The carbon and nitrogen isotope composition of Australian grasses in relation to climate


*Correspondence author. E-mail:


1. The carbon and nitrogen isotope composition of plants are known to be affected by environmental factors, especially water availability. While δ13C of C4 plants is generally assumed to be much less responsive to environmental variables than that of C3 plants, it is unclear whether the response of δ15N differs between the two photosynthetic pathways.

2. Focussing on differences in the response of members of the C3 and C4 photosynthetic pathways to climate variables, we examined the foliar δ13C and δ15N of grasses collected from natural vegetation in a wide range of climatic zones throughout Australia.

3. An index of water availability was clearly related to foliar δ13C and δ15N of both C3 and C4 grasses. There was a negative relationship between water availability and δ13C of C3 grasses (r2 = 0·21), similar to that documented extensively in other C3 plants. An opposite, positive relationship was found between water availability and δ13C in C4 grasses (r2 = 0·17), an effect that has been reported only infrequently. Accounting for differences in δ13C between the different C4 types (NADP-ME, PCK, NAD-ME and Aristida) resulted in a substantial increase in the fit of the model relating δ13C to water availability (R2 = 0·48).

4. There was a negative relationship between water availability and δ15N (r2 = 0·40), that was similar in both C3 and C4 grasses, but temperature had no effect on δ15N. This provides strong support for the theory that water availability is the dominant factor determining soil and plant δ15N via its effect on the ‘openness’ of the nitrogen cycle.

5. We also found significant differences in δ15N between the different C4 types, with the highest values for PCK, followed by NADP-ME and NAD-ME, and the lowest values for Aristida. The difference in δ15N between PCK and Aristida was large, at 5·1‰.

6. The importance of water availability as a predictor of δ13C in both C3 and C4 grasses suggests that variation in water availability should be considered when estimating C4 grass abundance based on δ13C measurements (e.g. in herbivore diets or as a contributor to biospheric carbon sinks).


Clear patterns can be seen in the stable carbon and nitrogen isotope composition (expressed as δ13C and δ15N respectively) of natural vegetation throughout the world, closely reflecting climate and other environmental factors. Water availability appears to be the most important factor, usually showing a strong negative relationship with both δ13C (e.g. Stewart et al. 1995; Swap et al. 2004; Weiguo et al. 2005) and δ15N (e.g. Handley et al. 1999; Schuur & Matson 2001; Swap et al. 2004). In the case of δ13C however, the negative relationship with water availability is usually only present in plants using the C3 photosynthetic pathway. In C4 plants, δ13C is much less variable and relationships with climatic factors are usually absent (Van Der Water et al. 2002; Swap et al. 2004).

In C3 plants, δ13C is primarily influenced by the ratio of intercellular to ambient concentrations of CO2 (ci/ca) (Farquhar et al. 1989a). Plants regulate ci/ca by opening and closing their stomata, in response to a range of environmental factors, including light (Yakir & Israeli 1995), nutrients (Raven & Farquhar 1990) and water availability (Winter et al. 1982). A close relationship exists between ci/ca and plant water use efficiency (WUE), which means that δ13C can provide an estimate of the integrated long-term WUE of a plant (Ehleringer 1989; Farquhar et al. 1989b). The relationship between water availability and δ13C of natural vegetation arises in two ways. First, at an individual level, plants respond to decreasing water availability by increasing their WUE, and hence δ13C (Farquhar et al. 1989b). Secondly, plants adapted to more arid environments tend to have higher WUE, and hence δ13C, than plants adapted to more mesic environments, even when grown in the same environment (Anderson et al. 1996).

In C4 plants, the effect of ci/ca on δ13C is similar to that in C3 plants, but greatly diminished, and confounded by post-photosynthetic fractionation due to ‘leakiness’ of the bundle sheath cells to CO2 (Farquhar 1983). Depending on the degree of leakiness, the slope of the relationship between δ13C and water availability can range from positive (e.g. Weiguo et al. 2005) to negative (e.g. Wang et al. 2005), but in general, δ13C in C4 plants tends to vary much less in response to environmental factors than in C3 plants (Henderson et al. 1992). Perhaps the largest source of variation in δ13C in C4 plants is biochemical subtype, with C4 plants generally divided into three groups on the basis of biochemistry: NADP-ME, NAD-ME and PCK, named after their respective C4 acid decarboxylases. NADP-ME species tend to have the highest δ13C, followed by PCK and then NAD-ME species (Hattersley 1982).

