Landscape Ecol (2011) 26:999–1010

DOI 10.1007/s10980-011-9624-0

RESEARCH ARTICLE

Landscape composition influences roe deer habitat selection

at both home range and landscape scales
Nicolas Morellet • Bram Van Moorter • Bruno Cargnelutti •

Jean-Marc Angibault • Bruno Lourtet • Joël Merlet •

Sylvie Ladet • A. J. Mark Hewison

Received: 21 July 2010 / Accepted: 21 June 2011 / Published online: 7 July 2011
Ó Springer Science+Business Media B.V. 2011

Abstract Understanding how patterns of habitat substitutable habitat for woodlands by providing roe
selection vary in relation to landscape structure is deer with similar resources. We observed a functional
essential to predict ecological responses of species to response in the use of hedgerows, implying some
global change and inform management. We investi- degree of landscape complementation between hedge-
gated behavioural plasticity in habitat selection of roe rows and open habitats, which may in part compensate
deer (Capreolus capreolus) in relation to variable for lower woodland availability. We also expected
habitat availability across a heterogeneous agricultural selection for woodland to be highest at the wider spatial
landscape at the home range and landscape scales. As scale, especially when this habitat was limiting.
expected, woodland was heavily selected, but we However, our results did not support this hypothesis,
found no functional response for this habitat, i.e. no but rather indicated a marked influence of habitat
shift in habitat selection with changing habitat avail- composition, as both the availability and distribution of
ability, possibly due to the presence of hedgerows resources conditioned habitat selection. There was no
which were increasingly selected as woodlands were marked between-sex difference in the pattern of habitat
less abundant. Hedgerows may thus function as a selection at either scale or between seasons at the
landscape scale, however, within the home range,
selection did differ between seasons. We conclude that
N. Morellet (&)  B. Cargnelutti  J.-M. Angibault 
landscape structure has a marked impact on roe deer
B. Lourtet  J. Merlet  A. J. M. Hewison
Comportement et Ecologie de la Faune Sauvage, Institut habitat selection in agricultural landscapes through
National de la Recherche Agronomique, BP 52627, processes such as landscape complementation and
Castanet-Tolosan Cedex 31326, France supplementation.
e-mail: Nicolas.Morellet@toulouse.inra.fr

S. Ladet Keywords Behavioural plasticity  Functional

INRA, UMR 1201 DYNAFOR, BP 52627, response  Substitutable resource  Ungulates 
31326 Castanet-Tolosan, France Woodland fragmentation
S. Ladet
Université de Toulouse, UMR 1201 DYNAFOR,
31326 Castanet-Tolosan, France Introduction

B. Van Moorter
In modern agricultural landscapes, reduction and
Centre for Conservation Biology, Department of Biology,
NTNU, Realfagbygget, Høgskoleringen 5, 7491 fragmentation of natural habitats, such as woodland,
Trondheim, Norway has occurred over much of Europe and has been

123
1000 Landscape Ecol (2011) 26:999–1010

exacerbated by changing agricultural practices (Soth- resources in nearby patches of the same habitat, or by
erton 1998; Benton et al. 2003). Habitat destruction using substitutable resources in nearby patches of a
and fragmentation generates marked variation in different habitat (Dunning et al. 1992). Dunning et al.
habitat availability at various temporal and spatial (1992) also describe landscape complementation,
scales. The consequent variation in resource avail- when organisms require different habitat types to
ability may necessitate considerable plasticity in survive, for example when one habitat provides cover
resource selection of animals. Because habitat loss and a second provides forage, requiring the organism
and fragmentation alters the spatial distribution, the to travel frequently between patches. At the land-
quantity and the quality of resources available to scape scale, we expect the importance of landscape
individuals (Banks et al. 2007), it is important to supplementation and complementation to increase
understand how animals cope with these perturba- with increasing landscape heterogeneity, as organ-
tions in human dominated landscapes. isms make use of non-substitutable and substitutable
Animals require access to a range of resources and resources.
conditions (e.g. forage and refuge/cover) to realize In this paper we investigate habitat selection of roe
their fitness potential (Berryman and Hawkins 2006). deer (Capreolus capreolus) in a heterogeneous land-
The distribution of these resources and conditions is scape at home range and landscape scales, as the
expected to affect habitat use when it potentially conclusions concerning how animals exploit the
limits fitness. The limiting factor is the resource or available resources may depend on the spatial scale
condition that is available in quantities closest to the of analysis (Johnson 1980, Boyce 2006). Rettie and
minimum requirement of the organism (sensu Messier (2000) suggested that the most limiting
Liebig’s law of the minimum, see Odum 1959). factor (i.e. with the greatest potential to reduce
Animals should thus prioritize the use of habitats that fitness) should be selected for most strongly at the
provide best access to this limiting factor. We expect largest spatial scale, as avoiding the factors that are
individuals to alter their habitat selection when the most limiting at each successive scale will maximise
availability of resources and conditions changes, an individual’s fitness. Indeed, the influence of a
because these changes in availability can lead to limiting factor may persist over a broad range of
shifts in limiting factors (Hobbs and Hanley 1990). scales if its effects are not overcome at the coarsest
Indeed, Mysterud and Ims (1998) found that habitats scale at which it is encountered (Rettie and Messier
providing cover were less selected over those 2000). Roe deer is classically considered a species
providing forage when their availability increased. adapted to woodland living (Hewison et al. 1998),
When no habitat provides all necessary resources with woodland containing important resources like
simultaneously, such functional responses in habitat access to cover (Mysterud and Ostbye 1999) and
selection are likely to occur (Benhaiem et al. 2008, forage along edges with open areas (Tufto et al. 1996;
Godvik et al. 2009, Massé and Côté 2009). In this Saı̈d and Servanty 2005; Van Moorter et al. 2009).
situation, animals are expected to allocate time to Moreover, habitat selection of roe deer may be
different habitat types so as to obtain those resources influenced by perceived sources of risk such as
that satisfy their requirements. Therefore, changing predation, hunting and disturbance (Benhaiem et al.
habitat availability should lead to shifts in habitat 2008). Hence, we might expect woodland availability
selection. to be the most limiting factor. However, roe deer
The availability of a habitat can also influence appear to exhibit considerable plasticity in terms of
selection for other habitats providing the same habitat selection: following substantial habitat mod-
resource. When the availability of a habitat with a ification in Europe, roe deer populations have
given important resource decreases, we can expect expanded rapidly over recent decades (Andersen
selection for it to increase, but also selection for other et al. 1998a), colonising fragmented landscapes and
habitats with similar resources to increase. Such even open agricultural plains (Hewison et al. 2001).
habitats are considered substitutable habitats (in To date, our understanding of this plasticity in space
analogy with Tilman 1980). From a landscape use is limited.
perspective, landscape supplementation can occur if As roe deer is essentially a woodland species, we
organisms supplement their resource intake by using tested the first prediction that roe deer show a

123
Landscape Ecol (2011) 26:999–1010 1001

functional response regarding woodland, i.e. wood-

lands are more highly selected as they become less
available. Second, we expected selection for hedge-
rows to be affected by the availability of woodlands.
Hedgerows potentially provide access to similar
resources as woodlands and could prove crucial in
the roe deer’s adaptability to heterogeneous agricul-
tural landscapes. Therefore, we expected to observe
landscape supplementation between hedgerows and
woodlands, such that hedgerows are used to a greater
extent when woodland is scarcer. In addition, as the
hedgerows and agricultural fields of open landscapes
provide markedly different resources (i.e. non-sub-
stitutable), we expected to observe landscape com-
plementation (i.e. a functional response) such that
hedgerows are more highly selected when they are
more available. Finally, we investigated variation in
habitat selection across spatial scales, from the home
range to the landscape level. We expected selection Fig. 1 Map of the study site with the five main habitat types
and the three sectors
for woodland to be highest at the landscape scale,
especially when this habitat is very rare (i.e. most
limiting).
menziesii and oak. Meadows, cultivated fields and
hedgerows cover respectively 31.9, 36.2 and 5.8% of
the total area.
Methods
The roe deer is hunted by stalking during summer
(June–August) and drive hunts with dogs during
Study area
autumn–winter (September to February). Deer den-
sity was estimated at around 30 deer/100 ha in the
The study was carried out in an 11,000 ha heteroge-
central forest and around 10 deer/100 ha in the
neous agricultural landscape (N 43°130 , E 0°520 ) in
surrounding fragmented landscape (see Hewison
the Comminges region of south-west France (Fig. 1).
et al. 2007).
It is a hilly region (max. 380 m a.s.l.) which has
undergone substantial modification due to agricul-
tural intensification, with loss of hedgerows and Study population and data collection
copses, planting of new crops (corn, sorghum) and an
increase in average field size. This has resulted in a From 2002–2008, roe deer were caught during winter
mixed landscape of open fields and small woodland (from 16th November to 27th March) using drives of
patches (mean ± SD: 2.9 ± 7.6 ha), with a central 30–100 beaters and 4 km of long-nets at one of 10
larger forest of 672 ha. The primary land use is for capture sites. Each roe deer was sexed and aged
sheep and cattle grazing, with agricultural crops on (juveniles \1 year, adults [18 months). Then, 134
the increase. The human population is present deer were equipped with GPS collars (Lotek 3300)
throughout the site, in small villages and farms and released on site. Collars were programmed to
distributed along the extensive road network. The obtain a GPS fix every 4 h (first two winters) or every
climate is oceanic, with an average annual temper- 6 h (following winters) for 1 year. We performed
ature of 11–12°C and 800 mm precipitation. differential correction to improve fix accuracy (Adra-
The landscape is characterised by woodland patches dos et al. 2002), as only 50% of uncorrected fixes
(12.6% of the area) dominated by oak Quercus spp., were located within 14 m of their true position in
and a central forest (5.8% of the area) containing a our study area (Cargnelutti et al. 2007). We removed
mixed species habitat of Douglas-fir Pseudotsuga the first 8 days of monitoring, when behavioural

123
1002 Landscape Ecol (2011) 26:999–1010

13.0% (22.2)
alterations due to capture and handling may be

5.1% (12.0)
3.1% (6.5)
2.7% (2.3)
2.4 (17.5)
pronounced (Morellet et al. 2009), prior to habitat

0.0 (0.0)
0.0 (0.0)
0.1 (0.4)
0.1 (0.4)
0.0 (0.0)
Fields
selection analyses.

123
Landscape composition

71.7% (13.9)
11.0% (18.4)
We considered five habitat types: woodland, mead-

2.5% (0.0)
2.5% (0.0)
Meadows

0.0 (0.1)
0.0 (0.0)
1.7 (2.4)
6.1 (5.8)
0.0 (0.0)
0.0 (0.0)
ows, cultivated fields, hedgerows and other (all other

122
habitats combined, mainly roads, houses, trails and
water). The availability of these habitats was assessed

Table 1 Mean forage biomass (g) and percent cover (%) with standard deviation in brackets in the four habitat types and for two seasons
annually. Using aerial photographs, homogeneous

Autumn–Winter (November 2005)

61.4% (25.2)
36.0% (26.0)
21.4% (18.6)
21.3% (23.1)
polygons (i.e. plots containing one habitat type only)

Hedgerows
in the landscape were digitized manually in a

1.9 (2.3)
1.1 (4.1)
0.8 (1.5)
2.9 (4.6)
0.1 (0.7)
0.0 (0.0)
Geographic Information System (ArcView GIS 3.3,

76
ESRI 2002). From 2004, each polygon was assigned
to a habitat type annually by field observations during
summer. Because equivalent information was not

46.1% (30.8)
19.8% (25.1)
13.6% (20.8)
8.6% (15.0)
Woodlands
available for 2003, we used the 2004 habitat infor-

1.3 (1.3)
1.7 (4.8)
0.3 (0.6)
0.5 (1.2)
0.0 (0.1)
0.0 (0.0)
mation. Given that in the present analyses we

126
considered relatively stable habitat types, we do not
believe that this generated significant error.
To compare resources in terms of forage and

42.7% (34.6)
40.0% (35.3)
27.5% (30.9)
shelter among habitat types, we collected vegetation

32.1 (31.9)

3.9% (8.0)
0.0 (0.1)
0.0 (0.0)
0.0 (0.3)
0.2 (1.1)
0.0 (0.0)
samples and measured vegetation cover during May
Fields

and November 2005 (Table 1). Measurements were

109
performed on regularly spaced sampling plots of
25 9 25 cm occurring along the perimeter of 26

55.7% (32.0)
51.6% (33.4)
32.2% (32.6)
16.7% (28.7)
squares (700 m per side) distributed across the study
24.5 (18.4)
Meadows

area, concentrating mostly on areas where marked roe

0.0 (0.0)
0.0 (0.0)
1.9 (3.0)

0.0 (0.0)
0.0 (0.0)

deer were located (see Abbas et al. 2011). Plots were

104
spaced at 150 m intervals along each square’s
perimeter, but less if a new habitat was encountered
(except the habitat type ‘‘other’’) to ensure that all
73.1% (10.9)
67.7% (18.1)
58.1% (24.2)
48.8% (28.4)
Hedgerows

habitats were adequately sampled (minimum of 18

Spring-Summer (May 2005)

9.4 (9.5)
0.0 (0.0)
2.9 (3.9)
4.6 (4.5)
0.0 (0.0)
0.0 (0.0)

plots per square). We collected all green vegetation

from ground level to 120 cm on a 25 9 25 cm square
62

(Said et al. 2005). We identified plant species (or

families) and weighed them (balance Mettler
52.8% (26.0)
38.1% (25.9)
22.5% (21.9)
14.3% (18.0)

PM480 ± 0.01 g) after drying at 60°C for at least

Woodlands

3 days. We then calculated average forage biomass

3.0 (3.2)
0.0 (0.0)
0.7 (1.9)
0.5 (2.1)
0.0 (0.2)
0.0 (0.0)

per category of vegetation as follows: ligneous,

165

herbaceous, grass, fruits, crops and other for each

habitat type. On each plot, percent cover of all edible
vegetation was assessed visually within four height
Cover 120–200 cm
Cover 50–120 cm
Cover 20–50 cm

strata (0–20, 20–50, 50–120 and 120–200 cm). For

Cover 0–20 cm

each stratum, percent cover was described in three

Habitat type

Sample size
Herbaceous
Ligneous

classes (\5, 5–50 and [50%) as the projection of the

Crops
Fruits

Other
Grass

available vegetation on the ground within a plot of

2-m radius. To calculate the average percent cover

123
Landscape Ecol (2011) 26:999–1010 1003

per habitat type, we used the median of each class home range and landscape scales. The selection ratio
(i.e. 0.025, 0.275 and 0.75). for habitat h corresponds to the ratio between the
The resources available, in terms of forage and proportion of habitat type h used and the proportion
shelter, differ widely between habitat types and of habitat type h available. Selection ratios were
seasons (Table 1). Forage biomass is very high in ln-transformed, adding 0.001 to allow calculation of
cultivated fields and meadows during spring-summer ln-SRs for non-used habitat types (in the following,
and very low in autumn–winter. The forage resources ln-SR refers to these ln-transformed values). A ln-SR
are more diversified in hedgerows and woodlands of habitat type h which includes zero within its
than in other habitat types. Cultivated fields and confidence intervals indicates a use of habitat type
meadows provide little hiding cover during autumn– h which is proportional to availability; a negative
winter, whereas hedgerows and woodlands offer ln-SR value indicates avoidance of the habitat type h,
cover throughout the year. Even though there were while a positive value indicates selection. At the
some differences between woodlands and hedgerows landscape scale, use was defined as the habitat
(greater biomass of ligneous and herbaceous species, composition of individual home ranges and avail-
grasses and better cover in hedgerows than in ability as the habitat composition of the correspond-
woodlands, particularly during spring–summer), ing landscape sector. At the home range scale, use
these habitats offer very similar resources in terms was assessed as the proportion of locations within
of both forage and shelter, with lower inter-seasonal each habitat type and availability as the composition
variability compared to meadows and cultivated of the home range. Home ranges were calculated
fields. Although we measured forage and shelter in using the fixed kernel home range method at 95%
2005 only and absolute values of these parameters with an ad hoc factor (Worton 1989). To perform
likely vary across years, relative differences among analyses at the home range scale, peripheral locations
habitat types should remain consistent through time. that fell outside the defined home range were
removed (4.2% of the locations).
Statistical analysis We analysed roe deer ln-SRs in relation to season
in order to control for changes in roe deer social
The 10 capture sites were grouped into three land- organisation and variation within habitats over the
scape units based on landscape structure in terms of course of the year. Hence, we defined a spring-
woodland extent and the relative proportions of summer season (March–September) when male roe
meadows and cultivated fields (see Hewison et al. deer are territorial, and an autumn–winter season
2009). Thus, we identified a forest block (sector 1: (October–February) as the remainder of the year. We
with 57.8% woodland, 22.1% meadows, 11.3% included sex as a factor, as only bucks are considered
cultivated fields, 1.4% hedgerows and 7.4% other), territorial (Hewison et al. 1998) and energetic
a partially wooded area (sector 2: with 27.7% requirements associated with reproduction differ
woodland, 41.0% meadows, 22.3% cultivated fields, between the sexes (Sempéré et al. 1998). Finally, as
3.6% hedgerows and 5.4% other) and an open our study site contains a gradient of landscape
agricultural area with highly fragmented woodland structures so that the local availability of the different
(sector 3: with 11.9% woodland, 32.4% meadows, habitat types varies markedly between sectors, we
41.1% cultivated fields, 6.6% hedgerows and 8.0% included the sector of the individual’s home range as
other). a factor.
In this study, we excluded dispersing individuals We used general linear mixed models to investi-
(that left their original home range at 1 year-old). As gate variation in ln-SRs for the five habitat types,
only five individuals were monitored for more than including sex (two modalities), season (two modal-
1 year, we retained data from a 1-year cycle for each ities), and sector (three modalities) in the model, with
individual, selecting the year with the most available the roe deer’s identity as a random factor to control
data. Thus, we considered habitat selection of 108 roe for repeated observations (across habitat types and
deer: 33 in sector 1, 18 in sector 2 and 57 in sector 3. seasons) per individual (see Herfindal et al. 2009).
We analysed habitat selection using selection ratios We expected differences in selection for habitats
(ln-SR, Manly et al. 2002; Herfindal et al. 2009) at among sectors and seasons, and potentially also

