.

Results reveal that 66.9% of global vegetated land could experience the same direction of changes in eCO₂(ind) and eCO₂(dir) (i.e., “+ +” and “– –” in Fig. 3c) between the historical (1982–11996) and future (2086–2100, SSP5-8.5) period, while the remaining 33.1% could manifest reverse directions of changes (i.e., “+ –” and “– +”). The concurrent decrease in eCO₂(ind) and eCO₂(dir) (“– –”) appears to be the most pervasive case (48.5%), particularly over northern latitudes (Fig. 3c). Averaged across regions with simultaneous reductions in eCO₂(ind) and eCO₂(dir), changes in eCO₂(ind) and eCO₂(dir) explain 47.9% and 52.1% of reduction in net CO₂ effect on growing-season GPP (−4.87 gC m⁻² ppm⁻¹), respectively (Fig. 3d). Such concurrent decrease in eCO₂(ind) and eCO₂(dir) is also reflected in the sharper decrease in eCO₂(net) in northern lands (−2.69 gC m⁻² ppm⁻¹, or −82.0%) between the historical and future periods compared to the global mean (−1.65 gC m⁻² ppm⁻¹, or −75.7%), as estimated by CMIP6_SMA (Supplementary Table 2).

Additional analyses based on model simulations from the idealized 1% per year increasing CO₂ experiments show spatial patterns and trends in both indirect and direct CO₂ effects at the global mean level similar to those described above (Supplementary Text 4; Supplementary Figs. 9–11). Meanwhile, under such idealized scenario where radiatively-coupled mode is available, estimates of direct CO₂ effect based on the non-linear regression framework and those based on the climate analog approach were carefully compared against those obtained directly from factorial experiments (Supplementary Text 4). The high agreement among the three sets of estimates further supports the validity of our multiple non-linear regression framework and of the climate analog approach (Supplementary Figs. 12 and 13).

GPP in the Northern Hemisphere has increased steadily during the past decades³⁵ in response to the large and positive eCO₂(ind) and eCO₂(dir) (Supplementary Fig. 14), thus playing a critical contribution to the global terrestrial carbon sink^36,37. Therefore, the widespread and strong decrease in both indirect and direct effects of eCO₂ in the Northern Hemisphere resulting from our analyses rises concern about the future dynamic of the regional carbon sink and its capacity to keep the pace of anthropogenic emissions.

Mechanisms behind the decline in the indirect effect of eCO₂

To disentangle the possible mechanisms responsible for the declining eCO₂(ind), we explored its relationship with the expected changes in terrestrial water availability. To this aim, we first exploited the CMIP6 simulations to quantify the spatiotemporal variations in aridity conditions, here expressed in terms of surface (0–10 cm) soil moisture (SM_surf). Results indicate a projected widespread decline in terrestrial water availability by the end of the century compared to the current conditions (82.6% of global vegetated land exposed to a reduction in SM_surf, Fig. 4a). At the global level and based on multi-model means (i.e., CMIP6_SMA), we estimated a significant decrease in SM_surf during 2086–2100 by 7.3% (p < 0.01, t-test) compared to analogous estimates obtained for the period 1982–1996 (Fig. 4b). Similar drying patterns emerge for individual model runs (Supplementary Fig. 15), for total soil moisture (SM_total), for a widely used aridity index (defined as the ratio of annual precipitation to potential evapotranspiration, P/PET) (Supplementary Figs. 16 and 17). Previous studies focusing on dryness indices³⁸ and hydrological regimes^39,40 further corroborate such drying trends.

**Fig. 4: Sensitivity of indirect CO₂ effect on terrestrial water availability.**

To investigate the relationship between change in eCO₂(ind) and land surface drying/wetting, we averaged the change in eCO₂(ind) across gradients of mean annual SM_surf during 1982–1996 and the corresponding change in SM_surf (i.e., 2000–2014 versus 1982–1996, and 2086–2100 versus 1982–1996). SM_surf = 0.26 m³ m⁻³ generally corresponds to P/PET = 1 at the mean annual scale (Supplementary Fig. 17b), which is widely treated as the threshold between non-humid and humid regions^41,42. We found that eCO₂(ind) generally declines (enhances) with the land drying (wetting) in humid regions (SM_surf > 0.26 m³ m⁻³, Supplementary Fig. 4a) in both historical and scenario simulations (Fig. 4c, d). However, in water-limited conditions (SM_surf < 0.26 m³ m⁻³), the weakened negative eCO₂(ind) along with the land drying results in a negative relationship between changes in eCO₂(ind) and SM_surf (Fig. 4c, d and Supplementary Fig. 7). CO₂ and drought-related enhancement in growing-season water-use efficiency (WUE) (Supplementary Fig. 18), relax the water limitation to vegetation growth, especially over semi-arid climate zones^24,43,44,45, and may consequently limit the negative trend in eCO₂(ind) (Fig. 4d). In addition, for water-limited environments, a decrease in eCO₂(ind) occurs consistently under both land drying and wetting, indicating the possible importance of other factors, such as vegetation type and species diversity, in modulating the vegetation response to climate change.

Similar sensitivities to increasing water limitation have been obtained using SM_total and P/PET in place of SM_surf and referring to different temporal window lengths (Supplementary Figs. 16, 17, and 19). Such high consistency demonstrates the substantial independence of our results on the proxy of terrestrial water availability and the selection of time-window length.

Discussion

Our study provides multiple and coherent evidence that the indirect effect of eCO₂ on global vegetation carbon uptake via associated climate change has declined over the last three decades (Fig. 1 and Supplementary Fig. 3). The signal of the ongoing trends has been derived both from Earth system models’ simulations and from satellite observations using a statistical approach to disentangle direct and indirect CO₂ effects from time series analysis. In addition, results show that the positive indirect effect of eCO₂ that has stimulated the global GPP in recent years will most likely continue to decline in the future, particularly in northern high latitudes, and turn into negative values firmly under the high CO₂ emission scenario (Fig. 2). This epochal change in the sign of the indirect CO₂ effect may lead to a positive land carbon-climate feedback from the atmosphere’s perspective^34,46.

The interpretation of the CO₂ effects mediated by climate change is intrinsically complex because primary productivity is controlled by different environmental drivers in the different biomes, like low temperature in the boreal regions, incoming radiation in the humid tropics, or water availability in the arid regions²⁷. For this reason, analyses have to address multiple factors and their interactions at once^28,47. However, previous assessments were largely based on the analysis of a single climatic factor (e.g., temperature²², water availability⁴⁸, and VPD²³), while our study presents an attempt to integrate multiple drivers across the different World regions. For instance, the strong signal emerging in the boreal regions (Figs. 1f and 2c) can be partially attributable to the weakening temperature-vegetation relationship in northern ecosystems²², which may be related to the non-linear response of photosynthesis to temperature, increased extreme heat, increasing water limitation driven by the anticipation of phenology⁴⁹ and expansion of woody shrubs^50,51.

