IEEE Geoscience and Remote Sensing Magazine - June 2023 - 20

commonly exploit chlorophyll concentration estimation, as
this parameter can be determined from HSIs [140].
Water quality monitoring is most often carried out by
ONBOARD PROCESSING
THAT COULD SELECT
RELEVANT INFORMATION AT
THE SENSOR LEVEL MAY
OFFER EXTENDED
MONITORING CAPACITIES
BEYOND THE INITIAL
MISSION PERIMETER.
determining parameters, such as temperature and acidity
as well as chlorophyll a [141]. An increase in temperature
promotes the growth of algae, which translates into an
increase in the content of the phytoplankton biomass in
the water [92]. This is often indicative of an uncontrolled
growth of HABs [139]. In recent work, Sahay et al. showed
that it is possible to estimate chromophoric dissolved organic
matter, being the fraction of dissolved organic matter
that absorbs sunlight in the ultraviolet and visible region
of electromagnetic radiation, from remote sensing reflectance
in coastal waters of India [142]. The authors showed
that seasonal and spatial variability in the investigated area
allows their algorithm to retrieve the chromophoric dissolved
organic matter absorption in coastal areas by using
high-resolution ocean color monitors, such as Sentinel-3,
but also from HSIs [143]. In [144], Cherukuru et al. focused
on estimating the dissolved organic carbon concentration
in turbid coastal waters by using optical remote sensing observations.
Organic carbon is
a major component of dissolved
organic matter in the
aquatic system, and it plays
a critical role in the marine
carbon cycle. Overall, datadriven
HSI analysis may be
used for water environment
monitoring and to understand
its dynamics, leading
to better understanding of
the underlying biogeochemical
processes at a larger scale.
Ocean monitoring is also
tackled within NASA's aquatic Plankton, Aerosol, Cloud,
and Ocean Ecosystem mission carrying the Ocean Color
Instrument, which will be capable of measuring the color
of the ocean, from ultraviolet to SWIR [145].
Recently, Caribbean coasts have experienced atypical
arrivals of pelagic Sargassum, with negative consequences
both ecologically and economically [146]. Removing
Sargassum before its arrival, thanks to early detection, could
reduce the damage it causes [147]. It is known that floating
mats of vegetation alter the spectral properties of the water
surface; hence, deep learning-powered exploitation of HSIs
has been investigated for such tasks [148].
MARITIME SURVEILLANCE
Maritime surveillance aims at maintaining the security
of waters by monitoring traffic [149] and fishing [150] as
well as the elimination of smuggling [151], illegal fishing
[152], and pollution [153]. Remote sensing can help
to locate ships and verify their location with automatic
identification systems [150], allowing for inferring the legality
of a vessel's movement at scale [154]. Ship detection
20
techniques based on signal processing [155] and deep learning
approaches [156] are often based on SAR and MSIs,
with the former unaffected by clouds and fog [157]. Ships
are characterized by various sizes and shapes; thus, the
appropriate spatial image resolution [158] is pivotal to detect
them [159]. Also, fusing SAR and MSI data enables
locating vessels [158], whose positions can be tracked
[160]. Object detection based on satellite images is an important
element supporting the search for lost ships and
planes [161]. This task is important in both civil and military
matters for safety reasons as well as for potentially
quick assistance in the case of accidents [162]. The level
of difficulty of detecting planes and ships depends on the
background and their size [159]; hence, the spatial resolution
of the imagery is important [158]. Although remotely
sensed images allow identifying such objects at a global
scale, they are also challenging due to the lack of homogeneity
of the background [161].
Illegal fishing is a threat to coastal and marine ecosystems
as well as the economy [152]. Unregulated fish catches
reduce the stocks of fisheries and the lack of reporting
makes it impossible to monitor fisheries, which poses a
threat to fish species and leads to economic effects [150].
The detection of illegal fishing is commonly built upon the
detection of ships and monitoring their trajectories by using,
e.g., deep learning techniques over MSIs [152]. The Integrated
System for the Surveillance of Illegal, Unlicensed,
and Unreported Fishing is an example of a working system
that exploits SAR (Sentinel-1) and MSI (Sentinel-2) data for
this task [150].
OBJECTIVE AND QUANTIFIABLE SELECTION
OF ONBOARD AI APPLICATIONS
Implementing AI onboard an EO satellite is not a goal in
itself, and it must produce significant benefits when compared
to what on-ground data processing may provide. For
a space project with cost and planning constraints, the answer
is not obvious and must take into account different
aspects by considering a wide range of criteria to estimate
how much a solution complies with the satellite, system,
and mission constraints as well as what benefits may result.
The analysis of pros and cons must demonstrate that
having AI onboard an EO satellite is the best option to provide
operational added value for the final user in terms of
performance, timeliness/latency, autonomy, and extended
capacities as well as from an engineering, industrial, operational,
commercial, and scientific points of view. Additionally,
selecting appropriate onboard applications can
impact society at large through, e.g., the implementation
of sustainable development goals and climate change adaptation
and mitigation.
At the end-to-end level, onboard processing may improve
overall system reactivity by sending alerts upon the
detection of transient and short-duration phenomena,
thus providing rapid responses to events that require fast
decision making that is incompatible with a standard
IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE JUNE 2023
IEEE Geoscience and Remote Sensing Magazine - June 2023

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - June 2023
