Mortensen - 2012 Thesis

Agent-Based Modelling of Short-Term Juvenile Bull Shark
Movement in a Semi-Enclosed Gold Coast Estuary
Author
Mortensen, Jonas Brandi
Published
2012
Thesis Type
Thesis (Masters)
School
Griffith School of Environment
DOI
https://doi.org/10.25904/1912/1201
Copyright Statement
The author owns the copyright in this thesis, unless stated otherwise.
Downloaded from
http://hdl.handle.net/10072/367675
Griffith Research Online

https://research-repository.griffith.edu.au
Agent-based modelling of short-term juvenile bull shark
movement in a semi-enclosed Gold Coast estuary
Jonas Brandi Mortensen
B.Sc. (University of Copenhagen)
School of Environment
Science, Environment, Engineering and Technology
Griffith University
Submitted in fulfilment of the requirements of the degree of
Master of Philosophy
December 2011
I
STATEMENT OF ORIGINALITY
This work has not previously been submitted for a degree or diploma in any university. To the
best of my knowledge and belief, the thesis contains no material previously published by
another person except where due reference is made in the thesis itself.
....................................................................
Jonas Brandi Mortensen
14th December 2011
II
ACKNOWLEDGEMENTS
First and foremost I would like to extend my warmest thanks to my Principal Supervisor,
Professor Joe Lee, for believing in me and allowing me to undertake this research project
under his supervision. Thank you for your guidance and support over the past two years,
especially here in the last few dramatic months. Also thanks to Dr Guy Castle for acting as
my associate supervisor, and guiding me through the tricky waters of my Confirmation paper
back in the early days. Special thanks goes out to Jonathan Werry for helping me establish
this project with the Australian Rivers Institute, while also helping me with the shark catching
effort and supplying me with the necessary equipment needed for the acoustic tracking
campaign.
My most sincere thanks goes out to the DHI for supplying me with a free MIKE software
license over the course of my study, but most importantly I would like to thank Flemming T.
Hansen for all the ABM support he has provided me over the past two years, and for allowing
me to stay in his office for three months while teaching me how to use the MIKE ABM
module. I would also like to extend a special thanks to Michael Pothoff for his technical ABM
troubleshooting, and most importantly; for fixing that bug in the software that prevented me
from completing ABM simulations! Furthermore I would like to thank Anders Erichsen and
Thomas Uhrenholdt, as well as the many other DHI employees who helped me with technical
advice, while also making my stay in Hørsholm a pleasant one.
My warmest thanks and gratitude goes out to the Queensland Urban Fish Habitat
Management Research Program and the Gold Coast City Council for funding part of my
research, as well as the Danish State Government for sponsoring my tuition fees and living
costs for the duration of this project. Also I would like to direct a special thanks to the Bureau
of Meteorology (BOM), the Queensland Department of Resource Management (DERM) as
well as the Environmental Ecosystem Health Program (EHMP) of Southeast Queensland for
providing me with essential input data free of charge.
Furthermore, I would like to direct my warmest thanks to the following people for aiding me
with the many various aspects of this research project:
Associate Professor Charles Lemckert - For allowing me to use two very expensive
Acoustic Doppler Current Profilers free of charge. It goes without saying that without
III
your help the available validation data of the hydrodynamic model would have been
quite lacking.
Paul Maxwell - For helping me with the historical EHMP data, and assisting me in the
shark tracking campaign during the latest of hours, while being eaten alive by
mosquitoes. Also thanks for the many good laughs in lab.
Johan Gustafson - For his help with constructing a mounting frame for the two
ADCP’s, as well as the actual deployment and retraction of the instrument, plus his
ongoing moral support throughout the thesis.
Geoffrey Turner, Ian Underhill and Malcolm Duncan - For allowing me into the
Engineering bay, as well as provide excellent guidance and assistance in the
construction of a mountable hydrophone rig for the Portunus.
Joshua Reinke - For continuing the water quality campaign during my time away from
the Gold Coast, while aiding me on several occasions with the actual shark tracking.
Simon Kerville and Alan Richards - For tirelessly helping me out with essential lab
equipment for chlorophyll analysis, as well as saving me from trouble in the field.
All the many volunteers who have been helping me with the shark tracking campaign
and other assorted field work; Ciaran Morris, Alf Okkels Jakobsen, Sophie Olsson-
Pons, Thomas Baatz Andersen, Sarah Richmond, Carl Brown-Kenyon, Barbro Haug-
land, Sameer Shah and Louise Pointon.
Last but not least, I would like to thank my brother, Simon Brandi Mortensen, for providing
me with a place to call home these past three years that I have spent in Australia, while also
guiding me through the many pitfalls of hydrodynamic modelling and Matlab scripting. I can
honestly say that your constant moral support has been the glue that held me together through
the hardest of times, and I will be forever grateful to you.
Finally, my warmest of thanks goes out to my family and friends back home in Denmark for
believing and supporting me throughout this entire venture. Your continuous motivational
support has been invaluable and keeps pushing me forever forward.
IV
ABSTRACT
This project investigated the value and future potential of a coupled Eulerian-Lagrangian
agent-based modelling approach as an alternative method of investigating the movement and
habitat use of juvenile bull shark Carcharhinus leucas in small peri-urban estuaries. Through
the use of the MIKE21 modelling suite (DHI), a depth-averaged two-dimensional hydro-
dynamic model was developed and implemented as a means to capture the spatio-temporal
variation in hydrodynamics of the semi-enclosed Tallebudgera Creek estuary. This system
provides a suite of habitats comprising artificial residential canals, polyhaline and brackish
creek sections in a peri-urban setting. The hydrodynamic model served as the dynamic
foundation of a spatially heterogeneous agent-based model (ABM) developed for juvenile C.
leucas. The movement formulation of juvenile C. leucas was represented as a kinesis search
for optimal conditions, while a random walk model served as a control.
The hydrodynamic model performed satisfactorily in terms of capturing the variations of key
physical conditions of Tallebudgera Creek. Modelled values of surface elevation and flow
dynamics were in good agreement with measured data sets. Simulated mean levels of salinity
and temperature were likewise in good agreement with measured means; however, model
analysis revealed a high sensitivity to increased freshwater influxes, and a delay in model
response time.
Three neonate and juvenile individuals of C. leucas were captured and attached with acoustic
tags for tracking of movement in Tallebudgera Creek. Short-term continuous tracks of a
juvenile C. leucas were successfully collected as a means to relate observed movement to out-
puts of the hydrodynamic model and measurements of water quality, while consecutive data-
points of animal position served as validation data for the agent-based model.
Analysis of C. leucas track data revealed a high site preference for the middle reach of the
system over the course of the tracking campaign, even during periods when salinity levels
were < 1 PSU. However, an avoidance of high salinities > 27 PSU was evident. Significant
movement of the animal in a downriver direction only occurred after a period of increased
flow velocities and turbidity, suggesting that these parameters may play an important role in
directing shark movement in conjunction with salinity.
V
The agent-based models in their current developmental stage performed unsatisfactorily in
capturing observed movement, and their predictive ability was generally poor. The current
ABM formulation of C. leucas movement is therefore deemed insufficient to capture the
observed pattern of behaviour. However, unforseen technical difficulties originating from the
narrow and shallow nature of the Tallebudgera Creek system prevented a full assessment of
the ABM results.
Despite current technical issues that were impracticable to be resolved under the available
timeframe, this study represents a first attempt to construct and implement agent-based
modelling to investigate bull shark movement and habitat use in a spatially and temporally
dynamic hydrologic environment. It is predicted that once these technical difficulties are
overcome, agent-based modelling as a research tool holds great promise for future
investigation of the habitat ecology of C. leucas to benefit its conservation and management.
VI
TABLE OF CONTENTS
Statement of originality ............................................................................................................. II

Acknowledgements .................................................................................................................. III
Abstract ..................................................................................................................................... V
List of figures ........................................................................................................................... XI
List of tables .......................................................................................................................... XVI
Ethics approval ................................................................................................................... XVIII
Chapter I: General introduction and research aims ............................................................. 1

1.1. General introduction ........................................................................................................ 2
1.2 Research questions .......................................................................................................... 4
1.3 Research aims .................................................................................................................. 5
1.4 Hypothesis ....................................................................................................................... 5
Chapter II: Water quality assessment of Tallebudgera creek ............................................. 6

2.1 Introduction .......................................................................................................................... 7
2.1.1 Aims & purpose ....................................................................................................... 9
2.2 Methodology ........................................................................................................................ 9
2.2.1 Sampling design ....................................................................................................... 9
2.2.2 Temperature ........................................................................................................... 11
2.2.3 Salinity ................................................................................................................... 11
2.2.4 pH ........................................................................................................................... 11
2.2.5 Turbidity ................................................................................................................ 11
2.2.6 Dissolved Oxygen .................................................................................................. 12
2.2.7 Chlorophyll ............................................................................................................ 12
2.2.8 Sonde calibration & measurement techniques ....................................................... 13
2.2.9 Data analysis .......................................................................................................... 13
2.3 Results ................................................................................................................................ 14
2.3.1 Temperature ........................................................................................................... 14
2.3.2 Salinity ................................................................................................................... 16
2.3.3 Dissolved Oxygen .................................................................................................. 17
VII
2.3.4 pH ........................................................................................................................... 19
2.3.5 Turbidity ................................................................................................................ 21
2.3.6 Chlorophyll a ......................................................................................................... 22
2.4 Discussion .......................................................................................................................... 23
Chapter III: Hydrodynamic modelling of Tallebudgera Creek ........................................ 26

3.1 Introduction ........................................................................................................................ 27
3.1.1 Aims & purpose ........................................................................................................... 28
3.2 Methodology ...................................................................................................................... 28
3.2.1 Equational Framework ........................................................................................... 28
3.2.2 Transport equations ................................................................................................ 29
3.2.3 Heat exchange ........................................................................................................ 30
3.2.4 Model domain ........................................................................................................ 32
3.2.5 Input parameters..................................................................................................... 34
3.2.5.1 Offshore boundary data ...................................................................................... 35
3.2.5.2 Freshwater discharge .......................................................................................... 35
3.2.5.3 Meteorological data ............................................................................................ 35
3.2.5.5 Precipitation ....................................................................................................... 36
3.2.5.6 Wave radiation stress ......................................................................................... 36
3.2.5.7 Initial conditions ................................................................................................. 37
3.2.6 Calibration parameters ........................................................................................... 37
3.2.7 Validation data ....................................................................................................... 39
3.2.7.1 Historical data .................................................................................................... 39
3.2.7.2 ADCP data.......................................................................................................... 39
3.2.7.3 Temperature/Salinity data .................................................................................. 41
3.2.8 Calibration & data analysis .................................................................................... 42
3.3 Results .......................................................................................................................... 43
3.3.1 Surface elevation validation ................................................................................... 43
3.3.2 Flow velocity validation ........................................................................................ 45
3.3.3 Salinity validation .................................................................................................. 48
3.3.4 Temperature validation .......................................................................................... 53
3.4 Discussion .......................................................................................................................... 58
VIII
Chapter IV: Short-term movement of a juvenile bull shark in Tallebudgera Creek ...... 65
4.1 Introduction ........................................................................................................................ 66
4.1.1 Aims & purpose ........................................................................................................... 68
4.2 Methodology ...................................................................................................................... 68
4.2.1 Catching and tagging ............................................................................................. 68
4.2.2 Acoustic telemetry and tracking methods .............................................................. 70
4.2.3 Hydrophone calibration and shark position ........................................................... 71
4.2.4 MIKE21FM data .................................................................................................... 73
4.2.5 Data Analysis ......................................................................................................... 73
4.3 Results ................................................................................................................................ 75
4.3.1 Track I.......................................................................................................................... 76
4.3.2 Track II ........................................................................................................................ 79
4.3.3 Track III ....................................................................................................................... 83
4.3.4 Track IV ....................................................................................................................... 87
4.3.5 Track V ........................................................................................................................ 91
4.3.6 Track VI....................................................................................................................... 95
4.3.7 Collective results ......................................................................................................... 99
4.4 Discussion ........................................................................................................................ 105
Chapter V: Agent-based modelling of a juvenile bull shark on a short spatio-temporal

scale ........................................................................................................................................ 110
5.1 Introduction ...................................................................................................................... 111
5.1.2 Aims & purpose ......................................................................................................... 112
5.2. Methodology ............................................................................................................... 113
5.2.1 The coupled Eulerian-Lagrangian model framework .......................................... 113
5.2.2 The random walk model ...................................................................................... 114
5.2.2.1 Random walk movement equations .................................................................... 115
5.2.3 The kinesis model ................................................................................................ 115
5.2.3.1 Kinesis movement equations .............................................................................. 116
5.2.4 Model structure .................................................................................................... 120
5.2.5 Model setup & calibration effort.......................................................................... 121
5.2.6 Data analysis ........................................................................................................ 124
5.3 Results .............................................................................................................................. 126
5.3.1 Track I........................................................................................................................ 126
IX
5.3.2 Track II ...................................................................................................................... 129
5.3.3 Track III ..................................................................................................................... 134
5.3.4 Track IV ..................................................................................................................... 138
5.3.5 Track V ...................................................................................................................... 141
5.3.6 Track VI..................................................................................................................... 143
5.4 Discussion ........................................................................................................................ 145
5.4.1 Technical restrictions ................................................................................................. 145
5.4.2 Theoretical limitations ............................................................................................... 146
5.4.3 Concluding remarks ................................................................................................... 148
Chapter VI: General discussion and future directions ..................................................... 150

6.1 Summary .......................................................................................................................... 151
6.2 Evaluation of the hydrodynamic model ........................................................................... 151
6.2.1 Ecological modelling ................................................................................................. 152
6.3 Agent-based modelling .................................................................................................... 152
6.3.1 Future modelling directions ....................................................................................... 154
6.3.2 Possibilities for conservation management ............................................................... 155
References .............................................................................................................................. 157
Appendix I............................................................................................................................. 162
Temporal validation of Salinity and Temperature across all stations ................................ 162
Spatial validation of Salinity and Temperature across all sampling days .......................... 167
DVD Appendix...................................................................................................................... 179
X
LIST OF FIGURES
Figure 2.2.1: Satellite imagery depicting the lower to middle reaches of the Tallebudgera Creek
estuary, and the adjoining artificial canals. Plotted using a MGA-56 projection. Data source: Google
Earth. .......................................................................................................................................................................................8
Figure 2.2.2: The location of water quality stations throughout Tallebudgera Creek; see colour
legend for station type ..................................................................................................................................................... 10
Figure 2.3.1: A) Seasonal variation in temperature throughout the course of 2010. B) Spatial
variation in temperature throughout the course of 2010. The data set depicted is the assimilated data
set of EHMP measurements along with measurements made by this study ................................................... 15
Figure 2.3.2: A contour plot of the spatiotemporal variation of temperature throughout the sampling
period. The data set depicted contains only the measurements conducted by this study ........................... 15
Figure 2.3.3: A) Seasonal variations in salinity throughout the course of 2010. Note the abrupt
fluctuations between sample days. B) Spatial variation in salinity throughout the course of 2010,
showing a general decline of salinity proportional to distance from river mouth. Note the higher level
of variation, which is indicated by the increased width between the 20th and 80th percentile, as the
distance from the river mouth increases. The data set depicted is the assimilated data set of EHMP
measurements along with measurements made by this study. ............................................................................ 16
Figure 3.3.4: A contour plot of the spatiotemporal variations of salinity throughout the sampling
period. The data set depicted contains only the measurements conducted by this study. .......................... 17
Figure 2.3.5: A) Seasonal variations in dissolved oxygen throughout the sampling period. Note the
relative small difference between the 20th and 80th percentile levels across sampling days. B) Spatial
variations in dissolved oxygen across the tidal-dominated part of the system. Note the slight increase
of dissolved oxygen concentrations with increased distance to the river mouth. The data set depicted is
the assimilated data set of EHMP measurements along with measurements made by this study. .......... 18
Figure 2.3.6: A contour plot of the spatiotemporal variations of dissolved oxygen throughout the
sampling period. The data set depicted contains only the measurements conducted by this study. ....... 19
Figure 2.3.7: A) The seasonal variations of pH throughout the sampling period. Note how the
fluctuations in pH tend to follow the same general trend as salinity (Figure 2.3.3A). B) Spatial
variations in pH throughout the course of 2010, showing a general decline of pH levels proportional
to distance from the river mouth. The data set depicted is the assimilated data set of EHMP
measurements along with measurements made by this study. ............................................................................ 20
Figure 2.3.6: A contour plot of the spatiotemporal variation of dissolved oxygen throughout the
sampling period. The data set depicted contains only the measurements conducted by this study. ....... 20
Figure 2.3.7: A) The seasonal variations of turbidity throughout the sampling period. B) Spatial
variations in turbidity throughout the course of 2010, showing a general increase of turbidity levels
proportional to distance from the river mouth. Note that the scale of the y-axis differs between A) and
B). The data set depicted is the assimilated data set of EHMP measurements along with measurements
made by this study. ........................................................................................................................................................... 21
Figure 2.3.8: A contour plot of the spatiotemporal variations of turbidity throughout the sampling
period. The data set depicted contains only the measurements conducted by this study. .......................... 22
Figure 2.3.8: A) Seasonal variations in chlorophyll a throughout the sampling period of 2010. Note
that chlorophyll a concentration only exceeded 20 μg/l on one occasion. B) Spatial variations in
chlorophyll a throughout the course of 2010, showing a weak trend of increased chlorophyll a
concentrations upstream. The data set depicted is the assimilated product of EHMP measurements
along with measurements made by this study. ......................................................................................................... 23
XI
Figure 3.2.1: Illustration showing the bathymetry of the Model domain, plotted in a MGA-56
projection. The area inside the black box is the part of the non-tidal dominated freshwater zone that is
included in the model setup (see main text for further explanation). The red box marks the area shown
in Figure 3.2.3 below....................................................................................................................................................... 33
Figure 3.2.2: An example of the flexible mesh of the model. The green markers each indicate a smaller
sub-area within the mesh. The figure is plotted using MGA-56 projection coordinates (unit: meters).
................................................................................................................................................................................................ 34
Figure 3.2.3: Model domain of the created offshore Spectral Wave model. The red box marks the
Tallebudgera Creek entrance. The figure is plotted using MGA-56 projection coordinates (unit:
meters). ................................................................................................................................................................................. 37
Figure 3.2.4: The deployment locations of ADCP's and CTD recorders in the middle and upper
reaches of the system (MR & UR). Blue stations mark the 2007 GCCM stations. The figure is plotted
using MGA-56 projection coordinates (unit: meters). .......................................................................................... 40
Figure 3.2.5: Picture of the constructed wooden frame with the ADCP mounted on it. All metal used
was either lead or stainless steel in order to reduce any effect on the internal compass. ......................... 41
Figure 3.3.1: Modelled surface elevation (red line) plotted against measured values (blue line) during
the period of 29/3-2007 to 8/4-2007 at the MR-2007 calibration point.......................................................... 43
the period of 24/9-2010 to 19/10-2010 at the MR-ADCP calibration point. ................................................. 44
the period of 24/9-2010 to 19/10-2010 at the UR-ADCP calibration point. ................................................. 45
Figure 3.3.4: Simulated current speed (red line) plotted against measured values (blue line) during
the period of 24/9-2010 to 19/10-2010 at the UR-ADCP calibration point. ................................................. 46
the period of 24/9-2010 to 27/9-2010 at the UR-ADCP calibration point. .................................................... 46
Figure 3.3.6: Simulated current directions (red line) plotted against measured values (blue line)
during the period of 24/9-2010 to 19/10-2010 at the UR-ADCP calibration point. ................................... 47
the period of 24/9-2010 to 19/10-2010 at the MR-ADCP calibration point. ................................................. 47
Figure 3.3.8: Simulated current directions (red line) plotted against measured values (blue line)
during the period of 24/9-2010 to 19/10-2010 at the UR-ADCP calibration point. ................................... 48
Figure 3.3.9: Simulated salinity (red line) plotted against measured values (blue line) during the
period of 20/1-2010 to 1/2-2010 at the UR-CTD calibration point.................................................................. 49
period of 11/2-2010 to 29/3-2010 at the MR-CTD calibration point............................................................... 50
period of 11/2-2010 to 29/3-2010 at the UR-CTD calibration point. .............................................................. 50
Figure 3.3.12: (Left) Seasonal variation in station-averaged salinity (blue line) plotted against
corresponding simulated values (red line). (Right) Annually averaged salinity for each measuring
station in relation to river mouth distance, plotted against corresponding simulated values. EHMP
data is omitted from the two measured data sets. ................................................................................................... 51
Figure 3.3.13: Simulated temperature (red line) plotted against measured values (blue line) during
the period of 20/1-2010 to 1/2-2010 at the MR-CTD calibration point. ......................................................... 54
the period of 20/1-2010 to 1/2-2010 at the UR-CTD calibration point. ......................................................... 54
the period of 11/2-2010 to 29/3-2010 at the MR-CTD calibration point........................................................ 55
XII
Figure 3.3.17: (Left) Seasonal variation in station-averaged temperature (blue line) plotted against
corres-ponding simulated values (red line). (Right) Annually averaged temperature for each
measuring station in relation to river mouth distance, plotted against corresponding simulated values.
EHMP data is omitted from the two measured data sets ..................................................................................... 56
Figure 3.4.1: Simulated current speeds at a model bottleneck during low tide (top), incoming tide
(middle) and high tide (bottom). The bottleneck is located ~3 km upriver from the river mouth, while
the first ADCP calibration point is located ~3 km further upriver from the bottleneck. ........................... 60
Figure 4.2.1: Picture of the deployed gill net. Due to the narrow nature of the system, the gill net was
often capable of stretching almost across the full width of the estuary........................................................... 69
Figure 4.2.2: Juvenile bull shark suspended in the specially designed harness, while water flow across
the gills was maintained by directing the boat against the current. Note the combined acoustic/roto-
tag attached to the shark’s dorsal fin (red arrow). ................................................................................................ 70
Figure 4.2.3: The resulting surface plots from each calibration. Note the clear differences in distances
at maximum signal strength and maximum gain (16 and 36 respectively), between the three
calibrations. See main text for further information on the associated accuracy levels of each
calibration. .......................................................................................................................................................................... 72
Figure 4.3.1: Recorded animal locations during the track undertaken on the 28/01-2011, plotted using
an MGA-56 coordinate projection. See main text for track details. An animated movement path is
available for illustration in the DVD appendix, filed as "Track_1_28012010" in the Track_Videos
subfolder. ............................................................................................................................................................................. 77
Figure 4.3.2: Proportion of time spent at various temperature (top left), salinity (top right), current
speed regimes (middle left), swimming speed (middle right), depth (bottom left) and riverbank ratio
regimes (bottom right) for the track conducted on the 28/01-2010. ................................................................. 78
Figure 4.3.3: Proportion of time spent at various dissolved oxygen (top left), pH (top right), turbidity
(bottom left) and animal heading regimes (bottom right) for the track conducted on the 28/01-2010. 79
subfolder. ............................................................................................................................................................................. 81
speed (middle left), swimming speed (middle right), depth (bottom left) and riverbank ratio regimes
(bottom right) for the track conducted on the 03/02-2010. ................................................................................. 82
subfolder. ............................................................................................................................................................................. 85
Figure 4.3.10: Recorded animal locations during the track undertaken on the 06/03-2011, plotted
using an MGA-56 coordinate projection. See main text for track details. An animated movement path
is available for illustration in the DVD appendix, filed as "Track_4_06032010" in the Track_Videos
subfolder. ............................................................................................................................................................................. 89
XIII
(bottom left) and animal heading (bottom right) for the track conducted on the 06/02-2010. ................ 91
subfolder. ............................................................................................................................................................................. 93
subfolder. ............................................................................................................................................................................. 97
Figure 4.3.19: Total amount of recorded animal locations during all six track undertaken from the
28/01-2010 to the 21/03-2011, plotted using an MGA-56 coordinate projection. Note the
concentration of registrations in the middle reach. See main text for explanatory details. ................... 102
Figure 4.3.20: Total proportion of time spent at various temperature (top left), salinity (top right),
current speed (middle left), swimming speed (middle right), depth (bottom left) and riverbank ratio
regimes (bottom right) for the track conducted from the 28/01/2010 to the 21/03/2010. ....................... 103
(bottom left) and animal heading (bottom right) for the track conducted on the 20/03-2010. ............. 104
Figure 4.3.22: Frequency of animal heading relative to current direction during below-mean flow
velocities (red), and above-mean flow velocities (blue). ................................................................................... 104
Figure 4.4.1: Mean current speeds as predicted by the HD model, during the tracking campaign
period, spanning from 28/01/2010 to 21/03/2010. Note the patches of reduced flow velocities within
the middle reach area (marked by red arrows).................................................................................................... 107
Figure 4.4.2: Mean salinity as predicted by the HD model, during the tracking campaign period,
spanning from 28/01/2010 to 21/03/2010. Note how the area normally utilised by the animal (marked
by the black box) lies within the 7-14 PSU range. .............................................................................................. 108
Figure 5.2.1: Two-dimensional view (longitudinal and latitudinal) of a shark’s sensory range. The
sensory range is independent of the Eulerian model grid, and is often estimated from the body length
of the animal and model timestep. Adapted from Goodwin et al. (2006)..................................................... 114
Figure 5.2.2: Plot A: and plotted against the variable X with optimum ,
and and . Plot B: and plotted against the variable X
with optimum , and and ................................................................. 119
XIV
Figure 5.2.3: The weighted functional response of (left) and (middle) plotted
separately and combined (right). The optimum values depicted ( , and
) are for illustration purposes only. .................................................................................................................... 120
Figure 5.2.4: The artificial channel forcing shown for the middle reach of the model domain. Should
an agent move more than 10 metres away from the channel forcing, it will automatically shift its
heading towards the channel forcing in the next timestep................................................................................ 123
Figure 5.2.5: Simplified schematic of the middle reach area, highlighting the essence of the proposed
distance estimation methodology. Note the potential for a huge bias in the Euclidian distance
estimation (as illustrated), compared to distance estimation using the transect line as a reference.. 125
Figure 5.3.1: The best-fitted simulated agent locations as predicted by the KSTC-NC model vs.
observed animal locations throughout the track conducted on 28/01 – 2010. Displayed agent locations
correspond to the time of each animal registration, while agent locations in the time between animal
registrations are not displayed. ................................................................................................................................. 128
Figure 5.3.2: The start (point of release) and end locations of simulated agents predicted by the
KSTC-NC model vs. the first registered animal location on the track conducted on 28/01 – 2010.
Displayed start/end agent locations correspond to the specific agents that make it through the
simulation without getting stuck on a land value. The lack of release points in the upper reach is due
to the fact that agents released in this section did not make it through the entire simulation. ............. 129
Figure 5.3.3: The best-fitted simulated agent locations as predicted by the KSTC model vs. observed
animal locations throughout the track conducted on 03/02 – 2010. Note how the simulated agent
follows the tide downriver........................................................................................................................................... 131
Figure 5.3.4: The best-fitted simulated agent locations as predicted by the KSTC-NC model vs.
observed animal locations throughout the track conducted on 03/02 – 2010. Note how the agent has
very limited longitudinal travel. ................................................................................................................................ 131
Figure 5.3.4: KSTC-model results (median, 10th and 90th percentile values) of distance (top left), depth
(top right), salinity (middle left), temperature (middle right), swimming velocity (bottom left) and
current speed (bottom right) versus corresponding track values (black line) for the 3 rd of February
2010 plotted on a temporal axis................................................................................................................................ 132
Figure 5.3.5: The start (point of release) and end locations of simulated agents predicted by the KS
model vs. the first registered animal location on the track conducted on 03/02 – 2010. Displayed
start/end agent locations correspond to the specific agents that made it through the simulation without
getting stuck on a land value...................................................................................................................................... 134
Figure 5.3.5: The best-fitted simulated agent locations predicted by the KS model vs. observed animal
locations throughout the track conducted on 03/02 – 2010. Note the similarity between the best-fitted
movement locations predicted by the KS model versus the RW model depicted in Figure 5.3.6. ........ 135
Figure 5.3.6: The best-fitted simulated agent locations predicted by the KS model vs. observed animal
locations throughout the track conducted on 03/02 – 2010. Note the similarity between the best-fitted
movement locations predicted by the RW model versus the KS model depicted in Figure 5.3.5. ........ 136
Figure 5.3.7: KS-model results (median, 10th and 90th percentile values) of distance (top left), depth
(top right), salinity (middle left), temperature (middle right), swimming velocity (bottom left) and
current speed (bottom right) versus corresponding track values (black line) for the 27 th of February
2010 plotted on a temporal axis................................................................................................................................ 137
XV
Figure 5.3.9: The best-fitted simulated agent locations predicted by the KSTC-ZS model vs. observed
animal locations throughout the track conducted on 06/03 – 2010. ............................................................. 139
LIST OF TABLES
Table 3.2.1: The values and user-specified settings of the enabled modules in the final model setup.. 38
Table 3.3.1: Calculated results of the Quality index for each measuring station over the course of
2010 vs. corresponding simulated values. See Appendix I for the plotted data sets. .................................. 52
Table 3.3.2: Calculated results of the Quality index for each sampling day in relation to distance from
the river mouth vs. corresponding simulated values. See Appendix I for the plotted datasets. ............... 52
Table 3.3.3: Calculated results of the Quality index for each measuring station over the course of
2010 vs. corresponding simulated values. See Appendix I for the plotted data sets. .................................. 57
Table 3.3.4: Calculated results of the Quality index for each sampling day in relation to distance from
the river mouth vs. corresponding simulated values. See Appendix I for the plotted data sets. .............. 57
Table 4.3.1: Details of the tagged sharks. Note that MN1 (male neonate), despite being of greater
length than FN1 (female neonate), is a whole kilogram lighter than his female counterpart. ................ 75
Table 5.2.1: The model calibration parameters and their final values adopted in the three types of
movement models. .......................................................................................................................................................... 121
Table 5.3.1: Primary model results from the five applied model templates for the simulation
covering the track that was conducted on the 28/01 - 2010. RW = Random Walk Model, KS =
Kinesis Search for Optimal Salinity, KSTC = Kinesis Search for optimal salinity, temperature and
current flow, KSTC-NC = KSTC with the contribution of current flow on agent movement removed,
KSTC-ZS = KSTC with an optimum salinity of 0 PSU. ....................................................................................... 127
covering the 12-hr period prior to the track conducted on 28/01 - 2010. .............................................. 129
covering the track that was conducted on 03/02 - 2010. .............................................................................. 130
covering the 12-hr period prior to the track that was conducted on 03/02 - 2010. ............................ 133
XVI
XVII
ETHICS APPROVAL
All work associated with catching, handling/tagging and tracking of live bull sharks were
approved by the Animal Ethics Committee of Griffith University (ENV/1709/AEC 2010).
The fishing effort in this study was conducted under the Queensland Department of Fisheries
General Fisheries Permit No. 90306
XVIII
CHAPTER I
GENERAL INTRODUCTION AND RESEARCH AIMS
1
1.1. GENERAL INTRODUCTION
Nearly 90% of Australia’s population lives within 100 km of the coastline, and the coastal
population in many countries are currently growing at a fast rate (Martínez et al. 2007). The
increased population growth and urbanisation has created a significant pressure on the
various, and often fragile, coastal environments through the alteration of hydrology and
sedimentation, with increased nutrient and pollution loads from concentrated run-off in
urbanised areas with a high proportion of impervious surfaces (Lee et al. 2006, Halpern et al.
2008). Significant proportions of the natural habitats on urbanising coasts and estuaries have
been replaced by artificial waterways such as canals, with > 65% of global seagrass and
wetland habitat destroyed by human impact (Lotze et al. 2006). The bull shark Carcharhinus
leucas is one of the few shark species that relies on riverine environments during specific
stages of its life cycle (Pillans et al. 2005, Werry et al. 2011) and might therefore be
particularly vulnerable to environmental changes due to urbanisation of coastal environments
(Martin 2005).
Several studies have documented that C. leucas utilises riverine environments as a nursery for
neonate and juveniles (Pillans et al. 2005, Simpfendorfer et al. 2005, Werry et al. 2011).
While sub-adults and adults move much more freely in between the coastal and estuarine
environments, neonate and juvenile bull sharks tend to stay in the river system (Werry 2010).
Thus, to study the impacts of increased urbanisation on the life cycle of C. leucas, one must
first understand how juvenile C. leucas utilise the often highly fluctuating riverine
environment through their movement behaviour.
While recent acoustic telemetry studies provide increased knowledge and understanding of
the physical, chemical and biological drivers of the observed movement patterns of juvenile
C. leucas on local and regional scales (Heupel & Simpfendorfer 2008, Ortega et al. 2009,
Werry 2010), what drives C. leucas movement behaviour on a fine spatio-temporal basis is
still poorly known. In order to develop effective management and conservation strategies for
this highly mobile and potentially dangerous species, it is imperative that a detailed
understanding of its short-term movement behaviour and habitat selection pattern in different
stages of its life cycle is established (Heithaus et al. 2001). The advancement of acoustic
technology in the last decade has enabled significant insight into the short-term movement
patterns as well as fine-scale identification of habitats that potentially offer advantages to
2
shark survival and growth (Sundström et al. 2001), but an in-depth understanding of short-
term movement behaviour on an individual-based level is yet to be established.
Previous studies have identified several drivers and hypotheses linked to the observed
movement patterns of C. leucas. These range from 1) salinity preferences due to energetic
costs associated with osmoregulation (Heupel & Simpfendorfer 2008, Simpfendorfer et al.
2008); 2) freshwater influx and diel variations in the flow regime as a result of tidal
movements (Ortega et al. 2009); 3) specific habitat usage to reduce energy costs associated
with movement (Werry 2010); 4) changes in the distribution of prey as a result of
tidal/freshwater interaction (Ortega et al. 2009); and 5) change in movement patterns due to
seasonal changes in temperature (Heupel & Simpfendorfer 2008, Ortega et al. 2009).
Furthermore, the movement of C. leucas has been correlated with environmental parameters
such as turbidity, pH and dissolved oxygen (e.g. Ortega et al. 2009), but so far no underlying
physiological mechanism has been proposed to explain the observed correlations.
These initial findings help identify important drivers for the movement patterns of C. leucas;
however, they are limited in their ability to test proposed theories due to the obvious
difficulties in creating a manipulative study framework, where confounding factors can be
reduced to a minimum. In addition, there are noticeable issues involved with accurately
estimating various environmental parameters at the actual location of the shark and its
immediate surroundings in real time. Previous studies have normally followed a general
linear model (GLM) approach in order to estimate the effects of parameters such as
temperature, salinity, pH, turbidity and dissolved oxygen at the location of the shark (Heupel
& Simpfendorfer 2008, Ortega et al. 2009, Werry 2010). However, due to the non-uniform
advection/dispersion and transformation of these parameters across a complex spatial
domain, these estimates might prove to be too limited in their accuracy.
Recent advances in computational fluid dynamics now allow researchers to apply models
simulating hydrodynamic processes on a sufficiently detailed scale that is meaningful to fish,
while laboratory studies have identified and defined many sensory abilities of fish capable of
distinguishing elements of hydrodynamic fields (Coombs et al. 2001, Krother et al. 2002). In
addition, advanced numerical ecological models have been integrated into hydrodynamic
models and developed to accurately simulate important biological parameters such as primary
production, nutrient cycling, chlorophyll concentration and dissolved oxygen (Szylkarski et
al. 2004). The development and application of these techniques enable researchers to
3
dynamically simulate entire ecosystems on a relatively fine scale, and thus create a model
foundation for the implementation of aquatic agent-based models capable of simulating
individual fish behaviour in a dynamic environment. The agent-based modelling approach
allows the researcher to design, apply and test detailed theories of individual movement
behaviour through biologically meaningful equations describing the movement and
interaction rules for the organism of interest. Previous studies in developing agent-based
models for fish behaviour range from migration patterns of commercially important species
of stream fish (Railsback et al. 1999) such as brown and rainbow trout (Van Winkle et al.
1998), and juvenile salmon migration (Goodwin et al. 2006), to schooling behaviour (Reuter
& Breckling 1994) and bioenergetic growth models of individual fish cohorts (Humston et al.
2004). Developing a dynamic agent-based model for an euryhaline apex predator, such as C.
leucas, is however yet to be attempted. It is believed that through this research approach it is
possible to gain valuable insight into the biology and behaviour of C. leucas, thus allowing
resource managers to improve their management decisions in regards to this species in the
face of intense coastal urbanisation.
1.2 RESEARCH QUESTIONS