Our assessment shows that the increasing water limitation is a critical driver of the weakened indirect effect of eCO₂ on the global vegetation carbon uptake (Fig. 4 and Supplementary Figs. 16, 17 and 19). This phenomena may be driven by the detrimental effect of water scarcity on the resilience of global vegetation to climate variability and extremes⁵². An exception to this pattern occurs in water-limited environments, where the negative indirect effect of eCO₂ seems to weaken with the land drying (Fig. 4c, d), probably due to the enhancement of direct CO₂ effect on WUE and the resulting mitigating effect on water constraints²⁴ (Supplementary Fig. 18). A recent analysis focusing on global forests suggests that WUE and aridity index are closely and negatively related below a threshold value of aridity index≈1⁵³, partly supporting our finding.

The complex interplay between the rapid changes in climate conditions and the increasing risks of natural disturbances may also contribute to the transition from positive to negative indirect CO₂ effects. For example, warmer and drier conditions facilitate insect outbreaks, while warmer and wetter conditions increase disturbance from pathogens⁵⁴. The expected intensification of disturbances may amplify the negative effect of climate change on primary productivity by enhancing plant vulnerability and mortality rates⁵⁵. However, the abovementioned processes have not yet been fully considered in state-of-the-art dynamic vegetation models and ESMs that are used in major climate assessments like the Intergovernmental Panel on Climate Change (IPCC) Assessment Reports^56,57. The simplified representation of disturbances and mortality in these models may ultimately hamper our full understanding of the ongoing and future variations in carbon-climate feedback, and in turn, lead to the overestimation of the future terrestrial carbon sink due to the misestimation of the indirect effect of eCO₂. In fact, our study shows that CMIP6 model simulations clearly report a lower magnitude of the decline in global indirect CO₂ effect during the historical period compared with observation-based estimates (Fig. 1f). This could be attributable to the poor representation of water limitation and natural disturbances in current ESMs, the limited consideration of vegetation mortality associated with biotic agents and the non-linearity of the processes behind^57,58. Considering that disturbance regimes are expected to intensify in many parts of the globe because of climate change^54,58,59, the enhanced representation of the phenomena in ESMs would plausibly produce an even stronger decline in the global projection of the indirect CO₂ effect.

The weakened indirect effect of eCO₂ reported here, in addition to the concurrent decline in direct physiological effect, confirms the transitory nature of the strong growth benefits induced by climate warming and CO₂ fertilization, especially in the boreal regions^9,60. These findings are in accordance with the expectation about the future saturation of CO₂ fertilization effect and the increasingly negative effect of climate change on vegetation¹³. More importantly, the simultaneous reductions in indirect and direct effects of eCO₂ imply that eCO₂ may exert a less positive up to negative role on the terrestrial carbon uptake in the future (Fig. 3), consequently reducing the ecosystems’ capacity to sequester atmospheric CO₂. These phenomena may ultimately lead to an acceleration of climate change in the second part of the century, further challenging the efforts of humanity toward carbon neutrality. In addition, more frequent and severe climate extremes in a warming climate^61,62, e.g., increasing drought conditions, may further aggravate the decline in the indirect effect of eCO₂ (Fig. 4) as well as that in direct CO₂ fertilization⁶³. The intensification of a positive feedback loop between climate change and land CO₂ emission undoubtedly would limit the potential of terrestrial ecosystems to serve as carbon sinks and have great implications for the efficacy of land-based mitigation policies and for the societal efforts required for meeting climate mitigation targets. In this respect, our results contribute to a better understanding of global change impacts on terrestrial ecosystems under current and future conditions, and meanwhile, may help the development of more integrated and realistic mitigation strategies, by informing climate policies on the weakening of the fertilization effects of eCO₂ and associated amplification of climate warming.

Methods

CMIP6 simulations

To explore the indirect effect of eCO₂ on vegetation carbon uptake, we used outputs from an ensemble of seven Earth system models (ESMs) that participate in the carbon-climate feedback experiment (C4MIP) within the framework of the Coupled Model Intercomparison Project Phase 6 (CMIP6)³⁰ (https://esgf-node.llnl.gov/search/cmip6/): ACCESS-ESM1-5, CanESM5, CNRM-ESM2-1, E3SM-1-1, MIROC-ES2L, MRI-ESM2-0, and UKESM1-0-LL (Table 1). These models were selected because they provide simulations under different coupling modes required to disentangle the effects of eCO₂. We focused on the SSP5-8.5 scenario because C4MIP simulations are available only for the highest emission trajectory (CO₂ concentration is projected to reach 1135 ppm in 2100⁶⁴). The ESMs have full carbon cycles, which include carbon uptake by vegetation that varies in response to changes in atmospheric CO₂ concentration and climate⁴⁶. For each ESM, one biogeochemically-coupled experiment and one fully-coupled experiment in both historical (1982–2014) and future scenario (2015–2100) periods were analyzed within a factorial simulation framework (Table 2). In biogeochemically coupled experiments (“hist-bgc” and “ssp585-bgc” in the CMIP6 terminology), biogeochemical processes over land respond to eCO₂, whereas the radiative code experiences fixed CO₂. In the fully-coupled experiment (“historical” and “ssp585” in the CMIP6 terminology), both radiative and biogeochemical processes respond to eCO₂ (consistent with observations in the historical period). All other forcings (e.g., non-CO₂ greenhouse gases, aerosols, and land use) are identical for these two sets of experiments, i.e., time-varying in both radiative and biogeochemical processes. Furthermore, we used outputs from CO₂ individual forcing experiment (“hist-CO₂” in the CMIP6 terminology) conducted by the Detection and Attribution Model Intercomparison Project (DAMIP)⁶⁵ (Table 2). “hist-CO₂” experiment refers to the historical simulation driven only by observed changes in CO₂ concentration, with other forcings keeping temporally constant (e.g., non-CO₂ greenhouses gases, aerosols, and land use). Combining “hist-CO₂” with “historical” and “hist-bgc” enables to quantify the historical eCO₂(dir) through factorial simulations of CMIP6 runs. Unfortunately, CO₂ individual forcing experiment has not yet been extended to the future period, and only one ESM (i.e., CanESM5) took part in all three experiments. Because of these disadvantages, the abovementioned analysis applied for the estimation of global direct effects of eCO₂ was complemented by a more general regression framework extendible to the full ESM ensemble and to the different temporal periods (details in section “Quantifying indirect and direct effects of eCO₂ by model outputs”). Factorial simulations based on the abovementioned experiments (“hist-bgc”, “historical”, and “hist-CO₂”) were elaborated in the following sections.

