Decoupling of stomatal conductance, transpiration and photosynthesis in terrestrial plants under elevated temperature: a meta-analysis

Abstract

Elevated temperatures pose challenges to stomatal conductance, which regulates transpiration and photosynthesis. However, the coupling of stomatal conductance, transpiration and photosynthesis may shift with warming. Here, we synthesize evidence from a meta-analysis of 207 studies to assess leaf physiological responses to warming. On average, the responses of stomatal conductance are highly variable, exhibiting no consistent directional trend, whereas transpiration increases and photosynthesis decreases, reflecting a shift towards transpirational cooling. Stomatal conductance declines until temperatures exceed 5 °C above ambient, whereas transpiration remains stable under warming <3 °C. Beyond these two thresholds, both stomatal conductance and transpiration increase with further warming. The sensitivity of stomatal conductance, photosynthesis, and water-use efficiency to warming varies substantially among plant functional types, with distinct responses across life forms, phylogenetic groups, and photosynthetic pathways. Higher mean annual temperature amplifies the positive responses of stomatal conductance and transpiration to warming, whereas greater mean annual precipitation mitigates the warming-induced declines in photosynthesis. Elevated CO₂ exacerbates warming-induced declines in photosynthesis, while drought constrains transpirational cooling. Collectively, these findings highlight a progressive decoupling of stomatal conductance, transpiration and photosynthesis with warming, revealing complex trade-offs between plant water use, thermal regulation, and carbon assimilation.

Introduction

Stomata play a critical role in regulating transpiration (E) and photosynthesis (A_net)¹. Plants are generally assumed to regulate stomatal behavior in order to maximize carbon gain per unit water lost, thereby maintaining a strong coupling between stomatal conductance (g_s), E and A_net^2,3. This coordination, however, tends to weaken under elevated temperatures, which may compromise the balance between water loss and carbon gain^4,5, and violate predictions based on stomatal optimization assumptions. Global surface temperatures have risen rapidly in recent decades and are projected to increase by at least 1.5 °C by the end of this century, accompanied by more frequent and intense heatwaves⁶. These climatic changes challenge stomatal behavior, with profound implications for terrestrial carbon and water cycles⁷.

Elevated temperatures typically lead to an increase in vapor pressure deficit (VPD), which drives stomatal closure to minimize water loss, thereby decreasing A_net^8,9. Importantly, this response is primarily triggered by VPD rather than air temperature alone¹⁰. As temperature rises, stomata may instead open, leading to a decoupling of g_s and A_net, suggesting that leaf cooling via E becomes a more critical driver of stomatal regulation than the potential enhancement of A_net^11,12,13. Although E is often assumed to rely primarily on g_s, high temperatures can decouple E from g_s as elevated VPD increases the driving force of leaf water loss through cuticular pathways¹⁴. Such increased water loss may reflect a physiological prioritization of thermal safety over strict water conservation¹⁵, or may result from mechanical failure of the cuticle or guard cells^16,17. However, the thresholds underlying the decoupling of g_s, E and A_net under rising temperatures remain poorly understood, hindering our ability to accurately predict plant responses to future climate scenarios.

The impacts of elevated temperatures on g_s, E and A_net, as well as their coupling, may vary significantly across plant functional types. For instance, different growth forms, such as herbaceous and woody plants, exhibit distinct adaptive strategies shaped by key functional traits¹⁸, resulting in variation in thermoregulation patterns among plant growth forms¹⁹. Herbaceous species generally possess more acquisitive traits that enhance resource capture and promote rapid growth under favorable conditions, whereas woody plants, particularly trees, tend to adopt more conservative resource-use strategies²⁰. As a result of these divergent strategies, woody and herbaceous plants may exhibit different responses to warming. However, a recent meta-analysis indicates that, compared to herbs, woody plants can experience greater biomass gains in response to climate warming²¹. Beyond growth form, life form, phylogeny, leaf habit, and photosynthetic pathway further influence functional traits and water-use strategies^22,23,24. For instance, deep-rooted trees can sustain high water-use efficiency (WUE) under warming, while shallow-rooted shrubs are more vulnerable to water deficits caused by warming, owing to differences in water uptake depth and stomatal regulation²⁵. These variations underscore the role of functional traits in shaping the coupling dynamics of leaf gas exchange and highlight the diversity in plant strategies for coping with warming across different functional groups. However, relatively few studies have directly compared functional types in this regard, leaving important gaps in understanding how plants with different strategies respond to rising temperatures at the leaf level.

The extent to which g_s, A_net, and E remain coupled under elevated temperatures can be influenced by experimental protocols. The type of warming facility, such as open-top chambers (OTCs), infrared heaters, or growth chambers, can generate distinct microclimatic conditions that alter the balance between carbon gain and water loss. For instance, the lower wind speeds typically occurring within OTCs can increase boundary layer resistance, thereby affecting whole-plant transpiration²⁶. In addition, warming exacerbates surface soil water stress, driving plants to rely more heavily on deeper water sources²⁷. Warming-induced topsoil desiccation can lead to pronounced vertical decoupling of water and nutrient availability within the soil profile, which may reduce A_net and WUE because nutrients in the dried topsoil become inaccessible to plants²⁸. Consequently, rooting conditions, including soil depth and rooting volume, further shape plant physiological responses to warming. As a result, observed responses of g_s, A_net, and E to elevated temperatures may vary considerably across studies, reflecting methodological artifacts associated with experimental design. A clearer understanding of how these methodological factors interact is essential for accurately predicting plant functional responses to future warming scenarios.

In natural ecosystems, the coupling between g_s, A_net, and E under elevated temperatures can be shaped by climatic contexts. Factors, such as background temperature regimes, humidity, and soil water availability, interactively determine how stomata respond to warming. For example, both the mean and variability of xylem-specific hydraulic conductivity peak in warm, humid tropical regions and decline in colder, drier climates²⁹, potentially influencing plant water use. Photosynthesis in low-latitude ecosystems with high mean annual precipitation (MAP) shows greater sensitivity to moderate warming, due to narrower thermal tolerance ranges, compared with colder, high-latitude regions³⁰. Together, these climate-driven physiological differences generate region-specific patterns in plant carbon-water relations, further influencing the trade-offs between water conservation, carbon gain, and leaf cooling.

Furthermore, the effect of warming on A_net and E will depend on concurrent changes in atmospheric CO₂ and soil water availability, which regulate g_s and ultimately affect T_leaf^7,31. While warming increases evaporative demand because warmer air has a higher capacity to hold water vapor, elevated CO₂ generally lowers E by decreasing g_s^7,32. It remains uncertain whether the water-saving effect of elevated CO₂ can fully offset the increased evaporative demand driven by warming. Soil moisture can also modulate the impacts of warming. For instance, Reich et al.³³ demonstrate that in boreal tree seedlings, warming increases A_net in moist soils but reduces it in drier soils, where stomatal limitations prevail. Additionally, trees with adequate water supply are generally better equipped to withstand transient extreme heat events³⁴. Heat avoidance is predicted to be feasible only when soil water availability is sufficient to meet leaf water demand¹². Drought conditions exacerbate thermal damage during heatwaves, as drought-stressed plants exhibit reduced thermal safety margins and greater leaf damage compared to well-watered plants³⁵. Understanding the interactions of warming with elevated CO₂ and drought on plant water use and carbon uptake is essential for predicting vegetation responses to future climate scenarios and developing strategies to mitigate climate change impacts on ecosystems.

