Coronavirus Information for the UC San Diego Community

Our leaders are working closely with federal and state officials to ensure your ongoing safety at the university. Stay up to date with the latest developments. Learn more.

AOS Seminar: Dr. Kaitlin Frasier, "Big data in a context vacuum: Learning more from offshore passive acoustics"

10/31/2019 - 4:00pm to 5:00pm
330 Spiess Hall
Event Description: 

Dr. Kaitlin Frasier

Scripps Institution of Oceanography

OA Section Research Candidate Interview Seminar


Abstract. Passive acoustic monitoring (PAM) is a rich, prolific, versatile oceanographic data collection method, however the data are in some ways underutilized due to their volume and complexity. Data accumulation rates for current broadband (10 Hz to 200+ kHz) PAM sensors exceed 1 terabyte/hydrophone/month and low power hardware can record autonomously for over a year between service intervals. The very large datasets produced document a broad range of physical, biological and anthropogenic sound sources, but contextual information is typically lacking. PAM sensors are often deployed in deep, remote and/or otherwise difficult to monitor areas, and the vast majority of datasets collected are “unlabeled”, i.e. little to no associated information beyond the recorded sounds themselves are available. Popular “big data” analysis strategies rely on the availability of high quality labeled datasets to train and evaluate algorithms such as those appropriate for acoustic detection, classification and localization. Novel approaches are needed for effective use and interpretation of these large PAM datasets without knowledge of source characteristics or local subsurface oceanographic conditions. 

In this talk, automated methods for extracting insights from large underwater passive acoustic datasets are considered. Strategies include unsupervised and semi-supervised clustering of acoustic signals, statistical simulations of acoustic sources, cross-observational data synthesis, and pattern recognition through deep learning. These types of computational methods are beginning to expand PAM applications by facilitating quantitative offshore ocean observations with relatively simple sensor configurations, as well as efficient analysis of large recording datasets. Hydrophones are increasingly incorporated into ocean monitoring platforms, from autonomous underwater vehicles and floats, to large scale ocean observatories, creating new opportunities to advance underwater acoustic research. 


For more information on this event, contact: 
Art Miller
Event Calendar: 
Applied Ocean Science