Content uploaded by Freya M. E. Muir
Author content
All content in this area was uploaded by Freya M. E. Muir on Sep 02, 2022
Content may be subject to copyright.
73rd International Astronautical Congress (IAC), Paris, France, 18–22 September 2022.
Copyright ©2022 by Mr. Joe Gibbs. Published by the IAF, with permission and released to the IAF to publish in all forms
IAC-22-B4.IP.56
Mapping Coastal Dynamism from Space:
A Small Satellite Application for the Onboard Automatic Extraction of Coastal Boundaries
Freya M. E. Muir a, Joe Gibbs b, Nektarios Chari c, Georgios Titas d
aSchool of Geographical and Earth Science, University of Glasgow, f.muir.1@research.gla.ac.uk
bJames Watt School of Engineering, University of Glasgow, j.gibbs.1@research.gla.ac.uk
cJames Watt School of Engineering, University of Glasgow, 2310543c@student.gla.ac.uk
dJames Watt School of Engineering, University of Glasgow, 2331546t@student.gla.ac.uk
Abstract
The coast is a highly dynamic environment with complex interacting processes that are being impacted
significantly by climate change. An estimated two fifths of the world’s population live on or near the coast,
putting key transport infrastructure, businesses and communities at risk of increased erosion and flooding as sea
levels continue to rise. To redirect assistance and focus planning and adaptation strategies to those most at risk,
standardised and regular observations of how coasts are changing needs to be available to coastal managers
and policymakers. Due to inherent dynamism, the coast is logistically difficult and costly to survey on the
ground. Earth Observation has opened new opportunities for gathering repeated measurements of shoreline
positions and coastal vegetation boundaries to infer morphological changes and damages to coastal assets
from flooding and erosion. Many methodologies now exist to extract these coastal boundaries from freely
available satellite-based optical and synthetic aperture radar sensors, but there is no dedicated coastal monitor
which means current lead-times on obtaining useful data are long and data must be extracted manually with
each new image. We present the preliminary system architecture of a 3U CubeSat platform and full data
product for a coastal observation mission. Novel onboard processing routines regularly image the coast from
a low-Earth orbit, pre-process images, use spectral band indices to extract shoreline positions and coastal
vegetation edges and downlink this vector information to be used rapidly by stakeholders. By shortening the
lead-time significantly, a faster response to storm event risk and damage assessment can be obtained and a
more holistic approach to analysing coastal change can be taken. The data products OirthirSAT will provide on
a weekly basis can be easily visualised for a comprehensible understanding of coastal change, and fed into
modelled predictions of future coastal change.
Nomenclature
𝑮 𝑩 Gigabyte
𝑮𝑯𝒛 Gigahertz
𝑴 𝑩 Megabyte
𝑴𝑯𝒛 Megahertz
𝑴𝒃 𝒑𝒔 Megabits per second
𝑼CubeSat unit of 10 x 10 x 10 cm
𝑾Watt
𝒎𝑾 Ampere
𝒎𝑾 Megawatt
1. Introduction
Nanosatellite technology has grown increasingly accessi-
ble to small startup enterprises and student groups over the
past few years. This has seen an increase in the number of
university groups developing novel missions for CubeSats
and the growth of competitions to grant funding to student
bodies [1]. The UK Space Agency’s LaunchUK Nanosat
Design Competition is one such programme, aimed at devel-
oping young people’s engineering and project management
skills. Teams across the UK are empowered to design and
develop small satellite missions aligned with climate change
mitigation and adaptation. OirthirSAT is the winning entry
from the 2022 LaunchUK competition, proposed by a team
from the University of Glasgow to deliver repeatable, regular
data for monitoring coastal change across the UK.
1.1. Monitoring Coastal Change
Coastal environments are incredibly dynamic regions
with a complex interplay of processes related to waves,
IAC-22-B4.IP.56 Page 1 of 14
73rd International Astronautical Congress (IAC), Paris, France, 18–22 September 2022.
Copyright ©2022 by Mr. Joe Gibbs. Published by the IAF, with permission and released to the IAF to publish in all forms
Fig. 1 A preliminary render of the OirthirSAT nanosatellite
in orbit.
sediment and ecology taking place. Despite their dynamism,
an estimated 40% of the world’s population live within
100 km of the coast and below 50 m elevation [2]. The
Intergovernmental Panel on Climate Change (IPCC) has
identified that recent rates of global mean sea level rise have
almost doubled to 3 mm yr
-1
, compared to 20
th
century rates.
Even after achieving the strictest emissions targets, sea levels
are predicted to continue rising over the next century, from
0.29–0.59 m under a low emissions pathway and 0.61–1.10
m under the highest emissions pathway [3]. The increase in
global sea level is causing a range of significant impacts to
coastal environments and the communities that reside there.
Low-lying coastal regions are being permanently submerged
under higher mean tides, coastal flooding is increasing in
intensity and frequency, rates and extents of erosion are
increasing and ecosystems are changing or failing to survive
under the rapid environmental shifts [4, 5]. These changes
both directly and indirectly affect coastal populations, leading
to loss of life, damage and destruction of buildings and utilities
infrastructure, widespread economic losses, displacement of
settlements and loss of indigenous knowledge [6, 7]. While
global mean observations indicate an overall rise in sea
level, different coasts may experience different amounts of
sea level rise and therefore respond differently to climate
change-driven forces. This spatial variation in the physical
and ecological impacts of climate change are also reflected
in how different communities respond to these risks and
impacts. Populations with higher levels of deprivation, less
space to accommodate relocation of infrastructure, or less
economic privilege to invest in coastal change mitigation and
adaptation tactics will be more adversely affected [8].
With this variation in the environmental and social im-
pacts of coastal change, comes variation in how to manage
these impacts. Coastal managers require up-to-date informa-
tion on how coasts have been changing, and may change in
the future in order to properly inform management decisions.
In England and Wales, there is a recommendation for local
governments to develop Shoreline Management Plans, with
detailed histories of change which guide a series of options
and limitations for future developments and investments [9].
