SEARCH

SEARCH BY CITATION

Keywords:

  • carbon (C);
  • climate;
  • experiments;
  • meta-analysis;
  • mires and peatlands;
  • nitrogen deposition;
  • productivity;
  • Sphagnum

Summary

  1. Top of page
  2. Summary
  3. Introduction
  4. Description
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information
  • Peat bogs have accumulated more atmospheric carbon (C) than any other terrestrial ecosystem today. Most of this C is associated with peat moss (Sphagnum) litter. Atmospheric nitrogen (N) deposition can decrease Sphagnum production, compromising the C sequestration capacity of peat bogs. The mechanisms underlying the reduced production are uncertain, necessitating multifactorial experiments.
  • We investigated whether glasshouse experiments are reliable proxies for field experiments for assessing interactions between N deposition and environment as controls on Sphagnum N concentration and production. We performed a meta-analysis over 115 glasshouse experiments and 107 field experiments.
  • We found that glasshouse and field experiments gave similar qualitative and quantitative estimates of changes in Sphagnum N concentration in response to N application. However, glasshouse-based estimates of changes in production – even qualitative assessments – diverged from field experiments owing to a stronger N effect on production response in absence of vascular plants in the glasshouse, and a weaker N effect on production response in presence of vascular plants compared to field experiments.
  • Thus, although we need glasshouse experiments to study how interacting environmental factors affect the response of Sphagnum to increased N deposition, we need field experiments to properly quantify these effects.

Introduction

  1. Top of page
  2. Summary
  3. Introduction
  4. Description
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Peat bogs significantly affect the earth’s atmosphere (Frolking & Roulet, 2007) by sequestering CO2 in the form of partly decomposed organic material (peat), mostly formed by peat mosses (Sphagnum species; Rydin & Jeglum, 2006). Peat bogs are nutrient-poor wetlands that commonly rely on atmospheric inputs as their sole source of external nutrients, making the vegetation composition sensitive to increases in nitrogen (N) deposition (Bobbink et al., 2010). The vegetation comprises mainly ericaceous dwarf shrubs and cyperaceous graminoids of low productivity, rooting in a matrix of living and dead Sphagnum. Once competition from Sphagnum is reduced, or the nutrient limitation is mitigated, vascular plants may gain a competitive advantage and, being taller, outcompete the mosses due to their better competitive ability for light (Hautier et al., 2009). Shifts from a moss- to a vascular-plant dominated state can depress carbon (C) sequestration rates (Juutinen et al., 2010) and even mobilize the N and C stored in the underlying peat by stimulating microbial activity (Freeman et al., 2004). The peatland store is equivalent to between 34 and 46% of atmospheric CO2 (IPCC, 2007), and hence the ensuing C release into the atmosphere and the local environment may be substantial (Limpens et al., 2008).

Northern hemisphere peat bogs are experiencing a hitherto unprecedented combination of stresses such as increases in N deposition (Galloway et al., 2008), temperature, and drought frequency. Interactions between these factors may initiate a regime shift from a moss- to a vascular-plant dominated plant community, turning peat bogs from C sinks to C sources (Dise, 2009). Interactions between N deposition and other stresses are likely, as shown in a recent statistical meta-analysis of field experiments (Limpens et al., 2011). The analysis suggested that several variables affect the response of Sphagnum species to N deposition including N influx, presence of vascular plant, phosphorus availability, temperature, water table depth and species. The benefit of a meta-analysis of many experiments is that it reveals the influence of factors that are uncontrolled in a field situation. Experimental underpinning is needed, however, to understand the mechanisms underlying these statistical correlations (Dise & Phoenix, 2011).

The design of glasshouse experiments, although artificial, can remove some of the variables allowing a closer examination of the influence of factors, alone or in combination with others, on the response to N application. One of the difficulties of this approach, however, is that the small spatial scale in glasshouse experiments may cause a different response to N (Irvine et al., 2004). Such scale effects make translating the results from glasshouse to field difficult. This is particularly the case if processes operating at a larger scale (e.g. competitive interactions) start to dominate the processes at a small scale (e.g. physiological changes; Englund & Cooper, 2003). Comparing the outcomes of N-application experiments between glasshouse and field experiments, across different experimental spatial scales (from shoot to plot) and organizational complexity (with and without vascular plants), allows us to explore if the responses of Sphagnum to N application depend on the scale of experimentation and whether they are driven mainly by direct or indirect effects. We hypothesized that:

  • the direction of the response (qualitative response) of Sphagnum production and tissue N concentration to N and its sensitivity to environmental modifiers (N influx, presence of vascular plants, phosphorus availability, temperature, water table depth, species) is similar for both glasshouse and field experiments;
  • the strength of the response (quantitative response) of Sphagnum production is lower in glasshouse experiments because of superior growing conditions;
  • the quantitative response of Sphagnum production is determined by the organizational complexity (such as the presence of vascular plants), rather than by the scale (area) of the experimental unit.

Description

  1. Top of page
  2. Summary
  3. Introduction
  4. Description
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Data acquisition

Nitrogen application studies conducted on Sphagnum-dominated vegetation were located by searching the Web of Science and Google Scholar using the key words Sphagnum, nitrogen, peatlands, bogs, mires, and fertilis(z)ation in 2009. In addition, we used our contacts within the peatland researcher community to acquire additional data and information about the experiments. Authors were asked to share raw data, enabling a uniform calculation of treatment effects. For those cases where raw data were irretrievable (nine studies; Supporting Information Table S1), we extracted the data from published manuscripts. In addition, we included unpublished N concentration, production or stem growth data related to published experiments of co-authors of this paper. For multiple-year experiments we only used data from the last year. We selected all studies where the control was subject to the same temperature regime as the fertilization treatments. These selection criteria left us with 44 studies covering 14 countries across North America and Eurasia. The resulting dataset (Table S1) contains information of 222 experiments (115 glasshouse; 107 field) reporting on Sphagnum N concentration (119 glasshouse; 87 field) and 206 reporting on Sphagnum production.

From the 44 studies, we included three response variables and 14 explanatory variables. Response variables were: Sphagnum N concentration in the upper 3 cm of the shoot, Sphagnum production and stem growth. Explanatory variables were: experiment type; number of seasons that N was applied; presence of vascular plants; background N deposition; N application; N application rate; P application; precipitation; temperature; position above the water table (microhabitat); scale of the experimental unit; Sphagnum species; N dose concentration; and form and frequency in which N was applied. Mean and standard deviation (SD) of the response variables were calculated or extracted for all N-treatments per study, treating different species subject to the same treatment, or the same species subject to different treatments, as separate experiments (Gurevitch & Hedges, 1999). In 14 studies, stem growth had been measured instead of production. To maximize the number of experiments included in our analysis, we converted stem growth to production by using the relationship between stem growth and production derived from a subset of glasshouse and field experiments where both variables had been reported. Excluding the stem growth studies from our meta-analysis did not affect the conclusions but did slightly widen the 95% credible intervals.

Sphagnum responses to N application were standardized before the analyses, expressing the effect relative to the control. For each experiment, the effect size was calculated as the natural logarithm (loge) of the response ratio (rr) of Sphagnum production (PROD) or N concentration (N). The rr is defined as the mean of the experimental group (E ) divided by the mean of the control group (C ). The loge of the response ratio was used to linearize the metric and achieve a more normal distribution (Hedges et al., 1999). A negative Nlogerr (hereafter written as Nlnrr) indicates that applying N reduced the N concentration, whereas a positive Nlnrr indicates that applying N increased the N concentration relative to the control. Assuming treatment and control are independent, the variance (Var) of logerr is Var (logeE– logeC) and is calculated as (SDE2/nEE2) + (SDC2/nCC2) (Hedges et al., 1999), where n is the sample size. To better evaluate the relative importance of the explanatory variables they were standardized by subtracting the mean and dividing by two times SD before the analyses (Gelman, 2008). Regression coefficients are then standardized in our models and are directly comparable with each other, including untransformed binary variables (Gelman, 2008).

Statistical model building

We tested our hypotheses using a meta-regression approach (Gurevitch & Mengersen, 2010) using the standardized response of Sphagnum production (PRODlnrr) and Sphagnum N concentration (Nlnrr) to N application as the response variables. Explanatory variables (Table 1) were assessed for collinearity to ensure that modelled variables could be estimated independently. Additionally, we investigated the distribution of the variables to ensure relatively even distribution of data within the N application range, and within categories. When building submodels for testing variables that could not be tested in the main models due to overfitting (Thompson & Higgins, 2002; Lajeunesse, 2010), or too uneven distribution of data within categories (Harrell, 2001), we included N application rate and, if possible, other variables that were significant in the main models (Tables 2, 3; model b). The influence of covariates associated with the N treatments that could potentially bias our results (Table 1) was tested in submodels. For more information on explanatory variables and model choices see Supporting Information Notes S1.

Table 1.   Description and ranges of predictor variables and covariates used in our models (Tables 2, 3, S2)
VariablesDescriptionRange
In models (Tables 2, 3)GlasshouseField
  1. For a more extensive description of the variables see Table S1 and/or Notes S1. Note that some variables (Temperature, Microhabitat, Background N deposition and Number of seasons) were not fully comparable between experiment types.

Experiment typeGlasshouse (G) or field experiment (F)
Presence of vascular plantsPresent vs removed by clipping aboveground shoots (G & F)
Number of seasons6 months = 1 growing season (G), number of growing seasons (F)0.3–31–6
Background N depositionWet N deposition rate (gN m−2 yr−1) at collection site (G) or experimental site (F)0.12–2.80.09-2
N application rateN applied (gN m−2 yr−1) to experimental unit (G & F)0.25–11.260.46–15
P applicationP applied vs no P applied (G & F)Yes–noYes–no
TemperatureMean glasshouse temperature (G) or mean July temperature (F) in °C15–2510.6–20
MicrohabitatSpecies characteristic of (G) wet lawn vs dry hummock microhabitat (F)Lawn-hummockLawn-hummock
ScaleExperimental unit to which treatments where applied (G & F)Shoot, potPot, plot, field
Variables & Covariates
In submodels (Table S2)   
Sphagnum speciesDominant species in the experimental unit (G & F)
N dose concentrationN concentration (g l−1) of the fertilizer solution (G & F)0.0002–20.002–2
N formForm (NH4+, NO3, NH4NO3) in which N was applied (G & F)All threeAll three
N frequencyFrequency: low < 6 times yr−1 ≤ medium ≤ 0.3 times wk−1 (high) of N application (G & F)HighLow, medium, high
Table 2.   Results of three (a–c) Hierarchical Bayes Linear Models (HBLM) on the response of Sphagnum N concentration (Nlnrr, loge (treatment/control)) with standardized model coefficients (St. coef.)
 NSt. coef.UpperLowerPτ2Rob.
  1. Note that Number of seasons replaces Presence of vascular plants, which affected PRODlnrr (Table 2) but not Nlnrr. Exp, total number of experiments; N, number of experiments within category. Negative coefficients indicate that an increase in the predictor variable depresses the response of Sphagnum to N addition. Categorical levels are compared to the intercept, which for model b is set to field experiments, in the lawn microhabitat, without phosphorus application. Interaction terms show the differences in intercept, or slope for continuous variables, between groups within a variable. For example, the slope for N application rate in the field is the term Nitrogen (N) application rate, while the slope for glasshouse experiments is acquired by adding the terms Nitrogen (N) application rate and Experiment type (glasshouse) × Nitrogen (N) application rate. Upper and lower = 95% credible intervals; P, two-sided P-value derived from the posterior probability that the regression coefficient is zero; τ2, hierarchical variance (i.e. heterogeneity); Rob., robustness results assessed by a leave-one-out jack-knifing approach. 0, sensitive; +, robust; ++, very robust.

