Cumulative nitrogen input drives species loss in terrestrial ecosystems

Authors


An De Schrijver, Ghent University, Laboratory of Forestry, Geraardsbergse Steenweg 267, BE-9090 Gontrode-Melle, Belgium. E-mail: An.DeSchrijver@UGent.be

ABSTRACT

Aim  Elevated inputs of biologically reactive nitrogen (N) are considered to be one of the most substantial threats to biodiversity in terrestrial ecosystems. Several attempts have been made to scrutinize the factors driving species loss following excess N input, but generalizations across sites or vegetation types cannot yet be made. Here we focus on the relative importance of the vegetation type, the local environment (climate, soil pH, wet deposition load) and the experimentally applied (cumulative) N dose on the response of the vegetation to N addition.

Location  Mainly North America and Europe.

Methods  We conducted a large-scale meta-analysis of in situ N addition experiments in different vegetation types, focusing on the response of biomass and species richness.

Results  Whereas the biomass of grasslands and salt marshes significantly increased with N fertilization, forest understorey vegetation, heathlands, freshwater wetlands and bogs did not show any significant response. Graminoids significantly increased in biomass following N addition, whereas bryophytes significantly lost biomass; shrubs, forbs and lichens did not significantly respond. The yearly N fertilization dose significantly influenced the biomass response of grassland and salt marshes, while for the other vegetation types none of the collected predictor variables were of significant influence. Species richness significantly decreased with N addition in grasslands and heathlands [Correction added on 23 March 2011, after first online publication: ‘across all vegetation types’ changed to ‘in grasslands and heathlands’]. The relative change in species richness following N addition was significantly driven by the cumulative N dose.

Main conclusions  The decline in species richness with cumulative N input follows a negative exponential pathway. Species loss occurs faster at low levels of cumulative N input or at the beginning of the addition, followed by an increasingly slower species loss at higher cumulative N inputs. These findings lead us to stress the importance of including the cumulative effect of N additions in calculations of critical load values.

INTRODUCTION

Elevated inputs of biologically reactive nitrogen (N) are considered to be one of the most substantial threats to biodiversity (Sala et al., 2000). World-wide atmospheric deposition of NOx and NHx more than tripled between 1860 and the early 1990s (Galloway et al., 2004), resulting in a large impact on the N status of ecosystems (e.g. De Schrijver et al., 2008). Scientists predict a 2.4- to 2.7-fold increase in nitrogen- and phosphorus-driven eutrophication of terrestrial ecosystems by the year 2050, causing an unprecedented loss of biodiversity and subsequent homogenization of ecosystems (Tilman et al., 2001). Current rates of N deposition are thought to be partly responsible for the already significant losses of terrestrial plant biodiversity (e.g. Stevens et al., 2004; Allen et al., 2007; Verheyen et al., 2009).

Several attempts have been made to scrutinize the driving factors for species loss following excess N input (for an extensive review, see Bobbink et al., 2010). Theory predicts that, above a certain level of primary productivity, local species diversity declines as biomass production increases (e.g. Grime, 1973; Tilman, 1982). But despite the frequently observed increase in biomass and the general decrease in species richness (here defined as the number of species per unit area), much variation in response to N addition has been found among communities and, thus, generalizations across sites or vegetation types cannot be made (Gough et al., 2000; Clark et al., 2007). Clark et al. (2007) tried to unravel the underlying mechanisms of this variation in responses by compiling a database of initial plant community properties and environmental characteristics of 23 experiments across eight herbaceous community types in North America. None of the 12 presumed predictors appeared to explain > 25% of the variation in species richness, whereas collectively they explained only 56% of this variation. Moreover, merely 15% of the variation in production response could be explained in this study. Shaver et al. (2001) found that inter-annual variation in above-ground production after 15 years of fertilizer addition to tundra vegetation was more significant than the variation among applied treatments. Results such as these stress the importance of long-term experiments (Nordin et al., 2009).

To develop a predictive framework of net effects of N addition, broad generality can be achieved by combining data from multiple independent experiments carried out in different ecosystems. Therefore, we conducted a large-scale meta-analysis of in situ N addition experiments, focusing on the effects on the vegetation's biomass and species richness. To our knowledge, this study is the first large-scale meta-analysis of N addition experiments considering biomass and species richness across different continents and vegetation types. Several similar meta-analyses have been published, but they only focused on herbaceous ecosystems in North America (Gough et al., 2000; Suding et al., 2005; Clark et al., 2007) or on net primary production (LeBauer & Treseder, 2008). Our meta-analysis encompasses studies originating from North America and Europe (and one in Australia) that present data for six vegetation types (forest, heathland, grassland, freshwater wetlands, salt marshes and ombrotrophic bogs) in a broad range of added N (annual dose between 0.5 and 48 g m−2 and cumulative dose between 1.5 and 326.4 g m−2). The availability of studies from both continents has the advantage of encompassing a broader range in atmospheric N deposition (North America, 0.1–1 g m−2 year−1; Europe: 0.1–6.0 g m−2 year−1). Hence, we focus here on the relative importance of vegetation type, the local environment (climate, soil pH, wet deposition load) and the experimentally applied (cumulative) N dose on the response of the vegetation to N addition in terms of biomass and species richness.

