Meta-analysis reveals asymmetric root and microbial phenology shifts under global change Authors: Changxing Zhao, Zihao Qian, Shuang Liang, Shaopeng Wang, M. Luke McCormack, Weile Chen Published: 2026 ABSTRACT Global environmental change is reshaping phenological patterns across ecosystems, yet belowground phenological responses remain poorly characterized. Using a flux-based framework, we derived phenological metrics from 2100 site-years of paired root and microbial respiration time series across 163 sites worldwide. Our meta-analysis shows that root and microbial phenology typically shift asymmetrically in direction and magnitude under global change, except for a warming-induced symmetric advance in season onset. Warming and nitrogen addition extended root growing season, whereas microbial season lengthened under warming and reduced precipitation but shortened following partial vegetation removal. These phenological shifts were largely decoupled from changes in respiration fluxes, indicating a distinct dimension of ecosystem responses to global change. Such phenological adjustments critically influence carbon dynamics, as warming-driven extensions of the microbial season alone could increase global heterotrophic respiration by 2.5 ± 0.8 Pg C yr−1 (mean ± SE). Our results provide flux-based empirical guidance for representing root and microbial phenology as distinct, driver-sensitive processes in Earth system models, thereby improving projections of ecosystem carbon dynamics. Introduction Phenology, the timing of recurring biological events, acts as a fundamental regulator of ecosystem functioning and species interactions1,2. Widespread shifts of aboveground phenology under climate warming and other global change factors have been well documented through long-term ground-based observations and remote sensing3–6. These shifts have profound ecological consequences; for instance, over one-third of the recent increase in Northern Hemisphere carbon uptake has been attributed to extended vegetation growing seasons7, underscoring phenology’s pivotal role in climate-carbon feedbacks. In contrast, belowground phenological responses to global change remain poorly understood, as do their impacts on the global carbon cycle8–11. Belowground phenology mainly encompasses the seasonal dynamics of roots and soil microbial communities10,12. Although roots and soil microbes are tightly coupled, their phenological responses may differ fundamentally. Root phenology is likely closely linked to carbon allocation from aboveground tissues and may also be buffered by internal reserves13,14, whereas microbial phenology may respond more directly to soil temperature, moisture, and substrate availability11,15. These contrasting constraints suggest that root and microbial phenological responses may be coupled under some drivers yet decoupled under others. Yet, explicit assessments of their phenological responses remain scarce, especially under experimental manipulations other than warming. Consequently, current Earth system models typically represent root phenology using shoot-based parameters16–18, while those that explicitly incorporate microbial processes (e.g., CLM-microbe, CASA-CNP19,20) overlook key transitions, such as dormancy and reactivation. This scarcity of empirical evidence and oversimplification of model representations limit our ability to capture the holistic dynamics of belowground processes and to predict their trajectories under changing environments11. Methodological limitations have long constrained progress in belowground phenology. Conventional approaches, such as minirhizotron imaging, effectively capture the phenology of fine-root growth21–23 but are technically demanding and difficult to scale. Complementary to these approaches, the functional phenology framework characterizes seasonal dynamics using flux-based measurements and has been widely applied to leaf and soil microbial processes24–26. Within this framework, soil autotrophic and heterotrophic respiration time series provide a unique opportunity to quantify root and microbial phenology. Importantly, respiration measurements are relatively easy to implement, which has facilitated a growing number of studies27,28 and enables large-scale synthesis using standardized phenological metrics. Cumulative flux thresholds, such as 10–90% of annual flux, are commonly applied to identify the growing-season onset and end14,24. These thresholds approximate the transition points when biological activity rises above and declines back to background levels, delineating an ecologically meaningful active period that encompasses the majority of annual respiration flux. Theoretically, shifts in the duration or timing of this belowground active period (i.e., growing season) can occur even when total respiration fluxes remain unchanged, underscoring that phenological change may represent an independent and important dimension beyond flux magnitude11. By applying flux-based phenological metrics, we can assess whether phenological responses align with or decouple from changes in respiration fluxes. Moreover, this flux-based approach enables decomposition of total respiration changes into phenological components (e.g., growing-season extension) and rate components (e.g., increased respiration rates). Analogous to recent partitioning approaches for aboveground productivity7, such belowground partitioning can facilitate estimation of the relative contributions of these components to total autotrophic and heterotrophic respiration adjustments, providing key insight into how and why belowground carbon fluxes respond to global change. Here, we conduct a meta-analysis integrating 2100 site-years of co-located root and microbial respiration time series across 163 sites spanning major terrestrial biomes (Supplementary Fig. 1). Using a flux-based framework, we estimate the onset, end, and length of the growing season (10–90% cumulative thresholds) for both roots and microbes, and evaluated the effects of four major drivers: warming (+0.2 to +5.0 °C), reduced precipitation (−25% to −100%), nitrogen (N) addition (mineral fertilizer, +3.0 to +500 g N m⁻² yr⁻¹), and partial vegetation removal (e.g., grazing, mowing, and thinning). We further compare the phenological responses between co-located roots and microbes, examine whether the direction and magnitude of these responses were coordinated with those of respiration fluxes, and quantify the contribution of shifts in growing-season length to total flux variation. We hypothesize that (1) roots and microbes exhibit distinct phenological responses because their respiratory substrate sources and limiting environmental factors may differ; (2) phenological responses are not necessarily synchronized with respiration flux adjustments, as flux changes also depend on response rates, which may be regulated by environmental or physiological factors different from those affecting phenology; and (3) respiration flux responses are jointly regulated by phenological shifts and changes in flux rates, analogous to the combined control of leaf photosynthesis by phenology and photosynthetic capacity. Overall, this study uncovers an often overlooked dimension of belowground dynamics and provides process-based insights for improving ecosystem projections under global change. Results and discussion Roots and microbes differ in their phenological responses under global change Across all paired observations, roots and microbes showed largely decoupled phenological responses to global change. Changes in their growing-season length were overall positively but only weakly correlated (R² = 0.08, degrees of freedom (df) = 504, P < 0.001). Shifts toward the same direction occurred no more often than opposite shifts (48.0 vs. 52.0%, binomial test: P = 0.3983, Fig. 1a). Even among convergent shifts, the magnitudes of root and microbial responses were not positively, but rather negatively correlated in dual-positive cases (30.0% of all observations). Moderate positive correlations emerged only in dual-negative cases, which were the least frequent scenario (18.0%). Similarly, within each global change driver, no strong positive correlations were detected across root–microbe pairs, and convergent shifts were not more frequent than divergent ones (Fig. 1b–e). Such patterns were also observed in analyses of growing-season start and end dates (Supplementary Figs. 2 and 3). Nevertheless, when roots and microbes responded in the same direction, their response magnitudes were moderately correlated under drivers other than warming (Fig. 1c–e), suggesting context-dependent coupling of their phenological responses.Fig. 1Asymmetric responses of root and microbial growing-season length to global change drivers.Response directions and magnitudes were compared across studies, either pooling all four major treatments (a) or analyzing them separately (b–e). Paired responses in opposite directions are shown in red, and those in the same direction are shown in blue. Regression lines and statistics (significance: *P < 0.05, **P < 0.01, and ***P < 0.001) are presented for all pairs combined (purple), and separately for dual-positive (green) and dual-negative (brown) cases. Slopes are derived from two-sided multivariate meta-regression models that account for study-level heterogeneity and may therefore differ from the apparent trend of individual data points. No adjustment for multiple comparisons was applied. Sample sizes and proportions (in parentheses) for each category are indicated on the plots. Source data, as well as the detailed statistical results and exact P values, are provided as a Source data file. Asymmetric responses of root and microbial growing-season length to global change drivers. Response directions and magnitudes were compared across studies, either pooling all four major treatments (a) or analyzing them separately (b–e). Paired responses in opposite directions are shown in red, and those in the same direction are shown in blue. Regression lines and statistics (significance: *P < 0.05, **P < 0.01, and ***P < 0.001) are presented for all pairs combined (purple), and separately for dual-positive (green) and dual-negative (brown) cases. Slopes are derived from two-sided multivariate meta-regression models that account for study-level heterogeneity and may therefore differ from the apparent trend of individual data points. No adjustment for multiple comparisons was applied. Sample sizes and proportions (in parentheses) for each category are indicated on the plots. Source data, as well as the detailed statistical results and exact P values, are provided as a Source data file. When phenological responses were averaged across all sites, asymmetry between roots and microbes remained frequent. Nitrogen addition advanced root onset by 2.0 days (95% CI: −3.9 to −0.1), whereas microbial onset remained largely unchanged (−0.2 days, 95% CI: −1.4 to 0.9; Fig. 2e, f), likely reflecting the direct reliance of roots on mineral N fertilizer compared with microbes, which depend more on organic N from soil organic matter14,29–31. Reduced precipitation had minimal effects on root phenology but unexpectedly extended microbial growing seasons by 5.5 days (95% CI: 1.6–9.5; Fig. 2i, j), potentially due to drought-induced increases in root exudation that stimulate microbial activity32,33. Partial vegetation removal primarily altered microbial season length, whereas root phenology remained stable (Fig. 2m, n), likely because plants’ carbon reserves buffer moderate reductions in carbon allocation8,34. These asymmetric patterns persisted across onset—end thresholds ranging from 5–95% to 30–70% of cumulative flux (Supplementary Fig. 4) and after standardizing treatment effects to comparable intensities (Supplementary Fig. 5). Similar asymmetries were also observed under other perturbations, including acid rain, precipitation addition, litter removal, and nutrient enrichment beyond nitrogen (Supplementary Fig. 6). Collectively, these findings suggest that asymmetric root–microbe