Metapopulation processes in epiphytes inferred from patterns of regional distribution and local abundance in fragmented forest landscapes



    1. Department of Plant Ecology, Evolutionary Biology Centre, Uppsala University, Villavägen 14, SE-752 36 Uppsala, Sweden, and *Department of Mathematics and Statistics, PO Box 68, FI-00014 University of Helsinki, Finland
    Search for more papers by this author

    1. Department of Plant Ecology, Evolutionary Biology Centre, Uppsala University, Villavägen 14, SE-752 36 Uppsala, Sweden, and *Department of Mathematics and Statistics, PO Box 68, FI-00014 University of Helsinki, Finland
    Search for more papers by this author
  • and * HÅKAN RYDIN

    1. Department of Plant Ecology, Evolutionary Biology Centre, Uppsala University, Villavägen 14, SE-752 36 Uppsala, Sweden, and *Department of Mathematics and Statistics, PO Box 68, FI-00014 University of Helsinki, Finland
    Search for more papers by this author

Swantje Löbel (tel. +46 18 4712870; fax +46 18 553419; e-mail


  • 1Deciduous trees within a coniferous forest landscape provide habitat for many organisms. Single trees in deciduous forests form dynamic patches that emerge, grow and fall, but the stands themselves are also dynamic patches established after disturbances and replaced by conifers during succession. Increased dispersal distance, as imposed by landscape fragmentation, may lead to extinctions and reduced biodiversity among species dependent on this dynamic network.
  • 2We analysed regional frequency distributions, local abundances and spatial occupancy patterns of epiphytic bryophytes in 135 deciduous forest stands in a fragmented landscape in Sweden. We used generalized non-linear models to test whether these patterns could be assigned to metapopulation dynamics of individual epiphytes by investigating the relative importance of stand size, habitat quality, connectivity and landscape history on species occupancies and local abundances.
  • 3Most asexually dispersed species were regionally rare, and spatial species occupancy patterns suggest that this is caused by dispersal limitation. In sexually dispersed species, a strong rescue effect was indicated by a bimodal frequency distribution of the species, as well as by increasing local abundance with increasing connectivity to stands present today, or some decades ago. There was a strong positive relationship between regional frequency and local abundance of the species, and between species richness and forest stand size. Vicinity to forest edge negatively affected the local abundance of most species.
  • 4Our results clearly indicate a metapopulation structure. Sensitivity of epiphytes to habitat fragmentation is caused by decreasing forest sizes, habitat alteration at forest edges and increasing dispersal distances. Even in assumed good dispersers, increasing distances can significantly alter regional dispersal processes. A lower rescue effect leads to smaller stand population sizes with a larger extinction risk. Rapid reduction of the amount of habitat during the last decades and the expected time-lag in species extinctions suggest that epiphytes will further decline in the future, although there may still be time for restoration programmes to prevent extinction.


Habitat loss and fragmentation are considered to be major threats to biodiversity (Heywood 1995). Forest ecosystems all over the world have been subjected to intensive exploitation, and in present-day landscapes many forest species are restricted to remnants of a formerly more-or-less contiguous forest cover. In Fennoscandia, modern forestry has led to a great decrease of old-growth forests (Berg et al. 2002), with the lack of old trees and a low proportion of deciduous trees being regarded as particularly important causes of the decline of many populations of forest plants and animals (Berg et al. 2002). Many species need to track the dynamic patches of trees that emerge, grow and fall for their metapopulations to survive at the stand level (Snäll et al. 2003, 2005a). Because the typical successional trend in the region is from deciduous trees, which establish in great numbers after disturbances, to conifers (Engelmark & Hytteborn 1999), even deciduous forest stands can be thought of as dynamic patches in the coniferous forest landscape. The long-term metapopulation persistence of species depends on dispersal across such landscapes (Snäll et al. 2005a) and the increased dispersal distances that result from landscape fragmentation may lead to species extinctions and reduced biodiversity.

Metapopulation theory has increased our understanding of the dynamics of species in fragmented landscapes, and has become an important tool for nature conservation. According to this theory, the most critical aspects of habitat fragmentation are reduced patch sizes and increasing distance between habitat patches (Hanski & Gaggioitti 2004). However, empirical data suggest that the reduced habitat quality of small patches and edge effects may be more important than area or distance effects per se (Harrison & Bruna 1999; Moen & Jonsson 2003). In addition, metapopulation persistence of many species is likely to be significantly affected by the dynamics of their patches (Snäll et al. 2005a, b).

