Caja PDF

Comparta fácilmente sus documentos PDF con sus contactos, la web y las redes sociales.

Compartir un archivo PDF Gestor de archivos Caja de instrumento Buscar PDF Ayuda Contáctenos



De Angelo et al 2010 JWM Track identification .pdf



Nombre del archivo original: De Angelo et al 2010 - JWM - Track identification.pdf

Este documento en formato PDF 1.2 fue generado por 3B2 Total Publishing System 7.51n/W / Apogee Create Series3 v1.0, y fue enviado en caja-pdf.es el 28/01/2015 a las 21:53, desde la dirección IP 200.112.x.x. La página de descarga de documentos ha sido vista 1461 veces.
Tamaño del archivo: 770 KB (13 páginas).
Privacidad: archivo público




Descargar el documento PDF









Vista previa del documento


Journal of Wildlife Management 74(5):1141–1153; 2010; DOI: 10.2193/2009-293

Tools and Technology Article

Traditional Versus Multivariate
Methods for Identifying Jaguar, Puma,
and Large Canid Tracks
CARLOS DE ANGELO,1 National Research Council of Argentina (CONICET) and Asociacio´n Civil Centro de Investigaciones del Bosque Atla´ntico
(CeIBA), Yapeyu´ 23, CP 3370, Puerto Iguazu´, Misiones, Argentina
AGUSTI´N PAVIOLO, National Research Council of Argentina (CONICET) and Asociacio´n Civil Centro de Investigaciones del Bosque Atla´ntico
(CeIBA), Yapeyu´ 23, CP 3370, Puerto Iguazu´, Misiones, Argentina
MARIO S. DI BITETTI, National Research Council of Argentina (CONICET) and Asociacio´n Civil Centro de Investigaciones del Bosque Atla´ntico
(CeIBA), Yapeyu´ 23, CP 3370, Puerto Iguazu´, Misiones, Argentina

ABSTRACT The jaguar (Panthera onca) and puma (Puma concolor) are the largest felids of the American Continent and live in sympatry
along most of their distribution. Their tracks are frequently used for research and management purposes, but tracks are difficult to distinguish
from each other and can be confused with those of big canids. We used tracks from pumas, jaguars, large dogs, and maned wolves (Chrysocyon
brachyurus) to evaluate traditional qualitative and quantitative identification methods and to elaborate multivariate methods to differentiate big
canids versus big felids and puma versus jaguar tracks (n 5 167 tracks from 18 zoos). We tested accuracy of qualitative classification through an
identification exercise with field-experienced volunteers. Qualitative methods were useful but there was high variability in accuracy of track
identification. Most of the traditional quantitative methods showed an elevated percentage of misclassified tracks ( 20%). We used stepwise
discriminant function analysis to develop 3 discriminant models: 1 for big canid versus big felid track identification and 2 alternative models for
jaguar versus puma track differentiation using 1) best discriminant variables, and 2) size-independent variables. These models had high
classification performance, with ,10% of error in the validation procedures. We used simpler discriminant models in the elaboration of
identification keys to facilitate track classification process. We developed an accurate method for track identification, capable of distinguishing
between big felids (puma and jaguar) and large canids (dog and maned wolf) tracks and between jaguar and puma tracks. Application of our
method will allow a more reliable use of tracks in puma and jaguar research and it will help managers using tracks as indicators of these felids’
presence for conservation or management purposes.
L

KEY WORDS canids, discriminant function analysis, identification keys, jaguar, Panthera onca, puma, Puma concolor, track
differentiation.

Sign surveys are useful noninvasive tools to assess and
monitor elusive or rare species (Wemmer et al. 1996,
Gompper et al. 2006). Spoor counts have been used as
indicators of presence, relative abundance, and density
estimation of different species (Van Dyke et al. 1986,
Stander 1998, Crooks 2002, Wilting et al. 2006). Likewise,
many studies involving endangered species have used sign
surveys to obtain basic ecological information (e.g.,
distribution and habitat use; Perovic and Herran 1998,
Potvin et al. 2005, Markovchick-Nicholls et al. 2008) and as
preliminary or complementary assessment in ecological
research or conservation plans (Schaller and Crawshaw
1980, Rabinowitz and Nottingha 1986, Soisalo and
Cavalcanti 2006, Paviolo et al. 2008). Tracks are not only
one of the most commonly used signs but they may also be
useful in identifying other associated signs such as fecal
samples (Wemmer et al. 1996, Scognamillo et al. 2003,
Shaw et al. 2007, Azevedo 2008). However, one of the
problems of employing tracks and other signs is correct
identification of the species, particularly in areas where 2
similar species are likely to be found and where little or no
information about their presence is available.
Because track shape is affected by many factors (like
substrate quality or the pace of the animal) it is often
difficult to distinguish tracks of similar species, particularly
in areas where soil conditions make track printing difficult
L

1

E-mail: biocda@gmail.com

De Angelo et al. N Large Felid and Canid Track Identification

and only few tracks are found in each event (Fjelline and
Mansfield 1988, Wemmer et al. 1996, Grigione et al. 1999,
Lewison et al. 2001). Multivariate analyses have been used
to reduce subjectivity and improve accuracy in sign
recognition (Zielinski and Truex 1995, Zalewski 1999,
Harrington et al. 2008, Steinmetz and Garshelis 2008). In
addition, multivariate analyses of feline tracks have been
used for more demanding objectives like sex determination
and individual identification (Smallwood and Fitzhugh
1993, Riordan 1998, Sharma et al. 2003, Wilting et al.
2006, Isasi-Catala´ and Barreto 2008; but see Karanth et al.
2003 and Gordon et al. 2007).
Jaguars (Panthera onca) live in sympatry with pumas (Puma
concolor) along most of their distribution and both species
are the focus of many research and conservation programs
(Nowell and Jackson 1996, Sanderson et al. 2002, Conroy et
al. 2006, Shaw et al. 2007). These felids also share most of
their range with some large canids, such as pumas with gray
wolves (Canis lupus) and coyotes (C. latrans) in North
America, both felids with the maned wolf (Chrysocyon
brachyurus) in central South America, and both felids with
the domestic dog in most of their distribution.
Tracks of pumas and jaguars have been employed for
research, monitoring, and management (Smallwood and
Fitzhugh 1995, Hoogesteijn 2007, Shaw et al. 2007).
However, it is often difficult to differentiate between puma
and jaguar tracks because they are similar in size and shape
and both are frequently confused with big canids’ tracks
1141

STUDY AREA
Between 2004 and 2008, we obtained track records from
jaguars, pumas, and maned wolves kept in captivity. Our
task was carried out in collaboration with 18 zoos from 6
countries: Zoo Bata´n (Bata´n, Argentina), Zoologico Bosque
Guaranı´ (Foz do Iguac¸u, Brazil), Refugio Biologico Bela
Vista (Foz do Iguac¸u, Brazil), Zoolo´gico Itaipu´ Paraguay
(Hernandarias, Paraguay), Parque Ecolo´gico Urbano de Rı´o
Cuarto (Rı´o Cuarto, Argentina), Zoolo´gico de la Ciudad de
Buenos Aires (Buenos Aires, Argentina), Temaiken (Buenos Aires, Argentina), Zoolo´gico de Roque Sa´enz Pen˜a
(Roque Sa´enz Pen˜a, Argentina), Zoolo´gico Tatu´ Carreta
(La Cumbre, Argentina), Jardı´n Zoolo´gico de la Ciudad de
Co´rdoba (Co´rdoba, Argentina), Zoolo´gico Santa Fe de
Medellı´n (Medellı´n, Colombia), Parque Zoolo´gico Caricuao (Caracas, Venezuela), Parque Zoolo´gico Las Delicias
(Maracay, Venezuela), Lincoln Park Zoo (Chicago, IL,
USA), Sedgwick County Zoo (Wichita, KS, USA),
Caldwell Zoo (Tyler, TX, USA), Philadelphia Zoo
(Philadelphia, PA, USA), and Little Rock Zoo (Little
Rock, AR, USA).

METHODS
Sample Collection and Processing
We collected tracks from 28 jaguars, 29 pumas, and 8 maned
wolves (mainly ad of both sexes for all the species, although
we also included 2 tracks of a juv jaguar and 3 tracks of 2 juv
pumas of around 1.5 yr old). We used the same methodology
to obtain large dog tracks from urban areas (35 individuals,
most of them stray dogs). Authors and zoo personnel
recorded tracks by following general recommendations (see
Wemmer et al. 1996, De Angelo et al. 2008) instead of
standardized methodology, because our main objective was to
evaluate identification methods for tracks collected in the
field by different people, under diverse circumstances, and
using different protocols. For the same reason, we used tracks
1142

for as many individuals and zoos as possible, looking for all
kind of tracks (front, rear, right, and left feet), diverse
substrates, and different animal size and behavior (to favor
including track variability caused by animal pace and wt). We
collected all tracks from free-walking animals (i.e., the paw
was not pressed into the substrate by handlers). We obtained
all dog tracks and 78% of tracks from zoos from unprepared
substrates, whereas remaining zoo tracks were from sleekly
prepared surfaces on which animals walked freely to yield
better prints (mainly in zoos lacking an adequate surface for
printing tracks). Substrate conditions varied among zoos and
among the sites where we collected dog tracks, including
mud, dust, and sand (wet or dry), for all groups considered.
We tried to collect as many tracks as possible from each
individual. However, because we obtained few tracks in most
of the zoos, we tried to select 1 track/foot from each
individual, in order to have all individuals represented by the
same number of tracks. We used 3 methods for recording
tracks: plaster molds, digital photos with a metric rule as size
reference, and track contour in acetate paper. Multiplicity of
zoos, individuals, collectors, and methods generated a variety
of tracks of different qualities, from which we only discarded
for further analysis those tracks that had conspicuous
deformations precluding an accurate measurement.
To give digital format to collected data, we photographed
the plaster molds with digital cameras and scanned track
tracings, including a metric rule. Then, the first author
digitalized all tracks using the spline tool in AutoCADE
2004 (AutoDesk, Inc., Fresno, CA). For digitalization and
measurement, we scaled tracks according to the sizereference metric rule. Subsequently, we rotated and placed
tracks over a base line defined by a tangent line between the
outer heel lobes (Smallwood and Fitzhugh 1989; Fig. 1).
We analyzed 167 tracks, including plaster molds (n 5 98),
photographs (n 5 57), and track tracings (n 5 12) from the
3 groups of interest. We randomly separated 35 tracks out of
the total (10 jaguars, 10 pumas, and 15 big canids) to be
used only for independent validation for discriminant
models and identification keys (independent cases). We
used all other 132 tracks in traditional method validation
and for multivariate model and identification key development (46 tracks from jaguars, 46 from pumas, 33 from dogs,
and 7 from maned wolves).
M

(Smallwood and Fitzhugh 1989, Childs 1998, Shaw et al.
2007). For this reason, many authors have described
qualitative traits to differentiate puma from dog tracks
(Smallwood and Fitzhugh 1989, Shaw et al. 2007; Appendix
A), and to distinguish jaguar from puma tracks (Schaller and
Crawshaw 1980, Aranda 1994, Childs 1998; Appendix A).
Nevertheless, due to the subjectivity and exceptions found in
application of qualitative traits, some authors proposed
quantitative identification measures (Belden 1978, Smallwood and Fitzhugh 1989, Aranda 1994, Childs 1998, Shaw
et al. 2007; Appendix B). However, these quantitative
criteria were created and evaluated based on either few
known individuals or unknown field tracks classified by
qualitative traits.
In view of these shortcomings, we used a diverse set of
tracks from definitely known origin with the aim of 1)
evaluating performance of qualitative and quantitative
methodologies described to differentiate tracks of big canids,
pumas, and jaguars; and 2) developing accurate and robust
quantitative multivariate criteria to classify tracks of these
groups and species.

Evaluation of Traditional Identification Methods
To evaluate accuracy of traditional qualitative track
identification we made a classification exercise with 67
participants from Argentina, Brazil, and Paraguay in a
jaguar conservation meeting (Third Workshop of Collaborative Effort for Jaguar Monitoring in Atlantic Forest, 5–6
Oct 2005, Puerto Iguazu´, Misiones, Argentina). Although
participants considered themselves capable of identifying
tracks of these carnivores (jaguars, pumas, and big canids),
we started the exercise by showing examples of tracks from
each group and by reviewing traditional qualitative differentiation characters (Appendix A). From our track database,
we randomly selected 10 photographs of tracks from each
group and then asked volunteers to classify tracks into these
The Journal of Wildlife Management N 74(5)

Figure 1. Examples of typical big canid, puma, and jaguar tracks. We used measurements shown in this figure for the validation of quantitative
identification criteria and for construction of variables used in discriminant analysis to develop a multivariate identification method for tracks of these species.
We collected samples from 18 zoos from Argentina, Brazil, Colombia, Paraguay, United States, and Venezuela, and urban areas from Argentina, 2004–2008.
(a) Heel pad: a1: total width, a2: total length, a3: width at the third quarter of the total length (a2 needed), a4: total area, a5: total perimeter. (b) Toes: b1 (i to
iv): total width, b2 (i to iv): total length, b3 (i to iv): width at the first third of the total length (b2i–b2iv needed), b4 (i to iv): width at the second third of the
total length (b2i–2biv needed), b5: the higher distance between toes and the heel pad. (c) Track: c1: total width, c2: total length. TL1: tangent line created
between the base of toes i and iii, TL2: tangent line created between the base of toes ii and iv, Angle D: angle between the longitudinal axes of i and iv toes,
Angle X: inferior angle formed between TL1 and TL2 lines.

Construction and Validation of Multivariate Models
We created a series of measurements and combinations of
measurements for representing quantitatively each of the
described qualitative traits for track differentiation (Appendix A). We also incorporated existing quantitative criteria
(Appendix B) as they were originally described and with
some modifications (e.g., see VarA2 and VarB3 in Appendix
C). This procedure resulted in 155 measurements and
variables that included linear, angular, and area dimensions,
2
proportional variables composed by ratios among
measurements, and shape variables as described by Lewison
et al. (2001). To reduce the total number of variables, we
randomly selected 20 tracks from each group (jaguars,
pumas, and big canids) and assessed variable performance in
track differentiation using box-plot graphs and indepenL

De Angelo et al. N Large Felid and Canid Track Identification

dent-sample t-tests. We maintained as preselected variables
only those that showed significant differences between
groups in univariate tests. We also checked these variables
for redundancy by using Pearson correlations. From groups
of highly correlated variables (r . 0.8), we selected only the
variable that showed the largest univariate difference
between groups (the highest t value). When 2 correlated
variables showed similar univariate differentiation capability,
we selected the variable that featured more simplicity for
precise measurement (Zielinski and Truex 1995).
We combined preselected variables and a set of tracks from
each group (original training cases) in discriminant function
analysis (DFA). We created models to differentiate 1) big
canid tracks from those of big felids, and 2) puma tracks
from those of jaguars. To construct each model, we
conducted stepwise variable selection using Wilks’ lambda
criterion, choosing the minimum number of less correlated
variables but ensuring the highest group differentiation
(min. partial F to enter 5 3.84; max. partial F to remove 5
2.71; SPSS 2008, Steinmetz and Garshelis 2008). We
validated models by percentage of original training cases
correctly classified, percentage of cross-validated cases
correctly classified (leave-one-out classification), and percentage of independent cases (tracks not used for model
development) correctly classified (SPSS 2008, Steinmetz
and Garshelis 2008). Additionally, we evaluated model
performance by estimating probabilities of group membership for all cases, both for misclassified and correctly classified training and independent tracks (see Stockburger 1998).
Jaguars are bigger than pumas and body size is correlated
with track size (Crawshaw 1995, Sunquist and Sunquist 2002);
therefore, absolute size proves to be useful in track differentiation (Childs 1998, Brown and Lopez Gonzalez 2001).
However, both species show high variation in size along their
L

3 categories during a slide presentation of photographs of
the tracks. Finally, we marked the achievement of each
volunteer by the percentage of correctly classified tracks, and
contrasted their performance with 67 randomly simulated
classifications.
We selected 4 categorical variables proposed for differentiation of big canids from felids (i.e., claw marks, shape of
heel-pad front area, shape of heel-pad base, and shape of the
inner side of outer toes; Appendix A), and we assessed
percentage of correctly classified tracks by each feature. We
used 33 dog, 46 puma, and 46 jaguar tracks to validate
existent quantitative classification criteria (Appendix B). We
also included 7 maned wolf tracks to assess whether they
were classified in the same group as dog tracks. For each
criterion, we observed the number of correctly and
misclassified tracks. Following the statement by Aranda
(1994) that any of the toes could be used in his criterion, we
used all toes in each track to test it.

1143

distribution and their size ranges overlap extensively (Iriarte et
al. 1990, Sunquist and Sunquist 2002). Most of our jaguar
tracks were bigger than puma tracks (pad width for pumas: ¯x 5
5.06 cm, SD 5 0.79, n 5 46; and for jaguars: ¯x 5 7.14 cm, SD
5 1.05, n 5 46). Inclusion of all tracks in model training
caused classification errors only with the biggest puma and
smallest jaguar tracks due to an overrated importance of size in
track classification. These intermediate size tracks are
commonly the most difficult tracks to identify, even for
experts. Accordingly, we selected the 50th percentile containing smaller jaguar tracks and the 50th percentile containing
bigger puma tracks as model-training tracks for puma versus
jaguar discriminant models to reduce the relative importance of
size-related variables. In addition, we produced a model with
all tracks without including absolute measurements related
with track size.
From the final variables used in discriminant models, we
selected easier-to-measure variables to make simpler discriminant models. We used these simpler models as steps to
construct track identification keys (Steinmetz and Garshelis
2008). We determined the ranges for step decisions by the
function discriminant scores (DS) that represented 90–99%
of posterior probabilities of group membership (Stockburger
1998). We started each key with the easiest-to-measure
variables and the simplest model, and then we used more
complex models in each step (adding more variables), to
make the identification process easy to perform.
We took all measurements using Auto Cad to the nearest
0.01 cm for linear measurements, 0.01 cm2 for areas, and 1u
for angles. We assessed normality and homoscedasticity of
variables with the Shapiro–Wilk and Levene’s test and
applied reciprocal, natural logarithm, square, or cubic
transformation when needed. We used the Box’s M statistic
to test for the homogeneity of covariance matrices. We
conducted all statistical analyses using SPSSE for Windows
statistical package version Rel.11.0.1.2001 (LEAD Technologies, Inc., Chicago, IL).

RESULTS
Reliability of Previously Described
Identification Methods
Volunteers correctly classified 61.5 6 1.4% (x
¯ 6 SE) of
tracks, a larger percentage than that obtained by a random
classification (x
¯ 5 35.1%, SE 5 1.0%, n 5 67; Mann–
Whitney test: Z 5 29.45, P
0.001). However, nobody
correctly classified all tracks, and participants showed a wide
range of identification accuracy (37–87%) with higher
variation than that observed in random classification
(volunteers classification SD 5 11.2%, random classification
SD 5 8.1%; F66,66 5 1.9, P , 0.02; Sokal and Rohlf 1995).
Although 67% of tracks were correctly identified by .50%
of volunteers, a group of 10 hard-to-identify tracks
(including canid, puma, and jaguar tracks) were incorrectly
identified by .50% of volunteers.
Criterion described by Belden (1978; Appendix B) showed
.20% of dog and puma tracks misclassified but had better
performance identifying maned wolf tracks as canids and
jaguar tracks as felids (Table 1). Criterion described by
M

1144

Table 1. Classification accuracy of traditional quantitative methods
described to differentiate big canids versus big felids tracks (mainly
described for dog vs. puma tracks) and puma versus jaguar tracks. We made
our evaluation using tracks from 18 zoos of Argentina, Brazil, Colombia,
Paraguay, United States, and Venezuela, and urban areas from Argentina,
2004–2008 (dog n 5 33, maned wolf n 5 7, puma n 5 46, jaguar n 5 46).
See description of criteria in Appendix B.
% of misclassified tracks
Criterion
a

Belden (1978)
Smallwood and Fitzhugh
(1989)a
Shaw et al. (2007)b
Aranda (1994)c,d
Childs (1998)ac,e

Dog

Maned wolf

Puma

Jaguar

45

0

24

4

21

14

17
0
87
15

9
0
0
18

a

Criterion described to differentiate dog vs. puma tracks.
Criterion described only to characterize puma tracks.
c
Criterion described to distinguish puma vs. jaguar tracks.
d
Criterion evaluated independently for each toe print of the tracks.
e
Criterion described for rear tracks and evaluated only with 17 jaguar and
13 puma tracks, because many of the collected tracks did not have accurate
information in this aspect.
b

Smallwood and Fitzhugh (1989) presented better reliability
in dog and puma track differentiation, but showed a higher
error with maned wolf and jaguar tracks than the method
described by Belden (1978; Table 1). All puma and jaguar
tracks had .3.5 cm of heel-pad width as described by Shaw
et al. (2007) for puma (Table 1). To differentiate puma and
jaguar tracks, criterion by Aranda (1994; Appendix B)
showed the highest error, misclassifying .80% of puma
toes, although all jaguar toes were correctly identified
(Table 1). The ratio of pad area to track area of rear tracks
described by Childs (1998) had better classification accuracy
(Table 1), but we evaluated accuracy for 17 jaguar and 13
puma tracks only, and 23% of evaluated tracks fell into the
undefined intermediate range. Considering the total heelpad width (Appendix B), ranges between rear tracks of
puma and jaguar from zoos overlapped (puma range 5
4.07–5.50 cm; jaguar range 5 5.18–8.99 cm) with one puma
and one jaguar rear track in the overlapping range (Childs
1998, Brown and Lopez Gonzalez 2001; Appendix B).
Two categorical features showed the expected pattern with
.70% of puma and dog tracks correctly classified, but the
other 2 categorical variables misclassified .60% of dog
tracks (Fig. 2). Claw marks were absent in 93.5% of puma
tracks whereas 97% of dog tracks presented claw marks
(Fig. 2A). The front of the heel pad was flatter or concave in
84.8% of puma tracks and 72.7% of dog tracks presented
rounded or pointed shape in the front of the heel pad. Jaguar
and puma tracks showed similar classification percentages
using these variables, as did maned wolf and dog tracks
(Fig. 2A, B), but no variable allowed differentiation of
100% of tracks among these species. Contrary to what was
expected, 63.6% of dog tracks showed a tri-lobbed heel-pad
base and 72.7% presented a rounded inner part of the outer
toes (Fig. 2C, D). In addition, it was sometimes difficult to
classify, unambiguously, tracks into the categories defined by
these last 2 categorical variables (i.e., shape of the base of the
heel pad and inner shape of the outer toes).
The Journal of Wildlife Management N 74(5)

Figure 2. Percentage of tracks bearing different characteristics related with 4 categorical variables used for canid and felid track differentiation. A) Presence
or absence of claw marks; B) shape of the front of the heel pad; C) shape of the base of the heel pad; D) shape of the inner part of the outer toes. We evaluated
tracks from pumas (n 5 46), jaguars (n 5 46), dogs (n 5 33), and maned wolves (n 5 7) that we collected from 18 zoos from Argentina, Brazil, Colombia,
Paraguay, United States, and Venezuela, and urban areas from Argentina, 2004–2008. See a complete description of these characteristics in Appendix A.

Multivariate Model Construction and Validation
We preselected 35 variables that showed more differentiation capability between the groups we analyzed. Stepwise
DFA starting with these variables resulted in 3 discriminant
models (Table 2; Fig. 3) constructed by different combinations of 10 of these preselected variables (see Appendix C).
For big canid versus big felid track contrast, we selected a
model composed by 3 variables that proved to be .98%
accurate in all validation procedures (model A1; Table 2;
Fig. 3A). Classification errors of this model occurred with
one puma and one jaguar track that were classified as canid,
with group probabilities of 69% and 88%, respectively
(Table 3). However, 82% of training and 83% of independent tracks from felids and canids were correctly classified
De Angelo et al. N Large Felid and Canid Track Identification

with high membership probabilities (.95%; Table 3), with
pumas having lower probabilities than jaguars within the
felid group (Fig. 3A).
To differentiate puma and jaguar tracks, we selected 2
models (models B1, B2; Table 2; Fig. 3B, C). We generated
model B1 from a stepwise process starting with all
preselected variables, and model B2 only used sizeindependent variables. Model B1 included transformed
pad width (VarB1) as a size parameter, so we developed
this model using only bigger puma tracks (x
¯ 5 5.67 cm, SD
5 0.65 cm, range 5 4.99–7.46 cm, n 5 23) and smaller
jaguar tracks (x
¯ 5 6.35 cm, SD 5 0.53 cm, range 5 4.99–
7.02, n 5 23), whereas we used excluded tracks only for
model validation. Model B1 had higher group separation
1145

Table 2. Parameters, variables incorporated, discriminant functions, and validation results for discriminant models selected for felid versus canid (A1) and
for puma versus jaguar track differentiation (B1 and B2). To develop these models we collected samples from 18 zoos from Argentina, Brazil, Colombia,
Paraguay, United States, and Venezuela, and urban areas from Argentina, 2004–2008.
Validation
Groups

Model

Felids vs. canids

A1

Jaguars vs. pumas

B1

B2

Wilks’ lambda (l)
and associated x2

Discriminant functiona

DS 5 10.711 3 VarA1 + 1.563 3 VarA3 + 0.047
3 Angle X 2 16.126
DS 5 5.412 3 VarB1 + 2.806 3 VarB2 + 10.865
3 VarB4 + 0.342 3 VarB5 2 8.936 3 VarB7
2 0.261 3 VarB8 2 0.942 3 VarB9 2 15.765
l 5 0.282
DS 5 4.052 3 VarB2 + 7.710 3 VarB4 + 0.207
x24 5 110.19*
3 VarB5 2 3.525 3 VarB7 2 0.072 3 VarB8
2 1.672 3 VarB9 2 7.659

l
x23
l
x24

5
5
5
5

0.280
163.1*
0.206
63.97*

Original
cases

CrossIndependent
validation
casesb

98.5%

98.5%

100.0%

100%

93.5%

100.0%

96.7%

94.6%

95.0%

a

Function used to calculate the discriminant score (DS) to classify tracks. See Appendix C for variable descriptions.
Independent validation in model B1 consists of independent cases plus excluded cases.
* P , 0.001.
b

(i.e., lower Wilks’ lambda) and higher values in the 3
validation methods than model B2 (Table 2). No original,
excluded, or independent tracks were misclassified and 90%
of tracks were classified with .95% of membership
probabilities (Table 3; Fig. 3B). Model B1 incorporated 7
variables that needed 19 measurements (Table 2).
Model B2 incorporated 6 size-independent variables (19
measurements needed; Table 2). Although model B2 had
lower performance than model B1 (Table 2), it misclassified
only 1 jaguar and 3 puma tracks, but with medium to high
group probabilities (58–95%; Table 3; Fig. 3C).
We constructed 3 identification keys: one for the contrast
between canids and felids and 2 for identifying puma from
jaguar tracks (with and without size-related variables,
respectively).
Key 1.—Big canid versus big felid track identification.
Step A: Calculate VarA1 (see Appendix C)
A1. ,0.41: canid.
A2. .0.67: felid.
A3. 0.41–0.67: go to step B.
Step B: Calculate discriminant score using:
DS 5 12.532 3 VarA1 + 1.962 3 VarA3 2 12.458
B1. DS , 21.60: canid.
B2. DS . 0.01: felid.
B3. DS 5 21.60–0.01: use discriminant model A1.
This key used 2 variables incorporated in 2 steps (6
measurements needed to complete the key). This key failed
to classify 19% of training tracks and 20% of independent
tracks, which then required use of discriminant model A1 to
attain a final classification.
Key 2.—Puma versus jaguar track identification using
track size.
Step A: Measure the heel-pad width (measurement a1)
A1. ,4.5 cm: puma.
A2. .7.9 cm: jaguar.
A3. 4.5–7.9 cm: go to step B.
Step B: Calculate discriminant score using:
1146

DS 5 6.528 3 VarB1 + 0.077 3 VarB8 2 14.627
B1. DS , 22.40: puma.
B2. DS . 1.95: jaguar.
B3. DS 5 22.40–1.93: go to step C.
Step C: Calculate discriminant score using:
DS 5 5.890 3 VarB1 2 8.088 3 VarB7 + 0.024
3 VarB8 2 9.830
C1. DS , 21.90: puma.
C2. DS . 1.80: jaguar.
C3. DS 5 21.90–1.80: go to step D.
Step D: Calculate discriminant score using:
DS 5 6.539 3 VarB1 + 10.222 3 VarB4 2 10.762
3 VarB7 2 0.039 3 VarB8 2 6.029
D1. DS , 21.00: puma.
D2. DS . 1.55: jaguar.
D3. DS 5 21.00–1.55: use discriminant model B1.
This key used total heel-pad width in the first step but
because it was a one-variable-size-related step, we selected
an extreme range (the chosen values represent .99% of
membership probabilities in one-variable discriminant
model). The key required 4 variables (13 measurements),
and 41% of training, 4% of excluded, and 25% of
independent tracks remained unidentified and required the
use of discriminant model B1 for further discrimination.
Key 3.—Puma versus jaguar track identification without
using track size.
Step A: Calculate heel pad:track area ratio (VarB8;
Childs 1998)
A1. ,24.5%: puma.
A2. .50%: jaguar.
A3. 24.5–50%: go to step B.
Step B: Calculate discriminant score using:
DS 5 3.989 3 VarB7 2 0.094 3 VarB8 + 3.156
3 VarB9 2 2.142
B1. DS . 1.75: puma.
The Journal of Wildlife Management N 74(5)

B2. DS , 22.26: jaguar.
B3. DS 5 22.26–1.75: go to step C.
Step C: Calculate discriminant score using:
DS 5 7.885 3 VarB4 2 4.043 3 VarB7 + 0.077
3 VarB8 2 2.487 3 VarB9 2 4.477
C1. DS , 21.40: puma.
C2. DS . 1.50: jaguar.
C3. DS 5 21.40–1.50: use discriminant model B2.
The key required 4 size-independent variables (17 measurements), and 53% of training and 55% of independent
tracks remained unidentified and required the use of
discriminant model B2 for further discrimination.

DISCUSSION
Described Methods for Jaguar and Puma
Track Identification
Although not a real field-track recognition situation, the
identification exercise showed that qualitative traits are
useful for track distinction. At the same time, it also
revealed ambiguity and subjectivity in criteria application
resulting in high variation in classification accuracy. All
participants were trained together in track recognition
before the exercise; however, variation in the performance
among volunteers was higher than that observed in random
classifications. Although volunteers (mainly biologists and
park rangers) had experience in fieldwork, not all of them
had the same expertise in track recognition. Previous
experience in identifying tracks is probably the main reason
for this variability in participants’ classification rate. Several
authors have mentioned the relevance of field expertise in
sign identification (Wemmer et al. 1996, Stander et al.
1997, Childs 1998, Shaw et al. 2007), but the frequent
assumption that people with field experience possess sign
recognition skills does not always hold true (Lynam 2002).
Existing quantitative univariate criteria showed low
accuracy in track classification. The technique by Belden
(1978) had a high misclassification rate (.20%) and
criterion by Smallwood and Fitzhugh (1989), though more
accurate, also misclassified nearly 20% of puma and dog
tracks. Likewise, for jaguar and puma track recognition we
obtained low classification rates with criterion by Aranda

r

Figure 3. Distribution of tracks along the variable with the highest
differentiation ability between groups (horizontal axis), against distribution
De Angelo et al. N Large Felid and Canid Track Identification

of tracks along discriminant models (vertical axis) obtained to differentiate
A) canid versus felid tracks, B) puma versus jaguar tracks using sizedependent variables, and C) puma versus jaguar tracks using sizeindependent variables. We included both training and independent tracks
in the graphs (jaguar n 5 56; puma n 5 56; dog n 5 44; maned wolf n 5
11). Horizontal continuous line indicates the limit among groups predicted
by the model, and horizontal dashed lines indicate the centroid value for
each group. VarA1: width at the third quarter of the heel pad, divided by
the total heel-pad width. VarB5: actual area of the heel pad divided by the
squared area of the track (total track width 3 total track length) expressed
as percentage. We collected samples from 18 zoos from Argentina, Brazil,
Colombia, Paraguay, United States, and Venezuela, and urban areas from
Argentina, 2004–2008.
1147

Table 3. Discriminant score ranges and their approximately associated probability for each discriminant model that we developed to differentiate tracks of
canids, pumas, and jaguars. Number of misclassified and well-classified training and independent tracks are shown for each range. To develop these models,
we collected samples from 18 zoos from Argentina, Brazil, Colombia, Paraguay, United States, and Venezuela, and urban areas from Argentina, 2004–2008.

Model Groups
A1

B1

B2

Canids

Centroid
value
22.411

Felids

1.048

Pumas

21.920

Jaguars

1.92

Pumas

21.579

Jaguars

1.579

Discriminant score
,22.000
22.000
21.085
20.738
20.624
20.278
.0.650
,21.2
20.769
20.360
20.051
0.052
0.361
.1.2
,21.450
21.450
20.438
20.063
0.064
0.439
.1.450

to
to
to
to
to

21.086
20.739
20.623
20.277
0.650

to
to
to
to
to

20.361
20.052
0.051
0.360
0.769

to
to
to
to
to

20.439
20.064
0.063
0.438
1.450

Misclassified tracks

Group
probabilitya

Training

.99% canid
80–99% canid
55–79% canid
Undeterminedb
55–79% felid
80–99% felid
.99% felid
.99% puma
80–99% puma
55–79% puma
Undeterminedb
55–79% jaguar
80–99% jaguar
.99% jaguar
.99% puma
80–99% puma
55–79% puma
Undeterminedb
55–79% jaguar
80–99% jaguar
.99% jaguar

0
1
1
0
0
0
0
0
0
0
0
0
0
0
0
1
0
1
0
2
0

Well-classified tracks

Excluded Independent
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
0
0

0
0
0
0
0
0
0

Training
26
13
1
0
8
26
56
18
3
2
0
0
6
17
26
14
3
1
4
15
25

Excluded Independent

22
0
1
0
0
2
21

10
3
2
0
1
6
13
8
1
1
0
0
0
10
4
3
2
0
1
3
6

a
Ranges of approx. group membership probabilities estimated using a posteriori probabilities calculated by SPSS for all cases (Stockburger 1998, SPSS
2008).
b
Although 50% probability determines the limit between 2 groups, we considered that the difference in values around 50% is too scarce to be used as
criterion for unknown track classification.

1148

toward including only well-defined tracks (easiest to identify
in the field; Lynam 2002, Sharma et al. 2003). However,
when using zoo tracks (or tracks from dogs in urban areas)
instead of field tracks, we are possibly introducing a bias
related to differences between natural substrates and zoo
substrates. Additionally, we would expect differences in paw
characteristics and pace of animals walking inside zoo cages
or dogs in urban areas versus animals walking freely in field
areas. Geographic variation in animals and soil condition
may also influence the misclassification rate of these
methods. All of these methods were originally developed
in a restricted region (i.e., criteria by Belden and by
Smallwood and Fizthugh in part of the United States,
criteria by Aranda in part of Mexico, and criteria by Childs
in the Brazilian Pantanal; Appendix B), though we
evaluated these methods with tracks from animals of many
zoos from 6 countries. Nevertheless, and despite the low
accuracy, all of these quantitative traits were preselected as
variables for the multivariate analysis and many of them
were used in final discriminant models (Appendix C).
Multivariate Track Classification Method
All discriminant models had statistically significant group
differentiation capabilities, and independent track classification had 95% discrimination accuracy in all models. As
mentioned above, the set of tracks used for construction and
validation of models included high variability because it
incorporated tracks from males and females, from rear,
front, right and left foot, as well as tracks from many
substrates and of various qualities. Thus, the high
discriminant rates we achieved with our models suggest
L

(1994), which misclassified most puma track toes. Heel
pad:track area ratio showed better results for distinguishing
rear tracks, but we only evaluated a few tracks and it is not
always possible to determine rear and front tracks under
field conditions. Heel-pad width of rear tracks presents the
same limitation and depends on regional size variation of
these species. Well-defined ranges in a specific region could
probably render this criterion locally useful for preliminary
track identification in the field.
Only 2 categorical variables were suitable for discriminating big canid from big felid tracks. Nevertheless, none of
these variables demonstrated a 100% classification rate and
sometimes tracks could not be included with certainty in the
categories defined by these variables. Subjectivity in variable
assessment could introduce important biases in track
classification. Smallwood and Fitzhugh (1989) found similar
results in their exploratory evaluation, and we agree with the
assessment by those authors that presence and absence of
claw marks and the front of the heel pad were the most
useful categorical variables.
A likely explanation for misclassifications using previously
described quantitative and categorical techniques is that we
selected a set of tracks with different qualities from different
individuals, zoos, and substrates, instead of using only highquality tracks collected in standardized and controlled
conditions. Most of these methods were originally tested
with only a few known tracks from zoos or mostly with
tracks collected in the field (see Appendix B), and this
practice could not only present the problem of possible
misidentification of the field track, but also introduce a bias

The Journal of Wildlife Management N 74(5)

L

L

De Angelo et al. N Large Felid and Canid Track Identification

confidence in final decisions. Substrate type and track
quality should also be considered in decisions; results from
low-quality tracks should be taken with caution.
Animal age could also influence track classification when
using key 2 and model B1, which are size-dependent.
However, at the age of puma and jaguar dispersal (around
1.5–2 yr old; Sunquist and Sunquist 2002), most juveniles
reach a size similar to an adult animal, particularly their paw
size (Crawshaw 1995, Shaw et al. 2007). Juvenile tracks
included in our analysis were correctly classified by key 2 and
model B1. Considering the wide range of track size that we
analyzed and the reduction of the relative importance of size
made by excluding extreme-sized tracks, we feel confident
that key 2 and model B1 should be useful with dispersing
subadults’ tracks. Key 3 and model B2, which are sizeindependent, could be an alternative method to avoid
potential problems related with track size. Nevertheless, we
suggest not using any of the keys and models with tracks
,3.5–4 cm of pad width, to avoid confusion with younger
animals and smaller canid or felid species not considered in
our analysis. Tracks of kittens or predispersal pumas and
jaguars will often be found with their mother’s tracks, thus
facilitating their track recognition (Shaw et al. 2007).
Our multivariate analyses present 2 main limitations in
comparison with qualitative or univariate quantitative
methods. First, some measurements are difficult to get,
the process of obtaining them is time-consuming, and track
digitalization is needed (i.e., total area and perimeter of the
heel pad). Second, variable construction and discriminant
function computation require mathematical skills. However, advanced technology and software are available for easy
and precise image digitalization and for automatic
calculations. Moreover, the digitalization process is facilitated by new methods for track collection (e.g., digital
photographs or track-tracing and image scanner), and
identification keys simplify the measurement and identification processes. We also incorporated the keys and models
in a spreadsheet file uploaded as an online supplemental
material related to this article at ,www.wildlifejournals.
org. (Supplemental material 1). Because we constructed
identification keys by incorporating few measurements in
each step and keys classified .50% of tracks (ensuring
.90% of membership probabilities), we offer the spreadsheet as a guide to key use during the measurement
process. It may also be helpful for the final models if the
keys do not identify tracks, which does not necessarily
make multivariate classification practical for field track
identification, but it will help to reduce the time needed
for arriving at confident classification results.
L

that they are robust, and they could be used to classify
unknown tracks in field studies.
However, all models had 1 misclassified track in crossvalidation classification, and occasionally in original cases
reclassification, which emphasized the importance of taking
into account group membership probabilities when using
these models (Stockburger 1998; Table 3). Most tracks were
correctly classified in all models with probabilities .90%,
and most errors occurred in tracks classified as belonging to
an incorrect group with low or medium probabilities.
The big canid versus big felid discriminant model (A1) and
identification key (key 1) used size-independent variables;
thus, their application apparently does not have limitations
due to regional size variation. In addition, correct classification of both domestic dogs and maned wolf tracks as canids
suggests application of discriminant model A1 and identification key 1 to differentiate big felid tracks from those of any
large canid. We made a preliminary analysis with 6 coyote
and 8 gray wolf tracks obtained from field guides of North
American mammals (Childs 1998, Murie 1998, Paul and
Gibson 2005, Reid 2006) and in all cases the classification key
and A1 model correctly classified these tracks as canids (most
of them with .95% membership probabilities). However,
this method should be further evaluated with more tracks
from these canid species where they could live in sympatry
with pumas or jaguars.
In the comparison of puma versus jaguar models, we
incorporated heel-pad width in model B1 as a variable
directly related with track size as used by many authors
(Fjelline and Mansfield 1988, Zielinski and Truex 1995,
Childs 1998, Sharma et al. 2003). By excluding extreme-size
tracks from training sets, we reduced the relative importance
of track size, thereby improving overlapping-size track
classification and overall classification success. The smallest
jaguar tracks we used in the analysis had 4.99 cm and the
biggest puma tracks had 7.46 cm total heel-pad width. Both
of these tracks were correctly classified by model B1. Model
B2 used a similar combination of variables but the exclusion
of the heel pad diminished its classification accuracy.
In spite of the robustness and accuracy demonstrated by
these models, differences between zoo and field tracks could
be a source of error that we did not consider in our analysis.
For this reason, when identification of a species is regarded
as having high practical relevance (e.g., reporting a new
locality for the species or needing to identify the predator in
human–predator conflicts), we recommend having
2
tracks classified with medium or high probabilities of group
membership (.80%) before arriving at definitive conclusions. Additionally, given both the existence of track
deformation that could lead to misclassification and the
risk of several individuals from different species leaving
tracks in the same place, we suggest measuring and
classifying each track independently instead of using a mean
value for a group of tracks (Zielinski and Truex 1995).
Because of front and hind foot differences in puma and
jaguar tracks (Brown and Lopez Gonzalez 2001), we suggest
using as many tracks of a set as possible to compare
membership assignment probabilities and have more

Unknown Track Identification
Qualitative and categorical traits performed well for track
identification. In addition, qualitative criteria include not
only features of isolated tracks, but also features of a trail of
tracks from the same individual (e.g., Childs 1998,
Hoogesteijn 2007, Shaw et al. 2007). However, these traits
are subjective and their effectiveness frequently relies on
high-quality track impressions and observer expertise
1149