The causes of variation in δ15N are much less clearly understood than the causes of variation in δ13C. Although extensive fractionation may occur within plants (Evans 2001), foliar δ15N tends to reflect soil δ15N (Austin & Vitousek 1998; Handley et al. 1999), and a number of authors have suggested that it is the ‘openness’ of the nitrogen cycle that primarily influences soil δ15N (Austin & Vitousek 1998, Handley et al. 1999, Schuur & Matson 2001). In an open nitrogen cycle, gains and losses of nitrogen are large relative to the total nitrogen pool. This is typical of arid areas, where water, rather than nitrogen, tends to be limiting. In more mesic areas, nitrogen, rather than water, tends to be limiting, such that it is efficiently recycled, with little leaving the nitrogen cycle.

Accurate prediction of foliar δ13C and δ15N has numerous applications in ecological studies. Values of δ13C of plant and animal remains can be used to estimate the proportion of plant biomass or diet that was C4, but typical values of δ13C for C3 and C4 plants must be estimated (Vogel 1978; Witt et al. 1998; Cerling et al. 2006; Codron et al. 2007). Accurate estimates of δ13C of C3 and C4 vegetation are also required for use in global carbon budgets that estimate the relative importance of terrestrial and marine carbon sinks from δ13C of atmospheric CO2 (Lloyd & Farquhar 1994; Fung et al. 1997; Suits et al. 2005). Values of δ15N of animal remains can be used to estimate trophic level, as animals tend to be enriched by 1–5‰ with each increasing trophic level (Hobson & Montevecchi 1991; Kwak & Zedler 1997). However, variation in the δ15N signature of plant material at the base of the trophic structure must be accounted for.

While most previous studies of variation in plant δ13C and δ15N have focused on woody plants, grasses provide an opportunity to directly compare the influence of the C3 and C4 photosynthetic pathways on the response of δ13C and δ15N to environmental factors, within a single functional group. This is not as readily achieved with woody plants due to the limited occurrence of C4 photosynthesis within this group. In this study, we examine the variation in foliar δ13C and δ15N of grasses throughout the Australian continent, in relation to a range of environmental factors. Given the existing evidence for a close relationship between water availability and both δ13C and δ15N in other plants, we focus primarily on this factor, and evaluate three hypotheses about isotopic variation in relation to water availability:

  • 1 there is strong a negative relationship between δ13C and water availability in C3 grasses;
  • 2 there is no relationship between δ13C and water availability in C4 grasses; and
  • 3 there is a negative relationship between δ15N and water availability, similar in both C3 and C4 grasses.

Materials and methods

Sample collection and analysis

Between March 2003 and November 2004, 408 grass specimens were collected at 173 locations across Australia, encompassing a wide range of climatic conditions, with locations accurately determined using a global positioning system (Fig. 1). The vegetation at the collection sites was relatively undisturbed; that is, the native woody vegetation had not been cleared. While many of the samples were collected on land subject to pastoral grazing by sheep and cattle, disturbance due to grazing was typically mild. We included exotic grasses in our analysis, although these represent only a very small proportion of specimens (Table 1).

Figure 1.

 Locations of grass specimen collections within Australia. C3 and C4 specimens are indicated (grey and black circles respectively). Contour lines refer to the water availability index (actual evapotranspiration/potential evapotranspiration).

