pelagic stereo-bruvs: development and implementation of a ...€¦ · calibrated to scientific...
TRANSCRIPT
Pelagic stereo-BRUVs: Development and
implementation of a fishery-independent technique
to study pelagic fish assemblages
Julia Santana-Garcon
BSc. Marine Biology (James Cook University, Australia)
MASc. Conservation Biology (James Cook University, Australia)
This thesis is presented for the degree of
Doctor of Philosophy at the University of Western Australia
The UWA Oceans Institute and
School of Plant Biology
Faculty of Natural and Agricultural Sciences
2015
ii
iii
Abstract
Pelagic fish are ecologically important to marine ecosystems and are a highly
valuable resource, yet little is known about the population status and ecology of
many species. Research on the biology and ecology of pelagic fish largely relies on
fishery-dependent data from commercial and recreational fisheries. However, the use
of fishery-dependent data alone can lead to sampling biases due to gear selectivity
and heterogeneous fishing effort. Alternatively, fishery-independent surveys can
adopt more robust sampling designs, but often employ the same fishing gear and as
such, catchability and size-selectivity biases remain. Emerging technologies like
remote sensing, acoustic cameras and video-techniques are providing new non-
destructive options for cost-effective ecological sampling in vast and remote
environments such as the open ocean.
Baited Remote Underwater Video systems (BRUVs) have proven to be a robust
fishery-independent method to obtain estimates of biodiversity, relative abundance,
behaviour and, when stereo-cameras are used, size and biomass of a range of marine
species. Stereo-BRUVs had not been tested in the pelagic environment, but evidence
suggested that they could overcome some of the difficulties of sampling in the water
column. In this thesis, I develop and validate the use of pelagic stereo-BRUVs as a
non-destructive fishery-independent technique to study highly mobile species in the
pelagic environment, and I examine the strengths and limitations of this method in
order to optimise their implementation in future studies.
Pelagic stereo-BRUVs proved to be an effective tool to sample fish assemblages in
the water column at depths ranging between 5 and 120 metres, in coastal waters and
at exposed offshore sites along the coast of Western Australia. Differences in the
vertical distribution of fish assemblages were detected (Chapter 2), which confirms
the potential of this method to study the vertical distribution and diel migration
patterns of both pelagic and demersal fishes. Measures of univariate and multivariate
precision were used to optimise the sampling regime of pelagic stereo-BRUVs
(Chapters 2 and 3). A sample unit size (soak time) of 120 minutes and a sample size
of at least 8 replicates per treatment are recommended to study pelagic fish
assemblages in tropical or warm-temperate waters. In order to account for the spatial
iv
and temporal variability of pelagic systems and to facilitate future comparisons
across studies using this method, we encourage maximizing replication given the
resources available while standardizing the soak time. Pelagic stereo-BRUVs were
calibrated to scientific longline surveys, a standard technique used to sample highly
mobile species like sharks (Chapter 4). The proportion of sharks sampled by both
methods was comparable across a latitudinal gradient sampled. The estimates of
relative abundance (catch per unit of effort) obtained from each camera system were
equivalent to those from 5 to 30 longline hooks. Moreover, pelagic stereo-BRUVs
were used to assess the effects of a spatial fishing closure on highly mobile fish
species (Chapter 5). The spatial area closure studied was found to have no significant
effect on the species composition or the relative abundance of pelagic fish. These
findings suggest that small spatial area closures may not be adequate for the
conservation and management of highly mobile fish, even if they have proved
effective for reef-associated species. Highly mobile pelagic fish species, such as
tunas, mackerel and some shark species proved difficult to measure, as these species
were observed furthest from the camera systems. Thus I discuss methods to increase
the attraction rate of pelagic fish to the stereo-cameras to enhance the performance of
pelagic stereo-BRUVs in future studies. Finally, I explored the potential of stereo-
video to make behavioural observations in the pelagic environment (Chapter 6). The
first in situ behavioural observations of a presettlement schooling priacanthid were
recorded using pelagic stereo-BRUVs. Additional information on the fish length,
mean swimming speed and spacing among individuals were also possible given the
use of stereo-video systems. This work added to the limited body of literature on the
schooling behaviour of the early pelagic stages of demersal fishes.
In conclusion, I demonstrate that pelagic stereo-BRUVs are an effective, non-
destructive and fishery-independent sampling approach to study pelagic ecosystems.
This method can provide data on the species composition, behaviour, relative
abundance and size distribution of highly mobile fish assemblages at broad spatial
and temporal scales. I also demonstrate that pelagic stereo-BRUVs can be calibrated
to standard techniques, are suitable for studies on rare or threatened species and can
be implemented in areas closed to fishing. This thesis critically assesses the use of
pelagic stereo-BRUVs and collates all the information needed to implement, adopt
and develop this method in future studies.
v
Publications from this Thesis
Articles published
Chapter 2
Santana-Garcon, J., Newman, S.J., Harvey, E.S., 2014. Development and
validation of a mid-water baited stereo-video technique for investigating
pelagic fish assemblages. Journal of Experimental Marine Biology and
Ecology 452, 82-90.
Author contribution: JSG designed the experiment, collected and analysed the data,
and wrote the manuscript. ESH and SJN supervised the study and reviewed the
manuscript.
Chapter 4
Santana-Garcon, J., Braccini, M., Langlois, T.J., Newman, S.J., McAuley, R.B.,
Harvey, E.S., 2014. Calibration of pelagic stereo-BRUVs and scientific
longline surveys for sampling sharks. Methods in Ecology and Evolution, 5.
824–833. DOI: 10.1111/2041-1210X.12217.
Author contribution: JSG collected data, conducted statistical analyses, and wrote
the manuscript. MB and TJL assisted in some components of data analysis. ESH,
SJN and RBM supervised the study and provided logistical support for the field
expedition to take place. All co-authors reviewed the manuscript and provided
comments.
Chapter 5
Santana-Garcon, J., Newman, S.J., Langlois, T.J., Harvey, E.S., 2014. Effects of a
spatial closure on highly mobile fish species: an assessment using pelagic
stereo-BRUVs. Journal of Experimental Marine Biology and Ecology, 460,
153-161. DOI: 10.1016/j.jembe.2014.07.003.
vi
Author contribution: JSG collected data, conducted statistical analyses, and wrote
the manuscript. ESH, SJN supervised the study and facilitated the field expedition.
All co-authors reviewed the manuscript and provided comments.
Chapter 6
Santana-Garcon, J., Leis, J.M., Newman, S.J., Harvey, E.S., 2014. Presettlement
schooling behaviour of a priacanthid, the Purplespotted Bigeye Priacanthus
tayenus (Priacanthidae: Teleostei). Environmental Biology of Fishes 97(3),
277-283.
Author contribution: JSG designed the experiment, collected and analysed the data,
and wrote the manuscript. JML provided advice on the ecology of presettlement fish
and species identification. ESH and SJN supervised the study and all co-authors
reviewed the manuscript.
Appendix 1
Anderson, M.J., Santana-Garcon, J., 2015. Measures of precision for dissimilarity-
based multivariate analysis of ecological communities. Ecology Letters 18(1), 66-73.
Author contribution: MJA developed the statistical methods presented and wrote the
final manuscript. JSG provided data, wrote early drafts and contributed in final
revisions of the manuscript. Some of the work presented in this article that is
relevant to this thesis, has been condensed and presented in chapter 3.
vii
Acknowledgements
First and foremost, I would like to thank my supervisors Euan Harvey, Stephen
Newman, Tim Langlois and Gary Kendrick for their advice and support. I feel I had
the best guidance possible throughout my thesis. I am also thankful to all my co-
authors of the publications that have resulted from the chapters presented here. Their
input and manuscript reviews were much appreciated and improved my work
substantially.
Funding and logistical support throughout the project were provided by the School
of Plant Biology (University of Western Australia) and the Department of Fisheries
(Government of Western Australia). I was supported by a Scholarship for
International Research Fees, a University Postgraduate Award (International) from
the University of Western Australia, and a UWA Safety-Net Top-Up Scholarship.
Work presented here was undertaken under the approval of UWA Animal Ethics
(RA/3/100/1035). And fieldwork in Ningaloo Marine Park was approved within the
Department of Environment and Conservation permit SF008486.
I would like to acknowledge the contribution of many colleagues at the University of
Western Australia and Department of Fisheries WA that helped with fieldwork,
logistics and advice. I am grateful to the experienced skippers P. and T. Pittorini for
their assistance during the design of the deployment method for pelagic stereo-
BRUVs. I am also thankful to K. Cure, J. Goetze, T. Bond and S. Bennett that
reviewed early drafts of some chapters, and to H. Twose for her assistance with the
illustration of the camera system.
Many thanks to the Fish Ecology Group (a.k.a. ‘Botany-Annex-II team’), the
working environment and friendship made us stronger and I could not have asked for
a better team (including the best officemate) to share the experience of my PhD.
Finally, special thanks to Scott Bennett. This thesis would not have started and,
definitely, could not have been completed without his constant inspiration,
encouragement, support and many reviews. Thanks to all my friends and family for
their support and for sharing this amazing journey with me.
ix
Table of Contents
Abstract .............................................................................................................. iii
Publications from this Thesis................................................................................ v
Articles published....................................................................................................... v
Acknowledgements ........................................................................................... vii
Table of Contents ............................................................................................... ix
List of Figures ................................................................................................... xiii
List of Tables ..................................................................................................... xvii
CHAPTER 1 - General introduction ...................................................................... 19
1.1 Background and rationale ................................................................................... 19
1.2 Specific aims ....................................................................................................... 22
1.3 Study area .......................................................................................................... 26
CHAPTER 2 - Development and validation of a mid-water baited stereo-video
technique for investigating pelagic fish assemblages .......................................... 29
2.1 Abstract ............................................................................................................. 29
2.2 Introduction ....................................................................................................... 29
2.3 Methods............................................................................................................. 34
2.3.1 Study area............................................................................................................. 34
2.3.2 Sampling technique .............................................................................................. 34
2.3.3 Experimental design ............................................................................................. 34
2.3.4 Image analysis ...................................................................................................... 35
2.3.5 Data analysis ......................................................................................................... 36
2.4 Results ............................................................................................................... 37
2.5 Discussion .......................................................................................................... 43
CHAPTER 3 - Multivariate pseudo standard error as a measure of precision:
optimisation for sampling pelagic fish assemblages ............................................ 49
3.1 Abstract ............................................................................................................. 49
3.2 Introduction ....................................................................................................... 50
x
3.3 Methods ............................................................................................................ 52
3.3.1 Multivariate pseudo-Standard Error .................................................................... 53
3.3.2 Optimisation of sampling effort for multivariate analyses .................................. 56
3.4 Results ............................................................................................................... 56
3.5 Discussion .......................................................................................................... 57
CHAPTER 4 - Calibration of pelagic stereo-BRUVs and scientific longline surveys for
sampling sharks ................................................................................................. 61
4.1 Abstract ............................................................................................................. 61
4.2 Introduction ....................................................................................................... 62
4.3 Methods ............................................................................................................ 64
4.3.1 Sampling technique .............................................................................................. 64
4.3.2 Video analysis ....................................................................................................... 67
4.3.3 Statistical analysis ................................................................................................. 68
4.4 Results ............................................................................................................... 70
4.5 Discussion .......................................................................................................... 74
CHAPTER 5 - Effects of a spatial closure on highly mobile fish species: an
assessment using pelagic stereo-BRUVs ............................................................. 81
5.1 Abstract ............................................................................................................. 81
5.2 Introduction ....................................................................................................... 81
5.3 Methods ............................................................................................................ 85
5.3.1 Study area............................................................................................................. 85
5.3.2 Sampling technique .............................................................................................. 85
5.3.3 Experimental design ............................................................................................. 86
5.3.4 Image analysis ...................................................................................................... 87
5.3.5 Data analysis ......................................................................................................... 88
5.4 Results ............................................................................................................... 89
5.5 Discussion .......................................................................................................... 95
CHAPTER 6 - Presettlement schooling behaviour of a priacanthid, the
purplespotted bigeye Priacanthus tayenus (Priacanthidae: Teleostei) .............. 101
6.1 Abstract ........................................................................................................... 101
6.2 Introduction ..................................................................................................... 101
6.3 Material and methods ...................................................................................... 103
6.4 Results ............................................................................................................. 105
xi
6.5 Discussion ........................................................................................................ 107
CHAPTER 7 - General discussion ....................................................................... 111
Literature cited ................................................................................................ 119
Appendix 1 - Measures of precision for dissimilarity-based multivariate analysis of
ecological communities ................................................................................... 133
xiii
List of Figures
Figure 1.1 Classification of marine environments. Pelagic habitat is defined here as
the entire ocean water column, including coastal and oceanic waters. .............. 19
Figure 1.2 Flow diagram outlining the background, rationale and general structure
of this thesis. ...................................................................................................... 23
Figure 1.3 Study area off the coast of Western Australia. Colour-coded panels and
markers indicate the specific study areas for each chapter. ............................... 27
Figure 2.1 Map of the study area in Ningaloo Marine Park, Western Australia. Sites
are in Tantabiddi (TN1, TN2) and Coral Bay (CB1, CB2). ............................... 33
Figure 2.2 Deployment method and frame details of pelagic stereo-BRUVs. .......... 35
Figure 2.3 Principal Coordinates Analysis plot showing differences in fish
assemblages sampled at two depth treatments (Depth 1, surface and Depth 2,
deep) using pelagic stereo-BRUVs. Different symbols show the sites sampled in
Tantabiddi (TN1, TN2) and Coral Bay (CB1, CB2) in Ningaloo Marine Park,
Western Australia. .............................................................................................. 40
Figure 2.4 Mean relative abundance (MaxN) of fish taxa sampled at two different
deployment depths using pelagic stereo-BRUVs in Ningaloo Marine Park:
Coral Bay and Tantabiddi. The vertical axes are shown in logarithmic scale and
error bars represent standard error. Some species are shown grouped at a higher
taxonomic level and species that had only one individual recorded are excluded.
