Copy
View this email in your browser
 e-newsletter March 2021
 
 
 
 
 
The MERIDIAN logo and full name on a blue and grey split background.
Introduction from the PI

In the span of three years, MERIDIAN has established itself as a unique endeavour in Canada and internationally, advancing the use of Big Data, AI, Machine Learning in ocean science, to the benefit of our oceans. Key achievements of MERIDIAN include the development of several open-source software packages, for example Ketos, now actively used by a number of research groups in Canada and internationally for developing novel AI-based acoustic detection and classification algorithms. Another major contribution of the MERIDIAN project has been the development of a web-based Ocean Soundscape Atlas, an application allowing researchers and ocean managers to visualize ocean noise pollution created by vessel traffic, and to quantify its impact on endangered marine species. 

In its short lifetime, MERIDIAN has built a large number of partnerships and collaborations, working with academic researchers, government scientists and NGO groups on solving pressing issues, such as protecting the endangered North Atlantic right whale on Canada's east coast. Data scientists and technicians have been trained by MERIDIAN with unique skill sets that combine elements of ocean science and marine biology with advanced Data Science and Machine Learning. Moreover, MERIDIAN has been active in knowledge transfer, organizing workshops and - recently -  webinars to train Canada's ocean researchers in the use of data science and machine learning.

Below, you can read about our Winter Webinar Series and catch up with the latest developments on the Ocean Soundscape Atlas app. You can also read about how we are collaborating with DFO scientists on the management and analysis of AIS vessel tracking data, as well as the acoustic detection of North Atlantic right whales. Finally, you can read about our experimentation with deep generative models to synthetically create underwater sounds and what this can be used for.

Funded in 2017 under the CFI Cyberinfrastructure program, with provincial support from Nova Scotia, Québec and British Columbia, MERIDIAN was initially conceived for a period of 3 years and was extended for an additional year due to the pandemic. It is therefore with great pleasure that I can announce that we have been able to secure funding for a second phase of MERIDIAN, enabling the team to continue their work several years into the future. 

In the years to come, the MERIDIAN 2.0 team, supported by a recent CFI Innovation Fund award, the Nova Scotia Research fund, and Dalhousie University will be working on the “Artificial Intelligence Meets Oceans” project, building the Marine Artificial Intelligence Platform (MAIPL), a virtual laboratory that will make our AI toolkits accessible to a wide range of Canadian and international ocean researchers.

We thank you for your support during the first years of MERIDIAN and look forward to continuing to collaborate with the ocean science community in the coming years.


A black-and-white photograph of Dr. Stan Matwin.  Stan Matwin, MERIDIAN PI

 
 
 
 
 
 
News, news, news
 


Winter Webinar Series

This winter, MERIDIAN hosted a series of webinars addressing a wide range of topics related to the intelligent curation and analysis of underwater acoustics data and vessel tracking data:

1. Underwater soundscapes
2. What do we know about fish sounds?
3. Visualizations for preference inspection in group decision making
4. Acoustic detection and classification in the modern era of big data, deep learning, and open science principles
5. Machine learning for in-situ automated analysis of underwater acoustic data
6. Management of underwater acoustics data
7. Acoustic telemetry detection range modeling

With an average attendance level of over 40 attendees per webinar, the outreach effort was well received and attracted considerable interest. Recordings of all the webinars (as well as other webinars we have hosted in the past) can be found on our YouTube channel and the presentation slides can be found on our homepage.

 

Latest developments on the Ocean Soundscape Atlas

In the last few months, the Ocean Soundscape Atlas saw itself enhanced with the addition of new maps at high frequencies (1, 2, 5, 10 and 20 kHz) in the Estuary and Gulf of St. Lawrence. It is highly suspected that the shipping noise at those frequencies are affecting negatively the endangered St. Lawrence Estuary beluga whale population. Furthermore, a new 3D spectrogram has been added to the collection of tools. In addition to the usual time and frequency axes, the third axis is allowing the exploration of the distribution of the sound levels (via a risk of exceedance axis or a quantile axis). It is based around the concept of the Soundscape Cube designed by Yvan Simard et al. 2016. Also, weekly maps were added to facilitate the analysis by having an intermediate between daily maps and monthly maps.

The Ocean Soundscape Atlas seems to have caught the attention of many users around the world, who have made more than 250,000 requests on the website since July 2020. As part of an effort to increase awareness of the app, we have also created a promotional video: 

https://www.youtube.com/watch?v=C_wbpy7kXLI (French)
https://www.youtube.com/watch?v=DZnz38jb5dg (English)

We are presently working to incorporate some zones of interests into the Soundscape Atlas such as the Saguenay-St. Lawrence Marine Park. It will allow for instance to compute statistics on those predefined zones. It is also planned to add new datasets for the Saguenay fjord computed using 3D ray tracing models. We are also planning to add data for year 2018 in the Estuary and Gulf of St. Lawrence that will complement the existing 2013 data and allow for a comparison between a 5 years time interval. Finally, we are working on the backend to make sure that the application will be running for years to come and that it will be easy to maintain.

