Observed declines in leaf nitrogen explained by photosynthetic acclimation to CO₂

Significance

Forest responses to rising atmospheric CO₂ create critical uncertainties about ecosystem services under climate change. While negative trends in leaf nitrogen content have been widely observed and linked to increasing CO₂, the underlying mechanisms remain unresolved. Our research across European forests suggests that declining leaf nitrogen likely represents an adaptive response rather than a symptom of increasing nutrient limitation on ecosystem functioning. By combining ecoevolutionary theory with field observations, we demonstrate that elevated CO₂ allows plants to enhance their efficiency and decrease nitrogen investment in leaves for photosynthesis. These findings offer alternative perspectives on our understanding of terrestrial carbon–nitrogen cycle interactions and suggest forests may sustain primary productivity with lower nitrogen requirements than previously thought under future conditions.

Abstract

Widespread evidence of decreasing leaf nutrients has raised concerns about ecosystem productivity under global change. Interpreting trends in leaf nutrients has important implications for the fate of ecosystem services, particularly the role of forests in mitigating climate change and sustaining quality food sources. Here, we challenge the common interpretation that decreasing leaf nitrogen concentration (LNC) is evidence of increasing nutrient limitations on ecosystem primary productivity. Instead, we show that declines in LNC (4% decrease per 50 ppm CO₂ increase), observed across 409 European forest plots over 22 y, can be explained by reduced photosynthetic nitrogen demand. This regional trend is consistent with leaf acclimation to increasing atmospheric CO₂ according to optimality theory. This finding suggests that enhanced photosynthetic nitrogen use efficiency due to CO₂ fertilization may lead to less nitrogen uptake and/or reallocation of nitrogen for plant growth and other functions. Our results have large implications for understanding and simulating interactions between ecosystem nitrogen and carbon cycles and suggest nitrogen requirements for terrestrial carbon uptake under elevated CO₂ may be lower than previously thought.

Declines in leaf nitrogen concentration (LNC, mg g⁻¹) have been widely observed in long-term forest monitoring and elevated CO₂ experiments, and attributed to increasing atmospheric CO₂ (1–5). However, the mechanisms explaining this phenomenon and its implications for CO₂ fertilization of ecosystem functioning remain unresolved (4, 6). Higher CO₂ leads to more efficient photosynthesis, resulting in increased plant primary productivity and lower stomatal conductance (7). It is often suggested that this increased plant growth dilutes leaf nutrients and that lower transpiration reduces plant nutrient uptake (4, 8–10). This “limitation-centric” perspective implies that CO₂ fertilization may lead to weakening ecosystems and progressive nutrient limitations, thereby threatening the strength of the land carbon sink (11–14).

Conversely, decreasing LNC could result from plant acclimation to higher atmospheric CO₂, which reduces nitrogen demand for photosynthesis. Nitrogen is a key nutrient for the photosynthetic enzyme ribulose-1,5-bisphosphate (RuBisCO), and more efficient photosynthesis associated with higher CO₂ theoretically requires less leaf metabolic nitrogen for carboxylation (15). The nitrogen in excess of what is required in conditions of elevated CO₂ is therefore economized and can be allocated elsewhere (4, 10, 16). This “efficiency-centric” perspective implies that CO₂ fertilization, by enhancing photosynthetic nitrogen use efficiency (PNUE) and reducing LNC, may lead to less nitrogen uptake and/or liberate nitrogen for other beneficial uses, thus potentially mitigating some negative impacts of climate change on ecosystem functioning (17).

In this study, we leverage ecoevolutionary optimality theory as a framework (18–21) through which to examine whether declines in LNC align with an efficiency-centric perspective, offering an alternative to current limitation-centric perspectives. Specifically, we use the coordination hypothesis (22, 23) and the least-cost theory (18, 24) to estimate photosynthetic capacity (here in terms of the maximum rate of RuBisCO carboxylation, $V_{c,max25}$) and associated LNC based on the assumption that plants optimize light use efficiency to maximize gross carbon assimilation at the lowest enzymatic cost. These optimality principles, which link photosynthetic capacity and therefore leaf nitrogen investments for photosynthesis to experienced environmental conditions, find strong support in observations and have been widely tested for average growing conditions (15, 19, 20), seasonal cycles (25), warming and elevated CO₂ experiments (26) and to understand long-term photosynthetic trends (27, 28).

We hypothesize that a reduction in leaf nitrogen demand to synthesize RuBisCO, due to optimal acclimation of photosynthetic capacity to increasing atmospheric CO₂, explains regional trends in leaf nutritional status. To test this hypothesis, we use annual LNC timeseries (1995 to 2016) from 409 European forest plots spanning boreal, temperate, and Mediterranean climates and comprising 32 species, standardized to a regional mean. We interpret observations through an optimality-based model (20, 24) including a parsimonious soil water balance (29) to predict annual growing season photosynthetic capacity and associated LNC anomalies at each plot. The model incorporates historical seasonal environmental conditions, including atmospheric CO₂, air temperature (T_a), vapor pressure deficit (VPD), photosynthetically active radiation (PAR), and plant water stress. By integrating ecoevolutionary theory with long-term observational records, we can thus assess from first principles the degree to which observed regional trends in LNC can be explained by theoretical declines in leaf nitrogen requirements for photosynthesis due to acclimation to elevated CO₂. Further, by evaluating theoretical requirements for photosynthetic nitrogen under future scenarios, we can show that rising atmospheric CO₂ might not exacerbate plant nitrogen limitations as much as previously thought.

Results and Discussion

The theoretical regional decrease in LNC (−39 ± 5 μg g⁻¹ y⁻¹) that optimally acclimates to climate and increasing CO₂ matches the trend derived from observations (−38 ± 7 μg g⁻¹ y⁻¹) across the European study area (Fig. 1A). This trend, incorporating all climatic and CO₂ responses, represents a 4% decrease in LNC per 50 ppm CO₂ increase, consistent with previous regional and global reports (1–3, 5). The model’s alignment with observations provides a theoretical and tractable explanation of the historical LNC trend associated with increasing CO₂.

Fig. 1.

(A) Trends in LNC (mg g⁻¹) across Europe derived from observations (Obs., black) and the optimality model for photosynthetic capacity (Th., green). Markers represent regional median LNC across 409 forest plots and shading represents the SE of plot-level annual LNC values. Trend lines are fit using the Thiel-Sen estimator and are statistically significant with >99% confidence. Scatterplots show the correlation between observed and theoretical. (B) regional median annual LNC values (ρ = 0.54, P < 0.01) and (C) plot-level annual LNC values (ρ = 0.08, P < 0.01). We standardize LNC values to a regional mean (18 mg g⁻¹) to remove forest and species-specific variability and focus on regional temporal anomalies. LNC observations are based on ICP Forests data (http://icp-forests.net).

To benchmark the theory’s performance, we develop an independent data-driven approach, using random forest regression, that approximates relationships between LNC anomalies and environmental variables (SI Appendix, Fig. S1). The Pearson’s correlations between observed and modeled LNC are 0.54 and 0.64 for theoretical (Fig. 1B) and data-driven (SI Appendix, Fig. S1B) regional median LNC estimates, respectively; and 0.08 and 0.14 for theoretical (Fig. 1C) and data-driven (SI Appendix, Fig. S1C) plot-level LNC anomalies, respectively. The small difference in goodness-of-fit between theoretical LNC estimates and the data-driven benchmark indicates that the theory uses the majority of information contained in available environmental variables. This lends support to the optimality assumptions underlying the modeling of photosynthetic acclimation to climate and CO₂. The remaining unexplained variability may be attributed to unobserved factors affecting temporal LNC responses, in addition to uncertainties inherent in noisy LNC observations. To separate and compare contributions of climate and CO₂ on LNC regional anomalies, we examine LNC trends derived from models with long-term average versus annually variable environmental conditions as well as the random forest variable importance factors (Fig. 2). We find an average 3% decrease in LNC per 50 ppm CO₂ increase using the theoretical model incorporating only the CO₂ response, consistent with LNC and photosynthetic capacity sensitivities to CO₂ from meta-analyses of elevated CO₂ experiments (4, 5, 26). Increasing CO₂ has a larger effect on LNC temporal patterns than climate for both theoretical and data-driven approaches. This finding corroborates previous empirical analyses of long-term European forest LNC monitoring (5), which additionally demonstrated weak influence of nitrogen deposition on foliar nutrients and no LNC influence on tree growth. The theoretical regional decreasing LNC trend with increasing CO₂ and constant climate is three times the trend with constant CO₂ and variable climate (Fig. 2A), indicating that CO₂ explains about three quarters of the historical LNC regional trend. The significance of CO₂ in the random forest, with an average variable importance factor of 53%, also supports the finding that CO₂ drives the historical regional LNC trend (Fig. 2B).

Fig. 2.

