Driving factors behind the eutrophication signal in understorey plant communities of deciduous temperate forests

Authors


Correspondence author. E-mail: kris.verheyen@ugent.be

Summary

1. Atmospheric nitrogen (N) deposition is expected to change forest understorey plant community composition and diversity, but results of experimental addition studies and observational studies are not yet conclusive. A shortcoming of observational studies, which are generally based on resurveys or sampling along large deposition gradients, is the occurrence of temporal or spatial confounding factors.

2. We were able to assess the contribution of N deposition versus other ecological drivers on forest understorey plant communities by combining a temporal and spatial approach. Data from 1205 (semi-)permanent vegetation plots taken from 23 rigorously selected understorey resurvey studies along a large deposition gradient across deciduous temperate forest in Europe were compiled and related to various local and regional driving factors, including the rate of atmospheric N deposition, the change in large herbivore densities and the change in canopy cover and composition.

3. Although no directional change in species richness occurred, there was considerable floristic turnover in the understorey plant community and a shift in species composition towards more shade-tolerant and nutrient-demanding species. However, atmospheric N deposition was not important in explaining the observed eutrophication signal. This signal seemed mainly related to a shift towards a denser canopy cover and a changed canopy species composition with a higher share of species with more easily decomposed litter.

4.Synthesis. Our multi-site approach clearly demonstrates that one should be cautious when drawing conclusions about the impact of atmospheric N deposition based on the interpretation of plant community shifts in single sites or regions due to other, concurrent, ecological changes. Even though the effects of chronically increased N deposition on the forest plant communities are apparently obscured by the effects of canopy changes, the accumulated N might still have a significant impact. However, more research is needed to assess whether this N time bomb will indeed explode when canopies will open up again.

Introduction

Atmospheric nitrogen (N) deposition rates are markedly exceeding their historical background levels in industrialized regions of the world, and deposition rates will probably continue to rise in the 21st century (Dentener et al. 2006). Reductions in plant diversity and shifts in species composition through increased N deposition in ecosystems around the globe are common (Bobbink et al. 2010). For example, the effects of N enrichment on plant diversity in temperate grasslands have been well studied. Both experimental N addition studies (synthesized in, for example, Clark et al. 2007; De Schrijver et al. 2011) and observational studies along large deposition gradients (Stevens et al. 2004, 2010; Dupréet al. 2010; Maskell et al. 2010) indicate negative relationships between (cumulative) N addition and plant species richness. Therefore, N deposition would also be expected to impair forest plant diversity. Understorey plant communities support the majority of the plant diversity in temperate forests (Gilliam 2007). Moreover, levels of N deposition received by the understorey may be considerably higher compared with other vegetation types due to a higher aerodynamic roughness and intercepting surface of forest canopies (Erisman & Draaijers 2003). Yet, the effects of experimental N additions on forest understoreys seem less consistent compared with N additions in grassland (Gilliam 2006; Bobbink et al. 2010; De Schrijver et al. 2011). A recent meta-analysis of N addition experiments by De Schrijver et al. (2011) reported, for instance, a tendency towards decreasing biomass in the understorey and no significant effect of N addition on understorey plant species richness.

By contrast, many observational studies reported shifts in the understorey species diversity and composition and attributed those shifts to increased N deposition rates. In contrast to the N addition experiments on forest understoreys, which were mostly performed in North America, most of the observational studies were performed in Europe (cf. Gilliam 2006). For observational studies, two approaches have been used: (i) resurveys of permanent or semi-permanent plots (e.g. Thimonier, Dupouey & Timbal 1992; Thimonier et al. 1994; Lameire, Hermy & Honnay 2000; Kirby et al. 2005; Bernhardt-Römermann et al. 2007; Van Calster et al. 2007, 2008a) or (ii) changes in vegetation composition along large deposition gradients (e.g. Tyler 1987; Brunet, Diekmann & Falkengren-Grerup 1998; van Dobben & de Vries 2010). A positive relationship between the increasing frequencies and abundances of nitrophilous species and (assumed) increased N availability was generally found. Yet, both approaches have shortcomings due to the possible occurrence of temporal or spatial confounding factors (Diekmann et al. 1999). Studies along large deposition gradients may include substantial differences in soil, climate and species pools between the study sites, making it difficult to isolate the effects of N deposition. Resurvey studies, on the other hand, are generally performed in single forests or landscapes, but the increased N deposition levels between the two survey dates often parallel other ecological changes that have taken place during the last decades (e.g. Hopkins & Kirby 2007). Many ancient, deciduous forests in lowland Europe have been managed as coppice or coppice with standards for many decades, if not centuries (e.g. Peterken 1993; Rackham 2003; Szabó 2010). This silvicultural system has now been abandoned or was replaced by a high forest management system in most regions resulting in important changes in the canopy structure and composition, which may have a significant impact on the understorey plant communities (e.g. Van Calster et al. 2008a). Densities of large herbivores (including roe deer –Capreolus capreolus, red deer –Cervus elaphus and fallow deer –Dama dama) and wild boar (Sus scrofa) have increased during recent decades in many regions across north-western Europe (Fuller & Gill 2001; Ward 2005; Milner et al. 2006; Bláha & Kotecký 2008). This increase is explained by land-use changes, milder winters and changes in game management. Rising herbivore populations have a large impact on the composition of the forest understorey (e.g. Welander 2000; Kirby 2001; Rooney & Waller 2003; Rooney 2009; Royo et al. 2010). These concurrent ecological changes make it inherently difficult to isolate N deposition from other drivers of forest vegetation change (e.g. Dzwonko & Gawronski 2002; Hofmeister et al. 2009).

