Leaf structure (specific leaf area) modulates photosynthesis–nitrogen relations: evidence from within and across species and functional groups



1. Net photosynthetic capacity (Amax, defined as light-saturated net photosynthesis under near optimal ambient environmental conditions) of mature leaves often depends on the level of leaf nitrogen (N), but an assortment of relationships between these variables has been observed in studies of diverse plant species. Variation in leaf structure has been identified as an important factor associated with differences between the area- and mass-based expressions of the AmaxN relationship. In this paper we test the hypothesis that leaf structure, quantified using a measure of leaf area displayed per unit dry mass invested [specific leaf area (SLA)], is more than just a conversion factor, but itself can influence AmaxN relationships. We test this using several kinds of comparisons, based on field data for 107 species from sites representing six biomes and on literature data for 162 species from an equally diverse set of biomes.

2. Species and genera with thicker and/or denser leaves (lower SLA) consistently have flatter slopes of the AmaxN (mass-based) relationship than those with higher SLA. These and all other contrasts usually applied as well using area-based expressions, although such relationships were less consistent and weaker overall. A steeper slope indicates greater incremental change in Amax per unit variation in N.

3. Functional groups (e.g. needle-leafed evergreen trees, broad-leafed trees or shrubs, forbs) show the same patterns: groups with lower SLA have lower AmaxN slopes. Functional groups differ in mean leaf traits as well as in AmaxN relationships. Forbs have the highest SLA and mass-based N and Amax, followed by deciduous species (whether needle-leafed or broad-leafed, shrub or tree), with lowest values in evergreen species (again regardless of leaf type or functional group).

4. Interspecific variation in mass-based Amax is highly significantly related to the combination of leaf N and SLA (r2 = 0·86). At any value of leaf N, Amax increases with increasing SLA and at any value of SLA, Amax increases with increasing leaf N. Because this relationship, between Amax and the combination of N and SLA, is similar in two independent data sets, and as well, across broad taxonomic and geographic gradients, we hypothesize that it is universal in nature. Therefore, for broad interspecific contrasts among dicotyledons in any biome, we can reasonably well predict Amax based on the combination of SLA and leaf N. These findings have important implications for convergent evolution of leaf adaptation and great potential utility in models of global vegetation functioning.


The general association of net photosynthetic capacity (Amax) with leaf N levels within and among wild species has become well accepted (Mooney et al. 1981; Field & Mooney 1986; Reich, Uhl et al. 1991; Reich, Walters & Ellsworth 1991, 1992, 1997). Amax has been defined in these studies as light-saturated net photosynthetic rate under near optimal environmental conditions, including ambient concentrations of CO2. The physiological basis for the Amax–leaf N relationship involves the central role of N associated with the amount of ribulose-1,5-bisphosphate carboxylase oxygenase, as well as the role of N in other photosynthetically important leaf constituents (Mooney 1986). Because variation among C3 species in Amax appears to be largely owing to differences in biochemistry rather than to differences in CO2 supply (Field & Mooney 1986), variation in Amax among species should also reflect similar variation in related parameters such as Vcmax (the maximum rate of carboxylation). Thus, the so-called photosynthesis–N relationship has important implications, ranging from our understanding of the physiology of leaf function to our ability to predict global carbon balance patterns based on remote sensing of canopy chemistry and many other areas in between.

It has been apparent for at least several decades (El-Sharkawy & Hesketh 1965; Field & Mooney 1986) that variation in leaf structure, often quantified using specific leaf area (SLA, a measure of projected leaf area per unit dry mass), is related to patterns of variation in mass- and area-based photosynthesis–N relations. However, the linkage of SLA to N and Amax has not yet been fully characterized either within narrow or among broad species groups. By definition, leaves will have a lower SLA if they are denser (greater mass per volume) or thicker. Natural variation in both thickness and density has been shown to be responsible for variation in SLA across species (Abrams, Kubiskie & Mosteller 1994; Garnier & Laurent 1994). Prior studies have shown that species with higher SLA generally have higher mass-based N (Nmass) and Amax (Field & Mooney 1986; Reich, Walters & Ellsworth 1991, 1992, 1997) and for tropical trees, a steeper slope of the relationship between Amax and N (Reich, Walters et al. 1994). A steeper slope of the AmaxN relationship was also observed to be associated with higher SLA in a contrast of temperate tree species in major functional groups (broad-leafed deciduous angiosperms vs needle-leafed evergreen gymnosperms; Reich et al. 1995). These patterns lead to the hypothesis that response of Amax to variation in leaf N is modulated by leaf structure, with thicker and/or denser leaves having both a lesser Amax per unit N (Field & Mooney 1986) and a smaller change in Amax per variation in N (Reich & Walters 1994). Species and/or individuals with thicker, denser leaves may have lower Amax per unit N for a variety of reasons, including (1) differential intraleaf allocation of N, (2) limitations to Amax/N owing to intraleaf shading (Terashima & Hirosaka 1995), (3) limited Amax/N owing to the slow intercellular diffusion of CO2 (Parkhust 1994) and (4) selective pressures towards low Amax for species with thick, dense leaves (Field & Mooney 1986; Reich, Walters & Ellsworth 1991, 1992).