(A) Trends in LNC derived from observations, a random forest regression and theoretical model alternatives with i) annually varying growing season climate, atmospheric CO₂, and accounting for water stress ii) annually varying growing season climate and atmospheric CO₂ in well-watered (ww) conditions; iii) annually varying atmospheric CO₂ and long-term average growing season climate; iv) annually varying growing season climate and constant atmospheric CO₂ (380 ppm); (B) Feature importance in random forest regression to estimate LNC anomalies. Features represent environmental variables used in the theoretical model and include atmospheric CO₂ concentration, water stress, VPD, air temperature (T_a), and PAR. Error bars in (A) and (B) indicate SE of the regional trends.

Water stress emerges as the primary climatic factor explaining LNC anomalies (Fig. 2B). To further evaluate ecohydrological controls, we compare models assuming LNC variability is proportional to photosynthetic capacity in water-stressed (Fig. 1) versus well-watered conditions (SI Appendix, Fig. S2). The well-watered theoretical model can capture the regional LNC trend, however, its trend is slightly steeper (Fig. 2A) and we only find a significant correlation between observed and modeled LNC values when accounting for water stress (Fig. 1 B and C and SI Appendix, Fig. S2 B and C). These differences in goodness-of-fit indicate a need to account for the soil water balance, which determines the influence of plant water uptake and stress on leaf nitrogen demand and supply. The variable importance factors from the random forest are 20% for water stress and 14% for VPD (Fig. 2B), underlining the effects of plant hydraulic status on LNC regional spatiotemporal patterns.

Uncertainties in capturing the full variability in plot-level LNC anomalies, using both theoretical (Fig. 1C) and data-driven (SI Appendix, Fig. S1C) approaches, point to limited information in considered environmental factors, potential model deficiencies, and challenges associated with the quality of regional long-term LNC observations. Consequently, we are unable to further analyze and explain spatial patterns in LNC trends and plot-level LNC anomalies that may be driven by plant mechanisms affecting total LNC beyond nitrogen requirements associated with photosynthetic acclimation. Improving model performance will require investigation into additional factors influencing nitrogen supply to meet theoretical demand and understanding temporal variability in LNC associated with other functions such as defense and structure. These factors affect the proportionality between photosynthetic capacity and LNC at the plot versus regional scales. Nevertheless, our findings demonstrate that ecoevolutionary theory effectively explains the observed regional LNC trend. This underscores the role of photosynthetic acclimation to elevated CO₂ as a primary mechanism influencing LNC declines, and incorporating plant water stress into the theory enhances the agreement between historical observations and theoretical LNC estimates.

Examining observed trends in LNC through first principles theory of how CO₂ and climate interact to determine direct CO₂ fertilization effects on photosynthetic capacity is important to develop process understanding of ecosystem responses to global change. At the leaf level, optimal acclimation predicts that increasing atmospheric CO₂ (all else equal) reduces the maximum rate of photosynthesis needed to maintain relatively constant CO₂ draw down (Fig. 3A). As such, enzymatic requirements for photosynthesis decrease with increasing atmospheric CO₂, leading to decreasing leaf nitrogen investments for RuBisCO and increasing PNUE.

Fig. 3.

Theoretical relationships for optimal relative photosynthetic capacity (scaled $V_{c,max25}$) with increasing (A) CO₂; (B) air temperature (T_a, °C); (C) VPD (kPa); (D) PAR (mmol photons m⁻² s⁻¹) in well-watered conditions and all else set to the regional average climate values across 409 forest plots. Theoretical change (% 50 ppm⁻¹ CO₂) of leaf-level photosynthetic nitrogen demand (light blue) and PNUE (dark blue) with increasing atmospheric CO₂ across forest plots binned by their long-term average growing season (E) air temperature (°C) and (F) VPD (kPa). Values presented in subplots E and F are estimated by only varying CO₂ using constant growing climate variables held at each plot’s long-term average values and well-watered conditions. Solid lines represent the median trend and shading represents the 5th to 95th percentile range across forest plots.

Responses to climatic variables (Fig. 3 B–D) modulate trends in leaf photosynthetic nitrogen demand (Fig. 3 E and F). Theoretical change in leaf-level photosynthetic capacity to CO₂ varies with T_a and VPD, while PAR and water stress effects only scale its magnitude and do not directly interact with CO₂ (Eqs. 2 and 3 in Methods). With increasing CO₂ and all else equal, the theory predicts a quasilinear increase in magnitude of the negative trend with increasing average growing season temperature (Fig. 3E), and interactions between temperature and atmospheric water demand result in a nonlinear response (Fig. 3F). These theoretical trajectories exemplify potential compounding effects of elevated CO₂ and climate change on LNC as well as complimentary responses on PNUE, as displayed in the theory-data integration (Figs. 1 and 2).