A combination of a temporal and spatial approach allows the assessment of the relative contribution of N deposition compared with other ecological changes on forest understorey plant communities. Dupréet al. (2010) recently demonstrated the usefulness of a similar spatiotemporal approach to assess deposition effects in acidic grasslands. Diekmann et al. (1999) and Köchy & Bråkenhielm (2008) had previously used this approach in forest ecosystems, but included only a small number of regions (2) or plots (9), respectively, in their analyses. In the present study, data from 1205 (semi-) permanent vegetation plots taken from 23 rigorously selected understorey resurvey studies along a large deposition gradient across the temperate zone of Europe were used to assess (i) whether species richness and vegetation composition have changed during the period between the surveys, (ii) whether changes in richness and composition were larger in regions with higher N deposition rates and (iii) the relative importance of other ecological factors, notably canopy structure, canopy composition and grazing pressure, compared to N deposition.

Material and methods

Study Sites

The sample sites are all described as ancient, semi-natural deciduous temperate forest in Europe (cf. Peterken 1993; Hermy et al. 1999). Temperate zones on other continents were not considered because Ellenberg indicator values (Ellenberg et al. 1992), which are important for the indirect assessment of changes in environmental conditions, are not available. All records were from sites in which no stand-replacing management actions (e.g. clear-cuttings followed by replanting with conifers) have taken place since the date of the first survey. We looked for studies with data for at least c. 20 permanent or semi-permanent plots. These plots had to be independent (e.g. no subplots of a single larger plot), and the interval between the first and the last survey had to be at least c. 20 years. This large time interval is needed to account for the long life span of many (understorey) forest species (e.g. Ehrlén & Lethilä 2002). Plot-level presence/absence data of all species in the understorey layer (here defined as all vascular plant species <1 m) for both survey dates were available in all cases. Where possible, plot-level cover data for the shrub and tree layers were included as well.

Potentially suitable studies were found using Web of Science (http://www.isiknowledge.com) and by contacting researchers through the FLEUR-network (a European network of forest understorey researchers; http://www.fleur.ugent.be) in the different regions of the temperate forest zone. Data from 23 studies and eight countries were obtained, ranging from the United Kingdom to the Czech Republic and from Switzerland to mid Sweden (Fig. 1; Table 1). The soil types covered by the studies ranged from relatively poor, sandy soils (Zoerselbos, Be; Speulderbos, Nl) to rich, clay soils (Dalby, Se; Göttingen, Ge) and deep calcareous rendzinas or luvisols (Děvín and Milovice Wood, CZ). However, most study sites were located on moderately rich, loamy soils. We cannot independently assess the representativeness of our samples, but we consider that a large part of the potential variation has been covered because of the geographic and edaphic distribution spread of the samples.

Figure 1.

 Map showing the location of the 23 regions included in this study (the numbers refer to Table 1).

Table 1.   Ecological details of the 23 studies included in this meta-analysis. The ID of each study refers to Fig. 1
IDAuthor(s)Study region, CountryLatLongMAT*MAP*Plot size (range)Number of plotsSurvey year(s)Grazing*Atmospheric deposition*Management†
°N°E°Cmmm2OldRecentDensity (no. 100 ha−1)ChangeNmean (kg ha−1 year−1)OldRecent
  1. MAP, mean annual precipitation; MAT, mean annual temperature.

  2. *See text for a detailed description.

  3. †Management classes were defined as follows: 1: no management, 2: low intensity cuttings (i.e. removal of a small fraction of canopy trees) at a low frequency (i.e. <1× per 10 year) and 3: high intensity cuttings (i.e. removal of a significant fraction of canopy trees) at a higher frequency (i.e. more than 1× per 10 year).

  4. ‡Studies for which no canopy data were available.

 1T. Vandenbroeck, unpublished dataGaume, Be49.65.58.2852100431950s200816Stable17.032
 2 Baeten et al. (2010) Binnen-Vlaanderen, Be51.04.59.7798150471977–198020090Stable22.132
 3S. De Smet, unpublished dataZoerselbos, Be51.24.79.779810017198220088Stable24.211
 4‡ Cornelis, Rombouts & Hermy (2007) Herenbossen, Be51.14.89.7798196111198020048Stable21.932
 5‡ Lameire, Hermy & Honnay (2000) Vorte Bossen, Be51.13.49.7798150261977–198019980Stable22.321 and 2
 6 Baeten et al. (2009) Meerdaalwoud, Be50.84.79.7798125–225211954200018Stable18.332
 7‡ Van Calster et al. (2008a) Florenne, Be50.34.69.779810058195720058Stable19.722
 8 Van Calster et al. (2008a) Tournibus, Be50.34.69.7798100139196720058Stable20.922
 9 von Oheimb & Brunet (2007) Dalby, Se55.713.37.96521 (16 for canopy)741935200215Increase8.511
10‡O. Eriksson, unpublished dataTullgarn, Se58.117.16.85091001271971200314Increase8.322
11 Naaf & Wulf 2010 Elbe-Weser, Ge53.69.08.3761100–400501986–198920087Increase24.922
12R. Hédl, unpublished dataDěvín, CZ48.916.68.6490100–1000501953–19632002–200312Decrease14.32 and 31 and 2
13 Hédl, Kopecký & Komárek (2010) Milovice Wood, CZ48.816.78.6490500461953–195420062Increase13.32 and 31 and 2
14 Hédl (2004) Rychlebské hory Mts., CZ50.317.17.2976315211941–19431998–19991Increase13.021 and 2
15 Kirby & Morecroft (2010) Wytham Woods, UK51.8−1.39.96311004919741999100Increase14.511
16 Dierschke (2009) Göttingen, Ge51.510.18.5643250 (100–400)42198020010Stable18.811
17 Petřík (2009) Milíčovský les, CZ50.014.58.6516240 (50–625)1919862008–200910Stable13.521 and 2
18 Walther & Grundmann (2001) Switzerland, CH47.37.89.4782100–400371940–1965199818Stable17.822
19G. Decocq, unpublished dataHirson/Saint-Michel, Fr49.94.110.2869500–800221956–19651996–199818Increase18.922
20G. Decocq, unpublished dataAndigny, Fr50.03.69.9685500–800191957–19631995–199620Increase21.222
21P. Hommel, unpublished dataSpeulderbos, Nl52.35.79.3820100–250271957–19591987–19889Increase35.721
22‡K. Vanhuyse, unpublished dataLady Park, UK51.7−2.79.371932351979200910Decrease14.411
23 Bernhardt-Römermann et al. (2007) Münich, Ge48.311.77.57931001251986200310Decrease21.511