Table 1.   Number of specimens examined, broken down by photosynthetic pathway and genus. Genera marked with an asterisk are naturalized

Samples of live foliage were collected, pressed in newspaper and air-dried. In about 25% of cases, no live foliage was available, so senescent foliage was collected. A number of previous studies suggest that both foliar δ13C and δ15N may change as leaves senesce, however reported trends are highly variable, ranging from negative (Currin et al. 1995; Cloern et al. 2002; Turney et al. 2002) to neutral (Garten 1993; Currin et al. 1995; Kolb & Evans 2002) to positive (Gebauer & Schulze 1991; Närholm 1991; Cloern et al. 2002) in various taxa. While variation in δ13C and δ15N due to senescence is clearly a potential source of error in this study, we consider it likely that this would simply increase the overall variability in the data, and not bias our results in any particular direction. Mature leaves were selected from each sample, rinsed with deionized water, dried and ground to a fine powder. Approximately 4 mg of each powdered foliage sample was placed in tin capsules and analysed using an EA 1110 elemental analyser (CE Instruments, Rodano, Italy) coupled with an IsoChrom mass spectrometer (Micromass, Manchester, UK). δ13C and δ15N are expressed in per mil (‰) units, relative to the Vienna Pee Dee Belemnite and atmospheric N2 standards respectively.

Specimens were classified as possessing either the C3 or C4 photosynthetic pathway (Table 1). If the specimen could be identified to generic level, pathway classification was made on the basis of that identification using Watson & Dallwitz (1992 onwards). If a specimen could not be identified to genus, classification was made on the basis of its foliar δ13C. Those with δ13C less than −19‰ were considered C3, and those with δ13C greater than −19‰ considered C4 (Cerling et al. 1997). While the abundance of C4 grasses, in terms of both species numbers and biomass, is sometimes assumed to be related to water availability, with greater abundance of C4 grasses in arid areas (Chazdon 1978), in Australia this is not the case. In Australia, C4 grasses are more closely associated with high growing season temperatures (Hattersley 1983; Murphy & Bowman 2007b).

Where possible, each C4 specimen was assigned to one of the three C4 subtypes (NADP-ME, NAD-ME, or PCK), using the lists provided in Schulze et al. (1996a) and Watson & Dallwitz (1992 onwards). The subtypes are based on physiology and leaf anatomical characteristics as defined by Prendergast et al. (1987). Following Schulze et al. (1996a), we also included a fourth subtype, comprising members of the genus Aristida. While this genus has typical NADP-ME physiology, it is anatomically distinct from other NADP-ME grasses, lacking a suberised lamella in its bundle sheath cells (Hattersley 1992).

Environmental data

We examined variation in foliar δ13C and δ15N in relation to a number of climatic and environmental variables: a soil water availability index (WAI), mean annual temperature (MAT), vegetation height and canopy cover, chenopod abundance and proximity to coast. The derivation of these variables, and the rationale behind their selection, is provided below. These variables were selected because they were: (i) the most likely to have large effects, based on the published literature and (ii) were easily obtained, and relatively accurate, for each collection location.

Digital maps of mean annual areal actual evapotranspiration (AET) and areal potential evapotranspiration (PET), covering Australia at a resolution of 0·05°, were obtained from the Australian Bureau of Meteorology, Canberra. These maps had been constructed following Morton (1983), and were used to estimate a soil WAI, calculated as: 1 − AET/PET (Walker & Langridge 1997). MAT for each collection location was estimated using the computer program ANUClim 5.1 (Centre for Resource and Environmental Studies, Australian National University, Canberra). As well as location coordinates, ANUClim requires elevations, which were obtained from a second resolution digital elevation model of Australia, obtained from the Australian Spatial Data Directory ( WAI ranged from 0·14 to 0·73, equivalent to a range in annual rainfall of 155–1605 mm, and MAT ranged from 6 to 26 °C.

Given that the presence of a tall, dense tree canopy can increase δ13C of understorey plants (Van Der Merwe & Medina 1989; Buchmann et al. 1997), we measured the average height of the surrounding vegetation at each collection location using a clinometer, and estimated canopy cover using a semi-quantitative index, ranging from 0 (no canopy) to 5 (overlapping crowns), based on crown separation ratios (mean gap/mean width: McDonald et al. 1998).

