Carbon Isotope Discrimination of C3 Vegetation in Central Asian Grassland as Related to Long-term and Short-term Precipitation Patterns

The relationship between carbon isotope discrimination (13) of C3 vegetation and long-term (30 years) and short-term (growing period) precipitation was investigated. Different species of Stipa, a dominant grass genus in the (semi-)arid Asian steppes, and other C3 species were collected along aridity gradients in Inner Mongolia in 2005 (11 sites, 71 samples) and in the Republic of Mongolia in 2006 (40 sites, 45 samples). The data set was expanded with published and unpublished data of Stipa and other C3 species (11 studies covering 8 years, including 64 observations of Stipa, and 103 observations of other C3 species) and C3 community bulk-samples (11 samples). Weather data were geostatistically interpolated for all sampling sites and years. 13 of Stipa followed different relationships for the individual years when related to mean annual precipitation due to large anomalies between annual and long-term average precipitation patterns. However, the 13 response to rainfall converged when the (long-term) mean annual precipitation was replaced by year-specific mean daily precipitation during the growing period (P G). Remarkably, the 13-response to (P G) for C3 species as a whole (including herbaceous di-cots, semi-shrubs and grasses) and also the C3 community-level response were virtually identical to that of Stipa. The relation was also valid outside the geographical and climatic range where it was developed, giving proof of its robustness.


Introduction
The Central Asian Grassland is the largest contiguous biome of the world and in- 20 cludes the steppes of the Republic of Mongolia (1.3 million km 2 of a total land area of 1.6 million km 2 , Kerven et al., 1996) and of the Inner Mongolia autonomous region (0.8 million km 2 of a total land area of 1.2 million km 2 , Xiao et al., 1995a and citations therein) of the People's Republic of China. This region has a continental climate with most of the rain falling in the summer. By far, the largest part of this grassland is of the genus Stipa are a frequent and often dominant component of most types of grassland in the region.
Precipitation is highly variable in time and space in the Mongolian grassland. Mean annual precipitation (MAP) varies from several hundred millimeters per year (mm yr −1 ; mainly in the northern and eastern part of the region) to less than one hundred millime- 20 ters per year in the Gobi desert (Fig. 1). But, at a given site, hydrological conditions of one year may differ drastically from another year. For example, MAP at Erenhot is 158 mm yr −1 , but exceeds 277 mm yr −1 or falls bellow 44 mm yr −1 in every second year (NOAA NCDC Climate Data Online, 2007). These factors may exert strong influences on the carbon isotope composition of grassland via (i) effects on the relative abundance 25 of C3 and C4 species (which differ by about 10 to 18‰ in carbon isotope discrimination; Farquhar et al., 1989;O'Leary, 1981), and (ii) variation of 13 ∆ in C3 plants which results from variation of stomatal conductance or photosynthesis (Farquhar et al., 1989 The carbon isotope signal produced by grassland vegetation is imprinted in ecosystem carbon pools (such as soil carbon) and exchange fluxes (such as biosphere atmosphere CO 2 exchange), and is therefore potentially highly useful for biogeochemical studies and ecosystem reconstruction. However, in C3:C4 mixed grassland in arid environments, the separate effects of C3:C4 abundance and variation of 13 ∆ in C3 plants 5 on community/ecosystem 13 ∆ must be known, to allow interpretation of ecosystem or community carbon isotope signals in terms of C3:C4 variation. The present work concentrates on one aspect of this problem, namely the effect of aridity on 13 ∆ in C3 plants, and then discusses its implications for C3:C4 mixing models. The effect of aridity on 13 ∆ in C3 plants has generally been studied in aridity-transect 10 studies where the 13 ∆ of species was related to mean annual precipitation at sampling sites. The responses observed in theses studies were highly variable (e.g., Stewart et al., 1995;Schulze et al., 1991Schulze et al., , 1996Wang et al., 2003;Liu et al., 2005;Zheng and Shangguan, 2007). Here, we test the hypothesis, that the variability of the 13 ∆response of C3 plants to mean annual precipitation in (semi-)arid grasslands is related 15 to interannual variation of weather conditions. To test this prediction, we analysed the relationship between 13 ∆ of C3 species and short-term (growing period) and long-term ( Interactive Discussion typical steppe and desert steppe, with MAP ranging from around 280 mm yr −1 near Ulaanbaatar to 125 mm yr −1 and less in the Gobi area near Sainshand and Erenhot and rising again to 280 mm yr −1 near Xilinhot (Fig. 1).
