Oecologia DOI 10.1007/s00442-014-2917-7
Community ecology - Original research
Species richness and trait composition of butterfly assemblages change along an altitudinal gradient Annette Leingärtner · Jochen Krauss · Ingolf Steffan‑Dewenter
Received: 7 May 2013 / Accepted: 5 March 2014 © Springer-Verlag Berlin Heidelberg 2014
Abstract Species richness patterns along altitudinal gradients are well-documented ecological phenomena, yet very little data are available on how environmental filtering processes influence the composition and traits of butterfly assemblages at high altitudes. We have studied the diversity patterns of butterfly species at 34 sites along an altitudinal gradient ranging from 600 to 2,000 m a.s.l. in the National Park Berchtesgaden (Germany) and analysed traits of butterfly assemblages associated with dispersal capacity, reproductive strategies and developmental time from lowlands to highlands, including phylogenetic analyses. We found a linear decline in butterfly species richness along the altitudinal gradient, but the phylogenetic relatedness of the butterfly assemblages did not increase with altitude. Compared to butterfly assemblages at lower altitudes, those at higher altitudes were composed of species with larger wings (on average 9 %) which laid an average of 68 % more eggs. In contrast, egg maturation time in butterfly assemblages decreased by about 22 % along the altitudinal gradient. Further, butterfly assemblages at higher altitudes were increasingly dominated by less widespread species. Based on our abundance data, but not on data in the literature, population density increased with altitude, suggesting a reversed density–distribution relationship, with higher
Communicated by Konrad Fiedler. Electronic supplementary material The online version of this article (doi:10.1007/s00442-014-2917-7) contains supplementary material, which is available to authorized users. A. Leingärtner (*) · J. Krauss · I. Steffan‑Dewenter Department of Animal Ecology and Tropical Biology, Biocentre, University of Würzburg, Am Hubland, 97074 Würzburg, Germany e-mail: annette.leingaertner@uni‑wuerzburg.de
population densities of habitat specialists in harsh environments. In conclusion, our data provide evidence for significant shifts in the composition of butterfly assemblages and for the dominance of different traits along the altitudinal gradient. In our study, these changes were mainly driven by environmental factors, whereas phylogenetic filtering played a minor role along the studied altitudinal range. Keywords Alpine ecosystems · Biodiversity · Climate · Environmental filtering · Population density
Introduction Altitudinal species richness gradients are shaped by longterm speciation processes, dispersal, and extinction events (Mittelbach et al. 2007) as well as short-term ecological interactions with other organisms and the environment. Ongoing global change permanently alters community compositions by influencing the distribution and interactions of species, hence reorganising species richness gradients (Carnicer et al. 2012). There are many hypotheses explaining present-day species richness and abundance patterns along altitudinal gradients. The species-energy hypothesis predicts declining species richness with increasing altitude due to decreasing energy availability and decreasing population sizes towards the mountain tops (Chown et al. 2012). Another hypothesis explains the linear decrease of biodiversity along altitude gradients with the species–area relationship (Lomolino 2000). Thus, the number of species declines towards the summit (increasing altitude) because less area is available due to the conical shape of mountains (Jones et al. 2003). However, biodiversity does not always show a linear pattern with increasing altitude, but can peak at intermediate altitudes, possibly
13
Oecologia
a result of reduced biodiversity at low elevations due to human impact (Nogues-Bravo et al. 2008), highest productivity at mid-elevations and the mid-domain effect, which explains a species richness peak at mid-elevation by stochastic effects of randomly distributed species ranges that overlap more towards the centre of a geographical domain (Colwell et al. 2004). The local presence and absence of species is determined by the characteristics or traits of the species, thus species inhabiting habitats with similar environmental and climatic conditions often have similar traits. However, when environmental and climatic conditions change, dominant traits within communities presumably will shift due to species turnover or intraspecific shifts (Cornwell and Ackerly 2009). Parmesan (2006) reported northward and upward range shifts of cool-adapted species as a means to escape rising temperatures, while Lenoir (2008) observed that mountain species moved their optimum elevation further upwards than non-mountain