Connectivity and homogenisation of population sizes: an experimental approach in Lacerta vivipara


  • Jane Lecomte,

    Corresponding author
    1. Laboratoire Ecologie, Systématique et Evolution, UMR 8079, Université Paris Sud XI – Centre d’Orsay, Bâtiment 360, F-91405 Orsay Cedex, France;
      Jane Lecomte, Laboratoire Ecologie, Systématique et Evolution, UMR 8079, Université Paris Sud XI – Centre d’Orsay, Bâtiment 360, F-91405 Orsay Cedex, France. Tel: +33 1 69 15 7657, Fax: +33 1 69 15 7353, E-mail:
    Search for more papers by this author
  • Katia Boudjemadi,

    1. Laboratoire Fonctionnement et Evolution des Systèmes Ecologiques, UMR 7625, Université Paris VI, 7, Quai Saint Bernard, Bâtiment A 7ème étage, F-75252 Paris Cedex 05, France;
    Search for more papers by this author
  • François Sarrazin,

    1. Laboratoire Conservation des espèces, Restauration et Suivi des Populations, UMR 5173, Muséum National d’Histoire Naturelle, 61 rue Buffon, F-75005 Paris, France; and
    Search for more papers by this author
  • Karine Cally,

    1. Station Biologique de Foljuif, Ecole Normale Supérieure, Chemin du Château, F-77140 St. Pierre-lès-Nemours, France
    Search for more papers by this author
  • Jean Clobert

    1. Laboratoire Conservation des espèces, Restauration et Suivi des Populations, UMR 5173, Muséum National d’Histoire Naturelle, 61 rue Buffon, F-75005 Paris, France; and
    Search for more papers by this author

Jane Lecomte, Laboratoire Ecologie, Systématique et Evolution, UMR 8079, Université Paris Sud XI – Centre d’Orsay, Bâtiment 360, F-91405 Orsay Cedex, France. Tel: +33 1 69 15 7657, Fax: +33 1 69 15 7353, E-mail:


  • 1At regional scales, dispersal is known to prevent metapopulation extinction by buffering stochastic processes. Theory predicts that connectivity, through density-dependent dispersal rates, should spatially homogenize population density and synchronize local population dynamics in the long term. However, empirical evidence for the effect of connectivity on synchrony and local population dynamics remains scarce.
  • 2We experimentally manipulated connectivity in order to investigate the homogenisation effect on population size. The experimental design consisted of 16 patches of common lizard populations (Lacerta vivipara), half of which were connected by dispersal. The design allowed us to identify candidates for dispersal in unconnected patches.
  • 3We found that population sizes became spatially more and more homogeneous with time in connected patches, whereas extinctions or demographic explosions were observed in unconnected patches. Juvenile dispersal was density-dependent in connected patches but not in unconnected ones. These results suggest that the loss of connection modifies population functioning by influencing how dispersal is determined by local conditions.
  • 4Finally, population explosions in unconnected patches were followed by a sharp decrease in population size. So non-extinct, unconnected populations did not stabilize. This could be due to over-compensatory density dependence.
  • 5Population viability analysis models suggest that environmental stochasticity and catastrophic events, in addition to the density-dependent process, are required to explain population size variation and extinction.


Habitat loss is one of the most important extinction causes (Wilcove, McLellan & Dobson 1986) and can be characterized by two components: habitat destruction and habitat fragmentation. Habitat destruction results in a decrease of available habitat for specialist species, together with a global decrease of habitat quality. Habitat fragmentation occurs when habitat destruction has led to the relative isolation of remnant fragments. In fragmented habitats, connectivity can play a critical role in determining population dynamics. Connectivity depends on the spatial distribution of favourable and unfavourable habitats and on species dispersal ability in a given environment (Taylor et al. 1993). Moreover the demographic, genetic and behavioural consequences of isolation are directly linked to factors influencing dispersal. In this context, low connectivity acts at a local scale (intrapopulation movements, motivation of individuals to move) and a regional scale (interpopulation movements).