The mean temperature ranges from about −23 • C in winter in Ulaanbaatar to about 23 • C in summer in Zamyn-Uud. The local precipitation is highly variable in space and 5 time (Gong et al., 2004). Even climate stations within about 10 km show remarkably different characteristics. Although the distance between Zamyn-Uud (Republic of Mongolia) and Erenhot (Inner Mongolia) is only 10 km, the mean monthly precipitation in August is 0.52 mm day −1 for Zamyn-Uud and 1.29 mm day −1 for Erenhot (mean for last normal period   (Table 1) were included in the validation data base with the primary aim of increasing the temporal (and secondarily the geographic) representation. Published data covered a wide range of site conditions, with altitudes above 2000 m above sea level (especially Ivanov et al., 2007). As altitude affects the δ 13 C of C3 plants 10 (Körner et al., 1988), all data were normalized to an altitude of 1000 m above sea level, which is close to the average altitude of our sampling locations (mean: 1160 m a.s.l.; SD: 180 m). Altitude correction was performed by applying a correction of 1.15‰ per 1000 m following Maennel et al. (2007) with a mean absolute correction to 13 ∆ of 0.26‰. For publications containing no altitude data, this information was obtained 15 with Google Earth® using the coordinates of sample sites.

Isotope analysis
The vegetation samples were further dried in the laboratory for one hour in a forced air oven at 95 • C and thereafter for 48 h at 60 • C. Dried samples were ground with a ball mill. The samples were then combusted in an elemental analyser (NA 1110;Carlo 20 Erba, Milan) interfaced (ConFlo III; Finnigan MAT, Bremen) to an isotope ratio mass spectrometer (Delta Plus; Finnigan MAT). Carbon isotope data are presented as δ 13 C relative to the international VPDB standard: δ 13 C=(R sample /R standard )−1; where R sample and R standard are the ratios of 13 C/ 12 C in the sample and standard. Interactive Discussion of calibration ±0.06‰ SD). Solid internal laboratory standards (SILS), with similar C/N ratio as samples (wheat flour, C/N: 21.6), were calibrated against these references. One SILS was measured after every tenth sample. The precision for sample repeats was better than 0.2‰ for δ 13 C. C3 community-mean 13 ∆ was calculated as the biomass-or ground cover-weighted average of all C3 species in the community. The ordinary (un-weighted) mean 13 ∆ was calculated if information on ground cover or biomass fraction was missing. In this way C3 community-mean 13 ∆ was obtained from 35 data sets by Gong et al. (2007)

Carbon isotope composition of atmospheric CO 2
The calculation of 13 ∆ with Eq. (1) considered the fact that δ 13 C a has been decreasing continuously in the recent past, so that plants sampled in different years grew in the presence of CO 2 with (slightly) different δ 13 C. So, δ 13 C a was estimated for the year 15 when the sampled plant grew. A third order polynomial was developed from measured δ 13 C a to predict the annual average δ 13 C a for every year starting in 1959, similar to the relation reported in Geist et al. (2005):  Gat et al. (2001), Allison et al. (2003) and NOAA NCDC Climate Data Online (2007) for the stations Mauna Loa, Siple, Antarctica, Ulan Uul, Shetland Islands, Hegyhatsal (Hungary) and Ochsenkopf (Germany). The standard error of this regression was 0.09‰. The predicted δ 13 C a changed from −8.12‰ for 1996 to −8.48‰ for 2007. From this annual average the 5 mean δ 13 C a during the growing period was estimated by taking into account the relative seasonal variability as measured at Ulan Uul (Tans et al., 2005), which is a long-term measuring station located near the center of the research area (44 . On average δ 13 C a was less negative by 0.25‰ between April and August as compared to the annual average. Thus a δ 13 C a of −8.17‰ was assumed for the growing period 10 in 2005 and a δ 13 C a of −8.19‰ for the growing period in 2006. Other years were calculated accordingly.

