The migration of serengeti wildebeest implies fewer animals than we had thought
Nicholas Tinelli / Alama
It is generally estimated that the “great migration” of East Africa involves up to 1.3 million wildebeest. But in reality, less than 600,000 animals could move through the Serengeti Mara landscape each year, according to an AI analysis of satellite images.
The great migration sees the GNOU, the zebra and the antelopes move between food and the motive of reproduction in Kenya and Tanzania, while trying to dodge predators, in particular lions, crocodiles and hyenas.
The evaluation of the number of animals involved is a difficult task, traditionally carried out using crew air surveys. Researchers cannot, however, study a small area at a time, so they use statistical models to extrapolate densities in non -reduced regions, which can introduce errors since the herds are unevenly distributed and constantly in motion.
Satellite surveys can avoid these problems because a photo can cover hundreds of thousands of square kilometers, reducing the counting chances of double counting animals and removing the need for extrapolast projections. It is not practical to manually count the GNU in such large images, but animals can be counted instead using AI. “Automation via AI provides greater consistency and precision in the count,” explains Isla Duporge at the University of Oxford.
For the new study, Duporge and his colleagues have formed two in -depth learning models – U -Net and Yolov8 – to identify the GNOU using a set of images data in which 70,417 animals had been manually labeled. The two models were then applied more than 4,000 square kilometers of high resolution satellite imaging. The images were captured on August 6, 2022 and August 28, 2023.
The two models of AI gave similar results, counting 324,202 and 337,926 Gnoues in 2022, and 502,917 and 533,137 in 2023. The substantial difference between the 1920s and 2023 reflects the number of animals involved by the two times through the month. “”[But] What is encouraging is that the two in -depth learning models, which use very different approaches, have produced consistent results, “explains Duporge.” This consistency strengthens our confidence in the reliability of AI -based accounts to follow the populations of wildlife. “”
The previous estimate of 1.3 million produced from air surveys is in place, largely unchanged, since the 1970s. “Based on our results, I suspect that the real size of the population is closer to around 800,000 people, if we could count each individual without error,” explains Duporge. “I think the air count is overestimation and ours is a slight underestimation. We are missing under the trees and there will be outside the study area, but it is very surprising that we cannot be more than 533,137. ”
The lower counts do not necessarily mean that the wildebeest populations have collapsed. They may have changed their migration paths. But the wildebeest faces significant challenges, including the loss of housing and fragmentation due to agriculture. Precise estimates of the population are crucial to clarify targeted conservation efforts.
The research team previously formed an AI model to recognize the elephants of satellite data, but this is the first time that the approach has been used to carry out an individual mammal census in a large distributed population. The team is currently developing a similar method for detecting and has African rhinos.
“I believe that we should go to satellite methods and based on AI to count the populations of fauna, in particular for species which are largely distributed in large landscapes,” explains Duporge.
The researcher’s model code was made available on https://github.com/sat-wildlife/wildebeest
Subjects:
