The Rimutaka Forest Park Trust has been working alongside the Department of Conservation since 1988 to restore the forest clad hills on the eastern side of Wellington.
Kiwi were first released into the park in 2006 and the population has grown from 6 adult birds to more than 100 roaming the 22,500 hectare park and adjacent areas.
As kiwi numbers increased, the task of tracking each bird in rugged terrain to attach and maintain radio transmitters on their legs became exhausting, expensive, and unsustainable.
In 2011, acoustic recorders were deployed within the park. They recorded thousands of hours of nocturnal sounds. It quickly became apparent that the group would need something robust and automated to make sense of their enormous - and rapidly growing - dataset.
The team at Dragonfly Science began experimenting with some Python scripts and applied statistical sampling methodologies to create a web-based tool that combines machine learning with the human identification of sonograms and audio. From this, Songscape has evolved.
Songscape was used to successfully monitor the distribution of North Island brown kiwi in the park. Songscape was able to eliminate 95% of "dead space" from recordings and used human identification to further determine distribution and population size.
Songscape was able to sift through thousands of hours of raw data, seeking out high-probability male or female kiwi call signatures to create a manageable subset of 1 minute segments for online human assessment and eventual statistical analysis.
The results allow the Rimutaka Forest Park Trust to make better-informed decisions about the ongoing management of their growing kiwi population and provided insights into the biodiversity health of their habitat.