How do climate warming and species richness affect CO2 fluxes in experimental grasslands?


  • Hans J. De Boeck,

    1. Research Group of Plant and Vegetation Ecology, Department of Biology, Universiteit Antwerpen (Campus Drie Eiken), Universiteitsplein 1, B-2610 Wilrijk, Belgium;
    Search for more papers by this author
  • Catherine M. H. M. Lemmens,

    1. Research Group of Plant and Vegetation Ecology, Department of Biology, Universiteit Antwerpen (Campus Drie Eiken), Universiteitsplein 1, B-2610 Wilrijk, Belgium;
    Search for more papers by this author
  • Sara Vicca,

    1. Research Group of Plant and Vegetation Ecology, Department of Biology, Universiteit Antwerpen (Campus Drie Eiken), Universiteitsplein 1, B-2610 Wilrijk, Belgium;
    Search for more papers by this author
  • Joke Van den Berge,

    1. Research Group of Plant and Vegetation Ecology, Department of Biology, Universiteit Antwerpen (Campus Drie Eiken), Universiteitsplein 1, B-2610 Wilrijk, Belgium;
    Search for more papers by this author
  • Stefan Van Dongen,

    1. Research Group of Evolutionary Biology, Department of Biology, Universiteit Antwerpen (Campus Groenenborger), Groenenborgerlaan 171, B-2020 Antwerpen, Belgium
    Search for more papers by this author
  • Ivan A. Janssens,

    1. Research Group of Plant and Vegetation Ecology, Department of Biology, Universiteit Antwerpen (Campus Drie Eiken), Universiteitsplein 1, B-2610 Wilrijk, Belgium;
    Search for more papers by this author
  • Reinhart Ceulemans,

    1. Research Group of Plant and Vegetation Ecology, Department of Biology, Universiteit Antwerpen (Campus Drie Eiken), Universiteitsplein 1, B-2610 Wilrijk, Belgium;
    Search for more papers by this author
  • Ivan Nijs

    1. Research Group of Plant and Vegetation Ecology, Department of Biology, Universiteit Antwerpen (Campus Drie Eiken), Universiteitsplein 1, B-2610 Wilrijk, Belgium;
    Search for more papers by this author

Author for correspondence: Hans J De Boeck Tel: +32 3820 22 82Fax: +32 3820 22 71 Email:


  • • This paper presents the results of 2 yr of CO2 flux measurements on grassland communities of varying species richness, exposed to either the current or a warmer climate.
  • • We grew experimental plant communities containing one, three or nine grassland species in 12 sunlit, climate-controlled chambers. Half of these chambers were exposed to ambient air temperatures, while the other half were warmed by 3°C. Equal amounts of water were added to heated and unheated communities, implying drier soils if warming increased evapotranspiration. Three main CO2 fluxes (gross photosynthesis, above-ground and below-ground respiration) were measured multiple times per year and reconstructed hourly or half-hourly by relating them to their most important environmental driver.
  • • While CO2 outputs through respiration were largely unchanged under warming, CO2 inputs through photosynthesis were lowered, especially in summer, when heat and drought stress were higher. Above-ground CO2 fluxes were significantly increased in multispecies communities, as more complementary resource use stimulated productivity. Finally, effects of warming appeared to be smallest in monocultures.
  • • This study shows that in a future warmer climate the CO2 sink capacity of temperate grasslands could decline, and that such adverse effects are not likely to be mitigated by efforts to maintain or increase species richness.


There is still considerable uncertainty over how the carbon balance of ecosystems (the total of C inputs and outputs) will be affected by climate change. Information on this matter is crucial, as changing C balances can reinforce or buffer the greenhouse effect (Oechel et al., 1993). How much CO2 (the major constituent of the C balance) is taken up from, or released to, the atmosphere by plant communities depends on the magnitude of photosynthetic inputs and respiration outputs, respectively. Photosynthesis and respiration both depend greatly on climatological conditions (Davidson et al., 1998; Baldocchi & Amthor, 2001), but do not necessarily respond identically to changes in those conditions. When considering temperature, it is known that photosynthesis tends to reach an optimum at lower temperatures than respiration (Barnes et al., 1998). Under a future warmer climate (Houghton et al., 2001), this could lead to a stronger increase in CO2 outputs than in inputs, resulting in a positive feedback towards climate warming. Some experimental studies have reported such responses to increased temperatures (Mertens et al., 2001; Welker et al., 2004), although a multitude of confounding indirect effects limit the generality of such observations. For example, a warmer climate could stimulate plant growth (Zavaleta et al., 2003), also indirectly promoting below-ground heterotrophic activity (Zhang et al., 2005). Higher temperatures will also lead to drier conditions unless precipitation increases (Saleska et al., 1999). Drought would also affect both plant growth and below-ground activity, although the direction of the response can differ (Barr et al., 2002; Saleska et al., 2002; Marchand et al., 2004). Another indirect effect of climate warming on the carbon balance of ecosystems is shifts in phenology. If the growing season is prolonged under warming, as several studies have found (Myneni et al., 1997; Walther, 2003), seasonal changes in both above- and below-ground activity would influence total CO2 inputs and outputs (Marchand et al., 2004). A further complication arises from the possibility of acclimation. Both plants and heterotrophic communities are reported to be able to acclimate to higher temperatures (Luo et al., 2001; Wythers et al., 2005), although in the case of soil respiration, a dampening of the warming response may also be attributed to the depletion of readily decomposable substrates (Kirschbaum, 2004).

