1. We test how fine-scale (≤ 400 m2) connectivity to conspecifics influences metapopulation dynamics using a field experiment in central Amazonia with the epiphyllous (i.e. leaf-inhabiting) bryophyte Radula flaccida Gott. (Radulaceae). This is a natural model system with spatially structured ephemeral patches at the leaf and phorophyte scale.
2. The aim was to test how the rates of leaf colonization, local abundance, growth and extinction of bryophytes from both leaves and host phorophytes were affected by experimentally manipulated variation in connectivity to conspecifics. We also investigated the relative importance of local stochastic extinctions and deterministic extinctions at both spatial scales.
3. Approximately 3500 leaves on 70 phorophytes were censused six times over a 15-month period by utilizing a repeated block design with control, positive control and three pulse-style reduction treatments, which varied in connectivity to surrounding conspecifics.
4. An increase in stochastic extinction events was only accompanied by a reduction in colonization in the treatment in which focal and neighbouring phorophytes within the 400 m2 plots were experimentally denuded of their natural populations, suggesting that epiphyllous cryptogams are subject to fine-scale (within phorophyte) rescue effects. Negative density-dependent growth was also detected in within-leaf population dynamics, suggesting that resource limitation or intraspecific competition influences local population growth. Finally, stochastic extinctions from viable leaves occurred with nearly the same frequency as deterministic events (e.g. leaf fall), whereas at the phorophyte scale only stochastic extinctions were observed.
5.Synthesis. The experiment demonstrates that rescue effects occur at fine scales even for vagile plant taxa, such as cryptogams, which may inhabit spatially isolated substrates characterized by turnover rates as fast as their own population dynamics. Furthermore, the results highlight the importance of quantifying both stochastic and deterministic extinction modes, as underestimating either of these parameters leads to over optimistic projections of future metapopulation size.
Many species of plants and animals display geographic population structure at some spatial scale resulting from a combination of discontinuous habitat distribution and limited dispersal abilities (Andrewartha & Birch 1954; Hanski & Gilpin 1997). Understanding the nature of the population dynamics within and among habitat patches is a central theme of metapopulation ecology and contributes to understanding how distribution patterns and demographic processes may interact to influence regional persistence (Hanski 1991).
Levins (1969) theorized in his classic metapopulation model that patch colonization rates respond unimodally and extinction rates linearly to increasing the number of occupied patches, assuming that neither propagule rain nor rescue effect contributes to patch colonization–extinction rates, a prediction considered to be biologically inaccurate for many metapopulation systems (Hanski 1982; Gotelli 1991). Indeed, theoretical extensions of Levins’ model subsequently incorporated effects of surrounding occupied patches on colonization and extinction dynamics (Hanski & Gyllenberg 1993; Harding & McNamara 2002; Roy, Harding & Holt 2008) and have been supported by studies of disparate taxonomic groups (e.g. McCauley 1989; Thomas & Hanski 1997; Gonzalez et al. 1998; Forbes & Chase 2002; Snäll, Ehrlén & Hakan 2005). However, to our knowledge, no experimental demographic studies have yet directly tested how changes in neighbouring number of occupied patches (‘connectivity’ henceforth) impact the probability of local population colonization, population growth or stochastic extinction.
Metapopulation theory predicts that the colonization probability of a patch increases with increasing connectivity to surrounding occupied patches, as a result of restricted species dispersal range (Hanski 1999). Moreover, the rescue-effect hypothesis posits that immigration, which increases with increasing connectivity, reduces the probability of local stochastic extinction (Brown & Kodric-Brown 1977). However, studies of colonization and extinction probabilities in relation to connectivity have typically been theoretical or based on observational field data (Hanski 1999; Roy, Harding & Holt 2008), largely due to the demanding spatial and temporal effort necessary for conducting a metapopulation-scale experiment.