Despite the general tendency for ‘thicker’-leafed species or individuals to have lower mass-based Amax, quantitative relationships describing general patterns of AmaxN–SLA relationships have yet to be developed. In this paper we first present data that demonstrate the way in which the slope of the AmaxN relationship changes with variation in SLA for comparisons among species, genera and functional groups. We then show how these relationships in aggregate allow good prediction of Amax from the combination of SLA and N. These findings should contribute to the development of an integrated understanding of the role of SLA in determining AmaxN relations, and of the ways in which leaf structure (SLA) and chemistry (N) combine to determine Amax on both mass and area bases. The interrelations between SLA and Amax also can help us better reconcile apparent differences between photosynthetic rates when they are expressed on either mass or area bases.

To address the issues outlined above we present data from several sources. We first use both published and new data to examine variation in intraspecific and intragenus AmaxN relationships. To compare interspecific relationships within and among broad functional groups, we then use a new field data set (107 species-site combinations) obtained from ecosystems in six distinct biomes that vary broadly in climate and vegetation type (ranging from alpine tundra to desert to tropical rain forest) combined with an independent 162 species data set based largely on published literature for species in a variety of ecosystems and biomes.

Materials and methods

We used two sources of data and several types of data sets in this study. Both literature and new field data were obtained for multiple species (as described below). One type of data set is a large multiple species data set made up of both literature-compiled and field-measured data where each species is represented once on a given site (and infrequently more than once in total). This data set was also divided into functional group subsets to allow examination of AmaxN–SLA relationships within functional groups. Another type of data set uses measurements of multiple leaves for a given species or genus, where data were sufficient to enable evaluation of the AmaxN relationship for that species or genus. Comparisons of these relationships were then made among species or genera. In such cases, variation in leaf N, SLA and Amax were the result of variation in leaf age, among plants, and/or across microenvironmental gradients.


To compare within-species or within-genera photosynthesis–N relationships among species and genera characterized by different leaf structure (Figs 1 and 2), we used our own published and unpublished data, and published data from the literature. Such data sets were required to be of a sufficient sample size to enable regression of intragenus or intraspecies AmaxN relations. To combine with our field data, for a broader interspecific comparison, mean Amax, N and/or SLA values of young mature leaves (per species on a given site) were compiled from the published literature for 162 species in 185 species-site combinations. The multiple species literature data set used in this paper is an expanded (more than doubled) version of part of a data set used in an earlier report that focused on linkages between leaf, whole plant and stand characteristics as they related to variation in leaf life span (Reich et al. 1992).

Figure 1.

. The relationship between mass-based photosynthetic capacity and leaf N for several temperate and tropical woody species and genera differing in SLA. Regression equations, other statistics and data sources are in Table 1.

Figure 2.

. Variation among species in the slope of their mass- and area-based AmaxN relationships, plotted against their variation in SLA. Regression relationships and statistics: mass-based AmaxN slope = 1·80 + 0·061 × SLA, n = 42, r2 = 0·50, P < 0·0001; area-based slope = 3·79 + 0·023 × SLA, n = 35, r2 = 0·22, P < 0·004.


We also collected field data to provide mean Amax,N and SLA values for numerous species from several functional groups at sites representing six biomes and several associated ecotones. The sites were selected to provide a wide range of environmental conditions and terrestrial species and ecosystem types. We studied conifers, hardwood trees and shrubs, and forbs, located in alpine tundra and/or open subalpine forest–alpine meadow transition at high elevations (3200–3500 m) in Colorado, USA. In southern Wisconsin, USA we studied prairie and forest understorey forbs, hardwood and coniferous forest tree species, and swamp and bog species. We studied temperate forest species, including a number of common forest understorey forbs, broad-leafed deciduous and evergreen hardwood and evergreen coniferous forest tree species at the Coweeta Hydrological Laboratory, Otto, NC, USA. On the lower coastal plain of South Carolina, USA, we studied species from upland pine-dominated forests and forested wetlands. Desert shrubland and Pinyon–Juniper woodland vegetation were studied in the Sevilleta National Wildlife Refuge, NM, USA. A tropical rain-forest site was located near San Carlos del Rio Negro, Venezuela, in the northern Amazon basin. A total of 24 species was studied in mature stands of three adjacent primary rain-forest communities and in secondary successional stands growing on one of the three primary communities (Reich, Uhl et al. 1991; Reich, Walters et al. 1994). More detailed information about these sites, species, selection criteria and measurement protocol is provided in a companion paper that focuses upon site/biome contrasts (Reich, Ellsworth et al. 1998).

Among sites, plants from four functional groups were studied (needle-leafed trees, broad-leafed trees, broad-leafed woody shrubs and broad-leafed herbaceous species). Species from the first three groups were further subdivided by leaf habit (evergreen/deciduous) and leaf longevity into two classes: (1) deciduous or evergreen with leaf life span < 1 year or (2) evergreen with leaf life span > 1 year. In all but the tropical site this was equivalent to a pure evergreen/deciduous split.

Because leaf life span varies substantially among species, and leaf traits within species vary with leaf age as well (e.g. Field & Mooney 1983; Reich, Uhl et al. 1991), contrasts of photosynthetic rates, N concentrations and SLA among species were made using leaves of a similar ‘physiological’ age rather than a similar chronological age. We used fully expanded young to medium-aged leaves of all species, which corresponds to the period when many leaf traits are relatively stable (Reich, Uhl et al. 1991;Reich, Walters & Ellsworth 1991). To minimize the potentially confounding influence of shade on SLA and Amax, where possible we selected ‘sun’ leaves growing in relatively open conditions for all species at all sites. Measurements were made on open-grown plants in all herbaceous dominated communities and usually were made for open-grown trees or shrubs, or saplings or young trees in gaps, or for mature trees in the upper canopy. Although leaf light microenvironment has a large impact on leaf traits, especially SLA (e.g. Ellsworth & Reich 1992), the interspecific differences in leaf traits in this study were large enough (often 25–50-fold) that smaller intraspecific differences owing to variation in leaf or whole-plant microenvironment (usually less than twofold, Ellsworth & Reich 1992, 1993) probably would not have significant impact on the results.

Simultaneous measurements of photosynthetic CO2 assimilation and leaf water vapour conductance were made under ambient conditions with a portable leaf chamber and infra-red gas analyser operated in the differential mode (ADC model LCA-2, Hoddesdon, Herts., England). Measurements were made at mid- to late-morning (08.00–11.00 h local time) when the following conditions were met: near full sunlight, non-limiting vapour pressure deficits or temperatures. Thus, sampling was designed so that measurements were taken to reflect closely light-saturated leaf photosynthetic capacity in the field at ambient CO2 concentration (Reich, Uhl et al. 1991; Reich, Walters & Ellsworth 1991; Ellsworth & Reich 1992). Values of Amax were similar for plants measured in the field under near optimal ambient environmental conditions as for leaves measured under optimal steady-state conditions (Ellsworth & Reich 1992). We took at least 10 (but usually more) measurements per species from several individuals at each site, then averaged these for subsequent analyses.

After measuring gas-exchange rates, foliage was harvested. In some instances, the silhouette of foliage was fixed onto diazo paper in sunlight or traced by hand. The projected surface area of either the leaf tissue or its silhouette was assessed by a digital image analysis system (Decagon Instruments, Pullman, WA, USA). Total surface area was also calculated for all species based on their known shapes. The results of this study were similar if SLA was estimated using total rather than projected surface area (owing to enormous interspecific variation), although the quantitative relations differ slightly, because twice the projected surface area of needles underestimates their total surface area. Given that projected area was measured in our study (and reported in most published papers), while total surface area was estimated indirectly, data are expressed on a projected area basis. SLA is by definition related to the combination of leaf thickness and density, and has been shown to be correlated with one or both (Abrams et al. 1994; Garnier & Laurent 1994). Henceforth in this paper we will use the terms leaf thickness, density and SLA to convey roughly the same information.


The main interspecific data set was ‘species-based’ (257 species and 292 observations). In statistical analyses and the plotted data, individual data points represent a single species at a single site, using data averaged from all measurements made for that species-site combination. Thus, species (except for Figs 1 and 2) are represented only once from a given site. However, if a given species (n = 26) was examined in different locations, then it was included more than once in the data set.

Although transformations (logarithmic) were required in order to analyse properly the entire (or site-specific) interspecific field data set (Reich et al. 1997; Reich, Ellsworth et al. 1998), this was much less true for relationships within individual species, genera or functional groups, because their Amax data were generally near normally distributed, whereas for the entire interspecific data set they were not. In addition, funnel-shaped patterns of heteroscedasticity were much more pronounced in linear contrasts that crossed the entire range of plant functional types (Reich et al. 1997, Reich, Ellsworth et al. 1998) than when such contrasts were made for species or groups of species with comparable traits, as in this study. Results and their implications are similar in any case (data not shown). Thus, in this paper we largely use untransformed data typical of the literature in this field (Mooney et al. 1981; Field & Mooney 1986; Chazdon & Field 1987; Reich, Walters & Ellsworth. 1991; Ellsworth & Reich 1993). However, in analysing for net photosynthesis as a function of combined leaf traits using the entire data set, we used transformed data because the data were not even close to being normally distributed (Shapiro–Wilk W) and there was patterned heteroscedasticity in the residuals.

Data were analysed using linear and multiple regression (JMP Statistical Software, SAS Institute). We statistically compared relationships between functional groups and other species groupings using linear contrasts (separate and same slopes analyses). We used this technique to test the hypothesis that different equations describe these relationships in different functional groups. The projected or total surface area of entire individual leaves was either not significantly or weakly related to leaf traits at individual sites, for functional groups, or using pooled data. Thus, total area per leaf does not appear to be a particularly important leaf trait overall, compared with the others in this analysis and is not mentioned further.

For the main interspecific comparisons and analyses of this paper (Tables 2, 3 and 4, Figs 3 and 4), we used a data set that combined our own field data with the comparable data obtained from the literature. This data set included a total of 257 species, with 26 species measured in more than one site (for a total of 292 species-site combinations). Not all leaf traits of interest were measured in every case, thus sample sizes for specific analyses were somewhat lower (see Tables and Figures). Results were similar in both data sets (P > 0·05 for tests of data set differences) when analysed separately (see also Reich et al. 1997). Our field data set is also used in a companion paper (Reich, Ellsworth et al. 1998) that has the separate objective of asking whether or not interspecific leaf trait relationships are similar among biomes characterized by large differences in vegetation type and climate. In the present paper we use the combined data sets to address questions about the variation in AmaxN relations associated with variation in leaf structure, and about whether such relationships are similar among different plant functional groups. By combining the data sets we obtain a larger data set which enables more comprehensive comparisons of species arrayed into several different kinds of functional groups than otherwise possible.

Table 2.  . Mean values (one standard deviation, SD) for leaf traits within the following functional groups for data pooled geographically: forbs, broad-leafed shrubs, broad-leafed trees and needle-leafed trees. The latter three functional groups are further subdivided into two classes: (1) deciduous or evergreen with leaf life span < 1 year (deciduous or LL < 1 year) or (2) evergreen and with leaf life span > 1 year (evergreen and LL > 1 year) Thumbnail image of
Table 3.  . Regression statistics describing the relationship between Amax and leaf nitrogen content for several species groups. For mass-based regressions the dependent variable is Amass (nmol g–1 s–1) and the independent variable is Nmass (mg g–1). For area-based regressions, the dependent variable is Aarea (μmol m–2 s–1) and the independent variable is Narea (g m–2). Regressions are presented for the following functional groups for data pooled geographically: forbs, broad-leafed shrubs, broad-leafed trees and needle-leafed trees. The broad-leafed woody functional groups are further subdivided into two classes: (1) deciduous or evergreen with leaf life span < 1 year (deciduous or LL < 1 year) or (2) evergreen and with leaf life span > 1 year (evergreen and LL > 1 year). There were insufficient data available for deciduous needle-leafed trees for this purpose Thumbnail image of
Table 4.  . Regression statistics describing the relationship between Amax and leaf nitrogen content for species grouped into SLA classes (pooled among functional groups). For mass-based regressions the dependent variable is Amass (nmol g–1 s–1) and the independent variable is Nmass (mg g–1). For area-based regressions, the dependent variable is Aarea (μmol m–2 s–1) and the independent variable is Narea (g m–2). Species arranged by the following leaf structural classes: < 40, 40·1–90, 90·1–130, > 130 specific leaf area (cm2 g–1) Thumbnail image of
Figure 3.

. The interspecific relationship between photosynthetic capacity (Amax) and leaf N, on mass and area bases, for species within several functional groups. Regression equations, functional group definitions, and other statistics are in Tables 2 and 3. Closed circles represent the deciduous/shorter leaf life span subgroup and open symbols the evergreen/longer leaf life span subgroup for the two broad-leafed woody functional groups. For the mass-based relations, the axes differ among groups (otherwise data would be obscure in some groups), but the proportional scaling is constant (thus slopes can be compared). For the area-based relations, axes are identical except for the x-axis for the needle-leafed evergreens.

Figure 4.

