Large regional-scale variation in C 3 / C 4 distribution pattern of Inner Mongolia steppe is revealed by grazer wool carbon isotope composition

Introduction Conclusions References Tables Figures


grassland.
There is abundant evidence that the current distribution of C4 plants is primarily controlled by growing season temperature (Ehleringer et al., 1997;Collatz et al., 1998) and that this is related to the higher effective quantum yield of CO 2 fixation (Ehleringer and Bjorkman, 1977) and higher maximum photosynthetic rate of C4 plants at high temper- 15 ature (Sage and Kubien, 2003). Central East Asia has experienced one of the largest temperature increases on earth during the last decades (Chase et al., 2000;Yu et al., 2003), which should promote C4 plants. On the other hand, rising atmospheric CO 2 stimulates C3 photosynthesis more than C4. Nutrient limitation due to increased photosynthesis may again favour C4 species (Sage and Kubien, 2003) leading to a complex 20 web of interactions, which makes it difficult to predict the overall effect. Furthermore, the different influences act on different spatio-temporal scales. While CO 2 concentration changes globally, temperature varies regionally and nutrient limitation may differ down to the scale of urine patches. The effects may thus vary regionally or even locally and can only be assessed by taking scales into account. Furthermore, the seasonal tats (e.g., Ehleringer et al., 1997). Presumably such secondary controls have their strongest effect on C3/C4 abundance in those regions which have a growing season mean temperature that is near the C3/C4 crossover temperature (i.e. the temperature above which the quantum yield efficiency of C4 plants is higher than that of C3 plants, see Ehleringer et al., 1997). In this respect, the grassland of Inner Mongolia is of 5 particular interest, as the average temperature during the summer months (especially July) when most of the annual precipitation falls is close to the C3/C4 crossover temperature. There are several reports indicating increases in the number and abundance of C4 species with aridity in (Inner) Mongolia grassland (Pyankov et al., 2000;Wang, 2004) and desertification (Wang, 2002), although conflicting evidence has also been 10 presented (Wang, 2003;Ni, 2003).
Although the C3/C4 ratio is the main factor affecting δ 13 C of grassland, variation in 13 ∆ of C3 plants may also contribute, as the δ 13 C of C3 plants decreases with precipitation, moisture availability (e.g., Schulze et al., 1998;Liu et al., 2005;Wittmer et al., 2008) and with altitude (Körner et al., 1991;Männel et al., 2007). Variation in the 15 δ 13 C of C3 plants has to be taken into account when estimating the C3/C4 ratio from carbon isotope composition.
Another difficulty in regional-scale studies of the 13 C signal of grassland is a potentially high spatio-temporal variation: the life-span of the aboveground parts of grassland plants is short, reflecting growing conditions prevailing in a period of weeks to 20 a few months before sampling. This is aggravated in C3/C4 grassland due to the asynchronous development of C3 and C4 components resulting from their different thermal preferences. One possibility to overcome (integrate) these spatio-temporal variations is to collect representative bulk samples over a large area and to sample several times during the vegetation period. An alternative to this is to take advantage of the "sam- 25 pling" activity of livestock which are grazing the grassland. Provided that they do not select between photosynthetic types, and eventual post-ingestion fractionation of carbon isotopes is known, the δ 13 C of vegetation can be inferred from wool or other animal tissue (e.g., Jones et al., 1981;Schnyder et al., 2006;Norman et al., 2008). Geostatis-548 tical analysis can then be used to separate between the regional pattern and the noise of the individual samples, e.g. caused by small-scale peculiarities of individual animals, farms or years. The aim of this work was to answer the following questions by geostatistical analysis of δ 13 C derived from wool samples by taking into account the diet-wool shift and 5 environmental influences on the C3 and C4 end-members: (i) How is the C3/C4 pattern of the grassland of Inner Mongolia? (ii) How does this pattern relate to driver patterns like (growing period) temperature and/or precipitation? creases from 100 mm yr −1 in the western part to 400 mm yr −1 in the eastern part of the study area. Most of the precipitation (approximately 75%) falls during the growing period (April-September). The spatial variation of temperature follows altitudinal and latitudinal trends. Mean annual temperature and mean temperature of the growing period vary between 0-6 • C and 14-19 • C, respectively. Almost the whole area is 20 used for small ruminant livestock production (mainly sheep and cashmere goats) with little agriculture, mainly cultivation of maize. In general, animals graze from spring till winter depending on snow cover and availability of herbage. In overgrazed areas grazing is suspended during winter and/or early spring to allow for recovery of vegetation. Animals do not receive supplements on pasture. Pen fodder in the winter 25 and/or spring consists of hay from meadows, supplemented in some cases with small 549 amounts (<10% of the ration) of maize and/or industrial pasture from the major towns e.g., Hohhot or Baotou (information obtained from interviewing the local herders).

