Effects of Free-Air CO2 Enrichment (FACE) on experimental grassland communities



1. Experimental grassland communities (turves) were exposed to elevated (60 Pa) and ambient (35 Pa) CO2 partial pressures (pCO2) in a Free-Air Carbon Dioxide Enrichment (FACE) experiment between 30 March 1995 and 4 July 1996. The vegetation was cut once during the experiment prior to the final harvest (harvest 2).

2. No significant treatment effects on total plant biomass at the whole turf level were detected, although biomass was typically about 25% higher under fumigation in year 1 and about 15% higher in year 2.

3. Biomass for two of the six sown species was significantly higher at harvest 2 than at harvest 1. There were no significant differences between individual species’ biomass under the two CO2 treatments at either harvest 1 or 2 or in terms of overall cumulative biomass. However, in four of the five sown species in both years biomass tended to be higher in the fumigated than in the control rings (Cerastium holosteiodes, Phleum pratense, Plantago lanceolata and Poa trivialis). In contrast, Lolium perenne showed increased biomass under the control treatment relative to the fumigated treatment in both years. Owing to the high variance both within and between rings for each of the two treatments the statistical power of most, but not all, of the analyses carried out was poor.

4. The relative proportions of each species in the turves under fumigated and control treatments was broadly similar after the first summer, with differences in the second year being mainly owing to the negative response of L. perenne to CO2 fumigation.


One of the major challenges in ecology at present is how to predict the likely influences of atmospheric change on natural ecosystems. One factor that is likely to affect ecosystems is the increasing partial pressure of CO2 in the atmosphere owing to human activities, which is expected to continue to rise, potentially doubling present levels by the end of the next century (Houghton et al. 1995). Attempts to predict future vegetation responses to CO2 increase (and to other gaseous pollutants) have resulted in a diverse range of experimental facilities designed to test everything from individual plant to whole ecosystem level responses (Ashenden & McLeod 1993). Many of the early studies (Strain & Cure 1985), particularly those on agricultural species, concentrated on highly productive single crop systems with unlimited nutrient and water levels, and found that such systems were highly responsive to elevated CO2 concentrations. Studies on non-agricultural species, under naturally occurring nutrient and water levels, have revealed a much wider range of responses both at the individual plant level, intermediate levels and at the whole ecosystem level (see for example Curtis & Wang 1998).

Grime (1996) argues that, within the UK, recently established fast-growing clonal herbs and ephemeral species, both invasive and native, are currently increasing at the expense of slower growing species, in response to environmental change, and may be the species with the greatest capacity for biomass increase under high atmospheric CO2 levels. A study by Bazzaz et al. (1995) found that for both Abutilon theophrasti and Betula allaghaniensis, elevated CO2 intensified intraspecific competition, with a resultant shift towards genotypes able to compete for resources other than CO2. If competition between plant species in response to CO2 increase is similarly dependent on limitations imposed by the availability of resources, rather than direct responses of individual plants to CO2 (Reynolds 1996), the result may be a similar shift in multispecific communities towards more competitively dominant species under high CO2.

A number of studies on grassland communities have already been carried out, either on artificially created communities or on intact field turves. These studies have yielded a variety of results including (1) nil to moderate effects of elevated CO2 on total biomass productivity both above and below ground (Campbell & Hart 1995; Saebo & Mortensen 1995; Leadley & Korner 1996; Luscher et al. 1996; Reynolds 1996; Roy et al. 1996; Stewart & Potvin 1996; Wolfenden & Diggle 1996; Potvin & Vasseur 1997; Hebeisen et al. 1997; Luscher, Hendrey & Nosberger 1998), (2) a hierarchy of responses of plants within communities to elevated CO2 dependent on plant functional type, with legumes performing better than other dicotyledonous plants, which in turn perform better than grasses (Luscher et al. 1996; Stewart & Potvin 1996; Hebeisen et al. 1997; Potvin & Vasseur 1997; Luscher et al. 1998; although see Leadley & Korner 1996), (3) differential responses to CO2 over time (Saebo & Mortensen 1995; Hebeisen et al. 1997; Luscher et al. 1998) and (4) the, often significant, impact of other variables such as nutrient and water levels on the potential effect of CO2 (Reynolds 1996; Roy et al. 1996; Wolfenden & Diggle 1996; Hebeisen et al. 1997).