METHODS

Data sources and collection

Data sources

The search for data concerning N addition experiments focused on peer-reviewed articles that studied the impact of N addition on biomass, the number of species per unit area (further referred to as ‘species richness’) or both variables at the community level. Relevant studies were identified through searches of the bibliographic database of the ISI Web of Knowledge and the references therein (1955 to 2009). The search terms nitrogen, addition, vegetation and experiment* were used, singly and in various combinations. Studies were initially filtered by title and obviously irrelevant articles were omitted. Subsequently, the abstracts were studied with regard to possible relevance to the research questions. Further criteria used for inclusion into the final stage of the meta-analysis were as follows. First, only data from in situ N addition experiments were incorporated into the database. Ex situ N addition experiments or papers considering vegetation change as a consequence of atmospheric depositions were not considered. Second, we included studies performed in terrestrial ecosystems and classified them into forest understoreys, heathlands, grasslands and three types of wetland. The studies conducted in forests were only considered when the forest understorey (herb or shrub layer) response was studied; moreover, studies were only allowed if the forests were at least continuously forested (no clear-cuts) for the last 50 years. Forests under study were mainly coniferous (two deciduous: Gilliam et al., 2006 and Ostertag & Verville, 2002). Studied heathlands were communities where the dominant life-form was small-leaved dwarf shrubs with graminoids and forbs occurring in discontinuous strata (in accordance with Bobbink et al., 2010). Frequently, a ground layer of mosses and lichens was present. Also, Arctic dry heath tundra (Gordon et al., 2001; Gough et al., 2002), alpine tundra (Seastedt & Vaccaro, 2001) and moist non-acidic tussock tundra (Gough & Hobbie, 2003) communities were included. Grassland communities are dominated by grasses (Poaceae) and other herbaceous (non-woody) plants (forbs). Sedges (Cyperaceae) and rushes (Juncaceae) can also occur. Grasslands can be semi-natural, with a traditional extensive agricultural management (grazing or hay making) or natural, as steppe or prairie. Grassland studies were only included if the area had been continuously grassland for the last 20 years. We found studies conducted in three types of wetland: freshwater wetlands (in our dataset, comprising fen meadows, fens and wet sedge tundra; Shaver et al., 1998), salt marshes and ombrotrophic bogs. Third, only experiments in which inorganic N was added (to be comparable with N additions through atmospheric depositions) were included. Simultaneous additions of phosphorus or potassium were omitted, as the impact of these nutrients may interfere with the effect of N. Fourth, to unravel the relative impact of N addition from all other environmental variables being randomly assigned to treatments, only studies reporting results for both N-treated and control plots were selected. Fifth, we focused on articles reporting data on the impact of N addition on biomass or species richness or on both variables. Biomass was predominantly measured aboveground, although one study examined total biomass (Gough & Hobbie, 2003). Studies reporting total community biomass (including vascular plants, mosses and lichens) or subdivided per growth form (graminoids, forbs, shrubs, mosses, lichens) were retained. Some studies only reported the biomass of one or more growth forms (Table 1). We use the term species richness throughout this paper, referring to the number of species at a certain plot scale. We only retained papers reporting species richness on the community level, so species richness data for one or more growth forms reported separately were not considered.

Table 1.  Summary of the articles used in the meta-analysis.
ArticleArticle no.RegionVegetation typeBiomassSpecies richnessSoil acidityAtmospheric wet N deposition (g m−2 year−1)N additionNo. cases
Fertilization rate (g m−2 year−1)No. yearsForm
  1. Vegetation type: 1, forest; 2, heathland; 3, grassland; 4, freshwater wetland; 5, salt marshes, 6, bogs; Biomass: articles reporting data of biomass response: 0, studies on community level; 1, forbs; 2, graminoids; 3, shrubs; 4, bryophytes; 5, lichens; Species richness: articles reporting data on species richness response: x, data available. Soil acidity: 1, acid [pH(H2O) < 4]; 2, neutral [4 < pH(H2O)] < 6); 3, alkaline [pH(H2O) > 6]; Atmospheric wet N deposition data obtained from Holland et al. (2005) are indicated with H; /, no data available; –, no data gathered since no data on biomass or species richness response available on the community level.