123
1004 Landscape Ecol (2011) 26:999–1010

sexes; therefore, we considered those models which use of one habitat was conditional on the proportional
incorporated a difference in habitat selection between availability of that habitat in the home range of a given
sexes (two-way interaction between sex and habitat individual, we used the approach proposed by Myste-
type), among seasons (two-way interaction between rud and Ims (1998), regressing proportional use p(u)
season and habitat type) and among sectors (two-way against proportional availability p(a), and logit-trans-
interaction between sector and habitat type), and all forming both the predictor and predicted variables to
possible combinations of these hypotheses. We also make the expectation linear on the logit scale, using the
considered the model describing a common pattern of equation logit p(u) = a?b log(p(a)/[1 - p(a)]). We also
habitat selection among sexes, seasons and sectors considered sex and season in these models, but not
(model containing the factor habitat type only). Finally sector as this analysis already implicitly controlled for
we considered the constant model (i.e. no habitat variation among individuals in habitat availability
selection). We did not include higher order interac- across the landscape. We used generalized linear
tions, as we had no specific hypotheses to test at this mixed models to investigate whether variation in
level. The models were fitted with the lme function in proportional use of woodlands or hedgerows was
the nlme package (Pinheiro et al. 2009) in R for related to the logit of proportional availability of that
windows version 2.10.0 (R Development Core Team habitat, including sex and season and with the roe
2009). We considered nine models (Tables 2, 3) and deer’s identity as a random factor to control for
used the second order Akaike’s Information Criterion repeated observations per individual. The models were
(AICc; Burnham and Anderson 1998) and the Akaike fitted with the glmmPQL function in the MASS
weights to select the model with the most support. To package (Venables and Ripley 2002) in R.
interpret and visualise the effects incorporated in the In the above analysis, we considered woodlands and
retained models, we plotted the parameter estimates hedgerows separately. However, use of each of these
and their associated 95% confidence intervals by habitat types is likely dependent upon the availability
posterior re-sampling (n = 10,000) of the parameter of the other. We first used partial correlations to test
estimates from the fitted model in the Zelig package whether the use of hedgerows was affected by the
(Bailey and Alimadhi 2007) in R. availability of hedgerows (i.e. a functional response in
At the home range scale, we tested for a functional hedgerow use), whilst controlling for woodland avail-
response in habitat use when considering two con- ability. Partial correlation analysis is aimed at finding
trasting habitat types. We first opposed woodland relationships between two variables after removing the
habitat against all other habitat types combined and effect of a third (Whittaker 1990). Second, we tested
second, we opposed hedgerows against all other habitat whether the use of hedgerows was affected by the
types combined. To investigate whether proportional availability of woodlands, controlling for hedgerow

Table 2 Candidate general linear mixed models to investigate in interaction with habitat type, with the roe deer’s identity as a
habitat selection (variation in ln-SRs among the five habitat random factor (N = 108)
types) at the landscape scale, including sex, season and sector
Models AICc (K) DAICc (wi)

Habitat ? sector ? habitat 9 sector 2 914.1 (17) 0.0 (0.780)

Habitat ? sector ? season ? habitat 9 sector ? habitat 9 season 2 917.3 (22) 3.2 (0.160)
Habitat ? sector ? sex ? habitat 9 sector ? habitat 9 sex 2 919.5 (22) 5.4 (0.050)
Habitat ? sector ? season ? sex ? habitat 9 sector ? habitat 9 season ? habitat 9 sex 2 922.7 (27) 8.6 (0.010)
Habitat ? sex ? habitat 9 sex 3 498.2 (12) 584.1 (0.000)
Habitat ? season ? sex ? habitat 9 season ? habitat 9 sex 3 504.1 (17) 590.0 (0.000)
Habitat 3 508.0 (7) 593.9 (0.000)
Habitat ? season ? habitat 9 season 3 513.9 (12) 599.8 (0.000)
Constant 3 641.4 (3) 727.3 (0.000)
Model ranking is based on differences in the corrected Akaike’s Information Criterion (DAICc) and Akaike weights (wi). K refers to
the number of parameters

123
Landscape Ecol (2011) 26:999–1010 1005

Table 3 Candidate general linear mixed models for habitat interaction with habitat type, with the roe deer’s identity as a
selection (variation in ln-SRs among the five habitat types) at random factor (N = 108)
the home range scale, including sex, season and sector in
Models AICc (K) DAICc (wi)

Habitat ? sector ? season ? habitat 9 sector ? habitat 9 season 1 984.0 (22) 0.0 (0.990)
Habitat ? sector ? season ? sex ? habitat 9 sector ? habitat 9 season ? habitat 9 sex 1 993.6 (27) 9.5 (0.010)
Habitat ? sector ? habitat 9 sector 2 000.1 (17) 16.1 (0.000)
Habitat ? sector ? sex ? habitat 9 sector ? habitat 9 sex 2 009.5 (22) 25.5 (0.000)
Habitat ? season ? habitat 9 season 2 039.2 (12) 55.2 (0.000)
Habitat ? season ? sex ? habitat 9 season ? habitat 9 sex 2 046.2 (17) 62.2 (0.000)
Habitat 2 057.7 (7) 73.6 (0.000)
Habitat ? sex ? habitat 9 sex 2 064.4 (12) 80.4 (0.000)
Constant 2 265.7 (3) 281.7 (0.000)
Model ranking is based on differences in the corrected Akaike’s Information Criterion (DAICc) and Akaike weights (wi). K refers to
the number of parameters

