- Researchers in New Zealand combined sound data from acoustic monitoring devices with species occupancy models to assess the success of translocating an endangered New Zealand bird, the hihi, to invasive species-free locations.
- The scientists say in their paper that advances in acoustic monitoring and statistical techniques have made it possible to infer spatial and temporal changes in population dynamics without needing to track individual animals.
- As wildlife managers increasingly release animals back to their historic ranges, cost-effective, non-invasive data collection, automated pattern recognition, and analysis techniques that predict the likelihood of species occupying a given location over time could improve the success of the reintroduction process.
When you’re new to an area, it takes time to settle in. We all want a comfortable place to live, in a safe neighborhood, with convenient access to shops and schools yet without too much traffic.
It’s the same for birds.
How animals find their way around unfamiliar territory is a key question for wildlife managers who translocate animals from one place to another. Now, an international research team hopes to answer that very question by “eavesdropping” on a group of one of New Zealand’s highly endangered birds with acoustic recording devices.
The researchers used the devices to listen for the calls of a group of hihis (Notiomystis cincta), small vocal birds that had been driven extinct across mainland New Zealand in the 1880s but recently released in a protected area on the country’s North Island.
Bringing back lost species
Moving animals from a threatened area to a safer one or from an overcrowded area to one where the species may have been previously extirpated is a management technique that’s become more common as human activities eliminate native animals.
Humans have decimated animal populations through hunting, destroying their habitat, or bringing non-native predators and competitors into new areas, especially on islands. New Zealand was once home to more than 130 species of land birds, 93 of which were found nowhere else on Earth, but no native terrestrial mammals.
More than half of these birds are now threatened or extinct. Human settlers hunted the larger species and cleared vegetation to raise sheep. The rats, cats, and dogs they brought with them have devastated populations of hihis, as well as kiwis, rails and other native birds across the country.
Automated recording units
Translocation of endangered animals into new, presumably safer, environments, is risky; successful translocation requires keeping tabs on how the animals move in their new space, which is challenging, especially when the species are small or cryptic and the habitat is closed forest.
To understand where reintroduced animals go once they’re released, managers have tracked them with radio tags or, for smaller animals, marked them and physically followed them around. Both options tend to be expensive, time-consuming and temporary; they may also affect the animals’ behavior.
For this study, the researchers chose a new approach that combined sound data collected by automated recording units (ARUs) with a set of models to predict at a fine scale of how and where the birds would move.
They tested whether they could use the acoustic data and the models to ascertain the hihis’ movements, determine whether and where the birds were settling in and establishing territories, and identify the environmental features the birds preferred in choosing where to live.
Increasingly, scientists are setting out grids of passive acoustic monitoring devices to detect the presence of a particular species or the composition of species in a given area. These automated recording devices allow field researchers to overcome some of the logistical challenges of surveying and monitoring cryptic or nocturnal species such as bats around wind farms.
While tags enable researchers to find known individuals, using ARUs may help them monitor reintroduced birds more effectively than radio tracking in large areas of difficult terrain, co-author John Ewen, a senior research fellow at the Zoological Society of London, told Mongabay. “They can be deployed and work without need of people tracking radio-tagged individuals,” Ewen said. “They can also continue recording, giving more information than, for example, using vehicle or aerial methods to obtain snapshots of animal locations. Another thing to consider is the cost to individuals for wearing a radio tag, [which] can reduce movement behaviour and survival in some cases.”
Listening in
In 2017, the research team set up a grid of 31 ARUs across the fenced Rotokare Scenic Reserve to monitor 40 juvenile hihis that had just been released at two locations there as part of a larger reintroduction program. Earlier successful hihi translocations had found that the birds would broadly explore a new area, then would settle in territories, often near streams and supplemental feeders.
The ARUs listen to their surroundings 24/7, are non-invasive, and require no human presence after installation, allowing them to detect the hihis’ presence without impacting the birds’ behavior. But could they capture sufficient data to enable the researchers to assess the success of the reintroduction?
