Content uploaded by Thomas Lake
Author content
All content in this area was uploaded by Thomas Lake on Aug 27, 2015
Content may be subject to copyright.
Methods for Individual Based Modelling
of Harbour Porpoise
T. Nick Croft
Email: t.n.croft@swansea.ac.uk
Ian Masters
Email: i.masters@swansea.ac.uk
Marine Energy Research Group, Swansea University
Singleton Park, Swansea, SA2 8PP, United Kingdom
Thomas Lake
Email: 527562@swansea.ac.uk
Abstract—Individual Based Models (IBMs) can be used to
investigate emergent behaviours of groups and flocks of animals.
Early uses of IBMs showed plausible looking behaviours emerg-
ing from simple rules and were used in computer generated
animations and images. This type of model has since seen use
simulating behaviours of a range of animals ranging from clam
larvae to moose in a range of environments. The flexibility of the
model allows a range of environmental parameters to be included,
allowing the response of the simulated animals to the environment
to be investigated. The complexity of these models can vary
considerably, and a number of additions can be made to the basic
IBM structure. The detail and complexity of these models is in
part constrained by the availability of environmental data and
in part by available data of the animals to be simulated, which
can be particularly difficult to obtain in a marine environment. A
planned model for the behaviour of Harbour Porpoise (Phocoena
phocoena) is given as an example use of these techniques.
Index Terms—individual based model, marine energy, environ-
mental impact, boids, agents
I. INTRODUCTION
Individual Based Models (IBMs) simulate the actions of
individual members of a population in a given environment.
The individual members of the population are given a set
of rules that define their behaviour, with the behaviour of
each simulated individual being determined on a case by case
basis based on the state of the simulated population and/or
environment. The aggregate behaviours of individuals in this
model are potentially able to provide insight into the behaviour
of intelligent actors, given a suitable set of rules. Recently, this
type of model has been used in ecological models to evaluate
the interaction of animals with the environment around them
[1].
This contribution will discuss some approaches to IBMs, as
applied to the development of an IBM for the behaviour of a
Harbour Porpoise population located close to a proposed tidal
stream turbine deployment site. The model will be informed
by a range of data sets collected over the last few years from
the site off the Welsh coast. These data sets include details of
bathymetry, current flow, noise and information on sightings
of porpoises and other species. The general behavioural rules
for the Harbour Porpoise will be developed based on available
literature and will then be adjusted as required based on site-
specific observational data.
The concepts behind this type of rule based simulation
are simple, there are some obstacles to overcome. This con-
tribution will examine a number of concepts which can be
incorporated into individual based models and some of the
challenges that must be dealt with when trying to apply these
models to environmental scenarios.
II. ENVIRONMENTAL IMPACT
When evaluating potential deployment sites for marine
energy devices, it is important to consider what impacts that
deployment may have on the surrounding area - both in
terms of physical changes to the location and also in terms
of the effects on the local ecology and other marine users.
In the United Kingdom, developers are legally required to
consider the impact of their devices on the environment as
part of the consenting process [2], with similar obligations
applying elsewhere in the European Union [3], [4]. Typically
these reports are required to detail how the development will
effect the local wildlife in both the short- and long-term.
This is a significant part of the consenting process, typically
requiring significant investment of both time and funds to
complete. Illustrating this, a recent scoping report [5] identified
a wide range of areas to be assessed, including the following
ecological items:
•Coastal Processes and Sediment Transport
•Marine Water Quality and Air Quality
•Intertidal and Subtidal Benthic Ecology
•Fish, Marine Mammals and Coastal Birds
•Marine and Terrestrial Noise
In addition, the report also considered a number of other
areas, including the visual appearance of the deployment, the
economic impacts and how the deployment might affect other
marine users.
Many of the ecological areas identified above require a
baseline to be established in order to examine the potential
effects of a marine energy device on the environment. This
can include taking samples and measurements of geophysical
data and monitoring marine mammals and birds in the area.