A useful indicator of site soil conditions in Australia is the presence of chenopod (family Chenopodiaceae)-dominated vegetation, as it is very closely associated with saline and sodic soils (Beadle 1981). These soil types could be expected to affect plant δ13C and δ15N (Heaton 1987; Brugnoli & Lauteri 1991; Van Groenigen & Van Kessel 2002). Each collection location was classified as either chenopod or non-chenopod dominated. Chenopod sites were those where >50% of the vegetation within a 1 km radius was described as ‘chenopod shrubland’, according to the National Vegetation Information System version 1.0 (Department of Environment and Heritage, Australia,

Given that previous studies have found elevated foliar δ15N in close proximity to the coast (Virginia & Delwiche 1982; Heaton 1987; but see Vitousek et al. 1989 for absence of an effect), we calculated the distance between each collection location and the nearest coastline using a geographic information system. Collection locations were classified as coastal if they were within 20 km of the coast. This distance was larger than that used in previous studies (8 km, Virginia & Delwiche 1982; 2 km, Heaton 1987), simply because we had few collection locations close to the coast.

Statistical analysis

We examined variation in δ13C and δ15N using least-squares linear regression models. Water availability was thought to be the variable of primary importance, so models containing only this variable were constructed first. To examine the effect of additional environmental variables, we added these to the models containing water availability. All models were constructed in the computer program R (version 2.6.2) (Ihaka & Gentleman 1996).

Given that δ13C is influenced by very different processes in C3 and C4 plants, we analysed C3 and C4 specimens separately (Farquhar 1983; Farquhar et al. 1989a). In contrast, we could find no published evidence that processes influencing δ15N systematically vary between C3 and C4 plants, and for this reason, we used a single analysis for δ15N. Visual inspection of foliar δ13C of C4 grasses and foliar δ15N plotted against WAI suggested a nonlinear relationship, so we applied an inverse transformation to WAI prior to any analysis, i.e. WAI−1. The entire data set used in the analysis is provided in the Supporting Information.


Grass foliar δ13C

There were highly significant relationships between water availability and foliar δ13C for both C3 (< 0·001, r2 = 0·21) and C4 grasses (< 0·0001, r2 = 0·17). In the case of C3 grasses, the relationship was negative, with a 2·8‰ difference between specimens from the wettest and driest sites (Fig. 2a). In contrast, the relationship was positive for C4 grasses, with a 1·7‰ difference between specimens from the wettest and driest sites.

Figure 2.

 (a) δ13C and (b) δ15N of grass foliage collected throughout Australia, plotted against water availability index (WAI). Filled circles represent C4 specimens; empty circles represent C3 specimens. The lines represent the predictions of linear regressions of (a) δ13C and (b) δ15N against WAI: the solid line is for C4 specimens; the dashed line is for C3 specimens. For δ13C, separate regressions were used for C3 and C4 specimens, with r2 of 0·17 and 0·21 respectively. For δ15N, a multiple regression model was used: δ15N ∼ WAI ×× photosynthetic pathway, for which R2 was 0·42.

Within C4 grasses, there were highly significant differences in δ13C between the four C4 types (< 0·0001). While the Aristida, NAD-ME and PCK types had similar δ13C, the NADP-ME type had δ13C about 1‰ higher than the other types (Fig. 3a). Given that the NADP-ME type tends to be associated with high rainfall areas (Hattersley 1992; Schulze et al. 1996a), there is a possibility that the high δ13C values typical of this type may have been at least partly responsible for the positive relationship between δ13C and water availability in C4 grasses. However, it is clear that the high δ13C values of the NADP-ME type were not entirely responsible, because a significant positive relationship was present in each of the C4 types, except PCK (NADP-ME < 0·0001; NAD-ME < 0·01; Aristida P < 0·01).

Figure 3.

 Modelled foliar (a) δ13C and (b) δ15N of grasses of the four C4 types examined, according to the linear regression model: δ13C or δ15N ∼ WAI−1 + C4 type. WAI was assumed to be 0·33, the mean of the data set. R2 for was 0·48 and 0·46 for δ13C and δ15N respectively. Standard errors are shown.

There was little evidence that the relationship between δ13C and water availability varied between the four C4 types, as the interaction between water availability and C4 type was not significant (= 0·82; Fig. 4). A simple linear model incorporating just water availability and C4 type was able to explain a high proportion (48%) of the variation in foliar δ13C of C4 grasses.

Figure 4.

 Foliar δ13C of grasses of the four C4 types examined: (a) NADP-ME, (b) PCK, (c) NAD-ME and (d) Aristida, plotted against water availability index (WAI). The lines represent the predictions of a least-squares linear regression model: δ13C ∼ WAI−1 + C4 type. R2 for the regression was 0·48.

