Differential phosphorus and nitrogen effects drive species and community responses to elevated CO2 in semi-arid grassland


  • José M. Grünzweig,

    Corresponding author
      *Present address and author to whom correspondence should be addressed: J.M. Grünzweig, Department of Environmental Sciences and Energy Research, Weizmann Institute of Science, Rehovot 76100, Israel. E-mail: jose.gruenzweig@weizmann.ac.il
    Search for more papers by this author
  • Christian Körner

    1. Institute of Botany, University of Basel, Schönbeinstrasse 6, CH-4056 Basel, Switzerland
    Search for more papers by this author

*Present address and author to whom correspondence should be addressed: J.M. Grünzweig, Department of Environmental Sciences and Energy Research, Weizmann Institute of Science, Rehovot 76100, Israel. E-mail: jose.gruenzweig@weizmann.ac.il


  • 1Productivity of dryland communities is often co-limited by water and nutrients. Since atmospheric CO2 enrichment induces water savings by plants, elevated CO2 and nutrients could interact to reduce growth limitation, irrespective of the direct influence of CO2 on photosynthesis. We studied CO2 effects in model communities from the semi-arid Negev of Israel with 17 mostly annual C3 species at three CO2 concentrations and three nutrient treatments.
  • 2Community biomass increased at elevated (440 and 600 µL L−1) compared to pre-industrial CO2 (280 µL L−1) by 34% on average in the low-nutrient control, by 45% in the high P and by 50% in the high NPK treatment. Less evapotranspiration at elevated CO2 increased soil water content by 30–40% on average. Significant CO2–fertilization interactions indicated that plant responses to CO2 enrichment were constrained by nutrient availability.
  • 3Responses of biomass and water-use efficiency (dry-matter accumulation per cumulative evapotranspiration) to CO2 enrichment were non-linear and were saturated at 440 µL L−1 at low nutrient and high P supply. CO2 effects were further increased up to 600 µL L−1 only under full NPK fertilization.
  • 4The overall CO2 effect on biomass depended on the differential response of plant functional groups, with the P-dependent legume response dominating at low nutrient and high P supply, and the N-dependent grass response dominating at high NPK. With the exception of grasses, species responded differently to combinations of CO2 enrichment and nutrient addition, even within functional groups.
  • 5Biomass production was co-limited by CO2 and nutrients in this semi-arid seasonal community, with both effects possibly mediated by water availability. Nutrient losses associated with desertification will thus diminish potential gains in biomass due to elevated CO2. Growth stimulation by CO2 enrichment beyond close-to-current concentrations will only be seen under nutrient-rich conditions in semi-arid and possibly other drought-stressed grasslands.


Plant growth and survival in semi-arid ecosystems is strongly limited by water supply. The already scarce availability of water might further decrease in many areas as a consequence of desertification, which is particularly grave in semi-arid regions (Schlesinger et al. 1990; Warren, Sud & Rozanov 1996). Any factor increasing water availability is likely to increase plant production in these regions. Atmospheric CO2 enrichment often improves plant water status besides its direct impact on photosynthesis (Körner 2000). The mechanism for this effect in both C3 and C4 plants is a reduction in leaf diffusive conductance caused by increased substomatal CO2 concentration (Drake, Gonzàlez-Meler & Long 1997; Ghannoum et al. 2000; Wand et al. 1999), resulting in decreased transpiration and increased soil water content (Bremer, Ham & Owensby 1996; Field et al. 1997; Grünzweig & Körner 2001b; Niklaus, Spinnler & Körner 1998b). Volk et al. (2000) suggested that most CO2 effects on grassland productivity are caused by water savings and increased soil water content rather than by photosynthetic stimulation. Under dry conditions, water-saving effects should be particularly important for plant growth and ecosystem productivity. Therefore, dryland ecosystems are expected to respond more strongly to elevated CO2 than more humid ones (Lockwood 1999; Mooney et al. 1991; Smith et al. 2000). However, productivity in subhumid (Navas et al. 1995; Shaw et al. 2002) and semi-arid grasslands (Grünzweig & Körner 2001b) increased by 5–20% at elevated CO2, similar to that found in mesic communities (Mooney et al. 1999).

The lack of more pronounced growth stimulation in dryland communities might be related to co-limitation by CO2 and nutrients. Productivity in dry regions (at ambient CO2) is often co-limited by water and N (Hooper & Johnson 1999), contrary to the common belief of limitation by water only. This might have consequences for biomass production under desertification, which causes the nutrient status of ecosystems to deteriorate (Schlesinger et al. 1990). Water shortage does not only limit growth directly, it also restricts soil nutrient availability (Chapin 1991). This effect could potentially be overcome by addition of readily available nutrients. Fertilization increased CO2 responses of above-ground and/or total biomass in a number of grassland communities (Daepp et al. 2000; Joel et al. 2001; Owensby, Auen & Coyne 1994; Stöcklin, Schweizer & Körner 1998), whereas other studies with grasslands from various climatic zones (temperate, alpine, subhumid Mediterranean) showed no or a negative change in biomass response to CO2 enrichment (Hebeisen et al. 1997; Loiseau & Soussana 2000; Reich et al. 2001; Schäppi & Körner 1996; Shaw et al. 2002). Because of the water-saving effect of elevated CO2 we hypothesize that biomass accumulation in dryland communities is co-limited by CO2 and N (or nutrients in general). This would also mean that CO2 responses are nutrient-limited.