............................................................................................................................ 40
Figure 2.5 Species accumulation curve and relative abundance accumulation curve
(B) in 15 minute time intervals for soak times up to 3 hours. Error bars
represent standard error. ..................................................................................... 41
Figure 2.6 Mean precision and 95 % confidence intervals for (A) species richness
and (B) relative abundance data using pelagic stereo-BRUVs with three
different soak times. Samples simulated using bootstrapping procedures with
replacement. Lower values of precision indicate more precise data. ................. 42
xiv
Figure 2.7 Estimate of the cost, in time, for sampling fish assemblages using
different soak times with pelagic stereo-BRUVs. Cost includes the time for
fieldwork with 5 camera systems and video analysis. ....................................... 43
Figure 3.1 (a) Study area in Ningaloo Marine Park (Western Australia) and (b)
design of pelagic stereo-BRUVs for sampling mid-water fish assemblages. .... 53
Figure 3.2 Multivariate pseudo standard error as a function of sample size on the
basis of Bray-Curtis dissimilarities calculated on square-root transformed fish
abundance data sampled using pelagic stereo-BRUVs for a soak time of 60
minutes only, showing results (means with 2.5 and 97.5 percentiles as error
bars) from 10,000 resamples obtained using either the permutation or bootstrap
approach. ............................................................................................................ 57
Figure 3.3 Multivariate pseudo standard error as a function of sample size on the
basis of Bray-Curtis dissimilarities calculated on square-root transformed fish
abundance data sampled using pelagic stereo-BRUVs for all three soak times
(60, 120 and 180 minutes), using the double-resampling method, with
permutation-based means and bias-adjusted bootstrap-based error bars (with
10,000 resamples for each). ............................................................................... 57
Figure 4.1 Location of study sites along the coast of Western Australia.................. 65
Figure 4.2 Deployment design of the (a) scientific longline shots and (b) pelagic
stereo-BRUVs used to sample requiem sharks. ................................................. 66
Figure 4.3 Mean relative abundance of fish species sampled using scientific longline
surveys and pelagic stereo-BRUVs. Catch per unit of effort (CPUE) is shown as
catch per hour for 10 hooks (CPUE10) in longline samples and as MaxN per
hour in stereo-BRUVs. a Benthic sharks were caught in longlines due to the
deployment design setting of a proportion of hooks near the bottom................ 71
Figure 4.4 Mean species proportion of target species sampled using scientific
longline surveys and pelagic stereo-BRUVs. P-values show non-significant
differences between sampling methods. ............................................................ 71
xv
Figure 4.5 Relative abundance (CPUE) of target species along a latitudinal gradient
(32° - 24° S) sampled using scientific longline surveys and pelagic stereo-
BRUVs. Trendlines illustrate the ANCOVA result. .......................................... 72
Figure 4.6 Equivalence of sampling effort required to estimate relative abundance of
requiem sharks. Plot shows mean p-values from one-way PERMANOVAs
testing the differences between pelagic stereo-BRUVs (MaxN per hour) and
scientific longline surveys (catch*h-1 *hook-1) across different levels of
sampling effort (number of hooks). Values in the shaded area (P < 0.05) are
statistically significant. Grey line is shown for reference as the general pooling
cut-off (P > 0.25). ............................................................................................... 73
Figure 4.7 Comparison of kernel density estimate (KDE) probability density
functions for the length distributions of requiem shark species caught by pelagic
stereo-BRUVs and scientific longline surveys. Grey bands represent one
standard error either side of the null model of no difference between the KDEs
for each method. ................................................................................................. 76
Figure 5.1 Location of (A) the Houtman Abrolhos Islands and (B) the study sites
inside and outside the spatial area closure at Easter Group. The area shaded in
light grey shows the Reef Observation Area closed to fishing since 1994. ....... 85
Figure 5.2 Deployment design of pelagic stereo-BRUVs to sample fish assemblages
in the water column. ........................................................................................... 86
Figure 5.3 Total abundance (sum of MaxN) of fish taxa sampled using pelagic
stereo-BRUVs in areas open and closed to fishing at the Houtman Abrolhos
Islands. ............................................................................................................... 90
Figure 5.4. The mean species richness per site (A) and the mean relative abundance
per site of the main target groups (B) in areas open and closed to fishing at the
Houtman Abrolhos Islands. Error bars show standard error. ............................. 92
Figure 5.5 The mean range, distance of the individuals to the stereo-camera system,
for fish genus recorded using pelagic stereo-BRUVs. Grey silhouettes represent
demersal species and black silhouettes are pelagic and highly mobile fish. The
xvi
shaded area shows the field of view of the cameras where fish length can be
measured. ........................................................................................................... 94
Figure 5.6 Comparison of kernel density estimate (KDE) probability density
functions for the length distributions of pink snapper Chrysophrys auratus in
areas open and closed to fishing. Grey bands represent one standard error either
side of the null model of no difference between the KDEs for each method. ... 94
Figure 6.1 Deployment method (a) and details (b) of the pelagic stereo Baited
Remote Underwater Video system (stereo-BRUVs). ...................................... 104
Figure 6.2 Still photos of presettlement Priacanthus tayenus (mean total length 24.1
mm) captured from video recorded in Coral Bay, Western Australia. ............ 104
Figure 6.3 Quantitative parameters for three groups of presettlement Priacanthus
tayenus recorded using pelagic stereo cameras in Coral Bay, Western Australia
(group A, 8 individuals; group B, 9 individuals and group AB, 17 individuals).
(a) Mean Total Length (TL; mm); (b) mean Nearest Neighbour Distance (NND;
mm) and (c) mean NND divided by the average TL. Error bars represent
standard error. .................................................................................................. 107
xvii
List of Tables
Table 2.1 Taxonomic group and stage of specimens recorded using pelagic stereo-
BRUVs in Ningaloo Marine Park. Stage of individuals was identified as either
adult (AD) or juvenile (J). .................................................................................. 38
Table 2.2 PERMANOVA testing for the effects of deployment depth on the
structure of fish assemblages sampled using pelagic stereo-BRUVs. Results are
based on Bray-Curtis similarity matrices using dispersion weighted and square-
root transformed data. All tests were undertaken using 9,999 permutations
under the reduced model; when less than 100 unique permutations were
obtained, Monte Carlo P values were used and are shown in italics. Values
highlighted in bold are significant...................................................................... 39
Table 4.1 Summary of ANCOVA tests with method as factor and latitude as
covariate. Abundance data was collected with pelagic stereo-BRUVs and
scientific longline surveys along an 8-degree latitudinal gradient. P-values in
bold are statistically significant. ......................................................................... 73
Table 4.2 Precision estimates for target species sampled using pelagic stereo-
BRUVs and scientific longline surveys. Precision (p) was estimated as a ratio of
the standard error and the mean abundance per deployment. Note that lower
values of p indicate better precision. .................................................................. 74
Table 4.3 Summary of the lengths of target species measured on pelagic stereo-
BRUVs and caught on scientific longline surveys. Maximum (Max), minimum
(Min) and mean fork length (FL) are shown in millimetres. ............................. 75
Table 5.1 PERMANOVA based on log 5 Modified Gower dissimilarities to test the
effects of spatial area closures (Status) on the composition and relative
abundance of fish assemblages sampled using pelagic stereo-BRUVs. ............ 90
Table 5.2 Frequency of occurrence of species sampled using pelagic stereo-BRUVs
inside (Closed) and outside (Open) a spatial fishing closure at the Houtman
Abrolhos Islands. ............................................................................................... 91
xviii
Table 5.3 Summary of univariate PERMANOVA tests for species richness and
relative abundance of the main target fish groups sampled using pelagic stereo-
BRUVs. Analyses were based on Euclidean distance dissimilarities of the raw
data to test for differences in Status (areas open and closed to fishing) and Sites.
Fields in bold-faced show significant P-values. ................................................ 92
Table 5.4 Summary of length and range measurements at genus level combining
areas open and closed to fishing. The mean length (cm) represents fork length
and the mean range (m) is the average distance of individuals to the stereo-
camera system. The number of individuals measured [N] is shown in brackets
and (*) shows the genus of demersal target species. ......................................... 93
19
CHAPTER 1 - General introduction
1.1 Background and rationale
Importance of the pelagic environment
The pelagic marine environment (i.e. the entire ocean water column; Figure 1.1)
constitutes 99% of the biosphere volume, accounts for half of the photosynthesis and
supports, directly or indirectly, almost all marine life (Angel, 1993; Game et al.,
2009). It also has a direct impact on the pace and extent of climate change; oceans
play a major role in the global carbon cycle and absorb most of the carbon dioxide
entering the atmosphere through human activities (Field, 2002; Hays et al., 2005;
Siegenthaler and Sarmiento, 1993). Moreover, pelagic ecosystems provide essential
resources to humans and account for about 80% of the global fisheries catch (Game
et al., 2009; Grantham et al., 2011; Halpern et al., 2008; Sumaila et al., 2007; Worm
et al., 2006).
Figure 1.1 Classification of marine environments. Pelagic habitat is defined here as the entire ocean water column, including coastal and oceanic waters.
Pelagic ecosystems are affected by multiple threats including pollution, climate
change and overexploitation (Bianchi and Morri, 2000; Halpern et al., 2008;
Robison, 2009). Global pelagic fisheries are declining (Pauly et al., 1998), and
20
climate change is altering the global ocean circulation patterns and mixing,
interfering with primary and secondary production, and disturbing the stability of
pelagic ecosystems (Cheung et al., 2013a; Cheung et al., 2013b; Pinsky et al., 2013).
These processes have resulted in a loss of biodiversity and the depletion of several
target species from selected areas of the world’s oceans (Lewison et al., 2004; Myers
and Worm, 2003; Myers et al., 2007; Pauly et al., 2002). Ocean biodiversity loss
affects the stability of the ecosystem, increases the risk of population collapse and
reduces its potential for recovery (Worm et al., 2006). The conservation of marine
biodiversity and improvement in the sustainability of fisheries are major
environmental and economic challenges (Pauly, 2009).
Lack of data in pelagic ecosystem research
Pelagic fish are ecologically important to marine ecosystems and are a highly
valuable resource, yet little is known about the population status and ecology of
many species (Bakun, 1996; Freon et al., 2005). Although our understanding of
individual species has improved; our knowledge of the species diversity, abundance
and patterns at the community level is still poor (Evans et al., 2011; Santos et al.,
2013a; Worm et al., 2005). Research on the biology and ecology of pelagic fish
largely relies on fishery-dependent data from commercial and recreational fisheries
(Myers and Worm, 2003; Ward and Myers, 2007). However, the use of fishery-
dependent data alone can lead to sampling biases due to gear selectivity and
heterogeneous fishing effort that can discriminate among species and habitats
(Bishop, 2006; Murphy and Jenkins, 2010; Thorson and Simpfendorfer, 2009).
Alternatively, fishery-independent surveys use more robust sampling designs, but
often employ the same commercial fishing gear (e.g. longlines, gillnets, trawls) and
as such, catchability and size-selectivity biases remain (Simpfendorfer et al., 2002;
Ward and Myers, 2005b). Similarly, it is often not possible to use extractive methods
to sample within areas that are closed to fishing or where protected, endangered and
threatened species might be caught (Assis et al., 2007; Murphy and Jenkins, 2010).
Thus many studies have highlighted the need to improve our capacity to monitor the
ecology and health of pelagic ecosystems using non-destructive and fishery-
independent techniques (Claudet et al., 2010; Heagney et al., 2007; Murphy and
Jenkins, 2010; Ward and Myers, 2005a).
21
Potential of emerging new technologies
The high cost of obtaining representative samples in the pelagic environment, which
comprises large areas with high spatial and temporal variability, has often inhibited
the use of fishery-independent methods (Bishop, 2006). However, emerging
technologies like remote sensing, acoustic cameras and video-techniques are
providing new options for cost-effective ecological sampling in vast and remote
environments such as the open ocean (Axenrot et al., 2004; Bertrand et al., 2002;
Murphy and Jenkins, 2010; Trenkel and Berger, 2013). Baited Remote Underwater
Video systems (BRUVs) have been demonstrated to be a robust fishery-independent
method to obtain estimates of biodiversity, relative abundance, behaviour and, when
stereo-cameras are used, size and biomass of a range of marine species (Harvey et
al., 2013; Harvey et al., 2012c; Mallet and Pelletier, 2014; White et al., 2013;
Zintzen et al., 2011).
Baited stereo-video methods
Baited video techniques are increasingly used for assessing fish community structure
in a variety of environments including deep water (Bailey et al., 2007; Zintzen et al.,
2012), estuaries (Gladstone et al., 2012) and both tropical and temperate reefs
(Harvey et al., 2012a; Harvey et al., 2012c; Langlois et al., 2010; Unsworth et al.,
2014). Stereo-BRUVs use bait to attract fish to the field of view of the cameras so
that individuals can be identified, counted and accurately measured (Cappo et al.,
2003; Dorman et al., 2012). Stereo-BRUVs have not been tested in the pelagic
environment, but evidence suggests that they could overcome some of the difficulties
of sampling in the water column. For instance, pelagic fish are often recorded during
benthic BRUV deployments (Cappo et al., 2004; Jones et al., 2003) and, single-
camera mid-water BRUVs, although unable to measure fish lengths, have been
successfully trialled to survey pelagic ecosystems (Heagney et al., 2007). However,
further research would be needed to expand the use of stereo-BRUVs to the pelagic
environment.
Research question
A fundamental question that needs to be addressed is ‘are pelagic stereo-BRUVs an
effective fishery-independent technique to study pelagic fish assemblages?’ And if
22
so, ‘what is the optimal sampling procedure in order to characterise and analyse
changes in these ecological communities?’ In this thesis, I develop and validate the
use of pelagic stereo-BRUVs as a non-destructive fishery-independent technique to
study highly mobile species in the pelagic environment, and I examine the strengths
and limitations of this method in order to optimise their implementation in future
studies (Figure 1.2).
The purpose of this research project is to develop and assess an alternative sampling
approach to overcome some of the difficulties associated with studying pelagic
ecosystems. The overarching goal of the thesis is to contribute to the limited body of
knowledge on the demographics, ecology and behaviour of fish assemblages that
inhabit the water column. This first section provides a broad overview of the
background and rationale of the project and, I present the specific aims that are
addressed in subsequent chapters. Each chapter is written as a stand-alone
manuscript to facilitate publication, thus chapters have their own specific
introduction sections and consequently may include some elements of the
background information presented here.
