Dynamic connectivity assessment for a terrestrial predator in a metropolitan region

Abstract

Protecting wildlife movement corridors is critical for species conservation. Urban planning often aims to create corridors for animal movement through conservation initiatives. However, research on connectivity for urban wildlife is limited. Here, we assessed connectivity for coyotes (Canis latrans) dynamically across temporal scales and demographic traits, parametrized using the habitat selection of 27 global positioning system (GPS)‐collared coyotes in the city of Toronto, Canada. The habitat selection models accounted for human population density, impervious area, vegetation density, and distance to different linear features. Results indicated that (1) vegetation‐dense areas were key for connectivity in urban areas; (2) riverbanks, railways, and areas below power lines were predicted as movement corridors; and (3) commercial and industrial clusters strongly disrupted connectivity. Spatiotemporal differences in connectivity were associated with time of day and coyote social status but not with climate and biological seasonality or coyote age and sex. Residential roads were pivotal in the temporal dynamism of connectivity. The maintenance and enhancement of plant structural complexity along key infrastructure (for example, highways, waterways, and parking lots) should be considered when managing connectivity corridors in cities.

Full text

© 2024 e Authors. Frontiers in Ecolog y and the Environment published by Wiley Periodicals LLC on behalf of e Ecolog ical Society of America.
Dynamic connectivity assessment for a terrestrial
predator in a metropolitan region
Tiziana A Gelmi- Candusso1*, Andrew TM Chin2, Connor A Thompson3, Ashley AD McLaren4,5, Tyler J Wheeldon4,
Brent R Patterson3,4, and Marie- Josée Fortin1
Protecting wildlife movement corridors is critical for species conservation. Urban planning often aims to create corridors for ani-
mal movement through conservation initiatives. However, research on connectivity for urban wildlife is limited. Here, we assessed
connectivity for coyotes (Canis latrans) dynamically across temporal scales and demographic traits, parametrized using the habi-
tat selection of 27 global positioning system (GPS)- collared coyotes in the city of Toronto, Canada. The habitat selection models
accounted for human population density, impervious area, vegetation density, and distance to different linear features. Results
indicated that (1) vegetation- dense areas were key for connectivity in urban areas; (2) riverbanks, railways, and areas below power
lines were predicted as movement corridors; and (3) commercial and industrial clusters strongly disrupted connectivity.
Spatiotemporal differences in connectivity were associated with time of day and coyote social status but not with climate and bio-
logical seasonality or coyote age and sex. Residential roads were pivotal in the temporal dynamism of connectivity. The mainte-
nance and enhancement of plant structural complexity along key infrastructure (for example, highways, waterways, and parking
lots) should be considered when managing connectivity corridors in cities.
Front Ecol Environ 2024; 22(4): e2633, doi:10.1002/fee.2633
Maintaining animal movement between habitat patches is
essential to preserving animal population fitness, by
facilitating dispersal throughout the course of a species’ life
cycle, such as seasonal migrations and colonization of new
habitat patches (McGuire et al.2016), as well as to preserving
their ecological contribution to plant communities, by facili-
tating pollination and seed dispersal. Understanding how ani-
mals navigate the landscape is therefore key to the development
of effective wildlife conservation strategies, especially across
fragmented landscapes, where heterogeneity constrains animal
movement. The degree to which the landscape facilitates ani-
mal movement is described as connectivity (Singleton and
McRae2013). Connectivity assessments can therefore identify
potential areas where animal movement is more likely (move-
ment corridors) or where animal movement is obstructed
(movement barriers).
In urbanized landscapes, habitats are reduced, fragmented,
and disturbed by anthropogenic factors, leading to behavioral
adaptations in animals, including changes in movement pat-
terns (Ritzel and Gallo2020). In addition, varying configura-
tion patterns of the built environment and habitable areas,
fencing practices, and dense road networks further restrict the
movement of terrestrial mammals, which are constrained to
the available space in between urban infrastructure. Identifying
which features promote or hinder animal movement within
urban areas will also provide valuable knowledge for managing
human–wildlife conflict, the spread of zoonotic diseases, and
the seed dispersal of native and non- native plant species (Kays
et al.2015).
Fine- scale research on functional connectivity within urban
areas has been limited, especially for terrestrial mammals
(LaPoint et al.2015). To date, fine- scale functional connectiv-
ity has been assessed for the white- footed mouse (Peromyscus
leucopus), European hedgehog (Erinaceus europaeus), fisher
(Martes pennanti), and red fox (Vulpes vulpes) (Munshi-
South 2012; LaPoint et al. 2015; Kimmig et al. 2020; App
et al.2022). Important urban features include roadsides, ceme-
teries, and residential areas with high tree canopy cover for
white- footed mice in New York City (Munshi- South 2012);
backyard vegetation structure for hedgehogs in Zürich,
Switzerland (App et al.2022); forested wetlands and highway
verges for fishers in suburban Albany, New York (LaPoint
et al.2013); and longitudinal movement along motorways and
railways for red foxes in Berlin, Germany (Kimmig et al.2020).
