The effects of habitat and spatial features of wetland fragments on the abundance of two rallid species with different degrees of habitat specialization

Abstract

Capsule Water Rail Rallus aquaticus and Moorhen Gallinula chloropus abundance is affected by habitat (both species), site isolation (Water Rail) and elevation (Moorhen).

Aims To evaluate the effects of patch area, habitat, isolation, elevation and inter-specific interactions on the abundance of two under-studied species, Water Rail and Moorhen; i.e. a wetland specialist and a more generalist species, respectively.

Methods We modelled species' abundance at 28 sites in a wetland system in Trentino (Italy), using variables describing both habitat and spatial configuration of sites. We evaluated spatial autocorrelation and carried out an aic_c -based model ranking and averaging, whose results were cross-checked using hierarchical partitioning.

Results Water Rail abundance was positively affected by cover of reed-bed and other wetland herbaceous vegetation, and negatively by the distance to the nearest wetland. Moorhen abundance was positively affected by cover of reed-bed and negatively by elevation and cover of woody vegetation.

Conclusions Results matched the preference for reed-beds and other wetland herbaceous vegetation reported at the local scale for Water Rails, and highlighted a negative effect of spatial isolation. Our work further demonstrates how explanations of species occurrence and abundance in fragmented landscapes must consider variables describing both habitat quality and spatial configuration.

The spatial distribution and abundance of animal species is affected by many factors belonging to a wide variety of environmental categories. Climate, topography, habitat types and quality at different scales from landscape to micro-habitat, inter-specific interactions, isolation and accessibility of habitat patches, can concur to determine both presence and abundance of animal species (Hall et al. 1997, Penteriani et al. 2001, Martínez et al. 2008, Sebástian-González et al. 2010). Metapopulation theory predicts that both habitat size and habitat isolation are important in determining species' distribution and abundance (Hanski & Gaggiotti 2004). While patch size is likely to determine population size, which in turn affects extinction rates, the degree of patch isolation can affect colonization rates, and empirical evidence consistent with these assumptions has been reported from several systems (Hokit et al. 1999, Hames et al. 2001, Hanski & Gaggiotti 2004, Bennett & Saunders 2010). Patch quality may be a further determinant of species' occupancy and turnover rates (Summerville & Crist 2001, Thomas et al. 2001, Fleishman et al. 2002, Bennett & Saunders 2010). The response to habitat quantity, quality, fragmentation and isolation differs between species, with the relative importance of these factors depending on species traits (Hokit et al. 1999, Marshall et al. 2000, Johansson & Ehrlén 2003). Most specialized and low-dispersal species show distinctive ecological requirements and are often sensible to variation in the aforementioned factors, while species with broader niches and high mobility are often less affected by environmental traits and occur over a broader range of combinations of isolation, habitat types and features. Although all the aforementioned factors could be important in shaping the distribution and determining the abundance of a species (Bennett & Saunders 2010), studies simultaneously assessing their relative importance are infrequent, as most research focuses alternatively on one or a few type of factors. Species and communities occurring in isolated habitat(s) within unsuitable landscape matrices represent a favourite topic for studies considering different environmental traits. Investigating isolated habitat patches allows explicit testing of the relative importance of patch area, habitat features and isolation on species composition of biological communities or on occurrence and abundance of given species.