The first surveys were carried out between 1935 and 1986/89 and the recent surveys between 1987/88 and 2009. The time interval between the two surveys ranged between 17 and 67 years. The forests were either not managed (seven studies), only extensively managed (11 studies), or a mixture of both (five studies) at the time of the most recent survey (Table 1). Management frequency and intensity decreased since the time of the first survey in 10 study regions. The number of plot pairs per study ranged between 17 and 139, with an average of 52 per study. In total, understorey plant community data from 1205 plot pairs were included in the data base.

Calculation of the Response Variables

Nomenclature was standardized based on Ellenberg et al. (1992), and understorey data were transformed to presence/absence records to standardize the recording scale among studies. Next, three plant community descriptors were derived for both the plots in the old (o) and recent (r) surveys: the species richness (So, Sr) and the mean Ellenberg indicator values for light availability (mLo, mLr) and soil nitrogen availability (mNo, mNr) based on presence/absence data. In the absence of actual measurements of environmental variables, the use of Ellenberg indicator values to document environmental preferences and changes in environmental conditions is an acceptable, widely used, alternative (Diekmann 2003), especially when used within a single vegetation type (Wamelink et al. 2002) such as (ancient) forest (Dzwonko 2001). Considering the ordinal nature of the indicator values, calculation of mean indicator values is strictly speaking not fully appropriate, but the vast majority of plant ecologists use calculated means as they work very well (Diekmann 2003). Potential time-lags in the response of the vegetation due to changes in environmental conditions were accounted for by selecting only studies with a sufficiently long time interval (>17 years) between the two surveys. Ellenberg indicator values for light availability range from 1 (species can grow in very deep shade and rarely occurs in more open conditions) to 9 (species only occurs in open conditions). Soil nitrogen availability values range from 1 (species occurs on sites with very low N availability) to 9 (species only occurs on sites with very high N availability). It should be noted that the Ellenberg N values indicate more than N availability alone and reflect general nutrient availability (Schaffers & Sýkora 2000; Diekmann 2003; Ellenberg & Leuschner 2010). Hence, in the remainder of the text, we will denote Ellenberg N as soil nutrient availability and increasing mN values will be referred to as eutrophication. Indicator values for soil reaction (mR) were not used due to the strong, positive correlation with the mN values (rso = 0.71, < 0.001; rsr = 0.78, < 0.001; n = 1201 and with rs = Spearman rank correlation for the old and new survey, respectively). Hence, we cannot fully disentangle the effects of eutrophication and acidification. To determine the change in species richness and Ellenberg indicator values, response ratios were calculated according to Hedges, Gurevitch & Curtis (1999) as ln(Xr/Xo), with X being one of the three response variables. These response ratios are further denoted as RRS, RRL and RRN. Response ratio means per study and across all studies were calculated according to the weighting proposed in Hedges, Gurevitch & Curtis (1999). A fourth response variable was the plot-level floristic turnover. We calculated the Lennon dissimilarity (Lennon et al. 2001) in floristic composition between a plot in the first survey and the same plot in the second survey as min(b, c)/[min(b, c)+a], with a representing the number of species shared by both plots; b the number of species that occur in the plot only during the first survey and c the number of species that occur in the plot only during the second survey. The dissimilarity was modified to a similarity measure: similarity = 1 – Lennon index. This simple presence/absence-based index is less sensitive to differences in species richness between the plots than the commonly used Jaccard index (Koleff, Gaston & Lennon 2003) and is therefore more appropriate to determine real floristic turnover, which was appropriate for the purposes of the current study.

Calculation of the Explanatory Variables

The variables used to explain the changes in the understorey plant community were the average rate of atmospheric N deposition, the climate in the study region, the actual density of large herbivores and the change therein and the change in canopy cover and composition which may reflect changes in management.

The rate of N deposition was quantified using the EMEP data base (http://www.emep.int), which allows deposition data for the whole of Europe to be derived with a resolution of 50 km × 50 km. We extracted wet and dry deposition data of reduced and oxidized N and for the year 2000 (N2000, expressed in kg ha−1 year−1). This year was chosen as it represents the average of the interval in which the recent surveys were performed. De Schrijver et al. (2011) recently showed that the modelled EMEP data and locally observed N deposition data are very well correlated. Significant underestimations only occur at sites where nearby point sources (e.g. large animal husbandry farms) are present. However, the throughfall deposition on the forest floor will likely be between 1.5× and 2× higher than the open field N deposition due to the high aerodynamic roughness of forest canopies (ICP 2005).

To obtain a mean N deposition rate over the period between the two surveys (Nmean), we accounted for the variation in deposition rates over time by calculating the cumulative deposition between the two survey years (Ncum) using correction factors for the different decades, based on the year 2000 deposition rates (see Dupréet al. (2010) for more information on the correction factors and the calculation methods). Then, the Ncum was divided by the time interval between the two surveys. Nmean ranges between 8.3 and 35.7 kg ha−1 year−1 (Table 1). Sulphur (S) deposition also contributes to the potential acidifying deposition rate, but this rate (expressed in keq ha−1 year−1 and calculated as: N2000/14+ (S2000/32.06)*2) was very strongly correlated (rs = 0.93, < 0.001, n = 23) to the N2000 deposition values. Hence, the sulphur deposition variable was not included in the analysis.

Climate may influence the rate and nature of vegetation changes both directly (e.g. through its influence on germination and growth rates) and indirectly (e.g. through its influence on biogeochemical cycling). Therefore, we derived the mean annual temperature (MAT) and precipitation (MAP) for the period 1961–1990 for each of the study sites using the program NewLocLim v1.10 (FAO 2005; Table 1).

Local expert knowledge was used to estimate the present density of the three most common large herbivores in Europe (i.e. numbers of roe deer, fallow deer and red deer per 100 ha) in each study area and to indicate whether these numbers have increased (nine studies), decreased (three studies) or remained stable (11 studies) in the period between the two surveys (see also Table 1). To account for differences in the density estimates, densities were ln(x + 1) transformed. The trend variable was recoded into two dummy variables (HERBI−, HERBI+).

The change in the cover and composition of the canopy (including both the shrub and tree layers) was quantified using three variables: the change in the total cover of the canopy, the change in the shade casting ability of the canopy species and the change in the litter quality. Canopy data were available for all but five studies (numbers 4, 5, 7, 10 and 22 in Table 1). The total cover of the shrub and tree layers was calculated as the sum of the cover percentages of all species occurring in these layers. The change in cover was calculated using the response ratio: ln(Coverr/Covero), further denoted as RRcover. The shade casting ability and the litter quality were calculated as a cover weighted average of, respectively, the shade casting ability and litter quality index scores listed in Appendix A (see also Van Calster et al. 2008a; Baeten et al. 2009). The scores range between 1 (very low shade casting ability and very low decomposition rate) and 5 (very high shade casting ability and very high decomposition rate). Index scores were not available for all species, and so plot values were only used when >70% of the total cover was comprised by species with a known score, resulting in 787 plots that could be used for further analysis. Finally, response ratios for shade casting ability (RRshade) and litter quality (RRlitter) were calculated to determine the change in these variables. The three canopy response ratios were not correlated (rs RRlitter– RRcover = 0.08, rs RRlitter– RRshade = 0.03, rs RRshade– RRcover = −0.02; n = 787). Response ratio means per study and across all studies were calculated according to Hedges, Gurevitch & Curtis (1999).

Statistical Analysis

Linear mixed models were used to relate each of the four understorey response variables to the explanatory variables at the study level (number of years between surveys, Nmean, MAT, MAP, actual grazing pressure and trend in grazing pressure) and at the individual plot level (RRcover, RRshade, RRlitter and the initial Ellenberg values mLo and mNo). The initial Ellenberg values mLo and mNo were not included as explanatory variables for the responses RRL and RRN as they form the denominator of the respective RRs. The modelling was carried out using the lme function in the nlme library in R 2.10.1 (Pinheiro et al. 2009; R Development Core Team 2009).

We adopted an information-theoretic approach of data modelling in which sets of models are compared in a symmetric way, which avoids problems associated with multiple pairwise testing (cf. Bolker 2008; Bolker et al. 2009). Here, we used the Akaike Information Criterion (AIC) to compare the competing models. First, we related a response variable to each of the explanatory variables separately using mixed models with the intercepts varying randomly by study, that is, ‘Study ID’ as random effects term. A null model with the intercepts varying by study, but with no explanatory variables, was also calculated. The ΔAIC of a model was then calculated as the difference in AIC value for that model and the model with the lowest AIC value (best fit to the data). Models with ΔAIC > 4 may be considered clearly distinguishable, while all models with ΔAIC < 4 may be kept under consideration (Bolker 2008). Then, we constructed a second set of models with all possible combinations of the explanatory variables that proved to be equivalent in the previous modelling step (here ΔAIC < 4). For instance, if explanatory variables a, b and c were retained, all combinations were a+b+c, a+b, a+c, b+c, a, b, c. The ΔAICs for this set of models will be reported. Finally, the parameter values of the model that showed the best fit (in terms of AIC value) were re-estimated with restricted maximum likelihood estimation and reported.

Results

Across all studies, the species richness did not change over time, but significantly increased in eight studies and decreased in eight others (Fig. 2a). The mean (exp-transformed) response ratio for species richness was 0.97 (95% confidence interval: 0.87–1.09). The mean species number across all plots in both the old and recent surveys was c. 17. The mean Lennon similarity was 0.69 (Fig. 2b), which implies that on average one-third of the species in each plot pair has been replaced. An overview of the fifty most frequent species and their average change in frequency is given in Table 2. The species decreasing in frequency were mostly herb species, whereas ferns and seedlings of tree and shrub species increased. The Ellenberg indicator value for nutrient availability significantly increased in six studies and decreased in two studies (Fig. 2c), and across all studies, a significant increase was found (RRN: 1.03; 95% CI: 1.01–1.05). The Ellenberg indicator value for light availability significantly decreased in four and increased in three studies (Fig. 2d) and exhibited a (marginally significant) decrease across all studies (RRL: 0.99; 95% CI: 0.97–1.01). RRL and RRN were negatively correlated (rs = −0.15, < 0.001, n = 1201), suggesting that a decreasing proportion of more light-demanding species in the understorey plant community goes along with an increasing proportion of more nutrient-demanding species.

Figure 2.

 Species richness, soil nutrient and light availability, and community shifts for the 23 studies included in this study. Mean (±95% confidence interval) response ratios are given based on the plot values in the old and recent surveys for understorey species richness (RRS) (a), and the mean Ellenberg values for nutrient availability (RRN) (c) and light (RRL) (d). Panel b depicts the mean Lennon similarity coefficients between the understorey composition in the old and recent survey. Due to the bounded nature (between 0 and 1) of the Lennon similarity coefficients, the 95% confidence intervals were based on 2000 bootstrap resamples within each study. The studies are ranked according to increasing mean atmospheric N deposition, and a * indicates studies for which no canopy data were available.

