Research articlesContext-dependent colonization of terrestrial habitat ‘islands’ by a long-distance migrant bird
Abstract
Landscape context can affect how individuals perceive patch quality during colonization. However, although context-dependent colonization has been observed in aquatic environments, it has rarely been studied in terrestrial environments or at large spatial scales. In this paper, we assessed how landscape context influenced colonization rates in a large-scale (ca 7000 km2) terrestrial system where colonizers (Willow Warbler Phylloscopus trochilus) are capable of rapid, long-distance movements. Bioacoustic recorders were used to detect first song dates (an indicator of colonization or re-colonization) and settlement in 23 naturally replicated habitat patches. We compared support for three competing hypotheses describing colonization patterns that depend on landscape context (‘redirection’, ‘landscape-selection’ and ‘relative patch size’) with two patch-level hypotheses (patch ‘quality’ and ‘heterospecific attraction’). First song was earlier when habitat availability in the landscape was low, supporting the ‘redirection’ hypothesis. Settlement probability was best predicted by patch ‘quality’ and was lower in woodlands with a dense understorey. Results suggest that colonization of habitat patches by male P. trochilus after spring migration is spatially hierarchical. First, initial colonization depends on landscape context, and settlement is then determined by fine-scale vegetation characteristics. More broadly, we suggest that patterns observed in fragmented aquatic environments (e.g. ‘redirection’) can, in some circumstances, be extended to large-scale terrestrial environments.
1. Introduction
Understanding how landscape-scale habitat patterns influence colonization dynamics in fragmented landscapes is an important theme in ecology and conservation [1,2]. Theories of landscape connectivity broadly predict that during dispersal and colonization the importance of habitat composition and configuration is related to a species' mobility and degree of habitat specialism [3,4]. Colonization can also be context-dependent, where colonizers select patches based on the relative availability or quality of alternative habitat in the landscape [2,5,6]. For many taxa, our understanding of how landscape context influences colonization dynamics is limited to relatively coarse temporal scales, for example, annual colonization and extinction patterns [7,8], and this can mask fine-scale behavioural processes that are important during initial colonization [4]. This knowledge gap is especially true for highly mobile, terrestrial species that are capable of rapid, kilometre-scale movements.
Evidence from aquatic islands provides some of the strongest support for context-dependent colonization. In experimental coral landscapes, the number of colonizers is higher when focal patches have no adjacent patches, supporting the ‘propagule redirection’ hypothesis [6,9]. This hypothesis predicts that colonization rates are lower when isolated patches have more habitat in the surrounding landscape because colonists are redirected away from the focal patch. In the opposite direction, the ‘field of dreams’ hypothesis predicts that ‘if you build it, they will come’ and colonization rates are expected to be proportional to habitat amount, which leads to higher colonization rates in landscapes with more habitat [6,10]. Perceived patch quality can also be context-dependent. For example, predator presence in one patch can reduce the perceived quality of neighbouring patches in the local landscape [2,11,12].
In terrestrial environments, the effects of landscape composition and configuration on species turnover and community assembly are well-studied, but less is known about the influence of landscape context on colonization behaviour [4]. The reasons for this knowledge gap are two-fold: (i) it is often difficult to precisely detect colonization events and their timing (e.g. first arrival), despite technological advances such as global positioning system (GPS) tracking, and (ii) it is difficult to conduct studies over large spatial scales to ensure habitat patches are sufficiently replicated and spatially independent (i.e. colonizers should not move between focal patches).