This study will seek to address the following questions based on the in-depth investigation of
a juvenile specimen of C. leucas:
1. Do spatio-temporal changes in water quality parameters affect the short-term

movement patterns of the test subject, or is it largely driven by a physical habitat
preference?
2. If so, can this behaviour be explained by an instinctive drive within the animal to
lower energy costs during non-foraging periods?
3. Finally, can this behaviour be recreated as an emergent behaviour through an

agent-based model by developing equations that describes the energetic benefits
of locating optimal habitat in terms of water quality and other physical
characteristics of the system?
4
1.3 RESEARCH AIMS
This study will apply an advanced numerical Eulerian 2D model framework approach to
create a dynamic virtual representation of the chosen study site, Tallebudgera Creek,
southeast Queensland, Australia, to simulate the advection/dispersion of the essential physical
parameters on a fine spatio-temporal scale. Short-term tracks of the juvenile C. leucas
specimen will be compared and analysed in relation to the physical parameters provided by
the hydrodynamic model and used as base to develop an Eulerian-Langrangian coupled
agent-based model framework (ABM) capable of testing different theories of this particular
individual's movement behaviour. The ultimate aim of this study is to replicate and predict
the movement of the juvenile C. leucas in its habitat through testing of theories that are
proposed by this and previous studies on juvenile bull shark behaviour. In the following
chapters, the methodology and results of each component of this study will be presented and
discussed in turn, while the overall outcomes as well possible future research directions will
be discussed in Chapter VI.
1.4 HYPOTHESIS
Based on the results of previous studies (Heupel & Simpfendorfer 2008, Ortega et al. 2009,
Werry 2010) on the movement patterns of juvenile C. leucas, it is hypothesised that short-
term movement behaviour during non-foraging periods can be explained by the shark’s
instinctive goal to maintain optimal homeostasis and metabolism based on active selection of
(and movement between) energetically beneficial physical habitats and environmental
conditions.
For the purpose of this study, an energetic beneficial habitat, will be defined as a point in
time and space where an interplay of physical and environmental parameters reduces the
metabolic requirements of the animal (thus maximising growth during non-foraging periods)
relative to other available water quality regimes within the animal’s home-range.
5
CHAPTER II
WATER QUALITY ASSESSMENT OF TALLEBUDGERA CREEK
6
2.1 INTRODUCTION
Assessment of water quality is a fleeting term, as “water quality” covers a wide range of
physicochemical as well as biological parameters, ranging from simple measurements of
temperature and salinity to nutrient concentrations and suspended sediments to bacterial
counts and algal biomass. However, for the purposes of this study, assessment of water
quality will only encompass traditional measurements of temperature, salinity, dissolved
oxygen, pH, turbidity and chlorophyll a. As mentioned in Chapter I, all the above mentioned
parameters with the exception of chlorophyll a, have been reported to affect movement of
juvenile C. leucas (Heupel & Simpfendorfer 2008, Ortega et al. 2009, Werry 2010). It
follows that an investigation into the spatio-temporal dynamics of these parameters in
Tallebudgera Creek is essential for understanding how these dynamics might translate into
potential movement responses of C. leucas. Furthermore, the water quality campaign of
Tallebudgera Creek allows the study to gain important empirical insight into the drivers of
spatial and seasonal variations of the system, which in turn will serve the purpose of
establishing and validating a hydrodynamic model for the system (Chapter III).
The Tallebudgera Creek estuary (153.45°E, -28.10°S), is situated next to the suburb of
Burleigh Heads, Gold Coast, Australia, and drains a small coastal catchment of ~110 km2 by
discharging its runoff directly into the Pacific Ocean (Figure 2.1.1). The relatively small
estuary (~10.2 km in length) is classified as a tide-dominated and highly modified system,
with two artificial canals adjoining the natural system in its lower reaches (EHMP 2007).
While the estuary is in a modified state and under pressure from continuous urban
development, the general water quality level is classified as "B" or "good" by the
Environmental Ecosystem Health Program (EHMP) of Southeast Queensland (EHMP 2007).
A total of 78 fish species and 17 species of crustaceans has been recorded inside the
Tallebudgera Creek estuary in 1973 (Shine et al. 1973). However, there is to the author’s best
knowledge no recent surveys documenting the continued presence of all of these, or the
occurrence of additional, species.
7
Figure 2.2.1: Satellite imagery depicting the lower to middle reaches of the Tallebudgera Creek estuary, and
the adjoining artificial canals. Plotted using a MGA-56 projection. Data source: Google Earth.
The lower reaches of the estuary are characterized by high urbanisation and modification
while the middle and upper reaches still have 47% of area as unmodified riparian habitat
(EHMP 2007), with the remaining areas being used for rural residential purposes.
Tallebudgera Creek is popular for recreational fishing and boating/swimming activities, and
is known to be utilised by C. leucas as a nursery ground (Werry 2010). Due to the relatively
small size of the tide-dominated estuary and its semi-enclosed nature, Tallebudgera Creek is a
suitable site for studying the movement of resident juvenile C. leucas, as this species has a
tendency to stay inside estuaries at early life-stages (Werry 2010). Furthermore, only a
relatively small number of measuring stations for water quality sampling is needed to cover
the full spatial extent of the estuary.
8
2.1.1 AIMS & PURPOSE
As an integral part of this project, the main aim and purpose of the water quality campaign is
to collect temperature and salinity validation data for the hydrodynamic model (Chapter 3),
and secondarily to directly assess the general status as well as the seasonal variations of the
tide-dominated system of the Tallebudgera Creek estuary.
2.2 METHODOLOGY
2.2.1 SAMPLING DESIGN

In order to capture a representative picture of the spatiotemporal variation in water quality in
the tidal-influenced zone of Tallebudgera Creek, 13 measuring stations (Figure 2.2.2) were
sampled throughout the system on a weekly to biweekly basis over one year (Jan 2010 - Jan
2011). Due to the primary use of the sampling campaign as validation data for the
hydrodynamic model, it was not necessary to standardise sampling according to tidal effects,
as the model automatically accounts for this factor (Chapter 3). Sample sites were selected on
the basis of accessibility from land, while keeping stations no further than a maximum of
~1.5 km apart from the nearest neighbour. Using a handheld YSI 6820V2-1 multi-probe
sonde, salinity, temperature, dissolved oxygen, pH, and turbidity and were recorded 20 cm
below the water surface at each station as well as at 2 m depth if the sample site allowed for
it. While some sample stations were deeper than 2 m, this depth was sampled due to the
interpolated bathymetry scheme (explained in Chapter 3) on average being 2 m below mean
sea level. This allowed measurements to give an indication of the spatiotemporal change of
parameters at the average bottom depth of the system compared to surface values.
Furthermore, chlorophyll a content was measured at four select stations (Figure 2.2.2) in
order to get an indication of the primary productivity and biological condition of the system.
9
Figure 2.2.2: The location of water quality stations throughout Tallebudgera Creek; see colour legend for
station type
Five additional stations corresponding to the Ecosystem Health Monitoring Program (EHMP)
monitoring stations are present in the system (Figure 2.2.2). Environmental parameters are
measured at these stations at several depths once every month as part of an ongoing water
quality monitoring program for Southeast Queensland. The water quality parameters
mentioned above, with the exception of chlorophyll a, were measured by the EHMP using an
YSI 6920 sonde, which is identical to the YSI 6820 Sonde in terms of the sensors utilised.
The EHMP data was therefore deemed to have a sufficient level of compatibility with the
data from the other sample stations established for this study, and was assimilated into the
measured dataset for final analysis. Chlorophyll a was, however, determined by the EHMP
using a different methodology than that used in this study (explained in detail in section
2.2.6) and data were therefore not assimilated into the chlorophyll dataset but treated
separately. Furthermore the EHMP stations provide the study with monthly measurements of
10
ammonia, organic nitrogen, oxidised nitrogen, total nitrogen, filterable reactive phosphorous
and total phosphorous, thus allowing the measured chlorophyll content to be related to
nutrient availability.
2.2.2 TEMPERATURE
Temperature and conductivity were measured using an YSI 6560 conductivity/temperature
probe installed in a handheld YSI 6820V2-1 multi-probe sonde. The temperature probe
produces a temperature reading in degree Celsius with a resolution of 0.001 °C and an
accuracy of + 0.15 °C in the range of -5 to 50 °C (YSI 2006).
2.2.3 SALINITY
Salinity was calculated by measuring conductivity using the same conductivity/temperature
probe as mentioned above. Since conductivity is highly dependent on temperature, the
conversion to salinity was done by the built-in software in the YSI 6820V2 sonde, which
accounts for the temperature effect in accordance with standard algorithms (APHA 1971).
The measurements produced have a resolution of 0.01 ppt and an accuracy of + 1.0% of the
salinity reading, in the range of 0 to 70 ppt and -5 to 50°C range (YSI 2006).
2.2.4 PH
pH was measured by an YSI 6561 pH probe installed in the 6820V2 multi-probe sonde and
utilises a combination electrode consisting of a proton selective glass reservoir filled with
buffer at ~ pH 7 and an Ag/AgCl reference electrode. The pH probe reports values with a
resolution of 0.01 with a accuracy of + 0.2 units in the -5 to 50 °C range (YSI 2006).
2.2.5 TURBIDITY
Turbidity was measured by an YSI 6136 Turbidity probe installed in the YSI 6820V2 unit.
The probe consists of a light emitting diode which emits radiation near the infrared region of
the spectrum (830 to 890 nm) to illuminate the sample, while a highly sensitive photodiode
detects the scattered light at a 90° angle from the light source in accordance with the
International Standards Organisation (ISO) protocol. The output from the turbidity probe is
then processed by the 6820V2 software and is recorded in Nephlometric Turbidity Units
(NTU). The values are recorded by the sonde with a 0.1 resolution and has an accuracy of +
2% of the reading in the 0 to 1000 NTU and -5 to 50 °C range (YSI 2006).
11
2.2.6 DISSOLVED OXYGEN
Dissolved oxygen (DO) was measured by an YSI 6562 Rapid Pulse polarographic probe
installed in the YSI 6820V2 unit. The probe utilizes a Clark-type sensor that measures the
electrical current that is associated with the reduction of oxygen as it diffuses through the
Teflon-membrane, which is proportional to the partial pressure of oxygen in the sample.
Dissolved oxygen is measured in mg/L and is converted into percentage saturation (%) using
the sonde's temperature and conductivity reading. DO is measured with a resolution of 0.1%
with an accuracy of + 2% of the reading in the range of 0 to 200% air saturation and -5 to
50°C (YSI 2006).
2.2.7 CHLOROPHYLL
Chlorophyll content was measured by collecting triplicate 500 mL water samples at ~20 cm
beneath the water surface at four different locations throughout the study site (see Figure
2.2.2), and kept on ice until filtration of the sample was possible. Millipore nitrocellulose
(0.45 μm pore size) filters were used for the filtration of the samples, and 500 mL was filtered
under mild suction unless a high turbidity in the sample prohibited full filtration of the
sample. All samples were filtered under dark conditions to prevent undue chlorophyll
denaturation. After filtration, 3 to 5 drops of a saturated MgCO3 solution were added to the
filters to prevent acidity in the filter. The filters were then placed in 50 mL centrifuge tubes
with 10 mL of 90% aqueous acetone and stored in the dark at +5 °C. After at least 20 hours,
the samples are centrifuged for 5 min at 4200 rpm and the resulting supernatant decanted into
1-cm path length glass cuvettes, before the absorbance of each sample was measured at the
wavelengths of 750, 664, 647 and 630 nm using a Novaspec II spectrophotometer. The
measured absorbance values were used to calculate the amounts of chlorophyll a, b and c in
(μg/l) using the empirical relationships described in (Parsons 1984):
12
Where Ex is the absorbance at wavelength x and corrected by subtracting the absorbance of
the 750 nm (turbidity) reading. The total amount of chlorophyll in the sample was then
calculated using the equation:
Where Chla, Chlb, and Chlc are the previously calculated amounts of Chlorophyll a, b and c,
respectively, while is the volume of acetone in mL (10 mL), and is the volume of filtered
seawater in litres. The final value of was then determined by calculating the mean
value of the triplicate samples for each site.
2.2.8 SONDE CALIBRATION & MEASUREMENT TECHNIQUES

Before each sampling event, each sensor of the YSI 6820V2-1 sonde was checked for any
discrepancies when measuring known standard solutions for pH, turbidity and conductivity.
If any significant drift was noticed the sensor was calibrated according to the instructions of
the manufacturer (YSI 2006). The DO sensor was, however, always calibrated to the current
pressure level at mean sea level using the recorded pressure data supplied from the Bureau of
Meteorology (BOM) at Burleigh Heads. During transport to and from the study site the sonde
would be kept secure while fitted with a protected cup filled with tap water in order to
prevent the DO membrane from drying out. Before any samples were recorded the sonde
would be kept submerged at the given depth for a minimum of 5 minutes to allow sensors to
stabilise with the ambient water conditions.
2.2.9 DATA ANALYSIS

All processing of water quality data was performed using the commercially available
software package, MATLAB R2007b and R2010b. Prior to any statistical analysis each
measured water quality parameter at each station was tested for normality and
homoscedasticity using the Lilliefors test (Lilliefors 1967) and the Levene’s test, respectively
(Levene 1960). Data transformation using log and square root transformations were
undertaken where necessary and alpha, was adjusted for multiple tests using the Bonferroni
correction. Due to the spatio-temporal connectivity between measurements. It is unlikely
that the measured concentrations of a given water quality parameter are truly independent of
each other, however for the purpose of this study they will be assumed independent.
13
In order to test if there was a difference between surface vs. bottom values for each station,
the non-parametric Kruskal-Wallis test was applied to data sets that met the requirement of
homoscedasticity but was found to be non-normal, while one-way ANOVA’s were applied to
data sets meeting the assumptions of normality and homoscedasticity. Data sets failing to
meet the requirement of homoscedasticity but fitting a normal distribution were alternatively
tested with a Welch’s t-test (Welch 1951). The parametric Pearson correlation test was used
to determine the relationship between each annually averaged parameter and distance to the
river mouth.
In accordance with the report guidelines of the EHMP (EHMP 2007), temporal plots showing
the seasonal variation of each measured parameter were created by calculating the median as
well as the 20th and 80th percentile values using the measurements from all stations on each
sampling day. Likewise for the generation of spatial plots, the annual median together with
the 20th and 80th percentile values were calculated for all sample days at each station located
in the course of the natural river system. Finally, a series of 3D contour plots was created for
visual representation of the spatio-temporal dynamics of each measured parameter.
2.3 RESULTS
2.3.1 TEMPERATURE
Temperature ranged from 17.38 (July) to 33.14 oC (January) over the course of 2010,
meaning a potential increase in the rate of physiological reactions in elasmobranch fishes by a
factor of 2-3 between winter and summer conditions (Brett & Groves 1979). Drops in
temperatures were registered following heavy rainfalls and increased freshwater discharge
throughout the year, with the upper reach station being the most sensitive, dropping as much
as 5.2 degrees on one occasion, while the river mouth only decreased by 0.44 oC. The median
annual temperature for each measuring station was not significantly correlated with increased
distance to the river mouth (P = 0.066, R = -0.47), although there was a slight trend for water
temperature to decrease when moving upstream. However, during prolonged periods with no
significant rainfall in spring and summer, there were several registered cases of an inverse
relationship, where the shallower and slower moving water of the upper reaches were warmer
than the river mouth.
14
Figure 2.3.1: A) Seasonal variation in temperature throughout the course of 2010. B) Spatial variation in
temperature throughout the course of 2010. The data set depicted is the assimilated data set of EHMP
measurements along with measurements made by this study
There was no significant difference between temperature in the surface layer (~20cm depth)
and at 2 m depth for any stations throughout the year, indicating the system being generally
well-mixed by a conjunction of tidal movements and surface waves due to wind action as
well as regular freshwater discharge events that prevent any noticeable thermocline to be
established in the shallow water system. Only on five sampling events out of a total of 180
was there a difference equal or greater than 2 oC between top and bottom temperatures, while
the mean of the differences was only 0.09 oC.
Figure 2.3.2: A contour plot of the spatiotemporal variation of temperature throughout the sampling period.
The data set depicted contains only the measurements conducted by this study
15
2.3.2 SALINITY
Salinity levels were found to be highly variable throughout the year and across all measuring
stations in the estuary, with the river mouth ranging from 5.39 to 39.0 PSU, and the end of
the tidal dominated area, from 0.03 to 25.92 PSU. The seasonal pattern in salinity was
characterized by wide fluctuations between individual sampling dates (Figure 2.3.3) in
particular during summer and autumn months. The observed pattern can be explained by the
increased frequency of heavy rainfalls and the subsequent freshwater discharges during the
wet summer season which has a clear impact on the salinity levels throughout the system,
while salinity levels remain generally higher and more stable during the dry winter months.
There was a strong negative correlation between the annual medians of each station and
distance to the river mouth (P < 0.001, R = -0.97), with an average difference in salinity
between the river mouth and uppermost station of 26.63 PSU.
Figure 2.3.3: A) Seasonal variations in salinity throughout the course of 2010. Note the abrupt fluctuations
between sample days. B) Spatial variation in salinity throughout the course of 2010, showing a general decline
of salinity proportional to distance from river mouth. Note the higher level of variation, which is indicated by
the increased width between the 20th and 80th percentile, as the distance from the river mouth increases. The
data set depicted is the assimilated data set of EHMP measurements along with measurements made by this
study.
Only two stations had significantly different salinity levels between surface and 2 m depth
measurements (P < 0.01). However, each of these stations was located at the entrance to the
adjoining man-made canals, which are both relatively deep (> 4m) and may be assumed to
16
have an altered flow regime compared to the rest of the natural river system. There was an
average difference of 3.51 between top and bottom measurements for those stations that
were non-significantly different, which could indicate the occurrence of a halocline in the
system, though the lack of any apparent thermocline (as mentioned above) suggests any
stratification to be relatively weak. Furthermore, there was no single sample day where all
stations displayed a difference greater than 2.0 between top and bottom, indicating that any
halocline present was local rather than covering the entire estuary.
Figure 3.3.4: A contour plot of the spatiotemporal variations of salinity throughout the sampling period. The
data set depicted contains only the measurements conducted by this study.
2.3.3 DISSOLVED OXYGEN

Dissolved oxygen (DO) levels across stations and throughout the year of 2010 were generally
relatively high, with an overall mean of 7.19 ±1.44 mg/l (S.D., n=665). The annual median
DO was positively correlated with distance from the river mouth (P < 0.01, R = 0.78). No
seasonal variation in DO levels seemed to be evident, aside from slightly elevated DO
concentrations in the spring of 2010. Out of 665 measurements over 43 sampling days, only
4.6% of all measurements registered a DO level that was below 4.0 mg/l. Of these, 71% were
bottom measurements, while 45% were registered on the same day (9th of August).
17
Figure 2.3.5: A) Seasonal variations in dissolved oxygen throughout the sampling period. Note the relative
small difference between the 20th and 80th percentile levels across sampling days. B) Spatial variations in
dissolved oxygen across the tidal-dominated part of the system. Note the slight increase of dissolved oxygen
concentrations with increased distance to the river mouth. The data set depicted is the assimilated data set of
EHMP measurements along with measurements made by this study.
There was no registration of DO levels below 2.0 mg/l over the sampling period of 2010.
Only one station (EHMP stations excluded) had a significant decrease in DO between its
surface and bottom values over the course of 2010 (P < 0.01, n = 60) with a mean difference
between surface and bottom of 1.18 ±0.82 mg/l (S.D., n=60). The lack of hypoxia events (DO
< 2.0 mg/l) coupled with a relative small number of low DO concentrations (DO < 4 mg/l),
suggests that the tide-dominated part of the Tallebudgera Creek estuary is well mixed on a
spatiotemporal scale and not prone to hypoxic events. The shallow nature of the system is a
major contributor to this pattern.
18
Figure 2.3.6: A contour plot of the spatiotemporal variations of dissolved oxygen throughout the sampling
period. The data set depicted contains only the measurements conducted by this study.
2.3.4 PH
pH levels fluctuated around a mean of 7.66 ±0.47 (S.D., n=665), with an overall range
spanning from 6.37 to 8.66 over the course of 2010. pH was, as expected, negatively
correlated with distance to the river mouth (P < 0.01, R = -0.96) due to the more alkaline
nature of seawater dominating the river mouth while freshwater discharge and rainfall
generated run-off causing a drop in pH in the upper reaches of the system. No station
displayed a significant difference between surface and bottom pH levels, with the highest
observed difference between surface and bottom being 0.7, which was registered at the
uppermost station during the presence of a weak halocline (4.58 difference between surface
and bottom salinities). The seasonal variation of pH was expected to be strongly influenced
by freshwater discharge/rainfall events, with the pH levels dropping down to as little as 6.7 at
the river mouth the week after a ~96.0 m3/s discharge event.
19
Figure 2.3.7: A) The seasonal variations of pH throughout the sampling period. Note how the fluctuations in
pH tend to follow the same general trend as salinity (Figure 2.3.3A). B) Spatial variations in pH throughout the
course of 2010, showing a general decline of pH levels proportional to distance from the river mouth. The data
set depicted is the assimilated data set of EHMP measurements along with measurements made by this study.
Figure 2.3.6: A contour plot of the spatiotemporal variation of dissolved oxygen throughout the sampling
period. The data set depicted contains only the measurements conducted by this study.
20
2.3.5 TURBIDITY
Turbidity levels were generally low across stations and throughout the year, with a median
value of 2.00 NTU (n = 665), but with spikes up to 252.80 NTU following a discharge event
of ~103.5 m3/s. Turbidity was positively correlated (P < 0.01, R = 0.96) with distance to the
river mouth, indicating a general increase of suspended solids upstream. There was no
significant difference between surface and bottom layers throughout the year, and any
seasonal variation of turbidity was most likely the result of a change in frequency and
magnitude of freshwater input between summer and winter.
Figure 2.3.7: A) The seasonal variations of turbidity throughout the sampling period. B) Spatial variations in
turbidity throughout the course of 2010, showing a general increase of turbidity levels proportional to distance
from the river mouth. Note that the scale of the y-axis differs between A) and B). The data set depicted is the
assimilated data set of EHMP measurements along with measurements made by this study.
21
Figure 2.3.8: A contour plot of the spatiotemporal variations of turbidity throughout the sampling period. The
data set depicted contains only the measurements conducted by this study.
2.3.6 CHLOROPHYLL A
Measured levels of chlorophyll a was low for all stations throughout the year, with a mean
value of 0.31 ± 0.28 μg/l (S.D., n = 120) and a maximum value of 1.42 μg/l registered the 3rd
November 2011 at the uppermost station. There was no correlation (P = 0.11, R = 0.88)
between chlorophyll a and distance to the river mouth. However, data suggests that there is a
general tendency of an increase of chlorophyll a concentrations upstream. The values
measured by this study were in contrast to the chl a concentrations reported by the EHMP,
where the mean value across stations and throughout the year was 2.71 ± 5.32 μg/l (S.D., n =
65) and a maximum value of 42.25 μg/l reported the 16th April 2011 for the middle reach,
and will be discussed further in Section 2.4.
22
Figure 2.3.8: A) Seasonal variations in chlorophyll a throughout the sampling period of 2010. Note that
chlorophyll a concentration only exceeded 20 μg/l on one occasion. B) Spatial variations in chlorophyll a
throughout the course of 2010, showing a weak trend of increased chlorophyll a concentrations upstream. The
data set depicted is the assimilated product of EHMP measurements along with measurements made by this
study.
2.4 DISCUSSION
As stated in Section 2.1 the two primary objectives of the water quality measuring campaign
was: 1) acquisition of validation data for comparison with hydraulic model outputs (detailed
in Chapter III); and 2) assessment of general water quality conditions and the spatio-temporal
water quality dynamics of the Tallebudgera Creek system. While the data sets collected by
this study, because of the limited temporal and spatial extents, do not allow an in-depth
analysis of ecosystem health, several key aspects that appear to influence ecosystem health
can be identified.
Importantly for the purposes of this study, vertical stratification of the water column was
found to be insignificant throughout the natural parts of the system, although prolonged
stratification was evident in the adjoining artificial canals. These findings are unsurprising
due to the shallow nature of the unmodified part of the estuary compared to the canals (> 5
m). Due to their placement perpendicular to tidal currents and their lack of major freshwater
sources, the artificial canals are expected to be more prone to stratification. In turn, the lack
of any significant spatio-temporal stratification patterns within the natural part of the system
further justifies this study's application of a two-dimensional depth-averaged hydrodynamic
23
model for simulating the transport and dispersion of salinity as well as heat exchange
(Chapter I and Chapter III).
Due to the limited temporal resolution of the data sets, it was not possible to assess the diel
variation of dissolved oxygen (See Chapter III for a discussion on diel variation in salinity
and temperature). While oxygen levels are known to fluctuate as high as >15 mg/l over a 24
hr period, especially in organically enriched waters (Lamb 1985), the results presented in
Section 2.3.3 suggest that DO levels throughout the system and over the course of the year
are in compliance with the EHMP Water Quality Guidelines (EHMP 2007). It is interesting
to note that nearly all readings of DO that were > 4 mg/l occurred on the same sample day
and at all stations (even those located in the well-mixed waters of the river mouth). While this
result cannot be dismissed due to faulty equipment, it is possible that this "outlier" in the data
set was caused by the electrolyte solution of the DO sensor being partly dried during storage
before the sampling event. Furthermore, the persistent high levels of DO, coupled with
relatively low background concentrations of chlorophyll a, suggests that the system is
predominantly aerated by physical means, i.e. diffusion from the atmosphere, and was little
affected by eutrophication. Only on one occasion was the chl a level high enough to classify
as a bloom (42.25 μg/l, EHMP measurement), which occurred after a prolonged period of
increased freshwater discharge (~7x above base levels during dry periods). It is thus plausible
that this bloom was triggered by an increase of organic material and nutrients being brought
into the system with the freshwater discharge. However, three days later there was already no
sign of an increased chl a concentration in the system (< 1.5 μg/l), which suggests that the
bloom to either have dissipated or become localised in small patches not covered by sampling
stations.
As mentioned in Section 2.3.6, there was a notable difference in the chl a measurements
obtained through this study's field campaign as opposed to the EHMP measurements. Since
this study sampled on different days and at different sample sites than the EHMP, it is
difficult to determine whether this discrepancy between the data sets is caused by natural
variation originating from the spatio-temporal dynamics of the system, or the result of an
inadequate chl a extraction technique (Section 2.1.6).
Aside from the tide-induced mixing of the system, the main driver of spatio-temporal changes
in measured water quality parameters was without equal heavy rainfall and the subsequent
run-off and discharge from the catchment. Freshwater discharge events had, not surprisingly,
24
the greatest effect on salinity and turbidity, but were also found to cause a notable change in
temperature and pH levels. While salinity obviously is the most sensitive to increased
freshwater inputs, the system as a whole displayed a remarkable resilience to these abrupt
system-wide shifts by returning to a pre-discharge state in a matter of days; e.g. registered
turbidity levels of 252.8 NTU dropped to 4.8 NTU just 6 days after a ~100 m 3/s discharge
event had taken place on the 4th of May, while salinity also increased from 0.3 PSU to 9.8
PSU at the same location. The system's natural ability to return to a steady-state can likely be
contributed to its small volume, meaning that relatively few tidal cycles are required for
flushing the entirety of the system. However, the small size of Tallebudgera Creek also
means that less freshwater discharge is needed before the system is being pushed into a
phase-shift, which in turn ultimately can/will result in a higher frequency of sudden system-
wide shifts between oligohaline and meso/polyhaline water conditions compared to larger
and more voluminous estuaries. It is highly plausible that these sudden changes in salinity
and turbidity will impose a range of physiological challenges for species residing within the
estuary, and the potential implications of these abrupt phase-shifts will be discussed in detail
in the following chapters.
In conclusion, the results gathered from the water quality campaign in conjunction with the
EHMP data are in good general agreement in supporting the past EHMP assessments of
Tallebudgera Creek, which in the past 4 years have been given a rating of "Good" to
"Excellent" water quality (EHMP 2009).
25
CHAPTER III
HYDRODYNAMIC MODELLING OF TALLEBUDGERA CREEK
26
3.1 INTRODUCTION
Numerical modelling of shallow water systems has in the past few decades experienced rapid
development, thanks to the availability of increasingly powerful computers capable of
resolving the complex numerical models necessary for adequate representation of the relevant
physical processes (Dias & Lopes 2006b). While the complexity of those physical processes
involved still prompts for future research and development of the mathematical formulation
of these processes, depth-averaged 2D-models have proved to be reliable in capturing the
hydraulic properties of shallow water systems (Cheng et al. 1993, Inoue & Wiseman Jr 2000,
Umgiesser et al. 2003, Dias & Lopes 2006b, Zacharias & Gianni 2008).
Furthermore, advanced hydrodynamic modelling tools, such as MIKE by DHI (Warren &
Bach 1992, DHI 2009), have consistently been used by researchers and engineers to simulate
the hydraulic properties of a wide range of systems, e.g. estuaries (Szylkarski et al. 2004,
Mirfenderesk & Tomlinson 2007, Paliwal & Patra 2011), coastal lagoons (Dias & Lopes
2006, Zacharias & Gianni 2008), large offshore straits (McClimans et al. 2000) and entire
oceans (Almroth & Skogen 2010).
In previous field studies on C. leucas movement (Simpfendorfer et al. 2005, Heupel &
Simpfendorfer 2008, Ortega et al. 2009, Werry 2010), a recurrent issue is the difficulty in
obtaining measurements of the physical as well as the biological parameters at the actual
position of the animal. While recent developments in acoustic transmitters now allow in situ
measurements of parameters such as temperature and salinity, these transmitters remain
relatively expensive while also incapable of measuring the magnitude and direction of flow.
However, the rapid advances in computer technology and development of adaptable grid
discretisation schemes have recently increased model spatial resolution to a level that allows
governing 2D physical processes in narrow estuaries and adjacent near-shore areas to be
resolved. This has opened up new possibilities for comparing short- and long-term movement
of large fish species, such as C. leucas, with model predictions of a wide range of
hydrodynamic entities (e.g. Werry (2010)).
While the prospect of direct comparison between fish movement and fine-scale hydraulic
model outputs poses a unique opportunity to investigate fish responses relative to dynamic
shifts in the hydraulic regime, the implementation of a hydrodynamic model also provides the
foundation for any further model development, e.g. agent-based models. In recent years,
27
ecological studies, e.g. Humston (2004) and Goodwin (2006), have coupled the Eulerian
framework of hydrodynamic models with the Lagrangian framework of agent-based models,
in order to simulate fish movement in relation to hydraulic cues in various aquatic
environments. While agent-based models will be discussed in detail in Chapter V, the
remainder of this Chapter will deal with implementation and validation of a hydrodynamic
model for Tallebudgera Creek.