Table 2.   Overview of the fifty most frequent understorey species with their average study-level frequency in the recent surveys and the change in frequency compared to the old surveys. The species are ranked according to increasing change in frequency; tree and shrub species that occurred as seedlings in the understorey are marked with (TS), ferns are marked with (F)
SpeciesAverage frequency (%) in recent surveyChange in frequency (%) compared to old surveySpeciesAverage frequency (%) in recent surveyChange in frequency (%) compared to old survey
Ajuga reptans 10−5.8 Stachys sylvatica 15+2.0
Poa nemoralis 17−5.4 Urtica dioica 26+2.0
Mercurialis perennis 24−4.6 Melica uniflora 15+2.0
Convallaria majalis 17−3.2 Carpinus betulus (TS) 13+2.1
Paris quadrifolia 14−3.2 Glechoma hederacea 17+2.3
Primula elatior 12−2.7 Dryopteris filix mas (F) 24+2.4
Ranunculus ficaria 17−2.4 Arum maculatum 17+2.7
Lonicera periclymenum 22−2.2 Circaea lutetiana 17+2.8
Asarum europaeum 11−2.0 Stellaria holostea 13+2.9
Anemone nemorosa 39−1.8 Geum urbanum 30+3.1
Ranunculus auricomus 10−1.0 Athyrium filix-femina (F) 25+3.8
Polygonatum multiflorum 31−0.8 Sorbus aucuparia (TS) 17+3.9
Viola reichenbachiana 14−0.5 Coryllus avellena (TS) 16+4.2
Geranium robertianum 10−0.3 Brachypodium sylvaticum 18+4.8
Pteridium aquilinum (F) 19+0.2 Quercus robur (TS) 15+5.1
Adoxa moschatellina 11+0.3 Poa trivialis 16+5.3
Maianthemum bifolium 13+0.3 Oxalis acetosella 30+5.5
Scilla non-scripta 13+0.6 Rubus fruticosus coll.54+6.4
Carex sylvatica 23+0.8 Galeopsis tetrahit 12+7.1
Moehringia trinervia 13+0.9 Acer pseudoplatanus (TS) 21+7.3
Galium odoratum 20+1.0 Hedera helix 31+7.4
Milium effusum 23+1.1 Fagus sylvatica (TS) 30+8.4
Lamiastrum galeobdolon 30+1.2 Fraxinus excelsior (TS) 27+9.8
Galium aparine 12+1.4 Dryopteris carthusiana (F) 24+10.7
Deschampsia cespitosa 29+1.9 Dryopteris dilatata (F) 17+12.6

The mean (exp-transformed) RRcover across studies was 1.05 (95% CI: 0.95–1.16) and the canopy cover increased in nine and decreased in five studies (Fig. 3a). The RRshade and RRlitter exhibited a significant (1.04; 95% CI: 1.01–1.07) and marginally significant (1.03; 95% CI: 0.99–1.07) increase, respectively. Both the shade casting ability of the canopy and the litter quality index significantly increased in three studies and decreased in one (Fig. 3b,c). Scatterplots (not shown) of the values of the three canopy variables in the old surveys with their respective response ratios indicated that the largest increases took place in plots with low values in the old surveys. An overview of the ten most frequent tree and shrub species in the recent surveys and their changes in frequency and cover is given in Table 3. It appears that the increasing importance of shade casting species and/or species with a better litter quality is mainly due to the increases of Acer pseudoplatanus, Carpinus betulus and Fraxinus excelsior.

Figure 3.

 Canopy cover, shade casting ability and litter quality changes for the 18 studies for which canopy data were available. Mean (±95% confidence interval) response ratios (RRs) are given based on the plot values in the old and recent surveys for canopy cover (RRcover) (a), the shade casting ability of the tree and shrub species in the canopy (RRshade) (b) and the quality of the litter shed by the species in the canopy (RRlitter) (c). The studies are ranked according to increasing mean atmospheric N deposition.

Table 3.   Overview of the ten most frequent tree and shrub species (across the different studies) in the recent surveys, their average cover in the plots where they occurred and the trends in frequency and cover compared to the old surveys. The species are ranked according to decreasing frequency
SpeciesFrequency (%) in recent surveyChange in frequency (%) compared to old surveyCover (%) in recent surveyChange in cover (%) compared to old survey
Quercus robur/petraea53+438−2
Fraxinus excelsior 46+344+4
Coryllus avellana 44036−2
Acer pseudoplatanus 43+738+11
Carpinus betulus 31−138+11
Fagus sylvatica 27+352−1
Betula pendula/pubescens 20−218+5
Sorbus aucuparia 19−170
Ulmus glabra 17−140+1
Crataegus spp.16−2130

The results of the null models indicate that variation of the change in species richness (RRS) is more or less equally distributed at the study (41%) and plot level (59%). The variation partitioning in the Lennon similarity coefficients is comparable: 36% variation at the study level and 64% at the plot level. The linear mixed models with RRS and the Lennon similarity coefficients retained the same set of explanatory variables (Table 4), and also the best-fitting model was similar. The ΔAICs between the best-fitting models and the null models were 15.4 and 16.1 for RRS and the Lennon coefficient, respectively, which indicates that the explanatory variables do explain a significant part of the variation. The RRS and the Lennon similarity coefficient decreased with an increasing number of years between the two surveys, that is, plots in studies with longer time intervals between surveys lost more species and exhibited higher turnover. Species richness decreased most, and turnover was highest in plots where the light availability at the time of the first survey was relatively high. The mean N deposition rate exhibited a (weak) positive relationship with RRS and the Lennon similarity coefficient.