Finally, we use our theoretical framework to examine projected trends in optimal leaf-level photosynthetic nitrogen requirements across Europe under climate change scenarios, including middle road (SSP245) and fossil fuel development (SSP585) shared socioeconomic pathways (SSP), with a 26% and 38% increase in CO₂ between 2015 and 2050, respectively (Fig. 4A). The theory predicts a continued decrease in leaf photosynthetic nitrogen and the regional trend is about 1.6 (SSP245) and 2.6 (SSP585) times that of historical LNC observed under a 12% increase in CO₂ between 1995 and 2016. In future scenarios, climate alone is projected to contribute to about half of the total decrease in photosynthetic capacity and associated photosynthetic leaf nitrogen requirements, which is more substantial than in the historical period (Fig. 2). The effect of climate, relative to CO₂, on photosynthetic nitrogen is greater under SSP585, where atmospheric CO₂ levels are highest. This indicates that climate change may increasingly reduce photosynthetic capacity relative to CO₂. Overall, these results show that models that do not account for the downregulation of photosynthetic capacity with increasing CO₂ may overestimate projected nitrogen limitations on ecosystem photosynthesis.

Fig. 4.

Projected theoretical responses for SSP climate change scenarios aggregated over European forest plot locations using forcings from an ensemble of 17 models in the Coupled Model Intercomparison Project (30), including middle road (SSP245, green) and fossil fuel development (SSP585, purple). (A) Regional trends in relative photosynthetic leaf nitrogen requirements assuming photosynthetic capacity optimally acclimates to CO₂ and climate (solid line), only climate and CO₂ fixed at the 2015 level (dashed line), only CO₂ and climate fixed at the historical (1996 to 2015) average values for each model (dotted line). Thin solid lines represent the median and shading the SE from the model ensemble (colored) and LNC observations (gray). (B) Sensitivities of leaf-level RuBisCO-limited photosynthesis (A_c) to elevated CO₂ (% A_c %⁻¹ CO₂) assuming CO₂ and climate acclimation of photosynthetic capacity, only climate acclimation of photosynthetic capacity, and no downregulation of photosynthetic capacity.

Acclimation of photosynthetic parameters (e.g., maximum rate of carboxylation, $V_{c,max25}$) to elevated CO₂ is typically not simulated by land surface models (31, 32). To illustrate implications for predictions of photosynthesis under CO₂ fertilization, we calculate the theoretical sensitivity of leaf-level RuBisCO-limited photosynthesis to CO₂ for SSP245 and SSP585 scenarios, both with and without acclimation of photosynthetic capacity (Fig. 4B). Omitting optimal downregulation of leaf-level photosynthetic capacity to CO₂ and only acclimating to climate leads to 5- and 2.5-times greater increases in RuBisCO-limited photosynthesis compared to acclimating to both CO₂ and climate for SSP245 and SSP585 scenarios, respectively (Fig. 4B). These results suggest that improving mechanistic understanding of leaf nitrogen demand on photosynthetic capacity is critical for predicting and not overestimating CO₂ fertilization effects on ecosystem photosynthesis. Our findings underscore the need to integrate dynamic representations of ecophysiological responses within terrestrial biosphere models that reflect acclimation to elevated CO₂.

Ecoevolutionary optimality principles, such as those applied in this study, provide a framework for representing how leaf-level traits interact with climate and CO₂ to optimize resource use and tradeoffs, thereby maximizing performance in a given environment (21). Specifically, the coordination hypothesis (23) and least-cost theory (18, 24) underlying theoretical estimates of LNC describe the costs and benefits of water and nutrients for optimal photosynthetic capacity. In this framework, the optimal balance of the two key photosynthesis resources (water and nitrogen) is set by the ratio of their unit cost. As such, leaf-level traits are optimized to environmental conditions and determine plant-level allocations so that soil resource limitation should result in fewer leaves, rather than “diluted” (4, 8) leaves with suboptimal photosynthetic capacity. That is, photosynthetic responses are regulated by leaf nitrogen demand while whole-plant responses can be constrained by nitrogen availability (33). Further, within a more comprehensive causal regression framework across the European study region, which related tree growth to LNC, no evidence of dilution effects was found and nitrogen deposition emerged as a weak predictor of LNC (5). Our regional analysis, focused on leaf-level mechanisms, thereby provides a compelling alternative hypothesis to limitation-centric perspectives about CO₂ fertilization effects on LNC. Importantly, this study’s plant-level, efficiency-centric reasoning allows us to better understand the role that resource use strategies and their interaction with environmental conditions play in determining ecophysiological observations.