The geophysical data has value in itself, but can also be
incorporated into computer models of animal behaviour and
movement. This is useful as it potentially allows the quality
of the computer models to be improved by supplying more
accurate data but without incurring additional effort on the
part of the developer.
A. Marine Mammals
Of particular note with regards to marine energy devel-
opments are the protections afforded to marine mammals
and the requirements this can place on developers. Harbour
Porpoise and Grey Seals are both listed in Annex II of [6],
which requires that their habitats be designated as Special
Areas of Conservation. This requires EU member states to
prevent actions that would damage the habitat or otherwise
cause disturbance to protected species. In the United Kingdom,
disturbing a protected species without a license is a criminal
offence [7]. This means that developers must be particularly
diligent when considering the impact that a marine energy
device may have on marine mammals throughout device
deployment, operations and maintenance.
There are a number of ways of monitoring the usage of an
area by marine life, including active and passive acoustic mon-
itoring, radar, thermal imaging and visual observations [8]–
[11]. This data can provide insight into the usage of an area
by a particular species, with different methods having different
advantages depending on the species being monitored. It may
sometimes be possible to ‘tag’ members of a population in
order to gain monitor the movements of individuals over a
longer time period [12]–[14]. These tags typically provide data
containing the position of the tagged animal along with sen-
sory information about its surroundings. This could include,
for example, depth and temperature for diving marine life but
might be airspeed and magnetic field for migratory birds. This
approach can be logistically and legally difficult (Disturbing
protected species requires additional licensing as discussed
above), but can be used to investigate the conditions sought
out or avoided by an individual. It can also show significant
variations in behaviour between individuals [12], which can
highlight the challenges faced when deciding on suitable rules
for an individual based model.
Existing local information, such as recorded sightings,
strandings or bycatch data may also be able to provide
information regarding the presence of marine mammals [14],
[15], which may be particularly useful for species such as
Harbour Porpoise which are more timid and harder to distin-
guish visually [14]. Information regarding how animals use
a habitat and the number of animals in a local population is
important when exploring the potential influence of marine
energy devices.
B. Predicting impacts
These models have been used to examine the impact of
a variety of habitat changes on seabirds [16], among other
examples. They are also being examined for use to feed into
Environmental Impact Assessments [17]. Before the predictive
potential of a model can be considered, the ability of the model
to reflect existing situations must be examined. The measured
values and values predicted by a number of IBMs were
compared in [16], and showed that individual based models
can be used to accurately predict a number of scenarios.
III. INDIVIDUAL BASE D MOD EL S
Individual Based Modelling techniques aim to capture in-
formation about the behaviour of a population based on the
individual movements and behaviours of the individuals in that
population. In it’s simplest implementation individuals (also
referred to as “agents” or “boids” [18]) use information about
other individuals around them to make decisions regarding
their movement. This basic simulation can be built upon to
incorporate a simulated environment and more realistic repre-
sentations of senses. The boids can also be designed to react
differently to stimuli depending on their current behavioural
state or mood [19], and to incorporate simple memory models
to allow them to map the areas that they have visited [20]. The
wide range of possible modelling approaches encompassed
here presents a great deal of flexibility when considering a
problem, but presents it’s own difficulties - a more detailed
model may produce results that more closely fit the real be-
haviours observed, but will have a higher computational cost.
Detailed models require a greater understanding of both the
animals concerned and location being simulated, both of which
can be particularly challenging in a Marine environment.
A. Basic methodology
A simple individual based model is an iterative process:
1) For each boid in the population:
a) Get list of nearby boids
b) Make decision about movement
c) Update velocity/orientation/position
2) Record positions
3) Advance simulation clock
4) Repeat from 1 for next time step
The behavioural rules defined in the model are used in
step 1b to determine how each individual will move, which
typically involves consulting the list of nearby boids. As each
of the Nboids in the simulation needs to check the positions
of all other boids in order to determine it’s neighbours (1a), the
simulation in a naive implementation scales according to N2.