Like rising temperatures, the loss of species from communities and ecosystems is a globally observed phenomenon. The reported declines are almost exclusively caused by anthropogenic disturbances (Sala et al., 2000), such as the loss and fragmentation of suitable habitats, eutrophication, and also climate change (Thomas et al., 2004). The number of species in a community has been proved to affect many aspects of ecosystem functioning (Naeem & Li, 1997; Kennedy et al., 2002; Levine et al., 2002). In both experimental and natural systems, higher species richness (S) often coincides with higher productivity (Hooper et al., 2005; Gillman & Wright, 2006), which is attributed to a more complete use of available resources when more species are present (complementarity); increased positive species interactions (facilitation, often regarded as an aspect of complementarity); and the higher probability of a productive species being present in the community (the ‘selection effect’). Productivity is directly linked to photosynthesis and autotrophic respiration, and indirectly to heterotrophic below-ground respiration through litter production and root exudates (Larcher, 2003). The possibility of species richness affecting the C balance is further reinforced by linkages between above- and below-ground species richness reported in some studies (e.g. Stephan et al., 2000), suggesting that microbial communities may be affected by above-ground loss of species richness, which could affect heterotrophic respiration.

There are reasons to assume that the effects of higher temperatures and changing species richness are not additive. For example, in cases where warming exerts a largely negative influence on productivity, and thereby on CO2 fluxes, the increased probability of multispecies communities containing species better adapted to heat and drought may ‘insure’ these communities against adverse effects better than species-poor systems (Naeem & Li, 1997). Negative effects of warming could thus be mitigated by maintaining or increasing the number of species in communities. However, although their impact could be substantial, interactive effects of temperature and S on the C balance of plant communities have, to our knowledge, never been studied.

To investigate effects of climate warming, species richness, and their interaction on the CO2 fluxes of plant communities, we built an experimental platform containing grassland communities of different S levels, which were subjected to ambient temperatures or temperatures that were continuously 3°C higher. Precipitation was identical in the unheated and heated communities to ensure warming would also encompass lower soil water contents if evapotranspiration was higher. In previous papers, we reported on autumn physiology (Gielen et al., 2005); the ecophysiology of the individual grassland species (Lemmens et al., 2006); water use (De Boeck et al., 2006); and productivity (De Boeck et al., 2007). In the current study, we consider single-factor and interactive effects of climate warming and species richness on CO2 fluxes. More specifically, we investigate the following questions: (1) how do the different CO2 fluxes respond to changes in temperature and S level? (2) do warming and S level affect the CO2 sink or source capacity of grasslands? (3) are there interactive effects on CO2 fluxes of the two global changes studied?

Materials and Methods

Study site and experimental set-up

This study was conducted at the Drie Eiken Campus of the University of Antwerp (Belgium, 51°09′ N, 04°24′ E), where an experimental platform containing 288 artificially assembled grassland model ecosystems in containers was established in 2003. The climate of northern Belgium is characterized by mild winters and cool summers, with an average annual air temperature of 9.6°C, and mean monthly air temperatures between 2.2°C (January) and 17.0°C (July). Annual precipitation averages 776 mm, distributed equally throughout the year.