A set of variables generated by ESM simulations were used for the following analyses, including: monthly scale gross primary production (GPP), net primary production (NPP), autotrophic respiration (R_a), heterotrophic respiration (R_h), evapotranspiration (ET), maximum, minimum and mean air temperature (T_max, T_min, and T), precipitation (P), cloud cover (CL), relative humidity (RH), surface (0–10 cm) soil moisture (SM_surf, 0–10 cm), and total soil moisture (SM_total, depth depending on models, see Table 1). Considering that the hydrologically active soil depth varies greatly among the models (from 2 m in UKESM1-0-LL to 35.18 m in E3SM-1-1), SM_total as well as SM_surf was converted from the original gravimetric unit (kg m⁻²) to volumetric unit (m³ m⁻³) by dividing the gravimetric soil water content by the corresponding soil depth, Such conversion allows for the comparison of results obtained from different models and the development of more robust multi-model ensembles of soil moisture. Variations in SM_total and SM_surf were expressed in relative terms (%) with respect to their average values computed for the baseline period (e.g., 1982–1996)³⁹ (Fig. 4 and Supplementary Figs. 4, 15 and 16). Only for the ES3M-1-1 model, some of the abovementioned variables were not provided (NPP, T_max, T_min, and RH), and therefore they were retrieved by empirical formula and statistical approach (additional details reported in Supplementary Text 5). We additionally derived ecosystem respiration (R_eco) as the sum of R_a and R_h, and the net ecosystem production (NEP) was calculated as the difference between GPP and R_eco⁶⁶. Vapor pressure deficit (VPD), which directly relates to atmospheric water demand^19,67, was calculated based on Abbott and Tabony⁶⁸ for each grid-cell as follows:

$${{{{{\rm{VPD}}}}}}=0.6108{{{{{{\rm{e}}}}}}}^{\frac{17.27{{{{{\rm{T}}}}}}}{{{{{{\rm{T}}}}}}+237.3}}\left(1-\frac{{{{{{\rm{RH}}}}}}}{100}\right)$$

(1)

where T is given in °C, and the resulting VPD is in kPa. Furthermore, we used the ratio of mean annual P to potential evapotranspiration (PET) as the aridity index retrieved from the FAO Penman-Monteith algorithm⁶⁹. Details on the PET estimation are reported in Supplementary Text 6.

CMIP6 outputs were resampled to a common 0.5° × 0.5° global grid-cell using the bilinear method of interpolation. Moreover, for each temporal window (e.g., 1982–1996, 2000–2014, 2086–2100, and 1982–2014), we computed the associated multi-year mean growing season at the grid-cell scale. The growing season was defined as the period spanning months with average T > 0°C¹⁴ and—limitedly to arid and semi-arid ecosystems—cumulative P between 10% and 90% of the annual total P (Supplementary Fig. 20). The integration of a P threshold in the definition of the growing season for water-limited environments enables to account for possible inactive vegetation phase at T > 0 °C due to water deficit conditions⁷⁰. Areas characterized by an aridity index, quantified in terms of P/PET, <1 were labeled as arid and semi-arid ecosystems⁹. For arid and semi-arid grid cells located in the Southern Hemisphere, P accumulation was set to start in July and end in June of the next year. The resulting growing season was used as a reference period to aggregate the original monthly variables provided by CMIP6 to the growing-season scale. The robustness of our results is tested with respect to two alternative definitions of the growing season period: (1) T > 5 °C and cumulative P between 10% and 90% of the annual total P; (2) T > 5°C and cumulative P between 20% and 80% of the annual total P (Supplementary Fig. 2). In both cases, the P threshold is applied to arid and semi-arid regions only.

Observation-based products

We exploited the long-term GPP dataset (hereafter GPP_obs) based on near-infrared reflectance of vegetation (NIRv) retrieved from the Advanced Very High Resolution Radiometer (AVHRR) reflectance observations^32,71 (https://data.tpdc.ac.cn/en/data/d6dff40f-5dbd-4f2d-ac96-55827ab93cc5/). The satellite GPP dataset, provided at monthly temporal resolution and at 0.05° spatial resolution, has global coverage and spans the period 1982–2014 (Supplementary Fig. 21a). It has been largely validated in previous studies against ground measurements and compared with estimates derived from machine-learning upscaling approaches, light-use-efficiency models and processed-based models^32,72. To match the spatial and temporal resolution of ESMs output, satellite GPP data were resampled to 0.5° and integrated over the growing season derived from the CMIP6 simulations in the fully-coupled experiment, as described above, to increase consistency in the data-model comparison (Supplementary Fig. 20). The obtained satellite-based growing season GPP data were used to evaluate the ESMs performance in capturing global GPP dynamics (Supplementary Figs. 22 and 23).

Furthermore, to explore the observed impact of eCO₂ on vegetation carbon uptake (details in the following sections), we used monthly T_max, T_min, T, P, CL, actual water vapor (VP), and PET retrieved from the Climatic Research Unit (CRU v4.05) climate dataset⁷³ (https://catalogue.ceda.ac.uk/) delivered for the whole globe at 0.5° spatial resolution and covering the period 1982–2014.

We additionally derived monthly VPD values as the difference between the saturated vapor pressure (SVP) and VP for each grid-cell based on the following formulation:

$${{{{{\rm{VPD}}}}}}=0.6108{{{{{{\rm{e}}}}}}}^{\frac{17.27{{{{{\rm{T}}}}}}}{{{{{{\rm{T}}}}}}+237.3}}-{{{{{\rm{VP}}}}}}$$

(2)

where T is given in °C, VP and VPD are in kPa. Here we applied Eq. (2) instead of Eq. (1) to estimate VPD because RH is not available from the CRU dataset. All climatic factors were then aggregated at the growing-season temporal resolution.

We derived a global vegetated land mask from the annual land cover maps of the European Space Agency’s Climate Change Initiative⁷⁴ (https://www.esa-landcover-cci.org) acquired for the period 1992–2014 at 300 m spatial resolution, referring to a simplified aggregation scheme based on physiognomy alone. Land cover maps were resampled to 0.5° using the majority method to match the common spatial resolution. All grid cells (0.5 × 0.5° resolution) classified as vegetation class (including forest, grassland, shrubland, cropland, and wetland) throughout the 23 years were defined as vegetated areas and included in our analyses (Supplementary Fig. 24).