Biomass responses to CO2 enrichment were often non-linear, e.g. biomass accumulation was increased over a lower range of elevated CO2 concentrations, but was negligible over a higher range of concentrations (Zangerl & Bazzaz 1984). This was particularly obvious when below-ambient CO2 concentrations were included (Hättenschwiler & Körner 1996; Grünzweig & Körner 2001b), and was potentially caused by non-linearity or thresholds of photosynthesis and other physiological processes and by acclimation (Gill et al. 2002; Körner 2000). However, linearity of CO2 effects might depend on resource supply (Körner 2000). We expected non-linear CO2 effects and dependence of this pattern on nutrient availability.

Most, but not all legume species were often more responsive to CO2 enrichment than other functional groups (Grünzweig & Körner 2001a; Hebeisen et al. 1997; Joel et al. 2001; Stöcklin & Körner 1999). Legume growth depends heavily on N2 fixation (Lüscher et al. 2000; Niklaus et al. 1998a), which is controlled by water and P availability and carbohydrate supply to N2-fixing bacteria (Bordeleau & Prévost 1994; Schulze, Adgo & Schilling 1994). Legumes might profit from elevated CO2 by increased carbohydrate supply to symbiotic associations, as N2 fixation is a process with high respiratory costs for plants (Ainsworth et al. 2002; Lloyd & Farquhar 1996; Niklaus et al. 1998a). In some cases, carbohydrate supply appeared to be sufficient at ambient CO2 (Almeida et al. 2000; Serraj, Sinclair & Allen 1998), and the CO2 effect on legumes was explained by increased growth rate at elevated CO2 subsequently leading to greater N demand, which could be met by increased N2 fixation (Zanetti et al. 1996). Grasses and other non-N2-fixing plant species might not be able to meet their N demand at elevated CO2, and therefore greater responsiveness of legumes was expected.

Besides a generally low nutrient status, dry regions are typically low in P (Schlesinger et al. 1990), which commonly limits legume growth. Moreover, P limitation could have caused negative CO2 effects in a subhumid grassland under different global change scenarios (Shaw et al. 2002). Grass and community biomass production at ambient CO2 is often nutrient (NPK) limited, while legume growth can be depressed by N fertilization (Elisseou, Veresoglou & Mamolos 1995; Kirkham, Mountford & Wilkins 1996; Mitchley, Buckley & Helliwell 1996; Niklaus et al. 1998a). We hypothesized that the CO2 response of grasses and non-leguminous forbs depended on general soil fertility (mainly N), whereas legumes should reach full responsiveness to CO2 by P fertilization alone. In addition, we predicted that species will respond differently even within functional groups because of species-specific response potential and interspecific competition.

We tested the interactive effect of CO2 and P or NPK in species-rich grassland assemblages from the northern Negev under a typical dry-year scenario (40% below mean annual precipitation), and analysed biomass- and water-related responses at the species, functional-group and community level. The use of three CO2 concentrations (280, 440 and 600 µL L−1), including pre-industrial, allowed us to assess the linearity of CO2 responses.

Materials and methods

The grassland that served as the source for plants and soil is located in the hills of the northern Negev, Israel (400 m, 31°21′ N, 34°51′ E). The site is part of the Lehavim Long-Term Ecological Research area and of the Bedouin Demonstration Farm (Agricultural Research Organization), and has a long history of grazing by domestic animals. The region is part of the Irano-Turanian phytogeographical region, but the vegetation includes many species of Mediterranean and Saharo-Arabean origin (Feinbrun-Dothan & Danin 1991), and is dominated by scattered evergreen shrubs and ephemeral grassland species. Seeds (bulbils in the case of Poa bulbosa L. var. bulbosa) of the following 16 mostly annual species representing the ephemeral grassland patches were collected in late spring of 1996 and 1997 (Feinbrun-Dothan & Danin 1991): Aegilops kotschyi Boiss., Brachypodium distachyon (L.) Beauvois, Bromus fasciculatus C. Presl, Crithopsis delileana (Schultes et Schultes fil.) Roshev., Stipa capensis Thunb. (annual grasses); Poa bulbosa (perennial grass); Hymenocarpos circinnatus (L.) Savi, Medicago minima (L.) Bartal, M. truncatula Gaertner, Trifolium campestre Schreber (annual legumes); Carrichtera annua (L.) DC., Daucus subsessilis Boiss., Hedypnois cretica (L.) Dum.-Courset, Plantago coronopus L. ssp. commutata (Guss.) Pilger, P. cretica L., Scabiosa porphyroneura Blakelock (annual non-leguminous forbs). In summer 1997, soil was collected to a depth of 30 cm. Soil was light lithosol on calcareous bedrock with pH 7·9, mineral N 7·5 µg g−1, available P 3·0 µg g−1 (extracted with CO2-saturated water; Eidg. Forschungsanstalten 1996) and water-extractable K 4·7 µg g−1. Plant material and soil were shipped to the University of Basel, Switzerland. Ranunculus asiaticus L. (geophyte) emerged spontaneously from the diaspore bank in the soil during the experiment at similar rates in all containers and was left as part of the plant communities, increasing the number of species in model communities to 17. All species were of the C3 photosynthetic type.

Plants were grown on nutrient-poor 1 : 9 mixtures of Negev soil and calcareous marl (particle diameter 0–8 mm) amended with a base supply of inorganic slow-release full fertilizer (3, 1·8 and 2·3 g m−2 N, P and K, respectively, plus other macro- and microelements; Optima, Münchenstein, Switzerland). Forty-five containers (0·265 m × 0·165 m, 0·21 m deep) were each filled with 13 kg of this substrate, which permitted only slow growth, comparable with that at the site in the northern Negev (Osem et al. 2002). On top of this base supply, fertilizer treatments included the addition of liquid plant-available nutrient solutions corresponding to either 6 g m−2 P only, or 6 g m−2 N, P and K each (as NH4NO3, triple-superphosphate and KCl, Landor, Birsfelden, Switzerland). Each fertilizer was added in two applications of 3 g m−2 at the end of week 3 and week 8 after sowing. Control containers (labelled ‘low-nutrient control’) received only the base supply of nutrients. Each of the two fertilization treatments and the control consisted of 15 containers, five for each of the three CO2 treatments (45 containers in total). The substrate was subsampled for measurement of water content to determine soil dry-weight in containers.

The plant communities were grown in closed growth cabinets (1·2 m2 area; one cabinet for each of the three CO2 treatments), which were equipped with a transparent top cover to allow penetration of natural daylight. The elevated CO2 concentrations (440 and 600 µL L−1) were regulated by an infrared gas analyser (WMA-2, PP Systems, Stotfold, Hitchin, Herts, UK) and a computerized injection system. Pre-industrial CO2 concentration (280 µL L−1) was produced by a small computer-controlled bypass duct passing air in the growth cabinet through a soda-lime scrubber. CO2 treatments were maintained during day and night. The following temperature regime represented early, mid and late Negev winter, respectively: During the first 40 days, 15/10 °C (day/night) were applied, followed by 30 days of 12/8 °C and 30 days of 14/10 °C. In order to compensate for greater cloudiness in Basel as compared to the Negev, natural daylight was supplemented for 10 h per day with three daylight lamps (Power Star HQI-T 1000/D, Osram, Munich, Germany) per chamber, thus achieving a photosynthetically active photon flux density of 500–900 µmol m−2 s−1 at canopy level. Deionised water was provided at a total of 167 mm, which represented a dry year, 40% below the long-term average rainfall in this region of the Negev (Baram 1996). Water was applied at 2–11 mm every 1–10 days, representing the lower half of the range of rain frequency and rain amount per event during a dry year, thus preventing drainage from containers through excessive watering. Some drought-induced wilting was observed particularly in the larger high NPK-treated communities several times during the experiment, with plants at 280 µL L−1 appearing more affected than plants at elevated CO2.

Prior to sowing, legume seed coats were slightly scarified with sand paper to break dormancy (seed-coat impermeability) of these species. In addition, all seeds were stratified in moist soil at 10 °C in the dark for 3 days immediately after sowing to improve germination. Seeds were dispersed uniformly over the container according to the mean natural species densities in ungrazed hillside plots (Y. Osem, M. Sternberg and J. Kigel, personal communication). At harvest, plant density averaged 2400 ± 30 individuals m−2 among the 45 containers. Plant density was similar among all treatment combinations, except for a 17% higher density in high NPK-treated systems at 440 and 600 µL L−1 compared to those at 280 µL L−1 CO2 (P = 0·043 and 0·015, respectively; Tukey-Kramer HSD test), possibly caused by lower seedling mortality.

Fifteen containers (five for each of the three fertilization treatments) were located in each of the three growth chambers. In order to minimize possible chamber effects, we randomised CO2 treatments among growth chambers and positions of the containers within growth chambers every week throughout the 14-week growing period (sowing to peak of vegetative biomass).

The ecosystem water balance was monitored periodically by weighing containers. Soil water content was determined by subtracting the weights of dry soil and empty container from the total weight. Total cumulative evapotranspiration during the entire experiment was determined by subtracting soil water content at harvest from the sum of the total amount of water applied during the experiment and the soil water content at the beginning of the experiment (no water drainage from the containers was noted). The fraction of flowering plants (individuals with at least one flower at anthesis per total number of individuals) was determined weekly for all legumes, the early flowering forb Carrichtera annua and the grass Bromus fasciculatus. The experiment ended with a harvest at the time of early anthesis of grasses (the last functional group to flower) and close to maximum biomass (100 days after sowing). Above-ground plant parts were harvested separately for each species, and were divided into stems, dead leaves, green leaves and reproductive parts. Roots could not be separated into species, and were harvested in bulk. Before drying, leaf area was measured by a photoplanimeter (Li-3100, Licor, Lincoln, Nebraska, USA), and plant material was dried at 80 °C. The term ‘biomass’ (amount of living matter) was used here instead of the term ‘phytomass,’ although a small amount of dead matter was present in the communities.

