Warming and free-air CO2 enrichment alter demographics in four co-occurring grassland species

Authors


Author for correspondence: Mark Hovenden
Tel: +61 36226 7874
Fax: +61 36226 2698
Email: Mark.Hovenden@utas.edu.au

Summary

  • • Species differ in their responses to global changes such as rising CO2 and temperature, meaning that global changes are likely to change the structure of plant communities. Such alterations in community composition must be underlain by changes in the population dynamics of component species.
  • • Here, the impact of elevated CO2 (550 µmol mol−1) and warming (+2°C) on the population growth of four plant species important in Australian temperate grasslands is reported. Data collected from the Tasmanian free-air CO2 enrichment (TasFACE) experiment between 2003 and 2006 were analysed using population matrix models.
  • • Population growth of Themeda triandra, a perennial C4 grass, was largely unaffected by either factor but population growth of Austrodanthonia caespitosa, a perennial C3 grass, was reduced substantially in elevated CO2 plots. Warming and elevated CO2 had antagonistic effects on population growth of two invasive weeds, Hypochaeris radicata and Leontodon taraxacoides, with warming causing population decline. Analysis of life cycle stages showed that seed production, seedling emergence and establishment were important factors in the responses of the species to global changes.
  • • These results show that the demographic approach is very useful in understanding the variable responses of plants to global changes and in elucidating the life cycle stages that are most responsive.

Introduction

Plant species distribution and abundance are tightly regulated by climatic factors (Woodward, 1987), so global changes are likely to cause major shifts in the ranges and abundance of species (Thomas et al., 2004). As even co-occurring species vary considerably in their responses to environmental factors, human-induced global changes are likely to alter community and ecosystem structure and function. The species composition of plant communities is an important controller of ecosystem function, particularly water, carbon and nutrient cycling, as species vary in water and nutrient use, carbon assimilation and respiration rates and decomposition processes (Campell & Stafford Smith, 2000). Plant species composition is also an important determinant of ecosystem services, such as provision of habitat, fodder, water interception and carbon storage (Wardle et al., 2000; Hooper et al., 2005).

Any alterations of community structure caused by global changes must be the result of alterations of the population dynamics of the component species. Thus, knowledge of the demographic response of plant species to global changes is a key component of our ability to predict future ecosystem function. There have been many investigations of the impacts of global changes on the biology of plant species, both under glasshouse conditions and in field-based manipulations (Long et al., 2004; Aerts et al., 2006). However, few of these investigations have concentrated on more than one life cycle stage and no published studies have examined the impacts of global changes on the entire life cycle of any plant species. This is probably a consequence of the difficulties of collecting long-term demographic data and a lack of acknowledgement of the importance of obtaining a population-level perspective of global change impacts. Thus, it is not atypical to find in the literature examinations of one variable, say the growth rate of young plants, to a particular environmental factor that might be changing, say CO2 concentration. The results from such an examination are then used to predict the impact of future changes in that environmental variable on the long-term viability of the species (Navas et al., 1999; Poorter et al., 2003). Such examinations provide useful hypotheses for study but neglect the importance of other stages in the life cycle and the transitions between stages. This is especially important where a species may have particularly responsive or susceptible life cycle stage transitions (Schemske et al., 1994). Therefore, it is important to obtain an integrated assessment of the impact of global changes on the life cycle of a species if the true impacts of such changes are to be predicted. Population matrix models allow such an assessment as they allow us to link the impact of global changes on the individual with the growth and persistence of the population (Caswell, 2001). Construction and analysis of population matrices, combined with a technique called life table response experiment analysis (Caswell, 2001; Brys et al., 2004; Fréville & Silvertown, 2005), also indicate which life cycle stages or transitions are particularly important to population growth and/or susceptible to global changes (Schemske et al., 1994). Additionally, population matrices can be modelled, allowing predictions of population futures under various scenarios. Thus, population matrix models are a powerful tool for the investigation of the impacts of global changes on vegetation and ecosystems.

