Content uploaded by Rafael Barrientos
Author content
All content in this area was uploaded by Rafael Barrientos
Content may be subject to copyright.
Content uploaded by Rafael Barrientos
Author content
All content in this area was uploaded by Rafael Barrientos
Content may be subject to copyright.
Biodivers Conserv (2009) 18:405–418
DOI 10.1007/s10531-008-9499-9
1 C
ORIGINAL PAPER
The presence of rabbits adjacent to roads increases
polecat road mortality
R. Barrientos · L. Bolonio
Received: 17 December 2007 / Accepted: 26 September 2008 / Published online: 15 October 2008
© Springer Science+Business Media B.V. 2008
Abstract Road mortality is an increasing problem for terrestrial vertebrate conservation
due to the increase of both road numbers and vehicle Xow. We hypothesize that the proba-
bility of a predator being killed on the road is related to the presence of its prey adjacent to
the road, which is likely to be related to the use that these predators make of road verges.
We aim to identify the features of speciWc stretches of road where road-kills of a predator
(European polecat) occur in Mediterranean landscapes, including the presence of its main
prey (European rabbit) and landscape and road features. We compared 85 100 m long
stretches of road where at least one road-kill was recorded with 104 stretches without any
road-kill in a dry agricultural landscape in central Spain. We used regression analysis to
investigate the relationship between road-kill occurrence and the features in the 67% of the
cases. Road-kill stretches were characterised by greater numbers of rabbit burrows in the
road verges and by higher traYc Xow and speed (i.e. higher speed limit, lower proportion of
heavy vehicles, wider road and lower proportion of unbroken central lines). Road-kill
stretches also had more metres built over bridges and lower densities of people. We
validated our best model with a dataset (the 33% of the cases) not included in its develop-
ment, which correctly classiWed 82% of road-kill stretches and 89% of non-road kill
stretches. Our results highlight the need for taking into account food resource distribution
when studying causes of animal road-kills.
R. Barrientos (&)
Departamento de Ecología Funcional y Evolutiva, Estación Experimental de Zonas Áridas,
CSIC, General Segura 1, 04001 Almeria, Spain
e-mail: barrientos@eeza.csic.es
L. Bolonio
Grupo Ornitológico Alcedo, Facultad de Biología, Universidad de Alcalá de Henares,
N-II, km. 36.5, 28871 Alcala de Henares, Madrid, Spain
e-mail: luis.bolonio@terra.es
406 Biodivers Conserv (2009) 18:405–418
1 C
Keywords Carnivore conservation · Mustela putorius · Oryctolagus cuniculus ·
Resources distribution · Road verge · TraYc · Unbroken line
Abbreviations
AIC Akaike information criterion
AICc Corrected Akaike information criterion
IGN The Spanish national geographic institute
PC Principal component
PCA Principal component analysis
UTM Universal transverse mercator coordinate system
Introduction
Carnivores are commonly identiWed as species whose conservation is more diYcult to
guarantee due to their highly demanding habitat requirements (e.g. Ginsberg 2001).
Present-day conservation strategies for carnivores focus on the integration of these
species into multi-use landscapes dominated by people (Linnell et al. 2000; Iuell et al.
2003). Roads cause one of the most important anthropogenic impacts on wildlife
communities and, as the cause of direct mortality, may reduce dispersal and reproduc-
tive or colonization events (reviewed in Forman and Alexander 1998; Spellerberg 1998;
Trombulak and Frissell 2000; Forman et al. 2003). Even without road mortality, roads
can cause barrier eVects. For example in a study with another mustelid species, the
badger (Meles meles), Clarke et al. (1998) found that high traYc loads may discourage
badgers from attempting to cross major roads. Similar Wndings have also been found for
other species (reviewed in Forman and Alexander 1998; Spellerberg 1998; Trombulak
and Frissell 2000; Forman et al. 2003). However, roads can increase the foraging
opportunities for predators through the availability of road-kill carrion (reviewed in
Little et al. 2002). We hypothesize that prey availability in road verges can increase the
fatality risk for predators. We tested this hypothesis with a study of the European
polecat (Mustela putorius L.) and its main prey in several ecosystems, the European
rabbit (Oryctolagus cuniculus L.).
The polecat is a small carnivore widely distributed in Europe where it remains in low
densities throughout most of its range, particularly in the Mediterranean peninsulas
(Mitchell-Jones et al. 1999; Marcelli et al. 2003; Virgós 2003; Virgós et al. 2007b). For
several decades, the species has undergone a serious decline in population levels through-
out the majority of its distribution. This decline has been caused by a combination of
factors, some human related (i.e. hunting, road-kills), but competition with alien species,
hybridization processes or low eVective population sizes have also been implicated (Blandford
1987; Davison et al. 1999; Mitchell-Jones et al. 1999; Sidorovich 2000; Lodé et al. 2003).
Recent status assessments in some parts of Europe have suggested that the polecat has
declined or is at unfavourable conservation status (see Virgós et al. 2007b for the Iberian
Peninsula). Consequently, it is currently listed on Annexe V of the EC Habitats and Species
Directive and Appendix III of the Bern Convention. In Spain, the polecat was recently
listed as “almost threatened” (Virgós et al. 2007b).
It has been suggested that uncultivated landscapes are important habitat resources for
carnivores in the inhospitable agricultural ecosystems as they provide protection and
connectivity among suitable habitats (Virgós 2001). The pattern of use of road verges by
polecats is unclear as Blandford (1987) suggests that polecats make frequent use of this
Biodivers Conserv (2009) 18:405–418 407
1 C
habitat, whereas Rondinini et al. (2006), contrary to their own prediction, found that road
verges were used by polecats in proportion to their availability. Habitat use by polecats is
dependent on habitat structure and food dispersion, visiting prey-rich habitat patches most
frequently (Lodé 1994; Baghli et al. 2005; Zabala et al. 2005; Rondinini et al. 2006; Mestre
et al. 2007). In agricultural landscapes, roadside vegetation is commonly more developed
than surrounding Welds as vegetation growth is promoted by water running oV the road into
the ditch (review in Forman et al. 2003). In addition, roadside vegetation is rarely harvested
(Bellamy et al. 2000). Thus, if we assume that vegetation requirements of polecats could be
met by roadside strips, the probability of road-kill, mediated by the use of roadsides by
polecats, may depend on the presence of their prey. Rabbits, which commonly place their
warrens in road ditches (authors, personal observation), are the main prey for polecats in
several habitats, especially in Mediterranean ones (Roger 1991; Lodé 1997; Schröpfer et al.