The aim of the present study was to evaluate the effects of patch area, habitat type and quantity, geographical isolation, elevation (which is also an indirect measure of climate) and inter-specific interactions on the abundance of two secretive and poorly studied species of the rallids (Rallidae), an avian family including species tied to water, with both generalist and specialist taxa inhabiting different wet habitats (Cramp & Simmons 1980). By considering species with good flying ability, we avoid the problem of potential barriers to dispersal within the wider landscape. We used Water Rail Rallus aquaticus and Moorhen Gallinula chloropus as model species. Basing our study on species with very different degrees of ecological specialization, we hoped to assess the effects of the aforementioned factors on two definitely different biological models; despite the fact that they are rather closely related one to each other and have similar size, they have different niche breadths (Cramp & Simmons 1980, Pedrini et al. 2005) and could potentially show somewhat different responses to variation in habitat, area, isolation and presence of potentially interacting species. Water Rail is a typical inhabitant of well-vegetated wetlands, usually requiring flooded reed-beds, some clearings with open water, and access to shallow water or muddy surfaces (Jenkins & Ormerod 2002); it can be considered a wetland specialist, unable to occupy successfully other habitats in the landscape matrix within which wetlands are interspersed. Moorhen is a generalist species, inhabiting all kind of habitats providing access to water and patches of vegetation (Cramp & Simmons 1980, Brambilla & Rubolini 2003), and also occurs outside ‘true’ wetland sites (Pedrini et al. 2005). We also counted the number of Coots Fulica atra at study site. Coot is the largest rallid species inhabiting the region and a strictly territorial species, displaying aggressive interactions toward conspecifics and also other waterbirds; aggressive behaviour directed to other rallids has been reported, and Moorhen is the species most attacked by Coots (Cramp & Simmons 1980). On the other hand, Moorhens can use presence of Coots as a cue indicating habitat quality (Sebástian-González et al. 2010). Therefore, we considered Coot as a potential source of inter-specific effects, especially for Moorhen. The inter-specific interactions are more likely to occur within small wetland sites, such as those included in the present study (see later).

All these three species are mainly lowland birds, inhabiting areas at low and middle elevation (Cramp & Simmons 1980); the upper elevation limit of their distribution in the study area is around 1000 m asl for all the species (Pedrini et al. 2005).

MATERIALS AND METHODS

Study area

The Trento province covers 6206 km² in the central-eastern portion of northern Italy. Elevation ranges between 67 and 3769 m asl; 30% of the land lays below 1000 m, while 50% of the area is comprised between 1000 and 2000 m. Landscape is characterized by intensively cultivated and partly urbanized valley floors, while mountainsides are covered by woodlands interspersed with pastures and vineyards at lower elevation (<1000 m), by woodlands and secondary (anthropogenic) grasslands at middle elevation (1000–2000 m), and by alpine grasslands, rocks and snow-covered ground at higher elevation (>2000 m). Rainfall values range from 700 mm/year to 1500 mm/year (and locally higher). Human density is lower than in most Italian regions, being equal to 76.3 inhabitants/km² in 1999 (Provincia Autonoma di Trento 2001). Until some decades ago, wetland habitats were much more widespread, occupying large areas in river valleys; however, drying-out and intensification of agricultural practices led to widespread wetland destruction. Wetland sites are thus currently limited to residual patches, and most riparian habitats have been destroyed; this is the likely cause of the extinction of some other rallid species (regional population of Spotted Crake Porzana porzana and wetland population of Corncrake Crex crex [Pedrini et al. 2005, Brambilla & Pedrini 2011, Pedrini et al. 2012]). We considered 28 wetlands scattered over the whole Province (Fig. 1). Wetland area varies from 0.9 to 72.2 ha and, apart from high-elevation alpine lakes and bogs and from the Garda Lake, we have included in our work all the wetland sites larger than 1 ha in the Trento province.

Figure 1. Geographical location of the study wetlands within the Trento province.

Bird counts

Water Rails and Moorhens were counted by crepuscular censuses, following (Brambilla & Jenkins 2009). The 28 wetlands were surveyed in 2009 and 2010; each site was visited 1–3 times per year during the breeding season of the species, depending on area and accessibility. A few sites were visited only in one of the two years because of logistical constraints. We used playback of broadcasted vocalizations in April (May for sites with prolonged snow cover) to maximize the detection rate of the species while simultaneously reducing the risk of including migrants (Brambilla & Jenkins 2009). For Water Rail, we followed the protocol tested in previous works (Brambilla & Jenkins 2009) and consisting of playback stimulation with a male's song followed by a pair duet, repeated twice, for a total of 50 seconds of broadcasting. A comparable protocol was used for Moorhen, with 50 seconds of species' vocalizations. Coots were also counted at each site, without use of playback, by means of diurnal surveys. We used the maximum number of birds detected per year as an estimate of the number of individuals per site in a given year.