species. Therefore, a study of species’ traits that are sensitive to current environmental and climatic conditions, such as those along altitudinal gradients, can further our understanding of species’ responses to climate change (Diamond et al. 2011). Such gradients can be used as a space-for-time substitution to study potential responses of species to changing environmental and climatic conditions, assuming that species respond to climate change over time in the same way as ecosystems now vary along altitude (Dunne et al. 2004). Changes in traits along environmental gradients can be studied on three levels, i.e. an intraspecific level, an interspecific level and community assemblage level (Gaston et al. 2008). Intraspecific studies look at shifts in traits within individual species (Berner et al. 2004; Karl et al. 2008; Wagner et al. 2011), whereas the aim of studies on an interspecific level is to analyse trait variation between several species (Hawkins and DeVries 1996). Studies at the community assemblage level, in contrast to those at the interspecific level, do not use each species as a response, rather they use aggregated values of community assemblages along an environmental gradient (Chown and Klok 2003; Hoiss et al. 2012). Community traits which presumably change along altitudinal gradients are dispersal capacity, reproductive strategies and developmental time, but how these traits change with altitude and which trade-offs between different traits are made are still unknown. Therefore, the analysis of multiple traits of community assemblages along altitudinal gradients is crucial to the prediction of shifts in community composition due to future climate change (Diamond et al. 2011). In the study reported here, we used butterfly species as the model because they represent adequate indicators of environmental and climate changes (Thomas 2005) and
13
react more sensitively and decline more rapidly than birds and plants (Thomas et al. 2004). We analysed six traits (wing length, dispersal, egg number, egg deposition, egg maturation time and number of generations per year) within butterfly assemblages. Butterfly assemblages along altitudinal gradients are mainly determined by temperature-driven filtering processes (Pellissier et al. 2013). Previous studies show a decrease of host plant specialisation and a loss of interactions with ants in butterfly assemblages towards higher altitudes (Pellissier et al. 2012a, b). However, little information is available on which traits are favoured in cold conditions at high elevations. Therefore, we studied how altitude affects butterfly species richness on alpine meadows and which traits prevail in butterfly assemblages at different altitudes. We tested the following predictions: 1. Species richness is highest in lowlands and decreases with altitude. 2. Traits in butterfly assemblages, i.e., wing length, dispersal, egg number, egg deposition, egg maturation time and number of generations per year, change along the altitudinal gradient. 3. Butterfly assemblages at higher altitudes contain a higher proportion of species and individuals at low population densities and narrow geographical distributions than butterfly assemblages at lower altitudes.
Methods Study sites This study was carried out in the National Park Berchtesgaden and its surroundings, located in the “Berchtesgadener Alps”, a part of the Northern Limestone Alps in the southeast of Germany. The national park is characterised by alpine meadows and high mountains with an altitudinal gradient from 600 to 2,700 m.a.s.l. Alpine pastoral systems have an over 1,000 year old tradition in the National Park Berchtesgaden, but today only few alpine meadows are still traditionally managed while the others are abandoned. The annual mean temperature varies between +7 and −2 °C, and the annual mean precipitation varies between 1,500 and 2,600 mm depending on the altitude. As study sites, we selected 34 grasslands that ranged in altitude from approximately 600 to approximately 2,000 m.a.s.l (mean latitude/longitude of all grasslands: N 47°6′, E 12°9′) that were situated along two valleys with smooth slopes and two mountainsides with steep slopes. We did not use sites situated at >2,000 m.a.s.l. because only bare rocks and coarse gravel could be found at this level. Of the 34 grassland sites, 29 were located inside the
Oecologia