This can be reduced by splitting the simulated environment
into a number of partitions and maintaining lists of boids
for each. This reduces the cost of searching for neighbours
by reducing the list searched to the list associated with the
local partition for most cases. Boids that are approaching or
crossing the boundary between partitions will need to consult
the list of neighbouring partitions, but this can be minimised by
periodically rearranging the partitions to reflect the changing
distributions of boids. This approach also makes it possible
to parallelise the process by iterating through each partition
independently. In this case, the balancing required to minimise
boids near borders should also attempt to balance the number
of boids in each partition in order to keep the processing
time similar for all partitions. This partitioning process can be
fulfilled by a number of different algorithms, as explored in
[21], with the aim of splitting the simulated environment into
an appropriate number of partitions as quickly and consistently
as possible in order to minimise overheads.
B. Representing boids and boid movements
Before looking in more detail at the additional features
that can be incorporated into individual based models, it
is necessary to consider how the boids themselves will be
modelled. This has an impact on how the movement of the
boids is handled.
The simplest representation of a boid is that of a point
particle, followed closely by orientable point particles as in
[18]. Under these representations the position of each boid is
marked by a set of coordinates and orientation indicated by
a vector in two or three dimensions. Alternatively, a skeletal
model of the animal represented by the boid can be imple-
mented as in [19]. This approach allows more physiologically
realistic motion to be obtained, and allows the effect of
currents on the orientation and manoeuvrability of the animal
to be modelled by the boids. This also allows for detailed
outputs showing the physical movement simulated by the boid,
although this level of detail may be more useful in the field
of computer animation [18], [19] than for ecological studies
looking at the usage of a larger habitat.
The next consideration concerns the spatial movement of
the boids, and is dependent on both the representation of the
boid discussed above and the way in which behavioural rules
are to be implemented. A simple model might use a single rule
to set a velocity for the boid, which is then combined with the
timestep duration to calculate the position of the boid. If the
boid is being represented by a skeletal model as in [19] then it
may be more appropriate to calculate the acceleration due to
the forces acting on or exerted by the different portions of the
boid. This also allows the orientation of the boid to be altered
by external forces as well as it’s own intended movements. For
simpler representations, these accelerations may be set directly
by the behavioural rules rather than due to forces arising from
changes in the shape of the boid.
In other cases the precise movement of an individual may
not be considered important, and only the area of the simula-
tion occupied by a boid needs to be recorded. The simulated
environment in that instance is divided into a number of cells,
with boids moving directly from cell to cell with destination
cells selected by their behaviour rules as in [22]. This can
reduce the complexity of the model while still capturing
useful information. This assumes that the smallest significant
movement is on the same order as the cell sizes, and adds
constraints on the minimum timestep unless a minimum dwell
time per cell is imposed on the boids via their behavioural
rules. This discretisation of the simulated environment can also
simplify how environmental information is represented in the
simulation, which will be discussed further below.
C. Intentions, moods and modes
Under the simple model outlined above in section III-A,
a boid will react based on the information available to it at
that instant - it’s decisions aren’t directly influenced by it’s
decision in the previous timestep. This is the sort of decision
making seen in a number of models, such as those given in
[18], [23], [24]. An alternative model is to set a short term
goal, or intention, that determines how the boid will react
to a given set of inputs until the next event which causes
that intention to change. This is implemented in the model
given in [19] which assigns each boid an ‘intention’. The
model implemented in [20] takes a slightly different route, and
has different behaviours that are chosen based on the boid’s
internal state.
This modification to the basic concept allows for the simu-
lation of different individual behaviours, as well as simulating
movement. For the fish boids in [19] this included avoid
(collision avoidance), escape (evade predator), school (join or
remain in a school), eat, mate, leave (leave current school)
and wander. Each of these intentions resulted in different
movements and reactions to the environment, with the current
intention being stored with each boid to enable intentions to
be carried over between timesteps. This allows behaviours to
persist over longer time frames and allows for conditions to be
defined which determine when a particular intention or mode
will yield to different modes.
D. Memory models
In addition to tracking any current intentions between
timesteps, boids can be constructed with an ability to mem-
orise information. An example of this can be seen in [20],
where the boids (representing moose) remembered the state
of areas of the environment that they had visited and used
this when deciding where to move. This was implemented
in such a way that each boid only had access to the state
at the time it left each of the areas, such that a boid could
travel to an area and find it no longer suitable to satisfy
the boid’s intentions. Boid memories could be extended in
any way thought suitable for a model. This might be to
remember whether they have encountered another boid before,
to store the location of particular features of the environment
or locations where a boid has successfully fed. This may be
useful if a species being modelled is thought to habitually
return to certain areas.
A memory of previously visited locations was implemented
in [22], allowing the boids (representing panthers) to prefer
moving to familiar territory and causing them to establish
home ranges. This memory of previously visited locations
could be queried by other boids, allowing a boid to ‘track’
other boids that had recently passed it’s location. This could
be considered analogous to tracking the scent of an individual
through the environment. This model only stored a list of
coordinates and the time at which they were visited, as the
environment was modelled as a static set of information rather
than as areas with dynamically changeable properties.
E. Environmental properties
The simulated environment in an IBM need not be a sim-
ple, homogeneous environment. The simulation can include
environmental data, either generated to represent a generic
environment or taken from measurements of a real site. For a
real site, the available data is likely to be available in different
formats and at different resolutions which will need to be
reconciled. In the model implemented in [22], the environment
was divided into a grid of 30m ×30m cells. Each of these
cells (which they referred to as ‘pixels’) was assigned a value
for each of a number of properties, including land cover,
deer population density, road presence and human population
density among other factors. Where the data was available at
a lower resolution, the same value was assigned to all 30m
cells within the areas defined in the lower resolution data.
When each boid in the model made a movement decision,
the environmental values for each were then combined with
a familiarity weighting based on boid’s memory as discussed
above, with the most favourable resulting nearby cell chosen
as the destination for that boid. This method is simple to
implement for properties that can be represented as scalar
fields (For this purpose, a text label applied to an area can
be considered as a scalar). If a vector field can sensibly be
averaged over a cell then these can also be applied in a
similar manner.The properties associated with areas of the
simulation need not be confined to descriptions of the physical
environment, the local ecology can also be included this way.
Food sources and fauna can also be modelled by tracking the
quantity within each area, which can be used by boids as part
of their decision making process. These quantities need not be
static - they can be increased or reduced as appropriate based
on the presence of boids as in [20], or altered over time to
reflect seasonal variations. These properties can even include
prey if the boids being simulated are not thought to cause
migration by the prey over the simulated timescales.
F. Boid interactions
One of the features of individual based models is the ability
of the models to develop emergent behaviours that can mimic
flocking and schooling behaviours observed in nature. It was
observed in [18] that simple rules governing the interac-
tions between boids could give rise to ’Plausible looking’
behaviours, and this concept can be used to try and mimic
the observed behaviour and movement in real species. In
addition to flocking and schooling type behaviours and usage
patterns, mating behaviours and predator/prey interactions can
be investigated, as described in [19].
When interacting with other boids, any information stored
about each boid is potentially available for use by a boid when
deciding how to respond to a target. The information could just
be the position and velocity of the target boid (for flocking and
schooling behaviours) or could include species, gender, age or
other properties of the target. The model used in [22] used
gender as a factor to allow simulated males to track simulated
females while avoiding other males, while the model detailed
in [19] featured both mating behaviours based on gender and
predator/prey dynamics based on the species of the simulated
boids.