40
• 36 to 47 • 42 N) exceeded the study (target) area to evade boundary effects at the periphery of the study area during spatial analysis and interpolation. The current sampling position and altitude was measured with a mobile outdoor global positioning system. Bulked leaves were collected within approximately 1000 m 2 on a reduced scale of apparent species-level contribution to total standing biomass ("community-scale" sam-10 ples) on 112 sites within the study area and on 6 sites outside the periphery. At some sites (n=52), encompassing the entire aridity gradient in the sampling area, the leaves of the dominant and in parts the co-dominant species (contributing individually >5% to total aboveground biomass) were sampled separately (>3 different plants per sample), to obtain separate estimates of the δ 13 C of the C3 and C4 components of plant 15 communities. The wool samples resulted from the last shearing and therefore reflected the period in between two consecutive shearing (mostly one year), which generally take place in June, and the grazing grounds of one herder ("farm-scale" samples). Wool from 146 sheep dating from 1998-2007 was sampled on 82 sites (2003, 2004 and 2007) 20 within the study area and on ten sites beyond the periphery. Most wool samples were provided by local herders and originated from the recent shearing.
Additional samples were taken between late June and mid of September in 2005 and 2006 from a long term grazing experiment at 116 • 40 E, 43 • 34 N. The grazing experiment allowed comparing vegetation and wool directly and to assess selective 25 grazing because it covered a large gradient in stocking densities between 1.5 sheep per hectare, which allowed for selection, and 9.0 sheep per hectare, where the total aboveground biomass was grazed leaving bare soil. For further information on the 550 grazing experiment see Zhao et al. (2007). On 14 plots vegetation samples were taken and the wool from 90 sheep. For the latter, the sheep were shorn before they were moved to the pasture (mid of July 2005 and 2006, respectively) and at the end of pasturing (mid of September 2005 and 2006, respectively).

Sample preparation 5
The plant samples for each site were transferred to separate paper bags and dried behind the front window of the car. In addition plant samples were dried for 48 h at 60 • C and then were ground to homogeneity with a ball mill. 0.7-0.8 mg plant material in tin cups was analyzed for their isotopic composition. The wool samples were cleaned by the procedure of Schwertl et al. (2003) and 0.2-0.4 mg wool were then packed into tin cups for isotope analysis. For the wool originating from the years 2003, 2004 and 2005 an additional segmentation was carried out. The wool was cut into cm pieces, which were analyzed separately to quantify an isotopic shift between the summer wool segments and summer vegetation (n=14 for the grazing experiment; n=62 for repeated regional samplings).

Isotope and elemental analysis
The carbon isotope composition and carbon and nitrogen content of each sample was determined with an elemental analyzer (NA 1110; Carlo Erba, Milan) interfaced (ConFlo III; Finnigan MAT, Bremen) to an isotope ratio mass spectrometer (Delta Plus; Finnigan MAT). Carbon isotopic data are presented as δ 13 C, with 20 δ 13 C=(R sample /R standard )−1, where R is the 13 C/ 12 C ratio and standard is the Vienna Pee Dee Belemnite standard. Each sample was measured against a laboratory working standard CO 2 gas, which was previously calibrated against an IAEA secondary standard (IAEA-CH6, accuracy of calibration 0.06‰ SD). After every tenth sample a solid internal lab standard (SILS) with similar C/N ratio as the sample material (fine 25 ground wheat flour for plant samples; protein powder for wool) was run as a blind 551 control. The SILS were previously calibrated against an international standard (IAEA-CH6). The precision for sample repeats (SD) was 0.06‰ for plant samples and 0.11‰ for wool samples.