(Zielinski and Truex 1995, Lynam 2002, Sharma et al.
2003). Besides, finding good-quality tracks or complete
track sets is difficult in some areas due to substrate condition
and animal behavior (Sharma et al. 2003), and many surveys
employ volunteers or students with low expertise in track
recognition (e.g., Smallwood and Fitzhugh 1995, Wydeven
et al. 2004, Danielsen et al. 2005, Markovchick-Nicholls et
al. 2008, De Angelo 2009).
Multivariate track recognition using discriminant functions could be useful to solve problems related with poorquality tracks and observer subjectivity, and has been
successfully used to differentiate tracks of American martens
(Martes americana) from tracks of fishers (M. pennanti;
Zielinski and Truex 1995); those of mink (Mustela vison)
from those of polecats (Mustela putorius; Harrington et al.
2008); claw marks of Asiatic black bears (Ursus thibetanus)
from those of sun bears (Helarctos malayanus; Steinmetz and
Garshelis 2008); and also for gender discrimination of tiger
tracks (Panthera tigris; Sharma et al. 2003). Following a
similar approach, we developed a robust multivariate
classification method to recognize puma and jaguar tracks
with confidence. Because we used tracks collected by
different people in an unstandardized way, our method
might be used to identify tracks obtained in the field from
different sources and methods of collection. Additionally,
selected variables and discriminant models may help to
develop more accurate techniques for track differentiation in
standardized track surveys in the future (e.g., when using
smoothed plots; Wemmer et al. 1996). For example, a more
detailed analysis using more tracks from each individual will
improve ability for species differentiation, and independent
classification of front and rear tracks may help to
differentiate puma and jaguar tracks with more confidence.
This approach could help to develop similar methods for
male and female track identification of these species
(Sharma et al. 2003), or for identifying other sympatric
felid or canid species as well.
To apply our methods in unknown track samples from the
field we suggest researchers 1) use keys and models in tracks
with .4 cm of heel-pad width; 2) start with identification
keys to distinguish tracks with confidence while using the
minimum number of variables and mathematical calculation;
3) use complete discriminant models (Table 2) when
identification keys do not classify the track; and 4) compare
the resultant discriminant score with our classification
results (Table 3) to get an estimate of confidence limits.
In addition, we recommend using model B2 in areas where
pumas are large (heel-pad width 7 cm) or jaguars are small
(heel-pad width 5.50 cm) or when there is no information
on track size (as in photographs obtained without a scale of
reference).
L

M

MANAGEMENT IMPLICATIONS
L

Our multivariate classification method has
3 main
applications in jaguar and puma conservation and management. First, our method offers a systematic approach for
using jaguar and puma tracks with more confidence for
surveying species presence, relative abundance, and habitat
1150

use, without the need to rely on the opinion of experts in
track identification. Secondly, tracks could be used with
more confidence as a complementary source of information
for research (e.g., helping in camera-trap localization) and
conservation plans (e.g., detecting the use of biological
corridors). Finally, tracks are often additional evidence
collected for identifying predators in human–predator
conflicts (Leite et al. 2002, Hoogesteijn 2007, Shaw et al.
2007). Tracks may not only help identify the species in
attacks to livestock or humans, but also determine presence
of potentially dangerous animals in urbanized or recreational
areas. Management actions may differ depending on the
predator species (puma, jaguar, or big canid); thus, correct
track identification in these cases is extremely important.

ACKNOWLEDGMENTS
We thank Proyecto Yaguarete´ volunteers for their participation in the classification exercise and sample collection.
We are grateful to M. Avalos, P. Brandolı´n, G. Boaglio, E.
Catala´, L. Chiyo, A. Colma´n, P. Cruz, W. de Moraes, Y.
Di Blanco, J. Earnhardt, C. Ferrari, M. Guevara, M.
Mealla, S. Moro, N. Perez, V. Quiroga, M. Ramı´rez, R.
Rivero, C. Schloss, A. Sestelo, D. Villarreal, and all zoos’
authorities and personnel for helping in track acquisition
and processing. G. De Angelo and C. De Angelo gave us
technical assistance in software management. L. Fitzhugh,
R. Steinmetz, and Y. Jhala assisted with bibliography. We
are thankful to J. Earnhardt, C. Boiero, E. Gese, S.
Smallwood, D. Hanseder, and an anonymous reviewer for
comments and suggestions for improving this manuscript.
Financial support for this project was provided by National
Research Council of Argentina (CONICET), Fundacio´n
Vida Silvestre Argentina, Education for Nature Program of
the World Wildlife Fund (WWF), WWF-International,
WWF-Switzerland, and Lincoln Park Zoo.

LITERATURE CITED
Aranda, M. 1994. Differentiation of jaguar and puma footprints: a
diagnostic analysis. Acta Zoolo´gica Mexicana (n.s.) 63:75–78. [In
Spanish.]
Azevedo, F. C. C. 2008. Food habits and livestock depredation of sympatric
jaguars and pumas in the Iguacu National Park area, South Brazil.
Biotropica 40:494–500.
Belden, R. C. 1978. How to recognize panther tracks. Proceedings of the
Annual Conference of Southeastern Association of Fish and Wildlife
Agencies 32:112–115.
Brown, D. E., and C. A. Lopez Gonzalez. 2001. Borderland jaguars: tigres
de la frontera. University of Utah Press, Salt Lake City, USA.
Childs, J. L. 1998. Tracking the felids of the borderlands. Printing Corner
Press, El Paso, Texas, USA.
Conroy, M. J., P. Beier, H. Quigley, and M. R. Vaughan. 2006. Improving
the use of science in conservation: lessons from the Florida panther.
Journal of Wildlife Management 70:1–7.
Crawshaw, P. G., Jr. 1995. Comparative ecology of ocelot Felis pardalis and
jaguar Panthera onca in a protected subtropical forest in Brazil and
Argentina. Dissertation, University of Florida, Gainesville, USA.
Crooks, K. R. 2002. Relative sensitivities of mammalian carnivores to
habitat fragmentation. Conservation Biology 16:488–502.
Danielsen, F., N. Burgess, and A. Balmford. 2005. Monitoring matters:
examining the potential of locally-based approaches. Biodiversity and
Conservation 14:2507–2542.
De Angelo, C. 2009. El paisaje del Bosque Atla´ntico del Alto Parana´ y sus
efectos sobre la distribucio´n y estructura poblacional del jaguar (Panthera
The Journal of Wildlife Management N 74(5)