Table 4.   Outcome of the general linear mixed models with the response ratio of species richness (RRs) and the Lennon similarity coefficients between the old and recent surveys as response variables and the deposition (Nmean), number of years (no. years), initial Ellenberg indicator values (mLO, mNO), climate (MAT, MAP), grazing (density, HERBI−, HERBI+) and canopy variables (RRcover, RRlitter, RRshade) as explanatory variables. Each combination of the individual explanatory variables that proved to be equivalent in terms of explanatory power when used in single-variable models (i.e. ΔAIC < 4 than the model with the lowest AIC) is reported
RRSLennon
Variable(s)ΔAICd.f.Variable(s)ΔAICd.f.
  1. AIC, Akaike Information Criterion; MAP, mean annual precipitation; MAT, mean annual temperature.

  2. The parameter values of the model that showed the best fit (in terms of AIC value) are shown at the bottom of the table. See Materials and methods for details on the stepwise model building.

No. years + Nmean + mLo0.06mLo + Nmean + No. years0.06
No. years + mLo1.35mLo + Nmean0.95
Nmean + mLo3.05mLo + No. years2.05
No. years + Nmean5.75mLo7.84
No. years6.74Nmean + No. years11.35
Nmean8.94Nmean11.54
mLo9.84No. years11.94
VariableValueSEd.f. t-value P-valueVariableValueSEd.f. t-value P-value
Intercept0.3780.4187650.9040.367Intercept−0.1820.098765−1.8480.065
No. years−0.0120.00515−2.1660.047No. years−0.0020.00115−1.6010.130
Nmean0.0230.013151.7380.103Nmean0.0060.003151.9190.074
mLo−0.0810.028765−2.8430.005mLo−0.0270.007765−3.616<0.001

The variation of RRL and, to a lesser extent, of RRN largely occurred at the plot level (91% and 68%, respectively). For both response variables, only the canopy change variables were retained (Table 5). The ΔAICs between the best-fitting models and the null models were 12.6 and 11.5 for RRL and RRN, respectively. RRN increased with increasing canopy cover and increasing quality of the litter. RRL also increased with increasing quality of the litter and decreased with increasing canopy cover and increasing shade casting abilities of the canopy species.

Table 5.   Outcome of the general linear mixed models with the response ratio of Ellenberg values for nutrient availability (RRN) and light (RRL) as response variables and the deposition (Nmean), number of years (no. years), initial Ellenberg indicator values (mLO, mNO), climate (MAT, MAP), grazing (density, HERBI−, HERBI+) and canopy variables (RRcover, RRlitter, RRshade) as explanatory variables. Each combination of the individual explanatory variables that proved to be equivalent in terms of explanatory power when used in single-variable models (i.e. ΔAIC < 4 than the model with the lowest AIC) is reported
RRNRRL
Variable(s)ΔAICd.f.Variable(s)ΔAICd.f.
  1. AIC, Akaike Information Criterion; MAP, mean annual precipitation; MAT, mean annual temperature.

  2. The parameter values of the model that showed the best fit (in terms of AIC value) are shown at the bottom of the table. See Materials and methods for details on the stepwise model building.

RRlitter + RRcover0.05RRcover +  RRlitter +  RRshade0.06
RRlitter3.54RRcover + RRlitter1.55
RRcover6.04RRcover + RRshade6.15
   RRcover6.84
    RRlitter + RRshade7.25
   RRlitter9.44
   RRshade11.24
VariableValueSEd.f. t-value P-valueVariableValueSEd.f. t-value P-value
Intercept0.0400.0177642.3000.022Intercept−0.02160.010763−2.1800.030
RRlitter0.0370.0137642.8290.005RRlitter0.04180.0157632.8410.005
RRcover0.0200.0087642.3500.019RRcover−0.02790.009763−3.0370.003
      RRshade−0.0480.026763−1.8780.061

Discussion

During the last decades, large changes in the understorey vegetation of the studied ancient, semi-natural deciduous woodlands have taken place. Although no directional change in species richness occurred, there was considerable floristic turnover and species composition shifted towards more shade-tolerant and nutrient-demanding species. In contrast to the expectations, atmospheric N deposition was not important in explaining the observed eutrophication signal. This signal seems mainly caused by a shift towards a denser canopy cover and a changed canopy species composition with a higher share of species with more easily decomposed litter.

Below, we first discuss the ecological changes that have taken place in the studied forests during recent decades and elaborate the way in which these ecological changes relate to the observed shifts in the understorey plant communities. We end by interpreting our results in terms of a model recently developed by Smith, Knapp & Collins (2009), which presents ecological change as a response to chronic resource alterations.

Ecological Changes in Ancient, Semi-Natural Deciduous Forests in Europe

The range of open field N deposition rates (between 8.3 and 35.7 kg ha−1 year−1) included in this study is very similar to the range in N deposition rate across temperate forest in Europe (Holland et al. 2005). Bobbink et al. (2010) state that there is evidence for N deposition effects on understorey biodiversity in temperate forests at deposition rates <20 kg ha−1 year−1 and perhaps even as low as 10–15 kg ha−1 year−1. As the N deposition on the forest floor will probably be one-and-a-half to two times higher than the cited open field values (ICP 2005), the critical load value has been exceeded in many of the study regions. Therefore, large N-driven changes in the understorey plant community were expected.

The decrease in management intensity in our study sites since the time of the first surveys is mainly the result of two factors: (i) many of the ancient, semi-natural deciduous forests have been given a more protected status (e.g. under the EU Habitat Directive) during the last decades because of their conservation value; (ii) more importantly, the coppice or coppice with standards management system, which was very common in much of Europe, has largely been abandoned or replaced by high forest systems characterized by much longer rotations due to a changing socio-economic context (Kirby & Watkins 1998; Szabó 2010). The decreased management intensity, especially in semi-natural, deciduous forests, is probably a general trend across north-western and central Europe (e.g. Hopkins & Kirby 2007; Ellenberg & Leuschner 2010; Hédl, Kopecký & Komárek 2010). This trend is also exemplified by increasing stocks of wood in European forests; for example in Western European forests, standing volumes per ha have doubled since 1950 (Gold, Korotkovb & Sasse 2006).

