The role, behaviour and biology of animals
Forecasting bird food hunt using computer models
An ultra-light smart sensor fitted to a bird can tell whether the bird is travelling, hunting or eating. Researcher Bruce Yu developed this technique to get a better understanding of the behaviour of birds. This will help scientists figure out how birds are adapting to climate change.
“I have always been interested in animals,” says Bruce Yu. That was why he switched from geophysics to ecology as a young scientist. Now Yu is a researcher in the Experimental Zoology chair group, where he uses machine learning – a form of artificial intelligence – to figure out the behaviour of birds. Flying is a very energy-intensive form of locomotion and migratory birds cover large distances. That makes it interesting to study the flight movements of migratory birds. How do they make progress, how do they behave and how do they manage their energy? If they have to go to a lot effort to find food because it has become scarcer due to climate change, perhaps they will be too exhausted to feed their young. This is important information for nature conservationists.
How does Yu go about his work? “First, we observe the natural behaviour of a species of bird. Then we train the computer to analyse and identify the flight movements.” To do this, Yu enters all the data about wingbeats, speed and movements into a computer model. Dedicated algorithms teach the model to recognise the images automatically as certain categories of behaviour, for example hunting or eating. The essence of Yu’s research is to make sure this self-learning programme is so small that it can be stored in a data logger that is fitted to a flying bird. “We design the computer model to process the data directly ‘on board’ (on the bird, ed.) so as little raw data as possible has to be stored. That makes it a smart onboard tracker that lets us monitor a bird over a longer period and in more detail.”
‘Data loggers let us track the migratory birds for longer periods and in more’
Keeping the storage and processing of data as close as possible to the source is an example of ‘edge computing’. That is a huge improvement on the familiar transmitters that researchers have been using for years to track birds. Yu: “The advantage is you don’t have to send or store as much data. That saves battery energy and lets the data logger operate for longer.”
In addition, he says, edge computing lets you respond more quickly to the live data. That is already happening in agriculture, where this has a clear purpose. If a cow with a sensor is behaving abnormally and may be sickening, the data logger sends a message to the farmer’s mobile phone, letting them take action immediately.
Edge computing can also help nature conservationists tackle poachers as it lets large amounts of data be collected and passed on with little delay. Yu explains: “The data can be signals such as GPS locations, images of camera traps where people can be seen, or the behaviour of animals in the wild that you can interpret using movement sensors.
When that is combined with modern telecommunications, you can warn rangers much more quickly about active poachers than when using traditional methods.”
Yu uses the smart data loggers to understand the behaviour of the pied flycatcher, a migratory bird that broods in the Netherlands from the end of April to June. It sits on the eggs and makes short trips to get insects, catching them in full flight. How do the birds manage this feat, how often do they do it and how do they maintain their energy levels?
The computer model can distinguish between various kinds of behaviour, such as hunting, swallowing insects, flying from one brooding spot to another or travelling to another area. Yu analyses how many rapid catching manoeuvres the bird makes and how many are successful. “We want to find out how good the flycatchers are at catching the prey they feed to their chicks. That affects their brooding success. We see similar scores for males and females.”
Data from the data loggers shows how animals are adapting to climate change
The results give more information about the availability of food, the timing of brooding and the birds’ fitness. “They let you relate flight patterns to brooding behaviour,” explains Yu. Based on the information, the researchers can draw conclusions about the extent to which the pied flycatcher can adapt to altered circumstances, for example due to climate change.
The machine learning system in data loggers is more efficient than the existing systems. That is welcome, but Yu acknowledges that not every researcher will have the option of programming data loggers. He therefore believes it is important to build more computer models, not just for birds but for other kinds of animals too, which is why Yu is making his expertise available to other researchers working with data loggers.
What does the future hold? The on-board sensor fitted to the birds now has accelerometers like the ones in smartphones. Yu plans to add microphones soon. There have been very few studies listening in on migrating birds, says Yu. “If you listen in, you can study the communication between birds during migration. But you need more processing power to get useful information from the audio signals. What is more, storing sound takes up a lot of memory. Here too, we can use edge computing to limit the amount of raw data required. Rather than storing all the data, we let the logger process the data in real time on the bird.” He is already considering the next step: “Who knows, maybe we could add video.”
Share this article
WHO Bruce Yu, Researcher
TEAM Experimental Zoology
This project is part of the Next Level Animal Sciences (NLAS) innovation programme.
Participating researchers of Wageningen University & Research collaborate with various partners to develop new research methods and technologies within the field of animal sciences. NLAS consists of three research directions, namely sensor technology, complex cell systems and data and models.