Using footprints to identify and sex giant pandas

1. What they did

China’s giant pandas live at low density in steep, rugged country, so they are hard to count. The two usual ways of telling individuals apart both have problems: measuring the size of bamboo bites in droppings lacks accuracy, and DNA from droppings is costly and needs very fresh samples. Pandas look alike on camera, with no stripes or spots, and males and females are hard to tell apart even up close. The authors set out to adapt the footprint identification technique (FIT), already used for rhinos, cheetahs, tigers and mountain lions, to identify individual pandas and their sex from footprints.

• Collecting footprints (March 2014 to April 2016). They photographed footprints of 41 captive pandas at the China Conservation and Research Centre for the Giant Panda in Sichuan. Most prints were on fresh, prepared sand; some were on snow in the Wolong enclosures. They used only the left front foot, which leaves the clearest prints and has a sesamoid pad (a “sixth finger” used to grip bamboo) that adds detail. Each print was photographed from directly above with a carpenter’s scale in the trail. In total they had 521 usable footprints along 76 trails (a trail is an unbroken series of footprints from one animal).
• Measuring the prints. In the FIT software, a person marks seven points on each print (the centres of the five toe pads and the sesamoid pad, and the end of the carpal pad). The software then calculates 124 measurements of lengths, angles and areas.
• Identifying individuals. The model compares trails two at a time. If the two trails’ clusters of prints overlap, they likely belong to the same panda. The results are combined into a tree diagram that predicts how many individuals there are. They built the model with 30 pandas that had at least six prints per trail, then tested it on two snow trails and on nine pandas with fewer than six prints.
• Telling males from females. They used discriminant analysis on single footprints and on trails, and looked at whether sex differences change with age.

2. What they found

Individual identification

• The best model used 12 measurements. It worked well when the training set had more than 12 individuals; with smaller training sets, predictions became more variable.
• For the 30 pandas (477 footprints, 67 trails), the model predicted 28 individuals (93.3% accuracy). Seven of the 67 trails were misclassified (89.6%). Allowing for uncertainty, it predicted 28 to 30 individuals.
• Adding two snow trails from two known pandas (32 individuals), the model predicted 29 (90.6%), and correctly identified the two snow trails as separate animals.
• For nine pandas with fewer than six prints per trail, the model predicted eight (88.9%). The two trails wrongly joined had only two and three footprints.

Sex

• Using single footprints, sex was predicted correctly about 84% of the time with about 30 measurements, and above 75% with as few as five.
• Using whole trails (majority vote of their footprints), 91% of trails were assigned the right sex.
• Females and males under three years old had very similar feet. Older age classes showed a clear difference between the sexes, though the authors say the age-class samples were too small for firm conclusions.

3. What they concluded

• FIT can identify individual giant pandas and their sex from footprint photographs, and to some extent their age class.
• They suggest a training set of at least 20 individuals with an even sex ratio for identifying individuals.
• Where only about ten pandas are present, or only part of a population is sampled, even three to four footprints per trail may give accurate identification.
• Field use: footprints must show the five toe pads, the sesamoid pad and the carpal pad. Prints distorted on steep slopes cannot be used, so surveys should focus on ridges and valleys, and on snow. Staff should follow the direction of a footprint to find the whole trail, and collect 3 to 6 left front footprints, photographed from directly above with a scale and an ID slip recording GPS location, date and footprint ID.
• Finding fresh, clear footprints takes commitment and experience; the help of local trackers is often invaluable, and field staff need training to recognise panda footprints and the correct foot.
• They encourage combining FIT with camera trapping and other methods.

4. How tracking was or was not used

The study is built on footprints. The researchers photographed and measured left front footprints of known captive pandas, mainly on prepared sand and some on snow, and used them to build and test the identification model. No trailing of wild pandas was done. For field use, the authors recommend following the direction of a footprint to find the whole trail, and say that finding footprints comes with experience and that local trackers are often invaluable.

5. Tracker or observer certification

Not reported. The paper does not say who found, photographed or marked up the footprints, or how much experience they had. This matters because the landmark points on each print are placed by hand.

6. Other technologies, skills or innovations

• The footprint identification technique (FIT) add-in for JMP software (SAS). The authors say it can be made available free to conservation organisations after a training workshop and by application to JMP.
• Basic digital cameras or smartphones can be used to photograph footprints.
• A Reference Centroid Value (RCV), built from the other known individuals, keeps the pairwise comparisons stable when animals are added or removed.

7. Implications for conservation (from the paper)

• FIT is non-invasive, unlike methods that need capture, handling and tagging, which can affect animals.
• Where routine surveys already take place, collecting footprints should add little or no cost, making FIT a cost-effective way to identify and sex individuals and follow population changes.
• Since 2010, seven captive-born pandas have been reintroduced and five survived. Their GPS collars drop off after two years; FIT could monitor them in the long term, including range changes, activity and interactions with other pandas.
• The technique could empower local conservation practitioners, especially those with tracking skills or other traditional ecological knowledge, to monitor pandas themselves, and visitors could collect footprint images as citizen scientists.