Birds are among the most well-studied taxa in landscape ecology, but although landscape configuration plays an important role in shaping bird communities, the link between pattern and process is poorly understood [4]. Homing experiments show that displaced Ovenbirds Seiurus aurocapilla return faster to established territories when forest cover in the landscape is high [13,14]. This could suggest that the amount of habitat in the landscape is likely to be important during colonization because birds might view the landscape as more favourable [7], or because individuals can move more easily between potential territories. Gap-crossing experiments have also been used to quantify how individuals cross non-habitat and indicate that forest species prefer to move through forest and forest edge rather than cross non-forest areas [14–16]. Although this work has provided valuable insights into the movement behaviour of birds, such experiments have been criticized for being unrealistic [17], and any link between experimental movement behaviour (e.g. during translocations) and natural colonization behaviour remains speculative. Furthermore, there is usually a significant trade-off between spatial and temporal resolutions; where studies at large spatial scales tend to have low temporal resolution (e.g. monthly or yearly assessments of patch occupancy), and studies with high temporal resolution (daily counts of birds or translocation experiments) are usually limited to relatively small spatial scales. To meaningfully assess how landscape-scale habitat patterns influence colonization behaviour in highly mobile taxa such as birds there is a need to achieve both high spatial and temporal resolutions [4].
Entire populations of migratory birds move biannually between their breeding and wintering grounds. In spring in the northern hemisphere, individuals arrive on the breeding grounds and compete to secure the ‘best’ territories [18]. Changes in timing of arrival at the breeding grounds can have significant reproductive consequences due to phenological match or mismatch with resource availability [19,20]. Many migrant birds also show strong natal and breeding philopatry [18,21]. Thus, the amount of available breeding habitat (i.e. territory availability) in a landscape should correlate with the number of individuals that return annually to breed in a given area. Migration and subsequent selection of breeding territories occurs over a varied range of landscapes, presenting an ideal natural experiment to test the effects of local and landscape context on colonization behaviour. However, most previous work has focused on the effects of landscape context on habitat-use during migratory stopovers and it is unclear how landscape context influences initial selection of breeding territories. Bennett et al. [7] found that migrant bird communities responded to regional gradients of land-use, but were unable to disentangle which factors were most important because of high multicollinearity between measures of land-cover. Despite a lack of empirical evidence, it has been suggested that migrant birds select landscapes with high habitat availability during initial colonization [7] and subsequent settlement of territories is expected to depend on local, patch-level characteristics such as resource availability or vegetation structure. Recent global declines of migratory birds are also poorly understood, and land-use change along migratory routes has been implicated in the population declines of some species [22]. Understanding how large-scale habitat patterns influence colonization behaviour is therefore a conservation priority.
Here, we used a natural experiment and novel approach to test whether colonization of terrestrial habitat ‘islands’ is context-dependent when individuals are capable of rapid, large-scale movements that are independent of habitat availability (i.e. direct, nocturnal migratory flights). We used bioacoustic recorders to monitor migrant bird (Willow Warbler Phylloscopus trochilus L.) colonization or re-colonization (date of first male song) and settlement in discrete habitat patches (post-agricultural woodlands) across large spatial scales and with high temporal resolution. Habitat patches were of similar size but varied in the amount and configuration of habitat in the landscape (i.e. context). We explicitly tested three competing, a priori hypotheses that described how landscape context might affect colonization rates and settlement probability (‘redirection’, ‘landscape-selection’ and ‘relative patch size’), and compared support for these with two patch-level hypotheses (patch ‘quality’ and ‘heterospecific attraction’ [23]).
2. Methods
(a) Model species
To test whether colonization rates of habitat islands depend on landscape context, the study system must satisfy five key criteria: (i) the focal species must use habitats that can easily be characterized at coarse spatial scales (e.g. any wooded habitat in the landscape); (ii) territory densities must be proportional to habitat amount (this assumption is tested in electronic supplementary material, table S1 and figure S1); (iii) individuals must initially be absent from a patch; (iv) focal patches must provide suitable habitat for potential colonizers; and (v) the timing of colonization events must be detected with high precision and accuracy. To meet these criteria, we selected P. trochilus, a generalist woodland passerine that breeds in northern Eurasia (above approximately 41° N) and overwinters in sub-Saharan Africa [24] as a model species.
