Synergistic Interactions Between Warming and Elevated Atmospheric CO₂ Increase Soil Organic Carbon Sequestration

ABSTRACT

Warming and elevated CO₂ (eCO₂) are two potentially opposing climate‐carbon (C) feedback mechanisms that modulate the magnitude of the land C sink, with warming decreasing and eCO₂ increasing C sequestration. However, their net effect on soil organic C (SOC)—the largest terrestrial C stock—remains uncertain. Here, we quantify how warming, eCO₂, and their interactions influence SOC by using 5558 paired observations from 1392 global studies across ecosystem types. Our study shows that warming reduces SOC by 8.0%, primarily by suppressing aboveground C inputs (−1.0%) and decreasing microbial C use efficiency (−9.9%). Concurrent warming and eCO₂ increase SOC by 7.6%—a synergistic effect larger than eCO₂ alone (+4.8%), primarily contributed by croplands. This outcome may result from increases in plant C inputs and soil nitrogen availability under eCO₂, which facilitate microbial C assimilation and necromass formation (+3.9%). These processes promote the accumulation of mineral‐associated C (+9.9%) and offset the negative effects of warming. The combined effect of warming and eCO₂ is projected to increase SOC by 27.4 Pg C by 2100. Our findings highlight that synergistic interaction between warming and eCO₂ increases SOC sequestration and enhances SOC stability under future climate change.

Global meta‐analysis reveals that warming decreases SOC, whereas warming and elevated CO₂ interactions synergistically enhance SOC sequestration through increased plant C inputs, soil N availability, microbial assimilation, and mineral‐associated C formation, leading to a projected global SOC gain of 27.4 Pg C by 2100 under future climate change.

1. Introduction

Projections under different shared socioeconomic pathway (SSP) scenarios show that atmospheric CO₂ concentrations may reach 500–1000 ppm by 2100, accompanied by a global temperature rise exceeding 2°C, which threatens ecosystem stability and food security (Bongaarts 2024; Tao et al. 2024). In fully coupled carbon (C)‐climate models, the terrestrial C cycle feedback remains a critical but poorly constrained component of global climate change (Cox et al. 2000). The terrestrial C cycle is regulated by global change drivers, primarily warming and elevated atmospheric CO₂ (eCO₂) levels, which co‐occur and alter nutrient availability, plant productivity, and microbial activity, thereby modulating ecosystem C responses (Bai et al. 2023; Black et al. 2017; Jin et al. 2020; Ofiti et al. 2023). As soil organic C (SOC) constitutes the largest terrestrial C reservoir and changes in SOC exert strong control over climate stability, understanding SOC responses to warming and eCO₂ is essential for unraveling the terrestrial C‐climate feedback (Le Quéré et al. 2018; Xu et al. 2025).

Across global studies, the effects of warming and/or eCO₂ on SOC remain highly uncertain, with reported responses ranging from positive to negative (Bai et al. 2023; Carney et al. 2007; Ofiti et al. 2022; Pries et al. 2017; Teng et al. 2024). Therefore, there is an urgent need to clarify the overall effects of warming and eCO₂ on global SOC. Although the impacts of warming and/or eCO₂ on SOC have been extensively documented at the plant (Song et al. 2025; Terrer et al. 2021), soil (Pries et al. 2017; Schimel et al. 2015), and microbial (Liu, Liu, et al. 2021; SiMa et al. 2025) levels individually, integration on whole‐system C cycling at global scales remains limited (Zhou et al. 2016). Additionally, SOC is made up of different fractions, such as mineral‐associated organic C (MAOC, microbially derived and relatively stable) and particulate organic C (POC, plant‐derived and more labile). Those SOC fractions may respond differently to climate change (Davidson and Janssens 2006; van Groenigen et al. 2017). Yet, how these distinctions are driven by regulating plant input, microbial assimilation, and stable SOC formation remains underexplored. Such knowledge gaps, arising in part from limited understanding of plant–soil–microbial C responses, limit robust quantification of the soil C‐climate feedback and induce uncertainties in future climate change projections.

The impacts of warming and/or eCO₂ on SOC vary because they alter plant growth, soil physicochemical properties, and microbial activity in different ways, all of which regulate soil C responses (Liu, Fan, et al. 2025; Luo et al. 2001; Ofiti et al. 2023; Xue et al. 2016). eCO₂ enhances plant growth, thereby increasing primary productivity and, in turn, augmenting plant‐derived C inputs (Cai et al. 2020). However, this does not necessarily translate into increased SOC accumulation (Angst et al. 2021), as long‐term eCO₂ may enrich oligotrophic bacteria that decompose stable organic C and accelerate SOC decomposition (Jin et al. 2020). For example, global meta‐analyses show that despite higher net primary productivity under eCO₂, SOC accumulation often remains unchanged or even declines due to intensified microbial activity and increased diversity (Procter et al. 2015; van Groenigen et al. 2014). Warming further enhances microbial decomposition and reduces microbial C use efficiency (defined as the fraction of C taken up by microbes that is allocated to biomass growth rather than respired as CO₂), thereby promoting the transfer of belowground C to the atmosphere and leading to net SOC loss (Quan et al. 2019). However, when soil nitrogen (N) is sufficient, microorganisms allocate more C to biomass synthesis rather than releasing it as CO₂ through respiration, and the accumulation of microbially derived compounds, such as microbial necromass C, can in turn promote the long‐term stabilization of SOC (Bai et al. 2023; Liang et al. 2017; Tian et al. 2024). Therefore, controversies remain in predicting the impacts of warming and eCO₂ on SOC, highlighting the urgent need to evaluate their interactive effects on plant–soil–microbial C responses at the global scale.