Analyses

Environmental variables considered in the analyses for each site were as follows: elevation (m asl); wetland area (measured in ha); cover (measured in ha) of open water; reed-bed (Phragmites australis); wetland herbaceous vegetation different from reeds (mainly Carex spp. and, less frequently, Typha spp, Scirpus spp. and Juncus spp.); woody wetland vegetation; and maximum number of Coots (average between the two years for sites censused in both years). For Water Rail only, we also considered the distance to the nearest wetland (measured in km); this variable was not considered in Moorhen analysis as several other breeding pairs occurred outside our study wetlands (in many different sites, such as small ditches, ponds, rivers and channels, etc.), while this is unlikely to happen for Water Rail (see earlier). Average values of habitat variables per site are summarized in Table 1.

Table 1. Habitat variables considered in the study (n = 28).

Variable	Acronym	Mean ± se per wetland site
Elevation (m asl)	Elevation	436 ± 42
Wetland area (ha)	Area	11.12 ± 3.56
Open water (ha)	Open w.	5.57 ± 3.04
Reed-bed (ha)	Reed-bed	1.47 ± 0.53
Wetland herbaceous vegetation (ha)	Herbaceous	0.57 ± 0.27
Woody wetland vegetation (ha)	Woody	3.51 ± 1.10
Number of Coots	Coots	4.57 ± 1.13
Distance to the nearest wetland (km)	Dist. to w.	4.19 ± 0.85

Cover habitat variables, wetland surface and number of Coots were square-root transformed prior to analysis in order to reduce deviation from a normal distribution. Although our playback-based method should ensure very high detection rates, in order to take into account the possible effect of different survey effort we also included the number of visits among the tested predictors in the analysis. As its effect was negative for Moorhen, which is a nonsense, this variable was excluded from Moorhen analysis. The dependent variable was the square-root transformed maximum number of the two study species (averaged across the two years for sites censused in both years), respectively. Notably, there was no significant difference between 2009 and 2010 counts (t-test: P > 0.28 for Water Rail and > 0.72 for Moorhen), and the two counts were highly correlated (for Water Rail, r > 0.75, P < 0.001; for Moorhen, r = 0.5, P = 0.020), indicating rather limited variation between the two study years.

We modelled abundance of Water Rail and Moorhen at the 28 sites after evaluating the potential occurrence of spatial autocorrelation, which may have affected the results of regression analyses of data with strong spatial structure (Beale et al. 2010). Moran's I was calculated to assess the occurrence of spatial autocorrelation in the number of Water Rails and Moorhens per site. Both Water Rail (P > 0.6) and Moorhen (P > 0.05) abundance were not significantly affected by spatial autocorrelation. Among predictors, distance to the nearest wetland, elevation (the two variables clearly more affected by geographical position of study sites) and number of Coots were the only ones displaying significant Moran's I statistics.

We then used both generalized least squares (gls) models and glms to build models with and without considering potential spatial autocorrelation, respectively. Given that transformed abundances used for analyses approached a normal distribution, we used Gaussian error distribution, and iteratively re-weighted least squares (glms) and maximum likelihood (glss) as methods for model fitting. gls allows the incorporation of spatial structure into the error of the model, and is considered among the techniques with the best performance for the analysis of spatial data (Dormann 2007, Beale et al. 2010). Gaussian, spherical and exponential spatial correlation structure yielded identical results.

We carried out an aic_c -based model ranking; given that most parsimonious glms have consistently lower aic_c values than gls analogues, we continued analyses with glms only. Notably, stepwise backward procedures led to identification of the same model (identical variables and coefficients) for Water Rail under both glm and gls (positive effect of reed-bed cover and other herbaceous wetland vegetation, and negative effect of isolation), and nearly perfect identity between the two different statistical methods was found also for Moorhen (same variables and nearly identical coefficients: negative effect of elevation, positive effect of reed-bed cover).

After ranking all possible models, the ones with similar support (Δaic_c  < 2) were averaged according to a multi-model inference approach.