phenological responses may be a general feature under anthropogenic change.Fig. 2Shifts in root and microbial growing-season phenology under global change.Left two panels (a, b, e, f, i, j, m, and n): frequency distributions of shifts (days) in start (SOS) and end (EOS) of the growing season. Main histograms highlight the central range (−8 to +8 days), with insets showing the full distributions (−50 to +50 days). Mean shifts with 95% confidence intervals are indicated. Right two panels (c, d, g, h, k, l, o, and p): microbial–root offsets in SOS and EOS (defined as microbial SOS–root SOS, or microbial EOS–root EOS) under control and treatment conditions. Solid circles denote significant offsets based on a two-sided multivariate random-effects meta-analysis model (P < 0.05). Mean changes in offsets from control to treatment are shown, with significance assessed using two-sided paired control–treatment comparisons (*P < 0.05, **P < 0.01, and ***P < 0.001). Source data, as well as the detailed statistical results and exact P values, are provided as a Source data file. Shifts in root and microbial growing-season phenology under global change. Left two panels (a, b, e, f, i, j, m, and n): frequency distributions of shifts (days) in start (SOS) and end (EOS) of the growing season. Main histograms highlight the central range (−8 to +8 days), with insets showing the full distributions (−50 to +50 days). Mean shifts with 95% confidence intervals are indicated. Right two panels (c, d, g, h, k, l, o, and p): microbial–root offsets in SOS and EOS (defined as microbial SOS–root SOS, or microbial EOS–root EOS) under control and treatment conditions. Solid circles denote significant offsets based on a two-sided multivariate random-effects meta-analysis model (P < 0.05). Mean changes in offsets from control to treatment are shown, with significance assessed using two-sided paired control–treatment comparisons (*P < 0.05, **P < 0.01, and ***P < 0.001). Source data, as well as the detailed statistical results and exact P values, are provided as a Source data file. The only exception was that warming advanced the season onset of both roots and soil microbes, with mean shifts of −4.6 days for roots (95% CI: −7.5 to −1.6) and −4.0 days for microbes (95% CI: −6.8 to −1.2; Fig. 2a, b). Warming also induced a high proportion of convergent advances in growing-season onset, with 56.4% of cases showing dual-negative responses (Supplementary Fig. 2). These patterns indicate that early-spring warming often acts as a shared environmental cue, synchronously triggering early-season activation of both roots and microbes, consistent with warming-induced advances in aboveground greening35,36. In contrast, although warming tended to delay season end, the effects were not statistically significant for either roots or soil microbes, with a tendency for a larger delay in roots than in microbes (Fig. 2a, b). This difference suggests that autumn phenology may be governed by more complex and heterogeneous constraints than spring onset. Our respiration-based findings differ from production-based metrics reported by Liu et al.14, which found no warming-induced advancement in root growing-season onset, implying that early-season root respiration is more thermally sensitive than root emergence. The observed advances of root respiration likely reflect metabolic reactivation of overwintering roots37 and the energetic demands of new tissue formation25. Importantly, phenological dates derived from cumulative flux thresholds mostly exhibited the expected seasonal behavior, with positive slopes at season onset (89.5% of cases) and negative slopes at season end (91.1% of cases). Moreover, treatment-induced shifts were broadly consistent when phenology was alternatively defined using a widely used, conceptually distinct amplitude-based approach that identifies seasonal transitions based on the earliest and latest 10% of seasonal flux amplitudes, providing an independent check on effect sizes (Supplementary Fig. 7). This concordance indicates that our conclusions are robust to phenological definitions and supports the reliability and biological relevance of flux-based phenology for characterizing belowground seasonal dynamics. Our synthesis of warming-induced phenological shifts partially aligns with a recent meta-analysis using similar cumulative flux thresholds (Wang et al.)15. The advance in microbial onset we estimated (−4.0 days) was slightly smaller than their reported value (−5.8 days). For roots, we detected a statistically significant advance (−4.6 days), whereas Wang et al. reported a similar magnitude that did not reach significance (−5.0 days). These discrepancies likely reflect differences in dataset composition (Supplementary Table 5), underscoring the importance of broad empirical coverage and standardized inclusion criteria to robustly quantify belowground phenological responses under global change. The often-asymmetric phenological shifts in roots and microbes altered the temporal alignment of their activity, either narrowing or widening the gap between their respective growing seasons. Under baseline conditions, soil microbial growing seasons initiated 3.5 days earlier and terminated 3.8 days later than those of roots, resulting in an intrinsically longer microbial season by 7.2 days (Supplementary Fig. 8). Warming, nitrogen addition, and partial vegetation removal consistently contracted these microbial–root offsets in both season onset and end, thereby synchronizing the seasonal timing and reducing the difference in growing season length. In contrast, reduced precipitation exhibited the opposite effect, exacerbating the phenological mismatching and increasing the length differences between root and microbial activity (Fig. 2c, d, g, h, k, l, o, p). Collectively, these asymmetric responses reshape the temporal niches and functional relationships between root and microbial processes. Environmental and experimental conditions drive spatial variation in belowground phenological