1.2 Specific aims
Design, development and validation of pelagic stereo-BRUVs
In chapter 2, I develop and describe a deployment method to use stereo-BRUVs in
the water column and test their ability to sample pelagic fish assemblages.
Characterising ecological communities requires key decisions regarding the method
and sampling effort needed to analyse the assemblage of interest (i.e. size and
number of sampling units; Andrew and Mapstone, 1987). There is considerable
variation in the soak time (time that cameras remain underwater) used across studies
using baited video techniques in different environments (Gladstone et al., 2012).
Studies on reef fish assemblages use soak times between 20 and 60 minutes (Stobart
et al., 2007; Watson et al., 2010; Willis and Babcock, 2000), and deep-water studies
deploy cameras from 1 to several hours (Bailey et al., 2007; Jones et al., 2003;
Zintzen et al., 2012). However, standardising the sample unit size to be used in the
pelagic environment will facilitate comparison across studies. In order to optimise
and standardise the sampling regime for future studies using pelagic stereo-BRUVs,
23
Figure 1.2 Flow diagram outlining the background, rationale and general structure of this thesis.
BA
CK
GR
OU
ND
AN
D R
AT
ION
ALE
Importance of marine
pelagic environment
Pelagic ecosystem
research is data poor
Potential of emerging new
technology for ecological
sampling
Baited video techniques
ingreasingly used in various
marine environments
Food sustainability
and other resources
Ecologically importantto marine ecosystems
Threads:
- climate change
- pollution
- overexploitation
Loss of biodiversity,
depletion of target
species
Need for monitoring and
ecosystem-based management
Research on pelagic species presents
difficulties for collection of accurate
survey data
Lack of data on:
- ecology and demographics
- population status
- vertical distribution
- behaviour
Knowledge relies mainly on
Fishery-dependent
techniques
Extractivefishery-independent
techniques
Need for implementation of
non-destructive fishery-independent
methods
Not suitable to sample
threatened species or in
areas closed to fishing
Sampling biases:
- gear selectivity
- targeted sampling effort
A combination of
methods would improve
data quality
Progress and cost-reduction
of video technology and
data storage
- standardised
- fishery-independent
- permanent record
Provide data on
species richness,
relative abundance
and behaviour
if stereo-video Also provides data
on fish length,
biomass and
swimming speed
Stereo-BRUVs have
not been tested in the
pelagic environment
Potential of pelagic
stereo-BRUVs to study
pelagic fish assemblages
e.g. deep water, temperate and
tropical reefs, estuaries
RE
SE
AR
CH
QU
ES
TIO
N
Are pelagic stereo-BRUVs an effective fishery-independent technique
to study pelagic fish assemblages?
SP
EC
IFIC
AIM
S
Design, development and
validation of pelagic stereo-BRUVs
Optimisation of sampling effort
for multivariate analyses
Calibration of pelagic stereo-BRUVs
to standard sampling techniques
Monitoring the effects of spatial
fishing closures on pelagic fishStudying in situ behaviour of fish
in the pelagic environment
CHAPTER 2 CHAPTER 3 CHAPTER 4 CHAPTER 5 CHAPTER 6
- Optimisation of sampling based
on cost and precision (univariate):
- soak time
- replication
- Sampling of vertical distribution
of pelagic fish in the water column
- Optimisation of sampling based
on a measure of multivariate
precision
- Description of pseudo-multivariate
Standard Error (MultSE) as a
measure of precision
- Comparison to scientific longline
surveys
- Effort equivalence
- Gear selectivity
- Monitoring of highly-mobile
species at broad spatial scales
- Explore effects of protection on
highly mobile species
- Assess pelagic stereo-BRUVs to
monitor spatial area closures
- Contribute to ongoing debate on
the use of spatial closures for the
management of pelagic species
- Report in situ behavioural
observations of presettlement
juvenile fish
- Provide quantitative data on
behaviour using stereo-video
(fish length, swimming speed and
schooling parameters)
- Pelagic fish are often
recorded on benthic BRUVs
- Single-camera BRUVs
have been tested in the
mid-water (Heagney et al. 2007)
Poor data quality:
- unreported catch
- poor species
identification for
low-value catch
24
I explored the effects of different soak times and replication on the precision and cost
of sampling.
This chapter also investigates whether differences in the vertical composition of fish
assemblages can be detected by deploying pelagic stereo-BRUVs at different depths
in mid-water pelagic environments. The physical boundaries of fish communities in
pelagic ecosystems are not well defined and further research is needed to understand
the vertical distribution and diel migrations of pelagic and demersal species in the
water column (Gray, 1997). Fish are distributed throughout the mid-water at
different depths and there is evidence that differences in the deployment depth of
sampling systems can affect the estimates of abundance and species richness
(Heagney et al., 2007; Ward and Myers, 2006). The work encapsulated in Chapter 2
has been peer-reviewed and published in the Journal of Experimental Marine
Biology and Ecology (Santana-Garcon et al., 2014a).
Optimisation of sampling effort for multivariate analyses
Chapter 3 expands the work presented on the optimisation of sampling based on the
precision of the sampling technique. Precision refers to the repeatability of
measurements; the degree of concordance among multiple estimates of a given
parameter (such as the mean) for the same population (Andrew and Mapstone, 1987;
Cochran and Cox, 1957). To date, there is no measure of multivariate precision that
can be used to assist researchers in decision-making processes regarding adequate
sampling requirements, this is especially relevant for studies using multi-species
count data that are gathered to characterise ecological communities. I describe the
use of a pseudo-multivariate standard error (MultSE) as a measure of multivariate
precision and I use estimates of MultSE to determine the appropriate sample unit
size (soak time) and replication for studies utilising pelagic stereo-BRUVs and using
dissimilarity-based multivariate analyses. This chapter includes concepts developed
as part of a larger study undertaken in collaboration with Prof. Marti J. Anderson
from Massey University (New Zealand) that has resulted in a manuscript currently in
review for publication in the journal Ecology Letters (Anderson and Santana-Garcon,
In review; Appendix 1).
25
Calibration of pelagic stereo-BRUVs to standard sampling
techniques
In chapter 4, I calibrate the sampling ability of pelagic stereo-BRUVs to a standard
sampling technique used to study highly mobile species in the pelagic environment.
In particular, I compare the catch composition, relative abundance and length
distribution of fish assemblages sampled using pelagic stereo-BRUVs and scientific
longline surveys. The study is focused on sampling sharks of the family
Carcharhinidae (requiem sharks) and calibrates the sampling effort required for each
technique to obtain equivalent samples of the target species. Research on shark
populations is largely based on fishery-dependent data (Myers and Worm, 2003) and
fishery-independent studies that use fishing-based techniques, like the commonly
used scientific longline surveys (Simpfendorfer et al., 2002). Technical advances are
opening new opportunities for non-destructive research evaluations of these highly
mobile species (White et al., 2013). However, in order to understand the potential of
novel techniques it is necessary to compare and calibrate these against traditional
methods. The work presented in Chapter 4 has been published in the journal
Methods in Ecology and Evolution (Santana-Garcon et al., 2014d).
Monitoring the effects of spatial closures on pelagic fish
In chapter 5, I explore the effects of protection on pelagic species and assess the
potential of pelagic stereo-BRUVs as a monitoring technique for spatial area
closures. There has been extensive research examining the response of demersal
species to spatial fishing closures in coastal waters (Babcock et al., 2010; Ballantine,
2014; Claudet et al., 2008). However, the methods used have often overlooked or not
specifically targeted the mid-water pelagic environment and, consequently, the
response of mobile pelagic species to spatial closures is not clear and is the subject
of ongoing debate (Claudet et al., 2010; Davies et al., 2012; Game et al., 2009;
Kaplan et al., 2010). Moreover, the implementation of large offshore spatial closures
is increasing around the world (Davies et al., 2012; Kaplan et al., 2014;
Notarbartolo-Di-Sciara et al., 2008; Sheppard, 2010), thus there is a need to improve
our capacity to monitor changes in these marine communities using non-destructive
fishery-independent methods is critical (Claudet et al., 2010). This chapter aims to
contribute to the development of monitoring techniques and also to the ongoing
26
debate on the use of spatial area closures for the conservation and management of
highly mobile species both in coastal waters and the high seas. Chapter 5 has been
peer-reviewed and published in the Journal of Experimental Marine Biology and
Ecology (Santana-Garcon et al., 2014b).
Studying in situ behaviour of fish in the pelagic environment
Chapter 6 presents the potential of utilising pelagic stereo-BRUVs to study the in
situ behaviour of fish in the pelagic environment. I report behavioural observations
recorded on video of presettlement juvenile fish and, provide quantitative data using
stereo-video on their length, swimming speed and schooling parameters. The biology
of the pelagic phase of demersal fishes has been extensively studied, but direct
observations and quantitative data on their behaviour in the pelagic environment is
limited and only available for a small number of species (Leis, 2010; Leis and
Carson-Ewart, 1998; Masuda, 2009). Stereo-video techniques offer unique
opportunities and the ability to study the ontogeny of fish behaviour in their natural
environment as they provide a permanent record, can be used at depths beyond the
limits of diver-based studies and can sample both during the day or at night using
lights (Harvey et al., 2012a). This chapter is the result of collaboration with Dr Jeff
Leis from the Australian Museum, he supervised and reviewed my work on juvenile
fish behaviour, and this has been published in the journal Environmental Biology of
Fishes (Santana-Garcon et al., 2014c).
1.3 Study area
The work presented in this thesis was undertaken in Australia, off the west coast of
Western Australia (WA; Figure 1.3). Initial trials of the deployment method for
pelagic stereo-BRUVs were carried out in Perth metropolitan waters allowing the
prototypes to be refined before the main field campaigns. Fieldwork for chapters 2, 3
and 6 took place in the Ningaloo Marine Park (23° 48’ - 21° 48’ S). Ningaloo Reef is
a fringing coral reef that stretches for approximately 270 km adjacent to the semi-
arid North-West Cape of WA. The sites were ~35 m deep and between 1 and 2 km
offshore from the reef slope. Data for chapter 4 was collected across ~950 km along
the coastline of WA (32° - 24° S) at sites 15 - 80 km offshore and at depths ranging
from 35 to 106 m. Finally, fieldwork for chapter 5 was undertaken at the Houtman
27
Abrolhos Islands, 60 km offshore from the mid-west coast of WA (28° 15’ - 29° 00’
S). The sites were located off the reef edge in the Easter Group and were between 30
and 35 m deep.
Figure 1.3 Study area off the coast of Western Australia. Colour-coded panels and markers indicate the specific study areas for each chapter.
29
CHAPTER 2 - Development and validation of a mid-water baited stereo-video technique for investigating pelagic fish assemblages
2.1 Abstract
Understanding the abundance, demographics and composition of pelagic fish communities
has historically relied on fisheries catch data or destructive fishery-independent methods.
Here, we test and validate the use of a pelagic stereo-Baited Remote Underwater Video
system (BRUVs) as a non-destructive, fishery-independent approach to study pelagic fish
assemblages. We investigated whether differences in the vertical composition of fish
assemblages could be detected with pelagic stereo-BRUVs by sampling at different depths
in the water column. The effects of soak time and replication on the precision and cost of
sampling were explored to allow for the optimization and standardization of future pelagic
stereo-BRUVs studies. Pelagic stereo-BRUVs effectively identified 43 fish taxa from 18
different families in the mid-water, 5 and 20 m below the surface, in the Ningaloo Marine
Park (Western Australia). The fish assemblages sampled at the two mid-water depths were
significantly different demonstrating that this method could be used to investigate the
vertical distribution and diel migration patterns of both pelagic and demersal fishes.
Precision estimates under different sampling regimes showed that a soak time of 120 min
and a sample size of at least 8 replicates per treatment would be optimal for sampling using
pelagic stereo-BRUVs in tropical or warm-temperate areas. In order to account for the
spatial and temporal variability of the system and to facilitate future comparisons across
studies using this method, we encourage maximizing replication given the resources
available while standardizing the soak time. Pelagic stereo-BRUVs may provide a useful,
non-destructive method to improve our understanding on the ecology and behavior of fishes
in pelagic ecosystems.
2.2 Introduction
Pelagic fish are ecologically important to marine ecosystems and are a highly
valuable resource, yet little is known about the population status and ecology of
many species (Bakun, 1996; Freon et al., 2005). Although our understanding of
individual species has improved, our knowledge of the species diversity, abundance
30
and patterns at the community level is still poor (Angel, 1993; Evans et al., 2011;
Santos et al., 2013b; Worm et al., 2005). Research on pelagic fish presents
difficulties for the collection of accurate survey data (Heagney et al., 2007) and often
relies exclusively on fisheries catch data (Myers and Worm, 2003; Ward and Myers,
2007).
The use of fishery-dependent data alone has many shortcomings, as it can lead to
sampling biases in the size and type of fish it targets. Similarly, it is often not
possible to sample destructively within areas that are closed to fishing, thus
impeding the assessment of the effects of protection on pelagic fish assemblages
(Murphy and Jenkins, 2010). For many pelagic species, the implementation of
fishery-independent surveys is inhibited by the high cost of obtaining representative
samples from a vast habitat which has high spatial and temporal variability (Bishop,
2006). However, many studies have highlighted the need to further develop fishery-
independent methods to assess the ecology and health of pelagic ecosystems
(Claudet et al., 2010; Heagney et al., 2007; Ward and Myers, 2005b). Murphy and
Jenkins (2010) reviewed current and emerging fishery-dependent and independent
observational methods used in marine spatial monitoring to obtain population and/or
habitat data in order to assess marine biodiversity and population trends. The review
highlights the potential of emerging technologies like remote sensing, acoustic
cameras and Baited Remote Underwater Video systems (BRUVs), and concludes
that a combination of methods would be the most effective way to reduce biases and
increase the quality of data.