At a regional scale, connectivity can increase metapopulation viability by two main mechanisms. First, connectivity allows recolonization of suitable but extinct patches (Levins 1970; Hanski 1991). Secondly, dispersers can also reinforce non-extinct populations. Crowding or population extinction can be avoided by the ‘rescue effect’ (Brown & Kodric-Brown 1977). Both recolonization and the ‘rescue effect’ can counterbalance demographic stochasticity and buffer environmental stochasticity when environmental variations are not strongly positively spatially autocorrelated (Gilpin 1996; Stacey, Johnson & Taper 1997). These two mechanisms can lead to an increase of mean population size and of number of occupied sites, which in turn enhance metapopulation viability. Theory also predicts that the counterbalancing of stochastic processes could lead to spatially homogenized density, which could lead to synchronized dynamics in the long term (Holt & McPeek 1996; Haydon & Steen 1997; Paradis 1997). Models also suggest that this could undermine temporal population fluctuations, resulting in less complex dynamics on an evolutionary time scale (Gyllenberg, Söderbacka & Ericsson 1993; Doebeli 1995). The few empirical (Paradis et al. 1999), experimental (Holyoak & Lawler 1996) and theoretical studies (Ripa 2000; Ylikarjula et al. 2000) suggesting that connection could lead to synchronized dynamics indicate that the spatial scale of this synchrony varies with species characteristics, with dispersal distance considered, and landscape structure (e.g. patch number). Synchrony due to internal processes acting on a local scale should decrease extinction risks, while synchrony resulting from external processes (e.g. positive autocorrelation of environmental stochasticity) acting on a global scale should increase metapopulation extinction risks (Heino et al. 1997; Stacey et al. 1997). However, dispersal, which is generally supposed to synchronize local dynamics, has already been found to increase global extinction probability (Heino et al. 1997).

Connectivity not only influences dynamics at a regional scale but it can also modify individual behaviour, and influence local dynamics by processes other than the direct effect of arrival or departure numbers (Aars, Johannesen & Ims 1999; Boudjemadi, Lecomte & Clobert 1999; Debinski & Holt 2000). Whereas empirical evidence for the effect of connectivity on extinction/recolonization processes and on the ‘rescue effect’ is beginning to be well documented for animals and plants (McCullough 1996; Hanski & Gilpin 1997), evidence concerning the effect of connectivity on synchrony and on local population dynamics remains scarce (McGarigal & Cushman 2002).

In this paper, we test experimentally the homogenisation effect of connection on population size through dispersal. We used the common lizard (Lacerta vivipara) as a biological model for two main reasons. Although this species is not directly threatened in France (thereby permitting experimentation), it lives in habitats, peat bogs and heathlands, facing an increasing level of fragmentation throughout Europe. Moreover, we have a good understanding of its dispersal behaviour (Clobert et al. 1994; Massot et al. 2002) together with a good knowledge of its habitats (Pilorge 1987) and demography (Clobert et al. 1994), so our experimental results can be compared with natural situations.

Our experimental design was made up of 16 patches. In four pairs of patches (called connected patches), exchanges between patches were possible through dispersal corridors, while in the other ones (called unconnected patches), exchanges were actively prevented. This situation mimicked a recent fragmentation, either having or not having induced a total isolation of populations. We performed this treatment during 4 years in both ‘rich’ and ‘poor’ habitats. We also developed population viability analyses (PVA, Beissinger & McCullough 2002) to check the coherence of our results and to explore the contribution of different sources of stochasticity.


the species

The common lizard is a widespread species, ranging from the polar circle to parallel 40°N across all of Eurasia. This small, viviparous, lacertid lizard (adult snout–vent length 50–70 mm) inhabits fragmented habitats such as peat bogs and heathlands. In our experimental populations, individuals hibernate from mid-October to March. Males are the first to emerge from hibernation and mating occurs soon after female emergence, at the end of April. Parturition occurs 2 months later. Females lay on average five eggs (range 1–13) which hatch within an hour of laying. Neonates are then autonomous and may disperse within the first 2 weeks following birth. Three age classes can be distinguished: juveniles (neonates of the year), subadults (1-year-old individuals) and adults. Subadult females reproduce in lowland populations, but not in mountain populations (Heulin, Osenegg & Michel 1998). In this species, there is no apparent territoriality (Lecomte et al. 1994). Although individuals may interact strongly for mates or food, they do not seem to be territorial (Massot et al. 1992; Lecomte et al. 1994).

