Population spatial structure and migration of three butterfly species within the same habitat network: consequences for conservation

Authors


Michel Baguette (fax + 32 10 47 34 90; e-mail baguette@ecol.ucl.ac.be).

Summary

1. In this study we addressed the question of the generalization of population viability analysis by comparing migration of different butterfly species in the same habitat network. We compared (i) the population spatial structure and (ii) the migration between local populations in three butterfly specialist species living in the same habitat (chalk grassland) within the same habitat network (eight patches) at the same time.

2. For the three species, population structures within the study system were considered as metapopulations. However, their patch occupancy and migration patterns differed strongly, from a species with six small populations, one being apparently isolated (Cupido minimus), to a species with eight populations connected by individual movement with all others within the landscape (Melanargia galathea).

3. Patch area of both donor and receiver sites affected migration rate in M. galathea. Patch quality may interfere with this area effect in Aporia crataegi. Increasing distance between patches decreased both the probability of movements between local population and the number of moving butterflies. The distribution of migration distances differed between A. crataegi and M. galathea: this distribution followed an inverse-power function for A. crataegi, and an exponential-negative function for M. galathea.

4. These differences in population structure and migration patterns will have consequences for population viability analyses: the persistence of C. minimus cannot be expected if the habitat network is designed on the basis of the requirements of the two other species. Therefore, metapopulation persistence analyses have to remain species-specific. As such analyses are impossible to perform on every threatened species living in a given habitat, we propose a preliminary screening of the most fragile candidate based on its population structure and mobility.

Introduction

Loss and fragmentation of suitable habitats are the main causes of decline in many species (Caughley 1994). In this context, the establishment of suitable habitat networks is a key issue in biological conservation. Metapopulation biology is a conceptual framework for the processes allowing the persistence of species within such networks (Gilpin & Hanski 1991; Hanski & Gilpin 1997). Metapopulations are groups of local populations exchanging individuals across unsuitable habitats. Each population has its own probability of extinction and (re)colonization. The dynamics of each local population within the metapopulation will depend on (i) local processes (birth and death) at low migration rate and (ii) regional processes (immigration and emigration) at high migration rate (Hill, Thomas & Lewis 1996). Migration is really the ‘glue’ between local populations within a network of suitable habitats (Hansson 1991). Metapopulation structures have been found in many organisms: plants (Ouborg 1993), mammals (Verboom, Lankester & Metz 1991), birds (Verboom et al. 1991), amphibians (Sjögren 1991) and insects, including mainly butterflies (Thomas & Harrison 1992; Thomas, Thomas & Warren 1992; Thomas & Jones 1993; Baguette & Nève 1994; Hanski, Kuussaari & Nieminen 1994; Hanski & Kuussaari 1995; Nève et al. 1996; Lewis et al. 1997).

Several theoretical models have been developed in order to make quantitative predictions on metapopulation persistence (Akçakaya 1994; Hanski & Thomas 1994; Hanski 1994). Such models, parameterised with field data, have been tested successfully on butterfly metapopulations (Hanski & Kuussaari 1995; Thomas & Hanski 1997). Because models test the influence of the spatial arrangement of suitable habitats on metapopulation survival, this approach can provide guidelines on landscape management for the long-term persistence of threatened species. However, as the models are parameterised with species-specific data, the question remains about the applicability of such predictions from one species to others. This question is crucial: it is often impossible to perform detailed analyses of population viability on every threatened species living in a given habitat.