The use of BRUVs has increased in recent years as it provides a standardized, non-
destructive and fishery-independent approach for estimating biodiversity indices and
relative abundance measures of a range of marine species (Harvey et al., 2007;
Langlois et al., 2010; Stobart et al., 2007; Stoner et al., 2008; Watson et al., 2010;
White et al., 2013; Willis and Babcock, 2000). This technique uses bait to attract
individuals into the field of view of a camera so that species can be identified and
individuals counted (Dorman et al., 2012). When stereo-camera pairs are used,
precise length and biomass estimates can be obtained (Cappo et al., 2006; Harvey
and Shortis, 1995; Harvey et al., 2010). The use of stereo-BRUVs to estimate
diversity, relative abundance and size structure of fish communities has been tested
and compared to other sampling techniques (Cappo et al., 2004; Ellis and Demartini,
31
1995; Harvey et al., 2012c; Langlois et al., 2010; Watson et al., 2010). Baited video
techniques have proven to be a robust method for assessing fish community structure
in deep water (Bailey et al., 2007; Zintzen et al., 2012), estuaries (Gladstone et al.,
2012) and tropical or temperate reefs (Langlois et al., 2010; White et al., 2013).
Despite the increase in fishery-independent techniques used to assess demersal fish
assemblages, the development and trial of such methodologies to survey pelagic
species remain limited. Recent evidence supports the use of stereo-video as a
possible tool for pelagic fish monitoring, with the potential to overcome some of the
difficulties associated with surveying pelagic ecosystems, if the technique can be
suitably adapted, developed and validated. Pelagic fish are often observed during
benthic BRUVs deployments (Cappo et al., 2004; Jones et al., 2003), and single-
camera mid-water BRUVs have been successfully trialed to survey pelagic
ecosystems (Heagney et al., 2007). Moreover, the use of pelagic (or mid-water)
stereo-camera pairs would provide accurate length and biomass measurements of
pelagic fish (Harvey et al., 2003; Santana-Garcon et al., 2013), as well as, the
opportunity to obtain behavioral data to better understand pelagic ecosystems
(Santana-Garcon et al., 2013).
One of the biggest challenges when implementing a sampling program for pelagic
fish assemblages is the definition of the pelagic community itself. Benthic
ecosystems are relatively easy to define by their horizontal distribution and very
abrupt boundaries in habitat (i.e. reef to sand) which provide structure to the fish
community (Habeeb et al., 2005). In pelagic ecosystems, these physical boundaries
are not well defined and there is also the vertical dimension to consider as pelagic
fish are distributed throughout the water column at different depths (Gray, 1997;
Holling, 1992). Differences in the deployment depth of sampling systems may
therefore affect the estimates of abundance and species richness in the pelagic
environment (Heagney et al., 2007; Ward and Myers, 2006). This parameter, which
potentially affects the sampling ability of BRUVs, is not well understood. Therefore,
pelagic stereo-BRUVs should allow for camera systems to be deployed and remain
at a predetermined depth using anchored or drifting systems. The deployment of
these systems at different mid-water depths could provide a powerful fishery-
independent technique to better understand pelagic fish assemblage composition, as
well as, the vertical distribution and diel migration patterns of both pelagic and
32
demersal fishes. A fundamental question that needs to be addressed for pelagic
stereo-BRUVs to be an effective monitoring and assessment methodology is ‘can
they detect differences in the vertical composition of pelagic fish assemblages?’
Another challenge with defining the structure of pelagic fish communities is
determining the spatial and temporal scales at which they need to be sampled in
order to capture the diversity and relative abundance of fishes living within them
(Habeeb et al., 2005). For BRUVs that remain stationary, this can be determined by
the time that cameras are left recording (soak time) in order to capture the majority
of species present in the area. Soak time is known to have a strong influence in the
estimates of abundance obtained from fishery-dependent techniques such as longline
operations (Ward et al., 2004). For sampling using baited video techniques, there is
considerable variation in the soak times used across studies in different environments
(Gladstone et al., 2012). Deep water BRUVs use soak times between one and several
hours (Bailey et al., 2007; Jones et al., 2003; Zintzen et al., 2011) and studies on reef
fish assemblages tend to deploy cameras for 20 to 60 min (Stobart et al., 2007;
Watson et al., 2010; Willis and Babcock, 2000). In order to optimize resources and
standardize the use of mid-water BRUVs in the pelagic environment, we explore the
performance of the method under various sampling regimes (i.e. soak times and
number of replicates) and evaluate the associated sampling costs.
Precision has been commonly used to optimize sampling effort in studies involving
univariate data (Bartsch et al., 1998; Downing and Downing, 1992; Pringle, 1984).
For example, Gladstone et al., (2012) assessed the precision of the species richness
and abundance estimates under different soak time regimes to optimize the sampling
effort of benthic BRUVs in estuarine environments. The precision of a sampling
technique refers to the repeatability of its measurements, the degree to which
repeated observations under unchanged conditions lead to the same result (Cochran
and Cox, 1957). Precision is an attribute of the sampling procedure and can be
assessed relatively easily from characteristics of the sample data (Andrew and
Mapstone, 1987). It is usually expressed numerically by measures of imprecision
like standard deviation, variance, coefficients of variation and most commonly, as a
ratio of the standard error (SE) and the mean (Andrew and Underwood, 1989;
Downing and Downing, 1992; Hellmann and Fowler, 1999). Given the resources
33
available, any method should use the sampling effort that yields the greatest
precision (Pringle, 1984).
The aims of this study are therefore to understand and validate the use of pelagic
stereo-BRUVs as an effective fishery-independent approach to study pelagic fish
assemblages. In particular, we describe the design and use of pelagic stereo-BRUVs
and present the results of a pilot study in the Ningaloo Marine Park (Western
Australia) to test the effects of deployment depth on the ability of this technique to
survey pelagic fish in the water column. Furthermore, the effects of soak time and
replication on the precision and cost of sampling are explored to allow for the
optimization and standardization of future studies. The advantages, limitations and
requirements for future development of this emerging sampling technique are also
discussed.
Figure 2.1 Map of the study area in Ningaloo Marine Park, Western Australia. Sites are in Tantabiddi Passage (TN1, TN2) and Coral Bay (CB1, CB2).
34
2.3 Methods
2.3.1 Study area
This study was conducted in March 2012 at two locations, Coral Bay and
Tantabiddi, in Ningaloo Marine Park (23° 48’ S - 21° 48’ S; Figure 2.1). Ningaloo
Reef is a fringing coral reef which stretches for approximately 270 km adjacent to
the semi-arid north-west cape of Western Australia. The sites sampled at both
locations were 35 m deep and between 1 and 2 km offshore from the reef slope. Site
depth was recorded to the nearest 0.5 m using the depth sounder onboard.
2.3.2 Sampling technique
The pelagic stereo-BRUVs used in this study were designed to be deployed,
anchored and to remain at a predetermined depth in the water column. Two Sony
CX12 high definition digital cameras were mounted 0.7 m apart on a galvanized
steel frame designed for mid-water deployment (Figure 2.2). The cameras were
converged inwards at 8° to gain a maximum field of view and to allow for fish
length measurements to be made (Harvey et al., 2010), although these length
measurements are not assessed here. The bait consisted of 800 g of crushed pilchards
(Sardinops sagax) in a wire mesh basket suspended 1.2 m in front of the cameras.
The bait arm acts as a rudder and keeps the camera system facing downstream of the
current. The use of ballast and sub-surface floats effectively reduces movement from
surface waves and allows for control over deployment depth. The deployment
system presented here was developed in collaboration with commercial fishermen
and allows for the effective deployment of several camera systems from a wide
range of boat types and sizes. These pelagic stereo-BRUVs have been used in depths
ranging from 5 to 200 m (Santana-Garcon et al., 2014d), but the deployment method
presented here can also be adapted to greater depths
2.3.3 Experimental design
The experimental design consisted of 3 factors: Location (2 levels, random: Coral
Bay and Tantabiddi), Site (2 levels, nested in Location, random) and Depth (2 levels,
fixed: Depth 1 – surface and Depth 2 – deep). Systems deployed at Depth 1
(surface) were designed to remain in the mid-water at ca 5 m depth (6.88 ± 0.26 m,
35
Mean ± SE), 30 m above the bottom. Systems at Depth 2 (deep) were deployed to ca
20 m (22.01 ± 0.23 m), at about 15 m above the bottom. For each depth, six replicate
deployments were conducted at each site. Overall, 48 mid-water stereo camera
systems were deployed and each recorded for 3 h. Samples were collected during
daylight hours, allowing at least 1 h between sampling and sunrise or sunset, to
avoid possible crepuscular variation in fish assemblages (Myers, 2013). A minimum
distance of 500 m was allowed between replicates at any time to avoid or minimize
potential overlap of bait plumes and to reduce the likelihood of fish moving between
replicates. Deployment depth was interspersed at each site so that different replicates
sampled the two depths simultaneously.
Figure 2.2 Deployment method and frame details of pelagic stereo-BRUVs.
2.3.4 Image analysis
The video images were analyzed using the software ‘EventMeasure (Stereo)’
(SeaGIS Pty Ltd). All fish recorded were quantified and identified to the lowest
taxonomic level possible. To avoid repeat counts of individual fish re-entering the
field of view, a conservative measure of relative abundance (MaxN) was recorded as
the maximum number of individuals of the same species appearing at the same time
(Priede et al., 1994). For the analysis of the influence of soak time on the relative
abundance and species composition, Cumulative MaxN was recorded at 15 minute
time intervals in order to record the number of new species and individuals observed
at each time period.
36
2.3.5 Data analysis
Deployment depth
Multivariate assemblage data were analyzed using non-parametric permutational
analysis of variance (PERMANOVA) in the PRIMER-E statistical package
(Anderson et al., 2008). The species abundance data was pre-treated with ‘dispersion
weighting’ to reduce the weight given to species that presented high spatial
clustering (Clarke and Gorley, 2006). Down-weighting was only applied to species
that exhibited significant evidence of clumping (Clarke et al., 2006b). A Bray-Curtis
coefficient on square root transformed data was chosen to allow mid-range and low
abundance species to exert an influence on the calculation of similarity (Clarke and
Warwick, 2001). A total of 9,999 permutations were computed to obtain P
significance values. Factor interactions with high P values (> 0.25) were pooled
(Underwood, 1997) and when the number of unique permutations was less than 100,
a Monte Carlo P value was used to interpret the result (Anderson et al., 2008). In
order to provide a suitable visual complement to the output given in PERMANOVA,
distance among centroids was calculated and plotted using Principal Coordinates
Analysis (PCO).
Optimization of sampling effort
The relationship between soak time and the number of species or individuals
observed by a sampling technique can be represented by species accumulation
curves and cumulative abundance curves. Species richness and relative abundance
curves were plotted for 15 minute time intervals up to 180 min of soak time.
The univariate species richness and total abundance (MaxN) raw data were used to
calculate the mean precision of the sampling technique for different sample sizes
(number of deployments) and for 3 soak times: 60, 120 and 180 min. For the
calculation of precision, data from only one of the depth treatments (surface) was
used to avoid confounding effects among treatments. However, all other data
combinations were explored and produced equivalent results to those presented here.
Precision (P) was calculated as:
P = SE / Mean
37
where Mean is the sample mean and SE the standard error . Bootstrapping
procedures were used to simulate samples from 2 to 12 replicates using the 24
replicates available. Replicates were selected randomly with replacement and the
procedure repeated 1000 times for each sample size. Mean precision and the 2.5 and
97.5 percentiles were calculated and plotted using the packages ‘reshape2’
(Wickham, 2007) and ‘ggplot2’ (Wickham, 2009) in R software version 2.15.2 (R
Core Team, 2012).
The cost differences among the three soak times and number of replicates were
estimated as a combination of the time required in the field and in the lab for video
analysis (see Gladstone et al., 2012). Field time was estimated for the deployment of
5 camera systems from a small boat (i.e. 5.5 m in length) allowing 30 min for the
deployment and retrieval of 5 systems including the travel among deployment sites.
Furthermore, for each day of fieldwork, 2 h were added to account for the
preparation of gear and travel time to the field locations. A maximum of 9 h was
allowed for fieldwork days, including preparation and time in the field. Times
required for video analysis in the lab were highly variable depending on the number
of fish and species present in the video but were most commonly about 1.5, 2 and 2.5
h for soak times of 60, 120 and 180 min respectively. The values presented here are
only used as a guide and were used to compare among sampling efforts. This time
calculation will change significantly across studies depending on the field location
and the resources available.
2.4 Results
From the 48 pelagic stereo-BRUVs deployed, 144 h of video were analyzed and a
total of 6,988 individuals of 43 taxa were identified representing 18 families. Of the
43 taxa defined, 31 were confidently identified to species, 6 were grouped to genus
and 6 to either family or a higher taxonomic level (Table 1). Species were grouped to
higher taxonomic levels when they could not be identified confidently across all
videos. Some of the specimens observed were juvenile fish in their pelagic phase
(17.3 %) and a large proportion of the individuals were pooled under the category of
‘various small pelagics’ (70.4%). This category included fish schooling in high
numbers from the family Clupeidae and small carangids from the genus Decapterus,
38
Selar and Selaroides among others; their identification to species level was not
possible for the present study. Juvenile fish can be difficult to identify from video
alone given their small size and their similar morphology at this size so some species
were grouped as ‘unidentified juveniles’ for the analysis.
Table 2.1 Taxonomic group and stage of specimens recorded using pelagic stereo-BRUVs in Ningaloo Marine Park. Stage of individuals was identified as either adult (AD) or juvenile (J).