experimental design


The experiment took place in the biological station of Foljuif (Saint-Pierre-lès-Nemours, France) from August 1995 to July 1999. The experimental design was made up of 16 patches. Pairs of enclosed patches (10 m × 10 m) were connected by two one-way corridors (Fig. 1a). All connections had the same orientation to limit microclimatic and biotic heterogeneity. Nylon nets covered each patch to avoid avian predation. Plastic walls delimiting the enclosures were buried in the ground (30 cm) to limit predation by snakes and small rodents. Enclosures had a standardized habitat, providing lizards with sites for hiding and thermoregulation (Boudjemadi et al. 1999). The two one-way corridors ended with a pitfall trap, allowing identification of individuals attempting to disperse (Fig. 1b). A transparent plastic sheet hermetically closed each corridor end. The corridor length of 20 m was chosen so that the distance between centres of two opposite patches corresponded to the upper limit of the confidence interval of an adult home range diameter (i.e. 30 m, Massot & Clobert 1995). Individuals moving such a distance have a very small probability of returning to their site of origin (Strijbosch, VanRooy & Voesenec 1983; Bélichon, Clobert & Massot 1996). More justification on corridor structure can be found in Boudjemadi et al. (1999).

Figure 1.

Representation of the experimental design and of dispersal corridors in an experimental unit. In (a), black areas at the end of corridors correspond to dispersal pitfall traps. Other black areas correspond to pitfall traps situated in enclosures. Four units are connected by dispersal, and four are not. The level of connection is crossed with habitat type (‘rich’ or ‘poor’ habitats). The two habitats are separated by 250 m of woodland, so this protocol is repeated one more time. ‘C’ indicates connected units and ‘UC’ unconnected units. Candidates for dispersal are caught when falling into the pitfall traps at the end of corridors (b). In connected units they are introduced into the population they are going to, whatever the direction considered. In unconnected units, they are returned to the population from when they came.

Experimental treatment

In 8 of the 16 patches, lizards attempting to disperse were allowed to enter the patch they tried to reach. These patches were called ‘connected patches’. In the remaining ones, individuals attempting to disperse were returned to their patch of origin. These patches were called ‘unconnected patches’. A validation experiment showed that the artificial return of candidates for dispersal in unconnected patches was not different from the behaviour of an individual facing a ‘natural’ barrier (Boudjemadi et al. 1999).

Eight patches were located in a grassland corresponding to high level of resources for the common lizard. The eight remnant patches were located in a wood clearance, 250 m from the grassland, which is a suboptimal habitat for this species (Boudjemadi et al. 1999). We call the grassland the ‘rich’ habitat, and the wood clearance the ‘poor’ habitat. Our aim was not to investigate the general interactions between connection and habitat quality. Because of the lack of replication at habitat level, we limited our objective to testing whether the effect of connection could depend on habitat.

Lizard introduction

During the first week of July 1995, we transplanted 96 gravid adult females, 64 adult males, 80 subadult females and 80 subadult males from natural populations of Mont Lozère (1420 m a.s.l., 44°30′N, 3°45′E) to the station at Foljuif (80 m a.s.l., 48°28′N, 2°67′E). We kept gravid females in the laboratory until parturition to determine family relationships. All other lizards were also maintained in the laboratory until the first female hatched in order to limit heterogeneity in the date of introduction. In the laboratory, gravid females were isolated in individual terraria (15 cm × 20 cm × 15 cm high) while adult males and yearlings were kept in groups (47 cm × 130 cm × 35 cm high, 15 lizards per terrarium). More details on laboratory conditions can be found in Boudjemadi et al. (1999).

All animals were measured, weighed and individually marked by toe-clipping. Six adult females with their litter, four adult males and five subadults of each sex were randomly released in each of the 16 patches in order to achieve a similar age and sex structure to natural populations (Massot et al. 1992).

Data collection

We monitored individuals during all of the active season (August–October in 1995, March–October in 1996, 1997 and 1998). Four traps were distributed within each patch in addition to those located at the end of corridors (Fig. 1a). We collected individuals falling in pitfall traps once a day, and performed hand captures once a month in all patches.

Philopatric individuals were defined as individuals recaptured at least once within the enclosure (Fig. 1a) in a given year, but never in a corridor pitfall trap. ‘Dispersers’ were individuals recaptured at least once in a corridor pitfall trap in a given year. Capture rate was equal to 1 in corridor traps but not within the enclosure. In order to consider the same capture rate for dispersers and philopatric individuals, ‘dispersal rate’ was defined as the number of recaptured dispersers divided by the total number of recaptured individuals. When analysing juvenile dispersal rate, the statistical unit was the clutch, since juveniles of the same clutch could not be considered as independent (Massot et al. 1994).

Population size’ was the total number of individuals captured at least once within a year inside a patch. This was either an underestimate of the real population size at the start of each season or an overestimate of the real population size at the end of the season. In a previous (Boudjemadi et al. 1999) and more recent (Le Galliard, Ferrière & Clobert, in press) analysis, we demonstrate that the capture probability per visit to a patch was over 0·5 for all ages and sexes. As we performed more than 10 visits per year per patch, it is reasonable to assume that the annual capture probability almost equals 1. Furthermore, no patch or treatment effects on capture probability were found, allowing us to compare population sizes across patches and treatments without bias.

An analysis of the first year following introduction showed that the connection directly modified dispersal patterns (Lecomte & Clobert 1996; Boudjemadi et al. 1999). Two dispersal periods occurred in unconnected patches instead of one in connected patches. During the first year following introduction, connection increased juvenile survival rate in the rich habitat and decreased it in the poor habitat.

data analysis

Homogenizing connection effect

We first examined the effect of connection on the difference of population sizes between two patches of the same unit. However, if the difference in population size between two connected patches has a clear meaning, this is not the same for unconnected patches since patches of different units (see Fig. 1a) are closer in term of distance – i.e. more likely to face similar microenvironments – than those belonging to the same unit. We therefore computed population size differences between all potential pairs of unconnected patches. We assumed that population size differences followed a Poisson distribution, and corrected for overdispersion by the dscale option (Genmod procedure, SAS 1990). We then examined year, connection, habitat type and patches effects. Because the way we computed differences among unconnected patches could induce non-independence, we performed a bootstrap examining all the potential two by two combinations of unconnected patches and realized the same analysis as above for each combination.

We also conducted three complementary approaches to assess the robustness of the results from the previous analysis. We compared intra-annual, interpatch variance of population size using Bartlett's tests (after transforming for normality). We also compared interannual, intrapatch variance of population size in connected and unconnected situations. Theory predicts that isolated populations have divergent trajectories so that the variance of population size becomes higher than the mean population size. This should lead to a log-normal distribution of population sizes (Caswell 2001). In connected patches, as demographic stochasticity should be compensated by dispersal, their population size should follow a Poisson distribution. We first assumed a Poisson distribution in both connected and unconnected patches and performed the global model Population size = Habitat_Type * Year (Genmod procedure, SAS 1990). We then calculated the overdispersion indicator (c = Deviance/d.f., McCullagh & Nelder 1989) which is a measure of the data fit to the assumed distribution. We hypothesized that c would be far from 1 (good fit) in unconnected patches compared to connected ones. We finally compared c-values between connected and unconnected situations using a chi-square test.

Linear and generalized linear model analyses

anovas were performed using the GLM procedure (SAS 1990). Underlying assumptions of the linear model were systematically verified. Logistic regressions were performed using the Genmod procedure (SAS 1990), assuming a logistic distribution with a binomial link function, and were corrected for overdispersion when necessary (statistics noted inline image). For each variable, type of analysis, statistical unit (patch, clutch or individual) and selected model are given. We systematically considered interaction of the first order in the general model, except when sample sizes were too small. The notation Y = A * B * C refers to the model Y = A + B + C + A * B + A * C + B * C containing all the main effects of A, B and C and their first order interactions. To account for patch differences in their population dynamics through time, we nested the patch effect within treatment and habitat. This nested effect, patch(connection * habitat), was then used as the error term in anova. Year effect was also systematically accounted for.

population modelling

Stochastic two-sex models of structured populations (Legendre et al. 1999; Caswell 2001) were used to predict the consequences of connectivity on populations’ viability and to investigate extinction probabilities. Simulations were performed with the ULM computer program (Legendre & Clobert 1995; Ferrière et al. 1996).

The life cycle graph described four age classes – yearlings, 2 years old, 3 years old and 4 and more years old – in a prebreeding census (Caswell 2001). Demographic parameters were sex-dependent. Survival rates, respectively, varied with age and sex according to previous estimates (see Massot et al. 1992; Ronce, Clobert & Massot 1998). Males and females started breeding at age 1 and proportion of pregnant females varied with age. Dispersal rates varied with age as measured in the field. All models included demographic stochasticity on all parameters: binomial distribution for rates (survival, sex ratio, mating and pregnancy events, potential breeders) and Poisson distribution for number of matings and fecundity. The number of mating was the minimum of the number of potentially breeding females, and the number of females that potentially breeding males could mate with.