2000).
Road mortality has been implicated as one of the major factors preventing the mainte-
nance of healthy and stable populations of polecats in Mediterranean landscapes (Virgós
et al. 2007b). In other geographic regions, where the ecology of the polecat is better
known, road fatalities are also a major cause of polecat deaths (Blandford 1987; Birks
1997; Mitchell-Jones et al. 1999; Lodé 2003; Kristiansen et al. 2007). Despite these
Wndings and that road-kills seem to be concentrated along certain stretches of road
(Blandford 1987), no studies have analysed habitat and road-related variables at speciWc
locations where road-kills occur. Road mortality patterns in small animals seem to be
species-speciWc (Clevenger et al. 2003). However there is a general paucity of knowl-
edge regarding the relationship between road features and small-animal fatalities
(Ginsberg 2001). TraYc characteristics and road features play an important role in
road-kills. For instance, the probability of road-kill mortality increases with both traYc
volume and speed (Van Langevelde and Jaarsma 2004). However, some road eVects such
as barrier eVects can adjust the level of mortality. For example Clarke et al. (1998) found
that roads with high traYc loads have proportionately fewer badger road-kills due
increased avoidance of these roads by the badgers. Unfortunately, this complex relation-
ship is hard to capture without radio-tracking studies. Road width is another important
feature, animals needing more time to cross wider roads, leading to a decrease in the
probability of a successful crossing (Van Langevelde and Jaarsma 2004). This is particu-
larly important for small animals (review in Forman et al. 2003). However, few of the
studies that have focused on small vertebrates have analysed road-kill in detail at speciWc
points where road-kills have occurred in order to develop predictive models of fatalities
(but see Saeki and Macdonald 2004; Ramp et al. 2005, 2006).
The aim of our study was to identify the features of speciWc stretches where polecat
road-kills occur and to provide the basis of a strategy to prevent this mortality. SpeciWcally,
we focused our research on three basic characteristics: (1) The presence of prey near the
road. In our study area, rabbits make up the bulk (mean 43.6% year-round) of polecat diet
by biomass (Cuesta 1994). Consequently, we expected road-kills to be associated with
those stretches of road where rabbits were more abundant. (2) Roadside vegetation.
Polecats are expected to make greater use of vegetated patches (Lodé 1994; Baghli et al.
2005; Zabala et al. 2005; Rondinini et al. 2006) and consequently the number of road-kills
is expected to be positively associated with greater amounts of vegetation. (3) Road and
traYc characteristics. We expected that stretches of road with higher speed limits and traYc
volume would be linked with larger numbers of polecat road-kills (Van Langevelde and
Jaarsma 2004).
408 Biodivers Conserv (2009) 18:405–418
1 C
Materials and methods
Study area and road kill survey
This study was carried out in the Tajo valley, Toledo province, central Spain, from August
2002 to July 2004. The Tajo basin ranges between 350 and 850 m in altitude. The climate is
Mediterranean and during the study period annual rainfall averaged 340 mm. The average
daily maximum temperature was 27.1°C (August) and the minimum 3.6°C (December).
The dominant land cover is dry crops such as cereals, fallow and olive Welds (altogether dry
crops cover the 54.6%). The remaining land cover consists of irrigated crops (mainly maize
Welds) (2.1%), urban areas (10.1%) and non-cultivated land (33.2%) (dominated by scrub-
land composed of Retama sphaerocarpa and Stipa tenacissima). The study area was
bounded to the North by motorway N–V (40°02⬘N, 04°26⬘W), by the city of Toledo
(39°51⬘N, 04°01⬘W) to the East, by the Montes de Toledo mountain range (39°35⬘N,
04°37⬘W) to the South and by the roads CM-4015 and CM-4102 (39°48⬘N, 04°38⬘W) to
the West. The extent of road development and traYc Xow in the study area can be consid-
ered to be representative of that found in most of central Spain.
We surveyed polecat road-kills on all of the two-lane paved roads in the study area using
a car. These roads total 246 km divided into two routes of 121 and 105 km. The number of
surveys conducted was 41; two surveys per month except for 7 months when we could only
carry out one survey. Thus, the total distance covered was greater than 10,000 km.
The survey was conducted by two observers. Whilst one drove at 40–50 km/h, the other
looked for carcasses on the road surface. Polecats were often found in the middle of the
road, although we cannot discount that some of the animals killed in the centre of the road
were moved to the ditch at the side by motorists or just by the crash. To assess the degree to
which we overlooked carcasses in the ditches, we randomly surveyed over 100 km of
verges on foot, along all the road classes. We found only two polecat carcasses not detected
by the previous vehicle survey. Despite surveys done by foot being more reliable, surveys
were conducted by car as much larger distances could be covered. Furthermore, the proba-
bility of missing carcases in our car surveys was likely to be constant across the study area.
All the carcases were removed from the road surface in order to avoid double counts.
Undercounting of carcases due to removal by scavengers was likely to be negligible as the
only scavenger common in the area is the magpie (Pica pica) which is not able to remove
polecat carcasses. As such, we could monitor for months some carcasses at identiWable
locations.
We divided the two sampling routes into 100 m stretches each delimited by two hecto-
metre posts. To select random stretches for their inclusion in the dataset, we chose at
random one stretch in every 20, discarding those with road-kills. Thus, if no polecat was
found during any survey in a selected stretch, this was considered to be a “non-road-kill
stretch” and it was included in the dataset. The stretches where we found any road-kill were
considered “road-kill stretches”. Although sampling was not paired, non-road-kill stretches
were distributed along the study area in proportion to the total length of every road type.
Overall, we randomly chose 104 non-road-kill stretches in order to compare their charac-
teristics with those of the 85 stretches where at least one road-kill was recorded. Before
constructed the model, we randomly divided the whole dataset into two, one made up of
126 cases (57 road-kill stretches and 69 without them) to use to develop the models. The
second dataset was composed of 63 cases (28 road-kill stretches and 35 without them) and
was used to validate the models.