At the start of the experiment (2003), the platform consisted of 12 sunlit, climate-controlled chambers (2.25 m2 ground area) facing south, half at ambient temperatures (unheated) and the other half continuously at ambient temperatures +3°C (heated). Each year, in November, two chambers (one heated and one unheated) were removed for destructive harvesting and root analysis. Each chamber had an individual air-control group with an electrical heating battery, and was linked to a central refrigeration unit by isolated pipes. The conditioned air was distributed evenly throughout the chambers by means of aerators with regulated flow. The aluminium-frame chambers were covered with a colourless polycarbonate plate (4 mm thick) and polyethylene film (200 µm thick) at the sides, both UV-transparent and with a total light transmission of 86%. Each chamber (the blocking variable in the statistical analyses) contained the same series of 24 different grassland communities of varying species richness: nine monocultures, nine S = 3 communities and six S = 9 communities. These communities are the experimental unit in this study, and were placed in PVC containers (24 cm inner diameter, 60 cm deep). We opted for species from three functional groups, which were represented equally at each S level: three grass species (Dactylis glomerata L., Festuca arundinacea Schreb., Lolium perenne L.); three nitrogen-fixing dicots (Trifolium repens L., Medicago sativa L., Lotus corniculatus L.); and three nonN-fixing dicots (Bellis perennis L., Rumex acetosa L., Plantago lanceolata L.). These C3 species were selected according to three criteria: presence in European temperate grasslands, perennial life cycle, and preference for clay or loam soil. In addition, they represented different productivities, and different temperature and drought resistances. Species representative of the three functional groups were used to create each of the S = 3 communities, with each species combined only once with any other species (Table 1). Out of the three possible sets of nine different S = 3 communities that met these criteria, only one randomly chosen set was used. Therefore not every species occurred together with every possible species pair, but the probability that the combination of specific pairs and single species would cause unexpected interactions (e.g. with warming) was low. Each of the six S = 9 communities had a different internal arrangement, to ensure that each species interacted to the same extent with any other species over the totality of the six S = 9 communities.

Table 1.  Species composition of the 24 communities of varying species richness (S) in each chamber
S = 1S = 3S = 9
  1. Species: Da, Dactylis glomerata; Be, Bellis perennis; Tr, Trifolium repens; Fe, Festuca arundinacea; Ru, Rumex acetosa; Me, Medicago sativa; Lp, Lolium perenne; Pl, Plantago lanceolata; Lo, Lotus corniculatus.

1DaDa, Be, Trall
2BeFe, Ru, Meall
3TrLp, Pl, Loall
4FeDa, Ru, Loall
5RuFe, Be, Loall
6MeLp, Ru, Trall
7LpDa, Pl, Me 
8PlFe, Pl, Tr 
9LoLp, Be, Me 

Each community contained 30 individuals planted in a hexagonal grid at 4 cm distance, with interspecific interactions maximized. Similar plant densities were used in other experiments (e.g. Van Peer et al., 2004), and are deemed realistic for temperate grasslands. Before planting in the containers in June 2003 (which took approx. 3 wk), plants were sown in small seedling pots in April 2003. The soil used in the experiment (76.3% silt, 14.8% clay and 8.7% sand; field capacity 0.39 m3 m−3; pH 6.45, carbon content 1.6%) was collected from an agricultural field and sieved (0.5 mm mesh size) to remove stones and large organic material. No fertilizer was added to this rich agricultural soil. Plants were treated regularly to avoid fungal infection and insect damage, and weeding was done manually throughout the experiment. For further information regarding the experimental set-up, we refer to Lemmens et al. (2006).

The containers with the plant communities were embedded in the soil to ensure realistic soil temperatures. Water could drain freely from the containers, while capillary rise of soil water from deeper layers and lateral water inflow were prevented. The heated chambers received the same amount of water as the unheated chambers, so that any enhanced consumption would result in (aggravated) soil drought. A more detailed description of the water supply is given by De Boeck et al. (2006).


Ecosystem CO2 fluxes were measured during 15 periods, each lasting approx. 1 wk, over the course of two full years (November 2003–October 2005). The flux measurements were performed between 9 and 18 h, using a transparent, 60-cm-high polymethyl pentene cuvette that could be fitted tightly on the plant containers, coupled to an EGM-4 infrared gas analyser (PP Systems, Hitchin, UK) or, before June 2004, a CIRAS infrared gas analyser (PP Systems). Two small fans inside the cuvette guaranteed well mixed air during measurements. After placing the cuvette on a container and the stabilization of the CO2 flux, net ecosystem exchange of CO2 (NEE) was recorded, followed by a measurement of total ecosystem respiration (Re) on the same community by darkening the cuvette using an opaque black cloth, preventing photosynthesis. Gross photosynthesis (Pgross) could then be calculated as Pgross = NEE − Re. Leaf respiration is partly inhibited in the light (Atkin et al., 2000), which could cause a slight overestimation of Pgross. However, as we are studying relative differences between temperature treatments and S levels, we are confident that the applied method was suitable. Below-ground respiration (Rb) was measured in small PVC chambers (8 × 5 × 1.5 cm), permanently installed on a strip of bare soil inside the communities. The aerating hole in these chambers, ensuring mixing with the outside air to avoid build-up of CO2 inside the chambers, was closed with terostat (Henkel KGaA, Düsseldorf, Germany) while measuring Rb. This enabled us to calculate above-ground respiration (Ra) as Ra = Re − Rb. Fluxes were measured in eight centrally located communities per chamber, inside six chambers (three unheated and three heated), so that all different communities of each S level were sampled in each temperature treatment. Measurements were repeated once during each weekly period.

Photosynthetic photon flux density (PPFD) was measured with a quantum sensor (JYP 1000, SDEC, Reignac sur Indre, France) both inside the cuvette during measurements, and in the chambers (stored every half hour). Air temperature (Tair) was measured continuously inside each chamber with a QFA66 sensor (Siemens, Berlin, Germany), and these data were stored every 30 min. Soil temperature (Tsoil) was measured at 5 cm depth inside four chambers with thermocouple probes (TME Ltd, Goring, UK) installed in eight communities representing a range of open to dense canopies, and stored each hour with a datalogger (Delta-T Devices Ltd, Cambridge, UK). During the measurements, Tsoil was recorded with a soil thermometer (Hanna Instruments, Woonsocket, RI, USA) at 5 cm depth. Profile probe tubes (55 cm long), which fitted with a PR1 soil moisture sensor (Delta-T), were installed in the 48 communities used to measure CO2 fluxes. Soil water content (SWC) was measured on average once a week (always before watering) (De Boeck et al., 2006), and was corrected using the calibration determined in the laboratory for the specific soil used. Above-ground biomass was determined by cutting all plant communities twice a year (at the end of May and the beginning of November) approx. 3.5 cm above the soil surface, drying the plant parts for at least 2 d at 70°C and weighing them. Finally, vegetation height was determined as the average of six height measurements (the height where the highest plant part touches a ruler) within a community.

Reconstruction of CO2 fluxes

As fluxes were not measured continuously, they had to be reconstructed by linking each flux with its main environmental driver. Photosynthesis is affected mainly by PPFD, and is described by the function:

Pgross = (QE × PPFD)/((1 – (PPFD/1500)) + ((QE × PPFD)/Pmax)) (Eqn 1)

with QE the quantum efficiency and Pmax the gross primary productivity at a PPFD of 1500 µmol m−2 s−1. The regression was done with the pooled measurements of each measurement period, but separately for S level and temperature treatment, yielding six regressions per period.

Respiration, both above- and below-ground, is driven primarily by temperature, along the exponential function:

Ra × Q10((T−10)/10)(Eqn 2)

with a respiration rate at 10°C, Q10 the relative change in R if the temperature increases by 10°C (making it a measure of the temperature sensitivity of respiration), and T the appropriate temperature (air or soil). Because the temperature range in one period was limited, we applied the function to the respiration data for 1 yr, but separately for each S level and temperature treatment. As the temperature sensitivity of respiration is not constant throughout the year (Janssens & Pilegaard, 2003), the function overestimated respiration in some parts of the year, while underestimating it in other parts. We therefore added day, day2 and day3 as covariates to equation 2 for each temperature and S level separately. By doing this, we ensured that seasonal changes in temperature sensitivity were included in the calculations of respiration.

After determining the relationship between the CO2 fluxes and their main drivers, we calculated the three fluxes each hour or half hour using the automatically recorded meteorological data. On the few occasions that such data were missing (through malfunctions), we used data from the nearby meteo station of Brasschaat. Confidence intervals were obtained in Bayesian analyses (see below). The yearly CO2 fluxes (Pgross, Ra, Rb) per S level and temperature treatment were subsequently calculated, and added together (with Pgross being negative), yielding the yearly CO2 balance.

Statistical analysis

The analyses of the CO2 fluxes consisted of two parts. First, the nonlinear model was fitted for each temperature treatment and diversity level separately, using the above models for photosynthesis and respiration, respectively. Second, based on model parameters, yearly CO2 fluxes were reconstructed. In order to be able to generate confidence intervals, we applied Monte Carlo Markov Chain (MCMC) simulations in a Bayesian framework. Bayesian statistics are becoming increasingly popular in many areas of research because of their flexibility and ability to deal with models of high complexity. In Bayesian analyses, inferences are made in probabilistic terms and are based on the so-called posterior distributions. Posterior distributions reflect the current stage of knowledge about the model parameters and contain all possible sources of uncertainty. In complex models, MCMC simulations are often used to determine the posteriors because analytical solutions may be impossible to obtain. The main advantage of using MCMC in a Bayesian framework for the present analyses is that the determination of the yearly CO2 fluxes and their uncertainty can be obtained, taking all sources of variation into account, without having to make often unrealistic distributional assumptions of the model parameters. In all the analyses, we used uniform priors to reflect the absence of prior knowledge. All posteriors were obtained by running five independent chains for 10 000 iterations, discarding the first 5000. All analyses were performed in the freeware openbugs. Visual inspection of the MCMCs did not reveal any problems with convergence. We refer to Gelman et al. (2003) for a general introduction to Bayesian statistics. To test whether relations between the primary environmental variables (PPFD and T) and CO2 fluxes were influenced by the soil water content or vegetation height, we applied residuals analysis. Linear regressions between residuals and SWC or vegetation height were subsequently tested for significance in spss 13.0 (SPSS Science, Woking, UK). Biomass data were square-root transformed for normality, and analysed with general linear model univariate analysis, with temperature treatment, species-richness level and year as fixed factors (no chamber effect was detected). We divided the yearly Pgross and Ra by the yearly biomass production, per temperature treatment and S level. Because we were combining results from two completely different statistical methods, it was difficult to determine statistical significance for these ratios. The best possible significance estimate was obtained by dividing the 95% confidence intervals calculated for the fluxes by the biomass, with nonoverlapping confidence intervals indicating significant statistical differences (a highly conservative method). The significance threshold was 0.05 throughout all analyses.



