Evidence for widespread thermal acclimation of canopy photosynthesis

Abstract

Plants acclimate to temperature by adjusting their photosynthetic capacity over weeks to months. However, most evidence for photosynthetic acclimation derives from leaf-scale experiments. Here we address the scarcity of evidence for canopy-scale photosynthetic acclimation by examining the correlation between maximum photosynthetic rates (A_max,2,000) and growth temperature ($\overline{T_{\mathrm{air}}}$) across a range of concurrent temperatures and canopy foliage quantity, using data from >200 eddy covariance sites. We detect widespread thermal acclimation of canopy-scale photosynthesis, demonstrated by enhanced A_max,2,000 under higher $\overline{T_{\mathrm{air}}}$, across flux sites with adequate water availability. A 14-day period is identified as the most relevant timescale for acclimation across all sites, with a range of 12–25 days for different plant functional types. The mean apparent thermal acclimation rate across all ecosystems is 0.41 (−0.38–1.04 for 5th–95th percentile range) µmol m⁻² s⁻¹ °C⁻¹, with croplands showing the largest acclimation rates and grasslands the lowest. Incorporating an optimality-based prediction of leaf photosynthetic capacities into a biochemical photosynthesis model is shown to improve the representation of thermal acclimation. Our results underscore the critical need for enhanced understanding and modelling of canopy-scale photosynthetic capacity to accurately predict plant responses to warmer growing seasons.

Analysis of the FLUXNET2015 dataset provides observational evidence for widespread thermal acclimation of canopy-scale photosynthesis and its timescales across diverse biomes, improving its representation in land surface models.

Main

The carbon uptake capacity of terrestrial ecosystem photosynthesis shows large spatiotemporal variation¹. Air temperature (T_air) is one of the key factors determining this variation². Given recent warming of 0.1–0.3 °C per decade³, a better understanding of ecosystem responses to T_air is needed. While the instantaneous temperature dependence of photosynthesis has been a major focus of research^4–6 and is represented in vegetation and land surface models^7–9, the slower process known as thermal acclimation, through which plants maintain or enhance their photosynthetic efficiency in response to warmer growth temperatures^10–14, is less well understood^15,16. Several studies have indicated that leaves acclimate to thermal growing conditions within weeks to months, although the relevant timescales for different plant types remain uncertain^17–20. The potential mechanisms of this (non-genetic) acclimation include changes in key biochemical parameters (electron-transport potential and carboxylation capacity)^12,14,21, the sensitivity of stomatal conductance to atmospheric vapour pressure deficit (VPD)^22–24 and enzymatic heat tolerance^10,14.

Widespread evidence of thermal acclimation at the leaf and canopy scales indicates that the optimal temperature (T_opt) of photosynthesis adjusts in accordance with the prevailing T_air averaged over the time frame most relevant for acclimation ($\overline{T_{\mathrm{air}}}$)^{12,14,21,25,26}. Yet the extent to which the maximum carbon assimilation rate under high light (A_max) acclimates to $\overline{T_{\mathrm{air}}}$ under natural conditions is less clear, particularly since most experiments are conducted on seedlings under highly controlled growth conditions^13,27. Given that T_opt is well-documented to increase with rising $\overline{T_{\mathrm{air}}}$, it is crucial to understand whether A_max can also acclimate to $\overline{T_{\mathrm{air}}}$, since only their simultaneous enhancement can lead to consistent increases in photosynthesis^28,29. While some process-based photosynthetic models have incorporated T_opt acclimation, A_max acclimation has not been adequately represented in models^30,31. Demonstrating the presence of thermal acclimation at the canopy scale, quantifying its relevant timescales and rates across ecosystems and assessing the accuracy of photosynthetic models in representing these acclimation processes are essential for understanding how thermal acclimation can mitigate the potentially detrimental effects of warming on the future terrestrial carbon sink¹⁶.

In this study, we define evidence for thermal acclimation of canopy photosynthesis as a positive adjustment in canopy-scale A_max in response to elevated $\overline{T_{\mathrm{air}}}$. Following the definition used in leaf-scale studies³², canopy-scale A_max is defined as the photosynthetic assimilation rate of the canopy measured under high light, ample water and ambient CO₂. We derive A_max from light response curves of half-hourly or hourly eddy covariance carbon fluxes obtained from >200 FLUXNET2015 flux sites (Methods). While canopy-scale T_opt has been shown to acclimate to elevated $\overline{T_{\mathrm{air}}}$ in several previous studies^25,26,33, our focus here is solely on thermal acclimation of canopy-scale A_max. To facilitate consistent analysis across different light conditions, we standardize A_max to photosynthetic photon flux density (PPFD) equivalent to 2,000 μmol m⁻² s⁻¹ (denoted as A_max,2,000; Methods). Given the limited number of A_max,2,000 samples for individual flux sites, we infer the thermal acclimation of A_max,2,000 across spatial gradients by leveraging the large range of climates sampled by the FLUXNET2015 sites. We examine the correlation between A_max,2,000 and $\overline{T_{\mathrm{air}}}$ when averaged over different time windows to identify the most relevant timescale (τ) for thermal acclimation, as indicated by peak correlation. Finally, we evaluate a biochemical model of canopy-scale C₃ photosynthesis^4,31, incorporating recent advances in parameterizing temperature dependence acclimation¹² and modelled optimality-based leaf photosynthetic capacity³⁴, to assess its ability to reproduce the observed thermal acclimation rates.