shifts Local vegetation type and environmental conditions strongly influence belowground phenological shifts. For instance, small-statured vegetation generally exhibited greater phenological responses than tall-statured vegetation (Supplementary Fig. 9). We further found that the dominant drivers of phenological responses differed between roots and microbes and varied among global change factors. Under warming, for instance, root phenological shifts were most strongly associated with soil physical properties, such as the volumetric fraction of coarse fragments (Fig. 3a), which may buffer soil temperature dynamics. In contrast, microbial phenology was more strongly linked to soil carbon-to-nitrogen ratios (Fig. 3a), suggesting greater sensitivity to substrate quality. These contrasts highlight distinct physiological sensitivities that shape root and microbial phenology across drivers, while also underscoring the need for additional empirical data to resolve context dependency at local scales.Fig. 3Key drivers of shifts in growing-season length across sites.a–d SHapley Additive exPlanations (SHAP) values from machine learning models, showing the seven most influential predictors of growing-season changes. Mean absolute SHAP values (left of each panel) indicate predictor importance, and corresponding R² values show model explanatory power. e–h SHAP dependence plots illustrating relationships between the top predictors and shifts in growing-season length, with LOESS-smoothed trends. Shaded bands indicate the 95% confidence intervals. Abbreviations of predictors: coarse (volumetric fraction of coarse fragments), BD (bulk density), TN (total nitrogen), SOC (soil organic carbon), soil fungi abundance (fungal proportion of fungal + bacterial pool), BBC (bacterial biomass carbon), FBC (fungal biomass carbon), F:B ratio, soil biodiversity (integrated diversity of microbial and macrofaunal species), bio1 (annual mean temperature), bio3 (isothermality), bio9 (mean temperature of driest quarter), bio10 (mean temperature of warmest quarter), bio15 (precipitation seasonality), bio16 (precipitation of wettest quarter), bio17 (precipitation of driest quarter), bio19 (precipitation of coldest quarter), GPP (gross primary productivity), Srad (solar radiation), duration (treatment duration), frequency (respiration measurement frequency), intensity (treatment level or deviation from ambient). Source data are provided as a Source data file. Key drivers of shifts in growing-season length across sites. a–d SHapley Additive exPlanations (SHAP) values from machine learning models, showing the seven most influential predictors of growing-season changes. Mean absolute SHAP values (left of each panel) indicate predictor importance, and corresponding R² values show model explanatory power. e–h SHAP dependence plots illustrating relationships between the top predictors and shifts in growing-season length, with LOESS-smoothed trends. Shaded bands indicate the 95% confidence intervals. Abbreviations of predictors: coarse (volumetric fraction of coarse fragments), BD (bulk density), TN (total nitrogen), SOC (soil organic carbon), soil fungi abundance (fungal proportion of fungal + bacterial pool), BBC (bacterial biomass carbon), FBC (fungal biomass carbon), F:B ratio, soil biodiversity (integrated diversity of microbial and macrofaunal species), bio1 (annual mean temperature), bio3 (isothermality), bio9 (mean temperature of driest quarter), bio10 (mean temperature of warmest quarter), bio15 (precipitation seasonality), bio16 (precipitation of wettest quarter), bio17 (precipitation of driest quarter), bio19 (precipitation of coldest quarter), GPP (gross primary productivity), Srad (solar radiation), duration (treatment duration), frequency (respiration measurement frequency), intensity (treatment level or deviation from ambient). Source data are provided as a Source data file. Further, soil microbial factors often influenced root phenology, while microbial phenology was often shaped by plant carbon fixation drivers (e.g., gross primary productivity, solar radiation, and soil physicochemical properties, Fig. 3a–d and Supplementary Fig. 10), indicating reciprocal plant–microbe feedbacks in seasonal timing. Notably, the most influential local factors often exhibited nonlinear relationships with growing-season length shifts (Fig. 3e–h). These shifts in growing season length were achieved through changes in the onset, the end, or both (Supplementary Fig. 11). Based on the fitted model, we extrapolated the global patterns of root and microbial growing-season shifts under the four major global change drivers (Supplementary Fig. 12). The resulting maps provide a foundation for incorporating belowground phenology and their response patterns under global change into Earth system models. Experimental conditions further modulated belowground phenological responses, indicating that accurate projections of belowground phenology require context-dependent parameterization accounting for both environmental and experimental heterogeneity. Factors such as treatment duration, intensity, and measurement frequency played important roles (Fig. 3a–d). Under stressed conditions, including reduced precipitation or partial vegetation removal, longer treatment periods could either attenuate or amplify shifts (Fig. 3g, h, and Supplementary Fig. 13), reflecting a dynamic balance between response accumulation and acclimation. We note that a subset of studies used relatively coarse temporal resolution or incomplete seasonal coverage (see “Methods”), which may introduce uncertainty in identifying precise phenological transition dates. Future studies would benefit from year-round, high-temporal-resolution measurements to fully capture belowground phenological dynamics, especially given evidence that belowground activity can extend beyond aboveground-defined growing seasons38. Shifts in growing season length do not mirror flux responses Compared with the frequently quantified changes in belowground respiration