To a certain extent the results from CO2 studies depend on the facility used. A meta-analysis of the effects of elevated CO2 on woody plants showed differential responses to CO2 between plants grown in growth chambers (growth enhancement + 19%) and those grown in greenhouses or open top chambers (+ 54%) (Curtis & Wang 1998). Unfortunately, many of the results presented are also from pseudoreplicated experiments (Jasienski, Thomas & Bazzaz 1998). The FACE system used here, like open top chambers, allows for in situ, field-based studies whilst avoiding complications that may arise from chamber effects on temperature, irradiation and wind. In the experiment reported here six naturally co-occurring grassland species were grown together in an artificial sown turf to investigate the effects of CO2 in a FACE system on an establishing grassland community. All six species used in the experiment had a low s-radius, can be considered as competitive or ruderal, and might therefore be expected to respond positively to elevated CO2 (Thompson 1994). The aim of the experiment was, in particular, to test whether productivity was higher in an establishing sward under elevated CO2 and whether this led to a shift in the competitive balance of the species in favour of the forbs.

Materials and methods

The Free-Air Carbon Dioxide Enrichment (FACE) facility used here is located at Eschikon (8°41’Ε, 47°27’N), 20 km north-east of Zurich (Switzerland) at an altitude of 550 m above sea level. This facility has been used in a number of other investigations of the responses of grassland ecosystems to enhanced CO2 and nitrogen (Luscher et al. 1996, 1998; Zanetti et al. 1996; Hebeisen et al. 1997). Climatic data for the site over the experimental period were obtained from a meteorological station near (200 m) the experimental site and are as given in Table 1.

Table 1.  . Meteorological data (mean monthly temperature and total monthly precipitation) for the FACE experiment in Eschikon for the experimental period from March 1995 to July 1996 Thumbnail image of

FACE technology (Lewin et al. 1992) was used for CO2 fumigation. Three FACE rings, each with a diameter of 18 m, were installed in an open field in April and May 1993. The rings were fumigated with CO2-enriched air while three control areas were at ambient CO2. The distances between fumigated and control rings were at least 100 m to prevent fumigation effects on the control rings. Fumigation was begun in spring when leaf temperature reached 8 °C (air temperature at 5 °C at 2 m on clear days in March) and was stopped at the same temperature threshold in November. CO2 fumigation was also stopped when the temperature was below this threshold in March and November and at night at all times. This procedure was selected because plant growth is rather poor below 8 °C, and it is known that effects of elevated pCO2 are reduced at low temperatures (Long 1994). Records of the CO2 partial pressure across the fumigated rings have shown the 1 min averages to be within the target concentration of 60 ± 6 Pa for 90–94% of the fumigation time.

The species used in the experiment were chosen, in part for their probability of occurrence together under field conditions [i.e. in an MG7 grassland (Rodwell 1992)] as well as for the fact that they are all primarily competitive species (sensuGrime 1977; Grime et al. 1988). The species included four grass species Lolium perenne L. (cv. ‘Taya’), Poa trivialis L., Dactylis glomerata L. and Phleum pratense L. and two dicotyledonous species with contrasting growth forms, Cerastium holosteiodes Fries and Plantago lanceolata L. The experiment was carried out in white tubs (210 mm × 145 mm × 150 mm high), filled with a standard soil/sand mix with holes to allow adequate drainage. To ensure an average density of one seed per cm2, 50 seeds per species were scattered over the soil surface of the tubs (an area of c. 300 cm2) and covered with a thin layer of soil (c. 1 cm depth). After sowing, the six tubs were placed adjacent to one another in a hole dug inside each of the six FACE rings and packed round with soil on all sides to equalize the level of the soil inside the tubs with the natural soil surface in the rings, on 30 March 1995.