national park and five were in close vicinity. These latter five grasslands sites were expressly chosen outside the national park to cover a larger altitude gradient because within the park we could not find suitable grasslands between 600 and 800 m.a.s.l. Almost all grasslands below the timberline were surrounded by coniferous forests. The selection criteria for the 34 grasslands were: (1) location along an altitudinal gradient with about 250-m altitude difference between the grasslands; (2) extensively used or not used; (3) permission of the farmers. Data collection Butterflies (Lepidoptera: Hesperioidea and Papilionoidea) were recorded from 8 May to 10 September in 2009. Sites located at elevations of >1,200 m a.s.l. were sampled five times during the study period, while those at 17 °C on cloudy days. All grasslands were visited at regular time intervals that spanned the sampling period in a random sequence. To analyse the effect of flower cover on butterfly occurrence, we estimated the cover of all flowering plant species as a percentage of the whole study site (60 × 60 m) after each transect walk. We used the mean value of the flower cover over all transect walks per study site for statistical analysis. We also measured air temperature with temperature loggers (Thermochron iButtons DS1921G#F5; Maxim Integrated Products, Inc., Sunnyvale, CA) at each study site 1 m above the ground in 2-h intervals. We only used temperature data between 16 June and 7 September 2009 for statistical analyses, as we only had data for all sites within this period. For trait analyses of butterfly species, we used published literature data on the following six trait values: wing length, dispersal capacity, egg number, the way of egg deposition, egg maturation time and the number of generations per year [Electronic Supplementary Material (ESM), Appendix
1, Table S1]. In contrast to intraspecific studies with traits of a single species, it was impossible for us to measure the traits of all sampled species for our community assemblage analyses. We therefore used the average wing length of the forewing of male butterfly species as a proxy for butterfly body size (Higgins and Riley 1971) because male and female wing length is strongly positively correlated (Komonen et al. 2004). The dispersal capacity of butterfly species was classified into migration categories where each butterfly species was assigned to one migration category, ranging from 1 = sedentary to 9 = migratory (Settele et al. 1999). The egg number per species represents the total number of eggs potentially laid per individual (Settele et al. 1999; Sonderegger 2005). The mode of egg deposition was divided into eggs which were singly laid and eggs which were laid as egg packages (clutches) (Schweizerischer Bund für Naturschutz 1994; Weidemann 1995; Sonderegger 2005). The egg maturation time represents the days from hatching of the adult female until the first egg deposition (Settele et al. 1999; Sonderegger 2005). For data analyses of egg maturation time, we only used records without hibernation and aestivation to have comparable data sets. The two butterfly species Gonepteryx rhamni and Nymphalis antiopa always hibernate before they lay their eggs; therefore, we excluded these two species from the analyses of egg maturation time. We used the number of generations per year to calculate the proportion of multivoltine species per study site, which also included bivoltine species (Settele et al. 1999; Stettmer et al. 2007). For the analysis of population density we used literature data provided for each species as individuals per area in nine population density classes ranging from an extremely high density of 1,000 individuals per hectare to an extremely low density of two individuals per square kilometre (Settele et al. 1999). Furthermore, we used our own sampled butterfly abundance data (individuals per hectare for each species) to calculate mean population densities of butterfly assemblages per study site in order to compare literature and field data. To calculate the geographical distribution of butterfly assemblages we took a distribution index from the literature that indicates the range of a butterfly species in relation to the total area of Europe calculated in percentage (Kudrna et al. 2011). In a few cases data on specialised alpine butterfly species were missing, so we included external expert opinion or used estimated data from closely related species (ESM, Appendix 1, Table S1). Statistical analyses Statistical analyses were performed with the software R 2.15.1 for Windows ® Core Team 2012). We used general linear models (GLMs) with Type III sums of squares, with the explanatory variables altitude and flower cover. The
13
Oecologia
