Early Online Release | Oceanography
of platforms, humans can no longer keep pace with the demand
for video annotation and the ancillary data that come with it.
Machine learning is now playing a central role in processing that
information. At the time of this writing, the VARS annotation
pipeline has been improved by using computer models trained
on approximately 900,000 localizations of over 1,600 expertly
curated concepts to assist with image annotation and identifica
tion (Figure 6; VARS-ML). In an effort to federate and coordi
nate this line of research, FathomNet offers a publicly accessible
platform for sharing images and accessing artificial intelligence
and machine learning tools to accelerate the analysis of ocean
visual data (Katija et al., 2022; Crosby et al., 2023). A compan
ion program, FathomVerse, a free mobile game, offers an inter
active science community experience where players engage
with real ocean images collected by researchers and robots from
around the world. Participants who play the game contribute
to improving computer algorithms used to chronicle ocean life
while learning about the animals they see, which is proving to be a
technologically novel way to expand participation in ocean explo
ration and discovery.
Machine learning and artificial intelligence can also be used
aboard remotely operated and autonomous platforms to process
visual and other sensory data in real time. Without any human
intervention, vehicles can adapt to dynamic environmental con
ditions by leveraging physical, chemical, and biological cues that
enable them to track marine life over extended periods (e.g., Zhang
et al., 2021a, 2021b) and navigate complex terrain in the absence of
detailed maps (e.g., Troni et al., in press a, in press b). The power
and potential of machine learning and artificial intelligence is only
beginning to alter our ability to observe the ocean holistically.
In the years ahead, this area of rapid innovation will undoubt
edly transform data acquisition, analysis, and dissemination both
ashore and at sea. This technology is also an effective means for
engaging the next generation of ocean enthusiasts. Robots super
charged with artificial intelligence offer something for everyone.
Whether it is the science they enable, the imagery they produce,
the computational capability that makes them “smart,” the mis
sions they undertake, or just the impressiveness of the machines
themselves, people are simply fascinated by robots.
THE BIOLOGICAL CARBON PUMP
AND VERTICAL MIGRATION
World Wars I and II sparked a revolution in ocean engineering.
Submarines were proving to be very effective at sinking combatants
and ships carrying supplies to aid the war effort, and a technological
advance was needed to detect and intercept them. Sonar (SOund
Navigation And Ranging) offered an answer while also providing
a way to gauge the depth of the seafloor. As the technology was
refined, a reflective layer was sometimes detected in the water col
umn that could be so dense it gave a false sense of the actual depth
of the seafloor, even to the extent that ships traveling in uncharted
waters reported the presence of phantom shoals. Stranger still, that
feature was usually observed to move in rhythm with the time of
day, rising at night and descending during the day. The deep scat
tering layer (DSL), as it came be known, was later associated with
dense aggregates of animals (e.g., Ritche, 1953; Dietz, 1962).
The advent of sonar had revealed something amazing: diel verti
cal migration. Animals who spent daylight hours in the twilight of
the deep rose at night to feed, and drew organic carbon with them
when descending back to the depths during the day. This behavior
accelerates the transport of carbon from surface to deep waters—a
phenomenon known as the biological pump—contributing to the
ocean’s role in modulating climate while also providing food for
animals and microbes throughout the water column and on the
FIGURE 5. Time-series obser
vations document the displace
ment of several midwater animals
toward the surface in response
to a shoaling oxygen minimum
zone (after Robison et al., 2017).
Hake and Chiroteuthis images ©
2025 MBARI; Tomopterid image
Rob Sherlock © 2007 MBARI