In this study, we produced a global fLNR and ({V}_{{c}_{{max }}}^{25}) map using an RF model trained primarily by remote sensing and in situ observations and examined seven ({V}_{{c}_{{max }}}^{25}) models based on 5 competing hypotheses with regard to their assumptions on fLNR. Our results suggested that the global average fLNR was 18.2 ± 6.2%, and the global distribution of fLNR was dominated by the interaction between fLNR and leaf traits (i.e., LMA and LPC), followed by regional influences from climate (i.e., VPD and PAR) and soil characteristics (i.e., soil pH and sand percentage). We used RF fLNR distribution and its relationships with environmental covariates to evaluate five empirical and two optimal ({V}_{{c}_{{max }}}^{25}) models, and found that the models showed different degrees of inefficacy in reproducing RF fLNR. Here, we discuss the mechanisms underlying the detected fLNR responses to leaf traits, climate, and soil characteristics and propose future directions to improve the simulation of fLNR and ({V}_{{c}_{{max }}}^{25}) in models.
Negative correlation between fLNR and LMA
Our finding that fLNR is negatively related to LMA agrees with a previous meta-analysis that found fLNR decreases by 0.54 ± 0.08% with a 1 g/m2 increase in LMA based on a univariate regression10, though another study reported that the negative relationship between fLNR and LMA was non-significant using a smaller dataset11. Using the global dataset, we found a relatively small sensitivity of fLNR to LMA (−0.19 ± 0.001% per 1 g/m2) when accounting for climate and soil (Fig. 3b).
Higher LMA is the result of plants allocating more biomass and nitrogen to building cell walls, which may cause a reduction in CO2 diffusion into the mesophyll as well as relative nitrogen allocated to RuBisCO38. Leaves with greater LMA are tougher and usually have a longer leaf lifespan11,36. Therefore, the negative correlation between fLNR and LMA highlights the trade-off between photosynthesis and persistence along the leaf economic spectrum: on one end, leaves invest more nitrogen in RuBisCO to increase the photosynthetic capacity and enhance carbon uptake; on the other end leaves invest more nitrogen in structural biomass to improve leaf longevity and lengthen the carbon uptake period. The latter is especially true for evergreen species that have greater LMA and smaller fLNR than deciduous and herbaceous species10. The coordination of fLNR and LMA is also consistent with a recent analysis highlighting the role of LMA in determining the variation and predictability of LNC in ecosystem models39.
In addition, we found that LPC increases fLNR in tropical evergreen forests and mixed forests, which tend to be more phosphorus limited40. Our result is consistent with previous studies reporting coupled leaf photosynthetic capacity (i.e., ({V}_{{c}_{{max }}}^{25}) or maximum photosynthetic capacity (Amax)) and LPC for tropical species41,42. This result indicates potential widespread adjustments of plants nitrogen use by phosphorus investment for photosynthesis and plant growth43 in tropical and mixed forests. In addition, we note that the productivity of some grasslands44,45 and boreal forests46,47 has also been reported to be limited by phosphorus availability, however, we did not detect a strong positive dependence of fLNR on LPC globally for these ecosystems in our study. The difference potentially suggests that the phosphorus limitation of grasslands and boreal forests is not as prevalent as that for tropical and mixed forests (though some mixed forests are in the boreal region).
Climate and soil impacts on fLNR
The response of fLNR to climate is often implicitly included in ({V}_{{c}_{{max }}}^{25}) models. We found that fLNR was sensitive to annual VPD globally. Several studies have reported that plants in arid environments (i.e., high VPD) tend to have a higher Amax and LNC48,49 as plants enhance photosynthetic capacity to maintain a given assimilation rate with lower stomatal conductance and reduced water loss. Such a response to aridity has been described using the least-cost theory19,21. Our results show that other than Amax and LNC, fLNR also increases with VPD, consistent with a recent study reporting higher nutrient use efficiency for plants in semi-arid ecosystems of the African Sahel49. We note that an earlier study reporting differently that a dry site has a smaller ({V}_{{c}_{{max }}}^{25})/LNC ratio (i.e., smaller fLNR) than a wet site19, though it used annual precipitation, not VPD to define aridity.