A wide range of taxonomic groups are restricted to ephemeral habitats, in which their metapopulation dynamics are generally slower than the turnover rates of the patches they inhabit. Such examples include host–parasite systems (Antonovics 2004), seasonal rock pool communities (Barrett & Husband 1997), intertidal zone mosaics (Paine & Levin 1981) and tree- (Snäll, Ehrlén & Hakan 2005) and leaf-inhabiting cryptogams (Zartman & Shaw 2006). The long-term persistence of such taxa may be influenced as much by patch longevity as by connectivity. However, to empirically disentangle rates of patch loss from stochastic metapopulation extinctions, extensive data sets are generally required. Moreover, distinguishing between the relative importance of stochastic and deterministic extinctions due to deterministic patch destruction is important, as failing to do so may overestimate future metapopulation size (Snäll, Ehrlén & Hakan 2005; Hodgson, Moilanen & Thomas 2009).
Trees constitute habitat patches for many phanerogamic and cryptogamic plant taxa in both temperate and tropical ecosystems (Benzing 1990). While some of these species are typically restricted to the main trunks and branches (epiphytes), many cryptogamic taxa, particularly in tropical regions, are epiphyllic, that is they occupy leaf surfaces (Zartman & Illkiu-Borges 2007). The ephemeral nature of trees (hereafter phorophytes) and leaves, and their well-defined hierarchical spatial structure (e.g. among phorophytes and among leaves within phorophytes), makes them suitable habitats for experimental demographic studies aimed at testing metapopulation theoretic predictions across different spatio-temporal scales.
The main objective of this study was to test theoretic predictions for metapopulation dynamics at two spatial scales using a field experimental approach with the epiphyllous bryophyte Radula flaccida as the model species. We address two main questions: (i) How do reductions in fine-scale (400 m2) connectivity to surrounding species occurrences affect the probability of patch colonization, stochastic extinction and population growth and (ii) What is the relative importance of deterministic extinction resulting from leaf or phorophyte fall and stochastic extinctions from intact leaves or phorophytes on the metapopulation dynamics of the focal species?
Materials and methods
Study area and study species
The field experiment was conducted in the Tupé Sustainable Development Reserve (RDS Tupé) located in the state of Amazonas, Brazil, c. 20 km north-west of the state capital, Manaus, on the northern margin of the Rio Negro (03°02′43.72′′S, 60°15′28.40′′W) (Fig. 1). The RDS Tupé covers an area of c. 12 000 hectares in size and is characterized by upland (terra-firme) and seasonally inundated blackwater (igapô) forest types (Silva 2005). Precipitation in the RDS Tupé is strongly seasonal and typical of areas within the central Amazonian Region. It increases markedly between November and May and declines sharply during the dry season from July to September.
Our model species, Radula flaccida Gott. (Family Radulaceae), is an epiphyllous liverwort of upland tropical forests world-wide (Lücking 1995). It is monoicous (antheridia and archegonia are produced on the same gametophyte) and produces asexual propagules in the form of discoid gemmae arising on the leaf margins. Auto-compatibility has not been verified experimentally for this bryophyte species; however, the prolific production of sporophytes on the same gametophyte (C.E. Zartman., pers. obs.) suggests that gametophytic selfing commonly occurs. R. flaccida is very common in central Amazonian upland forests on the leaves of understorey (≤ 3-m high) trees (Zartman & Ilkiu-Borges 2007), and dispersal of spores and gemmae is most likely augmented by the sporadic and forceful winds and rains common to the region. Both richness and abundance of epiphyllous bryophyte communities of central Amazonia drop off dramatically above 5 metres from the ground (Zartman 2002), most likely due to the extended dry season in the region which typically precludes their establishment in the mid- and upper-canopy strata. Our study species is easily identifiable in the field, eliminating the necessity for destructive sampling for verification. Although previous demographic experiments conducted on this particular species in the context of forest fragmentation suggest that dispersal limitation occurs at the meso-scale (100–1000 m; Zartman & Shaw 2006), we are not presently aware of any other published studies on the metapopulation dynamics of epiphyllous bryophytes.
