The Great Wildebeest Migration is some of the remarkable natural spectacles on earth. Every yr, huge herds of wildebeest, accompanied by zebras and gazelles, travel 800-1,000 km between Tanzania and Kenya searching for fresh pasture after the rain.
This prolonged tour is the engine of the Serengeti-Mara ecosystem. The migration feeds predators comparable to lions and crocodiles, fertilizes the land and maintains the grasslands. Countless other species and human livelihoods tied to rangelands and tourism, depend on it.
Because this migration shapes the complete ecosystem, it can be crucial to understand how many animals are involved. Changing numbers wouldn’t only impact wildebeest, but would also impact predators, vegetation and the tens of millions of people that depend on this landscape.
For a long time, aerial photography has been the first tool for estimating the dimensions of the wildebeest population in East Africa. Airplanes fly in straight lines (transects) a couple of kilometers apart and use these strips to estimate the overall population. This dedicated and painstaking work, using a time-tested method, has given us an estimate of about 1.3 million wildebeest.
In recent years, conservation scientists have begun testing whether satellites and artificial intelligence (recognizing patterns in large data sets) can provide a brand new strategy to monitor wildlife. Previous work showed that other species – Weddell seals, Beluga whales and elephants – may very well be identified on satellite images using artificial intelligence.
In 2023 we are going to showed that migrating wildebeest may very well be detected from satellite images using deep learning. This study proved that it is feasible to observe large aggregations of mammals from space. The next step was to maneuver from just tracking animals to estimating their populations – using satellites not only to detect them, but to count them on a big scale.
Our current one study was carried out through the collaboration of biologists, distant sensing specialists and machine learning scientists. We analyzed satellite images of the Serengeti-Mara ecosystem from 2022 and 2023, covering greater than 4,000 km².
Using deep learning models
Images were captured at very high spatial resolution (33–60 cm per pixel), with each wildebeest represented by fewer than nine pixels. We analyzed the pictures using two complementary deep learning models: a pixel-based U-Net and an object-based one YOLO model. Both were trained to identify wildebeest from above. Using them together allowed us to cross-validate detections and reduce potential bias. The images were taken originally and end of August and correspond to the several stages of the dry season migration. As expected, smaller flocks were observed earlier this month.
In each years, the models detected fewer than 600,000 wildebeest within the dry season area. Although these numbers are lower than some previous aerial estimates, this shouldn’t necessarily be interpreted as evidence of population decline, and we recommend more survey effort to find out the relative error biases in each approach. While some animals will inevitably be missed under trees or outside the imaged area, it’s unlikely that such aspects could account for a whole bunch of hundreds more animals. To confirm that the fundamental herds were covered, we validated the survey scope using collared wildebeest GPS tracking data and ground-based observations from organizations that monitor herd movements within the region.
These results provide the primary satellite-based dry season census of the Serengeti-Mara migration. Rather than replacing aerial photography, they supply a complementary perspective on seasonal population dynamics. The next step is to coordinate aerial and satellite surveys in parallel. In this manner, each method can assist refine the opposite and create a more complete picture of this extraordinary migration.
Future directions
Satellite monitoring will not be a panacea. Images are expensive and are sometimes obscured by cloud cover. And they’ll never capture every individual on the bottom (the identical goes for aerial photography). But the benefits are convincing. Satellites can capture a snapshot of vast landscapes at a single time limit, eliminating much of the uncertainty created by extrapolating local counts.
The approach is scalable to many other species and ecosystems. And as more high-resolution satellites (with imaging distances of lower than 50cm) are being launched, we are able to now revisit the identical spot on Earth multiple times a day, making wildlife monitoring closer to real time than ever before.
Beyond population counts, satellites also open up a brand new scientific frontier: the study of large-scale collective movements. The wildebeest migration is a classic case of emergence behavior: there is no such thing as a leader, but order still prevails. Each animal follows easy cues, comparable to where the grass is greener or where a neighbor is moving, and together hundreds form an enormous, coordinated journey.
With high-resolution satellite data, scientists can now explore the basic physics that govern the way in which animals move together in large groups. But how do dense waves of movement propagate across the landscape, what scaling rules might govern the spacing and orientation patterns, and the way do these collective patterns influence the functioning of ecosystems?
Our results show how satellites and AI may be used not just for monitoring wildlife populations, but in addition for applications that transcend population counting and uncover the mechanisms of collective organization in animal groups.
