Content uploaded by Tommy Lennartsson
Author content
All content in this area was uploaded by Tommy Lennartsson on Sep 29, 2015
Content may be subject to copyright.
3060
Ecology,
83(11), 2002, pp. 3060–3072
q
2002 by the Ecological Society of America
EXTINCTION THRESHOLDS AND DISRUPTED PLANT–POLLINATOR
INTERACTIONS IN FRAGMENTED PLANT POPULATIONS
T
OMMY
L
ENNARTSSON
1
Swedish University of Agricultural Sciences, Department of Conservation Biology, Box 7002, 750 07 Uppsala, Sweden
Abstract.
In order to estimate how much population extinction risk may be affected
by local fragmentation, population viability analyses were performed in six populations of
the endangered grassland herb
Gentianella campestris
in Sweden. The populations had been
experimentally reintroduced to grasslands that were locally fragmented by juniper shrubs.
The sites represented three sizes of grassland and two levels of grassland abundance in the
landscape. Five years’ demographic data were evaluated in a stochastic matrix population
model, and studies of seed set, pollinator abundance, and inbreeding depression were per-
formed in order to examine possible links between population viability and pollination. In
all six sites, plants with reduced capacity of self-pollination (due to herkogamy) showed
strongly reduced population viability in locallyfragmented grassland habit, withpronounced
extinction thresholds at certain levels of local fragmentation. Population viability was
reduced because of inbreeding depression and reduced seed production in combination,
both caused by pollinator deficit in locally fragmented grasslands. Plants with high selfing
capacity had low population viability over the entire local fragmentation gradients. Selfing
yielded high seed set in the absence of pollinators and was advantageous in fragmented
parts of the grasslands. However, selfing had negative effects in nonfragmented parts of
the grasslands, because it decreased the chances of cross-pollination and because selfed
progeny had reduced fitness compared to outcrossed offspring. A comparison among the
six sites indicated that the negative effects of local fragmentation were amplified by reduced
size of the grassland sites and by reduced abundance of grassland habitat in the landscape.
To my knowledge, this is the first quantitative estimate of increased extinction risk in
fragmented plant populations.
Key words: extinction risk; extinction threshold;
Gentianella campestris;
grassland; habitat frag-
mentation; herkogamy; inbreeding depression; matrix population model; plant demography; pollen
limitation; pollinator abundance; population viability analysis.
I
NTRODUCTION
One fundamental aim of conservation is to counter-
act the ongoing habitat fragmentation, because frag-
mentation can be expected to increase extinction risk
of local populations (Schemske et al. 1994) and meta-
populations (Hanski and Ovaskainen 2000). Theoreti-
cal models predict extinction thresholds, caused by
abrupt disruption of habitat connectivity at certain lev-
els of fragmentation (Gardner et al. 1987, With and
King 1999). Such nonlinear effects of fragmentation
are important for risk assessment and conservation
planning, because populations close to thresholds may
have lower viability than expected (Andre´n 1999). In
plants, negative effects of fragmentation are suggested
by observations of reduced pollination and seed pro-
duction (Jennersten 1988, Aizen and Feinsinger 1994,
Groom 1998, Morgan 1999), recruitment (Jules 1998),
increased inbreeding depression (Heschel and Paige
1995, Taylor et al. 1999), and reduced species diversity
(Dzwonko and Loster 1989, Launer and Murphy 1994)
in small and isolated habitat fragments. However, no
Manuscript received 5 October 2001; revised and accepted 4
April 2002.
1
E-mail: tommy.lennartsson@nvb.slu.se
empirical studies have so far quantified reduced plant
population viability in fragmented habitats by linking
single fitness measures to the species’ whole life cycle.
Such demographic analyses of habitat fragmentation
are essential if we want to know how much habitat
fragmentation may reduce population viability in com-
parison to other environmental factors or at which de-
grees of fragmentation extinction thresholds may occur.
In general, lack of empirical data strongly restricts our
insights into the effects of fragmentation in nature
(Bierzychudek 1981, Bond 1995, Didham et al. 1996,
Kearns et al. 1998, Harrison and Bruna 1999).
Habitat fragmentation results in patchy distributions
of organisms and, because of interpatch distance or
matrix hostility, decreased mobility of individuals be-
tween the patches. This leads to decreased chances of
recolonization of extinct populations (Andre´n 1999).
In addition, the population in each patch can be ex-
pected to experience an increased risk of extinction,
because of inbreeding and genetic erosion (Aizen and
Feinsinger 1994, Oostermeijer et al. 1994), reduced
population size in small patches (Lennartsson 2000),
and edge effects (Bender et al. 1998). Plants usually
have low spontaneous mobility, and the most imme-
diate effects of fragmentation are changed interactions
November 2002 3061
VIABILITY OF FRAGMENTED PLANT POPULATIONS
F
IG
. 1. Schematic detail of a locally fragmented grassland
of the type used in this study. Circles represent
Juniperus
communis
shrubs; ‘‘F’’ indicates local grassland fragments in
which groups of
Gentianella campestris
were established.The
scale at the right shows the degree of local fragmentation
(percentage of grassland that is not covered by shrubs) for
each 3-m section of the fragmentation gradient. The study
was conducted in six different areas in the province of Upp-
land in central Sweden.
between the plants and more mobile organisms, es-
pecially pollinators (Kearns et al. 1998), seed vectors
(Santos et al. 1999), and herbivores and seed predators
(Jennersten and Nilsson 1993, Zabel and Tscharntke
1998). Several studies have demonstrated effects of
fragmentation on patch occupancy of insect popula-
tions (Hanski et al. 1995), on insect abundance, in par-
ticular on specialist species (Zabel and Tscharntke
1998, Kreuss and Tscharntke 2000), and on species
with restricted mobility (Thomas 2000).
The plant’s reproductive system influences how sen-
sitive a population is to disrupted plant–pollinator in-
teractions, and isolation may select for increased au-
togamy (Olesen and Jain 1994, Johnston and Schoen
1996). Self-pollination may decrease a plant’s sensi-
tivity to pollinator deficit in terms of seed set, but not
necessarily in terms of inbreeding (Johnston and
Schoen 1996).
Depending on their mobility, different organisms ex-
perience their environment as fragmented at different
spatial scales of habitat heterogeneity (Thomas 2000).
Usually, fragmentation is studied at the landscape
(metapopulation) scale across habitat patches (popu-
lations), but fragmentation at the local scale, within
patches, may operate in a similar fashion (Hanski and
Gilpin 1991). The grassland community studied here
has undergone both landscape-level fragmentation due
to ceased grazing (Lennartsson and Svensson 1996) and
local fragmentation due to establishment of bushes dur-
ing periods with poor grazing (Fig. 1).