The accuracy of the applied models is reflected in the size of the 95% confidence intervals. These are relatively large (Figs 2,3) because of the considerable variation between the statistical units (the communities). When comparing modelled and measured fluxes, R2 values were 0.70 and 0.64 for Pgross; 0.43 and 0.35 for Ra; and 0.50 and 0.32 for Rb, in years 1 and 2, respectively, and for both temperature treatments and S levels combined. Unexplained variance (1 – R2) was relatively high because the differently composed communities served as statistical repeats inside each S level, a technique typically used in diversity experiments (e.g. Van Peer et al., 2004). In a separate analysis on above-ground CO2 flux data collected in July and August 2005, we accounted for the variation in height (a proxy for above-ground biomass) between the different communities by means of residuals analysis. This lowered the unexplained variance in the relationships between PPFD and Pgross, and between Tair and Ra, by over 50% (Fig. 1). Because height measurements were carried out only during certain periods, this residuals analysis could not be implemented. However, the analysis does show that most of the unexplained variance could be attributed to the unavoidable between-subject differences present in biodiversity experiments. These differences do not alter the calculated relationships (as in Fig. 1a), but because the model-generated confidence intervals do reflect the subject differences, these intervals are statistically very conservative (i.e. expanded).

Figure 2.

Total reconstructed CO2 fluxes from November 2003–October 2004 (left panels) and November 2004–October 2005 (right panels) in unheated (open symbols, dashed line) and heated communities (closed symbols, solid line) at three levels of species richness (S). The three fluxes, gross photosynthesis (circles), soil respiration (squares) and above-ground respiration (triangles), are depicted with 95% CI. Averages are connected and symbols are slightly shifted with respect to the x-axis for clarity.

Figure 3.

Yearly gross photosynthesis (circles) and above-ground respiration (triangles) divided by yearly above-ground biomass production, depicted with 95% CI and spanning November 2003–October 2004 (left panels) and November 2004–October 2005 (right panels) in unheated (open symbols, dashed line) and heated communities (closed symbols, solid line) at three levels of species richness (S). Averages are connected and symbols are slightly shifted with respect to the x-axis for clarity.

Figure 1.

Example of a residuals analysis on gross photosynthesis (Pgross) data, to account for the variation in vegetation height between different communities at one species richness (S) level. (a) The original relationship (dashed line) between photosynthetic photon flux density (PPFD) and Pgross; (b) the linear relationship (dashed line) between observed and predicted Pgross values, and the associated 1 : 1 line (solid line). (c) Linear relationship between vegetation height and the residuals of Pgross. (d) Improved linear relationship (dashed line) between observed and newly predicted Pgross values, and the associated 1 : 1 line (solid line).


Annual gross photosynthesis decreased in response to warming by 19% in the first year and by 5% in the second year (averaged across all S levels), although only the former difference was statistically significant. Photosynthesis was substantially and significantly higher in multispecies communities than in monocultures in both years, with increases of 41% in year 1 and 79% in year 2 (Fig. 2). No significant differences were observed between S = 3 and S = 9 communities. The impact of warming was the least pronounced at S = 1, and significant differences between the temperature treatments were observed only in the multispecies systems (year 1). A residuals analysis showed that SWC could not account for unexplained variance of the PPFD – Pgross relations. In heated communities, gross photosynthesis was 16 and 39% higher per unit biomass in years 1 and 2, respectively (Fig. 3). Despite generally overlapping confidence intervals, there was at least a trend towards increased photosynthesis per g biomass, especially in year 2, also considering the nonsignificant Pgross decrease and a highly significant biomass decrease (P < 0.005; Table 2). The 45% higher photosynthesis per g biomass in monocultures compared with multispecies communities, a result of decreased Pgross but further decreased biomass (P < 0.005), was reflected by nonoverlapping confidence intervals in year 2, while there was only a trend in year 1.