Assessing indirect/direct effects of eCO₂ from model outputs

Following similar approaches reported in literature^10,16,75,76, the effect of eCO₂-induced climate change on growing-season GPP (i.e., the indirect effect of eCO₂, expressed as eCO₂(ind)) was derived from factorial simulations of multiple CMIP6 experiments by calculating the difference between the trend in growing-season GPP generated in the fully-coupled mode and that in the biogeochemically-coupled mode normalized by the increase rate of atmospheric CO₂ concentration:

$${{{{{{\rm{eCO}}}}}}}_{2}({{{{{\rm{ind}}}}}})=\frac{{{{{{\rm{\delta }}}}}}{{{{{{\rm{GPP}}}}}}}^{{{{{{\rm{FULL}}}}}}}-{{{{{{\rm{\delta }}}}}}{{{{{\rm{GPP}}}}}}}^{{{{{{\rm{BGC}}}}}}}}{{{{{{{\rm{\delta }}}}}}{{{{{\rm{CO}}}}}}}_{2}}$$

(3)

where ${{{{{\rm{\delta }}}}}}{{{{{{\rm{GPP}}}}}}}^{{{{{{\rm{FULL}}}}}}}$ and ${{{{{{\rm{\delta }}}}}}{{{{{\rm{GPP}}}}}}}^{{{{{{\rm{BGC}}}}}}}$ are the trends in growing-season GPP in the fully-coupled experiment (i.e., “historical” and “ssp585”) and the biogeochemically-coupled experiment (i.e., “hist-bgc” and “ssp585-bgc”), respectively; ${{{{{{\rm{\delta }}}}}}{{{{{\rm{CO}}}}}}}_{2}$ represents the trend in atmospheric CO₂ concentration and is prescribed by CMIP6^64,77. The statistical significance of the trends was evaluated using the nonparametric Mann–Kendall test. The absolute signal (term ${{{{{\rm{\delta }}}}}}{{{{{{\rm{GPP}}}}}}}^{{{{{{\rm{FULL}}}}}}}-{{{{{{\rm{\delta }}}}}}{{{{{\rm{GPP}}}}}}}^{{{{{{\rm{BGC}}}}}}}$) was normalized to the unit of gC m⁻² ppm⁻¹, to eliminate the impact of the difference in increasing rate of atmospheric CO₂ concentration in various periods (e.g., 1982–1996, 2000−2014, and 2086–2100). The term eCO₂(ind) excludes the direct physiological effect of eCO₂ and the effects of non-CO₂ forcing agents on GPP, as these components have been removed from the factorial simulations (Eq. (3)). The approach enabled us to separately quantify the eCO₂(ind) for different reference temporal period (e.g., 1982−1996, 2000–2014, and 2086–2100) at grid-cell level. For global-scale eCO₂(ind) estimates, the GPP terms reported in (Eq. (3)) were obtained by spatial average weighting each grid-cell value based on its area (Fig. 1a). The same methodology is applied consistently for all global-scale and regional-scale aggregated metrics described in the following sections. The analyses were complemented by applying Eq. (3) to NPP, R_a, R_h, R_eco, and NEP (in place of GPP) to comprehensively evaluate the response of distinct carbon fluxes to CO₂ radiative forcing (Fig. 2d, e and Supplementary Fig. 6).

We quantified the direct effect of eCO₂ on growing-season GPP (i.e., eCO₂(dir)) within a multiple non-linear regression framework applied to simulations obtained from the CMIP6 fully-coupled experiment. Such approach was specifically designed since radiatively-coupled experiments ideally required to derived eCO₂(dir) from factorial simulations were not available. To derive robust fitting functions, we first performed a collinearity test based on the variance inflation factor (VIF), to preliminary select what drivers to include in the multiple regression. Results show that CO₂, T_min, P, VPD, and CL, show no/weak collinearity (VIF < 10) in most parts of the globe (CRU: 97.0 ~ 100%; CMIP_SMA: 86.7 ~ 100%) and thus were all retained in the predictor set (Supplementary Fig. 25). Considering that climate exerts non-linear control on terrestrial carbon uptake^78,79, non-linear terms (e.g., interaction and quadratic terms) were incorporated into the regression model in addition to linear terms. Following the modeling framework described in Chen et al.⁸⁰, we used stepwise regression at the grid-cell scale to reduce redundant predictors. The model form most often identified across all grid cells was ultimately adopted to each grid-cell, enabling a consistent analysis at the global scale. The adopted model is described by the following equation:

$${{{{{\rm{GPP}}}}}}=\beta \left({{{{{{\rm{CO}}}}}}}_{2}\right)+{C}_{1}\left({{{{{\rm{P}}}}}}\right)+{C}_{2}\left({{{{{\rm{VPD}}}}}}\right)+{C}_{3}\left({{{{{{\rm{T}}}}}}}_{\min }\cdot {{{{{\rm{VPD}}}}}}\right)+{C}_{4}\left({{{{{\rm{P}}}}}}\cdot {{{{{\rm{CL}}}}}}\right)+{C}_{5}+\varepsilon$$

(4)

where β, C₁, C₂, C₃, C₄, and C₅ represent the regression coefficients, and ε is the residual error term. Therein, β (gC m⁻² ppm⁻¹) refers to the sensitivity of GPP to CO₂, and thus reflects the term eCO₂(dir). Such an approach enabled us to disentangle the direct physiological effect of eCO₂ on GPP by factoring out the potentially confounding effects of climatic factors. All variables in Eq. (4) were taken from CMIP6 model simulations under “historical” and “ssp585” experiments. Regressions were estimated on annual anomalies (i.e., annual values minus the mean signal for a given period) for all variables, thus removing the background effects on vegetation but preserving those originating from interannual variations⁹. We performed an additional set of modeling experiments to test the model sensitivity on different hydrological variables. To this aim, we expressed the interannual variations in GPP within a non-linear regression that incorporates soil moisture in place of P (details in Supplementary Text 7). Test results based on the Akaike Information Criterion (AIC), the corrected Akaike Information Criterion (AICc), and the Bayesian Information Criterion (BIC) suggest that non-linear regression based on soil moisture has no substantial improvement in model performance (Supplementary Fig. 26) and leads to larger inter-model spread compared to the original one (i.e., Eq. (4)) (Supplementary Fig. 27 and Table 3). We therefore retained the regression framework based on P as defined in Eq. (4) for subsequent analyses.

For CanESM5, we estimated eCO₂(dir) by the use of an alternative method based on the outputs from three sets of factorial experiments available for the historical period as follows:

$${{{{{{\rm{eCO}}}}}}}_{2}({{{{{\rm{dir}}}}}})=\frac{{{{{{\rm{\delta }}}}}}{{{{{{\rm{GPP}}}}}}}^{{{{{{\rm{CO}}}}}}2}-({{{{{\rm{\delta }}}}}}{{{{{{\rm{GPP}}}}}}}^{{{{{{\rm{FULL}}}}}}}-{{{{{\rm{\delta }}}}}}{{{{{{\rm{GPP}}}}}}}^{{{{{{\rm{BGC}}}}}}})}{{{{{{\rm{\delta }}}}}}{{{{{{\rm{CO}}}}}}}_{2}}$$

(5)

where ${{{{{\rm{\delta }}}}}}{{{{{{\rm{GPP}}}}}}}^{{{{{{\rm{CO}}}}}}2}$ is the trend in growing-season GPP in the CO₂ individual forcing experiment (“hist-CO₂”). Estimates based on this approach were compared against those generated by the abovementioned regression model to test the robustness of Eq. (4) in quantifying global mean eCO₂(dir) and its change (Fig. 3a).