Aerts et al. (1992)1Sweden/Lapland60 /0.1/0.92/41NH4NO34
Aerts & Berendse (1988)2The Netherlands20, 2, 3 13.58.7/202NH4NO33
Bassin et al. (2007)3Switzerland30 21.0 H1.5/3/7.5/153NH4NO34
Berendse et al. (2001)4Finland/Sweden/Switzerland60 /0.4/0.6/1.2  NH4NO33
Bergamini & Pauli (2001)5Switzerland34x21.2 H102NH4NO32
Bobbink (1991)6The Netherlands30x31.7103NH4NO34
Bowman et al. (2006)7USA (Colorado)30, 1, 2 10.1 H67NH4NO39
Bowman et al. (2008)8Slovakia30 11.52/6/153NH4NO33
Britton & Fisher (2007)9Scotland2 x11.2 H54NH4NO33
Bubier et al. (2007)10Canada (Ontario)60 11.01.66NH4NO31
Carroll et al. (2003)11UK30 32.03.5/142NH4NO32
Clark & Tilman (2008)12USA (Minnesota)3 x20.4 H1/3.4/9.523NH4NO33
Clark et al. (2007)13USA (Colorado/Kansas)30x2/30.1/0.51016/10NH4NO34
Cole et al. (2008)14Scotland30 21.1 H12/242NH4NO32
Darby & Turner (2008)15USA (Louisiana)50 /0.5 H55.2/111.6/223.2/ 446.4/892.81(NH4)2SO45
Davis et al. (1999)16USA (Minnesota)30 20.55/152NH4NO36
Davy & Bishop (1984)17UK20, 2, 4, 5 31.6 H/3.5NaNO35
Dougherty et al. (1990)18USA (Louisiana)50 30.4 H8.4/16.8/33.61NH4NO33
Foster & Gross (1998)19USA (Michigan)30x20.6482NH4NO32
Gilliam et al. (2006)20USA (West Virginia)10x10.93.56NH4SO41
Gordon et al. (2001)21Norway2 x10.11/58NH4NO32
Gough & Hobbie (2003)22Alaska20, 1, 2, 3, 4, 5 30.1 H104NH4NO36
Gough et al. (2002)23Alaska20, 2, 3, 4, 5x10.1 H108NH4NO36
Gunnarsson & Rydin (2000)24Sweden64 1/3/5/103NH4NO319
Heijmans et al. (2001)25The Netherlands60, 3, 4 11.453NH4NO33
Hejcman et al. (2009)26Czech Republic32 36NH4NO31
Henry et al. (2006)27USA (California)30 30.1 H76Ca(NO3)21
Huenneke et al. (1990)28USA (California)30, 1, 2 30.1 H102NH4NO36
Hurteau & North (2008)29USA (California)10, 1, 3x30.1/0.41.2/2.42NH48
Kelley & Epstein (2009)30Alaska20, 1, 2, 3, 4, 5 30.1 H103NH4NO37
Lamb et al. (2007)31Canada (Alberta)30 30.1 H5.442NH4NO31
Madan et al. (2007)32Norway2 x20.10.5/53NH4NO32
Morgan (2007)33Australia30 1/51NH4NO31
Nordin et al. (1998)34Sweden12, 3, 4 10.41.5/2.5/51NH4NO315
Ostertag & Verville (2002)35Hawaii1 x2/105/11NH4NO3+ urea2
Pauli et al. (2002)36Switzerland40x21.2102NH4NO32
Plassmann et al. (2009)37UK30, 4 31.10.75/1.52NH4NO34
Power et al. (1998)38UK23 0.77/1.548(NH4)2SO42
Rainey et al. (1999)39USA (Massachusetts)10 10.85/ 157NH4NO32
Roem et al. (2002)40The Netherlands2 x14.7105NH4NO32
Samuel & Hart (1998)41USA (Wyoming)30, 1, 2 20.2 H2.2/3.414NH4NO312
Seastedt & Vaccaro (2001)42USA (Colorado)30x20.1154NH4NO3 + NH4SO36
Shaver et al. (1998)43Alaska40, 2, 3, 4 20.1 H105/9NH4NO315
Thormann & Bayley (1997)44Canada (Alberta)4/60 /0.2151NH4NO35
Tomassen et al. (2004)45Ireland62 10.4 H2/4/83NH4NO3/(NH4)2SO46
Turner et al. (1997)46USA (Kansas)30 20.5101NH4NO34
van der Hoek et al. (2004)47The Netherlands40 26.0201NH4NO34
van Wijnen & Bakker (1999)48The Netherlands50 31.85/253NH4NO34
Verhoeven & Schmitz (1991)49The Netherlands40, 4 /2.1201NH4NO36
Virtanen et al. (2000)50England34 //(NH4)2SO4/NaNO32
Wedin & Tilman (1996)51USA (Minnesota)3 x20.4 H1/2/3.4/5.4/9.5/17/27.212NH4NO37
Wilson & Tilman (2002)52USA (Minnesota)3 x20.52/9.5/178NH4NO33