The Tasmanian free-air CO2 enrichment (TasFACE) global change impacts facility is located on basalt-derived soils in a native, species-rich, low-altitude grassland in the southern midlands of Tasmania, Australia (Hovenden et al., 2006). The plant community is dominated by native perennial grasses but also contains many native forbs as well as introduced grasses and forbs, both annual and perennial. The TasFACE facility aims to determine the impacts of elevated CO2 and warming, in both isolation and combination, on the Australian temperate grassland ecosystem. Central to this aim is the elucidation of plant population processes that lead to changes in plant community composition. Thus, the aim of this investigation was to determine whether elevated CO2 and warming affected the population growth rate of four significant species: the two dominant native perennial grasses, Austrodanthonia caespitosa and Themeda triandra, and the two most abundant and invasive introduced species, Hypochaeris radicata and Leontodon taraxacoides, both of which are perennial forbs.

Materials and Methods

The TasFACE climate change impacts facility consists of 12 plots randomly arranged in a 2 × 2 factorial design with three replicates of each treatment combination located in a native lowland grassland in the southern midlands region of Tasmania, Australia. Experimental treatment of the native grassland commenced in February 2002 and is continuing. Full site, experimental, technical design and performance details are given in Hovenden et al. (2006). Briefly, elevated CO2 and warming were applied in 1.5-m-diameter circular plots, with CO2 concentration in the elevated CO2 plots increased to 550 µmol mol−1 by pure-CO2 fumigation free-air CO2 enrichment (FACE; Miglietta et al., 2001), which operated from sunrise to sunset. The system uses proportional control of CO2 concentration ([CO2]) in the centre of each plot by manipulating CO2 supply with electropneumatic flow control valves (CKD USA, Rolling Meadows, IL, USA). Valves are controlled via a microprocessor-based control system running a proportional integration device algorithm (Hendrey et al., 1993, 1999). CO2 control is excellent, with the central [CO2] being within 10% of the set-point for 86.7 ± 0.3% (mean ± SE) of the time. The pure-CO2 injection system also ensures that [CO2] is stable both temporally and spatially (Hovenden et al., 2006). Thus, the plants within the TasFACE experiment do not experience large daytime excursions in [CO2] that can lead to undesirable effects (Holtum & Winter, 2003). Continuous elevation of [CO2] might elicit a different response to elevation only during daylight (Bunce, 2005). At the TasFACE site, nocturnal soil CO2 emissions and wind interact to cause variation from one night to the next in the [CO2] below c. 20 cm (Hovenden et al., 2006). Soil CO2 emissions also cause significant vertical stratification of [CO2] in this experiment. Both of these natural phenomena would be disrupted by nocturnal fumigation, so it was decided not to fumigate at night. Thus, there were no differences in mean night-time [CO2] between the control and FACE plots. Mean daytime [CO2] in the control plots was 372 ± 0.3 (mean ± SE) µmol mol−1 compared with 549 ± 0.1 µmol mol−1 in the elevated CO2 plots.

Warming of 2.0°C was achieved by using overhead ceramic infrared (IR) heat lamps (Salamanda ESE250 240 V 250 W Emerson Solid Ceramic Infrared Emitter; Delta-T, Melbourne, Australia) that operated continuously, supplying 140 W m−2 of infrared radiation. The IR lamps produced no visible radiation and, on average during the growing season (April–December), elevated canopy temperature by 1.98 ± 0.05°C (mean ± SE) and soil temperature at 5 mm by 0.82 ± 0.04°C. Night-time warming was more pronounced than daytime warming because of the influence of direct solar radiation. Thus, canopy temperature was elevated by an average of 1.54 ± 0.20°C during the day and 2.61 ± 0.14°C during the night. Similarly, soil warming at 5 mm depth was 1.79 ± 0.10°C during the night but only 0.14 ± 0.013°C during the day. Warming and elevated CO2 treatments interacted to affect soil moisture content, with the mean soil water potential for the 3-yr period being –113 ± 19 kPa in unwarmed control plots, –112 ± 18 kPa in warmed control plots, –91 ± 13 kPa in unwarmed elevated CO2 plots and –126 ± 12 kPa in warmed elevated CO2 plots. Differences in soil moisture were most pronounced during periods of intermediate water availability and disappeared under wet or dry conditions.