predictors altitude and flower cover showed no significant correlation (r = −0.3) and thus could be included in one model. We removed one study site from the species richness and abundance analysis, as flower cover at this study site was identified to be an outlier that had a strong influence on the results. Model simplification was performed with likelihood ratio tests by removing non-significant terms from the model. The residuals of the models were normally distributed and met the assumptions of variance homogeneity. We used the butterfly abundance data to calculate the community density (individuals per hectare) for each study site along the altitudinal gradient. We calculated species richness estimators using the software EstimateS (ver. 8.2; R. K. Colwell, available at: http://purl.oclc.org/estimates) to verify that sampling effort on the study sites was sufficient. We used the eight 4-min sections (sub-samples) per transect walk as replicates for species richness estimation. We divided the recorded species richness by the estimator ACE (abundance-based coverage estimator of species richness) to obtain species saturation per study site. Species saturation was high and similar between the study sites (mean 83 %, range 60–100 %). For our analysis we used the species richness and not the calculated estimators because of the high detection rates. We used a two-sample t test to compare the mean species saturation of five transect walks with the mean species saturation of six transect walks (t32 = 0.21, P = 0.84, mean of 5 walks mean 83 %, mean of six walks 82 %). As the variances of the five and six walks’ mean were equal, we pooled the data. To test for spatial autocorrelation we calculated Moran’s I with the ‘correlog’ function of the R package ‘ncf’ (Kissling and Carl 2008). Temperature changes along the altitudinal gradient were calculated with a GLM. For community traits we calculated species- and abundance-based means for wing length, migration categories, egg number, egg maturation time, population density and distribution per site. Additionally, we used the ratio of egg number to wing length of each butterfly assemblage to correct for the effect that larger butterfly species lay more eggs (Garcia-Barros 2000; Berger et al. 2008). We also calculated the proportion of egg deposition in clutches and the proportion of multivoltine species per butterfly assemblage. Species life-history trait values were averaged when ranges were available in the literature. Thus, our community-based trait analyses do not take into account intraspecific trait variation along the elevational gradient. We used regression analyses with species-based and abundance-weighted trait values as response variables, and altitude as explanatory variable, to test the shifts of life-history traits along the altitudinal gradient. For the population density and the distribution analyses of the butterfly assemblages we performed
13
regressions with the species-based and abundance-based means per study site as response variables and altitude as predictor. Relationships between the response variables were analysed with Pearson’s correlations. Traits among phylogenetically more closely related species can be more highly correlated than those with distantly related species. We prepared polytomous, ultrametric trees using a published taxonomy of butterfly species (ESM Fig. S1) (Segerer et al. 2011) to analyse the phylogenetic relatedness within the studied butterfly assemblages at the study sites along the altitudinal gradient. We used the function ‘as.phylo’ of the R package ‘ape’ (Paradis et al. 2004) to convert the butterfly taxonomy into a tree. For the computation of branch lengths we used the function ‘compute.brlen’ of the R package ‘ape’ with the ‘Grafen’ method (Grafen 1989). The phylogenetic relatedness within butterfly assemblages was estimated with the net relatedness index (NRI) (Webb et al. 2002). For the null model, the number of species which occurred at each site was randomly selected from all species found in our study. The sampling probability of each species was abundance weighted, and the sampling was repeated 500 times per site. We then calculated the mean and standard deviation for every site. Species assemblages with a higher NRI are closer related than species with a lower NRI.
Results Species richness and abundance Altogether we recorded 4,691 individuals belonging to 67 butterfly species on the 34 grassland study sites. Fourteen of the sampled butterfly species were restricted to alpine regions, including 11 species of the genus Erebia, and five of these were only found above 1,500 m.a.s.l. (Aricia artaxerxes, Boloria pales, Erebia epiphron, E. gorge, E. pandrose) (ESM, Appendix 1, Table S1). The number of butterfly species linearly declined with increasing altitude (Fig. 1a). Flower cover was positively correlated with butterfly richness (Table 1). At 2,000 m.a.s.l. we recorded on average half of the butterfly species compared to species richness found at 600 m.a.s.l. Butterfly abundance was not significantly related to altitude and flower cover (Table 1). Further, Moran’s Index was not significantly different from zero for the tested distance classes 1–5 km with intervals of 1 km, indicating that our data set was not spatially autocorrelated. In contrast to the butterfly species richness, our analysis of butterfly community density showed no altitudinal pattern (Fig. 1b). The temperature decreased by 0.45 °C per 100 m of altitude during the summer months (Fig. 2).
