Competitive status influences tree-growth responses to elevated CO2 and O3 in aggrading aspen stands


† Author to whom correspondence should be addressed. E-mail:


  • 1Competition effects on growth of individual trees were examined for 4 years in aggrading, mixed-clone stands of trembling aspen (Populus tremuloides Michx.) at the Aspen-FACE free-air CO2 and O3 enrichment facility in northern Wisconsin, USA. During each growing season stands received one of four combinations of atmospheric [CO2] (ambient vs∼56 Pa) and [O3] (ambient vs∼1·5 × ambient).
  • 2Non-destructive measurements of annual tree growth were compared within and among clones and treatments in relation to an index of competitive status based on the difference between a tree's height and that of its four nearest neighbours. Competitive status strongly influenced tree growth, and the positive growth response to elevated [CO2] was greater for competitively advantaged individuals than for disadvantaged individuals of most clones.
  • 3The magnitude of O3 effects on growth depended on clone and competitive status: for some clones, negative O3 effects were stronger with competitive advantage while others showed stronger O3 effects with competitive disadvantage. The interactive effects of CO2 and O3 differed among clones, with negative effects of O3 amplified or ameliorated by elevated CO2, depending on clone and competitive status.
  • 4Treatments modified competitive interactions by affecting the magnitude of growth differences among clones. These modifications did not alter clone rankings of competitive performance, but when CO2 and O3 were both elevated, the differences in competitive performance among clones decreased.


Based on trends in emissions and other human activities, much of the Northern Hemisphere will probably experience substantially increased concentrations of carbon dioxide (CO2) (Keeling et al. 1995) and tropospheric ozone (O3) (Fowler et al. 1999). The impact of these atmospheric changes on terrestrial ecosystems may be considerably modified by an array of biotic interactions (Bazzaz & McConnaughay 1992; Bryant, Taylor & Frehner 1998; Lindroth, Kinney & Platz 1993). Plant–plant interactions could influence responses in natural and managed forests, where competition among individuals is omnipresent and assumed to be an important agent shaping community structure and function (Grace & Tilman 1990; Grime 1979). Free-air CO2 enrichment (FACE) technology can manipulate atmospheric conditions for whole stands of vegetation (Hendrey et al. 1999) and provides the opportunity to examine competition effects on plant responses. The Aspen-FACE experiment in northern Wisconsin, USA is a long-term field study with temperate hardwood stands exposed to altered concentrations of CO2 and O3 (Dickson et al. 2000).

One overarching hypothesis of the Aspen-FACE experiment is that O3 stress will decrease the potential benefit of elevated CO2 for tree growth. Elevated CO2 generally increases photosynthesis and growth (Ceulemans & Mousseau 1994, Saxe, Ellsworth, & Heath 1998; Woodward, Thompson, & McKee 1991), while increasing tropospheric O3 typically has negative effects on these parameters (Bortier, Ceulemans & Temmerman 2000; Clark et al. 1996; Hogsett et al. 1997). Mounting evidence indicates that tree growth under elevated CO2 is mediated by the availability of other limiting resources (Comins & McMurtrie 1993; Medlyn et al. 2000; Oren et al. 2001), and competition may thus compromise growth and survival of individual trees by decreasing availability or acquisition of limiting resources. Tree size is an effective integrator and metric of an individual's competitive status (Lorimer 1983). Size relationships among competitors may affect relative resource availability in two ways: (1) larger individuals interfere with the acquisition of available resources by smaller individuals; and (2) larger individuals have more stored resources, decreasing total resource availability for smaller competitors while giving larger individuals greater stress tolerance and capacity for growth response to treatment. For this study, these functional considerations underlie our hypothesis that competitively advantaged trees will be proportionally more responsive to elevated CO2 and less susceptible to O3 stress than competitively disadvantaged trees.

Differences in size and inherent growth potential interact to determine competitive outcomes within any given neighbourhood. Size relationships among competitors are probably a spatially autoregressive phenomenon: as growth proceeds, disparate size relationships among adjacent competitors are reinforced through time by net differences in absolute growth, even when competitors have similar relative growth rates. While treatment-induced or genotypic growth differences may dampen or reinforce this trend, disparities in absolute growth among competitors can ultimately lead to differential fitness and survival. Based on previous studies (Coleman et al. 1995; Karnosky et al. 1998; Kubiske et al. 1998; Kull et al. 1996), we anticipate treatment-mediated shifts in competitive rankings among aspen clones as a consequence of differing sensitivities to CO2, O3, and their interaction. We expect that the treatment effects on genotypic competitive potential will be manifested through relative growth responses.

