Real-time classification of Serengeti wildebeest behaviour with edge machine learning and a long-range IoT network

Aug 5, 2026

Upload by Eden

Animal populations are increasingly shaped by environmental disturbance from human activity, land-use change, and global warming, which can modify migration routes, space use, activity budgets, and wildlife behaviour. Fine-scale understanding is essential to locate disturbance hotspots, reduce impacts, and anticipate population-level consequences. This study presents a low-cost tracking system combining open-source electronics, edge machine learning, and a long-range IoT network for real-time behaviour and location classification of Serengeti wildebeest.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages16

Reading Time40 min

AccessFree to download

Introduction

Environmental disturbance and wildlife population dynamics

Indirect effects on behaviour and activity patterns

Remote monitoring and telemetry approaches

Behaviour identification from movement and limitations

DOI: xxx/xxxx R ES EA RC H A RT IC L E Real‐time classification of Serengeti wildebeest behaviour with edge machine learning and a long‐range IoT network Cyrus M. Kavwele1,2,3 J. Grant C. Hopcraft2 Juan M. Morales2 Gerald Nyafi4 Nancy Kimuya5 Colin J. Torney1 1School of Mathematics & Statistics, University of Glasgow, Glasgow, United Kingdom 2School Of Biodiversity, One Health & Veterinary Medicine, University of Glasgow, Glasgow, United Kingdom 3 Department of Natural Resources, Karatina University, Nyeri, Kenya 4 Information Communication Technology, Tanzania National Parks Authority, Arusha, Tanzania 5 Frankfurt Zoological Society, PO Box 1, Arusha, Tanzania. Correspondence Colin J. Torney Email: [colin.torney@glasgow.ac.uk](colin.torney@glasgow.ac.uk) Abstract Globally, animal populations are facing increasing levels of environmental disturbance. Human activ‐ ity, land‐use change, and global warming are altering migration routes, space use, activity budgets, and the behaviour of many wildlife species. Understanding impacts on wildlife at a fine scale is essential to identify locations of increased disturbance, mitigate its effects, and predict potential population level outcomes. In this work, we introduce a low‐cost animal tracking system that in‐ tegrates open‐source electronics, edge machine learning, and an Internet of Things (IoT) network, to provide real‐time information on the location and behaviour of animals. The system employs an on‐board machine learning algorithm to identify distinct behaviours and then transmits classi‐ fication outputs along with location data over a long‐range network. We deployed the system on wildebeest ( Connochaetes taurinus) in Serengeti National Park, Tanzania, a highly social migratory ungulate population that is ecologically and economically vital to the region. Analysis of the trans‐ mitted data showed activity readings were consistent with location data and revealed biologically meaningful fluctuations in daily activity patterns. Our system introduces a new dimension to study‐ ing animal behaviour and movement ecology by offering immediate insights into the behaviour and location of collared animals. K EYW O R D S animal behaviour, conservation technology, machine learning, remote tracking 1 INTRODUCTION Understanding the effects of environmental disturbance on the popu‐ lation dynamics of wildlife species is one of the most significant chal‐ lenges in ecology. Environmental changes, such as resource degradation, increasing anthropogenic presence, or climate change, can significantly reduce the survival and reproduction rates of populations (Davis et al. 2018, Waters and Turner 2022, Turvey and Crees 2019) . Direct impactson mortality are relatively well studied and measurable (DeCesare et al. 2014, Thaxter et al. 2017, Rentsch and Packer 2014), however changes to the climate or landscape can induce altered patterns of space use and behaviour in animals (Sih 2013) . While non‐lethal, these effects can create large physiological and energetic demands that will ultimately lead to increased mortality rates and reduced reproduction (Cooke et al. 2014) . The challenge of monitoring and mitigating the indirect effects of disturbance on population dynamics is especially acute for migratory species. Globally, animal migrations are increasingly impeded by human dominated landscapes (Tucker et al. 2018), they are challenging to ob‐ serve, and often the landscapes they inhabit are difficult to manage, since they span regional and national borders. Wildlife species may respond to altered habitats and increasing hu‐ man activity in various ways. To minimize encounters with humans, wildlife species may infrequently use habitats that are adjacent to human‐dominated landscapes (Wittemyer et al. 2008, Veldhuis et al. 2019, Kavwele et al. 2022) . Landscapes perceived to be high risk may lead to behavioural responses that involve increased walking or v

FAQs

What problem does the study address in wildlife monitoring?

It addresses how environmental disturbance affects wildlife behaviour and population dynamics, and the need for fine-scale, actionable monitoring to detect disturbance impacts.

How does the proposed system classify wildebeest behaviour in real time?

The system uses an on-board machine learning algorithm to identify distinct behaviours and then transmits behaviour classifications together with location data over a long-range IoT network.

Where was the system deployed and what did the data show?

It was deployed on collared wildebeest in Serengeti National Park, Tanzania, and transmitted data showed activity readings consistent with location data and biologically meaningful daily activity fluctuations.

Aug 5, 2026

Upload by Eden