G. Device representations
In order to examine the potential effects of marine en-
ergy devices on a population, it is necessary to incorporate
the effects of the device into the simulated environment. If
the simulation permits free movement of boids and detailed
bathymetry then it may be appropriate to include a full 3D
model of the device and any support structures, and their
corresponding effects on local wave climate and currents.
If the simulation has been designed around a grid with
simplified environmental properties, such as in [20], [22] then
the effects of the device can be incorporated into those area
based properties in the appropriate locations. This represents a
computationally simpler method, but doesn’t necessarily take
into account the physical obstacle represented by the device,
which may be particularly relevant for tidal range type devices
or marine energy devices with significant support structures.
As an alternative to the above extremes, it may be possible
to incorporate aspects of the two into a larger model by
splitting the simulation into a number of domains. Domains
further from the device(s) can be represented in the simplified,
cell based representation discussed, with boids then migrating
to/from a full 3D model in the vicinity of the device area.
This could be considered analogous to mesh refinement in the
vicinity of a solid part in CFD type simulations.
When deciding on device representations, the physiological
effects on the species represented by the boids should be
considered and this can be used to guide the representation
used. Considering the example of tidal stream turbines and
adult Killer Whales, work done in [25] showed that the
chance of significant injury for a head-on collision between
the whale and the turbine design discussed. Conversely, work
investigating the effects of a turbine transit on fish shows that
fish transiting a turbine disk can be subject to rapid changes
in pressure which may have an effect on the swim bladder,
which presents a chance for injury or death depending on the
species and turbine design [?].
H. Behavioural rule selection
Deciding upon a set of rules that allow the boids to mimic
a particular species is crucial if an individual based model is
to be of use to developers and regulators. When devising a set
of rules, the rules can be selected based on what we think the
species concerned will be considering, or rules can be selected
based on the correlation of observed behaviour and a selection
of environmental factors thought to be influential [26], [27]. In
[27], 3003 different mathematical models were tested. These
models used 13 recorded environmental variables, varying
from time of day to bed gradient. The models were then ranked
according to their Akaike Weights, which provides a relative
measure of how well a model fits observed or measured data.
The weights assigned only show the strength of each model
relative to the other models being tested. This approach could
be used to select which sets of environmental data to include in
an individual based model, but could also be used to compare
the results of individual based models to observational data.
As an alternative to the statistical top-down approach, the
rules chosen can be devices using existing literature literature
and any site data available. For some species, it may be
possible to perform tests in a lab environment to examine
responses to different conditions [28]. These rules can then
be adjusted based on the results of the models being tested.
Where data is available that the models can be verified against,
this adjustment could be driven by statistical methods as
described above, or by using a genetic algorithm approach
to ‘breed’ a suitable rule set. This would require a method
of encoding rules into a genome (using ‘rule enabled’ flags,
weights to different rules, adjustable numerical values within
specific rules or some combination of these) and a method
of comparing the model results to the field data that can be
used as a fitness function. Improved rule selections can then
be generated based on successful genomes.
I. A more comprehensive model
Taking into account some of the features described above,
we can now outline a more complex model.
1) For parallel simulations:
a) Check if current partitions are suitable.
If so go to step 2
b) Redefine partitions
c) Transfer boids to new partitions
d) Transfer any appropriate environmental data to new
partitions
2) For each boid in the population:
a) Get list of nearby boids
b) Look up local environmental data
c) Check and update current behaviour mode
d) Make decision about movement
e) Update position/Apply acceleration
3) Record positions
4) Update time dependent environmental data
5) Update presence dependent environmental data
6) Handle emigration and immigration
7) Advance simulation clock
8) Repeat from 1 for next time step
In addition to the steps given in section III-A, the boids
investigate the local environment, check for a stored inten-
tion and update it if required before making it’s movement
decision. After each boid has made it’s movements, the
environment is updated to take into account the passge of
time (4) and the actions of the boids, such as feeding and
hunting (5). After this, the simulation clock is advanced and
the process repeated as before.