. The relationship between mass-based leaf N (mg g–1, Nmass) and specific leaf area (cm2 g–1, SLA) for species in several functional groups. Regression relationships and statistics: for needle-leafed trees, Nmass = 9·47 + 0·054 × SLA, n = 32, r2 = 0·12, P < 0·05; (for all needle-leafed species, Nmass = 8·91 + 0·069 × SLA, n = 35, r2 = 0·34, P < 0·0003); for broad-leafed trees: Nmass = 8·41 + 0·099 × SLA, n = 72, r2 = 0·46, P < 0·0001; for broad-leafed shrubs: Nmass = 9·30 + 0·078 × SLA, n = 50, r2 = 0·52, P < 0·0001; for herbaceous species: Nmass = 20·25 + 0·077 × SLA, n = 54, r2 = 0·24, P < 0·0001.


Variation in the AmaxN relationship shows a consistent pattern in relation to variation in SLA regardless of whether intra- or interspecific or functional group relations are examined. When variation in intraspecific mass-based AmaxN relations is examined among woody species, we find that species with higher SLA tend to have higher Amax and N, a higher Amax per unit N at comparable N and a steeper slope of Amaxvs N (Fig. 1, Table 1). Tree species with long-lived leaves, such as Juniperus in North America and Pinus in Europe (Fig. 1), and Licania and Micrandra in South America (data not shown, see Reich, Walters et al. 1994), have dense, thick foliage (low SLA) and low mean mass-based N and Amax. They also have a low slope of the AmaxN relationship (Fig. 1). Species with intermediate SLA, such as Acer rubrum in North America and Vismia japurensis in South America, have intermediate mean mass-based N and Amax and slope of AmaxN. Fast-growing species with high SLA, such as the pioneer Cecropia in South America, have the steepest slope of AmaxN. The differences among species in the slope of the AmaxN relationship are large (> 50-fold).

Table 1.  . Regression statistics for several species and genera describing the mass-based relationship between Amax and leaf nitrogen concentration. The dependent variable is Amass (nmol g–1 s–1) and the independent variable is Nmass (mg g–1). Within-species relationships are based on data from numerous individuals and leaves from a single site. Relationships within a genus were based on data from numerous individuals and leaves of (1) six species (Piper), (2) seven species, including 14 different populations of one species (Quercus) and (3) 14 species, including eight populations within one species (Pinus). All regressions significant at P < 0·0001 Thumbnail image of

A similar pattern occurs when mass-based AmaxN relationships are compared among genera that differ in SLA. For example, among three genera (each with data for six to eight species), the slope of the mass-based AmaxN regression increased with SLA (Table 1, Fig. 1). Pinus sp. (SLA, range from 15 to 90 cm2 g–1), had a lesser slope than Quercus sp. (SLA range 65–135 cm2 g–1) and Piper sp. (SLA range 200–300 cm2 g–1) had the greatest slope.

In attempting a broader assessment of variation in intraspecific AmaxN relations, we regressed the AmaxN slope derived from individual species relationships (n = 42) against species variation in mean SLA. Variation among species in their mass-based AmaxN slope is significantly related to their differences in SLA (Fig. 2a). On average, the AmaxN slope varies 10-fold among species across the range of SLA. The patterns of variation among species in their intraspecific AmaxN relationships (such as in Fig. 1) are also found on an area basis (data not shown, see Reich, Walters et al. 1994 for example of area-based multiple species contrasts) but differences among species are smaller and less consistently related to variation in SLA. For instance, broad variation among species in the area-based AmaxN relationship is significantly (P = 0·004) but weakly (r2 = 0·22) correlated with variation in SLA (Fig. 2b). Although it may appear that the slope of the AmaxN relationship should be similar on mass and area bases this in fact is not the case (see Discussion).

Mean leaf traits for functional groups were similar among biomes (data not shown). Hence, data are presented contrasting functional groups using data pooled from all biomes. Mean leaf traits differ markedly among functional groups, and there is also substantial variation within such groups (Table 2). Mean SLA and mass-based leaf N and Amax (Nmass and Amass, respectively) are much higher in herbaceous forbs than in any of the woody species groups. Within each of the woody functional groups (shrubs, broad-leafed trees and needle-leafed trees), further subdivisions based on leaf habit and longevity had clear differences: evergreen species with leaf life-spans greater than a year had lower SLA, Nmass and Amass than species that are either deciduous or are evergreen but have a leaf life span of less than a year (Table 2). In the broad-leafed woody groups (shrub and trees), subgroups differing in leaf habit and longevity did not differ in area-based leaf N (Narea) but did differ in area-based Amax (Aarea). For a similar leaf Narea, groups with higher SLA and leaf Nmass had higher Aarea. In contrast, for needle-leafed conifers, evergreen species had much higher Narea than deciduous species and both groups had similar Aarea, despite large differences in Amass.