In this study, we analyzed functional connectivity for coy-
otes (Canis latrans) in Toronto, Canada, given that they are
wide- ranging terrestrial mesopredators and have the potential
to be an umbrella species (Breckheimer et al.2014). Coyotes
can thrive in highly developed landscapes by displaying strong
selection for urban parks and natural fragments, spending as
much as 75% of their time in these areas (Gehrt et al.2009;
Mitchell et al.2015; Thompson et al.2021). With home ranges
between 3 and 26 km2 and the capacity to move at rates of up
to 30 km per day (Gehrt et al.2009; Mueller et al.2018), urban
coyotes exploit both natural and anthropogenic resources
1Department of Ecology and Evolutionary Biology,University of Toronto,
Toronto, Canada *(tiziana.gelmicandusso@utoronto.ca);2Toronto and
Region Conservation Authority, Vaughan, Canada;3Department of
Environmental and Life Sciences,Trent University, Peterborough,
Canada;4Ontario Ministry of Natural Resources and Forestry,
Peterborough, Canada;5Government of Northwest Territories, Fort Smith,
Canada
RESEARCH COMMUNICATIONS 1 of 9

Front Ecol Environ doi:10.1002/fee.2633
TA Gelmi- Candusso et al.
(Bateman and Fleming2012), adjust their behavior in response
to landscape fragmentation and human presence through tem-
poral and spatial avoidance (Tigas et al. 2002), and differen-
tially select between urban zones (Mitchell et al. 2015). In
addition, coyotes are socially complex, forming territorial
packs with offspring dispersing yearly to form new packs or
join existing packs (Gese et al.1996). Therefore, factors such as
diel cycles, biological seasons, and social status influence coy-
ote habitat selection and how they travel through the urban
matrix (Mitchell et al.2015; Ellington et al.2020), potentially
affecting the degree of connectivity in urban areas, as reported
for other species (Zeller et al.2020).
With the aim of identifying urban features promoting or hin-
dering connectivity for coyotes and of assessing the dynamism of
that connectivity across temporal scales and demographic traits,
we analyzed functional connectivity within the metropolitan
Toronto region using an omnidirectional connectivity approach
(McRae et al. 2016) at a 30- m resolution. We described the urban
landscape using satellite data, census data, and the continuous
distance to various linear features as covariates, and assessed coy-
ote selection for these covariates using location data from 27
coyotes. Subsequently, we generated a resistance map for the
study area following the differential selection of the urban land-
scape covariates. To assess the spatiotemporal dynamism of con-
nectivity, we integrated into the analysis the selection of the
urban landscape in interaction with temporal and demographic
factors: namely, diel cycles, biological seasons, climate seasons,
social status, and age/sex of individuals. Ultimately, we identified
key features and areas where animal movement should be man-
aged to reduce the fragmenting effect of urban infrastructure,
and how connectivity may change in response to temporal and
demographic dynamics.
Methods
Study area
We conducted our study in the densely populated Greater
Toronto Area (GTA; 4334 people per km2, according to the
2016 census (Statistics Canada 2017) and the urban portion
of the Regional Municipality of Peel (2348 people per km2,
2016 census) in Ontario, Canada (Figure1). e study area
contained >1500 urban landscaped- green areas (eg cemeteries,
orchards, golf courses, parks, residential yards) scattered
throughout a diversely developed landscape. A distinctive feature
of the study area was the presence of multiple power line
corridors that travel transversely across the region and pass
through open green areas. In addition, the area was charac-
terized by ve river valleys with forested riverbanks running
longitudinally through areas of high human population density,
intersecting the study area. e region surrounding the study
area was characterized by a comparatively low- density human
population (<300 people per km2), protected natural areas,
and agricultural land.
Figure 1. Global positioning system (GPS)- collar location data of coyotes (n = 27) sampled between 2012 and 2021 (black circles) superimposed on a
land- cover map of the study area within the Greater Toronto Area (GTA) (gray border) and its surroundings, reclassified from the Southern Ontario Land
Resource Information System (v3.0). Online visualizations of maps are available at https:// story maps. arcgis. com/ stori es/ 92fe4 ecaf4 8e41f 69022 09fe5
ae5abcc.
2 of 9 RESEARCH COMMUNICATIONS

Front Ecol Environ doi:10.1002/fee.2633
Dynamic urban functional connectivity analysis
Coyote movements
Coyotes (n = 27; Figure1) were monitored between 2012 and
2021 for 245 ± 136 days (mean ± standard deviation)
(AppendixS1: TableS1). Coyotes were live- trapped with padded
foothold traps by the Ontario Ministry of Natural Resources
and Forestry following methods approved by their Wildlife
Animal Care Committee (protocols 75- 12, 75- 13, 75- 14) or
captured with nets by the Toronto Wildlife Centre; coyotes were
then tted with self- releasing GPS- collars (Lotek Wildcell SG,
Newmarket, Canada) that recorded location data, which were
resampled following the median sampling frequency in order
to maintain a constant sampling frequency for each individual
(1–3 hours; AppendixS1: TableS1). e data were well balanced
in terms of demographic traits (12 females/15 males, 19 adults/
eight subadults, 22 residents/15 transients; Appendix S1:
Table S1). Residents and transients were distinguished based on
movement characteristics following ompson et al. (2021).