Elevated CO₂ causes climatic warming (Nottingham et al. 2020). Warming, in turn, accelerates the mineralization of soil C pools, increasing the risk of C loss and reinforcing the positive feedback to atmospheric CO₂ (Wang, Guo, et al. 2022). Moreover, eCO₂ typically enhances the priming of old SOC rather than newly added C (Vestergård et al. 2016). At the same time, warming‐induced stimulation of C‐utilization pathways may accelerate the decomposition of both fresh plant‐derived and native SOC inputs. Together, these processes have the potential to intensify SOC losses over time (Miao et al. 2021). However, eCO₂ also enhances photosynthesis and plant productivity, increases the input of root exudates and suberin, the latter being chemically recalcitrant macromolecules that promote C stabilization and can thereby mitigate warming‐induced C losses, acting as a negative feedback to climate change (García‐Palacios et al. 2021; Ofiti et al. 2023; Terrer et al. 2021). This altered input–output balance may stabilize or even slightly increase SOC in the short term, particularly in systems with a strong net primary productivity response (Ofiti et al. 2023). Yet, given the complexity of the individual effects of warming and eCO₂, and the limited availability of long‐term experiments involving multiple global change factors, considerable uncertainty remains about the extent to which eCO₂ offsets or exacerbates warming‐induced SOC losses (Bradford et al. 2016; Crowther et al. 2016).

This study aims to elucidate soil C responses to warming and eCO₂ by synthesizing a global experimental dataset across diverse ecosystems. To this end, we integrate data from key processes involved in soil C cycling, including C inputs, SOC formation, stabilization, and soil C outputs—while clarifying plant–soil‐microbial interactions. We then identify the key factors shaping the C cycle and investigate the pathways of SOC formation and their responses to warming and eCO₂. Finally, we project the trajectory of global SOC change through 2100 under high‐emission, high‐warming scenarios. We find a synergistic effect of warming and eCO₂ on SOC sequestration, mainly via MAOC accumulation, offering implications for future C sequestration potentials under climate change.

2. Methods

2.1. Data Collection

We searched peer‐reviewed journal papers published before April 2025 in the Web of Science, Google Scholar, and China National Knowledge Infrastructure (CNKI) databases using the search terms listed in Table S1. Previous meta‐analyses within the scope of our search were incorporated after duplicates were removed. The number of publications included is summarized in Table S2. Studies were included if they met the following criteria: (i) written in English or Chinese with full text accessible; (ii) control and treatment groups located at the same site, with treatment intensity clearly documented; (iii) means, standard deviations (or errors), and replicate numbers reported or calculable; (iv) experimental duration of at least one growth period or cycle; (v) treatments limited to warming, eCO₂, or their combination, excluding other global change factors. In total, 5558 paired observations from 1392 publications were included: 791 studies on warming, 675 on eCO₂, and 116 on their combined effects, all focusing on top soils (0–30 cm). The selection process followed PRISMA guidelines (Figure S1). Finally, we recorded the ranges of warming magnitude (ΔT) and CO₂ increase (Δppm) to characterize experimental treatments (Figure S2).

Variables describing the soil C cycle response to warming and eCO₂ were categorized into four groups: plant (e.g., aboveground biomass, belowground biomass), microbial (e.g., microbial biomass C, microbial necromass C, microbial C use efficiency), ecosystem (e.g., net ecosystem productivity, gross ecosystem productivity, ecosystem respiration), and soil (e.g., SOC, POC, MAOC, dissolved organic C) (Table S3). Missing values were imputed using spatial interpolation with the missForest algorithm to ensure data completeness for random forest analysis and meta‐analysis (Stekhoven and Bühlmann 2012). This approach, like other random forest‐based techniques, makes no distributional assumptions and is well‐suited for multivariate non‐normal data and complex interactions or nonlinear relationships among variables. We also collected data on enzymes related to C (e.g., cellobiohydrolase, β‐glucosidase), N (e.g., N‐acetyl‐β‐glucosaminidase, leucine aminopeptidase), and P (e.g., acid phosphatase, alkaline phosphatase). For studies reporting results only in graphical format, numerical data were extracted using WebPlotDigitizer 4.8 (https://automeris.io/WebPlotDigitizer/).

2.2. Statistics

The natural log response ratio (lnR) was used to assess the effect size of warming and/or eCO₂ on each variable, following standard meta‐analysis practice (Hedges et al. 1999).

$$\mathrm{lnR}=\ln \left(\frac{X_T}{X_C}\right)=\ln X_T–\ln X_C$$

where X _T and X _C represent the arithmetic means of the treatment and control groups, respectively. When the SD was unavailable or could not be derived from the SE, it was set to one‐tenth of the mean (Lin et al. 2023). If the study reported the standard error (SE), the corresponding standard deviation (SD) was calculated as follows:

$$\mathrm{SD}=\mathrm{SE}\times \sqrt{n}$$

where n represents the sample size. The variance ($v$) of each lnR was then calculated as:

$$v=\frac{{\mathrm{SD}}_T^2}{N_T{X}_T^2}+\frac{{\mathrm{SD}}_C^2}{N_C{X}_C^2}$$

where SD _T and SD _C are the SDs of the treatment and control groups, and N _T and N _C are their respective sample sizes.

