landscape ecology of eastern coyotes based ...people.cst.cmich.edu/gehri1tm/bio 691 - landscape...
TRANSCRIPT
Ecological Applications, 18(4), 2008, pp. 1014–1027� 2008 by the Ecological Society of America
LANDSCAPE ECOLOGY OF EASTERN COYOTES BASED ON LARGE-SCALEESTIMATES OF ABUNDANCE
ROLAND W. KAYS,1,4 MATTHEW E. GOMPPER,2 AND JUSTINA C. RAY3
1New York State Museum, CEC 3140, Albany, New York 12230 USA2Department of Fisheries and Wildlife Sciences, University of Missouri, Columbia, Missouri 65211-7240 USA
3Wildlife Conservation Society Canada, 720 Spadina Avenue #600, Toronto, Ontario M5S2T9 Canada
Abstract. Since their range expansion into eastern North America in the mid-1900s,coyotes (Canis latrans) have become the region’s top predator. Although widespread acrossthe region, coyote adaptation to eastern forests and use of the broader landscape are not wellunderstood. We studied the distribution and abundance of coyotes by collecting coyote fecesfrom 54 sites across a diversity of landscapes in and around the Adirondacks of northern NewYork. We then genotyped feces with microsatellites and found a close correlation between thenumber of detected individuals and the total number of scats at a site. We created habitatmodels predicting coyote abundance using multi-scale vegetation and landscape data andranked them with an information-theoretic model selection approach. These models allow usto reject the hypothesis that eastern forests are unsuitable habitat for coyotes as theirabundance was positively correlated with forest cover and negatively correlated with measuresof rural non-forest landscapes. However, measures of vegetation structure turned out to bebetter predictors of coyote abundance than generalized ‘‘forest vs. open’’ classification. Thebest supported models included those measures indicative of disturbed forest, especially moreopen canopies found in logged forests, and included natural edge habitats along water courses.These forest types are more productive than mature forests and presumably host more prey forcoyotes. A second model with only variables that could be mapped across the regionhighlighted the lower density of coyotes in areas with high human settlement, as well aspositive relationships with variables such as snowfall and lakes that may relate to increasednumbers and vulnerability of deer. The resulting map predicts coyote density to be highestalong the southwestern edge of the Adirondack State Park, including Tug Hill, and lowest inthe mature forests and more rural areas of the central and eastern Adirondacks. Together,these results support the need for a nuanced view of how eastern coyotes use forested habitats.
Key words: abundance; Adirondack State Park, New York; Canis latrans; eastern coyote; fecal DNA;landscape ecology; noninvasive survey.
INTRODUCTION
Ecologists increasingly recognize the importance of
studying animal abundance over broad spatial scales to
provide a better understanding of relationships between
animals, natural habitats, and human disturbance
(Baum et al. 2003, Karanth et al. 2004). Conducting
these landscape-scale studies is problematic, however,
for animals that are difficult to census. The abundances
of mammalian carnivores, for example, are rarely
estimated because of their low densities, use of large
areas, and shy nature. As a result, landscape models for
these species generally derive from small scale studies of
a few individuals, or broader scale categorical (presence–
absence) rather than continuous (population size) data
(Fuller et al. 2001, Zielinski et al. 2005). These small-
scale or coarse data may fail to capture many of the
nuances of habitat use or be unrepresentative of larger
areas. Here we show how noninvasive survey techniques
may be used to assay the abundance of coyotes (Canis
latrans) across a large geographic landscape, and how
these data can be used to compare diverse predictive
models of their landscape use.
The assemblage of mammalian predators in north-
eastern North America has undergone extensive reor-
ganization over the past 150 years (Ray 2000).
Following the extirpation of native large carnivores,
coyotes expanded their range into the Northeast, such
that over the past half-century the species has become
widespread across the region (Parker 1995, Gompper
2002). The ecological consequences of this range
expansion remain unclear, and this lack of clarity is
amplified by uncertainty regarding the fundamental
landscape ecology of coyotes. Given their origin in open
landscapes of central and western North America the
relative suitability of eastern forest habitat for coyotes
has been repeatedly questioned (Crete et al. 2001, Richer
et al. 2002). Although coyotes have been recorded in
virtually all habitat types in New England they have
been predominantly associated with more open, human-
Manuscript received 23 February 2007; revised 5 November2007; accepted 19 November 2007. Corresponding Editor: M.Friedl.
4 E-mail: [email protected]
1014
modified habitats (Oehler and Litvaitis 1996, Way et al.
2004). The hypothesis that eastern forests are poor
habitat for coyotes is supported by results from eastern
Quebec where, compared with coyotes in forested areas,
individuals living in rural areas are more active at dens
(Tremblay et al. 1998), have smaller home ranges (Crete
et al. 2001) and higher densities (Richer et al. 2002),
presumably because of differences in available food. On
the other hand, Vermont and New York animals
preferred forests over rural areas (Person and Hirth
1991, Kendrot 1998) and coyotes in New Brunswick
forests showed no signs of poor nutrition (a sign of
suboptimal conditions [Dumond and Villard 2000]).
Furthermore, forest coyotes have marginally higher
annual survival than rural animals and demonstrate no
apparent selectivity for open habitats within their home
range (Crete et al. 2001). Overall, it is unclear from
previous research whether forest in northeastern North
America should generally be considered quality habitat
for coyotes.
The extent to which coyotes use the forests that
dominate eastern North America is important when
gauging their ecological role in the region and directly
reflects on broader conservation issues and evolutionary
questions. Some have suggested that Eastern coyotes
have filled a wolf (C. lupus) niche in the region because
of their regular predation on ungulates (Mathews and
Porter 1992, Ballard et al. 1999). However, if forests
have low carrying capacity for coyotes, as suggested by
Crete et al. (2001), then their ecological impact there
may be reduced, with the wolf niche remaining open in
the heavily forested regions typical of much of
northeastern North America. Thus, from a conservation
perspective, the potential for eastern coyotes to fill the
top-predator niche has a direct bearing on the proposed
reintroduction of wolves to the region (Mladenoff and
Sickley 1998, Paquet et al. 1999). Additionally, if eastern
coyotes have adapted to forested habitats, such a niche
shift from their origin as open-country animals raises
questions about the evolutionary mechanisms underly-
ing this adaptation, as has previously been discussed in
regards to apparent increases in body size in parts of
eastern North America (Thurber and Peterson 1991,
Lariviere and Crete 1993, Peterson and Thurber 1993,
Gompper 2002, Kyle et al. 2006).
To more fully understand the landscape ecology of
coyotes in forested regions of northeastern North
America, we report here on our survey of coyote
abundance across rural and forested landscapes in
northern New York. Our use of a noninvasive censusing
approach enabled us to estimate the variability in
abundance over large areas and thereby build land-
scape-level models of habitat use. These abundance-
based models allow us to evaluate the general suitability
of rural vs. forested habitats within the study area, as
well as the more specific features that explain patterns of
coyote abundance in a region. In particular, if eastern
forests represent unsuitable habitat we predict that
coyote abundance will be correlated negatively with
forest cover across the landscape and positively withanthropogenic features that describe the open rural
landscapes of this region.
METHODS
Field surveys
Fieldwork took place from 1998 to 2000 across 54sites in and around Adirondack State Park (ADK,
;25 000 km2) in northern New York, USA. ADK is thelargest park in the contiguous United States and
contains a broad range of habitat types, managementunits, human population densities, and resource extrac-
tion intensities (Jenkins and Keal 2004). Field sites werespread throughout the park and surrounding regions in
a variety of forested and open-canopy landscapes. Thesesites varied in their extent of anthropogenic modifica-
tion, from those characterized by relatively low levels offragmentation and human use to those situated insuburbanized landscapes or subject to intensive resource
extraction (forestry) and agriculture. We selected 54 sitesin three general landscape types broadly representative
of northern New York: interior forest landscapescharacteristic of much of the mid-elevation wildland
habitats of central ADK (n¼31 sites), logged landscapeswhere active forest extraction activities were taking place
(n¼ 12 sites), and suburban/agricultural landscapes (n¼11 sites) on the outskirts of the study area where the
landscape has undergone the most extensive changes inthe region due to a long history of human use.
At each site a 5-km transect was marked along hikingtrails or unpaved roads. We only used transects that had
forest cover immediately around them, although somesites were more broadly surrounded by rural or
suburban landscapes. We surveyed coyote presencealong each transect using two noninvasive techniques:
camera traps and scat surveys, with the later supple-mented by analyses of fecal-DNA. Data from camera
traps was found to be unreliable for identifying thepresence of coyotes (Gompper et al. 2006) and thus alldata presented in this study are derived from scat
surveys. Each 5-km transect was cleared of scats and wasthen walked once monthly for three consecutive summer
months. All non-bear (Ursus americanus) scats werecollected in paper bags and stored immediately at
�208C.
