movement ecology

How to deal with the growing wolf population in Switzerland?

by

The newspaper NZZ am Sonntag recently published an interview with our in-house predator specialist and movement ecology group leader Gabriele Cozzi about the current wolf situation in Switzerland and the associated research opportunities. The interview was conducted by Atlant Bieri and originally published in german. Below, we provide an english translation.

 

NZZ am Sonntag: Mr Cozzi, this summer a shepherdess and her dog were growled at by a wolf in Graubünden. There have also been more encounters between humans and wolves elsewhere recently. How dangerous is that?

Gabriele Cozzi: In such an encounter, a wolf possibly growls because it is frightened and thus signals its presence. In its language, this means: “Don’t come near me!” If you follow this signal, nothing can happen. In this situation, a wolf would only attack if it feels attacked and sees no escape options. But that hardly ever happens, you almost have to step on its tail.

 

As a defence, the woman called loudly and made herself appear as big as possible. The wolf then left. Does that always work?

She reacted correctly and used the language of the wolf. A loud voice is equivalent to growling. It means: “I am strong and could be a problem for you. You better leave me alone.” The louder you are, the more dangerous you appear to the wolf. For him, this means he could hurt himself unnecessarily if he attacks.

 

Is it also useful to stand up?

Yes, body size is important. In wolf language, it is synonymous with strength. That’s why you should not crouch or bend down to pick up stones as a weapon. It is better to move your hands above your head or to break branches from trees and waving them. But one should not directly attack the wolf. He might think that it is now too late to flee, and he must fight back.

 

Why is running away not a good idea?

It activates a hunting instinct in the wolf: the image of a fleeing prey. This can be observed very well in cats. A mouse dummy that does not move is ignored. Only when you let it slide across the floor does the cat’s hunting instinct kick in and it strikes. It’s the same with wolves. If I run away, I am the prey. In an encounter, the following message must always reach the wolf: a person is something big that is potentially strong and dangerous. Full stop. Then nothing will happen. You can walk slowly backwards, but always keep your eyes on the wolf without staring.

 

Why not?

Staring signals a challenge: whoever averts their gaze first submits. In an encounter, one should neither enter a competition nor be submissive, but rather remain calm and show self-confidence.

 

On another day, the same woman was surprised by three wolves at once. This time the animals attacked the woman’s dog. Why do wolves behave so aggressively towards dogs?

Wolves see dogs as nothing more than another wolf, but a wolf that is degenerate and weak. Moreover, he is an intruder in their territory. That is why dogs rather than their owners are attacked.

 

Sometimes, however, this also means that the dog is killed.

Yes, that can happen. Large predators are not squeamish about direct competition for food. It also happens that wolves kill other wolves. In Africa, lions kill hyenas without eating them afterwards. It is simply a matter of eliminating the competition. Basically, predators don’t like other predators.

 

In August, a group of hikers in Sufers came across two adult wolves that approached within a few metres. Later, the pups also followed the hikers. Why the approach and the subsequent pursuit?

In Africa, zebras are often seen walking directly towards a pride of lions. That means: “I see you and know what you’re doing.” It’s the same with wolves. Especially with parents, they may scout us to find out if we are a danger to the offspring.

With puppies it’s different. For them, everything is a game. They go after anything that moves. I have seen young African wild dogs chasing giraffes. This can be very dangerous for the young dogs, but they need to make experiences to understand it.

 

In North America, the authorities advise hiking tourists to always carry bear spray (pepper spray) with them. Would that also be conceivable in Switzerland?

Absolutely. It works very well in North America. All it takes there is a short blast and the wolf would takes off. Nobody likes to have pepper spray in their face. Using bear spray would only require a small change in our way of thinking. But maybe there is still some resistance at the moment.

 

Wolves seem to appear in settlements for no reason. Why?

You have to distinguish between migrating individuals and resident packs. Migrating individuals travel long distances every day and don’t really know where they are going. They don’t want to go to the city for sure. But in the densely populated landscapes, they only have to turn left once instead of right and they’re already in a village.