Results were analysed at the community, functional group and species levels (functional groups were ‘grasses’, ‘legumes’ and ‘forbs’; grasses included five annuals and one perennial). At the community level, variables were analysed with three-way factorial anova (with the factors CO2, fertilization and functional group) for above-ground biomass and its components, and with two-way anova (CO2 and fertilization) for other variables. At the functional-group level, three-way anova (CO2, fertilization and species) was conducted. Where needed, data were log-transformed prior to analysis, in order to homogenise variance. Angular transformation was performed for fractions. Multiple comparisons were performed with the Tukey-Kramer honestly significant difference (HSD) post hoc test.


biomass and leaf area at the community level

Phosphorus and NPK fertilization significantly increased above-ground, below-ground and community biomass per unit ground area irrespective of CO2 treatment (Fig. 1). Compared with the low-nutrient control, total community biomass was 40% larger following high P addition alone and 110% larger following high NPK addition.

Figure 1.

Above-ground (upper part of chart) and below-ground (lower part of chart) biomass at three CO2 concentrations (280, 440, 600 µL L−1) and three fertilization treatments (low-nutrient control, high P, high NPK). Mean ± SE, n = 5 experimental communities. Means of CO2 treatments within the same fertilization treatment that were not indicated by a common letter were significantly different at P ≤ 0·05. Fertilization effect: at each CO2 concentration, control, P and NPK are significantly different from each other, except for below-ground biomass at 440 and 600 µL L−1, where no difference was observed between control and P.

The CO2 effect on biomass depended on the CO2 range considered and on nutrient availability (significant interaction of CO2 and fertilization; Table 1). Tested separately for each nutrient treatment, biomass was significantly greater at 440 than at 280 µL L−1 CO2 in all nutrient treatments (increase in total biomass by 34, 48 and 38% in the control, high P and high NPK treatments). This response was saturated at 440 µL L−1 in the control and the high P treatment (no further increase from 440 to 600 µL L−1; Fig. 1). NPK fertilization, however, led to a significant increase in biomass by 17% at 600 compared to 440 µL L−1 (+61% compared to 280 µL L−1).

Table 1. F-values and probability levels from three-way anova (above-ground biomass and its components) and two-way anova for variables at the community level. Main factors were CO2, fertilization and functional group. Only above-ground dry mass could be separated into functional groups
VariableCO2 (d.f. = 2)Fert. (d.f. = 2)Group (d.f. = 2)CO2 × Fert. (d.f. = 4)CO2 × Group (d.f. = 4)Fert. × Group (d.f. = 4)CO2 × Fert. × Group (d.f. = 8)
  • *

    P≤ 0·05,

  • **

    P≤ 0·01,

  • ***

    P≤ 0·001.

Above-ground biomass 60***139*** 973*** 3·6**20***122***1·5
Stem fraction  3·9*  1·4 672*** 0·5 0·7  2·41·3
Dead-leaf fraction  1·7 23*** 572*** 1·7 6*** 46***1·4
Green-leaf fraction  1·7  2·3 452*** 0·7 3·3*  5·3***0·5
Reproductive fraction  0·5 11*** 170*** 1·8 0·7  3·9**0·5
Dead/total leaf ratio  2·2 17***1026*** 1·2 9·5*** 39***1·7
Below-ground biomass 48*** 55***  4·6**   
Total production 83***297*** 11***   
Leaf area index 13***140***  4·2***   
Cumul. evapotranspiration  4·6* 51***  1·3   
Water use efficiency146***383*** 14***   
Soil water content 36***128***  1·5   
Below-/above-ground ratio  0·9 63***  1·8   
Below-ground bulbil mass  4·7*  0·2  0·8   

All combinations of elevated CO2 and fertilization yielded more than would correspond to the sum of the separate responses to each factor. This was most obvious for the combination of 600 µL L−1 CO2 and high NPK, which resulted in a 220% increase over the combination of 280 µL L−1 CO2 and low-nutrient control. The separate responses summed up to an increase of only 160% (+35% from 600 vs. 280 µL L−1 CO2 in the low-nutrient control, +100% from NPK fertilization vs. control at 280 µL L−1 CO2).

Leaf area index (LAI) remained below 0·7 without additional fertilization, and increased by 70–210% when high P and high NPK were applied (Table 2). Leaf-area responses to elevated CO2 were nutrient-dependent (CO2–fertilization interaction; Table 1), and they were smaller than growth responses (17% at 440 µL L−1 and 27% at 600 µL L−1 compared to 280 µL L−1 CO2) in the low-nutrient control, but greater than the growth response in the high P-only treatment (80% at both 440 and 600 µL L−1). Surprisingly, CO2 enrichment had no effect on LAI when combined with high NPK. This means that CO2 enrichment increased LAI primarily when nutrition stimulated an exclusive legume response (high P-only treatment; see below).