Table 2.  Above-ground biomass per temperature treatment (unheated and heated); species-richness level (S); and measurement year (from the beginning of November until the end of October)
SAbove-ground biomass (g m−2)
  1. Averages for each temperature treatment are also shown.



The small decreases of above-ground respiration in response to warming (6 and 8% in the first and second years, respectively) were not significant. However, in both measurement years, Ra increased significantly with rising S. Similarly to Pgross, the difference in Ra was largest between monocultures and the multispecies communities (88 and 156% in years 1 and 2, respectively), although in year 1 we also recorded a 28% higher Ra in S = 9 compared with S = 3 communities (Fig. 2). No significant difference in response to the temperature treatment in the different S levels was observed. There was a trend towards a higher Ra per g biomass in heated vs unheated communities (32% in year 1; 38% in year 2), although confidence intervals were overlapping (Fig. 3). The Ra per g biomass was similar across S levels. Soil respiration was not affected by either temperature treatment or S, nor could any trends be detected (Fig. 2). Following residuals analysis, we conclude that unexplained variance of the T – R relations could not be attributed to SWC.

CO2 balance

Although confidence intervals were large (see ‘Models’), especially relative to the size of net fluxes (which are typically close to zero), some trends could nevertheless be detected. The experimental grasslands in this study acted as a net sink for CO2, although only significantly in year 1. Furthermore, there was a trend towards a decreased CO2 sink capacity under warming in year 1 because of lower photosynthesis and largely unchanged respiration. The net effect of S was unclear, but the smallest differences between the temperature treatments seemed to exist at the monoculture level (Fig. 4), with a near-significance S × temperature treatment interaction (year 1).

Figure 4.

Reconstructed total net CO2 fluxes (net uptake is negative) (a) November 2003–October 2004; (b) November 2004–October 2005 in unheated (open bars) and heated communities (closed bars) at three levels of species richness (S). Fluxes are depicted with 95% CI (cut off at zero).

Yearly course

We consider only temperature treatment here, as all S levels showed a comparable course over the year, and climatological factors are more important in shaping such year-round courses. As an example, we show the course of CO2 fluxes in 2005 in S = 3 communities (Fig. 5). In winter, the communities acted as a CO2 source, with the tipping point towards a sink around the spring equinox. Although Pgross was increased in heated compared with unheated communities during winter, Ra was also higher, leading to slightly higher CO2 losses. No clear indications were observed for an earlier onset of spring growth in warmed communities. In late spring and summer, less CO2 was taken up on a net basis under warming, because of lower photosynthesis (especially in mid-summer) while Ra was reduced less. Mowing the above-ground biomass (c. day 200) had a clear effect on above-ground fluxes, while Rb remained unaffected. Around the autumn equinox, all fluxes started decreasing in size, although the onset of autumn reduced Rb less than the above-ground CO2 fluxes. No signs of an extended growing season in response to warming were recorded.

Figure 5.

Reconstructed CO2 fluxes for three-species communities from November 2004–October 2005 in unheated (dashed line) and heated (solid line) communities. Left panels show gross photosynthesis (bottom curve), soil respiration (middle curve) and above-ground respiration (top curve); right panels show net CO2 flux (net uptake is negative). Daily averages were smoothed by 10-point adjacent averaging to enhance clarity. Mowing was done at days 0 and 201.


How do different CO2 fluxes respond to changes in temperature and S level?

Gross photosynthesis was decreased slightly under simulated climate warming. This finding is supported by ecophysiological measurements of these (heated) communities, which revealed increased midday stress in summer with subsequent downregulation of photosystem 2 (B. Gielen et al. unpublished data), and decreased stomatal conductivity (Lemmens et al., 2006) in response to the drier conditions (De Boeck et al., 2006). When expressed on a per unit biomass basis, photosynthesis actually increased in heated communities, apparently in contrast with the above-mentioned ecophysiological measurements. However, because above-ground biomass production declined under warming, and considering that biomass and leaf area index in grasslands are related (Spehn et al., 2000), canopies were less dense under warmer conditions. This probably stimulated the photosynthetic rates per unit biomass as fewer leaves were shaded (Larcher, 2003), while the total photosynthesis was lower because less green mass was present.