availability. We performed linear models of the models containing the 2 other two-way interaction
residuals of the dependent variable Y against the terms, between habitat type and sector and between
residuals of the independent variable X, after first habitat type and season (wi = 0.16). Hence, at the
removing the effect of variable Z. We calculated the landscape scale, the model with the greatest support
residuals of variable Y (and then the residuals of supported the hypothesis of differences in habitat
variable X) from a generalized linear mixed model of selection among sectors (Fig. 2). In particular, habitat
the variable Y (and then variable X) against the selection was very pronounced in the forest sector,
variable Z, including individual roe deer identity as a where the roe deer showed a strong selection for
random factor. For instance, we generated a linear woodlands and an avoidance of all other habitat
model of the residuals of proportional use of hedge- types. Roe deer habitat selection in the partially
rows (hedgerows(u1)) against the residuals of propor- wooded and open agricultural sectors was similar,
tional availability of hedgerows (hedgerows(a)), after with the exception of hedgerows that were avoided in
removing the effect of woodland availability. We
defined hedgerows(u1) as the residuals of a generalized
1

linear mixed model of the proportional use of hedge- _

_
rows against the logit of the proportional availability of _ _ _
_ _ _
_
_ _
woodlands (including individual identity as a random
0

_ _ _ _
log(selection ratio)

_ _ _
factor). In a similar way, we defined hedgerows(a) as _ _
_
the residuals of a generalized linear mixed model of the
-1

_ _
proportional availability of hedgerows against the logit _
_
of the proportional availability of woodlands (includ- _
-2

ing individual identity as a random factor). _

_ _
-3

_ Sector 1
Results Sector 2
-4

Sector 3

Habitat selection at the landscape scale Woodlands Fields Hedgerows Meadows Other

Habitat types
In the analysis of roe deer ln-SRs at the landscape
Fig. 2 Habitat selection ratios of roe deer at the landscape
scale (Table 2), the best model contained the two-
scale for the three sectors. The dashed line represents a level of
way interaction between habitat type and sector use that is proportional to availability and error bars represents
(wi = 0.78,), which compared favourably with the 95% confidence intervals by posterior re-sampling

123
1006 Landscape Ecol (2011) 26:999–1010

the partially wooded sector and used proportionally to avoided cultivated fields and the ‘‘other’’ habitat
availability in the open agricultural sector. Habitat category in both seasons. Habitat selection of roe deer
selection was indeed less pronounced in these sectors, in the partially wooded sector was intermediate: no
with roe deer using woodlands, cultivated fields and habitat type was significantly selected for in this
meadows proportionally to availability while avoid- sector; woodlands, hedgerows, meadows and culti-
ing the ‘‘other’’ habitat type. vated fields were all used in proportion to availability
during spring-summer, while cultivated fields and the
Habitat selection at the home range scale ‘‘other’’ habitat category were avoided during
autumn–winter.
In the analysis of roe deer ln-SRs at the home range
scale, the best model contained 2 two-way interaction The functional response in habitat use
terms, between habitat type and sector and between of woodlands and hedgerows
habitat type and season (Table 3). There was no
substantial support for the other models. Hence, at the For the logistic regression of the proportional use of
home range scale, the best model supported the woodlands against the logit of woodland availability
hypothesis of differences in habitat selection among (logitwood), none of the interactions were statisti-
sectors and among seasons (Fig. 3). In the forest cally significant (logitwood 9 season 9 sex: t =
sector, in both seasons, woodlands were used pro- -1.35, df = 102, P = 0.178; sex 9 season: t =
portionally to availability, whereas all other habitats 0.36, df = 103, P = 0.717; logitwood 9 season:
were avoided. In fact, as the home ranges in the forest t = 0.39, df = 103, P = 0.697; logitwood 9 sex:
sector were mainly made up of woodland, the t = -0.01, df = 103, P = 0.992) nor was the main
observed avoidance of the other habitats was likely effect of sex (t = 0.11, df = 106, P = 0.909). How-
due to their relative scarcity within the home ranges ever, the use of woodlands increased with its
and should be interpreted with caution. In the open availability (logitwood: t = 38.75, df = 106, P \
agricultural sector, roe deer strongly selected wood- 0.0001; Fig. 4), with an intercept of a = 0.41 ± 0.06
lands and hedgerows (the latter particularly during and a slope of b = 0.98 ± 0.03. This situation where
spring-summer), but used meadows proportionally to a [ 0 and b C 1 implies that the proportion of
availability. In contrast, the roe deer of this sector woodland habitat used was always greater than the
1.0
1.0

AW season
SS season
0.8

Sector 1
_
Use of woodland (%)
0.5

_ _ Sector 2
_ _
_
log(selection ratio)

Sector 3
_ _ _ _ _
_ _ _ _ _ _ _
0.6
0.0

_ _ _ _
_ _ _ _ _ _
_ _ __ _
_ _
_ _ _ _ _
0.4
-0.5

_ _ _
_ _ _
__ _ _
_ _
_ _ _
-1.0

0.2

_ _ _ AW season
_
SS season
-1.5

0.0

Woodlands Fields Hedgerows Meadows Other 0.0 0.2 0.4 0.6 0.8 1.0
Habitat types Availability of woodland (%)

Fig. 3 Habitat selection ratios of roe deer at the home range Fig. 4 Logistic regression of the use of woodland against the
scale for the three sectors and the two seasons (SS spring– logit proportion of the availability of that habitat in the home
summer, AW autumn–winter). The dashed line represents a range for the two seasons (SS spring–summer, AW autumn–
level of use that is proportional to availability and error bars winter). The dotted line represents a level of use that is equal to
represents 95% confidence intervals by posterior re-sampling availability

123
Landscape Ecol (2011) 26:999–1010 1007

proportion available (Mysterud and Ims 1998). Fur-

0.30
thermore, roe deer showed a more marked selection Female NT
for woodland during autumn–winter than during

0.25
Female T

Use of hedgerows (%)

spring-summer (season: t = -2.53, df = 106,
P = 0.013) (Fig. 4). Male NT

0.20
For the logistic regression of the proportional use of Male T

0.15
hedgerows against the logit of hedgerow availability
(logithedge), the higher order interaction did not

0.10
significantly affect use of hedgerows (logithedge 9
season 9 sex: t = 1.20, df = 81, P = 0.234) nor did