IV. FUTURE WOR K
There are a number of areas around the United Kingdom
that present viable opportunities for marine energy devices
to be deployed [29]–[31]. One of these locations is being
considered as a study location for an individual based model
of harbour porpoise. The location on the south west coast of
Wales has been studied extensively over a number of years
in order to evaluate it’s suitability for the deployment of
marine energy devices, and a range of data which could be
incorporated into an individual based model.
The data available for the site includes both measured and
calculated sets of data. The bathymetry and coastal geography
are available from measured data. This bathymetry has also
been used to create a computational fluid dynamics model
to calculate the flow of water through the area, with the
bathymetry modified to better define some of the local features
not captured in the measured bathymetry. There are also sets of
ADCP transects available [32], and recordings of underwater
noise [33] in the area. In addition to the data regarding
the physical area, there are also observational records of
local marine mammal populations, including Harbour porpoise
(Phocoena phocoena) and Grey seals (Halichoerus grypus)
over a period of several years.
This data will be used to help develop an individual based
model for Harbour Porpoise behaviour. The behavioural rules
for the model will be informed by existing studies and
compared to visual observations of Harbour Porpoise in the
area. Once a model has been established and compared to
existing observations, the effect of deploying a marine energy
device can be examined by adding a turbine to the model
as described above and comparing the changes in simulated
post-deployment behaviour to any changes observed in the
behaviour of the porpoise at the site.
REFERENCES
[1] A. J. McLane, C. Semeniuk, G. J. McDermid, and
D. J. Marceau, “The role of agent-based models in
wildlife ecology and management,” Ecological Modelling, vol.
222, no. 8, pp. 1544 – 1556, 2011. [Online]. Available:
http://www.sciencedirect.com/science/article/pii/S0304380011000524
[2] “The Infrastructure Planning (Environmental Impact Assessment) Reg-
ulations,” 2009, as amended.
[3] “Council Directive 85/337/EEC of 27 June 1985 on the assessment of
the effects of certain public and private projects on the environment,”
pp. 40–48, o.J. L. 5.7.1985.
[4] “Directive 2011/92/EU of the European Parliament and of the Council
of 13 December 2011 on the assessment of the effects of certain public
and private projects on the environment,” pp. 1–21, o.J. L. 26, 28.1.2012.
[5] Tidal Lagoon Swansea Bay, “Proposed Tidal Lagoon Development
in Swansea Bay, South Wales,” Tidal Lagoon Swansea Bay Ltd.,
Environmental Impact Assessment Scoping Report, October 2012.
[Online]. Available: http://infrastructure.planningportal.gov.uk/projects/
wales/tidal-lagoon- swansea-bay/?ipcsection=folder
[6] “Council Directive 92/43/EEC of 21 May 1992 on the conservation of
natural habitats and of wild fauna and flora, as amended.”
[7] “The Conservation (Natural Habitats, &c.) Regulations,” 1994, as
amended.
[8] S. A. Gauthreaux and J. W. Livingston, “Monitoring bird migration
with a fixed-beam radar and a thermal-imaging camera,” Journal
of Field Ornithology, vol. 77, no. 3, pp. 319–328, 2006. [Online].
Available: http://dx.doi.org/10.1111/j.1557-9263.2006.00060.x
[9] M. G. Barradell, “Fine Scale Use of Ramsey Sound, Pembrokeshire,
West Wales, by Harbour Porpoise (Phocoena phocoena),” BSc Disser-
tation, Coastal & Marine Environment Research Unit, Pembrokeshire
College/University of Glamorgan, 2009.
[10] I. Fairley, M. Broudic, R. Malki, M. R. Willis, I. Horsfall, and I. Masters,
“Results from a multi-disciplinary survey campaign at a tidal stream
site in west wales,” in Proc 4th International Conference on Marine
Research for Environmental sustainability, 2011.