In this study we addressed the question of the generalization of population viability analysis by comparing migration of different butterfly species in the same habitat network. We compared (i) the population spatial structure and (ii) the migration between local populations, in three butterfly specialist species living in chalk grassland at the same time. As we assumed that adult size was related to maximal migration distance, we selected one small-, one medium- and one large-sized species. As migration is the process connecting local and regional population dynamics, we investigated to what extent populations of each species were connected by exchanges of individuals within the same network of suitable habitat patches. We also compared how the migration rates of the different species depended on physical factors of the landscape (patch area, distance among patches). Metapopulation dynamics models have so far used exclusively negative-exponential functions to describe the distribution of migration distances. However, Hill, Thomas & Lewis (1996) have shown that an inverse-power function could fit the empirical distribution of migration distances better than the negative-exponential function. We compared the use of these two functions on different species, in order to find the best model describing the migration distance distribution between sites. Finally, we investigated whether or not a population viability analysis based on a given species could be generalized to the others.

Materials and methods

The species

The three species studied are classified as chalk grassland specialists because their larval development is restricted to chalk grasslands by the presence of their larval food-plants and the need for particular microclimatic conditions. The small species studied was the little blue Cupido minimus Fuesslin. Adults have a mean wing-span of 25 mm; larvae feed on flowers of Anthyllis vulneraria L. The medium-sized species was the marbled white Melanargia galathea L. Adults have a mean wing-span of 48 mm; larvae food-plants are grasses, mainly Festuca spp. and Brachypodium pinnatum L. The large species is was the black-veined white Aporia crataegi L. Adults have a mean wing-span of 68 mm; larvae feed on young shoots of Prunus spinosa L. and Crataegus monogyna L. growing in and at the margin of chalk grasslands.

Study system and field methods

The study system (3 × 3 km) was located in southern Belgium, in the valley of the Viroin river. It consists of a flood plain with fertilized grasslands, although some chalky hills with fragments of xeric chalk grasslands are scattered within the landscape. The whole area was searched for suitable habitats (defined by their physiognomy and the presence of food-plants) and adult butterflies. We located eight patches of chalk grasslands on three hills: La Montagne aux Buis (patches Mb1, Mb2 and Mb3), Le Fondry des Chiens (patches Fc1 and Fc2) and Les Abannets (patches Ab1, Ab2 and Ab3) (Fig. 1). Migration between patches located on the same hill was more likely to occur than migration between hills: the chalk grasslands were connected on each hill by pathways, open areas within forests that were used by adult butterflies as corridors, while the different hills were separated by unsuitable habitats (mainly fertilized grasslands).

Figure 1.

Figure 1.

Map of the study site. Shaded areas indicate suitable habitat patches (chalk grasslands), located on three hills limited by the surrounding lines. Habitats between the hills are unsuitable for butterflies (mainly fertilized pasture).

Each patch was visited daily (weather permitting) between 28 May and 28 July 1996, and standard transects were walked at each visit. The field method was MRR: mark–release–recapture. All adult butterflies observed along the transects were netted and marked individually using a thin-point permanent pen on the underside of the left wing. The date, location of the patch and individual mark were recorded at each recapture. Patch area and between patch-distances (measured as the straight line between the centres of the patches) are shown in Table 1.

Table 1.  Patch surface (ha) and distance between patches (m)
SurfaceAb1
0·94
Ab2
0·62
Ab3
0·94
Fc1
2·42
Fc2
0·30
Mb1
0·56
Mb2
0·90
Mb3
1·01
Distance
Ab1
Ab2304       
Ab3158220      
Fc1607762751     
Fc2586660705197    
Mb120401809188325682442   
Mb217931589163523502240326  
Mb315311334137320931989559264 

Data analysis

The proportion of residents in a given patch was computed as the fraction of same-site recapture events within the patch/total number of recaptures of butterflies marked in that patch. As stressed by Hill, Thomas & Lewis (1996), a high proportion of residents will indicate that local processes (birth and death) are more important to local dynamics than regional processes (immigration and emigration) and the system can be considered as a metapopulation. Conversely, if most individuals move between patches the system can be described as a single, large patchy population (Harrison 1991; Hill, Thomas & Lewis 1996).