We established a randomized complete block experimental design with a minimum inter-block distance of 500 m along a 2-km trail. Within each block, five 400-m2 plots were established with a minimum inter-plot distance of 50 m. In each of the 15 plots, 2–5 study phorophytes were selected, giving a total of 70 phorophytes. On the experimental plots, the following pulse-style treatments (Bender, Case & Gilpin 1984) were applied: (i) control – no changes, (ii) positive control – random elimination of unoccupied leaves on non-study phorophytes in a proportion similar to the number of occupied leaves removed in the complete elimination treatment, (iii) half neighbour removal – elimination of 50% of the occupied leaves on the non-study phorophytes, (iv) full neighbour removal – elimination of 100% of the occupied leaves on the non-study phorophytes and (v) complete elimination – removal of only occupied leaves (no unoccupied leaves) on all phorophytes within the study plot (Fig. 2). The positive control was used to test for the microclimatic influences on R. flaccida population dynamics potentially resulting from leaf removal.
To minimize inherent variation in epiphyll density among plots, the areas selected for plot establishment met the criteria of having, prior to treatment, 8–10 understorey phorophytes with ≥ 10 leaves occupied by the focal species. The confounding effects of interspecific competition, as well as oversaturation of available microhabitat space from the beginning of the experiment, were avoided by selecting focal phorophytes, which were only inhabited by the study species on no more than half of the total number of leaves. Visual screening for the occurrence of R. flaccida colonies above the height of the focal phorophytes was conducted in each treatment by climbing nearby trees to 20 m in height and scanning leaves from above with the aid of binoculars to ensure a minimal density of the focal species immediately above the plots.
Species censuses and statistical modelling
Six demographic censuses of R. flaccida were conducted within a 15-month period as follows: July 2008, September 2008, December 2008, February 2009, June 2009 and October 2009. During the course of the experiment, a total of 3542 leaves were assessed on 70 phorophytes of which 1428 were occupied by R. flaccida during at least one census. At each census, the total cover (cm2) of gametophytes (local abundance henceforth) on the marked, occupied leaves was measured by overlaying a transparent paper grid and summing the number of 1-cm squares occupied. In the first and last censuses, all unoccupied leaves were tallied for each focal phorophyte to obtain data for estimating colonization probability. The censuses provided data on four response variables, which were modelled using the random effects generalized linear modelling framework (Pinheiro & Bates 2000), where phorophyte was the random effect.
The first response variable was the probability of bryophyte extinction from intact leaves (control treatment: N = 468; full experiment: N = 3052). In all treatments, with the exception of complete elimination, this variable was measured from the second census onwards for all leaves occupied in censust − 1, which suffered extinction from intact substrates in census. For the complete elimination treatment, this variable was measured for populations, which were colonized after the treatment was applied but subsequently became extinct on intact leaves during the course of the study. We assumed that the probability of stochastic extinction (E) from an intact leaf s in census t from a leaf occupied in t − 1 followed a Bernoulli distribution with mean Es. Specifically,
where κ is the random effect ‘intercept parameter’ for phorophytes, θj is a parameter for the fixed effect of block j in comparison with block one, γk is a parameter for the fixed effect of treatment k in comparison with the control treatment, φ is a parameter that tests for an effect of number of months between censuses, M, and ζ is a parameter that tests for an effect of the natural log of the local species abundance on leaf s, ln(As), in the preceding census t − 1.
The second variable was probability of bryophyte extinction from an intact whole phorophyte (control treatment: N = 60; full experiment: N = 280). The model was similar to eqn 1, except that phorophytes instead of leaves constituted the observation units. This also meant that the model did not include any random effect, only an ordinary ‘intercept’ parameter.