fluxes, phenological shifts represent a critical yet often overlooked dimension of ecosystem responses to global change. We found that changes in growing-season length were only weakly correlated with changes in respiration flux (roots: R2 < 0.01, df = 516, P < 0.001; microbes: R2 = 0.03, df = 516, P < 0.001). In addition, the frequency of convergent responses in flux and phenology did not differ significantly from 50% for roots (binomial test: P = 0.7584) or microbes (binomial test: P = 0.2720) (Fig. 4a, b), indicating no general tendency toward coordinated shifts. Even among convergent cases, the magnitudes of change were usually weakly associated or uncorrelated (Fig. 4a, b). Indeed, phenological shifts and respiration flux responses were regulated by largely distinct sets of drivers. Soil-related factors were more strongly associated with phenological responses (Fig. 3), whereas respiration flux responses were more evenly associated with climatic and soil variables (Supplementary Fig. 14). This divergence in dominant associations provides a plausible explanation for the weak coupling between phenological shifts and respiration flux responses. Collectively, these results indicate that increases or decreases in respiration flux do not necessarily correspond to longer or shorter growing seasons, highlighting the need to jointly consider temporal and physiological dimensions when evaluating belowground responses to environmental change.Fig. 4Responses of phenology and respiration flux to global change.a, b Treatment effects on growing-season length and respiration flux. Paired responses in opposite directions are shown in red, and those in the same direction are in blue. Regression lines and statistics (significance: *P < 0.05, **P < 0.01, ***P < 0.001) are shown for all responses (purple) and separately for dual-positive (green) and dual-negative (brown) categories. Slopes are derived from two-sided multivariate meta-regression models that account for study-level heterogeneity and may therefore differ from the apparent trend of individual data points. Sample size and proportion (in parentheses) for each category are indicated. c, d Distributions of treatment effects on growing-season length and respiration flux. Solid circles indicate statistically significant effects (P < 0.05), with sample size (n, number of site-years) reported for each treatment. Error bars indicate 95% confidence intervals. Source data, as well as the detailed statistical results and exact P values, are provided as a Source data file. Responses of phenology and respiration flux to global change. a, b Treatment effects on growing-season length and respiration flux. Paired responses in opposite directions are shown in red, and those in the same direction are in blue. Regression lines and statistics (significance: *P < 0.05, **P < 0.01, ***P < 0.001) are shown for all responses (purple) and separately for dual-positive (green) and dual-negative (brown) categories. Slopes are derived from two-sided multivariate meta-regression models that account for study-level heterogeneity and may therefore differ from the apparent trend of individual data points. Sample size and proportion (in parentheses) for each category are indicated. c, d Distributions of treatment effects on growing-season length and respiration flux. Solid circles indicate statistically significant effects (P < 0.05), with sample size (n, number of site-years) reported for each treatment. Error bars indicate 95% confidence intervals. Source data, as well as the detailed statistical results and exact P values, are provided as a Source data file. Beyond this decoupling, phenological responses were generally smaller in both relative magnitude and variance across studies than respiration flux responses (Fig. 4c, d). The low variability likely reflects the smoothing effect of cumulative threshold-based metrics39. Yet, the relatively large consistency in response direction indicates intrinsic ecological significance rather than methodological artifact, suggesting that respiration phenology represents a more conservative dimension of belowground adjustment than respiration flux. In contrast, the relatively high variability in respiration flux responses has been documented previously27,40,41, likely due to the frequent influence of secondary processes. For instance, warming can enhance nitrogen mineralization, reducing the need for fine-root growth, thereby dampening root respiration responses despite increased mass-specific metabolic rates42,43. Shifts in belowground growing-season length shape respiration flux Using flux-based metrics, we quantified the relative contributions of phenological shifts (i.e., the extension or shortening of the growing season) and metabolic adjustments (i.e., enhanced or reduced respiration rates) to respiration flux responses under global change (Fig. 5). The contributions of rate adjustments were often counterbalanced by those of phenological shifts (e.g., changes of root respiration flux under warming, Fig. 5a). Consequently, an apparently stable annual or growing-season respiration flux under global change does not necessarily indicate stable activity; it may instead reflect higher rates occurring over a shorter season, or lower rates sustained over an extended season.Fig. 5Respiration flux changes driven by phenological shifts.Bar plots show the relative contributions of phenological shifts (changes in growing-season length) and respiration rate changes to total respiration flux variations under warming (a), nitrogen addition (b), reduced precipitation (c) and partial vegetation removal (d), with percentages above each bar indicating total flux change. Global maps depict simulated changes in root and microbial respiration fluxes attributable to phenological shifts. Positive contributions (>0) indicate that growing-season length and respiration flux changed in the same direction, whereas negative contributions (<0) indicate opposite changes. Values above each map denote the mean