This search yielded 231 articles of which only 52 could be retained for inclusion in the meta-analysis (Table 1). Most studies were performed in North America and Europe, and one study was situated in Australia (Fig. 1). For each study, the vegetation type and the studied growth forms were categorized. Six vegetation types (forest understoreys, heathlands, grasslands, freshwater wetlands, salt marshes and bogs) and five growth forms (forbs, graminoids, shrubs, bryophytes and lichens) were distinguished. Vegetation types and growth forms were categorized following the authors' description. For each study, information on geographic region, vegetation type, studied growth forms, soil acidity, atmospheric wet N deposition and experimental design is summarized in Table 1. We also report the number of cases in Table 1: different sites, vegetation types, growth forms and levels of N dose within a single paper were included as individual cases. We derived 241 individual cases, of which 204 provided data on the biomass response (96 cases at the community level and 108 cases involving one growth form), and 37 contained data on the species richness response. For papers reporting data from multiple monitoring years, only data from the last year were used. The most common N form in the selected addition experiments is ammonium nitrate (NH4NO3) (see Table 1).

Figure 1.

Map of studies included in this meta-analysis across North America, Europe and Australia (not drawn to scale for clarity). Numbers correspond to the article numbers in Table 1.

Collection of environmental predictor variables

To study the relative importance of vegetation type, local environment and the applied (cumulative) N dose, we collected predictor variables known to affect the sizes of biomass and/or species richness. We gathered data on soil pH to have a proxy for the initial buffering status of the soil. Although Clark et al. (2007) found soil cation exchange capacity (CEC) to be the strongest contributor to their multivariate explanation of species richness response, we could not use this variable because of its unavailability in almost all studies. If soil pH values or an indication of soil pH could not be extracted from the article, the authors were contacted (availability: 76/96 cases for the biomass dataset, 37/37 cases for the species richness dataset). The pH(KCl) and pH(CaCl2) values were recalculated into pH(H2O) according to Čapka et al. (2009). In many cases, we only had a rough indication of soil pH. Therefore, we classified the pH values into three biogeochemically relevant buffering classes: aluminium buffer range [pH(H2O) ≤ 4.5], cation exchange buffer range [4.5 < pH(H2O) < 6] and carbonate buffer range [pH(H2O) > 6] (see also Fig. 1 in Bowman et al., 2008).

Furthermore, we collected region-specific data on wet atmospheric N deposition (Table 1). We first searched the original or related publications from the same authors or sites for region-specific values (Appendix S1 in Supporting Information provides an explanation of the procedure followed). For 19 studies, however, no wet deposition data could be obtained following this procedure. Instead, wet deposition data modelled by Holland et al. (2005) were used. For papers reporting wet deposition data, we double-checked the data from Holland et al. (2005) by performing a regression analysis between reported wet deposition values and values from Holland et al. (2005). This regression was used to rescale the values from Holland et al. (2005) in the papers that did not report deposition data (see Appendix S1). The availability of N deposition data amounted to 97/98 cases for the biomass dataset (one missing case from Australia) and 35/37 cases for the species richness dataset (two missing cases from Hawaii).

Additionally, for each study, we gathered site latitude, longitude and altitude data. These values were combined for deducing mean (1961–90) annual temperature (MAT), mean annual precipitation (MAP) and potential evapotranspiration (PET) using nearest neighbour interpolation of the 10 nearest weather stations per site with the New-LocClim 1.10 software (FAO, 2005). Subsequently, we calculated a proxy for precipitation surplus (MAP/PET).

The cumulative amount of N added to the fertilized plots was calculated by summing the yearly amount of added N (g m−2 year−1) over the duration (number of years) of fertilization. Both cumulative N addition and annual amount of added N (further referred to as ‘N fertilization rate’) were retained as predictor values. Finally, although Gough et al. (2000) did not find plot size to affect differences in the magnitude of response to fertilization, we included plot size as a predictor variable.

Data analysis

Following standard meta-analytic procedures, the metric of the effect size E was calculated as E= ln(xN/xC) with xN being the mean biomass or species richness in the N fertilized plots and xC being the mean biomass or species richness in the control plots (Hedges et al., 1999; Borenstein et al., 2009). Zero values indicate no change in the response variable due to N addition, whereas positive (or negative) values correspond to an increase (or decline). The use of the natural logarithm linearizes the metric ratio (it treats deviations in the numerator the same as deviations in the denominator) and provides a more normal sampling distribution in small samples (Hedges et al., 1999). For each case, we calculated E and a 95% bootstrapping confidence interval (CI) with n= 1000 iterations using Resampling Stats v.4.0 (http://www.resample.com/). Due to a lack of standard deviation or error estimates in some studies, the bootstrapping approach used was more appropriate than the standard meta-analytical procedures (Hedges et al., 1999).