Neither vegetation height nor canopy cover had an effect on δ13C of C3 grasses (= 0·98 and 0·22 respectively), but both had a clear negative effect on δ13C of C4 grasses (< 0·0001 and <0·01 respectively). For C4 grasses, δ13C decreased by 1·4‰ between the shortest and tallest vegetation, and by 0·75‰ between the most open and closed vegetation. The presence of chenopod-dominated vegetation had no clear effect on δ13C of C4 grasses (= 0·34).

Grass foliar δ15N

There was a highly significant negative relationship between water availability and foliar δ15N (< 0·0001; r2 = 0·40), with a 9·5‰ difference between specimens from the wettest and driest sites (Fig. 2b). Additionally, there was strong evidence that the slope of the relationship between δ15N and water availability was steeper in C3 compared with C4 grasses (< 0·01). Over the entire range of water availability examined, foliar δ15N of C3 grasses decreased by 4·4‰ more than that of C4 grasses (Fig. 2b).

Within C4 grasses, there were highly significant differences in δ15N between the four C4 types (< 0·0001). The most conspicuous difference was that the PCK type had substantially higher δ15N than the other three types (Fig. 3b). δ15N was lowest in Aristida and intermediate in the NADP-ME and NAD-ME types. The magnitude of the effect was large, with a 5·1‰ difference between the PCK and Aristida types for a given level of water availability (Fig. 3b). There was little evidence that the relationship between δ15N and water availability varied between the four C4 types, as the interaction between water availability and C4 type was not significant (= 0·07).

Of the other variables examined, only the effect of proximity to the coast was significant (< 0·001). Contrary to expectation, grasses from sites greater than 20 km from the coast had foliar δ15N about 2·1‰ higher than those from coastal sites, opposite to the effect reported in the literature (Virginia & Delwiche 1982, Heaton 1987). Inland sites generally have lower water availability than coastal sites, so it may be that the observed effect is an artefact of the relationship between water availability and foliar δ15N.


Foliar δ13C

A negative relationship between water availability and foliar δ13C of C3 species has been extensively documented in a range of environments (Stewart et al. 1995; Korol et al. 1999; Van Der Water et al. 2002; Swap et al. 2004; Weiguo et al. 2005). While most of these studies have focused on woody plants, our results show that a similar pattern is present in C3 grasses in Australia. Of greater significance, however, is the opposite, positive relationship we have clearly demonstrated between water availability and foliar δ13C of C4 grasses. This relationship has been only infrequently reported (Table 2), and certainly never on such a large geographical scale.

Table 2.   Summary of recently reported relationships between water availability (annual rainfall) and foliar δ13C and δ15N. Across each rainfall gradient examined, the differences in δ13C and/or δ15N are specified. Differences in δ13C and/or δ15N were often visually estimated from figures in the source literature, so are approximate only, rounded to the nearest 0·5‰
IsotopePathwayLife formsRainfall gradient (mm)Isotopic difference (‰)LocationReferences
δ13CC3Grasses220–640−4·5Chinese Loess PlateauWeiguo et al. (2005)
Trees, shrubs350–1520−4NE AustraliaStewart et al. (1995)
Trees, shrubs230–970−2·5S AfricaSwap et al. (2004)
Trees220–1800−2·5N AustraliaSchulze et al. (1998)
Trees630–1600−2SE Australia, New ZealandKorol et al. (1999)
Trees, shrubs160–560−3 to 0SW United StatesVan Der Water et al. (2002)
Trees, shrubs, forbs, grasses130–7700PatagoniaSchulze et al. (1996b)
C4Grasses50–5500 to 2NamibiaSchulze et al. (1996a)
Grasses230–6500S AfricaSwap et al. (2004)
Single shrub species160–2500SW United StatesVan Der Water et al. (2002)
Single grass species370–670−2Chinese Loess PlateauWeiguo et al. (2005)
δ15NC3Trees220–1800−5·5N AustraliaSchulze et al. (1998), Austin & Sala (1999)
Trees, shrubs200–1300−6·5S AfricaSwap et al. (2004)
Trees500–5500−4HawaiiAustin & Vitousek (1998)
C3 + C4All20–2950−9GlobalHandley et al. (1999)
C4Grasses200–1300−1S AfricaSwap et al. (2004)