Habitat selection model
We analyzed habitat selection by coyotes in the urban land-
scape using the step- selection function (SSF) method developed
by Fortin et al. (2005). We then estimated the relative selec-
tion strength (RSS; Avgar et al. 2017) for each covariate
included in the habitat selection model (Figure2). Four urban
landscape covariates were included as xed factors: vegetation
density (normalized dierence vegetation index or NDVI),
human population density, impervious surface, and linear fea-
tures (AppendixS1: Table S2a; Figures S1 and S2). To measure
the spatiotemporal dynamic responses of coyotes, we included
the interaction of the xed variables with three temporal
scales (diel cycles, biological seasons, and climate seasons)
and three demographic traits (coyote age, sex, and social
status) (Appendix S1: Table S2b). All data processing and
analysis for the habitat selection models were conducted in
R (v4.1.0; R Core Team 2021).
Step- selection function
From consecutive GPS- collar locations, we calculated the
turning angle and step length with the steps_by_burst()
function from the R package amt (Signer et al. 2019). Aer
tting the distributions to observed step lengths and turning
angles, we generated nine random available steps for each
Figure 2. (a) Relative selection strength (RSS) of the environmental attributes (ie vegetation density, human population, impervious surface, and linear
features: low- , medium- , and high- traffic roads; hiking paths; and public service lines) in interaction with coyote social status (SS), diel cycles (DC), biolog-
ical seasons (BS), and climate seasons (CS). (b) Habitat suitability map from the habitat selection analysis. (c) Workflow for estimating connectivity from
GPS- collar location data and satellite images using the RSS to estimate resistance values and the habitat suitability map to define source nodes. Online
visualizations of maps are available at https:// story maps. arcgis. com/ stori es/ 92fe4 ecaf4 8e41f 69022 09fe5 ae5abcc.
RESEARCH COMMUNICATIONS 3 of 9

Front Ecol Environ doi:10.1002/fee.2633
TA Gelmi- Candusso et al.
observed step using the random_steps() function from the
R package amt (Signer et al. 2019). We standardized the
xed variables included in the model and extracted their
values at the endpoint of each step. We checked for col-
linearity between covariates using the vif() function from
the R package DAAG (Maindonald and Braun 2020). To
t the SSF, we used the glmmTMB() function from the R
package glmmTMB (Magnusson et al.2021) to run a Poisson
generalized linear mixed model with random intercepts for
individual ID and step ID, and random slopes for individ-
uals, which is the likelihood equivalent of a conditional
logistic regression (Mu et al. 2020).
Model specification
We conducted model selection using a combination of
Akaike’s information criteria (AIC) and null hypotheses
testing, comparing the models to a null model including
only random intercepts for individual ID and step ID. e
model containing only the xed variables describing the
urban landscape with the lowest AIC score was used as the
base model for selecting from the models containing inter-
action eects with temporal scales and demographic traits.
Subsequently, we selected the models including interaction
eects that had an AIC ≤ 2 as compared to the base model
and at least one signicant interaction eect (P < 0.05).
Relative selection strength
Our habitat selection analysis estimated the relative prob-
ability of an animal occurring in a unit of space (hereaer,
space unit). Each covariate included in the model inuenced
this probability dierently along the animal’s path. erefore,
the RSS function was applied to account for the fact that
individual coyotes encounter dierent values of these covar-
iates on each space unit within their path (Avgar et al.2017).
e RSS used the coecient estimate of each covariate,
given by the habitat selection model, to assess the probability
of an animal being present in a space unit as the covariate’s
value changed, with all other covariates held constant.
Consequently, to measure the RSS, we generated a dataset
of 200 equally spaced values within the covariate’s scaled
range. When interaction eects were included in the habitat
selection model, the RSS was given by:
where Δhi is the dierence between the mock value at loca-
tion x1 and the reference value (ie the average attribute
value across the landscape), βi is the coecient estimate of
the attribute under analysis, βij is the estimate of the inter-
action eect, and hj is the value of the interaction component
at location x1 (1 or 0, as the interaction components were
in binary form). e regression between the RSS values was
estimated across each scaled covariate using a relative selec-
tion curve (Figure2a), which was used to predict the
resistance of each cell in the study area across temporal
scales and demographic traits (Appendix S1: Figure S3).
Connectivity analysis
Despite their generalist and highly adaptive nature, coyotes
still use natural areas more than other areas within cities
(Gehrt et al. 2009; ompson et al. 2021), meaning that they
likely tend to stay within highly suitable areas. erefore, to
dene the habitat patches acting as source nodes in our con-
nectivity analysis, we estimated the habitat suitability index
(HSI) across our study area. To do so, we followed the pro-
portional hazard model described by Manly et al.(2002), using
the SSF output (Figure2b). We considered habitat patches
(ie source nodes) as cells (30 m × 30 m) within our study
area that had HSI values in the top 1%. ese habitat patches
were used in the connectivity model as sources and targets
for the current ow traveling through the resistance matrix.