Biodivers Conserv (2009) 18:405–418 409
1 C
Variables associated with road-kills
Landscape variables were assessed within a circular area based on the central point of the
road section and extending 50 m along roadside on either side of the road. This scale was
chosen as we started on the premise that rabbit presence is a key factor determining the use
of road verges by polecats and therefore it will also aVect the probability of polecats being
killed on the road. The chosen scale for landscape variables matches the size of rabbit home
ranges in Mediterranean habitats, which is about one hectare (Lombardi et al. 2007). The
presence of fresh rabbit carcases can be another important factor attracting polecats on to
roads (Birks and Kitchener 1999). As such, it is worth mentioning that rabbit carrion
reached up to 1.2 carcases/km and survey on some stretches during our study (Authors,
unpublished data). It is worth mentioning that the landscape composition is quite homoge-
neous at small scales as the area is dominated by extensive agriculture and non-cultivated
lands and that no changes in agricultural use were detected while carrying out the study.
This size of buVer zone also matched the 100 m measurement unit used for the road-related
variables. Data obtained from a stretch of road where road works occurred during the study
were excluded from the datasets. Following similar criteria to Ramp et al. (2005), variables
were selected for inclusion in the models when they were known from literature to be
related to animal fatalities, to be resources used by polecats or when they were manage-
ment variables not speciWc to our study area (see Table 1 for variables studied and their
deWnitions). Thus, variables such as dry Welds, non-cultivated land or other land use types
were incorporated in order to investigate the potential inXuence of landscape on polecat
mortality through variable resource distribution eVects (e.g. foraging, protection). The
number of rabbit burrows and the distance from the road to the nearest burrow were used as
measurements of the local abundance of rabbits. The study area is mainly deforested, prob-
ably leading to polecats to concentrate their movements in those patches with greater
amounts of vegetation, especially riverbanks (Zabala et al. 2005; Rondinini et al. 2006;
Mestre et al. 2007). To test for such a vegetation eVect on road-kills, we included the
percentage of woody vegetation cover and the distance to the nearest stream as model
variables. The distance to the nearest house and town were used as proxies for human
encroachment (i.e. greater presence of humans in the Weld, intensiWcation of agricultural
practices, alteration of ecosystems, pollution, etc.). Vehicle speed and traYc Xow rate can
aVect the likelihood of collisions with animals (Malo et al. 2004; Van Langevelde and
Jaarsma 2004; Seiler 2005), and we therefore included the speed limit, percentage of heavy
vehicles, road width, percentage of unbroken line (where overtaking is forbidden) and
traYc intensity as model variables. The lengths of sections of road with embankments,
going over bridges and the distance to the nearest non-wildlife underpass (mainly culverts)
were also included to control for the potential inXuence of structures which may funnel or
facilitate animal movements at particular crossing points (Yanes et al. 1995; Malo et al.
2004). Finally, we included road class as a categorical factor, encompassing a range of road
traits. This categorical variable consisted of Wrst or second class roads, based on the auton-
omous govern classiWcation of the two-lane paved roads. The characteristics of Wrst class
roads, measured in 63 random 100 m long stretches, were (mean §SE): 10.2 §0.1 m for
road width, 43.2 §4.6% unbroken line and a road surface of generally good quality. In
contrast, second class roads, measured in 41 random stretches, were narrower (6.5 §
0.2 m), had more stretches with bends (i.e. higher percent of unbroken line: 49.6 §6.6%)
and had road surfaces of lower quality. We initially included variables that were likely to
be correlated as an attempt to incorporate the best set of predictors in the models. However,
410 Biodivers Conserv (2009) 18:405–418
1 C
where correlation between variables was found we excluded the one with the least explana-
tory power (see “Statistical analyses”).
Altogether 14 quantitative local-scale variables (six landscape and eight road-related; see
Table 1 for their measurement units) and one categorical variable were used. The quantitative
variables were counted, measured or scored visually (percentage of vegetation coverage) and
always by the same researcher (R.B.). TraYc data were obtained from Junta de Comunidades
de Castilla-La Mancha and Diputación de Toledo. We used the monthly mean for traYc
variables. Finally, the three landscape large-scale variables were measured using a 1:50,000
topographic map of Spain, IGN, using the coordinates Universal Transverse Mercator coordi-
nate system (UTM) of the mid-point of a stretch as its location.
Statistical analyses
Quantitative diVerences between stretches with and without road-kills were evaluated using
unpaired t-tests (e.g. Seiler 2005) and Bonferroni step-down correction (Holm 1979). Tests
were calculated using STATISTICA 6.0 (Statsoft 2003).
We used a Factor Analysis with Principal Component Analysis (PCA) and the Varimax
normalized factor rotation to build a correlation matrix to explore the degree of association
among variables. For the subsequent model development, and in order to minimize multi-
collinearity (correlations >0.6) among independent variables, we only included the most
highly correlated variable with each of the PC factors, discarding the remaining variables.
Due to the strong associations of the variables selected according to PCA factor loadings,
seven variables were removed before the model process: NON CULT, OTHER, WOODY,
N_BURROWS, HEAVY, WIDTH and PASS (Appendix 1). We included road type
Table 1 The variables measured in each 100 m stretch used to study the factors that diVerentiate stretches
with polecat road-kills from those without them
Variable DeWnition
Landscape local-scale variables
DRY % Cover of dry crops within a radius of 50 m around both sides of the road
NON_CULTIV % Cover of non cultivated lands within a radius of 50 m around both sides of the road
OTHER % Cover of other land uses within a radius of 50 m around both sides of the road
WOODY % Cover of woody plants in within a radius of 50 m around both sides of the road
M_BURROW Distance (m) to the nearest rabbit burrow on any side of the road
N_BURROWS Number of rabbit burrows within a radius of 50 m around both sides of the road
Landscape large-scale variables
TOWN Distance (m) to the nearest town
HOUSE Distance (m) to the nearest isolated house on any side of the road
RIVER Distance (m) to the nearest river
Road-related local-scale variables
SPEED Speed limit on the stretch of road
TRAFFIC Monthly mean number of vehicles per day along the stretch
HEAVY Mean % of heavy vehicles per day along the stretch
WIDTH Road width (m)
PASS Distance (m) to the nearest non-wildlife underpass
EMBANKMENT Total metres of embankments of > 1 m height and > 60° slope on both sides
of the 100 m stretch
BRIDGE Length (m) of the 100 m stretch built over bridges.
