modelling mortality and dispersal: consequences of parameter generalisation on metapopulation...
TRANSCRIPT
Modelling mortality and dispersal: consequences of parameter
generalisation on metapopulation dynamics
Gwenaelle Mennechez, Sandrine Petit, Nicolas Schtickzelle and Michel Baguette
Mennechez, G., Petit, S., Schtickzelle, N. and Baguette, M. 2004. Modelling mortalityand dispersal: consequences of parameter generalisation on metapopulation dynamics.�/ Oikos 106: 243�/252.
Modelling dispersal is a fundamental step in the design of population viability analyses.Here, we address the question of the generalisation of population viability analysismodels across landscapes by comparing dispersal between two metapopulations of thebog fritillary butterfly (Proclossiana eunomia ) living in similar highly fragmentedlandscapes (B/1% of suitable habitat in 9 km2). Differences in dispersal patterns wereinvestigated using the virtual migration (VM) model, which was parameterised withcapture�/mark�/recapture data collected during several years in both landscapes. TheVM model allows the estimation of 6 parameters describing dispersal and mortality aswell as the simulation of dispersal in the landscapes. The model revealed largedifferences in the VM parameter estimates between the two landscapes andconsequently, simulations indicated differential rates of emigration and dispersalmortality. Furthermore, results from crossed-simulations i.e. simulations performed inone of the landscape but using parameter estimates from the other landscape emphasizethat dispersal parameters are very specific to each metapopulation and to theirlandscape. Hence, we urge conservation biologists to be cautious with such parametergeneralisations, even for the same species in comparable landscapes.
G. Mennechez, N. Schtickzelle and M. Baguette, Biodiversity Research Centre, Univ.Catholique de Louvain, Croix du Sud 4, BE-1348 Louvain-la-Neuve, Belgium([email protected]). �/ S. Petit, Centre For Ecology and Hydrology, MerlewoodResearch Station, Grange-over-Sands, Cumbria, UK, LA11 6JU.
All over the world, species face loss and fragmentation
of their natural habitats by human activities (Wilcox and
Murphy 1985, Saunders et al. 1991). In western Europe,
specialist species are now restricted to small remnants of
suitable habitat surrounded by a more or less hostile
matrix. In well known taxonomic groups like butterflies,
most specialist species have been shown to be dramati-
cally declining (Warren 1993, Van Swaay and Warren
1998, Maes and Van Dyck 2001). Nevertheless, such
species with narrow ecological requirements may persist
in fragmented landscapes as metapopulations, i.e. as-
semblages of local populations connected by dispersal
and potentially submitted to extinctions and recolonisa-
tions (Gilpin and Hanski 1991, Hanski and Gilpin 1997,
Hanski 1999).
The metapopulation paradigm is an operational
framework to predict the long-term persistence of
species living in highly fragmented landscapes (Hanski
et al. 1996, Hanski and Gilpin 1997, Baguette and
Stevens 2004). Different kinds of population viability
analysis (PVA) models (notably patch occupancy mod-
els, structured population models), currently developed
to assess the fate of endangered species in real conditions
or under different potential management plans, are
derived from metapopulation biology (Akcakaya and
Sjogren-Gulve 2000). The parameterisation of such
Accepted 2 December 2003
Copyright # OIKOS 2004ISSN 0030-1299
OIKOS 106: 243�/252, 2004
OIKOS 106:2 (2004) 243
models, and especially of the structured populations
models, is highly demanding in accurate estimates,
therefore requiring large and comprehensive empirical
data-sets on the focal species. For example, the analysis
of the viability of the bog fritillary Proclossiana eunomia
by Schtickzelle and Baguette (2004) is based on a long
term study in which a patchy population was monitored
during 11 consecutive years. Of course, it is unimagin-
able to extend such long term studies to masses of
endangered species and to all landscapes contexts. A
crucial question in conservation biology is therefore to
what extent models constructed for a specific species in a
particular landscape can be extrapolated to similar
landscapes or to related species (Wahlberg et al. 1996,
Rodriguez and Andren 1999, White 2000). Without
potential for a generalisation of models from one system
to another, we have to admit that their usefulness may be
limited in the more general perspectives of biodiversity
conservation. In this paper, we address this question of
the generalisation of PVA across landscapes by compar-
ing dispersal rates �/ a key parameter in the metapopula-
tion functioning �/ between two highly fragmented
landscapes.
Few attempts have been made to investigate differ-
ences in dispersal rates between metapopulation inhabit-
ing different landscapes (Mennechez et al. 2003). Surely,
such comparisons are not so trivial and require efficient
tools (Wahlberg et al. 2002). The virtual migration
model (VM) developed by Hanski et al. (2000) is a
multi-population model designed to estimate dispersal
and survival of individuals in metapopulations. This
model is well adapted to compare the functioning of
metapopulations since it allows the comparison of
movements (1) between species within the same network
of habitat patches or (2) within the same species between
different networks of habitat patches. The VM model
has already been used successfully to quantify the
dispersal rates in the bog fritillary butterfly as well as
their variation between sexes and across years (Petit et
al. 2001) and to compare metapopulation structure and
movements in five species of checkerspot butterflies
(Wahlberg et al. 2002).
Here we investigate differences in dispersal patterns of
P. eunomia between two highly fragmented landscapes of
the uplands of the Belgian Ardenne. We first parame-
terised the VM model using data collected in the two
landscapes during several years in order to estimate
dispersal and mortality. In a second step, dispersal was
simulated in both landscapes to obtain a general picture
of inter-patch movements, given their respective para-
meter estimates. Finally, using the parameter estimates
obtained in one landscape, we simulated dispersal and
survival in the other landscape, in order to assess
whether parameter generalisation will induce changes
in the prediction of dispersal pattern.
