7 marca, 2023

A new specialist has joined the scientists of Mammal Biology Institutute in Białowieża. This expert can look through 80 thousand forest photos in only 24 hours and recognise a species in less than a second. A fast learner and an overachiever – meet the new AI supercomputer purchased with the help of Żubr Fund. 

Without new technologies it would be hard to efficiently protect the environment – even in the oldest of European forests: Puszcza Białowieska. That’s why, in 2022, Żubr Fund donated to Białowieski National Park 1 mln zlotys for the purchase of highly specialized equipment.

This year’s partner is the Mammal Research Institute of the Polish Academy of Science which received from Żubr Fund a sum of 100 thousand zlotys for the purchase of a supercomputer. This high-end hardware will use AI solutions to accelerate research into endangered animals.

Cutting edge technology

Why do certain species die out? There can be many reasons: shrinking of habitat, disease, traffic, pollution of the environment… What is needed to answer this question is knowledge based on data.

Once, to observe wild animals, scientists had to wander through forests armed only with binoculars and a notebook. Today they have at their disposal automatic sensors, installed in the natural habitat. Sound detectors record the sounds of animals as well as the background noise (for example car engines or guns), wildlife cameras take thousands of photographs. Thanks to that we can observe deer gathering by a waterhole, listen in high frequency to bats feeding or see the first brown bear in years to appear in Puszcza Białowieska. Those tools are the ears and eyes of scientists in places where access is difficult for humans. 

Terabytes of knowledge

Sensors provide researchers with a gigantic amount of data: for example, 1 season could generate up to 25 thousand hours of recordings. These materials must then be looked at, listened to, interpreted and selected. In the case of photographs as much as half of them are “empty” meaning: without any animals in them. This process requires lots of time and work for many people.  

And that’s where the supercomputer enters the picture. This high-end hardware will use AI solutions to process data and draw conclusions from it. AI algorithms used in Mammal Biology Institute not only spot animals, but also recognise species. For example, they can almost perfectly assess if the specimen in the photograph is a fox. And also they keep on learning since they are based on neural networks.

Faster, more efficient, more precise

Możliwości, jakie daje AI, są ogromne. Sieci neuronowe przyśpieszają proces żmudnej i trudnej analizy danych. Pomagają określić liczebność populacji (np. trudnych do obserwacji AI makes so much more possible. Neural networks accelerate tedious and time-consuming analysis of data. They help assess species’ population (for example, of insects which are difficult to observe) and their habitat. They mark areas where man vs animal conflict occurs. They help create patterns of animal activity for chosen periods (24 hours or seasonal).

With such knowledge researchers can create specific recommendations concerning endangered species. For example, they can recommend conservation of an area with old trees in which a rare species of woodpecker resides or they can decide to move a feeding facility due to parasite infestation.

Thanks to Żubr Fund’s financial help the supercomputer will allow the Mammal Research Institute of the Polish Academy of Science to operate an even larger number of sensors in an even larger area. It will allow more research questions to be asked and more knowledge to be gained. Knowledge which will result in even more efficient nature conservation in Puszcza Białowieska.  

The supercomputer in numbers

• 32 core CPU processor
• 128 GB of RAM memory
• Processing of photographs, footage and sound recordings up to 100 times faster than by humans
• 1 second to spot animal presence in a photograph and to assess its species.
• Up to 94% precision in recognising species
• About 86 thousand photographs processed in a 24 hour period.