Packs, on the other hand, know their territories very well, and if they find something to eat near settlements, they are very likely to come back. This was a problem in Vättis, for example, where the Calanda pack often appeared. There, meat bait had been put out for fox hunting. The wolves quickly learned that they could easily get food here.

 

What does that mean for settlements?

Don’t leave food lying around. This can quickly lead to problems and unwanted encounters. In Turkey, where I researched bears and wolves, there is a town with a big rubbish dump. Bears and wolves come there every night to feed. It’s like a McDonald’s for them.

 

If wolves approach a settlement too often, they are declared problem wolves and can be shot. Would there be an alternative to this?

You can scare the animals away with rubber shot. But such measures, where the animals have to learn something, always take time. It is clear that this method would be very time-consuming for the gamekeepers and therefore not justifiable everywhere. Ultrasound could be used, similar to a cat or marten deterrent. In recent years, attempts have also been made to restrict the movements of predators by placing urine and droppings of conspecifics in strategic locations. We have to learn to be creative here.

 

Farmers in mountain areas have the same problem. They want to keep the wolf out of their pastures. Would there be more possibilities than fences and guard dogs?

We should try to exploit the technological possibilities. GPS transmitters, for example, could help to record the movements and preferred locations of resident packs. In the alpine region, packs have territories of about two hundred square kilometres. But certain locations are only heavily used during certain times of the year, such as the breeding season. The more information we have, the easier it will be to develop preventive measures.

 

There are currently about a hundred wolves in Switzerland, and the population is growing. Will more farm animals be killed every year?

The number of farm animals killed will probably increase. That is pure mathematics. But the number will not necessarily increase linearly: twice as many wolves does not necessarily mean twice as many lost farm animals. This because in Switzerland there are many wild animals such as deer, roe deer, chamois and wild boar, which play a much bigger role as prey than our sheep and goats.

 

But last year even a donkey was killed by wolves. Isn’t the situation deteriorating?

We should not condemn the wolf because of a donkey. At the end of the day, donkeys are part of its possible prey spectrum. It is important to maintain a rational mindset and take decisions based on facts and not emotions. Only in this way will the relationship between humans and wolves be sustainable.

A collaborative success story – how tourism can help research and benefit from it

by

Over the past few years, as part of a collaborative effort between the University of Zurich (Switzerland) and the Botswana Predator Conservation (BPC) and supported by the Botswana Department of Wildlife and National parks, we have equipped dispersing African wild dogs with GPS/Satellite radio collars. The aim of the project is (i) to follow dispersers after emigration from the natal group and to investigate the effect of landscape characteristics on dispersal distance, time and movements, and (ii) to gather crucial demographic parameters such as mortality rate, settlement success, and reproductive success after settlement in a new territory.

Recently, an unusually large coalition of eight brothers born in 2018 has emigrated from their natal pack inhabiting the Third Bridge – Budumtau – Xini region of Moremi Game Reserve. Thanks to the GPS data regularly sent to a base station via the Iridium satellite system, we have been able to remotely follow their movements. After emigration, they covered over 175 km in only five days before hitting the permanent swamp that surrounds the Kwedi Concession in the northern side of the Okavango Delta. During the past month they have been stationary in an area of about 180 km2 stretching between Vumbura Plains Lodge and Mapula Lodge. But dots on a map represent only a small part of the story… Are the eight brothers still together or have they split? What have they been doing? Have they met unrelated females and formed a new pack?