These coastal managers often work with coastal scientists
who perform monitoring and modelling of selected coastal
regions to estimate the environmental response to sea level
rise, erosion and flooding. Simplistic measures of change
have been found in analysing line-based features such as
the water’s edge and the most seaward boundary of vege-
tation. By measuring lateral changes in the shoreline over
time, information on changes to both beach morphology
and water level can be inferred [10]. Traditional techniques
to collect shoreline data to inform this work has however
been costly, logistically difficult and spatially and temporally
limited. Approaches such as digitisation of historical maps
and aerial imagery, aerial photogrammetry and LIDAR, and
ground-based GNSS surveying, all require significant manual
effort and cost, and due to this are restricted to single beach
systems and irregular surveying schedules [10, 11]. While
great strides have been made in using the past shoreline his-
tory to inform future change [12], the UK still has a spatially
and temporally gap-filled record of coastal change and no na-
tionalised or centralised way of distributing this information
and analysis, leading to inefficiency in coastal management
[13]. An example of the potential and drawbacks of simplis-
tic coastal change indicators is seen in Scotland’s Dynamic
Coast project, which made use of the rates of change between
past shoreline positions to inform a national model of coastal
retreat under different sea level rise scenarios. The work
produced a national understanding of coastal change histories
and predicting up to £1.2B of damage to coastal infrastructure
by 2050 under a high emissions scenario [14]. However, the
amount of retreat or advance predicted was highly sensitive
to the rates of change, which for some sites were informed by
shoreline observations from the 1970s with positional errors
of up to 10 m [15]. The requirement for national, regular,
rapid and simplistic measures of coastal change to inform
climate change response along our coasts is therefore evident.
Recent partial solutions to these requirements have been
found in satellite-based Earth Observation data. The current
IAC-22-B4.IP.56 Page 2 of 14
73rd International Astronautical Congress (IAC), Paris, France, 18–22 September 2022.
Copyright ©2022 by Mr. Joe Gibbs. Published by the IAF, with permission and released to the IAF to publish in all forms
constellation of moderate resolution imagers providing peo-
ple with freely available regular observations of the Earth’s
surface has radically improved the spatial and temporal cover-
age of coastal data. Dedicated coastal algorithms and portals
such as CoastSat, the Luijendijk et al. shoreline dataset, and
the USGS Coastal Change Portal, are making use of freely
available multispectral satellite imagery to extract measures
of coastal change such as shoreline position migration and
elevation change, to infer how coasts have changed in the
past and may be influenced by changes in the future [5, 16,
17]. However, as revolutionary as these resources have been
in extracting coastal change measurements from satellite im-
agery, their approaches remain site-specific and user-driven.
These methods require an individual to both have some level
of expertise in satellite imagery and programming, and an
investment of time and money in setting up a new site for
observation. Satellite images must be downloaded and pro-
cessed individually by the user before useful information on
coastal change can be extracted. Even with the efforts of
the Coastal Channel Observatory in parts of England and
Dynamic Coast in Scotland, the UK still lacks a holistic,
centralised measure of national coastal change that is updated
automatically [13, 18]. The UK Hydrographic Office and
Defra are currently trying to address this issue by putting
together a framework for automatically extracting national
shoreline positions from Sentinel-2 data. This displays the
desire for this kind of simplistic and satellite-derived dataset
for monitoring coastal change [19]. However, it remains
unclear how often or difficult repetitions of the method will
be to obtain updates, or whether any updates are planned for
the purposes of tracking change through time. Meanwhile,
coastal storms and erosion progressively worsen the risk
of damage to properties, as sea levels continue to rise [20].
Freely available satellite imagery from sources such as the
Sentinel and Landsat missions have opened the door to more
rapid assessments of coastal change on national to global
coverage. Nonetheless, a dedicated coastal platform is now
required that cuts down on the lead time between image
capture and coastal decision-makers holding processed data
to perform analyses with.
We present here a design for a 3U CubeSat dedicated
to monitoring changes to the UK coastline. This platform
design aims to present a national standardised approach
to coastal monitoring by automatically processing imagery
onboard and extracting coastal boundaries (between water
and land) to be downlinked to Earth. In doing so, we hope to
achieve a factor of 3 reduction in lead time between image
capture and data delivery (13 days to 4 days), and a factor
of 166 reduction in data downlink cost (1 GB image to 6
MB vector file). With recent advances in AI approaches to
data processing [21], primarily for feature extraction from
images using deep learning techniques, OirthirSAT intend
to reduce data downlink budgets and the associated costs by
processing coastal images in-situ. Artificial neural networks
(ANNs) have been demonstrated to be adept at identifying and
classifying types of terrain using satellite imagery. Recent
work deploying an ANN on shoreline extraction [16] and
holistically nested CNN on coastal vegetation [22], both using
satellite imagery, demonstrates that this approach is feasible
and could be readily applied onboard. The OirthirSAT
mission seeks to demonstrate not only AI techniques for
image segmentation and classification but to exploit the
potential of pre-processed data to reduce data budgets.
1.2. Mission Overview
Nanosatellites including CubeSats have limited link bud-
gets due to design characteristics including limited power
budgets and reduced antenna gains compared to larger coun-
terparts [23]. In this paper, a low-cost mission is proposed to
provide ready-to-use coastal boundary data nationally over a
shorter delivery time frame. The proposal was selected as the
winner of the LaunchUK Nanosat Design Competition and
will proceed through to the detailed design phase to refine
the platform and payload.
1.3. Mission Objectives
The OirthirSAT mission objectives can be refined down
to two key objectives:
1)
Delivery of ready-to-use coastal edges of the UK using
onboard processing.
2) Release of the processed data within 4 days.
To realise these aims, a suitable orbit and platform must be
defined to ensure that the space segment is able to capture not
only the required data, but also to process it in the required
time period to enable fast access to the formatted data.
1.4. Orbit Design
Two key requirements constrain the orbit selection for the
OirthirSAT mission which relate to the quality of the images
and the frequency of data collection respectively. Given the
novelty of performing calibration and processing on-board,
extensive validation is necessary using similar open-source
satellite imagery. The common imagery from Sentinel and
Landsat provide data with ground resolution of 30 m/pixel.
Multispectral imagers for CubeSats with the same capability
exist off-the-shelf such as the MultiScape50 CIS from Simera-
Sense or the Mantis imager from Dragonfly Aerospace which
achieve a ground resolution of approximately 30 m/pixel
at altitudes of 500 km. The current applications of coastal
IAC-22-B4.IP.56 Page 3 of 14
73rd International Astronautical Congress (IAC), Paris, France, 18–22 September 2022.
