Oceanography | Vol. 38, No. 2
46
centered at the seven ADEON bottom lander sites during a
three-year period. Volume backscatter data were gridded both
horizontally (100 m) and vertically (5 m; Figure 4c) to produce
variogram range estimates, the distance over which data are spa
tially autocorrelated, providing a proxy for scatterer patch size
and representative distance (Legendre and Fortin, 1989). Patch
horizontal lengths were consistently 2–4 km among the seven
ADEON locations (Blair et al., 2021).
A second study compared the spatial and temporal autocor
relations of vessel survey and stationary backscatter data using
two approaches. First, virtual backscatter transects were cre
ated by advecting stationary echosounder data using measured
current velocities from the vessel-mounted acoustic Doppler
current profiler during the FSASs at each site. This was done
during the same night an FSAS occurred, so spatial autocorrela
tion could be estimated for both data types. Next, the tempo
ral autocorrelation of the two-week-long time series of hourly
backscatter (Figure 4d,e) centered in time on each applicable
FSAS date for three sites (VAC, HAT, and JAX) was converted
into a distance estimate to compare with the FSAS variogram
ranges (Figure 4e). This methodology allowed for longer time
periods (up to two weeks instead of 12 hours) to be analyzed
and for associated autocorrelation patterns to be detected. The
resulting autocorrelation distances from the stationary systems
(0.8–3.4 km) were similar to those (1.3–3.8 km) from vertically
integrated FSAS data from the same three sites (Blair, 2023).
The spatial characteristics of epi- and mesopelagic scattering
layers are rarely measured in the horizontal dimension, yet they
are imperative information for the design and implementation
of monitoring and management programs for pelagic ecosys
tems (Horne and Jacques, 2018). These findings demonstrate the
importance of considering scale when designing active acoustics
monitoring networks and sampling protocols. Comparing scales
of space and time in the dynamic ocean is a nontrivial task, and
it remains unknown whether the characteristics measured along
the US eastern continental shelf are representative of shelf-slope
environments in other regions.
Acoustic Propagation Modeling
Soundscape modeling is among a number of considerations used
for policy decisions related to ocean sound. It is important to
know the performance accuracy of soundscape models (Heaney
et al., 2024), and measurements from the ADEON project are
extremely valuable for this purpose. For the acoustic modeling
component of ADEON, a wind and shipping soundscape model
was developed for the Atlantic OCS. This permitted evaluation of
the spatial and temporal distributions of the soundscape beyond
the data collected at the lander locations. Acoustic propagation in
the ocean is sensitive to temperature and salinity fields, bathyme
try, seafloor sediment type, and sea surface roughness (a function
of wind speed) (Jensen et al., 2011). The soundscape modeling
approach consisted of three steps: (1) identify the distributions of
sources contributing to sound in the region and collect the rele
vant environmental information, (2) compute the acoustic prop
agation loss from all sources to all receiver positions, and (3) sum
the contributions and compute the SPL.
The regional SPL was computed for the years 2018, 2019,
and 2020 for decidecade bands centered at 20, 50, 100, 200, and
400 Hz. A single snapshot and a monthly average of the SPL
for 50 Hz at the seafloor is shown in Figure 5 panels a and b,
respectively. The temporal observation window was three hours
for 2018 and 2020 and 10 minutes for 2019. The 2019 model
was generated first, and the 10-minute temporal observation
window proved computationally expensive with an extensive
storage requirement; thus, the observation windows for 2018
and 2020 were expanded to three hours. This massive model
ing product dataset is served to the public on the ADEON web
site (https://ADEON.unh.edu) as explained in the visualiza
tion section below. One observation of this modeling study was
that the SPL on the seafloor was often 3 dB higher than that at
10 m depth, due to the downward refraction of shipping sound
(Heaney et al., 2024).
The wind and shipping sound levels for each of the lander
positions were computed with a higher resolution time obser
vation window of five minutes. Sediment uncertainty, oceano
graphic variability, and shipping source depth and level uncer
tainties were incorporated using a Monte Carlo framework. The
sediment uncertainty drives the modeled SPL, permitting an
estimate of the local sediment characteristics when compared
with the observed data. Figure 5c shows the modeled 125 Hz
decidecade band SPL (5th, 50th, and 95th percentiles) along
with the measurements for the WIL site for the first week of
January 2019. The percentiles relate to the weekly mean SPL dis
tribution across the sediment types. The data match the 5th per
centile model across the ensemble with only a few passing ships
above the wind noise floor. The comparison of the SPLs using
the best sediment value for BLE (sediment grain size parame
ter, phi = 5.68) is shown in Figure 5d. The short time duration
peaks are nearby passing surface ships, and the slowly varying
low SPL regions are wind levels. The differences between the
two sites can be attributed to the number of passing ships and
the sediment (WIL having more ships and sediment with higher
acoustic impedance, and BLE having both fewer ships and lower
impedance sediment).
Ecosystem Modeling
The ADEON ecological modeling component focused first on
describing the temporal abundance patterns of marine mam
mals across the entire study region (Figure 6a). This informa
tion was then used to quantify the variability in marine mammal
distribution via call density as it related to changing oceano
graphic conditions. Both the diversity in calling marine mam
mals as well as the species-specific detection rates were analyzed
concurrent with the lander and remotely sensed oceanographic