Carbon isotope discrimination of plants
The δ 13 C of plants (δ 13 C P ) differs from the δ 13 C of air (δ 13 C a ) depending on the 5 discrimination 13 ∆ (Farquhar et al., 1989): with δ 13 C a continuously decreasing over time. Mean annual δ 13 C a was predicted applying a third order polynomial and corrected for seasonal variation in δ 13 C a following Wittmer et al. (2008). 10

Meteorological data
For year-specific data on precipitation and temperature we followed the procedure by Wittmer et al. (2008) and corrected long-term average high resolution (2 km×2 km) maps (The Climate Source LLC, 2002) obtained by the PRISM method (parameterelevation regressions on independent slopes model; Daly et al., 2002)

Estimation of C4 fraction
The relative contribution of C4 plants (P C4 ) to aboveground biomass at each site was estimated from 13 ∆ of the community using a two-member mixing model: where 13 ∆ c denotes community, and 13 ∆ 3 and 13 ∆ 4 are the end-members of the mix-5 ing model (local 13 ∆ of pure C3 and C4 vegetation communities, respectively). Regionalized estimates of 13 ∆ 4 (see Results) were obtained from δ 13 C of C4 vegetation samples, whereas 13 ∆ 3 was derived from study-area specific data accounting for precipitation during the growing period as reported by Wittmer et al. (2008 and citations therein) and for altitude following Männel et al. (2007). Combining the equations of 10 both studies yielded an estimate for 13 ∆ 3 : with A denoting altitude (m a.s.l.) and P denoting the mean precipitation (mm day −1 ) during the growing period. The constant 14.4‰ is the minimal reference 13 ∆ 3 as reported by Wittmer et al. (2008) for 1000 m a.s.l. 15 In principle, recent P C4 can be estimated from wool considering the isotopic offset with respect to vegetation: in studies with pure or mixed C3 or C4 diets the δ 13 C of hair was 2 to 3‰ enriched with 13 C relative to diet (Jones et al., 1981;Sponheimer et al., 2003a;Männel et al., 2007). This offset is termed "vegetation-wool shift", S V W (S V W =δ 13 C wool −δ 13 C vegetation ), and results from 13 C fractionation during digestion or 20 metabolism of animals (De Niro and Epstein, 1978). Hence, the 13 ∆ of the community can be estimated from Eq. (1) Männel et al. (2007) was established for pure C3 grassland, with considerably higher digestibility than that found in Inner Mongolia (Wang et al., 2008). In consequence, we assessed S V W by paired sampling across the study area and in the grazing experiment. We compared the δ 13 C of vegetation with that of the most-recently grown 1-cm long wool segment from the sheep actually graz-5 ing this vegetation. From the grazing experiment, the whole wool was used, which was grown while a sheep was grazing on an individual plot but the first cm was discarded to account for metabolic turnover of body carbon originating from the prior winter period.

Statistical methods and geostatistical analysis
Linear regressions were used to evaluate the datasets. The coefficient of determina-10 tion was tested with a two-sided test for significance of the regression. Hypothesis testing on equal means of groups were carried out using Student's t-test, customarily performed against a 95% confidence interval if not stated otherwise. Kernel densities (Silverman, 1986) were calculated using Gaussian kernels to yield estimates of the density distribution. Bandwidth of density estimation of different sized data sets was 15 defined via Silverman's "rule of thumb". To allow for comparison of differently sized data sets integral density was adjusted to unity. Statistical spread is denoted as standard deviation (SD) or 95% confidence interval (CI 95% ). All these procedures followed standard protocols (Sachs and Hedderich, 2006) and were carried out using GNU R 2.7.2 (R Development Core Team, 2008). 20 Geostatistical analyses (see Rossi et al., 1992 and citations therein) were conducted with the auxiliary packages geoR (Ribeiro and Diggle, 2001) and gstat (Pebesma, 2004). The semivariance (γ) of a parameter under consideration is the half mean quadratic difference of the parameter values of points which are separated by a certain distance (called lag) yielding the empirical semivariogram (x-axis: lag, y-axis: semi-25 variance). For calculation of the lag, the coordinates of sampling points had to be transformed from geographic coordinates to UTM coordinates with the package PBSmapping (Schnute et al., 2008;UTM zone=50). A theoretical semivariogram was fitted 554 to minimize weighted least squares, with weights calculated from the ratio of pairs (at least 30) within a class to mean lag. This gives more weight to those classes, which are based on many data pairs and which are more important for interpolation. The theoretical semivariogram delivers three parameters, the nugget effect, the sill and the range. The nugget effect quantifies the small-scale variation including data uncertainty. The 5 sill quantifies the total variation caused by the nugget effect and the variation due to the spatial pattern. The nugget to sill ratio hence reflects the ratio of random (unexplained) to total variation. The range quantifies the distance of autocorrelation caused by the extension of the pattern elements. The quality of the fit between the experimental and the theoretical semivariogram is expressed as root mean squared error (RMSE).
Spatial interpolation to construct maps was then carried out for a uniform rectangular grid (200×200 nodes) by ordinary block kriging for 5×5 km 2 blocks, using the theoretical semivariogram. Samples beyond the periphery of the study area were included because this greatly reduces the error at the margins of the study area. The quality of the predictions from the resulting maps is given as the (block) krige standard deviation 15 averaged for the study area.