In Great Britain, P. trochilus breeds in coniferous and broadleaf woodland and scrub, with an estimated 2.2 million territories across the island [25]. Birds prefer woodland patches larger than 0.5 ha with intermediate canopy cover and vegetation ranging in height from roughly 3 to 6 m [26,27]. Historical reporting rates show that P. trochilus begins to arrive in Great Britain in the last week of March and the population is fully installed by the end of April [28]. Migration occurs mainly at night and birds travel at average speeds ranging from 40 to 85 km per day, but speeds of up to 218 km per day have been recorded [29,30]. Males arrive before females and are highly vocal when establishing territories, producing a loud, easily detected song that can be repeated up to six times or more per minute [31]. Males of breeding age are highly philopatric, but adult interannual survival is relatively low at around 40% and thus individual turnover is high [21,32]. Once on the breeding grounds, males compete for the best territories, with higher quality territories occupied earlier in the season. Territory sizes range from approximately 0.2 to 0.3 ha [33,34].
(b) Habitat islands
To obtain a set of habitat islands for the study, a subset of post-agricultural broadleaf woodlands was selected from 101 patches used by the Woodland Creation and Ecological Networks (WrEN) research project (electronic supplementary material, figure S2). The WrEN project (http://www.wren-project.com) comprises a large-scale natural experiment that aims to evaluate how landscape structure and patch characteristics influence biodiversity responses to habitat creation [35]. Patch size is often a stronger predictor of bird species occurrence than other factors such as landscape context [36,37]. Since our primary interest was the influence of landscape context on the timing of colonization, we controlled for patch area by selecting small woodlands of similar size (0.5–2.6 ha), but which varied in the amount and configuration of suitable habitat in the surrounding landscape (see fig. 7a in [38]). Using these criteria and information from previous bird surveys in the 101 WrEN study sites [37], we selected a sample of 23 woodlands previously occupied by P. trochilus for inclusion in the study. Territory mapping (i.e. mapping of singing males) over three visits in April, May and June 2015 [37] indicated that each woodland held on average one territory. A further 12 previously unoccupied woods that satisfied the size criteria and which appeared to offer suitable habitat (based on a visual assessment) were also selected in an attempt to increase the sample size, but none of these was subsequently occupied by P. trochilus and they were excluded from further analysis (electronic supplementary material, figure S2).
Patches were typically greater than or equal to 3 km apart and were considered spatially independent because male P. trochilus dispersal distances (i.e. distance between breeding and natal site) are typically less than 1 km [33]. Patch age (years since woodland establishment) ranged from 10 to 160 years (median 90 years), and age was correlated significantly with mean tree diameter at breast height (DBH) (r = 0.78, p < 0.001, n = 23 patches).
(c) Detecting colonization
We used Solo audio recorders [39] to record male P. trochilus song in each patch. Pilot tests were used to calibrate microphones so that songs could be detected at distances of approximately 50 m (electronic supplementary material, figure S3), which is equivalent to a sampling area of 0.3 ha (i.e. one P. trochilus territory).
A single recorder was deployed in the centre of each woodland patch during the last week of March 2016 in advance of birds arriving, and set to record audio continuously (24 h, 7 days) until 1 May 2016. On the day of deployment, we used song playback to ensure that no territorial males were present in the study sites. P. trochilus song was broadcast using an mp3 player and handheld speaker in 10-s bursts at 1-min intervals for 3 min. No birds responded to song playback, indicating that patches were vacant when data collection commenced.
To identify initial colonization (or re-colonization of patches occupied in the previous year) and subsequent settlement we used a sub-sampling approach. Three continuous 10-min blocks of audio were randomly sampled from the 3-h period after sunrise per day of audio recording (i.e. 30 min of audio per day). We then manually examined audio spectrograms for P. trochilus song and, after the first detection, we counted the number of calls in the sub-sampled audio for 14 consecutive days to determine if the patch had been settled. Patches with no songs on two consecutive days of the 14 were considered not settled. We also evaluated how song rates in a patch changed over time after the first song date, expecting the pattern to be consistent with fig. 4 in [19] if there was minimal turnover (i.e. an initial steep rise in song rates followed by a gradual decline). We also compared first arrival dates with countrywide data from the British Trust for Ornithology [28] to ensure that results were representative of those across Great Britain in 2016.