A positive relationship between water availability and foliar δ13C of C4 plants is clearly supported by physiological theory. In C4 plants, the effect of ci/ca on δ13C is confounded by post-photosynthetic fractionation due to ‘leakiness’ of the bundle sheath cells to CO2 (Farquhar 1983). The degree of leakiness (Φ) affects the slope of the linear relationship between ci/ca and δ13C. When Φ is below about 0·35, δ13C is positively related to ci/ca, i.e. opposite to the relationship in C3 plants, and Φ tends to remain below this level under a wide range of environmental conditions and moderate levels of stress (Henderson et al. 1992; Williams et al. 2001). For this reason the relationship between water availability and δ13C of C4 species is usually positive, opposite to the direction of the relationship in C3 species (Schulze et al. 1996a; Wang et al. 2005).

Within the C4 photosynthetic pathway, we found that the NADP-ME type had δ13C substantially higher than the other three types. This is clearly consistent with the literature, with numerous previous studies reporting high δ13C values for NADP-ME species (Hattersley 1982; Schulze et al. 1996a; Cerling et al. 2003). Several previous studies have also reported that the PCK type tends to have slightly higher δ13C than the NAD-ME and Aristida types, although we did not detect this difference. Importantly, a model of foliar δ13C of C4 grasses, incorporating both C4 type and water availability, performed well (R2 = 0·48), substantially better than the best model of foliar δ13C of C3 grasses (r2 = 0·21). Given that the relative abundances of the four C4 types are closely related to water availability (Hattersley 1992; Schulze et al. 1996a), it is likely that mean foliar δ13C of C4 grasses in a given geographical area could be reliably estimated, and such estimates are commonly used in isotopic reconstructions of animal feeding ecology and global carbon budgets, described under Implications, below.

Numerous authors have suggested that the relationship between rainfall and foliar δ13C of C3 plants is nonlinear (Schulze et al. 1998; Korol et al. 1999; Miller et al. 2001; Van Der Water et al. 2002; Weiguo et al. 2005). In these studies, δ13C increased rapidly at low levels of rainfall and then levelled out at higher rainfall. From Australia, there have been conflicting results. In eastern Australia, Stewart et al. (1995) found a relationship between foliar δ13C of woody species and annual rainfall that was clearly linear. In contrast, over a similar rainfall gradient in monsoonal northern Australia both Schulze et al. (1998) and Miller et al. (2001) found that the relationship between foliar δ13C of Eucalyptus trees and annual rainfall clearly levelled out at about 500–600 mm. They suggested that this corresponded to the limit of a strong monsoonal influence, and above this level, soils tend to be uniformly saturated during the summer growing season when foliar δ13C is determined. The lack of C3 specimens from intermediate levels of water availability in our study makes it impossible to determine whether the relationship we observed between δ13C and water availability was also nonlinear. Given that our specimens were collected from a range of areas across the Australian continent, rather than from an unreplicated water availability gradient, it is unlikely that a single environmental feature, such as the limit of monsoon influence (e.g. Schulze et al. 1998; Miller et al. 2001), or a boundary between major soil formations (e.g. Van Der Water et al. 2002) would result in a nonlinear relationship.

A number of studies have shown that δ13C of plants growing beneath dense canopies tends to be lower than plants growing in more open vegetation. This has been termed the ‘canopy effect’ and has been attributed to both the 13C depletion of ambient CO2 during soil respiration (Van Der Merwe & Medina 1989) and reduced light availability (Buchmann et al. 1997). We found little evidence that vegetation height or canopy cover affected foliar δ13C of C3 grasses, but clearly that of C4 grasses was affected, with a 0·7-1·3‰ decrease in δ13C between the shortest/most open and tallest/most closed vegetation. This effect possibly reflects an increase in shade with increasing vegetation height, demonstrated experimentally by Buchmann et al. (1996). This explanation is more likely than 13C depletion of ambient CO2 due to soil respiration, which is typically associated with closed canopy vegetation that was not present in our data set. The effect is likely to have been detectable in C4 but not C3 grasses because of their much greater sample size (350 vs. 58).