In addition, the positive relationship between PAR and fLNR for non-forests (Fig. 3c) provides a potential explanation of the light acclimation of photosynthesis, as several studies have found that leaf and ecosystem Amax can be enhanced by intermediate to long-term average PAR50,51,52. For non-forest ecosystems, our results suggest that photosynthetic light acclimation emerges as plants increase fLNR in response to increasing annual PAR. However, for forests (except EBF) the results suggest that photosynthetic light acclimation may emerge more due to the increase in LNC as we did not detect a positive response of fLNR to light (Fig. 3c).
Soil characteristics have been reported to influence Amax and LNC37, but we found no studies that have examined the impact of soil characteristics on fLNR. Among the eight soil properties we examined, we found positive responses of fLNR to soil pH and soil sand percentage, followed by small influences of bulk density and silt for certain ecosystems (i.e., croplands, needle leaf forests). pH influences the ability of soil to hold on to nutrients, including Ca2+, K2+, and Mg2+, that are essential to plant growth. A higher pH means more available nutrient cations as acid soils replace nutrient cations with H+. Several studies have reported a positive effect of pH on Amax37, non-temperature standardized (V_{c_{max}})20, and LNC39. Soil sand percentage had a positive impact on fLNR, possibly because sandy soils tend to be less fertile53 and thus stimulate plants to use their nitrogen more efficiently for photosynthesis and growth. The global influence of soil on fLNR was generally smaller than leaf traits and climate, but our analysis indicated that on 11.9% of the vegetated surface, soil characteristics contributed more than 15% of the changes in fLNR (Fig. 3a).
Notably, our study found that the soil nitrogen content has a limited impact on the spatial variation of fLNR (Fig. 3). The result implies that processes such as nitrogen deposition/addition are unlikely to affect plants fLNR. The soil nitrogen map we used was upscaled from ground observations of soil profiles in the World Soil Information Service (WoSIS) database. About 47.4–81.4% of the soil profiles in WoSIS are collected from the 1980s to 2020s54, when there were strong N deposition effects55. Therefore, we expect the N deposition effect has been implicitly included in our analysis. We acknowledge that some studies have suggested N deposition influenced leaf nitrogen content and photosynthesis56,57, however, the influence is limited to certain biomes, deposition load range, and time after the deposition. It is unclear whether these localized and time-dependent effects can influence the global variation of fLNR.
Uncertainty in the derivation of fLNR and ({V}_{{c}_{{max}}})
fLNR was derived based on Eq. (1) (see “Methods”) that mechanistically links ({V}_{{c}_{{max }}}^{25}), LNC, and fLNR, with the assumption that specific activity of RuBisCO (α25) and mass ratio of RuBisCO to nitrogen (fNR) are relatively constant values. The average uncertainty of RF fLNR was about 4.20 ± 2.20% (Supplementary Fig. 3). The uncertainty of fLNR was propagated from several sources including RF ({V}_{{c}_{{ma}x}}^{25}), α25, fNR, and LNC (Supplementary Fig. 3). Among them, the α25 ranges between 47.34 and 60 μmol CO2/g RuBisCO/s, and fNR ranges between 6.11 and 7.16 g RuBisCO/g N4. Our uncertainty test showed that the influence of α25 and fNR uncertainties on global fLNR were only around 1.13 ± 0.39% and 0.80 ± 0.27%, respectively (see “Methods”; Supplementary Fig. 3). Physiologically, α25 is a value that reflects the change in active sites of RuBisCO and the kinetic constant of the enzyme RuBisCO (k25). The number of active sites of RuBisCO is often regarded as a fixed value (set at 6 × 1023/mol RuBisCO) for vegetation on the land surface5, but there are reports showing that k25 varies with species9, leaf ages58, and temperature59. While these dependencies are elusive due to limited observations, previous studies have reported that k25 negatively correlates with LNC60 and LMA61. The negative relationship between k25 and LMA or LNC is potentially caused by the relatively lower drawdown of CO2 from intercellular spaces to the chloroplast as increased LMA increases mesophyll resistance. In that case, the negative dependence of k25 and α25 on LNC and LMA might account for part of the negative dependence of fLNR on LMA that we found (Fig. 3b), though the negative influence of LMA on α25 was weak and within the range of uncertainty, we quantified (Supplementary Fig. 3).