UNBROK_LINE % of the length of the 100 m stretch of the road where overtaking is forbidden.
CLASS Categorical variable: 1 Wrst, 0 second class regional road
Biodivers Conserv (2009) 18:405–418 411
1 C
(i.e. Wrst or second class) as categorical factor in our models. Before starting the modelling
process, we randomly set aside 33% of the data (i.e. 63 cases, of which 28 were road-kill
stretches and 35 were without road-kills) to use them for model validation. Therefore
model development used 67% of the data (i.e. 126 cases, of which 57 were road-kill
stretches and 69 were without road-kills). We conducted a regression analysis utilising a
binomial distribution (road-kill versus non-road-kill stretches) and logit link function. We
used the best subsets procedure and the Akaike’s information criterion (AIC) to identify the
set of models best explaining the occurrence of road-kills in the 126 cases. We used this
technique because it yields consistent results independent of the order in which variables
are included in the model and allows models with diVerent numbers of parameters to be
directly compared with each other (Burnham and Anderson 2002). We used AIC values
corrected for small sample size (i.e. AICc) as the ratio between the number of observations
and estimator variables was under 40 (Burnham and Anderson 2002). The models were
developed using STATISTICA 6.0 and the AIC values were corrected with Microsoft
EXCEL 2003 (Microsoft) following Burnham and Anderson (2002). The models with the
lowest AICc represent the best compromise between a maximal Wt and a minimal number
of explanatory variables (i.e. statistical parsimony). To evaluate the relative explanatory
power of competing best models, we calculated the Akaike weights (i). The evidence
ratio was calculated to compare the Akaike weights of the best model and competing ones
to determine to what extent one was better than another (Burnham and Anderson 2002). In
order to estimate the relative importance of every variable included in any of the best mod-
els we calculated the sum of Akaike weights of the models where these variables were
included (Burnham and Anderson 2002). The signiWcance of variables included in the best
model was assessed using the log-likelihood ratio test because it is the most suitable for
small sample sizes (Moya-Laraño and Wise 2007). Model validation using data not used in
the model development was carried out using simple logistic regression (e.g. Seiler 2005).
Results
Road-kill pattern
A total of 107 polecat road-kills were recorded in 85 stretches (2 stretches with 3 casualties,
18 with 2 and 65 with only 1) over a period of 24 months. DiVerences between road-kill
and non-road-kill stretches based on t-tests are shown in Table 2.
Road-kill stretches had signiWcantly more rabbit burrows being the nearest one at
shorter distance. These stretches also had higher speed limits and traYc intensities, but had
lower frequencies of heavy vehicles. Road sections with fatalities were signiWcantly wider
and had lower percentages of unbroken line (i.e. were straight). Road-kill stretches were
also closer to streams and had lower percentages of other land uses (e.g. urban, irrigated
crops).
The regression procedure with the best subset analysis provided a set of 13 models
which could be considered as plausible models according to their AICc (i.e. the diVerence
between their AICc and the lowest one was less than two; Burnham and Anderson 2002).
The six best models are shown in Table 3. Our results do not give clear support for a single
model amongst these six, indicated by the small diVerences in AICc values and the compa-
rable values of Akaike weights (Table 3). Importantly, however, all the models included
M_BURROW, BRIDGE, SPEED and UNBROK_LINE (i.e. i= 0.5994). Less inXuen-
tial were TRAFFIC (i= 0.05262), HOUSE (i= 0.5174), TOWN (i= 0.2994) and
412 Biodivers Conserv (2009) 18:405–418
1 C
RIVER (i= 0.0832). We selected model no.1 (Table 4) as the best one because it repre-
sents the best compromise between a maximal Wt and a minimal number of explanatory
variables. The log-likelihood ratio test suggests that the most important variable associated
with the occurrence of polecat road-kills was the presence of rabbit burrows close to the
roadside.
Table 2 The results for the unpaired t-tests (mean §SD) comparing the 17 variables quantifying habitat and
road-related features between road-kill and non-road-kill stretches
The table shows the P-values with Bonferroni step-down correction (Holm 1979). See Table 1 for variable
codes
Variable Non-road-kill stretches Road-kill stretches Unpaired t-test
Mean §SD Mean §SD t-value P-value
DRY 49.14 §3.69 50.97 §3.75 0.35 0.7305
NON_CULT 32.12 §3.24 42.39 §3.49 1.94 0.0535
OTHER 18.75 §3.27 6.71 §1.80 ¡3.06 0.0275
WOODY 10.54 §1.54 11.45 §2.24 0.34 0.7315
M_BURROW 22.07 §0.75 9.53 §0.92 ¡10.68 <0.0001
N_BURROWS 3.02 §0.69 27.69 §2.74 9.54 <0.0001
TOWN 2475.96 §205.72 2518.82 §158.18 0.16 0.8735
HOUSE 272.12 §27.38 315.88 §31.27 1.06 0.2921
RIVER 437.50 §45.02 284.70 §39.15 ¡2.50 0.0133
SPEED 92.98 §12.53 99.53 §2.13 4.76 <0.0001
TRAFFIC 2445.66 §204.05 3847.08 §264.74 4.26 <0.0001
HEAVY 13.76 §0.81 9.46 §0.44 ¡4.34 <0.0001
WIDTH 8.75 §0.20 10.17 §0.10 6.09 <0.0001
PASS 91.40 §2.51 84.40 §3.36 ¡1.70 0.0902
EMBANKMENT 10.77 §3.46 20.12 §5.33 1.52 0.1301
BRIDGE 0.64 §0.35 0.88 §0.50 0.42 0.1301
UNBROK_LINE 45.72 §3.80 30.59 §3.87 ¡2.77 0.0062
Table 3 The set of six models that best separated stretches with polecat road-kills from those without them
The AICc is the diVerence in AICc values compared to the estimated best model (lowest AICc) what allows
the ranking of models from an estimated best (top of the table) to the worst. AICc weight is the estimated
probability t hat a model is the best model in the set. Evidence ratio indicates to what extent one model is bette
r
than another
Model no. Variables contained in the model KAICc
(i)
AICc
weight (i)
Evidence
ratio
1 M_BURROW + HOUSE + TRAFFIC + BRIDGE
+ UNBROK_LINE + SPEED
6 0 0.1461 0.00
2 M_BURROW + TOWN + HOUSE + TRAFFIC
+ BRIDGE + UNBROK_LINE + SPEED
7 0.0418 0.1430 2.11
3 M_BURROW + TRAFFIC + BRIDGE
+ UNBROK_LINE + SPEED
5 1.1552 0.0820 78.18
4 M_BURROW + TOWN + HOUSE + RIVER
+ TRAFFIC + BRIDGE + UNBROK_LINE
+ SPEED
8 1.2021 0.0832 82.40
5 M_BURROW + TOWN + HOUSE + BRIDGE
+ UNBROK_LINE + SPEED
6 1.3809 0.0732 99.47
6 M_BURROW + HOUSE + RIVER + TRAFFIC
+ BRIDGE + UNBROK_LINE + SPEED
7 1.4174 0.0719 103.13
Biodivers Conserv (2009) 18:405–418 413
1 C
Model validation
Simple logistic regression analysis was used to validate the predictive capacity of our best
model (Table 4). The use of the independent dataset showed that the model correctly
predicted 82.1% of road-kill stretches and 88.6% of non-road-kill ones. Thus, our model
succeeded in distinguishing between the occurrence and non-occurrence of polecat road-kills.
Discussion
Our results highlight the need for taking into account the distribution of food resources
when studying animal road-kills. The presence or abundance of resources near roads may
determine the use that animals make of roadsides and this, in turn, may aVect their road-kill
rates. In our polecat–rabbit system, we obtained a model where the main feature of
stretches where road-kills were recorded was the presence of the polecat’s main prey next
to the road, measured as the distance of the nearest rabbit burrow to the roadside. Higher
vehicle speeds (i.e. higher speed limits and smaller percentages of unbroken line) and
higher traYc Xows were also associated with road-kill stretches. Road-kill stretches also
had greater lengths over bridges and were farther from the nearest house. Our best model
had a high success rate in predicting polecat road-kill stretches using the independent data-
set (82–89% of cases correctly classiWed).
Rabbits make up the bulk of polecat diet in many, mainly Mediterranean, regions
(reviewed in Lodé 1997; see also Birks and Kitchener 1999; Virgós 2002). These lagomorphs
are mainly hunted by polecats in their warrens (Cuesta 1994), although they can be consumed
as carrion as well (Birks and Kitchener 1999). Thus, a signiWcant cause of polecat road-kills
must be their hunting movements among rabbit warrens on roadsides, although other causes
not identiWed here (seeking mates, dispersal, territory marking, etc.) could also be important
sources of polecat casualties (e.g., see Sleeman 1988 for a study on stoats). There are three
potential reasons for rabbits to establish warrens along roadsides: (1) Both in cereal crops and
sandy uncultivated areas, rabbits build their warrens between tree and shrub roots, seeking
protection against warren collapse by using the roots as supporting structures (Palomares
2003; Gea-Izquierdo et al. 2005). However, human activities have eliminated roots from
Welds for agricultural reasons and other warren supporting/protecting structures, such as gran-
ite rocks or walls, are scarce in the study area (Gea-Izquierdo et al. 2005). Building warrens
inside road embankments could be an adaptation to avoid their warrens collapsing by using
the embankments as supporting structures. (2) Ploughing destroys rabbit warrens (Calvete
et al. 2004). Consequently, rabbits avoid cultivated crops as areas to place their warrens and
Table 4 The best model after regression analysis with best subset procedure following AICc criterion
including the variables that best separated road-kill stretches from non-road-kill ones
Parameter Estimate SE Log likelihood-2P
CONSTANT ¡9.787977 7.109625
M_BURROW ¡0.144964 0.028240 35.272 <0.001
HOUSE 0.001590 0.000920 3.155 0.076
TRAFFIC 0.000257 0.000122 4.926 0.026
BRIDGE 0.410774 0.235557 3.899 0.048
UNBROK_LINE ¡0.018162 0.007623 6.305 0.012
SPEED 0.115488 0.072240 5.346 0.021
414 Biodivers Conserv (2009) 18:405–418
1 C
positively select non-ploughed ecotones (e.g. road verges) (Calvete et al. 2004; Gea-Izquierdo
et al. 2005). (3) The rabbits are important prey for almost 30 predator species in the Iberian
Peninsula but the activity of several predators decreases near roads, particularly in cultivated
landscapes (Virgós 2001; Bautista et al. 2004), suggesting that road verges could act as safer
breeding zones for rabbits. This eVect could be reinforced by the absence of shooting near
roads, which is forbidden by law.
Although scarce, stretches of road built over bridges were prone to be black spots.
There are three potential explanations that we suggest for this: (1) streams are important
ecotones selected by rabbits to place their warrens (Calvete et al. 2004) and stretches of
road built over bridges could be visited preferentially by polecats during hunting (Zabala
et al. 2005). (2) Although it was discarded in the pre-modelling process (see Appendix
1), woody coverage was positively related to the lengths of sections over bridges,
probably because the occurrence of riparian woodlands is a dominant component of
vegetation coverage in agricultural deforested areas. Utilisation of riparian vegetation is
an essential part of polecat movements in the agricultural matrix due to the protection it
provides (Zabala et al. 2005; Rondinini et al. 2006; Mestre et al. 2007). (3) Once a
polecat is trapped on a stretch of road that is diYcult to exit (e.g. on a bridge), the animal
will move in an erratic way along the section dramatically increasing its chances of
being run over.
The role of diVerent habitat types in species survival in relation to landscape structure
(suitable amounts of habitat, connectivity among patches or the geometry of patches) is a
topic under continuous debate (Andrén 1994; Gutzwiller 2002). In agricultural landscapes,
where most of the natural habitat has been transformed, the residual semi-natural
vegetation along rivers or road verges is a key tool in the conservation biology of many
species, including carnivores (Virgós 2001). If this kind of habitat is large enough, the
eVect of habitat loss or fragmentation can be minimized (Andrén 1994; Gutzwiller 2002).
The non-agricultural habitat patches distributed across the agricultural matrix is capable of
maintaining high levels of biodiversity because it can provide abundant food through the
edge eVect, allows animals to forage close to vegetation cover safe from humans or preda-
tors (Macdonald 1995) and maintains the connectivity among suitable patches (Andrén
1994; Gutzwiller 2002). In fact, some prey species inhabiting agricultural landscapes are
settling in non-arable patches, such as road verges (e.g. Bellamy et al. 2000). Thus, the
location of prey may become more predictable, and these patches will be visited more
frequently by predators, leading to a greater number of road-kills of the predator as seems
to be the case in our polecat–rabbit system.