For deducing the effect of N addition on biomass, calculations were performed at the community level for all studies and separately for each vegetation type. For studies reporting data on different growth forms, calculations were made for each growth form. To study the effect of N addition on species richness, bootstrapping calculations were performed for all of the studies together and for each vegetation type when at least five replicates were available.

To explore the relative importance of the predictor variables, namely (1) soil pH, (2) yearly site-specific wet N deposition, (3) MAT, (4) MAP/PET, (5) cumulative N addition, (6) N fertilization rate, (7) plot size and (8) vegetation type, on the response of biomass and species richness to N addition, we used multilevel analysis (Mixed procedure in SPSS 15.0). The biomass (in cases studying biomass response) or species richness (in cases studying species richness response) of the control plots was also added to the models. We added a random effect term ‘study’ to the models to address the likelihood that cases obtained from the same study share temporally or spatially autocorrelated characteristics. First, we constructed a null model with the intercepts varying randomly per study (random effect term study). To avoid overfitting, the seven predictors were first evaluated based on the −2 log likelihood information criterion (i.e. deviance; Hox, 2002) as stand-alone predictors. Subsequently, extra predictors were added one-by-one to the first model with the lowest deviance containing only one predictor. If the deviance decreased significantly (chi-square tested P < 0.05; Hox, 2002), this procedure was repeated. To avoid overfitting and for model simplification, only variables with a P-value < 0.05 were considered for the final multilevel model. To detect possible multicollinearity (Graham, 2003) between the different predictor variables, variance inflation factors (VIF) were calculated according to Quinn & Keough (2002). VIF values were low (< 3), indicating low collinearity. Only the final models are presented in the results.

Calculation of the intra-class correlation (according to Hox, 2002) showed that 64% and 47% of the variance in the effect sizes of biomass and species richness, respectively, was caused by the random effect term ‘study’. We then estimated the amount of variation explained by adding the predictor variables to the null model (further referred to as reduced study-level variance) through calculations of the ratio of the difference in residuals between the null model (inline image) (i.e. with only the random factor) and the final model (inline image) over the residuals of the null model (Hox, 2002):

image

RESULTS

The mean effect size of N addition on biomass across all vegetation types was 0.182 (ranging from −0.072 to 0.282). The variation among vegetation types was large (Fig. 2). The effect of N addition on the biomass of the forest understorey was negative, but not significantly. Significant positive effects of N addition on biomass were found for grassland and salt marshes, whereas heathland, freshwater wetland vegetation and bogs did not show any significant response. Significant divergent effects were also found within growth forms (Fig. 3): graminoid response was clearly positive, whereas the biomass of bryophytes decreased significantly following N addition. Forbs and shrubs did not show any significant response.

Figure 2.

Effect of N addition on biomass per vegetation type. Mean effect size, 95% confidence interval (CI) calculated by bootstrapping (see data analysis) and number of cases included in the analysis are shown. If the 95% CIs of the effect sizes do not overlap with 0, responses are significant at P < 0.05 and are marked with *.

Figure 3.

Effect of N addition on biomass per growth form. Mean effect size, 95% confidence interval (CI) calculated by bootstrapping (see data analysis) and number of cases included in the analysis are shown. If the 95% CIs of the effect sizes do not overlap with 0, responses are significant at P < 0.05 and are marked with *.

From the multilevel analysis, we found that none of the collected predictor variables had significant influence on the effect size of biomass. Looking into the data by vegetation type, we found a significant influence of the N fertilization rate (g m−2 year−1) for grassland (F= 6.7, d.f. = 40, P= 0.013) and salt marshes (F= 9.4, d.f. = 11, P= 0.010), reducing the study-level residual variance by 14% and 48%, respectively. For the other vegetation types, no significant relationships were found.

Species richness significantly decreased with N addition across all vegetation types (Fig. 4). The mean effect size of species richness across all studies was −0.322 (ranging from −0.424 to −0.227). When vegetation types were considered separately, species richness only decreased significantly following N addition in heathland and grassland. The species richness of the understorey vegetation in forest ecosystems was not significantly affected by N addition. For salt marshes, freshwater wetland and bogs, no conclusions regarding species richness could be drawn because of a lack of sufficient replicates.

Figure 4.

Effect of N addition on species richness per vegetation type. Mean effect size, 95% confidence interval (CI) calculated by bootstrapping (see data analysis) and number of cases included in the analysis are shown. If the 95% CIs of the effect sizes do not overlap with 0, responses are significant at P < 0.05 and are marked with *.