Material and methods
The species
In Belgium, the bog fritillary butterfly (Proclossiana
eunomia ESPER; Lepidoptera, Nymphalidae) is a spe-
cialist species living in a very restricted habitat (bogs and
unfertilised wet meadows) where its only host plant
Polygonum bistorta L. grows. Because of changes in
agricultural practices in the middle of the 20th century,
the spatial extent of this transitional habitat, which was
maintained by traditional agro-pastoral practices, has
been dramatically reduced (Baguette et al. 2003). Nowa-
days, suitable patches of habitat remain scattered along
rivers where they form networks supporting P. eunomia
metapopulations at the landscape scale (Neve et al.
1996). Besides the loss and fragmentation of wet hay
meadows, there is also the problem of their degradation
induced by the abandon of extensive management
practices. Bistort patches that represent an early transi-
tional stage are either unmanaged and progressively
invaded by nitrophilous vegetation and scrubs or over-
grazed or overcut. Previous studies have shown that
adult butterflies fly in one generation from late May to
early July with males emerging a few days before females
(protandry, Baguette and Neve 1994). Mating system is
polygynous and male mate-locating behaviour is patrol-
ling: males continuously fly looking for unmated females
(Baguette et al. 1996).
The landscapes
Our study of survival and dispersal parameters in
metapopulations of the bog fritillary was conducted on
the ‘‘Plateau des Tailles’’ upland (20�/20 km; 50814?N,
5847?E), southern Belgium. The selected landscapes,
Lierneux and Wibrin, have both a total area of 9 km2.
They were selected because they were (1) close to each
other (ca 10 km) and 2) located at the same altitude (ca
350�/400 m). The composition of these two landscapes
was typical of the Ardenne uplands i.e. a mosaic of
grasslands, Norway spruce (Picea abies Karst) planta-
tions and deciduous forests, although forests were more
widespread in Wibrin (68%) than in Lierneux (50%). The
proportion of suitable habitat for P. eunomia �/ wet
meadows with Polygonum bistorta �/ was similar in both
landscapes, i.e. less than 1% of the total area.
Habitat patch networks within the landscapes differed
from each other in the number of suitable patches as well
as in their spatial configuration (Fig. 1). The mean patch
size was similar in both landscapes (0.429/0.43 ha and
0.319/0.25 ha in Lierneux and Wibrin respectively) but
the mean proximity index �/ a measure of the degree of
habitat fragmentation which decreases when fragmenta-
tion increases (McGarigal and Marks 1994) �/ indicated
244 OIKOS 106:2 (2004)
that habitat fragmentation was slightly higher in Wibrin
(MPI�/9.9 vs MPI�/20.7 in Lierneux).
The capture�/mark�/recapture experiment
Capture�/mark�/recapture (CMR) experiments were car-
ried out in 1993, 1994 and 1997 in Lierneux and in 1999
and 2001 in Wibrin, following the same protocol. During
the flight season, patches were visited daily, weather
permitting. Within patches, encountered butterflies were
netted and marked with an individual number on the
underside of the left hind wing with a thin-point
permanent pen. For each (re)capture, we recorded the
butterfly mark, the date, the patch location and the sex
of the individual. Butterflies were immediately released
at the location of their capture.
The capture histories obtained from the CMR studies
were used as an input to the VM model and to calculate
population size. As inter-patch movements were very few
for females in Wibrin �/ due to lower capture probabil-
ities of females which is related to the mating behaviour
(Schtickzelle et al. 2002) �/ only males were taken into
account in the present analysis. Although CMR studies
were performed in different years in Lierneux and
Wibrin, results can really be compared between the
two landscapes. Indeed, there were no differences in
climatic conditions between the two landscapes (one-way
ANOVAs: for maximum temperature, P�/0.10 and for
insolation, P�/0.36).
Virtual migration model and data analysis
The VM model was developed by Hanski et al. (2000) to
describe survival and dispersal in metapopulations living
in networks of at least 10 patches. It was constructed
under some simple biological assumptions described in
previous papers (Hanski et al. 2000, Petit et al. 2001,
Wahlberg et al. 2002). Consequently, we only give here a
general outline of the biological model; more details �/
about the statistical model and the algorithm used for
the parameter estimation �/ are to be found in Hanski et
al. (2000).
The model is based on discretised histories of marked
butterflies with events occurring in the following order in
the course of a day time. First, consider a butterfly
within a habitat patch j. Its survival probability until the
next day or until it emigrates �/ whichever occurs first �/
is Fp. If the butterfly survives it may stay in the patch or
emigrate with the probability oj calculated as a power
function scaling the patch area Aj
oj�hA�zemj (1)
where h and zem are positive parameters that describe
the propensity to emigrate from a habitat patch of unit
size (�/1 ha in the present study) and the scaling of the
emigration to patch area respectively.
Next, if the butterfly emigrates, it may die or survive
to dispersal and reach another patch. The probability of
a individual to survive dispersal from patch j, Fmj, is
assumed to increase as a sigmoid function of the
Fig. 1. Patch networkconfiguration within thelandscapes of Lierneux (A) andWibrin (B). Patch networks arecomposed of 12 and 15 habitatpatches ranging from 0.04 to1.2 ha (totalling ca 5.04 ha ofsuitable habitat) and from 0.08to 1.05 ha (totalling ca 4.69 haof suitable habitat) in Lierneuxand Wibrin respectively.Suitable habitat patches(indicated in black) are locatedalong rivers.
OIKOS 106:2 (2004) 245
connectivity of patch j, Sj, as measured by the following
equation:
Fmj�S2
j
l� S2j
(2)
where l is a positive parameter �/ that corresponds to the
square of the population connectivity for which the
probability of surviving dispersal is equal to 0.5 �/ giving
an indication on the mortality during dispersal and
where Sj is calculated by:
Sj�X
k"j
e�adjk Azim
k (3)
In which djk is the Euclidian distance between patches j
and k, and a and zim are two other model parameters
(�/0). Parameter a weights the effect of the distance on
dispersal and zim is a constant scaling the dependence of
the immigration on the patch area.
Finally, the model assumes that individuals that
survive dispersal will immigrate into the potential target
patches depending on their relative contribution to the
connectivity of patch j. Consequently, the probability of
an individual leaving patch j to end up into patch k is:
cj;k�e�adjk A
zim
k
lSj
� Sj
(4)
Of course, these events recur in the subsequent time
intervals until the individual dies. Thus, in the following
time interval (e.g. day 2. . .) the same individual may or
not survive in the patch, (re-)emigrate and eventually
reach another patch. For each butterfly, it is possible to
compute the probabilities of the observed daily events
and the likelihood of the data set is calculated as the
product of the likelihood of individual capture histories.
To conclude, VM model has six parameters which
are estimated via maximum likelihood technique: mp
(�/1�/Fp), the daily patch mortality, h, the daily
propensity to emigrate from a 1-ha size patch, a, the
dependence of dispersal to distance, l, giving an
indication on mortality during dispersal, zem and zim
scaling respectively the dependence of emigration and
the immigration on patch area.
In order to obtain a general image of dispersal and
reduce the potential random effects due to limited
number of movements between patches in each year
(especially in Wibrin), to be able to compare mean
parameters between the two landscapes VM model was
parameterised using pooled yearly CMR data-sets for
Lierneux and for Wibrin separately: capture histories
were put end to end (Schtickzelle 2003, Schtickzelle and
Baguette 2003). This was possible as experimental
conditions are similar for all years: (1) sampling was
done only during favourable climatic conditions; (2) the
catch effort is kept as constant as possible and (3)
weather conditions (maximum temperature and insola-
tion) did not significantly vary between years for each
network (one-way ANOVA and Bonferroni mean com-
parisons: on the eight comparisons of weather variables,
only insolation 1993 and insolation 1994 are different at
the 0.05 level).
Since confidence intervals of the parameters are not
symmetrical, when comparing mean parameters, we
assumed as others (Wahlberg et al. 2002, Schtickzelle
and Baguette 2003) significant difference only if their
confidence limits do not overlap.
Simulations of inter-patch movements within
networks
Using the respective mean parameter estimates and
mean local population sizes (see below for the estimation
of population size), we simulated (with VM simulation
model) dispersal and mortality in Lierneux (Lierneux/
Lierneux simulations) and in Wibrin (Wibrin/Wibrin
simulations) to evaluate the potential difference between
the two landscapes. To allow comparisons between
predictions, CMR conditions were considered to be
similar in both landscapes; (1) all butterflies were
assumed to emerge the first day; (2) a 30-day period
was simulated with CMR experiment performed every
day. For each landscape, simulations were performed 10
times.
In a second step, we generated crossed-simulations i.e.
simulations (1) in Lierneux using parameter estimates
based on CMR data from Wibrin (Lierneux/Wibrin
simulations) and (2) in Wibrin using parameter estimates
based on CMR data from Lierneux (Wibrin/Lierneux
simulations) to evaluate the consequences of generalising
parameters from a metapopulation to another.
Results of VM simulations are notably given in terms
of (1) number of butterfly-days spent in and outside the
natal patch and (2) number of successful and unsuccess-
ful (i.e. leading to the death of the butterfly) dispersal
events.
Estimation of population size
Yearly population sizes were estimated using Con-
strained Linear Models methodology (CLM, Lebreton
et al. 1992, Schwarz and Seber 1999). CLM methodol-
ogy allows to fit regression models with various factors
on the 3 demographic parameters (survival 8, catch-
ability P and recruitment B). The best model is selected
by the way of the Akaike’s Information Criterion
modified for small samples (AICc: Burnham and
Anderson 1998) which makes a trade-off between the
fit of the model and its parsimony (in terms of number of
parameters). The best model gives the factors signifi-
cantly acting on demographic parameters as well as the
246 OIKOS 106:2 (2004)
estimates for these parameters and the total population
size.
Total population sizes for the core area of Lierneux
have already been estimated (Schtickzelle et al. 2002).
For Wibrin, data-sets did not contain enough recaptures
to allow model selection to be achieved, but good
parameter estimates could nevertheless be obtained by
using a biologically adequate model. We used the model
8 s�tlinps�1Bs�[tlin�tlin2] �/ where the daily survival de-
creases linearly through the flight period with a constant
sex difference, the catchability varies between days with a
constant sex difference and the recruitment rate follows a
parabola with males emerging, reaching their abundance
peak and dying before female (protandry) �/ as it has
been found to be the best description for the bog
fritillary (Schtickzelle et al. 2002) and a closely related
species with a similar ecology, the cranberry fritillary
Boloria aquilonaris (Baguette and Schtickzelle 2003).
For some patches, CMR data sets were insufficient to
obtain accurate estimates with the selected model. We
therefore estimated total population size by multiplying
the number of captured butterflies in the patch by a
conversion coefficient (Hanski et al. 1994). This coeffi-
cient was inferred from patches for which local popula-
tion size was calculated using CLM; it equals the
estimated number of individuals divided by the number
of individuals marked.
Results
CMR study and estimation of population sizes
The pooled data-set total to 632 marked individuals and
1061 recaptures in Lierneux and 172 marked individuals
and 446 recaptures in Wibrin. On average, the number of
recaptures per individual and the capture probability
estimated in the VM model were lower in Lierneux
(1.68 times; 0.479/0.04) than in Wibrin (2.59 times;
0.719/0.01); this does not prevent comparing the two
landscapes as VM parameter estimates are corrected for
capture probability (Hanski et al 2000).
Estimates of the total male metapopulation size varied
substantially between years in both landscapes (Fig. 2).
Despite the fact that the total area of suitable habitat
was similar in both landscapes (Fig. 1), butterfly density
was on average 2.5 times higher in Lierneux with 55.8
butterflies per ha of suitable habitat versus 21.2 butter-
flies per ha in Wibrin.
The CMR experiment revealed that butterflies moved
frequently between patches: the fraction of the number
of dispersal events/total number of recaptures �/ an
indicator of dispersal rate �/ was on average 0.20 (214/
1061) in Lierneux and 0.11 (49/446) in Wibrin. The
largest dispersal distance recorded during the study was
2.37 km (1993) in Lierneux and 5.13 km (2001) in
Wibrin.
VM parameters estimation and model goodness-of-
fit
To check the adequacy of the model with the data,
goodness-of-fit tests �/ based on comparison between
expected and observed numbers of immigration, emigra-
tion and residency events as described in Hanski et al.
(2000) �/ were performed for each patch. Of the 81
goodness-of-fit tests, only 9 (11%) revealed a significant
lack of fit, which suggested a globally good overall fit.
When a significant difference between the predicted and
observed number of emigrants and immigrants was
detected, the model always underestimated the dispersal
events (Table 1).
When comparing the mean parameter estimates
between the two landscapes (Fig. 3), it appears that
they all differed significantly (no overlap of confidence
intervals between the two sets of parameter estimates).
The daily propensity to emigrate from a patch of unit
size (h) was much higher in Lierneux than in Wibrin.
The effect of patch size on emigration, expressed by the
parameter scaling the emigration (zem), was significant in
both landscapes (i.e. significantly different from 0) but
roughly five times smaller in Lierneux than in Wibrin
(0.194 vs 1.032). As showed in Fig. 4, which illustrates
the probability of emigrating from patches as a function
of their size (Eq. 1), it means that males tended to
emigrate more from smaller patches in Wibrin than in
Fig. 2. Metapopulation size estimates for each year in Lierneuxand Wibrin.
Table 1. VM model goodness-of-fit tests for the numbers ofresidents, emigrants and immigrants in Lierneux and Wibrin.The table gives the patch codes for which there is a significantlack of fit. The difference between the predicted and observednumber of dispersal events is indicated in brackets (observed �/
predicted).
Lierneux Wibrin
Residents 2 (�/11), 4 (�/10) 3 (�/5)Emigrants 4 (�/12) 3 (�/2)Immigrants 3 (�/19), 6 (�/8) 2 (�/2), 7 (�/3)
OIKOS 106:2 (2004) 247
Lierneux; the reverse is true for large patches. The
immigration scaling (zim,) was significantly different
from 0 in both landscapes, which indicates that butter-
flies are more prone to immigrate into big patches than
into small ones. This effect of the receiver patch area on
immigration was much smaller in Lierneux (0.585 vs
2.426). Results also showed that the value of the
parameter describing the effect of distance on isolation
(a) was significantly higher in Lierneux: on average
butterflies tend to move shorter distances in Lierneux.
However, in both landscapes, the average range of daily
dispersal distances was more than several hundred
meters (inferred from a: 240 m in Lierneux and 470 m
in Wibrin). Butterflies in Lierneux have a lower prob-
ability of daily mortality within habitat patches (mp) than
in Wibrin while for the same patch connectivity, they
suffered a higher mortality when dispersing (Fig. 5).
Simulation of survival and dispersalSimulation of dispersal within the two landscapes using
their respective VM parameter estimates
Results of VM simulations are presented in Table 2.
Here, we compare results obtained for Lierneux/Lier-
neux and Wibrin/Wibrin simulations. The model pre-
dicted that the majority of butterfly-days was spent
inside the natal patch in both landscapes. However, the
proportion of butterfly-days spent outside the natal
patch was significantly higher in Lierneux (Mann-
Whitney test; U�/2, PB/0.001). The predicted number
of dispersal events was 7 times higher in Lierneux than in
Wibrin. Considering the mean metapopulation sizes
(estimated 275 in Lierneux and 103 in Wibrin), the
predicted mean number of dispersal events per indivi-
dual was 1.749/0.08 in Lierneux vs 0.669/0.08 in Wibrin.
The difference was significant (Mann-Whitney U-test;
U�/0, PB/0.001). The model also predicted a much
lower proportion of successful dispersal events in
Lierneux i.e. a higher mortality rate during dispersal
(Mann-Whitney U-test; U�/100, PB/0.001).
Crossed-simulations of dispersal within the two
landscapes
First, when running the model with the VM parameters
extrapolated from Wibrin using the patch configuration
etamitse rete
maraP
etamitse rete
maraP
etami tse rete
maraP
Lierneux Wibrin
0
0.2
0.4
0.6
0.8
1
Within-patch mortality, 1-φ
etamitse ret e
maraP
etamitse rete
maraP
etamitse rete
m ar aP
0
1
2
3
4
5
Distance dependence, α
Lierneux Wibrin
0
1
2
3
4
5
Scaling of immigration, ζim
0
0.1
0.2
0.3
0.4
0.5
Daily emigration rate , η
0
1
2
3
4
5
Scaling of emigration, ζem
0
0.2
0.4
0.6
0.8
1
Migration mortality, λ
Fig. 3. Comparison between Lierneux and Wibrin landscapesof mean VM parameter estimates (with 95% confidenceintervals) for P. eunomia males. See text for details on theseparameters.
Fig. 4. Daily probability of emigration as a function of patcharea for the Lierneux and Wibrin landscapes.
0.0
0.2
0.4
0.6
0.8
1.0
0.0 0.5 1.0 1.5
Lierneux
Wibrin
Patch connectivity (Sj)
la
sre
psi
d g
niru
d ytil
atro
m yli
aD
Fig. 5. Daily mortality of individuals emigrating from patchesas a function of patch connectivity, Sj, for the Lierneux and theWibrin landscapes.
248 OIKOS 106:2 (2004)
and the mean metapopulation size of Lierneux, simula-
tions indicated a much higher number of butterfly-days
spent in the system than with parameter estimates from
Lierneux (Lierneux/Lierneux vs Lierneux/Wibrin:
Mann-Whitney U-test; U�/100, PB/0.001, Table 2).
The model predicted also a higher proportion of
butterfly-days spent outside the natal patch (Mann-
Whitney U-test: U�/100, PB/0.001) but a lower number
of dispersal events (Mann-Whitney U-test: U�/86.5,
PB/0.006). In addition, simulations predicted a much
higher proportion of successful dispersal events (Mann-
Whitney U-test: U�/0, PB/0.001).
Secondly, when running the model with the VM
parameters extrapolated from Lierneux using the patch
configuration and the mean metapopulation size of
Wibrin, simulations predicted a lower number of butter-
fly-days spent in the system than with parameter esti-
mates from Wibrin (Wibrin/Wibrin vs Wibrin/Lierneux:
Mann-Whitney U-test; U�/98, PB/0.001, Table 2). The
model also predicted a lower proportion of butterfly-
days spent outside the natal patch (Mann-Whitney
U-test: U�/100, PB/0.001). While the predicted number
of dispersal events was significantly higher, the propor-
tion of successful dispersal events decreased significantly
(Mann-Whitney U-tests: U�/0, PB/0.001 and U�/100,
PB/0.001 respectively).
Discussion
We compared here dispersal and mortality of the bog
fritillary butterfly Proclossiana eunomia between two
fragmented landscapes of Belgian Ardenne using the
VM model. As our main objective was to test the
possibility to generalise dispersal between similar land-
scapes when constructing PVA models, the study land-
scapes were both (1) highly fragmented with similar
proportion of suitable habitat (less than 1% of suitable
habitat; albeit there was a difference in their fragmenta-
tion level, it remained slight), (2) located within the same
upland characterised by a agricultural land/forest matrix
and (3) close to each other to ensure similarity in
environmental and historical factors affecting P. euno-
mia populations. Furthermore, as the studies were
carried out in different years in Lierneux and Wibrin,
we made sure our comparison was valid by checking that
climatic conditions (maximum temperature and insola-
tion), which are known to affected strongly butterfly
movements (Shreeve 1992), were similar between the two
landscapes. It is worth noting that modelling studies
rarely take as many precautions when entering into the
model previously published parameter estimates i.e.
when using generalized parameter estimates (Wahlberg
et al. 1996, Thomas et al. 2002).
Similarities in dispersal pattern between fragmented
vs continuous landscapes
Our results show that all the six VM parameter estimates
are clearly different between Lierneux and Wibrin land-
scapes. However, comparisons with results obtained
under more continuous habitat conditions bring to light
some similarities in dispersal patterns between the two
fragmented landscapes (Mennechez et al. 2003, Schtick-
zelle and Baguette 2003). First, the dispersal propensity,
i.e. the daily propensity to emigrate from 1-ha patch is
much lower in fragmented landscapes. Secondly, the
emigration probability clearly decreases when patch size
increases in fragmented landscapes.
Using the same methodology, i.e. the VM model, we
estimated to 43% (vs less than 20% and 10% in Lierneux
and Wibrin respectively) the dispersal propensity of
P. eunomia in a continuous landscape totalling ca 45.5
ha of suitable habitat in 12 patches (G. Mennechez,
unpubl.). Furthermore, a recent study using ad hoc
Table 2. Results (mean9/SD, given 10 replicate simulations) of VM simulations in Lierneux and Wibrin landscapes using theirrespective mean population size (N) and the mean VM parameter estimates from Lierneux and Wibrin. The VM simulation modelcalculates the number of butterfly-days spent inside and outside the natal patch and the number of successful and unsuccessfuldispersal events.
Parameter values from: Landscape configuration
Lierneux (N�/275) Wibrin (N�/103)
Lierneux Wibrin Wibrin Lierneux
Number of butterfly-days spentin the natal patch 12089/72 13829/60 5419/78 4379/55outside the natal patch 7329/85 12539/78 1839/45 729/24in the system (i.e. total) 19409/88 26359/70 7249/95 5099/62
% butterfly-days spent outside the natal patch 389/3 489/2 259/5 149/4
Number of dispersal eventssuccessful 2109/6 1899/6 429/6 189/4unsuccessful 2699/5 19/5 269/5 1029/1total 4799/8 1909/8 689/8 1209/4
% successful dispersal events 449/5 999/5 629/5 159/3
OIKOS 106:2 (2004) 249
measures confirmed that dispersal in P. eunomia was
lower in the fragmented system (Mennechez et al. 2003):
39% of recaptures involved inter-patch movements in the
fragmented system instead of 64% in the more contin-
uous one. This reduction of dispersal propensity may be
explained in the light of the dispersal cost hypothesis:
due to predation risks and uncertainty to reach a
suitable patch when dispersing, individuals may be
reluctant to leave patch under highly fragmented condi-
tions. Recent behavioural analyses in P. eunomia sup-
ported this assumption (Schtickzelle and Baguette 2003).
Furthermore, we have also shown previously that
mortality experienced by dispersing individuals was
effectively higher in fragmented vs continuous land-
scapes (Schtickzelle and Baguette 2003, G. Mennechez,
unpubl.). Here, simulations confirmed the cost of dis-
persal in fragmented landscapes since only ca 40�/60% of
the dispersal events are successful, i.e. lead to an
immigration.
VM estimates also indicated here a significant effect of
patch size on emigration in both fragmented landscapes,
which confirms results obtained previously (Petit et al.
2001) in the Pres de la Lienne: emigration probability
decreases when patch size increases. Such relations were
shown in many other butterflies (Hill et al. 1996;
Kuussaari et al. 1996, Baguette et al. 2000). An
explanation often reported for the increasing emigration
rate with decreasing patch area is that butterflies are
more likely to encounter patch boundaries in small
patches and consequently to leave (Kuussaari et al. 1996,
Baguette et al. 2000). However, in a previous study we
showed that the effect of patch size on dispersal rates
depended on the fragmentation level of the landscape
(Mennechez et al. 2003): the comparison of dispersal
pattern between a highly fragmented landscape and a
continuous one revealed an effect of patch size on
dispersal rates in the fragmented landscape only. We
therefore suggest that differences in costs and benefits of
emigration from different patch size exist, leading to
area-dependant dispersal in fragmented landscapes: in
large patches, individuals may be reluctant to leave as
they find resources (nectar, mates. . .) while they may
take the risk of migrating from small patches where
resources are limited. Furthermore, these empirical
results match the predictions of the simulation model
of movement developed by some of us (Schtickzelle and
Baguette 2003): based on the behavioural difference
observed at patch boundary between landscapes with
contrasted level of fragmentation, we developed an
individual-based model that explores the relationship
between patch area, patch permeability and emigration
rate. The simulations showed that the value of zem
increases with habitat fragmentation. In other words,
the decrease in emigration rate with increasing patch
area is more pronounced in fragmented landscapes
where patch boundaries are less permeable to movement.
As a consequence of (1) patch area dependant
emigration and (2) the small size of suitable habitat
patches in both fragmented landscapes (mean patch
sizesB/1 ha), the global level of dispersal observed here
remains high in the two landscapes. Indeed, simulations
indicated that ca 25 to 40% of butterfly-days are spent as
immigrants in a new population. From earlier works on
movement pattern of P. eunomia within fragmented
landscapes, we already knew that the rate of inter-patch
movements was high in this specialist species, particu-
larly within the same river basin (Baguette and Neve
1994, Neve et al. 1996, Petit et al. 2001). Most exchanges
occurred between neighbour patches: the value of a-
parameter indicates that mean daily dispersal distances
were inferior to 500 m. By moving from patch to patch
scattered along rivers (stepping stone movements),
individuals may however travel long distances during
their lifetime. The longest travelled distance found
during this study was more than 5 km, meaning that
most patches within the networks were connected,
therefore allowing metapopulation dynamics and persis-
tence of the species in such a landscape.
Differences in dispersal and mortality patterns
between the two fragmented landscapes
Despite the above-mentioned similarities between the
two fragmented landscapes in comparison with contin-
uous landscapes, it appears that dispersal and mortality
patterns differed highly between Lierneux and Wibrin.
Concerning emigration, our results reveal that the
propensity to disperse was lower and the effect of patch
area on emigration higher in Wibrin, the more fragmen-
ted landscape. This is not surprising since h has been
shown to decrease and zem to increase with fragmenta-
tion (Schtickzelle and Baguette 2003, G. Mennechez,
unpubl.). However, we have to admit that the difference
in the fragmentation degree between the two landscapes
is very low and therefore this effect may need to be
qualified. Those two parameters affect the probability to
emigrate from patch (Eq. 1) and consequently a differ-
ential emigration rate is revealed by the VM model
between the two landscapes: simulations predicted 2.5
time less dispersal events in Wibrin than in Lierneux.
Density-dependence of dispersal which has been
demonstrated in other butterfly species (Kuussaari et al
1996, Brunzel 2002) might also be a likely explanation
for the difference in the dispersal patterns since, as we
pointed out earlier, butterfly densities were on average
2.5 times higher in Lierneux than in Wibrin. In P.
eunomia, Baguette et al. (1998) showed that at low
female density, males were more likely to move away
from patch in search of new mates. Since the probability
for a male to encounter a potential mate is very low in
smaller patches at low density �/ it was the case in
250 OIKOS 106:2 (2004)
Wibrin �/, the probability to leave such patches should
therefore be high. On the other hand, the relation
between densities and emigration could be inversed in
larger patches. At high densities �/ it was the case in
Lierneux*/the presence of many males could motivate
individuals to leave habitat patches: Petit et al. (2001)
showed a positive relationship between male movements
and population size. However, from our results it is not
possible to ascertain whether the observed difference is
purely attributable to landscape structure or to butterfly
densities.
Once dispersing, individuals moved longer distances
in Wibrin, which is probably related to differences in the
spatial configuration of patch network between the two
landscapes. They also encounter a lower risk of mortality
in Wibrin as indicated by the value of l (0.017 vs 0.803
in Wibrin and Lierneux respectively). Simulations pre-
dicted that only ca 40% of dispersal events lead to death
in Wibrin vs ca 60% in Lierneux where the connectivity
as measured by Sj is higher. Insofar as there is no clear
difference in the quality of the matrix between the two
landscapes (fertilized pastures/afforestation areas in both
cases), there is no obvious reason for such a differential
dispersal mortality. Possible explanations could lie on
difference in predation rates or in other sources of
mortality in the matrix.
Finally, immigration scaled significantly to area of
receiver-patch in both landscapes, butterflies tending to
immigrate more into large than into small patches.
However, area of receiver-patches affected more immi-
gration in Wibrin than in Lierneux. As for emigration,
this difference might be related to differences in frag-
mentation degree. In a more highly fragmented land-
scape, the effect of receiver-patch area on immigration is
probably stronger than in a less fragmented landscape,
butterflies having a real interest �/ in term of cost-
benefice �/ to settle and stay in a large patch where nectar
and mate resources are sufficient. Moreover, this effect
of fragmentation could be here reinforced by lower
butterfly densities recorded in Wibrin.
Consequences of generalising dispersal patterns
between landscapes
Simulation results were highly affected by either the
configuration of the patch network in which the model
was run or the set of parameter estimates used. First, the
size and the spatial arrangement of habitat patches
strongly constrained the emigration rate (function of hand Ai scaling by zem) and the survival during dispersal
(function of the connectivity and l). For example, the
VM parameter set from Wibrin resulted in 99% or 62%
of successful migrating events (i.e. 1% or 38% of death
during dispersal) according to the landscape configura-
tion used when running the model, Lierneux or Wibrin
respectively. Secondly, our results clearly show that
depending on the parameter set used to run the VM
model, i.e. generalised or not, simulation results were
significantly different. For the Lierneux patch config-
uration, simulations indicated that parameter general-
isation lead to an under-estimation of dispersal and to
an over-estimation of survival during dispersal. This is
related to the fact that VM parameters from Wibrin
expressed lower probabilities to leave patches and a
lower mortality during dispersal. On the contrary, for
the Wibrin patch configuration, simulations indicated
that parameter generalisation leads to an over estimation
of dispersal and an under estimation of survival during
dispersal. In other words, dispersal parameters seems
very specific to a particular metapopulation and to the
landscape in which it evolves. Hence, we have to face the
fact that generalisation of parameters in PVA models
may create a problem especially when it is conducted
between too dissimilar conditions. Conservation biolo-
gists can find themselves obliged to use best guesses of
demographic and dispersal parameters from the litera-
ture when constructing PVA, because of the lack of
required data (Conroy et al. 1995, Akcakaya 2000, White
2000). However, ‘‘incorrect parameter estimates will
result in unreliable model ouput’’ (Conroy et al. 1995).
To minimise such risk, we recommended to be rigorous
with (1) the environmental conditions under which the
estimation of parameters was performed �/ they have to
be as similar as possible to those under which the PVA
model will be conducted �/ and (2) the incorporation of
uncertainties in the data when building the model of
PVA as outlined notably by Akcakaya and Sjogren-
Gulve (2000).
Conclusion
Modelling dispersal is a fundamental step to the
construction of spatially-structured population viability
analysis models (Akcakaya 2000, Morris and Doak
2002). However, in some cases required data on dispersal
can not be available but in the literature. The message of
our paper is a call for caution with such parameter
generalisations, even within the same species between
comparable landscapes. We believe that the urgency to
provide efficient tools to predict endangered species
persistence in human-made landscapes must not drive
conservation biologists to make dangerous generalisa-
tions.
Acknowledgements �/ We thank Benedicte Gerard, GregoryPlace, Marc Dufrene, Gabriel Neve and Laurent Warge for fieldwork assistance. This work was supported by the EC fundedtraining and mobility of researchers network FRAGLAND(‘‘Survival and Evolution of Species in FragmentedLandscapes’’) and by a grant from the Office of Scientific,Technical and Cultural Affairs (Belgian Federal Government)(contract OSTC-SPSDII EV10/16A 2000�/2004). Butterfly
OIKOS 106:2 (2004) 251
capture licences and sites access were granted from theMinistere de la Region Wallonne. N.S. is ‘‘collaborateurscientifique du F.N.R.S.’’. This is contribution BRC030 of theBiodiversity Research Centre.
References
Akcakaya, H. R. 2000. Population viability analyses withdemographically and spatially structured models. �/ Ecol.Bull. 48: 23�/38.
Akcakaya, H. R. and Sjogren-Gulve, P. 2000. Populationviability analyses in conservation planning: an overview.�/ Ecol. Bull. 48: 9�/21.
Baguette, M. and Neve, G. 1994. Adult movements betweenpopulations in the specialist butterfly Proclossiana eunomia .�/ Ecol. Entomol. 19: 1�/15.
Baguette, M. and Schtickzelle, N. 2003. Local populationdynamics are important to the conservation of metapopula-tions in highly fragmented landscapes. �/ J. Appl. Ecol. 40:404�/412.
Baguette, M. and Stevens, V. M. 2003. Local populations andmetapopulations are both natural and operational cate-gories. �/ Oikos 101: 661�/663.
Baguette, M., Convie, I. and Neve, G. 1996. Male density affectsfemale spatial behaviour in the butterfly Proclossianaeunomia . �/ Acta Oecol. 17: 225�/232.
Baguette, M., Vansteenwegen, C., Convie, I. et al. 1998. Sex-biased density-dependent migration in a metapopulation ofthe butterfly Proclossiana eunomia . �/ Acta Oecol. 19: 17�/
24.Baguette, M., Petit, S. and Queva, F. 2000. Population spatial
structure and migration of the three butterfly species withinthe same habitat network: consequences for conservation.�/ J. Appl. Ecol. 37: 100�/108.
Baguette, M., Mennechez, G., Petit, S. et al. 2003. Effect ofhabitat fragmentation on dispersal in the butterfly Proclossi-ana eunomia . �/ Comptes Rendus Biol. 326: S200�/S209.
Brunzel, S. 2002. Experimental density-related emigration in thecranberry fritillary Boloria aquilonaris. �/ J. Insect Behav. 15:739�/750.
Burnham, K. P. and Anderson, D. R. 1998. Model selection andinference. �/ Springer-Verlag.
Conroy, M. J., Cohen, Y., James, F. C. et al. 1995. Parameterestimation, reliability, and model improvement for spatiallyexplicit models of animal populations. �/ Ecol. Appl. 5: 17�/
19.Gilpin, M. and Hanski, I. 1991. Metapopulation dynamics:
empirical and theoretical investigations. �/ Academic Press.Hanski, I. 1999. Metapopulation ecology. �/ Oxford Univ. Press.Hanski, I. and Gilpin, M. 1997. Metapopulation biology:
ecology, genetics and evolution. �/ Academic Press.Hanski, I., Kuussaari, M. and Nieminen, M. 1994. Metapopu-
lation structure and migration in the butterfly Melitaeacinxia . �/ Ecology 75: 747�/762.
Hanski, I., Moilanen, A. and Gyllenberg, M. 1996. Minimumviable metapopulation size. �/ Am. Nat. 147: 527�/541.
Hanski, I., Alho, J. and Moilanen, A. 2000. Estimating theparameters of survival and migration of individuals inmetapopulations. �/ Ecology. 81: 239�/251.
Hill, J. K., Thomas, C. D. and Lewis, O. T. 1996. Effects ofhabitat patch size and isolation on dispersal by Hesperiacomma butterflies: implications for metapoppulation struc-ture. �/ J. Anim. Ecol. 65: 725�/735.
Kuussaari, M., Nieminen, M. and Hanski, I. 1996. Anexperimental study of migration in the Glanville fritillarybutterfly Melitaea cinxia . �/ J. Anim. Ecol. 65: 725�/735.
Lebreton, J. D., Burnham, K. P., Clobert, J. et al. 1992.Modeling survival and testing biological hypotheses using
marked animals: a unified approach with case studies.�/ Ecol. Monogr. 62: 67�/118.
Maes, D. and Van Dyck, H. 2001. Butterfly diversity loss inFlanders (north Belgium): Europe’s worst case scenario?�/ Biol. Conserv. 99: 263�/276.
McGarigal, K. and Marks, B. 1994. FRAGSTATS: spatialpattern analysis program for quantifying landscape struc-ture. �/ USDA Forest Service General Tech. Rep. PNW-351,Oregon State Univ.
Mennechez, G., Schtickzelle, N. and Baguette, M. 2003.Metapopulation dynamics of the bog fritillary: comparisonof demographic parameters and dispersal between a con-tinuous and a highly fragmented landscape. �/ Land. Ecol.18: 279�/291.
Morris, W. F. and Doak, D. F. 2002. Quantitative conservationbiology: theory and practice of population viability analysis.�/ Sinauer associates, Inc. Publishers.
Neve, G., Barascud, B., Hughes, R. M. et al. 1996. Dispersal,colonisation and metapopulation structure in the vulnerablebutterfly Proclossina eunomia . �/ J. Appl. Ecol. 33: 14�/22.
Petit, S., Moilanen, A., Hanski, I. et al. 2001. Metapopulationdynamics of the bog fritillary butterfly: movements betweenhabitat patches. �/ Oikos 92: 491�/500.
Rodriguez, A. and Andren, H. 1999. A comparison of Eurasianred squirrel distribution in different fragmented landscapes.�/ J. Appl. Ecol. 36: 649�/662.
Saunders, D., Hobbs, R. J. and Margules, C. R. 1991. Biologicalconsequences of ecosystem fragmentation: a review.�/ Conserv. Biol. 5: 18�/32.
Schtickzelle, N. 2003. Metapopulation dynamics and viability ofthe bog fritillary butterfly Proclossiana eunomia . Ph.D.thesis, Biodiversity Research Centre, Univ. Catholique deLouvain, Louvain-la-Neuve, Belgium.
Schtickzelle, N. and Baguette, M. 2003. Behavioural responsesto habitat patch boundaries restrict dispersal and generateemigration-patch area relationships in fragmented land-scapes. �/ J. Anim. Ecol. 72: 533�/545.
Schtickzelle, N. and Baguette, M. 2004. Metapopulationviability analysis of the bog fritillary using a structuredpopulation model. Oikos 104: 277�/290.
Schtickzelle, N., Le Boulenge, E. and Baguette, M. 2002.Metapopulation dynamics of the bog fritillary butterfly:demographic processes in a patchy population. �/ Oikos 97:349�/360.
Schwarz, C. J. and Seber, G. A. F. 1999. Estimating animalabundance: review III. �/ Stat. Sci. 14: 427�/456.
Shreeve, T. G. 1992. Monitoring butterfly movements. �/ In:Dennis, R. L. H. (ed.), The ecology of butterflies in Britain.Oxford Univ. Press, pp. 120�/138.
Thomas C. D., Wilson R. J., Lewis O. T. 2002. Short-termstudies underestimate 30-generation changes in a butterflymetapopulation. �/ Proc. R. Soc. Lond. B. 563�/569.
Van Swaay, C. A. M. and Warren, M. S. 1998. Red data book ofEuropean butterflies. �/ Council of Europe Publishing,Strasbourg.
Wahlberg, N., Moilanen A, Hanski I. 1996. Predicting theoccurrence of endangered species in fragmented landscapes.�/ Science 1536�/1538.
Wahlberg, N., Klemetti, T., Selonen, V. et al. 2002. Metapopu-lation structure and movements in five species of checker-spot butterflies. �/ Oecologia 130: 33�/43.
Warren, M. S. 1993. A review of butterfly conservation incentral southern Britain. I. Protection, evaluation andextinction on prime sites. �/ Biol. Conserv. 64: 25�/35.
White, G. C. 2000. Population viability analysis: data require-ments and essential analyses. �/ In: Boitani, L. and Fuller, T.K. (eds), Research techniques in animal ecology: controver-sies and consequences. Columbia Univ. Press, pp. 288�/331.
Wilcox, B. A. and Murphy, D. D. 1985. Conservation strategy:the effects of fragmentation on extinction. �/ Am. Nat. 125:879�/887.
252 OIKOS 106:2 (2004)