Enhanced understanding of leaf acclimation to elevated CO₂, derived from our theory-data integration, can guide ecosystem model development by providing a more dynamic representation of photosynthetic capacity (21) and improve predictive capacity of the land carbon sink (6). Optimality-based estimates of photosynthetic parameters offer a parsimonious and generalized representation of environmental mechanisms driving leaf nitrogen demand and photosynthesis. Additionally, we show that climate interactions and hydraulic status influence direct effects of CO₂ fertilization on photosynthetic leaf nitrogen requirements and these interactions may have buffering or compounding effects. Numerous factors control complex ecosystem carbon–nitrogen interactions that are not fully represented and captured in models (34). There are substantial uncertainties and knowledge gaps about interactions between environmental conditions, photosynthesis, and the allocation of carbon and nitrogen from leaf to canopy and ecosystem levels. These gaps underscore the need for assessing multiple hypotheses in Earth system models to understand how our findings affect ecosystem carbon uptake and productivity. Nevertheless, our results suggest that downregulation of photosynthetic capacity in response to elevated CO₂ can improve models of leaf-level photosynthesis and reduce the likelihood of overestimating CO₂ fertilization effects. This dynamic response reduces leaf nitrogen requirements for photosynthesis and can therefore potentially alleviate nutrient limitations and availability for ecosystem functioning. Further, existing model simulations, incorporating photosynthetic acclimation to CO₂, indicate that rising atmospheric CO₂ might not exacerbate plant nitrogen limitations as much as previously thought. Instead, they suggest enhanced PNUE likely allows nitrogen to be more available for other pools within the ecosystem, and ultimately enhance ecosystem carbon stocks (17). Our findings, linking environmental conditions including elevated atmospheric CO₂ to observed LNC through optimality theory, offer an alternative perspective on interactions between terrestrial carbon and nitrogen cycles to improve understanding of expected declines in nutritional quality of plants (9, 35) and ecosystem carbon uptake (36).

Conclusions

This study reconciles theory and observational evidence to show that decreasing LNC is likely driven by photosynthetic acclimation to more favorable environmental conditions under elevated atmospheric CO₂. This efficiency-centric perspective, which also finds strong support in experimental observations (26), nuances the prevailing narrative that elevated CO₂ exacerbates vegetation nitrogen demand and limitations in terrestrial ecosystem carbon uptake (11, 13, 37). Instead, we suggest that elevated CO₂ may lead to a shift in nitrogen investments, from photosynthetic enzymes in leaves to other functions, and/or reduced nitrogen uptake. That is, acclimation that optimizes photosynthetic efficiency may drive nitrogen reallocation, representing a potential critical mechanism by which elevated CO₂ can beneficially influence interactions between ecosystem carbon and nitrogen cycles. Our ecophysiological interpretation of long-term trends in LNC offers an alternative avenue to incorporate process-based representations of CO₂ effects on photosynthetic capacity in Earth system models, using a parsimonious and generalizable principle for optimal photosynthetic nitrogen requirements. These results, which provide observational evidence of the effects of increased atmospheric CO₂ on photosynthetic capacity, significantly advance understanding of terrestrial carbon and nitrogen cycles and their interactions under long-term global change.

Materials and Methods

LNC observations.

For ground truth, we use a 22-y annual timeseries (1995 to 2016) of plot-level LNC (mg g⁻¹) derived from intensive European forest monitoring (Level II plots) in the International Cooperative Programme (ICP) on Assessment and Monitoring of Air Pollution Effects on Forests operating under the UNECE Convention on Long-range Transboundary Air Pollution (for details about data collection see ICP sampling manual Part XII: http://www.icp-forests.org/pdf/manual/2016/ICP_Manual_2017_01_part12.pdf). We consider that samples collected from different species within a same forest location each represent unique plots. For this study, we use annual composite plot-level data (see details in ref. 5) and select forest plots that have at least six annual composite LNC values over at least a 10-y period. The resulting dataset comprises 4,228 LNC datapoints from a total of 409 European forest plots across temperate (338 plots), boreal (25 plots), and mediterranean (46 plots) climates and includes 32 species. The dominant species in the dataset are Picea abies (104 plots), Fagus sylvatica (79 plots), Pinus sylvestris (68 plots), and Quercus petraea (34 plots). Plot-level mean LNC values range from 7 to 29 mg g⁻¹. To remove forest- and species-specific variability, we standardize LNC data to a common mean by dividing values by each plot’s mean LNC and multiplying by 18 mg g⁻¹ (the average of the plot-level LNC means). As such, we focus on annual LNC anomalies to study regional temporal trends in the European study area. We quantify regional LNC trends over the study period using the nonparametric Theil-Sen estimator. We report trends in units of μg g⁻¹ y⁻¹ and changes to elevated atmospheric CO₂ as % LNC 50 ppm⁻¹ CO₂ to facilitate comparisons with previous literature. While inherent errors of LNC sampling and the temporal distribution and spatial representativeness of LNC measurements may introduce biases and uncertainties, we ensure the generalizability of our findings by successfully reproducing trend results from a previous analysis based on the same underlying dataset (5).