(d) Hypotheses
The ‘propagule redirection’ hypothesis [6,9] predicts that when habitat availability in the landscape is low, focal patches will receive more individuals over time. This is based on the expectation that individuals randomly sample from the available habitat in a landscape. In the system used here, this would be indicated by earlier first song dates in more isolated patches because the probability of receiving an individual is higher per unit time (figure 1a). In this case the individual birds are not ‘propagules’ per se but an adult male of breeding age, and we refer to this hypothesis as ‘redirection’.
Figure 1. Hypothesized relationships (see Methods) between landscape context, patch characteristics and first song date (a–e), and settlement probability (f–j).
Alternatively, if the ‘initial cue to settlement’ [7] is based on landscape units (i.e. not individual patches) that appear more favourable in general, then focal patches might be colonized earlier when the amount of woodland in the landscape is higher. We call this the ‘landscape-selection’ hypothesis, which would be consistent with the predictions made by Bennett et al. [7]. The expectation is that individuals select landscapes with high habitat availability, with subsequent sorting into territories based on local, territory-level factors (e.g. food availability, vegetation characteristics, etc.). Thus, when habitat availability in the landscape is high, focal patches will be colonized earlier (figure 1b). The direction of this predicted relationship could also be caused by the effects of philopatry. Landscapes with more habitat (and therefore a higher number of potential territories) will receive more individuals returning to their previous breeding territories or close to their natal sites, which should lead to earlier colonization of patches with more habitat in the surrounding landscape. The two mechanisms (i.e. ‘landscape selection’ versus philopatry effects) are not mutually exclusive, however. It is therefore important to acknowledge that our study design cannot disentangle these if first song date is earlier in focal patches with more habitat in the landscape.
Larger patches have higher woodland bird abundance and occupancy [36,37]. Individuals might therefore select patches based on their relative size compared to other patches in the surrounding landscape, with relatively large patches occupied earlier than relatively small patches. The same pattern might also occur due to random sampling (e.g. a target area effect [40,41]), where relatively large patches are more likely to ‘capture’ individuals than relatively small patches in the same landscape due to probability alone. Focal patches might therefore be colonized earlier if they are large relative to nearby patches in the landscape, which we call the ‘relative patch size’ hypothesis (figure 1c).
Measures of patch ‘quality’, including vegetation characteristics such as canopy cover, understorey cover, tree DBH and tree diversity can be used to describe habitat suitability for breeding P. trochilus [26,27]. Since the 23 patches used here were occupied by P. trochilus in the previous year and were generally similar in structure (i.e. broadleaf post-agricultural woodlands), and because we assumed that initial colonization occurred rapidly at dawn after nocturnal flights, we predicted that patch-level factors would have no effect on the timing of initial colonization (figure 1d).
Manipulative experiments have demonstrated that migrant birds are attracted to the vocalizations of other woodland birds (heterospecific attraction) when establishing territories [23], perhaps because social information from heterospecifics is indicative of aspects of patch quality, such as predation risk [42]. To test for the existence of heterospecific attraction we calculated an index of acoustic complexity in the patch on the day of first P. trochilus song, expecting that patches would be colonized earlier as soundscape complexity (i.e. bird song richness) increased (figure 1e).
We attempted to control for patch size during site selection as far as possible (i.e. patches were all of similar size), but since there was some small variation between patches we tested for an effect, expecting no relationship between patch size and colonization rates (not illustrated). Finally, our null model for initial colonization predicted that woodlands further north and east would be colonized later (not illustrated) based on knowledge of large-scale P. trochilus migration patterns in the UK [29].