We recorded seed production, seedling emergence, seedling survival and adult survival of four perennial species that were abundant in replicate plots of each treatment combination from spring (September) 2003 until summer 2006. The species were the two dominant native perennial grasses, including one C3 species, Austrodanthonia caespitosa (Gaudich.) H.P.Linder, and one C4 species, Themeda triandra Forsskal, and the two most dominant invading species, both herbaceous dicots from the family Asteraceae, Hypochaeris radicata L. and Leontodon taraxacoides (Vill.) Mérat. Other species were also observed, but these either disappeared from some treatments over the course of the study or were present in insufficient numbers of individuals or in too few replicate plots to allow analysis of the full life cycle for 3 yr. Seed production was estimated by counting the number of plants of each species per plot and the number of inflorescences produced per plant weekly for each species throughout the flowering period. One inflorescence was collected from at least 10 different plants of each species in each plot once seeds were set. The average number of seeds per inflorescence was counted to estimate seed production per plant. Seedling emergence and survival were both recorded monthly using a grid of one hundred 5 × 5 cm cells in the centre of each plot. As each seedling emerged, its presence in the grid was recorded and the seedling encircled by a hoop made from wire coated in coloured plastic to allow the fate of each seedling to be determined. Adult survival was determined by counting the numbers of adult plants in each plot every 3 months. All data were converted to densities to calculate stage transitions for each species in each year and each plot using the model life cycle given in Fig. 1. These four species do not possess a persistent seed bank in this grassland, which was determined by exhaustive soil sampling at various times of year, and therefore there is no seed-to-seed transition in the model. This life cycle model was appropriate for all four species as all the species were perennials without a seed bank and the seedling establishment period was too short for newly emerged seedlings to reproduce before the census. During monitoring over 3 yr, we noticed no asexual reproduction in any of these four species. The starting adult population density for A. caespitosa was 105 ± 17 (mean ± SE) in unwarmed control plots, 89 ± 6.7 in warmed control plots, 112 ± 25 in unwarmed elevated CO2 plots and 97 ± 14 in warmed elevated CO2 plots. The corresponding figures were 40 ± 16, 51 ± 21, 60 ± 19 and 67 ± 24 for T. triandra, 47 ± 26, 15 ± 7.5, 25 ± 2.0 and 117 ± 54 for H. radicata and 105 ± 48, 33 ± 3.0, 83 ± 30 and 40 ± 25 for L. taraxacoides. Life cycle stage transitions were calculated for each plot in each year and then averaged across years to produce an average stage transition, or vital rate, for each plot from the 3 yr of observations. Data were converted to transitions per individual, thus standardizing for variation in initial population density. These data were entered into a square projection matrix, A, for each plot. The population projection is given independently for each plot by applying the formula Nt+1 = ANt, where Nt and Nt+1 are the vectors whose elements ni(t) correspond to the number of individuals in stage i at time t and time t + 1, respectively, and A is a nonnegative square matrix whose elements aij are the numbers of stage i individuals produced per stage j individual per unit time (Caswell, 2001; Fréville & Silvertown, 2005).

Figure 1.

Three-stage life cycle applied to all four species. The four life cycle stage transitions are shown as arrows labelled G (germination), E (establishment), S (survival) and R (reproduction).