Microbial necromass C (mg kg⁻¹ soil) was estimated from fungal GluN and bacterial MurN contents (Fu, Chen, Ma, et al. 2025):

$$\text{Fungal\hspace{0.17em}\hspace{0.17em}necromass}\mathrm{C}=\left(\frac{\text{GluN}}{179.17}–\frac{1.63\times \text{MurN}}{251.23}\right)\times 179.17\times 10.8$$

$$\text{Bacterial necromass}\mathrm{C}=\text{MurN}\times 31.3$$

$$\text{Microbial necromass}\mathrm{C}=\text{fungal necromass}\mathrm{C}+\text{bacterial necromass}\mathrm{C}.$$

The corresponding SD of $\text{microbial necromass}\mathrm{C}$ was calculated using error propagation (Lorber 1986):

$${\mathrm{SD}}_{\mathrm{MNC}}=\sqrt{{\mathrm{SD}}_{\mathrm{FNC}}^2+{\mathrm{SD}}_{\mathrm{BNC}}^2}$$

where SD_MNC is the SD of microbial necromass, and SD_FNC and SD_BNC are the SDs of fungal and bacterial necromass C, respectively.

A weighted random‐effects model meta‐analysis (Gurevitch and Hedges 1999) was then used to estimate the overall impact of global warming and/or eCO₂ on soil C cycling. The weighted mean response ratio (lnRR) was calculated as follows:

$$\text{lnRR}=\frac{\sum \left(\ln R_j\times w_j\right)}{\sum w_j}$$

where lnR _j is the effect size of the j‐th comparison and w _j its weight:

$$w_j=\frac{1}{v_j}$$

with $v_j$ defined as:

$$v_j=v+{\tau }^2$$

where $v$ is the within‐study variance and τ ² the between‐study variance. Because some studies contributed multiple effect sizes, the variable study was treated as a random effect in mixed‐effects models to account for non‐independence. The 95% confidence interval (CI) of lnRR was calculated as:

$$95\%\mathrm{CI}=\ln \mathrm{RR}\pm 1.96\times S_{\text{lnRR}}$$

where $S_{\text{lnRR}}$ is the standard error of lnRR. Effects were considered statistically significant if the 95% CI excluded zero. For interpretation, effect sizes were converted to mean percentage change (MPC):

$$\mathrm{MPC}=\left(e^{\text{lnRR}}–1\right)\times 100\%$$

We assessed the quality of this meta‐analysis using the checklist developed by Koricheva and Gurevitch (Table S4). Publication bias was evaluated by calculating the Rosenthal fail‐safe number (FSN) with the fsn function (Table S5) and by conducting Egger's test, where p < 0.05 indicates significant bias (Egger et al. 1997). The results showed no evidence of publication bias. Funnel plots were also constructed using the funnel function (Figure S3). Additionally, a Jackknife sensitivity analysis was performed to assess the influence of individual studies on the overall robustness. Each study was assigned a unique identifier, and in each iteration, one study was systematically removed from the dataset (Figure S4). Results confirmed that no single study disproportionately influenced the overall effect. All analyses were performed in R version 4.4.1 and ArcGIS10.8.2.

2.3. Environmental Drivers of Variability in Soil C

Partial correlation analysis was used to assess the relationships between SOC and three groups of factors: (i) soil properties (soil C/N, total N, total P, moisture, and pH), representing factors associated with SOC stabilization; (ii) plant variables (total biomass, belowground biomass, and aboveground biomass), representing C inputs; and (iii) soil enzyme activities (acid phosphatase, alkaline phosphatase, cellobiohydrolase, β‐glucosidase, N‐acetyl‐β‐glucosaminidase, leucine aminopeptidase), mediating organic matter decomposition. For each SOC‐predictor relationship, a zero‐order Spearman correlation was first calculated, followed by partial Spearman correlations obtained by separately controlling for each of the other two functional groups. The specific covariates included in each partial correlation analysis are listed in Table S6. Differences between the zero‐order and partial correlation coefficients were interpreted as changes in the SOC‐predictor association after accounting for covariation with the specified covariates (Figure 4a).

FIGURE 4.

Factors influencing SOC responses in terrestrial ecosystems. (a) Partial correlations of SOC change with three categories of factors under warming, eCO₂, and warming + eCO₂: Soil physicochemical properties (SI), plant traits (PA), and soil enzyme activity (EM). All predictors were calculated as response ratios to ensure cross‐study comparability. The outer ring indicates factors correlated with SOC. Sector colors denote the strength and direction of the correlation, with black borders indicating significance (p < 0.05). Changes between zero‐order and partial correlations indicate changes in the SOC‐predictor association after accounting for the specified covariates. (b) Relative importance of response ratios of environmental factors for SOC predicted by random forest models under warming and/or eCO₂. Bars marked with ** and * denote significance at p < 0.01 and p < 0.05, respectively. (c) Relationships between SOC and the key predictor (soil total N) under warming and/or eCO₂. Solid lines represent significant linear regressions (p < 0.001) with 95% confidence intervals (shaded). Acid P, acid phosphatase; AGB, aboveground biomass; Alkaline P, alkaline phosphatase; AP, soil available P; BG, β‐glucosidase; BGB, belowground biomass; CB, cellobiohydrolase; DOC, dissolved organic C; LAP, leucine aminopeptidase; MAP, mean annual precipitation; MAT, mean annual temperature; MBC, microbial biomass C; MBN, microbial biomass N; NAG, N‐acetyl‐β‐glucosaminidase; NH₄ ⁺, ammonium; NO₃ ⁻, nitrate; SM, soil moisture; TB, plant total biomass; TN, soil total N; TP, soil total P; Δppm, CO₂ concentration change; ΔT, Temperature change.