The effects of habitat fragmentation at the community level are poorly understood (Debinski & Holt 2000), and may depend on the taxonomic group, spatial scale and habitat type (Gonzalez et al. 1998; Golden & Crist 1999). Studies on short-lived animals (Gonzalez et al. 1998) and epiphyllous plants (Zartman 2003) have suggested that altered metapopulation processes in fragmented landscapes significantly change species composition and diversity. The effects of habitat fragmentation on the metapopulation dynamics of different community members may depend on species traits which determine dispersal ability (Herben & Söderström 1992; Hanski 1999). Further complexity is added by the fact that the time-lag between passing the extinction threshold and regional metapopulation extinction can be long, implying that the effects of habitat fragmentation may not be detected until many years after they have been triggered (Ovaskainen & Hanski 2002). Epiphytes, which have been shown to be resistant to stochastic local extinctions (Snäll et al. 2005a), may show such behaviour (Snäll et al. 2004b; Berglund & Jonsson 2005).

The core–satellite hypothesis (Hanski 1982; Hanski & Gyllenberg 1993) suggests that the importance of metapopulation processes in natural species assemblages can be assessed from regional frequency distributions and local abundance patterns among the species. It predicts that, within a region, most species are either core species (abundant within local patches and occurring in > 90% of all patches) or satellite species (sparse within patches and occurring in < 10% of all patches) (Hanski 1982). This bimodality is explained by including a rescue effect, whereby recurrent immigration reduces the rate of local extinction (Hanski & Gaggiotti 2004). Empirical studies that actually link the metapopulation dynamics of individual species with patterns of regional frequency and local abundance are largely missing (but see Gonzalez et al. 1998). Inferences about metapopulation processes can, however, be made from spatial species occupancy patterns by applying the incidence function model approach (Hanski 1994), and rescue effects may be detected by analysing the impact of patch connectivity (sensu Hanski 1999) on local species abundance.

In the boreo-nemoral and boreal forest regions, epiphytic bryophytes are confined to deciduous trees with a nutrient-rich bark, a habitat that is both declining and dynamic. Epiphytic bryophytes are an important target group for nature conservation: some of them are threatened (Gärdenfors 2000), while others are used as indicators of the presence of red-listed species (Nitare 2000). Diaspores are produced both sexually and asexually, with the latter type commonly distinctly larger and, thus, expected to have shorter dispersal distances. Many epiphytic bryophytes are sensitive to habitat alteration (Snäll et al. 2003), and poor dispersal has been indicated by several studies (Snäll et al. 2004a, 2005a). Regional studies have shown that the probability of occurrence of some epiphytes can largely be explained by the age of forest stand patches and by connectivity to dispersal sources in the past landscape (Rose 1992; Snäll et al. 2004b). Many authors have suggested that competition among bryophytes is of minor importance for community structure and rarely leads to species exclusion (Rydin 1997; Kimmerer & Driscoll 2000; Ojala et al. 2000; Löbel et al. 2005). This, together with much unoccupied space on host trees, suggests a ‘non-interactive’ community structure. So far, most research has focused on single epiphyte populations, with studies at the community level being rare. In addition, studies have tested the effects of dispersal on species occupancies, but not on local species abundances. This is unfortunate because declining local abundances may be detected long before species extinctions from forest patches, and thus may serve as an early warning system.

Here, we first test whether regional frequency distributions and local abundances of epiphytic bryophytes confirm predictions from metapopulation theory. Second, we test the relative importance of forest stand size, quality and connectivity on species occupancy patterns and local abundances in order to determine whether these patterns can be attributed to dispersal processes and metapopulation dynamics of single epiphyte species. Third, we assess possible time-lags in species responses to fragmentation by testing the relative effects of the present and past landscape structure on species occupancies and local abundances.

Materials and methods

study system

We considered all epiphytic bryophytes that specialize on deciduous trees with a nutrient-rich bark (Acer platanoides, Fraxinus excelsior, Populus tremula, Prunus padus, Quercus robur, Salix caprea, Tilia cordata and Ulmus glabra; hereafter ‘host trees’). Some of the studied epiphytes occasionally also grow on shady boulders enriched by litter of these trees, but such occurrences were not included. The bark of other tree species within the region, namely Picea abies, Pinus sylvestris, Betula pubescens and Alnus glutinosa, has distinctly lower pH, as well as low water capacity, low nutrient content and the presence of tannin and resin (Barkman 1958) and does not therefore support these epiphytes. However, even host trees differ in their habitat quality, which generally increases with increasing bark pH (e.g. Löbel et al. 2005). Our measurements show that surface pH is highest in Acer (mean 6.4, SD 0.6, range 5.1–7.5, n = 29), intermediate in Fraxinus (mean 6.1, SD 0.5, range 4.9–7.6, n = 689), Ulmus (mean 6.0, SD 0.5, range 5.0–7.5, n = 210) and Populus (mean 6.1, SD 0.3, range 5.5–6.8, n = 14), and relatively low in Tilia (mean 5.4, SD 0.5, range 4.3–6.8, n = 60), consistent with previous rankings, and all higher than values reported for Quercus robur (Barkman 1958; Sjörs 1971). Host tree species, however, also differ in longevity: in Sweden (Anderberg & Anderberg 2005) Quercus > Tilia > Fraxinus > Ulmus (often several hundred years) > Acer (rarely more than 150 years) > Populus (stem longevity is rarely more than 100 years). Within the coniferous forest landscape, therefore, host tree stands form distinct habitat patches for epiphytes.

Our study was conducted in the boreo-nemoral forest region in eastern Sweden (Fig. 1; 60°05′ N, 18°20′ E). Today, the landscape consists of a mosaic of managed and unmanaged forests with a total of 135 host tree stands, ranging in area from 0.01 to 15 ha. During the 16th−18th centuries the area was more densely inhabited, mainly by crofters. In the 19th century many fields were abandoned and developed into forests. Forestry was extensive until the 1970s. Thereafter, many old deciduous forests have been cut and replaced by forests dominated by Picea abies (Eriksson 1997). However, compared with other regions in Sweden the area is still relatively rich in deciduous host trees. In 1997, the conservation authorities mapped all deciduous forest stands within the region and evaluated their nature conservation value (Eriksson 1997). Using these maps and infra-red aerial photographs, we identified stands of host trees. These were visited and positioned by GPS (Global Positioning System). We used GPS-coordinates, the aerial photographs and stand maps when digitizing the exact stand borders in a GIS (Geographic Information System, ArcGis 9.0). We considered all stands where the host trees had a minimum diameter at breast height (1.3 m, d.b.h.) of 15 cm as potentially suitable forest patches. Isolated groups of host trees were considered as separate stands if they were separated by at least 25 m of wetland or forests with Picea abies, Pinus sylvestris, Betula pubescens or Alnus glutinosa.

Figure 1.

The study landscape with the surveyed deciduous forest stand patches and Fraxinus stands in 1977, and its location in eastern Sweden.

forest stand characteristics

We calculated the total stand area A (ha), the stand perimeter O (m), the coordinates of the stand centroid, and the edge-to-edge distance to all other host tree stands for each suitable forest stand using the GIS. We used the shape index R proposed by Patton (1975) and Laurance & Yensen (1991) to describe forest stand shape:

R = O/200(Aπ) 1/2.(eqn 1)

The higher R the more irregular is the forest stand shape; R = 1 if the stand is circular. The amount of edge habitat increases with increasing R for a given stand area (Laurance & Yensen 1991).

In each stand, all host trees of ≥ 15 cm d.b.h. were counted, and the tree species was noted. We measured the d.b.h. of all host trees in stands with ≤ 100 host trees, of every 5th tree in stands with 100–1000 host trees, and of every 10th tree in stands with > 1000 host trees: the smallest forest stand had three host trees and the largest 1614. Single deciduous trees surrounded by conifers were rare, and those that were found had few, if any, epiphytic specialists.

In all stands, we placed eight circular sample plots (7 m in radius) in a regular grid with a mesh size of L = (A/8)1/2, where A is the stand area in m2 (Anon. 1999); the smallest mesh size allowed for was 14 m. The grid was randomly projected on the map. We recorded several environmental variables in each plot to describe forest stand quality: we estimated soil moisture on a four-level ordinal scale (Anon. 1997). Because shady, large boulders enriched by host tree litter are a suitable substrate for some of the epiphytes (see above), we estimated the cover of boulders (%), and noted the dominant boulder size class (three height classes: 0–0.5 m, 0.5–1.5 m, > 1.5 m). The total basal area of trees ha−1 was measured with a relascope (a standard sighting instrument used in forestry to measure tree basal area). We further noted the height (cm) and cover (%) of the shrub, field and ground layer, and the type of understorey vegetation; but as none of epiphyte species was affected by these variables, they are not reported.

The historical landscape was inferred from infra-red aerial photographs taken in 1977. It was difficult to separate most host tree species from deciduous species that do not support epiphytes (Betula, Alnus), but it was possible to distinguish Fraxinus excelsior. We therefore used the 1977 distribution of Fraxinus to indicate patches suitable for the studied epiphytes at that time (Fig. 1). Although this does not provide a complete picture of the historical landscape structure, Fraxinus (together with Populus) was the most common host tree and (together with relatively uncommon Acer) the species with highest occupancy of obligate epiphytes (Snäll et al. 2004b). There is also a clear positive correlation (not shown) between the forest stand quantities for all trees used in the analysis and the corresponding quantities for Fraxinus. From the photographs, the stand area and the average height and crown diameter of Fraxinus in 1977 were recorded for each stand present in 1977. The Fraxinus volume and density were calculated based on these variables (see Snäll et al. 2004b for details). As indicated in Fig. 1, the area of deciduous trees has declined drastically since 1977.

regional frequency, occupancy and local abundance of epiphytes

We define the ‘regional frequency’ as the number of forest stand patches occupied by a species and these can be summarized in a ‘regional frequency distribution’. ‘Species occupancy’ describes the spatial distribution of occupied forest stand patches by a species (presence/absence). ‘Local abundance’ refers to the number of host trees with the species within a stand.

We calculated two measures of local abundance for each forest stand. The first measure (‘absolute local abundance’) is the number of occupied host trees in a stand. In stands with ≤ 100 host trees, we counted all occupied trees. In stands with > 100 trees we noted species presence/absence for each 5th tree, in stands with > 1000 trees for each 10th tree, and extrapolated the number of occupied host trees to the whole stand. The absolute local abundance scales the rate of diaspore output (emigration) from a stand. The second measure (‘relative local abundance’) is the proportion of occupied trees in a stand. It is determined by immigration from surrounding stands and by colonization–extinction dynamics among trees within stands.

To test for generalities in species occupancies, regional frequencies and local abundances of species with similar dispersal strategies, we grouped the epiphyte species into two main dispersal strategy groups. We distinguished between species with predominantly sexual dispersal (spores: mainly monoecious species) or asexual dispersal (diaspores, such as gemmae or gemmae-like branchlets, or fragments: mainly dioecious species). Sexual dispersal does occasionally occur in species with predominantly asexual dispersal, and vice versa. Critical specimens were collected and identified by microscopy. Nomenclature follows Söderström & Hedenäs (1998).

statistical analysis

Species–area relationships, regional frequency distributions, distribution–abundance relationships

We investigated the species–area relationship by fitting linear regression models predicting species richness as a function of forest stand area and number of host trees in a stand. Preliminary analysis suggested an exponential relationship between species number (S) and area (A); therefore, all variables were log-transformed (ln S = z ln A + c). To disentangle the effects of habitat area and habitat diversity on species richness, we further fitted a multiple regression model, including both host tree number and host tree species richness (as a measure of habitat diversity).

The relationship between regional frequency and local abundance of the species (sometimes referred to as ‘distribution–abundance relationship’, e.g. Hanski 1999, or ‘abundance–occupancy relationship’, e.g. Gaston et al. 1997, 1998) was tested by fitting linear regression models predicting regional frequency as a function of mean (absolute) local abundance of the species. Both variables were log-transformed. We also tested the logistic model proposed by Hanski & Gyllenberg (1997).

Patterns in the regional frequency distribution of the species were assessed by plotting the number of species occurring in 0–10%, 10–20%, … , 90–100% of all stands. We used Tokeshi's (1992) test for bimodality to determine whether species frequency distributions were significantly uni- or bimodal:

image(eqn 2)

where N is the total number species, i and j are the number of species in the left-most (nl) and right-most (nr) frequency classes, respectively, and h is the frequency interval.

Bimodality is judged based on the probability of obtaining the observed number of species in the rarest species group, P0−10% (‘satellite species’, Hanski 1982) and commonest species group P90−100% (‘core species’) under the null hypothesis of a random distribution. If P0−10% < 0.25 and P90−100% < 0.25, the frequency distribution is bimodal. P0−10% < 0.05 and P90−100% < 0.05 indicates a strong bimodal pattern. If P0−10% < 0.05 and P90−100% ≥ 0.5, or vice versa, the frequency distribution is unimodal (Tokeshi 1992). Because the frequency classes by which core and satellite species have been defined by Hanski (1982) are somewhat arbitrary, we experimented with their limits. The judgement of bimodality, however, was quite robust to choosing a broader (15%, 85%) or smaller limit (5%, 95%) for the frequency classes.

Species occupancy and relative local abundance

We tested the effects of host tree number, maximum tree diameter, patch quality, forest edge, present connectivity, stand size and Fraxinus density in 1977, and historical connectivity on both species occupancy and species relative local abundance (for 14 and 20 species, respectively – the other species were either too rare or too common for modelling). We chose the variables describing the forest stand habitat first – as they should have the most direct effects – followed by the present connectivity variable, and finally the variables describing the historical landscape. The order of inclusion of variables was the same for all species. We experimented with this order, but it did not significantly change our results. Variables were tested for each species separately. Although all significant environmental variables together (host tree species composition, total basal area of trees, soil moisture and frequency of large boulders) can be combined to illustrate ‘patch quality’, their effects were modelled separately. ‘Forest edge’ refers to the effect of the shape index, R, reflecting the irregularity of the forest stand.