Model a: overall differences
Exp = 206
Intercept (field)870.310.410.21<0.010.07 
Experiment type (glasshouse)119−0.010.15−0.170.88  
Model b: interactions with environmental variables (Fig. 2a)
Exp = 206
Intercept (field, lawn species, no P) 0.410.500.33<0.010.02++
Experiment type (glasshouse) −0.080.03−0.180.13  
No. of seasons 0.180.260.10<0.01 ++
Background N deposition −0.22−0.09−0.35<0.01 ++
Nitrogen (N) application rate 0.350.470.24<0.01 ++
Nitrogen (N) application rate^2 −0.37−0.23−0.52<0.01 ++
Phosphorus (P) application −0.050.04−0.140.27  
Temperature 0.110.28−0.050.18  
Microhabitat (hummock) −0.18−0.08−0.28<0.01 ++
Experiment type (glasshouse) × no. of seasons 0.020.22−0.170.82  
Experiment type (glasshouse) × Background N deposition 0.280.420.13<0.01 ++
Experiment type (glasshouse) × Nitrogen (N) application rate 0.010.14−0.130.92  
Experiment type (glasshouse) × Nitrogen (N) application rate^2 0.260.430.09<0.01 ++
Experiment type (glasshouse) × Phosphorus (P) application −0.17−0.04−0.30<0.01 +
Experiment type (glasshouse) × Temperature −0.080.11−0.280.40  
Experiment type (glasshouse) × Microhabitat (hummock) 0.360.500.22<0.01 ++
Model c: effect of scale (Fig. 2c)
Exp = 206
Intercept (Scale - shoot)630.280.330.23<0.010.02++
Scale - pot710.030.10−0.040.37  
Scale - plot720.080.150.010.02 +
Nitrogen (N) application rate 0.360.430.30<0.01 ++
Nitrogen (N) application rate^2 −0.19−0.11−0.27<0.01 ++
Table 3.   Results of three (a–c) Hierarchical Bayes Linear Models (HBLM) on the response of Sphagnum production (PRODlnrr, loge (treatment/control)) with standardized model coefficients (St. coef.)
 NSt. coef.UpperLowerPτ2Rob.
  1. Exp, total number of experiments; N, number of experiments within category. Negative coefficients indicate that an increase in the predictor variable depresses the response of Sphagnum to adding N. Categorical levels are compared to the intercept, which for model b is set to field experiments, without vascular plants, in the lawn microhabitat without phosphorus application. For interpretation interaction terms see legend Table 2. Upper and lower = 95% credible intervals, P, two-sided P-value derived from the posterior probability that the regression coefficient is zero; τ2, hierarchical variance (i.e. heterogeneity); Rob., robustness results assessed by a leave-one-out jack-knifing approach. 0, sensitive; +, robust; ++, very robust.

Model a: overall differences
Exp = 222
Intercept (field)107−0.070.01−0.140.080.04 
Experiment type (glasshouse)1150.020.11−0.080.74  
Model b: interactions with environmental variables (Fig. 3a)
Exp = 222
Intercept (field, no vascular plants, lawn species, no P) 0.250.400.09<0.010.02++
Experiment type (glasshouse) −0.32−0.15−0.49<0.01 ++
Presence vascular plants −0.52−0.35−0.68<0.01 ++
Background N deposition −0.43−0.23−0.62<0.01 ++
Nitrogen (N) application rate −0.15−0.05−0.25<0.01 ++
Phosphorus (P) application 0.270.430.12<0.01 +
Temperature −0.010.18−0.200.92  
Microhabitat (hummock) −0.160.00−0.320.04 0
Experiment type (glasshouse) × Presence vascular plants 0.550.840.27<0.01 ++
Experiment type (glasshouse) × Background N deposition 0.410.630.20<0.01 ++
Experiment type (glasshouse) × Nitrogen (N) application rate 0.030.16−0.100.64  
Experiment type (glasshouse) × Phosphorus (P) application −0.24−0.05−0.430.01 0
Experiment type (glasshouse) × Temperature 0.100.32−0.120.37  
Experiment type (glasshouse) × Microhabitat (hummock) 0.190.47−0.080.15  
Microhabitat (hummock) × Temperature −0.26−0.02−0.500.03 0
Model c: effect of scale (Fig. 3c)
Intercept (Scale - plot - no vascular plants) 0.320.490.15<0.010.03+
Scale - field5−0.160.10−0.430.21  
Scale - pot17−0.30−0.11−0.49<0.01 0
Scale - shoot62−0.38−0.20−0.56<0.01 +
Presence vascular plants (plot)62−0.59−0.39−0.78<0.01 ++
Scale - pot × Presence vascular plants (pot)670.450.730.17<0.01 +
Nitrogen (N) application rate9−0.12−0.06−0.19<0.01 ++

In order to test if glasshouse and field experiments differ in their overall response to N application, we fitted a model with experimental type as the only variable (Tables 2, 3; model a). To explore if glasshouse experiments are reliable proxies for field experiments for the assessment of the interactions between N deposition and environment, we used the models for the response of production (PRODlnrr) and Sphagnum N concentration (Nlnrr) from Limpens et al. (2011), adding interactions between the explanatory variables and experimental type (Tables 2, 3; model b). To test interactions with species we ran separate models for those species for which we had a substantial amount of data covering a broad range of our explanatory variables for both glasshouse and field experiments (Table S2). To investigate if there is a linear trend between Sphagnum response and scale we ran a separate model (Tables 2, 3; model c).