We assessed the relative importance of plant, soil, and microbial properties in explaining SOC variation by evaluating the increase in mean squared error (MSE) for each predictor in the random forest model. The statistical significance of the predictors was tested using the rfPermute R package. Given that changes in N may have significant effects on the formation, transformation, and stability of SOC, we chose to include the response ratio of soil total N. Besides, key predictors also included microbial biomass C, microbial biomass N, plant biomass, dissolved organic C, soil pH, soil moisture, soil total P, soil available P, soil C/N, ammonium, nitrate, warming magnitude (ΔT), CO₂ change (Δppm), experimental duration, mean annual temperature, and mean annual precipitation. After evaluating predictor significance, only the top thirteen variables were retained in the final model (Figure 4b). To further quantify relationships, linear regressions were conducted between SOC and the most influential drivers (Figure 4c).

Structural Equation Modeling (SEM) was used to test two hypotheses: (1) warming and/or eCO₂ affect soil C primarily through plant–soil‐microbial interactions, and (2) SOC changes under combined warming and eCO₂ are predominantly mediated through the MAOC pathway. To minimize potential bias from multicollinearity, variance inflation factors (VIFs) were calculated using the car package, and variables with the highest collinearity were iteratively removed until all VIF values were below 5 (Fu, Chen, Ma, et al. 2025). This procedure yielded the most parsimonious model describing plant–soil‐microbial interactions (Table S7). Variable selection was guided by ecological relevance: the response ratio of soil total N was included as the dominant factor regulating soil C, clarifying its role in modulating C responses; microbial biomass C was included as the major precursor for MAOC formation; and SOC was defined as the final node, integrating contributions from MAOC and POC, the two dominant SOC fractions. This structure enabled simultaneous assessment of the regulatory roles of upstream drivers (response ratios of soil N, plants, and microbes) and the contributions of downstream fractions to SOC. SEM path fitting was conducted using the piecewiseSEM package. The best‐fitting model was selected based on a non‐significant chi‐squared test (p > 0.05), low Akaike Information Criterion (AIC), and high Comparative Fit Index (CFI > 0.9). Missing paths identified by direct separation tests were added to the final model (Figure 2).

FIGURE 2.

Piecewise structural equation models of the direct and indirect drivers of SOC change in terrestrial ecosystems. (a) Warming. (b) eCO₂. (c) Warming + eCO₂. Red arrows indicate positive effects; blue arrows indicate negative effects. Solid lines indicate direct effects, while dashed lines indicate indirect effects. Numbers on arrows are standardized path coefficients derived from partial regressions, with arrow width proportional to effect size. The percentage of explained variance expresses the relative importance of each factor. Asterisks denote significance (*p < 0.05, **p < 0.01, ***p < 0.001). Marginal R ² (R ²m) and conditional R ² (R ²c) represent the variance explained by all predictors, with and without the random effect of sampling site, respectively. AGB, aboveground biomass; BGB, belowground biomass; MAOC, mineral‐associated organic C; MBC, microbial biomass C; POC, particulate organic C; SOC, soil organic C; TN, soil total N.

2.4. Global Upscaling of Soil C Predictions

Four widely adopted and robust machine learning algorithms—random forest, extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), and gradient boosted decision trees (GBDT)—were implemented to predict SOC responses (Ling et al. 2025). To account for the non‐independence among observations originating from the same study, a study‐level grouped five‐fold cross‐validation strategy was applied, in which all observations from the same study were assigned to the same fold. Model performance was evaluated based on root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R ²). Among the evaluated algorithms, GBDT consistently showed the best predictive performance for warming, eCO₂, and combined warming + eCO₂ scenarios (Table S8). Therefore, GBDT was selected as the unified modeling approach for subsequent global‐scale prediction of SOC responses. Future climate data (mean annual precipitation and mean annual temperature) were obtained from the WorldClim database, while soil data (soil moisture, soil C/N, and soil total N) and plant data (aboveground vegetation and root C) were derived from CMIP6 (Coupled Model Intercomparison Project Phase 6, ESGF). Detailed information on the future environmental covariates is provided in Table S9. All covariates were harmonized before model application by resampling them to a common spatial resolution (0.1°) and reprojecting them onto a unified pixel grid (EPSG:4326, WGS84). Continuous variables were resampled using bilinear interpolation to ensure spatial consistency among predictor layers. CMIP6 provides future projections generated by multiple global climate models (GCMs) under different Shared Socioeconomic Pathways (SSPs). As our study focuses specifically on global SOC changes under warming, eCO₂, and their combined effects, we selected the SSP5‐8.5 scenario, which assumes a radiative forcing of 8.5 W/m² by 2100 in the absence of strong greenhouse gas mitigation. This choice enables evaluation of SOC changes under the most severe climate conditions, providing insights into potential soil C changes over the next century.

2.5. Quantification of Uncertainties

Extrapolating empirical relationships driving SOC responses to warming and eCO₂ at a global scale inevitably introduces prediction uncertainty. For example, soils with low total N were underrepresented in the dataset, leading to potentially larger prediction errors and greater uncertainty. In the uncertainty predictions shown in Figure 6, our goal was not only to highlight uncertainty due to limited representation of key predictors (e.g., soil total N), but also to account for low sampling density in extreme climate regions (e.g., hot deserts and tundra) and in peatlands. To address this, temperature and precipitation were included as additional covariates in the model. Using global spatial distribution maps of temperature, precipitation, soil total N, soil C/N, and plant biomass, we extrapolated model predictions across regions where climate and soil conditions were poorly represented in warming and eCO₂ experiments (Figures S5–S7), and quantified the associated uncertainty using study‐level bootstrap resampling. Overall, the uncertainties in our predictions reflect the standard deviation of GBDT predictions across bootstrap iterations, with higher uncertainty generally expected in regions with insufficient sampling and underexplored soil‐climate conditions.

