Researchers at Verum Group have successfully developed machine learning approaches for individually identifying roroa – great spotted kiwi (Apteryx haastii) by their calls.
A study examining the voices of kiwi in the Paparoa Range has shown it is possible to identify unique voices of individual birds. The study has shed new light on their territorial roaming and other interactions.
In mid-July 2020, Verum Group’s Behavioural Ecologist Dr Laura Molles returned to the Paparoa Range bush to collect the kiwi call recordings from the audio recorders the Verum Group team set-up in June 2020. Verum Group researchers analysed over 2000 hours of recordings.
Laura has counted 650 calls of female and male kiwi recorded across 10 of the 11 sites. The next step was measuring the detailed characteristics of all the songs to distinguish the different individuals. This gave the researchers an idea of how many kiwi are out there.
However, Laura says they had to wait until the nesting season started, around September, to know exactly who was who. "Combined with the data from the transmitters carried by some of the kiwi, we will be able to find nests and match the calls with the individuals in each nest."
Laura says the study will hopefully allow them to monitor the Great Spotted Kiwi population in a non-invasive way, avoiding the need to capture the birds. "We hope that individual recognition will allow non-invasive monitoring, but we first need to understand how robust and practical the vocal ID method is. That’s what we’re doing now by studying a group of birds already fitted with transmitters for management purposes."
Being able to follow individuals based on their voice ID will give the researchers a better understanding of social interactions between individual kiwi. "Birds have a range of communication needs, but most of it is centred around either finding a mate or telling off a neighbour."
In June 2020, Laura was assisted by Ray Beckford, Atarau Sanctuary Manager, to set up a network of 10 AudioMoths recorders, which were also paired with commonly-used DOC audio recorders to assess their effectiveness. The AudioMoths are small, open-source recorders used to record wildlife sounds. Placed on ridgelines with 1km diameter recording areas, the devices recorded between 45 minutes after sunset and before sunrise when kiwi are most likely to be calling.
Using recordings from already identified individuals to build a database for a machine learning process, the software was then trained to recognise existing birds calls and identify new ones. Several recordings of the same bird captured by different recorders shed light on a bird's use of its territory.
See our latest news items on kiwi monitoring:
February 2023: Dr Laura Molles – listening to birds
March 2022: Using song to count and identify individual birds in the wild
When you hear birds calling, you’re listening to a conversation. Often, the birds you hear might know each other as individuals, and they may recognise one another based on song alone. Conservation can benefit from this voice recognition too. Being able to identify individual birds from their song could allow us to monitor individuals, neighbourhoods, and populations with minimal disturbance.
July 2020: Verum Group studies kiwi calls
In mid-July, Verum Group’s Behavioural Ecologist Dr Laura Molles returned to the Paparoa Range bush to collect the kiwi call recordings from the audio recorders the Verum Group team set-up in June. Verum now have two weeks of recordings, with over 2000 hours to analyse.
Laura was interviewed by CNN on bird dialects, Birds aren't all singing the same song. They have dialects, too.
For more information about kiwi monitoring please contact Dr Laura Molles.
See our published ecology research papers to learn more about kiwi monitoring.
Find out more information about our kiwi santuary at Atarau: Atarau Sanctuary website