Sampling dependence and hierarchical Bayes linear model (HBLM)

The linear mixed-model of a meta-analysis can be expressed as: y = Xβ + δ + ε (y, vector of effect size estimates (logerr); X, design matrix with the explanatory variables; β, vector of parameters (including an intercept term and the effects of the explanatory variables); δ, identity matrix with τ2 (i.e. the residual heterogeneity) along the diagonal). Residual heterogeneity is the variability among experimental outcomes that is not explained by the explanatory variables in the model. ε is the sampling variance–covariance matrix that is assumed to be known and has experiment-specific variances on the diagonal. As indicated earlier, we calculated the response ratio for different N application rates by dividing the mean of the experimental group by the mean of the control group. Because many studies used multiple N application rates and only one control treatment, the same samples were used as control for more than one experimental group, creating sampling dependence in the responses that had to be dealt with (Gurevitch & Hedges, 1999). We controlled for the sampling dependence by including co-variances between related experiments in ε as off-diagonal blocks (Hedges et al., 2010) in a hierarchical Bayes linear model (HBLM; Kulmatiski et al., 2008; Stevens & Taylor, 2009). For further details on calculations of co-variances between experiments and comparison of HBLM with the method of moments estimation see Limpens et al. (2011). Controlling for sampling dependence is crucial in a study such as ours as our dataset had many multiple-treatment studies (Table S1). We performed the analyses in R (R Development Core Team, 2011), employing the metahdep package (Stevens & Nicholas, 2009). In a HBLM set-up metahdep uses a noninformative normal prior on β|(τ), and a log-logistic prior on τ. See Stevens & Taylor (2009) for computational details. Regression coefficients are given with 95% credible intervals, meaning that β lies within the interval with a posterior probability of 0.95. Credible intervals were calculated as two times the posterior SD of the coefficients. Two-sided P-values for the coefficients are also provided for a more familiar interpretation of significant effects.

Model checking

In order to examine sample size bias, we plotted effect size vs variance and number of replicates (Supporting Information Fig. S1a–d). In addition, we checked potential bias of our results owing to differences in within-study variance of the two experiment types. Barring one outlier (Fig. S1c), mean within-study variance and SD were almost identical between field and glasshouse experiments. For general model checking we used residual analyses. To check for robustness in our significant coefficient estimates (i.e. < 0.05), a leave-one-out jack-knifing approach was used where we re-ran models after removing each observation, or study, sequentially from the initial dataset (Efron & Tibshirani, 1993). The result was considered weakly robust (+; Tables 2, 3) when the coefficient remained significant for all runs where one experiment was left out (i.e. Nexperiment − 1 runs), and robust (++; Tables 2, 3) when the coefficient remained significant after a whole study was left out (i.e. Nstudy − 1 runs). The effect of extreme data points was investigated by comparing the model results before and after removing these extremes. For the jack-knifing approach we used a statistical definition of study. We collected data of 44, mainly published, sources. Some of these sources reported experiments that had been performed at multiple sites. Others reported multiple experiments that had been run under different conditions in the glasshouse. At this substudy level we had not 44 but 56 studies that we considered statistically independent and that were used for the jack-knifing approach.

Exploring N treatment artefacts

A major criticism of N-application experiments is that the level of the experimental treatment (in this case, N-application rate) is often substantially higher than the background N deposition received by even highly impacted ecosystems, leading to treatment artefacts (Pearce & van der Wal, 2008). Limpens et al. (2011) showed that the increase in Sphagnum tissue N concentration with N influx across field experiments was indistinguishable from that measured across a natural gradient of N deposition (Bragazza et al., 2005), suggesting that N application side effects, if present at all, were negligible. To test if the same applied to glasshouse experiments, we compared the relationships between nonstandardized Sphagnum N concentration and N influx between glasshouse and field experiments. For field experiments N influx (gN m2 yr−1) was defined as the sum of background N deposition and N application rate; for glasshouse experiments N influx equalled N application rate. The analysis was performed by fitting a generalized least square regression (GLS) to account for the within-study correlation, using the R package nlme (Pinheiro et al., 2011). For more information see Notes S1.

Results

  1. Top of page
  2. Summary
  3. Introduction
  4. Description
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Sphagnum N concentration response

The relationship between Sphagnum N concentration and N influx in glasshouse experiments was almost identical (both in trend and scatter) to that in field experiments for a wide range in N influx (Fig. 1). This result was sustained when experiments at high N influx (10 g N m−2 yr−1) were omitted. The choice of a nonlinear model (N concentration = intercept + β × loge(N influx) was based on previous work (Bragazza et al., 2005; Limpens et al., 2011) to better model N saturation. Fitting a linear model (i.e. not transforming N influx with loge) did not yield any difference between glasshouse and field experiments either (= 0.37), was less successful in predicting N concentration at low N influx, and generally displayed a worse fit (Akaike Information Criterion (AIC)linear = 648, AICnon-linear = 645). Our results suggest that artefacts associated with glasshouse experiments are negligible compared to the variability inherent to field experiments.

image

Figure 1. Relationships between Sphagnum nitrogen (N) concentration in the top 0–3 cm of the shoot and N influx (the sum of background wet deposition and applied N) for glasshouse (circles) and field (triangles) experiments. The fitted model is: N concentration = μ + 2.7 ± 0.3 × loge (N influx) (< 0.01), and is illustrated by lines for glasshouse experiments (dashed line, μ = 11.7, = 67) and field experiments (solid line, μ = 11.4, = 69). The nonsignificant experiment type × N influx interaction (= 0.42) was removed for simplicity but the main effect of experiment type was kept for illustrative purpose, that is, the difference between the lines, even though their effects were tiny and not significant.

Download figure to PowerPoint

Nitrogen application increased Sphagnum N concentration relative to the control (Nlnrr), irrespective of experiment type (Table 2, model a). Moreover, the variation in effect size of Nlnrr did not significantly differ between experiment types (SD glasshouse = 0.28, SD field = 0.25; = 0.14). Examining the interaction effects between experiment type and environmental variables, such as microhabitat, background N deposition, N application rate and P application, did reveal a number of differences between glasshouse and field experiments, however (Table 2, model b). The largest interaction effect was with microhabitat (Fig. 2a). In glasshouse experiments, Sphagnum species characteristic of dry microsites (hummocks) increased more in N concentration than Sphagnum species characteristic of wet microsites (lawns), even at similar water table depth (not shown). In the field experiments, hummock species showed an opposite response, increasing less in N concentration than lawn species per unit of applied N. The response of Nlnrr to background N deposition also differed between experiment types. In glasshouse experiments, Nlnrr was not affected by the background N deposition the mosses had been subject to before the start of the experiment. For field experiments, however, Sphagnum subject to a high background N deposition increased less in N concentration per unit of applied N, than Sphagnum subject to low background N deposition. The interaction with P application was strongly driven by a few experiments and was, therefore, not robust.