FIGURE 6.

Global patterns of predicted relative SOC responses to warming and/or eCO₂. Under the SSP5‐8.5 scenario, projected global SOC responses to warming, eCO₂, and their combined effects by 2100 are shown (left panel), along with corresponding latitudinal variations (right panel).

3. Results and Discussion

3.1. eCO₂ Overrides Warming‐Induced SOC Loss

By synthesizing 5558 paired observations from 1392 published studies conducted between 1990 and 2025 (Figure 1a), we found that SOC decreased under warming (−8.0%; 95% CI: −8.4 to −7.7%), primarily due to reductions in net ecosystem productivity (−9.5%; −10.6 to −8.4%) and aboveground biomass (−1.0%; −1.8 to −0.3%). Reduced plant inputs directly inhibited POC formation (−8.8%; −10.1 to −7.4%), while the reduction in microbial biomass (−7.5%; −10.1 to −4.7%) and in microbial C use efficiency (−9.9%; −12.7 to −7.1%) led to a decrease in MAOC content (−4.0%; −5.2 to −2.6%) (Figure 1b). Unlike our study, Bai et al. and Liang reported a limited effect of warming on SOC (Bai et al. 2023; Liang 2025), likely due to the fact that previous studies primarily focused on overall SOC responses or microbial respiration responses, which may not fully resolve the differential responses of SOC fractions and microbial‐mediated C stabilization processes under prolonged warming (Liang 2025; Zhang et al. 2025). By compiling a larger and more comprehensive global dataset, our study further revealed the cumulative effects of long‐term warming on SOC dynamics (Figure S8). eCO₂ increased plant C inputs, boosted microbial biomass (+11.4%; +10.8 to +12.0%), and promoted the accumulation of both POC (+3.6%; +1.3 to +5.9%) and MAOC (+9.9%; +8.1 to +11.7%), thereby enhancing SOC responses (+4.8%; +4.3 to +5.3%) (Figure 1c). When combined, warming and eCO₂ intensified root inputs, which stimulated microbial biomass (+36.9%; +28.7 to +45.8%, indicated by PLFA content) and increased the accumulation of microbial necromass C (+20.4%; +5.4 to +37.7%), enhancing MAOC accumulation (+5.5%; +1.7 to +9.5%), which in turn ultimately increased SOC responses (+7.6%; +6.7 to +8.5%) (Figure 1d). Our results therefore highlight that plant–soil‐microbial interactions regulate SOC responses under warming and/or eCO₂.

FIGURE 1.

Responses of the soil C cycle to warming and/or eCO₂ at the global scale. (a) Global distribution of experimental sites in studies of the effects of warming and eCO₂ on SOC; sample sizes are shown in brackets, (b) Responses of the soil C cycle to warming, (c) eCO₂, and (d) warming + eCO₂. Blue values indicate significant increases, red values indicate significant decreases, and black values indicate no significant response (p > 0.05). “na” denotes insufficient data. CUE, microbial C use efficiency; DOC, dissolved organic C; SMB, Soil microbial biomass; MAOC, mineral‐associated organic C; PLFA‐derived microbial biomass; POC, particulate organic C; SOC, soil organic C.

Specifically, eCO₂ increased SOC by increasing net ecosystem productivity by 4.7% (+3.0 to +6.3%) and POC accumulation by 3.6% (+1.3 to +5.9%) (Figure 1c). Microorganisms decomposed plant‐derived C into low‐molecular‐weight compounds, which were incorporated into microbial biomass (+11.4%; +10.8 to +12.0%) and necromass (+3.9%; +0.6 to +7.4%) (Figure 1c) or were directly adsorbed onto minerals. Both pathways promoted MAOC formation (p < 0.05) and enhanced SOC accumulation (p < 0.001) (Figure 2b). Concurrently, eCO₂ reduced soil pH (−1.0%; −1.4 to −0.5%; Figure S9b), and the resulting acidic conditions may favor the formation of structurally complex and decomposition‐resistant microbial necromass, such as chitin‐rich fungal cell walls that resist degradation (Malik et al. 2018; Yeasmin et al. 2020). These microbial residues subsequently could bind to soil minerals and thereby contribute to the accumulation of MAOC (+9.9%; +8.1 to +11.7%; Figure 1c). However, the extent of this accumulation is likely influenced by soil texture and mineralogical composition (Zhou et al. 2024).

We found that the synergy between warming and eCO₂ not only offsets warming‐induced SOC loss but also increases SOC (Figure 1). Compared with the relatively large and slow‐cycling MAOC pool, the more labile and environmentally sensitive POC pool is theoretically expected to respond more readily to short‐term climate perturbations and thereby regulate SOC responses (SiMa et al. 2025; Zhou et al. 2024). However, contrary to this expectation, under combined warming and eCO₂, SOC gains were primarily reflected in increased MAOC accrual (p < 0.05) (Figure 2c). This pattern likely arose because, although eCO₂ enhanced plant‐derived C inputs, POC—lacking mineral protection—was highly susceptible to microbial decomposition. Warming intensified microbial activity (e.g., elevated extracellular enzyme production and priming effects) that accelerated POC turnover, such that decomposition outweighed the input stimulation from eCO₂, ultimately resulting in no significant POC accumulation (Figure 1d). Warming reduced soil moisture by 7.8% (−8.3% to −7.3%; Figure S9a), prompting plants to allocate more C to roots (+5.9%; +4.6% to +7.3%) to improve water uptake (Figure 1b), which, in combination with eCO₂, enhanced root‐derived C inputs and provided a sustained substrate supply that stimulated microbial growth, biosynthesis, and turnover, thereby enhancing microbial necromass C accumulation (+20.4%; +5.4% to +37.7%) and facilitating its stabilization within MAOC (+5.5%; +1.7% to +9.5%; Figure 1d). A similar effect occurred under eCO₂ alone (Figure 1c). Notably, the decomposition products of POC under warming and eCO₂ are not necessarily fully mineralized to CO₂, a portion of low‐molecular‐weight organic compounds (e.g., organic acids and amino‐containing compounds) can be assimilated by microbes and transformed into microbial necromass or be directly adsorbed onto mineral surfaces (Liu, Pold, et al. 2021; Liu, Tang, et al. 2025; SiMa et al. 2025). These processes jointly increased the MAOC to SOC ratio by 6.0% (+1.8 to +10.4%).