Genetic analyses
Species identification analyses were conducted on all
scats collected from a subset of sites (n ¼ 31 sites) thathad five or more suspected coyote scats. DNA was
extracted with Qiagen QIAmp DNA Stool Mini kit(Qiagen, Valencia, California, USA). To calibrate
subsequent PCR, we quantified the amount of DNA ina subset of 23 fecal samples using a Nanodropspectrophotometer (NanoDrop Technologies, Willming-
ton, Delaware, USA). To confirm species of origin, weused the MDO mitochondrial DNA PCR primers to
June 2008 1015COYOTE FOREST USE
amplify regions of genomic DNA derived from coyote
(Gompper et al. 2006; J. E. Maldonado, unpublished
data). We sequenced non-coyote DNA samples on an
ABI Prism 3700 DNA Analyzer (Applied Biosystems,
Foster City, California, USA) using the same mitochon-
dria-specific primer pair for the sequencing reactions.
Species identifications were made through alignment of
generated sequences with either a candidate species’
sequence found in Genebank, or with sequence derived
from the New York State Museum tissue collection. In
total, we tested 472 fecal samples and obtained sufficient
genetic material to confirm species of origin for 377
(79.9%), of which 335 (88.9%) were coyote.
DNA from coyote fecal samples was subsequently
genotyped using three microsatellite primer pairs
(FH2001, FH2062, and FH2140 [Mellersh et al. 1997,
Kohn et al. 1999]) to PCR-amplify canid-specific
tetranucleotide repeat microsatellite markers. We select-
ed these three loci because they had a high polymor-
phism information content, providing a low probability
of random match between multilocus genotypes within
populations (Kohn et al. 1999). Although three loci are
fewer than recommended for typical density estimates
(Waits et al. 2001), these loci provide sufficient precision
for the objectives of our study since we are not
attempting to compare the source individual for a fecal
sample across sites, but rather require only to derive an
accurate index of population abundance from the
relatively small number of fecal samples collected from
each site. After preliminary trials, we slightly modified
the 2140 primer to address artifactual ‘‘þA’’ addition to
PCR products by using the following primer sequences:
forward, 5-GTATGATGAGGGGAAGCCA-3; reverse,
5-GTTTCTTTGACCCTCTGGCATCTAGGA-3. The
forward primer of each pair was labeled with a
fluorescent dye as follows: 2001 (6-FAM), 2062
(HEX), and 2140 (NED). The final concentrations of
PCR components in 25-lL reactions were: 1X buffer
(Qiagen); 0.2 mmol each dNTP; 1.5 mmol MgCl2; 0.5
lmol each primer; 0.02 U/lL Taq polymerase (Qiagen);
‘‘TaqStart’’ antibody (Invitrogen, Carlsbad, California,
USA), 14:1 (antibody : Taq) molar ratio, for ‘‘hot start’’
PCR initiation; and 2.0 lL fecal DNA extraction (as
prepared from Qiagen Stool kit above). We ran negative
controls (lacking template DNA) with each set of
reactions. The PCR thermalcycling program used was
an initial cycle of 958C for 3 minutes; followed by 40
cycles of 948C for 30 s, 588C for 30 s, 728C for 45 s; and a
final cycle of 728C for 2 minutes. For 2062 primers, a
608C annealing temperature was used. For PCR
amplifications carried out using MDO primers, 0.4 lmol
of each primer was used (final concentration), and the
thermalcycling program was an initial cycle of 958C, for
3 minutes; 35 cycles of 948C for 30 s, 538C for 30 s, 728C
for 60 s; and a final cycle of 728C for 5 minutes. The
allele sizes of amplified genetic markers were determined
using an ABI 3100 Genetic Analyzer and Genotyper
v3.7 software (Applied Biosystems, Foster City, Cal-
ifornia, USA) at the Wadsworth Center, New York
State Department of Health, Albany, New York, USA.
Tissue samples from 39 adult coyotes obtained from
trappers in Otsego County, New York (;30 km south of
ADK) were used to assess the appropriateness of the loci
for the noninvasive analyses. We assumed these
individuals to be unrelated and used the program
GIMLET 1.3.3 (Valiere 2002) to calculate the probabil-
ity of identity (PID) (Paetkau and Strobeck 1994, Waits
et al. 2001) for unrelated (PIDun corrected for samples
size) and sibling (PIDs) individuals, thereby giving lower
and upper bounds for PID. Programs MICRO-CHECK-
ER 2.2.3 (Van Oosterhout et al. 2004) and GENEPOP
3.4 (Raymond and Rousset 1995) were used to quantify
heterozygosity and examine for possible Hardy-Wein-
berg and linkage disequilibrium.
Each coyote fecal sample was genotyped using
replicated amplifications that required each allele to be
seen at least twice prior to identifying a consensus
genotype for the sample. Likewise, homozygote geno-
types were amplified twice. If genotypes were not
confirmed by seeing each allele twice after two ampli-
fications the sample was removed from our analysis.
This approach is similar to the filtering method used by
Adams and Waits (2007), who found a two-amplifica-
tion approach for homozygotes to ultimately produce
results comparable to more conservative repeated
amplification approaches, while significantly reducing
the effort and cost. This approach is particularly
appropriate for our study because at each site we
compare only a relatively small number of scat samples.
Furthermore, the use of tetranucleotide repeat micro-
satellites facilitated the identification of false alleles and
thus the grouping of identical and near-identical
genotypes. We calculated two measures of the mean
per-replicate probability of allelic dropout using pro-
gram GIMLET 1.3.3 (Valiere 2002). Following Prugh et
al. (2005),
PDj� ¼Djk
Ahetj
or PDj ¼Djk
Aj
where Djk is the number of amplifications with a missing
allele at locus j in replicate k, Ahetj is the number of
genotypes heterozygous at locus j, and Aj represents the
total number of consensus genotypes (including homo-
zygotes). False allele probabilities were calculated as
PFj ¼Fjk
Aj
where Fjk is the number of amplifications resulting in a
false allele at locus j in replicate k. Under a rule that
requires seeing each allele twice to confirm a genotype,
ðPDjÞ2 and ðPFjÞ2 represent the likelihood of a dropout
or false allele for each loci when replicated, and the sum
of these measures, Perr, represents the total probability
of an erroneous genotype at locus j. Perr, can be summed
across loci to calculate the total probability of an
ROLAND W. KAYS ET AL.1016 Ecological ApplicationsVol. 18, No. 4
erroneous multilocus consensus genotype for any sample
(Prugh et al. 2005).
Only samples for which all loci were resolved with all
alleles seen at least twice were included in calculations of
population size. Population size per site was calculated
as the number of unique genotypes per transect and then
regressed as a dependent variable against the number of
fecal samples collected and fully genotyped per site. The
ensuing regression equation was used to extrapolate the
number of coyote individuals per site based on the
number of putative coyote scats (Gompper et al. 2006)
collected per site.
Habitat modeling
We used an information-theoretic approach for
habitat model development and selection (Burnham
and Anderson 2002), using the number of coyote
individuals as the response variable and employing
multiple linear regression techniques. We formed five a
priori hypotheses to explain how local and landscape
habitat attributes would influence coyote abundance
based on relationships reported in the literature and our
experience in the study area (Table 1). To frame these
hypotheses we selected variables reflecting forest type,
open areas, edge habitats, and anthropogenic distur-
bance, since these have been shown to be important for
within-home-range habitat selection of eastern coyotes
(Caturano 1983, Major and Sherburne 1987, Litvaitis
and Harrison 1989, Person and Hirth 1991, Brundige
1993, Oehler and Litvaitis 1996, Kendrot 1998, Trem-
blay et al. 1998, Crete et al. 2001, Richer et al. 2002,
Way et al. 2004). Included among these were direct
metrics of vegetation structure measured at each site,
reflecting the openness and disturbance of a forest more
accurately than regional GIS layers. GIS layers were
downloaded from the New York State GIS clearing-
TABLE 1. Variables used for landscape models arranged under headings that represent five primary hypotheses that might explainhow local and landscape habitat attributes influence coyote abundance.
Variable type and name Variable description Scale Source and/or method
Local vegetation structure
HEIGHT mean height of trees � 10 cm dbh local field measurementVOLCWD mean volume of coarse woody debris local field measurementCANOPEN mean canopy openness local field measurementBASNAG mean basal area of snags local field measurement
Land cover: natural edges
dtWATER distance to nearest lake, river, orstream
local GIS (NYS 1:24 km Hydrography NetworkCoverages)
SHORE shoreline density for lakes, rivers, andstreams
landscape GIS (NYS 1:24 km Hydrography NetworkCoverages)
WETLAND proportion of area in wetland landscape GIS (NY-GAP Land Cover Map 8.6)NATFRAG average size of natural fragment as
delimited by roads, agriculture, orhuman developments
landscape GIS (NY-GAP Land Cover Map 8.6)
Land cover: forest type
FORCOV proportion of area in forest cover landscape GIS (NY-GAP Land Cover Map 8.6)CON proportion of area in conifer forest landscape GIS (NY-GAP Land Cover Map 8.6)DEC proportion of area in deciduous
forestlandscape GIS (NY-GAP Land Cover Map 8.6)
MIX proportion of area in mixed forest landscape GIS (NY-GAP Land Cover Map 8.6)
Anthropogenic
dtLOGRD distance to nearest logging road local GIS (digitized from USGS 100-km maps ascompiled from USGS 1:24 kmtopographic maps dated 1945–1980)
dtPAVED distance to nearest paved road local GIS (NYS ALIS roads layer)dtHOUSE distance to nearest house local GIS (NYS Office for Real Property Services
from the year 2000)HOUSE house density landscape GIS (NYS Office for Real Property Services
from the year 2000)PAVED density of paved county roads landscape GIS (NYS ALIS roads layer)LOGRD logging road density landscape GIS (digitized from USGS 100-km maps as
compiled from USGS 1:24 kmtopographic maps dated 1945–1980)
Physical
TRI terrain ruggedness index landscape calculated from digital elevation model(Riley et al. 1999)
SNOW snowfall average, winters 2002–2003and 2003–2004
landscape GIS (National Operational HydrologicRemote Sensing Center 2004)
Notes: Local-scale parameters represent averaged measures from nine stations distributed along each transect. Landscape-scaleparameters derive from GIS assessments of the landscape around trails using four different buffer scales: 0.5, 1, 5, and 10 km.
June 2008 1017COYOTE FOREST USE
house except for the GAP land use data which came
from the New York State Gap Analysis Program
(Cornell University, Ithaca, New York) and the snowfall
data, which was an average of actual snowfall from
winters 2002–2003 and 2003–2004 (National Operation-
al Hydrologic Remote Sensing Center 2004). No
remotely sensed snow data were available from the
exact years of our field surveys, so we consider these
averaged data an index of typical snowfall patterns for
the region. We avoided false impressions of local
accuracy by only using the snow variable after averaging
over larger scales (5-km and 10-km buffers). We
calculated the terrain ruggedness index (TRI) from a
digital elevation model following Riley et al. (1999). We
included a hypothesis that topography and snow cover
would impact summer coyote density since such factors
were highly variable in the study area and have been
shown to influence carnivores in general (Carroll 2007)
and were hypothesized to have slowed coyote coloniza-
tion of the area (Fener et al. 2005).
We had no direct measures of prey abundance across
the study area. Coyote diet across northeastern North
America consists of a combination of white-tailed deer
(Odocoileus virginianus) and smaller prey (chiefly lepor-
ids), as well as seasonal fruit (Hamilton 1974, Brundige
1993, Patterson et al. 1998, Gompper 2002; R. Kays, J.
Ray, and M. Gompper, unpublished data). Measures of
small prey abundance across the study area do not exist,
and indirect measures of deer abundance through
hunting returns (Nesslage and Porter 2001) were too
coarse and geographically limited for the purposes of
our analyses. We did not use measures of greenness,
derived from MODIS satellite imagery, as a surrogate
for prey abundance as others have (e.g., Mace et al.
1999, Carroll 2007), because of its high correlation with
deciduous cover in this landscape, but used instead
measures of vegetation structure that have been shown
to be related to prey abundance.
We initially considered 32 landscape variables, but
discarded many either because they were highly corre-
lated with others (Pearson’s correlation . 0.70), or
because inherent heteroscedacity. We transformed (log,
natural log, or square root) a subset of the final variable
set to stabilize and normalize their variances. In all, we
classified the 21 habitat variables a priori to test the five
classes of hypotheses (Table 1). Nine of these were local-
scale variables representing the averages of measures
taken at each of nine stations along each transect. The
remainder were landscape-scale variables measured
using ARCGIS in buffers surrounding each trail at four
scales (500 m, 1 km, 5 km, and 10 km), except for
snowfall, which was only averaged over 5-km and 10-km
scales. We selected the larger scales as an approximate
upper extreme for the radius of a coyote home range
(Brundige 1993) and the two smaller scales to capture
more fine-grained habitat selection. Models with inter-
action terms were not included to restrict the number of
candidate models and because we knew of no strong
biological explanations for putative interactions.
We used a multi-scale, three-stage modeling process to
test hypotheses regarding coyote abundance and to
select top models. First, we grouped variable sets into a
priori models representing the five hypotheses listed in
Table 1, as well as five additional combinations of
hypotheses drawn from the literature: all land cover, all
land cover þ vegetation structure, all land cover þphysical, vegetation structure þ anthropogenic, and
anthropogenic þ physical. These models were then
tested with data collected at the four scales, except for
the local vegetation data, which could not be extrapo-
lated past the local scale, giving a total of 37 total
variable sets. For each set, we used a combined process
of stepwise selection (P to enter at 0.05 and leave set at
0.10) and Akaike’s Information Criterion (AIC; Burn-
ham and Anderson 2002) to derive one model for each
variable set. For the latter, we ranked all possible models
of up to three terms based on AIC adjusted for small
sample size (AICc). We then used the AICc differences
between the best model and the other candidate models
(Di ¼ AICci� minimum AICc) to determine the relative
ranking of each model. In all cases, the result generated
from the stepwise analysis selected a model for which Di
� 2 (Burnham and Anderson 2002) we selected the mostparsimonious model (fewest parameters) that included
variables from the given hypothesis. At the end of stage
one we had models describing the explanatory power of
37 different hypothesis-scale combinations to predict
coyote abundance.
In the second stage, we ranked the candidate models
against one another based on AICc and Di as previously
described. We also calculated AICc weights, to assess the
strength of evidence that any particular model was the
best model in our set. The ranking of models in stage
two allowed for the direct comparison between hypoth-
eses and between scales.
In the third stage, we produced one best model to
describe coyote abundance in the region. To do so, we
investigated whether any of the top models that emerged
from stage 2 could be improved by forcing additional
variables from the other top models (�2 AIC), which
often represented variables from different hypotheses or
different scales (following Holloway andMalcolm 2006).
We then rescaled AIC values relative to the best model
(i.e., the model with the minimumAIC), and recalculated
AIC weights to settle on a final model. We examined the
relative importance of each variable by summing the
AICc weights across the top models emerging from stage
2 for any model including the variable in question
(Burnham and Anderson 2002). To map our model
results, we repeated the three-stage modeling process
using only variables for which we had region-wide GIS
coverage (i.e., excluding vegetation structure measures).
All analyses were performed using SAS (version 8.2; SAS
Institute, Cary, North Carolina, USA).
ROLAND W. KAYS ET AL.1018 Ecological ApplicationsVol. 18, No. 4
RESULTS
Coyote abundance estimates
Thirty-six Otsego County coyotes obtained from
trappers were genotyped at all three loci and three
individuals were genotyped at two loci. The number of
alleles per locus varied from four to nine, and mean
allele frequency varied from 0.11 to 0.25 (Tables 2 and
3). Average observed and expected heterozygosity was
0.65 and 0.74, respectively. Departures from Hardy-
Weinberg equilibrium were not significant (P . 0.05) for
each locus, and no cases of linkage disequilibrium were
detected (P . 0.05). The PIDun using all loci was 2.92 3
10�7 and the PIDs was 3.96 3 10�3.
We collected a total of 645 fecal samples from the 54
study sites. Four of the sites had no putative coyote
feces. We attempted to extract DNA from 472 of those
samples from 31 sites with more than five suspected
coyote’s feces. A subsample of extracts from 23 scat
samples had an average of 1.58 ng/uL of DNA. Of the
472 samples extracted we were able to genetically
identify the species of origin for 377 samples, with two
additional samples giving a genotype of unknown
species origin, and 93 failing to amplify. This resulted
in an 80.3% amplification rate for the MDO species ID
primer. Most (88.9%) of the identified feces were from
coyotes (Fig. 1). We attempted to genotype all 335
coyote feces for the microsatellite loci, reamplifying
samples until each allele was seen twice or the sample
failed twice. Marker 2001 amplified at a rate of 78.6% (n
¼2010 attempted amplifications), marker 2062 amplified
at a rate of 82.1% (n¼ 2016), and marker 2140 amplified
at a rate of 82.6% (n ¼ 2346). Average per-locus drop-
out rate was 0.129 6 0.014 (mean 6 SE), and the
average per-locus, per-replicate false allele rate was
0.0029 6 0.0011 (n ¼ 2382). Probability of obtaining a
false three-locus genotype following replication was
3.53% (Table 4), implying approximately nine incorrect-
ly genotyped samples from our pool of 259 samples.