Table 2.  Biomass, leaf-area and water-relation responses to CO2 enrichment at the end of the experiment (100 d after sowing). Soil water content is an average over the second 50 d of the experiment. Mean ± SE, n= 5 experimental communities. Means of CO2 treatments within the same fertilization treatment that were not indicated by a common letter were significantly different at P≤ 0·05 (multiple comparisons were conducted, if CO2 × fertilization was statistically significant, see Table 1)
VariableFertilizationCO2 concentration (µL L−1)
Leaf area index (m2 m−2)Control0·53 ± 0·02b0·63 ± 0·05ab0·68 ± 0·02a
P0·88 ± 0·06b1·58 ± 0·17a1·60 ± 0·23a
NPK1·66 ± 0·12a1·72 ± 0·04a1·74 ± 0·06a
Cumulative evapotranspiration (ET) (kg H2O m−2)Control 155 ± 1 148 ± 1 149 ± 2
P 160 ± 1 159 ± 2 157 ± 2
NPK 165 ± 1 163 ± 2 165 ± 2
Water use efficiency (g total dry wt. kg−1 H2O ET)Control 1·9 ± 0·1b 2·7 ± 0·1a 2·7 ± 0·1a
P 2·4 ± 0·04b 3·6 ± 0·1a 3·5 ± 0·1a
NPK 3·5 ± 0·1c 4·9 ± 0·1b 5·7 ± 0·1a
Soil water content (kg H2O m−2)Control12·8 ± 0·316·3 ± 0·617·3 ± 0·5
P 9·3 ± 0·411·7 ± 0·713·8 ± 0·8
NPK 6·4 ± 0·8 8·7 ± 0·3 8·8 ± 0·4
Below-ground/above-ground ratioControl0·77 ± 0·040·91 ± 0·060·81 ± 0·05
P0·67 ± 0·030·59 ± 0·020·58 ± 0·07
NPK0·45 ± 0·020·48 ± 0·020·46 ± 0·01
Below-ground bulbil dry wt. (g m−2)Control  33 ± 6  39 ± 5  42 ± 8
P  32 ± 5  38 ± 7  41 ± 8
NPK  20 ± 3  44 ± 7  41 ± 2

water use at community level

Cumulative evapotranspiration (ET) over the 100-days growing period increased up to 10% following high P and NPK additions as a consequence of larger biomass (leaf area in the case of P; Table 2). Elevated CO2 reduced cumulative ET by 0·5–4·5%, independently of nutrient treatment (no interaction; Table 1). This relatively small decrease in ET needs to be seen in light of the larger LAI present at elevated CO2.

Increased ET in the high P and high NPK treatments resulted in 25 and 50% reduced mean soil water content as compared to that in the low-nutrient control (Table 2). On the other hand, reduction in ET at elevated CO2 resulted in a less depleted soil water content, to a similar extent at all nutrient treatments (Table 1). Soil was on average 30% wetter at 440 than at 280 µL L−1 and 40% wetter at 600 than at 280 µL L−1. At harvest, soil water content correlated negatively with cumulative ET across all CO2 and nutrient treatments (r2 = 0·98).

Water-use efficiency (WUE) expressed as the ratio of total net biomass accumulation to cumulative ET increased in the high P and NPK treatments, because production was greater than the increment of ET (Table 2). WUE increased by 40–60% at elevated CO2, depending on nutrient additions (Tables 1 and 2).

biomass at the functional-group and species level

Functional groups responded differently to fertilization (significant interactions at the community level; Table 1), as fertilizing grasses and legumes with high P alone resulted in a 10 and 280% increase, respectively, in above-ground dry matter, and fertilizing with NPK improved growth by 210 and 30% compared to the low-nutrient control (Fig. 2). Grasses appeared to suppress legumes once well supplied with nutrients (mainly N) other than P. Non-leguminous forbs were not affected by fertilization of P alone, but responded strongly to high NPK addition (+110%). All grass species were similarly affected by fertilization (no species–fertilization interaction), whereas legume species responded to fertilization in the same direction, but to different extents (species–fertilization interaction; Table 3, Fig. 3). Forb responses differed greatly among species, with some forbs, such as Carrichtera annua, gaining substantially from high NPK addition, while others, such as Scabiosa porphyroneura, were unresponsive.

Figure 2.

Above-ground biomass in grasses, legumes and forbs at three CO2 concentrations (280, 440, 600 µL L−1) and three fertilization treatments (low-nutrient control, high P, high NPK). Mean ± SE, n= 5 experimental communities. Note different scales on the y-axis.

Table 3. F-values and probability levels from three-way anova of above-ground variables at the functional-group level. Main factors were CO2, fertilization and species
VariableCO2Fert.SpeciesCO2 ×  Fert.CO2 ×  SpeciesFert. ×  SpeciesCO2 × Fert. × Species
  1. P≤ 0·05; **P≤ 0·01; ***P≤ 0·001.