In this study, I estimated how local and landscape
level fragmentation influenced reintroduced popula-
tions of the endangered field gentian (
Gentianella cam-
pestris
L. Bo¨rner) in terms of population extinction risk
and plant–pollinator interactions. I compared maternal
lines with different capacity of selfing (due to herko-
gamy) in order to test the advantages and disadvantages
of self-pollination and cross-pollination, respectively,
in environments with different degrees of fragmenta-
tion.
M
ATERIALS AND
M
ETHODS
Species and sites
The field gentian (
Gentianella campestris
) is a bi-
ennial grassland herb, growing in unfertilized semi-
natural grasslands of high conservation value (Natur-
va˚rdsverket 1987, Lennartsson and Svensson 1996). It
has declined dramatically over its whole range due to
habitat loss and is now included in the lists of threat-
ened plants in most of the countries in which it occurs
(Ingelo¨g et al. 1993). The species is pollinated by bum-
ble bees only, of which six different species were ob-
served pollinating during this study. In many popula-
tions, 85–95% of the seed embryos develop seeds in
absence of pollinators, due to spontaneous selfing and
high self-compatibility (Lennartsson et al. 2000). In
some populations, all or a proportion of the plants have
herkogamous flowers, i.e., the anthers and the stigma
are spatially separated within the flower, which reduces
seed set to
;
45–60% in absence of pollinators. The
two herkogamy types show strong heritability (Len-
nartsson et al. 2000).
In Sweden, many populations experience within-site
habitat fragmentation, because shrubs, mainly junipers,
have established during periods without grazing or with
low grazing intensity. The bushes have often invaded
from an adjacent forest edge, which creates a local
fragmentation gradient over the site (Fig. 1). For this
study, six locally fragmented, cattle-grazed grasslands
were used, all of dry to mesic herb-rich type (Pa˚hlsson
1994).
Experimental setup
In 1993–1994,
Gentianella campestris
was reestab-
lished at the six grasslands, by using seeds from one
population (occurring at a different grassland than the
six used in this study) containing both herkogamous
and nonherkogamous individuals. The six grasslands
3062
TOMMY LENNARTSSON
Ecology, Vol. 83, No. 11
F
IG
. 2. Life cycle graph of
Gentianella campestris
. The transitions
a
ij
,
having values between 0 and 1, show the probability
of the transition from stage
j
to stage
i. F
ij
shows fecundity transitions, which may have a value
.
1. Seed bank 1 and seed
bank 2 are the first year’s and the second year’s seed bank, respectively.
were situated in six different areas in the province of
Uppland in central Sweden. Three of the areas (‘‘land-
scapes’’) were grassland-rich (12–15% seminatural
grassland within 1.5 km), and three were grassland-
poor (2–5%). The six grassland sites furthermore rep-
resented three size classes (2.5, 5, and 15 ha). One site
of each size class was situated in each of the two land-
scape types (i.e., grassland-poor and grassland-rich, re-
spectively). In each grassland, the
Gentianella
popu-
lation was established along the whole local fragmen-
tation gradient (Fig. 1) by choosing 13–15 habitat frag-
ments per site. The fragments were essentially located
along a transect, with the smallest fragments (
;
10 m
2
)
corresponding to the highest degree of local fragmen-
tation (
;
75% habitat loss) and the largest fragments
(
;
2500 m
2
) to the lowest local fragmentation (
;
5%;
see Fig. 1). The vegetation height, litter layer, and soil
type in the selected local fragments were typical for
Gentianella campestris
sites. In each local fragment
seeds were sowed from 15 herkogamous and 15 non-
herkogamous self-pollinated mother plants (30 seeds/
mother) in 1993, all from the same original population.
All seeds from one mother plant were sown in an in-
dividual 40
3
40 cm square. When flowering plants
developed in 1995 all but three per mother plant were
removed to obtain groups with equal initial sizes (90
plants) in all local habitat fragments. This implied re-
moval of 0–6 offspring per mother plant. Because of
the species’ strictly biennial life cycle, the procedure
was repeated in 1994–1996 to ensure flowering adults
every year. The litter layer was scratched prior to sow-
ing to facilitate the initial establishment, but the pop-
ulations were thereafter only affected by normal graz-
ing, without any manipulations of the environment.
Population viability
In order to link mother plants with their progeny, the
seed rain from 15 random mother plants per herkogamy
group per local fragment and year (1995–1998) was
controlled by sowing each mother’s seeds in a 30
3
30 cm square around the mother plant. This mimics
roughly the normal dispersal distance of the species.
For each such seed family, demographic data were col-
lected during 1995–1999, for all life-cycle transitions
(Fig. 2). Data on production of seedlings and winter
survival of rosettes (
a
43
, Fig. 2) were collected in June.
Another census in late August provided data on sur-
vival of seedlings to the rosette stage, growth of ro-
settes and adults, summer survival of adults, and seed
production of adults. The seed bank persistence was
estimated by annually sowing seed batches (
;
100
seeds from three plants per herkogamy group per grass-
land site) in cold frames and by monitoring the ger-
mination from the batches in the first, second, and third
spring after sowing (Lennartsson and Oostermeijer
2001). This germination experiment showed that 25–
35% of the seeds from the batches germinated the first
spring, 4.5–6% the second, and 0.4–0.5% the third
spring after sowing. By combining these germination
data (specific for each year) with the number of seeds
produced 1, 2, and 3 yr back, it was possible to estimate
how many of the August rosettes could be assumed to
originate from last year’s adults (
F
34
, Fig. 2), from first
year’s (
a
31
) and from second year’s seed bank (
a
32
),
respectively. No seedlings appeared after the third
spring, and therefore all nongerminated seeds were
considered dead. Cold frames were used to facilitate
monitoring of germination, but the same patterns of
germination and seed mortality were observed in the
field. Rosette size was given as an index: the number
of leaf pairs times the rosette diameter (in centimeters).
To avoid a 1-yr time lag in the transitions, the adult
plants in year
t
were assumed to produce dormantseeds
(
F
15
, Fig. 2) and rosettes (
F
34
) directly in year
t
1
1
(Caswell 2001).