As a fundamental part of this study, the aim of this chapter is to implement a MIKE 21FM
hydrodynamic model of the Tallebudgera Creek estuary. The two main objectives are: 1) to
validate the model to a satisfactory degree, so that hydraulic model outputs may be used to
analyse short-term tracks of C. leucas (Chapter IV); and 2) use the validated hydrodynamic
model as a foundation for developing an agent-based model capable of replicating observed
short-term movement of C. leucas (Chapter V).
3.2 METHODOLOGY
The commercially available MIKE21 modelling software package was used to create a 2-D
hydrodynamic model of the tidal part of Tallebudgera Creek. This model is capable of
simulating the effect of freshwater discharge vs. tide-driven water flow while accounting for
the transport processes of salinity and temperature (Warren & Bach 1992, DHI 2009). In the
following sections, a short description encompassing the most fundamental properties of the
model framework and setup is given.
3.2.1 EQUATIONAL FRAMEWORK

The MIKE 21 FM Hydrodynamic Module is based on an unstructured flexible mesh
approach to resolving the two-dimensional flow simulation of the Tallebudgera Creek estuary
and the nearby offshore environment surrounding it (Warren & Bach 1992). The numerical
scheme adopts a finite-volume approach to solving the depth-integrated incompressible
Reynolds averaged shallow water equations, while using 2nd order transport equations to
resolve the advection/dispersion of salinity and temperature. Furthermore, a range of separate
turbulence closure schemes can be applied to resolve sub-grid scale mixing processes (DHI
2009).
28
The local continuity equation is:
Eq. 3.1
While the two horizontal momentum equations for the x- and y-components are, respectively:
Eq. 3.2
Eq. 3.3
Where t is time, while x, y, z are the Cartesian co-ordinates, is the surface elevation, d is the
still water depth and is the total water depth. The depth-averaged velocity
components are given by and in the x and y directions, respectively. f is the Coriolis
parameter and g is gravitational acceleration. is the density of water and is the reference
density of water. , , and are the constituents of the radiation stress tensor,
while is the vertical turbulent viscosity. is atmospheric pressure and is the magnitude
of a discharge from point sources, while and is the corresponding velocity by which the
water mass concerned is discharged into the ambient water. are the lateral stresses, which
include viscous friction, turbulent friction and differential advection, while are the surface
and bottom stresses (DHI 2009).
3.2.2 TRANSPORT EQUATIONS

The two-dimensional transport of temperature, , and salinity, , follows the general
transport-diffusion equations:
Eq. 3.4
Eq. 3.5
In the above equations, and are the depth averaged temperature and salinity, respectively,
is the total water depth while and are the depth averaged velocities in the and
directions, respectively. is a source term due to heat exchange with the atmosphere and is
29
the horizontal diffusion term. Finally and are respectively the temperature and salinity
of a point source (DHI 2009).
3.2.3 HEAT EXCHANGE

Heat exchange with the atmosphere is calculated on the basis of four major physical
processes: latent heat flux, sensible heat flux, net short-wave radiation and net long-wave
radiation. In the heat exchange calculation, the model assumes that all processes, except net
short-wave radiation, occur at the water surface (DHI 2009). In order to define an absorption
profile for the short-wave heat flux, a modified Lambert-Beer equation is used to
approximate light attenuation throughout the water column:
Eq. 3.6
Where is the light intensity at depth below the surface, while is the light intensity just
below the surface. is a user-specified constant that accounts for the fraction of light energy
that is absorbed near the surface and is the user-specified light extinction coefficient (DHI
2009). For two-dimensional calculations, the source term is given by:
Eq. 3.7
In equation 3.7 is the evaporative heat loss (or latent heat flux), is the sensible heat flux
due to convection, is the short-wave radiation and is the long-wave radiation.
is the reference density of water and is the specific heat capacity of water (DHI 2009).
The latent heat flux is given by:
Eq. 3.8
Where = 4370 and is the latent heat constant encompassing the latent heat
vaporization constant L and the Dalton number . is the wind speed 2 m above the
water surface while and is the temperature at 0°C. is the
relative humidity in percent. is the temperature of the water surface and is air
temperature. and are user specified constants with default values of 0.5 and 0.9,
respectively (DHI 2009).
30
The heat flux due to convection, ( , depends on the boundary layer between the
water surface and the atmosphere. The model assumes that this layer is predominantly
turbulent and thus implies the following relationship:
Eq. 3.9
Where is the air density , is the specific heat

capacity of air while and is the sensible transfer
coefficients for heating and cooling, respectively. is the wind speed 10 m above the
water surface (DHI 2009).
The intensity of the short-wave radiation, which mainly consists of light with wavelengths
between 4000 to 9000 Ǻ, depends on the distance to the sun, declination angle and latitude,
extraterrestrial radiation, cloud cover and the amount of water vapour in the atmosphere (DHI
2009). The net short-wave radiation is:
Eq. 3.10
Where is reflection coefficient and is the average hourly short wave radiation.
Long-wave radiation consists of wavelengths between 9000 and 25000 Å and is dependent on
cloud cover, air temperature, the vapour pressure in the air and the relative humidity (DHI
2009). The net outgoing long-wave radiation is given by Brunt's equation:
Eq. 3.11
Where is the vapour pressure at dew point temperature measured in mb. is the Stefan
Boltzman constant and is . is the number of hours of sunshine
and is the maximum number of hours of sunshine. The coefficients and are
defined by the values of:
For a more detailed overview of the equational framework of the hydrodynamic model the
reader is referred to the MIKE 21 scientific documentation (DHI 2009).
31
3.2.4 MODEL DOMAIN
The bathymetry measurements used for the model's grid generation were supplied by DHI,
and originally surveyed by Gold Coast City Council in 2001. As mentioned in the section
above, the model applies a flexible mesh to spatially represent the physical bathymetry of the
system. The mesh created for the Tallebudgera model consists of a mixture of triangular and
quadrangular grid cells. The resolution of the flexible mesh, or the actual size of the
individual grid cell, varies from relatively small quadrangular cells each with an area of 50
m2 to larger offshore triangular cells each with an area of 12000 m2. In total the entire mesh
consists of 34253 individual calculation cells that are solved and updated for each model time
step. A variable time step interval ranging from 0.01 to 30 seconds is used in the calculation
of both the shallow water equations, and the transport equations are determined so that the
Courant-Friedrich-Lévy (CFL) number is less than a critical CFL number in all
computational nodes. Figures 3.2.2 and 3.2.3 below provide an overview of the entire model
area, and an example of the structure of flexible mesh grid, respectively.
32
Figure 3.2.1: Illustration showing the bathymetry of the Model domain, plotted in a MGA-56 projection. The
area inside the black box is the part of the non-tidal dominated freshwater zone that is included in the model
setup (see main text for further explanation). The red box marks the area shown in Figure 3.2.3 below.
33
Figure 3.2.2: An example of the flexible mesh of the model. The green markers each indicate a smaller sub-area
within the mesh. The figure is plotted using MGA-56 projection coordinates (unit: meters).
In order to capture the effects of freshwater discharge on the temperature, salinity and
hydrodynamics of the system, the model domain was created to include a connection between
the point source of freshwater discharge data (See section 3.2.5.1) and the tide-dominated
area of the system (See black box in Figure 3.2.2).
3.2.5 INPUT PARAMETERS

In order to accurately simulate the hydrodynamics as well as the heat exchange between the
water and the atmosphere, a range of input data was required. In the following sub-sections,
each input parameter in the model design will be described.
34
3.2.5.1 OFFSHORE BOUNDARY DATA
The tidal data utilised by the offshore boundary of the Tallebudgera model domain are
extracted from DHI's Global Tide model, which represents the major diurnal (K1, O1, P1 and
Q1) and semi-diurnal tidal constituents (M2, S2, N2 and K2) with a spatial resolution of
0.25° × 0.25° and is based on altimetry data gathered from the NASA/CNES TOPEX-
/POSEIDON satellite. Boundary values for salinity and temperature are extracted as daily
values from the Global HYCOM model, which is a data-assimilative hybrid isopycnal-
sigma-pressure (generalized) coordinate ocean circulation model (the Hybrid Coordinate
Ocean Model or HYCOM). Due to the large spatial resolution of the HYCOM model, the
entire offshore area of the Tallebudgera Creek lies within a single HYCOM calculation cell,
and thus it is assumed that salinity and temperature vary in time, but are constant along the
boundary. For further documentation on the HYCOM model the reader is referred to
Wallcraft (2009).
3.2.5.2 FRESHWATER DISCHARGE

Freshwater discharge1 is handled as a point source in the model, with the source located in
the very top of the river in the model domain (see Figure 3.2.2). The freshwater discharge
data are measured and supplied by the Queensland Department of Environment & Resource
Management and provides a point measurement every six hours of freshwater discharge in
cubic meters along with corresponding salinity and temperature.
3.2.5.3 METEOROLOGICAL DATA

In order to calculate the heat exchange between the atmosphere and water surface (as
described in section 3.2.3), meteorological data from the Coolangatta Weather Station,
supplied by the Australian Bureau of Meteorology (BOM), was incorporated into the model
with the assumption that it is representative of the Tallebudgera Creek area, which is located
ca. 9 km away from the weather station. The time series of data used in the Heat exchange
module includes: air temperature, relative humidity, wind speed and wind direction - all with
a temporal resolution of 30 min. The heat exchange module assumes a default constant value
of 70% for the clearness coefficient, which is a term for the given cloud cover conditions, but
due to the lack of readily available cloud cover data for the area, this parameter was treated as
a calibration parameter throughout the model development.
1
Based on or contains data provided by the State of Queensland (Department of Natural Resources and Water) [2011], on the
condition that the State gives no warranty in relation to the data (including accuracy, reliability, completeness, currency or suitability)
and accepts no liability (including without limitation, liability in negligence) for any loss, damage or costs (including consequential
damage) relating to any use of the data. Data must not be used for direct marketing or be used in breach of the privacy laws.
35
3.2.5.5 PRECIPITATION
Rainfall was incorporated into the HD model as a time series of measured daily rainfall for
the Tallebudgera Creek area supplied by the Queensland Bureau of Meteorology. The model
does not take any run-off into account from the adjacent downstream catchments, meaning
that rainfall will only "fall" within calculation cells and not in the surrounding land values
that is not covered by the model mesh. It is however assumed that the majority of the
Tallebudgera systems total run-off originates from the much larger upper catchment, and is
thus captured in the freshwater discharge measurements by the DERM flow gauge station
located ~15 km from the river mouth (Uppermost point in figure 3.2.2).
3.2.5.6 WAVE RADIATION STRESS

In order to account for wave-induced longshore currents immediately offshore to the
Tallebudgera Creek river mouth, the second order stresses due to breaking of short period
waves was included in the hydrodynamic model. The wave radiation stresses Sxx, Syy and
Sxy (m3/s2), were calculated over the course of 2010 by utilizing the MIKE21 Spectral Wave
model (SW) to calculate the wave transformation and breaking processes occurring from
offshore to the Tallebudgera Creek river mouth (figure 3.2.4). The mesh of the model domain
was generated using bathymetry data obtained from the DHI, while offshore wave boundary
given by significant wave height, peak wave period and wave direction was obtained from
the Gold Coast Seaway wave buoy located approximately 10 km north of Tallebudgera Creek
river mouth. The setup of this wave model assumes that the wave conditions at the Gold
Coast Seaway are representative of the conditions at Tallebudgera, and constant along the
offshore boundary of the model domain.
36
Figure 3.2.3: Model domain of the created offshore Spectral Wave model. The red box marks the Tallebudgera
Creek entrance. The figure is plotted using MGA-56 projection coordinates (unit: meters).
3.2.5.7 INITIAL CONDITIONS

The initial conditions of temperature and salinity throughout the entire model domain were
generated by using historical EHMP measurements to produce an area map with spatially
varying temperature/salinity regimes according to the historical data. The generated initial
fields were then applied to a model setup simulating one month prior to the actual start date
of the full 2010-2011 model setup. This allowed the initial fields of temperature and salinity
to move into a quasi-stationary equilibrium through tidal current induced mixing and heat
exchange with the atmosphere. The resulting spatial maps containing data of surface
elevation, temperature, salinity, total water depth as well as u- and v-velocities was then
extracted and used as initial values for the full 2010-2011 model run.
3.2.6 CALIBRATION PARAMETERS

Certain physical variables and constants needed for the hydrodynamic model setup (Table
3.2.1) were beyond the scope of this study to obtain measurements of, and were thus treated
37
as calibration parameters in order to obtain satisfactory simulation results. Table 3.2.1
presents the list of the various calibration constants and solution settings specified in the
model setup, while spatially varying parameters are explained in more detail below.
Table 3.2.1: The values and user-specified settings of the enabled modules in the final model setup.
Shallow water equations Value Unit

Time integration Low order, fast algorithm -
Space discretisation Low order, fast algorithm -
Flooding and drying Value Unit
Drying depth 0.005 meter
Flooding depth 0.05 meter
Wetting depth 0.1 meter
Eddy Viscosity Value Unit
Horizontal eddy viscosity 0.02
Heat Exchange Value Unit
Time integration Low order, fast algorithm -
Space discretisation Higher order -
Constant in Dalton’s law 0.5 dimensionless
Wind coefficient in Dalton’s law 0.9 dimensionless
Sun constant, a, in Ångstrøm's law 0.295 dimensionless
Sun constant, b, in Ångstrøm's law 0.371 dimensionless
Standard meridian for time 153.45 Degrees, East
Beta in Beer's law 0.3 dimensionless
Light extinction coefficient 1
Clearness Coefficient 70 Percentage
The dispersion coefficient (m2/s) that accounts for the sub-grid mixing processes was through
a trial and error process found to be highly related to cell size, and thus a spatial dispersion
coefficient map was created by defining a unique dispersion coefficient for each calculation
cell through the following standardized equation:
(Eq.3.12)
Where is the dispersion coefficient while is the area of element cell i. Any instability
found in individual cells caused by the above formulation was manually decreased to a value
where they no longer produced any instability in the transport equations.
Bed resistance was described in the model setup by defining a spatial map of the model
domain with a value of the Manning’s number, for each cell. The default value
38
for the entire domain was set to 50, and later decreased to lower values (thus increasing bed
resistance) for local areas where high current velocities caused numerical instability during
model simulation.
3.2.7 VALIDATION DATA

For the purpose of validating the outputs of the hydrodynamic model, a wide range of
measurements were obtained throughout 2010. The following sub-sections give a brief
explanation to the origins and/or methodology used to obtain the validation data.
3.2.7.1 HISTORICAL DATA

The initial calibration of the HD model relied on historical data obtained from two data points
located in the middle and upper reaches of the estuary (Figure 3.2.4). The data covers a
period from 29/03/2007 to 01/05/2007 and was provided by the Griffith Centre for Coastal
Management (GCCM). The data consisted of surface elevation (meters), temperature (°C)
and salinity (ppt) with a temporal resolution of 15 min, and logged using Greenspan CTD's.
Due to the lack of supplementary information of this data, e.g. deployment depth, specific
Greenspan CTD type, calibration method and accuracy levels, it only served as a secondary
validation period compared to the data obtained through the efforts of this study.
3.2.7.2 ADCP DATA

As a means to obtain validation data for the hydrodynamic model outputs, two Nortek
Aquadopp Current Profilers (ADCP) were deployed in the middle and upper reaches of the
Tallebudgera system (figure 3.2.4) in the period 23/9-2010 to 20/10-2010. Both ADCP’s
were deployed in the middle of the river stream at ca. 4 m depth. The Aquadopp Profiler uses
three acoustic beams slanted at 25° to accurately measure the current profile in a user
selectable number of cells. The internal tilt and compass sensors tell the current direction and
the high-resolution pressure sensor gives the depth and the tidal elevation, with an accuracy
of ±1% of the measured value (Nortek 2008).
39
Figure 3.2.4: The deployment locations of ADCP's and CTD recorders in the middle and upper reaches of the
system (MR & UR). Blue stations mark the 2007 GCCM stations. The figure is plotted using MGA-56 projection
coordinates (unit: meters).
Both ADCP’s were set to an acoustic frequency of 2.0 MHz, with a bin size of 0.2 m and a
total of 20 bins. Sampling frequency was set to 600 seconds. Two custom-made wooden
frames were constructed for the purpose of keeping the ACDP’s in place on the riverbed and
preventing tilting of the instrument (Figure 3.2.5).
40
Figure 3.2.5: Picture of the constructed wooden frame with the ADCP mounted on it. All metal used was either
lead or stainless steel in order to reduce any effect on the internal compass.
Each frame was constructed using marine grade stainless screws to prevent drift of the
internal compass, and weighted down by 10 kg of lead dive weights. Furthermore, a 15 kg
cinder block was attached as an anchor to each frame by the end of an 8 m rope. Prior to
deployment, sensors and compass headings were tested and calibrated in accordance to the
Nortek user manual and a small westward drift of ~3.0 degrees were registered in the
compass bearings. Furthermore, the head of each ADCP was covered in zinc oxide cream to
prevent any biofouling. Data extraction and processing was done using the Nortek Storm
software (Nortek 2008), and Matlab 2007b.
3.2.7.3 TEMPERATURE/SALINITY DATA

Two Odyssey Salinity/Conductivity recorders were used for long-term deployment at
locations in the upper and middle reaches of the system in the periods of 19/1-2010 to 6/2-
2010 and 11/2-2010 to 30/3-2010. Reported accuracy levels lied within ±3% of the measured
value in 0-65 °C and 0-80 mS/cm conductivity range, respectively.
41
Prior to deployment both recorders were calibrated according to the Odyssey user manual,
and mounted inside an open PVC pipe housing that allowed ample water to flow through
while also acting as a protective casing for the recorder. Recorders were deployed ca. 0.3m
below the lowest predicted tide levels for the planned deployment period and set to
continuously log temperature and conductivity every 10 min throughout their deployment
period. A weekly measurement of temperature/conductivity using the YSI 6920 Sonde
(described in Chapter 1) at each deployment location was made in order to check for any
potential drift in the Odyssey recorders during deployment.
Data extraction was done using the supplied Odyssey software, and conductivity
measurements were then converted into salinity for comparison with model outputs using the
Unesco standard algorithms for computation of fundamental properties of seawater (Fofonoff
1983). YSI temperature/salinity data (described in detail in Chapter II) were furthermore
utilised to validate spatial and seasonal variations in the model output of temperature and
salinity.
3.2.8 CALIBRATION & DATA ANALYSIS

Calibration of surface elevation and current velocities was achieved by varying the Manning
number throughout the domain, while also continuously optimizing the mesh resolution of
the domain, until a satisfactory fit between simulated and observed values was achieved.
Salinity and temperature were calibrated by varying the dispersion coefficient throughout the
domain (section 3.2.6) until the best possible fit was obtained under the limitations of
applying a 2D model to simulate transport/diffusion processes that are inherently three-
dimensional in nature. Temperature was furthermore calibrated by varying the user-specified
constants (table 3.2.1) in the heat exchange module of the model setup.
Model outputs in terms of surface elevation, current speed, current direction, temperature and
salinity were extracted from the simulation result files at the corresponding locations and
time periods of the measured validation data. In accordance with DHI standards, all
individual datasets were analysed by calculating a model Quality Index consisting of the
mean, bias, standard deviation, root mean square error (RMS) and the correlation coefficient
between simulated and measured variables. Bias was defined and calculated as the mean of
the difference between simulated values and corresponding measured values. RMS error was
defined as the square root of the mean of the squared differences between modelled and
measured values. Time series as well as spatial plots for each type of dataset were generated
42
using Matlab R2007b and the built-in MIKE Plot Composer functionality for visual
comparison between simulated and measured values.
3.3 RESULTS
3.3.1 SURFACE ELEVATION VALIDATION

Modelled surface elevation for the initial calibration period of 2007 was in good agreement
with measured values at the MR-2007 Calibration point (MR-2007), with a correlation
coefficient of 0.962 along with bias and root mean square (RMS) values of 0.069 an 0.121,
respectively (Figure 7). The bias indicates that there is only a small difference on average
between measured and simulated data. The relatively high RMS value (compared to the bias
value), however, suggests that there is a high variation between differences. The positive bias
is because the tidal wave trough is slightly underestimated in the model as seen from below
figure. This is most commonly due to limitations in the accuracy of the measured bathymetry
of the system. Limitations in grid resolution through narrow gaps in the river could also affect
this.
Figure 3.3.1: Modelled surface elevation (red line) plotted against measured values (blue line) during the
period of 29/3-2007 to 8/4-2007 at the MR-2007 calibration point.
43
For the UR-2007 calibration point (UR-2007), the correlation coefficient was slightly lower
(R = 0.918), with bias and RMS values of 0.068 and 0.166, respectively. The same general
pattern therefore seems to be present at both the UR-2007 and the MR-2007 calibration point.
Surface elevation for the 2010 middle reach calibration point (MR-ADCP) was similar to
2007 in demonstrating good agreement with measured values, with a correlation coefficient
of 0.963 and a bias of 0.036. The RMS value of 0.102 does, however, indicate relatively high
variability in the differences between simulated and measured values. The simulated vs.
measured values for the Upper reach calibration point (UR-ADCP, Figure 3.3.3) displays the
same trend as MR-ADCP, with a correlation coefficient 0.969 together with bias and RMS
values of 0.183 and 0.203, respectively. The increased bias and RMS values of the UR-
ADCP calibration point suggest that model accuracy in terms of tidal prediction decreases
with increased distance from the offshore boundaries of the system.
period of 24/9-2010 to 19/10-2010 at the MR-ADCP calibration point.
As evident from Figures 3.3.2 and 3.3.3, a small flood event occurred in the system between
the 12th and 13th of October due to a heavy rainfall event, leading to freshwater discharge
levels at the upper reach of > 110 cumecs. As from the above-mentioned figures, the model
performs well in capturing the elevated water level at both calibration points, confirming the
44
model’s ability to handle sudden flooding events, with only minor discrepancies from
measured levels.
period of 24/9-2010 to 19/10-2010 at the UR-ADCP calibration point.
3.3.2 FLOW VELOCITY VALIDATION

Simulated current speeds for the UR-ADCP calibration point (Figure 3.3.4) for the entire
measuring period display a weak correlation of 0.28 with measured values, together with
bias and RMS values of 0.011 and 0.020, respectively. However, there is a strong correlation
(R = 0.980) in the period 10/10 to 14/10, where increased discharge levels caused an increase
in current speed. The results shown in Figure 3.3.5 suggest that during dry periods the model
predicts a current speed that fluctuates between 0.4 cm/s and 7 cm/s, corresponding to a very
weak tidal signal. The abrupt-fluctuating measured values suggest an almost non-existent
tidal influence in this part of the system in the model, ultimately causing the low correlation
coefficient between data sets. It is also important to note that the bias of 0.011 corresponds to
a mere 1.1 cm/s difference in mean current speeds. The model’s ability to accurately simulate
current speeds in this end of the system, however, was confirmed during high discharge
events (the 10th to 14th of October).
45
Figure 3.3.4: Simulated current speed (red line) plotted against measured values (blue line) during the period
of 24/9-2010 to 19/10-2010 at the UR-ADCP calibration point.
of 24/9-2010 to 27/9-2010 at the UR-ADCP calibration point.
Simulated current directions for the UR-ADCP calibration point (Figure 3.3.7) were found to
be affected by the same issue as described above with a correlation coefficient of 0.56
together with bias and RMS values of -10.89 and 65.23, respectively. As it is evident from
Figure 3.3.7, the predicted current directions followed a clear tidal pattern during dry periods,
whereas measured values fluctuated abruptly with only a weak tidal signal present. The
abrupt fluctuations are likely an indication of small local eddy vortexes, which are below the
mesh resolution of the model grid, and thus not possible to capture directly. During the
discharge event, the correlation between data sets increases to 0.75 while the relative values
of the bias and RMS drop to 4.77 and 36.74, respectively. Taking the drift of ~3.0 degrees of
46
the ADCP compass into account (Section 3.2.7.2), the bias is reduced even further; however,
the high RMS value still indicates a high variability in the differences.
Figure 3.3.6: Simulated current directions (red line) plotted against measured values (blue line) during the
While the reason remains unclear, the ADCP deployed in the Middle reach failed measuring
reliable data, possibly due to the instrument sinking into the fine sediment and blocking the
beams responsible for measuring the u- and v-components. The measured values of current
speed and direction are displayed against simulated values in Figure 3.3.8 and 3.3.9,
respectively.
of 24/9-2010 to 19/10-2010 at the MR-ADCP calibration point.
47
As it is evident from Figure 3.3.8, measured values of current speeds are extremely high,
topping at ~1.6 m/s during dry periods, which is nearly 15x higher than the predicted levels,
and peculiarly dropping below 0.5 m/s during the flooding event. Furthermore, neither
measured values for current speed nor direction display any persistent tidal signal, adding to
the growing evidence of equipment failure.
Figure 3.3.8: Simulated current directions (red line) plotted against measured values (blue line) during the
The lack of reliable validation data for the Middle reach unfortunately prevents assessment of
model performance and accuracy in terms of flow velocities for this part of the system.
However, due to the good agreement between predicted and measured values in the Upper
reach, it is plausible that model performance for the Middle reach is in good agreement as
well, because of its closer proximity to the tidal forcing from the offshore model boundary.
3.3.3 SALINITY VALIDATION

Due to a large freshwater discharge event taking place in between the two sampling periods
for salinity and temperature validation data, the two periods can be regarded as a temporal
snapshot of the system in a dry-period state (First period) and a wet-period state (Second
period), respectively. The different nature of the two sampling periods allowed for an
opportunity to assess model performance in terms of salinity during these contrasting system
states.
Simulated levels of salinity for the first period at the MR-CTD calibration point was found to
be in good agreement with measured data, with a correlation coefficient of 0.84 together with
48
bias and RMS values of 1.96 and 2.22, respectively. As indicated by the bias value, the
model tends to simulate slightly elevated levels of salinity compared to measured values, and
it follows from Figure 3.3.9 that a constant discrepancy between the measured and simulated
values occurs around the low tide mark.
Figure 3.3.9: Simulated salinity (red line) plotted against measured values (blue line) during the period of
20/1-2010 to 1/2-2010 at the UR-CTD calibration point.
Simulated salinity for the calibration point located in the Upper reach was in less good
agreement with measured values, with a correlation coefficient of 0.49 together with bias and
RMS values of -10.21 and 10.34, respectively. Tidally caused fluctuations in salinity were hardly
present in the measured data set, whereas the model predicted salinity level considerably
lower, but with a clear tide-induced fluctuation. While not a definitive answer, this result
suggests that the Upper reach CTD was placed too deep in the water column where salinity
levels were high, while average background levels of freshwater discharge during that period (>
0.05 cumecs) only affected salinity in the surface layer. Due to the depth-averaged solution
technique of the model, this kind of variation caused by stratification was therefore not
captured.
49
11/2-2010 to 29/3-2010 at the MR-CTD calibration point.
The performance of the model dropped in the second sampling period for the Middle reach,
with a correlation coefficient of 0.78 together with bias and RMS values of -1.90 and 4.63,
respectively. As evident from Figure 3.3.10, both data sets display a clear tide-dominated
pattern in fluctuations, but with a relatively big difference in the magnitude. While these results
indicate that the simulated saltwater intrusion into the system was small compared to observed
salinity levels, the good agreement between simulated and measured surface elevation (Section
3.3.1) dictates otherwise.
11/2-2010 to 29/3-2010 at the UR-CTD calibration point.
50
The UR-CTD calibration point had a correlation coefficient of 0.57, together with bias and
RMS values of 0.059 and 1.784, respectively. While the mean value between measured and
simulated was in good agreement with a very small bias, it is evident from Figure 3.3.11 that
the model failed to capture the two observed saltwater intrusions accurately. Taking the
results from the Middle reach as well the first sampling period into account, it is plausible
that the sub-grid variation of salinity levels increases when the system leaves its steady-state
(dry periods), thus making the comparison between point measurements and depth-averaged
model cells more difficult.
Figure 3.3.12: (Left) Seasonal variation in station-averaged salinity (blue line) plotted against corresponding
simulated values (red line). (Right) Annually averaged salinity for each measuring station in relation to river
mouth distance, plotted against corresponding simulated values. EHMP data is omitted from the two measured
data sets.
On a larger temporal scale, the model performed remarkably well with an average correlation
coefficient of 0.95, and the results for each measuring station (described in Chapter 2) are
summed up in Table 3.3.1. With an average correlation coefficient of 0.98, the model
performed just as well for the comparison between simulated and predicted salinity levels on
each sampling day in relation to each station’s location in the system. The results for the
spatial comparison are summarised in Table 3.3.2 and figures for both the temporal and
spatial analysis are available in Appendix I.
51
Table 3.3.1: Calculated results of the Quality index for each measuring station over the course of 2010 vs.
corresponding simulated values. See Appendix I for the plotted data sets.
TABLE 3.3.1: TEMPORAL QUALITY INDEX - SALINITY

Station Correlation Simulated Mean Measured Mean Bias RMS
1 0.87 33.11 31.93 1.18 0.68
2 0.94 32.52 31.32 1.20 0.67
3 0.96 30.33 29.41 0.92 0.52
4 0.95 29.36 27.51 1.85 0.65
5 0.99 23.26 23.48 -0.21 0.70
6 0.97 19.33 19.89 -0.56 0.19
7 0.97 12.99 14.15 -1.16 1.22
8 0.95 11.98 12.07 -0.09 2.46
9 0.96 8.60 9.51 -0.92 1.85
10 0.95 7.32 8.07 -0.75 1.15
11 0.92 4.74 5.29 -0.55 3.25
As evident from the results presented in Tables 3.3.1 and 3.3.2, the model performs quite well
on a seasonal scale while also capturing the spatial variation between the river mouth station
and the end of the tidal dominated zone on most occasions. Despite the discrepancies
between the simulated and measured values on a fine temporal scale, these results suggest
that the issues witnessed on a short temporal scale are not severe enough to significantly
affect model performance on a larger scale. However, fine temporal variations in
environmental parameters may still affect animal behaviour, which will be discussed in
further detail in the following chapters.
Table 3.3.2: Calculated results of the Quality index for each sampling day in relation to distance from the river
mouth vs. corresponding simulated values. See Appendix I for the plotted datasets.
TABLE 3.3.2: SPATIAL QUALITY INDEX - SALINITY