[11] A. Gallus, M. D¨
ahne, U. K. Verfuß, S. Br¨
ager, S. Adler, U. Siebert,
and H. Benke, “Use of static passive acoustic monitoring to assess the
status of the ‘critically endangered’ baltic harbour porpoise in german
waters,” Endangered Species Research, vol. 18, no. 3, pp. 265–278,
October 2012, german Oceanographic Museum.
[12] D. Thompson, “Assessment of Risk to Marine Mammals from
Underwater Marine Renewable Devices in Welsh waters (on behalf
of the Welsh Government), Phase 2: Studies of Marine Mammals
in Welsh High Tidal Waters, Annex 1 Movements and Diving
Behaviour of Juvenile Grey Seals in Areas of High Tidal Energy,”
http://www.marineenergypembrokeshire.co.uk/publications-resources,
July 2012, jER3688.
[13] T. C. Guilford, J. Meade, R. Freeman, D. Biro, T. Evans, F. Bonadonna,
D. Boyle, S. Roberts, and C. M. Perrins, “GPS tracking of the foraging
movements of manx shearwaters puffinus puffinus breeding on skomer
island, wales,” Ibis, vol. 150, no. 3, pp. 462–473, 2008. [Online].
Available: http://dx.doi.org/10.1111/j.1474-919X.2008.00805.x
[14] A. Read and A. Westgate, “Monitoring the movements of harbour por-
poises (Phocoena Phocoena) with satellite telemetry,” Marine Biology,
vol. 130, pp. 315–322, 1997.
[15] G. B. Stenson, S. Benjamins, and D. G. Reddin, “Using bycatch
data to understand habitat use of small cetaceans: lessons from an
experimental driftnet fishery,” ICES Journal of Marine Science: Journal
du Conseil, vol. 68, no. 5, pp. 937–946, 2011. [Online]. Available:
http://icesjms.oxfordjournals.org/content/68/5/937.abstract
[16] R. A. Stillman, A. D. West, R. W. G. Caldow, and S. E.
A. L. D. dit Durell, “Predicting the effect of disturbance on
coastal birds,” Ibis, vol. 149, pp. 73–81, 2007. [Online]. Available:
http://dx.doi.org/10.1111/j.1474-919X.2007.00649.x
[17] A. West and S. l. dit Durell, “Severn estuary bird food monitoring. phase
1,” Centre for Ecology and Hydrology, Natural Environment Research
Council, Tech. Rep., October 2006.
[18] C. W. Reynolds, “Flocks, herds and schools: A distributed behavioral
model,” ACM SIGGRAPH Computer Graphics, vol. 21, no. 4, pp.
25–34, 1987. [Online]. Available: http://www.red3d.com/cwr/boids/
[19] X. Tu and D. Terzopoulos, “Artificial fishes: Physics, locomotion,
perception, behavior,” in Proceedings of the 21st annual conference on
Computer graphics and interactive techniques. ACM, 1994, pp. 43–50.
[20] H. Saarenmaa, N. D. Stone, L. Folse, J. Packard, W. Grant, M. Makela,
and R. Coulsen, “An artificial-intelligence modeling approach to sim-
ulating animal habitat interactions,” Ecological Modelling, vol. 44, pp.
125–141, 1988.
[21] G. Vigueras, M. Lozano, J. Ordu˜
na, and F. Grimaldo, “A comparative
study of partitioning methods for crowd simulations,” Applied Soft Com-
puting, vol. 10, no. 1, pp. 225–235, January 2010. [Online]. Available:
http://www.sciencedirect.com/science/article/pii/S156849460900088X
[22] P. C. Cramer and K. M. Portier, “Modeling Florida panther movements
in response to human attributes of the landscape and ecological settings,”
Ecological Modelling, vol. 140, pp. 51–80, 2001.