image

Figure 2. Response of Sphagnum nitrogen (N) concentration (Nlnrr, loge (treatment/control)) for: (a) glasshouse and field experiments with species characteristic of dry (hummocks, open symbols) and wet (lawns, filled symbols) microhabitats, (b) glasshouse (circles) and field (triangles) experiments for individual species (Sphagnum fallax, S. fuscum, S. magellanicum) and (c) for different scales of the experimental unit (shoot, pot and plot). Positive values for the response ratio, that is, above the thin horizontal line, indicate an increased production relative to the control treatment. Symbols represent fitted means conditioned on the variables in models b and c (Table 2). For tests at species level we added species interaction terms to model b (Table S2). Variables not shown in the figure are kept fixed at their mean value. Confidence in estimates is illustrated by 95% credible intervals. The numbers below the intervals indicate the number of experiments in each group.

Download figure to PowerPoint

Analyses based on individual species (Table S2) showed that S. fuscum was likely the driver behind the microhabitat × experiment type interaction (Table 2, model b). This species had the lowest response value in the field but the highest in the glasshouse, while S. fallax and S. magellanicum had very similar values for both experiment types (Fig. 2b, Table S2).

The scale of the experimental unit hardly affected the N concentration response. Although the data suggested a smaller increase in N concentration in shoot than in plot experiments (Fig. 2c; Table 2, model c), the effect was small and depended on a few experiments and/or studies. Note that none of the experiments using field as experimental unit reported N concentration.

The effect of N dose on Nlnrr did not differ between experiment types (Table S2). Analysis of N form was unreliable as a consequence of unequal distribution of sample sizes (Table S2).

Sphagnum production response

Without controlling for differences in environment, species or design (Table 3, model a), the only difference between the outcome of glasshouse and field experiments was the smaller variation of the production response (i.e. the variation of effect sizes for PRODlnrr) in glasshouse experiments (SD glasshouse = 0.28, SD field = 0.60; < 0.01). After including environmental variables in the model (Table 3, model b), heterogeneity in the outcomes among the experiments was better accounted for and differences in experiment type became evident. While N application in glasshouse experiments had a neutral to negative effect on PRODlnrr, adding similar rates of N in field experiments either stimulated or depressed Sphagnum production relative to the control, depending on the presence of vascular plants (Fig. 3a; Table 3, model b). Although the sample size for glasshouse experiments with vascular plants is low and the interaction experiment type × presence of vascular plants should be interpreted with some caution, the effect is consistent between experiments and studies. For the other interactions with experiment type, only the effects of background N deposition and P application differed between the field and glasshouse experiments (Table 3, model b). The negative effect of background N deposition was less pronounced in glasshouse experiments, and it was not driven by a few experiments or studies. By contrast, the interaction between experiment type and P application was less robust and should be treated with caution.

image

Figure 3. Response of Sphagnum production (PRODlnrr, loge (treatment/control)) for: (a) glasshouse and field experiments without (open symbols) and with (filled symbols) vascular plants, (b) glasshouse (circles) and field (triangles) experiments for individual species (Sphagnum fallax, S. fuscum, S. magellanicum) and (c) for different scales of the experimental unit (shoot, pot, plot and field) without (open symbol) and with (filled symbol) vascular plants. Note: these categories could be distinguished for the pot and plot scale only. Positive values for the response ratio, that is, over the thin horizontal line, indicate an increased production relative to the control treatment. Symbols represent fitted means conditioned on the variables in models b and c (Table 3). For tests at species level we added species interaction terms to model b (Table S2). Variables not shown in the figure are kept fixed at their mean value. Confidence in estimates is illustrated by 95% credible intervals. The numbers below the intervals indicate the number of experiments in each group.

Download figure to PowerPoint

In order to explore whether the differences between glasshouse and field experiments could be attributed to the response of certain species, we fitted a model containing all explanatory variables significant in model b for a data subset including three species (S. fallax, S. fuscum, S. magellanicum) for which we had enough data (Table S2). The difference between glasshouse and field experiments was most pronounced for S. magellanicum, yielding a significant species × experiment type interaction (Fig. 3b, Table S2). While adding N to S. magellanicum in glasshouse experiments depressed production relative to the control, adding N at the same rate in field experiments stimulated production, resulting in a positive PRODlnrr.

The scale of the experimental unit interacted with N application, generally giving a more negative response at smaller scales (Fig. 3c; Table 3, model c). The effects in this model were not very robust, however, and should be treated as tentative. Note that the fitted model was adapted to only fit the interaction with vascular plants at pot and plot scale – the only scales for which we had experiments with and without removal of vascular plants.

The effect of N dose was assessed for data subsets containing all experiments for which information on the N concentration of the fertilization solution was available (c. 90% of all experiments). The analyses showed no effect of experiment type (interaction coefficient estimate: − 0.08, = 0.85; Table S2). Likewise, no clear effect of N form could be detected; However, as most experiments used NH4NO3, these results should be treated with caution (Table S2). Finally, we tested if the difference between glasshouse and field experiments could be related to differences in absolute production because one could argue that the variability of low production rates may be less than that of high production rates. However, when comparing production in control treatments between glasshouse and field experiments, there was no indication of such bias (mean ± SD; glasshouse: 215 ± 155 g m−2 yr−1, field: 229 ± 207 g m−2 yr−1).

Discussion

  1. Top of page
  2. Summary
  3. Introduction
  4. Description
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

How similar are the outcomes of glasshouse and field experiments?