onca) y el puma (Puma concolor). Dissertation, Universidad de Buenos
Aires, Buenos Aires, Argentina. [In Spanish.]
De Angelo, C., A. Paviolo, Y. Di Blanco, and M. Di Bitetti. 2008. Guı´a de
huellas de los mamı´feros de Misiones y otras a´reas del Subtro´pico de
Argentina. Ediciones del Subtro´pico, San Miguel de Tucuma´n,
Argentina. [In Spanish.]
Fjelline, D., and T. M. Mansfield. 1988. Method to standardize the
procedure for measuring mountain lion tracks. Pages 49–51 in R. H.
Smith, editor. The Third Mountain Lion Workshop. The Wildlife
Society and Arizona Game and Fish Department, Prescott, USA.
Gompper, M. E., R. W. Kays, J. C. Ray, S. D. Lapoint, D. A. Bogan, and J.
R. Cryan. 2006. A comparison of noninvasive techniques to survey
carnivore communities in northeastern North America. Wildlife Society
Bulletin 34:1142–1151.
Gordon, C., A. M. Stewart, and E. Meijaard. 2007. Correspondence
regarding ‘Clouded leopards, the secretive top-carnivore of South-East
Asian rainforests: their distribution, status and conservation needs in
Sabah, Malaysia’. BMC Ecology 7:5.
Grigione, M. M., P. Burman, V. C. Bleich, and B. M. Pierce. 1999.
Identifying individual mountain lions Felis concolor by their tracks:
refinement of an innovative technique. Biological Conservation 88:25–32.
Harrington, L. A., A. L. Harrington, and D. W. Macdonald. 2008.
Distinguishing tracks of mink Mustela vison and polecat M. putorius.
European Journal of Wildlife Research 54:367–371.
Hoogesteijn, R. 2007. Manual on the problem of depredation caused by jaguars
and pumas on cattle ranches. Jaguar Conservation Program - Wildlife
Conservation Society. ,http://www.savethejaguar.org/jag-conservation..
Accessed 6 May 2008.
Iriarte, J. A., W. L. Franklin, W. E. Johnson, and K. H. Redford. 1990.
Biogeographic variation of food habits and body size of the America
puma. Oecologia 85:185–190.
Isasi-Catala´, E., and G. R. Barreto. 2008. Identificacio´n de individuos de
jaguares (Panthera onca) y pumas (Puma concolor) a partir de morfometrı´a
de sus huellas (Carnivora: Felidae). International Journal of Tropical
Biology 56:1893–1904.
Karanth, K. U., J. D. Nichols, J. Seidensticker, E. Dinerstein, J. L. D.
Smith, C. McDougal, A. J. T. Johnsingh, R. S. Chundawat, and V.
Thapar. 2003. Science deficiency in conservation practice: the monitoring
of tiger populations in India. Animal Conservation 6:141–146.
Leite, R., T. G. de Oliveira, R. C. de Paula, and C. Indrusiak. 2002.
Manual de identificac¸a˜o, prevenc¸a˜o e controle de predac¸a˜o por carnı´voros.
Instituto Brasileiro do Meio Ambiente e dos Recursos Naturais, Brasilia,
Brasil. [In Portuguese.]
Lewison, R., E. L. Fitzhugh, and S. P. Galentine. 2001. Validation of a
rigorous track classification technique: identifying individual mountain
lions. Biological Conservation 99:313–321.
Lynam, A. J. 2002. Me´todos de trabajo de campo para definir y proteger
poblaciones de gatos grandes: los tigres indochinos como un estudio de
caso. Pages 55–71 in R. A. Medellı´n, C. Equihua, C. L. Chetkiewicz, P.
G. J. Crawshaw, A. Rabinowitz, K. H. Redford, J. G. Robinson, E. W.
Sanderson, and A. B. Taber, editors. El jaguar en el nuevo milenio.
Ediciones Cientı´ficas Universitarias, Me´xico Distrito Federal, Me´xico.
[In Spanish.]
Markovchick-Nicholls, L., H. M. Regan, D. H. Deutschman, A.
Widyanata, B. Martin, L. Noreke, and T. Ann Hunt. 2008.
Relationships between human disturbance and wildlife land use in urban
habitat fragments. Conservation Biology 22:99–109.
Murie, O. J. 1998. A field guide to animal tracks. Second edition.
Houghton Mifflin, New York, New York, USA.
Nowell, K., and P. Jackson. 1996. The wild cats: status survey and
conservation action plan. International Union for Nature ConservationCat Specialist Group, Gland, Switzerland.
Paul, W. J., and P. S. Gibson. 2005. Wolves. In S. Hygnstrom, P. Curtis,
G. Yarrow, R. Schmidt, and S. Vantassel, editors. Internet center for
wildlife damage management. Distance education and wildlife damage
management at the University of Nebraska. ,http://icwdm.org/handbook/
carnivor/Wolves.asp.. Accessed 8 Aug 2008.
Paviolo, A., C. De Angelo, Y. Di Blanco, and M. Di Bitetti. 2008. Jaguar
population decline in the Upper Parana´ Atlantic Forest of Argentina and
Brazil. Oryx 42:554–561.
Perovic, P., and M. Herran. 1998. Distribucio´n del jaguar (Panthera onca)
en las provincias de Jujuy y Salta, Noroeste de Argentina. Mastozoologı´a
Neotropical 5:47–52. [In Spanish.]
De Angelo et al. N Large Felid and Canid Track Identification

Potvin, M. J., T. D. Drummer, J. A. Vucetich, D. E. Beyer, Jr., R. O.
Peterson, and J. H. Hammill. 2005. Monitoring and habitat analysis for
wolves in upper Michigan. Journal of Wildlife Management 69:1660–
1669.
Rabinowitz, A., and B. G. Nottingham, Jr. 1986. Ecology and behaviour of
the jaguar (Panthera onca) in Belize, Central America. Journal of Zoology
210:149–159.
Reid, F. 2006. Peterson field guide to mammals of North America. Fourth
edition. Houghton Mifflin, New York, New York, USA.
Riordan, P. 1998. Unsupervised recognition of individual tigers and snow
leopards from their footprints. Animal Conservation 1:253–262.
Sanderson, E. W., K. H. Redford, C. L. B. Chetkiewicz, R. A. Medellin,
A. R. Rabinowitz, J. G. Robinson, and A. B. Taber. 2002. Planning to
save a species: the jaguar as a model. Conservation Biology 16:58–72.
Schaller, G. B., and P. G. Crawshaw, Jr. 1980. Movement pattern of jaguar.
Biotropica 12:161–168.
Scognamillo, D., I. E. Maxit, M. Sunquist, and J. Polisar. 2003. Coexistence
of jaguar (Panthera onca) and puma (Puma concolor) in a mosaic landscape in
the Venezuelan llanos. Journal of Zoology 259:269–279.
Sharma, S., Y. Jhala, and V. B. Sawarkar. 2003. Gender discrimination of
tigers by using their pugmarks. Wildlife Society Bulletin 31:258–264.
Shaw, H. G., P. Beier, M. Culver, and M. M. Grigione. 2007. Puma field
guide. The Cougar Network, Concord, Massachusetts, USA.
Smallwood, K. S., and E. L. Fitzhugh. 1989. Differentiating mountain lion
and dog tracks. Pages 15–18 in R. H. Smith, editor. Proceedings of
Mountain Lion Workshop, third. The Wildlife Society and Arizona
Game and Fish Department, Prescott, USA.
Smallwood, K. S., and E. L. Fitzhugh. 1993. A rigorous technique for
identifying individual mountain lions Felis concolor by their tracks.
Biological Conservation 65:51–59.
Smallwood, K. S., and E. L. Fitzhugh. 1995. A track count for estimating
mountain lion Felis concolor californica population trend. Biological
Conservation 71:251–259.
Soisalo, M. K., and S. M. C. Cavalcanti. 2006. Estimating the density of a
jaguar population in the Brazilian Pantanal using camera-traps and
capture–recapture sampling in combination with GPS radio-telemetry.
Biological Conservation 129:487–496.
Sokal, R. R., and F. J. Rohlf. 1995. Biometry: the principles and practice of
statistics in biological research. Third edition. W. H. Freeman, New
York, New York, USA.
SPSS. 2008. SPSS 14.0 statistical algorithms. SPSS. ,http://support.spss.
com/ProductsExt/SPSS/Documentation/Statistics/algorithms/index.
html.. Accessed 23 Aug 2008.
Stander, P. E. 1998. Spoor counts as indices of large carnivore populations:
the relationship between spoor frequency, sampling effort and true
density. Journal of Applied Ecology 35:378–385.
Stander, P. E., I. Ghau, D. Tsisaba, I. Oma, and I. IUi. 1997. Tracking and
the interpretation of spoor: a scientifically sound method in ecology.
Journal of Zoology 242:329–341.
Steinmetz, R., and D. L. Garshelis. 2008. Distinguishing Asiatic black
bears and sun bears by claw marks on climbed trees. Journal of Wildlife
Management 72:814–821.
Stockburger, D. W. 1998. Multivariate statistics: concepts, models and
applications. The Psychology Department at Missouri State University.
,http://www.psychstat.missouristate.edu/multibook/mlt00.htm.. Accessed 26 Aug 2008.
Sunquist, M., and F. Sunquist. 2002. Wild cats of the world. University of
Chicago Press, Chicago, Illinois, USA.
Van Dyke, F. G., R. H. Brocke, and H. G. Shaw. 1986. Use of road track
counts as indices of mountain lion presence. Journal of Wildlife
Management 50:102–109.
Wemmer, C., T. H. Kunz, G. Lundie-Jenkins, and W. J. McShea. 1996.
Mammalian sign. Pages 157–176 in D. E. Wilson, F. R. Cole, J. D.
Nichols, R. Rudran, and M. S. Foster, editors. Measuring and
monitoring biological diversity—standard methods for mammals. Smithsonian Institution Press, Washington, D.C., USA.
Wilting, A., F. Fischer, S. Abu Bakar, and K. E. Linsenmair. 2006.
Clouded leopards, the secretive top-carnivore of South-East Asian
rainforests: their distribution, status and conservation needs in Sabah,
Malaysia. BMC Ecology 6:16.
Wydeven, A., A. Sabor, R. Schultz, R. Megown, S. Boles, and E.
Wiedenhoeft. 2004. Guidelines for carnivore tracking during winter in
Wisconsin. Wisconsin’s Volunteer Carnivore Tracking Program, Wis1151

Big Canid and Big Felid Tracks
Summary of traits described by Smallwood and Fitzhugh
(1989), Childs (1998), Brown and Lopez Gonzalez (2001)
and Shaw et al. (2007). These criteria were mainly described
for dogs and pumas, but most of these criteria are used for all
big canid versus big felid track differentiation (see Fig. 1 for
tracks comparison): 1) Presence of claw marks: canids have
nonretractable claws so it is more common to find claw marks
in the front of canid toes than in felid tracks; 2) Relative size
between toes and heel pad: canids have bigger toes than felids in
relation to the heel pad; 3) Heel-pad size and shape: canids
have a triangular heel pad, usually with rounded front and
with no-lobed or slightly bi-lobed back part. Felids have
bigger and more squared or rounded heel pad, with a flatter or
concave front part and tri-lobed back part; 4) Toes’ position:
the 2 middle toes (ii and iii) and the 2 outer toes (i and iv) of
canid tracks tend to be in the same line and at the same
distance to the heel pad, making the track nearly symmetric.
Sometimes, this toe configuration shows an X shape in the
space left by the toes and a mound of soil between toes and
the heel. In felids, the second toe tends to be positioned more
forward than the others, making nonsymmetric tracks; 5)
Outer toes’ shape and direction: felids’ outer toes point to the
front of the track. In canids, the first and the fourth toes
usually point to the sides. In addition, canid outer toes usually
show an angular shape in their inner edge, whereas those
edges in felids’ toes are more rounded; 6) Pattern of tracks:
almost erratic in dogs but more regular and linear in felids.

L

APPENDIX A: QUALITATIVE TRAITS
FOR TRACK DIFFERENTIATION

Dog Versus Puma Tracks
Belden (1978).—Ratio of the width of the widest toe to
the heel pad (see Fig. 1: max. between b1i–b4iv divided by
a1): if the ratio exceeds 0.44 the track belongs to a dog, and
below 0.44 the track belongs to a puma.
Smallwood and Fitzhugh (1989).—Angle formed by the
longitudinal axis of the 2 outer toes (Angle D in Fig. 1): if
the angle is 29u, the track probably belongs to a puma, if
the angle is 30u, the track is most likely from a dog
(following modifications proposed for this criterion by Shaw
et al. 2007). To construct this range, Smallwood and
Fitzhugh (1989) used 19 dog tracks and 48 puma tracks
identified in the field by qualitative criteria.
Shaw et al. (2007).—Even young puma kittens have a
heel-pad width of 3.5 cm (a1 in Fig. 1). Tracks with
,3.5 cm of heel-pad width are most likely from smaller
felid species or from dogs.
L
M

consin Department of Natural Resources. ,http://www.dnr.state.wi.us/
org/land/er/mammals/volunteer/.. Accessed 17 Jul 2008.
Zalewski, A. 1999. Identifying sex and individuals of pine marten using
snow track measurements. Wildlife Society Bulletin 27:28–31.
Zielinski, W. J., and R. L. Truex. 1995. Distinguishing tracks of closelyrelated species: marten and fisher. Journal of Wildlife Management
59:571–579.

Puma Versus Jaguar Track
Summary of features described by Schaller and Crawshaw
(1980), Aranda (1994), Childs (1998), and Brown and
Lopez Gonzalez (2001; see Fig. 1 for track comparison): 1)
Track size and shape: jaguars have bigger and more rounded
tracks than pumas; 2) Heel-pad size and shape: jaguars have
more rounded and bigger heel pad than pumas. Pumas have
3 prominent lobes in the base of the heel pad that are less
pronounced in jaguars. The middle lobe should be wider in
jaguar heel pads; 3) Toes’ shape and position: jaguars have
more rounded toes, but puma toes are more elongated and
pointed. Jaguar toes are positioned around the heel pad.
Puma toes tend to be farther from the heel pad.

Puma Versus Jaguar Tracks
Aranda (1994).—Any of the toes is divided into thirds.
The width at one-third is divided by the width at two-thirds
(see Fig. 1: b3 divided by b4 for any of the toes i–iv).
Confidence limits range from 0.71 to 1.03 for jaguar and
0.60 to 0.76 for puma. Values .0.76 mean jaguar track and
values ,0.71 mean puma track, with an undefined range of
0.71–0.76. Aranda (1994) used 10 tracks from 2 captive
jaguars and 8 tracks from 2 captive pumas, plus 70 jaguar
and 45 puma tracks from the field (Mexico) identified by
qualitative criteria.
Childs (1998) a.—Percent ratio of heel-pad square area
(width 3 length), to total track square area of the rear track
(VarB8 in Appendix C). Jaguar range 5 41.2–56.7%, puma
range 5 28.1–38.5%. Childs (1998) described this range
using 15 jaguar and 8 puma tracks from the field (Pantanal,
Brazil) identified by qualitative criteria, but there is no
information about the number of rear tracks out of the
total.
Childs (1998) b.—This author suggests that maximum
width of rear feet heel pad (a1 in Fig. 1) could be useful to
differentiate species and shows the range from Pantanal
(Brazil) tracks: jaguar 5 7.0–8.8 cm, puma 5 4.5–7.0 cm.
Brown and Lopez Gonzalez (2001) also mentioned this
criterion and provided data from Southern United States
and Northern Mexico: jaguar 5.1–8.9 cm, puma 4.1–5.6 cm.
Shaw et al. (2007) mention values between 4.1 cm and
6.3 cm for puma rear tracks from Arizona and California,
USA. Childs (1998) described this range using 15 jaguar
and 8 puma tracks from the field (Pantanal, Brazil)
identified by qualitative criteria, but there is no information
about the number of rear tracks out of the total. Brown and
Lopez Gonzalez (2001) and Shaw et al. (2007) do not
mention the source of their information.

APPENDIX B: EXISTING
QUANTITATIVE CRITERIA

APPENDIX C: VARIABLE
DESCRIPTIONS

Summary of quantitative criteria described in the literature
for dogs versus pumas and puma versus jaguar track
differentiation (see Fig. 1 for track comparison):

Description of variables incorporated in discriminant models
and identification keys using measurements from Figure 1.
We used these variables with tracks from dogs, maned wolfs,

1152

The Journal of Wildlife Management N 74(5)

pumas, and jaguars, collected from 18 zoos from Argentina,
Brazil, Colombia, Paraguay, United States, and Venezuela,
and urban areas from Argentina, 2004–2008. We included
the complete list of evaluated measurements and variables in
the Online Supplemental Material 2 (Fig. S1 and Table S1)
uploaded as online supplemental material of this article at
,www.wildlifejournals.org..
VarA1.—Width at the third quarter of the heel pad
(total length of the heel pad needed), divided by total heelpad width:
VarA1 5 a3 / a1
VarA2.—Average of toe total width, divided by heelpad total width (modified from Belden 1978):
VarA2 5 [(b1i + b1ii + b1iii + b1iv) / 4] / a1
VarA3.—Reciprocal transformation of VarA2:
VarA3 5 1 / VarA2
Angle X.—See Figure 1.
VarB1.—Natural logarithm of heel-pad total width:

VarB3.—Average of the first third width divided by
second third width of each toe (modified from Aranda
1994):
VarB3 5 [(b3i / b4i) + (b3ii / b4ii) + (b3iii / b4iii) +
(b3iv / b4iv)] / 4
VarB4.—Square transformation of VarB3:
VarB4 5 (VarB3)2
VarB5.—Actual area of the heel pad divided by the
squared area of the track (total track width 3 total track
length) expressed as percentage:
VarB5 5 [a4 / (c1 3 c2)] 3 100%
VarB7.—The higher distance between toes and the
heel pad divided by the heel-pad width:
VarB7 5 b5 / a1
VarB8.—Squared area of the heel pad divided by the
squared area of the track, expressed as percentage (from
Childs 1998):
VarB8 5 [(a1 3 a2) / (c1 3 c2)] 3 100%

VarB1 5 ln(a1)
VarB2.—Cubic transformation of the heel-pad shape
factor described by Lewison et al. (2001):
2

VarB2 5 [(4 3 p 3 a4) / (a5 )]

3

De Angelo et al. N Large Felid and Canid Track Identification

VarB9.—Mean of length:width ratio of toes 2 and 3:
VarB9 5 [(b2ii / b1ii) + (b2iii / b1iii)] / 2
Associate Editor: Gese.

1153


Documentos relacionados


Documento PDF de angelo et al 2010 jwm track identification
Documento PDF de angelo et al 2011 oryx monitoring of large carnivores
Documento PDF paviolo et al 2009 protection affect puma abundance and activity pattern
Documento PDF tesis doctoral ver nica quiroga 2013 jaguar y puma en chaco
Documento PDF de angelo et al 2013 biol cons a management landscape for jagaurs
Documento PDF quiroga et al 2014 jaguar in argentinean chaco oryx


Palabras claves relacionadas