According to data collected earlier, density influences juvenile dispersal positively and juvenile survival and fecundity negatively (Massot et al. 1992; Lecomte et al. 1994). We simulated the effect of density on juvenile dispersal using a Beverton–Holt function (all juveniles are not affected in the same way by density, i.e. contest competition) whereas adult survival and fecundity were simulated through a Ricker function (all individuals are affected in the same way by density, i.e. scramble competition). The coefficient of the Ricker function was determined from natural systems (Massot et al. 1992; J. Lecomte et al. unpublished data).

Environmental stochasticity was considered at two levels: (i) a random variation of mean survival and fecundity (survival rates were drawn from a beta distribution and fecundity rates from a Gaussian distribution computed from observed temporal variations –Bauwens, Heulin & Pilorge 1986– and truncated to 0); and (ii) catastrophic events on survival following different impact frequencies and strength. Catastrophes occurred according to a binomial distribution with various level of probabilities. For each combination of parameters, 1000 population trajectories were drawn over 4 years – i.e. experiment duration – via Monte Carlo simulations. Extinction probabilities were computed as the number of extinct trajectories over the total number of simulated trajectories.


We first tested whether connection homogenizes population sizes through a density-dependent dispersal rate. PVA models were then developed to identify if, besides connection and density dependence, other sources of stochasticity (environmental and/or catastrophic), should be involved to explain the observed patterns of population size fluctuations.

Habitat explained much of the important population size variance (r2 = 24%), while year and treatment explained only a further 3% (Table 1). Mean population sizes were 6·31 (SD = 4·22) for poor and unconnected patches; 2·65 (SD = 2·87) for poor and connected ones; 24·56 (SD = 27·12) for rich and unconnected patches and 18·31 (SD = 12·18) for rich and connected ones. While some populations became extinct (four in the poor vs. three in the rich; four in the connected vs. three in unconnected patches), others increased strongly since 1995 (Table 1). Although connection did not significantly affect mean population size, it might have affected its variance.

Table 1.  Total population size before reproduction in each experimental population. For connected units, the two populations of the same unit have the same number. Unconnected patches are designated by a letter. ‘P’ refers to poor habitats and ‘R’ to rich habitats. ‘C’ refers to connected units, and ‘UC’ to unconnected units
1aP–C20  4  0  0  0
1bP–C20  3  0  0  0
2aP–C20  8  6  1  0
2bP–C20  7  3  5  5
4aR–C2024  1  1  0
AP–UC201210  0  0
BP–UC2012  711  3
CP–UC20  8  2  4  2
DP–UC20  8  711  4
GR–UC2015  1  0  0
HR–UC2013  4  0  0

connection and homogenisation of population size

To test whether connection homogenizes population size, we used four complementary approaches. We compared in connected and unconnected situations: (1) population size differences, (2) intra-annual, interpatch variance of population size, (3) intrapatch, interannual variance of population size, and (4) distribution of population size in connected and unconnected patches.

Population size differences among connected patches were of smaller magnitude than those among connected patches. This was close to significance as a main effect (n = 37, patch scale, log-linear analysis, inline image = 3·74, P = 0·06), but was clearly influenced by the year (Connection * Year effect, inline image = 8·36, P = 0·04) such that the overall effect of connection was clearly significant (inline image = 12·10, P = 0·017). Habitat type did not influence this result (Connection * Habitat effect inline image = 1·53, P > 0·1). Bootstrapping confirmed the above results since more than 75% of the combinations (much more than expected by chance alone) yielded a significant effect of connection.

In comparing intra-annual, interpatch population size, we separated the two types of habitat, because population sizes were very low in the poor habitat compared with the rich one (Table 1). In the rich habitat, population-size variance increased significantly in unconnected patches, but not in connected patches. In 1999, however, this pattern disappeared (Bartlett's test, Table 2, Fig. 2a). The same overall pattern was observed in the poor habitat, but the observed tendencies were never significant (Table 1, Fig. 2b). These results suggested that connection spatially decreased population size heterogeneity.