To generate the resistance matrices, we estimated the resist-
ance value for each cell within our study area as the sum of the
inverse RSS value of each environmental attribute at each cell
(30 m × 30 m; AppendixS1: FigureS3). We incorporated the
resistance contribution of each environmental attribute sepa-
rately, using the RSS instead of the HSI, to be able to incorpo-
rate the dynamism attributed by the coyote response to each
environmental attribute and their spatiotemporal changes in
value across the study area, as well as for each temporal and
demographic factor.
To estimate the resistance value R at each cell i, we used the
intercept and slope of the log- RSS curve of each covariate to
estimate the RSS value of each environmental covariate given
its value at each cell i, following:
where b0 is the intercept of the log- RSS curve reflecting the
variation in selection strength with respect to the scale of the
covariate (Figure2a), bx is the effect size of the covariates on
the log- RSS curve, Xi is the value of the covariate at cell i, by is
the effect size of the temporal scale or demographic trait added
as an interaction effect to the model (eg social status), Yi is the
binary value of the interaction covariate (eg resident = 0, tran-
sient = 1), and bx:y is the interaction effect of the non-
environmental covariate.
We assessed connectivity between the source nodes and
across the resistance values of our study area (Figure2c) using
the Omniscape package in Julia (Landau et al.2021). Omniscape
applies circuit theory omnidirectionally, evaluating connectiv-
ity, in terms of current flow, between every possible source
patch within a moving window (Landau et al.2021). The mov-
ing window radius was set to 15 km, which represented the
maximum daily net displacement of coyotes we monitored
(unpublished data). The normalized output was calculated by
dividing the cumulative current flow by the flow potential of
(Equation1),
RSS
=Δhi
(
βi+β
ijhj
[
x1
])
(Equation2),
R
i=∑
∞
n=i
exp
(
1−
(
b0+bxXi+byYi+bx:yXiYi
))
4 of 9 RESEARCH COMMUNICATIONS

Front Ecol Environ doi:10.1002/fee.2633
Dynamic urban functional connectivity analysis
the cell. Areas with high normalized current flow values repre-
sent areas where current flow is channeled (ie predicted move-
ment corridors), whereas areas with low values identify areas
acting as barriers to movement.
Results
Habitat selection
e model with the lowest AIC score included all of the
urban landscape covariates included in the analysis, and revealed
that coyotes signicantly selected areas with higher vegetation
index values and lower human population density values, as
well as areas with less impervious surface and areas closer to
hiking trails and public service roads, while avoiding low- and
medium- trac roads and to a lesser extent high- trac roads
(Figure2a; Appendix S1: Table S3). Including the interaction
with diel cycles, social status, biological seasons, and climate
seasons improved the explanation of model variability, but
including age or sex did not. However, interactions with bio-
logical seasons and climate seasons were not signicant
(Figure2a; AppendixS1: Table S3), and therefore these inter-
action factors were excluded from the connectivity analysis.
In contrast, diel cycles and social status both had a signicant
inuence on selection. During daytime, coyotes selected areas
with denser vegetation and areas farther away from low- trac
roads or hiking trails. Social status also inuenced habitat
selection by coyotes, given that transient individuals exhibited
a weaker avoidance of areas with a greater proportion of
impervious surfaces than resident individuals (Figure2a).
Across all models, the most suitable habitat areas were
located at the intersection of rivers along river valleys, as well
as smaller patches in golf courses and large urban parks (green
areas in Figure2b). The least suitable areas were located in
large commercial and industrial clusters (dark blue areas in
Figure2b).
Connectivity analysis
e connectivity assessment showed river valleys, smaller
streams with protected vegetation, linear parks or cemeteries,
golf courses, power line corridors, vegetation strips along
highways, and railways to be key for connectivity (Figure3a
[i, ii, iv]). In addition, the model output identied a dis-
tinction between nature- rich and nature- poor residential
areas (Figure3a [iii, iv]), with dense vegetation along res-
idential roads and in yards being key features for connec-
tivity. Connectivity values were lowest in large commercial
and industrial clusters (Figure3a [iii]).
Connectivity across the landscape and the predicted move-
ment corridors changed in the two models when diel cycles
(Figure3b) and coyote social status (Figure3c) were included.
In the diel cycle model, our results showed that residential
areas, especially nature- rich ones, became more permeable to
coyote movement during nighttime, with some predicted
movement corridors shifting to shorter but less sheltered
paths. Similarly, in the social status model, nature- rich residen-
tial areas were more permeable for transient coyotes than for
resident coyotes, thereby predicting movement corridors for
transient coyotes through less suitable areas.
Discussion
Urban features important for connectivity
Our ndings suggest that vegetation cover was the key factor
for coyote connectivity in our study area, as connectivity
values were highest when urban features included more
dense vegetation (ie higher NDVI values). For example, three
area types—residential areas composed of yards with diverse
vegetation (Figure3a [iv]); forested riverbanks (Figure3a
[ii]); and open green areas with patches of complex vege-
tation structure, such as golf courses (Figure3a [ii])—all
had higher connectivity values. erefore, common practices
(such as lawn maintenance), or common applications across
cities (such as wetland removal and river canalization), reduce
the number of sheltered paths available for coyotes to move
around a city. Because vegetation structural complexity is
key for the connectivity of urban small mammals (Munshi-
South 2012; LaPoint et al. 2013; App et al. 2022), we can
expect such practices to be detrimental to the connectivity
of local smaller mammals as well.