Although the primary focus of the study was on the timing of initial colonization, we also repeated the analyses for settlement probability, since not all sites were considered settled (see Results). Settlement was expected to depend on patch-level factors at the territory scale, and we therefore predicted that settlement probability would not depend on landscape context (figure 1f,g). However, individuals might decide to move to neighbouring patches if their initial patch appears poor relative to others in the landscape, and we predicted that settlement probability would therefore be lower when patches were small relative to their neighbours (figure 1h). Settlement probability was expected to be higher in better ‘quality’ patches (e.g. with an open understorey) (figure 1i) and in patches with higher bird song complexity (figure 1j).
(e) Predictor variables
Two variables were used to test our three primary hypotheses of interest. The amount of habitat (any woodland) in the landscape was used to test the ‘redirection’ and ‘landscape-selection’ hypotheses, and the ratio between the focal patch size and the median patch size of any woodland in the landscape was used to test the ‘relative patch size’ hypothesis. Habitat amount, patch size and relative patch size were calculated from Forestry Commission spatial data [43] at seven scales (nested Geographic Information System buffers surrounding the focal patch with radiuses of 250, 500, 1000, 1500, 2000, 2500 and 3000 m). To test the effects of vegetation structure we used four variables, which were per cent canopy cover, tree mean DBH, tree species richness and per cent understorey cover (obtained during field surveys; tables S2 and S3). Acoustic complexity was used to test the ‘heterospecific attraction’ hypothesis and was calculated as the mean acoustic complexity index [44] (electronic supplementary material, table S2) for the 3 × 10-min sub-sampled audio files from the date of first P. trochilus song in a patch. To validate the index we regressed the averaged acoustic complexity value against raw bird species richness per site (n = 23) in 2015, confirming that there was a positive (although noisy) relationship (electronic supplementary material, figure S4).
(f) Modelling approach
We used generalized linear models (GLMs) to quantify the relative effects of each predictor on first P. trochilus song date and settlement probability. Geographical position (i.e. northing and easting) was expected to have the strongest effect on first song date, and our null model included northing and easting only as continuous predictors. GLMs were fitted by maximum likelihood, and residual diagnostics suggested a Poisson error structure was more appropriate than a Gaussian error structure for first arrival dates. For settlement probability, we used GLMs with a binomial error structure, and an intercept only model was used as the null. To avoid overfitting the data (i.e. modelling the residual variation because the ratio between n and the number of parameters is too low) and because we were primarily interested in the relative magnitude of effects, separate models were constructed for each predictor of interest (but note that northing and easting were included as covariates in all models of first song date). We selected a single ‘scale of effect’ [38] for each landscape variable using univariate generalized linear models and corrected Akaike Information Criterion (AICc), selecting the most appropriate scale by evaluating those models (after validation [45]) with the lowest AICc and largest effect size for the predictor of interest. Predictors were mean centred and scaled by one standard deviation to compare relative effect sizes (β). AICc was used to compare goodness of fit between each model and the null, and 95% confidence intervals for effect sizes were bootstrapped from 500 resamples of the data. Likelihood-ratio-based pseudo-R-squared values were calculated using the r.squaredLR function in the MuMIn R package [46]. Multicollinearity between predictors in models of first song date (i.e. northing, easting and the predictor of interest) was assessed using variance inflation factors (threshold < 5) and model validation followed [45]. A correlation matrix for all predictors is in the electronic supplementary material, table S4, and results for the null models are given in the electronic supplementary material, tables S5 and S6. R statistical software was used for all analyses [47].
3. Results
The median arrival date was 16 April (range 5–22 April; electronic supplementary material, table S3 and figure S5). Arrival dates were consistent with British Trust for Ornithology countrywide data in the same year (electronic supplementary material, figure S5), indicating that results were representative. Although P. trochilus was detected in 23 woodlands, birds did not settle in five of them. In settled patches, song rates increased over time after the first detection before gradually declining, and the shape of the relationship was generally consistent with those observed in [19] (electronic supplementary material, figure S6). Data were often noisy, however, and in some instances males may have been replaced during the study period (e.g. electronic supplementary material, figure S6f,g).