The third response variable was probability of bryophyte colonization of a remaining intact leaf following the respective treatments (control treatment: N = 450; full experiment: N = 2699). The response was modelled as stochastic extinctions (eqn 1), except that the colonization model lacked the two last terms: number of months between censuses (M) and local species abundance (As). This colonization variable made use of the data from the complete tallying in censuses 1 and 6. First, all leaves occupied in census 1 were removed from the data set. Among the remaining leaves, those occupied in census 6 must have resulted either from colonization events of unoccupied leaves present in census 1 or from colonization events of leaves that had emerged between censuses 1 and 6. In effect, we modelled the probability of bryophyte occurrence on leaves among the data collected in census 6, but because these data on leaves occupied in census 1 had been removed, these occurrences were the result of colonization events since census 1. Although this approach may not provide accurate estimates of colonization probabilities, comparisons of colonization probabilities among treatments (Fig. 2) are considered robust.
The fourth response variable was local bryophyte population growth on occupied leaves (control treatment: N = 393; full experiment: N = 2547), which assumed relative exponential growth. It has been previously used in modelling the expansion of individual colonies (ramets) of the epiphytic bryophyte Neckera pennata Hedw (Wiklund & Rydin 2004). Specifically, we calculated a measure of relative abundance growth (or decline) to be modelled from the abundance data as follows:
where Gq is the relative abundance growth rate of a local population on leaf q, Aq,t −1 is the local abundance on leaf q in the preceding census, and Aq,t is the abundance measured at t, the time length in months between the censuses (between 2 and 4 months). We assumed that the monthly relative abundance growth (or decrease) (G, eqn 2) on an occupied leaf q followed a normal distribution with residual variance σ2 and a mean that was modelled using the right-hand side of eqn 1. For all statistical analysis, we used r version 2.10.1 (R Development Core Team 2011) with the add-on library lme4 version 0.999375-34 (Bates, Martin Maechler & Bolker 2010).
Non-manipulated metapopulation dynamics (the control treatment)
Local population extinctions took place nearly as often for stochastic reasons from viable leaves (77/468 = 0.16) as for deterministic reasons as leaves fell off (88/468 = 0.19).
We observed five whole-phorophyte R. flaccida extinctions resulting from checks of fourteen occupied trees sixty times among censuses two to six. One of these whole-phorophyte extinctions occurred from a shrub which became colonized during the experiment, and the other extinctions occurred from phorophytes which were occupied from the start of the experiment. No whole-phorophyte R. flaccida extinctions took place due to phorophyte mortality.
We recorded 71 colonization events in census six among the 450 leaves that were unoccupied or which emerged since census one. The local population abundance on occupied leaves in the first census (N = 80) ranged from 1 to 82 cm2, with a mean (SE) of 9.45 cm2 (16.7). During the course of the experiment, the monthly relative growth on occupied leaves (N = 393) ranged between -1.64 and 0.97, with a mean (SE) of 0.037 cm2 (0.313).
Manipulated metapopulation dynamics (the experiment)
In total, we observed 505 R. flaccida extinction events among the 3052 checks of leaves in census two to six. The block term did not explain any of the modelled responses, and it was therefore not included in the final models that are described later. In comparison with the control treatment, the probability of extinction was higher in the treatment where all occupied leaves on the focal phorophyte and surrounding neighbours within the 400 m2 plots had been removed in the first census (Table 1). The extinction probability also decreased with increasing local R. flaccida abundance on the leaf, and increased the longer time had passed between censuses.
Table 1. The final generalized linear mixed model for probability of stochastic Radula flaccida extinction from an intact leaf. Number of phorophytes: 65. Variance of phorophyte random effect: 0.98. Model deviance = 2454. Null deviance = 2595
2. Complete elimination (γ2)
3. 100% elimination (γ3)
4. 50% elimination (γ4)
5. Positive control (γ5)
Months between censuses (φ)
ln(R. flaccida abundance) (ζ)
Focal phorophytes were checked 280 times in censuses two to six, and among these, we observed 20 stochastic extinctions from whole phorophytes. Just as the probability of stochastic extinction from leaves, the probability of whole-phorophyte stochastic extinction was higher in the treatment, where all occupied leaves on the focal phorophyte had been removed in the first census (Table 2). Moreover, the extinction risk decreased with increasing total R. flaccida abundance among the leaves on the phorophyte, and also increased the longer time that had passed between the censuses (Table 2).