0.05
the two-way interaction between hedgerow availabil-
ity and sex (logithedge 9 sex: t = 1.09, df = 82,

0.00
P = 0.277). However, the use of hedgerows was
dependent upon its availability, but in a different way 0.00 0.05 0.10 0.15
between seasons (logithedge 9 season: t = -2.51, Availability of hedgerows (%)
df = 83, P = 0.014). The different effect of season
between the sexes on the use of hedgerows (sea- Fig. 5 Logistic regression of the use of hedgerows against the
logit proportion of the availability of that habitat in the home
son 9 sex: t = -2.58, df = 83, P = 0.012) became a range for the two seasons (SS spring–summer, AW autumn–
non-significant trend (t = -1.83, df = 82, P = winter, with normal lines and bold lines respectively) and
0.071) after the removal of one male exhibiting an sexes. The dotted line represents a level of use that is equal to
exceptional high use of hedgerows (30%). The inter- availability
cept (aNT = 0.88 ± 0.24 for autumn–winter and
aT = 0.47 ± 0.17 for spring-summer) and the slope availability of woodlands (slope = -0.39 ± 0.06,
(bNT = 1.26 ± 0.09 for autumn–winter and bT = t = -6.53, df = 181, P \ 0.0001, R2 = 0.19), while
1.04 ± 0.06 for spring-summer) of the logit of hedge- controlling for hedgerow availability.
row availability corresponds to the situation with a [ 0
and b [ 1 (b C 1 during spring-summer), implying
that the strength of selection for hedgerows increased Discussion
with the availability of hedgerows, at least during the
autumn–winter season (Mysterud and Ims 1998). We In this paper, we investigated how roe deer select
observed some variation in this general pattern, as their home range within the landscape and their
males tended to prefer hedgerows more strongly than patterns of habitat selection within their home range
females during autumn–winter, whereas the selection in relation to the resources provided by the five main
for hedgerows was the same for males and females habitat types of the landscape (Table 1). In fact, these
during spring-summer (Fig. 5). habitat types provided contrasting resources in terms
of forage and shelter among seasons, although
Substitutable habitats hedgerows and woodlands were favourable habitats
throughout the year and provided similar resources.
First, we tested whether the use of hedgerows was In contrast, meadows and cultivated fields were more
affected by the availability of hedgerows, while favourable during the spring-summer season, whereas
controlling for woodland availability. The propor- cultivated fields provided almost no resources during
tional use of hedgerows increased linearly with the autumn–winter (Abbas et al. 2011). The comparison
proportional availability of hedgerows (slope = of roe deer habitat selection patterns in three sectors
0.85 ± 0.05, t = 15.91, df = 181, P \ 0.0001, R2 = differing in landscape composition allowed us to
0.58), while controlling for woodland availability. investigate the functional response in habitat selec-
Second, we tested whether hedgerows could be con- tion at two spatial scales. Indeed, our study site
sidered as a substitute for woodlands, while controlling included a marked gradient in terms of woodland and
for hedgerow availability. The proportional use of hedgerow availability, with a decrease in the propor-
hedgerows decreased linearly with the proportional tion of woodland from sector 1 to sector 3, but an

123
1008 Landscape Ecol (2011) 26:999–1010

increase in the proportion of hedgerows. As expected available. Note that this observed lack of support for
and shown elsewhere, roe deer in the Comminges Rettie and Messier’s (2000) hypothesis supposes that
strongly selected woodland (Hewison et al. 2001) the absolute requirements of roe deer for woodland are
which provides important resources (i.e. cover and not inferior to the minimum available in the studied
access to forage at the forest edge). We found no landscapes (i.e. less than 11.9%). However, this seems
differences in these patterns between sexes, which unlikely, as roe deer home range size increases
was expected for this species with low sexual size considerably as the availability of woodland in the
dimorphism (Andersen et al. 1998b) and similar landscape decreases across the study area, suggesting
feeding habits between sexes (Cransac et al. 2001). the need for substantial wooded areas within their
Despite the importance of woodlands, we found no home ranges (Cargnelutti et al. 2002).
functional response for this habitat in roe deer, i.e. Van Moorter (2008, see also Gaillard et al. 2010)
habitat selection for woodland remained constant has shown that it is not only the effect of a factor on
across the landscape (Mysterud and Ims 1998). The an animal’s fitness that is important, but also the
limited behavioural response to variation in woodland factor’s distribution. If animals can make their home
availability can be explained by the presence of range coincide with an important habitat, as was the
hedgerows, which were selected more strongly when case for roe deer in sector 1 whose home range
woodlands were scarce. Indeed, roe deer showed a mostly fell inside woodland, we should expect no
functional response towards these hedgerows, with a selection for that habitat within the home range (see
more marked selection for this habitat as its avail- also Kie et al. 2002). In contrast, where the low
ability increased. However, when controlling for availability of woodland does not allow animals to
woodland availability, selection for hedgerows dif- match their home range with this habitat, the home
fered somewhat, as the roe deer used hedgerows, but range would necessarily include other habitats,
proportionately less than availability. This was due to resulting in a lower level of selection for that habitat
the dependence of the use of hedgerows on the at the landscape scale (as was the case in sector 3).
availability of woodlands, with a decrease in the However, in this situation, these animals would then
selection for hedgerows with increasing woodland preferentially use that habitat (in our case, wood-
availability. Therefore, hedgerows act as a substitut- lands, but also hedgerows) within the home range.
able habitat for woodlands when the availability of Our results support these hypothesized effects of
this latter habitat type decreases, providing roe deer habitat distribution on hierarchical habitat selection.
with similar resources to woodlands (Table 1). From There was no marked difference among seasons in
a landscape perspective, we found a marked pattern the pattern of habitat selection at the landscape scale,
of landscape supplementation between woodland and however, within the home range, selection did differ
hedgerows, with the importance of hedgerows among seasons. Not surprisingly, the habitat with
increasing with the openness of the landscape. In strongest seasonal variation in selection was agricul-
addition, we observed a functional response in the use tural fields, as they also showed the strongest
of hedgerows, implying a degree of landscape seasonal variation in terms of the resources they
complementation between hedgerows and open hab- provide, being largely barren during the winter, but
itats, which may in part compensate for the loss of rich in crops during the summer. We have already
woodland availability. shown that levels of nitrogen and phosphorous were
Our results did not support Rettie and Messier’s higher in deer faecal samples in the more open
(2000) hypothesis that the more limiting fac- sectors of the landscape compared to the forest
tor(s) should be selected for at the wider spatial scale environment, suggesting the use of high-quality
(i.e. landscape scale). Rather, we found the opposite; at feeding habitats provided by the cultivated agricul-
the landscape scale, animals in sector 1 (with 58% tural plain (Hewison et al. 2009). Hence, this pattern
available woodland) showed stronger selection for of habitat selection, more oriented toward cultivated
woodland than animals in sector 2 and 3 (with only 28 fields during spring-summer, could be linked to the
and 12% available woodland, respectively). However, search for nutrient rich foods. However, roe deer are
at the home range scale, we observed a stronger actively hunted and are subject to a variety of sources
selection for woodland when this habitat was less of disturbance or risk which likely influence animal