Sampling day Correlation Simulated mean Measured mean Bias RMS
1 0.99 26.18 27.39 -1.21 0.68
2 0.99 26.72 28.60 -1.89 0.67
3 0.83 8.67 2.38 6.28 0.52
4 0.99 20.04 19.36 0.68 0.65
5 0.99 14.02 11.08 2.95 0.70
6 0.99 12.21 10.96 1.25 0.19
7 0.99 18.98 17.40 1.58 1.22
8 0.99 20.04 21.13 -1.09 2.46
52
TABLE 3.3.2: SPATIAL QUALITY INDEX - SALINITY (Continued)
Sampling day Correlation Simulated mean Measured mean Bias RMS
9 0.99 23.70 26.36 -2.67 1.85
10 0.91 8.03 1.67 6.36 1.15
11 0.98 13.88 12.90 0.98 3.25
12 0.99 22.31 22.97 -0.66 2.52
13 0.99 21.67 21.21 0.46 2.29
14 0.99 20.09 23.47 -3.37 1.91
15 0.98 27.12 28.50 -1.39 1.17
16 0.98 24.81 25.45 -0.65 2.45
17 0.99 22.88 23.73 -0.85 1.32
18 0.99 27.35 28.33 -0.98 0.55
19 0.98 31.52 31.90 -0.39 0.13
20 0.96 32.21 31.88 0.33 2.62
21 0.99 26.69 27.30 -0.61 2.21
22 0.99 16.55 16.98 -0.43 3.59
23 0.99 14.74 16.40 -1.66 20.23
24 0.99 22.37 23.79 -1.42 15.90
25 0.99 18.89 20.54 -1.65 6.64
26 0.99 17.33 18.19 -0.86 12.81
27 0.99 4.91 3.39 1.51 2.68
28 0.99 16.14 15.90 0.24 2.26
29 0.99 8.57 7.21 1.36 2.03
30 0.98 13.72 13.50 0.21 5.81
3.3.4 TEMPERATURE VALIDATION

The simulated vs. measured fluctuations in temperature for the first period at the MR-CTD
calibration point were in poor agreement with a correlation coefficient of only 0.39. Upon
inspection of Figure 3.3.13, it is evident that the cause of the weak correlation is due to the
fact that the model predicted temperature fluctuations to be purely diurnal, while measured
values at times fluctuated semi-diurnal, resulting from the tidal cycle. The presence of a tidal-
influenced semi-diurnal temperature fluctuation in the measured data set suggests that there
was a noteworthy difference between water temperatures of the upper and lower reaches of
the system during this time period, which in turn was not captured by the model. Despite a
low correlation coefficient, a small bias value of 0.11 indicates that the simulated mean
temperature over the sampling period is in good agreement with the measured mean.
However, a high RMS value (relative to the bias) of 1.29 suggests high variability in between
individual differences, which is confirmed by Figure 3.3.13.
53
Figure 3.3.13: Simulated temperature (red line) plotted against measured values (blue line) during the period
of 20/1-2010 to 1/2-2010 at the MR-CTD calibration point.
A correlation coefficient of 0.41 for the UR-CTD calibration point in the first sampling
period indicates poor agreement with observed temperature levels, and unlike the MR-CTD
calibration point, a relatively high bias of -1.78 together with a RMS value of 2.27 were
registered. Upon inspection of the data sets in Figure 3.3.15, it becomes evident that the
measured temperature values did not fluctuate in a set diurnal or semi-diurnal pattern, which
explains the low correlation coefficient with the simulated diurnal fluctuations of the model.
The lack of a diurnal fluctuation pattern adds to the evidence that the CTD of the Upper reach
calibration point was deployed too deep (as mentioned in Section 3.3.3) and thus failing to
capture the immediate heat exchange between the atmosphere and surface layer.
of 20/1-2010 to 1/2-2010 at the UR-CTD calibration point.
54
Simulated fluctuations in temperature for the second period at the MR-CTD calibration point
was in good agreement with measured values, with a correlation coefficient of 0.88 together
with bias and RMS values of 0.37 and 1.04, respectively. The agreement between data sets
similarly increased for the UR-CTD calibration point, with a correlation coefficient of 0.73
together with bias and RMS values of 0.69 and 1.93, respectively. As evident from Figure 3.3.16,
the simulated temperature levels between day and night varies as much as nearly 4 °C, while
measured differences between day and night conditions typically only varies ~1.5 o C.
Considering that the simulated model values were predicted for a roughly 200 m 2 large depth-
averaged cell, it is unlikely that a fluctuation of 4 oC over a single diurnal cycle is representative
of the actual conditions. Possible reasons for these model predictions will be discussed in
further detail in Section 3.4.
of 11/2-2010 to 29/3-2010 at the MR-CTD calibration point.
Similar to the predicted levels of salinity, the model performed quite well at a larger temporal
scale, with an average correlation coefficient of 0.93.The results for each measuring station
are summarised in Table 3.3.3. Spatially, the model performed less accurately, with an
average correlation coefficient of 0.73. The results for the spatial comparison are summarised
in table 3.3.4 while graphed figures for both the temporal and spatial analysis are available in
Appendix I. Since there were only 11 comparison points and little to no variation in
temperature between the river mouth and the uppermost sampling station on most occasions,
the spatial correlation analysis for temperature was prone to be adversely affected by even
small outliers deviating from the mean difference (>2 oC). This is in contrast to the spatial
55
correlation analysis for salinity, which displayed a high level of spatial variation throughout
the year.
Figure 3.3.17: (Left) Seasonal variation in station-averaged temperature (blue line) plotted against corres-
ponding simulated values (red line). (Right) Annually averaged temperature for each measuring station in
relation to river mouth distance, plotted against corresponding simulated values. EHMP data is omitted from
the two measured data sets
As evident from the results presented in Tables 3.3.3 and 3.3.4, the model performs quite well
on a seasonal scale while also capturing the spatial variation between the river mouth station
and the end of the tidal dominated zone on most occasions. Similar to the case with salinity in
Section 3.3.3, the discrepancies between simulated and measured temperatures on a fine
temporal scale further suggest that the observed model discrepancies on a short temporal
scale are not severe enough to significantly affect model performance on a larger scale.
56
Table 3.3.3: Calculated results of the Quality index for each measuring station over the course of 2010 vs.
corresponding simulated values. See Appendix I for the plotted data sets.
TEMPORAL QUALITY INDEX - TEMPERATURE

Station Correlation Simulated Mean Measured Mean Bias RMS
1 0.91 24.48 23.77 0.71 2.23
2 0.94 24.47 23.60 0.86 2.21
3 0.92 24.13 23.75 0.38 1.74
4 0.93 24.15 23.68 0.47 1.49
5 0.94 24.13 23.87 0.26 0.96
6 0.94 23.77 23.60 0.18 0.10
7 0.96 23.42 23.64 -0.22 1.28
8 0.95 23.21 23.55 -0.34 0.95
9 0.96 23.16 23.38 -0.21 1.50
10 0.94 23.12 23.40 -0.27 1.94
11 0.89 23.00 23.14 -0.13 2.54
Table 3.3.4: Calculated results of the Quality index for each sampling day in relation to distance from the river
mouth vs. corresponding simulated values. See Appendix I for the plotted data sets.
SPATIAL QUALITY INDEX - TEMPERATURE

Sampling day Correlation Simulated Mean Measured Mean Bias RMS
1 0.95 29.44 29.41 0.03 2.23
2 0.74 27.79 26.44 1.35 2.21
3 0.91 25.05 25.36 -0.31 1.74
4 0.64 27.83 28.88 -1.05 1.49
5 0.91 26.36 25.43 0.93 0.96
6 0.82 24.05 23.57 0.49 0.10
7 0.84 26.06 26.20 -0.14 1.28
8 0.77 25.65 26.51 -0.86 0.95
9 0.55 24.41 24.33 0.08 1.50
10 0.88 20.92 19.97 0.95 1.94
11 0.93 21.44 21.99 -0.55 2.54
12 0.90 21.79 21.03 0.76 1.33
13 0.67 19.80 20.37 -0.57 1.28
14 0.41 18.51 20.02 -1.51 1.35
15 0.81 19.84 19.89 -0.05 1.47
16 0.93 20.09 19.53 0.57 1.15
57
SPATIAL QUALITY INDEX – TEMPERATURE (Continued)
Sampling
day Correlation Simulated Mean Measured Mean Bias RMS
17 0.83 20.80 20.04 0.76 1.35
18 0.58 23.05 22.65 0.41 2.10
19 -0.33 22.26 21.15 1.11 1.77
20 0.27 22.20 21.28 0.92 1.78
21 0.86 23.06 22.69 0.37 1.31
22 0.77 21.55 21.92 -0.37 0.04
23 0.71 24.58 25.08 -0.50 0.13
24 0.81 23.93 23.03 0.90 0.19
25 0.95 26.54 25.87 0.67 0.00
26 0.85 26.75 26.09 0.65 0.16
27 0.70 21.91 23.30 -1.39 0.55
28 0.86 24.56 23.52 1.04 0.78
29 0.64 24.06 24.03 0.03 1.03
30 0.88 27.67 27.85 -0.18 0.50
3.4 DISCUSSION
The HD model captured measured surface elevation to a satisfactory level (Figures 3.3.1 to
3.3.3). However, there was a tendency for the model to overestimate the trough of the ebb
tide, which in turn resulted in a positive bias between simulated and measured surface
elevation. Furthermore, there was a slight drop in model accuracy from middle reach to
upper reach calibration points. Since the initial model calibration by the application of a bed
roughness sensitivity analysis of the model (Section 3.2.9) did not reveal a notable change in
the model prediction of trough elevation (ebb tide), it is unlikely that insufficient model
representation of bed roughness is the cause of the discrepancies.
It is more plausible that the main reason for the overestimation of the ebb tide elevations and
observed decline of accuracy with increased distance to the offshore boundaries is due to
limitations in accuracy of the measured bathymetry of the creek. Current and accurate
bathymetry is one of the most fundamental pre-requisites for successful modelling (Cheng et
al. 1991). With regards to the bathymetry made available for this study, there are three
primary limitations. 1) The model bathymetry data dates back to 2001, which means that
nearly a decade of sediment transport, in particular around the river mouth, could have altered
several characteristics of the bottom contour, while flooding events during the wet season are
58
likely to have eroded parts of the upper reach; 2) due to relative high sediment disposition
rates at the river mouth, the Gold Coast City Council annually dredge the main channel to
keep the river mouth open (with an annual average of 38000 m3 sand removed, GCCM,
unpubl. data). While the bathymetry adopted resembles the immediate post-dredging
conditions, it is likely that model accuracy will be affected as sediment is redeposited over
time in the natural system, resulting in changed flow dynamics; 3) bathymetry data was
measured using LIDAR remote sensing (Light Detection And Ranging), which means the
accuracy of measured bathymetry data is likely to decline for the upper reach compared to the
wider downriver sections, due to increased tree and mangrove foliage obstructing
measurements along the river edges.
While the aforementioned issues with the bathymetry offer a plausible explanation for the
observed discrepancies in surface elevation, limitations in model grid resolution through
narrow and shallow gaps in the river could also affect model performance in terms of
simulating surface elevation. As mentioned in section 3.2.4, the smallest grid cells were 5x10
m in size (width*length), which can be considered quite small in comparison to the 30x30 m
cells utilised in a similar HD model of the nearby Gold Coast, Broadwater estuary
(Mirfenderesk & Tomlinson 2007). However, due to flooding and drying of model grid cells
over the course of tidal cycles, in particular neap tides, the narrow and shallow parts of the
system are on occasion represented by relatively few inundated grid cells, which can
ultimately result in a decrease of model accuracy in areas upriver from these bottlenecks.
Inaccurate bathymetry definition and narrow sections within a model domain was likewise
reported as a cause for model discrepancies in Dias and Lopes (2006).
While a decrease in cell size can help eradicate the potential problems associated with model
bottlenecks, it is important to note that a decrease in cell size and the subsequent increase of
total number of cells has a significant impact on required simulation times. Therefore, there is
ultimately a compromise between cell grid resolution, model performance and the associated
simulation times. In summary, since problems with inaccurate bathymetry data along the
river often has a cumulative negative effect on model performance, it is common that the
correlation between modelled and measured surface elevations decreases upstream, while the
severity of the decrease is likely to be enhanced by the presence of bottlenecks. Finally,
meteorological events such as large low-pressure systems, which are not represented in the
model formulation, could potentially cause some of the observed discrepancies between the
simulated and measured water levels. For example, in the immediate period after the
59
significant flooding event that took place between the 12th and 13th of October 2010, the
model underestimated the actual water level compared to measured levels.
Figure 3.4.1: Simulated current speeds at a model bottleneck during low tide (top), incoming tide (middle) and
high tide (bottom). The bottleneck is located ~3 km upriver from the river mouth, while the first ADCP
calibration point is located ~3 km further upriver from the bottleneck.
60
The failure of the middle reach ADCP unfortunately prevented a direct comparison between
measured and simulated flow velocities for this part of the estuary. However, since
Tallebudgera Creek is a tide-dominated estuary, the predominant flow velocities throughout
the system are undoubtedly closely linked to the time and magnitude of the tide. Also, the
good agreement between measured and simulated surface elevations for the middle reach area
suggests that simulated current velocities are in good agreement as well. The model
overestimated the current speeds for the upper reach area during the low-discharge period
prior to the flooding event. Because the currents are so small in the upper reach during dry-
period conditions, it is almost impossible to model these accurately with the aforementioned
model constraints in regards to bathymetry accuracy and model grid resolution. However, the
good agreement between measured and simulated current velocities during the flooding event
shows how model performance in terms of flow velocities increases when the number of
"flooding and drying" cells (cells that are continuously flooded during high tide and
subsequently dried out during low tide) decreases as an effect of increased freshwater
discharge.
As evident from the results, the model showed a delayed response time and to some degree
an underestimated amplitude in oscillations of salinity levels immediately after a discharge
event had pushed the system away from its dry-period quasi-stationary equilibrium. Despite
having a good representation of surface elevation and flow velocities, the slow response time
of the model before returning to "normal" salinity levels likely reflects the following two
issues: 1) insufficient model resolution of subscale turbulence dispersion processes that drive
the mixing of fresh and saline water masses, and thus determine how quickly the system is
restored to "normal" conditions after a rainfall event. Due to the limitation of the spatial
resolution of Tallebudgera Creek, it is not possible to resolve these subscale processes by
utilising the MIKE21FM Smagorinsky eddy-viscosity turbulence model formulation to a
satisfactory degree, and thus subscale dispersion processes had to be resolved by assuming a
temporally-constant, but spatially-varying dispersion coefficient formulation (Section 3.2.6).
In a MIKE21 water quality study on a shallow lagoon, Lopes et al. (2008) notes that using a
spatially-varying dispersion coefficient is more accurate; however, the results of this study
indicates that a dispersion coefficient that varies spatio-temporally is needed to improve
model performance. 2) While it was established in Chapter 2 that there was no significant
difference between top and bottom salinities at individual measuring stations on an annual
scale, one cannot rule out the possibility that stratification occurred in the deeper parts of the
61
system (> 3 m) over shorter temporal scales (especially during local events of small to
medium rainfall events), which in turn could bias the comparison between modelled and
measured values due to the depth-averaged nature of the HD model. With the relatively large
difference in density between oligohaline creek water and offshore euryhaline water, it is
plausible that the first few saltwater intrusions following a major freshwater discharge event
could cause brief haloclines to form in parts of the system. The depth-averaged MIKE21
model has in the past been successfully implemented for simulation of salinity and
temperature in a shallow lagoon (Zacharias & Gianni 2008). However, with an average and
maximum depth of 0.4 and 2 m respectively, it is likely that any vertical differences in
salinity are miniscule compared to potential differences in Tallebudgera Creek. While an
implementation of a 3D HD model would allow a test of this hypothesis, and possibly result
in an improvement of the aforementioned dispersion processes, it was outside the timeframe
and scope of this study to conduct this investigation. Furthermore, it is possible that the
computed evaporation scheme utilised in the model setup is insufficient to capture the actual
evaporation taking place, which in turn could affect modelled salinity levels. Zacharias &
Gianni (2008) utilised calculated time series of evaporation in their study of a shallow
hypersaline lagoon in southern Greece, and it is possible that an adaptation of their method
could improve model performance.
Despite the abovementioned problems, the model performed well in resolving oscillations of
salinity once a quasi-stationary condition was present (Figure 3.3.6), and captured observed
salinity gradients on most sampling days satisfactorily (R > 0.90). In summary, the model
limitations with the simulation of salinity seem to be linked to a fast response to increased
freshwater inputs while the subsequent response to saltwater intrusion remains slow, with an
underestimation of the amplitude in salinity oscillations. However, the model’s capability to
resolve mean values on a larger spatio-temporal remains good.
Diel oscillations in temperature were often overestimated by the model. However, on larger
temporal scales, the model performed well in capturing mean temperature levels. Since the
model utilises the same equations for solving the advection and dispersion of both
temperature and salinity, the observed discrepancies between modelled and measured
temperature fluctuations can likely be explained, at least in part, by the abovementioned
issues in terms of the model’s dispersion formulation. However, in the case of temperature,
other important limitations in the current model setup are evident, which in turn will affect
model performance: 1) it was not possible to obtain cloud cover data for the Tallebudgera
62
Creek area, and thus the clearness coefficient was defined as the default value in the model
setup. Due to this obvious limitation, it is likely that the model will overestimate diurnal
fluctuations in temperature when cloudy conditions prevail in the natural system. In a similar
study, Zacharias & Gianni (2008) estimated the clearness coefficient through measurements
of solar radiation and rainfall, and it is possible that an adaptation of their method in future
development of the model could improve performance of the heat exchange module. 2) Time
series of wind speed/direction, air temperature and relative humidity are assumed to be
constant throughout the domain, and are furthermore measured ~10 km south from the
system domain. While there is no better meteorological data available for the Tallebudgera
area, it is plausible that differences between utilised data and actual conditions differ enough
to contribute to the apparent discrepancies in the simulation of temperature. Wind speeds can
in particular be expected to decrease when moving from the coastal river mouth to the more
sheltered and vegetated areas of the upper reaches. However, the effect of spatially-varying
wind speeds is unknown due to lack of data. 3) Offshore boundary values of temperature rely
on the performance of the HYCOM model, from which daily values are extracted. With a
grid resolution of 0.25° × 0.25° it follows that any discrepancies between the sea surface
temperatures predicted by the HYCOM model and actual offshore conditions will be
transferred to the Tallebudgera Creek model.
It follows from the above that there is a variable degree of uncertainty attached to the use of
modelled temperature and salinity. In the case of salinity, this uncertainty increases in the
immediate time after larger discharge events. This impact naturally has implications for the
functioning of any future implementation of an agent-based model that relies on salinity as a
driver for movement. While this issue cannot be disregarded (Dias & Lopes 2006), it is
important to note that due to the narrow nature of Tallebudgera Creek, the salinity gradient is
primarily longitudinal in nature, with little to no variation in salinity at any given cross-
section of the river. This means that any favourable habitat in terms of salinity would during
post-discharge events be situated slightly further downriver in the model compared to its true
position in the system.
In conclusion, despite the aforementioned assumptions and limitations of the model setup in
regards to the accuracy of simulated parameters, the model still met an overall satisfactory
level of accuracy, and was sufficiently representative of actual conditions in the system. The
model was found to have a delayed response in salinity after large freshwater discharge, but
predicted mean levels were in good agreement with measured data. Diel oscillations of
63
temperature were at periods overestimated by the model, but predicted mean temperature
over time was in good agreement with measured data sets. While improvement of model
performance is possible through an implementation of higher quality data, in particular more
refined bathymetry data, one has to appreciate the fact that this work was solely supported by
publically available data, making the model in its current setup quite cost-effective.
Furthermore, the good agreement in surface elevation between both 2007 and 2010 data
suggests that the model in its current form can be readily reapplied for future studies of a
similar nature in Tallebudgera Creek by using only current datasets for the period of interest.
64
CHAPTER IV
SHORT-TERM MOVEMENT OF A JUVENILE BULL SHARK IN
TALLEBUDGERA CREEK
65
4.1 INTRODUCTION
In the past decades, elasmobranch species have faced increased fishing pressures causing a
substantial decrease in populations (Graham et al. 2001), while rapid increases in
urbanisation of coastal and estuarine regions over the same period (Lee et al. 2006, Martínez
et al. 2007) have increased the pressure on the natural nursery grounds of estuary-dependent
species, such as C. leucas. Once a common cosmopolitan species in the tropical and
subtropical regions, the bull shark is now listed as ‘near-threatened’ by the International
Union for Conservation of Nature (IUCN), and as a species with slow growth rates
(Branstetter & Stiles 1987, Schmid & Murru 1994) and late sexual maturity (Voight &
Dietmar 2011), it is less able to compensate for increased mortality and is therefore likely to
be more vulnerable to extinction (García et al. 2008). The need for conservation strategies for
C. leucas is imminent, and further investigation in how anthropogenic disturbances and
environmental change affect the pattern of habitat use in their riverine nurseries is needed in
order to formulate effective management strategies for this species.
Heupel & Simpfendorfer (2011) found that mortality rates of young C. leucas in estuaries
were low compared to other juvenile sharks in neighbouring marine nursery grounds, and
hypothesised that this was due to a lower predation risk from other sharks, coupled with a
decreased competition for resources as other juvenile shark species were absent from
estuaries. It is, however, unclear how juvenile bull sharks cope with the increased impacts of
urbanisation on their natural nursery grounds, and if this reduces the low-mortality value of
these systems.
The physical characteristics of riverine environments include a significantly smaller size and
volume relative to the open coastal environment. Riverine ecosystems are therefore also often
more prone to environmental fluctuations in both time and space. Thus, it is likely that
neonate and juvenile C. leucas inside riverine habitats are faced with more frequent changes
in flow and water quality, which in turn might translate into a response in movement and
changes in their distribution pattern (Heupel & Simpfendorfer 2008, Ortega et al. 2009).
With changes in hydrology as a common effect of urbanisation, the transport and advection-
dispersion of salinity is likely to be affected as well (Lee et al. 2006). Salinity in particular
has been correlated with the movement and distribution of young C. leucas (Simpfendorfer et
al. 2005, Heupel & Simpfendorfer 2008, Ortega et al. 2009, Werry 2010). While the
66
physiology behind the osmoregulatory capabilities of the bull shark at different life stages has
been investigated extensively in previous studies (Thorson et al. 1973, Pillans & Franklin
2004, Pillans et al. 2005, Pillans et al. 2006), the metabolic costs associated with
osmoregulating are yet to be established. Pillans et al. (2006), however, reported that juvenile
bull sharks face a greater osmoregulatory challenge when moving from 75% to 100%
seawater than their larger counterparts (> 2 m TL), which suggests that juvenile bull sharks
might avoid euryhaline waters as a means to reduce energetic costs. This notion is further
backed by the results of field-studies, which have reported a 7-20 salinity preference for
juvenile C. leucas, and avoidance of salinities < 7 PSU (Simpfendorfer et al. 2005, Heupel &
Simpfendorfer 2008, Ortega et al. 2009).
While field-studies of C. leucas have identified several other potential drivers of movement
in estuaries (Chapter I), these short-term tracking studies have normally relied on sampling
from a boat 50-100+ m away from the animal, or by inferring correlation with measured
parameters from one or more fixed points in the study site (Heupel & Simpfendorfer 2008,
Ortega et al. 2009, Werry 2010). This limitation introduces a potential bias in the analysis due
to the non-uniform advection/dispersion and transformation of these parameters across the
system. Further, it is often very difficult to assess spatio-temporal gradients of the various
parameters over the course of a tracking event, and how they might affect animal movement
and distribution. Furthermore, most short-term tracking studies only track an individual
animal once, which prevents an assessment of how significant changes in hydrology over
time and space affect the fine-scale movement and habitat selection of each individual.
By combining replicated short-term acoustic tracking of individual C. leucas with the outputs
of a hydrodynamic model (Chapter III), it is possible to investigate the spatio-temporal
variation of the hydrology of a system on a relatively fine scale, and allows for an
examination of how the tracked animal responds to changes in hydrology through time and
space. The assessment of individual variations in movement patterns relative to changes in
hydrology will provide more detailed knowledge of C. leucas movement responses to
environmental change, and may prove to be a key to predicting movement and distribution
through the application of an agent-based model.
67
As a means to better understand individual variation in movement of juvenile C. leucas,
replicated short-term acoustic tracking of individual C. leucas was conducted and correlated
with the outputs of a hydrodynamic model of Tallebudgera Creek as well as traditional water
quality sampling. However, the main purpose of this study is to obtain a detailed track record
of juvenile C. leucas inside the Tallebudgera Creek, which in turn can be used as validation
data for the development of an agent-based model that is capable of replicating and predicting
short-term movement of this species (Chapter V).
4.2 METHODOLOGY
4.2.1 CATCHING AND TAGGING

Capture of sharks occurred in the period of 13/1 to 17/1-2010, and was approved by the
Animal Ethics Committee of Griffith University (ENV/1709/AEC 2010). Protocol for
capturing and tagging followed Werry (2010). Ten baited drum lines with a breaking tension
of 1000 kg were set in the middle reach of the creek and sea mullet, Mugil cephalus, a local
and common resident in the system was used as bait on non-barbed 8/0 tuna hooks. The lines
had lengths varying from 5 to 15 m and were secured in position by a large float at the top
and a 7 kg brick as an anchor. After initial deployment, all lines were checked every two
hours and re-baited if necessary. In addition, a gill net with a mesh size of 8x8 cm and length
of 20 m was stretched across the creek at selected locations. The gill net was continuously
deployed and relocated in the middle and upper reaches of the system, with an average
deployment time of 1 hour at each location (Figure 4.2.1).
68
Figure 4.2.1: Picture of the deployed gill net. Due to the narrow nature of the system, the gill net was often
capable of stretching almost across the full width of the estuary.
Captured sharks were immediately removed from the gill net and restrained in a harness
specially designed for shark capture and handling by Werry (2010) (see Figure 4.2.2). There
was no by-catch of other species except a single stingray that was released immediately after
capture. Upon a capture event, the surface salinity and temperature were measured using an
YSI 6820V2-1 sonde (see section 3.3), and the captured shark kept in a 3 m long harness in
order to control and restrain movement for the duration of the tagging. Each shark was
marked externally with a numbered roto-tag as well as a VEMCO VR16-4H-R64K coded
acoustic transmitter with a set frequency for future tracking purposes. The roto-tag and VR16
pinger were secured to each other using 2 mm thick stainless steel wire, thus allowing the two
tags to be applied to the shark simultaneously. A hole was drilled through the approximate
centre of the dorsal fin using an electrical power drill (drill diameter 0.5 cm), through which
the combined roto-VR16 tag was attached by securing the standard locking mechanism of the
roto-tag.
69
Figure 4.2.2: Juvenile bull shark suspended in the specially designed harness, while water flow across the gills
was maintained by directing the boat against the current. Note the combined acoustic/roto-tag attached to the
shark’s dorsal fin (red arrow).
After applying the tag, various morphological dimensions of the shark were measured using a
soft plastic measuring tape, including: pre-caudal length (PCL), fork length (FL), total length
(TL) and clasper length for males. The weight of each shark was measured using a handheld
scale. Effort was made to perform tagging and measurement of morphological dimensions as
fast as possible and took on average less than 10 minutes per individual. After all data had
been recorded, the shark was kept in the harness for a few additional minutes while the boat
moved at a slow cruising speed to keep a steady flow of water through the gills of the shark,
thus allowing it to regain strength before release in the immediate area where it was caught.
4.2.2 ACOUSTIC TELEMETRY AND TRACKING METHODS

All tracking events took place on a 4.3 metre Stacer Seahorse boat equipped with a Yamaha
30 HMH, 22 kW, 2-stroke engine. For short-term tracking of the tagged animals, a VEMCO
VR60 Ultrasonic Receiver was used together with a VH65 Omni-directional hydrophone and
a V10 directional hydrophone. Together these allowed for initial detection of the tagged
animals using the VH65, followed by directional detection of the animals using the VH10.
The VH10 unit was mounted on a 3 m long aluminium pole, reaching ca. 30 cm beneath the
keel of the boat. The top piece of the pole was so constructed as to allow for a 360° rotation
of the hydrophone. A Suunto Commander lockable compass was mounted to the rotating top
piece in order to make accurate recordings of heading of the directional hydrophone. Once an
70
animal was located in the system, a recording of hydrophone heading and signal strength of
the hydrophone as well as hydrophone gain was made every 5-10 min for the following six
hours.
For each hydrophone recording, the position of the boat was marked using a Garmin 60
handheld GPS navigator with an estimated accuracy of ca.10 m. While tracking, the boat
engine was kept off as much as possible while maintaining a distance of at least 30 m from
the animal. However, maintaining said distance was sometimes made impossible when the
animal turned around and started heading towards the boat. Whenever it became necessary to
follow the shark with the engine on, it was done so with the lowest speed possible. Any
visible boating and/or fishing activity during the tracking effort were recorded for possible
inclusion in future statistical analyses. Throughout the tracking, salinity, temperature,
dissolved oxygen, pH, turbidity and chlorophyll concentration were recorded at the location
of the research vessel at each fix-point, using one or two YSI 6820V2-1 sondes (For more
information, see Water Quality measurements). Since stratification is believed to play a
minor role in the generally well-mixed and shallow waters of Tallebudgera Creek, only the
surface water (ca. 20-30 cm depth) was sampled during tracking.
4.2.3 HYDROPHONE CALIBRATION AND SHARK POSITION

The calibration of the V10 directional hydrophone and the VEMCO VR60 Ultrasonic
Receiver was done by recording the range of signal strengths at various gain with the distance
known from the signal source (V-16 tag deployed ca. 50 cm under water). The calibration
process was done for three separate sections of the Tallebudgera estuary that differed in the
type of bottom substrate (sand, mud and rocky) in order to establish potential differences in
the performance of the hydrophone. As a means to establish and quantify the relationship
between a given signal strength and hydrophone receiver gain to an unknown distance, a
biharmonic spline interpolation was done on each recorded dataset (Sandwell 1987).The
interpolated values were plotted in a 3D surface plot in order to visualize the relations, and
are shown in Figure 4.2.3.
71
Figure 4.2.3: The resulting surface plots from each calibration. Note the clear differences in distances at
maximum signal strength and maximum gain (16 and 36 respectively), between the three calibrations. See main
text for further information on the associated accuracy levels of each calibration.
The individual calibrations revealed a clear difference in hydrophone performance in relation

to bottom substrate type. The accuracy of the sand bottom calibration was found to be ±5 m
up to a total distance of ca. 70 m before accuracy abruptly dropped to ±15 m, whereas the
mud bottom calibration decreased from ±5 to ±15 m after exceeding a total distance of ca. 40
m to the transmitter. The rocky bottom had an accuracy of ±5 m within ca. 20 meters of the
transmitter, but then dropped to ±15 m and as much as ±25 m after ca. 40 m.
The position of the shark relative to the boat was calculated post-tracking by converting the
recorded signal strength at a given hydrophone gain to a corresponding distance between the
signal source originating from the V-16 transmitter and the hydrophone. Due to the registered
differences in the acoustic properties of the system, each distance from the boat to animal fix-
point was calculated using the calibration scheme appropriate to the animal’s location; e.g.
distances to fix-points located in areas known to have a muddy substrate would be calculated
utilizing the signal/gain relationship determined by the calibration over muddy substrates.
The actual position of the animal was estimated using the GPS coordinates of the boat, the
compass heading of the hydrophone and the estimated distance between the signal source and
receiver as described above. The V-16 transmitter does not give any estimation on the depth
of the shark, and thus it was only possible to estimate the animal’s horizontal position in the
system. Videos of the movement path of each track were created using MATLAB R2007b,
72
and is available for viewing on the supplied DVD along with other video outputs of this study
(Chapter V).
4.2.4 MIKE21FM DATA

Model outputs from the validated hydrodynamic model (Chapter III), such as temperature,
salinity, current speed and direction together with total water depth were extracted at the
corresponding points in time and space for each animal fix-point. The average values of the
above-mentioned parameters were calculated for the entire model domain over the duration
of each tracking event in order to create spatial time-averaged maps, which allowed
assessment of the general hydrological conditions of the system. By dividing each model
output parameter into specific value ranges, the total water volume for each range was
calculated in order to assess the amount of available habitat in the specific water quality
ranges, over the duration of each tracking period. The offshore area of the model domain was
not included in the volume calculations, since it was not considered to be part of utilised
habitat by the neonate and juvenile animals living inside the system.
4.2.5 DATA ANALYSIS