Table 2.  Comparison of variance of total population size before reproduction in connected and unconnected units by type of habitat and by year using Bartlett's tests
YearRich habitatPoor habitat
χ 2P χ 2P
1995Same number of individuals (20) in each population
Figure 2.

Comparison of intra-annual, interpatch variations of population size in connected (white bars) and unconnected units (grey bars): (a) rich habitat; (b) poor habitat.

Because of its particular distribution, we could not perform any anova on interannual, intrapatch variance of population size. However, in the rich habitat, the variance seemed to reach larger values in unconnected patches, especially in patches avoiding extinction (Fig. 3). In the poor habitat, population sizes were low, extinctions were frequent, and consequently their variances were quite homogeneous.

Figure 3.

Comparison of interannual, intrapatch variance of population size in connected (white bars) and unconnected (grey bars) units.

Population-size distribution was expected to follow Poisson and log-normal distributions in connected patches and in unconnected patches, respectively (see Methods). The assumption of a Poisson distribution gave a significantly better fit in unconnected patches than in connected ones (patch scale, log-linear analysis, Poisson distribution assumed, Population size = Habitat * Year, d.f. = 30; connected patches: Deviance = 150, Deviance/d.f. = 5·0; unconnected patches: Deviance = 453, Deviance/d.f. = 15·1, test of deviance comparison: inline image= 152·25, P < 0·0001). Again, this result was consistent with an homogenizing effect of connection on population size.

connection, population size and dispersal

Subadult dispersal rate did not depend on sex (logistic analysis, sex effect: inline image = 0·5307, P = 0·4663). Grouping males and females, subadults dispersed more in the poor habitat than in the rich one and the effect of connection on subadult dispersal was year dependent (patch scale, logistic analysis, 378 recaptured subadults, 67 dispersers, d.f. = 37; Connection effect: inline image = 2·8181, P = 0·0932; Habitat effect: inline image = 6·9735, P = 0·0083; Year effect: inline image = 26·7842, P = 0·0001, Connection * Year effect: inline image = 12·1483, P = 0·0069).

Adult males and females attempted to disperse more in the poor habitat than in the rich one (patch scale, logistic analysis, 430 recaptured individuals, d.f. = 49 and 40; Habitat effect: male –inline image = 3·7699, P= 0·0522; female –inline image = 4·7783, P= 0·0288). Furthermore, adult females attempted to disperse more in unconnected patches than in connected ones (Connection effect: inline image = 4·5335, P = 0·0332).

The second period of juvenile dispersal observed in unconnected patches in 1995 (Boudjemadi et al. 1999) was not observed in the following years. Juvenile dispersal rate depended on connection and population size (Table 3, Connection * Population size effect: inline image = 4·7695, P = 0·0290). It significantly increased with population size in connected patches but not in unconnected patches (connected patches, patch scale, logistic analysis, 549 juveniles, 73 dispersers, d.f. = 16; Habitat effect: inline image = 10·1995, P = 0·0141; Year effect: inline image = 11·0641, P = 0·0114; Population-size effect: inline image = 5·0219, P = 0·0250; unconnected patches, as above, 654 juveniles, 97 dispersers, d.f. = 21; Habitat effect: inline image = 5·5993, P = 0·0180; Year effect: inline image = 9·1339, P = 0·0276). This suggested that juvenile dispersal could regulate population size in connected patches.

Table 3.  Influence of connection (C), habitat type (H) and population size (PS) considering year (Y, 1995–98) on juvenile dispersal rate (DR). ‘UC’ corresponds to unconnected units, ‘P’ to the poor habitat
Variabled.f.inline imagePSense of the effect
  1. Patch scale, logistic analysis, 48 patches, 1203 juveniles, 170 dispersers, d.f. = 40, Deviance/d.f. = 2·27. Global model: DR = C * H * Y * PS with first and second order interactions, selected model: DR = C + H + Y + (PS) + PS * C.

Connection1  4·32070·0377UC > 0
Habitat1  6·46320·0110P > 0
Year310·34360·015997 > 96 > 95 > 98
Population size1  0·50600·4769>0
Population size * Connection1  4·76950·0290C > 0, UC ∼ 0

population viability analysis

The values of population variance and extinction probabilities obtained by population viability analysis showed that density dependence and demographic stochasticity were not sufficient to simulate the observed values in rich habitat (Fig. 4). In some cases, adding catastrophes allowed us to match observed values of extinction rate, mean average and variance of population sizes. In these cases, population-size variances were found to be greater for unconnected populations than for connected ones.

