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


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