The δ 13 C of vegetation and hair
The isotopic composition of carbon pools varied greatly in the sampling area. Furthermore, the range differed strongly between types of pools: the δ 13 C of whole vegeta-20 tion varied between −12.5‰ and −28.4‰ (Fig. 2a), whereas the site-means for wool ranged between −13.1‰ and −23.7‰ (Fig. 2b). The δ 13 C of the C3 components varied between −23.3‰ and −28.5‰ (mean: −25.0‰), the δ 13 C of the C4 components of vegetation between −12.5‰ and −15.8‰ (mean: −14.2‰). Evidently, the δ 13 C of the C3 and C4 components exhibited variations three to four times smaller than that of 25 whole vegetation (Fig. 2a) indicating a variation in the C3/C4 ratio (see below).

555
Bulk vegetation and wool differed markedly in their density distributions of δ 13 C ( Fig. 2a, b). These differences corresponded almost exactly to the isotopic shift, which was observed by paired sampling: on average, wool was enriched by 2.7‰ (n=76, CI 95% =0.7‰) relative to vegetation sampled at the same location. There was no indication of preferential grazing in the grazing experiment, when the shift was regressed ei-5 ther against stocking densities or against P C4 derived from vegetation analysis (P >0.05 and P >0.75, respectively). Correcting for this independently derived shift lead to similar density distributions for vegetation derived from either vegetation itself or from wool (Fig. 2b). The main difference between the density distributions then is that the extremes were lost when using wool, reflecting the larger temporal and spatial integration 10 as compared to the sampling area area of vegetation. The density peak for δ 13 C of vegetation derived from wool (−23.4‰) was close to the median (−23.6‰) and the mean (−22.8‰) of δ 13 C from bulk vegetation.

The relationship between δ 13 C and environmental gradients
The response of δ 13 C to the environmental variables was the same for vegetation 15 and wool with no significant differences in slope of the regression, which were very highly significant for mean July temperature, MAP and altitude (Table 1) with similar strength. However, the correlations were rather weak and effects were not independent, since there were also close correlations between the mean July temperature and MAP (R 2 =0.75, P 0.001) as well as between mean July temperature and alti-20 tude (R 2 =0.41, P 0.001) and MAP and altitude (R 2 =0.26, P 0.001). As a result, the effects of environmental variables on δ 13 C of vegetation could not be separated, and each response function of δ 13 C to an environmental variable included direct and indirect effects.  6‰; n=19), a common perennial xerophytic C4 grass in Inner Mongolia grassland, and other C4 species (mean 5.5‰; n=27; see also the two shoulders in the density 5 distribution of δ 13 C of C4 component in Fig. 2a). However, no significant relationship between 13 ∆ 4 and normal-period (P >0.60) or annual precipitation (P >0.75), normalperiod (P >0.80) or annual temperature (P >0.10), or plant available soil water (P >0.23, calculated after Allen et al., 1998 andSchnyder et al., 2006) was found. As no indication for a relevant regional variation was evident the mean 13 ∆ 4 was used to calculate 10 P C4 from 13 C of wool samples considering the isotopic shift between vegetation and wool.
On average of all wool samples, C4 plants accounted for 19% of aboveground plant biomass, but P C4 varied widely between sites (0 to 85%). C4 abundance increased by 6% with an increase of 1 • C of mean July temperature (normal period, P 0.001), 15 decreased by 6% for each 100 mm yr −1 increase in MAP (P <0.05). Although all these relationships were significant, the scatter was large in all of them (R 2 between 0.04 and 0.10).