Our models assumed that local processes (births and deaths) were the dominant factors in determining population growth and therefore we ignored both immigration into and emigration from the population. For each plot, we calculated the asymptotic population growth rate, λ, as the dominant eigenvalue of the projection matrix, A, for that plot. In order to determine the importance of the four vital rates to population growth, we analysed the response of population growth rate to a small, fixed change in each of the four life stage transitions: germination, establishment, survival and reproduction. This is termed matrix sensitivity to aij (sij) and is a measure of the relative importance of proportional changes in each vital rate to population growth, λ. The value of sij is calculated as:

image( Eqn 1)

(where vi and wj are respectively, the left and right eigenvectors of the matrix A; <vw>, their scalar product (Silvertown & Charlesworth, 2001). Because of differences in scales of different transitions, particularly fecundity, matrix elasticity is often used to overcome problems of comparing numerically differing vital rates. Elasticity (eij) is a modified index of sensitivity and is calculated as:

image(Eqn 2)

Elasticity represents the proportion of λ that results from the transition aij and sums to unity within a matrix (Silvertown & Charlesworth, 2001). Analyses of λ and matrix elasticities and sensitivities were calculated from individual plot matrices as per Caswell (2001) using Poptools software (Hood, 2006). These results were analysed by analysis of variance using general linear model procedures in the SAS statistical software package, version 9.1 (SAS Institute Inc., 2003).

A life table response experiment (LTRE) analysis was used to determine how treatment-induced alterations of individual vital rates contributed to the overall difference in population growth rate (Caswell, 2001) using exactly the procedure described in Fréville & Silvertown (2005). This procedure estimates the change in λ that would occur from the effect of a treatment on a single vital rate, with all other vital rates held constant. Calculating this effect on λ for each of the separate stage transitions identifies the relative importance of treatment-induced alterations of vital rates for the total, population-level response of the species. To calculate these values, treatment effects for each species were determined by creating an average projection matrix for each CO2× warming combination by calculating the arithmetic means of the projection matrix for each plot in each treatment combination. Thus, four treatment projection matrices were created, with transition values as per Table 1. For each species, the effect of each treatment on λ (i.e. λT – λC, where λΤ and λC are the population growth rate in treated and control plots, respectively) was estimated as:

Table 1.  Average vital rates of four species in warmed and unwarmed conditions in elevated CO2 and control plots in the Tasmanian free-air CO2 enrichment (TasFACE) experiment between spring (September) 2003 and spring 2006
CO2WarmingGerminationEstablishmentSurvivalReproduction
  1. Values are mean ± standard error.

  2. Values for germination, establishment and survival are probabilities. Values for reproduction are mean seeds produced per established plant.

Austrodanthonia caespitosa
ControlUnwarmed 0.44 ± 0.080.45 ± 0.090.55 ± 0.14100.9 ± 83.8
Warmed 0.39 ± 0.070.48 ± 0.120.67 ± 0.03 16.6 ± 3.0
Elevated CO2Unwarmed 0.36 ± 0.170.63 ± 0.010.59 ± 0.17  9.3 ± 3.4
Warmed0.026 ± 0.0020.21 ± 0.010.81 ± 0.04 19.5 ± 4.0
Themeda triandra
ControlUnwarmed 0.05 ± 0.020.61 ± 0.110.80 ± 0.16 22.0 ± 4.3
Warmed 0.22 ± 0.070.18 ± 0.180.91 ± 0.08 16.5 ± 3.8
Elevated CO2Unwarmed 0.19 ± 0.080.42 ± 0.080.86 ± 0.07 16.0 ± 3.9
Warmed 0.30 ± 0.180.70 ± 0.030.80 ± 0.11 16.6 ± 1.8
Hypochaeris radicata
ControlUnwarmed 0.45 ± 0.340.52 ± 0.050.55 ± 0.20101.2 ± 74.6
Warmed 0.04 ± 0.010.29 ± 0.010.46 ± 0.06103.5 ± 80.2
Elevated CO2Unwarmed 0.34 ± 0.170.66 ± 0.040.45 ± 0.11359.1 ± 258.8
Warmed 0.13 ± 0.040.41 ± 0.030.19 ± 0.17 39.3 ± 30.2
Leontodon taraxacoides
ControlUnwarmed 0.45 ± 0.090.50 ± 0.050.48 ± 0.13104.5 ± 65.1
Warmed0.018 ± 0.0020.23 ± 0.060.56 ± 0.36195.6 ± 104.0
Elevated CO2Unwarmed 0.65 ± 0.170.72 ± 0.070.61 ± 0.07 61.1 ± 17.9
Warmed 0.03 ± 0.040.29 ± 0.080.52 ± 0.18 33.0 ± 8.6
image(Eqn 3)