The pattern of migration between patches was analysed in two ways. First, the probability of detected individuals moving between patches was described by use of logistic regression (SAS 1990; Proc Logistic), using movement/no movement between pairs of patches as the dependent variable. log10 area of both donor and receiver patches and log10 distance between patches were used as potential predictors. Secondly, the effect of patch characteristics on the absolute number of movements was analysed by multiple regression (SAS 1990; Proc Reg). The dependent variable was the absolute number of individuals moving between sites (log-transformed) for all pairwise combinations (56 possible cases). Independent variables (the same as above) were added in a forward stepwise selection procedure. Neither analysis represents the ‘true’ dispersion of individuals around the source of migration, as recaptures originate from a few points only. However, this procedure is the most pragmatic method available for the study of migration of mobile organisms, spending most of their life in suitable habitat patches.

For each species, the inverse-power function and the negative-exponential function were fitted to the observed distribution of migration distances. Linearly transformed data were analysed using regression analysis weighted by the number of recaptures.

Results

Mrr study

A total of 291, 4041 and 323 butterflies was marked, for A. crataegi, M. galathea and C. minimus, respectively. Aporia crataegi was recorded in the eight patches, but was more abundant in patches Ab2, Fc1 and Mb3 (Table 2). Melanargia galathea was also recorded in all patches (Table 2). Cupido minimus was recorded mainly in patches Ab2 and Fc1 and was absent from patches Mb1 and Mb2 (Table 2). A total of 47, 766 and 72 per capita same-site recaptures (SSR) was recorded for A. crataegi, M. galathea and C. minimus, respectively. The fraction of SSR/total number of per capita recaptures is an indication of species’ residence; this fraction was 58% (47/81), 64% (766/1173) and 91% (72/79) for A. crataegi, M. galathea and C. minimus, respectively.Tables 3, 4 and 5 show the distribution of per capita recaptures for each species. Movements between the three hills were recorded for both A. crataegi and M. galathea, but only between Les Abannets and Le Fondry des Chiens for C. minimus. For A. crataegi, M. galathea and C. minimus, respectively, a total of 34, 432 and seven individuals moved between sites; the longest distances moved were, respectively, 1589 m (one individual), 2568 m (two individuals) and 762 m (three individuals).

Table 2.  Proportion (%) of individuals marked in the eight patches (n is the number of butterflies)
Patch codeA. crataegi (n = 291)M. galathea (n = 4041)C. minimus (n = 323)
Ab18122
Ab2221556
Ab35164
Fc1242130
Fc2332
Mb1470
Mb2250
Mb332216
Table 3.  Distribution of A. crataegi per capita recaptures
Receiver site
Donor siteAb1Ab2Ab3Fc1Fc2Mb1Mb2Mb3
Ab1221     
Ab22314   1
Ab321      
Fc1251114   
Fc2   3    
Mb1       1
Mb2 1     1
Mb3     2 31
Table 4.  Distribution of M. galathea per capita recaptures
Receiver site
Donor siteAb1Ab2Ab3Fc1Fc2Mb1Mb2Mb3
Ab175251942423
Ab21310130233 5
Ab326431047 217
Fc15136206382310
Fc2 221610  1
Mb145   10724
Mb212   122622
Mb31564 1921234
Table 5.  Distribution of C. minimus per capita recaptures
Receiver site
Donor siteAb1Ab2Ab3Fc1Fc2Mb1Mb2Mb3
Ab1
Ab2 581 1   
Ab3 2      
Fc1 3 12    
Fc2
Mb1
Mb2
Mb3       2

Emigration and immigration

Because of the small number of movements recorded for C. minimus (only four movements between sites were recorded for 56 possible combinations), the relationship between emigration/immigration processes and patch area was analysed for A. crataegi and M. galathea only. The eight patches were pooled in three size classes: small (0·30–0·62 ha, n = 3), medium (0·90–1·01 ha, n = 4) and large (2·42 ha, n = 1). Both emigration and immigration rates decreased with increasing patch area for M. galathea (Fig. 2). However, in absolute numbers, more emigrants came from the large patch compared with medium or small patches (number from the large patch = 77, from the medium = 59 per patch, and from the small = 39 per patch). The situation was different for A. crataegi: both emigration and immigration rates were highest from small patches, but smallest from medium patches (Fig. 2). The low emigration rate in medium-sized patches was mainly due to an excess of residents in patch Mb3 (31 individuals, 38% of the recaptures and 66% of the same-site recapture).