Table 2. The final generalized linear model for probability of Radula flaccida extinction from an intact phorophyte. Model deviance = 100.6. Null deviance = 144.1
2. Complete elimination (γ2)
3. 100% elimination (γ3)
4. 50% elimination (γ4)
5. Positive control (γ5)
Months between censuses (φ)
ln(R. flaccida abundance) (ζ)
In total, we observed 404 R. flaccida colonization events in census six among the 2699 leaves that were unoccupied in census one or that emerged since census one. In comparison with the control treatment, colonization probability was lower in the treatment where all occupied leaves on the focal phorophyte had been removed in the first census (Table 3). There was no effect of leaf removal on surrounding phorophytes (positive control). Local population growth was not affected by the variation in connectivity, as evaluated by 2547 measurements of abundance on occupied leaves (Table 4); however, local growth was negatively correlated with increasing local abundance (Fig. 3).
Table 3. The final generalized linear mixed model for probability of Radula flaccida colonization between censuses one and six. Number of phorophytes: 70. Variance of phorophyte random effect: 1.7. Model deviance = 2044. Null deviance = 2058
2. Complete elimination (γ2)
3. 100% elimination (γ3)
4. 50% elimination (γ4)
5. positive control (γ5)
Table 4. The final generalized linear mixed model for the monthly relative abundance growth of Radula flaccida on occupied leaves. Number of phorophytes: 60. Variance of phorophyte random effect: 0.002. Residual variance (σ2): 0.096. Model deviance = 1283. Null deviance = 1760
ln(R. flaccida abundance) (ζ)
Organisms inhabiting ephemeral substrates, such as epiphylls in Amazonian rain forests (Reich et al. 2004), must maintain short generation times to counteract their high extinction probabilities. Our demographic study takes advantage of the uniquely foreshortened life cycles of tropical epiphylls by accruing a large amount of demographic data on notoriously elusive population parameters such as local colonization, population growth and local extinction resulting from stochastic or deterministic reasons. In the light of the gap between metapopulation theory and supporting evidence from experimental studies, our study provides valuable empirical insights into the relationship between fine-scale connectivity and plant demography, as well as the relative importance of different causes of extinction.
We show that short-distance connectivity to surrounding conspecifics is most important in determining the rates of colonization and stochastic extinction among leaves. Specifically, reduced within-phorophyte connectivity concomitantly decreases colonization and increases extinction rates. In contrast, we could not detect any effect of reduced connectivity to conspecifics on neighbouring phorophytes on the demographic rates of epiphylls within the temporal and spatial scale of the experiment. Such a result is in accordance with metapopulation theory, suggesting that reduced connectivity leads to decreased colonization probability and increased stochastic extinction risk (Hanski 1999). Leaves of an individual phorophyte apparently act as ‘islands within an island’ in which the epiphylls are subject to internal rescue effects (Holt 1992), as the risk of bryophyte extinction from occupied leaves increases if the connectivity to surrounding occupied leaves on the focal phorophyte is reduced. One possible explanation is that spillover of diaspores (gemmae and spores) from neighbouring leaves of the same phorophyte is disproportionately more important in reducing extinction risk, thereby contributing more to within-phorophyte population maintenance, than migration events of diaspores among neighbouring phorophytes. This apparent lack of a ‘propagule rain effect’ (Gotelli 1991) within the framework of the experiment comes as a surprising result as it suggests dispersal limitation on a fine scale even for plants such as epiphyllous bryophytes characterized by high colonization rates (Zartman & Shaw 2006) and, in many cases, taxa with inter-continental scale distributions (Lücking 1995).
The increase in stochastic extinction which accompanied reduced species occupancy treatments could alternatively be caused by increased demographic stochasticity of recently colonized leaves, as reflected by the fact that extinction risk increased with decreasing local abundance. However, because the variable local abundance was included in the model, small populations resulting from recent colonizations would be more likely to become extinct than older small populations if this explanation is probable. An alternative explanation may be that differences in the quality of leaf characteristics, in which young (recently colonized) leaves are less optimal patches for local epiphyll persistence, may also contribute to the demographic differences between treatments.