We then separately analyzed soil C responses in grasslands, forests, and farmlands to warming and eCO₂, as these ecosystems had relatively large sample sizes. Under warming, the largest declines in MAOC (−12.1%; −21.3 to −1.9%) and SOC (−15.7%; −16.2 to −15.2%) occurred in grasslands (Figure 3a). This pattern was likely due to grasslands' C inputs being dominated by easily decomposable roots and labile litter (Bai and Cotrufo 2022), long‐term warming accelerating N mineralization (thereby reducing plant productivity and C inputs) (Liu, Wang, et al. 2017; Wang et al. 2019), and increased microbial metabolic costs (lowering microbial C use efficiency by 15.1% and raising respiratory C losses by 6.7%) (Figure 3a). Moreover, the temperature sensitivity (Q₁₀) of soil organic matter decomposition in grasslands is higher than that in forests, indicating that forests are more resistant to warming‐induced decomposition (Liu, He, et al. 2017). This resistance reflects the low temperature sensitivity of MAOC in forest soils—particularly those with high clay content—due to the chemical protection provided by mineral surfaces (Davidson and Janssens 2006). Consistently, MAOC responses remained unchanged under warming in forest ecosystems (Figure 3a). Combined warming and eCO₂ increased MAOC (+5.6%; +1.1 to +10.2%) and SOC (+8.3%; +7.4 to +9.3%) in croplands, whereas POC, MAOC, and SOC showed no significant change in grasslands, indicating that global SOC gains (+7.6%; +6.7 to +8.5%) under combined eCO₂ and warming were largely attributable to croplands (Figures 1d and 3c). Intensive soil management in agricultural ecosystems—particularly high N fertilization that alleviates N limitation—enhanced belowground C inputs, promoted microbial assimilation and mineral association, and thereby increased MAOC and SOC accumulation (Figure 3c). These results highlight ecosystem‐specific management strategies: supplemental irrigation and fertilization may be required to enhance soil C accumulation in grasslands; in croplands, optimizing N management to boost plant C inputs and combining it with conservation practices like no‐tillage and cover cropping can further strengthen SOC accumulation.

FIGURE 3.

Responses of the soil C cycle to warming and/or eCO₂ in croplands, grasslands, and forests. (a) Warming. (b) eCO₂. (c) Warming + eCO₂. Red indicates cropland; blue indicates grassland; green indicates forest. Symbols indicate response direction: +, significant increase (p < 0.05); −, significant decrease (p < 0.05); [], no significant response (p > 0.05). CUE, microbial C use efficiency; DOC, dissolved organic C; MAOC, mineral‐associated organic C; POC, particulate organic C; SOC, soil organic C; PLFA‐derived microbial biomass; SMB, Soil microbial biomass.

3.2. Nitrogen Mediates eCO₂ Mitigation of SOC Loss Under Warming

To elucidate how SOC responses are driven under warming and/or eCO₂, we examined key variables related to soil C, including plant and microbial traits, soil properties, and soil enzyme activities (Figure 4). Partial correlation analysis indicated that, after controlling for plant biomass and enzyme activity, the response ratios of soil total N, total phosphorus (P), moisture content, and pH remained strongly correlated with SOC under warming and eCO₂ (p < 0.05; Figure 4a). Random forest analysis further identified the response ratio of soil total N as the most important predictor associated with SOC variation under warming and eCO₂ (Figure 4b). In particular, increases in soil total N responses were strongly associated with higher SOC responses (warming: R ² = 0.17, p < 0.001; eCO₂: R ² = 0.60, p < 0.001; warming + eCO₂: R ² = 0.71, p < 0.001; Figure 4c). This association may partly arise from the intrinsic stoichiometric coupling between C and N in soil organic matter, while also reflecting coordinated shifts in soil C and N cycling under warming and/or eCO₂. Notably, the offsetting effect of eCO₂ on SOC losses under warming strengthened with soil N content (Figure S10), suggesting that soil N responses are closely associated with plant–soil–microbial interactions underlying SOC stability under concurrent warming and eCO₂.