contribution and the proportion of grid cells with concordant versus opposing directions of change. Values below each panel indicate globally integrated carbon fluxes resulting from phenological shifts and their standard errors. Source data are provided as a Source data file. Respiration flux changes driven by phenological shifts. Bar plots show the relative contributions of phenological shifts (changes in growing-season length) and respiration rate changes to total respiration flux variations under warming (a), nitrogen addition (b), reduced precipitation (c) and partial vegetation removal (d), with percentages above each bar indicating total flux change. Global maps depict simulated changes in root and microbial respiration fluxes attributable to phenological shifts. Positive contributions (>0) indicate that growing-season length and respiration flux changed in the same direction, whereas negative contributions (<0) indicate opposite changes. Values above each map denote the mean contribution and the proportion of grid cells with concordant versus opposing directions of change. Values below each panel indicate globally integrated carbon fluxes resulting from phenological shifts and their standard errors. Source data are provided as a Source data file. We further extrapolated the phenology-derived flux adjustments from experimental sites to global terrestrial biomes using random forest models. Among the four major treatments, warming induced the largest phenology-mediated changes in both autotrophic and heterotrophic respiration (Fig. 5). Notably, warming extended the microbial growing season across nearly all terrestrial regions (Supplementary Fig. 12). Assuming no compensatory adjustments in substrate inputs or microbial efficiency, this pervasive seasonal prolongation alone, without accounting for the counteracting effects of rate adjustments, corresponds to an estimated global increase of 2.5 ± 0.8 Pg C yr⁻¹ (mean ± SE) in heterotrophic respiration (Fig. 5a), potentially offsetting up to 80% of the contemporary terrestrial carbon sink44. However, this estimate should not be directly interpreted as a change in soil carbon storage, as variations in carbon inputs (e.g., litterfall, root exudation) were not examined here. Such a provisional estimate, based on currently available experimental evidence, should be interpreted as a first-order approximation intended to constrain the potential magnitude of phenological effects within the global carbon cycle. Uncertainty remains substantial due to the sparse and uneven geographic distribution of existing studies, particularly in key biomes, such as the tropics (Supplementary Fig. 15). Despite these limitations, presenting such initial estimates provides a necessary quantitative reference point and helps motivate future empirical efforts aimed at refining and validating these projections. Implications for Earth system modeling and future research Our meta-analysis provides comprehensive evidence of widespread and often asymmetric phenological responses between co-located roots and soil microbes under global change. For both roots and microbes, their phenological shifts may not mirror changes in respiration flux adjustment, but could fundamentally alter autotrophic and heterotrophic respiration fluxes, with quantifiable and substantial implications for the terrestrial carbon balance. Based on these findings, we highlight two priorities for future research and model development. First, advancing flux-based phenology in global observation efforts will require full-season, high-frequency measurements of soil respiration that explicitly partition autotrophic (root) and heterotrophic (microbial) components. Developing and standardizing such flux-based phenological metrics would facilitate cross-biome comparisons and improve the upscaling of local observations to global models. Additional empirical studies, especially in currently underrepresented regions, are also needed to strengthen the evidence base underlying these estimates. Moving beyond annual mean fluxes toward high-frequency seasonal observations is a necessary step for translating these first-order estimates into robust, predictive tools for Earth system science. Second, our results demonstrate that seasonal carbon flux responses cannot be fully captured by rate-based formulations alone, highlighting the need for explicit representations of phenological timing in models. Next-generation Earth system models would benefit from moving beyond single soil- or aboveground-proxy phenology toward explicit representations of root and microbial phenology. By deriving phenological metrics directly from root and microbial respiration time series, we provide indicators that are directly aligned with model state variables and thus readily applicable for model parameterization and calibration. Such representations should allow growing-season length to vary independently from respiration rates and permit distinct phenological sensitivities to different global change drivers, reflecting the decoupled responses observed here. Incorporating these features would improve model fidelity in simulating seasonal carbon fluxes and help reduce uncertainty in ecosystem projections under future climate scenarios. Methods Data compilation We performed a systematic literature search spanning 1990–2024 across Web of Science, Google Scholar, and the China National Knowledge Infrastructure, following PRISMA guidelines (Supplementary Fig. 16). Full search terms are provided in Supplementary Table 1. During title and abstract screening, we excluded reviews, meta-analyses, and studies outside terrestrial ecosystems. Full-text screening retained studies that satisfied all of the following criteria: (1) in situ manipulative experiments testing at least one global change factor on soil respiration; (2) partitioning of autotrophic and heterotrophic respiration via trenching or girdling; (3) regular respiration measurements (typically monthly or more frequent; Supplementary Fig. 17); and (4) partitioning of autotrophic and heterotrophic components of soil respiration via trenching (including deep-collar and other in situ root exclusion methods) or girdling (including clipping for small-statured vegetation), with at least one component reported. Studies reporting only one component (either autotrophic or heterotrophic respiration) were included only for estimating the overall effect size but excluded from comparisons of root and microbial phenological responses. Studies reporting total respiration but using simulation approaches to estimate autotrophic or heterotrophic respiration were excluded. In total, 190 peer-reviewed studies from 163 sites were included (Supplementary Fig. 1 and Data 1), yielding 2100 site-year respiration records. Theses and dissertations were excluded to ensure all data were extensively peer-reviewed. Major experimental treatments included warming (+0.2 to +5 °C; 29 studies), reduced precipitation (−25 to −100%; 30 studies), nitrogen addition (inorganic, 3–500 g N m⁻2 yr−1; 41 studies), and partial vegetation removal (e.g., grazing, mowing, understory removal, thinning; 41 studies). Additional treatments encompassed phosphorus addition (10 studies), organic fertilizer (7 studies), inorganic compound fertilizer (14 studies), precipitation addition (15 studies), acid rain (9 studies), wildfire (15 studies), and litter removal (18 studies) (Supplementary Fig. 6). Multi-level treatments (e.g., high/low nitrogen levels) were treated as independent observations if they met the inclusion criteria. To avoid confounding effects, data from multi-factorial experiments (e.g., warming × N addition) were excluded from single-factor analyses. Furthermore, each multi-factor combination was not analyzed as a separate category due to insufficient sample sizes (n < 5) for robust statistical analysis. Because management practices strongly influence phenology, agroecosystems (croplands) were also excluded from the phenological analysis. Site-specific climate, soil, vegetation, and experimental metadata were extracted directly from publications or supplemented using public databases (Supplementary Table 2)45–52. Phenological metrics extraction Seasonal respiration data were extracted from text, tables, or figures (digitized using WebPlotDigitizer, www.arohatgi.info/WebPlotDigitizer), yielding 22,071 microbial and 18,243 root respiration observations, including 18,099 matched pairs. For each seasonal respiration time series, cumulative annual respiration was calculated, and the start (SOS) and end (EOS) of the growing season were defined as the dates when cumulative respiration reached 10% and 90% of the annual total, respectively (Supplementary Fig. 18). The 10–90% thresholds are widely used to capture transitions from dormancy to sustained biological activity (SOS) and to the onset of dormancy (EOS), enabling consistent and comparable estimation across diverse biomes and experimental designs14,24,53,54. These phenological metrics were calculated using treatment-level mean seasonal curves because plot-level time series were not consistently available across studies. Growing-season length (LOS) was calculated as the difference between EOS and SOS. The resulting dataset comprised 983 SOS, 965 EOS, and 913 LOS estimates for roots, and 999 SOS, 979 EOS, and 920 LOS estimates for microbes. Phenological shifts were calculated as treatment–control differences in SOS, EOS, and LOS. Studies with unrealistically late SOS or early EOS were excluded (Supplementary Table 3). We evaluated the statistical consistency and biological validity of these phenological metrics. First, to evaluate whether our results were sensitive to the choice of cumulative flux thresholds, we repeated the meta-analyses using alternative threshold combinations for SOS (5–30%) and EOS (70–95%) at 5% increments. Second, we assessed whether the slopes at the calculated SOS and EOS were consistent with the expected directions of these metrics to verify that the identified phenological dates were embedded within coherent seasonal transitions. Seasonal flux curves were smoothed using a generalized additive model55 (GAM; mgcv package), with the maximum degrees of freedom limited to nine. Sites in which the local flux derivative at SOS or EOS did not match the expected sign (~10%) were not excluded, provided that the identified SOS and EOS were mostly located within the expected rising or falling portion of the broader seasonal trajectory when evaluated over an extended temporal window (from the beginning of measurement toward the seasonal peak, or from the peak to the end of measurements). To further validate the phenological estimates, the timing of SOS and EOS was compared with an alternative common method, in which SOS and EOS were defined as the first and last days when flux rates reached 10% of the seasonal amplitude. We compared both the absolute timing of SOS and EOS and their treatment-induced shifts between two approaches (Supplementary Fig. 7). Further, although SOS and EOS are cyclical events, we did not use circular statistics because these metrics were determined separately for each calendar year26,56. For Southern Hemisphere sites, day-of-year values were shifted by ±182 days to align seasonal timing with corresponding Northern Hemisphere seasonal phases. This treatment prevented seasonal trajectories from crossing calendar boundaries, yielding phenological metrics mathematically equivalent to those obtained with a circular-statistics approach57,58. Moreover, based on a subset of data spanning calendar boundaries, we found that continuous approaches over multiple years (see Supplementary Note 1) had only minor effects on the estimated phenological shifts. Given the limited availability of datasets suitable for such continuous calculations, we report results derived from the calendar-year method. Meta-analysis of phenological shifts The mean effect sizes of phenological shifts within each treatment were weighted by sample size (number of