After accounting for geographical location, first song dates were significantly earlier in patches with low amounts of woodland in the surrounding landscape within 2 km (table 1; figures 2a and 3). The effect size was relatively large (although with high uncertainty, table 1) and first song was delayed by approximately five days as woodland cover in a 2 km radius increased from 5% to 30%. The positive direction of this effect was consistent with the ‘redirection’ hypothesis (figure 1a), and the variance explained (36%) was 11 to 12% higher than alternative models (table 1). Based on AICc alone, however, the ‘redirection’ hypothesis had only marginally more support than the null, but it was more strongly supported than alternative hypotheses (table 1).
Figure 2. (a) Relationship between P. trochilus first song date and habitat amount within a 2 km radius of a woodland patch (holding northing and easting at their mean values), and (b) relationship between settlement probability and understorey cover per cent (Domin scale: electronic supplementary material, table S2). Circles are observed values and semi-transparent lines are bootstrapped 95% confidence intervals from the GLMs. Figure 3. Illustration showing the relationship between the amount of woodland in the landscape (green; within 2 km radius) and first P. trochilus song date (day in April 2016) in four example focal woodlands (coloured red, not to scale) at similar latitudes (near 56° N). P. trochilus artwork by Andreas Trepte, www.photo-natur.net (modified under CC BY-SA 2.5).
| hypothesis (figure 1) | focal predictor | β | −95% | +95% | p | ΔAICc from H0 | R2 |
|---|---|---|---|---|---|---|---|
| a,b | proportion any woodland (2 km radius) | 0.12 | 0.01 | 0.23 | 0.04 | −1.35 | 0.36 |
| c | relative patch size (2 km radius) | 0.01 | −0.12 | 0.13 | 0.92 | 2.95 | 0.23 |
| d | canopy cover per cent | 0.04 | −0.09 | 0.16 | 0.58 | 2.65 | 0.24 |
| d | tree species richness | 0.02 | −0.09 | 0.13 | 0.68 | 2.79 | 0.24 |
| d | understorey cover per cent | 0.00 | −0.12 | 0.10 | 0.93 | 2.95 | 0.23 |
| d | tree DBH mean | −0.01 | −0.12 | 0.10 | 0.91 | 2.95 | 0.23 |
| e | acoustic complexity index | −0.01 | −0.12 | 0.10 | 0.91 | 2.94 | 0.23 |
| — | patch size ha (ln transformed) | 0.04 | −0.10 | 0.17 | 0.59 | 2.67 | 0.24 |
Contradicting expectations, there was no evidence to suggest that the relative size of the focal patch in the landscape or bird song complexity in a patch affected first song dates. Patch ‘quality’ had no detectable effects on first song date, as expected.
Probability of settlement depended on patch-level vegetation characteristics and declined as understorey vegetation cover increased (figure 1b and table 2). Probability of settlement was also lower when there was more woodland in the surrounding landscape, but the model was only weakly supported based on AICc. Alternative hypotheses for settlement probability had minimal support and effect sizes were small (table 2).
