Press Release: Camera trap studies may systematically overestimate populations of elusive animals, new research suggests
Camera trap studies, a widely used method to estimate the abundance of elusive species such as snow leopards, tigers or jaguars, may be less accurate than previously believed. A new study testing the accuracy of the popular method found that one in eight photos of individual snow leopards were misidentified. As a result, population estimates based on these images were inflated by around a third. This research, published today in the journal Scientific Reports by Snow Leopard Trust scientists and their colleagues at Nordens Ark and the Swedish University of Agricultural Sciences, suggests that some endangered species could be less abundant than currently assumed.
“Our results show that we are not as good at identifying snow leopards in photographs as we previously believed, and that we tend to overestimate the number of individuals, mistaking the same cat as two different individuals in different photos. In our experiment, we only included high quality photographs of known animals, while in real world situations the photographs are often of lower quality, making mistakes even more likely.
This suggests that there may be fewer snow leopards than what current population estimates claim,” says Dr. Örjan Johansson, the study’s lead author and a scientist with the US-based conservation organization, the Snow Leopard Trust. Johansson, who is based at the Swedish University of Agricultural Sciences, added, “More experiments are needed. The same could well be true for other species that are estimated with camera traps. For example, such experiments would help better assess how reliable the current estimate of the global tiger populations is.”
Certain species such as snow leopards, tigers and jaguars are very difficult to observe in the wild, which is why researchers have often relied on camera trap studies to estimate their abundance. In these studies, remote-sensor cameras are deployed in a target species’ habitat; taking photos of animals as they pass by. Once these images are recovered, individual animals are identified by trained experts based on their markings, which are as unique as fingerprints. Since not all animals living in a given study area are likely to be photographed, researchers then use statistical models to calculate an overall population estimate based on the number of individuals photographed.
Underlying this method is the assumption that individual animals are accurately identified from photographs, since any systematic misidentifications would skew the resulting population estimates. However, until now, there was almost no empirical evidence to verify this assumption – which is surprising, considering that many of the species that are surveyed using camera traps are threatened or endangered, and accurate population and demographic estimates are critical for their conservation.
To test this assumption, Johansson and colleagues camera trapped 16 known captive snow leopards in zoo enclosures. They included multiple photo captures of each cat, then asked eight observers to identify the individuals in the photographs. Four of these observers were experts who had identified snow leopards for scientific purposes before, while four were non-experts who had not previously identified individuals in photographs.
On average, the observers misclassified 12.5% of the capture events. When such errors were made, it was much more common that observers identified two different photographs of the same individual as two different individuals, thereby creating a ‘new’, non-existent snow leopard, referred to as a ghost, rather than that combining photographs of two different individuals into one. In subsequent calculations, using a statistical tool called the capture-recapture method, the total population was overestimated by an average of 35%.
As a result of the study, the Snow Leopard Trust has created an online training tool that has been adopted and launched by the Global Snow Leopard & Ecosystem Protection Program – a high level inter-governmental alliance of all snow leopard range countries. Using this app (camtraining.globalsnowleopard.org) researchers or observers can practice identifying snow leopard individuals using the same data set as in our experiment. Here, observers can evaluate themselves and incorporate their error rates in future estimates to account for this uncertainty.
“Camera trap surveys are a key component of our population estimates for endangered species such as tigers, cheetahs or snow leopards, so it’s critically important that we can rely on their accuracy. While this is only one study, the fact that mistakes are more common than we believed, and that they appear to be systematic and result in overestimates of populations, should be of concern to all of us. Put simply, there may well be fewer of these iconic animals left than we assume,” says Dr. Charu Mishra, Executive Director of the Snow Leopard Trust.
Read the study at: https://www.nature.com/articles/s41598-020-63367-z.pdf
Acknowledgments
We are deeply grateful to the eight anonymous classifers for devoting their time to this study. We would also like to thank the Helsinki, Ätheri, Kolmården, Nordens Ark, Orsa Bear Park, Köln and Wuppertal Zoos for allowing us to deploy the camera in their exhibits and to the staf at the zoos for all their help. H. Andrén, H. Hemmingmoore, and B. Söderström provided comments improving the manuscript. Open access funding provided by Swedish University of Agricultural Sciences.