Marine predators are crucial components of marine food webs. To conserve these species it is necessary to understand how they respond to environmental challenges. Researchers from Macquarie University are using machine learning to automatically detect behaviors from accelerometers. This data can be used to monitor behavioral changes over time and identify major drivers behind population changes.
Dr. Monique A. Ladds and Prof. Robert Harcourt, Macquarie University
The conservation challenge
Marine predators play a large role in the functioning of the marine ecosystems. Increasing sea-surface temperatures and increased fishing pressures have led to ecosystem changes that are altering prey density and distribution. Understanding the working interaction between the ecosystems and human populations is difficult, and finding ways of protecting these animals is complicated. It is essential to the survivorship of these species that researchers can predict their responses to these challenges. Activity budgets developed for marine predators are a useful tool for monitoring their responses to these challenges. Developing activity budgets allow researchers to monitor marine predators foraging effort, in turn informing about prey availability.
To understand where conservation strategies may be implemented conservation managers first need to understand how marine predators respond to challenges. Foraging for marine predators requires traveling great distances and diving to depth in order to find food. The effort used to forage successfully over time will change in response to changing environmental factors. Accelerometers and machine learning can help to monitor these responses. Accelerometers attached to seals record behaviour, and artificial intelligence can automatically classify those behaviours. The automatic classification of behaviours allows researchers to monitor marine predators over time, discovering where and how they are spending their time.
To ensure that researchers gain the greatest recognition from the accelerometers, they tested a suite of machine learning algorithms (logistic regression, support-vector machines, random forests and gradient boosting machines). For each algorithm, they used a training and testing set which were built from captive experiments, where each behaviour was manually coded from video of the animal wearing the accelerometer. Using the training set each algorithm was taught the pattern of each behavior. Then the algorithm was used to predict behaviours from the test set. This resulted in accuracies of over 80%.
The major limitation faced by the researchers was how to define foraging. In the wild, marine predators would actively pursue live prey but researchers were unable to recreate these scenarios in captivity. Therefore it was difficult to predict this category with great accuracy.
The results from this study have been shared through peer-reviewed journals and are already being taken up to identify the behaviours of other species, such as sharks. The use of accelerometers to identify behaviours can be used to make activity budgets to monitor animals over time and identify how they adapt to changes in their environment.
The Marine Predator Research Group at Macquarie University works with collaborators around Australia and across the world and always happy to talk with new enthusiasts.
Find out more
- Macquarie University Marine Predator Research Group website
- Peer reviewed article: Seeing It All: Evaluating Supervised Machine Learning Methods for the Classification of Diverse Otariid Behaviours
- Peer reviewed article: Super machine learning: improving accuracy and reducing variance of behaviour classification from accelerometry