The rate of movement (ROM) was estimated between fix-points assuming the animal swam
in a linear direction between consecutive fix-points. Similarly, the total distance covered over
the course of a track was calculated by the same assumption of linearity. This assumption
undoubtedly leads to an underestimation of the ROM and consequently also biasing the total
distance covered by the animal. The estimated ROM was further investigated using the non-
parametric Spearman rank correlation test as a means to detect possible effects of measured
water quality parameters and simulated model outputs on ROM. Due to the underlying bias in
the estimation of ROM and the associated problems with comparing water quality
measurements sampled ca. 30-40 m away from the actual location, the results of the
statistical tests were only intended as a guide to potential movement pattern.
By dividing both measured and simulated parameters into specific value ranges (as
mentioned in section 4.2.4), the proportion of time the animal spent in a given value range
was calculated for each track. Using the calculated volume proportions of temperature,
salinity, water depth and current speed (section 4.2.4), the predicted proportion of time spent
in a given parameter range was calculated by assuming that the size of the volume of each
parameter range was proportional to time spent. The observed vs. predicted proportions of
time spent was then tested using a -test, as a means to establish if there are any significant
73
differences between the two. Since it was not possible to calculate definite water volumes for
dissolved oxygen, pH and turbidity, these parameters were not included in the -test.
Furthermore, an adapted version of the Manly’s alpha preference index (Manly 1972), was
applied to datasets where volume values were available, in order to determine if the animal
displayed preferential behaviour relative to available water quality regimes. The adaptation of
Manly’s equation used was:
(Eq. 4.1)
Where and are the proportion of time spent in the and water quality regime,
while . Similarly, and are the volume proportions of the and
water quality regime relative to the total volume of the system. The value of ranges from 0
to 1, where 0 indicates no preference while values close to 1 indicates a high level of
preference, while .
Animal position in relation to the width of the river was determined by defining and
calculating a river width ratio for each fix-point, in order to establish if the animal had a
latitudinal preference in relation to the riverbank. The ratio was defined as:
(Eq. 4.2)
Where is the shortest distance from a fix-point to either adjacent riverbank, while is
the total width of the river at the given location. ranges from 0 to 1, where values
approaching 1 indicates that the animal is positioned near the middle of the river.
A local nearest-neighbour convex-hull construction of home ranges was calculated on the

basis of the total amount of recorded animal fix-points (Getz et al. 2007). However, due to
the elongated and narrow nature of the system combined with the widespread range of fix-
points throughout the system, the above-mentioned method produced unrealistic results by
including land areas as part of the home range. Instead, the home range of the animal was
estimated using more traditional and simplistic means by calculating the percentages of fix-
points in relation to the distance to the river mouth.
74
4.3 RESULTS
Three sharks were caught and tagged over the course of this study. All sharks were caught by
gill net (as described in section 4.2) in the upper reach of the river where the mean water
depth was > 2 m. All three sharks were caught in the net from the shallow banks of the river.
Two sharks were caught during the daylight hours with an outgoing tide, while the third was
caught on an incoming tide shortly after sunset. All sharks were caught in a salinity range of
12 to 15 PSU and at 30°C (Table 2.). Two of the sharks were neonates (a male and female)
and most likely young-of-the-year with fresh umbilical scars evident. The female neonate
(FM1) had a total length of 74.0 cm and a weight of 3.9 kg, while the male neonate (MN1)
had a total length of 78.2 cm and a weight of 2.9 kg. The third shark was a young juvenile
male (JM1) with a total length of 108.0 cm and a weight of 9.0 kg. Clasper length was 6.2 cm
and not calcified. Despite an intensive fishing effort, no sharks were caught using baited
lines.
Table 4.3.1: Details of the tagged sharks. Note that MN1 (male neonate), despite being of greater length than
FN1 (female neonate), is a whole kilogram lighter than his female counterpart.
Shark name: FN1 JM1 MN1

Tag number 1074582 1074580 1074581
Roto-tag number 151 122 120
Tag frequency 84 78 81
Time of capture 17-01-10 12:35 17-01-10 13:20 17-01-10 19:04
Catch Method Gill net Gill net Gill net
Catch Location (CL) Upper reach Upper reach Upper reach
Salinity at CL (PSU) 12.70 14.60 14.65
Life stage Neonate Juvenile Neonate
Sex Female Male Male
Pre-caudal length (cm) 57.00 84.00 59.00
Fork length (cm) 64.50 95.00 65.30
Total length (cm) 74.00 108.00 78.20
Weight (kg) 3.90 9.00 2.90
Clasper length NA 6.20 4.10
Of the three animals only JM1 proved possible to track properly in the following months due
to the following reasons: 1) the tag of the male neonate was found dislodged from the animal
upon the first tracking event; 2) in most instances the female neonate was found to move in
extremely shallow water in the Upper reaches and often within mangrove coverage making
the tracking effort near impossible from boat as performance of the hydrophone decreased
significantly in these conditions (See section 4.2.3). A total of six movement tracks based on
75
the juvenile male (JM1) were recorded throughout the study, and each track will be described
and analysed in turn, and finally followed by an assessment of the collective results.
4.3.1 TRACK I
The track on the 28th of January took place from 14:45 to 20:35 with the majority of the track
occurring on an incoming tide. As the tide was coming in, the shark spent roughly 3 hr
moving around in the same section of the middle reach, which can be characterized as a
relatively deep and wide section compared to the rest of the system. As dusk approached and
roughly an hour before the high tide peaked, the shark started moving upriver until it reached
the approximate border of the tide-influenced zone, where it lingered for about an hour.
When the tide started moving out, the shark returned to the same area it had been utilising
earlier. During the tracking period the shark was estimated to have travelled a total of 6.22
km, assuming unidirectional travel between recordings. Estimated swimming velocities were
relatively low, with the animal spending 80% of the track travelling with a speed of 0 to 0.5
m/s, while predominantly having a heading along with the current (>50%)
As evident from Figure 4.3.2, the animal spent all of its time in the 30-32 °C temperature
range, despite this regime only encompassing 35.3% of the total water volume, with the
remaining volume of the system being of colder temperatures. This translated into a
significant difference (P < 0.01) between observed and expected proportions of time spent.
Furthermore, the calculated value Manly’s α in relation to the 30-32 °C regime was 1.0,
indicating an affinity for high temperatures.
The animal spent > 50% of its time in the salinity regime of 9-18 PSU. Only 5.3% percent of
the total water volume was within this regime. In contrast, 82.0% of the total water volume
was within the 27-36 PSU range, but with no animal registrations within this zone. This
translated into a significant difference (P < 0.01) between observed and expected proportions
of time spent in the four defined salinity regimes (Figure 4.3.2). Furthermore the calculated
value of Manly’s α in relation to the 9-18 salinity regime was 0.73, indicating an affinity for
this salinity range.
Despite spending more than 50% of its time in current speeds in the 0.1-0.2 m/s regime,
which corresponded to 21.9% of the total water volume, there was no significant difference
between observed and expected proportions of time spent. Manly’s α of 0.8 did, however,
76
indicate a preference for this current speed regime. Nearly 90% of time spent was at locations
of 1-3 m water depth, corresponding to 52.2% of available habitat in terms of depth. This
translated into a significant difference in observed vs. expected proportion of time spent at a
given depth, and a Manly α value of 0.76 suggests that the animal had an affinity for this
depth range. This result is backed by the animal spending close to 60% of its time close to the
middle of the river channel (0.6-1.0 Riverbank ratio).
As mentioned in section 4.2.5, it was not possible to determine a value for Manly’s α and/or
any statistical difference in observed vs. expected proportion of time spent relative to
dissolved oxygen, pH and turbidity, due to the lack of modelled volume data for these
parameters. However, as shown in Figure 4.3.3, the animal spent all its time in areas with
oxygen levels >5 mg/l, and the majority of its time in turbidity of 0-2 NTU, while remaining
in the 7.5-7.75 pH range >90% of the time.
TRACK I: 28/01 – 2010 (14:45 to 20:35)
Figure 4.3.1: Recorded animal locations during the track undertaken on the 28/01-2011, plotted using an
MGA-56 coordinate projection. See main text for track details. An animated movement path is available for
illustration in the DVD appendix, filed as "Track_1_28012010" in the Track_Videos subfolder.
77
Figure 4.3.2: Proportion of time spent at various temperature (top left), salinity (top right), current speed
regimes (middle left), swimming speed (middle right), depth (bottom left) and riverbank ratio regimes (bottom
right) for the track conducted on the 28/01-2010.
78
Figure 4.3.3: Proportion of time spent at various dissolved oxygen (top left), pH (top right), turbidity (bottom
left) and animal heading regimes (bottom right) for the track conducted on the 28/01-2010.
4.3.2 TRACK II
The track on the 3rd of February took place from 9:20 to 15:11, with a turning tide occurring
at around 12:00. The shark lingered around the same area as in the first track, moving both
with and against the tide, with no apparent sign of being affected significantly by tidal
current. While not leaving the middle reach area (Figure 4.3.4), its level of activity was
relatively high, with an estimated distance of 5.56 km travelled during the six hours of
tracking (Figure 10). The apparent affinity for this general middle reach area suggests that
this particular animal has a specific habitat preference for this section of the system, and
might be a display of territorial behaviour. Estimated swimming velocities were elevated
compared to the previous track, with the animal spending nearly 50% of the track travelling
with a speed of 0.25-0.75 m/s, while displaying no clear preference of swimming direction
relative to the current (Figure 4.3.6).
79
As evident from Figure 4.3.5, the animal spent all of its time in the 28-30 °C temperature
range, despite that this regime only encompassed 14.0% of the total water volume. The
remaining volume of the system was of colder temperatures (84.9% being 26-28 °C). This
did not translate into a significant difference (P > 0.05) between observed and expected
proportions of time spent. However, the calculated value of Manly’s α in relation to the 28-30
°C regime was 1.0, which indicates an affinity for higher temperatures.
The animal spent all of its time in the salinity regime of 18-27 PSU, which corresponded to
only 8.2% percent of the total water volume. In contrast, 86.0% of the total water volume was
within the 27-36 PSU range, but with no animal registrations within this zone. This translated
into a significant difference (P < 0.01) between observed and expected proportions of time
spent (Figure 4.3.5). Furthermore the calculated value of Manly’s α in relation to the 18-27
salinity regime was 1.0, which indicates an affinity for this salinity range throughout this
particular track.
The animal was found to spend more than 80% of its time in current speeds in the 0-0.2 m/s
regime, which corresponded to the prevailing current speeds in 81.8% of the total water
volume, which led to no significant difference between observed and expected proportions of
time spent. Manly’s α at 0.69 did, however, indicate a weak preference for this current speed
regime, compared to 0.31 for higher flow regimes (0.2-0.5 m/s).
Well above 60% of time spent was at locations with 3-4 m water depth, corresponding to
21.7% of available habitat in terms of depth (only 11.8% of the total area). This translated to
a significant difference in observed vs. expected proportions of time spent at a given depth,
and a Manly α value of 0.66 suggests that the animal had an affinity for this depth range. This
result is further backed by the animal spending close to 60% of its time close to the middle of
the river channel (0.6-1.0 Riverbank ratio). As evident from Figure 4.3.6, the animal spent all
of its time in oxygen levels >5 mg/l, while remaining in turbidity between 0-2 NTU and in
the 7.5-7.75 pH range.
80
TRACK II: 03/02 – 2010 (9:20 to 15:11)
81
(middle left), swimming speed (middle right), depth (bottom left) and riverbank ratio regimes (bottom right) for
the track conducted on the 03/02-2010.
82
4.3.3 TRACK III

The track on the 27th of February took place from 13:45 to 19:34, with a turning tide at
around 14:20 and thus the majority of the track took place during an incoming tide. The
juvenile was again located in the same area as illustrated in Figure 10, and spent the first
three hours patrolling this area before dusk (~ 17:10), then started heading downriver against
an incoming tide. The shark evidently kept close to the shallow banks instead of the deeper
parts in the middle section of the river while travelling downriver (Figure 4.3.7). Sticking
close to the shallow mangrove-covered river bank might help to shelter the animal from other
sharks. Since flow usually decreases at shallow river banks compared to deeper parts of the
river, this behaviour could also be a means of minimising energetic costs when swimming
against the current. Within the same track, the shark travelled back with the tide to its
preferred area, and after lingering there for an additional hour, it went downriver once more,
again following a path close to the river bank. The estimated total distance travelled during
83
this track was 5.14 km. The estimated swimming velocities for this track were relatively low,
with the animal spending 60% of the track travelling with a speed of 0-0.25 m/s, while
spending equal amounts of time moving along and against the current.
Salinity levels in the middle reach of the system had dropped due to increased freshwater
discharge in the preceding weeks, which was reflected by the animal spending > 80% of its
time in the salinity of 0-9 PSU, corresponding to 12.4% percent of the total water volume
(compared to 0.6% and 2.4% for Track I and II, respectively). The observed behaviour
translated into a significant difference (P < 0.01) between observed and expected proportions
of time spent in the four defined salinity regimes (Figure 4.3.8). Furthermore, the calculated
value of Manly’s α in relation to the 0-9 salinity regime was 0.78, suggesting an affinity for
this salinity range during this track.
As evident from Figure 4.3.8, the animal spent > 50% of its time in the 28-30 °C temperature
regime, despite that this regime only encompassed 12.0% of the total volume, with the
remaining volume of the system being of colder temperatures. However, there was no
significant difference (P > 0.05) between observed and expected proportions of time spent.
The calculated value of Manly’s α in relation to the 28-30 °C regime was on the other hand
0.96, once again indicating an affinity for high temperatures.
Unlike previous tracks, the animal spent more than 40% of its time in current speeds in the
0.2-0.3 m/s regime, which corresponded to only 8.6% of the total water volume, but did not
prove to be significantly different in terms of observed and expected proportions of time
spent. Manly’s α of 0.56 did, however, indicate a slight preference for this current speed
regime. As shown in Figure 4.3.9, the animal spent all its time in oxygen levels >6 mg/l, and
in 0-2 NTU turbidity, while primarily staying in the 7.25-7.75 pH range (>70% of the time.)
Once again, the animal spent nearly 50% of its time at locations where the water depth was 2-
3 m, corresponding to 22.9% of available habitat in terms of depth. This translated into a
significant difference (P < 0.01) in observed vs. expected proportions of time spent at a given
depth, while a Manly α value of 0.48 suggests a slight affinity for this depth range compared
to others. Once again this result is backed by the animal spending almost 40% of its time
close to the middle of the river channel (0.8-1.0 Riverbank ratio).
84
TRACK III: 27/02 – 2010 (13:45 to 19:34)
85
86
4.3.4 TRACK IV
The track on the 6th of March took place from 12:03 to 14:58 with the tide turning to an
outgoing tide as the track commenced. The week prior to the 6th of March had heavy rains,
and a large runoff from the catchment and hinterland had caused a substantial change in the
salinity gradient of the system, while increased flow rates caused relatively high levels of
turbidity (up to 30.2 NTU in the middle reach). For the first time during the tracking
program, the juvenile shark was located outside the area where it was previously found
(Figure 4.3.10). The shark was initially located ~3.6 km from the river mouth, where
salinities were between 10-14 PSU. The shark continued to move further downriver on the
outgoing tide, lingering around in salinities of 26-28 PSU, before turning around and
swimming against the outgoing tide for ~ 1 km. However, the shark was later recorded
moving further downriver again, and was finally recorded very close to the marine
87
recreational boating area near the river mouth, where more than seven boats were engaged in
high-speed water sport activities at the time.
Unfortunately, due to equipment breakdown during the tracking event, it was not possible to
complete a full 6-hour track. The animal travelled a total distance of 3.33 km, while spending
>50% of the track travelling with a speed of 0.25-0.50 m/s, and predominantly having a
heading along with the current (>70%).
As evident from Figure 4.3.11, the animal spent 65% of its time in the 28-30 °C temperature
range, while this regime encompassed 39.4% of the total volume, with the remaining volume
of the system being of colder temperatures. This did not, however, translate into a significant
difference (P > 0.05) between observed and expected proportions of time spent. The
calculated value Manly’s α in relation to the 28-30 temperature regime was 0.74, indicating
an affinity for this temperature regime.
The animal spent 65% of its time in the salinity regime of 27-36 PSU, corresponding to
49.1% percent of the total water volume. In contrast, the animal spent 30% of its time in the
9-18 PSU range, which only corresponded to 2.7% of the total water volume. This translated
into a significant difference (P < 0.05) between the observed and expected proportions of
time spent in the four defined salinity regimes (Figure 4.3.11). Furthermore, the calculated
value of Manly’s α in relation to the 9-18 salinity regime was 0.89, indicating an affinity for
this salinity range.
Despite spending nearly 70% of its time in current speeds in the 0.1-0.2 m/s range, which
corresponded to 28.4% of the total water volume, there was no significant difference between
the observed and expected proportions of time spent. Manly’s α of 0.54 did, however, suggest
that the animal had a slight preference for this current speed regime during this track.
Forty-five percent of time spent was at locations with 0-1 m water depth, corresponding to
only 8.0% of the total water volume (however, 0-1 m deep areas encompass 28.4% of the
total area). This translated into a significant difference (P < 0.01) in the observed vs. expected
proportions of time spent at a given depth, and a Manly α of 0.74 further suggests that the
animal had an affinity for the 0-1 m depth range. This result is backed by the animal spending
close to 70% of its time in relative close proximity to the riverbank (0-0.6 Riverbank ratio).
88
Like in previous tracking events, the animal spent all its time in oxygen levels >5 mg/l. While
the animal spent the majority of its time in turbidity of 2-6 NTU, the turbidity levels
downriver were considerably less than those at the animal’s normal location in the middle
reach during this track (5.4 vs. 30.2 NTU). The animal spent most of its time in the 8.0-8.25
pH range, which is likely to be an effect of elevated pH levels throughout the system caused
by high increases in freshwater discharge.
TRACK IV: 06/03 – 2010 (12:03 to 14:58)
89
90
left) and animal heading (bottom right) for the track conducted on the 06/02-2010.
4.3.5 TRACK V
The track on the 20th of March took place from 9:30 to 16:00, with most of the tracking
occurring during an outgoing tide. The juvenile shark was again registered at the same middle
reach area as previous tracks. Measured turbidity levels were still elevated compared to the
normal range of 0-2 NTU and ranged from 2-9 NTU, while salinity in the immediate area
ranged from 0-9 PSU during the tracking period. With the change to an outgoing tide at
approximately 10:40, the shark started moving downriver while sticking close to the shallow
banks, as had been observed in previous tracks. The salinity in the downstream section
ranged from 6-20 ppt. After travelling an estimated 1.2 km downriver, the signal to the
animal was lost and not regained before an hour later. At this point it had moved back to the
same area as it had been in the first place, where it remained for the remainder of the track.
The total distance travelled by the juvenile shark during the track was estimated to be 7.73
km.
91
The animal spent >50% of the track travelling with a speed of 0.25-0.50 m/s, and
predominantly moving perpendicular to the current (nearly 60%). Once again the animal
spent all its time in oxygen levels >5 mg/l, and spent the majority of its time in turbidity
between 4-8 NTU, while remaining in the 7.0-7.5 pH range >80% of the time.
As evident from Figure 4.3.14, the animal spent >50% of its time in the 22-24 °C temperature
range, despite that this regime only encompassed 15.6% of the total volume, with the
remaining volume of the system being of warmer temperatures. While this observed
behaviour translated into a Manly's α value of 0.82 in relation to the 22-24 temperature
regime, there was no significant difference between the observed and expected proportions of
time spent in the five defined temperature regimes.
The animal spent all of its time in the salinity regime of 0-9 PSU, corresponding to only
16.0% percent of the total water volume being within this regime. This translated into a
significant difference (P < 0.01) between the observed and expected proportions of time spent
in the four defined salinity regimes (Figure 4.3.14). Furthermore, the calculated value of
Manly’s α in relation to the 0-9 salinity regime was 1, which suggests the animal had a strong
preference for this salinity range during this track.
The animal spent more than 90% of its time in current speeds in the 0-0.2 m/s regimes, which
also corresponded to 90.7 % of the total water volume. There was no significant difference
between the observed and expected proportions of time spent. Manly’s α of 0.8 did, however,
indicate a preference for this current speed regime, compared to higher flow velocities.
While there was a significant difference (P < 0.1) in the observed vs. expected proportions of
time spent at a given depth regime, Manly’s α for each regime did not exceed 0.31, which
suggests that the animal had no particular affinity for any depth range. Similarly, the
calculated riverbank ratios for this track did not suggest that the animal had any preference
for particular positions relative to the middle of the river.
92
TRACK V: 20/03 – 2010 (9:30 to 16:00)
93
94
4.3.6 TRACK VI
This track took place on the night between the 20th and 21st March during the timeframe of
21:13 to 03:00. The tide peaked and shifted to an outgoing tide at ~ 23.30. The animal was
located in the area downstream of the middle reach area where it had been found on four
other occasions. The shark stayed in this general area for almost three hours, but as the tide
turned the shark moved into the adjoining upstream area where it normally lingered, and
stayed there for two hours before moving to the same downstream section as before. At the
end of the six-hour track the animal was estimated to have covered a total distance of 7.16
km. Estimated swimming velocities were almost identical to the previous track (although
slightly decreased), with the animal spending 50% of the time travelling with a speed of 0.25-
0.50 m/s. The animal showed no clear preference for swimming along or against the current,
but predominantly swam perpendicular to the current (Figure 4.3.18). The animal spent all its
time in oxygen levels >7 mg/l, and spent the majority of its time (> 70%) in turbidity between
2-4 NTU, while remaining in the 7.25-7.75 pH range >80% of the time.
95
Not surprisingly, the animal spent approximately the same amount of time (>60%) in the 22-
24 °C temperature range, as it did in the track earlier the same day. Due to the fact that the
volume of this temperature regime had decreased to 12.8%, this translated into an increased
Manly’s α value of 0.90 in relation to the 22-24 temperature regime (0.82 in the previous
track). As in the previous tracks, there was no significant difference between the observed
and expected proportions of time spent in the five defined temperature regimes.
The animal spent 85% of its time in the salinity regime of 0-9 PSU, corresponding to 11.5%
percent of the total water volume. This translated into a significant difference (P < 0.01)
between the observed and expected proportions of time spent (Figure 4.3.17). However, the
calculated value of Manly’s α in relation to the 0-9 salinity regime was only 0.52. There was
a significant difference (P < 0.01) between the observed and expected proportions of time
spent in various current speed regimes; however, the values of Manly’s α did not suggest any
strong affinity for a particular current speed regime. Similarly, for the depth regimes there
was no particularly affinity for a specific regime, and there was no significant difference
between the observed and expected proportions of time spent at a given depth.
96
TRACK VI: 21/03 – 2010 (21:13 to 03:00)
97
98
4.3.7 COLLECTIVE RESULTS

As mentioned in section 4.2.5, it was not possible to calculate reliable estimates of the
animal’s home range using the local nearest-neighbour convex-hull construction of home
ranges as proposed by Getz et. al (2007). However, from above mentioned results, and further
emphasised by Figure 4.3.19, the animal seemed to have a clear affinity for the middle reach
of the system, with ~90% of all registrations made within this area, while occasionally
utilising both up- and downriver sections of the system. At no point was the animal registered
beyond the tide-dominated zone, or in the adjoining artificial canals.
Throughout the six recorded tracks, the animal's estimated rate of movement (ROM) was
quite low, with nearly 80% of the time spent swimming between 0.0-0.5 m/s, with an
average swimming speed of 0.32 ± 0.25 m/s (S.D., n=188). As evident from the collective
results presented in Figure 4.3.20 and 4.3.21, the animal spent the majority of its time in
current speeds of 0-0.2 m/s, while spending near-equal amounts of time swimming along and
99
against the current. However, when current speeds exceeded the mean ambient current speed
(Figure 4.3.22), the animal decreased its frequency of movement perpendicular to the current,
whereas in low current speeds the animal spent more time moving perpendicular to the
current. This suggests a directional response of the animal in respect to flow velocities,
possibly as a means to stay in the same general location under changing flow conditions.
ROM was significantly negatively correlated with current speed (P < 0.05); however, the low
correlation coefficient of -0.16 suggests only a weak relationship.
The collective results for depth (Figure 4.3.20) suggest that the animal had an overall affinity
for intermediate to deep depths within the 1-4 m interval, while only spending modest
amounts of time (< 10%) in shallow water. It is likely that the lack of time spent at locations
that exceed a depth 4 m is due to the fact that the majority of these areas are located near the
river mouth and adjoining artificial canals, rather than an actual avoidance of deep holes.
Taking the collective riverbank ratios into account (Figure 4.3.20), the evidence of the animal
having a higher affinity for the middle, and thus deeper, sections of the system is further
supported. ROM was negatively correlated with both depth (P<0.01) and the riverbank ratio
(P<0.01) with correlation coefficients of -0.29 and -0.41, respectively. While the correlation
coefficients are relatively low, this result still suggests that the animal lingers in the deeper
parts of the system by decreasing its movement speed and generally moving faster closer to
the riverbanks, which in turn could indicate that the animal forages for prey close to the
mangrove sheltered riverbanks.
Throughout the six tracks, the animal spent the majority of its time (>50%) in the 0-9 PSU
regime, while spending only 6% of its time in the 27-36 PSU regime, despite the latter being
the salinity regime that encompassed the largest volume of habitat in the system on average
(~55%). The lack of a significant correlation between ROM and salinity (P > 0.05) further
suggests that the animal had an affinity for a broad range of salinities and that it did not
utilise a change in rate of movement as a strategy to avoid areas with too high or too low
salinity. However, a directional response to change in salinity remains a possibility, and
serves as a possible explanation for the animal being found downriver on the 6th of March
(Section 4.3.4).
The collective results for temperature reveals an almost uniform distribution of proportion of
time spent at the various temperature regimes (Figure 4.3.20), which is likely the result of the
observed system-wide temperature fluctuations in between tracks, thus clouding any potential
100
short-term temperature affinity. ROM was negatively correlated with temperature (P < 0.05,
R= -0.18), and while the correlation coefficient is quite low, this could indicate that the
animal displayed a slight behavioural response to temperature by decreasing its swimming
velocity in higher temperatures (>30 °C).
The animal spent the majority of its time (57%) in turbidity levels of 0-2 NTU, with
decreasing amount of time spent with increasing turbidity level, suggesting an overall affinity
for low turbidity levels (Figure 4.3.21). With almost 50% of time spent in pH levels less than
7.5-7.75, the overall proportion of time spent in the various pH regimes reflected the animal’s
primary choice of habitat in the middle reach mixing zone between freshwater discharge and
tidal movements. This indicates that pH could be a factor in habitat selection, among other
abiotic factors. While dissolved oxygen levels were relatively high throughout all tracks (> 5
mg O2/l), the animal spent the most time in the 7-8 mg O2/l regime (Figure 4.3.21), indicating
an affinity for higher oxygen levels. There was no significant correlation between ROM and
turbidity, pH or dissolved oxygen, which could be due to the lack of any strong spatial
gradients of these parameters on each given tracking event. However, the scarcity of these
data sets could also likely be a factor in the calculated correlation; e.g. at no point was the
animal registered in conditions below 4 mg O2/l, above 10 NTU or in potential harmful pH
levels, making it very difficult to relate each parameter’s effect on ROM.
101
ALL TRACKS: 28/01/2010 to 21/03/ 2010
Figure 4.3.19: Total amount of recorded animal locations during all six track undertaken from the 28/01-2010
to the 21/03-2011, plotted using an MGA-56 coordinate projection. Note the concentration of registrations in
the middle reach. See main text for explanatory details.
102
Figure 4.3.20: Total proportion of time spent at various temperature (top left), salinity (top right), current
speed (middle left), swimming speed (middle right), depth (bottom left) and riverbank ratio regimes (bottom
right) for the track conducted from the 28/01/2010 to the 21/03/2010.
103
left) and animal heading (bottom right) for the track conducted on the 20/03-2010.
Figure 4.3.22: Frequency of animal heading relative to current direction during below-mean flow velocities
(red), and above-mean flow velocities (blue).
104
4.4 DISCUSSION
A significant limitation of this study is the fact that only a single individual was studied, and
thus it lacks the ability to confidently generalise the observed movement behaviour for C.
leucas. While a larger number of tracked individuals could have increased the general
validity of the observed movement patterns, and allowed an assessment of intraspecific
variation in terms of habitat choices and utilisation, this study still provides interesting insight
into the potential variability of an individual’s behaviour in relation to temporal changes in
water quality regimes. It is nevertheless important to emphasise that this study does not
attempt to validate the results’ general applicability, and that any conclusions based on the
results presented in section 4.3 are valid for the tracked individual only.
In contrast to the reported preferred salinity range of 7-20 PSU for juvenile bull sharks in the
Caloosahatchee River (Florida, USA), and avoidance of salinity ranges < 7 and > 20 PSU
(Simpfendorfer et al. 2005) (Ortega et al. 2009), the tracked juvenile in this study was found
to prefer a wider salinity range of 0-27 PSU. Considering the large amount of time (> 50%)
spent in the 0-9 PSU regime when more saline conditions were available downriver, the
range of tolerable salinity levels is wider for this juvenile, with no evidence of immediate
avoidance of lower salinities. However, the lack of registrations above 28 PSU relative to the
large volume of water within this range still suggests an avoidance of high salinities. While
purely speculative, it could be hypothesised that due to Tallebudgera Creek being more
sensitive to freshwater discharges compared to larger estuaries, the high frequency of drastic
shifts in salinity in this system could require resident young C. leucas to have a wide salinity
tolerance range. This early life-stage adaptation could in turn aid young C. leucas to better
utilise the full extent of the space-limited estuary, rather than congregate in the same limited
salinity range with larger conspecifics. As cannibalism has been reported within this species
(Snelson et al. 1984), it is possible that avoidance of especially larger conspecifics is a strong
driver for movement than salinity in size-limited habitats.
As evident from Figure 4.3.19, the tracked animal had a high affinity for the middle reach
area, despite salinities in this area being highly variable over time, and at times was < 1 PSU.
Only on two occasions did the shark leave this area, with one being a rather quick upriver
migration on an incoming tide, before returning to its usual home range with the turning of
the tide. This upriver excursion might be a foraging trip for prey, while utilising the tide to
105
minimize energetic costs associated with this type of directed movement. As mentioned in
section 4.3.4, the other occasion when the animal was found outside of its usual home-range
was shortly after a freshwater discharge event had shifted the system into a state of elevated
flow, higher turbidity and decreased salinity. Since this was the only track conducted in the
time immediately after a system-wide shift in water quality, it is difficult to assess if the
unusual downriver position of the animal was the result of avoidance of unfavourable
conditions, e.g. elevated turbidity levels, or simply being driven downriver by elevated flow
instead of spending energy to maintain position at the middle reach. Likewise, it is difficult to
assess the contribution of decreased salinity levels to the downriver migration. However, two
weeks later when flow velocities and turbidity levels had decreased to pre-discharge levels in
the middle reach, but salinity levels remained low at 0-3 PSU, the animal spent the following
12 hours at its usual middle reach location. This suggests that while salinity might contribute
to movement and habitat selection in a more passive manner, elevated flow and/or turbidity
levels could possibly have a higher impact on movement associated with avoidance strategies
in the event of a system-wide shift.
As evident from the collective results in section 4.3.7, the animal did not show any clear
preference in terms of swimming direction relative to the current, although it did spend a
slightly larger amount of time travelling with, rather than against, the current. When taking
the affinity for a specific home-range into account, the near-equal amount of time spent
travelling with and against the current suggests that the animal predominately attempts to
maintain its position in the middle reach. This may be due to the relatively low flow
velocities throughout the middle reach area. This behaviour likely shifts during elevated flow
regimes, as might have been the case on the track conducted the 6th of March (Section 4.3.4).
The apparent site-affinity could partly be due to the fact that this section is relatively deeper
and wider compared to the surrounding sections. Werry (2010) reported an association of C.
leucas with deep holes in the Gold Coast canals, and suggested the animal might use deep
holes as a means to maintain their position against the current and conserve energy.
106
Figure 4.4.1: Mean current speeds as predicted by the HD model, during the tracking campaign period,
spanning from 28/01/2010 to 21/03/2010. Note the patches of reduced flow velocities within the middle reach
area (marked by red arrows).
As evident from Figure 4.4.1, the physical characteristics of the middle reach do indeed
provide patches of reduced mean flow velocities compared to the adjoining areas. This
suggests that part of the animal’s site affinity could be linked to the presence of a more
heterogeneous bathymetry that provides patches of preferred reduced flow conditions.
While salinity did not seem to be a strong short-term driver for directed movement, it is
possible that it is a major driver for habitat selection on a larger temporal scale. When
considering the range of mean predicted salinity levels over the entire tracking campaign
(28/01 to 21/03 - 2010) throughout the system (Figure 4.4.2), it is evident that the mean
salinity levels (7-14 PSU) for the middle reach area lies within the preferred salinity range
reported by Simpfendorfer et al. (2005) and Ortega et al. (2009).
107
Figure 4.4.2: Mean salinity as predicted by the HD model, during the tracking campaign period, spanning from
28/01/2010 to 21/03/2010. Note how the area normally utilised by the animal (marked by the black box) lies
within the 7-14 PSU range.
It may be hypothesised that the reason why the animal stayed within the middle reach when
salinity levels were below 7 PSU is that the energetic costs associated with moving downriver
in search for more saline habitats outweighs the costs of maintaining its current position until
preferred salinity levels return with the incoming tide. However, when other variables comes
into play, such as increased turbidity and flow in conjunction with a drop in salinity, the
energetic costs of staying put in the same location are likely to overweigh the costs of
downriver migration, as was witnessed in Track IV.
In terms of predicted mean temperatures for the middle reach during the period of 28/01 to
21/03-2010, there was only a 0.3°C difference in predicted mean temperatures between the
middle and uppermost reaches, while the area at the river mouth was ~1.0°C lower than at the
middle reach. Due to the lack of any notable spatial variation throughout the system, the
impact of temperature on movement and habitat selection is not easily assessed. It seems
unlikely that temperature plays a major role in habitat selection on a short-term scale,
although it is difficult to rule out entirely as a long-term driver, as ROM was negatively
correlated with temperature.
As mentioned in section 4.3.7, the lack of spatio-temporal data for dissolved oxygen, pH and
turbidity makes it very difficult to assess how these parameters affected movement. While
108
there is evidence that suggests turbidity has been a driver in the downriver migration
witnessed on the 6th of March, the contribution of DO and pH remains elusive. Due to DO
levels being generally high throughout the system over the course of 2010 (Chapter 2,
Section 2.3), it is plausible that DO have a limited effect on movement as long as levels
remain high. However, as the bull shark is a ram-obligate shark species (Weihs et al. 1981), it
is plausible that hypoxic conditions could result in an increase of swimming velocities, since
this behavioural response has been reported for other ram-obligate shark species under
hypoxic conditions (Parsons & Carlson 1998, Carlson & Parsons 2001).
Although a lack of proper data resolution in respect to the abovementioned parameters