Results and discussion

Evidence for thermal acclimation of canopy photosynthesis

By binning T_air and the fraction of absorbed photosynthetically active radiation (fAPAR) to control for the confounding effects of concurrent temperature and seasonal changes in canopy foliage quantity and the development of the photosynthetic system on A_max,2,000, our analysis reveals a pervasive positive correlation between A_max,2,000 and $\overline{T_{\mathrm{air}}}$ (see Methods for the derivations of τ for each plant functional type (PFT)) under conditions of adequate water availability as indicated by a high ratio of actual to potential evapotranspiration (ET/PET) (Fig. 1). This correlation is observed both spatially across multiple sites (Fig. 1a) and temporally within individual sites (Fig. 1c). We use linear mixed-effect models (LMMs) to obtain the regression coefficients of $\overline{T_{\mathrm{air}}}$ when estimating A_max,2,000 (A_max,2,000 ∼ $\overline{T_{\mathrm{air}}}$ + (1∣site)), which we define as the apparent thermal acclimation rate (γ_T, μmol CO₂ m⁻² s⁻¹ °C⁻¹). The concept of apparent rates is used here as the A_max,2,000 response rate to $\overline{T_{\mathrm{air}}}$ may be influenced by other covarying environmental conditions¹⁹, including the growth PPFD ($\overline{\mathrm{PPFD}}$) and VPD³⁵ (Supplementary Fig. 1). To account for the potential impact of adaptation¹²—the modification of A_max,2,000–$\overline{T_{\mathrm{air}}}$ relationships across different species and populations within a species growing at different sites—sites are treated as random intercepts within the LLMs (see Extended Data Fig. 1a for an example). Cropland sites are included in the PFT-based analyses but excluded from cross-site analyses.

Fig. 1. Relationships between A_max,2,000 and $\overline{T_{\mathbf{air}}}$.

a, γ_T values over fAPAR and T_air bins across flux sites. Black dots indicate significant (two-sided, P < 0.05) correlations between A_max,2,000 and $\overline{T_{\mathrm{air}}}$ in the LMM (A_max,2,000 ≈ $\overline{T_{\mathrm{air}}}$ + (1∣site)). b, PFT-specific γ_T values. PFTs are arranged in descending order on the basis of their mean γ_T values. In the box plots, the central lines represent the median γ_T values, the upper and lower box limits represent the 75th and 25th percentiles, and the upper and lower whiskers extend to 1.5 times the interquartile range, respectively. Letters represent statistically significant differences in the average γ_T values as determined by Tukey’s honestly significant difference test (two-sided, P < 0.05), which adjusts for multiple comparisons. The numbers in parentheses represent the sample size for each PFT. c, Partial correlation coefficients (partial r) between A_max,2,000 and $\overline{T_{\mathrm{air}}}$, when controlling for $\overline{\mathrm{PPFD}}$, T_air and fAPAR, across individual longer-term (>5 yr) flux sites. Colours in b and c indicate different PFTs, including CRO, DBF, EBF, ENF, GRA, MF, WET and all natural biomes combined (ALL).

Extended Data Fig. 1. An example of the response of A_max,2000 to $\overline{T_{\mathit{air}}}$, T_air, and fAPAR under a fAPAR-T_air bin pair.

Cross-site A_max,2000, $\overline{T_{air}}$, T_air, and fAPAR samples are collected when 0.56 < fAPAR ≤ 0.58 and 15 °C < T_air ≤ 16 °C. a–c, Relationships between A_max,2000 and $\overline{T_{air}}$ (a), A_max,2000 and T_air (b), and A_max,2000 and fAPAR (c). The black lines represent the best fits between A_max,2000 and $\overline{T_{air}}$ as a linear mixed-effect function (A_max,2000 ∼ $\overline{T_{air}}$ + (1∣Site), two-sided test, P < 0.001) (a), A_max,2000 and T_air as a linear function (A_max,2000 ∼ T_air, two-sided test, P > 0.05) (b), and A_max,2000 and fAPAR as a linear function (A_max,2000 ∼ fAPAR, two-sided test, P > 0.05) (c).

Detectability of thermal acclimation in canopy photosynthesis is quantified as the percentage of T_air–fAPAR bins showing a positive γ_T. Our cross-site analysis for natural ecosystems finds positive γ_T values in 87% of the T_air–fAPAR bins (938 in total) (Fig. 1a), with 65% of these positive relationships being statistically significant (P < 0.05), indicating that thermal acclimation is widespread across biomes. Averaged over all T_air–fAPAR bins, γ_T is 0.41 ± 0.62 (mean ± s.d.) μmol CO₂ m⁻² s⁻¹ °C⁻¹, with a 5th to 95th percentile range of −0.38–1.04 μmol CO₂ m⁻² s⁻¹ °C⁻¹. The average of positive γ_T values is 0.57 ± 0.30 m⁻² s⁻¹ °C⁻¹. The PFT-based analysis also shows strong evidence of thermal acclimation, with mean γ_T values decreasing as follows: croplands (CRO, 0.81) > deciduous broadleaf forests (DBF, 0.58) > wetlands (WET, 0.57) > evergreen needle-leaf forests (ENF, 0.54) > mixed forests (MF, 0.42) > evergreen broadleaf forests (EBF, 0.39) > grasslands (GRA, 0.34) (Fig. 1b and Extended Data Fig. 3). Furthermore, 92% of FLUXNET2015 sites with observations spanning 6 years or more show positive partial correlations between A_max,2,000 and $\overline{T_{\mathrm{air}}}$ after controlling for potential confounding factors of $\overline{\mathrm{PPFD}}$, T_air and fAPAR (Fig. 1c), indicating widespread acclimation to seasonal temperature variations at individual flux sites. Sites showing a negative correlation are mainly located in the tropics (Extended Data Fig. 4a).