The multilevel analysis showed a strong significant effect of the cumulative N fertilization dose (F= 58.47, P < 0.001, d.f. = 25) on the effect size of species richness, reducing the study-level residual variance by 70%. The relationship between the cumulative N dose and the effect size of species richness was found to be linear and negative (see Fig. 5). By excluding the study with the largest cumulative N dose (326 g m−2; Wedin & Tilman, 1996), the results remained highly significant (F= 28.09, P < 0.001, d.f. = 26). Moreover, when grassland plots were considered separately, the response ratio of species richness showed significant negative correlation with cumulative N dose, reducing study-level residual variance by 81% (F= 61.07, P < 0.001, d.f. = 15). Considering heathland vegetation cases separately, the response ratio also showed significant negative correlation with cumulative N dose, but it was less significant and reduced study-level residual variance by only 43% (F= 6.98, P= 0.027, d.f. = 9). None of the other predictor variables were found to be a significant determinant for the effect size of species richness.

Figure 5.

The effect of cumulative N dose (g N m−2) on the effect size of species richness for different vegetation types (triangles, forest; closed circles, heath; open circles, grassland; cross, freshwater wetland). Significance is derived from a multilevel model with study as a random factor: F= 58.47, P < 0.001, d.f. = 25, reduced variance compared to the null model: 70%.

DISCUSSION

The relative change in species richness following N addition was significantly driven by the cumulative N dose, and this was observed across all soil types, climatic conditions and N deposition levels. We found a strong negative relationship between the cumulative N dose and the effect size of species richness, reducing residual study-level variance by nearly 70% across all studied vegetation types (forest understoreys, heathland, grassland and one freshwater wetland). The effect size of species richness decreased linearly with the cumulative N dose, which means, because the effect size is a logarithmic value, that species richness declines much faster at low cumulative N inputs than at higher doses. However, our results show that not all vegetation types respond uniformly to N addition: whereas grassland and heathland communities suffer species loss, the diversity in the understorey of forest ecosystems is not significantly affected by N fertilization. From studies reporting data on changes in biomass following N addition, it can be concluded that, in accordance with Gough et al. (2000), considerable variation characterizes the magnitude of the response. Only the biomass of grassland and salt marshes was significantly increased by N addition, whereas the biomass of forest understoreys, heathlands, freshwater wetlands and bogs did not show any significant response.

For an extensive discussion on the different mechanisms through which N deposition affects plant diversity, we refer to Bobbink et al. (2010). Their review consistently shows that N accumulation in the ecosystem is the main driver of changes to species composition across a range of ecosystem types. For acidic grasslands in the UK, Stevens et al. (2004) found a significant decrease of 1 species per 4 m2 quadrant for every 0.25 g m−2 year−1 of chronic N deposition, and thereby suggestedthat the long-term cumulative N deposition load might be the main cause of this species loss. Also, Duprèet al. (2010) found a significant relationship between several biodiversity measures and cumulative N deposition in re-surveys of semi-natural grasslands on nutrient-poor, acidic soils in Great Britain and Germany. In meta-analysis studies of N addition experiments across North America, Clark et al. (2007) and Suding et al. (2005) did not find a significant contribution from the N fertilization rate or duration of N addition to explain the high diversity in species richness responses following N fertilization. Together, this highlights the importance of cumulative rather than actual N deposition for species assembly in terrestrial vegetation communities.

The relationship between species richness and productivity often follows a unimodal curve (e.g. Grime, 1973). However, the meta-analysis of Gough et al. (2000) did not show species richness to follow productivity as predicted by this so-called ‘hump-shaped’ relationship. We could not test the link between productivity and species richness because only a limited number of studies report both variables. But from the studies reporting data on changes in biomass following N addition, we can conclude that many ecosystems did not show any significant response. This means that factors other than N availability (e.g. other nutrients, water, light) are limiting for plant growth (Aerts et al., 1992; Hurteau & North, 2008). Also, the number of plant species decreases with an increasing number of added limiting soil resources, which is predicted by the theory of niche dimensionality (Harpole & Tilman, 2007). In closed-canopied forest ecosystems, light is generally the primary limiting factor for understorey species (Gilliam, 2006), so biomass and species diversity are not affected by N addition (Gilliam, 2006). The question is, however, what will happen following the creation of canopy gaps that increase light availability?

