HerdHover logo

The Project

Blair and Felicitas flying

Revealing the complex dynamics of collective anti-predator behavior

Living in groups is one of many tactics animals use to avoid predation. Group membership can offer a range of anti-predator benefits, from risk dilution to improved predator detection to coordinated defensive behaviors. The Herd Hover project examines the complex dynamics of collective antipredator behavior in wild herds of ungulates in Kenya. We are particularly interested in how information is transferred within ungulate herds, how herds make movement decisions, and how the complex savannah environment affects individual and group-level behavior. Understanding these complex phenomena requires high-quality datasets that capture the behavior of all group members continuously and simultaneously at high temporal resolution. Until recently, such datasets were out of reach for field biologists, but recent advances in drone and machine learning technology are enabling us to push the boundaries of what is possible for field studies of wildlife behavior.

African buffalo herd at Mpala Conservancy

Methodology

We conduct our fieldwork in Laikipia, Kenya at Mpala and Ol Pejeta Conservancies. We use DJI Phantom drones to film ungulate groups from a bird’s-eye view. Aerial filming enables us to capture footage of all individuals in a herd continuously and simultaneously. We also use a Sensefly eBee+ drone to capture aerial images of the environment in which we film herds. We use photogrammetry to stitch these photos into detailed 3D reconstructions of the savannah landscape.

3D landscape reconstruction

We are using deep convolutional neural networks to automatically track animals in our drone footage and estimate the body postures of individuals. By combining these tracks and posture estimates with our 3D habitat models, we will learn how individual- and group-level movement and behavior are affected by social and environmental factors.

Detection and tracking of a Grevy's zebra herd using machine learning
Automated posture estimation of an individual Grevy's zebra

For inquiries about the HerdHover project, please email blair.costelloe@orn.mpg.de