Deriving indirect effect of eCO₂ from observations

To further corroborate our model-based findings, we investigated the direct and indirect components of eCO₂ and their changes through an observation-based approach. The method is based on the pixel-level assessment of the changes in GPP under similar climate conditions but different atmospheric CO₂ concentrations. For this purpose, we used the climate analog approach³³ to identify couples of years with different atmospheric CO₂ concentrations but similar climate conditions (i.e., climate analogous (CA) years) in each time period (e.g., 1982–1996, and 2000−2014) based on the CRU v4.05 climate dataset. Temporal climate analogs are derived from the Mahalanobis distance, which is a multivariate distance independent of the scale of the climate variables⁸¹. For the period 1982−1996, as an example, we first identified the two sub-periods 1982–1988 and 1989–1996, and then identified for each grid-cell the years in which the climate condition is most similar between the two sub-periods. We calculated the Mahalanobis distance based on eleven climate variables derived from the CRU v4.05 dataset and identified as key determinants for climate analog analysis in previous studies^33,81,82,83. The selected variables include: mean annual CL, mean annual VPD, mean annual T, total annual P, annual P/PET, mean T, and total P for December–February (DJF) and June-August (JJA), T seasonality (represented by the standard deviation of monthly T), and P seasonality (represented by the coefficient of variation in monthly P). In order to reduce the dimensionality of the data space, we applied a principal component analysis to all climate variables and discarded the principal components with variance <0.01³³. The Mahalanobis distance was then computed between all the possible 56-member couples of years at the grid-cell scale based on the following equation:

$${{{{{{\rm{MD}}}}}}}_{{{{{{\rm{ij}}}}}}}=\sqrt{\mathop{\sum }\nolimits_{k=1}^{N}\frac{{\left({x}_{{jk}}-{x}_{{ik}}\right)}^{2}}{{\sigma }_{k}^{2}}}$$

(6)

where j and i belong to the first and second sub-period, respectively; x_jm and x_im are the values of the principal component k in the year j and i, N represents the number of retained principal components, and σ_k² refers to the standard deviation of the principal component k. Low MD scores represent similar climate conditions between the two sub-periods, high MD scores the opposite. For each grid-cell, the minimum Mahalanobis distance (MD_min) was then selected, and its corresponding couple of years were identified as potential CA years.

We assessed the statistical significance of the obtained MD_min. Considering that the chi distribution provides a null distribution for (non-squared) Mahalanobis distances, the obtained MD_min can be expressed probabilistically as percentiles of a chi distribution with degrees of freedom corresponding to the number of dimensions in which MD_min was measured (i.e., N in Eq. (6)). Following Mahony et al.³³, we considered the 95th percentile of the associated chi distribution to identify the upper threshold of the representative analog. MD_min whose corresponding percentile is lower than the abovementioned threshold, indicates a statistically similar climate between those two years (i.e., CA years).

Climate analogs were found not significant for a minority of grid cells (4.2%), and these areas were therefore excluded from the following analyses. For the remaining grid cells, we estimated the direct physiological effect of eCO₂ on GPP as follows:

$${{{{{{{\rm{eCO}}}}}}}_{2}({{{{{\rm{dir}}}}}})}_{{{{{{\rm{obs}}}}}}}=\frac{{\Delta {{{{{\rm{GPP}}}}}}}_{{{{{{\rm{obs}}}}}}}^{{{{{{\rm{CA}}}}}}}}{\Delta {{{{{{\rm{CO}}}}}}}_{2}^{{{{{{\rm{CA}}}}}}}}$$

(7)

where $\Delta {{{{{{\rm{GPP}}}}}}}_{{{{{{\rm{obs}}}}}}}^{{{{{{\rm{CA}}}}}}}$ is the change in growing-season GPP_obs computed between CA years, and $\Delta {{{{{{\rm{CO}}}}}}}_{2}^{{{{{{\rm{CA}}}}}}}$ is the corresponding variation in atmospheric CO₂ concentration acquired from the Earth System Research Laboratory of NOAA⁸⁴ (https://www.esrl.noaa.gov/gmd/ccgg/trends/). The analysis was complemented by the estimation of the combined direct and indirect effect of eCO₂ on GPP quantified as:

$${{{{{{{\rm{eCO}}}}}}}_{2}({{{{{\rm{net}}}}}})}_{{{{{{\rm{obs}}}}}}}=\frac{\Delta {{{{{{\rm{GPP}}}}}}}_{{{{{{\rm{obs}}}}}}}}{\Delta {{{{{{\rm{CO}}}}}}}_{2}}$$

(8)

where $\Delta {{{{{{\rm{GPP}}}}}}}_{{{{{{\rm{obs}}}}}}}$ and $\Delta {{{{{{\rm{CO}}}}}}}_{2}$ represent the change in the mean growing-season GPP_obs and CO₂ concentration, respectively, computed between the two sub-periods. We finally derived the observed indirect effect of eCO₂ on vegetation photosynthesis via associated climate change in a given period by combining Eq. (7) and Eq. (8) as follows:

$${{{{{{{\rm{eCO}}}}}}}_{2}({{{{{\rm{ind}}}}}})}_{{{{{{\rm{obs}}}}}}}=\frac{\Delta {{{{{{\rm{GPP}}}}}}}_{{{{{{\rm{obs}}}}}}}}{\Delta {{{{{{\rm{CO}}}}}}}_{2}}-\frac{\Delta {{{{{{\rm{GPP}}}}}}}_{{{{{{\rm{obs}}}}}}}^{{{{{{\rm{CA}}}}}}}}{\Delta {{{{{{\rm{CO}}}}}}}_{2}^{{{{{{\rm{CA}}}}}}}}$$

(9)

The approach assumes that both direct and indirect eCO₂ effects on GPP are annual and do not have legacy effects, and that eCO₂ is the dominant factor of climate change in the short-term. The above-described analyses were carried out at the grid-cell scale and separately for the periods 1982–1996 and 2000–2014. Changes between the two periods were then used to quantify the historical variations in the indirect effect of eCO₂ from observations (Fig. 1e, f).

To further test the robustness of our methods, a set of additional experiments were produced. First, the observed direct effect of eCO₂ was estimated using an alternative approach based on multiple non-linear regression model (i.e., Eq. (4)) in combination with CRU v4.05 climate dataset and GPP_obs. The obtained results, expressed as eCO₂(dir)_obs-RM, were confronted with climate analog-derived eCO₂(dir)_obs estimates (i.e., obs-RM and obs in Fig. 3a). Second, the climate analog approach presented above and applied to observations was also implemented with CMIP6 model outputs in fully-coupled experiments to verify the consistency with results obtained from factorial simulations described in Eq. (3) (Supplementary Fig. 28).

Statistical analysis

To explore the dynamics of the indirect effect of eCO₂ on vegetation carbon uptake during the historical period, we quantified the temporal changes in model-based eCO₂(ind) and observation-based eCO₂(ind)_obs retrieved for the two independent periods 1982–1996 and 2000–2014. The significance of the emerging changes was assessed through t test. Results presented in the main text refer to analyses conducted over 15-year time windows. Results obtained for different temporal window lengths (12 and 16 years) are quantified as well to verify the robustness of our results (Fig. 1a and Supplementary Fig. 1).

In exploring the projected changes in eCO₂(ind) from CMIP6 model simulations, we refer to six 15-year consecutive and independent periods, namely 2011–2025, 2026–2040, 2041–2055, 2056–2070, 2071–2085, 2086–2100 (Fig. 2a). We considered a series of not overlapped temporal windows to eliminate the possible impact of autocorrelation. To better disentangle the signal of future variation in eCO₂(ind), we compared the eCO₂(ind) during the last 15 years (2086–2100) against that one originating from the first 15 years of the historical period investigated here (1982–1996) and assessed the statistical significance of the change through t test.

Furthermore, to properly represent the generality of the relationships between the direction and extent of changes in eCO₂(ind) and local aridity conditions, we performed binned average analysis across environmental gradients. Such spatial averaging minimizes the uncertainty originating from spatial heterogeneity (e.g., the difference in topography and vegetation type) that may randomly affect the control of climate change on vegetation carbon uptake (Figs. 1d and 4c, d).

Data availability

All datasets used in this study are publicly available as referenced in Methods. Source data are provided with this paper.

Code availability

The custom MATLAB (R2023a) codes written to read and analyze data and generated figures are publicly available at https://doi.org/10.5281/zenodo.10451254.