Extended Data Fig. 3. The PFT-specific thermal acclimation rates (${\gamma }_{\mathrm{T}}$).

a–g, PFT-specific ${\gamma }_{\mathrm{T}}$ for croplands (CRO) (a), deciduous broadleaf forests (DBF) (b), evergreen broadleaf forests (EBF) (c), evergreen needle-leaf forests (ENF) (d), grasslands (GRA) (e), mixed forests (MF) (f), wetlands (WET) (g). Numbers (%) in parentheses represent the detectability of positive ${\gamma }_T$ values, which is defined as the percentage of the number of bins displaying a positive ${\gamma }_{\mathrm{T}}$ over the total number of bins. Black dots indicate significant (P < 0.05) correlations between A_max,2000 and $\overline{T_{air}}$.

Extended Data Fig. 4. Analyses of the partial correlation coefficients between A_max,2000 and $\overline{T_{\mathit{air}}}$ derived from long-term flux sites and their relationships with the site-level average $T_{\mathit{air}}$ and variability of $T_{\mathit{air}}$.

a, Geographic distribution of partial correlation coefficients between A_max,2000 and $\overline{T_{air}}$ controlling for $\stackrel{\circledR }{\mathit{PPFD}}$, fAPAR and T_air across sites with observations spanning over five years. b, Relationship between partial correlation coefficients and the site-level averages of $\overline{T_{air}}$. c, Relationship between partial correlation coefficients and the site-level standard deviation of $\overline{T_{air}}$. The black lines in b and c represent the predicted mean values from linear regression models, and the grey shaded areas indicate their 95% confidence intervals. P-values are determined through two-sided Pearson’s correlation significance tests. The “Forest” biome category includes evergreen needle-leaf forests, deciduous broadleaf forests, and mixed forests. Other PFTs are croplands (CRO), evergreen broadleaf forests (EBF), grasslands (GRA), and wetlands (WET).

The potential confounding effect of factors other than $\overline{T_{\mathrm{air}}}$ on A_max,2,000 appears to be minimal as the detectability of thermal acclimation remains high across diverse conditions. The binning approach has proved effective in previous studies for analysing relationships between variables of interest while controlling for confounding factors^35–37. The effects of concurrent T_air and seasonal changes in fAPAR on A_max,2,000 under T_air–fAPAR bin pairs are shown to be very weak (Extended Data Fig. 1b,c). To ensure our findings are not skewed by light acclimation³⁵, we consider the detectability of thermal acclimation when incorporating $\overline{\mathrm{PPFD}}$ into LLMs (89%; Extended Data Fig. 2a) and controlling for $\overline{\mathrm{PPFD}}$ through partial correlation (85%; Extended Data Fig. 2b). The impact of VPD is probably limited, as its negative effect on A_max has been accounted for during the derivation of A_max (equation (3) in Methods) and has been further mitigated by ET/PET filtering. After filtering, there is a positive relationship between A_max,2,000 and VPD (Supplementary Fig. 1c). Any negative VPD impact on A_max,2,000 is expected to reinforce, not diminish, the observed widespread thermal acclimation. Diffuse radiation is expected to increase A_max by penetrating into deep canopy layers where light is limited^38,39. However, this effect does not confound the relationship between A_max,2,000 and $\overline{T_{\mathrm{air}}}$ (Supplementary Fig. 2) since the conditions of diffuse radiation on the days of A_max measurements do not necessarily show a strong positive correlation with $\overline{T_{\mathrm{air}}}$ (Supplementary Table 1). Additionally, our findings remain robust with respect to the metric choice; detectability is 88% when A_max is unstandardized to a specific PPFD level and 87% when PFTs are treated as random effects within LLMs (Extended Data Fig. 2c,d).

Extended Data Fig. 2. The relationships between canopy photosynthetic capacities and $\overline{T_{\mathit{air}}}$ over fAPAR and T_air bins.

a, The partial effect of $\overline{T_{air}}$ on A_max,2000 when $\stackrel{\circledR }{\mathit{PPFD}}$ is also incorporated in the modelling (A_max,2000 ∼ $\overline{T_{air}}$ + $\stackrel{\circledR }{\mathit{PPFD}}$ + (1∣Site)). b, Partial correlation coefficients (partial r) between A_max,2000 and $\overline{T_{air}}$ when controlling for $\stackrel{\circledR }{\mathit{PPFD}}$ (A_max,2000 ∼ $\overline{T_{air}}$∣$\stackrel{\circledR }{\mathit{PPFD}}$). c, The cross-site thermal acclimation rate (${\gamma }_{\mathrm{T}}$) is calculated based on A_max (A_max ∼ $\overline{T_{air}}$ + (1∣Site)). d, The cross-site ${\gamma }_{\mathrm{T}}$ is calculated using plant function types (PFTs) as random intercepts (A_max,2000 ∼ $\overline{T_{air}}$ + (1∣PFT)). Numbers (%) in parentheses represent the detectability of positive ${\gamma }_T$ values, which is defined as the percentage of the number of bins displaying a positive ${\gamma }_{\mathrm{T}}$ over the total number of bins. Black dots indicate significant (P < 0.05) correlations.