Family Stage Family (cont.) Stage
Species (AD, J) Species (AD, J)
Acanthuridae Monacanthidae
Acanthurus mata AD Aluterus monoceros J
Belonidae Aluterus scriptus AD, J
Tylosurus crocodilus AD Cantherhines pardalis J
Caesionidae Eubalichthys caeruleoguttatus J
Caesio teres AD Paramonacanthus choirocephalus J
Carangidae Other J
Alectis sp. J Myliobatidae
Alepes sp. AD Myliobatidae sp. AD
Atule mate AD Nomeidae
Carangoides fulvoguttatus AD Nomeidae spp. J
Elagatis bipinnulata AD Priacanthidae
Gnathanodon speciosus J Priacanthus tayenus J
Seriolina nigrofasciata J Rhincodontidae
Other J Rhincodon typus AD
Carcharhinidae Scombridae
Carcharhinus albimarginatus AD Acanthocybium solandri AD
Carcharhinus amblyrhynchos AD Grammatorcynus bicarinatus AD
Carcharhinus plumbeus AD Scomberomorus sp. AD
Carcharhinus obscurus/brachyurus AD Thunnus albacares AD
Galeocerdo cuvier AD Thunnus tonggol AD
Other AD Other Thunnus spp. AD
Echeneidae Sphyraenidae
Echeneis naucrates AD Sphyraena barracuda AD
Remora remora AD Sphyrnidae
Fistulariidae Sphyrna mokarran AD
Fistularia sp. J Tetraodontidae
Lutjanidae Lagocephalus sceleratus AD
Lutjanus bohar AD Unknown
Lutjanus fulviflamma AD Unidentified juveniles J
Various small pelagic (mainly clupeids and small carangids)
AD, J
39
Deployment depth
The three-way PERMANOVA determined that species composition differed
significantly between locations and depths (both P < 0.05), among sites (P < 0.001)
and there was a significant interaction (P < 0.001) between each of these factors
(Table 2.2). Although the PCO plot shows a greater distance between the assemblage
centroids at the different locations (Coral Bay and Tantabiddi; PCO1 axis) than
between the two depths (PCO2 axis), there is a clear separation between depths at
each site (Figure 2.3). Most species were recorded at both depths but the assemblage
differences reported between treatments were mainly driven by species that only
occurred at one of the two depths sampled (Figure 2.4). The species that were
associated with surface deployments alone (Depth 1) include Tylosurus crocodilus,
Elagatis bipinnulata and the scombrid Acanthocybium solandri. Sharks of the
family Carcharhinidae were mostly sampled in the deeper deployments (Depth 2)
except for Carcharhinus plumbeus and Galeocerdo cuvier which, in this study, were
exclusive to the deployments near the surface. Also unique to the shallower depth
treatment was a whale shark Rhincodon typus and the specimens of Remora remora
associated with it. Exclusively sampled on the deeper deployments, 15 m above the
bottom were Alepes sp., Acanthurus mata, Caesio teres, Grammatorcynus
bicarinatus and Lagocephalus sceleratus. Species more commonly associated with
the reef such as Lutjanus fulviflamma and Lutjanus bohar, were also reported in the
deep treatment only. Tuna species in the genus Thunnus were identified at both
depths but the species Thunnus tonggol and Thunnus albacares were only recorded
at the deeper deployments.
Table 2.2 PERMANOVA testing for the effects of deployment depth on the structure of fish assemblages sampled using pelagic stereo-BRUVs. Results are based on Bray-Curtis similarity matrices using dispersion weighted and square-root transformed data. All tests were undertaken using 9,999 permutations under the reduced model; when less than 100 unique permutations were obtained, Monte Carlo P values were used and are shown in italics. Values highlighted in bold are significant.
Source of variation df MS Pseudo-F P (perm) Location 1 30262 4.757 0.011
Depth 1 7916.6 2.473 0.035
Site (Location) 2 6361.6 4.948 <0.001
Pooleda 3 3201.6 2.490 <0.001
Residual 40 1285.6 Total 47 a Pooled: (Site(Location) x Depth Treatment) + (Location x Depth Treatment)
40
Figure 2.3 Principal Coordinates Analysis plot showing differences in fish assemblages sampled at two depth treatments (Depth 1, surface and Depth 2, deep) using pelagic stereo-BRUVs. Different symbols show the sites sampled in Tantabiddi (TN1, TN2) and Coral Bay (CB1, CB2) in Ningaloo Marine Park, Western Australia.
Figure 2.4 Mean relative abundance (MaxN) of fish taxa sampled at two different deployment depths using pelagic stereo-BRUVs in Ningaloo Marine Park: Coral Bay and Tantabiddi. The vertical axes are shown in logarithmic scale and error bars represent standard error. Some species are shown grouped at a higher taxonomic level and species that had only one individual recorded are excluded.
41
Optimization of sampling effort
The mean species richness and mean relative abundance (MaxN) was estimated
cumulatively in 15 min time intervals (Figure 2.5). The slope of the species
accumulation curve is steep as soak time increases, the slope of the curve is reduced
after ca 90 min but the trend is still increasing at a reduced rate at 180 min. The
abundance curve follows a similar pattern, but shows a steeper increase with the
mean number of individuals rising steadily in 15 minute intervals throughout the
time sampled.
Figure 2.5 Species accumulation curve and relative abundance accumulation curve (B) in 15 minute time intervals for soak times up to 3 hours. Error bars represent standard error.
The mean precision and 95 % confidence intervals were calculated for the species
richness and abundance data for the three different soak times (60, 120 and 180 min)
and for sample sizes ranging from 2 to 12 replicates (Figure 2.6). Note that lower
values of precision indicate more precise data. The effect of soak time on the mean
precision remains consistent across the different sample sizes with precision
improving with the increase of soak time. The improvement in precision is greater
42
when soak time is extended from 60 to 120 min than in the period from 120 to 180
min. Precision also improves with an increase in replication, reaching an optimum at
8 replicates with only minimal marginal improvement beyond that number of
replicates. The increase in sample unit size and replication not only reduces mean
precision values, but also reduces the range of their 95% confidence intervals to a
great extent. The values of precision calculated from the species richness data are
lower (more precise) than those from the abundance data across all soak times and
sample sizes.
Figure 2.6 Mean precision and 95 % confidence intervals for (A) species richness and (B) relative abundance data using pelagic stereo-BRUVs with three different soak times. Samples simulated using bootstrapping procedures with replacement. Lower values of precision indicate more precise data.
The cost (working time) of sampling increases as the sample size increases for the
three different soak times (Figure 2.7). As expected, cost also escalates with soak
time. Since the time-cost was calculated for fieldwork using 5 camera systems, the
field cost increases every 5 replicates. For example, in order to increase from 10 to
12 replicates, with 180 min soak time, one extra day of fieldwork is required. The
43
time allocated to video analysis increases linearly with sample size and also slowly
increases for longer soak times.
Figure 2.7 Estimate of the cost, in time, for sampling fish assemblages using different soak times with pelagic stereo-BRUVs. Cost includes the time for fieldwork with 5 camera systems and video analysis.
2.5 Discussion
Pelagic stereo-BRUVs were effectively used to sample fish in the mid-water
environment identifying 43 taxa from 18 different families. The technique allowed
us to count and identify not only fish attracted to the bait, but also other fish in the
area and those swimming within the field of view by chance (e.g. the whale shark
Rhincodon typus). Moreover, small schooling fish and juveniles in their pelagic
phase were attracted to the physical structure and aggregated around the camera
system which, in some instances, attracted bigger pelagic fish to the field of view.
This effect of acting as a fish aggregation device enhances the sampling ability of
BRUVs in such an open environment as it concentrates individuals from a broader
area within a distance that allows for their identification and measurement (Bailey et
al., 2007).
Video techniques also provide a permanent record of the fish species occurring in an
area. The imagery recorded reduces the need for skilled observers in the field as
species can be identified in the laboratory by various observers and, video clips or
captions sent to experts for confirmation (Cappo et al., 2003). Despite having a
44
permanent record, the identification of some pelagic fish to the species level from
video alone have proved challenging and some had to be grouped at higher
taxonomic levels. This may be due to the adaption of pelagic fish to be almost
invisible in the mid-water as their coloration is countershaded (i.e. dark on the dorsal
side and pale on ventral side) and reflective which redirects light to make them
appear somewhat transparent. Furthermore, a wide range of taxa share similar traits
such as a streamlined fusiform body, pointy snouts and deeply forked and lunate
caudal fins. The difficulty in identifying pelagic species is also reported in other non-
extractive sampling methods like hydroacoustics (Lawson et al., 2001) and
underwater visual census, which are often not well suited to survey the pelagic
environment (Claudet et al., 2010).
In this study we classified a large proportion of the individuals observed under the
category ‘various small pelagics’ (70.4 %) which included mainly small carangids
and clupeids. Classification to lower taxonomic levels was not consistently possible
across all replicates due to the similarity between species, their small size and rapid
motion. This group fits within the definition of ‘small pelagic fish’ by Freon et al.
(2005), which defined them as shoaling epipelagic species preying on plankton,
characterized by high horizontal and vertical mobility in coastal areas and that, as
adults, are usually 5 to 30 cm in length. Small pelagic fish are gregarious and, as
observed in our study, often several species of similar size and body shape gather in
single schools (Freon and Misund, 1999). Pooling these species in our analysis was
considered suitable as they share specific traits in relation to their habitat, biology,
behavior and ecology (Freon et al., 2005). However, small pelagic fish often
constitute the bulk of the biomass in marine ecosystems; hence studies that require
more detail on these species should consider this limitation of pelagic stereo-BRUVs
and could use them in combination with a range of other sampling techniques.
The fish assemblages sampled at the two mid-water depths were significantly
different. These differences were mainly driven by species that during our study
were recorded at only one of the deployment depths. Benthic BRUVs have been used
to describe the changes in community structure at different depths (Bailey et al.,
2007; Goetze et al., 2011; Zintzen et al., 2012). The present study only tested the
effects of deployment depth in shallow water (35 m) by comparing mid-water
deployments at 5 and 20 m. However, it validates the use of this technique in the
45
future to investigate the distribution patterns along broader depth ranges. Vertical
migrations of fish are known to be common and driven by changes in the distribution
of prey and predator species and by levels of light that modify species detection
(Axenrot et al., 2004; Freon et al., 2005). In order to gain a better understanding on
how these processes affect pelagic and demersal fish communities, depth structure
and diel migrations could be explored using pelagic stereo-BRUVs at different
depths during the day and night.
An increase in soak time will affect the estimates of diversity and relative abundance
obtained from pelagic stereo-BRUVs as longer sampling times record greater
numbers of species and individuals. This has also been reported for estimates from
pelagic longline surveys (Ward et al., 2004) and in other studies using baited video
techniques (Willis and Babcock, 2000). The increase in species richness and
abundance in relation to longer soak times could be the result of the bait plume
dispersal attracting fish from a greater distance (Sainte-Marie and Hargrave, 1987).
However, standardizing soak time across deployments would provide comparable
estimates of richness and abundance (Cappo et al., 2006; Cappo et al., 2003; Willis
and Babcock, 2000). Soak time can also affect the behavior of fish, as longer
deployments allowed cautious individuals that first remain in the distance to get
closer to the bait, which in turn improves their identification and measurement. For
example, some sharks in the genus Carcharhinus that could not be confidently
identified to species level during the first hour of deployment, subsequently
approached the bait later in the video allowing identification to species level. Soak
time in the past was mainly limited by the capacity and cost of the recording media
(Harvey et al., 2010). Currently, while storage capacity has increased and costs have
decreased, soak time mainly depends on the resources available as increasing soak
time incurs greater costs both in the field and the succeeding laboratory analysis. In
addition, if detecting biodiversity (i.e. rare fish) was an objective then cost would
increase as longer soak times would be required.
In order to optimize the use of pelagic stereo-BRUVs in the Ningaloo Marine Park,
we estimated the precision of the data under different sampling regimes. In these
waters and at the depths sampled, a soak time of 120 min and a sample size of at
least 8 replicates per treatment are optimal for sampling the pelagic assemblage. We
believe that similar regimes will be suitable for sampling other warm-temperate and
46
tropical coastal environments. Different indices of assemblage composition showed
distinct levels of precision under the same sampling regime. The species richness
data produced more precise estimates than the abundance data so studies might adapt
their sampling effort depending on their objectives. In general, precision improved
when soak time was extended from 60 to 120 min, but longer deployments did not
significantly improve the quality of the data any further. Moreover, as expected,
precision improved rapidly when increasing the number of replicates (Andrew and
Mapstone, 1987; Gladstone et al., 2012). It is important to increase spatial and
temporal replication in order to account for the high variability of the processes that
determine the distribution of fish in the pelagic environment (Bakun, 1996). We
encourage maximizing replication given the resources available while standardizing
the soak time to facilitate future comparisons across studies using pelagic stereo-
BRUVs.
The cost of sampling increased for bigger sample sizes and longer deployment times.
Since the cost of sampling for 180 min of soak time was greater than that for 120
min, but the quality of the data was not significantly enhanced, we would
recommend using the resources to increase sample size rather than soak time. In
addition, increasing the number of camera systems available in future studies will
increase replication at no greater field time-cost.
The main advantage of using pelagic stereo-BRUVs to investigate the ecology of
pelagic fish is that they are a non-destructive and fishery-independent technique that
can be standardized and used to compare fish assemblages across treatments, spatial
areas and over time. Moreover, stereo-video can obtain accurate length
measurements of the pelagic fish observed (Santana-Garcon et al., 2013). Pelagic
stereo-BRUVs may have the potential to monitor the effects of fishing closures on
pelagic fish in the same way as benthic BRUVs have been used to study such effects
on demersal target species (Goetze et al., 2011; Harvey et al., 2012b; McLean et al.,
2010; Watson et al., 2009). In addition, remote video techniques can be used in areas
and at depths beyond the limits of diver based studies (Cappo et al., 2006), as well
as, for monitoring fish assemblages both during the day and at night using lights
(Harvey et al., 2012a). Pelagic stereo-BRUVs provide a window to the mid-water
that allows for in situ behavioral studies that would otherwise take place in the lab
(Santana-Garcon et al., 2013). The behavior of pelagic fish in their natural
47
environment is largely unknown and could be revealed using remote stereo-video as
has occurred in deep water studies (Zintzen et al., 2011).
All sampling techniques have some inherent biases and limitations (Murphy and
Jenkins, 2010), pelagic stereo-BRUVs are no exception. As discussed, the
identification to species level can be difficult, especially for juvenile fish and small,
highly mobile species. Another shortcoming of stationary camera systems is that
only a relatively restricted volume of water is sampled, but this is improved with the
use of bait and by deploying several systems simultaneously to increase spatial
replication (Harvey et al., 2007). Despite attracting fish to the field of view, this
technique produces zero inflated data due to the high spatial and temporal variability
of the pelagic environment. Generally, zeros play a special role in the analysis of
assemblage data and analyzing assemblage matrices with many zeros may have
limited statistical power or require more complex analysis to identify patterns and
changes in the community (Clarke et al., 2006a). Consequently, in order to reduce
the patchiness and variability of the data, greater sampling effort might be needed
when sampling the mid-water in comparison to other environments such as reef
habitats.