Figure 4.

Values and standard deviation (error bars) of mean population size (a), variance of population size (b) and extinction probability (c) of non-extinct populations obtained by population viability analyses (open symbols) or experiment (filled symbols) in connected (triangle) or unconnected (square) situations. For model descriptions, see Methods. DD are models with density dependence (Ricker's parameter c varies from 0·005 to 0·5); DD + E are models with environmental stochasticity on survival then density dependence on parameters (to model environmental stochasticity, standard deviation on survival and fecundity varies from 0·1 to 0·15 and from 0·5 to 0·75, respectively); DD + E + CAT are models with environmental stochasticity, density dependence and catastrophic events (to model catastrophic events, survival rates were multiplied by a reduction factor varying from 0·1 to 0·2; catastrophes frequencies varied from 0·09 to 0·19 which are extreme values observed in natural populations). All experiment data are those observed in the rich habitat during the last year of the experiment.


Connection seemed to decrease population variance through a density-dependent dispersal rate. As a result, connection could enhance synchrony in the long term. The underlying process appeared to be partly year and partly habitat dependent, implying that connection interacts with the overall population-dynamics profiles, i.e. excluding the direct effect of dispersal. In unconnected patches, populations became rapidly extinct or increased and then sharply declined. As predicted, juvenile dispersal was density dependent in connected patches but surprisingly not in unconnected patches.

Observed extinction rate, size and variance of population sizes as well as differences between connected and unconnected populations could be mimicked only by models including both environmental stochasticity and catastrophic events.

relevance of the experimental design

Most of the characteristics of the experimental design and particularly density at introduction and distances between patches were described and discussed in Boudjemadi et al. (1999). The results of this study strongly argue in favour of the relevance of our design. First, the mean size of non-extinct populations in the rich habitat was similar to the number of individuals introduced in 1995, indicating that the number of introduced individuals corresponded to the carrying capacity of the rich habitat. In the poor habitat, however, all populations declined, and carrying capacity was probably less than the 20 individuals initially introduced into each patch. This might have influenced the results for the poor habitat. Second, the distance between the centres of two opposite patches was 30 m. This could have been too low to mimic dispersal. Nevertheless, observed juvenile dispersal rates (about 10%) were more similar to interhabitat dispersal rates measured in nature (about 10%; Clobert et al. 1994) than to intrahabitat ones (30–60%; Clobert et al. 1994).

connection, homogenisation of population sizes and synchrony of dynamics

As predicted by theory (Holt & McPeek 1996; Haydon & Steen 1997), differences in population size between patches seem to be lower in connected situations than in the unconnected one. Dispersal is presented as the mechanism responsible for synchronized dynamics by buffering stochastic processes (Gilpin 1996; Stacey et al. 1997). The density-dependent juvenile dispersal rate that we observed in connected patches may correspond to such a mechanism. This is consistent with the known effect of juvenile dispersal on density regulation in the common lizard (Massot et al. 1992).

However, we did not expect juvenile dispersal to be density-independent in the unconnected unit. This result strongly suggests that the absence of connection destabilizes population functioning by modifying dispersal determinism. In the first year of the experiment, such a modification of the dispersal determinism caused a second period of juvenile dispersal involving individuals whose phenotype was similar to that of philopatric individuals (Lecomte & Clobert 1996; Boudjemadi et al. 1999). We hypothesized that the return of frustrated dispersers in unconnected patches induced the departure of individuals that would have been philopatric in connected ones. It might also have altered social interactions such that density is not perceived the same way when a population is connected as when it is not. For example, unconnected populations increased markedly in size only in the rich habitat before a sudden decline. These observations are consistent with an over-compensatory response to density, possibly resulting from the nature of the dispersers. Interestingly, such a destabilizing pattern is expected when scramble competition occurs (Begon, Harper & Townsend 1996). This definitively raises the question of the influence of connection on local competitive processes.