Statistical methods
Linear and linearized regressions were used to evaluate the datasets. The coefficient of determination was tested with a two-sided test for significance of the regression. In 15 addition the 95%-confidence interval for the samples and the 95%-confidence interval for the regression were calculated to allow for comparison between data sets. A pairwise comparison of means was used to test whether the species differed regarding MAP and ∆. These statistical procedures followed standard protocols (Sachs and Hedderich, 2006). 20 Geostatistical analyses (for theory see Webster and Oliver, 2004;Nielsen and Wendroth, 2003) were conducted with package geoR (Ribeiro and Diggle, 2001) of the software GNU R 2.6 (R Development Core Team, 2007). The semivariance of a parameter under consideration (e.g., precipitation) is the half mean quadratic difference of the parameter values of points which are separated by a certain distance (called lag). For  , 2007, UTM zone=50). Semivariances were then grouped by lag classes and semivariances and lags within a group were averaged yielding the empirical semivariogram (x axis: lag, y axis: semivariance). A theoretical semivariogram was fitted to minimize weighted least squares, with weights calculated from the ratio of pairs within a class to mean lag. This gives more weight to those classes, which are based on 5 many data pairs and which are more important for interpolation. The quality of the fit was controlled by calculating the Nash-Sutcliff-Index. Spatial interpolation to construct maps was then carried out for a rectangular grid by ordinary point kriging, based on the theoretical semivariogram. The quality of the predictions from the resulting maps is given as the krige standard deviation averaged for the sampling locations.

Meteorological data
The growing period in the sampling area starts in April with only one growth cycle (no regrowth after cutting or heavy grazing). Sampling took place in early July (in 2005) or at the end of July/beginning of August (in 2006). Sampled plant material thus included biomass grown in the period of April to mid of July (2005) or April to end of 15 July/beginning of August. We will call this "growing period" in the following and index it with "G". Most of the rainfall and hence most of the plant growth usually results in this period. For 40 meteorological stations in and around the sampling area, longterm mean precipitation from April to September correlated closely with MAP (r 2 =0.98) and contributed 74% to MAP. Normally only little growth occurs in September, which re-20 ceives only 10% of MAP in an average year. Hence, peak above-ground living biomass in ungrazed areas usually occurs between late July and late August (Xiao et al., 1995b). The effective water availability for plants in different years and studies was estimated as the mean daily precipitation during the respective growing period P G mm day −1 ). This accounted for the fact that sampling did not occur on exactly the same dates in the 25 different studies, but provided a common denominator for comparison of data gathered in different years.