Biodiversity assessments need to be made at the community, habitat and landscape
levels in order to predict changes over time (Gray, 1997). Most methods used in the
pelagic environment are looking at very specific questions or at very broad scales
(i.e. ocean wide); yet pelagic stereo-BRUVs could help explore patterns and
processes occurring at the assemblage or community level. Fish diversity in the
pelagic environment is generally low in comparison to that of the benthic habitats
and the boundaries of ecosystems are more loosely defined and difficult to determine
(Angel, 1993; Gray, 1997). During this study, we observed extreme temporal
variability in both the assemblage composition and abundance. For example, the
same site that recorded no fish or very low abundance across replicates one day had
high diversity and abundance the next day. Consequently, it is very important to
include temporal replication and sample simultaneously across different treatments
as we did in this study. Also, when possible, replicates should be deployed at the
same time across sites and locations. Due to logistical and resourcing constraints, we
sampled the two locations on different weeks so some of the differences observed
across locations might be driven by the temporal variability of the system.
48
In comparison to the common benthic BRUVs (Cappo et al., 2001), the logistics and
use of pelagic stereo-BRUVs in the field are very similar. The fact that the main
structure is suspended in mid-water and that only a ballast is deployed to the bottom
facilitates their use on complex habitats (Merritt et al., 2011). A singular trait of the
pelagic camera systems is that the bait arm acts as a rudder and makes the cameras
face downstream of the current so that fish following the bait plume are observed
approaching the cameras. Another unique feature of pelagic stereo-BRUVs is the use
of sub-surface floats that stabilize the image by reducing the effect of wave motion.
The use of bait in benthic BRUVs has been well investigated (Dorman et al., 2012;
Harvey et al., 2007), but further research is needed to explore the role of bait in
pelagic deployments, as well as the effect of other visual or acoustic cues to attract
pelagic fish to the field of view.
In conclusion, pelagic stereo-BRUVs are a novel technique that can be used to
effectively investigate pelagic fish assemblages. Future research should further
explore the potential and limitations of this technique in comparison to other
sampling methods. The results from this study can provide guidance to standardize
and optimize the resources available in future sampling programs. Pelagic stereo-
BRUVs used in combination with other techniques will improve our understanding
of the ecology and behavior of pelagic ecosystems.
49
CHAPTER 3 - Multivariate pseudo standard error as a measure of precision: optimisation for sampling pelagic fish assemblages
Prelude
The research presented in this chapter includes concepts developed as part of a larger
study undertaken in collaboration with Prof. Marti J. Anderson (MJA) from Massey
University (New Zealand). In this chapter, I have condensed the results from
Anderson and Santana-Garcon (In review; Appendix 1) that are directly applicable to
this thesis. As a result of this collaboration, MJA developed the statistical method
and the R script that I have used for the analysis of my data. The chapter has kept the
language as in the submitted manuscript (e.g. using we instead of I) to acknowledge
the work undertaken by both authors.
3.1 Abstract
The accurate and precise description of biological patterns is essential to most
sampling techniques. However, unlike in univariate statistics, there is no measure of
multivariate precision described to assess sampling effort in multivariate ecological
studies. We propose the use of a pseudo multivariate dissimilarity-based standard
error (MultSE) as a direct analogue to the univariate standard error, and as a useful
approach to assess the adequacy of sampling effort for multivariate studies. We use
this measure of multivariate precision to optimise the sampling unit size and
replication for studies using pelagic stereo-BRUVs to sample fish assemblages. The
results suggest that a soak time of 120 minutes and a sample size of at least 8
replicates per treatment is an appropriate sampling effort to obtain precise data using
pelagic stereo-BRUVs in tropical or warm-temperate waters. These findings are in
line with previous recommendations based on univariate precision estimates. We
suggest that multivariate ecological studies should use measures of multivariate
precision, such as MultSE, in order to optimise their sampling regime.
50
3.2 Introduction
The use of multivariate analysis in ecological studies is becoming increasingly
important. Ecologists often need to sample whole assemblages of species at once to
test hypotheses concerning the effects of experimental factors or environmental
impacts (Anderson, 2001). The accurate and precise description of biological
patterns is essential to most sampling techniques used in ecology (Andrew and
Mapstone, 1987). However, unlike in univariate statistics, for multivariate data there
is no obvious measure of multivariate precision that can be used on pre-existing data
to assist researchers in the decision-making processes regarding an adequate
intensity of sampling (Anderson and Santana-Garcon, In review; Appendix 1).
The precision of a sampling technique refers to the repeatability of its measurements;
the degree of concordance among multiple estimates of a given parameter (such as
the mean) for the same population (Cochran and Cox, 1957). Precision is most
affected by sampling effort, which depends on the sampling unit size and the number
of replicates (Andrew and Mapstone, 1987). An increase in the sample unit size (e.g.
quadrat size, transect length, soak time) often involves an increase in the cost of
sampling (Bartsch et al., 1998; Gladstone et al., 2012; Langlois et al., 2010; Pringle,
1984). For instance, for video techniques, longer soak times limit the number of
replicates that can be undertaken per day in the field, and the associated time
required for subsequent video analysis increases. Thus, given the resources available,
any method should use the sampling effort that yields the most adequate precision
(Pringle, 1984).
Precision can be used to optimise sampling effort in studies involving univariate data
(Bartsch et al., 1998; Downing and Downing, 1992; Gladstone et al., 2012; Pringle,
1984). To measure species richness or the relative abundance of a single species
(univariate), methods exist to allow researchers to assess (e.g. from a series of
replicates obtained in pilot investigations) the adequacy of any given choice of
sampling-unit size and/or number of replicates (see review in Andrew and Mapstone,
1987). In particular, for univariate data one may calculate a value of precision for a
given sample of n units of a given size. Univariate precision can be calculated from a
sample as the standard error divided by the mean; namely, , where ySEp /=
51
, s is the sample standard deviation, is the sample mean and n is the
number of sampling units. Note that, as values of p decrease, precision improves.
Measures of precision can be calculated from pilot data, a plot of p vs. n can be
drawn, and we expect a gradual decrease and levelling-off in the value of p with
increasing n in such a plot. One may then consider that a value of n around this
apparent asymptote in p (i.e., where the slope of the curve becomes very small)
would be reasonable to use for future investigations of the population, as further
increases in n would not result in substantial improvements in precision.
For instance, the precision of univariate data obtained under different sampling
regimes has been assessed in order to optimise the sampling effort of a novel video
technique used in pelagic environments (Santana-Garcon et al., 2014a). Pelagic
Baited Remote Underwater stereo-Video systems (pelagic stereo-BRUVs) are a non-
destructive fishery-independent method that provide estimates of species
composition, relative abundance, length measurements, and behavioural observations
of pelagic fishes (Cappo et al., 2003; Harvey et al., 2003; Heagney et al., 2007;
Santana-Garcon et al., 2014a; Santana-Garcon et al., 2014c). The parameters
potentially affecting the sampling ability of pelagic stereo-BRUVs include soak time
(time that cameras are left recording underwater) and the number of replicate
deployments. The replication required to use this sampling technique in the pelagic
environment has been optimised under different soak time regimes using estimates
of univariate precision for species richness and abundance data (see Santana-Garcon
et al., 2014a). However, the precision and optimisation of the sampling of these fish
assemblages using multivariate data was yet to be explored.
We propose the use of a pseudo multivariate dissimilarity-based standard error
(MultSE) – a direct analogue to the univariate standard error – as a useful quantity
for assessing sample-size adequacy for multivariate ecological studies. This can be
calculated easily from a single set of pilot data or on residuals after conditioning on
some more complex sampling design or model. We further propose a double-
resampling technique (permutation followed by bootstrapping) in order to quantify
uncertainty in MultSE values with increasing numbers (or sizes) of sampling units. In
this study, we calculate the MultSE for data on pelagic fish assemblages sampled
nsSE /= y
52
using pelagic stereo-BRUVs and compare it to particular univariate precision results
obtained from the same dataset and presented in Santana-Garcon et al. (2014a).
The aim of this study is to demonstrate that MultSE can be used as a way of
optimising the sampling regime of multivariate studies in much the same way as
precision has been used in a univariate context. Moreover, we aim to optimise
sampling of pelagic fish assemblages using pelagic stereo-BRUVs by identifying an
appropriate soak time and replication beyond which there is no substantial
improvement in MultSE. Finally, we compare the resulting optimal sampling effort
to that obtained from univariate precision estimates.
3.3 Methods
The data presented was collected near Tantabiddi Passage, in the Ningaloo Marine
Park (21°53’ S, 113°56’ E). Ningaloo Reef is a fringing coral reef which stretches
for approximately 270 km adjacent to the semi-arid north-west cape of Western
Australia. The two sites sampled at this location were ~35 m deep and between 1 and
2 km offshore from the reef slope (Figure 3.1a). The pelagic stereo-BRUV system
used in this study (Figure 3.1b) is described in detail in Santana-Garcon et al.
(2014a). The systems were designed to remain in the mid-water at ca. 5 m depth
(7.17 ± 0.54 m, Mean ± SE), 30 m above the bottom and were deployed for a soak
time of 3 hours. The data presented herein was obtained from 12 pelagic stereo-
BRUVs deployments; 6 replicates conducted at each site. All samples were collected
during daylight hours, allowing at least one hour between sampling and sunrise or
sunset to avoid possible crepuscular variation in fish assemblages (Myers, 2013).
The video images were analysed using the software ‘EventMeasure (Stereo)’
(SeaGIS Pty Ltd). All fish recorded were quantified and identified to the lowest
taxonomic level possible. To avoid repeat counts of individual fish re-entering the
field of view, a conservative measure of relative abundance (MaxN) was recorded as
the maximum number of individuals of the same species appearing at the same time
(Priede et al., 1994). MaxN was recorded at 60, 120 and 180-minute soak time
intervals.
53
Figure 3.1 (a) Study area in Ningaloo Marine Park (Western Australia; site 1 is TN1 and site 2 is TN2 as per defined in Chapter 2) and (b) design of pelagic stereo-BRUVs for sampling mid-water fish assemblages.
3.3.1 Multivariate pseudo-Standard Error
As a direct analogue to the univariate measure of precision, we consider a
multivariate measure of pseudo standard error, as follows (MultSE, first described by
Anderson, 2001; full description of the method in Anderson and Santana-Garcon, In
review). Let D be a ( ) matrix of dissimilarities among all pairs of
sampling units, i = 1, ..., n and j = 1, ..., n. Using Huygens’ theorem (e.g. Anderson,
2001; Legendre and Anderson, 1999), the sum of squared inter-point dissimilarities
divided by the number of sampling points (n) is directly interpretable as a
multivariate measure of pseudo sums of squares in the space of the chosen
dissimilarity measure:
(1)
The quantity in equation (1) is also equivalent to the sum-of squared distances from
individual sampling points to their centroid in the space of the chosen resemblance
nn × }{ ijd
ndsn
i
n
ij
ij /)1(
1 1
2∑ ∑
−
= +=
=
54
measure (Anderson, 2001). Furthermore, dividing this by (n – 1) yields a quantity
that is directly interpretable as a multivariate measure of pseudo variance in the
space of the chosen dissimilarity measure:
(2)
Indeed, in the case of a single variable and Euclidean distance, s is equal to the usual
classical univariate sum-of-squares and v is equal to the usual classical unbiased
measure of the sample variance.
Note however that the word “pseudo” has been used throughout here to distinguish s
in (1) above from the classical multivariate sums-of-squares-and-cross-products
(SSCP) matrix and v in (2) above from the classical multivariate sample variance-
covariance matrix. The classical matrices will clearly take into account correlation
structures in the multivariate (Euclidean) space of the original variables, which these
pseudo measures do not. To see this, note that in the case of multivariate data and
Euclidean distance, s in (1) is equal to the trace of the classical SSCP matrix and v in
(2) is equal to the trace of the classical sample variance-covariance matrix.
Finally, based on the above, a multivariate pseudo standard error therefore can be
calculated as:
(3)
Although the univariate measure of precision takes the standard error divided by the
sample mean, this is not possible in the present case because the sample mean is
clearly a vector here (being multivariate). However, we consider that m in (3) is
perfectly suitable as a measure of precision within a given study, because the
estimate of the mean (generally obtained as the sample average for univariate
analysis) is unbiased for any sample size. The only consequence of failing to divide
by the mean is to produce a value which is not standardized in any way, hence can
only be used within the context of a given study, and not compared among studies.
Thus, although some chosen value of precision for univariate analysis might be
suggested as a rule of thumb to be applied to studies across a given field (e.g. p = 0.2
or p = 0.5, etc.), the value of m in (3) is dependent on the scale of the dissimilarity
ndvn
i
n
ij
ijn/
)1(
1 )1(
2)1(
1 ∑ ∑−
= +=−
=
nvm /=
55
measure chosen, so no such generalization for its value across multiple studies is
apparent. However, m will still nevertheless be quite useful for examining relative
precision for different sample sizes within a given study.
We propose that, given a set of pilot multivariate data having a total of N sampling
units, one may draw a random subset (without replacement) of size n = 2, 3, 4, ..., N
and for each of these random subsets, a value of m can be calculated, which may be
denoted in each case by , , ... . Repeat this process for a very large
number of random draws (say 10,000), then calculate the mean values obtained
under permutation for each sample size as , , ... . A plot of vs. k
for k = 2, 3, ..., N will then provide the graphic we require to assess precision with
increasing sample size. We propose that these means be calculated using sampling
without replacement (i.e., using a permutation method) specifically to ensure that the
values obtained will be unbiased.
Next, we propose that a bootstrapping method be used to draw error bars on the plot.
It is clear that the number of possible permutations for n = N is 1 (i.e., this is simply
the full set of data) and further that the number of random subsets without
replacement that would yield unique values for (equivalent to simply
leaving out one observation at a time) is also limited to just N possible values.