Road-kills occur along those stretches of road that support higher traYc densities and
where vehicles go faster as predicted the traYc Xow theory (Van Langevelde and Jaarsma
2004). Road-kills are predicted in our study by higher speed limits, a lower density of
heavy vehicles (i.e. allowing faster driving), wider roads and a lower amount of unbroken
line (i.e. straight stretches). An animal crossing is successful if an acceptable gap in the
traYc Xow appears at the start of the crossing (Van Langevelde and Jaarsma 2004). TraYc
speed determines traYc mortality because lower vehicle speeds increase the probability of
a successful road crossing as they provide both the driver and the animal with a greater
amount of time to react and avoid road-kill (Van Langevelde and Jaarsma 2004). The pole-
cat is a small mammal. About 1 kg in weight and a body length of 50 cm (Blandford 1987).
This is the reason why it is rarely included in the Weld-of-view whilst driving at high speed.
Furthermore, polecats are often crepuscular or nocturnal (e.g. Marcelli et al. 2003), so are
active on roads when visibility for drivers is lower. Finally, the particular morphology of
polecats, with short legs and a long body (Blandford 1987), likely adaptations to hunting
Biodivers Conserv (2009) 18:405–418 415
1 C
rabbits in their warrens (Cuesta 1994), means that this species requires more time to cross a
road, decreasing its probability of a successful crossing, particularly on wider stretches
(Van Langevelde and Jaarsma 2004).
Finally, the least signiWcant predictor in the best model was the distance to the nearest
isolated house, this being greater for the road-kill stretches The distance to the nearest town
was also included in some of the best models, with a similar pattern. We interpret these
results as human encroachment leading to a decrease in suitable habitat for polecats (Zabala
et al. 2005).
One weakness of our model was that we did not know the density of live polecats in the
areas studied and therefore could not quantify the eVect of this variable on the incidence of
road-kills. Several studies have found a positive relationship between population density
and road-kills (e.g. Gehrt 2002; Seiler 2005; but see Klocker et al. 2006). However, our
model correctly predicted 82% of road-kill stretches which suggests either that the road-kill
pattern is density independent or that some of the variables that we measured were
correlated with live polecat density, and thus we estimated polecat density indirectly. For
example, it might be expected that polecat densities are higher in areas where its main prey,
the rabbit, is more abundant. To estimate densities of live animals requires costly monitor-
ing eVort and would therefore not be feasible in most cases. The variables we measured are
much easier and cheaper to obtain and the model developed was successful, and we there-
fore suggest that this model is a valuable tool for predicting polecat road-kills.
Management implications
The management measures we suggest must be easy and cheap to carry out as the road-kill
stretches in this study represented only 3.8% of the total road network. First, the number of
road-kills could be signiWcantly reduced along stretches of road with high numbers of
warrens in road embankments. This could be done by fencing oV the road and connecting
both sides with underpasses to permit the normal transit of both rabbits and their predators.
Second, the establishment of rabbit warrens along the roadside could be managed. ArtiWcial
refuges are currently being used to reintroduce or maintain rabbit populations elsewhere
(e.g. Gea-Izquierdo et al. 2005) as rabbits are declining in several areas (Virgós et al.
2007a). These artiWcial warrens could be used as part of a management strategy to dissuade
rabbits from settling in road embankments, by placing the artiWcial warrens away from the
roads. Even simple earth mounds can be used as a cheaper alternative to building artiWcial
warrens, as they are commonly used by rabbits to build their own warrens (authors,
personal observation). Third, traYc speed could be reduced along stretches of roads with
high numbers of road-kills by the use of lower speed limits combined with the laying of
speed bumps. This would be particularly necessary on straight stretches of road with high
roadside rabbit warren densities. Lower vehicle speeds will certainly increase the success-
ful crossing rate for this mustelid (Van Langevelde and Jaarsma 2004).
Acknowledgements We are very grateful to I. Cardiel, R. Jiménez, P.M. García, D. Martínez, O. Frías, J.A.
Calvo, J.A. Lemus for their help in the Weld. We thank F. Valera, R. Václav, E. Virgós, J. Moya-Laraño, G.
Hernández-Milián, J. Birks, D.P. Sleeman and an anonymous referee for their helpful suggestions. J. Moya-
Laraño, Mark T. Bulling and B. Nicholls checked the English. TraYc data were provided by M. Vidal (Junta de
Comunidades de Castilla-La Mancha), A. J. Cervantes and J. Cobas (Diputación de Toledo) and A. Oliver
(TELVENT). The Dirección General del Medio Natural of Toledo Province (Junta de Comunidades de Castilla-
La Mancha) partially supported our research by means of TO-01-03 and TO-01-04.
416 Biodivers Conserv (2009) 18:405–418
1 C
Appendix 1
The results of the principal component analysis run prior to the development of the models in order to inves-
tigate the multicollinearity among independent variables
References
Andrén H (1994) EVects of habitat fragmentation on birds and mammals in landscapes with diVerent propor-
tions of suitable habitat: a review. Oikos 71:355–366. doi:10.2307/3545823
Baghli A, Walzberg C, Verhagen R (2005) Habitat use by the European polecat Mustela putorius at low density
in a fragmented landscape. Wildl Biol 11:331–339. doi:10.2981/0909-6396(2005)11[331:HUBTEP]
2.0.CO;2
Bautista LM, García JT, Calmaestra RG, Palacín C, Martín CA, Morales MB et al (2004) EVects of weekend
road on the use of space by raptors. Conserv Biol 18:726–732. doi:10.1111/j.1523-1739.2004.00499.x
Bellamy PE, Shore RF, Ardeshir D, Treweek JR, Sparks TH (2000) Road verges as habitat for small
mammals in Britain. Mammal Rev 30:131–139. doi:10.1046/j.1365-2907.2000.00061.x
Birks JDS (1997) A volunteer-based system for sampling variations in the abundance of polecats (Mustela
putorius). J Zool (Lond) 243:857–863
Birks JDS, Kitchener AC (1999) The distribution and status of the polecat Mustela putorius in Britain in the
1990s. The Vincent Wildlife Trust, London
Blandford PRS (1987) Biology of the polecat Mustela putorius, a literature review. Mammal Rev 17:155–198.