In vegetation types other than forest, competition for light can be an important mechanism involved in biomass changes and plant diversity loss due to eutrophication. Using experimental grassland plant communities, Hautier et al. (2009) demonstrated that the addition of light to the understorey of grassland plant communities could prevent biodiversity loss caused by eutrophication. Grassland studies in both Europe (e.g. Bobbink, 1991; Stevens et al., 2006) and North America (e.g. Tilman, 1987) have highlighted strong reductions in species richness due to grass species that tend to outcompete forbs after N fertilization. An excess of bioavailable N reduces plant diversity by favouring competitive species that can respond quickly to increased N availability (Tilman, 1993; Stevens et al., 2006, 2009). The initial presence of certain species or functional groups probably modulates many responses to N addition. Forbs and grass biomass responded differently to N fertilization: graminoids clearly increased in biomass across studies, which is in accordance with other studies (e.g. Bobbink, 1991; Wedin & Tilman, 1996; Stevens et al., 2006; Duprèet al., 2010), whereas forbs did not show a significant response. Many grass species are probably able to produce a higher biomass and litter accumulation following N addition (Foster & Gross, 1998). Accordingly, the diversity of forbs and low-statured perennial grasses generally decreases through intense competition for light more than competition for soil resources (Hautier et al., 2009). Graminoids are also able to intercept proportionately more light compared with forbs (McLaren & Turkington, 2010). The same process of grass invasion is occurring in heathland ecosystems, with the biomass of graminoids showing a positive response to increased N availability, at least when canopy openings are being created by, for example, heather beetle attacks or winter injury (Bobbink et al., 2010). The biomass of ericaceous shrubs, however, remains unchanged or even decreases following N addition (e.g. Aerts et al., 1991; Power et al., 1998). Ericaceous shrubs have an N-conserving strategy because they have adapted to low-nutrient environments, whereas many graminoids are N-extravagants and thrive better in environments with faster nutrient turnover (Chapman et al., 2006). Furthermore, forb and shrub species richness has been proven to decline as a consequence of their sensitivity to high concentrations of NH4+ (Krupa, 2003; van den Berg et al., 2005; Kleijn et al., 2008), whereas certain graminoids seem to be able to cope with high NH4+ concentrations (e.g. van den Berg et al., 2005).

Heathland ecosystems, in contrast to grasslands, mainly occur in low-nutrient environments (Alonso et al., 2001), which might partly explain the differences in response between both vegetation types. In some cases (see Fig. 2), a negative response of biomass following N addition occurred, which can be caused by an increased abundance of herbivores through changed interactions between plants, parasitic fungi and herbivorous insects (Hatcher, 1995; Nordin et al., 2009). For example, Nordin et al. (1998) reported elevated concentrations of free amino acids in plant tissues of Vaccinium myrtillus following N addition, which resulted in increased damage by natural enemies (fungi and insects). Second, opposite responses of the cover and biomass of individual species or functional groups characterized by different nutrient utilization strategies can result in an unchanged or a decreased biomass (Gough et al., 2002) or species richness following N addition. In bogs, Sphagnum species, for instance, decrease in biomass as a consequence of an increased vascular plant cover (Heijmans et al., 2001) as a consequence of a high sensitivity to high NH4+ concentrations.

Although in some cases species density of a community responds differently following N addition, significant alterations in species composition can thus occur. Several studies show a decline in bryophyte biomass with an increase in vascular plant biomass (e.g. Virtanen et al., 2000; van der Wal et al., 2005), which indicates that bryophytes are limited by competition for space or light (Virtanen et al., 2000). The effects of reduced light arising from N pollution can be as important to mosses as direct toxicity from N deposition (van der Wal et al., 2005). Stevens et al. (2006) did not find a relationship between bryophyte abundance and N deposition, but individual bryophyte species showed species-specific responses to N. This was also concluded by van der Wal et al. (2005). According to the latter study, different sensitivities of mosses to both toxicity and shading effects of elevated N prevent generalization and can lead to replacement by competitive species within moss communities. Consequently, N addition can cause significant shifts in bryophyte species composition (Virtanen et al., 2000; Stevens et al., 2006). Also, Duprèet al. (2010) found a decline in bryophyte species richness in relation to cumulative N depositions.

Besides the direct effects of elevated NO3- or NH4+ concentrations in the soil solution, N fertilization with NH4NO3 or (NH4)2SO4 can cause soil acidification, a depletion of buffering base cations such as Ca2+, Mg2+ and K+ and an increase of Al3+ concentrations (Fig. 1 in Bowman et al., 2008). Some forb and shrub species seem to be sensitive to Al3+ (e.g. de Graaf et al., 1998), whereas certain graminoids growing on acidic soils appear to have a larger tolerance for low pH and high Al3+ levels than species found on less acidic soils (Göransson et al., 2009). Bryophyte species appear to be very sensitive to soil acidification as well, and according to Virtanen et al. (2000), soils with pH lower than 4.5 (aluminium buffer range, when aluminium becomes bio-available in high concentrations) are beyond the tolerance limits of most grassland bryophytes. According to Bobbink et al. (2010), soil acidification certainly plays a supporting role in species loss following N addition. Also, Stevens et al. (2006) points to soil acidification as being co-responsible for the strong decline in species richness along an N deposition gradient in the UK. Several other authors indicated that soil acidification may have played a significant role in the observed declines in species richness (e.g. Brunet et al., 1998; Carroll et al., 2003; Duprèet al., 2010), of course also caused by sulphur on top of N depositions. Besides the acidifying effect of N addition, additional soil acidification can be caused by sulphur depositions. This finding, however, does not exclude a possible simultaneous effect of soil acidification during the experiment. The information available in the original publications was insufficient to determine whether soil acidification occurred. For the majority of studies, we could only consider the initial soil pH range (before the experiment). Some studies controlled for this possible soil acidification by using CaNO3 as fertilizer (Henry et al., 2006) or adding crushed limestone (Wedin & Tilman, 1996). As shown by Bowman et al. (2008), N addition can cause soils to acidify and drop into another buffer range, having significant consequences for the soil nutrient status and plant-available cations. Clark et al. (2007) found soil CEC, to be a better predictor for the observed variation in species loss than soil pH. It remains a challenge for future research to unravel the relative importance of acidification and eutrophication for species loss following N input.