Oecologia Fig. 1 a Species richness of butterfly species (y = 28.9 − 0.009x) decreased with altitude (n = 33 study sites). The size of the filled circles in the butterfly graphic corresponds with flower cover (minimum = 0.41 %, maximum = 4.39 %). See Table 1 for statistical details. b Community density of butterflies (field data) did not change along the altitudinal gradient (F1, 32 = 2.53, r2 = 0.07, P = 0.12)
Table 1 General linear models output (Type III sums of squares) with butterfly species richness and abundance as response variables and flower cover and altitude as explanatory variables (see Fig. 1a) Response
Predictor
df
F
P
Butterfly richness
Flower cover Altitude
1, 30 1, 30
5.65 18.01
0.02
Community ecology - Original research
Species richness and trait composition of butterfly assemblages change along an altitudinal gradient Annette Leingärtner · Jochen Krauss · Ingolf Steffan‑Dewenter
Received: 7 May 2013 / Accepted: 5 March 2014 © Springer-Verlag Berlin Heidelberg 2014
Abstract Species richness patterns along altitudinal gradients are well-documented ecological phenomena, yet very little data are available on how environmental filtering processes influence the composition and traits of butterfly assemblages at high altitudes. We have studied the diversity patterns of butterfly species at 34 sites along an altitudinal gradient ranging from 600 to 2,000 m a.s.l. in the National Park Berchtesgaden (Germany) and analysed traits of butterfly assemblages associated with dispersal capacity, reproductive strategies and developmental time from lowlands to highlands, including phylogenetic analyses. We found a linear decline in butterfly species richness along the altitudinal gradient, but the phylogenetic relatedness of the butterfly assemblages did not increase with altitude. Compared to butterfly assemblages at lower altitudes, those at higher altitudes were composed of species with larger wings (on average 9 %) which laid an average of 68 % more eggs. In contrast, egg maturation time in butterfly assemblages decreased by about 22 % along the altitudinal gradient. Further, butterfly assemblages at higher altitudes were increasingly dominated by less widespread species. Based on our abundance data, but not on data in the literature, population density increased with altitude, suggesting a reversed density–distribution relationship, with higher
Communicated by Konrad Fiedler. Electronic supplementary material The online version of this article (doi:10.1007/s00442-014-2917-7) contains supplementary material, which is available to authorized users. A. Leingärtner (*) · J. Krauss · I. Steffan‑Dewenter Department of Animal Ecology and Tropical Biology, Biocentre, University of Würzburg, Am Hubland, 97074 Würzburg, Germany e-mail: annette.leingaertner@uni‑wuerzburg.de
population densities of habitat specialists in harsh environments. In conclusion, our data provide evidence for significant shifts in the composition of butterfly assemblages and for the dominance of different traits along the altitudinal gradient. In our study, these changes were mainly driven by environmental factors, whereas phylogenetic filtering played a minor role along the studied altitudinal range. Keywords Alpine ecosystems · Biodiversity · Climate · Environmental filtering · Population density
Introduction Altitudinal species richness gradients are shaped by longterm speciation processes, dispersal, and extinction events (Mittelbach et al. 2007) as well as short-term ecological interactions with other organisms and the environment. Ongoing global change permanently alters community compositions by influencing the distribution and interactions of species, hence reorganising species richness gradients (Carnicer et al. 2012). There are many hypotheses explaining present-day species richness and abundance patterns along altitudinal gradients. The species-energy hypothesis predicts declining species richness with increasing altitude due to decreasing energy availability and decreasing population sizes towards the mountain tops (Chown et al. 2012). Another hypothesis explains the linear decrease of biodiversity along altitude gradients with the species–area relationship (Lomolino 2000). Thus, the number of species declines towards the summit (increasing altitude) because less area is available due to the conical shape of mountains (Jones et al. 2003). However, biodiversity does not always show a linear pattern with increasing altitude, but can peak at intermediate altitudes, possibly
13
Oecologia
