Study species and their field surveys
Maculinea are highly specialised myrmecophilous butterflies, requiring specific foodplants and specific Myrmica host ants to complete their life cycle (Thomas 1995). The host ants are typically scarce but widely distributed, while the foodplants are highly abundant but occur in patches, which can thus be regarded as Maculinea habitat patches (Nowicki et al. 2005c, 2007; Anton et al. 2008). Consequently, Maculinea often form classic metapopulation systems (Nowicki et al. 2007; Dierks and Fischer 2009; but see Nowicki et al. 2009). M.
nausithous and M. teleius typically occur sympatrically in wet meadows, sharing the same larval foodplant, Sanguisorba officinalis, which is also a primary nectar source for their adults (Elmes and Thomas 1992; Thomas 1995). Other nectar plants occasionally used by both species, including Vicia cracca, Betonica officinalis, Cirsium arvense, and Veronica longifolia (Thomas 1984; Sielezniew and Stankiewicz-Fiedurek 2013; authors’ unpubl. data) grow commonly within grasslands and fallow lands, but very rarely occur in forests.
Both species were surveyed with mark-recapture methods in six metapopulations located in the Czech Republic, Germany, Poland, and Slovenia (Fig. 1). Butterflies were captured with entomological nets, individually marked with numbers written on the underside of their hind wing using permanent markers, and immediately released at the place of capture. For each capture we recorded the butterfly number, its species and sex, as well as the habitat patch number. The sampling was conducted daily to every second day (with few gaps due to unfavourable weather) between 9:00 and 17:00. Sampling intensity on particular habitat patches was adjusted to their area and butterfly numbers in order to ensure uniform capture probabilities across all the patches within a particular metapopulation.
In each metapopulation an intensive survey, in which mark-recapture sampling was conducted at a large number of habitat patches throughout the entire flight period, i.e. roughly from early July to late August, was performed in 1 year which was different for each metapopulation (see Table 1). Although mark-recapture studies were also carried out in the investigated metapopulations in other years, they were not comprehensive enough for dispersal analysis, because they were limited to too few local populations (cf. Stettmer et al. 2001; Nowicki et al. 2013). A clear exception in this respect was the metapopulation near the Czech town of Přelouč, which was intensively surveyed for seven consecutive years (Nowicki and Vrabec 2011). However, except for 1 year these surveys were restricted to the relatively small core fragment of the metapopulation. Thus for the sake of consistency in the spatial extent of the investigated metapopulations, we have not included these surveys in the present analysis, especially that the Přelouč metapopulation was the smallest.
Table 1 Characteristics of the investigated metapopulations of Maculinea butterflies
The information about the study sites is summarised in Table 1. The investigated metapopulations clearly differed in their matrix composition. In three of them (Přelouč, Kraków, Teisendorf) the matrix consisted predominantly of open lands, including meadows, fallow lands and fields. The remaining three (Dečin, Steigerwald, Slovenske Gorice hereafter Gorice) had a matrix dominated by forest. The proportion of forest within the minimum convex polygons encompassing all the Maculinea habitat patches in a particular metapopulation was below 20 % in the former group, while in the latter it reached ca. 50–70 %. We note that the proportion of forest was little changed if we extended the minimum convex polygons to include 500-m or 1-km buffer zones around each habitat patch. Obviously, a high proportion of forest in the matrix does not necessarily imply that dispersing butterflies often need to cross forest fragments, if most of them are located in marginal parts of a metapopulation. Hence, in order to account not only for the amount of forest in the matrix, but also for its location versus habitat patches, we calculated the proportions of forest along cross sections of the lines linking the centres of habitat patches in each metapopulation. Nevertheless, the results were almost identical to the proportions of forest in the matrix area (Table 1).
Other inhospitable landuse types, such as urban areas and waters (rivers and lakes), had a consistently low proportion of the matrix area, reaching only a few percent at all the study sites. The investigated metapopulations had fairly similar spatial dimensions, which corresponded to similar ranges of potential inter-patch movement distances, except for the Přelouč one, which was approximately half the size (Fig. 1; Table 1). Moreover, habitat patch sizes were also comparable across all the metapopulations, although very large patches, exceeding 10 ha, existed only in Kraków and Steigerwald (Table 1).
In an earlier study we found that dispersal within metapopulations of Maculinea butterflies is negatively affected by their strong spatial isolation (Bonelli et al. 2013). We also demonstrated positive density-dependence of emigration rate (but not of other dispersal parameters), leading to its sharp increase at densities exceeding carrying capacity (Nowicki and Vrabec 2011). In this context, it is important to stress that none of the investigated metapopulations was strongly isolated, with neighbouring metapopulations being located 2–4 km away in each case (Table 1). Such isolation distances are close to the maximum movement distances recorded for Maculinea butterflies (Nowicki et al. 2005b; see also the “Results” section). Consequently, inter-metapopulation movements are likely to occur, but only sporadically, and thus it is valid to restrict dispersal analyses to the investigated metapopulations (cf. Bonelli et al. 2013). Besides, the available data from other years indicate that in the years used in the present study butterfly abundances in all the metapopulations were at their normal levels below carrying capacities. Consequently, neither of the aforementioned effects is likely to influence our estimates of dispersal parameters.