We used generalized linear models (GLM) with a binomial error distribution and a logit link function (McCullagh & Nelder 1989). Multiple starting models were built using the recorded stand variables as predictors (host tree number, maximum tree diameter, patch quality variables, forest edge). We included biologically reasonable two-way interactions, but none of them was significant. Starting models were simplified using stepwise variable selection minimizing the AIC (Akaike's information criteria, Akaike 1974), defined as AIC =−2l + κp, where l is the maximized log likelihood, p is the number of parameters in the model, and κ was set to 4, which is equivalent to the use of p = 0.05 (McCullagh & Nelder 1989). Models predicting relative local abundance were generally overdispersed. We therefore estimated the dispersion parameter (McCullagh & Nelder 1989) from the data. We used likelihood ratio tests (based on the χ2 statistic for binomial models and on the F-statistic for quasibinomial models) for significance.

We then tested two different model extensions to account for the present and historical connectivity, respectively. We first tested whether the present connectivity (sensu Hanski 1999) to surrounding stands improved the model. Connectivity was defined as

image(eqn 3)

The indices i and j refer to the focal and surrounding stands, respectively. pj = 1 if the species occurred in the surrounding stand j, and pj = 0 if it did not occur. The effect of the surrounding stand is quantified by the log-normal function of the distance dij (m) between the centroids of the stands i and j. The tail of the log-normal function is fatter than that of the negative exponential function used by Hanski (1999), i.e. it predicts more frequent long-distance dispersal events. Functions similar to the log-normal function have been found to fit better to the dispersal of plants (Clark et al. 1999). We also tested the edge-to-edge distance between stands, but it explained slightly less variation. The maximum distance dij was 500 m as we did not have information about species occurrences beyond 500 m of the periphery of the investigated stands. The parameter α controls the rate of decay, γ is a scaling parameter and Qj is the absolute local species abundance in stand j. The parameters α and γ were estimated from the data. Again, we used likelihood ratio tests (based on the χ2 statistic for binomial models, and on the F statistic for quasibinomial models) to judge the significance of stand connectivity. Because of the non-linearity, however, we did not calculate exact P values for the connectivity variable (Snäll et al. 2003). The expanded models including the present connectivity variable thus were generalized non-linear models.

Second, to analyse the effect of the past landscape structure, we tested whether the stand area, the Fraxinus density in 1977 or the Fraxinus volume of the stand in 1977 improved the models significantly. We also fitted a historical connectivity variable in the same way as the present one; in this case Qj denotes the area in 1977 of the surrounding stand j, and pj is set to 1.

We used the free software R 1.8.1 (R Development Core Team 2004) with the add-on libraries geoR version 1.4–5 (Ribeiro & Diggle 2001), and MASS version 7.1–13 (Venables & Ripley 1999) for the statistical analyses. We wrote our own function to fit the connectivity variables (Ki). This function combines the standard function for fitting generalized linear models (glm[]) with an optimization function (optim[]). The optimization function searches among (i.e. tests) different sets of α and γ, aiming at finding the set that minimizes model deviance. For each set tested Ki is calculated. Next, Ki is added as a term in the model and this model is finally fitted using glm[]. The set of α and γ reported is thus the set that gives lowest model deviance.


speciesarea relationships, regional frequency distributions, distributionabundance relationships

We recorded a total of 28 obligate epiphytes, 14 sexually and 14 asexually dispersed species. Forest stand area explained a large part of the variation in species richness among stands (Fig. 2a, inline image = 0.41). However, the explanatory power of the model relating species richness to the host tree number was even higher (Fig. 2b, inline imageinline image = 0.57). The multiple regression model, including both host tree number and host tree species richness, did not have a higher explanatory power (inline image = 0.57); host tree species richness did not have a significant effect on epiphyte richness (P = 0.41).

Figure 2.

Relationship between (a) obligate epiphytic bryophyte species richness and forest stand area (ha) and (b) obligate epiphytic bryophyte species richness and host tree number of forest stands (ln S = z ln A + c in each case). 95% confidence intervals of the models are shown.

We observed a strong positive relationship between regional frequency and absolute local abundance (Fig. 3, linear regression, log-transformed) for both sexually and asexually dispersed epiphytes. The logistic model gave only a slightly better fit (R2 = 0.95, P < 0.001). Regression lines of sexually and asexually dispersed species did not differ significantly.

Figure 3.

Relationship between the regional frequency and local abundance of the species with the natural logarithm of average absolute local abundance plotted against the natural logarithm of number of occupied forest stand patches. 95% confidence interval of the linear regression model is shown. The logistic model only gave a slightly better fit (R2 = 0.95, P < 0.001).

The regional frequency distribution of the species was bimodal, with eight species occurring in fewer than 10% of the stands, and five species occupying more than 90% of all forest stand patches (Fig. 4a). In separate analyses of the dispersal types, a strong bimodal frequency distribution pattern was observed for sexually dispersed species (Fig. 4b), whereas asexually dispersed species showed a significantly unimodal pattern (Fig. 4c).

Figure 4.