Thermal acclimation capability can be influenced by the level and variability of $\overline{T_{\mathrm{air}}}$, as well as by species and PFTs^27,40–42. We observe negative effects of $\overline{T_{\mathrm{air}}}$ on A_max,2,000 when fAPAR falls below 0.7 and T_air exceeds 25 °C (Fig. 1a). Limited transpiration, due to a low amount of leaves, may not cool the canopy sufficiently under elevated T_air, making ribulose-1,5-bisphosphate (RuBP) regeneration a limiting process for canopy photosynthesis at high canopy temperature¹³. The reduction in A_max,2,000 with $\overline{T_{\mathrm{air}}}$ may be attributed to reduced stomatal conductance under high VPD²³ (Supplementary Fig. 3f) and/or decreased maximum quantum yield of photosystem II in response to elevated temperature^5,34,43. Additionally, under these conditions, the range of $\overline{T_{\mathrm{air}}}$ (the difference between the 90th and 10th percentiles; 3.8 °C) is significantly narrower than among the rest (8.4 °C) (two-tailed t-test, P < 0.01) (Supplementary Fig. 3b). Our site-level analyses also show that the correlation between A_max,2,000 and $\overline{T_{\mathrm{air}}}$ is positively associated with $\overline{T_{\mathrm{air}}}$ variability and negatively with $\overline{T_{\mathrm{air}}}$ (Extended Data Fig. 4b,c), which aligns with previous studies indicating that plants grown under low $\overline{T_{\mathrm{air}}}$ variability and/or high $\overline{T_{\mathrm{air}}}$ show reduced acclimation potential^27,40,44. Conversely, leaf-scale experiments indicate that the acclimation rates of light-saturated net assimilation rates (A_net) under different measurement temperatures are similar⁴¹, suggesting a limited impact of T_air on A_max,2,000. Moreover, EBF is the dominant PFT for the bin pairs with high T_air (Supplementary Fig. 4b). There is some evidence that tropical evergreen forests have a limited capability for physiological acclimation because these forests are adapted to relatively stable thermal conditions and/or thrive under high $\overline{T_{\mathrm{air}}}$ that is beyond the range limit for acclimation^33,45,46. The under-representation of EBF in the FLUXNET2015 database⁴⁷ may also lead to uncertainties in the estimation of γ_T for this biome.

The observed widespread thermal acclimation of A_max,2,000 (Fig. 1) contrasts with the varying sign of the response of leaf A_net to $\overline{T_{\mathrm{air}}}$, which can be positive, negative or neutral^{27,40,41,48,49}. This discrepancy may stem from the fact that, unlike A_max, A_net is not necessarily measured under ample water conditions^27,32 and water stress is known to affect the capacities of plant thermal acclimation²². In water-limited situations, plants typically reduce water loss through transpiration by decreasing stomatal conductance⁵⁰, resulting in decreased A_net.

Timescale of thermal acclimation of canopy photosynthesis

The timescale for canopy photosynthetic acclimation, as measured by the correlation coefficient (r) between A_max,2,000 and $\overline{T_{\mathrm{air}}}$ over different periods within concurrent T_air and fAPAR bins, varies across PFTs (Fig. 2 and Supplementary Fig. 5), increasing from GRA (12 d) to CRO (16 d), ENF (20 d), DBF (21 d) and finally WET (25 d). The τ value obtained across all sites is 14 d (Fig. 2f). For EBF, an optimal τ cannot be determined using A_max,2,000, even over an extended period of 180 d (Supplementary Fig. 5a). The enhanced vegetation index (EVI) that is derived from reflectance data in the near-infrared, red and blue spectral bands can characterize canopy structure, which closely relates with the canopy photosynthetic capacity⁵¹. We use a τ value of 13 d for EBF as identified by remote-sensing EVI for subsequent analysis (Methods and Supplementary Fig. 5b).

Fig. 2. Timescales for thermal acclimation of canopy photosynthesis.

a–f, The timescale for CRO (a), DBF (b), ENF (c), GRA (d), WET (e) and ALL (f). The x axes represent the number of days over which T_air is averaged to derive $\overline{T_{\mathrm{air}}}$. The y axes represent the 5-day moving average of positive Pearson correlation coefficients (r) between A_max,2,000 and $\overline{T_{\mathrm{air}}}$ over fAPAR and T_air bins. The τ value is the length of time frame for which r peaks.

Our estimate of an average of 14 d as τ for thermal acclimation of canopy photosynthesis falls within the range of leaf-scale τ, which varies from days to months depending on species and growth conditions^10,18,20,52. Studies that identify τ for photosynthetic acclimation using observational data across a spectrum of time frames are rare. A modelling study reports that a 15 day timescale for acclimation optimally predicts hourly eddy covariance flux measurements⁵³. It is important to note that A_max,2,000 can show positive correlations with $\overline{T_{\mathrm{air}}}$ over both the optimal τ value and other time frames close to the optimal, due to the potentially high correlation among $\overline{T_{\mathrm{air}}}$ calculated over different short-term periods.

