Tutorials and workshops
We are excited to offer 4 different tutorials on the first day of DCLDE 2022: 3 full day tutorials and 1 half-day tutorials. Tutorial registration provides for full-day participation and includes lunch.
Some tutorials are allowing virtual participation, and registration for virtual participation is open. Please check below and contact tutorial organizers directly if you are interested in participating virtually.
Connect the dots! Multi-target tracking in acoustics
Full day (Supporting virtual participation)
Organizers: Pina Gruden and Yvonne Barkley- University of Hawaii at Manoa, Selene Fregosi- Pacific Islands Fisheries Science Center
Location: 3rd floor Bluefin room
Target tracking is an active area of research relevant to a range of disciplines, including underwater acoustics. For example, many passive acoustic systems output bearing and/or location estimates, which are often processed manually to track individual or closely spaced acoustic sources (e.g., marine mammals). This is time-consuming and introduces subjectivity into localization results. Target tracking methods can speed up the processing and produce objective results by automatically connecting these data points. However, the challenge lies in precisely estimating the trajectories over time when measurements are noisy, measurement origin is uncertain, other non-target sources are present and data points are inconsistent due to animal behavior, such as when tracking multiple vocalizing animals with variable call rates.
This workshop will discuss target tracking in two parts. First, morning tutorial presentations will introduce the theory behind target tracking techniques and formulate them within the Bayesian framework. Basic single target tracking concepts, including Kalman and particle filters will be introduced. Additionally, we will discuss multitarget tracking techniques (MTT) and their applications, including traditional and non-traditional MTT. The afternoon practical portion will discuss the results of a challenge, which will be shared in advance with workshop participants. The challenge will include the basic code to guide participants through examples of target tracking techniques using a simulated dataset and a subset of the conference dataset. We also encourage participants to apply their own approaches to tracking the same datasets to foster further discussion and promote group learning. All levels are welcome to attend.
Pre-requisites: To participate in the challenge portion of the tutorial, participants will need to be familiar with MATLAB or other programming language and should download the challenge dataset in advance.
Contact: Pina Gruden- pgruden <at> hawaii.edu
An introduction to using and implementing the Low-Frequency Detection and Classification System (LFDCS) with archival and near real-time acoustic datasets
Full day (Supporting virtual participation)
Organizers: Genevieve Davis and Julianne Wilder- Northeast Fisheries Science Center, Mark Baumgartner- Woods Hole Oceanographic Institution
Location: 3rd floor Yellowtail room
The Low-Frequency Detection and Classification System (LFDCS) is a software system built for automated detection and classification of low-frequency baleen whale vocalizations in archival and real-time acoustic data. In the first half of this workshop, we will cover the fundamentals of the desktop version of the LFDCS, including the generation and attributes of pitch tracks (contour lines that trace tonal sounds), discriminant function analysis, building a call library, browsing/exporting autodetections and analysis results, and species-specific analysis protocols. In the second half of the workshop, we will cover the use of the LFDCS to detect and classify vocalizations recorded and transmitted via satellite in near real-time by autonomous buoys and ocean gliders equipped with the programmable digital acoustic monitoring (DMON) instrument. Following an introduction to the near real-time analysis protocol, participants will engage in interactive exercises analyzing pitch track data to confirm the presence of fin, sei, humpback, and North Atlantic right whales. Participants will have access to reference guides for the desktop LFDCS and near real-time analysis protocol. The real-time protocol is currently being used to analyze data from past and active deployments of autonomous platforms that are uploaded to a publicly accessible website, Robots4Whales (robots4whales.whoi.edu). The uses of the near real-time system include monitoring shipping lanes, fishing grounds, wind energy construction areas, migratory hotspots, and aiding visual surveys. Detections confirmed by human analysis help to improve conservation efforts by providing scientists, industries, and the public with near real-time information on whale presence.
Pre-requisites: Each participant should bring their own Mac laptop/charger, hard drive with acoustic data, and have IDL and the LFDCS program downloaded onto their laptop. For specific instructions on how to download LFDCS prior to the meeting, participants can email Genevieve.Davis <at> noaa.gov. Participants would benefit from reading the following paper, which describes the LFDCS, prior to the tutorial: Baumgartner, M.F. and S.E. Mussoline. 2011. A generalized baleen whale call detection and classification system. Journal of the Acoustical Society of America 129:2889-2902 (available at robots4whales.whoi.edu).
Contact: Julianne Wilder- julianne.wilder <at> noaa.gov
Half-day – AM (Supporting virtual participation)
Organizer: Doug Gillespie- University of St. Andrews
Location: 6th floor Ballroom 2
PAMGuard software is widely used for DCL tasks for a wide variety of species and applications. Rather than go over training material which people have seen before and is available online, we will hold a ‘clinic’. This has some elements of a standard tutorial, but the participants set the agenda by bringing their own (PAM related) problems and aspirations. These topics will then be discussed by the entire group, combining expert input from a lead PAMGuard developers with the real-world experience of other PAMGuard users. With luck, we’ll solve some problems, help those planning future research, and share ideas for processing the mountains of data already collected. Of course, sometimes we come up against the limitations of what’s possible with any software and we can discuss that too. As well as helping you with your research, I’m hopeful that the discussions will lead to ideas for future PAMGuard developments. Whatever happens, I’m sure we’ll all come away wiser. The workshop is open to anyone with an interest in PAM processing, but will be most suitable for people who are already familiar with PAMGuard. Problems and discussion topics should be emailed in advance to Douglas Gillespie firstname.lastname@example.org, so that the PAMGuard team can prepare material and we may ask the participant to help prepare a couple of slides or share some data.
Pre-requisites: Familiarity with PAMGuard. Not everyone is required to contribute problems – Passive listening is allowed! Bring a laptop if you want to show your own data and analysis setup.
Contact: Doug Gillespie- dg50 <at> st-andrews.ac.uk
Deep learning acoustic detection and classification with Ketos Half-day – PM (Sorry, not supporting virtual participation) Organizers: Fabio Frazao, Sadman Sakib, Bruno Padovese, Paul Nguyen Hong Duc, Oliver Kirsebom- MERIDIAN, Dalhousie University
Location: 3rd floor Stingray room
In this workshop you will be introduced to Ketos, an open-source Python package for developing deep learning models for sound detection and classification. Using publicly available acoustic datasets compiled by the HALLO project, we will be training deep neural networks at detecting the vocalizations of killer whales and classify some of their stereotypical calls. We have created the workshop with two recipients in mind, practitioners and developers. The typical practitioner is a marine bioacoustician who uses detection and classification (DC) systems to analyze hydrophone data, but has little experience with machine learning. The typical developer, on the other hand, is a machine-learning expert who develops DC systems for use in marine bioacoustics. We expect you to have some previous programming experience, preferably in Python, which is the programming language used in the workshop. You do not need to be an expert programmer, but familiarity with basic programming concepts such as functions, loops, if statements, etc. would be an advantage. The workshop will cover the basic steps of loading and running a pre-trained deep learning model, as well as the more complex tasks of building training and test sets, training a deep learning model from scratch, or adapting it to a new acoustic environment. We will also demonstrate how a trained Ketos model can be run within the popular PAMGuard software. We hope the workshop will be of interest to both practitioners and developers. Pre-requisites: Participants should have programming experience, preferably in Python, the programming language used in the workshop. You do not need to be an expert programmer, but familiarity with basic programming concepts such as functions, loops, if statements, etc. would be an advantage. Participants must bring their own laptop. They should attempt to install Ketos and run the provided test scripts ahead of the workshop (following guidelines provided by the organizers). Participants should also download training and test datasets provided by the organizers ahead of the workshop.
Contact: Oliver Kirsebom- oliver.kirsebom <at> dal.ca
Passive Acoustic Monitoring Density Estimation: raising AWARENESS of latest developments and discussing practical challenges.
Full day (Supporting virtual participation)
Organizers: Danielle Harris, Tiago Marques, Len Thomas- University of St Andrews, Luis Manual Matias, Andreia Pereira- University of Lisbon, Miriam Romagosa, Mónica Almeida e Silva- University of the Azores
Location: 3rd floor Stingray room
Density estimation from passive acoustic monitoring data (PAM DE) is a relatively new, but rapidly developing field. This workshop will give participants the opportunity to (1) hear the latest in PAM DE developments (2) discuss issues and opportunities relating to PAM DE using motivating case studies provided by the workshop organizers and participants and (3) have extended time to ask their own questions. We plan to poll potential participants to offer potential case studies for discussion but we will also use the AWARENESS project as a main case study. The project AWARENESS was aimed at assessing fin whale vocal behaviour for robust PAM DE. Results from this project showed that fin whale vocalisation rates can change abruptly in just a few years, as well as gradually over a couple of decades, and be density- and food-dependent; that call source levels, which affect detectability, vary seasonally; and that variability in source levels estimated from different instruments may affect call localisation and tracking. Findings from AWARENESS and similar studies highlight the huge variability in vocalization production rates and acoustic properties of several marine mammal species, inevitably leading to the questions: How can we account for the range of variability in marine mammal vocal behaviour in PAM DE analyses? To what extent does this variability threaten the reliability of PAM results to inform conservation and management efforts? In this workshop, we hope to stimulate discussion around these and other questions related to the practical implementation of PAM DE methods and if the momentum arises, we could consider writing a review paper on these findings.
PLEASE NOTE: At previous DCLDE workshops, we have held introductory tutorials ahead of the main workshop to lay out the fundamentals of PAM DE. For this tutorial, we plan to make the introductory slides freely available ahead of the tutorial for any interested participants.
Pre-requisites: Participants should be familiar with the introductory PAM DE material, which the organizers will make available online in advance of the tutorial.
Contact: Danielle Harris- dh17 <at> st-andrews.ac.uk or Adreia Pereira- afpereira <at> fc.ul.pt