The response of Sphagnum N concentration to N application was similar between glasshouse and field experiments, suggesting that glasshouse studies are suitable for detailed physiological investigations on Sphagnum N uptake and assimilation. The response of Sphagnum production, however, diverged between the experiment types in several important respects. Glasshouse experiments overestimated the negative N-effect on production compared to field experiments in the absence of vascular plant shoots, but underestimated the effect of adding N with vascular plant shoots present. The discrepancy between glasshouse and field experiments was particularly evident at species level where, on average, adding N stimulated production of S. magellanicum in the field, but depressed production in the glasshouse. The above implies that an important component of the N-effect mediated through vascular plants cannot be studied in glasshouse experiments until we understand the mechanisms underlying the interaction with vascular plants better. Similar discrepancies between glasshouse and field experiments when assessing direct and indirect effects of perturbations can be found in other studies, although such effects have hitherto received little attention in plant ecology. For example, in a recent meta-analysis on the effects of water and light on plant performance, Holmgren et al. (2011) showed that physiological (direct) effects were very similar between glasshouse and field experiments, whereas seedling survival (indirect effect) was not.

We hypothesized that the responses of Sphagnum production (PRODlnrr) and tissue N concentration (Nlnrr) in the glasshouse would be modified by the same variables that operate in field experiments. For some variables, the effect was consistent between approaches: in both glasshouse and field experiments production decreased more at high than at low rates of N application. Most other variables, such as presence of vascular plants (PRODlnrr, not significant in glasshouse), temperature (PRODlnrr, not significant in glasshouse), P application (PRODlnrr, weaker effect in glasshouse), background N deposition (PRODlnrr, weaker effect in glasshouse) and microhabitat (Nlnrr, positive in glasshouse, negative in field) were of less significance in the glasshouse than in the field. The reasons behind the lack of sensitivity in the glasshouse remain speculative. Either these variables are mainly controlled for in glasshouse experiments (background N deposition), take effect at time scales beyond the duration of most glasshouse experiments, are mediated through other factors controlled for in glasshouse experiments (presence of vascular plants, temperature, microhabitat), are not fully comparable between the experiment types (temperature, microhabitat), or results reflect a limited amount of data. Sensitivity analyses suggested that the absence of a P effect in the glasshouse is driven by a few influential experiments only. The other effects were fairly robust, however, suggesting alternative explanations. The most interesting difference with field experiments, the vascular plant effect aside, is the absence of an interaction between N application and temperature in glasshouse experiments. Assuming we can equate the effect of average glasshouse temperature with that of mean July temperature, the absence of a temperature effect in glasshouse experiments suggests that the temperature sensitivity reported for field fertilization experiments (Limpens et al., 2011), is likely an indirect effect mediated through other (environmental) factors and not a direct effect of temperature-induced changes in respiration or photosynthetic activity (Limpens et al., 2011). The above clearly highlights the current lack of experiments targeted at elucidating interactions between N application and other environmental variables. In view of the potential consequences of such interactions for the C sequestration potential of peat bogs (Limpens et al., 2011), this lack of mechanistic knowledge is worrying.

Why do glasshouse experiments give different results?

Glasshouse experiments underestimated the (indirect) effect of adding N compared to field experiments when vascular plant shoots are present, indicating that the vascular plant effect found in field experiments at sparse cover is not so much a consequence of competitive interaction, but, rather, it is more likely to be mediated through other factors controlled for in glasshouse experiments. The presence of vascular plant shoots in the field could potentially aggravate the effect of N by increasing interception of snow (Dorrepaal et al., 2003) dry N deposition (Limpens et al., 2004) or, perhaps, enable more intensive competition with algae (Gilbert et al., 1998) by altering the microclimate (Chong et al., 2012). Why the direct effect of adding N in the absence of vascular plant shoots was overestimated in the glasshouse compared to the field remains speculative. The contrasting response cannot be attributed to differences in overall production, at least not under our assumptions that 6 months in the glasshouse equalled one growing season in the field.

One explanation could be that the frequent application of low doses of N characteristic of glasshouse experiments has a more negative effect than infrequent application of high doses common in field experiments. As we found no indication of interactions with N dose and N frequency, this explanation seems unlikely. Another explanation could be that N uptake, or N mineralization, was higher in the glasshouse than in the field, resulting in N saturation and associated negative effects occurring more rapidly (Limpens et al., 2011). Although the relationship between N influx and the response of Sphagnum N concentration seemed to be more linear in the glasshouse, suggestive of more efficient N uptake, there was no consistent difference in unstandardized Sphagnum N concentrations between glasshouse and field experiments (Fig. 1).

Alternatively, the deviating microclimate (low light intensity, low relative humidity, absence of seasonality) in glasshouse experiments may have interacted with the N effect, particularly because Sphagnum performance is sensitive to microclimate (Chong et al., 2012). Although most Sphagnum species saturate photosynthesis at low to intermediate photosynthetically active radiation (PAR) up to 550 μmol m−2 s−1 (Marchall & Proctor, 2004; Hájek et al., 2009), light intensities in glasshouse experiments are often lower. The latter particularly applies to glasshouse experiments conducted in winter and climate chamber experiments where the only source of light is artificial. As such we cannot exclude that at low light different N allocation patterns (Harley et al., 1986) or, perhaps, changes in leaf morphology (Manninen et al., 2011) may have interacted with N addition, even though recent evidence for S. capillifolium suggests otherwise (Bonnett et al., 2010) and photosynthetic capacity is not negatively affected by N application in glasshouse or field experiments (Granath et al., 2009, 2012). Interactions between N influx and microclimate may also explain the contrasting response of production between glasshouse and field experiments observed for S. magellanicum (Fig. 2b). For this species, adding the same amount of N depressed production in the glasshouse, but stimulated production in the field, despite controlling for differences in environmental variables. Changes in the sign of responses clearly indicate that the production response to N in the glasshouse was constrained compared with responses in the field. Glasshouse experiments focusing on the effect of adding N on Sphagnum at contrasting irradiance levels (Bonnett et al., 2010) may help to explain some of the uncertainty regarding the underlying mechanisms.

Do scale or organizational complexity matter?