References:
Simard, Y., Bandet, M., Gervaise, C., Roy, N., & Aulanier, F. (2016, July). Soundscape cube: A holistic approach to explore and compare acoustic environments. In Proceedings of Meetings on Acoustics 4ENAL (Vol. 27, No. 1, p. 070012). Acoustical Society of America.

 
 
 
Collaborative project with DFO on management of AIS vessel-tracking data

AIS (Automated Identification System) is an automatic tracking system used by seafaring vessels equipped with transcievers to report voyage related info, such as position and speed. This project with DFO seeks to employ AIS vessel traffic data collected from land-based receivers for tracking the spread of non-indiginous species in Canadian waters. MERIDIAN contributes to this project by providing tools and support for the processing of vessel traffic data. This will enable not only a more in-depth understanding of how non-indiginous species may spread across ecosystems via anthropogenic sources, but also provide an AIS data framework that can be used for future research applications.

Some of the challenges of working with AIS data are managing the volume of AIS messages and quality control filtering of erroneous messages. MERIDIAN approaches these challenges by providing a database of extracted messages that can be used for ship activity assessment and characterization in the Atlantic, Pacific, and Arctic Canadian regions. In addition to the database are tools for visually representing vessels and ecoregions of interest, processing and filtering vessel tracks from the raw message data, and exporting data in a GIS-suitable format

 

Synthezising underwater sounds with deep learning

In current years, passive acoustic monitoring is benefitting from ongoing advances in deep learning research, which is enabling the development of improved acoustic detectors and classifiers. However, performance remains limited by the scarcity of labeled data to train deep learning models, and it is often not feasible or expensive to acquire more data. To mitigate this problem a promising solution is to use a strategy called data augmentation which is an inexpensive way to acquire more labeled data by synthetically generating them.

Among suitable augmentation strategies, generative methods are capable of generating completely new samples, greatly enhancing the amount and variety of data in a training dataset. For speech synthesis, there are several lines of research currently being explored such as Generative Adversarial Networks (GANs) and autoencoders. The MERIDIAN data analytics team is working on a particular branch of deep generative methods called autoregressive models that are capable of synthesizing raw audio by sequentially generating individual audio samples until the entire audio is complete. They are called autoregressive models because each generated sample is fed back into the model in order to generate the next one. 

To showcase what we are already able to achieve, here is a generated sample of a North Atlantic right whale upcall (right) and an original authentic recording (left) so you can see, listen, and compare yourself. (Click the image to listen to the audio.)

 
 
 
 
 

New Data Scientist: Nicole 

Nicole joined the MERIDIAN team in January this year on a 4-month position funded through a DFO-AI initiative. Nicole graduated from Dalhousie University in 2020 with a Master of Science, specializing in atmospheric physics, and holds a Diploma in Meteorology. Throughout her academic career she had the opportunity to work in laboratories at Dalhousie University where she analyzed atmospheric aerosol properties and tested air quality as well as at the Canadian Nuclear Laboratory where she investigated cosmic ray muon scattering through dense materials. She also had the opportunity to work in the field where she was on board the R/V Sharp. The ship sailed along the northeastern coast of the USA and Canada sampling ambient aerosols in order to investigate their physical properties and influence by meteorology. At MERIDIAN, Nicole is experimenting with novel representations of audio data and novel neural network architectures, aiming to further improve the performance of our North Atlantic right whale upcall detector.

 
 
 
 
A black-and-white photo of Dr. Oliver Kirsebom.  Oliver Kirsebom, Senior Staff Scientist

  
Matthew Smith, AIS Data Manager

  Bruno Padovese, Data Scientist

  Pierre Mercure-Boissonnault, Data Analytics and Management Expert
 
MERIDIAN is funded by
The logos of MERIDIAN's various sponsors.
CC BY 4.0 logo
With the exception of logos and third party images or where otherwise indicated, this work is licensed under the Creative Commons 4.0 International Attribution License.

Want to change how you receive these emails?
You can update your preferences or unsubscribe from this list.

 






This email was sent to <<Email Address>>
why did I get this?    unsubscribe from this list    update subscription preferences
Faculty of Computer Sciences, Dalhousie University · 6050 University Ave · Halifax, NS B3H 4R2 · Canada

Email Marketing Powered by Mailchimp