In summary, the experimental design consisted of six species grown together in each of six tubs (providing a pooled mean) per ring, in three rings per treatment under two treatments (+ CO2 and ambient). Harvests were taken from the turves on two separate occasions. At harvest 1 (1 October 1995), plants were cut to a height of c. 4 cm and sorted into the constituent species before drying to constant weight in an oven at 70 °C for 72 h. At harvest 2 (4 July 1996) all plants were cut at soil level, sorted and weighed, as above. The recorded data were dried shoot biomass for each species.


Treatment effects on total biomass at each harvest and on the cumulative total biomass (harvest 1 + 2) in the turves were analysed using both a repeated measures analysis of variance (to take into account the effects of enhanced CO2 on sequential harvests of the turves) and simple analysis of variance. The mean values of all six replicates in each of the rings at both harvests 1 and 2 were used in the analysis. Student's t-tests were used to test for treatment effects on the mean biomass of each of the five species at each harvest and also on the cumulative total biomass of each species (harvest 1 + 2). Statistical power analyses (Zar 1996) were carried out on the results of both the simple analyses of variance and the t-tests to identify how sensitive the experiment was to potential differences between the treatments. Spearman's Rank correlation was used to establish whether ranks of individual species varied between treatments at both harvests 1 and 2.


No D. glomerata plants were found in any of the turves under either elevated CO2 or control conditions probably owing to poor quality seed and subsequent lack of germination, as this was also the case in two concurrent experiments using the same seed stock (Norton et al. 1998).

Both the repeated measures analysis of variance and the simple analysis of variance showed no significant treatment effects on total plant biomass at either harvest 1 or 2 (F1,4 = 1·3, P = 0·32) (Fig. 1) although biomass was typically 25% higher under elevated CO2 at harvest 1 and 15% higher at harvest 2. There was similarly no significant treatment effect on cumulative biomass. A significant difference between the total biomass harvested at each of the two harvests (F1,4 = 50·8, P < 0·01) shown by the repeated measures analysis of variance was owing in part to the higher biomass of both P. pratense and P. lanceolata at harvest 2 (Fig. 2). However, the two harvests would be expected to be different for a number of reasons, including different lengths of growing period with the initial harvest taking place after establishment from seed and the second harvest from established plants, and different harvesting methods (i.e. to within 4 cm of soil surface at harvest 1 and to soil surface at final harvest). Hence the use of simple analysis of variance alongside the repeated measures analysis.

Figure 1.

. Mean total biomass + SE for the three rings under each of the two treatments (control, ambient CO2 and fumigated, + CO2) at harvest 1 and harvest 2 together with the cumulative total biomass harvested.

Figure 2.

. Mean biomass + SE for each of the five individual species (+ other species which invaded the turves) for the three rings under each of the two treatments (control, ambient CO2 and fumigated, + CO2) at (a) harvest 1 and (b) harvest 2 together with (c) the cumulative total biomass harvested.

At both harvests the turves were dominated by P.lanceolata which comprised c. 60% of the total biomass at harvest 1 and 50% at harvest 2 (Fig. 3). Student's t-tests on the mean biomass of each species produced per turf also showed no significant treatment effects (Figs 1 and 2) despite the fact that in several cases there was no overlap between the biomass produced under the two treatments. For P. pratense at harvest 1, P. trivialis (Fig. 4) and C. holosteiodes (present in tiny amounts under both treatments) at harvest 2 and for the cumulative biomass of P. trivialis, mean biomass was higher in all three of the fumigated rings than in any of the control rings. However, the reverse was true for the biomass of L. perenne both at harvest 1 (Fig. 4) and for overall cumulative biomass.

Figure 3.

. Mean percentage biomass of each of the five component species in the turves at (a) harvest 1, (b) harvest 2 and for (c) the cumulative total biomass harvested from each of the three rings under each of the two treatments.

Figure 4.

. Mean biomass of (a) L. perenne and (b) P. pratense at harvest 1 and (c) P. trivialis at harvest 2, in each of the three rings under each of the two treatments.

As a result of the high variance within and between rings within each of the two treatments, and the low level of true replication (three rings per treatment), statistical power analyses on the results of each simple analysis of variance on total biomass harvested (harvest 1 and 2 and cumulative total) indicated that in each case power was < 0·3, i.e. there was a greater than 70% chance of having committed a type II error in the analyses. Similarly, for all but one of the t-tests carried out on the mean biomass of each species produced per turf, statistical power was < 0·5, and more usually < 0·3. In the case of the final biomass of L. perenne produced, despite the visual differences between the two treatments apparent in Fig. 2, the statistical power analysis estimated the power of the t-test which showed no significant differences between treatments to be greater than 90%.

Spearman's rank correlation showed a significant correlation between the ranks of the five species (+ other species which invaded the turves) comprising the turves at harvest 1 (r = 0·94, n = 6, P < 0·05). However, at harvest 2 (r = 0·83, n = 6, NS) and for cumulative biomass (r = 0·83, n = 6, NS) the ranks of the species in the turves were not correlated between treatments, mainly owing to the lowered productivity of L. perenne under fumigation (rank 5) compared to the control (rank 3).


Despite increases of between 15 and 25% under fumigation, in terms of both total biomass and the biomass of four out of five species, it was impossible to detect a significant effect of CO2. This may reflect an actual lack of response to elevated CO2, or may result from the lack of statistical power owing to low numbers of replicates. The inability to detect significant treatment effects in the experiment reported here reflects a problem commonly encountered in the use of experimental facilities designed to investigate the impacts of global environmental change. As replication at the treatment level is generally low (owing to cost limitations on the provision of CO2 for fumigation), high variability within a data set, particularly if the variability is uneven between treatments, leads to non-significant results (see also Roy et al. 1996; Zanetti et al. 1996). In the case of the FACE experiment, although costs limit the number of CO2 fumigated rings, an increase in the number of control rings (presumably considerably cheaper to maintain) would at least allow for a more accurate measure of the performances of plants/communities under control conditions against which to compare the responses to fumigation. Alternatively the number of subsamples within a ring should be high in order to reduce the variance, which would mean fewer scientists carrying out more highly replicated experiments, rather than a very broad range of small-scale experiments. Population studies, in particular, are space demanding, and the results from this experiment indicate that the levels of variability possible in mixed species studies lead to a requirement for very high levels of replication in order to achieve significant results. Certainly in the experiment reported here, it appears that intrinsic variability was more important than the effects of CO2, despite differences of between 15 and 25% between fumigated and control rings. Similar results have been gained from a concurrent experiment using identical communities carried out in a UV-B facility (Norton et al. 1998), where high variability within treatments outweighed UV-B effects. Interestingly, statistical tests on the biomass of L. perenne in the UV-B experiment gave the most statistically powerful results, as was the case in this experiment. This indicates that of the species used, L. perenne is the least variable and the most amenable to experimentation where replication at the treatment level is low. Statistical power analysis is itself subject to discussion (see Gerard et al. 1998) but even without it, it is clear that three true replicates severely limits the ability to detect significant differences as well as limiting the statistical tests which can legitimately be employed. Far too many studies to date present the results of pseudoreplicated experiments (Jasienski et al. 1998).

Whilst not significant, the effect on plant biomass under elevated CO2 found in this study is in the same direction as the results from other studies on grassland communities (Jackson et al. 1994; Leadley & Korner 1996; Luscher et al. 1996; Roy et al. 1996; Reynolds 1996; Stewart & Potvin 1996; Hebeisen et al. 1997; Schenk et al. 1997; Luscher et al. 1998) and differs from nil or negative responses of grassland communities to CO2 found by Korner & Miglietta (1994), Clark et al. (1995), Wolfenden & Diggle (1996) and Schappi (1996a,b). In this study the community investigated was an establishing one (from seed), contrasting with all of the above studies which have been carried out on intact field communities, excavated turves or communities established from greenhouse grown plants. It may be expected that an establishing community of competitive (sensuGrime 1977; Grime et al. 1988) species would be more likely to show a positive response to elevated CO2 than a stable late successional community (as in Wolfenden & Diggle 1995; Grime 1996; Schappi 1996a,b).