Indeed, there was no apparent effect of connectivity on local population growth on occupied leaves. However, local growth did decrease with increasing abundance: a result that corroborates with the demographic dynamics of the epiphytic bryophyte species Neckera pennata (Roberge et al. 2011). Although intra- or interspecific competition for limited space is a possible explanation, no leaf was entirely covered during the course of the study, and we only selected phorophytes that were inhabited by the focal species. Other non-exclusive explanations could be intrinsic factors such as resource limitation (e.g. space) or extrinsic factors such as pathogens. However, the former is unlikely as the phorophytes selected at the start of the study were predetermined to be less than half to ensure adequate microhabitat space for future potential colonizations.
Local population extinctions were equally likely to be caused by stochastic reasons from viable leaves than by deterministic ones (i.e. when leaves were shed), suggesting that epiphylls follow ‘habitat-tracking’ dynamics. Thus, the metapopulation stability of epiphyllous bryophytes at the leaf scale is controlled by a combination of the two extinction modes, suggesting that metapopulation growth occurs only when the colonization rate is higher than the combined rates of stochastic extinction and leaf loss. Moreover, extinction risk decreases with increasing local abundance at the leaf scale as well as with increasing summed cover on all occupied leaves at the phorophyte scale. A possible explanation for the relationship at the leaf scale is that bryophyte taxa, being poikilohydric in nature, demand a constantly humid environment to perform basic physiological functions (Proctor 1990). Increased abundance of the tightly woven colonies would ensure longer intervals of water retention between rains, thus reducing the risk of drying and physiological hibernation: processes that can potentially lead to senescence and release from the leaf surface. A possible explanation at the phorophyte scale are internal rescue effects in which the increased number of occupied leaves at the within-phorophyte scale contributes to the colonization and subsequent persistence of local populations on neighbouring leaves of the same phorophyte.
The extinction dynamics observed at the phorophyte scale were in accordance with metapopulation theory (Hanski 1999) as stochastic extinctions were observed from phorophytes that remained viable and standing. In other sessile organisms, contrasting findings have been reported – local stochastic extinctions have been negligible, and the extinction rate controlled by the rate of patch destruction (patch-tracking dynamics; Snäll, Ribeiro & Rydin 2003) as such species are associated with more stable substrates than leaves (Snäll, Ehrlén & Hakan 2005). Other species as well have shown intermediate extinction dynamics with both stochastic and deterministic extinctions (Thomas 1994; Laube & Zotz 2007; K. Fedrowitz, M. Kuusinen & T. Snäll, unpublished data).
The importance of connectivity on metapopulation viability is a subject of long-term theoretical interest rooted in the concept that connectivity to conspecific populations plays an important role in metapopulation persistence (Gotelli 1991; Hanski & Gyllenberg 1993). The lack of a treatment effect from the neighbourhood removal experiment does not preclude an effect of changing connectivity at a range of spatial or temporal scales. However, within our experimental framework, which was designed to reflect the natural history of the focal species, we confidently conclude that internal (e.g. within-phorophyte) spatial dynamics are disproportionately more important to epiphyll persistence compared with the spatial dynamics of phorophytes in the immediate surroundings (400 m2). In the light of this result, the study has increased our understanding of the role of fine-scale connectivity using a bryophyte model system with easily delimited, spatially structured ephemeral patches.
We are grateful to Dr Edinaldo N. dos Santos Silva and Dra. Veridiana V. Scudeller of the Biotupé Project for assisting with community support in the Tupé Sustainable Reserve. We also thank two anonymous reviewers for their invaluable insights and editorial contributions. Financial support was provided by grants from the Foundation for the Advancement of Research in the state of Amazonas (FAPEAM -DCR 2008) and MCT/CNPq/MEC/CAPES No. 52/2010 – PROTAX as obtained by the lead author. TS was funded by FORMAS grant 2005-933.