(aijT and aijC, the aij values for treatment and control matrices, respectively; the derivative term is the sensitivity of λ to variation in aij, sij (Caswell, 2001).) We calculated sij from the arithmetic mean matrix of the control and treated plots. Thus, the product of the difference in the vital rate between treated and control plots and the sensitivity of λ to changes in that vital rate can be used to determine the contribution of treatment-induced alterations of each vital rate to population growth rate. We used this method to determine the effect on λ of changes in germination, establishment, survival and reproduction produced by warming, elevated CO2 and elevated CO2 and warming combined for each of the four species.

Results

The impact of elevated CO2 and warming on the finite rate of population growth, λ, was dependent upon species (Fig. 2). Elevated CO2 significantly reduced the population growth rate of the dominant perennial grass in the experiment, A. caespitosa (P < 0.02). While warming also appeared to reduce λ (Fig. 2), this reduction was not statistically significant (P = 0.10) and there was no significant CO2× warming interaction (P = 0.73). However, λ for A. caespitosa was < 1 in the warmed plots at elevated CO2, indicating population decline (Fig. 2). By contrast, λ of the other dominant perennial grass species, T. triandra, a C4 species, was not significantly affected by either elevated CO2 (P = 0.2) or warming (P = 0.2), nor was there a significant CO2× warming interaction (P = 0.2; Fig. 2). However, λ for T. triandra was highest in the warmed elevated CO2 plots and lowest in the control plots (Fig. 2). The population growth rates of both weed species, H. radicata and L. taraxacoides, were substantially reduced by warming (P < 0.02 for both). The population growth rate of L. taraxacoides was not affected by elevated CO2 or by a CO2× warming interaction (Fig. 2). For H. radicata, λ was greatest in the unwarmed elevated CO2 plots, but variation among plots meant that this was not significantly higher than in unwarmed control plots (P = 0.08).

Figure 2.

Asymptotic rate of population growth (λ) in unwarmed (open columns) and warmed (closed columns) conditions at control and elevated CO2 concentrations for four species from the Tasmanian free-air CO2 enrichment (TasFACE) experiment. Values are mean ± standard error. The dotted line on each graph indicates λ = 1, the point of population stasis.

Elevated CO2 significantly changed the matrix elasticity values for A. caespitosa only, with values for the other three species being unaffected by CO2 concentration (Fig. 3). Elevated CO2 increased the elasticity of the A. caespitosa population growth rate to changes in adult survival and decreased the importance of each of the other three transitions (Fig. 3). In fact, the elasticity of the A. caespitosa population growth rate to changes in adult survival increased from 16.1 ± 3.3% (mean ± SE) to 51.7 ± 12.8% upon exposure to elevated CO2 (P < 0.017). This indicates that elevated CO2 increased the relative impact on the A. caespitosa population growth rate of changes in adult survival and reduced the impact of changes in the other transitions. Warming had no significant effect on elasticity values of any of the four species, nor were there any significant CO2× warming interactions. Thus, elevated CO2 and warming did not significantly affect the relative importance of proportional changes in each stage transition to population growth rate in T. triandra, H. radicata or L. taraxacoides (Fig. 3).

Figure 3.

Elasticities of population growth rate to proportional changes in each stage transition at current and elevated [CO2] with (W) and without (U) warming for four species from the Tasmanian free-air CO2 enrichment (TasFACE) experiment. Values are mean ± standard error.

