Fig 1.
The figure shows the outcome of a pair-wise comparison of trails from the same individual (A) and two different individuals (B) based on a customized model in JMP software. The classifier incorporated into the model is based on the presence or absence of overlap between the ellipses. Note that the analysis is performed for each pairwise comparison in the presence of a third entity, i.e. the reference centroid value (RCV).
Fig 2.
This figure shows three puma footprints of poor (a), acceptable (b), and excellent (c) quality.
The main difference is the clarity of the outline of the heel and toes.
Fig 3.
An excellent quality puma footprint, photographed to the FIT protocol.
A metric scale is placed to the left and posterior edge of the footprint, and a photo ID slip containing details of the footprint and image is included. The image is taken from directly overhead.
Fig 4.
A puma footprint showing the placement of 25 landmark points (red circles) and 15 points derived from them and generated by the FIT script (yellow circles).
The landmark points and derived points are numbered in one sequence, providing 40 total points from which measurements (variables) of the footprint are made.
Table 1.
The number of variables extracted from the footprint images as lengths (L), angles and areas.
Table 2.
Footprint collection.
Fig 5.
The figure shows a model validation of the accuracy of classification in relation to the size of the test set against the training set.
The varying test set size was plotted against itself (red), the predicted value for each test size iteration (black) and the mean predicted value for each test size (blue). Optimal classification accuracy was obtained when the test set size was smallest relative to the training set. However, the robustness of the model was demonstrated by the predicted value being close to the expected value, even when the test set was considerably larger than the training set (24:11).
Fig 6.
This figure shows the classification dendrogram output from FIT.
Three trails of 77 were misclassified (indicated by an x), giving an overall accuracy of individual identification for correct trail placement of 96.11%.
Fig 7.
This figure shows the output when a relative likelihood slider is moved to the left, giving the relatively likelihood of 34 puma as 73.7%.
Fig 8.
This figure demonstrates that when the slider is moved to the right the likelihood of 36 cougars is 76.9%.
Fig 9.
Footprints collected from free ranging animals at Big Bend and Camino Cielo (labeled o) are classified correctly as two individuals different from each other and from all the others in the dataset.
Misclassified trails are again indicated by an x.
Fig 10.
The relationship between the number of variables used to classify by sex and the resulting percentage accuracy of classification, from which 20 was identified as an optimal number.
Fig 11.
Scatterplot showing the distribution of footprints by sex.
Red stars are footprints from females, and blue from males. The black squares and circles are from animals of unknown sex from Big Bend (classified as female) and Camino Cielo (classified as male) respectively.