When interspecific AmaxN relations are examined within the four broad functional groups (forbs, shrubs, broad-leafed trees and needle-leafed trees), a similar pattern emerges as for species and genera differing in SLA (Table 3). Within each broad functional group, the mass-based AmaxN relation is significant (P < 0·001, 0·54 < r2 < 0·66). At any given leaf Nmass, forbs have the highest Amass and needle-leafed trees the lowest, with broad-leafed trees and shrubs intermediate (Fig. 3). These differences mirror SLA differences among functional groups, with mean SLA high in forbs, intermediate in broad-leafed trees and shrubs, and lowest in needle-leafed trees. The regressions differ significantly in slope (P = 0·02). Forbs, with the highest mean SLA and Nmass, have the steepest slope of AmassNmass (Table 3). Broad-leafed trees and shrubs were intermediate in slope of AmassNmass and had similar mean and range of SLA and Nmass. Needle-leafed species had the lowest mean SLA and Nmass and the lowest slope of AmassNmass. Roughly similar conclusions can be drawn when Amax and N are expressed on an area basis (Fig. 3, Table 3). The area-based AmaxN slope is highest for the forbs, lowest for the conifers and intermediate for the broad-leaved woody plants. For all four groups the AmaxN fit is poorer on an area basis.

When AmaxN relationships are examined within functional groups further subdivided by leaf habit and longevity, similar patterns are seen (Fig. 3). Within the two woody broad-leafed groups, the mass-based slope of AmaxN is greater in the group with shorter-lived leaves with higher average SLA (Table 3, Fig. 3). For area-based relationships, these differences only are apparent in the broad-leafed tree group. For all species pooled, there is a strong AmaxN relationship on a mass basis (r2 = 0·73) and a very weak one on an area basis (r2 = 0·02).

When the total multiple species data set is divided into arbitrary SLA classes, ignoring species and functional groups, a similar pattern emerges as for comparisons of intraspecific and functional group relationships (Table 4). Species in the highest SLA class have a steeper slope of both mass- and area-based AmaxN relationships than those in lower SLA classes. On a mass basis, there were modest differences in the strength of the relationship, with the high SLA group having the tightest fit; i.e. highest r2 (0·51, the lowest was 0·28). In contrast, on an area basis, for the high SLA group there was a very strong correlation (r2 = 0·82), with weaker relationships for the intermediate SLA groups (r2 = 0·2–0·3) and essentially an almost flat line relationship in the lowest SLA class (r2 = 0·1).

Further evidence for the role of leaf structure in AmaxN relationships is provided by examination of interspecific data on photosynthesis per unit nitrogen (Aleaf N). Using all data pooled, Aleaf N is more tightly related to SLA (r2 = 0·53) than with N itself (r2 = 0·16), despite the fact that N is incorporated as part of the measure of Aleaf N. Thus, if SLA does not differ among species but N does, Aleaf N will not differ as much as if SLA differs and N does not. This indicates that Aleaf N is more closely related to leaf structure than to the actual N level per se. Similar conclusions are reached based on multiple regression analysis where both Nmass and SLA are included as independent variables and Aleaf N is the dependent variable. Although both Nmass and SLA were significant (whole model r2 = 0·56), the Fstat for N was 6·9 and for SLA was 92·5, indicating a much greater proportion of the variation in Aleaf N can be attributed to variation in SLA than to N.

Leaf Nmass scales linearly with SLA for all data pooled (P < 0·0001, r2 = 0·54) or for species separated into functional groups (0·12 < r2 < 0·52) (Fig. 4). The slopes of the Nmass–SLA relationship did not differ significantly (P < 0·6) among functional groups (separate slopes analyses) but there was a significant difference (P < 0·001) in the intercepts of these relationships (tested after removing the interaction term from the model). The difference in intercept indicates that for any given SLA, leaf Nmass tends to be highest in forbs and lowest in conifers.

Multiple regression analyses were conducted for both area- and mass-based Amax. Using multiple regression, Amass was highly significantly related to the combination of Nmass and SLA (Fig. 5, P < 0·001, r2 = 0·86). As either Nmass or SLA increases, so does Amass. Thus, predicting Amass from the combination of SLA and leaf N effectively captures leaf structure and chemistry as the predominant and underlying sources of variation in Amass (Fig. 6). Aarea was significantly related to the combination of Narea and SLA, but a much smaller fraction of variation was explained (see Fig. 5 legend for equation, P < 0·001, r2 = 0·46). Increasing SLA or Narea are associated with increasing Aarea. Examination of observed vs predicted Amass values (Fig. 6) demonstrates that the combination of Nmass and SLA predicts Amass approximately equally well for species from four distinctly different functional groups. Recognition of the generality of joint SLA–N‘regulation’ of Amass across biomes and functional groups should prove valuable as a conceptual framework and as a modelling tool.