a result of reduced biodiversity at low elevations due to human impact (Nogues-Bravo et al. 2008), highest productivity at mid-elevations and the mid-domain effect, which explains a species richness peak at mid-elevation by stochastic effects of randomly distributed species ranges that overlap more towards the centre of a geographical domain (Colwell et al. 2004). The local presence and absence of species is determined by the characteristics or traits of the species, thus species inhabiting habitats with similar environmental and climatic conditions often have similar traits. However, when environmental and climatic conditions change, dominant traits within communities presumably will shift due to species turnover or intraspecific shifts (Cornwell and Ackerly 2009). Parmesan (2006) reported northward and upward range shifts of cool-adapted species as a means to escape rising temperatures, while Lenoir (2008) observed that mountain species moved their optimum elevation further upwards than non-mountain species. Therefore, a study of species’ traits that are sensitive to current environmental and climatic conditions, such as those along altitudinal gradients, can further our understanding of species’ responses to climate change (Diamond et al. 2011). Such gradients can be used as a space-for-time substitution to study potential responses of species to changing environmental and climatic conditions, assuming that species respond to climate change over time in the same way as ecosystems now vary along altitude (Dunne et al. 2004). Changes in traits along environmental gradients can be studied on three levels, i.e. an intraspecific level, an interspecific level and community assemblage level (Gaston et al. 2008). Intraspecific studies look at shifts in traits within individual species (Berner et al. 2004; Karl et al. 2008; Wagner et al. 2011), whereas the aim of studies on an interspecific level is to analyse trait variation between several species (Hawkins and DeVries 1996). Studies at the community assemblage level, in contrast to those at the interspecific level, do not use each species as a response, rather they use aggregated values of community assemblages along an environmental gradient (Chown and Klok 2003; Hoiss et al. 2012). Community traits which presumably change along altitudinal gradients are dispersal capacity, reproductive strategies and developmental time, but how these traits change with altitude and which trade-offs between different traits are made are still unknown. Therefore, the analysis of multiple traits of community assemblages along altitudinal gradients is crucial to the prediction of shifts in community composition due to future climate change (Diamond et al. 2011). In the study reported here, we used butterfly species as the model because they represent adequate indicators of environmental and climate changes (Thomas 2005) and
13
react more sensitively and decline more rapidly than birds and plants (Thomas et al. 2004). We analysed six traits (wing length, dispersal, egg number, egg deposition, egg maturation time and number of generations per year) within butterfly assemblages. Butterfly assemblages along altitudinal gradients are mainly determined by temperature-driven filtering processes (Pellissier et al. 2013). Previous studies show a decrease of host plant specialisation and a loss of interactions with ants in butterfly assemblages towards higher altitudes (Pellissier et al. 2012a, b). However, little information is available on which traits are favoured in cold conditions at high elevations. Therefore, we studied how altitude affects butterfly species richness on alpine meadows and which traits prevail in butterfly assemblages at different altitudes. We tested the following predictions: 1. Species richness is highest in lowlands and decreases with altitude. 2. Traits in butterfly assemblages, i.e., wing length, dispersal, egg number, egg deposition, egg maturation time and number of generations per year, change along the altitudinal gradient. 3. Butterfly assemblages at higher altitudes contain a higher proportion of species and individuals at low population densities and narrow geographical distributions than butterfly assemblages at lower altitudes.
Methods Study sites This study was carried out in the National Park Berchtesgaden and its surroundings, located in the “Berchtesgadener Alps”, a part of the Northern Limestone Alps in the southeast of Germany. The national park is characterised by alpine meadows and high mountains with an altitudinal gradient from 600 to 2,700 m.a.s.l. Alpine pastoral systems have an over 1,000 year old tradition in the National Park Berchtesgaden, but today only few alpine meadows are still traditionally managed while the others are abandoned. The annual mean temperature varies between +7 and −2 °C, and the annual mean precipitation varies between 1,500 and 2,600 mm depending on the altitude. As study sites, we selected 34 grasslands that ranged in altitude from approximately 600 to approximately 2,000 m.a.s.l (mean latitude/longitude of all grasslands: N 47°6′, E 12°9′) that were situated along two valleys with smooth slopes and two mountainsides with steep slopes. We did not use sites situated at >2,000 m.a.s.l. because only bare rocks and coarse gravel could be found at this level. Of the 34 grassland sites, 29 were located inside the
Oecologia