Sampling sites, either own or from literature, were usually not located near meteoro-911 Interactive Discussion logical stations. Two data sets were used to estimate meteorological data for the sampling sites. (i) The long-term averages of the last normal period ) of precipitation data (monthly and annual means) were taken from high resolution maps obtained from The Climate Source Inc., Corvallis, Oregon. These maps have a pixel resolution of 0.02 • ×0.02 • (approximately 1.5×1.5 km 2 , Fig. 1), judged sufficient to locate the sam-5 pling sites. These data were created using the PRISM method (parameter-elevation regressions on independent slopes model; Daly et al., 2002), which accounts for topography, rain shadows, lake effects, temperature inversions, and more. (ii) For yearspecific data we assumed that the principal causes underlying the PRISM maps were also valid at the small scale but were modified by large-scale trends. Daily precipi-10 tation of 40 climate stations, provided by the NOAA Satellite and Information Service (NOAA NCDC Climate Data Online, 2007) were compiled. These stations were situated inside or just outside the sampling area. The latter were included because they improved the geostatistical interpolation at the periphery of the sampling area. To calculate P G , the difference between the normal-period mean daily precipitation during 15 the growing period (MPG) and P G was calculated for each meteorological station to quantify the anomaly, dP G . These anomalies were geostatistically interpolated for the whole research area and each year to represent the large-scale spatial pattern of the anomalies. This trend was then superimposed on high-resolution normal-period maps to obtain high-resolution maps for individual years; e.g., for the year 2005 P G05 was 20 given by: While annual precipitation was similar in 2005 and 2006, P G05/06 differed considerably (Fig. 2). In some locations anomalies in P G between the two years and between MGP and P G were as large as 0.6 mm day −1 , which was more than one third of the 25 total precipitation. The best geostatistical interpolation was obtained by ordinary kriging interpolation with Gaussian models (Fig. 2) Interactive Discussion the anomaly, which decreased from northwest to southeast. In 2005 the trend differed somewhat in direction, strength, curvature and the position of the zero isohyet compared to 2006. In 2006 the zero isohyet was near the center of the sampling area, the south-eastern part of the sampling area was drier than average and the north-western part was wetter. In 2005 the zero isohyet was near the northern border of the research 5 area, and most of the sampling area received less precipitation during the vegetation period than in a normal year (Fig. 2).

Carbon isotope discrimination of Stipa as related to mean annual precipitation
The gradient of MAP covered by the sampling transects was similar in the two years, 10 and ranged between approx. 130 and 290 mm yr −1 (Fig. 3). 13 ∆ among Stipa increased with MAP in both years, although the increase was only significant in 2006 (P <0.01). The slope of this relationship ( 13 ∆ versus MAP) was very similar in both years (P >0.05), but there was a significant offset between the years, so that 13 ∆ at a particular MAP was 1.1‰ higher in 2006 than in 2005. In both years, there was no 15 relationship between 13 ∆ and the C/N ratio of samples.
In both years, the effect of MAP on 13 ∆ coincided at least partially with species replacement along the aridity gradient: S. gobica and S. glareosa were only present in the most arid part of the transects (MAP<200 mm yr −1 ) and had lower than average 13 ∆ ( followed the same relationship with P G , with 13 ∆ (‰)=15.0+2.53 sqrt P G (Fig. 4). Thus, the difference between years (apparent when 13 ∆ was regressed against MAP) disap-5 peared when 13 ∆ was regressed against growing period precipitation. This suggested that interannual variations in 13 ∆ were related to interannual variations in precipitation.
The square root expression indicates that the effect was strong when P G was very low and decreased with increasing precipitation. Species did not differ in the relationship between 13 ∆ and P G (highlighted for S. gran-10 dis in Fig. 4). Thus, it seemed that species-replacement was controlled by long-term precipitation patterns, whereas the within-species response to short-term precipitation patterns controlled 13 ∆.

Comparison with published Stipa data
There were eight data sets (six published and two unpublished; see Table 1) for 15 which the relationship of 13 ∆ with actual precipitation could be calculated and compared with the results from our transects (see Table A2 in supplementary material, http://www.biogeosciences-discuss.net/5/903/2008/bgd-5-903-2008-supplement.pdf, for precipitation data). These studies reported a total of 62 data for different species of Stipa. The data corresponded well with the relation derived from the 2005 and 2006 20 samples (Fig. 5 top). Notably, this was also true for studies, in which Stipa experienced higher precipitation than the maximum encountered in our transect studies (Fig. 5

Printer-friendly Version
Interactive Discussion the notion that the relationship presented in Fig. 4 provides a general and unbiased prediction of the 13 ∆ of Stipa in the grasslands of Mongolia.