The primary objective of the present study was to determine how competition influences trembling aspen (Populus tremuloides Michx.) responses to elevated CO2 and O3 exposures. The second was to examine how these atmospheric constituents influence the competitive potential of aspen clones. We addressed these distinct aspects of competition by examining the role of genotype, atmospheric treatments and their complex of interacting effects on aspen growth. The assessment of competition effects on growth required evaluating the initial size of individuals and their competitive environment as growth determinants (Connolly, Wayne & Bazzaz 2001), as variation in either determinant confers differences in initial advantage to individual competitors (sensuBlack 1958). Within a stand, an individual's size directly affects its subsequent growth, while the individual's size in relation to competitors indirectly affects growth via competition effects. In this experiment we did not manipulate competitive interactions, but relied on (1) existing variation in size relationships among individuals; and (2) genotypic differences in growth rates, with both resulting in substantial variability in competitive environments among individuals.

Materials and methods

experimental design

Dickson et al. (2000) described the study background, FACE ring construction, and experimental design. The study site is in Oneida County, WI, USA (45·6° N, 89·5° W), with mixed, frigid, coarse, loamy Alfic Haplorthod topsoil. During this 4-year study, averages for rainfall and temperature during the growing season (May–September) were 536 mm and 16·0 °C, respectively. During 1998, periodic irrigation ensured establishment and survival of the young trees, but no irrigation has occurred since August 1998.

The design employs three replications of randomized complete blocks containing a 2 × 2 treatment factorial of ambient and elevated [CO2] (36 and 56 Pa, respectively) and ambient and elevated [O3] (ambient and 1·5 × ambient cumulative exposure, respectively). The FACE rings are spaced at least 100 m apart to minimize cross-contamination of treatment gases. Each 30 m diameter FACE ring is divided into sections utilized for separate studies. Atmospheric conditions at the ring centre are monitored and maintained by computerized systems controlling the release of gases on the upwind side of the plot from vertical vent pipes, evenly spaced along the ring perimeter. A buffer zone extends 5 m from the vents, a distance allowing adequate gas mixing to obtain homogeneous atmospheric conditions toward the plot centre. Atmospheric treatments began in 1998 and were applied during the daytime from May to September. One-minute sample averages indicate that elevated CO2 concentrations were within 20% of the target (56 Pa) more than 90% of the time, and daytime averages of elevated [O3] during the growing season were 54·5, 51·1, 48·9 and 52·8 p.p.b. compared with ambient [O3] of 34·6, 36·9, 36·0 and 36·6 p.p.b. for years 1998, 1999, 2000 and 2001, respectively.

plant material, growth measurements and competition indices

The eastern half of each ring contains the trembling aspen competition study. The aspen clones (216, 259, 271, 42E and 8L) have been studied extensively in the Great Lakes region, with results indicating that they span a range for O3 tolerance and phenological characteristics (Karnosky et al. 1998; Kubiske et al. 1998; Kull et al. 1996). In July 1997, rooted cuttings were planted as randomized pairs of each clone with individuals at a 1 × 1 m spacing. The portion of each ring excluding the buffer contains approximately 130 trees.

In early June 1998, and following leaf abscission at the end of each growing season, stem height was measured to the nearest centimetre, and basal diameter was measured to the nearest millimetre at 3 cm above soil surface for all core trees. Dimensions in June 1998 represent the initial size of individuals for the 1998 growth analysis, with final dimensions in 1998, 1999 and 2000 serving as initial sizes for the 1999, 2000 and 2001 growth analyses, respectively. To monitor growth non-destructively, an index of tree size was generated based on the product of diameter2 × height (D2H). Analysis of data from a set of aspen trees harvested from all rings in August 2000 showed that total above-ground biomass was linearly related to D2H (R2 = 0·95, E.P.M., E.L.K. and J.G.I., unpublished results).

Developing a method for quantifying the competitive status of individual trees required: (1) determining the spatial extent of an individual's competitive environment, the ‘neighbourhood’; and (2) developing an appropriate metric for the intensity of competition within neighbourhoods. Preliminary information from spatial statistics showed that an individual's growth was negatively correlated with that of its neighbours over a range of approximately 1·3 m, which included the eight nearest neighbours. This finding, that a tree's competitive environment was apparently dominated by its adjacent neighbours, agrees with results of competition studies in deciduous hardwood stands (e.g. Cole & Lorimer 1994; Hix & Lorimer 1990). Alternative neighbourhood types with membership including the four closest neighbours at the cardinal directions or all eight adjacent neighbours were evaluated. Sample size can be maximized by utilizing individuals both as focal plants and as neighbours of other individuals, but this violates the assumption of independence required for many statistics (Mitchell-Olds 1987). In the present study, competition indices developed for four- vs eight-member neighbourhoods were highly correlated with one another, and yielded similar results and interpretations. As a conservative strategy, this analysis utilized the competition index based on the four nearest neighbours to yield sufficient sample size (c. 70 trees per plot) while meeting independence assumptions, as individuals were tested either as focal plants or as neighbours, but not both.