Geographic variation of C4 abundance
The theoretical semivariogram followed a spherical model for P C4 (Fig. 3). The range 20 was approximately 260 km and the nugget-to-sill ratio was 0.4. Hence the nugget effect corresponds to an uncertainty of approximately 16% for P C4 . The nugget/sill ratio indicates further that 60% of the variation was caused by a spatial pattern (Table 2). This is much more than what could be explained with regression analysis (Table 1) assuming linear relationships to isolated environmental variables. 25 The map of P C4 as reflected in δ 13 C of wool generated by ordinary block kriging ranged only between 0 and 40% because block kriging separates the local variation 557 (noise) from the pattern (regional variation). The map (Fig. 4) is characterized by an increase of P C4 from 0% in the south-eastern part to around 40% in the western part of the study area near the Gobi desert and also by a "C3 favoring valley" with P C4 <20% from south-east to north-east. From Collatz et al. (1998)   The short temporal integration of vegetation samples likely leads to a bias of the results by the time of sampling as C4 develops considerably later than C3 in this area (Bai et al., 2004). This causes a high variability in δ 13 C of vegetation samples leading to a semivariance about five times as high as for wool samples (nugget: 690% 2 ver-20 sus 130% 2 ) and a semivariogram being almost a complete nugget (nugget/sill ratio of 73%) with a large RMSE (Table 2), which does not allow to retrieve a spatial pattern from vegetation analyses. Whole wool, however, integrates the isotopic signal of feed ingested between two successive shearing (one year) on an area that encompasses the grazing grounds and the haymaking areas of an entire farm (several km 2 ). be large. Inter-annual weather variability (e.g., dry years versus wet years; Wittmer et al., 2008) and a semi-arid situation, where moisture availability is almost entirely driven by precipitation but not by the spatially varying storage capacity of soil, certainly can alter the spatial patterns between years. Theoretically, inter-annual variations can be excluded by pooling the data (Schuurmans et al., 2007) by years prior to the calcu-5 lation of the empirical semivariogram. Though, this procedure neither improved the semivariogram (actually the nugget and the nugget/sill ratio were higher) nor the kriging prediction (all not shown). The main random variability hence mainly resulted from the intra-annual variation of the C3/C4 ratio but not from the inter-annual variation and this intra-annual variability was excluded by using wool instead of vegetation.