The demographic parameters were combined in a
stochastic matrix population model (Caswell 2001) to
estimate extinction risk over the local fragmentation
gradients in the six study populations. The life cycle
November 2002 3063
VIABILITY OF FRAGMENTED PLANT POPULATIONS
F
IG
. 3. Population viability of six populations of
Gentianella campestris
along gradients of local fragmentation. Each
data point represents one local grassland fragment. Filled symbols show herkogamous plants; unfilled symbols, nonherko-
gamous (self-pollinating) plants. Two measures of fragmentation are given: local proportion of remaining habitat (top; see
Fig. 1 for explanation) and log(local fragment size) (bottom). The six panels represent six grassland sites, ordered by size
(outer
y
-axis) and habitat fragmentation at the landscape level (outer
x
-axis). Mean time to extinction was calculated for each
local habitat fragment separately using a stochastic matrix population model, with the same initial population size for all
fragments (1000 rosettes, 1000 adults, 500 000 seeds in first year’s seed bank, and 125 000 seeds in second year’s seed bank).
Error bars show
6
SD
of 1000 iterations. The curves are fitted by using an unweighted three-step moving average.
of the species was transformed into a transition matrix
with six autumn–autumn transitions from one life stage
in year
t
to the following stage in year
t
1
1 (Fig. 2).
A stage-structured model is appropriate for
G. cam-
pestris
since it is a strict biennial species, that is, the
probability of flowering the second year is not size
dependent (Lennartsson 1997, Kelly 1989). For each
local habitat fragment, three matrices (1996–1997,
1997–1998, 1998–1999) were constructed and used in
the stochastic population model with 1000 iterations of
population growth during 800 yr. The model sampled
the three matrices in random order, but since 1999 was
exceptionally dry, the frequency of the 1998–1999 ma-
trix was adjusted to the normal drought frequency of
;
0.16 (Ultuna Climate and Bio-climate Station,
un-
published data
). In the model, the same initial popu-
lation size was used for all local habitat fragments:
1000 rosettes, 1000 adults, 500000 seeds in first year’s
seed bank, and 125 000 seeds in second year’s seed
bank. All populations became extinct during the sim-
3064
TOMMY LENNARTSSON
Ecology, Vol. 83, No. 11
F
IG
. 4. Seed set in 1995–1999 in six populations of
Gentianella campestris
along gradients of local fragmentation. Filled
symbols show herkogamous plants; unfilled symbols, nonherkogamous (self-pollinating) plants (see Fig. 3 for explanation).
Error bars indicate
6
1 year-based standard deviation.
ulated time period, and mean time to extinction could
therefore be used as an estimate of population viability.
The simulations were performed using computer soft-
ware developed by Kari Lehtila¨(
personal communi-
cation
).
The effects of pollination on population viability
The estimates of population viability over the local
fragmentation gradients were thus based on standard
demographic field data. In order to examine why (if at
all) some demographic parameters varied over the local
fragmentation gradients (yielding varying population
viability), a number of pollination experiments were
performed. The focus on pollination was motivated by
the assumption that local habitat fragmentation affects
plant population viability mainly through reduced pol-
lination in small and isolated local fragments. First,
reduced pollination may reduce the seed set (the pro-
portion of seed embryos that develop seeds), especially
in herkogamous plants. Second, reduced pollination
may lead to inbreeding depression in one or more life
stages, because of reduced cross-pollination.
Seed set
The natural seed set (seeds per seed embryo) was
followed during 5 yr, 1995–1999. Each year the seed/
embryo ratio was determined on two fruits per plant
on 30 plants (15 per herkogamy group) per local habitat
November 2002 3065
VIABILITY OF FRAGMENTED PLANT POPULATIONS
fragment (thus a total of
;
13500 plants and 27 000
fruits during the study period). In order to detect pollen
limitation, the seed set after natural pollination was
compared with seed set following facilitated self-pol-
lination. One flower per plant on 10 plants (five per
herkogamy group) per local fragment was treated with
additional self pollen 1995–1999. Newly opened flow-
ers were pollinated in the morning, and the flowers
closed spontaneously later during the day. In 1995
cross-pollination was performed also in the same way,
but since cross- and self-pollination yielded the same
seed set only selfing was performed the following
years. Earlier pollination experiments (Lennartsson et
al. 2000) also have shown that selfing and cross-pol-
lination yield equal seed set. No seeds were removed
from the populations.
Inbreeding depression
The demographic data showed the performance of
progeny produced by natural pollination over the local
fragmentation gradients. In order to analyze if any dif-
ferences in progeny performance between large and
small local fragments could be attributed to inbreeding,
natural pollination was compared with controlled self-
pollination (expected high inbreeding effects) and con-
trolled cross-pollination (expected low effects). This
comparison was performed annually during the study,
in two of the largest and two of the smallest local
habitat fragments per site (the largest and the second
largest local fragments at each site, plus the second
smallest and the third or fourth smallest). Natural pol-
lination was the normal situation for all plants in this
study, and detailed demographic data were thus ob-
tained for 15
1
15 mother plants per local fragment
per year (see
Population viability
, above). The self-
pollination treatment was obtained by self-pollinating
one flower per plant on five plants per herkogamy group
and local fragment (the same flowers as in the study
of seed set). The seeds from each fruit were sown in
a
;
20
3
20 cm square to enable monitoring of ger-
mination, survival of seedlings to autumn-rosette stage,
growth of rosettes, survival of rosettes to autumn-adult
stage, fruit production of adults, and fruit size of adults
(in terms of number of seed embryos per fruit). The
cross-pollination treatment was performedon one flow-
er per plant on five plants per herkogamy group per
local fragment. Newly opened flowers were pollinated
with dehisced anthers from other plants belonging to
the same herkogamy group. Data on progeny perfor-
mance were collected as for self-pollination.
Abundance of pollinators
The second week of August (the peak of flowering of
Gentianella campestris
) 1995–1999, the number of flow-
er-visiting bumble bees per hour was recorded during 1
d (10 h) in the central 10 m
2
of the two small and the
two large local habitat fragments per site. One person
alternated between the two small local fragments and
another person between the two large fragments, ob-
serving bumble bees in 15-min periods. One bumble bee
visit was defined as one bumble bee entering the plot
and visiting at least one
Gentianella
plant.
Statistical treatment
Differences in pollinator abundance between local
habitat fragments and between grassland sites were
tested for by using the nonparametric Mann-Whitney
U
test. The differences between local fragments were
analyzed for each year separately, using Bonferroni
correction for multiple comparisons. Progeny fitness
data were normally distributed and nested ANOVAs
were used to test for sources of variation. In the pol-
lination experiment, pollination treatment was consid-
ered a fixed effect and site (nested within treatment)
and cohort (nested within site within treatment) were
considered as random effects. The analyses of natural
pollination were performed in the same way, with de-
gree of local fragmentation as a fixed effect (Sokal and
Rohlf 1995:271). The study of progeny performance
following selfing, outcrossing, and natural pollination
was performed in two of the largest and two of the
smallest local habitat fragments per site. The difference
between the two small local fragments was first tested
for using Mann-Whitney
U
tests for each year and site
separately, and the same test was performed for the
differences between the two large local fragments.