References

Ahlström, A. et al. The dominant role of semi-arid ecosystems in the trend and variability of the land CO₂ sink. Science 348, 895–899 (2015).
Article ADS PubMed Google Scholar
Friedlingstein, P. et al. Global carbon budget 2022. Earth Syst. Sci. Data 14, 4811–4900 (2022).
Article ADS Google Scholar
Ballantyne, A., Alden, C., Miller, J., Tans, P. & White, J. Increase in observed net carbon dioxide uptake by land and oceans during the past 50 years. Nature 488, 70–72 (2012).
Article ADS CAS PubMed Google Scholar
Schimel, D., Stephens, B. & Fisher, J. Effect of increasing CO₂ on the terrestrial carbon cycle. Proc. Natl Acad. Sci. USA 112, 436–441 (2015).
Article ADS CAS PubMed Google Scholar
Keenan, T. et al. Recent pause in the growth rate of atmospheric CO₂ due to enhanced terrestrial carbon uptake. Nat. Commun. 7, 13428 (2016).
Article ADS CAS PubMed PubMed Central Google Scholar
Liu, Y. et al. Field-experiment constraints on the enhancement of the terrestrial carbon sink by CO₂ fertilization. Nat. Geosci. 12, 809–814 (2019).
Article ADS CAS Google Scholar
Jiang, M. et al. The fate of carbon in a mature forest under carbon dioxide enrichment. Nature 580, 227–231 (2020).
Article ADS CAS PubMed Google Scholar
Smith, W. et al. Large divergence of satellite and Earth system model estimates of global terrestrial CO₂ fertilization. Nat. Clim. Change 6, 306–310 (2016).
Article ADS Google Scholar
Wang, S. et al. Recent global decline of CO₂ fertilization effects on vegetation photosynthesis. Science 370, 1295–1300 (2020).
Article ADS CAS PubMed Google Scholar
Zhu, P. et al. Elevated atmospheric CO₂ negatively impacts photosynthesis through radiative forcing and physiology-mediated climate feedback. Geophys. Res. Lett. 44, 1956–1963 (2017).
Article ADS CAS Google Scholar
Green, J. et al. Large influence of soil moisture on long-term terrestrial carbon uptake. Nature 565, 476–479 (2019).
Article ADS CAS PubMed PubMed Central Google Scholar
Duffy, K. et al. How close are we to the temperature tipping point of the terrestrial biosphere? Sci. Adv. 7, eaay1052 (2021).
Article ADS CAS PubMed PubMed Central Google Scholar
Peñuelas, J. et al. Shifting from a fertilization-dominated to a warming-dominated period. Nat. Ecol. Evol. 1, 1438–1445 (2017).
Article PubMed Google Scholar
Cheng, L. et al. Recent increases in terrestrial carbon uptake at little cost to the water cycle. Nat. Commun. 8, 110 (2017).
Article ADS PubMed PubMed Central Google Scholar
Mitchard, E. The tropical forest carbon cycle and climate change. Nature 559, 527–534 (2018).
Article ADS CAS PubMed Google Scholar
Bala, G. et al. An estimate of equilibrium sensitivity of global terrestrial carbon cycle using NCAR CCSM4. Clim. Dyn. 40, 1671–1686 (2013).
Article Google Scholar
Gedalof, Z. & Berg, A. Tree ring evidence for limited direct CO₂ fertilization of forests over the 20th century. Glob. Biogeochem. Cycles 24, GB3027 (2010).
Article ADS Google Scholar
Reich, P., Hobbie, S. & Lee, T. Plant growth enhancement by elevated CO₂ eliminated by joint water and nitrogen limitation. Nat. Geosci. 7, 920–924 (2014).
Article ADS CAS Google Scholar
Yuan, W. et al. Increased atmospheric vapor pressure deficit reduces global vegetation growth. Sci. Adv. 5, eaax1396 (2019).
Article ADS CAS PubMed PubMed Central Google Scholar
Hickler, T. et al. CO₂ fertilization in temperate FACE experiments not representative of boreal and tropical forests. Glob. Change Biol. 14, 1531–1542 (2008).
Article ADS Google Scholar
Wang, T. et al. Emerging negative impact of warming on summer carbon uptake in northern ecosystems. Nat. Commun. 9, 5391 (2018).
Article ADS CAS PubMed PubMed Central Google Scholar
Piao, S. et al. Evidence for a weakening relationship between interannual temperature variability and northern vegetation activity. Nat. Commun. 5, 5018 (2014).
Article ADS CAS PubMed Google Scholar
Ding, J. et al. Increasingly important role of atmospheric aridity on Tibetan alpine grasslands. Geophys. Res. Lett. 45, 2852–2859 (2018).
Article ADS Google Scholar
Gonsamo, A. et al. Greening drylands despite warming consistent with carbon dioxide fertilization effect. Glob. Change Biol. 27, 3336–3349 (2021).
Article Google Scholar
Jiao, W. et al. Observed increasing water constraint on vegetation growth over the last three decades. Nat. Commun. 12, 3777 (2021).
Article ADS CAS PubMed PubMed Central Google Scholar
Xu, C. et al. Increasing impacts of extreme droughts on vegetation productivity under climate change. Nat. Clim. Change 9, 948–953 (2019).
Article ADS CAS Google Scholar
Seddon, A., Macias-Fauria, M., Long, P., Benz, D. & Willis, K. Sensitivity of global terrestrial ecosystems to climate variability. Nature 531, 229–232 (2016).
Article ADS CAS PubMed Google Scholar
Fu, Z. et al. Atmospheric dryness reduces photosynthesis along a large range of soil water deficits. Nat. Commun. 13, 989 (2022).
Article ADS CAS PubMed PubMed Central Google Scholar
Eyring, V. et al. Overview of the Coupled Model Intercomparison Project Phase 6 (CMIP6) experimental design and organization. Geosci. Model Dev. 9, 1937–1958 (2016).
Article ADS Google Scholar
Jones, C. et al. C4MIP – the coupled climate–carbon cycle model intercomparison project: experimental protocol for CMIP6. Geosci. Model Dev. 9, 2853–2880 (2016).
Article ADS CAS Google Scholar
Zhang, Y., Song, C., Band, L. & Sun, G. No proportional increase of terrestrial gross carbon sequestration from the greening Earth. J. Geophys. Res. Atmos. 124, 2540–2553 (2019).
Article CAS Google Scholar
Badgley, G., Field, C. & Berry, J. Canopy near-infrared reflectance and terrestrial photosynthesis. Sci. Adv. 3, e1602244 (2017).
Article ADS PubMed PubMed Central Google Scholar
Mahony, C., Cannon, A., Wang, T. & Aitken, S. A closer look at novel climates: new methods and insights at continental to landscape scales. Glob. Change Biol. 23, 3934–3955 (2017).
Article ADS Google Scholar
Friedlingstein, P. et al. Climate–carbon cycle feedback analysis: results from the C⁴MIP model intercomparison. J. Clim. 19, 3337–3353 (2006).
Article ADS Google Scholar
Ciais, P. et al. Five decades of northern land carbon uptake revealed by the interhemispheric CO₂ gradient. Nature 568, 221–225 (2019).
Article ADS CAS PubMed Google Scholar
Pan, Y. et al. A large and persistent carbon sink in the world’s forests. Science 333, 988–993 (2011).
Article ADS CAS PubMed Google Scholar
Sitch, S. et al. Recent trends and drivers of regional sources and sinks of carbon dioxide. Biogeosciences 12, 653–679 (2015).
Article ADS Google Scholar
Dai, A. Increasing drought under global warming in observations and models. Nat. Clim. Change 3, 52–58 (2013).
Article ADS Google Scholar
Berg, A., Sheffield, J. & Milly, P. Divergent surface and total soil moisture projections under global warming. Geophys. Res. Lett. 44, 236–244 (2017).
Article ADS Google Scholar
Cook, B. et al. Twenty-first century drought projections in the CMIP6 forcing scenarios. Earth’s Future 8, e2019EF001461 (2020).
Article ADS Google Scholar
United Nations Environment Program. World Atlas of Desertification. 182 (1997).
Ukkola, A. et al. Reduced streamflow in water-stressed climates consistent with CO₂ effects on vegetation. Nat. Clim. Change 6, 75–78 (2016).
Article ADS Google Scholar
Peters, W. et al. Increased water-use efficiency and reduced CO₂ uptake by plants during droughts at a continental scale. Nat. Geosci. 11, 744–748 (2018).
Article ADS CAS PubMed PubMed Central Google Scholar
Piao, S. et al. Characteristics, drivers and feedbacks of global greening. Nat. Rev. Earth Environ. 1, 14–27 (2020).
Article ADS Google Scholar
Kannenberg, S., Driscoll, A., Szejner, P., Anderegg, W. & Ehleringer, J. Rapid increases in shrubland and forest intrinsic water-use efficiency during an ongoing megadrought. Proc. Natl. Acad. Sci. USA 118(52), e2118052118 (2021).
Article CAS PubMed PubMed Central Google Scholar
Arora, V. et al. Carbon–concentration and carbon–climate feedbacks in CMIP6 models and their comparison to CMIP5 models. Biogeosciences 17, 4173–4222 (2020).
Article ADS CAS Google Scholar
Nemani, R. et al. Climate-driven increases in global terrestrial net primary production from 1982 to 1999. Science 300, 1560–1563 (2003).
Article ADS CAS PubMed Google Scholar
Li, W. et al. Widespread increasing vegetation sensitivity to soil moisture. Nat. Commun. 13, 3959 (2022).
Article ADS CAS PubMed PubMed Central Google Scholar
Buermann, W. et al. Widespread seasonal compensation effects of spring warming on northern plant productivity. Nature 562, 110–114 (2018).
Article ADS CAS PubMed Google Scholar
Pearson, R. et al. Shifts in Arctic vegetation and associated feedbacks under climate change. Nat. Clim. Change 3, 673–677 (2013).
Article ADS Google Scholar
Myers-Smith, I. et al. Complexity revealed in the greening of the Arctic. Nat. Clim. Change 10, 106–117 (2020).
Article ADS Google Scholar
Forzieri, G., Dakos, V., McDowell, N., Ramdane, A. & Cescatti, A. Emerging signals of declining forest resilience under climate change. Nature 608, 534–539 (2022).
Article CAS PubMed Google Scholar
Adams, M., Buckley, T., Binkley, D., Neumann, M. & Turnbull, T. CO₂, nitrogen deposition and a discontinuous climate response drive water use efficiency in global forests. Nat. Commun. 12, 5194 (2021).
Article ADS CAS PubMed PubMed Central Google Scholar
Seidl, R. et al. Forest disturbances under climate change. Nat. Clim. Change 7, 395–402 (2017).
Article ADS Google Scholar
Forzieri, G. et al. Emergent vulnerability to climate-driven disturbances in European forests. Nat. Commun. 12, 1081 (2021).
Article ADS CAS PubMed PubMed Central Google Scholar
Anderegg, W. et al. Tree mortality from drought, insects, and their interactions in a changing climate. New Phytol. 208, 674–683 (2015).
Article PubMed Google Scholar
Bonan, G. & Doney, S. Climate, ecosystems, and planetary futures: the challenge to predict life in Earth system models. Science 359, eaam8328 (2018).
Article PubMed Google Scholar
McDowell, N. et al. Pervasive shifts in forest dynamics in a changing world. Science 368, eaaz9463 (2020).
Article CAS PubMed Google Scholar
Senf, C. & Seidl, R. Mapping the forest disturbance regimes of Europe. Nat. Sustain. 4, 63–70 (2021).
Article Google Scholar
D’Orangeville, L. et al. Beneficial effects of climate warming on boreal tree growth may be transitory. Nat. Commun. 9, 3213 (2018).
Article ADS PubMed PubMed Central Google Scholar
Reichstein, M. et al. Climate extremes and the carbon cycle. Nature 500, 287–295 (2013).
Article ADS CAS PubMed Google Scholar
Frank, D. et al. Effects of climate extremes on the terrestrial carbon cycle: concepts, processes and potential future impacts. Glob. Change Biol. 21, 2861–2880 (2015).
Article ADS Google Scholar
Obermeier, W. et al. Reduced CO₂ fertilization effect in temperate C3 grasslands under more extreme weather conditions. Nat. Clim. Change 7, 137–141 (2017).
Article ADS CAS Google Scholar
Meinshausen, M. et al. The shared socio-economic pathway (SSP) greenhouse gas concentrations and their extensions to 2500. Geosci. Model Dev. 13, 3571–3605 (2020).
Article ADS CAS Google Scholar
Gillett, N. et al. The detection and attribution model intercomparison project (DAMIP v1.0) contribution to CMIP6. Geosci. Model Dev. 9, 3685–3697 (2016).
Article ADS Google Scholar
Spafford, L. & MacDougall, A. Validation of terrestrial biogeochemistry in CMIP6 Earth system models: a review. Geosci. Model Dev. 14, 5863–5889 (2021).
Article ADS CAS Google Scholar
Novick, K. et al. The increasing importance of atmospheric demand for ecosystem water and carbon fluxes. Nat. Clim. Change 6, 1023–1027 (2016).
Article ADS CAS Google Scholar
Abbott, P. & Tabony, R. The estimation of humidity parameters. Meteorol. Mag. 114, 49–56 (1985).
Google Scholar
Allen, R. G., Pereira, L. S., Raes, D. & Smith, M. Crop evapotranspiration-guidelines for computing crop water requirements. FAO irrigation and drainage paper no. 56 (1998).
Cook, K. & Vizy, E. Impact of climate change on mid-twenty-first century growing seasons in Africa. Clim. Dyn. 39, 2937–2955 (2012).
Article Google Scholar
Wang, S. & Zhang, Y. Long-term (1982–2018) global gross primary production dataset based on NIRv [Data set]. https://doi.org/10.6084/m9.figshare.12981977.v2 (2020).
Wang, S., Zhang, Y., Ju, W., Qiu, B. & Zhang, Z. Tracking the seasonal and inter-annual variations of global gross primary production during last four decades using satellite near-infrared reflectance data. Sci. Total Environ. 755, 142569 (2021).
Article ADS CAS PubMed Google Scholar
Harris, I., Osborn, T., Jones, P. & Lister, D. Version 4 of the CRU TS monthly high-resolution gridded multivariate climate dataset. Sci. Data 7, 109 (2020).
Article PubMed PubMed Central Google Scholar
European Space Agency (ESA). Land Cover CCI Product User Guide Version 2 (2017).
Skinner, C., Poulsen, C. & Mankin, J. Amplification of heat extremes by plant CO₂ physiological forcing. Nat. Commun. 9, 1094 (2018).
Article ADS PubMed PubMed Central Google Scholar
Park, S., Kim, J. & Kug, J. The intensification of Arctic warming as a result of CO₂ physiological forcing. Nat. Commun. 11, 2098 (2020).
Article ADS CAS PubMed PubMed Central Google Scholar
Meinshausen, M. et al. Historical greenhouse gas concentrations for climate modelling (CMIP6). Geosci. Model. Dev. 10, 2057–2116 (2017).
Article ADS CAS Google Scholar
Zhang, X. et al. Linear and nonlinear effects of dominant drivers on the trends in global and regional land carbon uptake: 1959 to 2013. Geophys. Res. Lett. 43, 1607–1614 (2016).
Article ADS CAS Google Scholar
Felton, A., Knapp, A. & Smith, M. Precipitation–productivity relationships and the duration of precipitation anomalies: an underappreciated dimension of climate change. Glob. Change Biol. 27, 1127–1140 (2021).
Article ADS CAS Google Scholar
Chen, Z., Wang, W., Cescatti, A. & Forzieri, G. Climate-driven vegetation greening further reduces water availability in drylands. Glob. Change Biol. 29, 1421–1679 (2023).
Article Google Scholar
Fitzpatrick, M. & Dunn, R. Contemporary climatic analogs for 540 North American urban areas in the late 21st century. Nat. Commun. 10, 614 (2019).
Article ADS CAS PubMed PubMed Central Google Scholar
Williams, J., Jackson, S. & Kutzbach, J. Projected distributions of novel and disappearing climates by 2100 AD. Proc. Natl Acad. Sci. USA 104, 5738–5742 (2007).
Article ADS CAS PubMed PubMed Central Google Scholar
Ordonez, A. & Williams, J. Projected climate reshuffling based on multivariate climate-availability, climate-analog, and climate-velocity analyses: implications for community disaggregation. Clim. Change 119, 659–675 (2013).
Article ADS Google Scholar
Masarie, K. & Tans, P. Extension and integration of atmospheric carbon dioxide data into a globally consistent measurement record. J. Geophys. Res. Atmos. 100, 11593–11610 (1995).
Article ADS CAS Google Scholar