High‐latitude regions such as boreal forests and tundra are strongly N‐limited, resulting in intensified negative soil C responses under warming (Marañón‐Jiménez et al. 2019). Low soil N availability constrained vegetation productivity in cold ecosystems, prompting microorganisms to increase their metabolic activity to acquire N. This accelerated organic matter decomposition, thereby weakening the ecosystem's resistance to warming‐induced soil C loss (Bai et al. 2023; Wang et al. 2025). eCO₂ reduced the plant leaf demand for the CO₂‐fixing enzyme (RuBisCO, N‐dependent) (Smith et al. 2024), and the saved N from photosynthetic apparatus was redirected to other tissues, such as roots, which promoted root biomass growth and increased microbial biomass C (Figure 1c). However, additional N was required for the construction and maintenance of added root tissue (Kuzyakov and Xu 2013). This stimulated demand occurred against the backdrop of limited soil N, triggering nutrient competition between plants and microorganisms. Such competition limited microbial proliferation—particularly of fungi, which generally require more N and thereby counteracting the potential stimulatory effect of eCO₂ on fungal growth, maintaining a balanced fungal‐to‐bacterial (F:B) ratio (Figure 5b). Moreover, this competition suppressed microbial respiration on a per‐unit biomass basis (Hu et al. 2001). In croplands, fertilization supplied the N needed for plant growth and mitigated soil microbial N mining, thereby facilitating SOC accumulation (Tang et al. 2023; Tang et al. 2025). Together, these findings suggest that eCO₂ alters plant‐microbe interactions by favoring plant N utilization, thereby slowing microbial decomposition and enhancing ecosystem C accumulation.

FIGURE 5.

Effects of warming, eCO₂, and warming + eCO₂ on soil microbial communities. Scatter points represent individual response ratios; circles with error bars denote means ±95% confidence intervals. Numbers in parentheses indicate observation counts. F:B, fungal‐to‐bacterial biomass ratio; G+, gram‐positive bacterial biomass; G–, gram‐negative bacterial biomass; G+/G–, gram‐positive to gram‐negative bacterial biomass ratio.

3.3. Spatiotemporal Effect of Soil Carbon Sequestration Under Warming and eCO₂

Extrapolating from the identified key climatic and biogeochemical drivers of SOC responses, our analysis showed that eCO₂ had substantial potential to mitigate future warming‐induced C losses (Figure 6). The global topsoil (0–30 cm) SOC stock was currently estimated at 783.5 Pg C (2025) (Padarian et al. 2022). Given the limited temporal variability in SOC responses (Figures S11 and S12), we adopt 2100 as a stable long‐term reference to assess the centennial‐scale potential for C sequestration. At the global scale, warming reduced SOC by 5.2% (−40.7 Pg; −0.5 Pg C year⁻¹), whereas eCO₂ increased SOC by 3.2% (+25.1 Pg; +0.3 Pg C year⁻¹). Their combined effect produced a net gain of 3.5% (+27.4 Pg; +0.4 Pg C yr.⁻¹). These results demonstrate that eCO₂ not only offsets warming‐induced soil C losses but also triggers a net global C gain. Therefore, accurate projections of the terrestrial C sink under climate change require simultaneous consideration of both warming and eCO₂ responses.

By 2100, warming was projected to increase SOC in several temperate regions of the Northern Hemisphere, particularly in the Eastern European Plain, parts of northern China, and northern North America. This may be attributed to the fact that these regions have a mean annual temperature of approximately 5°C–15°C, where moderate warming can enhance photosynthetic enzyme activity and plant C assimilation (Xu et al. 2025). Additionally, in managed croplands, no‐tillage, cover cropping, and fertilization can further maintain soil C storage (Teng et al. 2024). In contrast, SOC in high‐latitude areas above 50°N was expected to decline by approximately 5.0% (−39.2 Pg C), while arid and semi‐arid zones such as central Australia might lose approximately 7.0% (−54.8 Pg C; Figure 6a). Drought may amplify warming‐induced SOC losses, potentially due to altered microbial metabolic processes and accelerated decomposition of MAOC (Guo et al. 2026). A meta‐analysis showed that the temperature sensitivity of soil CO₂ emissions increased with latitude (Carey et al. 2016). Cold regions store vast amounts of SOC; warming‐induced increases in soil respiration and permafrost thaw were likely to release long‐sequestered organic C (Crowther et al. 2016; García‐Palacios et al. 2021). The limited thermal adaptability of these ecosystems further weakened their capacity to buffer warming‐driven C emissions, thereby amplifying their role in the C‐climate feedback (Xu et al. 2025). Under eCO₂, equatorial rainforest regions—particularly the Congo Basin and Southeast Asian tropics—showed an average SOC gain of 4.0% (+31.3 Pg C). SOC increased by 0.4% (+3.1 Pg C) in mid‐ to high‐latitude regions (Figure 6b,e). By contrast, arid and semi‐arid regions showed generally small and spatially heterogeneous SOC changes under eCO₂, with localized losses in parts of Central Asia, the Middle East, southwestern North America, and Patagonia (Figure 6b). Water availability constrains plant productivity and plant‐derived C inputs, while reduced soil moisture suppresses microbial activity and limits substrate diffusion, thereby reducing microbial decomposition (Wang, Jiao, et al. 2022; Schimel 2018). Given the limited observational constraints in arid regions, the projected SOC losses and their underlying mechanisms should be interpreted cautiously, as SOC estimates in these regions are subject to relatively high uncertainties (Figure S5), while the relative contributions of constrained C inputs and altered mineralization remain uncertain and require further validation. Under combined warming and eCO₂ conditions, global SOC showed a net positive response. In the temperate regions (30°–50°N), such as northern China, eastern Europe, Central Asia, and the central and northeastern United States, SOC increased by an average of 4.5% (+35.2 Pg C), while SOC in high‐latitude regions, including Siberia, northern Alaska, and northern Canada, increased by an average of 3.3% (+25.8 Pg C) (Figure 6c). Under combined warming and eCO₂, the negative impacts of warming on SOC in mid‐ to high‐latitude regions were substantially alleviated. Across most latitudes, SOC responses to the combined treatment remained positive (Figure 6f), indicating that synergistic interactions enhanced soil C sequestration. This outcome reflects coordinated plant–soil–microbial interactions: eCO₂ increases plant C inputs and soil N availability, facilitating microbial assimilation and necromass formation, promotes the mineral protection of organic matter, accelerates MAOC accumulation, and ultimately reduces SOC vulnerability to warming.