| hypothesis (figure 1) | focal predictor | β | −95% | +95% | p | ΔAICc from H0 | R2 |
|---|---|---|---|---|---|---|---|
| f,g | proportion any woodland (2.5 km radius) | −1.14 | −2.8 | −0.03 | 0.09 | −1.68 | 0.25 |
| h | relative patch size (250 m radius) | −0.37 | −1.36 | 0.63 | 0.63 | 1.74 | 0.04 |
| i | canopy cover per cent | −0.87 | −2.62 | 0.3 | 0.22 | 0.44 | 0.13 |
| i | tree species richness | −0.27 | −1.27 | 0.77 | 0.59 | 2.12 | 0.02 |
| i | understorey cover per cent | −1.32 | −2.83 | −0.25 | 0.04 | −3.66 | 0.36 |
| i | tree DBH mean | −0.59 | −1.71 | 0.41 | 0.25 | 1.05 | 0.09 |
| j | acoustic complexity index | −0.34 | −1.48 | 0.68 | 0.52 | 1.98 | 0.03 |
| — | patch size ha (ln transformed) | 0.40 | −0.63 | 1.72 | 0.49 | 1.89 | 0.04 |
4. Discussion
The importance of landscape context for community assembly and dispersal behaviour is well known, but the relationship between landscape context and colonization behaviour is less clear, particularly in terrestrial systems and for organisms that undertake very large-scale, rapid movements. In agreement with experimental work in aquatic environments [6], our results support the ‘redirection’ hypothesis (i.e. colonists are redirected away from the focal patch if habitat amount in the landscape is high) and we found little support for the ‘landscape selection’ [7] or ‘relative patch size’ hypotheses. Previous work in our study area also supports the ‘redirection’ hypothesis: we found that, after controlling for patch size and other factors, the relative abundance of woodland birds was lower in focal patches (n = 101) when the amount of broadleaf woodland within 1 km was higher (fig. 3b in [37]). As predicted, settlement probability depended on vegetation characteristics (understorey cover) at the local, patch-scale. This is consistent with P. trochilus preference for woodlands and woodland edges with a relatively open, grassy understorey, which facilitates ground nesting. This result provides some of the first empirical support for the hypothesis that landscape context acts as an initial cue for colonization, and that more local, patch-level factors then dictate whether the territory is eventually settled or abandoned.
Whether earlier first song dates in sparse landscapes are caused by a random sampling process (i.e. higher probability of ‘capturing’ a colonist in a focal patch when there is less habitat in the landscape), or whether individuals are actively selecting patches with low amounts of habitat in the surrounding landscape, cannot be determined using these results. Philopatry also plays an important mechanistic role in dictating where birds return to breed each year. Adult males will typically return to their previous territory if they bred successfully in the previous year, and first year male P. trochilus will return to within 1 km (on average) of their natal territory. Despite strong philopatry in P. trochilus, however, it cannot explain the observed relationship between first song date and the amount of woodland in the landscape based on our a priori expectations. Therefore, given the remarkable navigational skills of migrant birds and because many species are known to actively select vacant territories based on their relative quality (e.g. prospecting behaviour in Collared Flycatchers Ficedula albicollis [48]), we suggest that the relationship between first song date and amount of habitat in the landscape is unlikely to be due to a random sampling process, and propose that isolated patches become disproportionally attractive to early male P. trochilus in landscapes with less alternative habitat.
The relatively isolated agricultural woodlands studied here probably represent poor quality habitat patches for P. trochilus. Yet, when they are situated in landscapes with little alternative habitat, they appear to become relatively more attractive, because males arriving earlier. Paradoxically, this is despite P. trochilus' ability to travel more than 100 km per night during migration, which implies that individuals could avoid sparsely wooded landscapes with relative ease. This relationship would have been overlooked by studies of patch occupancy or abundance [8,37,49] because most patches were eventually occupied and thus would appear equally ‘attractive’ if surveys were made at coarse temporal resolutions or later in the season.
During migration and stopover, body mass gain in P. trochilus is higher when landscapes have more forest cover [50]. Assuming that the same is true on breeding territories, it therefore appears counterintuitive that first song date is earlier in the most isolated woodlands, since individuals colonizing these woodlands are likely to experience fitness costs. Perhaps isolated woodland patches are acting as ecological traps [51,52], leading individuals to trade off current versus future fitness potential by avoiding further, potentially risky long-distance migratory flights when the landscape appears to have limited habitat availability. Other migratory birds are also attracted to ‘ecological traps’ that arise from novel habitat patterns in the landscape, such as selectively harvested forest [52]. Considering recorded declines of P. trochilus across much of its range [22], there is a need to understand whether the patterns observed here are causing relatively high-quality individuals (as measured by earlier arrival date) to select suboptimal breeding territories.
Translocation experiments have shown that forest birds (including long-distance migrants such as S. aurocapilla) return faster to established territories when landscapes are more favourable and have more habitat [13,14]. This suggests that the same could be true during initial selection of breeding territories, and it might be expected that patches located in landscapes with more habitat would be occupied earlier than patches in sparse landscapes. However, our results are consistent with initial colonization occurring after nocturnal flights that are not constrained by habitat availability in the landscape, and colonization was earlier when habitat availability was low. In the case of translocation experiments, individuals probably take advantage of favourable habitat to travel by day and return rapidly to their existing territory. We should therefore be cautious when extrapolating findings from experimental work on movement ecology to natural behaviour.
Contradicting expectations, we detected no relationship between acoustic complexity in a patch and first song date or settlement of P. trochilus, providing no support for ‘heterospecific attraction’. Heterospecific attraction is well supported by evidence from other studies that have tested for its existence experimentally [23,53]. These have shown that migrant bird abundance tracks the abundance of other woodland bird species. However, we did not quantify P. trochilus abundance in this study (indeed most patches could support only one territory), and this could explain why we detected no effect, since we did not quantify post-colonization dynamics beyond settlement probability.
The lack of support for the relative patch size hypothesis during initial colonization was surprising given the apparent importance of patch size for P. trochilus and other woodland birds [7,36,37]. There is ongoing debate surrounding the importance of habitat amount versus habitat configuration (i.e. relative patch size in this study) for community assembly in fragmented landscapes [38,54]. This result suggests that habitat amount is more important than habitat configuration for P. trochilus during selection of territories after spring migration.
It is important to acknowledge that these results are from a single year and differences in weather conditions or population densities between years, for example, might also interact with landscape context to influence first song dates and settlement probabilities. Other factors not measured here could also affect first song dates and settlement, such as fine-scale measures of prey availability or the age of individual birds. Nonetheless, our arrival times were consistent with the historical average in Great Britain and those recorded in 2016, and results supported well-defined, a priori hypotheses.
5. Conclusion
We suggest that landscape context plays an important role during initial colonization (or re-colonization) of discrete habitat patches by a long-distance migrant bird after spring migration. These results represent an important step forward in understanding how landscape context influences the spatial and temporal dynamics of colonization by extending results from smaller aquatic systems to considerably larger scales and to a terrestrial environment under natural conditions. The non-invasive method used to detect first song dates also offers considerable promise for studying how landscape context influences the colonization behaviour of other highly mobile, acoustically active taxa that might be too small to track in time and space using existing tracking technology.
Ethics
The study was approved by The University of Stirling Biological and Environmental Sciences ethics committee.
Data accessibility
Data are available from the Dryad Digital Repository: http://dx.doi.org/10.5061/dryad.8qf122m [55].
Authors' contributions
R.C.W. co-designed the study, analysed the data, wrote the first draft of the manuscript and collected the data with L.W. K.J.P., E.F.-M., K.W. and N.A.M. co-designed the study and co-wrote the manuscript. All authors contributed substantially to manuscript revisions.
Competing interests
We declare we have no competing interests.
Funding
R.C.W. was funded by the IAPETUS Doctoral Training Partnership via the Natural Environment Research Council (grant number NE/L002590/1), the National Forest Company and Forest Research. The WrEN project (E.F.-M., K.W., N.A.M., K.J.P.) has been funded by the University of Stirling, Natural England, Forestry Commission, Scottish Natural Heritage, the Department for the Environment, Food and Rural Affairs, the National Forest Company, Forest Research, the Woodland Trust and Tarmac Ltd.
Acknowledgements
We are grateful to landowners for permitting access to the study sites. The authors thank Mark Whittingham, Luc Bussiere, Craig Osenberg, Tom Matthews and an anonymous reviewer for their valuable comments on an earlier version of the manuscript. We also thank James Christie for technical advice.
