How AI Helps us Understand & Protect Snow Leopards

Monitoring snow leopards is a time-consuming business - but a new AI solution developed in partnership between Microsoft and the Snow Leopard Trust could be a game changer and free up resources to invest in conservation action.

Since snow leopards are notoriously difficult to spot and observe, we have to resort to technology to find out where these cats are and how their populations are doing. Camera trap surveys are one of the few scientifically sound methods to do so.

The cameras used in these surveys are equipped with heat and motion sensors and are triggered by passing animals – giving the elusive “Ghost of the Mountain” an opportunity to take a selfie for science.

Photo: NCF India / Himachal Pradesh Forest Department / Snow Leopard Trust

Each camera trap survey lasts several months, covering areas of up 500 square miles and generating 200,000 to 300,000 images from 30 to 60 cameras. Once these photos are retrieved from the field, scientists sift through them to identify snow leopards.

“These cameras can’t distinguish between a snow leopard and other animals. The majority of photos they take are of goats, camels or horses that plant themselves in front of the lens and ruminate for hours. Sometimes they are triggered by a blade of grass swaying near a sun-heated rock. And once in a blue moon, a snow leopard sneaks by! We spend days just swiping through image folders looking for cats”, says Koustubh Sharma, a Senior Ecologist with the Snow Leopard Trust.

But thanks to a new Microsoft AI solution, this process is about to become a whole lot faster. A machine learning model developed by Microsoft engineers can identify snow leopards and automatically classify hundreds of thousands of photos in a matter of minutes.

Microsoft’s AI solution automatically scans thousands of camera trap photos and identifies those with a snow leopard in them. Photo by Microsoft / SLT

Microsoft engineers built the model with deep neural networks, an AI technology that learns patterns similar to how a human brain learns.

“Once we had built the model, we trained it with thousands of photos. Now, it has reached a point where it can tell reliably if there is a snow leopard in a given picture and help researchers save valuable time”, says Microsoft software engineer Mark Hamilton.

It’s a huge step forward – and yet only a part of the larger objective. The next step for the technology is automating the identification of individual snow leopards, based on their unique markings. “The question is, ‘Is the snow leopard in Picture 1,240 the same snow leopard in Picture 1,000,240?’” Mark Hamilton explains. “Right now, the process is labor-intensive and error-prone, and as you get more images, it’s like adding more pieces to a 40,000-piece jigsaw puzzle. We hope deep learning can help us find likely matches.”

The snow leopard’s individual fur patterns help scientists identify each cat on camera trap images – but the process is very time-consuming. In the future, AI might also help with this critical step. Photo by SLT

“We have manually identified individual snow leopards in photos throughout the years, and have been able to come up with reliable population estimates and trends for a few habitat areas”, says Koustubh Sharma. “But have a backlog of roughly 10,000 photos awaiting cat IDs. We hope that in good time, machine learning will alleviate the bottleneck, freeing up lots of time for us to invest in our conservation programs and leading to more precise data and better population estimates.”

The technological developments could not come at a better time. The snow leopard conservation community is currently undertaking a massive and ambitious project to jointly come up with a reliable global population estimate. The initiative has been labeled PAWS (Population Assessment of the World’s Snow Leopards) and is being coordinated by the Global Snow Leopard & Ecosystem Protection Program.

The global snow leopard population remains unknown – but the PAWS initiative could mark a bg step towards a solid estimate. Photo by Shan Shui / Panthera / Snow Leopard Trust

“The big idea behind PAWS is to align methods and share best practices in snow leopard population studies”, Koustubh Sharma says. “If we all follow the same standards and coordinate our efforts, we may finally be able to get a solid estimate of how many snow leopards are left in the world.”

All snow leopard range countries have endorsed PAWS standards and protocols. Dozens of conservation NGOs and research institutions are participating in the initiative. “Soon, all those researchers will be able to use the AI solution developed by Microsoft to speed up analysis. And once the software learns how to identify individual cats, we’ll collectively take another leap forward”, Koustubh Sharma says.


To highlight the partnership, Microsoft is currently running an awareness campaign in various online channels, with short videos and ads featuring Snow Leopard Trust researchers in action. The Snow Leopard Trust has received an AI for Earth grant from Microsoft to help cover computing and data costs, and Microsoft has donated the time and talents of their software engineers to develop the solution.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.