The great wildebeest migration, seen from space: satellites and AI are helping count Africa's wildlife The Great Wildebeest Migration is one of the most remarkable natural spectacles on Earth. Each year, immense herds of wildebeest, joined by zebras and gazelles, travel 800-1,000km between Tanzania and Kenya in search of fresh grazing after the rains . This vast, circular journey is the engine of the Serengeti-Mara ecosystem. The migration feeds predators such as lions and crocodiles, fertilises the land and sustains the grasslands. Countless other species, and human livelihoods tied to rangelands and tourism, depend on it.

'Great Migration' involves far fewer wildebeest than we had thought An estimate that as many as 1.3 million wildebeest move across the Serengeti Mara landscape each year has been cut down to size using AI East Africa's "Great Migration" is generally estimated to involve as many as 1.3 million wildebeest. But in reality, fewer than 600,000 of the animals might move across the Serengeti Mara landscape each year, according to an AI analysis of satellite images. The Great Migration sees wildebeest, zebra and antelopes move between feeding and breeding grounds in Kenya and Tanzania, while also trying to dodge predators including lions, crocodiles and hyenas. Lions' record-breaking swim across channel captured by drone camera Assessing the number of animals involved is a tough task, traditionally achieved using crewed aerial surveys. Researchers can only survey a small area at a time, however, so they use statistical models to extrapolate densities across unsurveyed regions, which can introduce errors given herds are unevenly distributed and constantly on the move.

ShopRunner has more than ten million product images that we use to train computer vision classification models. Moving those files around and processing them is a pain without good tooling. Just downloading them serially takes many days, and the occasional corrupted image can bring the whole process to a halt. Without good logging and error handling, it might then be necessary to start the process over until the next error is raised. Over time we built up techniques for parallelizing over files, handling errors, and skipping files that had already been processed.