restricts the ROM analysis, it is important to note that the bias associated with the calculation
of ROM will undoubtedly lead to a biased result in the correlation analysis. It follows
logically that movement with a high level of linearity in open unobstructed areas will have a
relative small bias associated with the calculation of ROM using the Euclidian distance
between points as a proxy for true distance travelled. However, more tortuous movement in
confined areas (such as Tallebudgera Creek) will in turn potentially grossly underestimate the
true ROM unless registration of animal position is made in quick succession. As mentioned
in section 4.2, this study utilised a variable 5-10 minute interval between registrations.
However, on multiple occasions while the animal was in its middle reach home range, it was
noted that it would swim in one direction before turning and returning down the same path
within a 5-10 minute interval. While effort was made to capture this back and forth
behaviour, it was near impossible to predict when or if the animal would backtrack at a given
point in time, and thus a great deal of detail could have been lost, which in turn would result
in an underestimation of ROM.
While the obvious limitations in the data set warrant further investigations into the movement
of C. leucas in Tallebudgera Creek, one has to bear in mind that the main objective of the
tracking campaign - the acquisition of movement data for validation of an agent-based model
(ABM) - was fulfilled to a satisfactory degree. Logically, a more detailed data set
encompassing repetitive tracks of several animals over a range of different water quality
conditions would improve confidence in the observed pattern on a more general level.
However, for the purpose of testing the ABM approach as a means to study and replicate the
movement patterns of C. leucas, a large and comprehensive sample size is not necessarily
mandatory.
109
CHAPTER V
AGENT-BASED MODELLING OF A JUVENILE BULL SHARK ON A
SHORT SPATIO-TEMPORAL SCALE
110
5.1 INTRODUCTION
As an apex predator potentially up to 3.5 m in total length (Voight & Dietmar 2011), the bull
shark fulfils an important ecological role in the top-down regulation of food chains in its
natural environment (Martin 2005, Heithaus et al. 2008). Due to the bull shark’s ability to
enter estuarine and freshwater systems, this species has always been living in close proximity
to human settlements. It is a potential dangerous species to humans and has been implicated
in many reported shark attacks (Werry 2010). As human populations continue to grow in
coastal areas, so does the pressure on habitats frequented by C. leucas, and there is an urgent
need for effective management of this keystone species (Werry 2010).
Many mobile aquatic species move extensively as they use different habitats for different
purposes, e.g. foraging, spawning and refuge. These species may thus be affected if any of
these habitats are degraded as a result of anthropogenic impact (Goodwin et al. 2001). While
habitat degradation in terms of water quality and hydrology can be modelled on a fine spatio-
temporal scale through the implementation of Eulerian models (Szylkarski et al. 2004, Dias
& Lopes 2006b, Zacharias & Gianni 2008), the resulting impact on fish movement and
behaviour remains elusive.
Movement of terrestrial and aquatic organisms has often been modelled by biologists using a
Lagrangian object-oriented model framework, which unlike the Eulerian framework, allows
for the individual representation of organisms (Nestler et al. 2005). However, many of these
early agent-based models do not account for spatial variations in habitat and fish movement
(Railsback et al. 1999). The inability of the Eulerian framework to simulate individual
entities and the lack of a spatial heterogeneous habitat representation in Lagrangian models
can, however, be overcome by coupling the two model frameworks.
The coupled Eulerian-Lagrangian framework allows for an accurate representation of

hydrology and water quality within a spatially complex system over time, while also capable
of simulating higher trophic levels such as fish on an individual level, which moves
independently of the Eulerian model grid (Nestler et al. 2005). As simulated agents within the
model domain can be made capable of reacting to Eulerian gradients, it is thus possible to
investigate potential effects of hydraulic cues on fish movement on a complex spatial scale
over time (Nestler et al. 2005).
111
From an ontological point of view, one can never truly know how a wide range of complex
environmental stimuli is perceived neurologically and translated into a movement response
by a fish. Thus, simplifying assumptions on how the fish perceives the outside environment is
a necessity in order to simulate possible movement responses to environmental cues without
first describing the complex nature of the neurological translation of perceived stimuli.
Although many agent-based models are relatively complex in nature (Jopp et al. 2011), many
of these models limit the number of possible environmental stimuli to only a few parameters
that are hypothesised to be the main drivers of movement, e.g. Humston et al. (2000)
simulated the migration of bluefin tuna solely through the use of temperature stimuli, while
Goodwin et al. (2006) simulated the downstream migration of juvenile salmon using flow
velocities, dissolved oxygen and temperature as primary drivers for movement.
While long-term agent-based models are required to simulate some measure of food-uptake
and growth in order to remain biological plausible, e.g. Humston et al. (2004), it is a
reasonable assumption that short-term movement are not necessarily driven by foraging, and
that growth of the individual is negligible given the short time-frame. Juvenile C. leucas have
been reported to move and distribute in relation to hydrology and physical cues by several
studies (Simpfendorfer et al. 2005, Heupel & Simpfendorfer 2008, Ortega et al. 2009, Werry
2010), which has also been backed by the findings of this study (Chapter IV). It is therefore
possible that short-term movement and distribution of this species can be modelled to a
plausible degree as has been demonstrated in some teleosts, e.g.: (Van Winkle et al. 1998,
Railsback et al. 1999, Humston et al. 2000, Goodwin et al. 2001, Humston et al. 2004,
Goodwin et al. 2006).

As the final component of this study, this chapter sought to develop an Eulerian-Lagrangian
agent-based model capable of replicating and predicting the short-term movement of juvenile
C. leucas using hydraulic and physical cues simulated by the hydrodynamic model as drivers
for movement. It was predicted that the outcome of this agent-based modelling approach
would help to better understand the drivers of C. leucas movement and with time, provide the
first brick of a more comprehensive ABM that accounts for the ontogenetic habitat use by C.
leucas in the estuarine-marine continuum.
112
5.2. METHODOLOGY
The MIKE ECO Lab integrated agent-based modelling module was utilised for developing an
individual-based model relating to the virtual habitat defined by the hydrodynamic model
described in Chapter III. The MIKE ECO Lab ABM module is still in a developmental stage
and not part of the commercially available MIKE software package provided by the DHI.
However, it has been tested reliable though application of in-house DHI projects, e.g.
(Andersen et al. in prep, Canal-Vergés et al. in prep, Hansen & Potthoff in prep). In the
following section a short description of the fundamental properties of the MIKE ABM
module will be given, before presenting each movement model in turn.
5.2.1 THE COUPLED EULERIAN-LAGRANGIAN MODEL FRAMEWORK

The MIKE ABM-module consists of a Lagrangian framework integrated into the Eulerian
framework of the hydrodynamic model (Chapter III) that simulates the physical dynamics of
the environment of the agents. The integrated Eulerian-Lagrangian framework makes it
possible for a simulated agent (or individual) to move independently of the Eulerian model
grid and potentially affect Eulerian variables such as dissolved oxygen or
phytoplankton/zooplankton concentrations through consumption, if such variables were
simulated in the Eulerian framework. In addition to manipulation of the Eulerian
environment, each agent can through user-specified equations react to other nearby agents to
simulate specific types of interaction between individuals, such as flocking/avoidance-
behaviour, predation, and mating.
Through the use of model built-in spatial functions, it is possible to define a sensory sphere
that can potential stretch across Eulerian grid cells (Figure 5.2.1). The implementation of a
sensory sphere enables the agent to detect the gradient Eulerian variables and/or the presence
of other Lagrangian agents within the radius of its sensory sphere. The size of the sphere can
be defined through a user-specified radius together with the angle of the agent’s field of view,
meaning that if needed an agent can be specified to only sense variables ahead of its direction
of orientation.
113
Figure 5.2.1: Two-dimensional view (longitudinal and latitudinal) of a shark’s sensory range. The sensory
range is independent of the Eulerian model grid, and is often estimated from the body length of the animal and
model timestep. Adapted from Goodwin et al. (2006)
Movement of agents in the horizontal plane in the model setup are expressed by assigning
values (constant or variable over time) as a horizontal vector, consisting of a speed (m/s) and
a direction (degrees). Movement in the vertical plane is defined through a positive
(swimming upwards) or negative (swimming downwards) speed, although in this study the
agents are limited to moving in the horizontal plane with a constant Z-value. In the current
version of the ABM module, it is possible to define a maximum of five movement vectors,
where the final movement is found by internally combining all vector sub-sections for each
timestep. The implementation of two or more movement vectors allows the agents to have
their final movement vector affected by several factors, e.g. the flow direction and velocity of
the water body. The length of the timestep determines the time spent travelling with a given
speed and direction before a new trajectory is calculated for each agent. The size of the
timestep is therefore an essential parameter for the performance of the movement models
tested, and will be discussed in more detail later in this chapter.
5.2.2 THE RANDOM WALK MODEL

Although it is unlikely that animal movement is truly random in nature, this assumption is
often made when modelling animal movement in the biological sciences (Codling et al.
2008), as it provides a reasonable ‘null’ model for hypothesis testing . The Random walk
114
model represents the simplest of assumptions by assuming that the animal shows no
behavioural sensitivity to stimuli from its external environment, and thus provides a suitable
control for comparison against movement models of increased sophistication (Humston et al.
2004), such as the kinesis model that is described in a later section (5.2.3).
5.2.2.1 RANDOM WALK MOVEMENT EQUATIONS

Direction (degrees) in the random walk model is defined as 360 multiplied with a random
number drawn from a uniform distribution ranging from 0 to 1.
Eq. 5.1
Equation 5.1 ensures that the frequency of chosen swimming directions will be uniformly
distributed around the compass rose, and thus no directional preference will be evident in the
agent’s movement.
Swimming velocity is defined as:
Eq. 5.2
Where and are the maximum and minimum swimming speeds respectively, and
is a random number drawn from a uniform distribution ranging from 0 to 1. The above
equation allows the agent to swim at a uniformly distributed range of swimming velocities
varying from to . The term is included in the formulation of since bull
sharks are obligate ram-ventilators and are required to swim continuously in order to
maintain adequate water flow over its gills. represents this basal level of swimming
activity. See section 5.6.0 for a further discussion of and .
In the random walk model, simulated agents will not be affected by any external cues besides
water flow velocity. No other movement rules apart from mechanical forcings that prevent
the agents from crashing into model boundaries will be implemented. See section 5.2.5 for a
description of these movement forcings.
5.2.3 THE KINESIS MODEL

The following model description builds upon the one-variable kinesis model first proposed by
Humston et al. (2000) as a way for blue-fin tuna to locate habitat that provided optimal
conditions in terms of temperature. Later this model was developed further to include two
variables; prey abundance and optimal salinity conditions for bonefish Albula vulpes
115
(Humston et al. 2004). For the purpose of this study, two types of kinesis search models will
be tested: 1) A one-variable kinesis search for optimal salinity and 2) a three-variable kinesis
search where it will be assumed that the animal responds to salinity, temperature and current
velocities, while naturally also being passively affected by flow velocities as would any
object in the water column.
Kinesis describes an agent’s reactive, yet non-directional, response to external or internal

stimuli relative to a preferred state, where the intensity of stimulation compared to the
preferred state will affect the change of speed (orthokinesis) and/or the frequency of
directional changes (klinokinesis) (Humston et al. 2000).
The mechanisms behind kinesis correspond to a minimalist assumption of cognitive

capabilities in the animal, where no spatial or temporal evaluation of sensory inputs is taken
into account: the agent solely reacts on the concurrent stimuli in its point of presence with no
"knowledge" of the surrounding environment, and thus not able to move towards a specific
direction based on external cues (Humston et al. 2004).
5.2.3.1 KINESIS MOVEMENT EQUATIONS

The agent’s swimming velocity components, and , are respectively defined as:
Eq. 5.3
Eq. 5.4
Where is a function describing the deterministic response to an external stimulus based on

the swimming velocities of the previous timestep, while is a stochastic function (Eq.
5.8) that determines the random component of the agent’s reactive response to a stimulus.
and are the flow velocities predicted by the hydrodynamic model at the agent’s location.
The equation of is identical to that of except is substituted for ,
and is defined as follows:
Eq. 5.5
is the swimming velocity in the previous timestep , and describes the influence of
inertia on movement, meaning that an agent moving at higher speeds is more likely to
continue in the same general direction in the next timestep. , and are
Gaussian equations describing the reaction to salinity, current speed and temperature in the
116
current timestep relative to a theoretical optimum. , and are, respectively, variable
weighting factors for salinity, current speed and temperature, which determine the relative
contribution of , and and is dependent on the difference between the
ambient and preferred levels of salinity, current speed and temperature. The functions ,
and are defined as follows:
Eq. 5.6
Eq. 5.7
Eq. 5.8
Where , C and are the given temperature, current speed and salinity, respectively, in the
present timestep, while , and denotes the respective optimal/preferred value by the
agent. , , and are shape parameters determining the Gaussian variance and
height, which in turn dictate the sensitivity of the kinesis response to external stimuli
(Humston et al. 2004). The stochastic component, , in Eq. 5.3 and 5.4, is defined as:
Eq. 5.9
Similar to , the stochastic component consists of three Gaussian equations,

, and , that describe the respective reactions to salinity, temperature and
current speed in a given timestep relative to a theoretical optimum, and weighted by ,
and , respectively. The stochastic variable in this study is defined as a random number,
, drawn from the standard normal distribution with a mean of 0 and variance of
(resulting in 99.7% of all numbers drawn being in between ±1), and subsequently scaled
according to to prevent any agent reaching swimming velocities that exceed the
specified maximum:
Eq. 5.10
The Gaussian equations , and appearing in Eq. 5.9 are defined as:
117
Eq. 5.11
Eq. 5.12
Eq. 5.13
In the above equations, determines the height of Gaussian curve while the variance
parameters , and are identical to the ones specified in Eq. 5.5, 5.6 and 5.7. Finally the
weighting parameters used in Eq. 5.5 and 5.9 are defined as:
Eq. 5.14
Eq. 5.15
Eq. 5.16
The value of , and ranges from 1 to , depending on the difference experienced

between the current external stimulus and preferred optimum of salinity, current speed and
temperature, respectively. The value of determines the range of possible weightings, and is
assigned the value of 5 following Humston et al. (2004), due to the inability of
parameterisation of from biological data.
It is evident from Eq. 5.11, 5.12 and 5.13 that (the variable X is used for the purpose
of illustration) will move towards its global minimum defined as which is reached
only when , meaning that the stochastic contribution to movement is minimal when
the agent is located in an optimal environment (Figure 5.2.).
118
Figure 5.2.2: Plot A: and plotted against the variable X with optimum , and
and . Plot B: and plotted against the variable X with optimum ,
and and .
(Eq. 5.6, 5.7 and 5.8), on the other hand, has its maximum value, , when ,
which will cause an increased contribution of inertia from and results in a decrease of
the likelihood of the agent turning to another direction when close to its optimal state. When
the agent is moving away from its optimal state, the relative contribution of will
decrease while increases (Figure 5.2.2), thus resulting in a more random movement.
This functional response is intended to mimic the behaviour of an animal in search for
optimal conditions without spatial "knowledge" of its surroundings, and only able to sense
the environmental conditions in its own point of presence. As Plots A and B in Figure 5.2.2
illustrate, the shape of the Gaussian functions is highly dependent on the specified optimum
as well as the shape parameters , and . Furthermore, Figure 5.2.2 illustrates that if
(Plot A) it will result in decreased swimming velocities when located close to the
optimal state, whereas when (Plot B) a relatively higher swimming velocity will be
maintained in optimal conditions.
Similarly, the functional response of and when reacting to more than one variable (X
and Y for illustration), results in a three-dimensional matrix (Figure 5.2.3). As evident from
Figure 5.2, the value of peaks when the optimum values for both X and Y are met
( , in this demonstration), while the stochastic contribution of is
119
reduced to its minimum value, and vice versa for values located away from the optimum
values.
Figure 5.2.3: The weighted functional response of (left) and (middle) plotted separately and
combined (right). The optimum values depicted ( , and ) are for illustration
purposes only.
5.2.4 MODEL STRUCTURE

The random walk model (RW), the kinesis search for optimal salinity model (KS) and kinesis
search for optimal salinity, temperature and current speed (KSTC) were applied to each of the
six registered animal tracking events (Chapter 4) with 100 agents released at the registered
start location of each track. The simulation period covered the full duration of the recorded
tracks in order to perform a direct comparison between observed and simulated movement
tracks.
In order to test the predictive abilities of the above-mentioned movement models, 100 agents
were randomly distributed within the model domain and had their movement simulated for 12
hours prior to the timestamp of the initial animal location for each track. The start location
versus end location of simulated agents allows for the assessment of the agents ability to react
to the prevailing environmental conditions throughout the system and locate the habitat where
the tracked animal was registered upon the start of a tracking event.
Table 5.1 provides a list of the final values of each calibration parameter in the three
movement models, and their values explained in turn in the following section.
120
5.2.5 MODEL SETUP & CALIBRATION EFFORT
The timestep of the ABM model setups were initially set to 30 seconds, but due to relative
high agent velocities that caused agents to crash into land boundaries in a single timestep, the
value was reduced to 10 seconds to reduce the amount of time travelled before making a new
movement decision, but at the cost of increased simulation times. was originally set to a
value of 3.0 m/s, which is well above the reported cruising speed of 0.56-0.80 m/s for C.
leucas (Weihs et al. 1981), while was set to 0.1 m/s. Initial model runs revealed that the
average speeds of agents were around 0.6-0.7 m/s using these values; however, due to the
very narrow nature of the model domain, the rare event of agents reaching the maximum
velocity often resulted in agents crashing into model boundaries. As a necessary means to
avoid the loss of agents, was consequently scaled down to 1.5 m/s.
Table 5.2.1: The model calibration parameters and their final values adopted in the three types of movement
models.
Parameter RW KS KSTC Unit Reference

Timestep 10 10 10 seconds -
1.5 1.5 1.5 m/s -
0.1 0.1 0.1 m/s -
- 13.5 13.5 ppt -
- - 20 °C -
- - 0.1 m/s -
- 20 20 - -
- - 10 - -
- - 0.5 - -
- - 5 - -
- - 0.7 - Humston et al. (2000)
- - 0.9 - Humston et al. (2000)
While exact optimum values for salinity, temperature and flow velocities have not been
established in the literature for juvenile C. leucas, the optimum value of salinity , was
selected in order to reflect the range of salinities in which the tracked individual was recorded
in 0-27 ppt as well as the salinity range of 7-20 recorded by the literature (Simpfendorfer et
al. 2005, Heupel & Simpfendorfer 2008, Ortega et al. 2009). The selected temperature
optimum of 20 °C was selected as a means to have the agents look for the “coldest” possible
water in the prevailing temperature conditions in a given simulation. The “cold-water”
121
optimum was chosen to reflect one of the model assumptions that the agents are not looking
for prey but habitat (in terms of water quality) where metabolic costs is minimum, e.g. it has
been reported that the euryhaline bat ray, Myliobatis californica, displays behavioural
thermoregulation by feeding in warmer waters and resting in cooler waters (Matern et al.
2000). The optimum value for current flow was set to 0.1 m/s due the observations made in
Chapter 4, where the majority of time spent was in the flow regime of 0.0-0.1 m/s while the
animal on one occasion was observed downriver after a relative high freshwater discharge
event, indicating a possible avoidance of elevated flow velocities (Chapter 4).
In order to assess the contribution of the above-mentioned optimums on the resultant

movement behaviour of the agents, each track and pre-track simulation (section 5.2.4) was
run with salinity optimums set to zero and compared with the model results where the
optimum was set (Table 5.2.1). In a similar fashion, the contribution of flow velocities on the
simulated movement was assessed by running each model scenario with the effects of current
flows removed from the equations, so that agents would move around in the system
unaffected by the movement of the water body.
Following Humston et al. (2000), the value of the shape parameters , , and was
chosen to reflect the variability of salinity, temperature and current flow through the system
on a spatio-temporal scale. The values of 0.7 and 0.9 for and , respectively, were
adapted from Humston et al. (2004) in order to have agents decrease their movement speed in
conditions that are close or equal to the specified optimum values, thus decreasing the
probability of moving out of the area and increase the time spent in optimal conditions.
As mentioned in section 5.2.2.1, all three types of models were subject to movement-forcing
rules, with the intention of preventing agents from crashing into model boundaries or land
values. The forcing rules consist of two general criteria formulated into an IF statement,
described as follows:
Eq. 5.17
Where is the ambient total water depth at the agent’s location, while is the
gradient magnitude of an artificial channel placed in the middle of the estuary within a 10 m
radius (Figure 5.4). is the direction to the highest gradient magnitude of the channel
forcing within a 60 m radius, and is the resultant direction of the and
components calculated from the movement equations specified in the above sections.
122
Figure 5.2.4: The artificial channel forcing shown for the middle reach of the model domain. Should an agent
move more than 10 metres away from the channel forcing, it will automatically shift its heading towards the
channel forcing in the next timestep.
The artificial channel consists of a one to two Eulerian-cell wide line that spans from the river
mouth to the limit of the tide-dominated zone in the upper reaches. It follows from the
movement rule in Eq. 5.17 that should an agent happen to move more than 10 metres away
from the channel, it will no longer be able to register its magnitude, and will shift its direction
towards the channel in the following timestep and continue this direction until it again is
within 10 metres of the channel. The logic behind this movement rule is to first and foremost
prevent agents from crashing into model boundaries and thus become stranded, while also
preventing them from entering water that is too shallow for them to inhabit. The obvious
downside to this movement rule is that it restricts freedom of movement and biases the
agent’s frequency of directions toward the middle of the river. However initial calibration
runs of the models revealed that without this movement rule in place, the vast majority (97-
100%) of released agents would crash and get stuck onto land values before the simulation
123
came to an end, and thus it was deemed necessary to keep the rule in place, despite its
obvious downsides.
5.2.6 DATA ANALYSIS

The specified outputs for each movement model consisted of ambient values of salinity,
temperature, current speed, total water depth, horizontal speed and direction of agents as well
as their X- and Y-coordinates for each timestep. Prior to data analysis, agents that stranded on
land values despite the enforced movement rules specified by Eq. 5.17, were excluded from
the dataset. Thus, the ‘agent success rate’ is defined as the number of agents that make it
through the entire simulation without getting stuck on cells containing dry land values.
Data for the corresponding times between simulated tracks and the observed animal locations
were extracted in order to calculate the distance from each agent to the registered location of
the animal. Due to the curved and bending nature of the system, the distance was calculated
by generating a mid-channel longitudinal transect line with an equidistant distance between
coordinate points (0.973 m), which spans from the river mouth to the end of the tide-
dominated zone. At model timesteps corresponding to recorded track timestamps, the closest
point on the transect point for each agent (and corresponding animal locations) was then
determined by calculating the Euclidian distance from agent locations to each point on the
transect line and finding the respective minimum distances for each agent. The final
estimated distances between simulated agents and animal locations were then defined as:
Eq. 5.18
Where is the number of the coordinate set corresponding to the transect point closest to
the ith agent at the jth timestep, while is the number of the coordinate set corresponding
to the transect point closest to the observed animal location at the jth timestep. is the
distance between transect coordinate points (0.973 m).
124
Figure 5.2.5: Simplified schematic of the middle reach area, highlighting the essence of the proposed distance
estimation methodology. Note the potential for a huge bias in the Euclidian distance estimation (as illustrated),
compared to distance estimation using the transect line as a reference.
A relatively minor bias is attached to the use of this method, because it "projects" all agents
and animal locations onto the middle of the river, but due to the narrow nature of the system
this method is deemed more accurate than calculating the Euclidian distance between points,
since it takes the shape of the system into account (Figure 5.2.5). The calculated distances for
all simulated agents were further processed by calculating the median distance as well the
90th and 10th percentile and plotted against the duration of the simulated track period in order
to assess the general performance of the model.
Similarly the median together with the 90th and 10th percentile was calculated for ambient
values of salinity, temperature, current speed, total water depth and horizontal swimming
velocity and plotted against the values extracted from the hydrodynamic model at the
observed location of the animal over the duration of each tracking event. For further
assessment of model performance, the Quality index (explained in Chapter 3) was calculated
for each of above mentioned parameters. The frequencies of simulated and observed
horizontal swimming directions between corresponding points in time were divided into four
125
bins corresponding to North (>315-45°), West (>45-135°), South (>135-225°) and East
(>225-315°) and compared using the χ2-test. Furthermore, using the test for Circular-Circular
correlation (Berens 2009), the sequential headings of the best-fitted track (meaning the agent
who proved to have the least distance to observed locations) was analysed against the
sequential headings of observed tracks.
The performance of the pre-track models were analysed by calculating the start distance of
each of the 100 randomly distributed agents to the registered start location of the given track
using the same calculation method as described above, and compare them to the end distance
of the agents after a 12-hour simulation. Agents that became stranded during the simulation
were, however, excluded from analysis. Since the agents’ end location is considered heavily
dependent on their start location, only descriptive statistics will be used to analyse these
results. In a similar fashion, start levels of salinity, temperature and current speed will be
compared to the levels at the end of the simulation in order to evaluate the agents’ ability to
locate habitat that is closer to the optimum conditions specified by the model equations.
5.3 RESULTS
Due to the substantial amount of data produced by the applied model templates for every
track and associated pre-tracks, only the most essential results of each model will be
presented in the following sections. Based on the collective model results summed up in
Tables 5.3.1 to 5.3.12 the best-fitting model template for a particular track/pre-track will be
defined and presented in turn, while remaining model results are available in the
“Appendixes” subfolder on the DVD-Appendix. Selected video outputs of model tracks are
likewise available for inspection in the attached DVD appendix in the “ABM_Videos” folder.
Specific ABM output videos are available upon request of the author.
5.3.1 TRACK I
Compared to the track conducted on the 28/01-2010, none of the applied model templates
were able to replicate the animal’s migration to the upper reach, resulting in a high mean and
minimum distance to actual animal locations for all model simulations (Table 5.3.1). The No-
Current control Kinesis search model for optimal salinity, temperature and flow velocities
(KSTC–NC) provided the best fit with measured data, with a minimum mean distance of
892.8 m to observed track locations (S.D. = 1282.5 m, n = 73). Inter-model performance
126
differed mostly in agent success rate, with the Random Walk (RW) model having the highest
agent success rate of 90 agents.
Table 5.3.1: Primary model results from the five applied model templates for the simulation covering the
track that was conducted on the 28/01 - 2010. RW = Random Walk Model, KS = Kinesis Search for Optimal
Salinity, KSTC = Kinesis Search for optimal salinity, temperature and current flow, KSTC-NC = KSTC with the
contribution of current flow on agent movement removed, KSTC-ZS = KSTC with an optimum salinity of 0
PSU.
TRACK: 28/01 -2010

Model Template: RW KS KSTC KSTC - NC KSTC - ZS
Agent Success Rate (n) 90 78 37 73 11
Mean Distance (m) 1324.2 1304.8 1299.0 1005.6 1287.3
Min. Distance (m) 1303.9 1266.3 1249.8 892.8 1239.4
S.D. (Mean) 1400.7 1401.7 1401.4 1282.5 1392.8
As mentioned in section 5.2.5, the KSTC-NC model allows agents to move around without
being affected by the u- and v-components of the current flow, and while this model template
is meant as a control to assess the contribution of current flow on simulated agent movement,
the better fit of the KSTC-NC model suggests that the simulated movement of current-
affected models are too sensitive in regards to current for flow. However, the overall inability
across all models to capture the observed upriver migration (Figure 5.3.1) suggests that this
behaviour cannot be explained sufficiently by a random walk or a kinesis search for optimal
conditions in terms of salinity, temperature and flow velocities alone.
127
Figure 5.3.1: The best-fitted simulated agent locations as predicted by the KSTC-NC model vs. observed animal
locations throughout the track conducted on 28/01 – 2010. Displayed agent locations correspond to the time of
each animal registration, while agent locations in the time between animal registrations are not displayed.
As tested by the Pre-track model scenario (described in section 5.2.4), the predictive abilities
of each model template were quite poor, as none of the models were able to achieve a
decrease in distance from the agents’ start-to-end location relative to the observed animal
location. The clear difference of mean start distance in between model types (Table 5.3.2) is
due to the fact that while the 100 randomly-selected release locations remain the same
throughout each model type, only the agents that make it through the entire simulation are
used for the calculation of the start and end mean distances. Thus the calculated distances are
highly dependent on the agent success rate. The Kinesis search for optimal salinity (KS)
model had the highest agent success rate of 83, while the KSTC-NC control model had the
lowest mean end distance to the observed animal location (Table 5.3.2). Upon inspection of
Figure 5.3.2, it becomes apparent that the agents of the KSTC-NC model linger in the same
general area of release, thus neither increasing nor decreasing their end distance to the
observed animal location.
128
Table 5.3.2: Primary model results from the five applied model templates for the simulation covering the 12-
hr period prior to the track conducted on 28/01 - 2010.
PRE-TRACK: 28/01 -2010

Mean Start Distance (m) 3173.4 2465.3 3318.0 3603.5 6077.2
S.D. (Start) 1918.3 1806.6 2217.2 2004.3 53.6
Mean End Distance (m) 5796.9 4055.1 5546.1 3614.0 7059.1
S.D. (End) 553.6 1785.7 996.4 1971.5 69.5
Figure 5.3.2: The start (point of release) and end locations of simulated agents predicted by the KSTC-NC
model vs. the first registered animal location on the track conducted on 28/01 – 2010. Displayed start/end agent
locations correspond to the specific agents that make it through the simulation without getting stuck on a land
value. The lack of release points in the upper reach is due to the fact that agents released in this section did not
make it through the entire simulation.
5.3.2 TRACK II
Compared to the previous track simulations, there was a noteworthy increase in model
performance for the track conducted on 03/02 -2010. The KSTC-NC control model once
again provided the best fit to the observed data set, with an agent success rate of 96 and a
minimum mean distance of 110.9 ± 146.4 m. The good fit between the control model and
129
measured data further supports the notion that the u- and v-components of the current flow
exert an overwhelming effect on the other model templates. It is, however, also plausible that
the good fit is an artefact of registered animal locations being concentrated in the same
general area, and that the lack of current influence on the KSTC-NC template causes
simulated agents to linger in the same general area as it was the case for the 28/01-2010
KSTC-NC Pre-track simulation.
track that was conducted on 03/02 - 2010.
TRACK: 03/02 -2010