Several indices of competitive status were compared, including ratios or residuals based on the dimensions (height and D2H) of each individual relative to its neighbourhood. The characterization of neighbourhood was unrelated to clone, with the mean dimensions of the four members of the neighbourhood used in the calculations, and neighbours missing due to mortality treated as zero values. The competition metric was chosen based on model fit in analysis of covariance (ancova, described below), according to (1) residual variance; (2) model-fitting statistics; and (3) covariance linearity. The mean difference between the height of an individual and the height of its neighbourhood (ΔH = heightindividual −heightneighbourhood) during the growing season was chosen as the competitive status index, CSI = (ΔHinitial +ΔHfinal)/2. During a given year, an individual taller than its neighbourhood height was competitively advantaged (CSI > 0), while an individual shorter than its neighbourhood height was competitively disadvantaged (CSI < 0).

growth analyses and statistical methods

Our study is a split-plot experiment, with the atmospheric treatment (ambient; +CO2; +O3; +CO2 + O3) representing the whole-plot level. Clones were chosen based on previous performance. Therefore clone and atmospheric treatments were fixed effects in analysis of variance (anova), whereas replication effects and replication × treatment effects were considered random. Data were analysed according to mixed-model anova or ancova (PROC Mixed, SAS Institute Inc., 1989–2001). Appropriate denominator degrees of freedom for F-tests and least-squares means (LS means) estimation were determined by Satterthwaite's approximation. The replication × treatment error (a) terms were either pooled or partitioned, based on tests of differences between the −2 restricted maximum likelihood indices associated with pooled and partitioned models (Littell et al. 1996). Pooled error (a) terms were appropriate for most analyses.

To address interannual variation that might affect competitive relationships, growth was assessed year-by-year, accounting for individual size at the start of each growth interval and seasonal CSI. Standard growth analysis requires accounting for initial size differences through covariance analysis or by evaluating growth adjusted for initial size. However, significant clonal differences were always observed in the slope of the relationship between initial size and annual growth (data not shown), which precluded using initial size as a covariate or a basis for adjustment across all clones. Accordingly, to evaluate genotypic, treatment and competition effects simultaneously, a simple method to standardize growth responses was devised to account for differences in initial size and complement the approach used to quantify competitive status. Performance under competitively neutral conditions in the ambient treatment served as the reference for assessing treatment and competition effects. For each year, competitively ‘neutral’ subpopulations in ambient conditions were identified, and included individuals having CSI values within ±20% of the median tree height for the entire population. Thus competitively neutral trees had heights within ±26, ±50, ±69 and ±82 cm of their neighbourhood for 1998, 1999, 2000 and 2001, respectively. Neutral populations spanned a range of initial sizes, yielding significant regressions of initial D2H (D2Hinitial) vs final D2H (D2Hfinal) in each year (P < 0·0001, R2 ranges among clones: 0·72–0·92, 0·78–0·93, 0·85–0·94 and 0·88–0·97 for years 1998, 1999, 2000 and 2001, respectively). Each year the regression equations were used to generate predicted final D2H values (D2Hpredicted) for the entire population of trees using D2Hinitial values, and standardized net growth (SNG) was calculated as SNG = (D2Hfinal −D2Hinitial)/(D2Hpredicted − D2Hinitial) to facilitate clone comparisons.

The SNG values reflect each clone's net growth in a given treatment and competitive situation relative to performance under competitively neutral, ambient conditions. Annual variation within SNG responses attributable to genotype, atmospheric treatments, CSI, and their interactions can be evaluated using ancova. Once significant covariance was demonstrated (P < 0·0001) and unequal slopes among clone and treatment combinations were verified, the full factorial of CSI and fixed effects was tested to explore all patterns of covariance possible within the experimental design. To evaluate average effects during the 4-year period and to condense the presentation of results, the annual SNG and CSI values were averaged for each individual, and those 4-year means were tested with ancova. Significant CSI × treatment effects can lead to misinterpretation of differences among whole-population means for clone × atmospheric treatments due to underlying interactive effects of competition. Different slopes in ancova were accounted for by testing fixed effects and generating LS means at the 10th, 50th and 90th quantiles of the covariate population to capture tree performance at CSI values representative of competitively disadvantaged, neutral and advantaged conditions, respectively. Standard errors (SE) of the LS means calculated by PROC Mixed are valid estimates of true standard errors derived from correct linear combinations of the variance components (Littell et al. 1996).