Grasses(d.f. = 2)(d.f. = 2)(d.f. = 4)(d.f. = 4)(d.f. = 8)(d.f. = 8)(d.f. = 16)
Above-ground biomass 30***841*** 308***4·8***0·8 2·50·9
Stem fraction 17***  3·6*  87***3·6**2·3* 6·3***1·2
Dead-leaf fraction 15***181*** 303***1·33·3** 0·70·4
Green-leaf fraction121*** 23*** 207***8·6***6·7*** 8·1***1·0
Reproductive fraction  1·2 25*** 605***0·80·9 4·8***1·6
Dead/total leaf ratio 57***132*** 248***1·23·3** 2·2*0·8
Legumes(d.f. = 2)(d.f. = 2)(d.f. = 3)(d.f. = 4)(d.f. = 6)(d.f. = 6)(d.f. = 12)
Above-ground biomass105***181*** 417***3·6**2·4* 4·8***1·3
Stem fraction  0·4  2·2  96***1·23·0** 1·01·6
Dead-leaf fraction  3·6* 27*** 110***0·33·0**12·0***0·7
Green-leaf fraction  2·1  0·8  71***1·33·0** 5·7***1·1
Reproductive fraction 13*** 12*** 112***3·1*8·2*** 2·3*2·0*
Dead/total leaf ratio  3·6* 27*** 112***0·33·2**12***0·6
Non-leguminous forbs(d.f. = 2)(d.f. = 2)(d.f. = 5)(d.f. = 4)(d.f. = 10)(d.f. = 10)(d.f. = 20)
Above-ground biomass  3·3* 33·0*** 103***1·62·4* 2·7**0·9
Stem fraction  4·2*  4·3*1137***0·41·0 3·4***1·1
Dead-leaf fraction  0·5  1·7 192***0·91·9 1·9*0·7
Green-leaf fraction  4·2*  0·1 750***0·13·4*** 2·4**1·0
Reproductive fraction  1·1  2·1 368***0·41·6 1·11·0
Dead/total leaf ratio  0·1  2·2 130***0·71·7 1·9*1·3
Figure 3.

Species above-ground biomass at three CO2 concentrations (280, 440, 600 µL L−1) and three fertilization treatments (low-nutrient control, high P, high NPK). The four species on the left are grasses, the ones in the middle are legumes, and the ones on the right are non-leguminous forbs. Mean ± SE, n= 5 experimental communities. Probabilities of two-way anovas are indicated for the two factors CO2 and fertilization and for the interaction between them (*P≤ 0·05; **P≤ 0·01; ***P≤ 0·001; ns, non-significant). Note different scales on the y-axis.

The CO2 response of grasses was nutrient-dependent, with a 5–20% increase at elevated CO2 in the low-nutrient control and the high P treatment, and 35–50% increase in the high NPK treatment (Table 3, Fig. 2). Legumes were more responsive to CO2 enrichment than the other functional groups, and the extent of this response was again nutrient-dependent (increase in above-ground biomass by 80–120% in the low-nutrient control and the high NPK treatment, and by 150–190% following high P addition alone). The CO2 response of forbs was not affected by nutrient treatments.

All grass species were stimulated similarly by elevated CO2 (Table 3, Fig. 3). CO2 enrichment significantly stimulated the growth of all legume species, yet the extent of the increase in biomass above 280 µL L−1 CO2 varied greatly among them (Table 3, Fig. 3). For example, the relative biomass increase at 600 vs. 280 µL L−1 CO2 in the low-nutrient control amounted to 40% in Hymenocarpos circinnatus and more than 300% in Trifolium campestre. Forb species responded to different extents and sometimes in different directions (Table 3, Fig. 3). Significant production gains at elevated CO2 were found only for Hedypnois cretica and Daucus subsessilis (for the latter species only in the high NPK treatment), while Carrichtera annua tended to lose about 30% above-ground biomass in all nutrient treatments. Many of the differences between 440 and 600 µL L−1 CO2 were non-significant, hence significant CO2 effects resulted largely from the pre-industrial reference CO2 concentration.


NPK fertilization considerably decreased dead-leaf fraction in grasses on the functional-group level (Table 3, Fig. 4) and in all species (data not shown). CO2 enrichment resulted in a general but relatively small increase in stem fraction at the community level (Table 1, Fig. 4). More significantly, CO2 enrichment reduced green-leaf fraction in grasses by 15–40% and increased dead-leaf fraction by 7–15% in all nutrient treatments (Table 3, Fig. 4). Consequently, the ratio of dead to total leaf mass increased, suggesting earlier leaf senescence. It appears that the lack of a CO2 effect on community LAI in the high NPK treatment (despite a large biomass response; Table 2, Fig. 1) reflected leaf senescence in grasses, as these communities had 80% grass dominance. Reproductive fraction was less than 20% in most grasses, but was increased at elevated CO2 in Bromus fasciculatus, the first grass species to flower (Table 4, Fig. 4). Changes in partitioning in legumes and forbs were small and affected only a few species. Variation in partitioning was not a simple correlate of plant size (data not shown), and thus might be a true response to CO2 enrichment, although some contribution by ontogenetic drift cannot be ruled out (McConnaughay & Coleman 1999).

Figure 4.

Partitioning of above-ground phytomass in grasses at three CO2 concentrations (280, 440, 600 µL L−1) and three fertilization treatments (low-nutrient control, high P, high NPK). Fractions of stem (S), dead leaves (D), green leaves (G) and reproductive structures (R) are the proportion of the plant-part mass to above-ground phytomass. Mean, n= 5 experimental communities.

Table 4.  Above-ground partitioning of biomass of the main species. A positive, statistically significant CO2 effect (two-way anova) was indicated by +, a negative effect by –, with one, two and three + or – signs indicating probability levels of ≤ 0·05, ≤ 0·01 and ≤ 0·001. x stands for a significant CO2–fertilization interaction, x(+) and x(–) for an interaction with a significant positive and negative CO2 effect, respectively, at least in one fertilization treatment
SpeciesStemsDead leavesGreen leavesReprod. partsDead/total leaf ratio
  1. n.a., not applicable; all leaves in Trifolium were green.