(a) Regional frequency distribution of (a) all obligate epiphytic bryophytes, (b) sexually dispersed epiphytic bryophytes and (c) asexually dispersed epiphytic bryophytes. P-values refer to the probability of obtaining the observed species number in the left-most (satellite species, P0−10%) and right-most (core species, P90−100%) frequency classes under the assumption of a random distribution (see Statistical analysis).

species occupancy patterns and relative local abundances

Generalized non-linear models of species occupancy among stands indicated differences in dispersal abilities among sexually and asexually dispersed epiphytes. Occurrence probabilities of most species increased with the number of host trees in a stand. The maximum tree diameter tended to be more important in asexually (effects in six of eight species, Fig. 5a) than in sexually dispersed species (effects in three of six species, Fig. 5b). In both groups, it was the only significant variable for species with low regional frequency (Orthotrichum pumilum, Anomodon longifolius). Species with occupancy too low to be statistically analysed occurred in stands with largest maximum tree diameter, and highest Fraxinus density in 1977.

Figure 5.

Generalized non-linear models of species occupancy of (a) sexually dispersed epiphytic bryophytes and (b) asexually dispersed epiphytic bryophytes. The reductions in deviance (%) for adding a variable (in the order from left to right) to the model are shown. Only significant variables (P ≤ 0.05 in likelihood ratio tests) were included in the models and presented in the graphs. The bar pattern indicates whether a variable had a positive or negative effect; open bars indicate nominal variables. The patch quality bar indicates the variation explained by the significant environmental variables as a whole (i.e. host tree species composition, total basal area of trees, soil moisture and frequency of large boulders combined), although their effects were tested separately. Forest edge refers to the effects of a shape index, R, reflecting the irregularity of the forest stand shape. As last variable, we added either the present stand connectivity or the Fraxinus density and historical connectivity in 1977.

Among sexually dispersed species, patch quality explained a considerable part of occurrence probabilities of Homalia trichomanoides, Neckera pennata, Frullania dilatata and Lejeuna cavifolia. The most important patch quality factors were the total basal area of trees and the Fraxinus density. Patch quality was relatively unimportant for most asexually dispersed species. Positive effects of forest edge (irregularity of the forest stand shape) were observed for the occurrence probability of Ulota crispa (Fig. 5a). Present stand connectivity significantly enhanced occurrence probabilities of six out of eight asexually dispersed species (Fig. 5b), but only of one out of six sexually dispersed species (Neckera pennata, Fig. 5a). For most asexually dispersed species, best fit was given by α = 0.19, which implies that the spatial scale of the aggregated structure was less than 200 m. For the sexually dispersed Neckera pennata, the best fit was given by α = 0.09, corresponding to a spatial aggregation structure up to about 450 m. Either the historical stand connectivity (0.05 ≤ α ≤ 0.07) or the Fraxinus density in 1977 was important for occurrence probabilities of four out of six sexually dispersed species (Fig. 5a, Homalia, Neckera pennata, Ulota, Frullania), but only two out of eight asexually dispersed species (Fig. 5b, Platygyrium repens, Homalothecium sericeum).

When relative local abundance was considered, different sexually dispersed species were affected by different factors, and often several factors were important (Fig. 6a). In many asexually dispersed species, however, maximum tree diameter was the most important variable (Fig. 6b). Patch quality was important for several species in both sexually and asexually dispersed groups, with the most important variables (not presented separately in Fig. 6) being total basal area of trees (Isothecium alopecuroides, Homalia, Metzgeria furcata, Neckera complanata) and frequency of large boulders (Lejeuna, Neckera complanata, Antitrichia curtipendula). In sexually dispersed species, the density of Fraxinus further enhanced relative local species abundances (Orthotrichum speciosum, Radula complanta, Pylaisia polyantha, Homalia, Neckera pennata, Frullania). Forest edge had negative effects on the relative local abundance of Homalia, Lejeuna (Fig. 6a), Anomodon, Metzgeria and Neckera complanata (Fig. 6b), and a positive effect on that of Ulota (Fig. 6a). Present stand connectivity was significant in six out of 10 sexually dispersed species (Fig. 6a, 0.01 ≤ α ≤ 0.4), but only in four out of 10 asexually dispersed species (Fig. 6b, 0.15 ≤ α ≤ 0.26). For some sexually dispersed species, historical connectivity was even more important (Neckera pennata, α = 0.13; Lejeuna, α = 0.10). The Fraxinus density in 1977 enhanced relative local abundances of Platygyrium, Homalothecium and Anomodon (Fig. 6b).

Figure 6.