The relation with P G even explained data obtained in 2001 (Table 1), which was an extremely dry year in Inner Mongolia. In that year more than half of the numerous lakes (more than 4000) of the province of Qinghai (in the west of Inner Mongolia) disappeared 5 (Reuters, 2001), and in the Republic of Mongolia the herders lost one quarter of their large-animal livestock such as horses and cattle (Retzer, 2007).

Comparison with other C3 species
The dependence of 13 ∆ on growing period precipitation in Stipa was also compared with that of 'non-Stipa' C3 species. The latter included 170 data from a total of 55 species, collected in eight different years (1996, 1999, 2001, 2002, 2003, 2004, 2005 and 2006). Again, the comparison indicated a fair agreement between the precipitation response of 13 ∆ in Stipa and that of the other species, although the scatter was somewhat larger in the latter. Still, 80% (135 out of 170) of all non-Stipa data fell inside the 95% confidence interval of Stipa (Fig. 5 bottom). Subdivison of the whole data set in 15 functional groups (monocots n=163 and dicots n=118; grasses n=143, forbs n=50 and shrubs and semi-shrubs n=68) also provided no evidence for a different precipitation response at the level of functional groups. However, dicots deviated from monocots insofar as their 13 ∆ was about 0.5‰ higher on average at a given precipitation level. In particular, leaves of Reaumuria soongorica, a short-statured woody shrub, had 1.8‰ 20 higher 13 ∆ than Stipa and other species at the same precipitation level. Yet, even in R. soongorica, the slope of the 13 ∆ versus P G relationship did not deviate from that of Stipa or that of the whole of other species.
3.5 The 13 ∆-response of C3 communities to growing period precipitation The C3 community-mean 13 ∆ responded to mean daily growing period precipitation 25 in the same way as did Stipa (Fig. 6)  which compressed the 95% confidence interval for the data points. The regression for the community data, calculated to obtain a robust relation over the entire range of precipitation, was virtually the same relation as that for Stipa, but the confidence interval of the regression remained narrow over the entire range due to the continuous support by data. 5 3.6 The relationship between precipitation anomalies and 13 ∆ anomalies The deviation of growing period precipitation from the long-term mean (precipitation anomaly) was directly related to the deviation between actual 13 ∆ discrimination and 13 ∆ predicted for mean conditions at the respective sites ( 13 ∆ anomaly). This effect was manifest at the level of C3 communities (Fig. 7), and species (data not shown). The 10 precipitation anomalies modified the community-mean 13 ∆ by ±1‰. The response at the community level and the large spatial extension of precipitation anomalies imply that this response was also effective at larger scales.

Discussion
4.1 Interannual variation of 13 ∆ results from growing period precipitation anomalies 15 This work reveals strong effects of interannual variation in precipitation on 13 ∆ of C3 plants in the grasslands of Mongolia. This effect was particularly evident in the genus Stipa for which a large data set was available from collections along precipitation gradients in two different years, and published data from several years. Moreover, the effect was also apparent at the level of communities and functional groups (grasses other 20 than Stipa, perennial forbs, and shrubs and semi-shrubs). An important consequence of the interannual variability of precipitation was that it caused interannual fluctuations in the relationship between (annually determined) 13 ∆ and MAP. This means, that a stable and unbiased relationship between 13 ∆ and MAP can only be obtained by relating long-term averaged 13 ∆ to MAP. Interannual variation of 13 ∆ (or δ 13 C) has been observed before, particularly in trees (Sparks and Ehleringer, 1997;Warren et al., 2001;Li et al., 2007), but also in grassland (Tsialtas et al., 2001;Mole et al., 1994), and was also explained by rainfall anomalies (Li et al., 2007;Mole et al., 1994) or other weather related factors (e.g., seasonal variation in water vapour deficit as reported by Sparks and Ehleringer (1997), variation in conditions from long-term mean conditions. This may also include interannual variability of soil moisture and atmospheric vapour pressure, which exert direct effects on 13 ∆ (Sparks and Ehleringer, 1997; Tsialtas et al., 2001). However, since these factors are usually (but not always, see Schulze et al., 1996) correlated with precipitation, their effect is (at least partially) included in the relationship of 13 ∆ with year-specific growing 15 period precipitation.