The timescale τ for photosynthetic acclimation to a changing environment reflects a trade-off between potential benefits (for example, carbon assimilation) and costs (for example, resource re-allocation)⁴⁸. A rapid adjustment in photosynthetic capacities is expected to enhance photosynthetic performance but is accompanied by higher costs in energy and resources¹⁵. The shorter τ observed in GRA and CRO are in line with the expectation that fast-growing plants with a high generation rate of new leaves might show shorter τ than slow-growing species due to their greater physiological plasticity⁵⁴. Conversely, we found larger τ values in forests and WET, indicating that these ecosystems require more time for acclimation; however, this longer acclimation period is potentially compensated for by a higher acclimation rate (Fig. 1b). The PFT-specific and cross-site τ values for the canopy photosynthetic capacity provide a credible basis for explicitly incorporating the timescale of thermal acclimation into vegetation and land surface models.

Representing acclimation in photosynthesis models

We further explore the representation of A_max,2,000 thermal acclimation in a biochemical model for C₃ canopy photosynthesis incorporated in the Breathing Earth System Simulator (BESS)⁵⁵, based on the Farquhar–von Caemmerer–Berry (FvCB) model⁴ (Methods). We test three alternative approaches, each under different resource-use allocation assumptions, to estimate maximum carboxylation rates (V_cmax, μmol m⁻² s⁻¹) standardized to 25 °C ($V_{\mathrm{cmax}}^{25\mathrm{C}}$). These approaches are: (1) assuming a temporally constant and PFT-specific $V_{\mathrm{cmax}}^{25\mathrm{C}}$ ($V_{\mathrm{cmax}\_\mathrm{PFT}}^{25\mathrm{C}}$), where plants do not actively regulate $V_{\mathrm{cmax}}^{25\mathrm{C}}$ through the growing seasons; (2) scaling leaf $V_{\mathrm{cmax}}^{25\mathrm{C}}$ by canopy phenology, as indicated by leaf area index (LAI) (LAI-scaled $V_{\mathrm{cmax}}^{25\mathrm{C}}$, $V_{\mathrm{cmax}\_\mathrm{LAI}}^{25\mathrm{C}}$); and (3) modelling acclimation to prevailing environments based on the eco-evolutionary optimality (EEO) theory^34,56 ($V_{\mathrm{cmax}\_\mathrm{EEO}}^{25\mathrm{C}}$) (Methods and Supplementary Texts 1 and 2). The FvCB model as applied here incorporates recent advances in parameterizing the temperature dependence of leaf photosynthetic capacities to represent T_opt acclimation¹² (Supplementary Text 1). We run the model using the site-level forcings from the FLUXNET2015 database and derive A_max,2,000 by setting PPFD equivalent to 2,000 μmol m⁻² s⁻¹. Canopy temperature is a key uncertainty in modelling canopy-scale photosynthesis^30,57. We evaluate model performance using three temperature approximations, including T_air, aerodynamic surface temperature and radiometric surface temperature⁵⁸. We finally use T_air to represent canopy temperature because it has comparable performance to the other two approximations and greater data availability (Supplementary Text 1 and Supplementary Fig. 8). For further analysis, we select estimated A_max,2,000 values from 65 C₃ sites excluding CRO and water-limited sites, where all three model variants show acceptable accuracy in estimating A_max,2,000 (coefficient of determination (R²) > 0.5) (Supplementary Table 2).

The BESS model variant incorporating optimality-based $V_{\mathrm{cmax}\_\mathrm{EEO}}^{25\mathrm{C}}$ more closely approximates the observed γ_T compared to the other two variants, $V_{\mathrm{cmax}\_\mathrm{PFT}}^{25\mathrm{C}}$ (BESS_PFT) and $V_{\mathrm{cmax}\_\mathrm{LAI}}^{25\mathrm{C}}$ (BESS_LAI) (Fig. 3). The Kolmogorov–Smirnov (K–S) test indicates that the cumulative distribution functions of γ_T between BESS_EEO and FLUXNET2015 observations are more closely aligned, despite significant differences between all three BESS model distributions and observations (P < 0.05) (Fig. 3b). BESS_PFT and BESS_LAI underestimate the median observed γ_T by 65% and 50%, respectively, while BESS_EEO overestimates it by 34% (Fig. 3a).

Fig. 3. Impact of leaf photosynthetic capacities on γ_T estimation.

a, Probability densities of γ_T values derived from FLUXNET2015 and three variants of the BESS model (BESS_PFT, BESS_LAI and BESS_EEO). The vertical lines represent the median γ_T values. b, The statistics of the two-sided K–S tests between FLUXNET2015 observations and three model variants.

The considerable underestimation of γ_T by BESS_PFT and BESS_LAI highlights the limitation in process-based photosynthetic models that incorporate only T_opt acclimation. To capture γ_T accurately, process-based models must also integrate seasonal variations in photosynthetic capacities resulting from thermal acclimation. The overestimation by BESS_EEO can be attributed to its higher predicted detectability (99%) of thermal acclimation than observed (92%) (Fig. 3a). When calculating $V_{\mathrm{cmax}\_\mathrm{EEO}}^{25\mathrm{C}}$, we assume that plants are not water-stressed following ET/PET filtering; a water-stress factor is not applied to scale $V_{\mathrm{cmax}}^{25\mathrm{C}}$ as described in ref. ⁴³ (Supplementary Text 2). Consequently, in this study, the EEO theory represents an idealized condition where carbon assimilation is optimized under the assumption of sufficient water availability. While plant light use efficiency can be reduced by physiological stress due to water scarcity⁵⁹, the absence of such water-stress constraints can lead to an overestimation of $V_{\mathrm{cmax}}^{25\mathrm{C}}$. Although ET/PET is an effective indicator of soil moisture, it may not fully correspond to plant physiological stress. Bridging the gap between existing water availability metrics and actual plant stress responses remains a challenge⁶⁰.