Aegilops kotschyix x(–) + +
Brachypodium distachyonx(+)+– – – + + +
Bromus fasciculatus+ x(–)+ + ++ + +
Crithopsis delileana  x(–)  
Stipa capensisx(+)+ + +– – – + + +
Hymenocarpos circinnatus+– –  – – –
Medicago minima   + 
Medicago truncatulax(+)    
Trifolium campestren.a.x(+)n.a.
Carrichtera annua  – – – x
Daucus subsessilis+ + + – – –  
Hedypnois cretica+    
Plantago cretica
Scabiosa porphyroneura    

The ratio of below-ground (mainly roots) to above-ground biomass was decreased by high P-only addition, and was further reduced by high NPK addition (Tables 1 and 2). In contrast, CO2 enrichment had no significant effect on below-ground/above-ground ratio irrespective of nutrient availability. The size of special below-ground storage organs (bulbils) was not affected by fertilization, but was increased by CO2 enrichment (Tables 1 and 2). Bulbil dry-matter was more than doubled when CO2 enrichment was combined with high NPK addition. This stimulation was mainly caused by Poa bulbosa, which contributed 60% to the total mass of such storage organs (data not shown).


Time of anthesis was similar among CO2 treatments in all species, except for Bromus, which showed earlier flowering by one week at elevated compared to pre-industrial CO2 (data not shown; Bromus production was 16% of that of all grass species combined). The fraction of flowering individuals averaged over the whole flowering period was increased at elevated CO2 for the legume species, which reflected their overall growth (size) response (P = 0·014, < 0·001, < 0·001 and < 0·001 for CO2 in two-way anova for Hymenocarpos circinnatus, Medicago minima, M. truncatula and Trifolium campestre). No CO2–fertilization interaction on flowering was found in these species.


non-linearity of co2effects

This study of semi-arid model communities revealed a notable non-linearity of CO2-effects, with large differences in plant responses between pre-industrial (280 µL L−1) and close-to-current CO2 concentration (440 µL L−1), and small differences between 440 µL L−1 and distant future CO2 concentration (600 µL L−1). However, CO2 effects were not saturated at 440 µL L−1 at high nutrient (NPK) availability. Similar CO2 saturation of biomass production and soil carbon storage at 400–450 µL L−1 and natural nutrient supply was reported from studies with different annual and perennial communities (Gill et al. 2002; Grünzweig & Körner 2001b; Hättenschwiler & Körner 1996; Zangerl & Bazzaz 1984). Hence, the results of the current study reflected largely past and perhaps current responses in the real world. This is a significant aspect in the interpretation of our data because most other studies used 350–370 µL L−1 as low CO2 and 560–720 µL L−1 as high CO2 (at ambient nutrient availability), a range of CO2 concentrations which is above the most responsive range covered in our study.

co-limitation of co2and nutrients at the community level

Atmospheric CO2 enrichment reduced water consumption and increased community biomass, WUE and soil water content in these semi-arid communities. These results were expected as water savings under CO2 enrichment should ease drought. Part of this effect might reflect improved nutrient availability because nutrient mobility and diffusion to root surfaces increase in wetter soil (Chapin 1991). In Californian grassland, N uptake by plants increased at elevated CO2 due to this effect (Hu et al. 2001; Hungate et al. 1997). Nevertheless, nutrient addition increased growth responses to elevated CO2, suggesting a limitation of CO2 effects by low nutrient availability. Increased nutrient supply appears to alleviate constraint on use of higher CO2 concentrations by plants under dry conditions. In addition to the effect of nutrient supply itself, water-mediated CO2 effects could be increased as a consequence of the drier soil in fertilized as compared to control communities. Irrigation experiments (at natural nutrient supply) showed greater relative biomass responses to elevated CO2 at low than at high water availability (Körner 2000; Poorter & Pérez-Soba 2001). Consequently, low soil water content following fertilization could have increased CO2 effects per se, i.e. irrespective of increased availability of nutrients. Interestingly, interactive effects of CO2 and fertilization were observed only in the primarily growth-related variables (biomass, LAI, WUE), but not in the water-related variables (ET, soil water content). Co-limitation by CO2 and nutrients was observed in tallgrass prairie during a dry year (Owensby et al. 1994), and in individually grown plants in low water treatments (Arp et al. 1998). These interactive responses of growth to CO2 enrichment and nutrient availability confirm our hypothesis of co-limitation of biomass accumulation by these factors in droughted communities, and reflect the known water-nutrient co-limitation in drylands (Hooper & Johnson 1999). Hence, the CO2 effect observed here was largely a water effect, similar to that shown for temperate grassland (Volk et al. 2000).