Dispersal analysis
In the original data sets many land fragments covered with S. officinalis foodplants were regarded as separate habitat patches based on, for example, different land ownership, even though they were directly adjacent to each other (cf. Hovestadt et al. 2011). Thus, to ensure that habitat patches are defined in a uniform way across all the metapopulations, for the purpose of dispersal analysis we have pooled together all the patches that were separated by less than 50 m. The 50-m threshold was adopted after Nowicki et al. (2007), who found that such a distance is enough make local populations of Maculinea butterflies demographically independent. Pooling together closely lying patches also allowed disregarding short-distance movements between them, which are likely to represent daily routine flights rather than genuine dispersal (Hovestadt et al. 2011).
The mark-recapture data collected were analysed with the Virtual Migration (VM) model (Hanski et al. 2000), which is a well-established standard for investigating dispersal in metapopulations. Since the rationale and a detailed description of the model have been provided elsewhere (Hanski et al. 2000; Petit et al. 2001), in the present paper we only briefly outline its features. Dispersal within a metapopulation is modelled using six parameters, which include: (i) mortality in habitat patches (μp); (ii) emigration propensity (η), defined as daily emigration rate scaled to 1 ha patch; (iii) emigration scaling with patch area (ζem,); (iv) immigration scaling with target patch area (ζim); (v) scaling of dispersal mortality with natal patch connectivity (λ); and (vi) distance dependence of dispersal (α).
Mortality in habitat patches is independent of dispersal and constrained to be constant across all the patches within a metapopulation. Emigration propensity reflects the emigration rate scaled to an imaginary 1-ha patch. Both emigration and immigration are assumed to depend on patch area (A
j
), with the power relationship being negative for emigration (\( E_{j} = \eta A_{j}^{{\zeta_{\text{em}} }} \), where ζem < 0, η represents daily emigration rate from 1 ha patch) and positive for immigration (\( I_{j} \sim A_{j}^{{\zeta_{\text{em}} }} \), where ζim > 0). The probability of successful dispersal (dispersal survival, φmj
) is modelled to increase sigmoidally with the natal patch connectivity (S
j
, defined as in Hanski 1994): \( \varphi_{{{\text{m}}j}} = {{S_{j}^{2} } \mathord{\left/ {\vphantom {{S_{j}^{2} } {(\lambda + S_{j}^{2} )}}} \right. \kern-0pt} {(\lambda + S_{j}^{2} )}} \). The square root of λ is thus the equivalent of patch connectivity, for which half of emigrants from the patch die during dispersal. The α parameter defines the dispersal kernel.
We opted for the negative exponential function (NEF) as the kernel (as in Hanski et al. 2000), in which mean dispersal distance (measured in km) corresponds to 1/α, as it was found to describe movements of Maculinea butterflies quite well in previous studies (Hovestadt and Nowicki 2008; Nowicki and Vrabec 2011). Nevertheless, the estimates of all other VM model parameters changed hardly at all, when we attempted the inverse power function (IPF), preferred as the kernel by some authors (Schtickzelle et al. 2006; Fric et al. 2010). The inter-patch movement distances used for fitting the kernel were measured between centres of patches, which is a standard approach (Hanski et al. 2000; Matter et al. 2005; Hovestadt et al. 2011). Alternative solutions, such as applying edge-to-edge distances or dividing centre-to-centre distances into within-patch and within-matrix fractions, for which separate kernels are fitted (see Matter et al. 2004), would in fact change very little as patch dimensions were typically small as compared with inter-patch distances.
The analysis was conducted using the VM2 program (Hanski et al. 2000). Its goodness-of-fit tests indicated that the VM model fitted our mark-recapture data well. The observed numbers of emigrants, immigrants, and residents were not significantly different from those predicted by the model, except for the small number of patches with few (<10) captures recorded. However, low numbers of captures are well known to bias the goodness-of-fit tests toward more significant values (Schtickzelle et al. 2006).
The VM2 program allows the estimation of the VM model parameters together with their 95 % confidence intervals. The parameter estimates are expressed in uniform units and thus they can be used for comparisons between metapopulations, with non-overlapping 95 % confidence intervals indicating statistically significant differences between the estimates (Schtickzelle and Baguette 2003; Schtickzelle et al. 2006). The accuracy of the model estimates is not affected by sample size, though their precision may be reduced in the case of a small sample (Nowicki and Vrabec 2011). It must be stressed that while the estimation of the VM model parameters requires spatial information (area and location) for all the habitat patches within a metapopulation, not all of them need to be sampled with mark-recapture. It is enough that the sampling has been conducted in at least ca. 10 patches (Hanski et al. 2000; Petit et al. 2001), which was the case in our metapopulations (Table 1). We derived the parameter estimates separately for both species in each metapopulation. In addition, we calculated the weighted means and their confidence intervals (Sokal and Rohlf 2012) for both species in ‘open-land matrix’ metapopulations (Přelouč, Kraków, Teisendorf) and ‘forest matrix’ metapopulations (Dečin, Gorice, Steigerwald), with weights being the numbers of butterfly captures.