123
Landscape Ecol (2011) 26:999–1010 1009

behaviour (Brown et al. 1999). Indeed, in the References

presence of risk, prey are expected to increase
vigilance at the cost of foraging or resting, and to Abbas F, Morellet N, Hewison AJM, Merlet J-l, Cargnelutti B,
Lourtet B, Angibault J-M, Daufresne T, Aulagnier S,
select habitats where they feel more secure but where Verheyden H (2011) Landscape fragmentation and plas-
forage may be poorer (Benhaiem et al. 2008). ticity in resource use generate spatial variation of diet
Woodlands and hedgerows provide shelter all year composition and quality in a generalist herbivore. Oeco-
round, while some cultivated fields may do so during logia. doi:10.1007/s00442-011-1994-0
Adrados C, Girard I, Gendner J-P, Janeau G (2002) GPS
at least part of the spring-summer period when crops location accuracy improvement due to selective avail-
are of sufficient height, but this is not so during the ability removal. C R Biol 325:1–6
rest of the year. Hence, the seasonal variations in Andersen R, Duncan P, Linnell JDC (eds) (1998a) The Euro-
habitat selection that we observed were likely linked pean roe deer: the biology of success. Scandinavian
University, Oslo
to variation in availability of both food resources and Andersen R, Gaillard J-M, Liberg O, San Jose C (1998b)
shelter in the different habitat types (Mysterud et al. Variation in life-history parameters. In: Andersen R,
1999). Duncan P, Linnell JDC (eds) The European roe deer: the
To conclude, agricultural landscapes have under- biology of success. Scandinavian University Press, Oslo,
pp 189–219
gone substantial modification, notably the reduction Bailey D, Alimadhi F (2007) ls.mixed: mixed effects linear
and fragmentation of woodland and hedgerows. model. In: Imai K, King G, Lau O (eds) Zelig: everyone’s
Agricultural practices, in particular crop rotation statistical software. Available in http://gking.harvard.edu/
and conversion of pasture to cultivated fields, may zelig
Banks SC, Piggott MP, Stow AJ, Taylor AC (2007) Sex and
also lead to pronounced modifications in landscape sociality in a disconnected world: a review of the impacts
structure, inducing modifications in the spatial distri- of habitat fragmentation on animal social interactions.
bution of habitat types across the landscape. From a Can J Zool 85:1065–1079
management perspective, it is becoming increasingly Benhaiem S, Delon M, Lourtet B, Cargnelutti B, Aulagnier S,
Hewison AJM, Morellet N, Verheyden H (2008) Hunting
clear that there is a need to maintain semi-natural increases vigilance levels in roe deer and modifies feeding
habitats such as woodlands and hedgerows as reser- site selection. Anim Behav 76:611–618
voirs of biodiversity in modern agricultural land- Benton TG, Vickery JA, Wilson JD (2003) Farmland biodi-
scapes (Burel et al. 2004). We have shown that, versity: is habitat heterogeneity the key? Trends Ecol Evol
18:182–188
although essentially a woodland species, roe deer Berryman AA, Hawkins BA (2006) The refuge as an inte-
shows pronounced behavioural plasticity in habitat grating concept in ecology and evolution. Oikos 115:
selection across a gradient of landscape openness, for 192–196
example, substituting woodlands by hedgerows in the Boyce MS (2006) Scale for resource selection functions.
Divers Distrib 12:269–276
most open areas of the landscape. Roe deer occur in Brown JS, Laundré JW, Gurung M (1999) The ecology of fear:
almost all natural habitats of Europe, showing optimal foraging, game theory, and trophic interactions.
substantial variation in social and spatial organisation J Mammal 80:385–399
in response to environmental heterogeneity (Ander- Burel F, Butet A, Delettre YR, de la Pena NM (2004) Differ-
ential response of selected taxa to landscape context and
sen et al. 1998a, b). Furthermore, their populations agricultural intensification. Landsc Urban Plan 67:
have expanded rapidly over recent decades and it 195–204
may be that these high levels of social and behav- Cargnelutti B, Reby D, Desneux L, Angibault J-M, Joachim J,
ioural plasticity (Hewison et al. 1998) are key Hewison AJM (2002) Space use by roe deer in a frag-
mented landscape some preliminary results. Rev Ecol
attributes contributing to their success in adapting (Terre Vie) 57:29–37
to a changing environment within most man-modified Cargnelutti B, Coulon A, Hewison AJM, Goulard M, Angi-
landscapes. bault J-M, Morellet N (2007) Testing GPS performance
for wildlife monitoring using mobile collars with known
Acknowledgments We would like to thank the local hunting reference points. J Wildl Manag 71:1380–1387
associations with the Fédération Départementale des Chasseurs Cransac N, Cibien C, Angibault J-M, Morellet N, Vincent J-P,
de la Haute Garonne for allowing us to work in the Hewison AJM (2001) Variations saisonnières du régime
Comminges, as well as numerous co-workers and volunteers alimentaire du chevreuil (Capreolus capreolus) selon le
for their assistance and two anonymous referees for their sexe en milieu forestier à forte densité (forêt domaniale de
constructive criticisms on an earlier draft of this work. Dourdan). Mammalia 65:1–12