The critical load concept has been designed to establish deposition levels that ecosystems can tolerate without harmful effects on, for example, biodiversity (Nillson & Grennfelt, 1988). According to this idea, a terrestrial ecosystem for which the critical load value has been determined to 1 g m−2 year−1 will not change in biodiversity when the atmospheric N deposition does not exceed this threshold. According to our results, however, after 10 years of N deposition at a rate of 1 g N m−2 year−1, the species richness of this ecosystem will have decreased significantly. According to the review of Bobbink et al. (2010), critical load values derived from empirical N addition studies may be lower with increased duration of the treatment and may not represent the real biological threshold for cumulative effects of N deposition over several decades. Clark & Tilman (2008) found great species losses at low N addition rates, due to fast losses of rare species. Nordin et al. (2005) found that changes in key ecosystem components occurred even at lower rates of N input than the recommended empirical critical load values in a region with low background N deposition. Also, Bobbink et al. (2010) mention that even relatively low N deposition inputs can have a significant impact on long-term plant species biodiversity. These observations are in obvious accordance to our study. We found the effect size of species richness to decrease linearly with the cumulative N dose, which means, because the effect size is a logarithmic value, that species richness decline follows a negative exponential relationship with the cumulative N input. Consequently, species loss occurs faster at low levels of cumulative N input or in the beginning of the addition, followed by an increasingly slower species loss at higher cumulative N inputs. These findings lead us to stress the importance of including the cumulative effect of N depositions in the calculations of critical load values, e.g. by using dynamic models (Bobbink et al., 2010). Otherwise, the concept of critical load values might not be adequate enough to protect ecosystem biodiversity in the long run.

CONCLUSION

Although the scientific community has delivered manifold evidence for the negative impact of a N excess on species richness, our findings, which are based on N-addition experiments, are the first to show a significant decrease in species richness as a consequence of cumulative N additions across several vegetation types, soil types and climatic conditions. Species loss occurs faster at low levels of cumulative N input or in the beginning of the addition, followed by an increasingly slower species loss at higher cumulative N inputs. These findings lead us to stress the importance of including the cumulative effect of N depositions in the calculations of critical load values, e.g. by using dynamic models (Bobbink et al., 2010). A considerable proportion of the variation in the biomass response remains, however, unexplained. Also, Clark et al. (2007) did not find strong predictor variables for the biomass response. In conclusion, it is important to note that the effects of long-term atmospheric N deposition, when compared with relatively short-term experimental N additions, remain to be quantified and deserve continuing interest in the field of ecology (Clark et al., 2007; Lovett et al., 2009).

ACKNOWLEDGEMENTS

We thank M. Davis, F. Gilliam, U. Gunnarsson, K. Gross, H. Henry, M. Hurteau, E. Lamb, J. Laundre, D. Milchunas, A. Nordin, H. Persson, S. Rainey, G. Shaver, D. Tilman, S. Wilson and J. Zedler for providing soil pH data or additional information. This paper was written while A.D.S. and P.D.F. held a post-doctoral and PhD fellowship, respectively, from the Research Foundation Flanders (FWO). E.A. and A.D. held a PhD scholarship from the Institute for the Promotion of Innovation through Science and Technology. K.W. and L.V.N. were paid by Ghent University with a post-doctoral fellowship (BOF) and as a teaching assistant, respectively. We thank the Research Institute for Nature and Forest (INBO) of the Flemish Government for the financial support, Maarten Hens and Martin Diekmann for valuable comments.

BIOSKETCH

The authors as a team have a broad knowledge on the topic of plant population and community dynamics in relation to various global change drivers, including land-use change, eutrophication and climate change. A.D.S., P.D.F. and K.V. conceived the initial ideas, all authors collected the data and A.D.S. and P.D.F. analysed the data and led the writing.

Editor: Martin Sykes

Ancillary