Figure: Movement trajectory of a dispersing coalition of eight male African wild dogs

Despite the collar sends us regular information, keeping up with the dogs over such large areas is almost impossible, unless we can capitalize on “the many eyes out there”. Tourists, guides, camp managers, all can contribute with their sightings towards research. No sooner said than done. We informed people at the lodges about the presence of the dogs in their area and asked them to report of any sighting and to send as many pictures as possible. Just a few days later we received the first information: a group of 9 dogs, including a collared dog, had been seen a few kilometres north of Vumbura Plains by the lodge staff. The pictures allowed identifying five of the original eight males and four unknown females. As expected, the males had indeed split and three brothers had probably gone a separate way. Or had they died during dispersal? An answer to the question arrived just a week later when a second sighting of 12 dogs was reported to us near Bushman Plains camp. Again, thanks to the pictures sent to us, we were able to identify the three missing brothers, who had clearly not died, among the 12 dogs. Future sightings will tell if this newly formed pack composed of 12 dogs will remain together or if some individuals will spin off (process known as secondary dispersal) searching for new mates with whom to build the own pack. But why would some undergo the risks of a second dispersal? Well, because of the eight males and four females only one of each sex will become dominant and reproduce. The others who will remain will help raise future pups but won’t directly reproduce. Therefore, some dogs may decide to take on the extra risk and continue dispersing. Bets are open, and chances are that the three brothers will part again. Time, and your sightings (!), will tell.

We, researcher, can benefits from any report and sightings, as it has been the case here. In return will be able to centralize all information and put together all pieces of the puzzle to share our knowledge with policy makers, stakeholders, and the tourism industry.

Please, keep sharing your sightings with us, of both collared and non-collared individuals, to help us protecting these amazing animals.

Biomechanically aware behaviour recognition using accelerometers

by

Accelerometers, Ground Truthing, and Supervised Learning

Accelerometers are sensitive to movement and the lack of it. Accelerometers are not sentient and must recognise animal behaviour based on a human observer’s cognition. Therefore, remote recognition of behaviour using accelerometers requires ground truth data which is based on human observation or knowledge. The need for validated behavioural information and for automating the analysis of the vast amounts of data collected today, have resulted in many studies opting for supervised machine learning approaches.

Ground-truthing. The acceleration data stream (recorded using the animal-borne data logger, bottom-left) is synchronised with simultaneously recorded video (near top right). Click on photo to view larger version. Photo credit: Kamiar Aminian.

In such approaches, the process of ground truthing involves time-synchronising acceleration signals with simultaneously recorded video, having an animal behaviour expert create an ethogram, and then annotate the video according to this ethogram. This links the recorded acceleration signal to the stream of observed animal behaviours that produced it. After this, acceleration signals are chopped up into finite sections of pre-set size (e.g. two seconds), called windows. From acceleration data within windows, quantities called ‘features’ are engineered with the aim of summarising characteristics of the acceleration signal. Typically, ~15-20 features are computed. Good features will have similar values for the same behaviour, and different values for different behaviours.

To automatically find robust rules to separate behaviours based on feature values, machine learning algorithms (e.g. Random Forest etc) are used. Here, candidate algorithms are trained (i.e. each algorithm is shown which datapoints correspond to which ground truthed behaviours). From here, algorithms are then tested by asking them to classify datapoints they haven’t seen yet into one of the behaviours. How well the algorithm does on testing data determines its ‘performance’. The model with the best performance wins and is selected for final use. Different ways of doing training and testing give rise to different forms of ‘cross-validation’.

 

Leveraging the Biomechanics Underlying Common Animal Behaviour

All animal behaviour is performed for a finite duration, following which the animal transitions to a different behaviour. The animal may be static for a while (e.g. resting), then begin foraging (involving movement), and then, perhaps perceiving threat, run (involving vigorous, periodic motion). Different behaviours may be performed in different postures (e.g. upright during vigilance, and horizontal while running).

We targeted an ethogram applicable to most animals: resting, foraging, and fast locomotion. This ethogram is a good match between covering most of an animal’s time budget and includes behaviours that an accelerometer is capable of ‘seeing’. For this however, features developed from the accelerometer signal must somehow be able to quantify posture, movement intensity, and movement periodicity. We reasoned that three well-engineered features – one each to quantify the posture, intensity, and periodicity – should be able to tell these three behaviours apart.