Generalized non-linear models of relative local abundance of (a) sexually dispersed and (b) asexually dispersed epiphytic bryophytes. Conventions are as in Fig. 5.


metapopulation processes in the fragmented forest landscape

Our data strongly suggest that metapopulation processes are important for the structure and diversity of epiphyte communities, and that even common species with high dispersal abilities display a metapopulation structure. The first support comes from the positive relationship between regional frequency and local abundance of the species and bimodal frequency distributions (Hanski 1982; Gotelli 1991; Hanski & Gyllenberg 1993; Gonzalez et al. 1998; Ovaskainen & Hanski 2004). Further evidence is provided by the effects of stand connectivity on species occupancies (Hanski 1994) and local abundances (Gaston et al. 1997). The strong species–area effect confirms the suggestion of Cook et al. (2002) that classical island theory remains an appropriate tool to study diversity patterns in fragmented landscapes, provided that only habitat specialists, which do not occur in the matrix, are considered. By contrast, at the scale of single trees, we observed only a very weak relationship between host tree size (diameter) and species richness of epiphytic bryophyte specialists (Löbel et al. 2005). This, and the lower explanatory power of forest stand area compared with host tree number, supports metapopulation dynamics as a likely explanation for the observed strong species–area relationship. The observed z-values lie towards the lower end of the range of z-values reported by many studies (0.2–0.3, e.g. Gilbert 1980). Similar to Rydin & Borgegård (1988), we did not find a z-value differing significantly from the theoretical value of 0.263, which would be obtained under equilibrium conditions (Preston 1962; MacArthur & Wilson 1967), even though our system is probably not in equilibrium. The low effect of host tree species richness (as a measure of habitat diversity) on epiphyte richness may be explained by the fact that all the studied bryophytes occur on all host species, and ‘prefer’ the same tree species.

Mechanisms underlying frequency distributions and relationships between regional distribution and local abundance are poorly understood, and are difficult to separate from each other (e.g. Gaston et al. 1997, 1998; Warren & Gaston 1997; Holt et al. 2002; McGeoch & Gaston 2002). Our study possibly provides the first example for a plant species assemblage where the patterns are most probably caused by metapopulation processes: metapopulation dynamics of epiphytic bryophytes are not only suggested by our analysis of species occupancy and local abundance patterns, but also by a long-term study of the colonization–extinction dynamics of the epiphytic moss Neckera pennata (Snäll et al. 2005a). Furthermore, the niche breadth hypothesis (Brown 1984) is very unlikely to explain species frequency and local abundance patterns in the epiphyte community: habitat quality explained less variation in species occupancy patterns of asexually than sexually dispersed epiphytes, suggesting a lower degree of habitat specialization of the first. This is supported by a study on spatial epiphyte richness patterns at a local, within-stand scale (Löbel et al. 2005), and germination experiments in the laboratory (S. Löbel, unpublished data).

The effect of stand connectivity on species occupancy indicates that regional rarity of most asexually dispersed species is a consequence of restricted dispersal (Hanski & Gagiotto 2004). The high a-values suggest that distances up to a few hundred metres between forest stands are large enough to prevent successful colonization. This is supported by dispersal experiments (Walser et al. 2001; Dettki & Esseen 2003), by a study of colonizations of trees (Snäll et al. 2005a), and by the combined results of studies of occupancy patterns (Snäll et al. 2003) and of spatial genetic structuring (Snäll et al. 2004a) in a particular landscape. An alternative explanation for the regional rarity of these species may be habitat specialization, as predicted by the niche breadth hypothesis (Brown 1984). This is, however, unlikely because patch quality explained less variation in species occupancy patterns of asexually than of sexually dispersed epiphytes. Stand connectivity did not influence species occupancies of most sexually dispersed species. However, for core species, we also observed significant effects of stand connectivity on relative local species abundances. This, and the bimodal species frequency distribution in sexually dispersed species, indicates that dispersal among forest stand patches enhances local population sizes within stands (rescue effect). Major impacts of habitat fragmentation on local abundances have been also suggested for epiphyllous bryophyte communities in rain forests (Zartman 2003). By contrast, the rescue effect is likely to play only a minor role in species with low dispersal ability, as supported by the small effect of connectivity on local abundance in asexually dispersed species.

Variables related to the historical landscape and stand age, e.g. the maximum tree diameter, were particularly important in explaining the occupancy and abundance of rare species, and species used as indicators for old-growth forests (Nitare 2000). This suggests that colonizations of the stands by these species occurred long ago, and that their local abundance was elevated by immigration, with both processes driven by dispersal from stands that were nearby in the past. Similar arguments have been put forth in studies of individual species (Snäll et al. 2003, 2004b), but in the present study we have identified a whole group of such species. We expect there to be long time-lags to extinction in these bryophytes (Ovaskainen & Hanski 2002), although the fact that ‘stand age’ was the only significant variable for occupancies of rare species requires attention. Dispersal events in these species seem to be rare in the present landscape, and regional extinctions are likely when the few occupied old trees fall.

