Right whale dataset - PAMGuard deep learning tutorial

Deep Learning and Right Whales - PAMGuard Tutorial Dataset

General Information

Overview

This is a dataset to be used in part one of the PAMGuard tutorial Deep Learning in PAMGuard. The dataset is from a low-frequency, single-channel acoustic recorder recording off the coast of North America (Cape Cod Bay and Great South Channel) and contains right whale and humpback whale calls. The tutorial text and other related files can be found at https://github.com/PAMGuardLearning/DeepLearningPamguar

Year of Data Collection

2009

Geographic Location of Data Collection

Cape Cod Bay and Great South Channel (off Massachusetts, USA)

Funding Sources that Supported the Collection of the Data

The Bureau of Ocean Energy Management funded MARU deployments and data collection (M10PC00087 for Georgia and North Carolina, M15AC00010 for Virginia, M14AC00018 for Maryland).

Recommended Citation for this Dataset

The deep learning model was created by and the dataset is discussed in the following publication:

Shiu, Y., Palmer, K.J., Roch, M.A. et al. Deep neural networks for automated detection of marine mammal species. Sci Rep 10, 607 (2020). https://doi.org/10.1038/s41598-020-57549-y

The full dataset can also be found on the NOAA website.

NOAA Northeast Fisheries Science Center. 2023. The North Atlantic Right Whale Annotations from Passive Acoustic Data Collected in the Western North Atlantic Ocean. NOAA National Centers for Environmental Information. https://doi.org/10.25921/2c09-ng58 [access date].

Data & File Overview

Description of Dataset

The dataset contains one hour of right whale data collected using a Marine Autonomous Recording Unit (MARU). The MARU was moored approximately 5m from the ocean floor and had a system sensitivity of around -146 dB re 1µPa/V with an effective dynamic range of 66 dB.

The dataset also contains pre-processed data from a four-day deployment. These data have been processed in PAMGuard software (www.pamguard.org) for right whale calls using two detectors: the deep learning detector and PAMGuard's right whale edge detector.

A deep learning model with PAMGuard metadata is included as a zip file. This can be imported directly into PAMGuard.

Note that these data are a subset of the DCLDE 2013 dataset, which can be found here. A comprehensive description of this dataset can be found in Shiu et al. (2020). The full dataset can also be found on the NOAA website.

Files

There are two directories: wav and viewer_mode. wav contains .wav files, which are uncompressed audio files.

The viewer_mode folder contains a PAMGuard database and associated detection files in the PAMBinary folder and subfolders. The NARWDCLDE2013_4Days - Viewer.sqlite3 file contains PAMGuard settings and some basic metadata. The detection .pgdf files are non-human-readable files that contain detection data such as detected clicks, frequency contours of whistles, moans, and other tonal sounds, and a time series of soundscape metrics. These files can be opened with PAMGuard software (www.pamguard.org), MATLAB, and R.

The folder structure is as follows:

├── wav                    #Three example days of acoustic recordings│   ├── NOPP6_EST_20090329_120000.wav│   ├── NOPP6_EST_20090329_121500.wav │   └── '''├── README.md                    #This readme file├── right_whale_model.zip                #Right whale deep learning model in PAMGuard compatible format.     └── viewer_mode                 #Processed data from a whole deployment    ├── NARWDCLDE2013_4Days - Viewer.sqlite3     #PAMGuard database    └── PAMBinary       ├── 20090228                 #Folders containing detection files      ├── 20090229       ├── '''

Sharing and access

These data are open source under Creative Commons Attribution 4.0 International. This means that these data can be used in other tutorials as long as the original authors are credited. Using these data for scientific purposes requires the permission of the authors.