While elasticity analysis indicated the relative importance of proportional changes in each transition to population growth, the experimental treatments did not produce proportional changes in the stage transitions. LTRE analysis determines the actual impact on λ of real treatment-induced changes in vital rates. Thus, the results presented in Fig. 4 show, for each species, the effect on λ of the treatment-induced changes in germination, establishment, survival and reproduction given in Table 1. Therefore, λ of A. caespitosa was decreased by negative effects of warming, elevated CO2 and their combination on germination and, more substantially, on reproduction, with the impact of changes in germination being most pronounced in the combined warming and elevated CO2 treatment, in which reductions in establishment also reduced λ (Fig. 4). By contrast, the population growth of T. triandra was largely unaffected by treatments (Fig. 2), as the positive effect on λ of increases in germination were cancelled by negative effects on λ of decreases in establishment and/or reproduction (Fig. 4). The negative effect of warming on germination also had a substantial impact on λ in both H. radicata and L. taraxacoides (Fig. 4). For L. taraxacoides, this was also true for seedling establishment such that the positive effect on λ of increased reproduction in warmed plots was overwhelmed by the negative effects of warming on germination and establishment (Fig. 4). Elevated CO2 in the absence of warming substantially increased reproduction of H. radicata and this had a positive impact on λ (Figs 2, 4), although variation among elevated CO2 plots (Table 1) meant that this effect was not significant (Fig. 2). In all species, adult survival was largely unaffected by experimental treatments and thus had little impact on simulated global changes on λ (Fig. 4).

Figure 4.

Life table response experiment (LTRE) results showing the effect on population growth rate (λ) of the treatment-induced alterations in each vital rate, as compared with that of the unwarmed control plots. Treatment codes: warming (W), elevated [CO2] (C) and their combination (W + C) for four species from the Tasmanian free-air CO2 enrichment (TasFACE) experiment.

Discussion

Here, the impact of experimental warming and elevated CO2 is shown on projections of population growth of four perennial plant species important in south-eastern Australian grasslands. While some of the impacts of global changes appear dramatic, the results are projections, not predictions. Matrix model analysis and projection provide unparalleled insight into the overall impact of experimental manipulation on the population as a whole and indicate which life cycle stages, if any, are particularly susceptible to environmental perturbations (Caswell, 2001). However, projections of population growth rely on a reliable definition of important life cycle stages and accurate estimation of all life cycle stage transitions or vital rates. Our models ignore immigration into and emigration from the population within each plot. The surrounding grassland is mowed to prevent seed dispersal input into plots from adjacent plants, so there is likely to be little immigration. As the experimental site is fenced, it is likely that there is little seed dispersal by vertebrates and therefore the underestimates are likely to be negligible for T. triandra and A. caespitosa. However, both H. radicata and L. taraxacoides have wind-dispersed seed and thus there was probably some emigration from the plots and therefore our estimates of population growth are underestimates. However, the underestimates are probably constant across treatments and therefore the comparisons remain valid. Further, we have only monitored the impact of simulated global changes on stage transitions for 4 yr and thus the projections must be viewed as being representations of population responses over this period. However, we were fortunate during the observation period to have a wide range of climatic conditions, including spring seasons belonging to each of the wettest 30% and driest 30% from the past century. There was no evidence that herbivory, seed predation, disease or other biotic interactions varied substantially across the years. Thus, while the observations are limited in temporal scope, they cover a substantial range of climatic conditions and thus the assessment of the impacts of elevated CO2 and warming on λ for these species is likely to be robust. Most importantly, however, use of matrix models and LTRE analysis has allowed us to elucidate the impacts of simulated global changes on the entire life cycle of these species, which is superior to analyses of photosynthesis, growth or single life cycle stages.

