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

Vista previa del archivo PDF de-angelo-et-al-2010-jwm-track-identification.pdf

Página 1 2 3 4 5 6 7 8 9 10 11 12 13

Vista previa de texto

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. (LEAD Technologies, Inc., Chicago, IL).

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


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

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


Maned wolf












Criterion described to differentiate dog vs. puma tracks.
Criterion described only to characterize puma tracks.
Criterion described to distinguish puma vs. jaguar tracks.
Criterion evaluated independently for each toe print of the tracks.
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.

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)