Mean Distance (m) 1140.7 1007.6 1001.1 179.9 1051.3
Min. Distance (m) 948.9 812.8 753.8 110.9 933.4
S.D. (Mean) 955.6 843.4 852.7 146.4 876.0
The contribution of current flow to the movement of simulated agents is visualised in Figure
5.3.3 (KSTC) and Figure 5.3.4 (KSTC-NC), from which it becomes evident that animal
movement in the current-affected KSTC-model is clearly affected by an outgoing tide,
whereas KSTC-NC agents hardly have any longitudinal travel in comparison. While this
apparent issue is discussed in detail in Section 5.4, an immediate explanation for this obvious
sensitivity to current flow is likely that the non-corrected u- and v-components of simulated
agents were too weak in magnitude compared to the relative contribution of current flow,
ultimately resulting in a flow-corrected agent u- and v-component that is dominated by the
speed and direction of the current. While the KSTC-NC model provided the best fit in terms
of distance to observed locations, it is important to note that this model template was intended
as a control through the removal of the contribution of current flow on the agents’ resultant u-
and v-component. This is in essence against the laws of physics, as any object suspended in a
water body will be affected by the movement of the water. Thus, the KSTC-NC results will
not be treated as a primary source for model validation, but rather function as a stress test of
the four other model templates.
130
Figure 5.3.3: The best-fitted simulated agent locations as predicted by the KSTC model vs. observed animal
locations throughout the track conducted on 03/02 – 2010. Note how the simulated agent follows the tide
downriver.
Figure 5.3.4: The best-fitted simulated agent locations as predicted by the KSTC-NC model vs. observed animal
locations throughout the track conducted on 03/02 – 2010. Note how the agent has very limited longitudinal
travel.
131
Figure 5.3.4: KSTC-model results (median, 10th and 90th percentile values) of distance (top left), depth (top
right), salinity (middle left), temperature (middle right), swimming velocity (bottom left) and current speed
(bottom right) versus corresponding track values (black line) for the 3 rd of February 2010 plotted on a temporal
axis.
132
As evident from Figure 5.3.4, the calculated distances for the KSTC model were in relative
good agreement with registered animal locations until approximately 12:00 pm, when the
current speeds started picking up along with the outgoing tide, resulting in a downriver
movement and increased distance to observed animal locations, as well as a slight increase in
agent swimming velocities. As an effect of the modelled downriver movement, ambient
levels of salinity slightly increased from 20 to 25 PSU while observed values remained
relatively constant around 20 PSU. Observed swimming velocities mostly remained within
the 10th and 90th percentile values of modelled velocities, while the mean total distance
travelled was 5.22 km compared to 5.66 km estimated from the recorded track, and there was
a significant circular-correlation (P < 0.05, R = 0.37) between sequential animal and best-
fitted agent headings. Not surprisingly, ambient depth levels for simulated vs. recorded
animal locations were in poor agreement, which primarily is due to the movement constraints
imposed by equation 5.17 (section 5.2.6), effectively preventing agents from getting too close
to the riverbank.
The pre-track results for each applied model template (Table 5.3.4) displayed a poor
predictive ability, with no model being able to simulate an overall agent migration toward the
registered location of the animal. However, the movement predicted by the RW and,
particularly the KS model, caused a congregation of agents in the lower part of the middle
reach as well as in the upper reach (Figure 5.3.5), which is also evident from a notable
decrease in the standard deviation of the mean end distance (Table 5.3.4).
hr period prior to the track that was conducted on 03/02 - 2010.
PRE-TRACK: 03/02 - 2010

Mean Start Distance (m) 2656.2 2032.1 2214.0 3242.8 Nil
S.D. (Start) 1770.2 1426.3 1762.9 1937.4 Nil
Mean End Distance (m) 2674.5 2192.7 2401.1 3243.1 Nil
S.D. (End) 820.4 865.9 1309.0 1913.1 Nil
133
Figure 5.3.5: The start (point of release) and end locations of simulated agents predicted by the KS model vs.
the first registered animal location on the track conducted on 03/02 – 2010. Displayed start/end agent locations
correspond to the specific agents that made it through the simulation without getting stuck on a land value.
5.3.3 TRACK III

For the model simulations covering the track conducted on the 27th of February, the KSTC-
NC model once again provided the best fit to observed animal locations, with an agent
success rate of 94 and a minimum mean distance of 295.2 ± 439.3 m, while the KS model
provided the next best fit in terms of both agent success rate and minimum mean distance
(Table 5.3.5).
TRACK: 27/02 - 2010

Mean Distance (m) 595.0 553.1 533.5 381.0 541.6
Min. Distance (m) 532.7 465.3 443.6 295.2 416.6
S.D. (Mean) 340.9 317.3 314.0 439.3 321.7
134
As it is evident from Table 5.3.5 there are only relatively small differences between models
in terms of mean distances to observed values (KSTC-NC excepted), which suggests that the
Kinesis search models (KS, KSTC and KSTC-ZS) in terms of model formulation does not
explain the observed movement better than the random walk (RW) control model for this
particular track simulation. This notion is further backed by the similarities between the best-
fit results from the KS model and RW model depicted in Figure 5.3.5 and 5.3.6, respectively.
Figure 5.3.5: The best-fitted simulated agent locations predicted by the KS model vs. observed animal locations
throughout the track conducted on 03/02 – 2010. Note the similarity between the best-fitted movement locations
predicted by the KS model versus the RW model depicted in Figure 5.3.6.
135
Figure 5.3.6: The best-fitted simulated agent locations predicted by the KS model vs. observed animal locations
throughout the track conducted on 03/02 – 2010. Note the similarity between the best-fitted movement locations
predicted by the RW model versus the KS model depicted in Figure 5.3.5.
Salinity and temperature levels for simulated agents were in good agreement with observed
levels (Figure 5.3.7), with a mean salinity and temperature of 7.9 PSU and 27.9 °C,
respectively. With a mean of 0.17 m/s, simulated swimming velocities were decreased
compared to the previous track simulation on the 3rd of February. This result is most likely
due to the relatively close proximity of the ambient salinity levels to the specified KS model
optimum of 13.5 PSU, causing agents to slow down in accordance with the Kinesis model
formulation as described in section 5.2.
The total mean distance travelled by simulated agents was 5.71 km as opposed to 5.14 km
estimated from registered animal locations, and there was a significant circular-correlation (P
< 0.05, R = 0.37) between sequential animal and best-fitted agent headings.
136
Figure 5.3.7: KS-model results (median, 10th and 90th percentile values) of distance (top left), depth (top right),
salinity (middle left), temperature (middle right), swimming velocity (bottom left) and current speed (bottom
right) versus corresponding track values (black line) for the 27 th of February 2010 plotted on a temporal axis.
137
Pre-track simulations for the 27th of February displayed the same behaviour as the above-
mentioned pre-track simulations, by having a poor predictive ability. Simulated agents across
model templates (KSTC-NC excluded) had a tendency to congregate at upstream and
downstream areas, with the predominant current flow and direction being the main driver for
longitudinal migration.
PRE-TRACK: 27/02 - 2010

S.D. (Start) 1991.0 1550.0 1755.3 1900.6 928.5
S.D. (End) 664.0 1498.7 710.9 1909.5 320.9
5.3.4 TRACK IV
The KSTC-ZS model (Salinity optimum: 0 PSU) performed the best for the track simulation
for the 6th of March in terms of the minimum mean distance. However, the random walk
control model had the best overall mean distance of 558.3 ± 229.2 m (n = 100). As evident
138
from Table 5.3.5, there was little difference in performance between model templates,
suggesting that the contribution of the Kinesis search equations (Section 5.2.1) for this
particular model scenario is approaching that of a random walk. Simulated swimming
velocities were, however, notably different between the RW and KSTC-ZC models, with the
mean and 90th percentile values being 0.52 and 0.73 m/s, respectively. The increased
swimming velocities predicted by the KSTC-ZC model were primarily caused by the ambient
salinity levels of 20-24 PSU, which in turn caused agents to move faster due to being far from
the salinity optimum of 0 PSU.
TRACK: 06/03 - 2010

Mean Distance (m) 558.3 582.7 596.8 815.3 582.2
Min. Distance (m) 494.0 450.8 480.0 706.5 436.0
S.D. (Mean) 229.2 253.4 259.0 497.9 254.1
Figure 5.3.9: The best-fitted simulated agent locations predicted by the KSTC-ZS model vs. observed animal
locations throughout the track conducted on 06/03 – 2010.
139
Mean total distance travelled was 4.51 km, as opposed to the recorded 3.36 km. The best-
fitted sequential agent headings compared to sequential animal headings had a circular-
correlation coefficient of 0.75, although it was not significant (P = 0.06).
The performance of the pre-track simulations for the 6th of March increased significantly
compared to previous pre-track simulations, with all model templates (KSTC-NC excluded)
predicting a notable decrease in both their mean end distance and corresponding standard
deviation (Table 5.3.8). The highest decrease between start and end distances (-1529.0 m)
was achieved by the KS-model, while also having the highest agent success rate (n = 90) and
largest reduction in standard deviation (-906).
PRE-TRACK: 06/03 - 2010

S.D (Start) 1701.8 2078.4 1748.7 1518.4 1772.1
S.D. (End) 891.3 1172.1 1001.2 1477.9 1096.9
As evident from the KSTC-NC model results listed in Table 5.3.8, there was no notable
decrease in neither mean end distance or its corresponding standard deviation, which once
again indicates that the predicted downriver migration of agents by the other model templates
was an effect of increased outgoing currents due to higher levels of freshwater discharge
during the pre-track period (See Section 4.3.4). The notion that the downriver migration is an
effect of current flows rather than a search for optimal salinity and temperature is further
backed by the fact that the RW model performs almost identical to the KS and KSTC model,
while the KSTC-ZS model departs significantly from its zero salinity optimum.
140
5.3.5 TRACK V
In terms of agent distances to observed locations, the best-fitted model (KSTC-NC excluded)
for the track conducted on the 20th of March 2010, was the KSTC-ZS model, with an agent
success rate of 100, and minimum mean distance of 380.2 ± 312.5 m (Table 5.3.9). The
KSTC model performed almost identically to the KSTC-ZS model, while simulated agents
from both the RW and KS model were driven downriver by the outgoing tide. Since KSTC
and KSTC-NC performed similarly despite having different salinity optimums (13.5 and 0
PSU respectively), these results suggest that the difference in model performance between the
RW/KS and KSTC/KSTC-ZS models are due to the addition of a kinesis search for optimal
temperature and current flow in the KSTC/KSTC-NC models.
141
TRACK: 20/03 -2010

Mean Distance (m) 1580.6 1440.2 434.7 335.3 406.9
Min. Distance (m) 1466.1 1220.0 383.4 287.4 380.2
S.D. (Mean) 1177.9 1058.5 326.5 359.4 312.5
The most notable difference between the KSTC-ZS and KSTC model was in terms of
predicted swimming velocities, where the mean and 90th percentile values were 0.26 (KSTC-
ZS) and 0.39 m/s, respectively. Again, this difference in predicted swimming velocities was
due to ambient salinity levels being practically 0 PSU, meaning the simulated agents of the
KSTC-ZS model were in their defined salinity optimum and thus would slow down in
accordance with the model formulation described in section 5.2. The difference in swimming
velocities between the two models resulted in a notable difference in the predicted mean total
distance travelled, with the KSTC-ZS model predicting a mean total distance of 3.63 km,
while the KSTC model predicted 5.57 km. Both models did, however, underestimate distance
travelled compared to the recorded value of 7.73km. The best-fitted sequential agent heading
142
for both models were significantly correlated with recorded animal headings (P < 0.01), with
correlation coefficients of 0.46 (KSTC-ZS) and 0.53 (KSTC).
Model performance dropped drastically for the pre-track simulations (Table 5.3.10), and
results resembled the previously described pre-track patterns for the 28/01, 03/02 and 27/02,
with simulated agents predominantly following the direction of the current flow. This further
supports the notion that the applied model templates are inadequate in their respective
formulations in regards to directing agents to move more independently relative to current
flow, as witnessed by the track data.
12-hr period prior to the track that was conducted on 20/03 - 2010.
PRE-TRACK: 20/03 -2010

S.D. (Start) 2061.5 1852.7 2020.2 1904.9 2245.9
S.D. (End) 1365.2 1696.0 1360.4 1889.6 1703.3
5.3.6 TRACK VI
All model templates (KSTC-NC excluded) performed similarly for the simulation covering
the 21st of March, and were reasonably fitted to observed animal locations. The best-fitted
simulation was achieved by the KSTC model, with a minimum mean distance of 577.1 ±
518.0 m. However, the similarity in performance between the RW and Kinesis models once
again suggests that simulated movement was primarily governed by prevailing flow
directions, rather than the kinesis movement equations.
TRACK: 21/03 - 2010

Agent Success Rate (n) 94 88.00 91 100 93
Mean Distance (m) 737.8 701.5 697.5 618.2 696.7
Min. Distance (m) 689.7 592.8 577.1 456.6 612.1
S.D. (Mean) 561.0 518.0 518.0 397.5 515.7
143
The mean total distance travelled in the KSTC simulation was 4.85 km with a mean
swimming velocity of 0.25 m/s as opposed to the 7.16 km travelled and mean velocity of 0.38
m/s estimated from recorded animal locations. With a correlation coefficient of 0.47., the
best-fitted sequential agent headings predicted by the KSTC model were significantly
correlated (P < 0.01) with registered animal headings. As evident from Table 5.3.12, pre-
track model performance remained poor, with no notable improvement from the start location
of simulated agents across model types to the location of agents at the end of simulation, and
thus again reflecting the applied models' inability to predict and locate the utilised habitat of
the juvenile bull shark, under the given model assumptions.
12-hr period prior to the track that was conducted on 21/03 - 2010.
PRE-TRACK: 21/03 - 2010

S.D (Start) 1648.6 1372.2 1628.7 1554.7 1604.7
S.D. (End) 1245.5 1020.7 1193.1 1562.6 1101.3
144
5.4 DISCUSSION
While neither of the proposed movement models was able to replicate observed movement
patterns of the track individual to a satisfactory degree, it should be noted that the proposed
model formulations to this date still are in development. The time-constraints imposed on this
project also did not allow for a full investigation of the potential performance of the proposed
models (section 5.2). Due to the developmental nature of the models, the results presented in
the previous section are affected by both technical restrictions and theoretical limitations, and
the nature of these two limitations will be discussed in turn.
5.4.1 TECHNICAL RESTRICTIONS

The technical difficulties that the project had mainly revolved around preventing agents from
stranding onto dry cells of the model. As mentioned in section 5.2.5, it was necessary to
impose a very restrictive movement rule (Eq. 5.17) as well as decreasing the maximum
swimming velocity, , to 1.5 m/s in order to keep agents from crashing onto dry cells over
a 10-second timestep. While this formulation was successful (in most simulations) in
preventing agents from stranding, it also produced very unfortunate side-effects on the
models, which can be summarised as: 1) severely limiting latitudinal movement, effectively
rendering the agents incapable of utilising the system to its full extent, while making a
comparison between observed latitudinal migration (riverbank to riverbank movement) and
simulated movement impossible; 2) by decreasing from 3.0 to 1.5 m/s the agents’
average swimming speed was as a result also decreased. This made agents more susceptible
to the effects of current speed and direction, which is evident by the results presented in
section 5.3. This undesirable effect was found to be enhanced by the equational formulation
of the Kinesis model, as agents located close to their defined optimums would slow down
their movement speed even further, which in turn resulted in an almost passive drift of
agents; 3) by forcing particles to move towards the channel-gradient (Figure 5.2.4) whenever
they were > 15 m away from it, this resulted in a reduction of the relative contribution of
model equations on movement, e.g. if an agent was located 16 m away from the channel
gradient, the movement rule of Eq. 5.17 would supersede the Kinesis-search equations and
make it swim toward the channel gradient with its current swimming velocity. Since agents
simulated by the Kinesis-search formulation has a decreased probability of turning in the
following timestep (depending on their swimming velocity in the previous timestep), agents
145
would on occasion continue in the same direction in the following timestep. This resulted in
agents swimming past the channel gradient and ending up > 15 m away from it on the other
side, which ultimately resulted in a nonsensical zigzagging of the agents; 4) very narrow parts
of the system (discussed in section 3.4), which were < 20 m in total width prevented a
decrease of the distance-response value to the channel gradient (15 m), since the likelihood of
agents stranding in these sections of the system would approach 100% at higher distance-
response values; 5) while a decrease in the model timestep (10 s) results in agents travelling
shorter distances in between time steps, and thus decreases the likelihood of hitting dry cells,
a reduction of the timestep to 2 s meant an extreme increase in simulation times (from 20
minutes to 36 hours). Furthermore, a reduction in the timestep would result in agents
adjusting their speed and direction more often, resulting in a more tortuous movement. While
any animal capable of movement inherently has the ability to adjust its speed and direction at
any given point in time, it seems unlikely that a large slow-moving animal such as C. leucas
would randomly change its speed and direction (following the definitions of the random walk
and kinesis search equations) every 2 seconds. Even with a 10-second timestep it can be
argued that the model are likely to overestimate the frequency and intensity of changes in
animal heading and velocity, which is in sharp contrast with the 25-min timestep used in the
original kinesis formulations by Humston et al. (2000) and Humston et al. (2004). However,
due to the abovementioned reasons, a minimal timestep was required to prevent animals from
crashing into land values. As a concluding remark on the technical restrictions, it is important
to note that most if not all of these issues arose from the flooding & drying tidal
characteristics of the system as well as the very narrow (relative to length) nature of
Tallebudgera Creek.
5.4.2 THEORETICAL LIMITATIONS

As mentioned in section 5.1 and 5.2, the models that have been tested in this study were
applied by making a range of simplifying assumptions about the behavior and biology of the
tracked animal. First and foremost the model does not include any predator-prey interactions
due to the assumption that the tracked individual was not foraging during tracking, thus
rendering any effects on movement behavior negligible. Young sandbar sharks, C. plumbeus,
which is considered a closely related species to the bull shark, have been reported to take up
to 80 hrs for full gastric evacuation while having a feeding duration of 8-9 hrs (Medved et al.
146
1985, 1988). In a review of feeding habits of sharks, Wetherbee et al. (1990) suggest that
sharks that are fasting after a large feeding event, are likely to encounter potential prey in this
period, but may not be attracted to it because they are satiated. While no detailed data exists
on the feeding durations and gastric evacuation rates of the bull shark (to my best
knowledge), it is plausible that this species follows a similar digestive pattern, with prolonged
non-foraging periods between feeding excursions. However, as there was no direct way to
determine whether the tracked animal was in a state of foraging, the validity of
abovementioned assumption relies solely on probability, e.g. if the tracked animal’s
feeding/non-feeding time allocations are identical to those reported for young sandbar sharks,
there would be a ~90% chance that the animal would be in a state of non-foraging at the start
of a given track, but only a 53% chance of the shark being in a non-foraging
state at the start of all six tracking events (assuming that consecutive tracks are independent).
In reality the chance of the animal being in non-foraging state throughout the entire tracking
period would be even lower, as the shark might shift from non-foraging to foraging at any
point during a tracking event. Furthermore there is the chance that the animal will perform
asynchronous opportunistic feeding events at any given point during its ‘non-foraging’
period.
Aside from assuming foraging behavior to be neglible in short-term movement, the model
also assumed that movement originating from interaction with conspecifics is negligble. In a
long-term study (18 months), Heupel & Simpfendorfer (2011) suggest that estuaries provide
a low-mortality nursery ground for neonate bull sharks due to lower predation risk and
decreased competition for food compared to coastal nurseries utilised by neonates of other
shark species. While the tracked animal was a juvenile, and potentially large enough to prey
on neonate conspecifics, Werry (2010) reported that sub-adult C. leucas make regular
excursions into Gold Coast estuaries from nearby coastal areas and linger in the system
before heading offshore again. The plausible presence of sub-adults inside the Tallebudgera
system is therefore likely to affect the movement and habitat selection of neonate and
juvenile C. leucas, although the actual contribution of predator-avoidance movement is
difficult to assess given the obvious limitions of the tracking record.
In addition to the potential effects on movement of foraging and species interaction, the
Kinesis model assumes that only salinity, temperature and flow velocities affect movement,
while disregarding other potential water quality influences such as dissolved oxygen, pH, and
turbidity. The results of Ortega (2009) reported a correlation between C. leucas ROM and
147
DO, pH and turbidity, and while it remains unclear what effect these parameters have on the
short-term movement of C. leucas, the track record of this study indicates that turbidity could
play a part in the displacement of bull sharks after a large freshwater influx (Section 4.4).
While the HD model (Chapter III) in its current state does not simulate any of the
abovementioned parameters, an implementation of the MIKE ECOLab module could allow a
future spatio-temporal representation of these parameters (Szylkarski et al. 2004), which in
turn could be related to a new and more detailed track record of C. leucas and eventually
support a more detailed agent-based model.
5.4.3 CONCLUDING REMARKS

Due to the technical problems described above, the ability of the applied movement models
to replicate the movement patterns is difficult to assess. It remains clear that the efforts made
to prevent agents from crashing onto land in turn had a detrimental effect on the actual
movement equations (Section 5.3), and the resultant decreased swimming velocities led to an
almost passive drift of agents with the current. The dominant effect of current velocities on
all models (KSTC-NC excluded) also clouded potential differences in inter-model
performance, which is reflected by the little to no difference between simulated movements
of the RW- KS- KSTC- and KSTC-ZS models between tracks. However, the results of the
no-flow KSTC-NC model revealed that agents only spread out across on a 200-300 m stretch
from their point of origin (See DVD, Appendix V for position plots). While this could
indicate that the kinesis search formulation performs poorly in replicating observed
movement patterns, it is likely that the movement rule of Eq. 5.17 was the main cause of the
simulated KSTC-NC pattern. Although inconclusive in nature, the results for all Pre-track
simulations (KSTC-NC excluded) for the 6th of March (Section 5.3.4) showed a stronger
displacement downriver of all agents, which is accordance with the observed results. While
this is an effect of increased outgoing flow due to greater freshwater influx during this
simulation period, it suggests that the theory of flow-induced movement of juvenile C. leucas
(Heupel & Simpfendorfer 2008) might be correct.
Although the current developmental stage of the applied models does not provide the ability
to conclude on model performance and thus, contribute significantly to the current
knowledge-base of C. leucas, this study has nevertheless established the framework needed
for future study of the bull shark in Tallebudgera Creek or similar water bodies, e.g.
residential canals. The Achilles heel of the ABM-development of this project was without
148
doubt the flooding and drying characteristics as well as the narrow and shallow nature of
Tallebudgera Creek, which necessitated the development of the channel-gradient movement
rule (Eq. 5.17). In the current stage of development, the overly-restrictive nature of Eq. 5.17
is essentially the only thing that stands in the way of a full assessment of the kinesis models
and the implementation of more advanced and sophisticated models that allow greater
incorporation of more biological information on the bull shark. Reformulation of Eq. 5.17 to
a less restrictive yet still effective means of keeping agents from crashing onto land has
proven difficult due to current model software limitations, as it is currently not possible to
assign dry cells with a value that the agents can detect and respond to. However, once this
problem has been overcome it is predicted that the MIKE ABM Module will provide an
excellent tool for testing and assessing theories of bull shark movement in Tallebudgera
Creek and similar water courses. This prospect will be discussed in more detail in Chapter
VI.
149
CHAPTER VI
GENERAL DISCUSSION AND FUTURE DIRECTIONS
150
6.1 SUMMARY
Over the course of this study a range of interdisciplinary techniques was applied as a means
to test an alternative approach to studying C. leucas movement. From a year-long water
quality campaign to hydrodynamic modelling and acoustic telemetry of live C. leucas, this
study has gone through all the essential stages of the construction, application and assessment
of a dynamic agent-based model. Although the final results of this study remains elusive, and
thus preventing a full assessment of proposed movement models at the present stage, there is
still immediate academic value to be gained from this project. The purpose of this Chapter is
to highlight and discuss the value of the outputs generated by this study, and discuss possible
directions for future research.
6.2 EVALUATION OF THE HYDRODYNAMIC MODEL

While the hydraulic properties of Tallebudgera Creek proved much more difficult to capture
through a 2D model framework, the end product met satisfactory standards, especially in
terms of flow and tide-dynamics (Section 3.4). Simulation of salinity and temperature
dynamics on a short-term scale did reveal room for improvement in the current model
formulation in terms of a finer resolution of model bathymetry and application of a 3D rather
than 2D model. However, the immediate value of the currently established HD model
remains clear, and the possibilities of utilising the model for future study are many. Since the
model has demonstrated an ability to simulate representative tide-dynamics for periods in
both 2007 and 2010 by using a 2001 bathymetry, one can in theory (and with caution) back-
calculate the spatio-temporal flow dynamics of Tallebudgera Creek for as long as the
freshwater discharge record allows. This in turn opens up for the possibility of relating
previous ecological studies in the Tallebudgera estuary with model outputs, and gain an
increased understanding of how hydraulics affect specific ecological properties of the system.
Similarly, any future ecological studies independent of this project could potentially be
strengthened by incorporating the existing and validated hydrodynamic model. Although the
shallow and narrow nature of Tallebudgera Creek originally imposed difficulties for the
calibration and validation of the HD model, the same nature of the system makes it close to
being ideal for in-depth study of local fish population dynamics in relation to system
hydraulics. The results of such studies could in turn be utilised as a data source for including
movement and distribution of prey in future formulations of agent-based models in this
151
system (See section 6.2.1). On an urban management level, the model also holds great
potential in terms of assessing potential flood levels under different discharge scenarios, since
it was demonstrated that the model captures such events to a satisfactory degree (Section
3.3.1).
6.2.1 ECOLOGICAL MODELLING

The existing HD model provides a solid base for further model expansion in terms of
developing a dynamic ecological Eulerian model through the MIKE ECOLab module,
capable of simulating water quality and primary productivity. The ECOLab module can be
fully integrated into the MIKE 21 HD model, coupled with an Advection-Dispersion model
that handles the transportation of the various ECOLab state-variables throughout the model
domain. The ECOLab module is in essence designed as a model-independent equation solver
for ecological modelling that allows for tailor-made process descriptions to be incorporated
directly into the model formulation.
In a study of the neighbouring Pimpama estuary, an ECOLab model template was developed
and validated in order to account for eutrophication effects on the benthic vegetation
(Szylkarski et al. 2004). Given the relative short distance to Tallebudgera Creek (36 km), it is
plausible that the custom-made Pimpama model template can be adapted to the Tallebudgera
Creek with only relative minor modifications needed. While it was the original intent of this
study to include ecological modelling of Tallebudgera Creek as a means to relate model
outputs of water quality to movement of C. leucas, the time-constraints of the project did not
allow the development of such a model. Notwithstanding being an extensive project in itself,
the implementation and validation of an ecological model coupled with the HD model would
allow a further investigation into how water quality affects fish distributions, as well as C.
leucas, on a fine spatio-temporal scale. Finally, any Eulerian variables simulated by an
ecological model could also be utilised as potential drivers for movement in an agent-based
model along with the outputs of HD model, which in turn could help investigate the effects of
DO, turbidity and pH on C. leucas movement.
6.3 AGENT-BASED MODELLING

Compared to other model approaches, agent-based models have the advantage of representing
observation and current knowledge in a form that is highly congruent with how we
understand interaction (Jopp et al. 2011), and have in the past proved useful for the study of
152
predator-prey interactions (Charnell 2008), schooling behaviour (Reuter & Breckling 1994)
and behavioural shifts under varying conditions (Railsback et al. 1999, Goodwin et al. 2001,
Goodwin et al. 2006, Peacor et al. 2007), as well as the formation of colonies and the
description of structural-functional development of modular organisms (Eschenbach 2005).
The nature of agent-based modelling does, however, demand a high level of understanding of
the species in question in order to create theoretically sensible model formulations. While
significant recent advances are evident in the knowledge base of C. leucas (Anderson et al.
2005, Pillans et al. 2005, Simpfendorfer et al. 2005, Pillans et al. 2006, Heupel &
Simpfendorfer 2008, Pillans et al. 2008, Ortega et al. 2009, Carlson et al. 2010, Werry 2010,
Heupel & Simpfendorfer 2011, Werry et al. 2011), vital pieces of information, especially
regarding bioenergetics of the bull shark in various water quality regimes, are still missing.
The kinesis search model formulation in this study was applied as a means to indirectly
simulate the proposed energetic benefits of staying within certain water quality ranges, but
without actually modelling associated metabolic costs. While this represents a minimalist
assumption of animal behaviour (Section 5.3), the apparent weakness of this formulation is
that it quickly loses its theoretical plausibility over longer time scales, since the model in its
current stage does not account for food uptake and growth of agents. Humston et al. (2004)
coupled a kinesis search model of fish cohorts with a bioenergetic growth model by including
spatio-temporal data on prey distribution as well as salinity in the collective model
formulation. It is certain that a further adaptation of Humston et al. (2004) would increase the
theoretical validity of the kinesis model proposed in this project on a longer temporal scale.
However, no spatio-temporal data currently exists for the distribution of common C. leucas
prey species (e.g. mullet) in Tallebudgera Creek, and it cannot be emphasised enough how
important such data will be for any future ABM-development of C. leucas.
Furthermore, as of this date and the author’s best knowledge, no laboratory-controlled

experiments conducted on the effects of salinity, temperature, flow velocities, pH and
dissolved oxygen on C. leucas metabolism, despite all of these parameters having been
reported to affect bull shark movement (Heupel & Simpfendorfer 2008, Ortega et al. 2009,
Werry 2010). The associated costs and immediate difficulties of such experiments are likely
causes for the apparent lack of such data, however, they could provide researchers with a
vital key to understanding observed C. leucas movement patterns. Furthermore, these studies
would aid in establishing the much-needed empirical relationships between varying water
quality regimes and energetic costs for future ABM development.
153
6.3.1 FUTURE MODELLING DIRECTIONS
As mentioned above, future ABM-development for C. leucas could significantly benefit from
directed empirical research of their bioenergetics over a range of water quality regimes, as well as the
distribution patterns of common prey species e.g. the sea mullet, Mugil cephalus. However, in
terms of actual development of the future bull shark ABM model structure, researchers need
to consider other options than a kinesis search formulation for capturing the dynamics of bull
shark movement. Sharks have been reported to utilise a variety of movement strategies both
between and within species, e.g. tiger sharks Galeocerdo cuvier, have been shown to perform
directed ‘walks’ on large spatial scales, which suggests that the animal has a spatial
knowledge of its home range (Papastamatiou et al. 2011). Furthermore, through their sensory
system sharks have been shown to use a range of signals to navigate , which includes but are
not limited to: water currents, temperature and geomagnetic fields (Klimley 1993,
Montgomery & Walker 2001, Meyer et al. 2005), while it has also been proposed they might
utilise olfactory cues to navigate over long distances (Papastamatiou et al. 2011).
In its current formulation, the kinesis search model assumes that the simulated agents have no
sensory abilities besides the ambient conditions of their current location, e.g. agents cannot
sense spatial-gradients in water quality, nor do they have information about different habitat
characteristics on a system-wide scale. Given the growing evidence of the sensory abilities of
sharks as a means to navigate, the current minimalist assumption formulation of the model
might well be an under-representation of the animal’s actual ability to utilise its sensory
system for locating favourable habitats. Therefore, any future ABM-development for C.
leucas should aim to explore the possibility of implementing a sensory range of simulated
agents, and maybe even “cognitive maps” of hotspots in the system that indicate favourable
foraging grounds.
However, as mentioned in section 6.2, in order for the agent-based model to be applicable on
a larger temporal scale, the emphasis in future ABM-development must lie in developing a
meaningful C. leucas-specific theoretical growth model. This first and foremost implies a
need for prey representation in the model formulation, but also long-term tracking records of
both C. leucas and common prey species, as a means to validate the model. While C. leucas
is considered a generalist feeder (Werry 2010), stomach contents of C. leucas suggests that
the species primarily feed on teleosts fishes (Snelson et al. 1984, Wetherbee et al. 1990).
Teleost prey species inside the Tallebudgera Creek system could in theory be modelled as
154
three anonymous functional groups: freshwater teleosts, estuarine teleosts and marine
teleosts. Each group will be described by a simple random walk model, but with movement
rules that would prevent them from entering water bodies outside of their given salinity
range, while utilising avoidance behaviour in the presence of simulated C. leucas agents. It
could then be hypothesised that the relative abundance and distribution of prey species in the
system would follow the spatial extent of the salinity gradient at any given time, which in
turn might provide adequate representation of natural prey migration dynamics. In terms of
gathering long-term movement records, the relatively small size and narrow nature of
Tallebudgera Creek makes the site ideal for strategic placement of remote listening stations,
with only a relatively small number of stations needed to cover the full extent of the system.
However, the number of long-term acoustic tags needed for both C. leucas and associated
prey species, coupled with Tallebudgera Creek being a popular destination for recreational
fishing (pers. obs.) imply that the immediate costs associated with such a project will be
substantial.
While any model must be validated by empirical evidence, one of the benefits of agent-based
modelling is that one can start testing theory and develop model formulation prior to
collecting data for validation. Thus, future ABM-development of Tallebudgera Creek could
essentially continue before additional field data become available. Such data would of course
be required at a later point for model validation.
6.3.2 POSSIBILITIES FOR CONSERVATION MANAGEMENT