The results of the glm selection and averaging procedures were cross-checked by a hierarchical partitioning analysis (Mac Nally 1996, Brambilla et al. 2006, Brambilla, Bassi et al. 2010), which provides estimates of the independent and joint explanatory power of each predictor, considering all possible models in a multivariate regression setting (Chevan & Sutherland 1991, Mac Nally 1996, 2000). Variables with large independent values are most likely to be important for habitat selection (Mac Nally 2000), and consequently relevant for management when models are used as practical tools (Brambilla & Rubolini 2005, Brambilla, Casale et al. 2010). When the variables deemed as most important by selection procedures and hierarchical partitioning are not in accordance, results of the selection procedure should be revised by selecting variables for the final model under the guidance of hierarchical partitioning results (Mac Nally 2000, Heikkinen et al. 2004). The independent contribution (expressed as percentage of the total independent contributions) of each explanatory variable may be used to validate the results of regression analyses (Radford & Bennett 2004, Brambilla et al. 2006).

We performed all analyses in r (www.r-project.org), using the package mumin (Barton 2011) to rank all possible models and to average the most supported ones, and the package hier.part (Walsh & Mac Nally 2005) to carry out the hierarchical partitioning analyses.

RESULTS

The average number of individuals per site ranged between zero and five for Water Rail, and between zero and six for Moorhen, respectively. The abundance of the two species was not correlated (r = 0.09, P = 0.660).

Top-ranked glms for Water Rail are reported in Table 2. Model averaging (Table 2) suggested that Water Rail abundance was positively affected by cover of reed-bed and other wetland herbaceous vegetation, while the distance to the nearest wetland negatively influenced Water Rail abundance (the same factors with the same effect have been also selected by both glm and gls backward procedure; see Materials and Methods). Elevation, number of Coots and other habitat-type variables had no effect on the abundance of Water Rail. Hierarchical partitioning strongly supported results of model selection, as the two variables with the highest independent contribution – i.e. cover of reed-bed (37.2%) and cover of other wetland herbaceous vegetation (25.7%) – were also the most important ones according to aic_c -based model ranking and averaging (all other variables had independent contribution lower than 19% in the hierarchical partitioning analysis).

Table 2. Most parsimonious glms and averaged model for abundance of Water Rail Rallus aquaticus.

	Intercept	Reed-bed	Dist. to w.	Herbaceous	aicc	Delta	ω
Model 1	0.22	0.52	–0.04	0.40	45.54	0.00	0.12
Model 2	0.11	0.46		0.41	46.57	1.03	0.07
Averaged model	0.18 ± 0.15	0.50 ± 0.14	–0.03 ± 0.03	0.40 ± 0.16
Variable importance		1.00	0.63	1.00
For each variable the relative coefficient is reported;. ω, variable weight; relative variable importance (measured considering the sum of the Akaike weights over all models in which that variable appears [Burnham & Anderson 2002]) is also reported; for variables' acronyms, see Table 1; all parameters as calculated by mumin in r.

Top-ranked glms for Moorhen are reported in Table 3. Model averaging suggested that Moorhen abundance was positively affected by cover of reed-bed and negatively by elevation (the same factors with the same effect have been also selected by both glm and gls backward procedure; see Materials and Methods); moreover, number of Moorhens was also negatively related to cover of woody vegetation, but the effect was weak and non-significant (Table 3), consistent with backward procedure and hierarchical partitioning.

Table 3. Most parsimonious glms and averaged model for abundance of Moorhen Gallinula chloropus.

	Intercept	Reed-bed	Woody	Elevation	aicc	Delta	ω
Model 1	1.44	0.36		–1.51 × 10^−3	58.77	0.00	0.15
Model 2	1.65	0.44	–0.11	–1.80 × 10^−3	60.62	1.85	0.06
Averaged model	1.50 ± 0.30	0.38 ± 0.17	–0.03 ± 0.07	–1.59 × 10^−3 ± 6.42 × 10^−4
Variable importance		1.00	0.28	1.00
For each variable the relative coefficients is reported; for variables' acronyms see Table 1; all parameters as calculated by mumin in r.

Hierarchical partitioning again strongly supported results of model selection: the two variables that had the highest independent contribution were cover of reed-bed (31.4%) and elevation (38.7%), which were also the most important variables according to aic_c -based model ranking and averaging (all other variables had independent contribution lower than 8% in the hierarchical partitioning analysis).