Conclusion

Photosynthesis can benefit from future warming through thermal acclimation, resulting in increased carbon uptake under conditions where water is not limiting. While leaf-scale acclimation is widely recognized, our study shows that the positive acclimation of canopy-scale photosynthetic capacity to growth temperature is a widespread phenomenon across various terrestrial biomes. We have shown that, on average, the canopy photosynthetic capacity acclimates to the growth thermal conditions of the preceding 14 days. Incorporating seasonal acclimation of photosynthetic capacities (the maximum carboxylation rate and the maximum electron-transport rate) is critical for achieving accurate simulations of photosynthesis in response to variations in temperature at timescales of weeks to months. Despite warmer growing seasons, water availability is increasingly constrained in many regions, potentially forcing plants to reduce photosynthetic capacity as a water conservation strategy. Improving the understanding of canopy-scale photosynthetic thermal acclimation in response to future conditions characterized by warming and variable water availability is therefore important.

Methods

Global database of ecosystem-scale carbon fluxes

We derive A_max from >200 eddy covariance sites from the global database FLUXNET2015, which covers a wide range of geospatial locations and PFTs^47,61 (Supplementary Table 2). FLUXNET2015 is an openly accessible database containing data on the net exchange of carbon (NEE), water and energy between the atmosphere and the biosphere and meteorological observations. Uniform processing approaches are implemented for the flux calculation and quality control across the sites⁴⁷. We use half-hourly or hourly NEE (NEE_VUT_USTAR50), its corresponding estimation of the uncertainty caused by friction velocity filtering (NEE_VUT_USTAR50_ RANDUNC) and gap-filled meteorological observations, including incoming radiation (SW_IN_F), air temperature (TA_F) and VPD (VPD_F) to derive A_max (refs. ^47,62) (described below). Sites are excluded if data are unavailable during the MODIS period from 2002 onwards (for example, US-LWW and US-Me4) or if the uncertainty estimation is missing (for example, CA-Man).

Derivation of ecosystem-scale A_max

We derive A_max from light response curves across the FLUXNET2015 sites according to the daytime flux partitioning methods detailed in refs. ³⁵,⁶³. We fit NEE using the following hyperbolic equation:

$$-\mathrm{NEE}=\frac{\alpha \beta R_{\mathrm{g}}}{\alpha R_{\mathrm{g}}+\beta }+\gamma$$ 1

where β (μmol CO₂ m⁻² s⁻¹) is the target variable of interest. Variables α, R_g and γ represent the ecosystem-scale quantum yield (μmol C J⁻¹), global radiation (W m⁻²) and ecosystem respiration (μmol CO₂ m⁻² s⁻¹), respectively.

To account for the potential influence of high VPD (hPa), β is scaled using an exponential function only when VPD exceeds 10 hPa. Thus, we obtain A_max as follows:

$$A_{\max }=\left\{\begin{array}{c}\hfill \beta ,\mathrm{VPD}\le 10\mathrm{hPa}\hfill \\ \hfill \beta \exp \left(-k\left(\mathrm{VPD}-10\right)\right),\mathrm{VPD}>10\mathrm{hPa}\hfill \end{array}\right)$$ 2

where β and k are fit parameters to the flux data. The ecosystem respiration term in equation (1), γ, is estimated using an Arrhenius-type function describing the temperature dependence of γ (ref. ⁶⁴), which is applied to night-time data by assuming that night-time NEE is equivalent to ecosystem respiration:

$$\mathrm{NEE}=R_{\mathrm{ref}}\exp \left\{E_0\left(\frac{1}{T_{\mathrm{ref}}-T_0}-\frac{1}{T_{\mathrm{air}}-T_0}\right)\right\}$$ 3

where R_ref and E₀ are the basal respiration rate (μmol CO₂ m⁻² s⁻¹) at a reference temperature (T_ref = 15 °C) and temperature sensitivity (°C), respectively. T₀ is a constant equal to −46.02 °C (ref. ⁶⁵).

In practice, E₀ is first estimated according to equation (3). With a fixed E₀, the remaining parameters of equations (2) and (3) (α, β, k and R_ref) are derived using a time window of 2–14 d. The specific time window depends on data availability and the A_max value is assumed invariant within the same fitting window. On average, 25% of estimated A_max values are flagged as medium or low quality because the parameter ranges are unreasonable and/or the curve fitting is unconstrained (Supplementary Fig. 6b) and are subsequently discarded³⁵. Additionally, A_max values that are constant for 14 consecutive days or more are excluded. More than 88% of the A_max values in the remaining dataset are fitted within a 2 d window (Supplementary Fig. 6a), indicating a sufficient sample size for most fitting. Here we derive A_max using the REddyProc R package (https://github.com/bgctw/REddyProc)⁶⁶, as A_max is not provided in the FLUXNET2015 database. We convert PPFD to R_g using a constant of 2.1 μmol J⁻¹ (ref. ⁶⁷). We standardize A_max to PPFD = 2,000 μmol m⁻² s⁻¹ (A_max,2,000) by setting R_g = 952 W m⁻² in equation (1) and calculating the corresponding assimilation rate. This approach can avoid any A_max values obtained from potentially unsaturated light conditions and ensure consistent levels of absorbed PAR³⁵.