[23] R. J. H. Herbert, J. Willis, E. Jones, K. Ross, R. H¨
ubner,
J. Humphreys, A. Jensen, and J. Baugh, “Invasion in tidal
zones on complex coastlines: modelling larvae of the non-native
Manila clam, Ruditapes philippinarum, in the UK,” Journal of
Biogeography, vol. 39, pp. 585–599, 2012. [Online]. Available:
onlinelibrary.wiley.com/doi/10.1111/j.1365-2699.2011.02626.x/pdf
[24] J. Alderman and S. A. Hinsley, “Modelling the third dimension: Incor-
porating topography into the movement rules of an individual-based
spatially explicit population model,” Ecological Complexity, vol. 4,
no. 4, pp. 169–181, December 2007.
[25] T. Carlson, B. Watson, J. Elster, A. Copping, M. Jones, M. Watkins,
R. Jepsen, and K. Metzinger, “Assessment of strike of adult killer
whales by an openhydro tidal turbine blade,” Pacific Northwest National
Laboratory, Tech. Rep. PNNL-22041, November 2012.
[26] S. Katsanevakis, “Modelling fish growth: Model selection, multi-model
inference and model selection uncertainty,” Fisheries Research, vol. 81,
pp. 229 – 235, 2006. [Online]. Available: http://www.sciencedirect.
com/science/article/pii/S0165783606002645
[27] S. Isojunno, J. Matthiopoulos, and P. G. H. Evans, “Harbour
porpoise habitat preferences: robust spatio-temporal inferences from
opportunistic data,” MARINE ECOLOGY PROGRESS SERIES, vol.
448, pp. 155–170, 2012. [Online]. Available: http://www.int-res.com/
abstracts/meps/v448/p155-170/
[28] V. I. Nikora, J. Aberle, B. J. F. Biggs, I. G. Jowett, and J. R. E. Sykes,
“Effects of fish size, time-to-fatigue and turbulence on swimming
performance: a case study of galaxias maculatus,” Journal of Fish
Biology, vol. 63, no. 6, pp. 1365–1382, 2003. [Online]. Available:
http://dx.doi.org/10.1111/j.1095-8649.2003.00241.x
[29] M. R. Willis, I. Masters, S. Thomas, R. Gallie, J. Loman, A. Cook,
R. Ahmadian, R. Falconer, B. Lin, G. Gao, M. Cross, N. Croft,
A. J. Williams, M. Muhasilovic, I. Horsfall, R. Fidler, C. Wooldridge,
I. Fryett, P. Evans, T. O’Doherty, D. O’Doherty, and A. Mason-Jones,
“Tidal turbine deployment in the bristol channel: a case study,” in
Proceedings of the Institution of Civil Engineers - Energy, vol. 163,
no. 900024, August 2010, pp. 93–105.
[30] ABPmer, “Atlas of UK Marine Renewable Energy Resources: Technical
Report,” Department for Business, Enterprise & Regulatory Reform,
Tech. Rep., March 2008.
[31] I. Fairley, S. Neill, T. Wrobelowski, M. R. Willis, and I. Masters,
“Potential array sites for tidal stream electricity generation off the
pembrokeshire coast,” in Proceedings of the 9th European Wave and
Tidal Energy Conference (EWTEC), 2011.
[32] I. Fairley, P. Evans, C. Wooldridge, M. Willis, and I. Masters,
“Evaluation of tidal stream resource in a potential array area via
direct measurements,” Renewable Energy, vol. 57, no. 0, pp. 70 – 78,
2013. [Online]. Available: http://www.sciencedirect.com/science/article/
pii/S0960148113000499
[33] M. R. Willis, M. Broudic, C. Haywood, I. Masters, and S. Thomas,
“Measuring underwater background noise in high tidal flow environ-
ments,” Renewable Energy, vol. 49, pp. 255–258, January 2013.