Download references

Acknowledgements

W.W. was supported by the National Natural Science Foundation of China (grant no. U2240218). G.F. was supported by the Horizon Europe Project ECO2ADAPT (grant no. 101059498).

Author information

Authors and Affiliations

National Key Laboratory of Water Disaster Prevention, Hohai University, Nanjing, China
Zefeng Chen & Weiguang Wang
Yangtze Institute for Conservation and Development, Hohai University, Nanjing, China
Zefeng Chen & Weiguang Wang
College of Hydrology and Water Resources, Hohai University, Nanjing, China
Zefeng Chen & Weiguang Wang
Department of Civil and Environmental Engineering, University of Florence, Florence, Italy
Giovanni Forzieri
European Commission, Joint Research Centre, Ispra, Italy
Alessandro Cescatti

Authors

Zefeng Chen
View author publications
You can also search for this author in PubMed Google Scholar
Weiguang Wang
View author publications
You can also search for this author in PubMed Google Scholar
Giovanni Forzieri
View author publications
You can also search for this author in PubMed Google Scholar
Alessandro Cescatti
View author publications
You can also search for this author in PubMed Google Scholar

Contributions

W.W. and Z.C. conceived and designed the research; Z.C. implemented the data analysis; G.F. and A.C. contributed analysis ideas; Z.C. interpreted the results and drafted the initial manuscript; W.W., G.F., and A.C. provided suggestions and further improved writing. All authors approved the final version of this manuscript.

Corresponding author

Correspondence to Weiguang Wang.

Ethics declarations

Competing interests

The authors declare no competing interests.

Peer review

Peer review information

: Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Supplementary Information

Peer Review File

Source data

Source Data

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions

About this article

Cite this article

Chen, Z., Wang, W., Forzieri, G. et al. Transition from positive to negative indirect CO₂ effects on the vegetation carbon uptake. Nat Commun 15, 1500 (2024). https://doi.org/10.1038/s41467-024-45957-x

Download citation

Received: 22 July 2023
Accepted: 08 February 2024
Published: 19 February 2024
DOI: https://doi.org/10.1038/s41467-024-45957-x

Comments

By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.