Timescale for thermal acclimation of A_max,2,000

We hypothesize that the most relevant timescale for thermal acclimation (τ) ranges between 2 and 60 d, according to the coordination hypothesis and observations^18,20,68. We conduct linear regressions between A_max,2,000 derived from the FLUXNET2015 sites and the daytime $\overline{T_{\mathrm{air}}}$ averaged over the 2–60 d before the time of A_max,2,000 measurements with a time interval of 1 d. On the basis of a previous study³⁵, savanna and shrubland sites are excluded from the analysis because they are frequently subject to water stress. Croplands are excluded from the cross-site analysis. Furthermore, we exclude the A_max,2,000–$\overline{T_{\mathrm{air}}}$ pairs collected during water-limited conditions, as indicated by the ratio of prevailing actual evapotranspiration to Priestley–Taylor potential evapotranspiration (ET/PET) < 0.7 (ref. ⁶⁹) and VPD > 20 hPa. Additionally, we only focus on growing seasons, characterized by fAPAR > 0.3 and T_air and $\overline{T_{\mathrm{air}}}$ > 0 °C. Daily fAPAR and LAI for each site were derived by interpolating the 8 d MODIS MOD15A2H products following ref. ³⁵. Low-quality data affected by cloud contamination are removed³¹. A total of 149,403 A_max,2,000 records are used for further analyses.

To remove the potential effects of concurrent T_air and fAPAR on A_max,2,000, we group A_max,2,000–$\overline{T_{\mathrm{air}}}$ pairs into different bins of T_air with 1 °C intervals and fAPAR with 0.02 intervals. This approach allows the analysis of changes in A_max,2,000 along $\overline{T_{\mathrm{air}}}$ gradients to be made while controlling for the instantaneous temperature dependence of photosynthesis and seasonal changes in leaf quantity and the development of the photosynthetic system. Pearson r between A_max,2,000 and $\overline{T_{\mathrm{air}}}$ that is averaged over different time frames (that is, 2–60 d with 1 d interval) is calculated for T_air and fAPAR bins. A positive r indicates the thermal acclimation potential of A_max,2,000. Only bins with sampling numbers larger than 10 and 20 for PFT-based and cross-site analyses, respectively, are retained. We examine the relationship between the average of the positive r values obtained from T_air and fAPAR bins and the time frames used to calculate $\overline{T_{\mathrm{air}}}$ for each PFT and cross sites (Fig. 2). Parameter τ is defined as the corresponding time frame when the 5 d moving average of the positive r reaches its peak. EVI, derived from MODIS reflectance data (MCD43A4) in the near-infrared, red and blue spectral bands⁵¹, is used to estimate τ for EBF for subsequent analysis, as an optimal τ cannot be identified for this PFT using A_max,2,000 (Supplementary Fig. 5).

Evidence for thermal acclimation of A_max,2,000

We use PFT-specific τ values for aggregating prevailing T_air to obtain $\overline{T_{\mathrm{air}}}$ (Fig. 1). We run LMMs, which include a random effect of different sites for removing the site-level adaptation effect, to explore the relationship between A_max,2,000 and PFT-specific $\overline{T_{\mathrm{air}}}$ (that is, A_max,2,000 ∼ $\overline{T_{\mathrm{air}}}$ + (1∣site)) (Extended Data Fig. 1a). The same data selection procedure and T_air and fAPAR binning scheme are used for the cross-site analysis (Fig. 1a and see earlier). The coefficient of $\overline{T_{\mathrm{air}}}$ estimated from LMMs is defined as thermal acclimation rate (γ_T). The sampling number, conditional and marginal correlation coefficients for the cross-site analysis are shown in Supplementary Fig. 6. The LMM is conducted with the R package lme4 (ref. ⁷⁰). For each site, the sampling number of A_max,2,000–$\overline{T_{\mathrm{air}}}$ pairs is insufficient to support the correlation analysis under the binning scheme³⁵. Instead, a partial correlation analysis is run between A_max,2,000 and $\overline{T_{\mathrm{air}}}$ controlling for $\overline{\mathrm{PPFD}}$, T_air and fAPAR on flux sites with observation lengths longer than 5 yr (Fig. 1c).

The prevailing conditions of T_air and PPFD often show a high correlation (Supplementary Fig. 1a). Therefore, we also include $\overline{\mathrm{PPFD}}$ as an additional predictor in the LMM (Extended Data Fig. 2a) and we analyse partial correlations between A_max,2,000 and $\overline{T_{\mathrm{air}}}$ controlling for $\overline{\mathrm{PPFD}}$ (Extended Data Fig. 2b) to eliminate the confounding effect of light acclimation³⁵. Additionally, we repeat LMMs with a different target variable (A_max) and random effect (PFT) to examine the robustness of the detectability of thermal acclimation (Extended Data Fig. 2c,d).