Copyright ©2022 by Mr. Joe Gibbs. Published by the IAF, with permission and released to the IAF to publish in all forms
monitoring require data weekly given the slow change of
coastal boundaries. To maintain similar applicability and
comparability of the output data, a weekly frequency is also
adopted for the mission. Commercial off-the-shelf (COTS)
CubeSat Imagers have a swath of 90–150 km in altitudes in
Low-Earth-Orbits (LEOs). The UK varies in across-track-
width between 150–550 km making the observation of both
coasts using a single pass impossible. Therefore, at least
two passes over the UK are necessary in a week to allow
observation of both the eastern and western coast. The
orbit was selected from the standard sun-synchronous-orbits
following the process suggested by Boain [24]. Multiple
candidate orbits exists as seen in Table 1. The indexing of the
orbits follows the format suggested by Boain [24] with the
first number corresponding to the revisit time and the second
one corresponding to the number of revolutions between two
consecutive visits. For instance, the orbit 2D31R refers to an
orbit with a revisit period of 2 days, over which the satellite
performs 31 revolutions.
Orbit
2D31R 3D46R 3D47R 1D15R
GSD (m/pixel) 25.0 27.9 22.0 33.9
Altitude (km) 417 466 368 566
Swath (km) 102 114 90.5 139
GT seperation (km) 831 560 548 1717
Maximum ONA (°) 44.9 30.6 36.6 56.5
Table 1 Orbit Selection Matrix.
The orbit 3D46R is selected for the mission as it offers the
most comparable GSD to open-source imagery, acceptable
ONA while allowing observation of both coasts on a weekly
basis. Launch opportunities and associated costs are a critical
factor to low-cost CubeSat projects. Thus, maintaining
flexibility in preliminary stages is key. Therefore, all feasible
candidate orbits are considered.
1.5. Payload Selection
The requirement for a GSD of 30 m and the volume
budget being limited to to a 3U CubeSat structure reduces
the available imaging payloads that could be used for the
mission. Other considerations are the cost of the imager and
the imaging configuration, number of bands and orientation of
the lens assembly as this will impact the configuration of the
platform. The Simera-Sense MultiScape50 was selected, as
it offers multispectral capabilities that fulfill the requirement
for data acquisition in the Blue (410 to 490 nm), Green (490
to 540 nm), Red (650 to 700 nm) and Near Infrared (750 to
900 nm) spectral bands. A fixed lens assembly is preferred
to eliminate motorised mechanisms and potential points of
failure. To fit the 3U volume budget, the Tuna Can CubeSat
form-factor is chosen to house the lens along the -Z axis.
Operating in push-broom configuration, the imager scans
4096 pixels at a time (approximate swath of 114 km) capturing
all spectral bands of interest continuously. Representative
data sizes reach 948 MB per pass-over for 4 bands and 12-
bit resolution configuration. Captured imagery data stored
within the imager payload FLASH storage require subsequent
processing to achieve quick and frequent data downlink to
ground. A dedicated processing and data handling unit is
included to extract the vectors from rasters.
The data processing unit proposed is the Nanoavionics
Payload Computer 2.0 (Xilinx Zynq-7015 Cortex-A9 at max
866 MHz). A combined FPGA and SoC configuration
provides the processing throughput required to deploy the
neural networks within the strict processing window and
before the next ground station revisit time. The selected
unit meets power, volume and mass requirements, while
the processing time requirements were evaluated through
benchmark comparisons [25] and total FLOPs calculations.
For a Cortex-M7 architecture, the total processing time per
pass (over the UK) amounts to 9 hours of processing, implying
a comparable or lower total processing time for the more
powerful Xilinx Cortex-A9 processor.
2. Processing Pipeline
Processing imagery captured so that coastal boundaries
may be extracted is performed in several steps or Levels, with
interstitial products from each Level labelled as such. Levels
0, 1A and 1B are concerned with telemetric, geometric and
radiometric corrections and checks to ensure the image to be
used is free from errors before beginning the main Level 2
processing routine (an example of which is seen in Figure
2). This main routine is made up of cloud identification
and masking, the definition of a buffer zone along the coast,
the coastal feature classification (water, sand, vegetation)
and the extraction of contours along the edges of classified
features representing the coastal boundaries. Cloud masking
is performed to avoid misclassification of and around these
pixels, and the buffer zone is defined to reduce the number
of pixels requiring classification, reducing the computation
time and the number of non-coast boundaries extracted.
OirthirSAT is designed to perform all pre-processing on the
space segment to reduce the required computation, lead time
and downlink costs of processing on the ground prior to
coastal boundary data delivery.
IAC-22-B4.IP.56 Page 4 of 14
73rd International Astronautical Congress (IAC), Paris, France, 18–22 September 2022.
Copyright ©2022 by Mr. Joe Gibbs. Published by the IAF, with permission and released to the IAF to publish in all forms
Fig. 2 Example of a Level 1B 30 m resolution raster image to
be fed into the main processing routine onboard the platform.
Sea runs bottom left to top right, and clouds with shadows
are situated in bottom left and top left corners.
2.1. Cloud Detection
Simple thresholding of the blue channel in the multispec-
tral raster image can be used to identify clouds, although there
is a potential for misclassification of snow as cloud in target
locations. Convolutional neural networks have been trained
to distinguish between snow and cloud pixels [26], but this
usually relies upon short-wave infrared (SWIR) sensors. It is
foreseen that this potential for misclassification of snow as
cloud will not cause problems for coastal boundary extraction
across the UK due to both the extreme rarity that snow will
fall and lie undisturbed across the beach and shoreline (rather
than on higher elevation slopes), and the fact that if snow
is present this may also obscure the true boundary position
and so should be masked out anyway. In order to further
minimise error introduced by cloud objects falling over the
boundary between water and land, identified cloud objects
are buffered by an arbitrary 1 pixel width (Figure 3).
2.2. Buffer Zone Definition
The coastal buffer zone represents an equally spaced
region of interest that runs along the current shape of the coast.
This masking of the buffer zone of interest is implemented to
increase onboard computational efficiency by only including
coastal pixels in the subsequent classification and boundary
extraction steps. Additionally, constraining the region of
interest to a coastal buffer zone ensures that spectrally similar
areas (such as the sandy edge of an inland lake or reservoir)
are not also extracted as a coastal boundary (Figure 4). This
buffer zone is defined using an ANN, trained on the ground,
that segments neighbouring water pixels into a collective
feature. This is then buffered inland and seaward by a constant
amount to give a total cross-shore distance across of 4 km,
Fig. 3 Example output from cloud identification and mask-
ing. Identified clouds and their shadows are shown in pink/-
dark pink.
defined using the maximum tidal range for the UK to ensure
any UK coastal boundary is captured within this range.
Fig. 4 Example output from the coastal buffer zone defini-
tion, showing a 4 km buffer running along the shape of the
coast with areas outwith this or beneath cloud masked to give
‘nodata’ values.
2.3. Classification
Coastal and land classification using multispectral and
hyperspectral imagers has become more prominent in recent
years, due in part to advances in the design of neural network
architectures. Convolutional neural networks (CNN) are
commonly used for classification problems using images,
with CNNs also recently applied to the mapping of coastal
wetlands [27]. The convolutional layer in a CNN breaks the
full target image into smaller kernels of sub-images, reducing
the dimensionality of the image while treating classified
objects spatially (e.g. water pixels being near to other water
pixels). Since OirthirSAT will be utilising a segmentation
IAC-22-B4.IP.56 Page 5 of 14
73rd International Astronautical Congress (IAC), Paris, France, 18–22 September 2022.
Copyright ©2022 by Mr. Joe Gibbs. Published by the IAF, with permission and released to the IAF to publish in all forms
Input
Layers
Fully Connected
Layers
Output
Layer
Fig. 5 Neural-Network Architecture for the baseline classi-
fication algorithm.
algorithm to extract a buffer-zone around the coastline, we
can implement a much simpler, and therefore much more
energy efficient operation to classify each pixel individually
based on the four bands. The simple multi-layer perceptron
proposed as the baseline solution for the OirthirSAT mission
is illustrated in Figure 5. It features 4input nodes and a single
output node due to each pixel containing information in RGB
and near-infrared bands. The output of the algorithm can be
one of four values, corresponding to water, sand, vegetation
or urban/other land cover.
2.4. Edge Detection and Extraction
Upon classification of water, land and vegetation features,
the boundaries between these features are vectorised as line
coordinates. The boundaries are delineated and labelled using
the labelling of the features from the previous classification
step (e.g. boundary between ‘water’ and ‘bare land’ or ‘water’
and ‘vegetation’ is the ’shoreline’). The boundary vectors are
digitised along the edges of the 30 m pixels in the raster image,
and then smoothed using a Gaussian kernel of size relative
to the 30 m pixels to produce smoothed connections between
the pixels. To tidy further the final boundary result disparate
line objects of less than a set threshold length are removed to
eliminate lines delineated around a single pixel and therefore
irrelevant to the larger continuous boundary lines. Figure 6
shows the final pixel classification and extracted shoreline
based on the coastline near Dornoch and Tain, featured in
Figures 2 to 4.
Fig. 6 Example output from the feature classification and
edge extraction. The different classes are separated by shades
of blue and the extracted boundary vector is marked in
orange. The smaller insets show the initial pixel-shaped and
subsequently smoothed boundary.
2.5. In-orbit processing
Typical coastal boundary extraction is a labour intensive
process, with images downloaded from web portals of mis-
sions including Sentinel and Landsat before being processed
by researchers. OirthirSAT propose to provide a single point
of access for researchers and policy makers by performing
large amounts of the processing in-orbit, before downlinking
small vector files containing shoreline boundaries, NDWI and
NDVI indicators as numerical data. Image files generated by
the Simera Sense MultiScape50 multispectral imager (MSI)
will be approximately 1 GB in size, placing significant time
and power constraints on the platform to generate a single
piece of information. It is proposed therefore to perform
processing at Levels 0 through to 2 onboard the nanosatellite,
to generate packaged data parcels of approximately 1–2 MB
each. This will not only reduce the power requirements to
downlink data but also reduce the cost of transmission which
can be costed on a per minute or per MB rate.
2.6. Reduction in Downlinked Data
CubeSats have limits to the amount of data they can
downlink due to the reduced space and power required for
conventional high gain antennas. Multispectral imagers, by
definition, produce large files due to the multiple spectral
IAC-22-B4.IP.56 Page 6 of 14
73rd International Astronautical Congress (IAC), Paris, France, 18–22 September 2022.
Copyright ©2022 by Mr. Joe Gibbs. Published by the IAF, with permission and released to the IAF to publish in all forms
Fig. 7 Processing workflow for the OirthirSAT platform,
highlighting the different stages of the onboard image process-
ing and classification to extract coastal boundaries. Optional
downlinking of interim rasters is marked with a dashed arrow.
bands used. For push-broom sensors, where a single large
image is produced, image file sizes can be in the order of
gigabytes, requiring longer contact windows to successfully
downlink data. OirthirSAT propose to process images in-
orbit and extract critical information automatically, to enable
rapid and responsive access to shoreline vector data. This
means performing the Level 0, 1 and 2 processing in orbit,
as shown in Figure 7. For the Simera Sense MultiScape-50
imager selected, a single pass of the east or west UK mainland
coastline, with a length of approximately 1215 km, at a GSD
of 30 m/pixel, 12 bit/pixel and a ground swath of 120 km
would generate an image file of approximately 948 MB. The
complete list of file sizes is presented in Table 2. Level
1A, 1B and 1C rasters, generated as auxiliary data through
radiometric, reflectance and geometric processing, combined
with the image data would keep the file size the same. Level
2 rasters generated using Normalised Difference Water Index
(NDWI) and Normalised Difference Vegetation Index (NDVI)
classifiers would generate a 618 MB file with 32 bit/pixel
depth. The associated binary classified rasters at 1 bit/pixel
depth creates an additional 19 MB file. The final shoreline
extraction process will output a coastline vector file of 2.9
MB, increasing to 3.2 MB when additional .dbf attributes are
added. While processed images could be downlinked when
required (assumed to be on a monthly cycle), the critical
shoreline edge data can be downlinked on each pass since the
3.2 MB is approximately 300 times smaller than the original
raw image.
3. System Budgets
The top-level system budgets for the platform are pre-
sented to demonstrate the feasibility of the proposed configu-
ration of the OirthirSAT platform. The complete specification
is detailed in later sections.