Growing season climate and CO₂ timeseries.

To estimate theoretical LNC values, we extract historical seasonal climate timeseries from gridded data products using Google Earth Engine. We use TerraClimate monthly gridded (1/24°) climate data (38) to calculate average annual growing season air temperature (T_a, °C), PAR (mol photons m⁻² s⁻¹), VPD (Pa), and potential evapotranspiration (m d⁻¹) for each plot. We calculate average annual growing season rainfall intensity (α, m d⁻¹) and frequency (λ, d⁻¹) using daily ERA5 aggregates of gridded (1/4°) atmospheric reanalysis of global climate produced by the European Centre for Medium-Range Weather Forecasts (39). We determine average atmospheric pressure (P_a, Pa) at each location based on elevation extracted from the GTOPO30 global digital elevation model (40). We obtain timeseries of average annual atmospheric CO₂ concentration from the Mauna Loa Observatory (41, 42).

Site characteristics and growing season.

For soil water balance calculations, we extract additional site soil and land cover characteristics from gridded products using Google Earth Engine. We determine the soil texture class at 30 cm depth from SoilGrids (43) for each location and use literature values of soil water retention parameters for each soil texture class (44). We obtain leaf area index (LAI, m² leaf m⁻² ground) at a 500-m resolution from MODIS product MCD15A3H V6 (45), selecting retrieval values flagged as cloudless and high quality between 2002 to 2020 within a 1.5 km radius of each plot. We determine the growing season period as the months of the year during which average LAI is at least 50% the LAI of the peak month and the peak as the month of the year with highest long-term average LAI.

Climate and CO₂ projections.

To evaluate LNC under future SSP climate change scenarios, we obtain global average annual atmospheric CO₂ and growing season climate for each plot from the Coupled Model Intercomparison Project (30) (CMIP6) using Pangeo’s CMIP6 Google Cloud Collection. We extract the historical, middle road (SSP245) and fossil fuel development (SSP585) scenarios from all models that provide monthly climate variables required in this study. We therefore use climate inputs from 17 models to estimate LNC under CMIP6 scenarios and the range of their predictions to quantify uncertainty in our theoretical projections. See SI Appendix, Table S1 for a list of model names and associated modeling centers. We extract atmospheric CO₂ for each scenario from the GFDL-ESM4 model by the National Oceanic and Atmospheric Administration Geophysical Fluid Dynamics Laboratory.

Optimality-based photosynthetic capacity.

We quantify photosynthetic capacity at leaf level by the maximum rate of carboxylation at a standard temperature of 25 °C ($V_{c,max25}$, mol CO₂ m⁻² s⁻¹). We estimate ambient temperature photosynthetic capacity ($V_{c,max}$) and the associated optimal rate of photosynthesis per unit leaf area (A, mol CO₂ m⁻² s⁻¹) based on the assumption that plants optimize leaf light use efficiency to maximize gross carbon assimilation at the lowest total resource (enzymatic and water) cost—refer to previous optimality model descriptions for details and validation (20, 24, 46). Previous studies also show that the amount of nitrogen allocated to the photosynthetic enzyme RuBisCO is proportional to $V_{c,max}$ at a standard temperature and not at growth temperature (19). In this study, we calculate optimality-based $V_{c,max}$ for each plot given average annual growing season environmental conditions (T_a, PAR, VPD, P_a) and for sun-lit leaves (fPAR = 0.91). We then standardize ambient temperature values to 25 °C ($V_{c,max25}$) following standard methods (47–50).