Plants using the C4 photosynthetic pathway (C4 plants) use a biochemical pathway to concentrate CO2 in their chloroplast-containing cells (Hatch & Slack, 1966). Photosynthesis in plants lacking this pathway (C3 plants) is therefore more limited by the availability of CO2 under current atmospheric conditions. Thus, photosynthesis, and therefore growth, of C3 plants are generally hypothesized to be stimulated by the increasing atmospheric concentration of CO2 to a greater extent than those of C4 plants (Ehleringer et al., 1997; Wand et al., 1999). Indeed, much research has gone into comparing the responses of C3 and C4 plants to the increase in CO2 both in artificial environments and in the field (Bowes, 1993). Most of this research has indicated that the growth of C3 species is indeed generally stimulated by elevated CO2 to a greater extent than that of C4 species (Wand et al., 1999). Hence, we were surprised to find that the population growth rate of A. caespitosa, one of the dominant C3 grass species of southern Australia, was substantially and significantly reduced by exposure to elevated CO2, most markedly in warmed conditions. The LTRE analysis indicated that elevated CO2 reduced seed production per plant, thus reducing the potential number of seedlings, which would limit recruitment and lead to population decline. The combination of warming and elevated CO2 also reduced seedling germination/emergence and establishment, compounding the effect of reduced seed production. The impact of elevated CO2 on seed production varies substantially among species, with highly negative to highly positive effects being reported (Jablonski et al., 2002). As an elevated CO2-induced lowering of seed production led to a reduction in population growth rate of A. caespitosa, this variation may well be of extreme ecological significance and lead to large changes in the relative abundance of co-occurring species. Indeed, elevated CO2-induced changes in seed production in a grassland community did result in changes in biomass and abundance of those species (Edwards et al., 2001). It is likely that treatment-induced alterations of soil water availability had a substantial impact on seed production, seed quality and germination/emergence. This is particularly likely in the warmed elevated CO2 plots in which soil moisture was substantially lower than in control plots. However, soil moisture differences are unlikely to be responsible for the reduction in reproduction in the unwarmed elevated CO2 plots, as soil moisture levels were highest in these plots. Thus, elevated CO2 appears to have direct effects on reproductive output in this species.

By contrast, the population growth rate of T. triandra, the dominant C4 grass of south-eastern Australia, was not affected by elevated CO2 or warming. Indeed, the combined elevated CO2 and warming treatment slightly increased the population growth rate of this species (P = 0.2). While seed production was less strongly suppressed by elevated CO2 in T. triandra than it was in A. caespitosa, it is likely that the germination/emergence stage transition is the key to the different responses of the two species, as both warming and elevated CO2 increased seedling emergence of T. triandra but not of A. caespitosa. As the stimulation of photosynthesis in C3 plants by elevated CO2 is more pronounced at higher temperatures (von Caemmerer, 2000), it might be expected that the combination of warming and elevated CO2 would confer a greater advantage on C3 species than on C4 species (Sage & Kubien, 2003). Our results indicate that this was definitely not the case, with T. triandra being the only species in which the population growth rate was increased by warmed elevated CO2 conditions, with that of all other species being substantially reduced. Thus, these results indicate that predictions of the effect of global changes based on understanding of physiology are not necessarily correct. It is also possible that soil moisture played an important role in the relative success of T. triandra over A. caespitosa. As T. triandra has a higher water use efficiency than A. caespitosa (Hovenden, 2003), it is possible that T. triandra was less affected by the warming-induced reduction in soil moisture at elevated CO2 and may even have benefited from altered competition from co-occurring C3 species.

Because global changes such as the increasing atmospheric concentration of CO2 produce a suite of secondary effects on factors such as nitrogen availability (Luo et al., 2004), it is possible that the way species respond to these secondary changes are stronger determinants of population success than the photosynthetic pathway. Elevated CO2 and warming are also likely to have altered the degree of competition amongst the co-occurring species, and this may be equally or more important in determining the effect of global changes on communities (Brooker, 2006). Our results also support previous observations that the way in which plants allocate resources is very important in determining their response to global changes (Coleman & Bazzaz, 1992; Schafer et al., 2003; Zavaleta et al., 2003a). Elevated CO2 caused H. radicata to substantially increase seed production, in contrast to the responses of the other three species, which affected population growth rate whereas the addition of warming to elevated CO2 substantially reduced seed production and resulted in a reduction of population growth rate.