Compared to α25 and fNR, the uncertainties in LNC and RF ({V}_{{c}_{{max }}}^{25}) incurred larger uncertainties in fLNR. We found that LNC alone caused changes of 3.35 ± 2.16% in fLNR and RF ({V}_{{c}_{{max }}}^{25}) caused 3.13 ± 1.50% (Supplementary Fig. 3). Our study is the first attempt to upscale in situ ({V}_{{c}_{{max }}}^{25}) to the globe using remote sensing, while similar studies have done that for other leaf traits33. The observations used for training RF were densely distributed in Europe and North America, while inner Asia, Southeast Asia, Africa, and high-latitude regions are much less constrained by observations (Supplementary Fig. 6a). In addition, we did not consider temperature acclimation when standardizing in situ (V_{c_{max}}) to ({V}_{{c}_{{max }}}^{25}) (Eq. (2)), in order to facilitate the comparison with models that only estimate ({V}_{{c}_{{max }}}^{25}). However, the uncertainty related to temperature scaling should be limited as acclimated and non-acclimated temperature scaling factors for (V_{c_{max}}) are similar under 30 °C62,63.
The choice of an LNC map is another source of uncertainty in the derivation of fLNR. There are several global LNC maps available other than the EB1728 map we used, namely AMM1833 and CB2031. Each product has been validated in their respective studies (Supplementary Table 3). To examine the uncertainty incurred by the choice of LNC maps, we calculated fLNR using each of the three LNC maps. The three resulting fLNR maps show similar spatial patterns (Supplementary Fig. 10), with the spatial correlation coefficients (r) between them ranging from 0.57 to 0.71 (p < 0.01). Examining the influences of environmental variables on fLNR, we found the fLNR based on EB17 and AMM18 demonstrated similar results—fLNR was primarily influenced by LMA, LPC, VPD, PAR, and soil pH. Meanwhile, the fLNR based on CB20 was mostly influenced by soil pH, LMA, VPD, air temperature, and soil sand percentage. Noting that the CB20 LNC map has lower R2 in its cross-validation compared to that of EB17 and AMM18 (Supplementary Table 3), we have more confidence in the fLNR maps based on EB17 and AMM18. In our study, we used EB17 as the principal LNC map since it demonstrated the highest R2 in validation (Supplementary Table 3) and it was more consistent with the other two LNC maps than AMM18 (Supplementary Fig. 9). We acknowledge that the AMM18 LNC map has a smaller RMSE in validation compared to EB17 though it has a slightly lower R2 (Supplementary Table 3). In this study, we do not identify which LNC map is more accurate but show that the choice between EB17 and AMM18 has a limited influence on our conclusion regarding the dominant controls for fLNR (Supplementary Fig. 10g, h).
Implication for modeling photosynthesis in ecosystem models
The accurate simulation of fLNR provides a reliable constraint on vegetation photosynthesis, though fLNR is often not an explicit variable in ({V}_{{c}_{{max }}}^{25}) and ecosystem models4. Our results suggest that the conventional PFT-specific method is not effective due to the large variation in fLNR within PFTs. The development of optimal ({V}_{{c}_{{max }}}^{25}) models is a step forward in estimating spatially varying fLNR, though they currently also show some degree of inaccuracies. We propose two directions moving forward:
(1) Improve optimal ({V}_{{c}_{{max }}}^{25}) models. We found that optimality models demonstrated some promising strengths, e.g., LUNA and EO detected the coordination between leaf traits and fLNR, however, they demonstrated different responses to climate and soil. The difference between LUNA and EO fLNR is perhaps due to two reasons: (a) the different optimization approach taken and (b) the representativeness of training and validation datasets.