national park and five were in close vicinity. These latter five grasslands sites were expressly chosen outside the national park to cover a larger altitude gradient because within the park we could not find suitable grasslands between 600 and 800 m.a.s.l. Almost all grasslands below the timberline were surrounded by coniferous forests. The selection criteria for the 34 grasslands were: (1) location along an altitudinal gradient with about 250-m altitude difference between the grasslands; (2) extensively used or not used; (3) permission of the farmers. Data collection Butterflies (Lepidoptera: Hesperioidea and Papilionoidea) were recorded from 8 May to 10 September in 2009. Sites located at elevations of >1,200 m a.s.l. were sampled five times during the study period, while those at 17 °C on cloudy days. All grasslands were visited at regular time intervals that spanned the sampling period in a random sequence. To analyse the effect of flower cover on butterfly occurrence, we estimated the cover of all flowering plant species as a percentage of the whole study site (60 × 60 m) after each transect walk. We used the mean value of the flower cover over all transect walks per study site for statistical analysis. We also measured air temperature with temperature loggers (Thermochron iButtons DS1921G#F5; Maxim Integrated Products, Inc., Sunnyvale, CA) at each study site 1 m above the ground in 2-h intervals. We only used temperature data between 16 June and 7 September 2009 for statistical analyses, as we only had data for all sites within this period. For trait analyses of butterfly species, we used published literature data on the following six trait values: wing length, dispersal capacity, egg number, the way of egg deposition, egg maturation time and the number of generations per year [Electronic Supplementary Material (ESM), Appendix
1, Table S1]. In contrast to intraspecific studies with traits of a single species, it was impossible for us to measure the traits of all sampled species for our community assemblage analyses. We therefore used the average wing length of the forewing of male butterfly species as a proxy for butterfly body size (Higgins and Riley 1971) because male and female wing length is strongly positively correlated (Komonen et al. 2004). The dispersal capacity of butterfly species was classified into migration categories where each butterfly species was assigned to one migration category, ranging from 1 = sedentary to 9 = migratory (Settele et al. 1999). The egg number per species represents the total number of eggs potentially laid per individual (Settele et al. 1999; Sonderegger 2005). The mode of egg deposition was divided into eggs which were singly laid and eggs which were laid as egg packages (clutches) (Schweizerischer Bund für Naturschutz 1994; Weidemann 1995; Sonderegger 2005). The egg maturation time represents the days from hatching of the adult female until the first egg deposition (Settele et al. 1999; Sonderegger 2005). For data analyses of egg maturation time, we only used records without hibernation and aestivation to have comparable data sets. The two butterfly species Gonepteryx rhamni and Nymphalis antiopa always hibernate before they lay their eggs; therefore, we excluded these two species from the analyses of egg maturation time. We used the number of generations per year to calculate the proportion of multivoltine species per study site, which also included bivoltine species (Settele et al. 1999; Stettmer et al. 2007). For the analysis of population density we used literature data provided for each species as individuals per area in nine population density classes ranging from an extremely high density of 1,000 individuals per hectare to an extremely low density of two individuals per square kilometre (Settele et al. 1999). Furthermore, we used our own sampled butterfly abundance data (individuals per hectare for each species) to calculate mean population densities of butterfly assemblages per study site in order to compare literature and field data. To calculate the geographical distribution of butterfly assemblages we took a distribution index from the literature that indicates the range of a butterfly species in relation to the total area of Europe calculated in percentage (Kudrna et al. 2011). In a few cases data on specialised alpine butterfly species were missing, so we included external expert opinion or used estimated data from closely related species (ESM, Appendix 1, Table S1). Statistical analyses Statistical analyses were performed with the software R 2.15.1 for Windows ® Core Team 2012). We used general linear models (GLMs) with Type III sums of squares, with the explanatory variables altitude and flower cover. The
13
Oecologia
predictors altitude and flower cover showed no significant correlation (r = −0.3) and thus could be included in one model. We removed one study site from the species richness and abundance analysis, as flower cover at this study site was identified to be an outlier that had a strong influence on the results. Model simplification was performed with likelihood ratio tests by removing non-significant terms from the model. The residuals of the models were normally distributed and met the assumptions of variance homogeneity. We used the butterfly abundance data to calculate the community density (individuals per hectare) for each study site along the altitudinal gradient. We calculated species richness estimators using the software EstimateS (ver. 8.2; R. K. Colwell, available at: http://purl.oclc.org/estimates) to verify that sampling effort on the study sites was sufficient. We used the eight 4-min sections (sub-samples) per transect walk as replicates for species richness estimation. We divided the recorded