To generate directly comparable relative growth estimates for each clone, the population was restricted to minimize competition effects and avoid initial size × clone effects already described. Sub-sample populations in each treatment included individuals that began and ended a growing season under competitively neutral conditions (i.e. within 20% of mean neighbourhood height yielded adequate representation of all clone × treatment combinations). This approach emphasized the role of genotype, substantially minimizing other sources of variation. While comparisons among clones for absolute or relative growth cannot be conducted for the entire population, for reasons already stated, these neutral populations were smaller and spanned more limited ranges of initial size than the entire population. Thus relative growth (RG), calculated as log(D2Hfinal) − log(D2Hinitial), showed consistent behaviour among clones in relation to initial size. In contrast to SNG, which is a measure of relative responsiveness to treatment and competition effects, RG allows direct comparisons of growth potential among clones. Each year, RG results were adjusted to the mean initial size across treatments using the appropriate linear transformation of initial size. The anova results for individual years and the 4-year average for RG were obtained using PROC Mixed.

Finally, treatment effects on competitive performance (CP) were evaluated for each year, based on changes in an individual's D2H (biomass surrogate) relative to its neighbourhood, where CP = ΔD2HindividualD2Hneighbourhood. Initial size ratios (SR) were calculated each year, as SR = D2Hindividual/D2Hneighbourhood, to serve as covariates for modelling CP results using a PROC Mixed ancova. In this manner, fixed-effects tests and LS means estimated for SR = 1, equivalent to competitively neutral initial conditions, yield CP values for each clone based on growth differences between competitors and neighbours. Average effects were evaluated by testing the mean of annual CP values for each clone from each ring. Annual and 4-year average CP values were ranked by clone, and the significance of crossover effects on clone competitive rankings among treatments were tested using the Van der Laan–de Kroon method (Hühn & Léon 1995; de Kroon & van der Laan 1981).


standardized net growth and competitive status

Competitive status index and its interactions with atmospheric treatments and clone significantly affected SNG in all years; ancova statistics for annual and 4-year average results are reported in Table 1 for reference. The overall positive response of SNG to CSI was consistent among years (P < 0·0001, Fig. 1). Although main effects of clone, CO2 and O3 were significant (P < 0·01), the complex nature of interactive effects can confound the interpretation of main effects. Across all clones and CSI conditions, +CO2 alone had neutral or stimulating effects on SNG. In general, the degree to which +CO2 enhanced growth depended on CSI, the relative enhancement of SNG being 15% greater in competitively advantaged than in disadvantaged conditions (CSI × CO2, P = 0·014). However, the CSI × CO2 effect was apparently due to responses of clones 216, 259, 271 and 42E, which exhibited stronger growth enhancements in +CO2 under competitively advantaged conditions than under disadvantaged conditions. In contrast, clone 8L showed relatively consistent responses to +CO2 that were minimally influenced by CSI. These clone differences probably contribute to the significant CSI × clone × CO2 × O3 effect (P = 0·032).

Table 1.  Levels of significance (P values) for covariance parameters according to mixed-model analysis of covariance (ancova). Standardized net growth (SNG) responses to CO2 and O3 in each growing season were analysed with a factorial of covariance effects utilizing competitive status index (CSI) as the covariate. Average effects of competition were analysed by testing the mean of annual values for SNG and CSI during the 4-year period, with those ancova results identified as the 4-year average
Covariance term19981999200020014-year average
CSI × CO2    0·714    0·027    0·009    0·261    0·014
CSI × O3    0·186    0·607    0·936    0·656    0·901
CSI × CO2 × O3    0·038    0·127    0·186    0·524    0·645
CSI × clone    0·443    0·771<0·001<0·001<0·001
CSI × clone × CO2    0·490    0·902    0·077    0·208    0·654
CSI × clone × O3    0·034    0·177    0·958    0·093    0·041
CSI × clone × CO2 × O3    0·286    0·227    0·341    0·402    0·032
Figure 1.

Standardized net growth (SNG) responses averaged during the 1998–2001 period for mixed-clone aspen stands exposed to combinations of ambient and elevated CO2 and O3. Bars represent least-squares mean estimates (LS means) ± 1 SE for individual clones, with the average response across clones identified as ‘All clones’. Shaded bars, ambient CO2 treatments; unshaded bars, elevated CO2 treatments; open bars, ambient O3 treatments; hatched bars, elevated O3 treatments. The competition status indices (CSI) for this analysis were means of annual CSI values during the 4-year period, with competitively advantaged (+) and disadvantaged (–) LS means calculated at ±90 cm values of the CSI covariate. The dashed horizontal lines denote SNG response in competitively ‘neutral’ (CSI = 0), ambient conditions, for reference. Analysis of covariance (ancova) results for fixed effects of atmospheric treatments, clone and their interactions under competitively advantaged, neutral and disadvantaged conditions are reported next to each panel. Table 1 reports the significance of covariance parameters for individual years and the 4-year average.

Generally, +O3 alone had neutral or negative effects on SNG (Fig. 1). The magnitude of the O3 effect on SNG also depended on clone, with a significant CSI × clone × O3 interaction (P = 0·041) that a