3.1. Power Budget
Power budgeting is dictated by the payload operation
frequency. With a 3-day revisit period and approximately 15
orbits per day, ample energy is available to avoid initiating
imaging periods under a (battery pack) energy deficit. The
power budget is constructed with reference to the 3-day
revisit, which is used to derive the duty cycle operation of
all subsystems. The nominal and peak power consumption
of each subsystem is provided in Table 3. Average power
consumption values can be calculated using the duty cycle
figures to evaluate the system energy margins.
The Payload Computer (data handling unit) is assigned
the highest duty cycle of 31.9%, assuming a worst-case
operation of 1 day (or 15 orbits) which relates to the case of
no image segmentation and classification of all pixels from
an image.
During the mission main phase, the routine operations
required to deliver the mission objectives will be regular and
repeatable (3 days revisit) over the 2 year mission lifetime.
States are defined as the CubeSat operational states from the
perspective of power budgeting, implying which subsystem
dominates the power budget at a time and indicating the un-
dergoing primary task. Those are presented in Table 4, with
the effective duty cycle and corresponding average power
of each state. Science Acquisition states split the scientific
operation periods into "Imaging and Downlink" and "Data
Processing". The former describes the actual imaging period
with respect to the 3-day revisit period, while the latter the
operation of the on-board processor, ultimately producing the
IAC-22-B4.IP.56 Page 7 of 14
73rd International Astronautical Congress (IAC), Paris, France, 18–22 September 2022.
Copyright ©2022 by Mr. Joe Gibbs. Published by the IAF, with permission and released to the IAF to publish in all forms
Processing
Stage
Description
Source File Size
[MB]
Raw Imagery Rasters from the MSI in R,G,B,NIR
spectral bands MSI 948
Auxiliary Image Data Level 1A, 1B, 1C rasters OBC storage 948
Processed Imagery Level 2 rasters Temporary OBC storage 618 / 19
Coastal boundary Vector marking coastal boundaries Classified raster edge detection 2.9 / 3.2
Table 2 Estimates of data file sizes following each stage of the in-orbit processing.
Subsystem
Baseline Power
(mW)
Peak Power
(mW)
PDH 2000 3500
MSI 2700 7000
COMMS S-band 420 7200
COMMS UHF 82.5 1400
ADCS 720 2950
OBC 400 1000
EPS 100 100
Total 6422.5 23150
Table 3 OirthirSAT subsystems power consumption.
vector files ready for downlink. Considering that UK-based
ground stations have been prioritised at this stage of the
design, there is overlap between imaging and communica-
tion with the ground station (primarily commands), which is
considered in "Imaging and Downlink" state. The final state
"Data Downlink" describes the case where only the communi-
cation subsystems are operable and are dominating the power
consumption, also corresponding to the expected ground
contact time to deliver the vector files. The period in which
the CubeSat is not operating in any of the above-described
states is considered as a "Charging Period", describing the
nominal or baseline power consumption by the rest of the
platform.
The use of states enables the identification of the number
of instruments that may operate in parallel, enabling the
design of a power budget that can meet the peak power
demands of each state. The state machine during routine
operations is shown in Figure 8.
Fig. 8 State diagram.
3.2. Mass Budget
The mass budget is typically constrained by the launcher
and/or the CubeSat deployer. Commercial deployers follow
the CubeSat Design Specification Standard Rev. 14 [28] sug-
gesting a maximum mass of 6 kg for 3U CubeSats. Margins
of 5% at subsystem level are used to take into account addi-
tional mass such as harness and fasteners. Higher margins are
considered for subsystems with higher level of uncertainty
such as the Payload Data Handling (PDH) and the Dragsail
(DSL). Table 5 summarises the mass budget and the margin
philosophy.
3.3. Volume Budget
Similarly, the volume of the spacecraft is restricted by the
deployer which is the spacecraft’s main interface. CubeSat
Design Specification Rev. 14 [28] is also adopted in this
case. The standard 3U CubeSat configuration with the tuna
can allows the lens to be accommodated externally on the
+Z face as seen in Figure 9, allowing more internal volume
IAC-22-B4.IP.56 Page 8 of 14
73rd International Astronautical Congress (IAC), Paris, France, 18–22 September 2022.
Copyright ©2022 by Mr. Joe Gibbs. Published by the IAF, with permission and released to the IAF to publish in all forms
Operating State Duty Cycle (%) Average
Power (mW)
Peak Power
(mW)
Charging State 67.0 1290 1900
Science Acquisition -
Data Processing
31.9 2260 10500
Science Acquisition -
Imaging and Down-
link
0.116 26 24000
Data Downlink 0.096 15 16000
Total 3590
Table 4 Operating States power consumption.
Subsystem Unit Margin
Mass with
Margin (kg)
EPS 5% 1.77
OBC 5% 0.065
ADCS 5% 0.582
PDH 10% 0.026
COMS 5% 0.361
MSI 5% 0.536
TCS 5% 0.053
STR 5% 0.414
DSL 20% 0.480
Total 4.29
Table 5 OirthirSAT Mass Budget.
to the remaining subsystems. Similar margin philosophy
is followed in the volume budget taking into account the
volume separation between the subsystems as well as the
higher uncertainty associated with specific subsystems. The
volume budget and margins are summarised in Table 6.
4. Platform Architecture
Multispectral imagers are common CubeSat payloads
with missions proposed in [29] and [30] defining platform
architectures for nanosatellites. The subsystems selected for
the OirthirSAT mission are presented in the following section
along with system budgets to demonstrate the feasibility of
the platform architecture.
Subsystem Unit Margin
OirthirSAT
Volume with
Margin (Us)
EPS 5% 0.396
OBC 5% 0.058
ADCS 5% 0.788
PDH 10% 0.099
COMS 5% 0.357
MSI 5% 0.557
TCS - *
DSL 20% 0.684
Total 2.94
Table 6 Volume Budget.
4.1. Electrical Power Subsystem
The electrical power subsystem (EPS) provides power
to the Nanosat by regulating the power input from the so-
lar panels, charging the battery pack and monitoring the
current consumption levels of the voltage buses. The AAC
ClydeSpace EPS (Starbuck Nano board, Optimus battery
pack and solar panels) is chosen for its high TRL of 9 and
extensive flight heritage, while also marginally outperforming
other candidates with respect to mass, volume and quiescent
power loss.
The OirthirSAT mission requires 3x3U solar panels,
positioned in the +X and -X (along travel) Nanosat faces. With
the orbit design suggesting a 15°beta angle, different solar
panel configurations are possible. For 4x3U panels positioned
on
±𝑋
and
±𝑌
, maximising the incident solar power requires
IAC-22-B4.IP.56 Page 9 of 14
73rd International Astronautical Congress (IAC), Paris, France, 18–22 September 2022.