4.2 Species, functional groups and communities share the same precipitation response of 13 ∆ The 13 ∆-response (that is the slope of the relationship between 13 ∆ and P G ) of S. grandis was the same as that of the other members of Stipa in the data base. It was also 20 very similar to the "mean" species response, and the response of C3 communities. Even in R. soongorica, which had a comparatively high 13 ∆ in all rainfall conditions, the slope of the relationship between 13 ∆ and year-specific growing period precipitation was the same as that of the "mean" species. Although the relationship between 13 ∆ and growing period precipitation has not been studied in any detail, the uniformity in 25 the 13 ∆ responses of species' or functional groups to rainfall, as seen here, was not expected: inter-specific differences in the 13 ∆-response to environmental parameters have been observed before (e.g., Handley et al., 1994), and the adaptive significance of different strategies of water use have been discussed and emphasized (e.g., Golluscio and Oesterheld, 2007). Yet, we acknowledge that, although the number of species in the data base was relatively large (55 C3 species, including 7 Stipa species), it was nevertheless a small fraction of the total flora of the grasslands of Mongolia. Moreover, 5 the data base was dominated by perennial grasses (over 50% of all data). But in this respect the data base reflects the species composition of most grassland communities of Mongolia. For instance, Stipa accounted for more than 40% and perennial grasses (including Stipa) for more than 70% of total aboveground biomass in the communities sampled in 2005. Perennial grasses from arid and semi-arid temperate grasslands 10 share great similarities in phenology, leaf structure, and root architecture and placement, which may explain the similarity in their water use (Golluscio and Oesterheld, 2007). Thus, the similarity in the precipitation response of 13 ∆ by C3 communities was related to the predominance of perennial grasses (particularly Stipa) in these communities and the similarity in water use strategies among grasses. These similarities 15 might also explain, why species-replacement along the aridity gradient did not affect the precipitation response of 13 ∆ (Fig. 4).
Nevertheless, there was significant scatter in the overall relationship between 13 ∆ and growing period precipitation. Although there was no difference between functional groups, the scatter may partly be due to differences between species within groups. 20 Differences between species have been interpreted in terms of differences in intrinsic water use efficiency (e.g., Condon et al., 1990;Meinzer et al., 1992;Ehleringer et al., 1992) and may be related to differences in phenology (e.g., Smedley et al., 1991), rooting pattern/depth (Golluscio and Oesterheld, 2007) and leaf anatomy (including leaf thickness and nitrogen content) (Farquhar et al., 1989;Schulze et al., 2006). Mech- 25 anisms also include special adaptations to arid conditions such as leaf shedding in response to drought as expressed in R. soongorica (Ma et al., 2005) and phreatophyte lifestyle. This also explains, why the precipitation response at the community level was associated with much less scatter than that at the species level (cf. Figs. 4 and 6 Interactive Discussion communities include species with a range of functional attributes that imply differences in 13 ∆. Finally, there exists large short-term, small-scale variability in precipitation in the region, which is not represented entirely when interpolating precipitation at collection sites from weather station data. For instance, the Erenhot and Zamyn-Uud weather 5 stations are separated by approximately 10 km but differ by an average of 15 mm in the monthly precipitation during the growing period. Such variations in rainfall are frequent even within one kilometre and increase in strength with continentality (Fiener and Auerswald, 2008 2 ). Therefore, errors in precipitation estimates are probably substantial and explain some of the scatter in the 13 ∆ versus precipitation relationship. The findings of this work are useful for estimation of