Modelling canopy photosynthesis of C₃ plants

We apply the photosynthesis module of the BESS model⁵⁷ to estimate canopy photosynthesis (A) and subsequently A_max,2,000, for each flux site. This allows a direct comparison to be made of the impacts of different empirical formulations of leaf photosynthetic capacities on thermal acclimation. The photosynthesis module is based on the FvCB model⁴, where A is determined as the lower CO₂ assimilation rate between the maximum rate of ribulose-1,5-bisphosphate carboxylase/oxygenase activity when light is saturated (A_c) and the electron-transport rate for RuBP regeneration when light is limited (A_j). For this study, the two-big-leaf scheme implemented in the BESS model is simplified to a one-big-leaf scheme. We have updated the parameters of temperature dependence of the maximum carboxylation rate (V_cmax, μmol m⁻² s⁻¹), the maximum electron-transport rate (J_max, μmol m⁻² s⁻¹), as well as the ratio of their values at 25 °C following ref. ¹². A detailed description of the canopy photosynthesis model can be found in Supplementary Text 1 (also see refs. ^31,55,57).

Leaf photosynthetic capacities

V_cmax is a key parameter in the FvCB model, particularly under light-saturated conditions⁴. Previous studies have shown that leaf biochemical components can acclimate to $\overline{T_{\mathrm{air}}}$ (refs. ^11,12,21). In this study, we compare three empirically derived variants of V_cmax at 25 °C ($V_{\mathrm{cmax}}^{25\mathrm{C}}$) within the FvCB model to evaluate their effectiveness in simulating the observed γ_T:

1. $V_{\mathrm{cmax}\_\mathrm{PFT}}^{25\mathrm{C}}$: this variant assumes a constant $V_{\mathrm{cmax}}^{25\mathrm{C}}$ value over the growing season, an assumption that is still widely used in vegetation models¹⁶. The prescribed top leaf $V_{\mathrm{cmax}}^{25\mathrm{C}}$ values are adopted from a look-up table based on PFTs and climatic zones compiled from the TRY trait database^31,71.

2. $V_{\mathrm{cmax}\_\mathrm{LAI}}^{25\mathrm{C}}$: leaf $V_{\mathrm{cmax}}^{25\mathrm{C}}$ varies seasonally, with its seasonality following LAI. This scheme, implemented in the previous version of the BESS model³¹, follows equation (4).

   $$V_{\mathrm{cmax}\_\mathrm{LAI}}^{25\mathrm{C}}=a\times V_{\mathrm{cmax}\_\mathrm{PFT}}^{25\mathrm{C}}+\left(1-a\right){\times V}_{\mathrm{cmax}\_\mathrm{PFT}}^{25\mathrm{C}}\times \frac{\mathrm{LAI}-{LAI}_{\min }}{{\mathrm{LAI}}_{\max }-{\mathrm{LAI}}_{\min }}$$ 4

   where LAI_min and LAI_max are the 5th and 95th percentile values of LAI over a growing season, respectively, and a is an empirical parameter set to 0.3 (ref. ⁵⁷).

3. $V_{\mathrm{cmax}\_\mathrm{EEO}}^{25\mathrm{C}}$: the calculation is based on EEO theory^19,34,56, specifically the coordination hypothesis^17,72 and the least-cost hypothesis^50,73. The coordination hypothesis proposes that plants actively coordinate resource allocation so that A_c tends to equal A_j on weekly to monthly timescales. The least-cost hypothesis proposes that plants minimize the combined costs (per unit assimilation) of maintaining the biochemical capacity for photosynthesis and the water transport capacity required to support it, through stomatal regulation. Combining the two hypotheses results in an optimal intercellular CO₂ concentration under representative conditions⁷⁴. Here we assume that $V_{\mathrm{cmax}\_\mathrm{EEO}}^{25\mathrm{C}}$ acclimates to prevailing conditions following the same timescale as A_max,2,000 (Fig. 2). The calculation is detailed in Supplementary Text 2 and ref. ³⁴.

Reporting summary

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

Author contributions

J.L. designed the study, performed the analysis and drafted the initial paper. J.L., Y.R., X. Luo, T.F.K. and P.G. participated in the early-stage discussion. I.C.P. substantially revised the paper. All co-authors commented on the results and contributed to the writing of the paper.

Peer review

Peer review information

Nature Plants thanks Mirindi Eric Dusenge, Marc Peaucelle and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Data availability

The dataset of FLUXNET2015 flux sites under the CC-BY-4.0 policy is publicly available for download at http://fluxnet.fluxdata.org. Remote-sensing canopy structure data from the MODIS MCD43A and MOD15A2H products are freely accessible at https://lpdaac.usgs.gov/products/mcd43a3v006/ and https://lpdaac.usgs.gov/products/mod15a2hv006/. BESS flux products are publicly available at https://www.environment.snu.ac.kr/data/.

Code availability

The corresponding R code scripts used in this study are available via Zenodo at 10.5281/zenodo.13854273 (ref. ⁷⁵). The code for the deviation of A_max from the FLUXNET2015 database is available via GitHub at https://github.com/trevorkeenan/inhibitionPaperCode. The code for modelling optimality-based V_cmax is available via GitHub at https://github.com/chongya/SVOM.

Competing interests

The authors declare no competing interests.

Extended data

is available for this paper at 10.1038/s41477-024-01846-1.

Supplementary information

The online version contains supplementary material available at 10.1038/s41477-024-01846-1.