Using an approach that requires seeing every allele
twice, we obtained 122 samples with usable genotypes
for all three microsatellites, 47 for two markers, 53 for
one marker, and 76 with no markers amplifying. In
addition, 37 samples amplified for a locus only once, and
were therefore not used because we were unable to verify
the genotype. Across the 25 sites with successfully
genotyped feces, mean number of coyote fecal samples
examined per site was 4.8 (range: 1–29) and mean
number of identified genotypes was 3.12 (range: 1–13).
The number of coyote genotypes was closely correlated
with the number of fecal samples (Fig. 2), indicating that
fecal counts can be used to census relative coyote
densities across sites. The relationship was best fit by the
power function y ¼ 1.0914x0.727 (r2 ¼ 0.93, P , 0.001).
Applying this formula to identify the number of coyotes
across all 54 sites revealed an average of 4.0 coyotes per
transect (64.0 [SD]; range 0–21).
Landscape models
The first stage of habitat model selection produced
candidate models for each scale resulting in 24 models
that described the explanatory power of 37 different
hypothesis-scale combinations predicting coyote abun-
dance (Table 5). Some model selection procedures
produced identical models for different sets (notably
WETLAND500 was repeated twice, SHORE10km three
times, VOLCWD þ CANOPEN þ BASNAG emerged
three times at different scales, as did CANOPEN þdtPAVED). In the case of land cover-forest hypotheses
at all scales, no significant models were produced.
Following stage 2 of the model selection process, four
models received substantial support (emerging from the
top nine hypotheses), with AIC scores within two units
of one another, all of which contained vegetation
TABLE 2. Variability of three microsatellites used to genotype Otsego County coyotes (n ¼ 39coyotes) obtained from trappers.
Locus No. alleles Allele frequency�
Heterozygosity
PIDun PIDsObserved Expected
FH2001 5 0.20 6 0.15 0.69 0.71 6.558 3 10�3 1.580 3 10�1
FH2062 4 0.25 6 0.11 0.64 0.71 7.890 3 10�4 6.744 3 10�2
FH2140 9 0.11 6 0.11 0.62 0.79 5.641 3 10�2 3.714 3 10�1
Note: Values represent allelic diversity and frequency, observed and expected heterozygosity,and the probability of identity of unrelated and related individuals.
� Mean 6 SD.
TABLE 3. Allele frequencies for three microsatellites amplified in Otsego County coyote tissuesobtained from trappers (n ¼ 39 coyotes).
Locus
Allelles
a b c d e f g h i
FH2001 0.038 0.077 0.218 0.397 0.269FH2062 0.141 0.385 0.295 0.179FH2140 0.250 0.069 0.083 0.069 0.014 0.125 0.333 0.042 0.014
June 2008 1019COYOTE FOREST USE
structure variables (vegetation structure, vegetation
structure þ land cover at all scales, and vegetation
structure þ anthropogenic at all scales). Three such
models demonstrated the highest AICc weights. Models
without vegetation structure variables had DAIC values
. 7 and AIC weights , 0.01. Models using the same
variables at different scales produced similar top models.
The exceptions were the inclusion of FORCOV at the 1-
km scale within the top model, whereas this variable was
unimportant at any other scale. STRSHORE replaced
WETLAND as an important variable at the two largest
scales.
In the final stage of the modeling process we produced
34 additional models by forcing additional variables into
the four top models emerging from stage 2. The top
three (Table 6) were within two AIC points of each
other, and therefore are viewed as roughly equal, while
all lesser models had Di . 2.6. We selected model 2 as
the optimal model because it is more parsimonious than
model 1 and was ranked only marginally lower. This
optimal model was:
Coyote abundance ¼ �12:16147þ 4:60206 3 VOLCWD
þ 6:73530 3 CANOPEN
� 16:84135 3 BASNAG
þ 15:20426 3 WETLAND1k
þ 14:88565 3 FORCOV1k:
This equation indicates coyote abundance is positively
associated with canopy openness and volume of coarse
woody debris at the stand level, and positively associ-
ated with amount of forest cover and wetland habitat
within a 1-km buffer of a trail, but negatively associated
with basal area of snags at the stand level. In alternative
models (Table 6), dtPAVED or SHORE10K were also
positively associated with coyote abundance. To rank
the relative importance of all variables across these 34
stage-3 models we summed the AIC weights for models
that included the variable in question, obtaining the
following: CANOPEN ¼ 1.0, WETLAND1K ¼ 0.699,
FORCOV1K¼ 0.674, SHORE10K ¼ 0.574, VOLCWD
¼ 0.559, BASNAG ¼ 0.559, dtPAVED ¼ 0.454.
When vegetation structure variables were eliminated
from the analyses to produce models that could be
mapped across the study area, the top nine models from
stage two were no longer considered, and variables such
as TRI, SNOW, and HOUSE, emerged as the most
important (model 10 and below; Table 5), with
dtPAVED remaining the only variable in common. In
the final stage of constructing models that could be
mapped we built 22 models, of which the top four are
presented in Table 6. All lesser models had Di . 2.4. The
model with the highest AIC score and weight emerging
from this process was:
Coyote abundance ¼ �9:72937þ 4:62711 3 dtPAVED
� 0:16438 3 TRI500
þ 0:01240 3 SNOW10k
� 3:76598 3 HOUSE5k
� 5:70695 3 dtHOUSE:
This equation shows that in the absence of vegetation
structure data, coyote abundance is positively associated
with distance to roads and negatively associated with
FIG. 1. Source species and percentage of 377 genotypedfecal samples collected from 54 independent transects in andaround the Adirondacks (New York, USA) during the summersof 2000–2002.
TABLE 4. Genotyping error rates for fecal samples.
Locus PDj�� n� PDj� n� PFj� n� ðPDjÞ2 ðPFjÞ2 Perr
2001 0.203 6 0.035 102 0.101 6 0.019 206 0.001 6 0.001 834 0.0102 1 3 10�6 0.010202140 0.143 6 0.022 132 0.092 6 0.015 205 0.003 6 0.002 960 0.0085 9 3 10�6 0.008472062 0.201 6 0.030 137 0.129 6 0.020 213 0.003 6 0.002 884 0.0166 9 3 10�6 0.01665
Total 0.0353 1.9 3 10�5 0.03533
Notes: PDj� is the per-sample dropout rate including heterozygotes only. PDj is the per-sample dropout rate including all data.PFj is the false allele rate for all samples and all replicates. ðPDjÞ2 and ðPFjÞ2 represent the probability of a dropout or falseallele, respectively, in the multilocus genotype under a see-each-allele-twice rule. Perr is the total probability of obtaining anerroneous genotype.
� Values shown are means 6 SE.� Number of fecal samples.
ROLAND W. KAYS ET AL.1020 Ecological ApplicationsVol. 18, No. 4
topographic relief at small scales, housing density at
larger scales, and distance to house, with a slightly
positive relationship to snow depth at large scales. The
relative importance of these variables, as represented by
summed AIC weights, were: dtPAVED ¼ 1.0, TRI ¼0.960, SNOW ¼ 0.734, HOUSE ¼ 0.703, dtHOUSE ¼0.683. Scale had no affect on these relative relationships.
Using this top model to map predicted coyote abun-
dance across northern New York shows a patchwork of
high and low densities, with lowest abundance toward
the center of ADK and higher abundance along its
periphery, especially in the southwest portion of the
park and in the Tug Hill area (Fig. 3).
DISCUSSION
Broadscale abundance surveys
We found coyotes to be virtually ubiquitous in the
rural and forested landscapes of northern New York,
but quite variable in their local abundance. This
variation would have been lost had we documented
only the presence or putative absence of coyotes since
only four of our 54 transects did not record coyote
FIG. 2. Regression plot of the number of coyote fecalsamples collected and genotyped at a site vs. the number ofcoyote individuals identified through genetic fingerprinting ofthe same samples. The relationship was best fit by the powerfunction y ¼ 1.0914x0.727 (r2 ¼ 0.93, P , 0.001).
TABLE 5. Best model results predicting coyote abundance from stages 1 and 2 of overall model selection criteria.