the relative abundance of C3 and C4 vegetation in community biomass from community 13 ∆ ( 13 ∆ community ), which requires knowledge of the 13 ∆ of the C3 ( 13 ∆ 3 ) and C4 ( 13 ∆ 4 ) members in a two-15 component mixing model (e.g., Still et al., 2003). Evidently, errors in the assessment of 13 ∆ 3 and 13 ∆ 4 cause errors in the estimation of C3:C4 abundance. In general, neglect of variation of 13 ∆ 3 or 13 ∆ 4 overestimates the true variation of the fraction of C3 (and C4) vegetation in communities, because all variation of 13 ∆ community is attributed to variation in the relative abundances of C3 and C4 plants. If variation of 13 ∆ 3 (or 13 ∆ 4 ) 20 is systematic, then neglect of this variation provokes a biased variation of the relative abundances of C3 and C4 plants. For instance, neglect of the effect of aridity on 13 ∆ of C3 plants would lead to an underestimation of the relative abundance of C3 plants in the dry section of the transect. As shown above the potential variation of 13 ∆ 3 is large in semi-arid and arid grassland, creating opportunities for large errors in estimation of the Interactive Discussion relative abundance of C3 and C4 plants from 13 ∆ community . In the present study aridity caused a variation of 13 ∆ 3 of up to ∼5‰ (Fig. 6). This is about one third to one half of the difference in 13 ∆ between C3 and C4 plants in arid and semiarid grassland (Schulze et al., 1996;Wang et al., 2005). Accordingly, the shift from the wet to the dry end of the aridity gradient has the same effect on 13 ∆ 3 , as a 33% to 50% replacement by C4 5 plants. A possible systematic variation of 13 ∆ community must also be considered when interpreting long-term records such as soil organic matter or sediments in terms of the C3:C4 abundance. Variation of 13 ∆ 4 would have a similar effect, although, variation in 13 ∆ 4 is generally smaller (e.g., Liu et al., 2005;Wang et al., 2005) et al., 1996;Ghannoum et al., 2002). Yet, there exist systematic differences in 13 ∆ between C4 metabolic types (e.g., Schulze et al., 1996;Ghannoum et al., 2002) and the relative abundance of metabolic types may change with aridity (Schulze et al., 1996).

15
This work demonstrates that rainfall anomalies cause large variations of the 13 ∆ versus MAP relationship, which are greatly reduced when 13 ∆ is related to growing period rainfall. Significantly, the 13 ∆-response to growing period rainfall was stable across years, and it was very similar for dominant species (Stipa members), functional groups (including herbaceous dicots, semi-shrubs and grasses), and C3 communities. More-20 over, the relation was also valid outside the geographical and climatic range where it was developed, giving proof of its robustness. Because of its generality and stability, the relationship between 13 ∆ and growing period rainfall allows an unbiased estimation of 13 ∆ of the C3 member for use in C3:C4 mixing models based on community C isotope composition. Schnyder, H., Schwertl, M., Auerswald, K., and Schäufele, R.: Hair of grazing cattle provides an integrated measure of the effects of the site conditions and interannual weather variability on δ 13 C of temperate humid grassland, Glob. Change Biol., 12, 1-15, 2006. Schulze, E. D., Ellis, R., Schulze, W., and Trimborn, P.: Diversity, metabolic types and delta C-13 carbon isotope ratios in the grass flora of Namibia in relation to growth form, precipitation 25 and habitat conditions, Oecologia, 106(3), 352-369, 1996. Schulze, E. D., Turner, N. C., Nicolle, D., and Schumacher, J.: Species differences in carbon isotope ratios, specific leaf area and nitrogen concentrations in leaves of Eucalyptus growing in a common garden compared with along an aridity gradient, Physiol. Plantarum, 127(3), [434][435][436][437][438][439][440][441][442][443][444]2006.  Sparks, J. P. and Ehleringer, J. R.: Leaf carbon isotope discrimination and nitrogen content for riparian trees along elevational transects, Oecologia, 109(3), 362-367, 1997. Still, C. J., Berry, J. A., Ribas-Carbo, M., and Helliker, B. R.: The contribution of C 3 and C 4 plants to the carbon cycle of a tallgrass prairie: an isotopic approach, Oecologia, 136 (3)