replicates)59:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\rm{W}}}_{{\mbox{r}}}\,=\frac{{{\rm{N}}}_{{\mbox{c}}}\,\times {{\rm{N}}}_{{\mbox{t}}}}{{{\rm{N}}}_{{\mbox{c}}}\,+{{\rm{N}}}_{{\mbox{t}}}}$$\end{document}Wr=Nc×NtNc+Ntwhere Wr is the weight, and Nc and Nt are the number of replicates in control and treatment groups. To minimize pseudo-replication, treatment intensity (e.g., fertilizer dose) and duration (years of manipulation) were modeled as nested random effects within study identity60: (~1|study id) + (~1|study id: treatment intensity) + (~1|study id: treatment durations). For partial vegetation removal (lacking intensity metrics), only study identity and duration were included. Weighted mean effect sizes and 95% confidence intervals were calculated using the rma.mv function in the metafor R package. Statistical significance was inferred when 95% confidence interval excluded zero. Publication bias was assessed using Egger’s regression, Rosenberg’s fail-safe number, and the trim-and-fill test (Supplementary Table 4). While some previous studies standardized responses by treatment intensity, our tests showed that phenological responses did not scale linearly with treatment magnitude. Moreover, normalizing by treatment intensity did not change our main conclusions (Supplementary Fig. 5). We therefore retained the unstandardized effect sizes in the main analysis for ecological interpretability, while providing the standardized results for reference. Quantifying phenology-driven respiration flux adjustments We applied a partitioning framework modified from Fang et al.61 to separate the contributions of phenological shifts and mean respiration rates to total respiration flux changes. Growing-season total flux in control (Fc) and treatment (Ft) plots was defined as the product of growing season length (L) and mean respiration rate (R). The relative change in F is expressed as:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{{\rm{F}}}_{{\mbox{t}}}}{{{\rm{F}}}_{{\mbox{c}}}}\,=\,\frac{{{\rm{L}}}_{{\mbox{t}}}}{{{\rm{L}}}_{{\mbox{c}}}}\times \frac{{{\rm{R}}}_{{\mbox{t}}}}{{{\rm{R}}}_{{\mbox{c}}}}$$\end{document}FtFc=LtLc×RtRc Taking natural logarithms gives:3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{ln}}\left(\frac{{F}_{{{\rm{t}}}}}{{F}_{{{\rm{c}}}}}\right)={\mathrm{ln}}\left(\frac{{L}_{{{\rm{t}}}}}{{L}_{{{\rm{c}}}}}\right)+\,{\mathrm{ln}}\left(\frac{{R}_{{{\rm{t}}}}}{{R}_{{{\rm{c}}}}}\right)$$\end{document}lnFtFc=lnLtLc+lnRtRc\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{Defining}}\,f={\mathrm{ln}}\left(\frac{{F}_{{{\rm{t}}}}}{{F}_{{{\rm{c}}}}}\right),{{\rm{l}}}={\mathrm{ln}}\left(\frac{{L}_{{{\rm{t}}}}}{{L}_{{{\rm{c}}}}}\right),{\mathrm{and}}\; r={\mathrm{ln}}\left(\frac{{R}_{{{\rm{t}}}}}{{R}_{{{\rm{c}}}}}\right),$$\end{document}Definingf=lnFtFc,l=lnLtLc,andr=lnRtRc,4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$100 \%=\frac{{\rm{l}}}{{\rm{f}}}\,+\,\frac{{\rm{r}}}{{\rm{f}}}$$\end{document}100%=lf+rf Here, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{l}{f}$$\end{document}lf is the relative phenological contribution (%) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{r}{f}$$\end{document}rf the relative respiration rate contribution (%) to flux change. The phenology-driven flux change (Δfluxp) in each study was calculated as:5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta {{\mathrm{flux}}}_{{{\rm{p}}}}=\Delta {\mathrm{flux}}\times \frac{l}{f}$$\end{document}Δfluxp=Δflux×lfwhere Δflux denotes the treatment–control differences in root or microbial respiration flux. Global variation in phenological shifts and their flux contributions We analyzed the effects of geospatial, climatic, edaphic, vegetation, and experimental factors on phenological shifts, as well as the phenological-derived flux change (Δfluxp) using machine learning models. From an initial pool of 42 predictors (Supplementary Table 2), a two-stage feature selection procedure was implemented to reduce overfitting. First, the Boruta algorithm62 (500 iterations) classified variables as “confirmed,” “tentative,” or “rejected,” retaining only confirmed features. Second, among retained variables, highly correlated pairs (|r| > 0.75) were pruned by removing the variable less correlated with the response. Stacked ensemble models predicting phenological responses were trained using the h2o.automl framework63. Model robustness and generalization performance were evaluated using five-fold cross-validation, repeated five times with different random seeds. Model performance based on cross-validation was summarized using the mean and standard deviation of the coefficient of determination (R²), root mean square error, and mean absolute error (Supplementary Table 6). This approach provides a robust assessment of predictive performance under limited sample sizes. Model and variable interpretability were assessed using SHAP (SHapley Additive exPlanations) values64,65. Site-level predictions were extrapolated globally at a 0.25° × 0.25° resolution using the ensemble AutoML framework and global environmental datasets (Supplementary Table 2). For extrapolation, we retained the eight most influential environmental predictors identified from cross-validated models. Global predictions were calculated as the ensemble mean of the top-performing model from each of five independent AutoML runs. Globally integrated carbon flux changes associated with phenological shifts were obtained by summing grid-level estimates across terrestrial regions. Uncertainty in global estimates was quantified as the standard deviations across the ensemble of model runs. All analyses were conducted in R v4.4.1 (https://www.r-project.org/). Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Description of Additional Supplementary Information Supplementary Data 1 Reporting Summary Transparent Peer Review File Supplementary Information Description of Additional Supplementary Information Supplementary Data 1 Reporting Summary Transparent Peer Review File Source data Source Data Source Data