Organizational complexity affected the response of Sphagnum production to N additions, but neither scale nor organizational complexity modified the response of Sphagnum N concentration. The response of Sphagnum N concentration to adding N was similar among shoot, pot or plot experiments, corroborating the direct cause–effect relationship between N deposition and tissue N concentration of bryophytes (Markert et al., 2003). The main process determining the tissue N concentration, N uptake by Sphagnum, is the same for a single shoot, hummock or carpet and, thus, relatively insensitive to the type of experiment, its scale or the organizational complexity (Levin, 1992; Englund & Cooper, 2003). By contrast, the response of production to N application turned out to be far more sensitive to differences in experimental approach, reflecting its more indirect relationship with N deposition (Manning et al., 2006). While removing vascular plant shoots, that is, reducing organizational complexity, affected the direction of the response of production to N application (qualitative response; Irvine et al., 2004), the scale of the experimental unit only affected the strength of the response (quantitative response). The negligible effect of scale in assessing qualitative responses to N application corroborates earlier work by Wiedermann et al. (2008) who found high correspondence between N effects in a medium-scale field N application experiment and a large-scale field survey along a N deposition gradient in Sweden in the presence of vascular plants.

Conclusion

Glasshouse and field experiments gave similar qualitative and quantitative estimates of changes in Sphagnum N concentration in response to adding N. However, glasshouse-based estimates of changes in production, even qualitative assessments, diverged from field experiments owing to a stronger N effect on the production response in the absence of vascular plants in the glasshouse, presumably caused by the artificial microclimate; and a weaker N effect in the presence of vascular plants compared to field experiments. Thus, although we need glasshouse experiments to study the effects of interacting environmental factors on the response of Sphagnum to increased N deposition, we need field experiments to properly quantify these effects.

Acknowledgements

  1. Top of page
  2. Summary
  3. Introduction
  4. Description
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

We are indebted to Dr John Stevens for his statistical advice on meta-regressions. We thank Uppsala University and Wageningen University for facilitating contact between the lead authors. Finally, a warm thanks to all researchers and sponsors associated with the N application experiments included in this meta-analysis: without their work and support, ours would not have been possible.