First, the EO model estimates photosynthetic capacity based on the first principles of photosynthesis, that plants minimize the relative carbon and water costs of photosynthesis per unit carbon assimilated while coordinating the electron transport rate-limited and RuBisCO-limited rates of photosynthesis to maximize photosynthesis while minimizing enzymatic and water costs20. While the LUNA model dynamically adjusts the fraction of nitrogen invested in different components of photosynthetic and metabolic processes in leaves (i.e., light capture, electron transport, carboxylation, and respiration) to maximize net photosynthesis (gross photosynthesis—photorespiration —dark respiration)22,23. LUNA and EO adopt cost functions that are formulated by different biological and environmental constraints to reach also slightly different goals of carbon gain (see “Methods”).
Second, our results suggest that subsets of global ({V}_{{c}_{{max }}}^{25}) demonstrate very different sensitivities to biological and environmental factors (Supplementary Fig. 5). For example, the ({V}_{{c}_{{max }}}^{25}) dataset used to parameterize the LUNA model identified PFT, LMA, LPC, and cation exchange capacity as the critical factors for ({V}_{{c}_{{max }}}^{25}), while the NS19 dataset20 was used to validate the EO model identified PFT, LNC, soil water content, and air temperature (Supplementary Fig. 5). The dataset we compiled combines the NS19, TRY, and LUNA datasets, and shows that leaf chlorophyll content, PFT, precipitation, and soil pH are the most important factors. Therefore, the choice of a subset for training model parameters might result in model biases—e.g., the underperformance of LUNA was likely caused by the relatively small dataset that was used for its parameterization rather than inherent issues of its structure. This thus suggests that using a more representative dataset of ({V}_{{c}_{{max }}}^{25}), such as the one used here, might improve optimal ({V}_{{c}_{{max }}}^{25}) model performance.
(2) Estimate fLNR empirically using key predictors. Based on the dominant controls we identified for fLNR, it is feasible to develop an empirical equation to estimate fLNR and then ({V}_{{c}_{{max }}}^{25}). fLNR is primarily determined by the coordination of leaf traits, which is represented by a negative relationship between fLNR and LMA, and a positive relationship between fLNR and LPC. The effects of VPD, PAR, soil pH, and sand also need to be considered (Fig. 3b). By adopting a multivariate linear model, we obtained fLNR = −0.19LMA + 42.2LPC + 4.76VPD + 0.25 pH + 0.0026PAR + 0.032sand + 2.4 for global vegetation and specific fLNR equations for each PFT (see “Methods”; Supplementary Table 2). In fact, EM3 and EM4 have attempted to use LNC and LPC to adjust fLNR15, while the low-biased fLNR from EM3 and EM4 might have resulted from the relatively small dataset of phosphorus-limited observations used to derive the model. Our results suggest that empirical ({V}_{{c}_{{max }}}^{25}) models can be improved by including climate and soil factors.
In addition, the data-driven ({V}_{{c}_{{max }}}^{25}) map provides a direct constraint on the spatial variations of vegetation photosynthetic capacity. We release the ({V}_{{c}_{{max }}}^{25}) map and its associated uncertainty (i.e., one standard deviation of estimates from bagged trees in the RF) to facilitate large-scale ecological and modeling studies (Supplementary Fig. 8). Since the remote sensing retrieval of leaf chlorophyll content is for top leaves35, and the phenological stages of trait samples are not often available, the ({V}_{{c}_{{max }}}^{25}) estimated in our study can be interpreted as a multi-year average ({V}_{{c}_{{max }}}^{25}) for top canopy leaves. The seasonality and the within-canopy variations of ({V}_{{c}_{{max }}}^{25},) are not accounted for in the map.
In conclusion, we have used observations from a range of sources to develop data-driven global maps of ({V}_{{c}_{{max }}}^{25}) and fLNR, which are critical to understanding vegetation nitrogen use and reducing the uncertainty in estimates of photosynthetic carbon assimilation. We find that the global distribution of fLNR is largely determined by LMA and LPC, in concordance with the leaf economic spectrum, as well as regional climate (i.e., VPD and PAR) and soil properties (i.e., pH and sand fraction). The new understanding and data presented in this study allow for future benchmarking and improvement of ({V}_{{c}_{{max }}}^{25}) and photosynthesis models, and provide insight into nitrogen use strategies of global vegetation.
Source: Ecology - nature.com