In our study, the decreasing management intensity is reflected in the increase in total canopy cover in most of the study regions and an increasing importance of more shade casting, late successional species such as Acer pseudoplatanus and C. betulus. Shrubs or small trees such as Corylus avellana, Sorbus aucuparia and Crataegus spp. tended to decrease (Table 3). Similar trends have been reported before in some of the study regions included in this paper (e.g. Meerdaalwoud: Baeten et al. 2009; Milovice: Hédl, Kopecký & Komárek 2010; Wytham: Kirby et al. 2005; Tournibus & Florennes: Van Calster et al. 2008a). Next to a higher litter input due to the increased canopy cover and change in management system, the litter generally also became more decomposable over the years due to the increasing importance of species with good litter quality such as Fraxinus excelsior, A. pseudoplatanus and C. betulus (cf. Jacob et al. 2009) in the studied forests. This has also been reported earlier for some of the study regions included (e.g. Dalby: Persson, Malmer & Wallén 1987). Furthermore, active litter removal was also a common practice in the past (e.g. Kirby & Watkins 1998).

The unexpected significant decrease in canopy cover in some study regions, where no management has taken place (Dalby), where management intensity has decreased (Rychlebské hory, Milíčovský les) or remained stable (Elbe-Weser), is related to a series of natural disturbances: canopy tree mortality due to diseases (e.g. tracheomycosis of oaks in Milíčovský les, Dutch elm disease in Dalby), storm damage (Elbe Weser), air pollution-related damage to beech (Rychlebské hory) and mortality due to old age (Elbe-Weser, Dalby and Milíčovský les). It is also likely that the canopy will become more open in other regions as in the next decades, more and more forests are gradually moving towards a canopy breakup stage (sensuPeterken 1996).

The trend towards increasing large herbivore densities is being observed across Europe (e.g. Fuller & Gill 2001; Ward 2005; Milner et al. 2006; Bláha & Kotecký 2008). The increase is explained by land-use changes, milder winters and changes in game management. Decreasing numbers in three of the studied forest landscapes are due to fencing (Lady Park Wood), targeted hunting (Munich) or the abandonment of a game preserve (Děvín).

In summary, the ecological changes in our study sites reflect some of the major trends that are affecting broadleaved woodland more generally across Europe, and so the changes seen in the understorey in this study are likely to be applicable more generally.

Changes in the Understorey

In 70% of the study regions, the number of species significantly increased or decreased, and almost 30% of the species in the plots has been replaced since the time of the first survey, which concurs with understorey changes reported in other resurvey studies performed in, for instance, North America (e.g. Taverna, Peet & Phillips 2005; Rogers et al. 2008). The large shift in composition could be partly due to the fact that most studies used semi-permanent plots, which might have introduced some relocation error, and because the observers differed between the old and recent surveys. However, the use of presence/absence data partly reduced such sampling error. The fact that the changes in the floristic composition were directional as indicated by the mean Ellenberg value shifts (Fig. 2) also suggests that these are real-world effects. Furthermore, the presence/absence data yield conservative estimates of the plant community change, and it is likely that shifts in species’ relative abundances have taken place as well.

The degree to which the species richness and composition changed over time was positively related to the time interval between the old and recent surveys (Table 4), which may be caused by the life span of many forest understorey species that can be as long as several decades (Ehrlen & Lethilä 2002). Therefore, community reorganization is more likely to be detected as the time interval between the two surveys increases. Similar results were, for instance, found by Dupréet al. (2010) in acidic grasslands.

In our study, plots with higher initial light availability (expressed as higher Ellenberg L values) showed lower similarity and larger reduction in species richness between the two survey dates. The replacement and filtering of light-demanding species due to a gradual canopy gap closure is likely to account for those patterns (Van Calster et al. 2008b). This corroborates the results of Kirby et al. (2005) who found a general decrease in understorey species richness, except in sites that were most severely hit by the 1987 storm in the south and east of England.

Unlike studies in grassland (e.g. Stevens et al. 2004; Dupréet al. 2010), high atmospheric N deposition rates did not have a negative effect on the understorey species richness. By contrast, there was even a (weak) positive effect on species richness, and plots exposed to higher N deposition rates tended to exhibit less floristic changes. Some experimental N addition studies also found positive (Hurteau & North 2008) or mixed (Ostertag & Verville 2002) effects on understorey species richness, whereas others found negative effects (Strengbom et al. 2001). Gilliam (2006) discusses N-mediated changes in various processes (e.g. competition, herbivory, mycorrhizal infection), which could all potentially affect forest understorey diversity and composition, but the rather idiosyncratic results of studies so far indicate that the understorey effects of adding a single limiting resource cannot yet be predicted at the community level (see also De Schrijver et al. 2011).

The shifts in Ellenberg values, which point to increased shading and nutrient availability, are consistent with those reported elsewhere (e.g. Kirby et al. 2005; Baeten et al. 2009; Keith et al. 2009). The importance of canopy variables to explain the changes in the Ellenberg indicator value for light (Table 5) is consistent with the expectation that more shady conditions, caused by increasing canopy cover and a higher relative importance of shade casting species in the canopy, reduce the survival chances of more light-demanding species in the understorey plant community. The positive relationship between increasing litter quality and the share of light-demanding species may be related to the difficulties these species experience in germinating and/or establishing on sites where a litter layer has accumulated (Sydes & Grime 1981; Eriksson 1995; Dzwonko & Gawronski 2002; Sayer 2006).