Previous meta-analyses have synthesized plant physiological responses to climate warming, including thermal acclimation of leaf respiration³⁶, single and combined global-change drivers on g_s³⁷, A_net³⁸ and leaf respiration^30,39,40. To the best of our knowledge, no study has yet comprehensively investigated the complex mechanisms underlying stomatal regulation and its interrelations with A_net and E under warming. Changes in g_s, A_net and E inevitably lead to variations in intrinsic WUE (WUE_i = A_net/g_s) and instantaneous WUE (WUE_t = A_net/E). These gaps make it challenging to forecast ecosystem processes in the context of future global warming and heatwaves. Here, we conducted a comprehensive meta-analysis to evaluate the effects of warming on stomatal behavior and its consequences for water loss, thermal regulation, carbon assimilation and WUE. The primary objectives of the present study were to: (1) quantify the overall effects of experimental warming on leaf physiological traits, and assess the degree of coupling or decoupling among g_s, A_net, and E under elevated temperatures; (2) evaluate the role of warming amount and duration in shaping plant physiological responses, with particular attention to thresholds at which g_s, A_net, and E decouple; (3) determine how plant functional types modulate leaf physiological responses to warming, thereby explaining cross-species variation in carbon-water trade-offs; (4) assess the influence of experimental protocols on leaf physiological responses, in order to identify potential methodological artifacts that may bias cross-study comparisons; (5) investigate how background climatic context mediates warming effects on leaf physiological responses; (6) explore the interactive effects of warming with elevated atmospheric CO₂ or altered soil water availability, to better predict plant responses under multifactor global change scenarios. Such work would provide a mechanistic foundation for refining carbon assimilation and E processes in dynamic global vegetation models and Earth system models, improving predictions of terrestrial carbon and water fluxes under future climate scenarios¹⁷.

Results

Responses to warming are highly variable and context-dependent

Overall, warming effects on g_s were highly variable, exhibiting no consistent directional trend across the full dataset (−1.4%; 95% CI: −7.1% to 4.7%; Fig. 1). Among all observations, 54.67% showed increases, 43.15% showed decreases, and 2.18% showed no significant change (Supplementary Fig. S1a), underscoring the context-dependent nature of g_s responses. In contrast, warming generally enhanced E by an average of 26.3% (95% CI: 12.1%–42.2%; Fig. 1), although individual responses were variable (61.4% were positive, 1.75% were neutral, and 36.84% were negative; Supplementary Fig. S1b). On average, A_net declined (−9.6%; 95% CI: −13.5% to −5.6%; Fig. 1), with a majority of observations showing decreases (61.74%; Supplementary Fig. S1c). This decrease in A_net contributed to the reduction in WUE_i (−11.0%; 95% CI: −16.4% to −5.2%; Fig. 1), which was observed in 57.49% of cases (Supplementary Fig. S1d). WUE_t decreased even more markedly (−30.2%; 95% CI: −37.7% to −21.8%; Fig. 1), with reductions being observed in 75.56% of the data (Supplementary Fig. S1e). T_leaf increased consistently by 19.3% (95% CI: 13.6% to 25.3%; Fig. 1), with nearly all observations (98.13%) reporting positive responses (Supplementary Fig. S1f). Meanwhile, Ψ_leaf declined (−16.7%; 95% CI: −22.0% to −11.1%; Fig. 1), with negative responses in 69.87% of observations (Supplementary Fig. S1g).

Fig. 1: Effect sizes of leaf physiological responses to experimental warming.

Numbers on the right indicate sample sizes. Effect sizes are considered significant when their 95% confidence intervals do not overlap zero. Trait: stomatal conductance (g_s), transpiration (E), net photosynthesis (A_net), intrinsic water-use efficiency (WUE_i) and instantaneous water-use efficiency (WUE_t), leaf temperature (T_leaf), and leaf water potential (Ψ_leaf). Source data are provided as a Source Data file.

Substantial heterogeneity among studies was attributable to variation in experimental design. Warming treatments ranged from 0.5 to 20 °C and employed diverse methodologies (e.g., open-top chambers, infrared heaters, greenhouses, growth chambers). These methodological differences significantly influenced observed responses and limited extrapolation to natural ecosystems. To address this, we conducted a refined analysis restricted to field-based experiments and moderate warming experiments (≤5 °C). In field-based experiments, g_s showed a significant mean decline (−7.0%; 95% CI: −12.8% to −0.8%; Supplementary Fig. S4a). In contrast, E showed no consistent directional response to warming (0.2%; 95% CI: −11.5% to 13.6%; Supplementary Fig. S4a). A_net declined (−7.4%; 95% CI: −12.7% to −1.8%; Supplementary Fig. S4a). The reductions in WUE_i (−4.0%; 95% CI: −13.4% to 6.5%) and WUE_t (−17.9%; 95% CI: −33.4% to 1.3%) were not statistically significant (Supplementary Fig. S4a). T_leaf exhibited a consistent positive response, increasing by 12.7% (95% CI: 6.6%–19.1%; Supplementary Fig. S4a), whereas Ψ_leaf decreased (−13.2%; 95% CI: −20.1% to −5.7%; Supplementary Fig. S4a). In moderate warming experiments, g_s showed no significant mean change (−4.3%; 95% CI: −10.5% to 2.3%), while the increase in E was not significant (5.5%; 95% CI: −3.9% to 15.7%; Supplementary Fig. S6a). A_net declined significantly (−4.9%; 95% CI: −8.3% to −1.3%), leading to corresponding reductions in WUE_i (−7.5%; 95% CI: −14.3% to −0.1%) and WUE_t (−15.6%; 95% CI: −23.8% to −6.6%). T_leaf (11.3%; 95% CI: 6.2%–16.6%) and Ψ_leaf (−14.7%; 95% CI: −21.4% to −7.4%) changed in the same direction as in the full dataset but with reduced magnitudes (Supplementary Fig. S6a).

Random forest models applied to the full dataset explained a moderate-to-high proportion of the variance in effect sizes (R² = 0.52–0.83; Supplementary Fig. S2a–g), with dominant predictors varying by specific trait. Plant functional type (PFT) emerged as the primary predictor for g_s (Supplementary Fig. S2a), A_net (Supplementary Fig. S2c), WUE_i (Supplementary Fig. S2d) and Ψ_leaf (Supplementary Fig. S2g), whereas warming facility/method most strongly influenced E (Supplementary Fig. S2b). Warming amount exerted the greatest control over WUE_t (Supplementary Fig. S2e) and T_leaf (Supplementary Fig. S2f). Secondary predictors also varied among traits: warming duration for g_s (Supplementary Fig. S2a) and Ψ_leaf (Supplementary Fig. S2g); PFT for E (Supplementary Fig. S2b), WUE_t (Supplementary Fig. S2e) and T_leaf (Supplementary Fig. S2f); warming amount for A_net (Supplementary Fig. S2c) and WUE_i (Supplementary Fig. S2d). Third-ranking predictors included warming amount for g_s (Supplementary Fig. S2a) and E (Supplementary Fig. S2b), warming duration for A_net (Supplementary Fig. S2c), warming time for WUE_i (Supplementary Fig. S2d) and Ψ_leaf (Supplementary Fig. S2g), and warming facility/method for WUE_t (Supplementary Fig. S2e) and T_leaf (Supplementary Fig. S2f).

Dependence on warming amount and duration

Warming amount and duration significantly influenced multiple leaf physiological traits (Fig.2; Supplementary Tables S1 and S3). Threshold-dependent responses were evident for g_s (Fig. 2a) and E (Fig. 2b). For g_s, a negative relationship occurred when plants were warmed by less than 5 °C (p = 0.0067; Fig. 2a); beyond this threshold, a significant positive correlation emerged (p = 0.0031; Fig. 2a). A similar threshold response was detected for E (Fig. 2b). Below 3 °C, warming had no significant effect on E (p = 0.3), while above it, the effect became strongly positive (p < 0.001; Fig. 2b). In contrast, A_net responses showed a negative correlation with increasing temperatures (p < 0.001; Fig. 2c). The responses of WUE_i exhibited a linearly negative response to warming amount (p < 0.001; Fig. 2d), while the responses of WUE_t showed a nonlinear response, with a steep decline above 5 °C of warming (p < 0.001; Fig. 2e). The response of T_leaf to warming amount was also best described by a segmented regression model, with a breakpoint at approximately 16 °C warming (Fig. 2f). T_leaf initially increased with warming (p < 0.001) but declined beyond this point (p = 0.0028; Fig. 2f), implying that plants might switch successfully to high transpirational cooling under extreme warming. However, data points above 16 °C all originated from a single study, and more experiments with extreme warming would be needed to evaluate the consistency of these observations. The absolute increase in T_leaf also showed a significant positive relationship with warming amount (p < 0.001; Supplementary Fig. S9). For Ψ_leaf, the response ratio remained relatively stable at low to moderate levels of warming, but declined sharply at higher warming levels (p = 0.0051; Fig. 2g).

Fig. 2: Effects of warming amount and duration on leaf physiological responses.

a–g responses as a function of warming amount; h–n responses as a function of warming duration. Points represent individual observations. Solid lines show the fitted mean response from the best-supported regression model, and shaded bands represent the 95% confidence intervals (CIs). The slope was tested using two-sided t-tests with classical degrees of freedom. Lines and CIs are only shown when the slope is statistically significant (p < 0.05). Trait abbreviations are as in Fig. 1. Source data are provided as a Source Data file.

To further assess the temperature dependence of these responses, we classified warming levels (ΔT) into three categories: low (ΔT < 3 °C), moderate (3 °C ≤ ΔT ≤ 5 °C) and high (ΔT > 5 °C; Supplementary Fig. S10 and Table S2). The impacts of the three temperature levels on E (test of moderators: Q_M = 10.2, p = 0.006) and T_leaf (Q_M = 56.74, p < 0.001) were significantly different (Supplementary Fig. S10). While low and moderate warming had little effect, high warming significantly enhanced both E and T_leaf (Supplementary Fig. S10). The effects of warming on A_net (Q_M = 7.79, p = 0.02) and WUE_t (Q_M = 62.06, p < 0.001) were also temperature-dependent, with high warming causing stronger declines than low or moderate warming (Supplementary Fig. S10). The impacts of warming on g_s (Q_M = 3.05, p = 0.22), WUE_i (Q_M = 2.55, p = 0.28) and Ψ_leaf (Q_M = 1.63, p = 0.44) were not significantly different across the three categories (Supplementary Fig. S10).

Warming duration also influenced leaf physiological responses (Fig. 2h–n). For g_s, a significant negative relationship with warming duration was detected (p = 0.02; Fig. 2h), with effect size declining sharply at shorter durations and stabilizing thereafter. Similarly, the positive effect of warming on E was strongest in short-term experiments and diminished with increasing duration, stabilizing under long-term warming (p < 0.001; Fig. 2i), whereas the responses of A_net (p = 0.014; Fig. 2j), WUE_i (p = 0.012; Fig. 2k) and WUE_t (p < 0.001; Fig. 2l) showed positive correlations with warming duration.

Significant interactive effects between warming amount and duration were detected for g_s (p = 0.05; Fig. 3a), A_net (p = 0.046; Fig. 3b), and Ψ_leaf (p = 0.006; Fig. 3c). In all three traits, high warming levels were associated with steep declines as warming duration increased, whereas low warming treatments showed relatively stable or slightly positive responses over time (Fig. 3). Intermediate warming levels exhibited moderate declines, falling between the low and high treatments. These patterns indicate that the negative impacts of warming on leaf gas exchanges were most pronounced under prolonged, high-intensity warming, while mild warming exerts minimal long-term effects.

Fig. 3: Interactive effects of warming amount and duration on leaf physiological responses.

a–c Model-predicted responses of stomatal conductance (g_s), net photosynthesis (A_net) and leaf water potential (Ψ_leaf) to warming duration at the three levels of warming amount. Lines show the fitted mean response from the linear model, including the warming amount × warming duration interaction, and shaded bands represent 95% confidence intervals around the fitted mean response. The p-values shown in each panel correspond to two-sided tests of the warming amount × warming duration interaction term in a regression model. Source data are provided as a Source Data file.

Dependence on plant functional types

Across all plant functional groups, warming effects varied substantially in magnitude and direction among traits (Fig. 4 and Supplementary Table S10). For g_s, the 95% CIs of both woody species and herbs overlapped with zero, indicating no consistent responses. The between-group difference was weak and only marginal (Q_M = 2.76, p = 0.097), with herbs tending to show a slight positive effect and woody species a negative one. Herbs and woody plants showed similar directional responses in E (Q_M = 2.08, p = 0.15), A_net (Q_M = 1.03, p = 0.31), WUE_i (Q_M = 1.32, p = 0.25), WUE_t (Q_M = 0.06, p = 0.81), T_leaf (Q_M = 0.0016, p = 0.97) and Ψ_leaf (Q_M = 0.71, p = 0.4; Fig. 4a).

Fig. 4: Effects of plant functional types (PFTs) on leaf physiological responses to experimental warming.

a–f Mean effect sizes with 95% confidence intervals (CIs) for comparisons between PFTs: herbs versus woody species, shrubs versus trees, broadleaf trees versus conifers, deciduous versus evergreen broadleaf trees, C3 herbs versus C4 herbs, and annual herbs versus perennial herbs. Points show mean effect sizes, and error bars indicate 95% CIs. Effect sizes are considered significant when their 95% CIs do not overlap zero. Numbers below indicate sample sizes. Between-group heterogeneity was assessed using the omnibus test of meta-regressions with an intercept. Significance levels are denoted as †0.05 <p < 0.1, *p < 0.05, **p < 0.01, and ***p < 0.001. Trait abbreviations are as in Fig. 1. Source data are provided as a Source Data file.

A marginally significant difference in g_s response to warming was observed between shrubs and trees (Q_M = 3.65, p = 0.056), with g_s response in shrubs varying without a consistent trend, whereas trees showed a significant decreasing trend (Fig. 4b). Warming effects on WUE_i differed significantly between the two groups (Q_M = 11.31, p < 0.001; Fig. 4b), with a significant reduction observed in shrubs but not in trees (Fig. 4b). By contrast, no significant differences were detected between shrubs and trees for E (Q_M = 0.8, p = 0.37), A_net (Q_M = 0.75, p = 0.39), WUE_t (Q_M = 2.51, p = 0.11) and Ψ_leaf (Q_M = 0.94, p = 0.33; Fig. 4b).

Broadleaf and coniferous trees exhibited significantly different responses to warming across several leaf physiological traits. In coniferous trees, warming significantly reduced g_s (Q_M = 16.09, p < 0.001), A_net (Q_M = 58.99, p < 0.001) and WUE_t (Q_M = 4.81, p = 0.028), whereas no significant changes were observed in broadleaf trees (Fig. 4c). Furthermore, the warming-induced increase in T_leaf was more pronounced in coniferous trees than in broadleaf trees (Q_M = 5.41, p = 0.02; Fig. 4c). The warming-induced decrease in Ψ_leaf was stronger in broadleaf trees than in coniferous trees (Q_M = 5.57, p = 0.018; Fig. 4c). No significant differences were found between broadleaf and coniferous trees in their responses of E (Q_M = 0.62, p = 0.43) or WUE_i (Q_M = 0.14, p = 0.71; Fig. 4c).

Within the broadleaf group, evergreen species exhibited a stronger reduction in A_net under warming compared to deciduous species, whereas deciduous species showed no significant change (Q_M = 10.85, p < 0.001; Fig. 4d). Although changes in WUE_i were not statistically significant in either group, WUE_i tended to increase in deciduous species but decrease in evergreen species, resulting in a significant difference between the two groups (Q_M = 9.97, p = 0.0016; Fig. 4d). Similarly, a warming-induced decrease in WUE_t was observed in evergreen species but not in deciduous species, and this difference was marginally significant (Q_M = 3.81, p = 0.051; Fig. 4d). Ψ_leaf decreased significantly in deciduous but not in evergreen species (Q_M = 5.79, p = 0.016; Fig. 4d and Supplementary Table S10). No significant differences between deciduous and evergreen species were found in their responses of g_s (Q_M = 0.069, p = 0.79) or E (Q_M = 0.54, p = 0.46; Fig. 4d).

Photosynthetic pathway also played a key role in shaping plant responses to warming (Fig. 4e and Table S10). In C4 herbs, g_s increased significantly under warming, while no such effect was observed in C3 herbs (Q_M = 27.27, p < 0.001; Fig. 4e). In contrast, A_net (Q_M = 118.36, p < 0.001) and WUE_i (Q_M = 14.76, p < 0.001) decreased significantly in C3 herbs but remained unaffected in C4 herbs (Fig. 4e). The photosynthetic pathway did not significantly alter the responses of E (Q_M = 0.51, p = 0.48), WUE_t (Q_M = 1.16, p = 0.28), T_leaf (Q_M = 0.42, p = 0.52) and Ψ_leaf (Q_M = 0.47, p = 0.49) to warming (Fig. 4e). Additionally, no significant differences were observed between annual and perennial herbs in most leaf traits, except for A_net, which declined under warming in perennial herbs but remained unchanged in annuals (Q_M = 17.58, p < 0.001; Fig. 4f).

Experimental factors regulating warming effects

To further investigate potential sources of variation in leaf physiological responses to warming, we assessed the effects of rooting condition, warming timing, and warming facility/method on effect sizes (Fig. 5 and Supplementary Table S14). Rooting condition had a significant influence on the warming responses of several traits (Fig. 5a). Although the warming effect on g_s was slightly negative in free-rooted systems, it was neutral to positive in potted systems, with marginally significant differences between them (Q_M = 2.96, p = 0.085; Fig. 5a). Compared with free-rooted plants, potted plants exhibited significantly higher warming-induced increases in E (Q_M = 5.13, p = 0.024; Fig. 5a). Unexpectedly, potted plants also had a greater increase in T_leaf compared with free-rooted plants, albeit only marginally significant (Q_M = 3.77, p = 0.052; Fig. 5a). Warming timing also influenced trait responses (Fig. 5b). While 24 h, nighttime, and daytime warming had no individually significant effects on g_s, their responses differed significantly among categories (Q_M = 8.83, p = 0.012; Fig. 5b): daytime warming tended to suppress g_s, whereas nighttime warming tended to enhance it. Daytime warming reduced A_net more strongly than either 24 h or nighttime warming (Q_M = 29.4, p < 0.001; Fig. 5b). For other traits, such as E (Q_M = 0.95, p = 0.62), WUE_i (Q_M = 0.53, p = 0.77) and T_leaf (Q_M = 2.34, p = 0.31), warming responses were relatively consistent across timing categories, with no significant between-group differences (Fig. 5b). Warming facility/method introduced further variability (Fig. 5c). Growth chamber experiments produced stronger warming-induced increases in E compared with OTCs and IR heaters (Q_M = 6.25, p = 0.044; Fig. 5c). Declines in WUE_i (Q_M = 16.66, p < 0.001; Fig. 5c) and WUE_t (Q_M = 11.54, p = 0.003; Fig. 5c) were not significant in heater-based experiments but were significant in growth chamber and OTC studies.

Fig. 5: Effects of experimental protocols on leaf physiological responses to warming.

a–c Mean effect sizes (%) with 95% confidence intervals (CIs) for rooting condition, warming time, and warming facility. Points show mean effect sizes, and error bars indicate 95% CIs. Effect sizes are considered significant when their 95% CIs do not overlap zero. Numbers below indicate sample sizes. Between-group heterogeneity was assessed using the omnibus test of meta-regressions with an intercept. Significance levels are denoted as †0.05 <p < 0.1, *p < 0.05, **p < 0.01, and ***p < 0.001. Trait abbreviations are as in Fig. 1. Source data are provided as a Source Data file.

Climatic factors regulating the effects of warming

To assess how background climate modulates leaf physiological responses to warming, we analyzed relationships between response ratios and MAT (Fig. 6a–g and Supplementary Table S18), MAP (Fig. 6h–n and Supplementary Table S19) and AI (Fig. 6o–u and Supplementary Table S20) using a subset of studies performed under field conditions. Along the MAT gradient, threshold-type responses were observed for several traits. For g_s, the warming effect was independent of MAT below 10.9 °C (p = 0.34), but above this threshold the warming response of g_s increased with MAT (p < 0.001; Fig. 6a). E showed a clear V-shaped response: the effect of warming decreased with MAT up to 11.3 °C (p = 0.0011) and increased at higher MAT (p = 0.0054; Fig. 6b). A similar pattern was observed for A_net; the slopes on both sides of the breakpoint (MAT = 12.5 °C) were relatively small but significant (p < 0.001; Fig. 6c). The warming effects on WUE_i declined with MAT below 18.6 °C (p = 0.0075) and showed no clear trend at higher MAT (p = 0.55; Fig. 6d). The warming effects on WUE_t also decreased with MAT at cooler sites (p < 0.001) but became non-significant above the breakpoint (18.6 °C; p = 0.84; Fig. 6e). The responses of T_leaf showed no detectable dependence on MAT (p = 0.077; Fig. 6f). The responses of Ψ_leaf peaked at 15.4 °C: the warming effect increased toward the breakpoint (p < 0.001) and then declined slightly at higher MAT (p = 0.027; Fig. 6g).

Fig. 6: Climatic controls on leaf physiological responses to warming.

a–g responses as a function of mean annual temperature (MAT); h–n responses as a function of mean annual precipitation (MAP); o–u responses as a function of aridity index (AI). Points represent individual observations. Solid lines show the fitted mean response from the best-supported regression model, and shaded bands represent the 95% confidence intervals (CIs). The slope was tested using two-sided t-tests with classical degrees of freedom. Lines and CIs are only shown when the slope is statistically significant (p < 0.05). Trait abbreviations are as in Fig. 1. Source data are provided as a Source Data file.

The warming effect on g_s was not related to MAP (p = 0.19; Fig. 6h). In contrast, the warming effect on E was negatively associated with MAP (p < 0.001; Fig. 6i), whereas the warming effects on A_net (p < 0.001; Fig. 6j) and WUE_i (p = 0.0016; Fig. 6k) were positively correlated with MAP. The warming effect on WUE_t showed no significant relationship below 354.4 mm MAP (p = 0.079) but increased significantly above this threshold (p < 0.001; Fig. 6l). T_leaf also exhibited a threshold-type response: no relationship below a breakpoint at 1413 mm (p = 0.71), but a decline in the warming effect at higher MAP (p < 0.001; Fig. 6m). The warming effect on Ψ_leaf showed no clear dependence on MAP (p = 0.14; Fig. 6n).

The warming effect on g_s remained unrelated to AI (p = 0.27; Fig. 6o). The warming effect on E showed a negative linear relationship (p < 0.001; Fig. 6p). The warming effect on A_net increased with AI up to a breakpoint at AI = 0.49 (p = 0.003) after which no significant trend was detected (p = 0.55; Fig. 6q). The warming effect on WUE_i increased with AI at the dry end (p = 0.0084) and then plateaued (p = 0.9; Fig. 6r), and the warming effect on WUE_t similarly rose at low AI (p = 0.0014) and then leveled off beyond an AI of 1.2 (p = 0.94; Fig. 6s). The warming effects on T_leaf (Fig. 6t) and Ψ_leaf (Fig. 6u) were stable at low AI (p = 0.55 and p = 0.71, respectively) but declined in wetter climates (p < 0.001 and p = 0.0025, respectively).

Combined effects of warming with elevated CO₂ and drought

To explore how other climate factors modulate leaf physiological responses to warming, we assessed the effect sizes of multiple traits under different atmospheric CO₂ concentrations and soil moisture regimes (Fig. 7 and Supplementary Table S21). g_s showed no consistent responses to warming under either aCO₂ or eCO₂. However, the magnitude of the responses differed between the two CO₂ treatments (Q_M = 5.95, p = 0.015; Fig. 7a). Although warming elevated E in all treatments, the difference between aCO₂ and eCO₂ was not significant (Q_M = 0.47, p = 0.49; Fig. 7a). Warming led to decreased A_net across all treatments, with A_net being decreased to a greater extent under eCO₂ (Q_M = 10.74, p = 0.001; Fig. 7a). Warming tended to reduce WUE_i under aCO₂, whereas the responses of WUE_i were minimal under eCO₂ (Q_M = 18.72, p < 0.001; Fig. 7a). No significant differences were found between aCO₂ and eCO₂ for the responses of WUE_t (Q_M = 1.4, p = 0.24; Fig. 7a). Warming increased T_leaf at both CO₂ levels, with a more pronounced increase observed under aCO₂ (Q_M = 5.79, p = 0.016; Fig. 7a). Warming reduced Ψ_leaf under both CO₂ levels, with a significantly larger decline under eCO₂ (Q_M = 6.55, p = 0.011; Fig. 7a).

Fig. 7: Effects of CO₂ concentration and water availability on leaf physiological responses to warming.

a ambient (aCO₂) vs elevated CO₂ (eCO₂). b Water availability: high (HW) vs low (LW). Points show mean effect sizes, and error bars indicate 95% confidence intervals (CIs). Effect sizes are considered significant when their 95% CIs do not overlap zero. Numbers below indicate sample sizes. Between-group heterogeneity was assessed using the omnibus test of meta-regressions with an intercept. Significance levels are denoted as †0.05 <p < 0.1, *p < 0.05, **p < 0.01, and ***p < 0.001. Trait abbreviations are as in Fig. 1. Source data are provided as a Source Data file.

g_s showed no consistent responses to warming under either well-watered or drought treatments. Nevertheless, differences in the magnitude of the g_s response to warming were evident, with a significant overall difference between the two watering treatments (Q_M = 5.8, p = 0.016; Fig. 7b). Warming enhanced E in both well-watered and drought-stressed plants, but the increase was significantly greater under well-watered conditions (Q_M = 22.95, p < 0.001; Fig. 7b). Warming reduced A_net in both water treatments, but no significant difference was detected between well-watered and drought groups (Q_M = 0.24, p = 0.62; Fig. 7b). Warming significantly reduced WUE_i in both water treatments, with a larger decrease under drought conditions (Q_M = 49.02, p < 0.001; Fig. 7b). The negative impacts of warming on WUE_t were comparable in both well-watered and drought-stressed plants (Q_M = 0.009, p = 0.98; Fig. 7b). Warming increased T_leaf in both water treatments, with a marginally significant group difference (Q_M = 3.6, p = 0.058; Fig. 7b), with a greater warming-induced increase in T_leaf under drought conditions. Ψ_leaf declined under both treatments, but the decrease was significantly more pronounced under well-watered conditions (Q_M = 40.61, p < 0.001; Fig. 7b).

Discussion

Warming decouples transpiration and photosynthesis from stomatal conductance

Based on a comprehensive dataset compiled from global warming experiments, we found that warming effects on g_s varied substantially among studies, with both significant increases and decreases reported, resulting in no consistent overall effect when averaged across all observations (Fig. 1a). In contrast, E did not track the responses of g_s, but instead showed a positive response to warming (Fig. 1a), suggesting that E was driven not only by g_s but also by additional factors, such as VPD⁴. Evidence for decoupling between g_s and E was also apparent in the subset analyses of field studies and of experiments with moderate warming (≤5 °C). In the subset of field experiments, g_s decreased but E showed no consistent changes (Supplementary Fig. S4a). The divergence between the full dataset and the field subset likely reflects the smaller warming magnitude in the field subset: the full dataset had mean/median ΔT of 4.57/4.00 °C, compared with 2.8/2.5 °C in the field subset; notably, E increased only when warming exceeded a 3.0 °C threshold. In experiments with moderate warming (≤5 °C), neither g_s nor E exhibited consistent overall responses, but g_s tended to decline while E tended to rise (Supplementary Fig. S6a), consistent with partial decoupling whereby increases in VPD under warming elevated E even as g_s weakly decreased. The decoupling of E from g_s reflects an adaptive cooling mechanism^31,41. For example, during an extreme heatwave, trees reduced T_leaf by an average of 2.8 °C relative to expectations through enhanced E⁴². In some broadleaf species, E-driven cooling can lower T_leaf by up to 9 °C¹². Consistent with these findings, our analysis showed that the increase in T_leaf was smaller than the rise in air temperature (Supplementary Fig. S9). However, this cooling effect came at the cost of greater water loss, as indicated by the warming-induced declines in Ψ_leaf (Fig. 1a), consistent with higher water use under warming⁴.

Such transpirational cooling helps maintain T_leaf within an optimal range, thereby sustaining A_net^4,19,41. However, our results showed that A_net was depressed at higher temperatures (Fig. 1a), consistent with previous studies demonstrating that plants in hot environments can achieve higher E even when A_net nearly ceases⁵. Notably, the funnel plot and Egger’s test indicate strong publication bias for A_net. Because our overall result shows that warming suppressed A_net, such bias likely reflects the selective reporting of studies with statistically significant declines in A_net. As a consequence, the magnitude of the negative warming effect on A_net may be inflated, and this result should be interpreted with caution. Collectively, these findings suggest that warming decouples A_net and E from g_s¹¹, as plants prioritize latent cooling over water conservation to reduce the risk of heat-induced leaf mortality and maintain the potential for future carbon gain¹⁹. This decoupling of carbon assimilation from water flux was further evidenced by pronounced reductions in WUE_i and WUE_t²⁶.

It is also noteworthy that the number of observations across traits in our dataset was uneven. While A_net was well represented, E was reported in far fewer studies, despite being commonly measured alongside A_net and g_s. This imbalance limits the robustness of conclusions regarding E. We recommend that future studies consistently report E together with A_net and g_s, and also include measurements of T_leaf and Ψ_leaf to enable more comprehensive assessments of the trade-offs among carbon uptake, water use, and thermal regulation.

Threshold responses of stomatal conductance and transpiration to warming

Stomatal regulation serves multiple functional roles, including limiting water loss during CO₂ uptake², preventing xylem cavitation under water deficits⁴³, and enabling transpirational cooling to mitigate heat stress⁴⁴. Our synthesis revealed that stomatal responses to warming ranged from decreases^45,46 to increases^12,35, reflecting the interplay of multiple mechanisms that generate diverse patterns across temperature regimes. Notably, g_s exhibited a threshold response to warming: under modest warming (<5 °C), g_s declined (Fig. 2a), consistent with findings that a 2 °C warming can significantly reduce g_s⁴⁷, likely driven by higher VPD and evaporative demand^8,9. In contrast, g_s increased when warming exceeded 5 °C, reflecting a shift toward transpirational cooling¹¹. A comparable threshold pattern was observed in experimental warming up to 42 °C, where g_s decreased with temperature but rose again at higher values, even when A_net approached zero or became negative⁴⁸. Furthermore, heatwaves are typically defined as periods of at least three or five consecutive days during which the daily maximum temperature exceeds the climatological average by more than 5 °C⁴⁹. Consistent with this definition, we found that warming beyond 5 °C sharply exacerbated the reductions in A_net and WUE_t (Supplementary Fig. S10), highlighting the detrimental impacts of heatwaves¹³, with the g_s threshold at 5 °C providing physiological support for this benchmark.

The response of E paralleled that of g_s, but with a lower warming threshold for E (3 °C versus 5 °C; Fig. 2b). This difference likely arises from water loss through the cuticle, which remains permeable even when stomata are closed⁴¹. As temperatures rise, cuticular conductance contributes increasingly to total water loss¹⁴, driven by both enhanced cuticular permeability and reduced water viscosity⁴¹. E remained stable below 3 °C warming, but increased sharply beyond this threshold (Fig. 2b), suggesting a shift from stomatal to cuticular dominance in water loss¹⁴. Reduced stomatal control and accelerated water movement through plants at higher temperatures suggest that evaporative cooling and protection of heat-sensitive metabolic processes become critical for plant survival.

In contrast, A_net declined consistently with increasing temperatures (Fig. 2c), reflecting a combination of limiting processes. At lower warming, stomatal constraints dominate¹⁰, but the influence diminishes at higher temperatures¹². Instead, non-stomatal limitations become increasingly important, including increased respiration³⁹, enhanced photorespiration³¹, Rubisco deactivation⁵⁰ and damage to electron transport capacity of photosystem II⁵¹. Collectively, these factors contribute to the sustained decline in A_net under warming conditions.

Our analysis shows that warming duration played a crucial role in shaping plant physiological responses. In short-term experiments, warming tended to increase g_s and E, while reducing A_net, WUE_i and WUE_t; however, these initial responses progressively weakened with longer warming durations (Fig. 2h–l). This pattern indicates partial acclimation of both carbon and water fluxes as warming persisted, likely driven by adjustments in biomass allocation. For instance, a previous meta-analysis reports that warming increased biomass allocation to roots, thereby enhancing access to deeper soil water²¹. Furthermore, we found significant interactions between warming magnitude and duration in shaping g_s (Fig. 3a). Under modest warming, g_s was suppressed after a short duration of warming, but tended to recover when warming lasted longer (Fig. 3a). In contrast, stronger warming initially stimulated g_s, but this effect diminished and often reversed at longer durations (Fig. 3a), consistent with high g_s triggering cavitation and hydraulic failure⁵². To prevent such failure, plants likely prioritize water conservation at the expense of carbon gain and evaporative cooling¹⁷.

Plant functional types modulate leaf physiological responses to warming

Our results showed that the warming effects on leaf physiological traits did not differ significantly between woody plants and herbs (Fig. 4a). However, notable differences appeared within woody life forms (Fig. 4b). For example, g_s of trees decreased in response to warming while shrubs showed no clear trend, allowing trees to maintain higher WUE_i (Fig. 4b). The same pattern was confirmed by the subset of field studies (Supplementary Fig. S14b). These results are consistent with recent findings that deep-rooted trees, owing to their greater access to soil water and tighter stomatal regulation, are less vulnerable to warming-induced water deficits and can maintain higher WUE. In contrast, shallow-rooted shrubs are more susceptible²⁵.

Within trees, responses further diverged between conifers and broadleaf species (Fig. 4c). Warming negatively impacted g_s in conifers but not in broadleaf species (Fig. 4c), suggesting contrasting stomatal regulation strategies. In conifers, such as Norway spruce, a down-regulation of g_s under warming conditions has been observed, reflecting their isohydric nature, characterized by tight stomatal control over E^32,46. This regulation limited declines in Ψ_leaf in conifers but incurred a greater reduction in A_net (Fig. 4c). Furthermore, T_leaf increased more in conifers than in broadleaf species (Fig. 4c), which is consistent with the higher leaf-air thermal coupling in conifers, allowing T_leaf to respond more directly to changes in air temperature⁵³. This difference is driven by variation in leaf morphology: the narrow leaves in conifers tend to be more aerodynamically coupled to the atmosphere, resulting in smaller temperature deviations from ambient air⁵⁴. In contrast, the broader leaves of deciduous species often exhibit greater temperature divergence from air temperature, influenced by g_s and the capacity for latent heat dissipation^55,56. Furthermore, A_net decreased more strongly in evergreen than in deciduous species (Fig. 4d), likely because deciduous trees display greater thermal acclimation of photosynthesis compared with evergreen trees under warming³⁰.

For herbs, we also observed functional differentiation. C3 and C4 herbs responded differently to warming (Fig. 4e). In C3 herbs, g_s showed no consistent trend, whereas in C4 herbs, g_s generally increased (Fig. 4e). However, the wider stomatal apertures in C4 herbs did not lead to higher E (Fig. 4e), likely due to differences in leaf wax composition⁵⁷ that affect cuticular permeance⁵⁸. Furthermore, warming depressed A_net in C3 herbs but not in C4 herbs (Fig. 4e). This contrast likely reflects the higher photosynthetic temperature optimum and the presence of a carbon-concentrating mechanism in C4 species, which minimizes photorespiration under warming^59,60. However, the warming-induced increase in T_leaf did not differ between C3 and C4 plants, consistent with the lack of significant differences in their leaf thermal stability¹⁹.

Variations in warming impacts across experimental protocols

Experimental protocols also influenced certain leaf physiological responses to warming (Fig. 5). For instance, g_s tended to increase in potted plants but decrease in free-rooted plants under warming, resulting in a stronger increase in E in the former (Fig. 5a). Several factors may explain these differences. First, in potted experiments, water availability is often artificially maintained despite root confinement, allowing plants to increase g_s and E without the risk of water stress. In contrast, free-rooted plants, which experience more natural soil conditions, may adopt more conservative water-use strategies to cope with potential soil drying under warming. Consistently, E responses were greatest in growth chamber studies, where plants were primarily potted (Fig. 5c). Second, the smaller soil volume in pots heats up more rapidly, increasing root-zone temperatures and potentially intensifying the need for evaporative cooling. Indeed, the increase in T_leaf in potted plants tended to be greater (Fig. 5a). However, this difference could also reflect the higher warming intensities often applied in pot-based experiments, and should therefore be interpreted with caution. Third, g_s responses in potted plants may be more sensitive to temperature due to simplified environmental conditions, whereas in free-rooted plants, stomatal regulation is influenced by multiple interacting factors (e.g., VPD, soil moisture, wind), which may buffer or override temperature-driven responses. For example, the IR heating methods used in field experiments may exacerbate such indirect effects by disproportionately increasing canopy and soil surface temperatures⁸.

Furthermore, it appears that nighttime warming tended to enhance g_s, whereas 24 h and daytime warming tended to suppress it (Fig. 5b). This contrasting pattern is likely related to differences in VPD. Daytime or continuous warming often causes a larger increase in VPD, thereby inducing stomatal closure. By contrast, VPD at night remains low even under heated treatments⁶¹. In addition, daytime warming also exerted the strongest negative effect on A_net compared with 24 h and nighttime warming, which may be explained by two mechanisms. First, relative to daytime warming, nighttime warming has been shown to exert a smaller negative influence on soil moisture⁶², thereby reducing indirect effects on A_net via soil water limitation. Second, nighttime warming may enhance respiratory carbon loss, leading to a depletion of nonstructural carbohydrates. This reduction in carbohydrate availability could relieve feedback inhibition on A_net, thereby promoting higher A_net during the following day⁶³.

Climate context modulates plant physiological responses to warming

The effects of warming on several plant traits varied with climate contexts. MAT exerted a threshold effect rather than a linear influence (Fig. 6). For instance, g_s showed no significant response to warming when MAT was below 10 °C, but increased linearly at higher MAT (Fig. 6a). This suggests that under warmer climates, plants enhance stomatal opening to facilitate evaporative cooling, thereby reducing the risk of leaf overheating and protecting photosynthetic machinery^5,64. Consistent with this, warming-induced increases in E were also evident when MAT exceeded 10 °C (Fig. 6b). Furthermore, it has been suggested that species adapted to warmer growth temperatures may benefit less from additional warming compared with species from cooler climates³⁹. Our findings partially support this hypothesis: warming stimulated A_net at both low and high MAT, but suppressed it at intermediate MAT (Fig. 6c). The positive response at high MAT is likely driven by reduced stomatal limitation, as g_s responded positively to warming at high MAT. Unexpectedly, although g_s responses to warming did not change with MAP (Fig. 6h), the responses of E to warming decreased with increasing MAP (Fig. 6i). This discrepancy may be explained by lower VPD in wetter environments⁶⁵, since VPD is an important factor influencing minimum leaf conductance and cuticular conductance¹⁴. With increasing MAP, the negative impacts of warming on A_net diminished (Fig. 6j), supporting the idea that soil moisture availability modulates leaf photosynthetic responses to warming³³.

The interactions of warming with elevated CO₂ and drought

Elevated CO₂ is well-documented to reduce g_s while improving WUE³. Based on this, we expected that warming would amplify E at elevated CO₂ due to the improved plant water status associated with higher CO₂. Contrary to our expectations, the impacts of warming on E were not different between eCO₂ and aCO₂ (Fig. 7a). A plausible explanation is that eCO₂ has only a modest effect on E under high VPD conditions^66,67. Interestingly, we found that the negative impacts of warming on Ψ_leaf were more pronounced under eCO₂ (Fig. 7a). This may be attributed to the greater leaf area produced by plants grown in eCO₂⁶⁸, resulting in increased whole-plant E. These results underscore the complex interactions between warming and CO₂ in shaping plant water use, highlighting the necessity for integrated approaches in predicting climate change impacts on vegetation water dynamics⁶⁷. It has been shown that the thermal optimum of A_net increases with eCO₂ because of the suppression of photorespiration under high measurement CO₂^69,70. Thus, warming is expected to benefit A_net under eCO₂. Unexpectedly, the negative impact of warming on A_net was greater under eCO₂ than under aCO₂. This is likely because photosynthetic acclimation to eCO₂ involves downregulation of photosynthetic capacity³, which can be exacerbated under warming.

We also found that the decoupling of g_s and E induced by warming became less pronounced under drought stress (Fig. 7b). Specifically, g_s tended to increase more with warming under drought conditions (Fig. 7b). In contrast, the warming-induced increase in E was smaller in drought-stressed plants compared to well-watered ones (Fig. 7b). This pattern suggests that drought constrained transpirational cooling despite a tendency for higher g_s, supporting the idea that heat avoidance through increased water loss may not be sustainable over extended periods in arid environments constrained by soil moisture availability⁵. These results are consistent with previous studies showing that drought reduces the capacity of plants to regulate T_leaf through transpirational cooling³⁴, resulting in a larger rise in T_leaf. For instance, it has been reported that T_leaf in poplars in dry soil were, on average, 1.1 °C lower than those of non-transpiring leaves, while in wet soil, the temperature difference reached up to 9 °C¹². Moreover, warming caused a greater decrease in WUE_i when combined with drought stress (Fig. 7b), suggesting an enhanced decoupling between g_s and A_net. This effect appears to be driven by a stronger warming-induced increase in g_s in drought-stressed plants compared to well-watered ones, without a corresponding increase in A_net. This aligns with the finding that reductions in A_net were similarly pronounced under warming alone and the combined warming and rainfall reduction treatment²⁶.

In summary, this study demonstrates that the responses of g_s, E, and A_net to warming are decoupled, nonlinear, and highly context-dependent. Rather than varying in concert, these traits diverged under warming: moderate warming typically suppressed g_s (up to 5 °C warming) but did not affect E (up to 3 °C warming), whereas higher warming stimulated both traits, highlighting the dual roles of stomatal regulation in conserving water and dissipating heat. In contrast, A_net, WUE_i and WUE_t declined progressively with temperature. These responses were further modulated by plant functional types, experimental protocols, climate contexts, and interactions with elevated CO₂ and drought. The observed decoupling between g_s, E, and A_net under moderate-to-high warming conditions challenges the theoretical foundation of stomatal optimization models and necessitates a re-evaluation of how terrestrial carbon and water cycles are represented in Earth system models. By identifying thresholds, trade-offs, and context dependencies in leaf gas-exchange responses, the present study provides a mechanistic basis for refining representations of g_s, A_net and E in dynamic global vegetation models and Earth system models, improving predictions of terrestrial carbon and water fluxes under future climate scenarios.

Methods

Data collection

We conducted a thorough search for journal articles using databases such as Web of Science (https://www.webofscience.com) and China National Knowledge Infrastructure (https://www.cnki.net). All searches were completed by May 2024. We employed various combinations of keywords including “warming” or “elevated temperature” + “leaf gas exchanges” or “stomatal conductance” or “transpiration” or “photosynthesis” or “water-use efficiency” or “leaf temperature” or “leaf water potential”. The selected studies had to meet specific criteria: (1) inclusion of a warming treatment along with a control treatment, (2) comparable baseline environmental and biological conditions between the warming and control treatments, and (3) leaf gas exchange parameters were measured during daytime and at growth temperature (Supplementary Fig. S20). Our search yielded 207 peer-reviewed articles, published from 1993 to 2024, that met these criteria (Supplementary Note 1).

Seven variables were extracted, including g_s, E, A_net, WUE_i, WUE_t, T_leaf and Ψ_leaf. For studies that reported A_net together with either g_s or E, but not WUE_i or WUE_t explicitly, WUE_i and WUE_t were calculated as WUE_i = A_net/g_s and WUE_t = A_net/E, respectively. In addition, we compiled associated environmental and experimental information, including plant species, warming amount, experimental duration, warming facility/method, warming time, rooting condition, longitude (°), latitude (°), MAP (mm yr⁻¹), and MAT (°C). Missing MAT or MAP data were obtained using the Climate Explorer (climexp.knmi.nl) based on the geographic locations. The AI, defined as the ratio of MAP to mean annual potential evapotranspiration, was extracted for each study site from CHELSA climatology raster layers based on geographic coordinates⁷¹. Lower AI values indicate drier, more arid conditions, whereas higher values represent wetter, more humid climates, respectively. Warming amount ranged from 0.2 to 20 °C. Because g_s (Fig. 2a and Supplementary Table S1) or E (Fig. 2b and Supplementary Table S1) exhibited a threshold-dependent pattern with increasing temperature, with the threshold of 5 and 3 °C, respectively, we further categorized warming amount (ΔT) into three levels: low temperature elevation (ΔT < 3°C), moderate temperature elevation (3 °C ≤ ΔT ≤ 5°C) and high temperature elevation (ΔT > 5°C). The dataset included 194 plant species, classified by growth form into herbaceous and woody plants. Woody plants were further divided by life form (trees and shrubs), phylogeny (broadleaf trees and conifers) and leaf habit (deciduous angiosperms and evergreen angiosperms). Herbaceous plants were categorized by photosynthetic pathways (C3 plants and C4 plants) and life cycle (annuals and perennials). Experimental protocols were grouped according to warming facility/method (OTCs, infrared heaters and growth chambers), warming time (daytime, nighttime or 24 h warming), and rooting condition (free-rooted and potted). Experiments conducted in greenhouses or glasshouses were classified as growth chamber studies.