Feature engineering. Three biomechanically meaningful features were engineered from acceleration data – one each to characterise posture, movement intensity, and periodicity.

 

Using this approach, we predefined a hierarchical tree-like scheme that classifies broader behavioural categories into increasingly specific ones up to the desired level of behavioural resolution. Each node of this tree uses one or more features tailored to the classification at that node. Robust machine learning algorithms find optimised decision boundaries to separate classes at each node.

We demonstrate the application of this approach on data collected from free-living, wild meerkats (Suricata suricatta). The model accurately recognised common behaviours constituting >95% of the typical time budget: resting, vigilance, foraging, and running.

 

Model Performance: Leave-One-Individual-Out (LOIO) Cross-Validation

The ultimate goal of many behaviour recognition studies is to build models that will accurately classify data from a new individual previously unseen by the model. Leave-one-individual-out (LOIO) cross-validation is most appropriate to characterise the model’s ability to do this. Here, training is performed using data from all individuals but one, and the left-out individual’s data is used as the testing set. This process is carried out until each individual’s data has been the testing set exactly once.

In other forms of cross-validation, such as validation splits (also called hold out) or 10-fold cross-validation, both training and testing sets contain datapoints extracted from a single continuous recording on the same individual. This violates these methods’ assumption that datapoints are independent and identically distributed, since they are extracted from the same time series. LOIO cross-validation, however, has been shown to mitigate the effects of non-independence of data in human neuroimaging studies. Only one other study has performed LOIO CV (for animal behaviour recognition), and ours is the first study to do so on data from free-ranging, wild individuals.

Model validation. When it comes to evaluating model performance, a crucial aspect that sets leave-one-individual-out cross-validation (CV) apart is that it can test how well the model performs on data from an individual unseen by the trained model. Other approaches, such as train-test split (hold-out) mix data from different individuals, and hence cannot evaluate the model’s capability to generalise to new individuals. Note that for the sake of clarity, we’ve shown all individuals to have an equal number (M) of datapoints; in general, this might not be the case.

 

Reporting Metrics for Each Behaviour Reveals Fuller Picture of Model Performance

Overall accuracy (i.e. the sum of diagonal elements divided by the sum of elements in the confusion matrix), alone can be misleading and uninformative when it comes to characterising model performance in animal behaviour recognition applications. This is because of the issue of imbalanced classes, where durations of continuously filmed behaviours are naturally unequal. This makes the detection of rarer behaviours problematic.

Thus, overall accuracy alone cannot reliably guide model selection. A good model is one that has good sensitivity and precision for each behaviour of interest. This automatically guarantees good overall accuracy, whereas good overall accuracy does not guarantee good behaviour-wise performance.

 

Benefits of Biomechanically ‘Aware’ Learning

In our paper ‘A novel biomechanical approach for animal behaviour recognition using accelerometers’, we show that the proposed biomechanically driven classification scheme performs better than classical approaches based on black-box machine learning. Further, it is better able to handle the issue of imbalanced classes. Biomechanical considerations in the model can help provide valuable feedback on processes further upstream that are inaccessible to classical machine learning, such as defining the ethogram. The interpretability of the model sheds light on why some classes get consistently misclassified.

Grouping behaviours by biomechanical similarity in a hierarchical classification scheme can allow model sharing between studies on the same species. This eliminates the need to build entire models from scratch every time a new set of behaviours are to be recognised, as would have to be done with classical machine learning approaches.

Finally, we recently showed in this Movement Ecology methodological paper that our classification framework can be extended to magnetometer data as well. This helped to understand the similarity and complementarity of accelerometers versus magnetometers for behaviour recognition.

 

To find out more about biomechanical approach for animal behaviour recognition, check out our Methods in Ecology and Evolution article, ‘A novel biomechanical approach for animal behaviour recognition using accelerometers’.

 

This article was shortlisted for the Robert May Prize 2019. You can find out more about the shortlisted articles here.