The contrasting responses of the two codominant perennial grass species have important implications for the structure and function of this ecosystem. It appears highly likely that A. caespitosa abundance in the community will diminish in the future, as the combined elevated CO2 and warming treatment substantially reduced the population growth rate to a point where population decline is likely. This contrasts sharply with the result for the co-occurring T. triandra, in which the population growth rate was greatest in the combined treatment. This indicates that T. triandra will probably become the dominant grass species and A. caespitosa will either become a minor component of the community or will disappear entirely. These two species have different photosynthetic pathways, water use efficiencies and net primary productivities, are structurally different and provide different nutritional benefits to herbivores (Garden et al., 2005; Waters et al., 2005). It is also likely that the two species have differing litter decomposition rates (Ball & Drake, 1997). Thus, a major alteration in the relative dominance of the two species would have implications for many aspects of ecosystem function including productivity, nutrient cycling, water use and herbage quality. As disturbance is an important factor in grassland ecosystems (Belsky, 1992), it is possible that a major disturbance such as fire would affect population growth rates. Future work should consider the importance of disturbance regimes in mediating the community responses to global changes.

Most field experiments investigating the impact of global changes on vegetation concentrate on the impact of a single factor, such as elevated CO2 or warming, but more recently results from experiments exposing communities to multiple global changes have shown that there are substantial interactions among factors (Shaw et al., 2002; Zavaleta et al., 2003a). The results shown here emphasize the importance of including multiple factors, as the impacts are not always predicted from separate simulations. Of particular note in this regard is the case of T. triandra, in which the reduction in seedling establishment caused by each of warming and elevated CO2 in isolation restrained population growth. Thus, it would be reasonable to predict that the combined treatment would also result in a reduction of establishment. However, the combination of elevated CO2 and warming did not reduce establishment, and this, in combination with the large increase in the germination transition, resulted in a slight increase in population growth in the warmed elevated CO2 treatment. Thus, the impact of the two factors combined on population growth could not have been predicted by adding the responses to the factors applied in isolation.

Differences among plant species in their response to global changes are particularly relevant when considering species invasions. It is generally believed that characteristics that contribute to the invasiveness of a plant, namely broad environmental tolerance, high relative growth rate and high fecundity, are the very traits that would be favoured in a warmer, high-CO2 world (Patterson, 1995). Previous research has demonstrated substantial impacts of elevated CO2 on selected invasive species (Saebo & Mortensen, 1998; Mooney & Hobbs, 2000a,b; Rejmanek, 2000; Smith et al., 2000; Ziska, 2003; Edwards et al., 2005), mostly indicating that elevated CO2 does increase weed invasion success, particularly when the invasive species were C3 plants (Ziska & Goins, 2006). Fewer studies, however, have considered the interaction between elevated CO2 and warming. Our results clearly demonstrate that warming caused a substantial and significant reduction of population growth of the invasive species at both current and elevated CO2 concentrations. This indicates that global warming may be a more important determinant of the success of invasive species than CO2 concentration. Indeed, our results indicate that both species are likely to be excluded from the grassland community by increasing temperatures.

Ecosystems are already responding to global changes (Abu-Asab et al., 2001; Walther et al., 2002) and one of the most important responses is the alteration of species composition and abundance, which has ramifications for most ecosystem processes and services (Chapin et al., 2000; Hooper et al., 2005; Loiseau et al., 2005). Because changes in community composition also produce changes in processes such as nutrient cycling, it is possible that the impact of global changes on ecosystem function will be mediated by shifts in species composition, as has already been demonstrated (Zavaleta et al., 2003b). Community composition is determined by the demographic processes controlling the population size of the component species. The various responses we have observed in four important grassland plant species are directly relevant for community composition, weed invasion and ecosystem function. We believe that the matrix model approach offers a useful tool for assessing global change impacts on species with relevance for future community composition and the results obtained suggest testable hypotheses concerning the responses of different plant functional groups and the importance of various life cycle stages in determining the ultimate ecosystem-level response to global changes.

Acknowledgements

We thank the Australian Federal Department of Defence for access to the Pontville Small Arms Range Complex. We thank Ms Michaela Nolan for help with field work and Dr Greg Jordan for discussions. This research was supported by the Australian Research Council Discovery Projects scheme.

Ancillary