species richness by the estimator ACE (abundance-based coverage estimator of species richness) to obtain species saturation per study site. Species saturation was high and similar between the study sites (mean 83 %, range 60–100 %). For our analysis we used the species richness and not the calculated estimators because of the high detection rates. We used a two-sample t test to compare the mean species saturation of five transect walks with the mean species saturation of six transect walks (t32 = 0.21, P = 0.84, mean of 5 walks mean 83 %, mean of six walks 82 %). As the variances of the five and six walks’ mean were equal, we pooled the data. To test for spatial autocorrelation we calculated Moran’s I with the ‘correlog’ function of the R package ‘ncf’ (Kissling and Carl 2008). Temperature changes along the altitudinal gradient were calculated with a GLM. For community traits we calculated species- and abundance-based means for wing length, migration categories, egg number, egg maturation time, population density and distribution per site. Additionally, we used the ratio of egg number to wing length of each butterfly assemblage to correct for the effect that larger butterfly species lay more eggs (Garcia-Barros 2000; Berger et al. 2008). We also calculated the proportion of egg deposition in clutches and the proportion of multivoltine species per butterfly assemblage. Species life-history trait values were averaged when ranges were available in the literature. Thus, our community-based trait analyses do not take into account intraspecific trait variation along the elevational gradient. We used regression analyses with species-based and abundance-weighted trait values as response variables, and altitude as explanatory variable, to test the shifts of life-history traits along the altitudinal gradient. For the population density and the distribution analyses of the butterfly assemblages we performed
13
regressions with the species-based and abundance-based means per study site as response variables and altitude as predictor. Relationships between the response variables were analysed with Pearson’s correlations. Traits among phylogenetically more closely related species can be more highly correlated than those with distantly related species. We prepared polytomous, ultrametric trees using a published taxonomy of butterfly species (ESM Fig. S1) (Segerer et al. 2011) to analyse the phylogenetic relatedness within the studied butterfly assemblages at the study sites along the altitudinal gradient. We used the function ‘as.phylo’ of the R package ‘ape’ (Paradis et al. 2004) to convert the butterfly taxonomy into a tree. For the computation of branch lengths we used the function ‘compute.brlen’ of the R package ‘ape’ with the ‘Grafen’ method (Grafen 1989). The phylogenetic relatedness within butterfly assemblages was estimated with the net relatedness index (NRI) (Webb et al. 2002). For the null model, the number of species which occurred at each site was randomly selected from all species found in our study. The sampling probability of each species was abundance weighted, and the sampling was repeated 500 times per site. We then calculated the mean and standard deviation for every site. Species assemblages with a higher NRI are closer related than species with a lower NRI.
Results Species richness and abundance Altogether we recorded 4,691 individuals belonging to 67 butterfly species on the 34 grassland study sites. Fourteen of the sampled butterfly species were restricted to alpine regions, including 11 species of the genus Erebia, and five of these were only found above 1,500 m.a.s.l. (Aricia artaxerxes, Boloria pales, Erebia epiphron, E. gorge, E. pandrose) (ESM, Appendix 1, Table S1). The number of butterfly species linearly declined with increasing altitude (Fig. 1a). Flower cover was positively correlated with butterfly richness (Table 1). At 2,000 m.a.s.l. we recorded on average half of the butterfly species compared to species richness found at 600 m.a.s.l. Butterfly abundance was not significantly related to altitude and flower cover (Table 1). Further, Moran’s Index was not significantly different from zero for the tested distance classes 1–5 km with intervals of 1 km, indicating that our data set was not spatially autocorrelated. In contrast to the butterfly species richness, our analysis of butterfly community density showed no altitudinal pattern (Fig. 1b). The temperature decreased by 0.45 °C per 100 m of altitude during the summer months (Fig. 2).
Oecologia Fig. 1 a Species richness of butterfly species (y = 28.9 − 0.009x) decreased with altitude (n = 33 study sites). The size of the filled circles in the butterfly graphic corresponds with flower cover (minimum = 0.41 %, maximum = 4.39 %). See Table 1 for statistical details. b Community density of butterflies (field data) did not change along the altitudinal gradient (F1, 32 = 2.53, r2 = 0.07, P = 0.12)
Table 1 General linear models output (Type III sums of squares) with butterfly species richness and abundance as response variables and flower cover and altitude as explanatory variables (see Fig. 1a) Response
Predictor
df
F
P
Butterfly richness
Flower cover Altitude
1, 30 1, 30
5.65 18.01
0.02