Modelno. Model type Scale K
AdjustedR2 AICc DAICc AICwt Variables in model
1 Veg struc þ Land cover-All local þ 1 km 4 0.352 164.813 0.000 0.156 CANOPEN, FORCOV1K,WETLAND1K
2 Veg struc local 4 0.351 164.892 0.079 0.150 VOLCWD, CANOPEN, BASNAG3–6 Veg struc þ Anthro local þ
all scales4 0.351 164.892 0.079 0.150 CANOPEN, dtPAVED
7 Veg struc þ Land cover-All 10 km 3 0.313 165.617 0.804 0.105 CANOPEN, SHORE10K8 Veg struc þ Land cover-All 5 km 3 0.291 167.341 2.528 0.044 CANOPEN, SHORE5k9 Veg struc þ Land cover-All 500 m 4 0.305 168.580 3.767 0.024 CANOPEN, FORCOV500,
WETLAND50010 Anthro þ Physical local þ 10 km 4 0.262 171.854 7.041 ,0.01 dtPAVED, SNOW10k, TRI10k11 Anthro þ Physical local þ 500 m 3 0.219 172.575 7.762 ,0.01 dtPAVED, TRI50012 Anthro local þ 5 km 4 0.237 173.685 8.872 ,0.01 dtPAVED, dtHOUSE, HOUSE5k13 Anthro þ Physical local þ 5 km 4 0.237 173.685 8.872 ,0.01 dtPAVED, dtHOUSE, HOUSE5k14 Anthro þ Physical local þ 1 km 3 0.200 173.870 9.057 ,0.01 dtPAVED, TRI1k15 Abiotic þ Landcover-All 1 km 4 0.226 174.439 9.626 ,0.01 WETLAND1K, FORCOV1K,
TRI1k16 Anthro local þ 10 km 4 0.224 174.581 9.768 ,0.01 dtPAVED, dtHOUSE, HOUSE10k17 Land cover-All 10 km 2 0.149 174.976 10.163 ,0.01 SHORE10K18 Land cover-Natural edge 10 km 2 0.149 174.976 10.163 ,0.01 SHORE10K19 Land cover-All þ Abiotic 10 km 2 0.149 174.976 10.163 ,0.01 SHORE10K20 Land cover-All 1 km 3 0.179 175.300 10.487 ,0.01 WETLAND1K, FORCOV1K21 Anthro local þ 500 m 3 0.172 175.722 10.909 ,0.01 dtPAVED, dtHOUSE22 Anthro local þ 1 km 3 0.172 175.722 10.909 ,0.01 dtPAVED, dtHOUSE23 Land cover-Natural edge 1 km 2 0.133 176.000 11.187 ,0.01 WETLAND1K24 Land cover-All 5 km 4 0.198 176.374 11.561 ,0.01 SHORE5k, WETLAND5k, DEC5K25 Land cover-All þ Abiotic 5 km 4 0.198 176.374 11.561 ,0.01 SHORE5k, WETLAND5k, DEC5K26 Physical 10 km 3 0.151 177.065 12.252 ,0.01 SNOW10k, TRI10k27 Land cover-Natural edge 5 km 3 0.145 177.450 12.637 ,0.01 SHORE5k, WETLAND5k28 Land cover-All 500 m 2 0.103 177.839 13.026 ,0.01 WETLAND50029 Land cover-Natural edge 500 m 2 0.103 177.839 13.026 ,0.01 WETLAND50030 Land cover-All þ Abiotic 500 m 3 0.138 177.899 13.086 ,0.01 WETLAND500, TRI50031 Physical 500 m 2 0.095 178.266 13.453 ,0.01 TRI50032 Physical 1 km 2 0.074 179.506 14.693 ,0.01 TRI1k33 Physical 5 km 2 0.051 180.867 16.054 ,0.01 TRI5k
34–37 Land cover-Forest all scales (no significant model)
Notes: All models are significant at P , 0.01 except models 34–37, which were not significant. K indicates the number ofvariables used in the model (including the intercept), AICc values are Akaike’s Information Criterion adjusted for small sample size,DAICc is the difference in AICc score from the top model, AICwt is the Aikaike weight. Variables are described in Table 1.
June 2008 1021COYOTE FOREST USE
presence. The transects that detected the most animals
probably reflect the frequent use of an area by two to
three packs, perhaps using areas nearby as travel
corridors or rendezvous sites, while the lower abundance
sites reflect rarer visits by coyotes, probably from one
broad-ranging pack. Understanding the patterns and
potential causes of this variation in coyote abundance
over the broader landscape is important for document-
ing the extent to which the species has become
established as the region’s top predator.
Large carnivores are generally wary, wide-ranging,
and relatively rare, which makes these species difficult to
detect, let alone quantify their abundance across large
landscapes. On the other hand, adaptable generalists like
coyotes can be present in almost all parts of their range,
such that measures of abundance (rather than presence/
absence) are needed to understand their habitat prefer-ences. Fortunately, coyotes indirectly reveal their
presence by leaving feces along trails, and the results
of this study confirm that the abundance of feces is
directly related to the number of individual coyotes
inhabiting the site. This simple and direct index of
coyote abundance allows us to identify the factors that
affect coyote abundance, and predicatively map these
relationships across the entire region, something rarely
done for top predators (Karanth et al. 2004). Although
several noninvasive surveying techniques have a similar
potential to overcome the limitations of presence/
absence data by allowing estimates of absolute or
relative abundance across broad landscapes withoutthe necessity of capturing the animal (Karanth and
Nichols 1998, Gompper et al. 2006, Herzog et al. 2007)
these techniques remain poorly developed for censusing
population size of species that do not have markings
that are individually distinct, or that exist over very large
landscapes, such as coyotes.
Coyotes in forested ecosystems
Coyotes entered eastern North America from the
Midwest, where they evolved in the open habitats of the
Great Plains (Parker 1995, Gompper 2002). Studies of
individual populations have provided conflicting in-
sights on the extent to which coyotes have adapted to
the forested systems that dominates the Northeast
(Person and Hirth 1991, Kendrot 1998, Crete et al.
2001, Richer et al. 2002). We found coyote abundance to
be positively related to the amount of forest cover within
the surrounding 1 km and negatively related to levels of
rural human settlement, thereby rejecting the hypothesis
that eastern forests are unsuitable habitat for eastern
coyotes.
More importantly, our abundance-based models
reveal that specific characteristics of the forest explain
more variation than coarse-scale habitat delineations
(open vs. forested). Our data always favored hypotheses
that included measures of vegetation structure measured
at the forest stand level. Hence, landscapes with the
highest abundance of coyotes were forests characterized
by open canopy and less woody structure (e.g., more
open canopy, less coarse woody debris) with abundantnatural edges (e.g., wetlands and shoreline). Thus at a
broad spatial scale, forested habitat was correlated with
higher densities of coyotes, but on a finer spatial scale,
coyotes were more abundant where the degree of forest
canopy openness increased. The natural edges and
disturbed forests favored by coyotes, with more open
canopies and less downed woody debris, may serve as
refuges from human hunters compared to areas with
more open landscapes. Such hunter-driven habitat
preference has been observed in ungulates (Kilgo et al.
1998, Millspaugh et al. 2000), and coyotes in northern
New York appear to face high hunting pressure(Kendrot 1998). Even more importantly, however,
natural edges and disturbed forest are likely to provide
increased prey availability relative to mature forests.
Habitat selection by coyotes is typically influenced byprey availability (Ozoga and Harger 1966, Hilton 1978,
Litvaitis and Shaw 1980, Andelt and Andelt 1981).
Although we were not able to directly measure this at
the scale of our predator surveys, the patterns of coyote
abundance we observed mirror likely patterns in the
distribution of preferred prey. The diet of coyotes in
northern New York (Hamilton 1974, Chambers 1987;
R. Kays, J. Ray, and M. Gompper, unpublished data) is
dominated by snowshoe hare and white-tailed deer, with
fruit (e.g., raspberry [Rubus spp.] and blueberry [Vacci-
nium spp.]), playing an important role in the summer.
These food items are more abundant in younger, more
TABLE 6. Best model results predicting coyote abundance from stage 3 of overall model selection criteria.
Model no. Variables in model AICc DAICc AICwt
All variables
1 VOLCWD, CANOPEN, BASNAG, WETLAND1K, FORCOV1K, SHORE10K 161.692 0 0.1482 VOLCWD, CANOPEN, BASNAG, WETLAND1K, FORCOV1K 162.303 0.610 0.1093 VOLCWD, CANOPEN, BASNAG, WETLAND1K, FORCOV1K, dtPAVED 163.613 1.920 0.057
Without vegetation structure variables
1 dtPAVED, HOUSE5K, dtHOUSE, TRI500, SNOW10K 168.995 0 0.2152 dtPAVED, HOUSE5K, dtHOUSE, TRI10K, SNOW10K 170.157 1.162 0.1203 dtPAVED, HOUSE5K, dtHOUSE, TRI1K, SNOW10K 170.822 1.827 0.08634 dtPAVED, HOUSE5K, dtHOUSE, TRI500 171.044 2.049 0.0772
Notes:AICc values are Akaike’s Information Criterion adjusted for small sample size, DAICc is the difference in AICc score fromthe top model, and AICwt is the Aikaike weight. Variables are described in Tables 1 and 5.