References

  1. Top of page
  2. Summary
  3. Introduction
  4. Description
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information
  • Bobbink R, Hicks K, Galloway J, Spranger T, Alkemade R, Ashmore M, Bustamante M, Cinderby S, Davidson E, Dentener F et al. 2010. Global assessment of nitrogen deposition effects on terrestrial plant diversity: a synthesis. Ecological Applications 20: 3059.
  • Bonnett SAF, Ostle N, Freeman C. 2010. Short term effect of deep shade and enhanced nitrogen supply on Sphagnum capillifolium morphophysiology. Plant Ecology 207: 347358.
  • Bragazza L, Limpens J, Gerdol R, Grosvernier P, Hájek M, Hajkova P, Iacumin P, Kutnar L, Rydin H, Tahvanainen T. 2005. Nitrogen content and δ15N signature of ombrotrophic Sphagnum plants in Europe: to what extent is the increasing atmospheric N deposition altering the N-status of nutrient-poor mires? Global Change Biology 11: 106114.
  • Chong M, Humphreys ER, Moore TR. 2012. Microclimatic responses to increasing shrub cover and its effect on Sphagnum CO2 exchange in an ombrotrophic bog. Ecoscience, in press.
  • Dise NB. 2009. Peatland response to global change. Science 326: 810811.
  • Dise NB, Phoenix GK. 2011. Peatlands in a changing world. New Phytologist 191: 309311.
  • Dorrepaal E, Aerts R, Cornelissen JHC, Callaghan TV, Van Logtestijn RSP. 2003. Summer warming and increased winter snow cover affect Sphagnum fuscum growth, structure and production in a sub-arctic bog. Global Change Biology 10: 93104.
  • Efron B, Tibshirani R. 1993. An introduction to the bootstrap. New York, NY, USA: Chapman and Hall.
  • Englund G, Cooper SD. 2003. Scale effects and extrapolations in ecological experiments. Advances in Ecological Research 33: 161202.
  • Freeman C, Fenner N, Ostle NJ, Kang H, Dowrick DJ, Reynolds B, Lock MA, Sleep D, Hughes S, Hudson J. 2004. Export of dissolved organic carbon from peatlands under elevated carbon dioxide levels. Nature 430: 195198.
  • Frolking S, Roulet NT. 2007. Holocene radiative forcing impact of northern peatland carbon accumulation and methane emissions. Global Change Biology 13: 10791088.
  • Galloway JN, Townsend AR, Erisman JW, Becunda M, Cai Z, Freney JR, Martinelli LA, Seitzinger SP, Sutton MA. 2008. Transformation of the nitrogen cycle: recent trends, questions, and potential solutions. Science 320: 889892.
  • Gelman A. 2008. Scaling regression inputs by dividing by two standard deviations. Statistics in Medicine 27: 28652873.
  • Gilbert D, Amblard C, Bourdier G, Francez A-J. 1998. Short-term effect of nitrogen enrichment on the microbial communities of a peatland. Hydrobiologia 373: 111119.
  • Granath G, Strengbom J, Rydin H. 2012. Direct physiological effects of nitrogen on Sphagnum: a greenhouse experiment. Functional Ecology 26: 353364.
  • Granath G, Wiedermann MM, Strengbom J. 2009. Physiological responses to nitrogen and sulphur addition and raised temperature in Sphagnum balticum. Oecologia 161: 481490.
  • Gurevitch J, Hedges V. 1999. Statistical issues in ecological meta-analyses. Ecology 80: 11421149.
  • Gurevitch J, Mengersen K. 2010. A statistical view of synthesizing patterns of species richness along productivity gradients: devils, forests, and trees. Ecology 91: 25532560.
  • Hájek T, Tuittila E-S, Ilomets M, Laiho R. 2009. Light responses of mire mosses – a key to survival after water-level drawdown? Oikos 118: 240250.
  • Harley PC, Tenhunen JD, Murray KJ, Beyers J. 1986. Irradiance and temperature effects on photosynthesis of tussock tundra Sphagnum mosses from the foothills of the Philip Smith mountains, Alaska. Oecologia 79: 251259.
  • Harrell FE. 2001. Regression modeling strategies: with applications to linear models, logistic regression, and survival analysis. New York, NY, USA: Springer-Verlag.
  • Hautier Y, Niklaus PA, Hector A. 2009. Competition for light causes plant biodiversity loss after eutrophication. Science 324: 636638.
  • Hedges LV, Gurevitch J, Curtis PS. 1999. The meta-analysis of response ratios in experimental ecology. Ecology 80: 11501156.
  • Hedges LV, Tipton E, Johnson MC. 2010. Robust variance estimation in meta-regression with dependent effect size estimates. Research Synthesis Methods 1: 3965.
  • Holmgren M, Gomez-Aparicio L, Quero JL, Valladares F. 2011. Non-linear effects of drought under shade: reconciling physiological and ecological models in plant communities. Oecologia. doi: 10.1007/s00442-011-2196-5.
  • IPCC. 2007. Solomon S, Qin D, Manning M, Chen Z, Marquis M, Averyt KB, Tignor M, Miller HL, eds. Climate change 2007: the physical science basis. Contribution of Working Group I to the fourth assessment report of the Intergovernmental Panel on Climate Change. Cambridge, UK & New York, NY, USA: Cambridge University Press.
  • Irvine RL, Crone EE, Jackson LJ, MacIsaac EA. 2004. Does scale affect ecological model predictions? A test with lake responses to fertilization. Ecological Applications 14: 11781188.
  • Juutinen S, Bubier JL, Moore TR. 2010. Responses of vegetation and ecosystem CO2 exchange to 9 years of nutrient addition at Mer Bleue bog. Ecosystems 13: 874887.
  • Kulmatiski A, Beard KH, Stevens JR, Cobbold SM. 2008. Plant–soil feedbacks: a meta-analytical review. Ecology Letters 11: 980992.
  • Lajeunesse MJ. 2010. Achieving synthesis with meta-analysis by combining and comparing all available studies. Ecology 91: 25612564.
  • Levin SA. 1992. The problem of pattern and scale in ecology. Ecology 73: 19431967.
  • Limpens J, Berendse F, Blodau C, Canadell J, Freeman C, Holden J, Roulet N, Rydin H, Schaepman-Strub G. 2008. Peatlands and the carbon cycle: from local processes to global implications – a synthesis. Biogeosciences 5: 14751491.
  • Limpens J, Berendse F, Klees H. 2004. How P affects the impact of N deposition on Sphagnum and vascular plants in bogs. Ecosystems 7: 793804.
  • Limpens J, Granath G, Gunnarsson U, Aerts R, Bayley S, Bragazza L, Bubier J, Buttler A, Van den Berg L, Francez A-J et al. 2011. Climatic modifiers of the response to N deposition in peat-forming Sphagnum mosses: a meta-analysis. New Phytologist 191: 496507.
  • Manninen S, Woods C, Leith ID, Sheppard LJ. 2011. Physiological and morphological effects of long-term ammonium or nitrate deposition on the green and red (shade and open grown) Sphagnum capillifolium. Environmental and Experimental Botany 72: 140148.
  • Manning P, Newington JE, Robson HR, Saunders M, Bradford MA, Eggers T, Ellis RJ, Gange AC, Marhan S, Kandeler E et al. 2006. Decoupling the direct and indirect effects of nitrogen deposition on ecosystem function. Ecology Letters 9: 10151024.
  • Marchall M, Proctor MCF. 2004. Are bryophytes shade plants? Photosynthetic light responses and proportions of chlorophyll a, chlorophyll b and total carotenoids. Annals of Botany 94: 593603.
  • Markert B, Breure A, Zechmeister HG. 2003. Bioindicators and biomonitors – principles, concepts and applications. Amsterdam, the Netherlands: Elsevier.
  • Pearce ISK, van der Wal R. 2008. Interpreting nitrogen pollution thresholds for sensitive habitats: the importance of concentration versus dose. Environmental Pollution 152: 253256.
  • Pinheiro J, Bates D, DebRoy S, Sarkar D, the R Development Core Team. 2011. nlme: linear and nonlinear mixed effects models. R package version 3.1-102.
  • R Development Core Team. 2011. R: a language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing. URL: http://www.R-project.org.
  • Rydin H, Jeglum JK. 2006. The biology of peatlands. Oxford, UK: Oxford University Press.
  • Stevens JR, Nicholas B. 2009. Metahdep: meta-analysis of hierarchically dependent gene expression studies. Bioinformatics 25: 26192620.
  • Stevens JR, Taylor AM. 2009. Hierarchical dependence in meta-analysis. Journal of Educational and Behavioural Statistics 34: 4673.
  • Thompson SG, Higgins JP. 2002. How should meta-regression analyses be undertaken and interpreted? Statistics in Medicine 21: 15591573.
  • Wiedermann MM, Gunnarsson U, Nilsson MB, Nordin A, Ericson L. 2008. Can small-scale experiments predict ecosystem responses? An example from peatlands. Oikos 118: 449456.

Supporting Information

  1. Top of page
  2. Summary
  3. Introduction
  4. Description
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Fig. S1 Scatterplots of effect size vs sample size and sample variance (i.e. within-study variance) for PRODlnrr (a, c), and Nlnrr (b, d).

Table S1 Data file with overview of data sources and explanatory variables

Table S2 Supplementary models for PRODlnrr and Nlnrr testing effects of species, N dose concentration, N form and depth of the water table

Notes S1 Additional methods information: model choices and variable description.

Please note: Wiley-Blackwell are not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing material) should be directed to the New Phytologist Central Office.

FilenameFormatSizeDescription
NPH_4157_sm_FigS1.pdf32KSupporting info item
NPH_4157_sm_TableS1-S2.xls89KSupporting info item
NPH_4157_sm_NotesS1.docx21KSupporting info item