Does the technology behind predictive texts, facial recognition, and our social media feeds offer the solution to struggling global wildlife conservation efforts?
By Katie Jones
Artificial intelligence (AI) is one of the fastest growing approaches to tackling the global biodiversity crisis. Machine learning, whereby computer systems learn to improve their accuracy and efficiency without human intervention, is leading these developments. Now, this technology is transforming conservation efforts in unexpected and fascinating ways.
An image speaks a thousand words (when AI is used)
Conservationists find themselves in a race against time to save declining endangered populations. Yet, in the absence of technological help, acquiring the data needed to form conclusions about the state of wild populations can take years. Researchers at Princeton University, Rensselaer Polytechnic Institute, the University of Illinois-Chicago, and the NGO Wild Me who are behind ‘Wildbook’, a so called ‘Facebook for animals’, may hold one solution to this. The Wildbook system processes photographs and can identify and count individual animals visible within them. This uses computer vision, a visual machine learning technique, for detection of any unique animal markings such as stripes, dots, or wrinkles that give away the species or even individual identities. Through processing the most abundant source of information on wildlife, Wildbook is providing conservationists with real-time tracking of populations and estimations of population sizes. Combining this software with automated camera traps could provide conservationists with a previously unimaginable supply of data, supporting the fight against species extinction. So far, Wildbooks have been developed for several species, including lynxes, seals, and sharks. In 2016, this system facilitated the first Grevy’s zebra species census using citizen taken photographs, with this data now being used as official International Union for Conservation of Nature (IUCN) census data.
What about when abundant images of individuals are not available, such as in protected areas where human activity is restricted? An alternative ‘from the ground up’ approach to AI-powered photo processing could offer a solution. Footprint Identification Technology (FIT), developed by conservationists working at WildTrack and UC Berkeley, uses visualisation software to identify species, individuals, and even the sex and age of individuals, just from photographs of their tracks. Conservationists working within protected areas need only to take photographs of animal tracks; FIT can then extract measurements of the angles, distances, and areas of footprints and run them through species-specific algorithms to achieve identification. Recording individuals and their movements in this way eliminates the need for contact between conservationists and animals. Hence, this tech has the potential to make conventional animal tracking methods like expensive GPS collars obsolete in many contexts. Disruption of animal behaviour is avoided and the risk of human-animal disease transmission is eliminated. This method also remains firmly rooted in animal tracking expertise, creating a space for the traditional ecological knowledge of indigenous peoples within the conservation of their landscapes.
Initial tests of FIT looking at cheetah, Africa’s most endangered large felid, successfully predicted the identity of individuals more than 90% of the time and facilitated the identification of 38 distinct individuals from 781 footprints. FIT’s algorithms are fully adaptable for use with different species and have already been developed for 15 endangered species so far, including rhino, pumas, cheetahs, and Bengal and Amur tigers.
Technological solutions to conservation hurdles are not limited to well-loved terrestrial animals like zebras and cheetah. Automated AI technology is even being used to understand the peril faced by underwater ecosystems. Over two-thirds of the ocean has never been mapped or even seen by humans, making conservation within the underwater realm especially challenging. Remotely operated vehicles equipped with underwater cameras have vastly increased the acquisition of marine ecology data over the last 10 years. However, limited processing of this huge amount of data has inhibited how deeply analyses have been able to delve. AI technologies are now taking on these data processing demands. The Koster Seafloor Observatory is providing researchers with a platform to learn more about the decline of deep-water coral species. Researchers can upload underwater footage and apply machine learning algorithms to automatically detect and classify captured cold water coral species. Intriguingly, the classification ability behind the algorithms is possible thanks to initial manual classification by volunteer citizen scientists. These preliminary classifications provide the training material the AI software learns from, allowing this innovative approach to overcome an otherwise extremely limiting technical bottleneck.
AI to prevent wildlife poaching
Global illegal wildlife trade is valued at an estimated £15 billion annually, with populations of critically endangered species being devastated by the vehement poaching which supports this market. Conventional guarding of protected areas by dedicated wildlife rangers is failing to keep pace. PAWS, ‘Protection Assistant for Wildlife Security’, is a novel AI system that could hopefully turn the tide on this extinction driver. The system processes geographic data, including topology and road locations, and records of historical poaching activities. It subsequently determines the poaching threat across an area, utilising predictive modelling to provide wildlife rangers with optimised patrol routes. In a similar way to how smartphone apps predict the fastest journey times on congested roads, this AI can predict poacher behaviour and suggest patrol routes that maximise the number of snares and traps which can be intercepted by rangers.
During a month-long trial of PAWS in Srepok Wildlife Sanctuary, Cambodia, a site noted for its importance for Southeast Asian tiger populations, twice as many snares were removed compared to normal. In this single month, 1,000 illegal snares, 42 chainsaws, 24 motorbikes, and 1 truck were removed. PAWS has since been integrated into an established database of poaching activity called SMART (Spatial Monitoring and Reporting Tool). SMART is already in use across 800 protected areas in 60 countries, giving opportunities for the widespread expansion of PAWS. Moreover, the potential of this tech does not stop with poaching as the same system is adaptable for tackling global illegal fishing and logging activities.
It is undeniable that AI is changing the face of wildlife conservation, and at a time when advancements in this field are more needed than ever. Through increasing the efficiency and speed of data collection, as well as increasing cost-effectiveness, technology-led solutions might be just what conservationists need to overcome 21st century drivers of extinction.
Anton, V., Germishuys, J., Bergström, P., Lindegarth, M., & Obst, M. (2021). An open-source, citizen science and machine learning approach to analyse subsea movies. Biodiversity data journal, 9. https://bdj.pensoft.net/article/60548/
Berger-Wolf, T. Y., Rubenstein, D. I., Stewart, C. V., Holmberg, J. A., Parham, J., Menon, S., … & Joppa, L. (2017). Wildbook: Crowdsourcing, computer vision, and data science for conservation. arXiv preprint arXiv:1710.08880. https://arxiv.org/abs/1710.08880
Eddy, T. D., Lam, V. W., Reygondeau, G., Cisneros-Montemayor, A. M., Greer, K., Palomares, M. L. D., … & Cheung, W. W. (2021). Global decline in capacity of coral reefs to provide ecosystem services. One Earth, 4(9), 1278-1285. https://www.cell.com/one-earth/fulltext/S2590-3322(21)00474-7
Jewell, Z. C., Alibhai, S. K., Weise, F., Munro, S., Van Vuuren, M., & Van Vuuren, R. (2016). Spotting Cheetahs: Identifying Individuals by Their Footprints. Journal of visualized experiments: JoVE, (111) https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4942022/
Kwok, R. (2019). AI empowers conservation biology. Nature, 567(7746), 133-135. https://pubmed.ncbi.nlm.nih.gov/30833739/
O’Brien, C. (18.04.17). Wildbook: a social network for wildlife. Mongabay.com. [Accessed 26.02.22] https://news.mongabay.com/2017/04/wildbook-a-social-network-for-wildlife/
TeamCore, Harvard John A. Paulson School of Engineering and Applied Sciences. PAWS: Protection Assistant for Wildlife Security [Accessed 26.02.22] https://teamcore.seas.harvard.edu/paws-protection-assistant-wildlife-security
WWF (2020) Living Planet Report 2020 – Bending the curve of biodiversity loss. Almond, R.E.A., Grooten M. and Petersen, T. (Eds). WWF, Gland, Switzerland. https://www.wwf.org.uk/living-planet-report