The optimality-based model for $V_{c,max}$ emerges from the simplification that electron transport rate–limited photosynthesis and RuBisCO carboxylation rate–limited photosynthesis are coordinated (23). That is, for the time-scale of this study’s forcing variables (growing season daily averages), we assume plants photosynthesize at the colimitation point of light- and RuBisCO-limited photosynthesis rates ($A_{\mathrm{j}}=A_c$), yielding an effective linear relation between carbon assimilation and light.

$$V_{c,max}={\phi }_0{I}_{\mathit{abs}}\left(\frac{\chi C_a+K}{\chi C_a+2{\mathrm{\Gamma }}^{*}}\right)\left(\frac{{\omega }^{*}}{8\theta }\right),$$ [1]

$$A={\phi }_0{I}_{\mathit{abs}}\left(\frac{\chi C_a-{\mathrm{\Gamma }}^{*}}{\chi C_a+2{\mathrm{\Gamma }}^{*}}\right)\left(\frac{{\omega }^{*}}{8\theta }\right),$$ [2]

where $C_a$ is ambient CO₂ concentration (mol mol⁻¹); ${\phi }_0$ is quantum yield efficiency of photosynthesis, a function of temperature and plant water stress (mol photons mol⁻¹ CO₂); $I_{\mathit{abs}}$ is absorbed photosynthetic photon flux density, a function of incoming PAR (mol photons m⁻² s⁻¹); ${\mathrm{\Gamma }}^{\ast }$ is the photorespiration CO₂ compensation point, a function of temperature (mol mol⁻¹); K is the Michaelis–Menten coefficient of RuBisCO-limited photosynthesis, a function of temperature and pressure (mol mol⁻¹); $\frac{{\omega }^{\ast }}{8\theta }$ is a correction factor and a function of $\chi$ that accounts for electron-transport rate limitations (20); and $\chi$ is the optimal ratio of leaf-internal to ambient CO₂ concentration, solved by the underlying theory (24) to balance the costs of maintaining carboxylation capacity and water transport or transpiration needed for photosynthesis, expressed as

$$\chi =\frac{{\mathrm{\Gamma }}^{*}}{C_a}+\left(1-\frac{{\mathrm{\Gamma }}^{*}}{C_a}\right)\frac{\xi }{\xi + \sqrt{VPD/P_a}},$$ [3]

with $\xi = \sqrt{\beta \frac{K+{\mathrm{\Gamma }}^{*}}{1.6{\eta }^{*}}}$; β =146 is the ratio of the unit costs of carboxylation and transpiration (unitless) (51); ${\eta }^{\ast }$ is the viscosity of water relative to its value at 25 °C (unitless) (52) and 1.6 is the ratio between the diffusivities of water vapor and CO₂ in air. We calculate temperature dependencies in the model ($K,{\mathrm{\Gamma }}^{\ast }$, ${\phi }_0$, $V_{c,max}$) according to equations and parameters in standard methods (47–51).

Plant water stress.

We couple the optimality model for photosynthetic capacity with an ecohydrological model to account for the influence of water balance components interacting with photosynthesis and therefore bring more edaphic and climatic conditions into the framework. We use a soil water bucket model forced with stochastic rainfall inputs, accounting for canopy interception, and in which soil water losses include infiltration, runoff, soil water evaporation, and plant transpiration (29). This minimalist water balance approach only requires soil texture (to determine soil water retention parameters), average rainfall intensity and frequency, potential evaporation, and leaf area index derived from gridded datasets, and it estimates probability distributions of soil moisture states. Based on estimated soil moisture probability distributions, this robust ecohydrological framework predicts average growing season plant water stress by accounting for the impact and likelihood of stressful low soil moisture states—refer to previous stochastic ecohydrological model descriptions for details and validation of soil water balance and plant water stress (29, 53). We apply the dynamic water stress factor to downregulate ${\phi }_0$ in Eq. 1, similarly to Stocker et al. (51) but with a process-based versus data-driven stress factor. As such, we convert $V_{c,max25}$ under well-watered conditions to actual average growing season photosynthetic capacity.

Theoretical LNC.

The fraction of leaf nitrogen invested in RuBisCO ($f_{\mathit{LNR}}$, g N in RuBisCO g⁻¹ N in leaf) depends on plant leaf nitrogen use strategies and relates $V_{c,max25}$ to LNC (54, 55)

$$V_{c,max25}=a_R\times f_{\mathit{NR}}\times f_{\mathit{LNR}}\times LMA\times LNC,$$ [4]

where $a_R$ is the maximum rate of carboxylation per unit RuBisCO protein (47.34 × 10⁻⁶ mol CO₂ g⁻¹ RuBisCO s⁻¹ at 25 °C); $f_{\mathit{NR}}$ is the fractional mass of the RuBisCO molecule to the mass of nitrogen in RuBisCO (6.25 g RuBisCO g⁻¹ N in RuBisCO), converting nitrogen concentration to protein concentration; and LMA is leaf mass per area (g leaf m⁻²), converting LNC to nitrogen per leaf area.