The ecophysiological measurements showed that photosynthesis was probably limited under high temperatures and dry soils. Adverse effects of warming on Pgross would therefore have occurred especially during the summer season. The yearly course of gross photosynthesis in the current experiment indeed showed that the largest Pgross decreases in heated communities were recorded in midsummer. Likewise, in a pan-European study on carbon fluxes during the warm 2003 summer, Reichstein et al. (2007) concluded that such a heat wave resulted in reduced photosynthesis. Furthermore, Knapp et al. (2002) showed in a 4-yr experiment on grasslands that CO2 uptake and productivity were negatively related to variability of SWC. As drying of the upper soil layer is faster in warmer conditions, and therefore in heated communities, and as this increases SWC variability, this effect may have further reduced photosynthesis and productivity under warming, especially considering that an important portion of roots is located in this top soil layer (Jones & Donnelly, 2004). Such an effect would again have been most pronounced in summer, when high temperatures stimulate rapid soil drying.

Above-ground respiration was not lowered under warmer conditions, while the respiratory flux per unit biomass was increased by a substantial amount. This was probably associated both with a direct temperature-induced stimulation (Barnes et al., 1998), and possibly with higher levels of abiotic stress resulting in increased maintenance respiration (Larcher, 2003). Below-ground respiration seemed largely unresponsive to warming. It is possible that any direct, temperature-driven respiration increases through improved metabolic efficiency of heterotrophs and higher root turnover (Edwards et al., 2004) were counteracted by secondary responses. The warming-induced drought (De Boeck et al., 2006) could, for example, have limited heterotrophic activity, in accordance with other studies (Gorissen et al., 2004; Harper et al., 2005). Furthermore, the decreased plant productivity observed both above- and below-ground (H.J.D.B. and co-workers, unpublished data), probably also counteracted direct warming-induced stimulation of soil respiration, with several studies highlighting the importance of productivity on below-ground respiration (Saleska et al., 2002; Zhang et al., 2005).

The CO2 fluxes in heated communities were possibly dampened by some degree of acclimation, as Vicca et al. (2007) in another study performed on the same communities observed thermal acclimation of both photosynthesis and total ecosystem respiration. Finally, while we found that SWC did not affect the relationship between the environmental drivers and the CO2 fluxes, this does not signal the absence of drought effects on the fluxes, but merely that these effects were uniform across the entire SWC range, and that no drought thresholds that would disproportionately affect fluxes were exceeded. Meteorological data from the nearby Lint meteo station show that precipitation amounts in both the first (808 mm) and second (733 mm) measurement year were close to average, with slightly wetter-than-average summers, while the average temperature was 11.0°C in both years. The fact that both temperatures and precipitation were very similar in both experimental years makes important drought stress differences between these years unlikely.

Both photosynthesis and above-ground respiration were higher in multispecies communities. This corresponds with our observations on above-ground biomass production, and is also in line with other studies on the relationship between species richness and productivity (Roscher et al., 2005; van Ruijven & Berendse, 2005). In another study, we found that complementarity (including facilitation) caused most of the productivity differences between S levels (H.J.D.B. and co-workers, unpublished data), while net selection effects were small or absent, even though shifts in species success were observed (De Boeck et al., 2006, 2007). Multispecies communities were probably able to capture more of the available light because of less-uniform and hence better-filled canopies (Cernusca, 1976; Spehn et al., 2000; Middelboe & Binzer, 2004), directly resulting in higher photosynthesis. The increased space filling also seemed reflected in the fact that Pgross in mixtures was lower per g biomass in such dense canopies.

Below-ground respiration was similar in communities of different S. Several studies have examined links between species richness and soil respiration, but most of the effects appear to be caused by increased above-ground productivity and subsequently increased litter production in multispecies systems (Zak et al., 2003; Dijkstra et al., 2005). These effects were probably suppressed in our experiment because of the half-yearly mowing and removal of above-ground biomass. Nevertheless, below-ground production was also increased in mixtures (H.J.D.B. and co-workers, unpublished data), and could have stimulated soil respiration. However, the lower SWC in mixtures (De Boeck et al., 2006) may have counteracted any such production-induced stimulation of below-ground respiration.

Do warming and S level affect the CO2 sink or source capacity of grasslands?

Despite the probably overly expanded confidence intervals, we can conclude that the grasslands in this experiment acted as net sinks for CO2. This sink capacity decreased, rather than increasing, under simulated climate warming, especially when also considering the harvested biomass, which was significantly lower in heated communities. Carbon flux studies have reported very varying responses to climate warming, depending mainly on which climatological factor is most limiting to plant growth in the ecosystem studied. While warming could mitigate constraints of low temperatures on metabolic activity in polar regions (Marchand et al., 2004), it could increase heat and drought stress in warmer regions (Llorens et al., 2003). In our temperate grasslands, warming seemed to benefit plant activity in autumn and winter, with increased photosynthesis and above-ground respiration. However, in late spring and summer, when soil moisture limits plant growth more than temperature, we observed a predominantly negative effect of warming, especially on photosynthesis. Studies on currently exceptionally warm summers in temperate regions have shown decreased productivity (Ciais et al., 2005) and inhibited CO2 fluxes (Reichstein et al., 2007) under such conditions. Our research shows that this is likely to occur during many summer seasons in a future warmer world. Even if warming prolongs the growing season (Myneni et al., 1997; Walther, 2003), or increases plant activity in winter, negative effects of warming during summer will probably be dominant in the yearly C balance, as the summer is the main growing season in temperate regions. The effects of species richness on the net CO2 flux were largely unsubstantial, although if taking the harvested biomass into account, multispecies communities were probably larger CO2 sinks than monocultures.

Are there interactive effects on CO2 fluxes of the two global changes studied?

In previous studies on these communities, we have observed consistent interactions between warming and species richness, regarding both water relations (De Boeck et al., 2006) and primary production (H.J.D.B. and co-workers, unpublished data). The differences between the two temperature treatments were most pronounced in the nine-species communities. A similar trend was observed regarding the net CO2 fluxes, although only in year 1 and probably reflecting the photosynthesis fluxes. It is clear that the ‘insurance hypothesis’ (Naeem & Li, 1997) did not apply to these communities, as the negative effect of warming was not dampened at higher species-richness levels. Tilman et al. (2006) argue that the insurance hypothesis may not apply in the short term, after an experiment showed that following an 8-wk drought, the absolute biomass loss was greater at higher species richness (Pfisterer & Schmid, 2002). Our results suggest that the insurance effect may not apply in the longer term if abiotic stress is increased. We have put forward several possible explanations for this (De Boeck et al., 2006; H.J.D.B. and co-workers, unpublished). It is, for example, possible that interspecific competition increased substantially under the higher levels of abiotic stress experienced in heated communities (Callaway et al., 2002; Michalet et al., 2006). This may also explain why the effects of warming were least pronounced, or even absent, at the monoculture level. These findings emphasize the need to study global changes simultaneously, as the responses to single changes are not necessarily additive.


We would argue that careful extrapolation of the current experiment is not unrealistic, for several reasons. First, although drought effects may have been stronger in this experiment compared with natural conditions caused by the absence of a water table, this effect was probably not defining, because in grasslands up to 80% of the roots can be found in the top 30 cm (Jones & Donnelly, 2004). This suggests that grasslands depend mostly on precipitation, and not so much on the water table. Second, although the communities were small in size, both experimental and modelling studies have shown that positive relationships between plant species richness and biomass production are robust and independent of spatial scale or species pools (Cardinale et al., 2004; Roscher et al., 2005). Individual plants interact mainly with close neighbours, making the small-scale distribution of plants of particular importance (Pacala, 1997). Third, in natural grasslands more species are often present, but the objective was to examine the course of S effects, and the strongest changes usually occur at low S levels (Spehn et al., 2005). Finally, we are confident that the direction of the responses will also be consistent in the longer term. We used a variety of grassland species from different functional groups, and opted to include species that also occur in warmer climates, such as F. arundinacea and M. sativa, as well as species that occur in humid grasslands, such as R. acetosa and L. perenne. This ensured that the competitive success of the species could differ depending on which (sets of) traits were beneficial under warmer conditions. Although we did observe shifts in the competitive success of the different species under warming, with species from warmer regions apparently gaining in importance (De Boeck et al., 2006, 2007), the community response was a decrease in production. This suggests that, regardless of compositional changes of temperate grassland communities in the future, warming will probably have a negative effect on community productivity. However, it is imperative that further studies are carried out, also in natural communities, to investigate whether the general effects of warming and species richness that we observed here are consistent.


Our data suggest that warming could cause a decline of net CO2 taken up and stored in temperate grasslands. Effects would be strongest in summer, when heat and drought stress could affect photosynthesis adversely. Lengthening of the growing season was limited or absent, and appeared unimportant in the total C budget. The negative impacts of warming on the sink capacity of temperate grasslands will probably not be mitigated by higher species richness, although multifactorial studies on a longer timescale are needed to improve confidence in this and to identify possible interactions with other global changes.


This research was funded by the Fund for Scientific Research – Flanders (Belgium) as project ‘Effects of biodiversity loss and climate warming on carbon sequestration mechanisms in terrestrial ecosystems’, contract # G.0434.03. N. H. J.D.B. and S.V. hold a grant from the Institute for the Promotion of Innovation through Science and Technology in Flanders (IWT-Vlaanderen). We thank F. Kockelbergh and W. De Boeck for technical assistance.