Foliar δ15N

The negative relationship between water availability and foliar δ15N that we have found throughout Australia is similar to that already recorded in northern Australia (Schulze et al. 1998), on other continents (Heaton 1987; Schuur et al. 2001, Aranibar et al. 2004, Swap et al. 2004) and in an extensive global synthesis (Handley et al. 1999). Many authors now agree that the mechanism is the increased ‘openness’ of nitrogen cycling in drier areas (Handley et al. 1999; Schuur et al. 2001), a concept first developed by Austin & Vitousek (1998). These authors suggested that where water is abundant, nitrogen tends to be limiting, and is tightly recycled. This limits the amount of nitrogen leaving the system through fractionating pathways that would cause the system to become enriched in 15N. In contrast, nitrogen is not recycled so tightly in drier systems, because water is the limiting factor, and nitrogen leaves the system more readily through fractionating pathways.

The cause of the pattern in northern Australia has been the subject of debate. Schulze et al. (1998) suggested that increasing foliar δ15N with increasing aridity may be due to higher grazing pressure at arid sites, resulting in an acceleration of nitrogen turnover and losses of 14N, although this hypothesis has been disputed by Austin & Sala (1999). These authors suggest that grazing intensity is in fact greater at wetter sites, and that the pattern is more likely to be caused by the direct effects of water availability on the ‘openness’ of the nitrogen cycle, consistent with the hypothesis of Austin & Vitousek (1998). Also working in northern Australia, Cook (2001) added to the debate by demonstrating that grazing has only a small effect on foliar δ15N of grasses and trees, certainly not sufficient to cause the large differences (∼6‰) observed across the northern Australian rainfall gradient. Cook (2001) also suggested that fire may play an important role, with frequent fires, typical of the high rainfall end of the gradient, possibly resulting in greater losses of 14N, and hence high δ15N. However, that the relationship between foliar δ15N and water availability holds across such a diverse range of climatic and vegetation zones as examined in this study, clearly lends support to the mechanism proposed by Austin & Vitousek (1998) and Handley et al. (1999), that is, increased openness of the nitrogen cycle in arid areas.

Recent work in the tropical forests of Hawaii by Houtton et al. (2006) has provided additional insights into the mechanism by which 14N is lost from relatively dry sites. Measuring the isotopic composition of both inputs and hydrological outputs, these authors found that hydrological outputs from drier sites were actually less enriched than inputs, suggesting that 14N must be preferentially lost via another pathway, presumed to be gaseous loss via denitrification of NO3. While nitrogen losses via denitrification were greater at wetter sites, more complete consumption of NO3 was assumed to minimize ecosystem enrichment with 15N. It remains unclear whether the mechanism proposed by Houtton et al. (2006) is also applicable outside tropical forest biomes, and particularly whether denitrification remains an important pathway of nitrogen loss in arid areas, where the globally maximum values of ecosystem δ15N tend to be found (Amundson et al. 2003).

Swap et al. (2004) concluded that the relationship between plant δ15N and rainfall in southern Africa was much stronger and steeper in C3 than in C4 plants, but provided no explanation for this phenomenon. Working in an arid region of southern Africa, Aranibar et al. (2008) more recently reported that δ15N was higher in C3 than in C4 plants. Unfortunately, from these two studies it is difficult to distinguish the direct effects of the photosynthetic pathway and growth form on δ15N, because C3 plants were almost always non-grasses and C4 plants were almost always grasses. Our finding that the response of δ15N to increasing water availability is more negative in C3 grasses than in C4 grasses provides evidence that the effect previously reported by Swap et al. (2004) and Aranibar et al. (2008) is directly related to photosynthetic pathway, rather than growth form. Discussions of the effects of photosynthetic pathway on plant δ15N are virtually absent from the published literature, although there is abundant evidence of differences in nitrogen metabolism between C3 and C4 plants, with substantially higher nitrogen use efficiency in C4 plants, largely due to a lower demand for Rubisco in leaves (Brown 1978; Schmidt & Edwards 1981; Sage & Pearcy 1987; Makino et al. 2003). Given these differences in nitrogen metabolism, and the large fractionations that can occur within plants (Robinson 2001), it is not unexpected that δ15N may differ between the C3 and C4 pathways. Similarly, differences in nitrogen metabolism may also contribute to the differences we found in δ15N between C4 types, an effect that, to the best of our knowledge, has not been previously reported. Work by Ghannoum et al. (2005) suggests that there are clear differences in nitrogen use efficiency and leaf nitrogen content between the NAD-ME and NADP-ME types, although we found very little difference in δ15N between those types. The most notable difference we found was between the PCK type and the other three types, yet there is little in the published literature to suggest that the PCK type differs substantially from the other types in terms of nitrogen metabolism. As such, we are unable to provide an explanation for this effect at this stage, and it clearly requires further investigation.

While it is well established that soil moisture is an important determinant of soil and plant δ15N (Austin et al. 1998, Handley et al. 1999; Schuur et al. 2001), the effect of temperature is poorly documented. Martinelli et al. (1999) were among the first to note that tropical forests are enriched in 15N relative to temperate forests, and Amundson et al. (2003) analysed a global data set to conclude that soil and plant δ15N were more closely correlated with MAT than mean annual rainfall. In contrast, we found no correlation between MAT and foliar δ15N. The lack of a relationship is probably due to the limited temperature range over which samples were collected in our study. In contrast, the study of Amundson et al. (2003) utilized an extreme temperature range, from tropical forests and deserts to high latitude boreal forest.


Our findings also have implications for the use of δ13C measurements of plant and animal remains in ecological studies. For example, in areas where most grass biomass is C4, δ13C measurements of herbivore remains are commonly used to estimate the contribution of grass to the diet (Vogel 1978; Witt et al. 1998; Cerling et al. 2006; Codron et al. 2007). Such studies require accurate estimates of typical δ13C values for C3 and C4 plants, so-called ‘end-members’. In the majority of cases, C3 and C4 end-members are each expressed as a single mean, without accounting for effects such as water availability (e.g. Witt et al. 1998; Sponheimer et al. 2003), while recently, researchers have begun to account for such effects and vary C3 end-members according to water availability (e.g. Murphy et al. 2007a). However, in only several instances have attempts been made to account for the effect of water availability on C4 end-members (e.g. Cerling & Harris 1999; Cerling et al. 2003). Our findings underscore the importance of accounting for variation in foliar δ13C of both C3 and C4 species in dietary studies, especially when attempting to make precise estimates of parameters such as diet-tissue enrichment factors, where errors of ±1‰ would be unacceptable (e.g. Cerling & Harris 1999; Murphy et al. 2007a).

Models of 13C discrimination by terrestrial plants form an important part of global carbon budgets. In these budgets, changes in the isotopic concentration of atmospheric CO2 are used to derive the location and magnitude of biospheric carbon sinks, particularly the relative contributions of the land and oceans (Battle et al. 2000). However, the approach relies on accurate estimates of 13C discrimination by terrestrial vegetation. While current models assume that discrimination by C3 plants shows high spatial and temporal variability, especially in response to water availability, discrimination by C4 plants is assumed to be either constant, or vary little (Lloyd & Farquhar 1994; Fung et al. 1997; Scholze et al. 2003; Suits et al. 2005). We have shown that at a continental scale, 13C discrimination by C4 plants is clearly influenced by water availability. Indeed, the effect of water availability on δ13C of C4 grasses (1·7‰) is of an order of magnitude similar to the effect of water availability on δ13C of C3 grasses (2·8‰; this study) and C3 plants more generally (2–4·5‰; Table 2). Given that the contribution of C4 grasses to terrestrial gross primary production is high at about 23% globally (Still et al. 2003), failure to account for the relationship between water availability and 13C discrimination by C4 grasses may represent an important, yet easily corrected, source of error in current global carbon budgets.


We thank Wendy Telfer, Lynda Prior and two anonymous referees for providing helpful comments on an early version of the manuscript, and Stephen Clayton, Sue Wood and Hilary Stuart-Williams for assisting with the isotope analysis. Grants from the Australian Research Council (DP0342788 and DP0878177) supported this work.