NIWA Oceanographer Dr. Giacomo Giorli (pictured above) has been working with NeSI's Computational Science Team to incorporate machine learning tools into his acoustic studies of Cook Strait whales. (Image by: Dave Allen, NIWA)

Machine learning for marine mammals

“By understanding the abundance and distribution of different marine mammals in New Zealand, we can inform conservation policy, management of marine resources, licensing for offshore activity and create better environmental impact assessments.”

The Problem:
Data on New Zealand marine mammal populations and their movements are difficult to obtain. Protecting these species from human activity is difficult without understanding their numbers and behaviour.

The Solution:
To take underwater recordings around New Zealand and use machine learning to identify marine mammals based on their calls.

The Outcome:
Using machine intelligence, the sounds from three whale species were categorised. Information about these species will be used to inform regulation on marine activity, as well as conservation efforts.

 

Dr Giacomo Giorli is an oceanographer at the National Institute of Water and Atmospheric Research (NIWA). There, his team tracks marine mammal populations around New Zealand through underwater acoustic monitoring.

“We’re trying to develop a way to classify underwater sound. There are more than 50 different marine mammal species in New Zealand, and most of the ones that vocalize underwater make unique calls,” Giacomo said.                                                    

The team’s research focused on three signal classes and was able to record and categorise two unidentified species of beaked whale calls, as well as the Cuvier’s beaked whale species. These calls were measured in three separate locations at the mouth of the Cook Strait. By identifying the species, Giacomo’s team will be able to map their biology, spread and abundance around New Zealand.

“It was a short project and we had a very specific task we wanted to accomplish. We were able to collect a lot of data in that time. Then we used machine learning to categorise our recordings into three classes. It was very efficient. We had a 95-98 per cent accuracy when defining these classes.”

To create this machine learning system, Giacomo worked with scientific programmer, Alexander Pletzer, one of NeSI’s Computational Science Team members based at NIWA. Giacomo created training data sets, while Alexander built the algorithm for the machine learning program to distinguish between the categories.

“Alexander was great. He was very knowledgeable about these artificial intelligence systems. His skillset complemented our team’s research. Through Alexander’s approach, we were able to get great results out of a small training data set.”

The algorithm filtered recordings to remove ambient noise before separating marine mammal calls by spectral shape.

Using machine learning to categorise different recordings is the first step to efficiently study all of New Zealand’s marine mammal populations using underwater acoustics. Manual processing of recordings for even one species can take a long time, which is why machine learning was used to sort the data.

“If we have to manually study one species from acoustics recordings collected over long period of time, it can take an entire year. Repeating that study for every marine mammal species that you might have recorded in your data would take a lot of extra time. It takes so long to analyse the data, so automatic tools become necessary to speed up the data analysis and reduce the cost of the research,” Giacomo said.

The next step for the team is to expand their research to identify more of New Zealand’s marine mammal population through acoustics. This will help study the distribution and occurrence of these populations around the country during different seasons.

This will also provide useful information to for a science-based regulation and consent of marine construction and offshore industries that can pose a threat to marine species. By knowing their spatial and temporal presence, these projects can take place at times of the year when there is no threat to protected species.

“By understanding the abundance and distribution of different marine mammals in New Zealand, we can inform conservation policy, management of marine resources, licensing for offshore activity and create better environmental impact assessments,” said Giacomo.

NeSI will play an important part in future projects with Giacomo and NIWA, as the machine learning algorithm is expanded to categorise new vocal calls. With ongoing support from NeSI team members and resources, New Zealand’s marine mammal population will be better protected.

Interested in learning how NeSI’s Computational Science Team can help you? Or have an example of NeSI supporting your work that we can profile as a case study? Get in touch by emailing support@nesi.org.nz.

Next Case Study

Joao Albuquerque NeSI Case Study University of Auckland

New Zealand 2100: The future through high-resolution wave modelling

"I’m lucky to have access to a cluster like NeSI’s. Without it, I’d be running this model for years, rather than about a month.”
Subject: