Context
In June 2022, Renewable UK released “Roadmap to net zero: a manifesto for a fully decarbonised power system by 2035”. This report provides both short and long-term recommendations that the UK government should follow in order to achieve full decarbonisation of power generation in the next 13 years, and splits recommendations into six key areas which include the “accelerating renewable deployment”.
Offshore Renewable Catapult (OREC) is committed to accelerating renewable energy deployment by reducing consenting times as part of the UK Government’s Energy Security Strategy. Marine mammal tracking reports for Offshore Energy project consenting have been of increased interest by local Governments, including Wales [1,3], and could become key to be able to reduce consenting times for offshore renewable energy projects.
Data
To achieve this in this challenge, passive acoustic data from 32 sonobuoys following a 4×8 grid arrangement (and separated around 8km each) at the Southern Gulf of St. Lawrence (Canada) is provided [4]. This area is known as one of the core habitats for the North Atlantic Right Whales (NARW)
Data was recorded with DIFAR buoys in multiplexed format by 32 buoys. This was de-multiplexed and stored in .wav format, and ordered as omni, sine (East-West), and cosine (North-South) channels. The de-multiplexed data was stored at a sampling rate of 8000 Hz. Channel numbers and corresponding locations are stored in the filename. The signals mainly consist of upsweep and down-sweep calls of NARW.
Each sonobuoy channel was reviewed independently and individual calls were detected and classified by call type. The objective of this process was to obtain a good set of calls for testing localisation algorithms, and detection of calls have been carried out with that objective in mind. The set should not be considered as an ideal ground-truth set for testing/evaluation of classification algorithms, although we certainly recommend people explore this as well.
Ground truth: Coincident shipboard and aerial NARW visual surveys as well as oceanographic surveys by Slocum ocean gliders were conducted within each sonobuoy array. A spreadsheet containing information of the different whale sightings is also included in the provided data. Furthermore, manual labelling of identical calls detected simultaneously at different sonobuoys is included.
Challenge Goals
Basic
The basic goal of the challenge is to use machine learning to localise and identify individual whales using signals from one or more buoys. Improved localisations could be implemented by detecting simultaneous calls in nearby sonobuoys.
Intermediate
The intermediate goal is to track individual whales throughout successive calls, including estimates of direction, speed and depth. Repetitions of unique calls should be identified within data from nearby sonobuoys and used to track whales throughout the 2-day 3.5-hour survey data.
Stretch
The more advanced goal of this challenge would be to estimate the density of NARW in the area during the survey time, as well as classifying different types of calls associated to either different marine mammal species or differences amongst individuals within the same species.
A full data description of the dataset can be found in the DCLDE2024 dataset report [4]. The dataset can be accessed through the DRDC website. Access details for this website can be found in Chapter 4 of the report.
[2]: https://www.marineenergywales.co.uk/developers/consenting-guidance/
[4]: http://www.dclde2024.com/data-sets/
This dataset is kindly made available to the DCLDE community by the research consortium consisting of Dalhousie University, Fisheries and Oceans Canada (DFO), Defence Research and Development Canada (DRDC), the Acoustic Data Analysis Centre (ADAC), the National Oceanic and Atmospheric Administration (NOAA) Northeast Fisheries Science Center (NEFSC), the New England Aquarium (NEAq) Anderson Cabot Center for Ocean Life (ACCOL), and the Canadian Whale Institute (CWI). The acoustic data was collected by the Royal Canadian Airforce (RCAF).