Similar to other carnivores (Kimmig et al.2020) and smaller
mammals (Munshi- South 2012; LaPoint et al. 2013; App
et al.2022) in other cities, linear features were also key for coy-
ote connectivity. The presence of linear features within green
areas, such as hiking trails or public service lines, further made
these green areas more permeable to animal movement. Public
service lines (such as railways and power lines) were key for
facilitating coyote movement across the landscape and across
all spatiotemporal sets of models. Our analysis suggests that
railways, by running uninterrupted across the urban matrix,
can be used by coyotes for longitudinal movement, especially
when vegetation shelter is present along the tracks and within
the protective fencing (Figure3a [i, iv]), but different types of
fencing will create species- specific degrees of barriers to trans-
versal movement. Similarly, power lines (Figure3a [i]), posi-
tioned in prescribed regulated open areas, create corridors of
green areas potentially important for longitudinal movement
of coyotes, as has been observed for other mammals, reptiles,
and birds preferring open habitat (Richardson et al.2017). Yet
power line corridors are frequently intersected by roads, which
diminishes their effectiveness for animal movement because of
the increased mortality risk associated with roadways (Gunson
et al.2011). Thus, these urban features offer great potential for
enhancing urban connectivity, provided that the segmentation
of these components is reduced, for example, by the installa-
tion of road underpasses and the creation of linear parks or
trails beneath power lines.
The key features hindering connectivity in our study area
were medium- and high- traffic roads and industrial/
RESEARCH COMMUNICATIONS 5 of 9

Front Ecol Environ doi:10.1002/fee.2633
TA Gelmi- Candusso et al.
commercial areas. For high- traffic roads, our model did not
consider the presence of underpasses and overpasses, which
certainly attenuate their fragmenting effect (Denneboom
et al.2021). Nonetheless, even in the absence of underpasses
and overpasses, our model shows that maintaining dense
green areas along highways (Figure3a [ii]) would promote
longitudinal movement of coyotes, as was found for shorter-
ranging predators in other cities (Kimmig et al. 2020).
Clusters of commercial/industrial infrastructure strongly
reduced connectivity across large areas (Figure3a [iii]). Such
clusters should be specifically addressed by management pol-
icies across cities, because they are existing areas where
movement paths can be created with only a few small
changes. For example, most commercial and industrial clus-
ters already contain large open spaces such as parking lots,
unused during times of closure and available for animal
movement, but these sites lack the necessary vegetation struc-
tural complexity to provide shelter and resource availability.
Interspersing green areas within parking lots, in the form of
hedgerows, or allowing public service features to cross
through, such as open corridors beneath power lines or linear
parks, would enhance functional connectivity.
Temporal changes in connectivity
During nighttime, coyotes’ overall selection of more vegetated
and less populated areas decreased. e largest temporal
dierence was found for low- trac roads, as these were
avoided during daytime but preferentially used during night-
time, a result that further explains the nighttime selection
of residential areas by coyotes in Toronto (ompson
et al. 2021) and in Denver, Colorado (Poessel et al. 2016).
Figure 3. Connectivity shown as normalized current density (NCD) between source nodes (ie areas within the top 1% of habitat suitability index [HSI] values).
Higher connectivity values predict movement corridors and lower connectivity values predict movement barriers. (a) Connectivity according to the
base model (only environmental attributes). Examples with corresponding orthophotos (City of Toronto2019): (i) power line and railway over a high-
density residential area; (ii) highway over a residential area between a ravine and a golf course; (iii) commercial area and nature- poor residential area;
(iv) nature- rich residential area, railway, and beach. Basemap credit: City of Toronto, Province of Ontario, Esri Canada, Esri, HERE, Garmin, SafeGraph,
FAO, METI/NASA, USGS, EPA, NPS, NRCan, and Parks Canada. (b and c) Difference in connectivity (normalized current density difference [NCDd])
between (b) night and day (diel cycle) and (c) transient and resident coyotes (social status). Online visualizations of maps are available at https:// story
maps. arcgis. com/ stori es/ 92fe4 ecaf4 8e41f 69022 09fe5 ae5abcc.
6 of 9 RESEARCH COMMUNICATIONS

Front Ecol Environ doi:10.1002/fee.2633
Dynamic urban functional connectivity analysis
In our connectivity model, this temporal difference in
selection translated into residential roads acting as key
movement paths at night for coyotes. This finding may be
applicable to other urban wildlife that tend to be more active
at night, especially those, such as opossums, raccoons, and
skunks, that use nocturnality to access areas with higher
human population density and greater impervious surface
cover (Gallo et al.2022). As residential roads are key move-
ment paths during the temporal peak of urban wildlife activ-
ity, increasing residential road permeability to animal
movement should be addressed, especially when residential
areas are located between remnant natural areas. To improve
animal movement across residential areas, our results sug-
gest that increasing vegetation complexity along roadsides or
introducing linearly shaped green areas, such as those shown
in Figure3a (i), would be beneficial. This finding may be
applicable to other urban predators as well, given that habi-
tat structural complexity affects the occurrence of both wild
felids and canids (Stobo- Wilson et al.2020). Furthermore,
speed limits and strategic maze- like one- way streets, as seen
in the downtown residential areas of our study area, may
help reduce traffic and mortality risk within these crucial
nighttime paths.
Our results support the incorporation of temporal scales,
such as diel cycles, into assessments of landscape connectivity.
This temporal effect was strongest in areas with a higher den-
sity of low- traffic roads, where humans are usually most pres-
ent, likely due to an avoidance of human daily activity patterns
(Tigas et al. 2002). Avoidance of human- associated landscape
features has been found for coyotes in Chicago (Magle
et al.2014) and, as shown in the present study and previously
for red foxes in Berlin (Kimmig et al. 2020), avoidance of
humans can also influence the movement paths selected by
animals to travel through the urban matrix. Therefore, given
the strong temporal patterns in human presence and abun-
dance in urban areas, future studies on urban connectivity
should consider the potential resolution lost and bias intro-
duced when data are pooled from different temporal periods,
especially for predominantly nocturnal animal species (Gallo
et al.2022).
Seasonality scales were not relevant for the connectivity
of coyotes, despite having an important effect on habitat
selection, as reported in previous work on urban systems
(Ellington et al. 2020; Thompson et al. 2021). Although
including biological seasons and climate seasons explained
variability in coyote habitat selection better than the base
model, the strength of the interaction effect was insufficient
to translate into changes in landscape connectivity. Having
year- round water/resource availability and diet adaptability
likely reduced the effect of seasonality for coyotes in our
study area. However, these factors may be important for
other urban wildlife more sensitive to seasonality, such as
those relying on foliage cover as shelter from predators
(Fullman et al. 2021), those strongly dependent on fruit
availability (Ciudad et al. 2021), or those in cities where
water availability is seasonal, as seen in non- urban environ-
ments (Osipova et al.2019).
Demographic effects on connectivity
Coyote age and sex had no eect on habitat selection or
landscape connectivity, but an individual coyote’s social status
inuenced both habitat selection and connectivity. Transient
coyotes exhibited weaker avoidance of impervious surfaces
than resident coyotes, further explaining the dierence in
the degree of avoidance of residential areas observed by
ompson et al. (2021).
In our connectivity model, weaker avoidance of impervious
surfaces by transient coyotes resulted in higher connectivity val-
ues for residential, industrial, and commercial areas. Analogous
to the pressure that human presence exerts on coyotes during the
daytime, the presence of territorial resident coyotes likely rele-
gates non- territorial transient coyotes to suboptimal habitat
(Mitchell et al.2015). Our results suggest that spatial partitioning
of the landscape by coyotes and their territory- defending behav-
ior (Gese2001) may be strong enough to alter landscape connec-
tivity for transient coyotes, which could increase their exposure
to humans and the potential for human–wildlife conflict. These
results highlight the importance of incorporating the demo-
graphic dimension into connectivity assessments (Drake
et al. 2022), especially for territorial animals such as coyotes,
foxes, and raccoons, for which spatial partitioning of the land-
scape and intraspecific avoidance are strong. For example, for
coyotes in our study area, residential areas were less important
for connectivity when social status was excluded as an interaction
effect. Instead, when social status was included in the assessment,
the results showed that residential areas were important for the
most mobile and exposed portion of the coyote population (ie
transient coyotes; Mitchell et al. 2015). Therefore, addressing
connectivity across residential, industrial, and commercial areas,
for example by promoting sheltered paths such as linear parks or
increasing vegetation density along other linear features, becomes
important because such changes may help mitigate human–wild-
life conflict as transient coyotes move through suboptimal, highly
urbanized habitat.
Conclusion
We identied the factors inuencing landscape connectivity
for coyotes at a ne- scale resolution (30 m × 30 m) over
an 80- km- wide metropolitan region, taking into account the
spatiotemporal dynamics of coyote habitat selection, in a
landscape characterized by a network of dierently permeable
linear features, and the combined presence of humans,
buildings, and vegetation at dierent scales. While linear,
densely vegetated areas across the urban matrix were con-
sistently important across spatiotemporal scales, habitat
selection shied as coyotes avoided human daily activity
and the territorial pressure of conspecics, creating spatio-
temporal patterns in the predicted movement corridors.
RESEARCH COMMUNICATIONS 7 of 9

Front Ecol Environ doi:10.1002/fee.2633
TA Gelmi- Candusso et al.
Overall, our findings suggest that, to improve connectivity,
cities can enhance existing potential movement corridors by
increasing plant structural complexity in key infrastructure,
including along public service lines and highways, in urban
parks, and within parking lots of extensively impervious areas
(such as industrial and commercial clusters).