ROLAND W. KAYS ET AL.1022 Ecological ApplicationsVol. 18, No. 4
open forest, as well as in more disturbed habitats in the
area (Oosting 1956, Severinghaus and Brown 1956),
with lower densities of such prey in large tracts of
closed-canopy forests (Bittner and Rongstad 1982,
Litvaitis et al. 1985). Deer may also be more vulnerable
near frozen lakes and forest edges in winter (Patterson
and Messier 2003), and if coyote hunting strategies for
killing adult deer follow those of wolves and elk (Cervus
elaphus), hunting in forested areas may be more
successful than in open habitats (Creel et al. 2006).
Our finding that forest type is more important in
predicting coyote abundance than more generalized
habitat categories (i.e., open vs. forest) may help explain
contradictory results of past research. Several studies
suggesting that forested habitats are suboptimal for
coyotes were carried out in boreal forest systems
(Samson and Crete 1997, Tremblay et al. 1998, Crete
et al. 2001). It may be that coyote use of these forested
systems differs significantly from their use of the
younger or more deciduous forests of northern New
York (Richer et al. 2002). Indeed, two local studies in
landscapes directly comparable to our study area found
coyotes preferred forests (Person and Hirth 1991,
Kendrot 1998). These differing results indicate the
importance of including detailed descriptions of forest
type when analyzing the habitat use of wide-ranging
carnivores such as coyotes.
Coyotes and human activities
The specific forest characteristics we found most
important in predicting coyote abundance were mea-
sures indicative of forest disturbance, especially the open
canopies and lower levels of coarse woody debris that
result from selective logging. The eastward expansion of
coyotes coincided with the forest conversion due to
logging and agriculture (Parker 1995, Fener et al. 2005).
Yet aside from Major and Sherburne’s (1987) report of
summer use of clearcuts, there has been little study of
coyote habitat use as a function of timber harvest.
Nearly all current eastern forests have undergone
periods of deforestation and reforestation associated
with logging and land conversion (Foster 1992, Foster et
al. 1998), and logging continues to be the primary
disturbance factor in some forests throughout Northeast
FIG. 3. Predicted coyote abundance across northern New York showing the highest densities (red) in areas with lower roaddensities lacking steep terrain, especially where snow depth can be greater. The 54 study sites where coyote abundance wasmeasured through fecal surveys are indicated in black; lakes and rivers are blue.
June 2008 1023COYOTE FOREST USE
North America (Jenkins and Keal 2004). Our results
show that coyotes are most abundant in these disturbed
forests, and suggest that this anthropogenic activity may
in turn have strong indirect effects on local predator
assemblages by influencing the dynamics of intraguild
competition (Crooks and Soule 1999).
Our hypothesis that aspects of human settlement
affect coyote distribution was supported, particularly
when local-scale variables related to vegetation structure
were excluded from the models. Coyote densities were
highest far from roads in areas with low house density,
but within these areas coyotes were actually more
common closer to houses. Among these models, distance
to roads was by far the most important parameter
(highest AIC weight). These nuanced findings probably
reflect the higher risk of mortality leading to lower
densities for coyotes in highly settled areas at larger
scales (Major 1983, Kendrot 1998, Bogan 2004), offset
at local scales by the abundant prey in disturbed habitats
near human residences. When vegetation structure
variables were included in the analyses, only one
anthropogenic variable (dtROADS) received enough
support to be included in our final stage of model
building, and had the lowest summed AIC weights in
this final stage.
There have been substantial changes to land cover and
land use patterns in northeastern North America since
European settlement. Over the past century, logging and
agriculture have declined in the region while rates of
wildland development and tourism have increased
(Foster et al. 1998, Jenkins and Keal 2004, Glennon
and Krester 2005). The combined result of these land use
changes in northern New York are extensive tracts of
maturing forests, especially on public lands. Our study
suggests that initial stages of reforestation are likely to
be associated with increased coyote populations, but as
forests mature and support less potential prey, coyote
numbers in these areas should decline. Private lands
where more active logging is taking place, on the other
hand, are likely to produce more coyote food, and
therefore support more coyotes.
One benefit of using large-scale landscape variables to
model coyote abundance is that the results can be
projected over large regions (Fig. 3). The results
graphically illustrate the relationships described by our
best model that could be mapped, namely higher
summer coyote densities away from settled regions,
especially where local terrain is moderate and snowfall is
high. This model fits with established coyote–human
interactions in rural areas and also suggests that, since
coyotes are not known to be seasonally migratory, deep
snow might indirectly constitute favorable summer
habitat for coyotes by making deer more vulnerable,
as has been observed in studies of coyote–deer
interactions (Patterson et al. 1998, Patterson and
Messier 2000, 2003). An important caveat, however, is
that our mapping was carried out without the fine-scale
vegetation structure variables that were most important
for predicting coyote abundance. Thus, additional
variation should be expected within the context of the
mapped predictions with higher coyote abundance in
more disturbed forests.
Top-down effects of large canids
With the extinction of the wolf in the region in the late
1800s (Coleman 2004), coyotes are now the top predator
in the region, adding two interesting perspectives to our
results. First, if the ecological effects of top predators
vary positively with their abundance, coyote-driven top-
down direct and indirect effects should be stronger
where their numbers are greater (Ripple and Beschta
2004, Ray et al. 2005). Second, the extent to which
eastern coyotes have filled the wolf niche has implica-
tions for proposed reintroductions of wolves to the
region.
Predators have the potential to drive community and
ecosystem processes (Ray et al. 2005), as has been
demonstrated for coyotes in southern California
(Crooks and Soule 1999). The mechanisms by which
these effects occur include direct and fear-mediated
changes to prey numbers and behavior as well as indirect
effects derived from the altered impact of prey on the
landscape (Post et al. 1999, Ripple and Beschta 2004,
Croll et al. 2005, Ray et al. 2005, Creel et al. 2006).
However the circumstances that determine if and when
(at what abundance) predators become drivers of
ecosystem processes are still poorly understood. If
greater coyote abundance equates to greater strengths
of trophic interactions in the Northeast we might expect
that the top-down effects of this recent arrival would be
most strongly observed in younger and more disturbed
forests with greater proportions of edge habitat, where
our models suggest higher numbers of coyotes. The
existence of these habitats is strongly influenced by the
extent of extractive logging in the Northeast, leading to
the testable prediction that coyote-driven top-down
effects are ephemeral; as habitats shift from open areas
to younger, more productive forest, the effects of
coyotes on the local animal and plant community
should increase. However as these younger forests
mature and become more closed-canopy and less
productive for coyote prey species, the top-down
influence of coyotes should also decrease.
The potential to reintroduce wolves into northeastern
forests is periodically raised (Paquet et al. 1999, Sharpe
et al. 2001, U.S. Fish and Wildlife Service 2003). The
ecological similarity between eastern coyotes and wolves
is a concern for any such plan, with some questioning
the extent to which eastern coyotes have already filled
the wolf niche (Mathews and Porter 1992, Ballard et al.
1999, Sage 2001). Although eastern coyotes eat more
ungulate prey than western coyotes, their consumption
of large prey is still much less than that of wolves, with
coyotes eating more fruit and small and medium-sized
prey instead, and never taking large prey such as moose
(Alces alces) (Potvin et al. 1988, Gompper 2002, Chavez
ROLAND W. KAYS ET AL.1024 Ecological ApplicationsVol. 18, No. 4
and Gese 2005). Our data show that eastern coyotes can
flourish in rural, human-dominated landscapes and also
take advantage of disturbed forests, presumably because
of the diverse food there; these relationships drive the
geographic patterns shown in our predictive map (Fig.
3). These patterns are clearly different than those
predicted for wolves, which are more sensitive to road
density and track ungulate populations, but are not
sensitive to densities of smaller prey or fruit (Harrison
and Chapin 1998, Mladenoff and Sickley 1998). Within
the Adirondack Park, wolves would presumably do
better than coyotes in the mature wilderness forest,
spatially tracking deer, beaver (Castor canadensis), and
the growing moose populations there. These predicted
differences in spatial ecology underscore the idea that
eastern coyotes have not filled the wolf niche in the
region, similar to the conclusion of Crete et al. (2001)
based on a comparison of the foraging ecology and
energetic physiology of the two species.
CONCLUSIONS
We have shown how noninvasive survey techniques
can be used over a large landscape to collect the
necessary data for modeling the habitat ecology of a
wide-ranging and difficult-to-observe predator species
based on population abundance. This approach reveals
not only that coyote abundance is greater in forested
landscapes, but also provides the means of testing
alternative models for how variance in these habitats
affects abundance, ultimately allowing a more nuanced
view of how coyotes use forested systems. Such an
approach provides a more detailed and biologically
relevant view of the spatial patterns of top predator
populations, and represents a necessary first step for
addressing several important fundamental and applied
questions such as how the arrival of a novel top predator
influences community and ecosystem structure and
function, and how the species is likely to respond to
future changes in eastern forests.