Grimm et al. (2006) succinctly summarised the benefits of agent-based models as being
“important both for theory and management because they allow researchers to consider
aspects usually ignored in analytical models: variability among individuals, local interactions,
complete life cycles, and in particular individual behaviour adapting to the individual’s
changing internal and external environment” (Grimm et al. 2006).
However, due the often complex nature of many agent-based models, they are also often
challenging to analyse, document and communicate compared to analytical models (Grimm
1999, Jopp et al. 2011). While this can raise potential issues for the use of ABM’s as a tool
for environmental managers and policy-makers in terms of formulating the results of an
ABM into conservation strategies, the versatility of ABM’s as tools for conservation
management remains. The ability to representatively simulate a dynamic virtual environment
and the resulting movement response of virtual fishes to different environmental stimuli
155
effectively allows the modeller to forecast fish movement to a range of different scenarios.
Goodwin et al. (2006) utilised a coupled Eulerian-Langrangian model to forecast downstream
migration behaviour of juvenile salmon through fish bypasses in three hydropower dams,
with a total of 20 structural and operational configurations. In a similar approach, the impact
of naturally occurring flood events as well as anthropogenic changes to the environment on
the movement and distribution of C. leucas in Tallebudgera Creek could potentially be
forecast through the use of an agent-based model coupled with an Eulerian framework.
Although the ABM for C. leucas in its current stage is far from being useful as a tool to
formulate conservation strategies for this IUCN ‘near-threatened’ species, it is predicted that
once the model has been properly validated, it could become a powerful tool in future
conservation management of this species.
156
REFERENCES
Almroth E, Skogen MD (2010) A North Sea and Baltic Sea Model Ensemble Eutrophication
Assessment. AMBIO: A Journal of the Human Environment 39:59-69
Andersen J, Hansen F, Potthoff M, Uhrenholdt T, Diem Vo H, Linden O (in prep) Combining
modelling of hydrodynamic processes and agent-based modelling: Toward a prototype
decision support yool for ballast water risk management. Journal of Marine Affairs, in
prep
Anderson WG, Good JP, Pillans RD, Hazon N, Franklin CE (2005) Hepatic urea biosynthesis in the
euryhaline elasmobranch Carcharhinus leucas. Journal of Experimental Zoology Part A:
Comparative Experimental Biology 303A:917-921
APHA (1971) Standard methods for the examination of water and wastewater / prepared and
published jointly by the American Public Health Association, American Water Works
Association, Water Pollution Control Federation ; joint editorial board, Michael J. Taras,
Arnold E. Greenberg, R.D. Hoak, Vol 1. American Public Health Association, Washington,
D.C.
Berens P (2009) CircStat: A MATLAB toolbox for circular statistics. Journal of Statistical
Software 31:1-21
Branstetter S, Stiles R (1987) Age and growth estimates of the bull shark, Carcharhinus leucas,
from the northern Gulf of Mexico. Environmental Biology of Fishes 20:169-181
Brett JR, Groves TDD (1979) Physiological Energetics. In: Hoar WS, Randall DJ, Brett JR (eds)
Fish Physiology, Vol Volume 8. Academic Press, p 279-352
Canal-Vergés P, Potthoff M, Flindt M, Rasmussen E (in prep) Macroalgae movements in shallow
waters, an agent-based model. Ecological Modelling, in prep
Carlson JK, Parsons GR (2001) The effects of hypoxia on three sympatric shark species:
Physiological and behavioral responses. Environmental Biology of Fishes 61:427-433
Carlson JK, Ribera MM, Conrath CL, Heupel MR, Burgess GH (2010) Habitat use and movement
patterns of bull sharks Carcharhinus leucas determined using pop-up satellite archival
tags. Journal of Fish Biology 77:661-675
Charnell M (2008) An individual-based model of a tritrophic ecology. Ecological Modelling
218:195-206
Cheng R, Barau J, Gartner J (1991) Tidal Hydrodynamics: Interfacing data analysis and
numerical modelling for tidal hydrodynamic phenomena., Vol 1. John Wiley & Sons, New
York
Cheng RT, Casulli V, Gartner JW (1993) Tidal, residual, Intertidal mudflat (TRIM) Model and its
applications to San Francisco Bay, California. Estuarine, Coastal and Shelf Science
36:235-280
Codling EA, Plank MJ, Benhamou S (2008) Random walk models in biology. Journal of The Royal
Society Interface 5:813-834
Coombs S, Braun C, Donovan B (2001) The orienting response of Lake Michigan mottled sculpin
is mediated by canal neuromasts. Journal of Experimental Biology 204:337-348
DHI (2009) MIKE 21 & MIKE 3 Flow Model FM, Hydrodynamic and Transport Module -
Scientific Documentation, DHI, Hørsholm, Denmark
Dias JM, Lopes JF (2006) Implementation and assessment of hydrodynamic, salt and heat
transport models: The case of Ria de Aveiro Lagoon (Portugal). Environmental
Modelling & Software 21:1-15
Dias JM, Lopes JF (2006b) Calibration and validation of hydrodynamic, salt and heat transport
models for Ria de Aveiro Lagoon (Portugal). Journal of Coastal Research SI 39:1680-
1184
EHMP (2007) Annual Technical Report, Ecosystem Health Monitoring Program (EHMP),
Queensland Government, Australia
157
EHMP (2009) Annual Technical Report, Ecosystem Health Monitoring Program (EHMP),
Queensland Government, Australia
Eschenbach C (2005) Emergent properties modelled with the functional structural tree growth
model ALMIS: Computer experiments on resource gain and use. Ecological Modelling
186:470-488
Fofonoff NP, Millard Jr., R.C. (1983) Algorithms for computation of fundamental properties of
seawater, Unesco, Paris, France
García VB, Lucifora LO, Myers RA (2008) The importance of habitat and life history to extinction
risk in sharks, skates, rays and chimaeras. Proceedings of the Royal Society B: Biological
Sciences 275:83-89
Getz WM, Fortmann-Roe S, Cross PC, Lyons AJ, Ryan SJ, Wilmers CC (2007) LoCoH:
nonparameteric kernel methods for constructing home ranges and utilization
distributions. PLoS ONE 2:e207
Goodwin RA, Nestler JM, Anderson JJ, Weber LJ, Loucks DP (2006) Forecasting 3-D fish
movement behavior using a Eulerian-Lagrangian-agent method (ELAM). Ecological
Modelling 192:197-223
Goodwin RA, Nestler JM, Loucks DP, Chapman RS (2001) Simulating mobile populations in
aquatic ecosystems. Journal of Water Resources Planning and Management 127:386-393
Graham KJ, Andrew NL, Hodgson KE (2001) Changes in relative abundance of sharks and rays
on Australian south east fishery trawl grounds after twenty years of fishing. Marine and
Freshwater Research 52:549-561
Grimm V (1999) Ten years of individual-based modelling in ecology: what have we learned and
what could we learn in the future? Ecological Modelling 115:129-148
Grimm V, Berger U, Bastiansen F, Eliassen S, Ginot V, Giske J, Goss-Custard J, Grand T, Heinz SK,
Huse G, Huth A, Jepsen JU, Jørgensen C, Mooij WM, Müller B, Pe'er G, Piou C, Railsback SF,
Robbins AM, Robbins MM, Rossmanith E, Rüger N, Strand E, Souissi S, Stillman RA, Vabø
R, Visser U, DeAngelis DL (2006) A standard protocol for describing individual-based
and agent-based models. Ecological Modelling 198:115-126
Halpern BS, Walbridge S, Selkoe KA, Kappel CV, Micheli F, D'Agrosa C, Bruno JF, Casey KS, Ebert
C, Fox HE, Fujita R, Heinemann D, Lenihan HS, Madin EMP, Perry MT, Selig ER, Spalding
M, Steneck R, Watson R (2008) A global map of human impact on marine ecosystems.
Science 319:948-952
Hansen F, Potthoff M (in prep) A generic Eulerian-Lagrangian tool for modelling ecological and
biological processes in the aquatic environment. . Environmental Modelling & Software,
in prep
Heithaus MR, Frid A, Wirsing AJ, Worm B (2008) Predicting ecological consequences of marine
top predator declines. Trends in Ecology & Evolution 23:202-210
Heithaus MR, Marshall GJ, Buhleier BM, Dill LM (2001) Employing Crittercam to study habitat
use and behavior of large sharks, Vol 209. Inter-Research, Oldendorf, Allemagne
Heupel MR, Simpfendorfer CA (2008) Movement and distribution of young bull sharks
Carcharhinus leucas in a variable estuarine environment. Aquatic Biology 1:277-289
Heupel MR, Simpfendorfer CA (2011) Estuarine nursery areas provide a low-mortality
environment for young bull sharks Carcharhinus leucas. Marine Ecology Progress Series
433:237-244
Humston R, Ault JS, Lutcavage M, Olson DB (2000) Schooling and migration of large pelagic
fishes relative to environmental cues. Fisheries Oceanography 9:136-146
Humston R, Olson D, Ault J (2004) Behavioral assumptions in models of fish movement and
their influence on population dynamics. Transactions of the American Fisheries Society.
133:1304-1328
Inoue M, Wiseman Jr WJ (2000) Transport, mixing and stirring processes in a Louisiana estuary:
A model study. Estuarine, Coastal and Shelf Science 50:449-466
Jopp F, Reuter H, Breckling B (2011) Modelling complex ecological dynamics, Vol 1. Springer-
Verlag Berlin, Berlin Heidelberg
158
Klimley AP (1993) Highly directional swimming by scalloped hammerhead sharks, Sphyrna
lewini, and subsurface irradiance, temperature, bathymetry, and geomagnetic field.
Marine Biology 117:1-22
Krother S, Mogdans J, Bleckmann H (2002) Brainstem lateral line responses to sinusoidal wave
stimuli in still and running water. J Exp Biol 205:1471-1484
Lamb JC (1985) Water quality and its control, Vol 1. Wiley & Sons, New York, 384pp
Lee SY, Dunn RJK, Young RA, Connolly RM, Dale PER, Dehayr R, Lemckert CJ, McKinnon S, Powell
B, Teasdale PR, Welsh DT (2006) Impact of urbanization on coastal wetland structure
and function. Austral Ecology 31:149-163
Levene H (1960) Robust Tests for Equality of Variances. Olkin, ed Contributions of probability
and statistics, Palo Alto, Calif: Stanford University Press:278-292
Lilliefors HW (1967) On the Kolmogorov-Smirnov test for normality with mean and variance
unknown. Journal of the American Statistical Association 62:399-402
Lopes JF, Silva CI, Cardoso AC (2008) Validation of a water quality model for the Ria de Aveiro
lagoon, Portugal. Environmental Modelling & Software 23:479-494
Lotze HK, Lenihan HS, Bourque BJ, Bradbury RH, Cooke RG, Kay MC, Kidwell SM, Kirby MX,
Peterson CH, Jackson JBC (2006) Depletion, degradation, and recovery potential of
estuaries and coastal seas. Science 312:1806-1809
Manly BFJ, Miller, P., Cook, L.M. (1972) Analysis of a selective predation experiment. The
American Naturalist 106:719-736
Martin RA (2005) Conservation of freshwater and euryhaline elasmobranchs: a review. Journal
of the Marine Biological Association of the UK 85:1049-1073
Martínez ML, Intralawan A, Vázquez G, Pérez-Maqueo O, Sutton P, Landgrave R (2007) The
coasts of our world: Ecological, economic and social importance. Ecological Economics
63:254-272
Matern SA, Cech JJ, Hopkins TE (2000) Diel movements of bat rays, Myliobatis californica, in
Tomales Bay, California: Evidence for behavioral thermoregulation? Environmental
Biology of Fishes 58:173-182
McClimans TA, Pietrzak JD, Huess V, Kliem N, Nilsen JH, Olaf Johannessen B (2000) Laboratory
and numerical simulation of the Skagerrak circulation. Continental Shelf Research
20:941-974
Medved RJ, Stillwell CE, Casey JG (1985) Stomach contents of young sandbar sharks,
Carcharhinus plumbeus, in Chincoteague Bay, Virginia. Fishery Bulletin 83:395-402
Medved RJ, Stillwell CE, Casey JG (1988) The Rate of Food Consumption of Young Sandbar
Sharks (Carcharhinus plumbeus) in Chincoteague Bay, Virginia. Copeia 1988:956-963
Meyer C, Holland K, Papastamatiou Y (2005) Sharks can detect changes in the geomagnetic field.
. Journal of the Royal Society Interface 2:129-130
Mirfenderesk H, Tomlinson R (2007) Numerical modelling of tidal dynamic and water
circulation at the Gold Coast Broadwater, Australia. Journal of Coastal Research SI
50:277-288
Montgomery JC, Walker MM (2001) Orientation and navigation in elasmobranchs: Which way
forward? Environmental Biology of Fishes 60:109-116
Nestler JM, Goodwin A, Loucks DP (2005) Coupling of engineering and biological models for
ecosystem analysis. Journal of Water Resources Planning & Management 131:101-109
Nortek (2008) Aquadopp Current Profiler - User Guide, Nortek AS, Vangkroken, Norway
Ortega L, Heupel M, Beynen P, Motta P (2009) Movement patterns and water quality
preferences of juvenile bull sharks (Carcharhinus leucas) in a Florida estuary.
Environmental Biology of Fishes 84:361-373
Paliwal R, Patra RR (2011) Applicability of MIKE 21 to assess temporal and spatial variation in
water quality of an estuary under the impact of effluent from an industrial estate. Water
Sci Technol 63:1932-1943
Papastamatiou Y, Cartamil D, Lowe C, Meyer C, Wetherbee B, Holland K (2011) Scales of
orientation, directed walks and movement path structure in sharks. Journal of Animal
Ecology 80:864-874
159
Parsons GR, Carlson JK (1998) Physiological and behavioral responses to hypoxia in the
bonnethead shark, Sphyrna tiburo: routine swimming and respiratory regulation. Fish
Physiology and Biochemistry 19:189-196
Parsons TR, Yoskiaki, M., Lalli, C. M. (1984) A manual of chemical and biological methods for
seawater analysis, Vol. Pergamon Press, Oxford
Peacor SD, Allesina S, Riolo RL, Hunter TS (2007) A new computational system, DOVE (Digital
Organisms in a Virtual Ecosystem), to study phenotypic plasticity and its effects in food
webs. Ecological Modelling 205:13-28
Pillans RD, Anderson WG, Good JP, Hyodo S, Takei Y, Hazon N, Franklin CE (2006) Plasma and
erythrocyte solute properties of juvenile bull sharks, Carcharhinus leucas, acutely
exposed to increasing environmental salinity. Journal of Experimental Marine Biology
and Ecology 331:145-157
Pillans RD, Franklin CE (2004) Plasma osmolyte concentrations and rectal gland mass of bull
sharks Carcharhinus leucas, captured along a salinity gradient. Comparative
Biochemistry and Physiology - Part A: Molecular & Integrative Physiology 138:363-371
Pillans RD, Good JP, Anderson WG, Hazon N, Franklin CE (2005) Freshwater to seawater
acclimation of juvenile bull sharks ( Carcharhinus leucas): plasma osmolytes and Na +/K
+-ATPase activity in gill, rectal gland, kidney and intestine. Journal of Comparative
Physiology B: Biochemical, Systemic, and Environmental Physiology 175:37-44
Pillans RD, Good JP, Anderson WG, Hazon N, Franklin CE (2008) Rectal gland morphology of
freshwater and seawater acclimated bull sharks Carcharhinus leucas. Journal of Fish
Biology 72:1559-1571
Railsback SF, Lamberson RH, Harvey BC, Duffy WE (1999) Movement rules for individual-based
models of stream fish. Ecological Modelling 123:73-89
Reuter H, Breckling B (1994) Selforganization of fish schools: an object-oriented model.
Ecological Modelling 75-76:147-159
Sandwell DT (1987) Biharmonic spline interpolation of GEOS-3 and SEASAT altimeter data.
Geophysical Research Letters 14:139-142
Schmid TH, Murru FL (1994) Bioenergetics of the bull shark, Carcharhinus leucas, maintained in
captivity. Zoo Biology 13:177-185
Shine R, Ellway CP, Hegerl EJ (1973) A biological survey of Tallebudgera Creek estuary.
Operculum 3:59-83
Simpfendorfer C, Freitas G, Wiley T, Heupel M (2005) Distribution and habitat partitioning of
immature bull sharks (Carcharhinus leucas) in a Southwest Florida estuary. Estuaries
and Coasts 28:78-85
Simpfendorfer CA, Heupel MR, Collins AB (2008) Variation in the performance of acoustic
receivers and its implication for positioning algorithms in a riverine setting. Canadian
Journal of Fisheries and Aquatic Sciences 65:482-492
Snelson FF, Mulligan TJ, Williams SE (1984) Food Habits, Occurrence, and Population Structure
of the Bull Shark, Carcharhinus Leucas, in Florida Coastal Lagoons. Bulletin of Marine
Science 34:71-80
Sundström LF, Gruber SH, Clermont SM, Correia JPS, de Marignac JRC, Morrissey JF, Lowrance
CR, Thomassen L, Oliveira MT (2001) Review of elasmobranch behavioral studies using
ultrasonic telemetry with special reference to the lemon shark, Negaprion brevirostris,
around Bimini Islands, Bahamas. Environmental Biology of Fishes 60:225-250
Szylkarski S, Dorge J, Toomey D (2004) Hydraulic and ecological modeling of the Pimpama river
estuary. In: Engineers Io (ed) National Conference on Hydraulics in Water Engineering
Conference, Gold Coast
Thorson TB, Cowan CM, Watson DE (1973) Body fluid solutes of juveniles and adults of the
euryhaline bull shark, Carcharhinus leucas, from freshwater and saline environments.
Physiological Zoology 46:29-42
Umgiesser G, Melaku Canu D, Solidoro C, Ambrose R (2003) A finite element ecological model: a
first application to the Venice Lagoon. Environmental Modelling and Software 18:131-
145
160
Van Winkle W, Jager HI, Railsback SF, Holcomb BD, Studley TK, Baldrige JE (1998) Individual-
based model of sympatric populations of brown and rainbow trout for instream flow
assessment: model description and calibration. Ecological Modelling 110:175-207
Voight M, Dietmar W (2011) Field guide for sharks of the genus Carcharhinus, Vol 1. Verlag Dr.
Friedrich Pheil, München, Germany. ISBN: 978-3-89937-132-1. 151 p.
Wallcraft AJ, Metzger, E.J. (2009) Software design description for the HYbrid Coordinate Ocean
Model (HYCOM) version 2.2, Naval Research Laboratory, Oceanograhy Division, Stennis
Space Center, Mississippi, USA
Warren IR, Bach HK (1992) MIKE 21: a modelling system for estuaries, coastal waters and seas.
Environmental Software 7:229-240
Weihs D, Keyes RS, Stalls DM (1981) Voluntary swimming speeds of two species of large
carcharhinid sharks. Copeia 1981:219-222
Welch BL (1951) On the comparison of several mean values: An alternative approach.
Biometrika 38:330-336
Werry JM (2010) Habitat ecology of the bull shark Carcharhinus leucas on urban coasts in
eastern Queensland, Australia. Ph.D dissertation. Australian Rivers Institute and School
of Environment, Griffith University at Gold Coast Campus, Gold Coast, Australia
Werry JM, Lee SY, Otway N, Hu Y, Sumpton W (2011) A multi-faceted approach for quantifying
the estuarine-nearshore transition in the life cycle of the bull shark, Carcharhinus leucas.
Marine and Freshwater Research 62:1421-1431
Wetherbee BM, Gruber SH, Cortes E (1990) Diet, feeding habits, digestion, and consumption in
sharks, with special reference to the lemon shark, Negaprion brevirostris. In:
Elasmobranchs as a living resources: advances in the biology, ecology, systematics and
the status of the fisheries., (H.L Pratt, Jr., S. H. Gruber and T. Taniuchi, eds.), p. 29-47. U.S.
Dep. of Commer., NOAA Tech. Rep,
YSI (2006) YSI user manual. In: 6-Series multiparameter water quality sondes. Appendix O,
Yellow Springs, Ohio, p 1-14
Zacharias I, Gianni A (2008) Hydrodynamic and dispersion modeling as a tool for restoration of
coastal ecosystems. Application to a re-flooded lagoon. Environmental Modelling and
Software 23:751-767
161
APPENDIX I
TEMPORAL VALIDATION OF SALINITY AND TEMPERATURE ACROSS ALL STATIONS
Appendix I, Table 1: Calculated results of the Salinity Quality index for each measuring station over the course
of 2010 vs. corresponding simulated values. See following pages for plotted data sets.
TEMPORAL QUALITY INDEX - SALINITY

Station Correlation P-value Simulated Mean Measured Mean Bias RMS
1 0.867 0.000 33.108 31.925 1.184 0.681
2 0.936 0.000 32.516 31.320 1.196 0.670
3 0.960 0.000 30.327 29.412 0.915 0.515
4 0.948 0.000 29.355 27.509 1.846 0.654
5 0.987 0.000 23.264 23.477 -0.213 0.698
6 0.973 0.000 19.330 19.888 -0.558 0.188
7 0.974 0.000 12.988 14.150 -1.162 1.216
8 0.950 0.000 11.975 12.066 -0.091 2.455
9 0.963 0.000 8.595 9.513 -0.918 1.853
10 0.945 0.000 7.318 8.070 -0.753 1.147
11 0.919 0.000 4.737 5.289 -0.552 3.249
Appendix I, Table 2: Calculated results of the Temperature Quality index for each measuring station over the
course of 2010 vs. corresponding simulated values. See following pages for plotted data sets.
TEMPORAL QUALITY INDEX - TEMPERATURE

Station Correlation P-value Simulated Mean Measured Mean Bias RMS
1 0.910 0.000 24.480 23.767 0.713 2.228
2 0.935 0.000 24.465 23.604 0.861 2.208
3 0.918 0.000 24.129 23.751 0.378 1.742
4 0.933 0.000 24.153 23.681 0.472 1.494
5 0.936 0.000 24.127 23.868 0.259 0.959
6 0.943 0.000 23.773 23.597 0.176 0.098
7 0.962 0.000 23.420 23.641 -0.221 1.280
8 0.953 0.000 23.207 23.548 -0.340 0.952
9 0.959 0.000 23.161 23.375 -0.214 1.498
10 0.940 0.000 23.124 23.398 -0.273 1.940
11 0.885 0.000 23.004 23.138 -0.134 2.540
162
Appendix I, Figure 1: Temporal validation plots for Temperature (left) and Salinity (right) throughout the
course of 2010 for measuring stations 1 to 3. RM-Dist is station distance to the river mouth in metres.
163
164
165
166
SPATIAL VALIDATION OF SALINITY AND TEMPERATURE ACROSS ALL SAMPLING DAYS
Appendix I, Table 3: Calculated results of the Salinity Quality index for each sampling day in relation to
distance from the river mouth vs. corresponding simulated values. See following pages for plotted data sets.
SPATIAL QUALITY INDEX - SALINITY

Sample Day Correlation P-value Simulated mean Measured mean Bias RMS
1 0.999 0.000 26.182 27.394 -1.211 0.681
2 0.996 0.000 26.715 28.601 -1.886 0.670
3 0.827 0.002 8.666 2.383 6.283 0.515
4 0.990 0.000 20.043 19.363 0.680 0.654
5 0.994 0.000 14.022 11.077 2.945 0.698
6 0.988 0.000 12.209 10.960 1.248 0.188
7 0.993 0.000 18.982 17.402 1.580 1.216
8 0.993 0.000 20.041 21.127 -1.086 2.455
9 0.990 0.000 23.698 26.364 -2.666 1.853
10 0.914 0.000 8.026 1.666 6.360 1.147
11 0.984 0.000 13.878 12.895 0.982 3.249
12 0.992 0.000 22.314 22.969 -0.655 2.515
13 0.992 0.000 21.671 21.208 0.462 2.294
14 0.996 0.000 20.092 23.465 -3.373 1.910
15 0.977 0.000 27.116 28.503 -1.387 1.170
16 0.983 0.000 24.805 25.450 -0.645 2.445
17 0.990 0.000 22.883 23.732 -0.849 1.321
18 0.992 0.000 27.352 28.334 -0.982 0.554
19 0.979 0.000 31.516 31.902 -0.386 0.131
20 0.956 0.000 32.212 31.881 0.331 2.616
21 0.997 0.000 26.689 27.299 -0.610 2.208
22 0.996 0.000 16.546 16.980 -0.434 3.585
23 0.993 0.000 14.738 16.401 -1.663 20.225
24 0.992 0.000 22.369 23.788 -1.418 15.899
25 0.995 0.000 18.887 20.539 -1.651 6.641
26 0.996 0.000 17.333 18.190 -0.858 12.809
27 0.997 0.000 4.906 3.392 1.513 2.677
28 0.989 0.000 16.137 15.900 0.237 2.259
29 0.991 0.000 8.569 7.205 1.364 2.025
30 0.980 0.000 13.717 13.503 0.214 5.808
167
Appendix I, Table 4: Calculated results of the Temperature Quality index for each sampling day in relation to
distance from the river mouth vs. corresponding simulated values. See following pages for plotted data sets.
SPATIAL QUALITY INDEX - TEMPERATURE

Sample Day Correlation P-value Simulated Mean Measured Mean Bias RMS
1 0.950 0.000 29.439 29.409 0.029 2.228
2 0.743 0.009 27.790 26.439 1.351 2.208
3 0.908 0.000 25.046 25.355 -0.309 1.742
4 0.644 0.032 27.833 28.880 -1.048 1.494
5 0.910 0.000 26.360 25.430 0.930 0.959
6 0.824 0.002 24.054 23.567 0.487 0.098
7 0.841 0.001 26.056 26.196 -0.141 1.280
8 0.765 0.006 25.648 26.509 -0.861 0.952
9 0.550 0.080 24.411 24.328 0.083 1.498
10 0.881 0.000 20.920 19.966 0.954 1.940
11 0.927 0.000 21.441 21.991 -0.550 2.540
12 0.897 0.000 21.787 21.028 0.758 1.334
13 0.671 0.024 19.802 20.367 -0.565 1.280
14 0.409 0.212 18.507 20.018 -1.511 1.347
15 0.807 0.003 19.839 19.885 -0.045 1.470
16 0.932 0.000 20.091 19.525 0.567 1.154
17 0.833 0.001 20.803 20.039 0.763 1.352
18 0.579 0.062 23.053 22.645 0.408 2.104
19 -0.333 0.317 22.261 21.146 1.114 1.773
20 0.272 0.419 22.197 21.275 0.922 1.776
21 0.862 0.001 23.056 22.685 0.371 1.309
22 0.768 0.006 21.551 21.918 -0.367 0.038
23 0.705 0.015 24.582 25.079 -0.497 0.128
24 0.811 0.002 23.928 23.025 0.903 0.190
25 0.952 0.000 26.539 25.867 0.672 0.002
26 0.850 0.001 26.745 26.092 0.653 0.162
27 0.700 0.017 21.906 23.298 -1.392 0.554
28 0.855 0.001 24.563 23.522 1.041 0.780
29 0.636 0.035 24.059 24.030 0.029 1.031
30 0.875 0.000 27.671 27.850 -0.179 0.502
168
Appendix I, Figure 5: Spatial validation plots for Temperature (left) and Salinity (right) on sampling days 1
to 3 with each Station plotted in relation to river mouth distance
169
to 6 with each Station plotted in relation to river mouth distance.
170
to 9 with each Station plotted in relation to river mouth distance.
171
Appendix I, Figure 8: Spatial validation plots for Temperature (left) and Salinity (right) on sampling days
10 to 12 with each Station plotted in relation to river mouth distance.
172
173
174
175
176
177
178
DVD APPENDIX
Please find additional result appendices (II-VI) on the attached DVD, along with videos of observed
C. leucas movement and selected ABM and HD-model videos. Additional videos of model outputs
are available upon request.
179

Mortensen - 2012 Thesis

Uploaded by

Copyright:

Available Formats

Mortensen - 2012 Thesis

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Mortensen - 2012 Thesis

Uploaded by

Copyright:

Available Formats

Agent-Based Modelling of Short-Term Juvenile Bull Shark

Movement in a Semi-Enclosed Gold Coast Estuary

Griffith Research Online

Jonas Brandi Mortensen

B.Sc. (University of Copenhagen)

Science, Environment, Engineering and Technology

Submitted in fulfilment of the requirements of the degree of

Jonas Brandi Mortensen

14th December 2011

Statement of originality ............................................................................................................. II

Chapter I: General introduction and research aims ............................................................. 1

Chapter II: Water quality assessment of Tallebudgera creek ............................................. 6

Chapter III: Hydrodynamic modelling of Tallebudgera Creek ........................................ 26

Chapter V: Agent-based modelling of a juvenile bull shark on a short spatio-temporal

Chapter VI: General discussion and future directions ..................................................... 150

1.2 RESEARCH QUESTIONS

1. Do spatio-temporal changes in water quality parameters affect the short-term

3. Finally, can this behaviour be recreated as an emergent behaviour through an

WATER QUALITY ASSESSMENT OF TALLEBUDGERA CREEK

2.2.1 SAMPLING DESIGN

2.2.8 SONDE CALIBRATION & MEASUREMENT TECHNIQUES

2.2.9 DATA ANALYSIS

2.3.3 DISSOLVED OXYGEN

3.1.1 AIMS & PURPOSE

3.2.1 EQUATIONAL FRAMEWORK

3.2.2 TRANSPORT EQUATIONS

3.2.3 HEAT EXCHANGE

Where is the air density , is the specific heat

3.2.5 INPUT PARAMETERS

3.2.5.2 FRESHWATER DISCHARGE

3.2.5.3 METEOROLOGICAL DATA

3.2.5.6 WAVE RADIATION STRESS

3.2.5.7 INITIAL CONDITIONS

3.2.6 CALIBRATION PARAMETERS

Shallow water equations Value Unit

3.2.7 VALIDATION DATA

3.2.7.1 HISTORICAL DATA

3.2.7.2 ADCP DATA

3.2.7.3 TEMPERATURE/SALINITY DATA

3.2.8 CALIBRATION & DATA ANALYSIS

3.3.1 SURFACE ELEVATION VALIDATION

3.3.2 FLOW VELOCITY VALIDATION

3.3.3 SALINITY VALIDATION

TABLE 3.3.1: TEMPORAL QUALITY INDEX - SALINITY

TABLE 3.3.2: SPATIAL QUALITY INDEX - SALINITY

3.3.4 TEMPERATURE VALIDATION

TEMPORAL QUALITY INDEX - TEMPERATURE

SPATIAL QUALITY INDEX - TEMPERATURE

4.2.1 CATCHING AND TAGGING

4.2.2 ACOUSTIC TELEMETRY AND TRACKING METHODS

4.2.3 HYDROPHONE CALIBRATION AND SHARK POSITION

The individual calibrations revealed a clear difference in hydrophone performance in relation

4.2.4 MIKE21FM DATA

4.2.5 DATA ANALYSIS

A local nearest-neighbour convex-hull construction of home ranges was calculated on the

Shark name: FN1 JM1 MN1

TRACK I: 28/01 – 2010 (14:45 to 20:35)

4.3.3 TRACK III

TRACK IV: 06/03 – 2010 (12:03 to 14:58)

4.3.7 COLLECTIVE RESULTS