Copyright ©2022 by Mr. Joe Gibbs. Published by the IAF, with permission and released to the IAF to publish in all forms
Fig. 9 Subsystem Arrangement in the 3U-Tuna Can Volume
Envelope.
the Nanosat to maintain a yaw at 45°. Considering the
incident solar power cosine loss [31], given in (1)
𝑃panel =𝑃0∗𝐼d∗cos(𝜃)(1)
a small loss exists due to the beta angle of
cos(
15°
)=
0
.
9659
for panels on
±𝑋
. By including an additional deployed panel
on either
±𝑋
face, the total incident power exceeds that of
4x3U panels. Consequently, it is preferable to increase power
input generation via additional deployed panels on
±𝑋
rather
than
±𝑌
. For the selected configuration, the solar panels
generate up to 16.9 W of power during the sunlit period of
each orbit, with an average of 5.7 W per orbit.
Battery packs are typically discharged during the eclipse
period when solar panels cannot supply any power. For this
preliminary design stage, worst-case scenarios are considered.
Since the solar panels’ power output varies throughout an
orbit, the battery pack supplies the remaining load. Hence,
at the worst case, when the solar panels momentarily fail
to support the load (i.e. 0 W power production) the battery
pack must support all power requirements. A 30 Wh battery
capacity was determined sufficient to ensure below 1 C (3.87
A) discharge under all peak current consumption scenarios
such as the one described above.
4.2. Attitude Determination and Control System
The ADCS is required to provide a maximum pointing
error of 1°during imaging mode and to be able to actuate
the nanosatellite to the maximum off-nadir pointing angle
required to image coastlines after factoring in orbit drift. The
pointing error requirement will lead to increased mapping
errors for off-nadir pointing commands. However this value
ensures that the imager swath covers the whole coastline
when within the 1°limit [31]. The CubeSpace CubeADCS 3-
Axis is an ADCS subsystem consisting of coarse sun sensors,
fine sun and Earth or nadir Sensors, magnetometers and
inertial measurement Unit. This model is TRL 9 after having
been used on previous CubeSat missions and expected to
launch again on the SERB mission [32]. To achieve the
knowledge error of 0.5°the ADCS is complemented with a
CubeSpace CubeStar Star Tracker. This is a low-power star
tracker able to provide vector measurements with an accuracy
of
(<
0
.
0154°
, <
0
.
0215°
, <
0
.
0610°
)
in the right ascension,
declination and roll axes. The star tracker and baffle are to
be mounted on along the x-axis of the CubeSat, facing the
direction of travel. An additional precautionary manoeuvre
is also to be detailed to ensure the operation of the star tracker
prior to imaging passes. The required off-nadir pointing
angle could lead to pointing commands moving parts of the
earth or sun into the field-of-view of the star tracker, disabling
it. As a contingency the nanosatellite will be able to perform
a short manoeuvre to orientate the x-axis into deep space,
enabling the full-state EKF to estimate gyroscope biases and
reduce knowledge error prior to the imaging pass.
In addition to the ADCS sensor suite, a GNSS module,
the AAC Clyde Space GNSS200, is included to enhance the
tracking capability of the ground station. This will reduce the
along and cross-track position errors compared with tracking
just using an orbit propagation model.
4.3. Communications system
The OirthirSAT nanosatellite splits communications be-
tween UHF and S-Band modules for tracking, telemetry and
control (TT&C) and data downlink. The data saving from the
downlink of vector files rather than images comes through the
reduced minimum required contact time with a ground station
IAC-22-B4.IP.56 Page 10 of 14
73rd International Astronautical Congress (IAC), Paris, France, 18–22 September 2022.
Copyright ©2022 by Mr. Joe Gibbs. Published by the IAF, with permission and released to the IAF to publish in all forms
to be able to downlink fully payload data. Both communi-
cation bands use amateur frequencies of between 2.40–2.45
GHz and 430–440 MHz for S-Band and UHF respectively.
OirthirSAT intend to use Dundee Satellite Station as the
primary ground station for the duration of the mission.
S-Band communication offers increased data rates which
makes it a popular choice for CubeSats due to the well
understood limitations of data downlink. While OirthirSAT
will perform onboard processing of EO data and hence,
provide a reduction in the size of data files downlinked
to ground stations, it is still preferable to have significant
downlink capability in the event that additional images are
required on the ground. A potential use case for images
would be to validate and tune the ANN algorithms using
data from the MSI and enable reprogramming of the payload
computer if required. The Endurosat S-Band Transmitter has
been selected due to the high maximum data rate of up to 20
Mbps.
OirthirSAT utilises the Endurosat UHF Transceiver II for
TT&C. The Nanoavionics UHF Antenna system was chosen
in part due to the mounting options and no interference
issues when mounted on the +Z face of the CubeSat. This is
required since the upper -Z face of the nanosatellite structure
is taken up by the dragsail which is presented in later sections.
This enables full operation of the multispectral imager while
maintaining communications downlink as the antenna won’t
interfere with the field of view.
4.4.1. Contact Time
Given the considered mission altitudes and a ground
segment based in Scotland, a daily ground contact time of
approximately 75 seconds is expected. With the current
S-band data rate and an approximated data output of 2.9 MB
at the end of the processing, the payload data transmission is
expected to last for a maximum of 2 seconds. The remaining
bandwidth can be used for the transmission of validation data
or for additional error checking tasks. While the primary data
product will consist of the vector marked coastal boundaries
detailed in Table 2, raw imagery will be required to be
downlinked during the initial validation mission phase. As
such, with estimated file sizes of 948 MB for a single pass
of the eastern or western UK coastline and allowing for a
error-margin of 30%, approximately 7 passes will be required
to downlink a raw image.
4.4. Dragsail
The increasing access to space is not without its draw-
backs. Perhaps of most importance is the issue of artificial
space debris, mostly comprising of defunct satellites that
Fig. 10 A preliminary render of a quasi-rhombic pyramid
dragsail deployed from a 3U CubeSat.
have not been disposed of at the end of life. Unchecked, this
problem is referred to as Kessler Syndrome [33], whereby
the amount of space debris reaches a critical point where
it begins to exponentially generate more debris. This has
severe implications for platforms such as the ISS but also for
future launches. To combat the space debris problem, ESA
has developed the Clean Space initiative, to provide training,
regulation and funding for technologies aimed at reducing
the amount of space debris in low earth orbit. One such
method of removing future space debris due to end-of-life
platforms is to de-orbit satellites at the end of their mission.
