The Atlantic Deepwater Ecosystem Observatory Network (ADEON) is an array of hydrophones deployed along the U.S. Mid- and South Atlantic Outer Continental Shelf, that collected four years of acoustic measurements of natural and human sounds in the region from 2017 to 2020. This paper focuses on the design and implementation of a web-based geospatial and acoustic visualization interface that allows anyone to easily explore the various massive datasets generated by the project in order to gain insight about the ecology and soundscape of the region.

The massive size of the ADEON datasets presents many accessibility and visualization issues. There are seven physical landers deployed in the network, each with multiple directional and omnidirectional hydrophones that record various frequency ranges. These hydrophones generated 73 terabytes of audio files. It is difficult for researchers to download all of this data, due to both bandwidth and physical data storage limitations, and finding what one is looking for in the files can be just as difficult, due to the overwhelming number and length of the recordings.

The recordings were also further processed to produce additional datasets, such as event detections based on different filters (e.g. dolphin clicks or seismic surveys). The project also had a soundscape modelling component that generated 5D (lat/long/depth/frequency/time) datasets of predicted sound energy levels for noise sources such as ships and surface winds. This visualization project sought to build a single, cohesive visual interface for researchers both within and outside of the ADEON project to easily explore the acoustic data and the derived datasets, without having to download and store the data or install any particular software.

Visitors to the public ADEON website can access an interactive map of the ADEON project region, which was built using the Leaflet JavaScript mapping interface. The map displays the locations of the hydrophone deployments, layers of contextual data (e.g. sea surface temperature and chlorophyll levels), and model derived soundscapes for various sources. Selecting individual deployment stations brings up linked visualization interfaces that provide multiple ways to explore the data associated with that station:

A tri-level spectrogram viewer presents linked spectrograms at three different time-scales, which enables rapid exploration of multiple years of raw recordings. The top level shows a few weeks on screen at a time, the middle--a few hours, and the bottom--a few minutes. Users can quickly scroll through months of data in the top level until something catches their eye. Then, they can click on a region of interest to center the other two levels on that time. Mousing over the bottom level shows the exact time and frequency at that point in the spectrogram. Selecting a rectangular region of interest within the bottom spectrogram allows the user to playback or download an audio clip of that particular time, and the frequency range can be filtered to remove distracting noises. The color map used in the spectrograms has been perceptually optimized to best reveal salient features (e.g. marine mammal noises), based on previous experimental research on designing for the human visual system.

Another tab contains a similar tri-level viewer that presents calculated deviations in sound levels for different frequency bins over time. These plots use a diverging blue-white-red color map to reveal times when the ocean around a station was unusually loud (or quiet) at various frequencies. The moving window size used to calculate the deviations can be switched from weekly, monthly, or quarterly in order to suppress or highlight various factors (e.g. temperature or sensor drift).

Finally, another tab presents detected events using a heat map visualization. Recordings were run through various detection filters to find instances of marine mammal noises, shipping tones, seismic surveys, etc. The plot uses color intensity to indicate the number of detections within each one-hour cell. The plot is 24 cells tall (i.e. each column of cells represents a single day) and can be toggled between either ‘continuous’ mode, in which the plot expands horizontally to fill the screen and displays all records in temporal order, or  ‘cyclic’ mode, which stacks each year of data on top of the other in a single plot 365 cells wide. The cyclic mode reveals patterns which repeat each year, and an ‘emphasis’ feature allows you to mouse over years labels to indicate the contribution of each year to the overall plot.

The plot can be adjusted vertically and horizontally to change when wrapping occurs, i.e., a pattern straddling the arbitrary midnight or December 31st/January 1st wrapping bounds would appear disconnected, but could appear whole if those bounds were adjust to noon and July 1st.

Contextual data bars at the top and bottom of the plot show relationships between the patterns in the plot and surrounding conditions such as temperature and chlorophyll, helping to reveal animal migration patterns and other interdependent relationships.

This suite of web-based visualization tools allows researchers, managers, and regulators to gain insight from the massive ADEON dataset and can help inform future, more targeted studies into the impacts of marine noise.