123
1010 Landscape Ecol (2011) 26:999–1010

Dunning JB, Danielson BJ, Pulliam HR (1992) Ecological Mysterud A, Ims RA (1998) Functional responses in habitat
processes that affect populations in complex landscapes. use: availability influences relative use in trade-off situa-
Oikos 65:169–175 tions. Ecology 79:1435–1441
ESRI (2002) Using ArcView GIS version 3.3: the geographic Mysterud A, Ostbye E (1999) Cover as a habitat element for
information system for everyone. Environmental Systems temperate ungulates: effects on habitat selection and
Research Institute, Inc., Redlands demography. Wildl Soc Bull 27:385–394
Gaillard J-M, Hebblewhite M, Loison A, Fuller M, Powell R, Mysterud A, Larsen PK, Ims RA, Ostbye E (1999) Habitat
Basille M, Van Moorter B (2010) Habitat-performance selection by roe deer and sheep: does habitat ranking
relationships: finding the right metric at a given spatial reflect resource availability? Can J Zool 77:776–783
scale. Philos T R Soc Lond B 365:2255–2265 Odum EP (1959) Fundamentals of ecology. WB Saunders
Godvik IMR, Loe LE, Vik JO, Veiberg V, Langvatn R, Mysterud Company, Philadelphia
A (2009) Temporal scales, trade-offs, and functional Pinheiro J, Bates D, DebRoy S, Sarkar D, The R Core Team
responses in red deer habitat selection. Ecology 90:699–710 (2009) nlme: Linear and nonlinear mixed effects models.
Herfindal I, Tremblay J-P, Hansen BB, Solberg EJ, Heim M, R package version 3, pp 1–96
Sæther B-E (2009) Scale dependency and functional Rettie WJ, Messier F (2000) Hierarchical habitat selection by
response in moose habitat selection. Ecography 32:849–859 woodland caribou: its relationship to limiting factors.
Hewison AJM, Vincent JP, Reby D (1998) Social organisation Ecography 23:466–478
of European roe deer. In: Andersen R, Duncan P, Linnell Saı̈d S, Servanty S (2005) The influence of landscape structure
JDC (eds) The European roe deer: the biology of success. on female roe deer home-range size. Landscape Ecol
Scandinavian University Press, Oslo, pp 189–219 20:1003–1012
Hewison AJM, Vincent J-P, Joachim J, Angibault J-M, Car- Saı̈d S, Pellerin M, Guillon N, Debias F, Fritz H (2005)
gnelutti B, Cibien C (2001) The effects of woodland Assessment of forage availability in ecological studies.
fragmentation and human activity on roe deer distribution Eur J Wildl Res 51:242–247
in agricultural landscapes. Can J Zool 79:679–689 Sempéré AJ, Mauget R, Mauget C (1998) Reproductive
Hewison AJM, Angibault J-M, Cargnelutti B, Coulon A, physiology of roe deer. In: Andersen R, Duncan P, Linnell
Rames J-L, Serrano E, Verheyden H, Morellet N (2007) JDC (eds) The European roe deer: the biology of success.
Using radio-tracking and direct observation to estimate Scandinavian University Press, Oslo, pp 161–188
roe deer density in a fragmented landscape: a pilot study. Sotherton NW (1998) Land use changes and the decline of
Wildl Biol 13:313–320 farmland wildlife: an appraisal of the set-aside approach.
Hewison AJM, Morellet N, Verheyden H, Daufresne T, Ang- Biol Conserv 83:259–268
ibault J-M, Cargnelutti B, Merlet J, Picot D, Rames J-L, R Development Core Team (2009) R: a language and envi-
Joachim J, Lourtet B, Serrano E, Bideau E, Cebe N (2009) ronment for statistical computing. R Foundation for Sta-
Landscape fragmentation influences winter body mass of tistical Computing. Vienna, Austria. ISBN 3-900051-07-
roe deer. Ecography 32:1062–1070 0, http://www.R-project.org
Hobbs NT, Hanley TA (1990) Habitat evaluation: do use/ Tilman D (1980) Resources: a graphical-mechanistic approach
availability data reflect carrying capacity. J Wildl Manag to competition and predation. Am Nat 116:362–393
54:515–522 Tufto J, Andersen R, Linnell JDC (1996) Habitat use and
Johnson DH (1980) The comparison of usage and availability ecological correlates of home range size in a small cervid:
measurements for evaluating resource preference. Ecol- the roe deer. J Anim Ecol 65:715–724
ogy 61:65–71 Van Moorter B (2008) Habitat selection at different spatial
Kie JG, Bowyer RT, Nicholson MC, Boroski BB, Loft ER scales: application to roe deer. Dissertation, Claude Ber-
(2002) Landscape heterogeneity at differing scales: effects nard Lyon I
on spatial distribution of mule deer. Ecology 83:530–544 Van Moorter B, Gaillard J-M, McLoughlin PD, Delorme D,
Manly BFJ, McDonald LL, Thomas DL, McDonald TL, Er- Klein F, Boyce MS (2009) Maternal and individual effects
ickson WP (2002) Resource selection by animals: statis- in selection of bed sites and their consequences for fawn
tical analysis and design for field studies. Kluwer survival at different spatial scales. Oecologia 159:669–678
Academic Publishers, Dordrecht Venables WN, Ripley BD (2002) Modern applied statistics
Massé A, Côté SD (2009) Habitat selection of a large herbivore with S. Springer, New York
at high density and without predation: trade-off between Whittaker J (1990) Graphical models in applied multivariate
forage and cover? J Mammal 90:961–970 statistics. Wiley, Chichester
Morellet N, Verheyden H, Angibault J-M, Cargnelutti B, Worton BJ (1989) Kernel methods for estimating the utilization
Lourtet B, Hewison AJM (2009) The effect of capture on distribution in home-range studies. Ecology 70:164–168
ranging behaviour and activity of the European roe deer
Capreolus capreolus. Wildl Biol 15:278–287

123
The author has requested enhancement of the downloaded file. All in-text references underlined in blue are linked to publications on Researc