doi:10.1111/j.1365-2907.1987.tb00282.x
Burnham KP, Anderson DR (2002) Model selection and multimodel inference. Springer, New York
Calvete C, Estrada R, Angulo E, Cabezas-Ruíz S (2004) Habitat factors related to wild rabbit conservation in
an agricultural landscape. Land Ecol 19:531–542. doi:10.1023/B:LAND.0000036139.04466.06
Clarke GP, White PCL, Harris S (1998) EVects of roads on badger Meles meles populations in south-west
England. Biol Conserv 86:117–124. doi:10.1016/S0006-3207(98)00018-4
Clevenger AP, Chruszcz B, Gunson KE (2003) Spatial patterns and factors inXuencing small vertebrate fauna
road-kill aggregations. Biol Conserv 109:15–26. doi:10.1016/S0006-3207(02)00127-1
Cuesta M (1994) Bioecología de los mustélidos en los Montes de Toledo. Disseration, Universidad Complu-
tense of Madrid
Davison A, Birks JDS, GriYths HI, Kitchener AC, Biggins D, Butlin RK (1999) Hybridization and the
phylogenetic relationship between polecats and domestics ferrets in Britain. Biol Conserv 87:155–161.
doi:10.1016/S0006-3207(98)00067-6
Forman RTT, Alexander LE (1998) Roads and their major ecological eVects. Annu Rev Ecol Evol Syst
29:207–231
Variable PC1 PC2 PC3 PC4 PC5 PC6
DRY 0.421258 ¡0.132407 0.230408 ¡0.839526 ¡0.030065 ¡0.007502
NON_CULT 0.219066 0.230037 ¡0.097520 0.809347 0.067512 0.218732
OTHER ¡0.820769 ¡0.107831 ¡0.182926 0.110852 ¡0.043183 ¡0.257860
WOODY 0.077028 0.633344 0.101267 0.369508 0.224009 ¡0.143506
M_BURROW ¡0.242736 0.066876 0.115347 ¡0.127656 ¡0.071437 ¡0.805967
N_BURROWS 0.147825 0.027444 ¡0.132319 0.182981 0.218188 0.786171
TOWN 0.263878 ¡0.006824 0.377190 0.529965 ¡0.054079 0.212921
HOUSE 0.230201 ¡0.354476 0.494688 0.072375 0.239758 0.215083
RIVER 0.126340 ¡0.286418 0.083116 0.025225 0.384410 ¡0.539122
SPEED 0.854368 0.029600 ¡0.110758 0.169153 0.018011 0.078418
TRAFFIC ¡0.161921 0.115909 ¡0.762606 0.150385 0.265780 0.115479
HEAVY ¡0.085134 0.117805 0.647747 0.082747 ¡0.053125 ¡0.223346
WIDTH 0.431084 ¡0.088140 ¡0.617466 0.248435 ¡0.061471 0.129852
PASS 0.008378 ¡0.750472 0.007064 ¡0.033088 0.118274 ¡0.143836
EMBANKMENT 0.126225 0.086953 ¡0.269614 ¡0.031974 0.783474 0.109793
BRIDGE 0.043028 0.775742 ¡0.042387 0.075463 0.153974 0.017040
UNBROK_LINE ¡0.443517 0.143221 0.132539 0.129783 0.563282 0.069696
Biodivers Conserv (2009) 18:405–418 417
1 C
Forman RTT, Sperling D, Bissonette JA, Clevenger AP, Cutshall DC, Dale VH et al (2003) Road ecology.
Science and solutions. Island Press, Washington
Gea-Izquierdo G, Muñoz-Igualada J, San Miguel-Aynaz A (2005) Rabbit warren distribution in relation to
pasture communities in Mediterranean habitats, consequences for management of rabbit populations.
Wildl Res 32:723–731. doi:10.1071/WR04129
Gehrt SD (2002) Evaluation o f spotlight and road-kill surveys as indicators of local raccoon abundance. Wildl
Soc Bull 30:449–456
Ginsberg JR (2001) Setting priorities for carnivore conservation, what makes carnivores diVerent? In:
Gittleman JL, Funk SM, Macdonald D, Wayne RK (eds) Carnivore conservation. Cambridge University
Press, Cambridge, pp 498–523
Gutzwiller KJ (ed) (2002) Applying landscape ecology in biological conservation. Springer, New York
Holm S (1979) A simple sequentially rejective Bonferroni test procedure. Scand J Stat 6:65–70
Iuell B, Bekker GJ, Cuperus R, Dufek J, Fry G, Hicks C et al (2003) Wildlife and traYc. A European
handbook for identifying conXicts and designing solutions. KNNV Publishers, Delft
Klocker U, Croft DB, Ramp D (2006) Frequency and causes of kangaroo-vehicle collisions on an Australian
outback highway. Wildl Res 33:5–15. doi:10.1071/WR04066
Kristiansen LV, Sunde P, Nachman G, Madsen AB (2007) Mortality and reproductive patterns of wild
European polecats Mustela putorius in Denmark. Acta Theriol (Warsz) 52:871–878
Linnell JDC, Swenson JE, Andersen R (2000) Conservation of biodiversity in Scandinavian boreal forest,
large carnivores as Xagships, umbrellas, indicators, or keystones? Biodivers Conserv 9:857–868. doi:10.
1023/A:1008969104618
Little SJ, Harcourt RG, Clevenger AP (2002) Do wildlife passages act as prey-traps? Biol Conserv 107:135–
145. doi:10.1016/S0006-3207(02)00059-9
Lodé T (1994) Environmental factors inXuencing habitat exploitation by the polecat Mustela putorius in
western France. J Zool (Lond) 234:75–88
Lodé T (1997) Trophic status and feeding habits of the European polecat Mustela putorius L. 1758. Mammal
Rev 27:117–184. doi:10.1111/j.1365-2907.1997.tb00447.x
Lodé T (2003) Sexual dimorphism and trophic constraints: prey selection in the European polecat (Mustela
putorius). Ecoscience 10:17–23
Lodé T, Pereboom V, Berzins R (2003) Implications of an individualistic lifestyle for species conservation:
lessons from jealous beasts. C R Biol 326:S30–S36. doi:10.1016/S1631-0691(03)00024-6
Lombardi L, Fernández N, Moreno S (2007) Habitat use and spatial behaviour in the European rabbit in three
Mediterranean environments. Basic Appl Ecol 8:453–463. doi:10.1016/j.baae.2006.09.004
Macdonald DW (1995) European mammals: evolution and behaviour. Harper Collins, Oxford
Malo JE, Suárez F, Díez A (2004) Can we mitigate animal-vehicle accidents using predictive models? J Appl
Ecol 41:701–710. doi:10.1111/j.0021-8901.2004.00929.x
Marcelli M, Fusillo R, Boitani L (2003) Sexual segregation in the activity patterns of European polecats
(Mustela putorius). J Zool (Lond) 261:249–255. doi:10.1017/S0952836903004151
Mestre FM, Ferreira JP, Mira A (2007) Modelling the distribution of the European polecat Mustela putorius
in a Mediterranean agricultural landscape. Rev Ecol (Terre Vie) 62:35–47
Mitchell-Jones AJ, Amori G, Bogdanowicz W, Kryntufek B, Reijnders PJH, Spitzenberger F et al (1999) The
atlas of European mammals. Academic Press, London
Moya-Laraño J, Wise DW (2007) Two simple strategies of analysis to increase the power of experiments with
multiple response variables. Basic Appl Ecol 8:398–410. doi:10.1016/j.baae.2006.09.014
Palomares F (2003) Warren building by European rabbits (Oryctolagus cuniculus) in relation to cover avail-
ability in sandy area. J Zool (Lond) 259:63–67. doi:10.1017/S0952836902002960
Ramp D, Caldwell J, Edwards KA, Warton D, Croft DB (2005) Modelling of wildlife fatality hotspots along
the snowy mountain highway in New South Wales, Australia. Biol Conserv 126:474–490. doi:10.1016/
j.biocon.2005.07.001
Ramp D, Wilson VK, Croft DB (2006) Assessing the impacts of roads in peri-urban reserves: road-based
fatalities and road usage by wildlife in the Royal National Park, New South Wales, Australia. Biol
Conserv 129:348–359. doi:10.1016/j.biocon.2005.11.002
Roger M (1991) Food-habits and prey availability in the polecat Mustela putorius. Rev Ecol (Terre Vie)
46:245–261
Rondinini C, Ercoli V, Boitani L (2006) Habitat use and preference by polecats (Mustela putorius L.) in a
Mediterranean agricultural landscape. J Zool (Lond) 269:213–219
Saeki M, Macdonald DW (2004) The eVects of traYc in the raccon dog (Nyctereutes procyonoides viverri-
nus) and other mammals in Japan. Biol Conserv 118:559–571. doi:10.1016/j.biocon.2003.10.004
418 Biodivers Conserv (2009) 18:405–418
1 C
Schröpfer von R, Bodenstein C, Seebass C (2000) Der Räuber-Beute-Zusammenhang zwischen dem Iltis
Mustela putorius L., 1785 und dem Wildkanichen Oryctolagus cuniculus (L., 1758). Z Jagdwiss 46:1–13.
doi:10.1007/BF02240659
Seiler A (2005) Predicting locations of moose-vehicle collisions in Sweden. J Appl Ecol 42:371–382.
doi:10.1111/j.1365-2664.2005.01013.x
Sidorovich VE (2000) The on-going decline of riparian mustelids (European mink, Mustela lutreola, polecat,
Mustela putorius, and stoat, Mustela erminea) in eastern Europe: a review of the results to date and an
hypothesis. In: GriYths HI (ed) Mustelids in a modern world. Management and conservation aspects of
small carnivore: human interactions. Backhuys Publishers, Leiden, pp 295–317
Sleeman DP (1988) Irish stoat road casualties. Ir Nat J 22:527–529
Spellerberg IF (1998) Ecological eVects of roads and traYc, a literature review. Glob Ecol Biogeogr 7:317–333.
doi:10.1046/j.1466-822x.1998.00308.x
Trombulak SC, Frissell CA (2000) Review of ecological eVects of roads on terrestrial and aquatic communi-
ties. Conserv Biol 14:18–30. doi:10.1046/j.1523-1739.2000.99084.x
Van Langevelde F, Jaarsma CF (2004) Using traYc Xow theory to model traYc mortality in mammals. Land
Ecol 19:895–907. doi:10.1007/s10980-004-0464-z
Virgós E (2001) Relative value of riparian woodlands in landscapes with diVerent forest cover for medium-
sized Iberian carnivores. Biodivers Conserv 10:1039–1049. doi:10.1023/A:1016684428664
Virgós E (2002) Mustela putorius, Linneaus, 1758. In: Palomo LJ, Gisbert J (eds) Atlas de los Mamíferos
Terrestres de España. Dirección General de Conservación de la Naturaleza-SECEM-SECEMU, Madrid,
pp 262–265
Virgós E (2003) Association of the polecat Mustela putorius in eastern Spain with montane pine forests. Oryx
37:484–487. doi:10.1017/S0030605303000863
Virgós E, Cabezas-Díaz S, Lozano J (2007a) Is the wild rabbit (Oryctolagus cuniculus) a threatened species
in Spain? Sociological constrains in the conservation of species. Biodivers Conserv 16:3489–3504.
doi:10.1007/s10531-006-9054-5
Virgós E, Cabezas-Díaz S, L ozano J (2007b) Mustela putorius, Linneaus, 1758. Ficha Libro R ojo. In: Palomo
LJ, Gisbert J, Blanco JC (eds) Atlas y Libro Rojo de los Mamíferos Terrestres de España. Dirección
General de Conservación de la Naturaleza-SECEM-SECEMU, Madrid, pp 297–298
Yanes M, Velasco JM, Suárez F (1995) Permeability of roads and railways to vertebrates: the importance of
culverts. Biol Conserv 71:217–222. doi:10.1016/0006-3207(94)00028-O
Zabala J, Zuberogoitia I, Martínez-Climent JA (2005) Site and landscape features ruling the habitat use and
occupancy of the polecat (Mustela putorius) in a low density area, a multiscale approach. Eur J Wildl
Res 51:157–162. doi:10.1007/s10344-005-0094-z