Dragsails have been proposed to act as de-orbiting devices
for spacecraft in LEO for some time [34], [35]. Since there
is non-negligible atmospheric drag in low-earth orbits, in-
creasing the effective area of a CubeSat would significantly
increase drag and reduce the altitude much earlier than if left
to the base drag of the platform. The University of Glasgow
have developed a prototype quasi-rhombic pyramid structure
[36], inspired by [37], for use as a dragsail. OirthirSAT plan
to utilise this system to de-orbit the 3U CubeSat at the end
of the data-gathering phase, aligning the programme to the
ESA Clean Space initiative. A render of the proposed design
deployed from a 3U CubeSat is presented in Figure 10.
5. Conclusions
This paper presents the preliminary design of a 3U Cube-
Sat and supporting infrastructure to provide dedicated, pro-
cessed data for the purposes of modelling the effects of
IAC-22-B4.IP.56 Page 11 of 14
73rd International Astronautical Congress (IAC), Paris, France, 18–22 September 2022.
Copyright ©2022 by Mr. Joe Gibbs. Published by the IAF, with permission and released to the IAF to publish in all forms
climate change on UK coastlines. With the support of the
UK Space Agency and several government organisations,
OirthirSAT will be able to generate dedicated datasets on UK
coastlines in a convenient, pre-processed format, reducing
the costs of downlink and reducing the power budget. The
proposed processing pipeline has been defined with file size
estimates made for each level of processing, showing how
the onboard processing can extract only the critical data to
measuring changes in coastlines.
Delivering this dataset will significantly decrease the
time taken for coastal researchers to build up a picture of
the effect of climate change on UK coastlines. Providing
this coastal data with national coverage and in a standardised
format will allow various institutions to query coastal change
observations easily, and perform their own analyses in a
more repeatable manner. By performing image processing
and classification onboard the space platform, OirthirSAT
additionally removes the independent repetition of processing
and analysing raster data on the ground. The timely nature of
the data will also help to identify sudden changes in coastlines
that could be indicators of potential flooding, severe erosion
and loss of assets. The mission architecture presented here
may be scaled up to larger constellations of nanosatellites for
performing similar processing tasks across the world.
The nanosatellite configuration presented meets the re-
quirements posed by the mission objectives, both in terms
of capturing, processing and downlinking data, but also in
providing the nominal operational states for accurate pointing,
charging and health monitoring provided by the ADCS, EPS
and OBC. Preliminary system budgets have been presented
in section 3 demonstrating adherence to requirements for a
positive power margin, volume and mass constraints for the
3U platform.
The OirthirSAT mission has been funded through the
LaunchUK programme and began the detailed design phase
in August 2022. The platform launch is currently projected
for late 2024.
6. Acknowledgements
This work has been undertaken as part of OirthirSAT’s
entry in the LaunchUK Nanosat Design Competition run by
the UK Space Agency and LaunchUK, the UK Government’s
Spaceflight Programme. The authors thank GU Orbit and the
OirthirSAT team for the support provided over the duration of
the preliminay design phase. OirthirSAT acknowledges the
support from Dr Kevin Worrall at the University of Glasgow
and Douglas McNeil from EOLAS Insight. Additional
thanks to Dundee Satellite Station for advice regarding the
communications systems.
References
[1]
UKSEDS, UK Students for the Exploration and De-
velopment of Space - Satellite Design Competition,
https://ukseds.org/aurora/?p=SDC2021
, Ac-
cessed: 2022-08-31, 2021.
[2]
CIESIN, “Percentage of Total Population Living in
Coastal Areas,” Columbia University, Palisades, Tech.
Rep., 2006, pp. 170–175.
[3]
IPCC, Sea Level Rise and Implications for Low-Lying
Islands, Coasts and Communities. 2019, pp. 321–446,
isbn: 9781009157964.
[4]
K. L. Jankowski, T. E. Törnqvist, and A. M. Fernan-
des, “Vulnerability of Louisiana’s coastal wetlands
to present-day rates of relative sea-level rise,” Na-
ture Communications, vol. 8, pp. 1–7, 2017, issn:
20411723.
[5]
A. P. Luijendijk, G. Hagenaars, R. Ranasinghe, F.
Baart, G. Donchyts, and S. Aarninkhof, “The State of
the World’s Beaches,” Scientific Reports, vol. 8, no. 1,
pp. 1–11, 2018, issn: 20452322.
[6]
L. M. Abadie, “Sea level damage risk with probabilistic
weighting of IPCC scenarios: An application to major
coastal cities,” Journal of Cleaner Production, vol. 175,
pp. 582–598, 2018, issn: 09596526.
[7]
IPCC, “Summary for Policymakers,” IPCC, Tech. Rep.,
2022, p. 37.
[8]
P. Sayers, E. C. Penning-Rowsell, and M. Horritt,
“Flood vulnerability, risk, and social disadvantage:
current and future patterns in the UK,” Regional Envi-
ronmental Change, vol. 18, no. 2, pp. 339–352, 2018,
issn: 1436378X.
[9]
R. C. Ballinger and W. Dodds, “Shoreline management
plans in England and Wales: A scientific and trans-
parent process?” Marine Policy, vol. 111, no. March
2017, p. 102 689, 2020, issn: 0308597X.
[10]
E. H. Boak and I. L. Turner, “Shoreline definition and
detection: A review,” Journal of Coastal Research,
vol. 21, no. 4, pp. 688–703, 2005, issn: 07490208.
[11]
F. J. Anders and M. R. Byrnes, “Accuracy of shoreline
change rates as determined from maps and aerial
photographs,” Shore and Beach, vol. 59, no. 1, pp. 17–
26, 1991.
[12]
A. F. Rennie et al., “The National Overview Synopsis
of the Dynamic Coast Research,” CREW, Tech. Rep.,
2021, pp. 1–13.
IAC-22-B4.IP.56 Page 12 of 14
73rd International Astronautical Congress (IAC), Paris, France, 18–22 September 2022.