Isotopic shift between vegetation and wool
Preferential grazing is one mechanism by which grazers can affect vegetation composition (Caswell et al., 1973), and hence carbon pools, fluxes and their isotopic composition. Yet, the present data indicated no preferential grazing of one photosynthetic type (C3 relative to C4, or vice versa). This was suggested by the fact that the observed 15 vegetation-wool shift (+2.7‰) was similar as shifts with pure C3 or C4 diets (average shift: +2.6‰±0.6‰ SD for hairs of different animal species in the studies of Minson et al., 1975;Jones et al., 1981;Sponheimer et al., 2003a;Ayliffe et al., 2004;Schwertl et al., 2005). If animals had grazed preferentially one component of vegetation, then the apparent shift should deviate with a shift <2.7‰, if the sheep preferentially grazed 20 the C3 component, and a shift >2.7‰ for preferential grazing of the C4 component. Preferential grazing would also result in an either right (=more negative; C3 preferred) or left (=less negative; C4 preferred) skewed density distribution of wool compared to vegetation. This was not the case as the density distributions for wool with applied shift and for vegetation apparently were centered (Fig. 2b). In the density distribution of 25 wool, however, the extremes on both sides of the vegetation density distribution were clipped due to the spatio-temporal integration by the animals. And finally, preferential grazing should cause a correlation between the shift and either stocking densities or 559 P C4 , which both determine the chance for selective grazing. Such relations were not found in the grazing experiment. Furthermore, if selection would happen opportunities would be small on an annual basis with only one growth cycle due to the high grazing pressure caused by the high stocking rate in the study area (Sneath, 1998). Differences in digestibility between C3 and C3 species can displace the signal in 5 the wool and thus influence the vegetation-wool shift. Nonetheless several studies reported that the nutritional quality and/or digestibility of C4 species is inferior to that of C3 species (e.g., Sponheimer et al., 2003b;Barbehenn et al., 2004) the opposite was also reported (Scheirs et al., 2000). Digestibility decreases strongly with age of the plant material (Fick et al., 1994). An alternating ranking of the digestibility can 10 thus be expected from the asynchronous growth of both photosynthetic types where the greening of the C4 species starts when the some aboveground parts of the C3 species are almost two months old. For that reason, further investigation on individual and combined digestibility of C3 and C4 species are needed. However, this has only minor influences on our results, as the reported differences in digestibility between C3 15 and C4 species rarely exceed 10-15% (e.g., Sponheimer et al., 2003b;Norman et al., 2008) and the overall C4 abundance in our study area is not higher than 40%. Postulating a direct effect of digestibility on the pattern of C4 abundance, this would lead to maximum error in the estimation of 6%. Anyhow, this would not have the potential to distinctly alter the C4 abundance pattern as derived from wool. Several studies reported variations of 13 ∆ 4 along environmental gradients e.g., a decrease in 13 ∆ 4 with increasing precipitation (Schulze et al., 1996) or, in contrast, an increase of 13 ∆ 4 with increasing soil water content (Chen et al., 2002). The absence of any relationship to climatic gradients (temperature, precipitation and soil water parameters) in our data set may thus be caused by compensating effects. Furthermore, C4 species of different metabolic types (see Pyankov et al., 2000) can differ significantly Interactive Discussion in 13 ∆ (e.g., Schulze et al., 1996). This was apparent in our data, as C. squarrosa (NAD-ME metabolic type, Pyankov et al., 2000) had a 1.1‰ higher 13 ∆ than other metabolic types (mainly NADP-ME). Several studies reported that the abundances of different metabolic types are related to precipitation gradients with abundance of NADP-ME plants increasing with increasing precipitation at the expense of NAD-ME 5 plants ( Wang et al., 2003;Cabido et al., 2008;Osborne, 2008). This should lead to an increase of 13 ∆ 4 with precipitation. This was not obvious in our data as C. squarrosa was a component of the C4 community all over the sampling area. Additionally, there was no apparent pattern in the occurrence of the different metabolic types (data not shown) with the exception of Chloris virgata, an annual grass of the PEP-CK metabolic type, which was restricted to the drier (western) part of the sampling area. However, C. virgata was never a dominant component the C4 community. Importantly, the present estimates of P C4 did not change very much when we varied the C4 end-member value of the C3-C4 mixing equation. If 13 ∆ 4 was set to 5.5‰ (100% NADP-ME species in C4 community), then the resulting mean P C4 was 18.2%. Conversely, if 13 ∆ 4 was set to 15 6.5‰ (assuming that the C4 community consisted entirely of NAD-ME species, such as C. squarrosa), then the mean P C4 was 19.8%. So, any reasonable error in estimation of 13 ∆ 4 had a negligible effect on the spatial variation of P C4 (Fig. 4).
4.4 Regional variation in C3/C4 composition is related to temperature gradients Competition between C3 and C4 photosynthetic types is controlled by ambient CO 2 20 concentration and temperature during the growing period (Ehleringer et al., 1997) with high CO 2 concentrations favoring C3 plants and high temperatures favoring C4 plants. The latter agrees with the conclusions of Pyankov et al. (2000) regarding the temperature-dependent distribution of C4 grasses in (Republic of) Mongolia. Due to the simultaneous increase in temperature and precipitation during the growing season 25 (Xiao et al., 1995) the temperature of the warmest month is effective for this competition. The warmest-month isotherm for the predicted crossover temperature follow-561 ing Collatz et al. (1998)  On the other hand, for the lower CO 2 concentration 30 years ago the crossover temperature was about 2 • C lower according to Collatz et al. (1998). Although on average both effects seem to compensate each other numerically, this was not the case for the 10 spatial pattern. The 20 • C crossover isotherm of the last normal period did not differentiate between areas of significantly different P C4 (means 19 vs. 14%) while the mean difference in P C4 between the areas above and below the present crossover isotherm of 22 • C was 13% (23 vs. 10%) and highly significant. Hence, the C4 abundance followed the crossover temperature isotherm calculated for the conditions during the sampling 15 period. The good agreement even despite a flat gradient in summer temperature indicates the dominant influence of differences in quantum yield on the regional pattern while the large variation found on smaller scales are hence likely to be caused by controls acting on smaller scales. To our knowledge, this is the first proof on a regional scale for the predicted crossover temperature following Collatz et al. (1998). It shows 20 that the present community reflects present conditions despite large changes in CO 2 concentration and temperature in the past.

Conclusions
This work demonstrates that wool is qualified for assessing the abundance of C4 plants in the Inner Mongolia grassland on the regional scale. This is advantageous because 25 it allows covering the whole growth cycle, many years and large areas within short sampling campaigns. An enormous variation on P C4 occurred despite the integration 562  20.5-25.0 • C 0.8/1 • C ±0.151 0.10 * * * 264 * , * * , * * * denote error probability smaller than 0.05, 0.01 and 0.001, respectively.   Mean krige standard deviation for the study area is 9.4%. Towns are the same as in Fig. 1. The 22 • C isotherm reflects meteorological conditions during the period 1998-2007 and it is equivalent to the predicted crossover temperature for the CO 2 concentration during this period.