The team left the ARUs in the field for a total of 32 days, programming them to record two periods of two hours each day. They then divided the sound data into nearly 16,000 15-minute recording periods, from which they could determine the presence or absence of hihi calls by training algorithms to recognize the sounds.
Machine learning for data processing
The researchers also tested the capacity of machine learning to find hihi calls among the mountains of sound data generated by the acoustic devices in the field. They created a pair of automated call recognition models, one with few false detections and another with a high number of true detections. They tested the models on a subset of the call data by manually inspecting the sound spectrograms to determine whether hihis were present in a given recording. The call detection models correctly assessed whether the birds were present or absent in each test recording period, which told the researchers where the hihis were and when they were there.
The hihi ‘stitch’ call sounds like two marbles being knocked together, lead author Oliver Metcalf told Mongabay, rendering it indistinct both to the ear and on a spectrogram and thus a challenging call type for AI detection. “Although species diversity is low in New Zealand, some of the native songbirds have an incredible vocal repertoire, which I think makes it as acoustically complex as almost anywhere,” he added. “If [automated detection] worked at Rotokare with hihi, I think it stands a really good chance of working anywhere.”
With the sound data from specific dates, times, and locations captured by the grid of ARUs, the researchers compared where and when the hihis were present and absent to possible drivers, including the animal’s proximity to running water, suggested by previous translocations.
They conducted this comparison using dynamic occupancy models, which estimate the probability of a species occupying a given location over time, by modeling extinction and colonization probabilities at the 31 sampling locations. The dynamic occupancy model enabled the team to look for changes in the birds’ patterns of occurrence over time and use these patterns to estimate habitat preferences and post-release behavior, said Metcalf, a doctoral student at Manchester Metropolitan University.
The researchers expected the female hihis to settle on breeding territories after about four weeks. Their analyses of the call data showed that as time went on, the spatial and temporal patterns of the calls suggested this was happening. For example, over time, hihi calls were detected at fewer sampling locations, though they continued to be detected consistently at the same sampling stations. Moreover, over time, they increasingly stayed closer to water courses within “preferred” areas.
“We found the hihi were initially pretty random in their movements around Rotokare, as you would expect from birds exploring a new home,” he said, “but towards the end of the study they had settled down onto territories, and they preferred to have territories in areas close to water.”
“The more we can learn what is perfect for hihi,” Ewen said, “the more we hope to reduce the intensive management support we need to provide to allow reintroduced populations to flourish.”
A learning process
Metcalf said that traditional survey methods could still monitor released animals accurately and at low cost in easily accessible areas, but as ARUs became more accurate, easier to use and less expensive, the presence/absence data they collect would really help the process.
Acoustic monitoring devices can have high initial costs, though these are dropping, he said, and producing an algorithm to detect species can be time-consuming. “If the recorders can be used multiple times or for long periods, the cost would actually be pretty cheap compared to getting people to collect a comparable amount of data,” he added, “and once the detection algorithm is working, it can be used over and over again, so data analysis can be very quick. Radio-tracking provides fantastically detailed information on individuals, but can be quite expensive and is more invasive, particularly for smaller species.”
“We are increasingly reintroducing species back to mainland restored forests,” Ewen said, which, unlike island reintroductions, allows released animals to mover farther and enter landscapes where threats, such as non-native predators, have not been removed or controlled. “We are still learning how the size, shape and connectivity of restored forest sites influence the outcome of reintroductions.”
Metcalf said he thought their listening and modeling technique could be applied more beyond conservation reintroductions. “I think this would be a fantastic way to cost-effectively monitor the impacts of large-scale development projects like windfarms on multiple species.”
Citation
Metcalf, O., Ewen, J., McCready, M., Williams, E., & Rowcliffe, M. (2019). A novel method for using bioacoustics to monitor post-translocation behaviour in an endangered passerine. Methods in Ecology and Evolution.
FEEDBACK: Use this form to send a message to the editor of this post. If you want to post a public comment, you can do that at the bottom of the page.