3.4. Limitations and Uncertainties

This study has limitations related to data sources and modeling approaches. Firstly, the current dataset lacks sufficient spatial coverage, particularly missing extreme dry sites (deserts) and wet sites (tropical forests), which hinders the accurate estimation of SOC responses across ecosystem types. Moreover, since warming and eCO₂ are not independent of other factors, such as drought and N deposition, we have not yet investigated their interactions, which may lead to an overestimation of the results. Although our dataset spans 1990–2025 and covers diverse environmental conditions and experimental methods, this inclusiveness may introduce heterogeneity. Notably, warming and eCO₂ not only alter microbial biomass but also shift the abundance of key functional genes and metabolic pathways, directly influencing SOC formation, decomposition, and stabilization (Zhou et al. 2012). Future work should integrate metagenomic, metatranscriptomic, and metabolomic analyses to elucidate the regulatory mechanisms governing SOC stability. Despite these limitations, our study offers critical insights into how warming and eCO₂ regulate SOC responses via plant–soil‐microbial interactions.

4. Conclusions

Our synthesis shows that warming generally reduces net ecosystem productivity, stimulates microbial activity, and lowers microbial C use efficiency, thereby increasing respiratory C losses and accelerating mineralization of labile C fractions, ultimately depleting SOC. By contrast, eCO₂ enhances plant C inputs, stimulates microbial biomass C, and—with sufficient N availability—promotes stabilization of organic matter into MAOC. The effects of combined warming and eCO₂ on soil C responses are enhanced by synergistic interactions, primarily in croplands. By 2100, these synergistic effects are projected to override the negative impacts of warming on SOC, particularly in mid‐ to high‐latitude regions, ultimately leading to a net positive SOC response. This further underscores the critical need to consider the interactive impacts of warming and eCO₂ in future projections of SOC responses, providing valuable insights for strategies aimed at enhancing global C sinks and achieving C neutrality.

Author Contributions

Ruiqi Yao: conceptualization, formal analysis, writing – original draft, data curation. Tida Ge: methodology. Zhongkui Luo: writing – review and editing. Yike Zhang: data curation, validation. Haoran Fu: formal analysis. Wolfgang Wanek: writing – review and editing, validation. Scott X. Chang: supervision, writing – review and editing. Guopeng Liang: investigation. Mingkai Jiang: visualization. Yiqi Luo: writing – review and editing. Qingxu Ma: writing – review and editing, funding acquisition, resources, conceptualization, supervision. Yongchao Liang: formal analysis. Lianghuan Wu: funding acquisition. David R. Chadwick: writing – review and editing. Davey L. Jones: writing – review and editing. Ji Chen: writing – review and editing. Yuanhe Yang: writing – review and editing. Manuel Delgado‐Baquerizo: writing – review and editing, validation. Cesar Terrer: writing – review and editing, validation.

Funding

This work was supported by National Key Research and Development Program of China (2023YFD2302200, 2023YFD1900601), National Natural Science Foundation of China (32573140, 32172674) and Zhejiang Provincial Natural Science Foundation (LZ23C150002).

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Data S1: References in meta‐analysis.

Table S1: Search terms used in the meta‐analysis to locate warming and/or eCO₂ experimental studies published before April 2025 in Web of Science, Google Scholar, and China National Knowledge Infrastructure (CNKI).

Table S2: Summary of papers used in the meta‐analysis after duplicate removal.

Table S3: List of abbreviations used in this study.

Table S4: Summary of meta‐analysis quality assessment criteria. Quality criteria are adapted from Koricheva and Gurevitch.

Table S5: The Rosenthal fail‐safe number (Nfs) and Egger's test were used to assess publication bias for SOC.

Table S6: Variables controlled for in the partial correlation analyses.

Table S7: Variance inflation factor (VIF) values of candidate predictors for structural equation modeling (SEM).

Table S8: Performance of four machine‐learning algorithms to predict SOC changes.

Table S9: Detailed information on future environmental covariates used for global soil organic carbon (SOC) prediction.

Figure S1: PRISMA 2020 flow diagram for systematic reviews including searches of databases and registers only.

Figure S2: Distribution of experimental treatments in the meta‐analysis.

Figure S3: Funnel plots for SOC effects.

Figure S4: Leave‐one‐out sensitivity analysis using the inverse‐variance weighted ln(response ratio)method.

Figure S5: Uncertainty in predicted SOC changes under warming, eCO₂, and warming + eCO₂ by 2100.

Figure S6: Uncertainty in predicted SOC changes under warming, eCO₂, and warming + eCO₂ conditions by 2040.

Figure S7: Uncertainty in predicted SOC changes under warming, eCO₂, and warming + eCO₂ conditions by 2070.

Figure S8: Relationship between SOC and warming duration. Solid line indicates a significant linear regression (p < 0.05) with 95% confidence intervals (shaded areas).

Figure S9: Effects of warming, eCO₂, and warming + eCO₂ on soil physicochemical properties and nutrient contents.

Figure S10: Relationship between the offsetting effect of eCO₂ on warming‐induced SOC losses (%) and soil total N.

Figure S11: Global patterns of predicted relative SOC responses to warming and/or eCO₂.

Figure S12: Global patterns of predicted relative SOC responses to warming and/or eCO₂.

Data Availability Statement

The data that support the findings of this study are openly available in Figshare at https://figshare.com, reference number https://doi.org/10.6084/m9.figshare.31563121.