Acknowledgements
We acknowledge the contribution of all employees of the Ontario
Ministry of Natural Resources and Forestry and the Toronto
Wildlife Centre involved in collaring and tracking coyotes in
the Greater Toronto Area (GTA), including A Wight. We ac-
knowledge the German Research Foundation (DFG) for sup-
porting this project through a research fellowship and the
University of Toronto for hosting. Funding was provided by
the Canada Research Chair in Spatial Ecology (to M- JF) and
DFG Research Fellowship GE 3103/1- 1 (to TAG- C). Author
contributions: TAG- C and M- JF conceived the study; TAG- C,
ATMC, and CAT contributed to the computational framework;
TAG- C implemented the computational framework, analyzed
the data, designed the gures, and draed and edited the man-
uscript; CAT, AADM, TJW, and BRP collected and curated
the coyote tracking data, and provided comments and edits to
the manuscript; M- JF supervised the development and ndings
of the work, and contributed to the writing and editing of the
manuscript and gures. All authors provided critical feedback
and helped shape the research, analysis, and manuscript.
Data Availability Statement
e data and code that support the ndings of this study
are openly available in Zenodo, at http:// doi. org/ 10. 5281/
zenodo. 10419385, reference number 10419385.
References
App M, Strohbach MW, Schneider A- K, and Schröder B. 2022.
Making the case for gardens: estimating the contribution of urban
gardens to habitat provision and connectivity based on hedgehogs
(Erinaceus europaeus). Landscape Urban Plan 220: 104347.
Avgar T, Lele SR, Keim JL, and Boyce MS. 2017. Relative selection
strength: quantifying effect size in habitat- and step- selection
inference. Ecol Evol 7: 5322–30.
Bateman PW and Fleming PA. 2012. Big city life: carnivores in urban
environments. J Zool 287: 1–23.
Breckheimer I, Haddad NM, Morris WF, et al. 2014. Defining and
evaluating the umbrella species concept for conserving and restor-
ing landscape connectivity. Conserv Biol 28: 1584–93.
City of Toronto. 2019. Orthophotos [digital resource: raster]. Toronto,
Ontario: Land Information Toronto. https:// mdl. libra ry. utoro nto.
ca/ colle ctions/ geosp atial- data/ toron to- ortho photos- 2019.
Retrieved 9 Mar 2023.
Ciudad C, Mateo- Sánchez MC, Gastón A, et al. 2021. Landscape
connectivity estimates are affected by spatial resolution, habitat
seasonality and population trends. Biodivers Conserv 30:
1395–413.
Denneboom D, Bar- Massada A, and Shwartz A. 2021. Factors affect-
ing usage of crossing structures by wildlife – a systematic review
and meta- analysis. Sci Total Environ 777: 146061.
Drake J, Lambin X, and Sutherland C. 2022. The value of considering
demographic contributions to connectivity: a review. Ecography
2022: e05552.
Ellington EH, Muntz EM, and Gehrt SD. 2020. Seasonal and daily
shifts in behavior and resource selection: how a carnivore navi-
gates costly landscapes. Oecologia 194: 87–100.
Fortin D, Beyer HL, Boyce MS, et al. 2005. Wolves influence elk
movements: behavior shapes a trophic cascade in Yellowstone
National Park. Ecology 86: 1320–30.
Fullman TJ, Wilson RR, Joly K, et al. 2021. Mapping potential effects
of proposed roads on migratory connectivity for a highly mobile
herbivore using circuit theory. Ecol Appl 31: e2207.
Gallo T, Fidino M, Gerber B, et al. 2022. Mammals adjust diel activity
across gradients of urbanization. eLife 11: e74756.
Gehrt SD, Anchor C, and White LA. 2009. Home range and land-
scape use of coyotes in a metropolitan landscape: conflict or coex-
istence? J Mammal 90: 1045–57.
Gese EM. 2001. Territorial defense by coyotes (Canis latrans) in
Yellowstone National Park, Wyoming: who, how, where, when,
and why. Can J Zool 79: 980–87.
Gese EM, Ruff RL, and Crabtree RL. 1996. Social and nutritional fac-
tors influencing the dispersal of resident coyotes. Anim Behav 52:
1025–43.
Gunson KE, Mountrakis G, and Quackenbush LJ. 2011. Spatial wildlife–
vehicle collision models: a review of current work and its application to
transportation mitigation projects. J Environ Manage 92: 1074–82.
Kays R, Crofoot MC, Jetz W, and Wikelski M. 2015. Terrestrial animal
tracking as an eye on life and planet. Science 348: aaa2478.
Kimmig SE, Beninde J, Brandt M, et al. 2020. Beyond the landscape:
resistance modelling infers physical and behavioural gene flow
barriers to a mobile carnivore across a metropolitan area. Mol Ecol
29: 466–84.
Landau VA, Shah VB, Anantharaman R, and Hall KR. 2021.
Omniscape.jl: software to compute omnidirectional landscape
connectivity. J Open Source Softw 6: 2829.
LaPoint S, Balkenhol N, Hale J, et al. 2015. Ecological connectivity
research in urban areas. Funct Ecol 29: 868–78.