Figure 5.

. Relationship between mass-based photosynthetic capacity (nmol g–1 s–1, Amass), mass-based leaf N (mg g–1, Nmass) and specific leaf area (cm2 g–1, SLA) for all species pooled, from multiple biomes and functional groups. For perspective, the data points are shown in relation to the plane representing the relationship between Amass, Nmass and SLA: log10Amass = – 0·66 + 0·844 × log10Nmass + 0·782 × log10 SLA; r2 = 0·86, n = 213; P < 0·0001. The interaction term was significant when included in the regression model, slightly lessening heteroscedasticity and marginally improving the amount of variation explained: log10Amass = – 2·03 + 2·019 × log10Nmass + 1·433 × log10 SLA – 0·551 × (log10 Nmass× log10 SLA); r2 = 0·87, n = 213, P < 0·0001. The area-based relationship (not shown) is log10Aarea = – 0·51 + 0·844 × log10Narea + 0·626 × log10 SLA; r2 = 0·46, n = 213; P < 0·0001.

Figure 6.

. Relationship between measured Amass and Amass predicted based on Nmass and SLA (as shown in gridded plane in Fig. 5, using the additive model) for species from four broad functional groups (see Fig. 5 legend for the equation without the interaction term).


These results indicate that species with higher SLA have a higher Amax per unit leaf N and also vary more in Amax per unit variation in N than those with lower SLA. This was true for comparisons of AmaxN relationships among species, among genera, among functional groups or for arbitrary SLA groupings. This suggests that SLA is more than just a conversion factor linking mass- and area-based expressions of Amax. Additionally, the combination of SLA and Nmass effectively predicted Amass within and among all plant groups analysed here. Thus, a simple model of Amass as a function of Nmass and SLA offers great promise for broad scale modelling. The use of area-based AmaxN relationships for global-scale modelling has been criticized because the variation in Aarea for a given Narea is so great (Woodward & Smith 1994). Clearly, use of mass-based, rather than area-based relationships, eliminates a substantial share of the unexplained variance; moreover, joint consideration of SLA and N would enable even better estimates of photosynthetic capacity.

Similarity in the AmassNmass relationship among species in several large data sets (see Field & Mooney 1986; Reich et al. 1992) supports the idea that this is a broad universal relationship among species. However, this broad relationship exists across only the entire array of plant species, because there are different AmassNmass relationships among individual species (i.e. comparing single species regressions, Figs 1 and 2, Table 1, see also Reich, Walters et al. 1994) and functional groups (Fig. 3, Table 3, see also Reich et al. 1995). Thus, owing to the influence of SLA on Amax per unit leaf N (Reich, Walters et al. 1994; Reich & Walters 1994), the well documented general linear Amax–leaf N relationship (Field, Merino & Mooney 1983; Field & Mooney 1986; Reich, Uhl et al. 1991; Reich et al. 1992; Reich 1993) although real, is in fact made up of a series of nested relationships with increasing slope as SLA (and usually leaf N) increase. This is true whether the contrasting AmaxN relationships are based on variation within individual species (Figs 1 and 2, see also Reich, Walters et al. 1994), functional groups (Fig. 3, see alsoReich et al. 1995) or among unrelated species placed into arbitrary SLA classes (Table 4).

Why is Amass related to Nmass? As previously well established, variation in Amass is related to variation in Nmass because of the central role of N in photosynthetic enzymes, other proteins and pigments (Field & Mooney 1986; Sage & Pearcy 1987; Evans 1989). Thus, within species or functional groups, variation in Amass often follows variation in Nmass (Field et al. 1983; Chazdon & Field 1987; Reich, Walters & Ellsworth 1991; Reich et al. 1995). Variation in Aarea also follows variation in Narea but primarily (1) when there is little variation in Nmass but marked variation in SLA (DeJong, Day & Johnson 1989; Ellsworth & Reich 1993), as is often the case across light microenvironments, or (2) when Nmass and SLA have no relationship or vary inversely among leaves, which occurs much more frequently within a species or a like group of species, than across widely divergent species, where Nmass and SLA usually vary in parallel (Reich et al. 1992, 1995; Reich, Walters et al. 1994).

For specific contrasts, the slopes of AmaxN relationships were sometimes much greater on a mass than an area basis (e.g. all species, forbs, shrubs, broad-leaved trees, needle-leaved trees, Table 3). Because the slope in each case describes the change in Amax per unit change in leaf N (μmol CO2 g N–1 s–1) it may seem intuitive that the slopes should be the same regardless of the basis of expression. In fact, the slopes of AmaxN relationships would be identical on mass and area bases if there was no variation in SLA associated with variation in leaf N (Reich & Walters 1994) but SLA tends to increase with increasing Nmass (see Fig. 4, also Reich & Walters 1994). As a result, the species with highest or lowest N is not necessarily the same on mass and area bases, and overall, the mass vs area AmaxN relationships assess different gradients of multiple leaf traits (detailed explanations are provided in Reich & Walters 1994; Reich, Walters et al. 1998).

Because the ratio of leaf intercellular CO2 (Ci) to ambient CO2 (Ca) concentration did not vary consistently in any pattern and was roughly similar among all species (Ci/Ca = 0·81 ± 0·005 for n = 96 species from our field data), we expect that variation in Amax under near optimal conditions was largely the result of differences in the biochemical efficiency of carboxylation rather than differences in CO2 supply to the intercellular air spaces. Yoshie (1986) also observed similar Ci/Ca ratios among a broad suite of species. Verifying this hypothesis requires future work, although preliminary measurements on a limited number of grassland and tree species does show a correspondence between Vcmax (measured at elevated CO2) and Amax (measured at ambient CO2 concentration) (D. S. Ellsworth, unpublished data).

Even for a given leaf Nmass, variation in Amass is related to variation in SLA. Why? The answer is not entirely clear, but it appears that Amass scales with SLA because decreasing SLA is associated with greater allocation of biomass to structural components of the leaf rather than metabolic components, with greater internal shading, and with potential diffusion limitations (Vitousek, Field & Matson 1990; Lloyd et al. 1992; Parkhurst 1994; Terashima & Hirosaka 1995). These restrict the potential for high Amass in thick or dense leaves (low SLA), because chloroplast stacking (Terashima & Hirosaka 1995) imposes a constraint owing to the need for light to get to all chloroplasts and gaseous diffusion limitations (Parkhurst 1994; Epron et al. 1995) impose a constraint owing to the need for CO2 to get to all internal cells. It is also possible that species that vary in SLA allocate N differentially to different leaf constituents, but data on this are scarce.

The combination of N and SLA enables good prediction of Amax among diverse species, functional groups and biomes. In essence, regardless of environment or genotype, for leaves of a given leaf Nmass, those that are thicker and/or denser will have lower photosynthesis per unit leaf N and hence lower Amass; whereas for leaves of a given SLA, those with higher Nmass will have also have higher photosynthesis per leaf N and hence higher Amass. The exact shape of the relationship is unclear. Increasing slopes of the linear AmaxN relations with increasing SLA (Figs 1, 2 and 3, Table 4) suggest a greater-than-additive interaction. However, multiple regression indicates a significant less-than-additive interaction, on either a logarithmic (see Fig. 5 legend) or linear basis (data not shown). It is not immediately clear how to reconcile these differences. These issues are being further explored using mixed models (A. Robinson and P. B. Reich, unpublished data).

Results shown in this paper have two main implications. First, recognition of the combined effects of Nmass and SLA on Amass should allow a clear understanding of the physiological variation in Amass, and a better ability to model it. Second, many ecosystem and global models currently have no consistent, quantitative way of identifying Amax for a given species or vegetation type. These values are in a sense estimated from the literature (Woodward & Smith 1994) or the modellers ‘best guess’ (e.g. Collatz et al. 1991; Melillo et al. 1993). In contrast, these same models often employ sophisticated physiological algorithms to model changes in realized net photosynthesis from the ‘bench-mark’ value of Amax: thus these models have a mismatch of model sophistication for variation in realized A, in contrast to the ability to generically generate a useful Amax. In the absence of detailed ACi curves, which are currently only available for a relatively small subset of species, largely grown in growth chambers (Wullschleger 1993), or of good means of estimating chloroplast CO2 concentrations (Epron et al. 1995), estimating Amax for field-grown plants at the ‘normal’ CO2 supply in ambient CO2 concentrations provides a useful step in bridging gaps between detailed physiological models and landscape scale information, such as canopy leaf N and vegetative community composition. The relationships shown in this paper could be used in physiologically based ecosystem and global-scale models to predict Amax for given species, functional types or community types with known or typical levels of SLA and/or N.


>We thank the numerous people who contributed in various ways to the collection of the data used in this paper. This research was partially supported by National Science Foundation Grants BSR 8819718, BSR 8857129, and IBN 9296005, and by the NSF Long-Term Ecological Research Program.