ACKNOWLEDGMENTS
Funding for this work came from the Wildlife ConservationSociety, the Geraldine R. Dodge Foundation, the NationalGeographic Society, Columbia University, the New York StateMuseum, the New York Biodiversity Research Institute, theNew York Department of Environmental Conservation, andthe National Science Foundation (DEB 0347609). The AlbanyPine Bush Preserve Commission, Adirondack Park Agency,New York State Department of Environmental Conservation,and Saratoga National Historical Park provided permitting andlogistic support. Field and laboratory work benefited from theassistance of J. Bopp, H. Bull, J. Cryan, A. DeWan, S.Franklin, R. Gill, C. Herzog, E. Hellwig, J. Isabelle, R. Keefe,N. Korobov, S. LaPoint, J. Richardson, L. Robinson, D.Ruggeri, and G. Woolmer. J. Malcolm and G. Holloway gavevaluable advice with habitat modeling. Numerous privatelandowners generously gave permission to conduct surveys,including the Adirondack League Club, International PaperCompany, The Nature Conservancy, Brandreth Park, RossPark, Domtar Industries, Gulf View Hunting Club, MinerAgricultural Institute, and the Dragoon and DeCosse families.
Thanks to two anonymous reviewers for their helpful commentsthat improved the manuscript.
LITERATURE CITED
Adams, J. R., and L. P. Waits. 2007. An efficient method forscreening faecal DNA genotypes and detecting new individ-uals and hybrids in the red wolf (Canis rufus) experimentalpopulation area. Conservation Genetics 8:123–131.
Andelt, W. F., and S. H. Andelt. 1981. Habitat use by coyotesin southeastern Nebraska. Journal of Wildlife Management45:1001–1005.
Ballard, W. B., H. A. Whitlaw, S. J. Young, R. A. Jenkins, andG. J. Forbes. 1999. Predation and survival of white-taileddeer fawns in north central New Brunswick. Journal ofWildlife Management 63:574–579.
Baum, J. K., R. A. Myers, D. G. Kehler, B. Worm, S. J. Harley,and P. A. Doherty. 2003. Collapse and conservation of sharkpopulations in the Northwest Atlantic. Science 299:389–392.
Bittner, L. B., and O. J. Rongstad. 1982. Snowshoe hare andallies. Pages 146–161 in J. A. Chapman and G. E. Feld-hammer, editors. Wild mammals of North America. JohnHopkins University Press, Baltimore, Maryland, USA.
Bogan, D. A. 2004. Eastern coyote (Canis latrans) home range,habitat selection, and survival rates in the suburban AlbanyPine Bush landscape of New York. Thesis. State Universityof New York at Albany, Albany, New York, USA.
Brundige, G. C. 1993. Predation ecology of the eastern coyote,Canis latrans var., in the Adirondacks, New York. Disser-tation. SUNY College of Environmental Science andForestry, Syracuse, New York, USA.
Burnham, K. P., and D. R. Anderson. 2002. Model selectionand multi-model inference. Springer, New York, New York,USA.
Carroll, C. 2007. Interacting effects of climate change,landscape conversion and harvest on carnivore populationsat the range margin: marten and lynx in the NorthernAppalachians. Conservation Biology 21:1092–1104.
Caturano, S. L. 1983. Habitat and home range use by coyotes ineastern Maine. Thesis. University of Maine, Orono, Maine,USA.
Chambers, R. E. 1987. Coyote and red fox diets in the centralAdirondacks. Abstract. Proceedings of the Northeast Fishand Wildlife Conference. Boston, Massachusetts, USA.
Chavez, A. S., and E. M. Gese. 2005. Food habits of wolves inrelation to livestock depredations in northwestern Minneso-ta. American Midland Naturalist 154:253–263.
Coleman, J. T. 2004. Vicious: wolves and men in America. YaleUniversity Press, New Haven, Connecticut, USA.
Creel, S., Jr., J. Winne, B. Maxwell, K. Hamlin, and M. Creel.2006. Elk alter habitat selection as an antipredator responseto wolves. Ecology 86:3387–3397.
Crete, M., J. Ouellet, J. Tremblay, and R. Arsenault. 2001.Suitability of the forest landscape for coyotes in northeasternNorth America and its implications for coexistence withother carnivores. Ecoscience 8:311–319.
Croll, D. A., J. L. Maron, J. A. Estes, E. M. Danner, and G. V.Byrd. 2005. Introduced predators transform subarctic islandsfrom grassland to tundra. Science 307:1959–1961.
Crooks, D. R., and M. E. Soule. 1999. Mesopredator releaseand avifaunal extinctions in a fragmented system. Nature400:563–566.
Dumond, M., and M. A. Villard. 2000. Demography and bodycondition of coyotes (Canis latrans) in eastern New Bruns-wick. Canadian Journal of Zoology 78:399–406.
Fener, H. M., J. R. Ginsberg, E. Sanderson, and M. E.Gompper. 2005. Chronology of range expansion of theCoyote Canis latrans, in New York. Canadian FieldNaturalist 119:1–5.
Foster, D. R. 1992. Land-use history (1730–1990) andvegetation dynamics in central New England, USA. Journalof Ecology 80:753–772.
June 2008 1025COYOTE FOREST USE
Foster, D. R., G. Motzkin, and B. Slater. 1998. Land-usehistory as long-term broad-scale disturbance: regional forestdynamics in central New England. Ecosystems 1:96–119.
Fuller, T. K., E. C. York, S. M. Powell, T. A. Decker, andR. M. DeGraaf. 2001. An evaluation of territory mapping toestimate fisher density. Canadian Journal of Zoology 79:1691–1696.
Glennon, M., and H. Krester. 2005. Impacts to wildlife fromlow density, exurban development: information and consid-erations for the Adirondack Park. Adirondack Communitiesand Conservation Program, Saranac Lake, New York, USA.
Gompper, M. E. 2002. The ecology of northeast coyotes:current knowledge and priorities for future research. WildlifeConservation Society Working Paper 17:1–47.
Gompper, M. E., R. W. Kays, J. C. Ray, S. D. LaPoint, D. A.Bogan, and J. A. Cryan. 2006. A comparison of non-invasivetechniques to survey carnivore communities in NortheasternNorth America. Wildlife Society Bulletin 34:1142–1151.
Hamilton, W. J. 1974. Food habits of the coyote in theAdirondacks. New York Fish and Game Journal 21:177–181.
Harrison, D. J., and T. G. Chapin. 1998. Extent andconnectivity of habitat for wolves in eastern North America.Wildlife Society Bulletin 26:767–775.
Herzog, C. J., R. W. Kays, J. C. Ray, M. E. Gompper, W. J.Zielinski, R. Higgins, and M. Tymeson. 2007. Using patternsin track plate footprints to identify individual fishers. Journalof Wildlife Management 71:955–963.
Hilton, H. 1978. Systematics and ecology of the eastern coyote.Pages 210–228 in M. Bekoff, editor. Coyotes: biology,behavior, and management. Academic Press, New York,New York, USA.
Holloway, G. L., and J. R. Malcolm. 2006. Sciurid habitatrelationships in forests managed under selection and shelter-wood silviculture in Ontario. Journal of Wildlife Manage-ment 70:1735–1745.
Jenkins, J., and A. Keal. 2004. The Adirondack atlas: ageographic portrait of the Adirondack Park. SyracuseUniversity Press, Syracuse, New York, USA.
Karanth, K. U., and J. D. Nichols. 1998. Estimation of tigerdensities in India using photographic captures and recap-tures. Ecology 79:2852–2862.
Karanth, K. U., J. D. Nichols, N. S. Kumar, W. A. Link, andJ. E. Hines. 2004. Tigers and their prey: predicting carnivoredensities from prey abundance. Proceedings of the NationalAcademy of Sciences (USA) 101:4854–4858.
Kendrot, S. R. 1998. The effects of roads and land use on homerange use, behavior and mortality of eastern coyotes (Canislatrans var.) in Northern New York. Thesis. SUNY Collegeof Environmental Science and Forestry, Syracuse, NewYork, USA.
Kilgo, J. C., R. F. Labisky, and D. E. Fritzen. 1998. Influencesof hunting on the behavior of white-tailed deer: implicationsfor conservation of the Florida panther. ConservationBiology 12:1359–1364.
Kohn, M. H., E. C. York, D. A. Kamradt, G. Haught, R. M.Sauvajot, and R. K. Wayne. 1999. Estimating population sizeby genotyping faeces. Proceedings of the Royal Society B266:657–663.
Kyle, C. J., A. R. Johnson, B. R. Patterson, P. J. Wilson, K.Shami, S. K. Grewal, and B. N. White. 2006. Genetic natureof eastern wolves: past, present and future. ConservationGenetics 7:273–287.