The mean and standard deviation (SD) of each treatment were either taken directly from tables or extracted from figures with GetData Graph Digitizer 2.26 software. If studies reported mean values with standard errors (SE), the SD was derived using the formula:

$${{{\rm{SD}}}}={{{\rm{SE}}}} {{\surd}} n,$$

(1)

where n is the sample size.

If neither SE nor SD was reported, an average ratio of SD to the mean value for the specific trait was calculated using the available data. The SD was calculated by multiplying the mean by this average ratio⁷².

Meta-analysis

Response ratios quantify the relative change in a trait under treatment compared to the control, providing a measure of the response to warming⁷³. The control treatment was considered the baseline for the warming treatments. The natural log-transformed response ratio (lnRR) was calculated as:

$${{{\mathrm{ln}}}} {{{\mathrm{RR}}}}={{{\mathrm{ln}}}} \left({\overline{X}}_{{{{\rm{t}}}}}/{\overline{X}}_{{{{\rm{c}}}}}\right)={{{\mathrm{ln}}}}\left({\overline{X}}_{{{{\rm{t}}}}}\right) - {{{\mathrm{ln}}}}\left({\overline{X}}_{{{{\rm{c}}}}}\right),$$

(2)

where $\overline{{X}_{{{{\rm{t}}}}}}$ and $\overline{{X}_{{{{\rm{c}}}}}}$ are means of the concerned variable in the treatment and control groups, respectively. The corresponding variance ($v$) was calculated as:

$$v=\frac{{S}_{{{{\rm{t}}}}}^{2}}{{n}_{{{{\rm{t}}}}}\overline{{X}_{{{{\rm{t}}}}}^{2}}}+\frac{{S}_{{{{\rm{c}}}}}^{2}}{{n}_{{{{\rm{c}}}}}\overline{{X}_{{{{\rm{c}}}}}^{2}}},$$

(3)

where ${n}_{{{{\rm{t}}}}}$ and ${n}_{{{{\rm{c}}}}}$ were sample sizes of a concerned variable in the treatment and control groups, respectively; ${S}_{{{{\rm{t}}}}}$ and ${S}_{{{{\rm{c}}}}}$ were SDs of a concerned variable in the treatment and control groups, respectively.

To prioritize reliable and precise evidence, we weighted response ratios by the inverse of their variance; thus, the initial weight (w) was determined as:

$$w=1/v.$$

(4)

For each study, a single “mean ± SD” per trait per species under each warming magnitude and duration was treated as one observation. Consequently, the dataset contained non-independent observations, because the same species could appear multiple times within a study under different warming magnitudes and durations. We evaluated potential publication bias and data quality for the response variables using funnel plots and Egger’s test (Supplementary Fig. S21). To address non-independence, we calculated the weighted mean of log-transformed RR $(\overline{{{\mathrm{lnRR}}}})$ with the 95% confidence interval (CI) with the mixed-effects model via the metafor package in R⁷⁴. In the model, “study” was treated as a random factor to account for within-study non-independence. Effects were identified when the 95% CI did not include zero. Percentage changes in traits induced by warming were then calculated as follows:

$${{\mathrm{Percentage}}}\,{{\mathrm{change}}}(\%)=\left({e}^{\overline{{{\mathrm{ln}}}\,{{\mathrm{RR}}}}}-1\right)\times 100.$$

(5)

To assess the relative importance of factors in leaf physiological responses to warming, we used random forest (RF) regression models implemented in the R package randomForest. For each trait, the RF model included six explanatory variables: rooting condition (RC), warming facility (WF), warming time (WT), warming amount (WA), warming duration (WD), and plant functional type (PFT). For the subset of field experiments (free-rooted plants exposed to IR heaters or OTCs), MAT, MAP, and aridity index (AI) were included as additional predictors. Each model was fitted with 500 trees, and the importance of predictors was quantified using the percentage increase in mean squared error (%IncMSE) when each variable was permuted. For each fitted model, we computed the coefficient of determination (R²) and root mean squared error (RMSE) to evaluate model performance.

We performed meta-regressions to examine the effects of various moderators, including growth form, life form, phylogeny, leaf habit, photosynthetic pathway, life cycle, rooting condition, warming time, warming facility and ΔT. When assessing the general leaf physiological responses to warming, as well as the effects of plant functional types and experimental protocols on these responses, we included only observations from the control and warming treatments. Observations that received other treatments, such as eCO₂ or drought, were excluded. However, when evaluating the effects of atmospheric CO₂ concentration or water availability on leaf gas exchange responses to warming, we included only studies that combined experimental warming with eCO₂ or water treatments. We categorized CO₂ treatments as either aCO₂ or eCO₂ and classified water treatments as either drought or well-watered. We fitted separate mixed-effects meta-regressions including each moderator as a fixed effect and an intercept, and evaluated between-group heterogeneity (Q_M) using an omnibus Wald-type χ² test to determine whether factor-level coefficients differ significantly from zero. If the null hypothesis is not rejected, it implies that the factor level does not differ significantly from the intercept, indicating a non-significant moderator effect^72,74,75. These analyses were conducted using the meta-regression function within the metafor package.

We examined the relationships between leaf physiological responses and either warming amount or duration using regression models in R. For warming amount, five candidate models were tested (linear, logarithmic, exponential, quadratic, and segmented regression) because plant responses to warming are often nonlinear and may exhibit thresholds. For warming duration, only three candidate models (linear, logarithmic, quadratic) were fitted. Segmented regression was not applied to warming duration due to limited range, which provided insufficient evidence for a meaningful breakpoint. Exponential models were excluded because they produced numerical instability and biologically unrealistic predictions at long durations. For each model, we extracted the slope, adjusted R², p-value, and Akaike’s Information Criterion (AIC). Model selection was based on statistical significance (p < 0.05) and minimum AIC. We also analyzed the interaction between warming amount and duration on leaf physiological responses.

For field-based experiments, we fitted regression models using three climatic predictors: MAT, MAP, and AI, to explore the relationships between climatic factors and leaf physiological responses. For each response variable, we evaluated different sets of models depending on the predictor variable. For MAT and AI, we considered five candidate models: linear regression, logarithmic regression, exponential regression, quadratic regression, and segmented regression with one breakpoint estimated using the segmented package. For MAP, we excluded the exponential form due to fitting instability, and thus evaluated only four models: linear, logarithmic, quadratic, and segmented regression. Model performance was evaluated using the AIC, adjusted R², and p-value of the slope. For segmented regressions, we additionally extracted the estimated breakpoint and reported regression slopes for each segment.

All analyses were conducted using four datasets: the full dataset, field-based experiments, moderate warming experiments (≤5 °C), and field experiments with moderate warming. The results from field-based experiments (Supplementary Figs. S3, S4, S11, S14, S17 and Tables S4, S5, S11, S15), moderate warming experiments (Supplementary Figs. S5, S6, S12, S15, S18 and Tables S6, S7, S12, S16), and field experiments under moderate warming (Supplementary Figs. S7, S8, S13, S16, S19 and Tables S8, S9, S13, S17) were provided in the Supplementary Information.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.