It is now well established that responses to elevated CO2 are species-specific. In this study, four of the five species which grew successfully in the turves (C.holosteiodes, P. pratense, P. lanceolata and P. trivialis) showed non-significant increases in biomass under CO2 fumigation, with mean total biomass at each harvest and as a cumulative total always at least 15% higher under fumigation. Conversely, the biomass of L. perenne was consistently lower in the fumigated rings at c. 30% of its productivity under control conditions.

Hunt et al. (1991) found that CO2 responsiveness of singly grown plants was common only for species with competitive, or closely related strategies, when grown under enhanced CO2 levels in greenhouses for 56 days as part of the Integrated Screening Programme (ISP, Grime 1985). It might be expected therefore that all of the species grown here would show a positive response to elevated CO2. The tendency for L. perenne (a competitive species) to perform poorly under CO2 fumigation in this study, although non-significant, could be a direct negative response to CO2, but is more likely to be the result of competition for resources other than CO2. It is possible that nutrients were limiting as the soil used was a 50/50 soil sand mix (John Innes No.1 equivalent) and L. perenne has a high nitrogen demand. However, as the two dominant species P. pratense and P. lanceolata were also physically large species, competition for light may also have affected the performance of L.perenne. Other FACE studies (Hebeisen et al. 1997; Luscher & Nosberger 1997) involving L. perenne have also revealed weak responses to doubled CO2.

A number of studies, including the ISP, have investigated the performances of some of the species used in this study under similarly doubled CO2 environments giving a range of results, e.g. L. perenne[above and Clark et al. (1995) no response, Roy et al. (1996) negative response, Luscher et al. (1998) slight to moderate positive responses, Saebo & Mortensen (1995) positive and negative biomass responses over time], P. lanceolata[Wolfenden & Diggle (1995) and Hunt et al. (1991) no biomass response] and P.pratense[Potvin & Vasseur (1997) no above-ground biomass response, Saebo & Mortensen (1995) positive and negative biomass responses over time, Mortensen & Saebo (1996) nil and positive biomass responses over time]. However, direct comparisons are made unfeasible by a number of factors varying between the experiments; these include soil type, pot size, nutrient levels, presence or absence of other species, type of facility used and levels of replication.


As well as investigating the effects of being grown in a community on the responses of particular species to elevated CO2, this study investigated the potential alterations in community structure under elevated CO2 as a result of those responses. The simple grassland community investigated here, established from seed under the different treatments, and grew through two springs with one cut, yet the basic composition of the resulting turves was very similar between the two treatments, with P. lanceolata and P. pratense the dominant species under both treatments (Fig. 2). The poor performance of L. perenne under CO2 fumigation caused an alteration in the ranked biomass of the different species in the turves. It is unclear to what extent this resulted from shifts in the competitive relationships between the species. The relatively small shift in community structure here contrasts with that of Potvin & Vasseur (1997) where community composition was shown to be markedly different under elevated CO2 after a year-and-a-half of exposure. In their experiment, the proportion of P. major in the communities studied decreased under ambient conditions (whilst the proportion of Agropyron repens increased) and remained stable under elevated CO2. In typical MG7 grasslands, P. lanceolata becomes dominant in older stands of L. perenne/T. repens sown grasslands. Here, establishment at the start of the experiment, allowed it to dominate from the start.

Although the lack of significant effects of CO2 reported here may reflect problems with the experimental set-up and with low levels of replication, it may also give a true picture of plant community responses to CO2. The need for large-scale, realistic, in situ, long-term, well-replicated experiments to be carried out under resource constraints that apply in natural ecosystems is reinforced by the findings of small-scale studies like this one.


We thank K. Ruegg for invaluable technical assistance and facilitation of the experiment. Financial support was provided by the NERC through its TIGER (Terrestrial Initiatives in Global Environmental Research) programme, award number T03087b6(LGF).