Since the difference small-small and large-large was
nonsignificant in all cases (
P
.
0.12), the two small
local fragments were pooled in the nested ANOVA, as
well as the two large local fragments.
The number of replicate sites (six) was too low to
allow any detailed analyses of differencesbetweensites
or between landscapes. For each site, a curve was fitted
through the estimates of population extinction risk over
the local fragmentation gradient, by using an un-
weighted three-step moving average (i.e., for a certain
local fragment, the curve shows the mean extinction
risk of that fragment and the two adjacent smaller
ones). Based on a graphical interpretation of the curves,
the local fragmentation gradients could be divided into
three segments: high degree of local fragmentation (9–
20 m
2
fragments), intermediate (40–120 m
2
fragments),
and low degree of local fragmentation (200–2500 m
2
fragments). The average extinction risk was calculated
for each segment and site, and Kruskal-Wallis non-
parametric ANOVA was used to analyze how the av-
erage extinction risk for each segment varied between
sizes of grassland sites and between landscapes. All
analyses were performed using SPSS 10.1 for Windows
(SPSS 2000).
R
ESULTS
The stochastic population model showed that pop-
ulation viability of herkogamous plants, in terms of
time to extinction, was reduced by 80–85% as a result
of local habitat fragmentation and exhibited distinct
3066
TOMMY LENNARTSSON
Ecology, Vol. 83, No. 11
T
ABLE
1. The mean number of bumble bee visits to
Gentianella campestris
plants per hour in large and small local habitat
fragments in six grassland sites during 5 yr in the province of Uppland in central Sweden.
Year†
Mean no. bumble bee visits/h and 10 m
2
(
SD
)
Site 1 (2.5 ha),
landscape 1
(4%)
Small Large
Site 2 (2.4 ha),
landscape 2
(15%)
Small Large
Site 3 (5 ha),
landscape 3
(5%)
Small Large
Site 4 (5.2 ha),
landscape 4
(12%)
Small Large
Site 5 (15 ha),
landscape 5
(2%)
Small Large
Site 6 (14 ha),
landscape 6
(14%)
Small Large
1995
1996
1997
0.20
(0.42)
0.20
(0.42)
0.20
(0.63)
0.30
(0.67)
0.70
(0.67)
1.00
(0.94)
0.10
(0.32)
0.20
(0.63)
0
0.50
(0.71)
0.90
(0.88)
0.90
(0.88)
0.10
(0.32)
0.20
(0.42)
0.20
(0.42)
1.20
(0.63)
0.50
(0.71)
1.30
(0.95)
0.10
(0.32)
0.30
(0.48)
0.10
(0.32)
1.10
(0.88)
1.00
(0.82)
1.00
(0.67)
0.30
(0.67)
0.30
(0.67)
0.20
(0.42)
0.60
(0.70)
1.50
(0.85)
1.70
(0.95)
0.20
(0.63)
0.40
(0.70)
0.20
(0.42)
1.20
(0.92)
1.50
(1.08)
1.56
(1.65)
1998
1999
0.10
(0.32)
0
0.80
(1.03)
0.30
(0.48)
0.20
(0.42)
0.10
(0.32)
0.80
(0.63)
0.60
(0.70)
0.10
(0.32)
0.30
(0.67)
1.30
(0.82)
0.30
(0.67)
0.20
(0.42)
0.20
(0.42)
0.80
(0.79)
0.90
(0.88)
0.10
(0.32)
0.20
(0.63)
1.60
(0.70)
0.50
(0.71)
0.20
(0.42)
0.10
(0.32)
1.40
(1.17)
1.00
(0.82)
All years‡ 1.4
(0.89)
6.2
(3.11)
0.12
(0.84)
0.74
(1.82)
0.18
(0.84)
0.92
(4.82)
0.18
(0.83)
0.96
(1.14)
0.22
(0.84)
1.18
(5.81)
0.22
(1.10)
1.34
(2.30)
Notes:
See
Materials and methods: Experimental setup
for information about sampling methods. For each of the six sites,
the area is given in hectares along with the percentage of grassland habitat in the landscape (within 1.5 km) in which it is
situated.
† Per-hour visitation rate of 10 h per year per local fragment. Means in bold indicate significant differences between small
and large local fragments for each year and site (
P
,
0.05 in Bonferroni-corrected Mann-Whitney
U
tests for each year and
site separately;
N
5
20 h in all tests).
‡ Per-day visitation rate of one day per year during 5 yr. Means in bold indicate significant differences between small and
large local fragments for each site (
P
,
0.05 in Bonferroni-corrected Mann-Whitney
U
tests for each site separately;
N
5
5 yr in all tests).
extinction thresholds (Fig. 3). Population viability of
nonherkogamous plants was not affected by local hab-
itat fragmentation. It exceeded the viability of herko-
gamous plants in locally fragmented habitats, but was
considerably reduced compared to the viability of her-
kogamous plants in nonfragmented habitats (Fig. 3).
Seed set of herkogamous plants decreased with in-
creasing local fragmentation in all six populations and
showed pronounced thresholds at 40–55% habitat loss
(Fig. 4). The thresholds corresponded to the thresholds
observed for population viability over the local frag-
mentation gradients. Addition of self-pollen to the stig-
ma produced fruits with minimum of 91% and maxi-
mum 97% of seed set, regardless of herkogamy type
and degree of local fragmentation. Also the cross-pol-
linated flowers in the study of inbreeding depression
obtained high seed set (91–95%). The pattern of seed
set over the local fragmentation gradients corresponded
to the abundance of bumble bees (Table 1). Thirty 1-d
studies of bumble bee abundance were performed (six
sites during 5 yr). In 29 of the studies more bumble
bees were observed in large local habitat fragments
than in small. The difference was significant in 19 cases
(Table 1). As an average for the whole study period,
4–6 times more bumble bees were observed in large
than in small local habitat fragments, and this differ-
ence was significant for all sites (per day visitation rate;
Table 1). There was a tendency for large grassland sites
to have a higher abundance of bumble bees than small
sites, especially where the grasslands were not locally
fragmented (Table 1). On average, 12.6 bumble bee
visits per day were recorded in the two 15-ha grassland
sites in absence of local fragmentation, as compared
to 6.8 visits per day in the two 2.5-ha sites (Mann-
Whitney
U
5
14.5,
n
5
20,
P
5
0.007).
The pollination experiment showed that offspring
produced by selfing experienced inbreeding depression
compared to outcrossed progeny in three out of six
steps in the life cycle: survival from seedling to mature
rosette, growth of rosettes, and fruit production of adult
plants (Fig. 5). This effect was significant for both her-
kogamy groups (Table 2). Natural pollination of her-
kogamous plants in locally fragmented habitats pro-
duced offspring that were similar to selfed offspring,
whereas pollination in nonfragmented habitats pro-
duced offspring similar to outcrossed progeny (Fig. 5).
Thus, rosette survival, rosette growth, and fruit pro-
duction was significantly affected by local habitat frag-
mentation (Table 2). Also in nonherkogamous plants
locally nonfragmented habitat was more favorable, but
the differences between fragmented and nonfragmented
were smaller (Fig. 5) and significant only for rosette
survival and rosette growth (Table 2).
A comparison between the local population viability
curves of the six sites indicated effects of both grass-
land area and proportion grassland habitat in the land-
scape. In the two largest grasslands the rapid decline
of population viability occurred where
;
45–65% of
the local grassland habitat was lost by fragmentation,
as compared to 30–45% in the two smallest grasslands
(Fig. 6). The corresponding figures for the medium-
sized grasslands were 35–55%. In landscapes where
grassland habitat was abundant, populations of her-
kogamous plants became extinct within
;
250–300 yr
November 2002 3067
VIABILITY OF FRAGMENTED PLANT POPULATIONS
F
IG
. 5. Six fitness parameters of
Gentianella campestris
offspring produced by controlled self-pollination, controlled
cross-pollination, natural pollination in a locally fragmented grassland habitat, and natural pollination in a locally non-
fragmented habitat (means
1
1
SD
). The pollinated plants were either herkogamous or nonherkogamous. (A) Germination
of four cohorts at six sites each. (B) Survival from seedling to mature rosette (four cohorts, six sites). (C) Survival from
mature rosette to fruiting adult (three cohorts, six sites). (D) Size of mature rosettes measured as number of leaf pairs times
rosette diameter in centimeters (four cohorts, six sites). (E) Fruit production of adult plants (three cohorts, six sites). (F)
Number of seed embryos per fruit (three cohorts, six sites).
if they were not affected by local fragmentation. Pop-
ulations in grassland-poor landscapes became extinct
within
;
200–250 yr. This difference was generated by
the matrix population model, because seed set varied
less between years in grassland-rich landscapes, than
in grassland-poor landscapes (see error bars in Fig. 4).
A mean extinction risk for all large local fragments
(fragment sizes of 200–2500 m
2
) was calculated for
each site. These means varied significantly between the
two landscape types (Kruskal-Wallis
x
2
5
9.9, df
5
1,
P
5
0.002), but not between the three size classes of
grassland sites (
x
2
5
1.1, df
5
2,
P
5
0.57). The cor-
responding means for the intermediate-sized local frag-
ments (40–120 m
2
) varied significantly with size of
grassland sites (
x
2
5
16.8, df
5
2,
P
,
0.001), but not
with grassland cover in the landscape (
x
2
5
0). The
3068
TOMMY LENNARTSSON
Ecology, Vol. 83, No. 11
T
ABLE
2. Nested ANOVA for variation in six fitness characters of
Gentianella campestris
offspring produced by two
experimental pollination treatments (Treatment, T) and by natural pollination in locally fragmented (70%) and nonfrag-
mented (5%) grassland habitat (Fragmentation, F) in six grassland sites (Site, S). The study was repeated on 3–4 cohorts
(Cohort, C).
Survival
Germination Rosettes Adults
Source of variation
SS
df
F
SS
df
F
SS
df
F
Experimental pollination (cross-pollination and self-pollination)
Herkogamous
T
S(T)†
C(S(T))‡
Error
25.6
71.8
163
893
1
10
36
192
3.56
1.58
0.97
0.83
0.022
3.99
0.77
1
10
36
192
372
0.019
27.5
0.19
312
29809
3585
1
10
24
144
0.006
0.025
49.9
Nonherkogamous
T
S(T)†
C(S(T))‡
Error
39.6
65.6
220
826
1
10
36
192
6.04
1.07
1.42
0.86
0.02
2.98
1.15
1
10
36
192
379
0.027
13.8
44.3
294
27265
3323
1
10
24
144
1.50
0.026
49.2
Natural pollination (in 70% and 5% locally fragmented habitat)
Herkogamous
F
S(F)§
C(S(F))
\
Error
29.4
223
206
5749
1
10
36
672
1.32
3.90
0.67
1.34
0.058
9.39
5.36
1
10
36
672
234
0.022
32.7
227
880
94645
14950
1
10
24
504
2.58
0.022
133
Nonherkogamous
F
S(F)§
C(S(F))
\
Error
24.7
226
448
4547
1
10
36
672
1.10
1.81
1.84
0.063
0.053
8.76
4.22
1
10
36
672
11.9
0.022
38.7
399
907
116695
16858
1
10
24
504
4.40
0.019
145
Notes:
The pollinated plants belonged to two herkogamy groups: herkogamous and nonherkogamous.
F
values in bold are
significant at the 0.05 level or lower.
† Nested ANOVA for site nested within pollination treatment.
‡ Cohort nested within site, nested within treatment.
§ Site nested within degree of fragmentation.
\
Cohort nested within site, nested within degree of fragmentation. Fixed factors in the nested ANOVA were pollination
treatments and fragmentation.
average extinction risks for the smallest fragments (9–
20 m
2
), finally, varied neither with size of grassland
sites (
x
2
5
3.5, df
5
2,
P
5
0.17), nor with grassland
cover in the landscape (
x
2
5
0.2, df
5
1,
P
5
0.69).
D
ISCUSSION
This study demonstrates dramatically reduced pop-
ulation viability and distinct extinction thresholds in
locally fragmented populations of the grassland bien-
nial
Gentianella campestris
. The drastic population ef-
fects could be explained by inbreeding depression and
reduced seed output in combination, both caused by
pollinator deficit in fragmented habitats. The effects of
local fragmentation differed strongly between maternal
families with different capacity of self-pollination. The
results provide rare empirical evidence of population
extinction risk being affected by pollen limitation (Har-
rison and Bruna 1999, Menges 2000) and inbreeding
depression (Johnston and Schoen 1996, Saccheri et al.
1998).
Seed production
Nonherkogamous plants obtained high seed set over
the whole local fragmentation gradient. Herkogamous
plants, having reduced capacity of self-pollination, ob-
tained reduced seed set in small fragments. Facilitated
selfing and cross-pollination yielded high seed set over
the whole fragmentation gradient, thus proving pollen
limitation as a consequence of local habitat fragmen-
tation. Seed set over the local fragmentation gradient
described thresholds at certain degrees of fragmenta-
tion. This confirms that the pollinators’ environment
abruptly becomes disconnected over a narrow range of
habitat loss, as predicted by theoretical models (e.g.,
Gardner et al. 1987, With et al. 1999). It is notable that
the disconnection here was poorly related to distance
between local fragments, since it occurred at a small
spatial scale (Fig. 1) in relation to the flight distances
of bumble bees (Bowers 1985, Steffan-Dewenter and
Tscharntke 1999). Instead, it might be caused by bushes
obstructing the pollinators’ mobility (Westerbergh and
Saura 1994) and by decreased attraction of small, shad-
ed, and floristically poor fragments (Bowers 1985, Jen-
nersten 1988). High seed set in locally nonfragmented
grassland, and low seed set in fragmented, agreed with
the abundance of pollinators in large and small local
habitat fragments, respectively.
November 2002 3069
VIABILITY OF FRAGMENTED PLANT POPULATIONS
T
ABLE
2. Extended.
Reproduction
Rosette size
SS
df
F
Fruits per plant
SS
df
F
Seed embroys/fruit
SS
df
F
3219
33.2
9353
3749
1
10
36
192
969
0.013
13.3
39.9
1.66
375
40.5
1
10
24
144
240
0.011
55.6
0.19
284
933
4352
1
10
24
144
0.007
0.73
1.29
3130
34.0
8178
3377
1
10
36
192
921
0.015
12.9
43.4
1.65
369
65.8
1
10
24
144
263
0.011
33.6
7.59
197
1431
3554
1
10
24
96
0.39
0.33
1.73
2630
87.1
15453
13841
1
10
36
672
302
0.020
20.8
85.6
4.78
1087
184
1
10
24
504
179
0.011
124
60.7
570
791
17925
1
10
24
504
1.07
1.73
0.93
222.3
69.1
13598
14569
1
10
36
672
32.2
0.018
17.4
1.13
4.17
1281
176
1
10
24
504
2.71
0.008
153
1.87
69.8
395
18283
1
10
24
504
0.27
0.42
0.45
F
IG
. 6. Comparison between the population
viability curves in Fig. 4. Areas of the grassland
sites are indicated by 2.5, 5, and 15 ha. Labels
2–5% and 12–15% indicate the proportion of
grassland habitat within 1.5 km from the site.
Inbreeding depression
Seed set thus suggests that self-pollination in com-
parison with herkogamy yields equal or higher fitness.
However, for both herkogamy types of
G. campestris
,
progeny produced by controlled selfing showed in-
breeding depression compared to outcrossed progeny.
Natural pollination in locally fragmented habitats pro-
duced inbred progeny in both herkogamy groups,
which indicates that few seeds were produced by out-
crossing. This corresponded with low pollinator abun-
dance in locally fragmented habitats. In absence of lo-
cal fragmentation, inbreeding depression was consid-
erably reduced in herkogamous plants, but less in non-
herkogamous plants. In particular, nonherkogamous
plants showed low fruit production in both ends of the
fragmentation gradient. This indicates higher outcross-
ing rates in herkogamous than in nonherkogamous
plants in absence of local habitat fragmentation. The
explanation for low outcrossing in nonherkogamous
plants in spite of high pollinator abundance was that
flowers of
G. campestris
closed during the day of pol-
lination. Earlier studies (Lennartsson et al. 2000) have
shown that nonherkogamous flowers close within a
maximum of 1.6 d, compared to 5 d in herkogamous
3070
TOMMY LENNARTSSON
Ecology, Vol. 83, No. 11
plants and thus that efficient self-pollination made the
nonherkogamous flowers unsusceptible to cross-polli-
nation more rapidly than herkogamous flowers.
One important result of this study is that local frag-
mentation resulted in inbreeding entirely because of
reduced pollen transfer. In contrast to many other stud-
ies, effects of genetic erosion in isolated groups of
plants (Steffan-Dewenter and Tscharntke 1999) can be
excluded because all local fragments can be assumed
to have approximately uniform genetic variation. The
populations were initially established using the same
number of mother plants. Since outcrossed (within lo-
cal fragments) progeny showed no fitness reduction in
any habitat fragments, it is not likely that any important
erosion of the original genetic material took place dur-
ing the study. Furthermore, reduced pollen transfer was
not an effect of reduced plant population density (Ku-
nin 1993), because this factor did not change notably
over the study period.
Population viability
High capacity for self-pollination in
G. campestris
was advantageous in terms of seed set, but carried costs
in terms of inbreeding depression in the following gen-
eration. The overall population viability is a function
of both seed production and progeny performance
throughout the life cycle. Selfing made seed set inde-
pendent of pollinators and could therefore partly neu-
tralize the negative effects of local fragmentation on
population viability. However, self-pollination de-
creased the chances of cross-pollination where polli-
nators were abundant, which resulted in evenly high
inbreeding effects and reduced population viability
over the whole local fragmentation gradient. In sum-
mary, the advantages of selfing and high seed set in
nonherkogamous plants exceeded the cost of inbreed-
ing depression in locally fragmented habitats. In non-
fragmented habitats, on the other hand, advantages of
cross-pollination and reduced inbreeding depression in
herkogamous plants were greater than advantages of
high and secure seed set (cf. Taylor et al. 1999).
The population viability curve for herkogamous
plants over the local fragmentation gradient corre-
sponded with both the pattern of seed set over the gra-
dient and with the abundance of pollinators in small
and large local habitat fragments. No significant frag-
ment effect on fitness was discovered when plants were
treated with the same pollination, which indicates that
possible environmental differences between small and
large fragments had little influence on population vi-
ability. It therefore seems clear that the observed var-
iation in population viability was caused mainly by
disrupted plant–pollinator interactions as a result of
local habitat fragmentation. The observed population
trends approximately corresponded to the estimated
population growth rates, that is, nonherkogamous pop-
ulations declined more than herkogamous, except in
the smallest habitat fragments.
Effects of fragmentation at different spatial scales
The number of replicate grasslands in this study is
too low to draw far-reaching conclusions about effects
of fragmentation at the landscape scale. However, pop-
ulation viability varied among sites in a highly sug-
gestive pattern, which was also supported by the ob-
served abundance of pollinators. The pattern suggests
that population viability of
G. campestris
was influ-
enced by fragmentation at different spatial scales: (a)
local fragmentation (within populations and sites), (b)
area of grassland site, and (c) fragmentation at the land-
scape level, in terms of grassland abundance in 3-km
squares.
Local fragmentation (a) had the strongest impact and
determined the general threshold shape of the popu-
lation viability curves. Area of grassland site (b) in-
fluenced at which degree of local fragmentation the
extinction threshold occurred. In the largest grasslands
the threshold occurred at an
;
50% higher degree of
local fragmentation than in the smallest sites (30–45%
habitat loss in the 2.5-ha sites, compared to 45–65%
in the 15-ha sites). Hence, with increasing grassland
area, the field gentian became less sensitive to local
habitat fragmentation. This was clearly an effect of
higher bumble bee frequencies in large grasslands, an
observation reported also from other studies (Bowers
1985).
Habitat fragmentation at the landscape level (c), fi-
nally, affected the maximum population viability in
absence of local fragmentation, i.e., above the extinc-
tion threshold. In sites situated in grassland-rich land-
scapes, both seed set and pollinator abundance varied
less between years than in grassland-poor landscapes.
Lower between-year variation in seed set resulted in
higher population viability when the demographic data
were entered in the stochastic population model. A less
varying pollination in grassland-rich landscapes can be
expected if bumble bee populations are favored by, for
example, floristic diversity (Bowers 1985) or if local
variations in bumble bee mortality are buffered by im-
migration from neighboring grasslands.
Implications for conservation
Most empirical and theoretical work on habitat frag-
mentation has focused on the landscape scale and meta-
populations (Hanski and Gilpin 1991). This study dem-
onstrates that local habitat fragmentation in a similar
fashion also can disrupt plant–pollinator interactions
(Kearns et al. 1998, McIntyre and Hobbs 1999) and
result in reduced plant population viability and extinc-
tion thresholds. Local fragmentation had a strong, di-
rect effect on plant population viability, and the results
indicate that this effect can be amplified by fragmen-
tation and habitat loss at larger spatial scales (cf. An-
dre´n 1994, Bender et al. 1998). The combined effects
of local- and landscape-level fragmentation increased
the extinction risk of
G. campestris
from very low to
November 2002 3071
VIABILITY OF FRAGMENTED PLANT POPULATIONS
a risk approximately corresponding to the IUCN threat
category ‘‘vulnerable’’ (IUCN 1994). Thus, fragmen-
tation can reduce population viability considerably,
even for the relatively large population of 2000 plants
that was used here as initial population size in the sto-
chastic population model. For reduced population sizes
(which can be expected to be an effect of habitat frag-
mentation) the extinction risk would be even higher
(Lennartsson and Oostermeijer 2001). Many Swedish
grassland sites are locally fragmented like the ones
used in this study, and many of the last populations of
Gentianella campestris
occur in small and floristically
poor grassland fragments in a more and more forested
landscape (Lennartsson and Svensson 1996). Frag-
mentation clearly threatened the success of this rein-
troduction experiment, which is striking since
G. cam-
pestris
has a high capacity of self-pollination and is
pollinated by a common group of pollinators. Surpris-
ingly few population viability analyses have included
pollination (Menges 2000), but as more studies are per-
formed we may find that pollinator deficit in frag-
mented habitats is an important threat to plant popu-
lations and plant diversity.
A
CKNOWLEDGMENTS
I thank Tom Juenger and Kari Lehtila¨ forhelpfulcomments,
Kari also for developing and providing computer software
for the stochastic model. I further thank all field assistants
for their patience, the Uppsala University Botanical Garden
for help with seed bank experiments, and Eric Menges and
two anonymous referees for valuable comments on the man-
uscript. The study was supported by the Foundation for Stra-
tegic Environmental Research (award 438-97-1) and by the
Swedish Council for Forestry and Agricultural Research
(award 34.0297/98).
L
ITERATURE
C
ITED
Aizen, M. A.,
AND
P. Feinsinger. 1994. Forest fragmentation,
pollination, and plant reproduction in a chaco dry forest,
Argentina. Ecology 75:330–351.
Andre´n, H. 1994. Effects of habitat fragmentation on birds
and mammals in landscapes with different proportions of
suitable habitat—a review. Oikos 71:355–366.
Andre´n, H. 1999. Habitat fragmentation, the random sample
hypothesis and critical thresholds. Oikos 84:306–308.
Bender, D. J., T. A. Contreras, and L. Fahrig. 1998. Habitat
loss and population decline: a meta-analysis of the patch
size effect. Ecology 79:517–533.
Bierzychudek, P. 1981. Pollinator limitation of plant repro-
ductive effort. American Naturalist 117:838–840.
Bond, W. J. 1995. Assessing the risk of plant extinction due
to pollinator and disperser failure. Pages 131–146
in
J. H.
Lawton and R. M. May, editors. Extinction rates. Oxford
University Press, Oxford, UK.
Bowers, M. A. 1985. Bumble bee colonization, extinction,
and reproduction in subalpine meadows in northeastern
Utah. Ecology 66:914–927.
Caswell, H. 2001. Matrix population models, construction,
analysis, and interpretation. Sinauer Associates, Sunder-
land, Massachusetts, USA.
Didham, R. K., J. Ghazoul, N. E. Stork, and A. J. Davis.
1996. Insects in fragmented forests: a functional approach.
Trends in Ecology and Evolution 11:255–260.
Dzwonko, Z., and S. Loster. 1989. Distribution of vascular
plant species in small woodlands on the western Carpathian
foothills. Oikos 56:77–86.
Gardner, R. H., B. T. Milne, M. G. Turner, and R. G. O’Neill.
1987. Neutral models for the analysis of broad-scale land-
scape patterns. Landscape Ecology 1:19–28.
Groom, M. J. 1998. Allee effects limit population viability
of an annual plant. American Naturalist 151:487–496.
Hanski, I., and M. E. Gilpin. 1991. Metapopulation dynam-
ics: brief history and conceptual domain. Pages 1–16
in
M.
E. Gilpin and I. Hanski, editors. Metapopulation dynamics:
empirical and theoretical investigations. Academic Press,
London, UK.
Hanski, I., and O. Ovaskainen. 2000. The metapopulation
capacity of a fragmented landscape. Nature 404:755–758.
Hanski, I., T. Pakkala, M. Kuussaari, and G. C. Lei. 1995.
Metapopulation persistence of an endangered butterfly in
a fragmented landscape. Oikos 72:21–28.
Harrison, S., and E. Bruna. 1999. Habitat fragmentation and
large-scale conservation: what do we know for sure? Ecog-
raphy 22:225–232.
Heschel, M. S., and K. N. Paige. 1995. Inbreeding depres-
sion, environmental stress, and population size variation in
scarlet gilia (
Ipomopsis aggregata
). Conservation Biology
9:126–133.
Ingelo¨g, T., R. Andersson, and M. Tjernberg. 1993. Red Data
Book of the Baltic Region. Part 1. List of threatened vas-
cular plants and vertebrates. Swedish Threatened Species
Unit, Uppsala, Sweden.
IUCN [International Union for the Conservation of Nature
and Natural Resources]. 1994. Red List categories. IUCN
Species Survival Commission, Gland, Switzerland.
Jennersten, O. 1988. Pollination in
Dianthus deltoides
(Car-
yophyllaceae): effects of habitat fragmentation on visita-
tion and seed set. Conservation Biology 2:359–366.
Jennersten, O., and S. G. Nilsson. 1993. Insect flower visi-
tation frequency and seed production in relation to patch
size of
Viscaria vulgaris
(Caryophyllaceae).Oikos 68:283–
292.
Johnston, M. O., and D. J. Schoen. 1996. Correlated evo-
lution of self-fertilization and inbreeding depression: an
experimental study of nine populations of
Amsinckia
(Bor-
aginaceae). Evolution 50:1478–1491.
Jules, E. S. 1998. Habitat fragmentation and demographic
change for a common plant: trillium in old-growth forest.
Ecology 79:1645–1656.
Kearns, C. A., D. W. Inouye, and N. M. Waser. 1998. En-
dangered mutualisms: the conservation of plant–pollinator
interactions. Annual Review of Ecology and Systematics
29:83–112.
Kelly, D. 1989. Demography of short-lived plants in chalk
grassland. I. Life-cycle variation in annuals and strict bi-
ennials. Journal of Ecology 77:747–769.
Kruess, A., and T. Tscharntke. 2000. Species richness and
parasitism in a fragmented landscape: experiments andfield
studies with insects on
Vicia sepium
. Oecologia 122:129–
137.
Kunin, W. E. 1993. Sex and the single mustard: population
density and pollinator behavior effects on seed-set.Ecology
74:2145–2160.
Launer, A. E., and D. D. Murphy. 1994. Umbrella species
and the conservation of habitat fragments—a case of a
threatened butterfly and a vanishing grassland ecosystem.
Biological Conservation 69:145–153.
Lennartsson, T. 1997. Om gentianornas gronings- och blom-
ningsfenologi. Svensk Botanisk Tidskrift 90:263–265.
Lennartsson, T. 2000. Management and population viability
of the pasture plant
Gentianella campestris
: the role of
interactions between habitat factors. Ecological Bulletines
48:111–121.
Lennartsson, T., and J. G. B. Oostermeijer. 2001. Demo-
graphic variation and population viability in
Gentianella
3072
TOMMY LENNARTSSON
Ecology, Vol. 83, No. 11
campestris
: effects of grassland management and environ-
mental stochasticity. Journal of Ecology 89:451–463.
Lennartsson, T., J. G. B. Oostermeijer, J. van Dijk, and H. C.
M. den Nijs. 2000. Ecological significance and heritability
of floral reproductive traits in
Gentianella campestris
(Gen-
tianaceae). Basic and Applied Ecology 1:69–81.
Lennartsson, T., and R. Svensson. 1996. Patterns in the de-
cline of three species of
Gentianella
in Sweden illustrating
the deterioration of semi-natural grasslands. Symbolae Bo-
tanicae Upsaliensis 31(3):169–184.
McIntyre, S., and R. Hobbs. 1999. A framework for con-
ceptualizing human effects on landscapes and its relevance
to management and research models. ConservationBiology
13:1282–1292.
Menges, E. S. 2000. Applications of population viability
analyses in plant conservation. Ecological Bulletines 48:
73–84.
Morgan, J. W. 1999. Effects of population size on seed pro-
duction and germinability in an endangered, fragmented
grassland plant. Conservation Biology 13:266–273.
Naturva˚rdsverket. 1987. Inventering av a¨ngs- och hagmar-
ker. Handbok. Naturva˚rdsverket, Solna, Sweden.
Olesen, J. M., and S. K. Jain. 1994. Fragmented plant pop-
ulations and their lost interactions. Pages 417–426
in
V.
Loeschcke, J. Tomiuk, and S. K. Jain, editors. Conservation
genetics. Birkha¨user, Basel, Switzerland.
Oostermeijer, J. G. B., M. W. van Eijck, and J. C. M. den
Nijs. 1994. Offspring fitness in relation to population size
and genetic variation in the rare perennial plant species
Gentiana pneumonanthe
(Gentiananceae). Oecologia 97:
289–296.
Pa˚hlsson, L. 1994. Vegetationstyper i Norden (Nordic veg-
etation types). Nordic Council of Ministers, Copenhagen,
Denmark.
Saccheri, I., M. Kuussaari, M. Kankare, P. Vikman, W. For-
telius, and I. Hanski. 1998. Inbreeding and extinction in a
butterfly metapopulation. Nature 392:491–494.
Santos, T., J. L. Telleria, and E. Virgos. 1999. Dispersal of
Spanish juniper
Juniperus thurifera
by birds and mammals
in a fragmented landscape. Ecography 22:193–204.
Schemske, D. W., B. C. Husband, M. H. Ruckelshaus, C.
Goodwillie, I. M. Parker, and J. G. Bishop. 1994. Evalu-
ating approaches to the conservationof rareand endangered
plants. Ecology 75:584–606.
Sokal, R. R., and F. J. Rohlf. 1995. Biometry: the principles
and practice of statistics in biological research. Freeman,
New York, New York, USA.
SPSS. 2000. SPSS 10.1 for Windows. SPSS, Chicago, Illi-
nois, USA.
Steffan-Dewenter, I., and T. Tscharntke. 1999. Effects of hab-
itat isolation on pollinator communities and seed set. Oec-
ologia 121:432–440.
Taylor, D. R., S. Trimble, and D. E. McCauley. 1999. Eco-
logical genetics of gynodioecy in
Silene vulgaris
: relative
fitness of females and hermaphrodites during the coloni-
zation process. Evolution 53:745–751.
Thomas, C. D. 2000. Dispersal and extinction in fragmented
landscapes. Proceedings of the Royal Society of London
Series B, Biological Sciences 267:139–145.
Westerbergh, A., and A. Saura. 1994. Gene flow and polli-
nator behavior in
Silene dioica
populations. Oikos 71:215–
224.
With, K. A., S. J. Cadaret, and C. Davis. 1999. Movement
responses to patch structure in experimental fractal land-
scapes. Ecology 80:1340–1353.
With, K. A., and A. W. King. 1999. Extinction thresholds
for species in fractal landscapes. Conservation Biology 13:
314–326.
Zabel, J., and T. Tscharntke. 1998. Does fragmentation of
Urtica
habitats affect phytophagous and predatory insects
differentially? Oecologia 116:419–425.