Copyright ©2022 by Mr. Joe Gibbs. Published by the IAF, with permission and released to the IAF to publish in all forms
[13]
E. D. Lazarus et al., “The UK needs an open data
portal dedicated to coastal flood and erosion hazard
risk and resilience,” Anthropocene Coasts, vol. 4, no. 1,
pp. 137–146, 2021, issn: 25614150.
[14]
F. M. Muir, M. D. Hurst, A. F. Rennie, and J. D. Han-
som, “Technical Annex Work Stream 2RA - National
Coastal Risk Assessment,” CREW, Tech. Rep., 2021,
pp. 1–13.
[15]
M. D. Hurst, F. M. Muir, A. F. Rennie, and J. D.
Hansom, “Technical Annex Work Stream 2 - Future
Coastal Change,” CREW, Tech. Rep., 2021, pp. 1–16.
[16]
K. Vos, K. D. Splinter, M. D. Harley, J. A. Sim-
mons, and I. L. Turner, “CoastSat: A Google Earth
Engine-enabled Python toolkit to extract shorelines
from publicly available satellite imagery,” Environ-
mental Modelling and Software, vol. 122, p. 104528,
2019, issn: 13648152.
[17]
USGS, Coastal change hazards portal,
https://
marine.usgs.gov/coastalchangehazardsportal/
,
Accessed: 2021-08-30, 2021.
[18]
Channel Coastal Observatory, National network of
regional coastal monitoring programmes - programme
design,
https://coastalmonitoring.org/programme_
design, Accessed: 2021-08-13, 2019.
[19]
UK Hydrographic Office, Creating coastlines using
data science,
https: //ukhodigital.blog.gov .
uk/2020/02/12/creating- coastlines- using-
data-science/, Accessed: 2022-06-27, 2021.
[20]
P. Sayers, C. Moss, S. Carr, and A. Payo, “Responding
to climate change around England’s coast - The scale
of the transformational challenge,” Ocean & Coastal
Management, vol. 225, no. July 2021, p. 106 187, 2022,
issn: 09645691.
[21]
M. J. Veyette et al., “AI/ML for mission processing
onboard satellites,” in AIAA SCITECH 2022 Forum,
2022, p. 1472.
[22]
M. S. J. Rogers, M. Bithell, S. M. Brooks, and T.
Spencer, “VEdge_Detector: automated coastal veg-
etation edge detection using a convolutional neural
network,” International Journal of Remote Sensing,
vol. 42, no. 13, pp. 4809–4839, 2021.
[23]
N. Saeed, A. Elzanaty, H. Almorad, H. Dahrouj, T. Y.
Al-Naffouri, and M.
-
S. Alouini, “Cubesat communica-
tions: Recent advances and future challenges,” IEEE
Communications Surveys & Tutorials, vol. 22, no. 3,
pp. 1839–1862, 2020.
[24]
R. J. Boain, “AB-Cs of Sun-Synchronous Orbit Mis-
sion Design,” in Advances in the Astronautical Sci-
ences Series, Proceedings of the 14th Space Flight
Mechanics Meeting, Univelt: Jet Propulsion Labora-
tory, NASA, 2005.
[25]
B. Sudharsan et al., “TinyML Benchmark: Executing
Fully Connected Neural Networks on Commodity
Microcontrollers,” in 2021 IEEE 7th World Forum on
Internet of Things (WF-IoT), 2021, pp. 883–884.
[26]
D. Varshney, P. Gupta, C. Persello, and B. Nikam,
“Snow and cloud discrimination using convolutional
neural networks,” ISPRS Annals of the Photogramme-
try, Remote Sensing and Spatial Information Sciences,
vol. 4, pp. 59–63, 2018.
[27]
C. Liu, R. Tao, W. Li, M. Zhang, W. Sun, and Q. Du,
“Joint classification of hyperspectral and multispectral
images for mapping coastal wetlands,” IEEE Journal
of Selected Topics in Applied Earth Observations and
Remote Sensing, vol. 14, pp. 982–996, 2020.
[28]
Alicia Johnstone, “CubeSat Design Specification,” Cal-
ifornia Polytechnic, San Luis Obispo, CA, Tech. Rep.,
2022.
[29]
A. Pellegrino et al., “HORUS: multispectral and multi-
angle cubesat mission targeting sub-kilometer remote
sensing applications,” Remote Sensing, vol. 13, no. 12,
p. 2399, 2021.
[30]
S. Song, H. Kim, and Y.
-
K. Chang, “Design and imple-
mentation of 3U cubesat platform architecture,” Inter-
national Journal of Aerospace Engineering, vol. 2018,
2018.
[31]
J. R. Wertz, W. J. Larson, D. Kirkpatrick, and D.
Klungle, Space mission analysis and design. Springer,
1999, vol. 8.
[32]
M. Meftah et al., “SERB, a nano-satellite dedicated
to the earth-sun relationship,” in Sensors and Systems
for Space Applications IX, International Society for
Optics and Photonics, vol. 9838, 2016, 98380T.
[33]
D. J. Kessler, N. L. Johnson, J. Liou, and M. Matney,
“The Kessler Syndrome: Implications to future space
operations,” Advances in the Astronautical Sciences,
vol. 137, no. 8, p. 2010, 2010.
[34]
P. Harkness and P. Roberts, “A drag sail for safe rapid
disposal of spacecraft from leo,” 2006.
[35]
P. C. Roberts and P. G. Harkness, “Drag sail for
end-of-life disposal from low earth orbit,” Journal
of Spacecraft and Rockets, vol. 44, no. 6, pp. 1195–
1203, 2007.
IAC-22-B4.IP.56 Page 13 of 14
73rd International Astronautical Congress (IAC), Paris, France, 18–22 September 2022.
Copyright ©2022 by Mr. Joe Gibbs. Published by the IAF, with permission and released to the IAF to publish in all forms
[36]
G. MacAskill, S. Messina, and I. Serrano Martín-
Sacristán, “Mechanical design and deployment of a
quasi-rhombic pyramid drag sail for safe de-orbit of a
3U cubesat,” in 4th Symposium on Space Educational
Activities, Universitat Politecnica de Catalunya, 2022.
[37]
M. Ceriotti, P. Harkness, and M. McRobb, “Variable-
geometry solar sailing: The possibilities of the quasi-
rhombic pyramid,” in Advances in Solar Sailing,
Springer, 2014, pp. 899–919.
IAC-22-B4.IP.56 Page 14 of 14