Figure 2.

Figure 2.

Effect of patch area on emigration and immigration rates.

Presence or absence of movements was checked for all pairwise combinations within the study site (56 possible cases). For A. crataegi, the logistic model classified 90·5% of presence/absence of movements correctly. The probability of movements between sites increased with the patch area of the donor site and decreased with distance between sites. The patch area of the receiver site had no significant effect on the probability of individuals moving between sites (Table 6).

Table 6.  Logistic regression of movements of A. crataegi between habitat patches. The model classified 90·5% of movements correctly. The Wald χ2 tests whether the coefficient is significantly different from 0. n1/n2: number of patches where a movement did/did not occur
 Parameter estimateSEWald χ2Pn1n2
Constant6·115·041·470·221739
Log10 distance–5·321·3715·160·0001  
Log10 area (donor)1·590·764·360·04  
Log10 area (receiver)0·450·710·390·53  

For M. galathea, the logistic model classified 87·7% of presence/absence of movements correctly. The probability of movements between sites decreased again with distance between sites. The effect of area of the donor patch was almost significant (P = 0·08). The patch area of the receiver site had no significant effect on the probability of individuals moving between sites (Table 7). For M. galathea, the high number of individuals moving between sites allowed the effect of patch characteristics on the absolute number of movements by multiple regression to be tested. The dependent variable was the absolute number of individuals moving between sites (log-transformed) for all pairwise combinations (56 possible cases). The number of individuals moving between sites decreased with increasing distance. Here again, the effect of area of the donor site was almost significant (P = 0·08). No significant effect of the patch area of the receiver site was detected (Table 8).

Table 7.  Logistic regression of movements of M. galathea between habitat patches. The model classified 87·7% of movements correctly. The Wald χ2 tests whether the coefficient is significantly different from 0. n1/n2: number of patches where a movement did/did not occur
 Parameter estimateSEWald χ2Pn1n2
Constant2·129·340·050·824412
Log10 distance–4·941·996·150·01  
Log10 area (donor)3·562·072·940·08  
Log10 area (receiver)0·180·720·060·80  
Table 8. anova results of multiple regression of the number of individual movements of M. galathea (log-transformed) between habitat patches. Independent variables were added using a forward stepwise selection procedure. Variables were incorporated in the model in the sequence shown. The coefficients of determination are based on the inclusion of the named variables and all preceding variables
Source of variationd.f.SSMSF P
Model36·692·2323·270·0001 
Error403·830·09   
VariableParameter estimateSEFPR2 
Constant3·750·37104·3950·0001  
Log10 distance–1·030·1367·190·00010·58 
Log10 area (donor)0·150·083·240·080·62 
Log10 area (receiver)–0·120·090·160·200·64 

Modelling migration

The inverse cumulative proportion of individuals moving certain distances is shown at Fig. 3. These data were fitted to a negative-exponential function. The probability (P) of an individual moving a certain distance D is:

Figure 3.

Figure 3.

Distribution of recapture frequency with distance.