The more frequent occurrence of nutrient-demanding species in the community, detected through the Ellenberg N indicator values, is not directly explained by variation in the N deposition rate. Instead, changes in the canopy seem to be primarily responsible for the observed eutrophication signal. The increased input (and maybe also reduced removal) of litter and the increasing importance of species with faster decomposing litter is likely to have increased the general nutrient availability in the studied forests (cf. Dzwonko & Gawronski 2002; Hofmeister et al. 2009). Common garden experiments have shown that tree species differ greatly in their impacts on soil acidity and fertility (e.g. Neirynck et al. 2000; Hagen-Thorn et al. 2004; Reich et al. 2005; Vesterdal et al. 2008) with consequent impacts on the understorey vegetation (e.g. van Oijen et al. 2005; Wulf & Naaf 2009; Kooijman 2010). Kooijman (2010), for example, specifically focused on litter-generated differences in N cycling under tree species with contrasting litter quality (beech and hornbeam), and the effect on the understorey species composition. The, albeit weak, negative correlation between RRL and RRN is consistent with a canopy-induced eutrophication signal as it suggests that the frequency of more nutrient-demanding species has particularly increased in plots where the canopy has become more closed. However, an indirect effect of N deposition on the forest understorey, caused by increasing forest productivity and the rates of canopy closure (e.g. Hedwall et al. 2010) or by changing the foliar nutrient contents and litter decomposition rates (e.g. May et al. 2005), cannot be excluded. Indeed, excluding the Elbe-Weser study from Fig. 3a, the results reveal a positive relationship between the N deposition rate and the canopy closure, but further research is needed to confirm this relationship. Nevertheless, interspecific variability in leaf traits will most likely continue to have a dominant impact on litter decomposition (Cornwell et al. 2008).

Although it is inherently difficult to disentangle acidification and eutrophication using Ellenberg indicator values (e.g. Diekmann & Dupré 1997), it seems that the latter process is more important to explain the patterns observed in this study. The RRN values equally increased in plots in more acidic (mRo ≤ 5) and more base-rich sites (mRo > 5). Also the response ratios for soil reaction (RRR) increased, but more so in the more nutrient-poor sites (mNo ≤ 5) than in the more nutrient-rich sites (mNo > 5) (results not shown). Bürger-Arndt (1994) found similar patterns and considered the increasing mR values in forests that are becoming darker to be an artefact caused by the selective loss of acid-tolerant species that are often light demanding.

Synthesis

Significant shifts have occurred in the understorey vegetation of semi-natural deciduous forest in temperate Europe during recent decades. Whereas no unidirectional shifts in species richness occurred, the relative proportion of nutrient-demanding and shade-tolerant species has clearly increased. Atmospheric N deposition may be one of the (indirect) drivers behind the change, but management-related alterations in the canopy structure and composition appear much more important. This finding is yet another example of the importance of understorey–overstorey interactions in forests (Gilliam 2007).

Our multi-site approach clearly demonstrates that one should be cautious when drawing conclusions about the impact of atmospheric N deposition based on the interpretation of plant community shifts in single sites or regions (e.g. Thimonier et al. 1994; Lameire, Hermy & Honnay 2000) due to other concurrent ecological changes. However, even though the effects of many decades of increased atmospheric N deposition are currently overruled by the effects of canopy changes, atmospheric N deposition may still have a significant impact.

Smith, Knapp & Collins (2009) recently proposed a hierarchical-response framework (HR-framework), conceptualizing ecological change as a response to chronic resource alterations. The forest understoreys under study have on the one hand experienced chronic increases of atmospheric N but on the other hand chronic decreases of light availability. These opposing trends in resource availability together with the longevity of forest understorey species and their often slow colonization rates (Verheyen et al. 2003; De Frenne et al. 2011) may help to explain the apparent resistance of forest understorey plant communities to species losses as a result of chronic N additions. This resistance may change, however, if the forest canopies are opened up again so that light becomes a less limiting resource. The HR-framework would suggest that a strong community reordering and possible species loss could be expected if the N that has built up over decades becomes available for plant growth. Initially, this might mean increases in species richness due to the increased occurrence of disturbance-adapted and non-forest species, but in the longer term, a decline in some forest specialist species through competition with competitive, light-demanding taxa such as Rubus fruticosus coll. and several graminoids could be expected. Chronic N deposition might therefore be regarded to as the building up of a ‘nitrogen time bomb’.

However, the accumulated N may not become readily available for plant growth due to microbial immobilization and subsequent storage in stable soil organic matter (SOM) pools. Due to, often long term, intensive coppice with standards management, the SOM stocks in many ancient forests in Europe are currently still building up (e.g. Luyssaert et al. 2010), generating a higher potential for N immobilization. For instance, McLauchlan et al. (2007) demonstrated that N availability for plant growth in a North-American hardwood forest subject to increased atmospheric N deposition has even declined over the past 75 years probably because the system was still recovering from a period of agricultural land use during the 19th century.

Clearly, more research is needed to better understand the current and future impacts that increased N deposition may have on forest understorey communities.

Acknowledgements

This research was financially supported by the Institute for Nature and Forest Research (INBO) and by the Research Foundation – Flanders (FWO) by funding the scientific research network FLEUR (http://www.fleur.ugent.be). While writing the paper, L.B and G.V. held a PhD grant from the Institute for the Promotion of Innovation through Science and Technology in Flanders (IWT-Vlaanderen), P.D.F. held a PhD grant from the FWO and R.H. and P.P. received support by the grants AV0 IAA600050812 and AV0 Z60050516. Finally, we greatly acknowledge Martin Diekmann, Frank Gilliam and two anonymous reviewers for extensive remarks and discussions on an earlier version of the manuscript.

Ancillary