We have also observed more dispersal attempts in the poor than in the rich habitat. Dispersal intensity appeared to be habitat-dependent (see also Massot et al. 2002) and therefore might change the effect of connection on the overall local dynamics. Both the nature of dispersers (Sorci, Massot & Clobert 1994; Léna et al. 1997) and dispersal determinism (Massot & Clobert 1995; Ronce et al. 1998) might be involved in determining the habitat-dependent connection effect.

connection, synchrony and extinction

Stochastic environmental variations at the global scale (Moran effect, Moran 1953) are thought to induce synchronization at the metapopulation level, whereas dispersal is expected to synchronize dynamics at a more local scale (Haydon & Steen 1997; Heino et al. 1997; Paradis et al. 1999; but see Hudson & Cattadori 1999). However, one of the basic assumptions required to ensure metapopulation viability is the asynchrony of local population extinction (Levins 1969). All sources of asynchrony, even including chaos (Allen, Shaffer & Rosko 1993), can potentially decrease the probability of global extinction (Heino et al. 1997). According to Heino et al. (1997), local synchronization induced by dispersal is also sufficient to increase global extinction probability. Although we cannot directly link extinction to connection, it is likely that the destabilization induced by the absence of a connection had a higher impact on extinction probability than the population-size autocorrelation induced by connection. The spatial and temporal scale at which dispersal homogenizes dynamics, and thus its influence on the probability of global extinction, may depend on the interaction between the dispersal power of a species in a given landscape and the level of local stochasticity. For a given distance of dispersal, the spatial scale of homogenisation may decrease if the impact of local stochasticity increases. In such conditions, dispersal may decrease local extinction risk without increasing global extinction probability. Moreover, if a low local stochasticity leads to an accumulation of homogenizing effects over time (Heino et al. 1997), the reverse should be observed when local stochasticity is high. Thus the high impact of local stochasticity should decrease both the temporal and spatial scale of dispersal-induced synchronization. The same conclusion could be drawn in the case of a non-positively spatially autocorrelated environmental stochasticity. From a conservation point of view, the effect of restoring connectivity when patches have recently been isolated might be more complex than previously thought.

population viability analysis

Using demographic parameters estimated on other data, our modelling approach provided an effect of connection on population size variation and extinction consistent with our experimental results. However, to find at the same time the mean population size, its variance and the relatively high probability of extinction, it was necessary to add both environmental stochasticity and catastrophic events to density dependence and demographic stochasticity. So, all sources of stochasticity had to be combined to explain observed trajectories. In the face of an important local environmental stochasticity, connection had an even stronger impact on damping population fluctuations and decreasing extinction probability. Catastrophic events could be due to predation effects that were not fully under control. The modelling exercise strengthens the importance of some of our results. It emphasizes the need to study the interplay between local sources of stochasticity in balancing the positive and negative effect of a dispersal-synchronized population dynamics.


Our experimental results suggest that connection stabilized population fluctuations by homogenizing population size. A density-dependent juvenile dispersal rate was likely to be involved in such a regulation. Preventing dispersal in unconnected systems also modified the determinism of dispersal. Moreover, the absence of interpopulation exchanges leads to extreme variations in population size. The rapid increase of some population sizes in unconnected systems was followed by a sudden decline in the last year of the experiment, suggesting that a delayed density-dependent regulation occurs. Population viability analysis underlined the role of combined action of all stochasticity sources to simulate the observed results.

In this experiment, it seems that dispersal, by damping local stochasticity, had a more important impact, via homogenisation of population sizes, than its potential synchronizing effect on population dynamics. This led us to think that connection can indeed decrease metapopulation extinction probability. However, we have to be cautious at this stage since only two replicates of each situation were studied. Indeed, as confirmed by the rarity of such studies, it is very difficult to perform experiments requiring several years and a lot of replicates (Burkey 1997; Debinski & Holt 2000), especially on vertebrates. Because of the wide range of species-specific responses (Debinski & Holt 2000; Schmiegelow & Mönkkönen 2002), it is only by the repetition of such experimental approaches for several groups of organisms that it will be possible to assess the generality of these results.


It is a great pleasure to thank all the people who made this study possible. The Ecole Normale Supérieure, the CNRS and the Ministry of Environment financially supported this research. Members of the biological station of Foljuif (belonging to the Ecole Normale Supérieure), the Cévennes National Park and the Office National des Forêts provided most facilities to do our experimental and field work. We especially want to thank Amélie Lefèbvre, Céline Téplitski, Yann Gautier, Sandra Lallemand and other students for having helped us to collect data.