functional groups drive differential community responses to p and n

The overall community response to CO2 enrichment and fertilization depended on the differential responses of grasses and legumes, and on the nutrient in question, i.e. P or N [addition of N with K is likely to be a N effect because K limitation is rare (Aerts & Chapin 2000)]. At low nutrient supply (control) and in the high P treatment, increased community biomass at elevated CO2 was caused mainly by the strong response of legumes. It seems that even the low-nutrient control supplied enough P to facilitate a relatively strong legume response to CO2 enrichment, a response which was enhanced at high P fertilization. Legume species at elevated CO2 most likely profited from increased N2-fixation rates even under dry conditions, as CO2 enrichment increases drought tolerance of legumes (Serraj et al. 1998). It was hypothesized that P availability to plants could also be increased at elevated CO2 (Gifford, Lutze & Barrett 1996), thus further stimulating legume growth, but there is little evidence for enhanced P uptake by legumes at natural P supply (Gavito, Schweiger & Jakobsen 2003; Staddon, Graves & Fitter 1999). Phosphorus fertilization increased community production response to elevated CO2 in calcareous grassland (Stöcklin et al. 1998), an effect that depended on the presence of legumes (Stöcklin & Körner 1999). Similar to these findings, P fertilization (in combination with K) had no effect on legume production at ambient CO2, but doubled production at elevated CO2 in Californian grassland (Joel et al. 2001).

High NPK supply mainly enhanced the biomass response to CO2 in grasses, with N supposedly being the essential nutrient for this effect, as shown in different grasslands (Daepp et al. 2000; Niklaus et al. 1998a; Owensby et al. 1994). Alternatively, it could be the consequence of N plus P addition, as P limitation reduced biomass production at elevated CO2 in a Californian grassland fertilized with N (Shaw et al. 2002). Legume growth was suppressed at high NPK probably because of competition from grasses. This also suggests that a long-term enhancement of legumes by CO2 enrichment under natural conditions is unlikely, as stimulation of legumes would ultimately enhance N-supply to grasses (Grünzweig & Körner 2001b; Vitousek & Field 1999).

The CO2-induced higher dead-leaf and lower green-leaf fraction in grasses indicated earlier leaf senescence and suggested an advancement of plant development. This is in line with earlier flowering and increased reproductive fraction in Bromus fasciculatus (the only grass species that flowered prior to harvest). Plants tend to show a functional equilibrium response in the sense that they reduce the site or amount of structures that produce a surplus of resources. Green-leaf fraction is reduced at elevated CO2 because of excessive production of non-structural carbohydrates (Grünzweig & Körner 2001a). More rapid plant development at elevated CO2 was obvious only in grasses in our study, and was also observed in wheat (Kimball et al. 1995), barley (Fangmeier et al. 2000) and several Mediterranean grasses (Navas et al. 1997). Enhanced development and earlier flowering is common in dry ecotypes of annuals (Aronson et al. 1992; Del Pozo et al. 2000; Ehrman & Cocks 1996), and can be an advantage in seasonally dry grasslands because more time and water can be used for grain filling. Improved water supply increased grain mass in wheat because of a longer grain-filling period (Li et al. 2000). This is of particular importance in arid and semi-arid climates where large variation in precipitation within and among years can halt rainfall and result in late-season drought (Noy-Meir 1973).

species-specific responses

Legume species differed considerably in their biomass response to CO2 enrichment and in the effect that nutrients had on this. Two mechanisms might cause different biomass responses of species to CO2 enrichment: (1) Species differ in their response potential. For example, Hymenocarpos was the least responsive legume species under all nutrient conditions. (2) Species differ in their interaction with competitors concerning resource availability. Depending on species-specific sensitivity and competitivess, legume species are affected to different degrees by competition for water, nutrients and light imposed by grasses.

Notably, the biomass response to elevated CO2 in grasses was small and not species-specific, similar to the consistent zero effects on vegetative growth of grasses in larger unfertilized communities (Grünzweig & Körner 2001a). Among non-leguminous forbs, which also showed little responsiveness to CO2, Carrichtera annua tended to lose terrain at elevated CO2, in line with earlier results for several members of the Brassicaceae (Grünzweig & Körner 2001a).

Earlier work showed large variation in reproductive output, seed quality and germination success among species in the CO2 treatment (Grünzweig & Körner 2000, 2001a), thus making future changes in species composition and biodiversity likely. These changes will depend on the CO2 concentration, the nutrient status of the ecosystem and the species present.


Plant responses to CO2 enrichment in this semi-arid community were mediated largely by water effects and were limited by nutrient availability. Therefore, the more desertification will deteriorate the nutrient status of ecosystems, the less will primary production respond to rising atmospheric CO2. The differential CO2 response of species, particularly the interactive one with nutrients, suggests consequences for species composition and abundance. Considering the entire postglacial period of past and future CO2 increase, many of the responses reported here might be close to their potential realization in the wild as we rapidly approach a CO2 concentration of 420–450 µL L−1. At these concentrations most of the responses explored in this study appeared to be saturated unless significant amounts of N or NPK were added.


We wish to thank Jaime Kigel, Marcelo Sternberg and Yagil Osem for data on species abundance, Eugene Ungar for climate data, and Jay Arnone and Jürg Stöcklin for statistical advice. We acknowledge the supervision of climate chambers by Lukas Zimmermann and Fritz Ehrsam, and the assistance with chemical analysis and harvests of Susanna Peláez-Riedl, Jaylin Durango, Sandra Djenadic and Brigitte Steullet. We gratefully acknowledge the helpful suggestions of two anonymous referees to an earlier draft of this paper.