P = e –kD

where k is the migration constant describing the shape of the species-specific exponential curve; 1/k is the average distance moved by individuals between patches. ln(P) was regressed upon distances; regressions were weighted according to the number of recaptures (R2 = 0·89, F1,6 = 46·64, P = 0·0005 for A. crataegi;R2 = 0·70, F1,12 = 27·95, P = 0·0002 for M. galathea). The value k was the slope of the regression:

ln(P) = –6·60(SE = 0·97)D(therefore1/k = 151 m)forA.crataegi
ln(P) = –3·63(SE = 0·69)D(therefore1/k = 275 m)forM.galathea

The same data were also fitted to an inverse-power function. The probability (P) of an individual moving a certain distance D was:

P = aD–norlnP = lna – nlnD

ln(P) was regressed upon ln(D) (R2 = 0·72, F1,5 = 13·14, P = 0·015 for A. crataegi;R2 = 0·83, F1,11 = 55·52, P = 0·0001 for M. galathea). The value n was the slope of the regression:

ln(P) = –4·84(SE0·47) – 1·18(SE = 0·33)lnD(therefore1/n = 847 m)forA.crataegi
ln(P) = –4·54(SE = 0·20) – 1·00(SE0·13)lnD(therefore1/n = 1000 m)forM.galathea

Both functions fitted the data. However, the inverse-power function fitted the data for M. galathea (R2 = 0·83 vs. 0·70) while the exponential function provided a better fit for A. crataegi data (R2 = 0·89 vs. 0·72).

Discussion

Spatial structure of populations and conservation

Migration of butterflies between habitat patches was observed for each species (Tables 3–5). The majority of recapture events were individuals remaining in the patch in which they were marked (58%, 64% and 91% for A. crataegi, M. galathea and C. minimus, respectively). Therefore, for the three species, population structures within the study system could be considered as metapopulations [sensuHanski & Simberloff (1997), ‘set of local populations within some large area, where typically migration from one local population to at least some other patches is possible’]. However, patch occupancy and migration patterns differed strongly, despite the fact that these three species were specialist butterflies living in the same habitat. Adults of M. galathea were observed in all the patches in the study system, and all the local populations were connected with each other by individual movements. Local populations of A. crataegi were found in the eight habitat patches; migration was detected from each local population to at least one other patch in the system. Adults of C. minimus were observed in six of the eight habitat patches and movements were detected only between three local populations; the resident fraction (same site recaptures/total recaptures) was extremely high (> 90%). The single small population located on the Montagne aux Buis in patch Mb3 (six individuals) was probably isolated from the other local populations of the system.

These three structures will have different consequences in a metapopulation persistence analysis. For M. galathea, the habitat network and its configuration are sufficient: if one local population, or even if all the patches located on the same hill, becomes extinct due to stochastic or demographic events, empty habitat(s) could be colonized quickly from other patches in the network. For A. crataegi, the recolonization could be slower, as all the patches are not interconnected by individual movements within the network. For C. minimus, the habitat network is clearly not sufficient: extinction of one or some local populations will be a serious threat to the persistence of the species within the system, due to its weak migration power. As metapopulation processes [extinction (re)colonization, migration] are more likely to occur within each hill than between hills, conservation of this species implies the creation of more closely related suitable patches on each hill.

Landscape structure, migration and conservation

Patch area strongly affected the migration rate for M. galathea: proportions of individuals emigrating from, and immigrating into, small patches were higher compared with larger patches. Large patch area of the donor site has been identified as a factor decreasing emigration in butterflies (Hill, Thomas & Lewis 1996; Kuussaari, Nieminen & Hanski 1996). This effect was explained by patch geometry: small patches have a relatively high perimeter to area ratio so butterflies are more likely to encounter patch boundaries than in larger patches. The effect of patch area on immigration is due to the fact that, for a comparable density of butterflies in small and larger patches, the number of immigrants is proportional to the patch diameter. The ratio patch diameter/patch area is higher in small patches than in large ones and the proportion of butterflies immigrating into the patch decreases with increasing patch area (Hill, Thomas & Lewis 1996).

Our results on A. crataegi suggest that patch quality may interfere with the effect of patch geometry, at least on emigration. The proportion of residents within the patch Mb3 (medium size, area = 1·01 ha) was extremely high (31 recaptures/32 individuals marked). In the larger patch Fc1 (area = 2·42 ha) the proportion of residents was only 11 recaptures/24 individuals marked. It should be noted that young shoots of P. spinosa and C. monogyna were extremely abundant in patch Mb3 in 1996, due to felling of trees in and around the chalk grassland during the winter 1995–96. This abundance of larval food-plants may have ‘fixed’ females in the patch, themselves ‘fixing’ males.