LaPoint S, Gallery P, Wikelski M, and Kays R. 2013. Animal behavior,
cost- based corridor models, and real corridors. Landscape Ecol 28:
1615–30.
Magle SB, Simoni LS, Lehrer EW, and Brown JS. 2014. Urban preda-
tor–prey association: coyote and deer distributions in the Chicago
metropolitan area. Urban Ecosyst 17: 875–91.
Magnusson A, Skaug H, Nielsen A, et al. 2021. glmmTMB: general-
ized linear mixed models using template model builder. Vienna,
Austria: The R Foundation for Statistical Computing.
Maindonald JH and Braun WJ. 2020. DAAG: data analysis and
graphics data and functions. Vienna, Austria: The R Foundation
for Statistical Computing.
Manly BFJ, McDonald LL, Thomas DL, et al. 2002. Resource selection
by animals. Dordrecht, the Netherlands: Springer.
8 of 9 RESEARCH COMMUNICATIONS

Front Ecol Environ doi:10.1002/fee.2633
Dynamic urban functional connectivity analysis
McGuire JL, Lawler JJ, McRae BH, et al. 2016. Achieving climate connec-
tivity in a fragmented landscape. P Natl Acad Sci USA 113: 7195–200.
McRae BH, Popper K, Jones A, et al. 2016. Conserving nature’s stage:
mapping omnidirectional connectivity for resilient terrestrial
landscapes in the Pacific Northwest. Portland, OR: The Nature
Conservancy. https://doi.org/10.13140/RG.2.1.4158.6166.
Mitchell N, Strohbach MW, Pratt R, et al. 2015. Space use by resident
and transient coyotes in an urban–rural landscape mosaic. Wildlife
Res 42: 461–69.
Mueller MA, Drake D, and Allen ML. 2018. Coexistence of coyotes
(Canis latrans) and red foxes (Vulpes vulpes) in an urban land-
scape. PLoS ONE 13: e0190971.
Muff S, Signer J, and Fieberg J. 2020. Accounting for individual-
specific variation in habitat- selection studies: efficient estimation
of mixed- effects models using Bayesian or frequentist computa-
tion. J Anim Ecol 89: 80–92.
Munshi- South J. 2012. Urban landscape genetics: canopy cover pre-
dicts gene flow between white- footed mouse (Peromyscus leu-
copus) populations in New York City. Mol Ecol 21: 1360–78.
Osipova L, Okello MM, Njumbi SJ, et al. 2019. Using step- selection
functions to model landscape connectivity for African elephants:
accounting for variability across individuals and seasons. Anim
Conserv 22: 35–48.
Poessel SA, Breck SW, and Gese EM. 2016. Spatial ecology of coyotes
in the Denver metropolitan area: influence of the urban matrix.
JMammal 97: 1414–27.
R Core Team. 2021. R: a language and environment for statistical com-
puting. Vienna, Austria: The R Foundation for Statistical Computing.
Richardson ML, Wilson BA, Aiuto DAS, et al. 2017. A review of the
impact of pipelines and power lines on biodiversity and strategies
for mitigation. Biodivers Conserv 26: 1801–15.
Ritzel K and Gallo T. 2020. Behavior change in urban mammals: a
systematic review. Front Ecol Evol 8: 576665.
Signer J, Fieberg J, and Avgar T. 2019. Animal movement tools (amt):
R package for managing tracking data and conducting habitat
selection analyses. Ecol Evol 9: 880–90.
Singleton P and McRae B. 2013. Assessing habitat connectivity. In:
Craighead FL and Convis CL (Eds). Conservation planning: shap-
ing the future. Redlands, CA: ESRI Press.
Statistics Canada. 2017. Census profile. 2016 census. Catalogue no.
98- 316- X2016001. Statistics Canada: Ottawa, Canada. https://
www12. statc an. gc. ca/ census- recen sement/ 2016/ dp- pd/ prof/
index. cfm? Lang= E. Released 29 Nov 2017.
Stobo- Wilson AM, Stokeld D, Einoder LD, et al. 2020. Habitat struc-
tural complexity explains patterns of feral cat and dingo occur-
rence in monsoonal Australia. Divers Distrib 26: 832–42.
Thompson CA, Malcolm JR, and Patterson BR. 2021. Individual and
temporal variation in use of residential areas by urban coyotes.
Front Ecol Evol 9: 344.
Tigas LA, Van Vuren DH, and Sauvajot RM. 2002. Behavioral
responses of bobcats and coyotes to habitat fragmentation and
corridors in an urban environment. Biol Conserv 108:
299–306.
Zeller K, Lewison R, Fletcher R, et al. 2020. Understanding the
importance of dynamic landscape connectivity. Land 9: 303.
This is an open access article under the terms of the Creative Commons
Attribution-NonCommercial-NoDerivs License, which permits use and distribu-
tion in any medium, provided the original work is properly cited, the use is
non-commercial and no modifications or adaptations are made.
Supporting Information
Additional material can be found online at http://online-
library.wiley.com/doi/10.1002/fee.2633/suppinfo
RESEARCH COMMUNICATIONS 9 of 9