Abstract | Keywords | Export | Availability | Bookmark

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

At a time when a sixth mass extinction is about to hit our planet, protection and conservation strategies are the best chances of survival of some wildlife populations. But for those strategies to be effective, the use of reliable monitoring techniques is essential to assess the distribution, dynamic and status of the targeted species. Considering the cost of direct observations and that of invasive high-tech tools, such as camera traps and GPS, collars can be, the use of tracks is a low-cost non-invasive alternative to study elusive species such as carnivores. 
In the present study, we evaluate the possibility of identifying the anteroposterior (front or hind) and mediolateral (right or left) position of spotted hyena tracks from their digital models created from field photography. Several combinations of data recording and feature extraction methods were tested so that we could compare the accuracy of prediction of their identification algorithm and determine which combination is the most reliable.
Track sampling, which consisted of photographing encountered tracks, took place in Hluhluwe-iMfolozi Park, in South Africa. 2D and 3D models of 80 tracks (20 from each position) were constructed using ImageJ and Photoscan software respectively. Landmarks were digitized on the models so that different types of measurements could be extracted by conducting either traditional or geometric morphometrics. Using extracted morphological features, Linear Discriminant Analyses (LDA) generated identification algorithms for each combination of methods. In total, the algorithms of 31 different scenarios were compared, each of which involved (i) a type of model (2D or 3D), (ii) a feature extraction method (traditional or geometric morphometrics), (iii) the types of landmarks used to characterize the form of the models (fixed, fixed and curve-sliders, or fixed and curve- and surface-sliders), (iv) a type of object on which statistical analyses were conducted (independent pads or entire track) , and (v) a type of variables taken into account by the algorithms (shape, size, or both).
Nine of the thirty-one scenarios were able to provide algorithms with accuracies of prediction > 95%. It appeared that the relative position of the pads within a track (i.e. the information provided by the “entire track” objects) as well as their sizes are two pieces of information that are essential for the position identification of spotted hyena track. However, before being able to establish which type of model and which type of landmarks provide the most accurate algorithm, the manipulator bias of each method should be quantified and used as a second evaluation criteria. The track modelling process should also be made more effective both in term of time and manipulator bias.

tracks --- digital 2D model --- digital 3D model --- traditional morphometrics --- geometric morphometrics --- position identification of tracks --- ecological monitoring --- spotted hyenas --- Crocuta crocuta --- Sciences du vivant > Sciences de l'environnement & écologie