Distance between patches is a crucial factor, which must be considered together with the migration power of the species in order to determine which spatial scale is appropriate to build habitat networks where metapopulation processes are likely to occur. We have seen before that the present network is not convenient for C. minimus. The logistic regressions showed that for A. crataegi, as for M. galathea, the probability of movements between sites decreased with distance. However, the distance between the more extreme sites in the network would not prevent recolonization in cases of local extinction. For M. galathea, the absolute number of migrants decreased with increasing distance between sites.

Patch area of the receiver site had no effect on the probability of movements between site, for both A. crataegi and M. galathea, as well as on the absolute number of butterflies migrating between patches for M. galathea.Hill, Thomas & Lewis (1996) found that the probability of migration between patches increased with the receiver patch area in a metapopulation of Hesperia comma. However, in their study system the area of the small patches was much smaller (0·01 ha). Our results show that, in our study system, with a smaller patch area of 0·30 ha migrant butterflies easily detect suitable habitat patches.

Models of migration distances and conservation

In the same habitat network, the distribution of migration distance differs strongly from one species to the other. For A. crataegi, the migration distances between patches fitted an exponential negative function (ENF) better than a inverse-power function (IPF) (R2 = 0·89 vs. 0·72). For M. galathea, the IPF provided a better fit (R2 = 0·83 vs. 0·70). The differences between the functions can be illustrated by the probability that an emigrant would move 2500 m, i.e. the longest distance between patches in the study system. For A. crataegi, the ENF gave a probability of 6·83 × 10–8, while the IPF gave a corresponding value of 2·33 × 10–2. For A. crataegi, the ENF gave a probability of 1·14 × 10–4, while the IPF gave a corresponding value of 2·67 × 10–2. ENF and IPF also differed strongly in their predictions of the average distances travelled by individual migrants: 151 m (ENF) vs. 847 m (IPF) for A. crataegi; 275 m (ENF) vs. 1000 m (IPF) for M. galathea. These differences in predictions show the need for more than one single function in order to describe migration distance distribution in theoretical models.

The parameters of migration used in this comparative study are relatively simplistic. More sophisticated methods of movements based on MRR results are now available, e.g. the virtual migration model (Wahlberg, Moilanen & Hanski 1996). This method allows the comparison of movements (i) between species within the same network of habitat patches or (ii) within the same species between different networks of habitat patches. The main advantage of this method is the estimation of the absolute rate of emigration and immigration. However, such methods need large data sets; in the present study only one species (M. galathea) had a sufficiently large sample size.

Potential use of the work for conservation

The message of this paper is that metapopulation persistence analyses have to remain species-specific: from the results of the present study, we cannot expect the persistence of the less mobile species (C. minimus) if the habitat network is designed on the basis of the requirements of the two other species. As detailed population viability analyses are impossible to perform on every threatened species living in a given habitat, we propose a preliminary screening of the most fragile candidate based on its population structure and mobility, following the same procedure as we used in the present study. Size of butterflies may not be a good predictor of their mobility, as M. galathea moved longer distances than the larger A. crataegi.

Acknowledgements

David Boughton, Andreas Erhardt and Gabriel Nève gave invaluable comments on a first draft of the manuscript. Sandrine Liégeois contributed to the field work. Léon Woué and people from the Centre Marie Victorin provided hospitality. Logistical assistance was supplied by the ‘Maison de l’UCL à Matagne-la-Petite’. Special capture licenses and financial support were given by the Ministère de la Région Wallonne (Belgium). This is contribution number 2 of the Biodiversity Center of the Université catholique de Louvain.

Received 6 November 1998; revision received 29 September 1999

Ancillary