Lariviere, S., and M. Crete. 1993. The size of eastern coyotes(Canis latrans): a comment. Journal of Mammalogy 74:1072–1074.
Litvaitis, J. A., and D. J. Harrison. 1989. Bobcat–coyote nicherelationships during a period of coyote population increase.Canadian Journal of Zoology 67:1180–1188.
Litvaitis, J. A., and J. H. Shaw. 1980. Coyote movements,habitat use, and food habits in southwestern Oklahoma.Journal of Wildlife Management 44:62–68.
Litvaitis, J. A., J. A. Sherburne, and J. A. Bissonette. 1985.Influence of understory characteristics on snowshoe harehabitat use and density. Journal of Wildlife Management 49:866–873.
Mace, R. D., J. S. Waller, T. L. Manley, K. Ake, and W. T.Wittinger. 1999. Landscape evaluation of grizzly bear habitatin western Montana. Journal of Conservation Biology 13:367–377.
Major, J. T. 1983. Ecology and interspecific relationships ofcoyotes, bobcats, and red foxes in western Maine. Disserta-tion. University of Maine, Orono, Maine, USA.
Major, J. T., and J. A. Sherburne. 1987. Interspecificrelationships of coyotes, bobcats, and red foxes in westernMaine. Journal of Wildlife Management 51:606–616.
Mathews, N. E., and W. F. Porter. 1992. Maternal defensebehavior in white-tailed deer. Pages 141–160 in A. H. Boer,editor. Ecology and management of the Eastern Coyote.University of New Brunswick, Fredericton, Fredericton, NewBrunswick, Canada.
Mellersh, C. S., A. A. Langston, G. M. Acland, M. A. Fleming,K. Ray, N. A. Wiegand, L. V. Francisco, M. Gibbs, G. D.Aguirre, and E. A. Ostrander. 1997. A linkage map of thecanine genome. Genomics 46:326–336.
Millspaugh, J. J., G. C. Brundige, and R. A. Gitzen. 2000. Elkand hunter space-use sharing in South Dakota. Journal ofWildlife Management 64:994–1003.
Mladenoff, D. J., and T. A. Sickley. 1998. Assessing potentialgray wolf restoration in the northeastern United States: aspatial prediction of favorable habitat and potential popu-lation levels. Journal of Wildlife Management 62:1–10.
National Operational Hydrologic Remote Sensing Center.2004. Snow data assimilation system (SNODAS) dataproducts at NSIDC. National Snow and Ice Data Center,Boulder, Colorado, USA.
Nesslage, G. M., and W. F. Porter. 2001. A geostatisticalanalysis of deer harvest in the Adirondack Park, 1954–1997.Wildlife Society Bulletin 3:787–794.
Oehler, J. D., and J. Litvaitis. 1996. The role of spatial scale inunderstanding responses of medium-sized carnivores toforest fragmentation. Canadian Journal of Zoology 74:2070–2079.
Oosting, H. J. 1956. The study of plant communities. Secondedition. W. H. Freeman and Company, San Francisco,California, USA.
Ozoga, J. J., and E. M. Harger. 1966. Winter activities andfeeding habits of northern Michigan coyotes. Journal ofWildlife Management 60:809–818.
Paetkau, D., and C. Strobeck. 1994. Microsatellite analysis ofgenetic variation in black bear populations. MolecularEcology 3:489–495.
Paquet, P. C., J. R. Strittholt, and N. L. Staus. 1999. Wolfreintroduction feasibility in the Adirondack Park. Preparedfor the Adirondack Citizens Advisory Committee on thefeasibility of wolf reintroduction. Conservation BiologyInstitute, Corvallis, Oregon, USA.
Parker, G. 1995. Eastern coyote: the story of its success.Nimbus Publishing, Halifax, Nova Scotia, Canada.
Patterson, B. R., L. K. Benjamin, and F. Messier. 1998. Preyswitching and feeding habits of eastern coyotes in relation tosnowshoe hare and white-tailed deer densities. CanadianJournal of Zoology 76:1885–1897.
Patterson, B. R., and F. Messier. 2000. Factors influencingkilling rates of white-tailed deer by coyotes in easternCanada. Journal of Wildlife Management. 64:721–732.
Patterson, B. R., and F. Mesier. 2003. Age and condition ofdeer killed by coyotes in Nova Scotia. Canadian Journal ofZoology 81:1894–1898.
Person, D. K., and D. H. Hirth. 1991. Home range and habitatuse of coyotes in a farm region of Vermont. Journal ofWildlife Management 55:433–441.
ROLAND W. KAYS ET AL.1026 Ecological ApplicationsVol. 18, No. 4
Peterson, R. O., and J. M. Thurber. 1993. The size of easterncoyotes (Canis latrans): a rebuttal. Journal of Mammalogy74:1075–1076.
Post, E., R. O. Peterson, N. C. Stenseth, and B. E. McLaren.1999. Ecosystem consequences of wolf behavioral response toclimate. Nature 401:905–907.
Potvin, F., H. Jolicoeur, and J. Huot. 1988. Wolf diet and preyselectivity during two periods for deer in Quebec: declineversus expansion. Canadian Journal of Zoology 66:1274–1279.
Prugh, L. R., C. E. Ritland, S. M. Arthur, and C. J. Krebs.2005. Monitoring coyote population dynamics by genotypingfaeces. Molecular Ecology 14:1585–1596.
Ray, J. C. 2000. Mesocarnivores of northeastern NorthAmerica: status and conservation issues. Wildlife Conserva-tion Society Working Paper 15:1–84.
Ray, J. C., K. H. Redford, R. S. Steneck, and J. Berger. 2005.Large carnivores and the conservation of biodiversity. IslandPress, Washington, D.C., USA.
Raymond, M., and F. Rousset. 1995. GENEPOP (Version 1.2):population genetic software for exact test and ecumenism.Journal of Heredity 83:248–249.
Richer, M. C., M. Crete, J. P. Ouellet, L. P. Rivest, and J.Huot. 2002. The low performance of forest versus ruralcoyotes in Northeastern North America: inequality betweenpresence and availability of prey. Ecoscience 9:44–54.
Riley, S. J., S. D. DeGloria, and R. Elliott. 1999. A terrainruggedness index that quantifies topographic heterogeneity.Intermountain Journal of Sciences 5:23–27.
Ripple, W. J., and R. L. Beschta. 2004. Wolves and the ecologyof fear: Can predation risk structure ecosystems? BioScience54:755–766.
Sage, R., Jr. 2001. Wolves in the Adirondacks? Perspectivesfrom the heart of the Adirondack Park. Pages 39–46 in V. A.Sharpe, B. Norton, and S. Donnelley, editors. Wolves andhuman communities: biology, politics, and ethics. IslandPress, Washington, D.C., USA.
Samson, C., and M. Crete. 1997. Summer food habits andpopulation density of coyotes, Canis latrans, in boreal forestsof southeastern Quebec. Canadian Field-Naturalist 111:227–233.
Severinghaus, C. W., and C. P. Brown. 1956. History of thewhite-tailed deer in New York. New York Fish and GameJournal 3:130–167.
Sharpe, V. A., B. Norton, and S. Donnelley. 2001. Wolves andhuman communities: biology, politics, and ethics. IslandPress, Washington, D.C., USA.
Thurber, J. M., and R. O. Peterson. 1991. Changes in body sizeassociated with range expansion in the coyote (Canis latrans).Journal of Mammalogy 72:750–755.
Tremblay, J. P., M. Crete, and J. Huot. 1998. Summer foragingbehaviour of eastern coyotes in rural versus forest landscape:a possible mechanism of source–sink dynamics. Ecoscience 5:172–182.
U.S. Fish and Wildlife Service. 2003. Final rule to reclassify/delist the gray wolf. Federal Register 68:15804–15875.
Valiere, N. 2002. Gimlet: a computer program for analyzinggenetic individual identification data. Molecular EcologyNotes 2:377.
Van Oosterhout, C., W. F. Hutchinson, D. P. M. Wills, and P.Shipley. 2004. Micro-checker: software for identifying andcorrecting genotyping errors in microsatellite data. MolecularEcology Notes 4:535–538.
Waits, L. P., G. Luikart, and P. Taberlet. 2001. Estimating theprobability of identity among genotypes in natural popula-tions: cautions and guidelines. Molecular Ecology 10:249–256.
Way, J. G., I. M. Ortega, and E. G. Strauss. 2004. Movementand activity patterns of eastern coyotes in a coastal, suburbanenvironment. Northeastern Naturalist 11:237–254.
Zielinski, W. J., R. L. Truex, F. V. Schlexer, L. A. Campbell,and C. Carroll. 2005. Historical and contemporary distribu-tions of carnivores in forests of the Sierra Nevada, California,USA. Journal of Biogeography 32:1385–1407.
June 2008 1027COYOTE FOREST USE