Here, we test whether the theoretical long-term variability in LNC is directly proportional to long-term variability in $V_{c,max25}$, assuming the supply of nitrogen in the soil is not depleted and that $f_{\mathit{LNR}}\times LMA$, for a given tree species in its environment, does not vary temporally with atmospheric CO₂ and climate. As such, the ecoevolutionary theory assumes that leaf nitrogen investments for RuBisCO are dependent on CO₂ and climate, and potential increases in LMA with elevated CO₂ are compensated by decreasing $f_{\mathit{LNR}}$. While decreases in LNC are widely observed in elevated atmospheric CO₂ experiments, leaf nitrogen per unit area are not significantly reduced because they are compensated by observed increases in LMA (4, 56). Variability in LNC due to other functions such as defense and structure are highly uncertain and may affect our proportionality assumption locally. As with the LNC observations, we remove spatial variability in theoretical estimates of LNC. We set $f_{\mathit{LNR}}\times LMA$ as a scaling factor for each plot that converts $V_{c,max25}$ to units of LNC (Eq. 4) and standardizes values to the common plot-level regional mean (18 mg g⁻¹). Our calculations do not require specific information about LMA and $f_{\mathit{LNR}}$ because we focus on temporal anomalies of LNC and evaluate regional trends. Consequently, we minimize uncertainties related to forest- and species-specific variability in these factors, while acknowledging that the model does not represent all potential local environmental effects and interactions.

Theoretical PNUE.

We calculate the theoretical leaf-level PNUE (g C g⁻¹ N s⁻¹) as the ratio of photosynthesis and leaf nitrogen invested in RuBisCO ($N_{\mathit{RuBisCO}}$). PNUE scales with CO₂ via optimality-based $\chi$ in the theoretical equations for $V_{c,max25}$ and photosynthesis.

$$PNUE=\frac{A}{N_{\mathit{RuBisCO}}}=\left(\frac{k_{\mathit{cat}}{n}_R{M}_C}{M_N{M}_R{N}_R}\right)\left(\frac{\chi C_a-{\Gamma }^{*}}{\chi C_a+K}\right),$$ [5]

where $k_{\mathit{cat}}$ is the catalytic turnover at 25 °C (3.5 mol CO₂ mol⁻¹ RuBisCO site⁻¹ s⁻¹); n_R is the catalytic sites per mol RuBisCO (8 mol RuBisCO sites mol⁻¹ RuBisCO); M_C is the molecular mass of CO₂ (44 g mol⁻¹); M_N is the molecular mass of nitrogen (14 g mol⁻¹) M_R is the molecular mass of RuBisCO (552,000 g mol⁻¹); and N_R is the nitrogen concentration of RuBisCO (0.0144 mol nitrogen g⁻¹ RuBisCO).

Data-driven benchmark for LNC anomalies.

We develop a random forest regression to relate the LNC anomalies (plot-level LNC observations standardized to a common regional mean) to environmental variables used in the optimality theory (PAR, T_a, VPD, water stress, CO₂). We use the random forest to benchmark the theoretical model’s performance because LNC measurements are inherently noisy and are influenced by several unknown factors not accounted for in the model. As such, the random forest provides a higher-bound estimate of the information the input variables used in the theoretical model can provide about regional LNC, based solely on a data-driven approach. We calculate the variable importance factors of the random forest model to assess the influence of environmental variables used in the model. To verify potential collinearity, we evaluate cross-correlations between variables (SI Appendix, Fig. S3). The Pearson’s correlation coefficients are all below 0.5 except for VPD and air temperature (ρ = 0.74) and VPD and PAR (ρ = 0.71). We use fivefold cross-validation by randomly splitting the LNC into five groups. We use four of the folds for model training and the fifth fold is held out for model testing, ultimately obtaining out-of-sample estimates for each location. Within each training set, we perform a grid search using fivefold cross-validation to determine optimal hyperparameter sets.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Python scripts to reproduce analysis (57) are deposited in Zenodo (10.5281/zenodo.15486462). Previously published data are used for this work. All study datasets are referenced in the manuscript.