The Department of Conservation (DoC) is using AI and machine learning technologies from AWS and Spark subsidiary Qrious along with AWS cloud facilities to improve its conservation of endangered birds — the kākāpō and the kiwi — and to share its techniques and its data with other conservationists.
Details of the projects were presented by DoC CTO Mike Edginton at the AWS Summit in Canberra earlier this month.
DoC shifted its compute and storage requirements into AWS in 2017, and Edginton said this had enabled it to consolidate valuable research data previously distributed across multiple systems.
He told Computerworld NZ that DoC had used the Couchbase open source distributed database to consolidate onto AWS masses of data on the critically endangered kākāpō previously held on individual researchers’ laptop and other DoC systems.
“There are now 200 kākāpō. Six months ago there was only about 137,” Edginton said. “We've had a really good breeding season, and an important part of the past breeding season has been the database we built using Couchbase that enables offline and asynchronous connection with devices in the field, and when we have connection, back into AWS.
He said the technology had enabled DoC to create a digital twin of every kākāpō that can be shared among researchers and with the wider community.
“Each bird has what is basically an RFID tag. We now know everything about a bird from the time the egg was laid through to its adult life. We know who it socialises with, we know how much food it eats every day, because we have scales out in the forest where they get on the scales and get a little treat when they get on there and get a bit of food.
“We now have essentially a digital twin of those birds that we can use for mating purposes and for general, animal husbandry.
“After 30 years of managing these birds intensively, we now have a lot of IP and we are now bringing all of that IP onto one platform on AWS.
“We have made this database open source and very shortly we'll publish the code so that other conservation agencies around the world can use that code to build similar systems for other species.”
Listening to the kiwis
DoC is also using AWS facilities to store and analyse sound recordings from thousands of listening stations around New Zealand to detect kiwi calls.
Analysis of the recording can tell researchers if kiwis were present and if they were male or female, adult or young, but this was a specialised skill, Edginton said.
“We've got 60+ terabytes of call data that goes back over 15 to 20 years,” Edginton said. “I wanted to ensure that data was secure because it’s been sitting on hard drives, and that we could share this code with the many community groups that are helping to protect the kiwi around New Zealand.”
DoC has engaged Qrious, a data analytics company, to develop a means of using AI to detect kiwi calls in the recordings.
“Instead of just listening to the call of the bird, Qrious have turned it into an image using standard AI image recognition tools in AWS,” Edginton said.
“And using that, we've been able to get a confidence rating of 100 percent that it is a bird and then be 87 percent sure that it's a kiwi call. The machine learning has also been able to hear calls a human ear couldn't hear because of background noise or because they were a long way off.”
He said DoC now intended to apply the same techniques to track the sounds of other species in the recordings.