Real-time conservation: AI shrinks wildlife tracking from months to days

Washington State University and Google have demonstrated that the AI model “SpeciesNet” can automate wildlife tracking with 90% accuracy compared to humans

The breakthrough study, led by Washington State University and Google and published in the Journal of Applied Ecology, has shown that AI can now address the massive “data bottleneck” in wildlife conservation. 

By using a fully automated system to process millions of camera-trap images, researchers reduced analysis time from nearly a year to just a few days, with results that match human accuracy.

Breaking the bottleneck

Wildlife researchers rely on camera traps—motion-activated cameras hidden in habitats—to monitor species like jaguars, wolves, and bears. Traditionally, this created a mountain of data that required human eyes to sort through.

• The old way:

  • A team of students and experts would manually review hundreds of thousands of photos. This painstaking process typically took 6 to 12 months before actual scientific analysis could even begin.

• The AI way:

  • Using a model called SpeciesNet, the team processed images from Washington, Montana, and Guatemala with zero human intervention. The processing time was slashed to roughly one week.

How accurate is the machine?

The goal wasn’t for the AI to be perfect in every single frame, but to see if it reached the same scientific conclusions as humans regarding where animals live and how environmental factors affect them.

• Success rate:

  • In 85–90% of cases, the AI’s ecological models aligned perfectly with those produced by human experts.

• Robustness:

  • Because ecological “occupancy models” rely on repeated sightings over time, the overall scientific conclusion remained solid even if the AI occasionally misidentified a single image.

• Limitations:

  • Human review is still vital for rare or easily confused species, where the AI still struggles to tell the difference between similar-looking animals.

A win for underfunded conservation

The shift to full automation is “transformative” for smaller conservation groups that lack the budget to hire large teams of data-labellers. Faster data processing means faster decision-making; if a predator moves into a new area or a rare species disappears, managers can now respond in near real-time rather than waiting a year for the data to be processed.

As lead author Daniel Thornton puts it: “We’re not trying to replace people. The goal is to help researchers get to answers faster so they can make better decisions about managing and conserving wildlife.”