June 2025 | Oceanography
49
An important visualization is the tri-level spectrogram, which
presents the acoustic recordings from a lander in an easy-to-use
exploratory interface (Figure 7b). Using a color scale perceptu
ally optimized to highlight marine mammal sounds, spectro
grams are pre-processed into image files sufficiently compressed
to be loaded faster than users can scroll, allowing for seamless
exploration. The top level of the spectrogram viewer displays
weeks to months of audio (depending on monitor resolution)
and allows users to quickly peruse the entire dataset, see trends,
and spot major events. The middle level shows roughly a day of
spectrogram data, while the bottom level shows a few minutes
at full resolution. The levels are linked, so clicking on one level
centers the other levels around the same time. Selections can be
made in the lower-level view, allowing in-browser playback or
download of sound files from specific time ranges, with options
to select and filter by frequency.
An event viewer presents marine mammal detections in
an interactive heatmap (Figure 7c). Users select an event type
(e.g., dolphin click) and view a plot of detected events over the
entire project duration. Alternatively, all years can be stacked
to produce a cyclic visualization that reveals repeated seasonal
patterns, with an option to interactively emphasize contribu
tions from each year. The heatmap can be shifted in direction to
center patterns. Clicking on individual heatmap cells switches
over to the spectrogram viewer, which jumps to the correlating
timestamp. Additional context is provided via a day/night indi
cator band and environmental data plots (e.g., chlorophyll). A
second lander can be selected to perform direct comparisons
within a single heatmap (using multiple colors).
Finally, the deviations viewer presents a similar tri-level
interface, but instead of spectrograms, it displays times and fre
quency ranges in which the soundscape was unusually loud or
quiet, based on a weekly, monthly, or quarterly moving win
dow analysis (Figure 7d). For the data displayed in the viewer,
recordings were processed into 60-second, decidecade fre
quency bins. Running means and standard deviations were cal
culated for each window length, and the number of standard
deviations above or below the running mean was mapped to a
diverging blue-white-red heatmap. See Butkiewicz et al. (2021)
for additional details. Since project completion, the visualiza
tion interface has been successfully used by the public as indi
cated by the project webpage visitor log, and it provides a valu
able tool to other researchers for studying a wide range of topics
from marine mammal behavior to extracting training data for
AI/ML detection applications.
SUMMARY
The ADEON team designed and deployed an ocean acous
tics observation network on the US OCS between Virginia
and Florida from November 2017 to December 2020. The
10
20
30
40
50
11/11/17
12/11/17
1/10/18
2/9/18
3/11/18
4/10/18
5/10/18
6/9/18
7/9/18
8/8/18
9/7/18
10/7/18
11/6/18
12/6/18
NASC (m2 nmi–2)
HAT NRS Med (125 kHz)
HAT NRS Large (125 kHz)
HAT RS Small (38 kHz)
HAT RS Large (38 kHz)
DISTRIBUTIONAL AND
BEHAVIORAL DATA
Presence/absence of prey from active
acoustics and predator from passive
acoustics and sightings
OCEANOGRAPHIC DATA
Sea surface temperature and
chlorophyll from satellite measurements
ENSEMBLE MODELS
Generalized additive
mixed models,
boosted regression
trees, Bayesian
approaches
SPATIAL AND
TEMPORAL PREDICTIONS
Probability of occurrence of mid-trophics
and top predators, time series of
abundance of mid-trophics and
top predators
FIGURE 6. (a) Ecosystem modeling framework. (b) Example of a daily spatial prediction of relative fin whale
call density on September 7, 2018, using preliminary fitted relationships from a generalized additive model.
(c) Acoustic backscatter can be apportioned to different taxonomic or size classes of scatterers (i.e., NRS
and RS, non-resonant and resonant scatterers respectively; medium NRS at 125 kHz [10–25 mm], large NRS
at 125 kHz [25–122 mm], small RS at 38 kHz indicative of small swim-bladdered fish, and large RS at 38 kHz
indicative of larger swim-bladdered fish based on animal total length; Miksis-Olds et al., 2021). These param
eters (representing the abundance or biomass of different types of zooplankton or fish) can then be used as
model input parameters to determine relationships between prey abundance and marine mammal predator
presence or vocal behavior. These are the time series of size classes from the HAT lander.
74°W
40°N
38°N
36°N
34°N
32°N
30°N
28°N
82°W
78°W
80°W
76°W
Fin
Activity
10