How FIT works

The footprint identification technique (FIT) is a tool for monitoring endangered species. It was developed by WildTrack with the help of bushmen trackers in Southern Africa. From digital images of footprints alone, FIT can identify at the species, individual, age-class or gender levels.

A full report of the technique as used for white rhino in Namibia has been published in the journal Endangered Species Research (ESR).

Alibhai, S.K., Jewell, Z.C. & Law P.R. (2008). Identifying white rhino (Ceratotherium simum) by a footprint identification technique, at the individual and species levels. Endangered Species Research 4: 219-225

There are two main stages in the FIT procedure; firstly the capture of footprints and conversion of a footprint into a geometric profile which serves as the data and secondly, analyses to which the data are submitted for the purpose of classification.

Identifying footprints (tracks) is a real challenge; there are four levels of complexity. Firstly each species has a different foot anatomy, and FIT must be flexible enough to produce a customised species algorithm; the set of variables which determine the characteristics of that species. Secondly, within each species, each individual has its own unique foot characteristics, analogous to our fingerprints. Thirdly, each individual has four feet, and it is necessary to identify which track results from each foot in order to standardise the analytical procedures. Lastly, each time the individual puts a foot on the ground, it leaves a slightly different track. This is determined by many factors, including the gait of the animal, substrate type, moisture levels and weather conditions. In order to account for the variation of tracks along the same trail (ie. made by the same animal), we collect 6-8 different tracks from each trail.

the front and back, left and right prints of a tiger

the front and back, left and right prints of a tiger

Digital images of the track made by that foot are collected, at a size of 1200×1600 or greater. Images are taken according to a standardised protocol. A scale is placed next to the track, a direct overhead view is used, and a photo-ID slip or voice-tag is attached to the image with details of data, track number, photographer etc. The resulting image and tag are uploaded to a laptop, and optimised using standard commercial image manipulation software.

The next step is the selection and positioning of anatomical ‘landmark’ points on the track. Landmark points are selected on the basis of foot anatomy to include those points which are clearly definable and repeatable across many tracks. After placing these landmark points a set of derived landmarks, geometrically constructed from the set of natural landmarks, is then defined. The derived points are positioned by JMP script into which their geometrical constructions have been entered. The full set of points is designed to allow all measurements that one anticipates might prove useful in discriminating between tracks. The geometric profile typically contains many measurements (more than 100 in application of FIT to the black rhino). The measurements consist of lengths, angles and areas (polygons approximating to the area).

The first step in the FIT analysis is the reduction of the total number of measurements to the set (the FIT algorithm) which provides a good discrimination at the level required. For this purpose, a library of footprints from known individuals is required. We collect prints from known wild animals, or from captive individuals for this one-off process to develop the species algorithm. Stepwise variable selection in simple discriminant analysis provides selection of those variables which will produce an effective classification. Once this step is complete, it is then possible to use FIT to discriminate for censusing (estimating an unknown population) or monitoring (following individuals in a known population). Normally the evaluation of the status of endangered species would begin with a census and then proceed with frequent monitoring updates.

For censusing we have developed a new Canonical Pairwise Comparison Technique (CPCT). For each pair of tracks, two groups are formed, one from each track, and in addition a third group, the Reference Centroid Value (RCV), consisting of all the footprints in the library other than those belonging to the two test tracks. The CPCT computes the two canonical variates for these three groups and inspects the corresponding plot in canonical space of group means and 95% confidence regions. If the confidence regions of the two test tracks overlap the tracks are from the same animal. If not, from different animals. Testing the accuracy of FIT for censusing the white rhino using three different methods we achieved accuracies of between 91 -99%.

CCTP_diagram-200-150For monitoring, we have developed the Canonical Elipse Reduction Technique (CERT). Here each known individual is represented by a set of footprints and a new, unknown, track is tested against these to see which it matches. JMP computes the first two canonical variates for these data and produces the corresponding 2-dimensional (canonical space) plot of group means and 95% confidence ellipses. All groups whose confidence ellipses do not overlap with the unknown track are set aside and the procedure repeated. If this routine ultimately results in a plot of the unknown track overlapping exactly one group, that group’s identity is assigned to the track. Otherwise the track remains unclassified. Testing the accuracy of FIT for monitoring white rhino using two different test methods gave accuracies of between 93% and 97%. Species identifications are in the range of 95 – 98% accuracy.

We have also successfully tested the use of Footprint identification as a recognition technique in mark-recapture studies. Here it provides an opportunity to sample footprints from a study site in which animals are reclusive, nocturnal, or otherwise difficult to observe directly.