the role of dispersal in the dynamic forest landscape

Metapopulation processes take place at two spatio-temporal scales: forest stands constitute dynamic patches at the landscape level and single trees within these also constitute dynamic patches (Snäll et al. 2005a). A full understanding of this system thus requires considering metapopulation dynamics on both spatial scales, as stressed by Holt (1992), as well as considering the dynamic nature of patches at both spatial scales. Demographic differences among host tree species add further complexity, and the longer expected lifespan of trees with ‘low-quality’ bark (Quercus, Tilia) could be another explanation why the density of ‘high-quality’ trees (Acer, Fraxinus) did not have an effect on occupancy and abundance of many asexually dispersed epiphytes. However, even at the within-tree level, there are local spatial dynamics: each tree grows through time, adding and losing limbs, and bark structure changes. Our own unpublished data indicate that stochastic extinctions of small epiphyte colonies from parts of the stem of single host trees, caused by small-scale bark-loosening, are common. Hence, local population dynamics on single host trees are further affected by the dynamic habitat of the epiphytes.

The total number of host trees, which determines the within-stand metapopulation carrying capacity (Hanski & Ovaskainen 2000), had a significant positive effect on occurrence probabilities of most species (Fig. 5), and explained a large part of the variation in species richness among stands (Fig. 2b). Tree falls can be the main cause of epiphyte extinction from individual trees (Snäll et al. 2005a), and they may even cause species extinction at the stand level. Generally, we expect lower epiphyte extinction probabilities from well-connected forest stands with high relative local abundances and from large stands with a high number of suitable host trees and thus a large local metapopulation size (Ovaskainen & Hanski 2004). Dispersal within forest stands, establishment of new host trees and recurrent colonization of trees from surrounding stands may promote metapopulation persistence in the stand over quite long time periods (Snäll et al. 2005a). However, as deciduous trees are replaced by conifers during succession, the long-term metapopulation persistence over large spatial scales depends on the establishment of new deciduous stands (Snäll et al. 2005a,b).

impact of habitat fragmentation on local processes

Several studies have provided evidence that changes in patch quality related to patch size and edge effects have important impacts on local processes in fragmented landscapes (e.g. Esseen & Renhorn 1998; Fahrig 1998; Debinski & Holt 2000; Harrison & Bruna 1999; Moen & Jonsson 2003). Different authors have suggested major effects of environmental factors on metapopulation dynamics in single epiphyte species (Heegaard & Hangelbroek 1999; Snäll et al. 2003, 2005a). In addition, epiphytes have been proposed as useful indicators of forest edge effects (Esseen & Renhorn 1998). Our study confirms this: occurrence probabilities and relative local abundances of many obligate epiphytic bryophytes were affected by patch quality, especially by the total basal area of trees. This and the study of Snäll et al. (2003) show that many obligate epiphytes are favoured by shady conditions, which we expect to be negatively correlated with the edge/interior ratio of stands. Furthermore, many species had lower local species abundances in irregular forest stands. Effects of forest edges are known to be complex, and it may be difficult to separate edge from size effects (Moen & Jonsson 2003). However, adding the shape index as first variable to our models did not significantly change the results of our analysis. As expected, it slightly lowered the effects of some environmental variables considered to be negatively correlated with forest edge, but not that of stand area. Therefore, we are confident that both area and edge are important factors for epiphyte metapopulation dynamics.


Our study suggests that epiphyte communities are sensitive to habitat fragmentation through alteration of both regional metapopulation and local processes. Large, shady stands are important refugia, where local metapopulation dynamics between trees may balance small-scale disturbances and deterministic species extinctions (caused by tree death). With decreasing forest stand sizes, the influx of diaspores from outside becomes increasingly more important for ‘buffering’ deterministic extinctions within the stand. In the long term, however, species are deemed to go extinct even from large forest stands due to natural succession. Thus, dispersal between stands is required for maintenance of species richness in the landscape. Our study showed that habitat insularity significantly alters regional dispersal processes, even of assumed good dispersers. Rapid reduction of the amount of habitat during the last decades and the expected time-lag in species extinctions suggests that most species may further decline in the future. However, the time-lag in epiphyte extinctions also implies that there is still a chance to prevent extinction through restoration programmes.

Current forestry practice promotes conifers. In the long term, conservation management efforts should focus on increasing overall numbers of deciduous trees in the landscape. New forest stands should be established in the vicinity of current stands, especially those with a large number of ‘high-quality’ host trees, namely Fraxinus and Acer (Snäll et al. 2004b). In the short term, the conservation of large, old-growth deciduous forests is important. These will serve as important dispersal sources, which increase overall dispersal rates across the landscape.


We thank Tommy Löfgren at NaturGIS AB for the infra-red aerial photograph interpretation. Bege Jonsson, Robert Freckleton and Bob Holt gave useful comments on the manuscript. Financial support was received from FORMAS.