DISCUSSION

The distribution and abundance of biological species may be affected by several factors, including climate, habitat type, spatial configuration of habitats and inter- and intra-specific interactions (McCoy & Mushinsky 1994, Katzner et al. 2003, Brambilla, Bassi et al. 2010).

Rallids are poorly investigated bird species whose elusive habits often have precluded deep investigation of their autoecology (Brambilla & Jenkins 2009). With the present study, we have provided a comparative analysis of two species belonging to this family with different degrees of ecological specialization. Water Rail in our study area represent a model particularly suited to assessing the effects of habitat types and site isolation on the abundance of a target species: it is a habitat specialist species tied to wetlands (Jenkins & Ormerod 2002), which in the study area occur as fragments embedded within an unsuitable landscape matrix (Pedrini et al. 2005). Therefore, it is likely that no occupancy of other sites occurred outside the wetland network we investigated. Wetland fragmentation allowed us to treat each site as a unit, and to assess the relative importance of wetland area, habitat composition and presence of potentially interacting species in determining the abundance of the study species. Moorhen is a more widespread species, which also inhabits other sites outside the wetland network investigated here, thus being less ideal for evaluating the effect of geographical isolation. On the other hand, Moorhens are even more likely to display inter-specific interactions with the larger Coot, being the species more frequently attacked by it (Cramp & Simmons 1980), but also being in some cases attracted by Coots, which can be used by Moorhens as a cue for site choice (Sebástian-González et al. 2010). Moorhen is thus a better candidate to test the effect of inter-specific interactions.

As expected, Water Rail abundance was strongly affected by habitat and isolation, while Moorhen abundance was related to reed-bed cover and elevation. Distance to the nearest wetland site is irrelevant for the latter species, which also inhabits other types of wet habitats interspersed within the landscape matrix (Pedrini et al. 2005).

Our results confirmed at the landscape scale what has been already found at the local scale for the Water Rail: the key importance of reed-bed habitats and the secondary importance of other herbaceous vegetation, which are likely to substitute reed-bed where it is lacking (Brambilla & Rubolini 2004), have been reported from other studies carried out at small spatial scales and using fine variables measured in the field (Jenkins & Ormerod 2002, Brambilla & Rubolini 2004, De Kroon 2004).

Results of the present study also highlighted how isolation of wetland patches negatively affected the number of Water Rails; such a pattern is rather frequent in terrestrial vertebrates (McCoy & Mushinsky 1994, Virgós 2001). At the regional scale, the conservation of the vulnerable Water Rail (Pedrini et al. 2005) depends on the preservation and correct management of reed-bed and, secondarily, of other herbaceous wetland vegetation, even in small wetland sites: total area does not affect species' abundance, and some small sites (area < 2 ha, including the smallest of our study areas) are occupied by the species. A similar preference for reed-bed, independent from wetland site, has been reported for Reed Buntings Emberiza schoeniclus breeding in wetland fragments (Pasinelli et al. 2008). On the contrary, wetlands located far from other ones are unlikely to be important for Water Rail, until their isolation is attenuated by wetland restoration in the landscape matrix. Reconstructing a network of wetlands in the drained landscapes of the Trento province would favour this and other regionally threatened species (Pedrini et al. 2005).

Although it is a generalist species, Moorhen also strongly benefits from reed-bed cover, whereas it is negatively affected by elevation – being probably tied to warmer areas at low elevation (Cramp & Simmons 1980) – and marginally also by cover of woody wetland vegetation; the latter playing a minor importance, as suggested by the results of model averaging procedure, backward procedure and hierarchical partitioning.

Our study further confirms how explanations of species occurrence and abundance in fragmented landscapes must consider variables describing both habitat quality and spatial configuration (Virgós 2001, Johansson & Ehrlén 2003).

ACKNOWLEDGEMENTS

We are very grateful to A. Bertocchi, A. Franzoi, O. Negra, K. Tabarelli de Fatis, R. Bergamini, M.C. Deflorian, S. Ferretti, E. Osele, M. Segata, G. Stefani, C. Tattoni, G. Volcan for help with fieldwork, and to G.F. Ficetola, M. Girardello, D. Rubolini and two anonymous reviewers for useful comments on a first draft of the manuscript.