Hence, we propose obtaining a further random subset as a random draw with
replacement (i.e., a bootstrap sample), of size n = 2, 3, 4, ..., N and for each of these
to calculate a value of m, denoted in each case by , , ... . Consider the
0.025 and 0.975 quantiles of the bootstrap distribution for each sample size n = k for
k = 2, 3, ..., N as and , respectively. Unlike the permutation
approach, the bootstrap distribution is known to be biased (e.g. Davison and
Hinkley, 1997). An empirical estimate of the bias, for each sample size k, is given
by:
(4)
Hence, bias-adjusted 0.025 and 0.975 quantiles are provided by
and , (5)
π2=nm
π3=nm
πNnm =
π2=nm
π3=nm
πNnm =
πknm =
π)1( −= Nnm
β2=nm
β3=nm
βNnm =
)025.0(βknq = )975.0(β
knq =
)( βπknknkn mmb === −=
})025.0({ knkn bq == +β })975.0({ knkn bq == +β
56
respectively. As our proposed method uses permutations to obtain mean values for
the plot, and a separate bias-adjusted bootstrapping approach to obtain appropriate
error bars, we shall refer to this overall process generally as a “double resampling”
method. We note also that clearly other choices of quantiles may easily be calculated
here in a similar fashion (e.g. 0.25, 0.75, etc.), depending on the context.
The R code (R Core Team, 2014) developed for the methods proposed is provided as
a supplement in Anderson and Santana-Garcon (In review; Appendix 1).
3.3.2 Optimisation of sampling effort for multivariate analyses
In order to understand the performance of the permutation and bootstrap method
described above, we assess the means and quantiles of MultSE with increasing
sample size using each method separately. This analysis is conducted only for the 60
min soak time data, and is based on the square-root transformed abundance estimates
and Bray-Curtis dissimilarities. Next, we used the double resampling approach to
estimate the mean and quantiles of MultSE for the three soak times (60, 120 and 180
min) for 1 to 12 replicate samples.
3.4 Results
First, we examined the means and quantiles of MultSE with increasing sample size
using either the permutation or the bootstrap method alone for the 60 minute soak
time. Both approaches show precision improving with increasing sample size.
However, the permutation method alone yields obvious constraints on the error bars
for larger sample sizes, while the bootstrap method alone yields means that are
biased downwards (Figure 3.2). Thus, this motivates the use of the double
resampling approach described.
Using the double resampling method shows that MultSE as a measure of multivariate
precision improves when soak time is extended from 60 to 120 minutes, but
extending soak time to 180 minutes does not improve the precision of the data
further. Moreover, MultSE also improves with replication and it stabilises when
replication reaches 7 or 8 deployments (Figure 3.3). The range of the 95 %
confidence intervals is reduced with the increase of replication more than with the
extension of the soak time.
57
Figure 3.2 Multivariate pseudo standard error as a function of sample size on the basis of Bray-Curtis dissimilarities calculated on square-root transformed fish abundance data sampled using pelagic stereo-BRUVs for a soak time of 60 minutes only, showing results (means with 2.5 and 97.5 percentiles as error bars) from 10,000 resamples obtained using either the permutation or bootstrap approach.
Figure 3.3 Multivariate pseudo standard error as a function of sample size on the basis of Bray-Curtis dissimilarities calculated on square-root transformed fish abundance data sampled using pelagic stereo-BRUVs for all three soak times (60, 120 and 180 minutes), using the double-resampling method, with permutation-based means and bias-adjusted bootstrap-based error bars (with 10,000 resamples for each).
3.5 Discussion
We have described a measure of precision for multivariate data (MultSE) based on
inter-point dissimilarities that is a direct analogue of the well-known univariate
standard error. A double resampling method allows unbiased measures of variation
in MultSE to be obtained. This estimate can be calculated from pilot data alone and
58
will allow for the optimisation of sampling in a broad range of ecological studies
using multivariate data.
The use of MultSE as a measure of precision to optimise the sampling of pelagic
stereo-BRUVs shows that precision is improved by extending camera deployments
from 60 to 120 minutes, but no appreciable precision is gained by deploying the
cameras longer than 120 minutes. Moreover, the levelling-off in MultSE occurs for
sample sizes around n = 7 or 8 replicates, indicating that these or larger sample sizes
would be appropriate when using this technique to study mid-water fish assemblages
in Ningaloo Marine Park.
The results obtained are in line with the findings from Santana-Garcon et al., (2014a)
that used the same dataset to calculate univariate precision and determined that
reasonably precise estimates of total abundance and species richness of pelagic fish
assemblages could be achieved using ca. 8 replicate deployments of pelagic stereo-
BRUVs each of 120 minutes duration. Both for the univariate and multivariate
precision, the idea of “levelling-off” of precision values must rest in the eyes of the
beholder. This approach is intended to be a heuristic diagnostic tool only, and not a
prescription for defining appropriate sampling designs. Nevertheless, the analyses of
MultSE in this case coincide well with the results obtained for the precision of either
of the univariate measures of richness and total abundance (Santana-Garcon et al.,
2014a). This coincidence might not always occur, however, and will depend on the
nature of the data and the characteristics that are emphasised by the chosen
dissimilarity measure (Anderson and Santana-Garcon, In review; Appendix 1).
This approach to calculate MultSE is further expanded and tested on more complex
designs in Anderson and Santana-Garcon (In review; Appendix 1). We recognise
some limitations to the method described; for instance, the measure takes no account
of the shape of the multivariate data cloud, only its overall dispersion (size). Future
developments should include better characterization of the actual shape of the data
cloud, which might be achieved, for example, through the use of multivariate kernel
density estimation (KDE) in the principal coordinate (PCO) space of the
resemblance measure. This will require enhanced computer power and programming
to cope with analyses of high-dimensional datasets like those typically found in
ecology. Future work should focus also on how such approaches might compare with
59
a more classical approach to measure precision, such as the determinant of a
normalized variance-covariance matrix in the PCO space.
A further limitation of the approach outlined here is its scale-dependency. Values of
MultSE will necessarily depend on the dissimilarity measure used, so cannot be
easily compared across studies. On the other hand, estimates of univariate precision
are standardised as they take the standard error divided by the mean, which allow for
precision values to be defined as acceptable and applied as a rule of thumb across
studies (e.g. precision values of 0.2 or 20% have been described as a ‘tolerable error’
in ecological studies (Bartsch et al., 1998). More work is needed to assess what
levels of multivariate precision (as measured by MultSE) might be acceptable for
different resemblance measures that are bounded and have a common scale, such as
Bray-Curtis. However, values of MultSE will still be useful for sampling
optimisation as it allows for the comparison of relative precision across sample sizes
within a study.
In conclusion, we suggest that multivariate ecological studies should use measures of
multivariate precision, such as the MultSE described here, in order to optimise their
sampling regime. As suggested by the univariate precision measures, estimates of
multivariate precision indicate that studies sampling with pelagic stereo-BRUVs
should use soak times of 120 minutes and sample sizes of at least 8 replicates.
Further calculations should be carried out in other areas but we believe that similar
sampling regimes will be suitable for sampling other tropical and warm-temperate
coastal environments.
61
CHAPTER 4 - Calibration of pelagic stereo-BRUVs and scientific longline surveys for sampling sharks
4.1 Abstract
1. Our understanding of the ecology of sharks and other highly mobile marine
species often relies on fishery-dependent data or extractive fishery-independent
techniques that can result in catchability and size-selectivity biases. Pelagic Baited
Remote Underwater stereo-Video systems (pelagic stereo-BRUVs) provide a
standardised, non-destructive and fishery-independent approach to estimate
biodiversity measures of fish assemblages in the water column. However, the
performance of this novel method has not yet been assessed relative to other standard
sampling techniques.
2. We compared the catch composition, relative abundance and length distribution of
fish assemblages sampled using pelagic stereo-BRUVs and conventional scientific
longline surveys. In particular, we focused on sharks of the family Carcharhinidae
(requiem sharks) to assess the sampling effectiveness of this novel technique along a
latitudinal gradient off the coast of Western Australia. We calibrated the sampling
effort required for each technique to obtain equivalent samples of the target species
and discuss the advantages, limitations and potential use of these methods to study
highly mobile species.
3. The proportion of sharks sampled by pelagic stereo-BRUVs and scientific
longline surveys was comparable across the latitudinal gradient. Carcharhinus
plumbeus was the most abundant species sampled by both techniques. Longline
surveys selected larger individuals of the family Carcharhinidae in comparison to the
length distribution data obtained from pelagic stereo-BRUVs. However, the relative
abundance estimates (catch per unit of effort) from the pelagic stereo-BRUVs were
comparable to those from 5 to 30 longline hooks.
4. Pelagic stereo-BRUVs can be calibrated to standard techniques in order to study
the species composition, behaviour, relative abundance and size distribution of
62
highly mobile fish assemblages at broad spatial and temporal scales. This technique
offers a non-destructive fishery-independent approach that can be implemented in
areas that may be closed to fishing, and is suitable for studies on rare or threatened
species.
4.2 Introduction
Emerging technologies are providing new options for cost-effective ecological
sampling. These technical advances dramatically increase the opportunities for in
situ ecological and behavioural research in vast and remote environments such as the
open ocean (Murphy and Jenkins, 2010). However, in order to understand the
potential of novel techniques it is necessary to compare and calibrate them against
traditional methods.
Studying the ecology and assessing the status of sharks is challenging due to their
generally high mobility, ontogenetic shifts in habitat preference and broad
geographic range (Dulvy et al., 2008). Our understanding on the biology and ecology
of sharks and other highly mobile marine species largely relies on fishery-dependent
data from commercial and recreational fisheries (Myers and Worm, 2003). The use
of fishery-dependent data alone can lead to sampling biases due to gear selectivity
and heterogeneous fishing effort that discriminate among species and habitats
(Murphy and Jenkins, 2010; Simpfendorfer et al., 2002). Alternatively, fishery-
independent surveys use more robust sampling designs, but often employ the same
commercial fishing gear (e.g. longlines, gillnets, trawls) and as such, catchability and
size-selectivity biases remain (McAuley et al., 2007a).
Scientific longline surveys are among the most commonly used fishery-independent
methods for studying the demography and ecology of shark populations
(Simpfendorfer et al., 2002). These surveys provide measures of relative abundance,
sex ratio and length distribution of a range of shark species (McAuley et al., 2007b).
Additionally, longlines allow the collection of samples for population biology
studies (e.g. genetics, age, growth, reproduction, diet) and the deployment of
conventional and electronic tags (Meyer et al., 2010). However, in order to obtain
length or biomass data from longline surveys, sharks must be caught, retrieved and
handled out of the water. The handling of sharks aboard research vessels aims to
63
maximise survival of individuals, but all captured fish are exposed to varying
degrees of physiological stress and physical trauma that can induce pre- or post-
release mortality (Skomal, 2007). Consequently, these extractive techniques may not
be suitable for sampling rare or threatened species, or used in areas that are closed to
fishing (Murphy and Jenkins, 2010).
Baited Remote Underwater Video systems (BRUVs) provide an alternative
standardised, non-extractive and fishery-independent approach that is widely used to
estimate biodiversity indices and relative abundance measures of a range of marine
species (Cappo et al., 2003; Langlois et al., 2012a; Santana-Garcon et al., 2014c),
including sharks (Brooks et al., 2011; Goetze and Fullwood, 2013; White et al.,
2013). This technique uses bait to attract individuals into the field of view of a
camera so that species can be identified and individuals counted (Dorman et al.,
2012). When stereo-camera pairs are used, precise length measurements can be made
and biomass estimated (Harvey et al., 2010). Pelagic stereo-BRUVs are a novel
method that sets camera systems at a predetermined depth in the water column as
opposed to the commonly used benthic deployment where stereo-BRUVs are set on
the seafloor. This deployment design allows pelagic stereo-BRUVs to estimate the
composition, relative abundance and length distribution of fish assemblages that
inhabit the water column (Heagney et al., 2007; Santana-Garcon et al., 2014a).
Methodological comparisons assist in validating the utility of innovative sampling
methods and to understand the advantages and limitations of different techniques.
The use of benthic BRUVs to survey demersal fish assemblages has been compared
to scientific fishing surveys including trawl (Cappo et al., 2004), trap (Harvey et al.,
2012c), hook and line (Langlois et al., 2012b), and longline (Ellis and Demartini,
1995). Additionally, benthic BRUVs and scientific longline surveys were compared
to estimate the diversity and relative abundance of sharks in the Bahamas (Brooks et
al., 2011). These studies found that the species composition determined with baited
video techniques differed to the catch of trawls (Cappo et al., 2004) but it was
comparable to some extent to the catch of traps and longlines (Brooks et al., 2011;
Harvey et al., 2012c). Estimates of relative abundance differed among techniques,
especially for rare species (Brooks et al., 2011; Harvey et al., 2012c). However,
differences in length distribution of species taken in traps and hooks or recorded on
stereo-BRUVs were not biologically significant (Langlois et al., 2012b). Pelagic
64
stereo-BRUVs have been used to assess pelagic fish assemblages (Santana-Garcon et
al., 2014a) but their performance has not been assessed relative to other sampling
techniques.
The present study aims to compare the catch composition, relative abundance and
length distribution of fish sampled by pelagic stereo-BRUVs and conventional
scientific longlines. The sampling effort required for each technique to obtain
equivalent relative abundance samples is determined for each of the target species. In
particular, surveys were conducted along a latitudinal gradient off the coast of
Western Australia to target sharks of the family Carcharhinidae, commonly known
as requiem sharks. Given that longlines used large hooks to target sharks, we
hypothesised that the methods would differ in total catch composition with pelagic
stereo-BRUVs providing data on a broader range of species. However, we expect
that both methods would generate comparable estimates of relative abundance and
length distribution for the targeted shark species. Finally, the advantages and
limitations of pelagic stereo-BRUVs and scientific longline surveys to study highly
mobile species are discussed.
4.3 Methods
4.3.1 Sampling technique
We conducted a longline and pelagic stereo-BRUVs survey in August 2012 at 10
sites over 950 km along the coastline of Western Australia (Figure 4.1). Sites were
15 to 80 km offshore at depths ranging between 35 and 106 metres, although most
sites were 40 to 60 metres deep. Data were recorded from 31 pelagic stereo-BRUVs
and 31 scientific longline deployments targeting requiem sharks. Three replicate
deployments of each method were conducted simultaneously at each site, with the
exception of one site in the Houtman Abrolhos Islands where four replicates of each
technique were undertaken. The number of replicates for each method was limited
by logistical constraints of this research expedition. During deployment, both
methods were interspersed following a straight line with a separation of at least one
kilometre between deployments of either method to avoid or minimize potential
overlap of bait plumes and, to reduce the likelihood of fish moving between
replicates.
65
Figure 4.1 Location of study sites along the coast of Western Australia.
Scientific longline surveys
Scientific longline surveys were conducted as part of the annual shark monitoring
and tagging program of the Department of Fisheries (Western Australia). Surveys
were designed to target requiem sharks. The longlines were 500 m in length and
comprised ~50 J-shaped hooks of size 12/0 baited with sea mullet (Mugil cephalus;
half a fish per hook) and attached to the main line via 2 m metal snoods (Figure
4.2a). Lines were designed to hold hooks approximately 8 metres above the seafloor.
However, hooks near the ballast remained closer to the bottom; this was confirmed
during retrieval as fragments of benthos such as sponges were occasionally caught
(Santana-Garcon pers. obs.). Longlines were set before dawn at ~5 am and soak time
ranged between 2.5 and 6 hours, depending on the time required for retrieval and
processing of the catch. Upon retrieval, all individuals caught were identified to the
66
species level and their fork length (FL) was measured. Catch per unit of effort
(CPUE), a measure of abundance where catch is standardised across deployments of
different sampling effort, was calculated as the catch of each longline divided by the
soak time (hours) and the number of hooks used. In the present study, we defined
CPUE10 as the catch per hour per 10 hooks as a measure to facilitate comparison
between methods (the number of hooks was chosen on the basis of results presented
herein).
Figure 4.2 Deployment design of the (a) scientific longline shots and (b) pelagic stereo-BRUVs used to sample requiem sharks.
Pelagic stereo-BRUVs
Pelagic stereo-BRUVs were adapted to match the deployment characteristics of
longlines so that both methods sampled at similar depths, for equal periods of time
and using the same bait. During the deployment of pelagic stereo-BRUVs, cameras
67
were placed in the mid-water, approximately 8 to 10 metres above the bottom
(Figure 4.2b). This technique uses ballast and sub-surface floats in order to anchor
the systems, enabling control over the deployment depth and reducing movement
from surface waves (Santana-Garcon et al., 2014a). The camera systems consisted of
two Sony CX12 high definition digital cameras mounted 0.7 m apart on a steel frame
and converged inwards at 8 degrees to allow the measurement of fish length (Harvey
et al., 2010). The bait consisted of 1 kg of mullet (Mugil cephalus; fish cut in halves)
in a wire mesh basket suspended 1.2 m in front of the cameras. As for longlines,
camera deployments were set before dawn, at 5 am in the morning, and soak time
ranged between 2.5 and 6 hours depending on the time required for longline
retrieval. Videos were analysed for the full length of the deployment. A blue light
(wavelength 450-465 nm) was fitted on the frame, between the cameras, in order to
illuminate the field of view during the sampling hours before dawn. Blue light
wavelength is thought to be below the spectral sensitivity range for many fish
species (Von Der Emde et al., 2004), and therefore, it is expected to have minimal
impact on fish behaviour (Harvey et al., 2012b).
4.3.2 Video analysis
Stereo-camera pairs were calibrated before and after the field campaign using CAL
software (SeaGIS Pty Ltd) following Harvey & Shortis (1998). The video images
obtained from pelagic stereo-BRUVs were analysed using the software
‘EventMeasure (Stereo)’ (SeaGIS Pty Ltd). All fish observed were quantified,
identified to the lowest taxonomic level possible and measured. However, for this
study, small pelagic fish species in the family Clupeidae and small carangids from
the genus Decapterus, Selar and Selaroides among others were excluded from the
analysis. A conservative measure of relative abundance, MaxN, was recorded as the
maximum number of individuals of the same species appearing in a frame at the
same time. MaxN avoids repeat counts of individual fish re-entering the field of view
(Cappo et al., 2003; Priede et al., 1994). MaxN per hour was used in order to
standardise sampling effort across all deployments due to variable soak times.
Length measurements (FL) were made from the stereo-video imagery for each
individual within 7 metres of the camera system recorded at the time of MaxN.
Individuals must be measured when their body is straight which can be difficult for
68
sharks given their swimming behaviour, as such, in order to improve the accuracy of
shark measurements, the length of each individual was determined from an average
of five measurements obtained in different video frames (Harvey et al., 2001a).
4.3.3 Statistical analysis
Comparison of catch composition
Differences in species composition between scientific longlines and pelagic stereo-
BRUVs were tested using one-way univariate permutational analysis of variance
(PERMANOVA; Anderson et al., 2008). Proportional data facilitates the
comparison of composition patterns sampled by each method as it standardises all
samples to the same scale (Jackson, 1997). Hence, for each of the five species of
requiem sharks recorded, we used proportional data to emphasise the contribution of
each species to the total number of individuals caught per deployment and method.
Proportional data were calculated from CPUE data across all replicates and were
arcsine transformed to normalise possible binomial distributions (Zar, 1999).
Euclidean distance was used to generate the dissimilarity matrices (Anderson et al.,
2011b), P-values were obtained using permutation tests (9,999 permutations) for
each individual term in the model and, Monte Carlo p-values were used to interpret
the result when the number of unique permutations was less than 100 (Anderson,
2001). Data manipulation and graphs across the study were undertaken using the
packages ‘reshape2’(Wickham, 2007), ‘plyr’ (Wickham, 2011) and
‘ggplot2’(Wickham, 2009) in R (R Core Team, 2013).
Catch comparison along a latitudinal gradient
The ability of the two methods to describe spatial patterns along a latitudinal
gradient (32° to 24° S) was compared. For each of the target species, a one-way
analysis of covariance (ANCOVA) was conducted with method as a factor and
latitude as a covariate. A significant interaction between latitude and method would
indicate that the methods were not comparable across the latitudinal range. Analyses
were based on Euclidean distance resemblance matrices calculated from arc sine
transformed proportional data. Statistical significance was tested using 9,999
permutations of residuals under a reduced model.
69
Equivalence of sampling effort
For the target species, the equivalent longline and pelagic stereo-BRUVs sampling
effort was determined by performing a series of statistical tests on the abundance
estimates obtained from BRUVs (MaxN per hour) and from a range of longline
effort data sets (1-50 hooks). Random samples of our data were taken with
replacement and the differences between methods were tested using univariate
PERMANOVAs based on Euclidean distance resemblance matrices of the raw
CPUE data, with method as a fixed factor (Anderson et al., 2011b). P-values were
obtained from 9,999 permutations using the ‘adonis’ function from the ‘vegan’
package (Oksanen et al., 2013) in R. This process was bootstrapped (1000 times) to
generate a distribution of p-values across sampling efforts for the target species.
Additionally, we compared the sampling precision of both techniques at the family
level and for each target species. The precision of a sampling method refers to the
repeatability of its measurements under unchanged conditions, it can be expressed
numerically by measures of imprecision like standard deviation, variance and most
commonly, as a ratio of the standard error (SE) and the mean (Andrew and
Mapstone, 1987). Here, we estimated precision (p) as p = SE / Mean, where the
mean and standard error were obtained from the abundance per deployment for each
sampling technique.
Comparison of length distributions
For the target species, length distributions obtained from pelagic stereo-BRUVs and
longline surveys were compared using kernel density estimates (KDE). The KDE
method is sensitive to differences in both the shape and location of length
distributions (Sheather and Jones, 1991). KDE analyses were conducted using the R
packages ‘KernSmooth’ (Wand, 2013) and ‘sm’ (Bowman and Azzalini, 2013)
following the method described by Langlois et al. (2012b). For each species, the
statistical analysis between the pairs of length distributions collected by each method
was based on the null model of no difference and a resulting permutation test. The
statistical test compared the area between the KDEs for each method to that resulting
from permutations of the data into random pairs. To construct the test, the geometric
mean between the bandwidths for stereo-BRUVs and longline data were calculated
70
(Bowman and Azzalini, 1997). If the data from both methods have the same
distribution, the KDEs should only differ in minor ways due to within population
variance and sampling effects (Langlois et al., 2012b). The ‘sm.density.compare’
function in the ‘sm’ package was used to plot the length distributions where the
resulting grey band shows the null model of no difference between the pair of KDEs.
4.4 Results
Comparison of catch composition
Scientific longline surveys used 1,671 baited hooks (125 hours) and caught 236
individuals of 18 different species. Pelagic stereo-BRUVs recorded 123 hours of
video in 31 deployments with a total of 124 individuals of 20 species identified. The
numerous small pelagic fish (TL < 250 mm) observed in the video were not included
in the species count or in the analyses. Teleost species were almost exclusively
sampled by pelagic stereo-BRUVs, while the semi-pelagic sharks were sampled by
both methods (Figure 4.3). Due to the deployment design of longlines, a proportion
of the hooks adjacent to the ballast were set close to the bottom and, consequently,
benthic sharks were almost exclusively sampled by this method.
The target shark species caught included sandbar Carcharhinus plumbeus, tiger
Galeocerdo cuvier, blacktip C. limbatus/tilstoni and milk Rhizoprionodon acutus
sharks, and Carcharhinus spp*. The latter combines four requiem species that could
not be confidently distinguished across all videos (bronze whaler C.brachyurus,
dusky C.obscurus, spinner C.brevipinna and spot-tail C.sorrah sharks). The common
blacktip C. limbatus and the Australian blacktip C. tilstoni sharks are also combined
here as there are no external morphological features that distinguish these species
(Harry et al., 2012).
For each of the target species, there was no significant difference in the proportion
sampled by either method (P > 0.05; Fig. 4). The sandbar shark C. plumbeus was the
most abundant species for both methods followed by Carcharhinus spp* (Figures
4.3 and 4.4). Using pelagic stereo-BRUVs, the third and fourth most abundant
species were G. cuvier and C. limbatus/tilstoni, whereas with longlines these species
71
were the fourth and third most abundant respectively. Rhizoprionodon acutus was
the least abundant and it was rarely recorded by either method.
Figure 4.3 Mean relative abundance of fish species sampled using scientific longline surveys and pelagic stereo-BRUVs. Catch per unit of effort (CPUE) is shown as catch per hour for 10 hooks (CPUE10) in longline samples and as MaxN per hour in stereo-BRUVs. a Benthic sharks were caught in longlines due to the deployment design setting of a proportion of hooks near the bottom.
Figure 4.4 Mean species proportion of target species sampled using scientific longline surveys and pelagic stereo-BRUVs. P-values show non-significant differences between sampling methods.
72
Catch comparison along a latitudinal gradient
Pelagic stereo-BRUVs and scientific longline surveys showed similar patterns of
abundance for all target species along the 950 km latitudinal gradient (Figure 4.5).
The ANCOVA revealed no significant interaction between method and latitude for
any of the target species (P > 0.05). The species proportional abundance did not
differ significantly between methods whereas latitude had a significant effect on the
distribution of C. plumbeus with a greater abundance present in the northern sites
(Table 4.1). Rhizoprionodon acutus and C. limbatus/tilstoni also showed a
significant effect of latitude in their distribution as they were only recorded north of
Shark Bay, the most northern sampling sites (~24° S). For G. cuvier and the species
complex (Carcharhinus spp*), there was no significant effect of latitude or method.
Figure 4.5 Relative abundance (CPUE) of target species along a latitudinal gradient (32° - 24° S) sampled using scientific longline surveys and pelagic stereo-BRUVs. Trendlines illustrate the ANCOVA result.
Equivalence of sampling effort
PERMANOVA tests on bootstrapped CPUE data and a range of sampling efforts
indicated that the relative abundance of requiem sharks obtained from each camera
system (MaxN per hour) is statistically comparable (P > 0.05) to a sample obtained
73
from 5 to 30 hooks with similarities peaking at 12 hooks (Figure 4.6). This range of
effort equivalence for requiem sharks is largely driven by C. plumbeus, the most
abundant species in this study. The range of equivalence varied among species, for
C. plumbeus effort equivalence ranged between 3 and 30 hooks, with similarities
peaking at 10 hooks. Carcharhinus spp* and G. cuvier showed no significant
difference between methods when MaxN per hour was compared to the catch of 1 to
50 hooks but similarities peaked at 24 and 21 hooks, respectively. Results for C.
limbatus/tilstoni and R. acutus were inconclusive due to the low abundance recorded
with both techniques.
Table 4.1 Summary of ANCOVA tests with method as factor and latitude as covariate. Abundance data was collected with pelagic stereo-BRUVs and scientific longline surveys along an 8-degree latitudinal gradient. P-values in bold are statistically significant.
Latitude Method La x Me Carcharhinus plumbeus <0.001 0.878 0.108 Carcharhinus spp* 0.486 0.291 0.571 Galeocerdo cuvier 0.350 0.226 0.208 Carcharhinus limbatus/tilstoni 0.022 0.682 0.303 Rhizoprionodon acutus 0.003 0.834 0.410
Figure 4.6 Equivalence of sampling effort required to estimate relative abundance of requiem sharks. Plot shows mean p-values from one-way PERMANOVAs testing the differences between pelagic stereo-BRUVs (MaxN per hour) and scientific longline surveys (catch*h-1 *hook-1) across different levels of sampling effort (number of hooks). Values in the shaded area (P < 0.05) are statistically significant. Grey line is shown for reference as the general pooling cut-off (P > 0.25).
74
Precision estimates of pelagic stereo-BRUVs and scientific longline surveys were
similar for the Carcharhinidae family, C. plumbeus, G. cuvier and C.
limbatus/tilstoni (Table 4.2). For Carcharhinus spp* and R. acutus, estimates
obtained from longline surveys were more precise. Note that, as the values of p
decrease, the precision of the sampling technique improves. We found that both
techniques were considerably less precise at sampling uncommon species compared
to the more abundant species. Precision values of 1 indicate that individuals of that
species were only recorded in one deployment.
Table 4.2 Precision estimates for target species sampled using pelagic stereo-BRUVs and scientific longline surveys. Precision (p) was estimated as a ratio of the standard error and the mean abundance per deployment. Note that lower values of p indicate better precision.
p BRUVs p Longlines
Family Carcharhinidae 0.236 0.206 Carcharhinus plumbeus 0.273 0.238 Carcharhinus spp* 0.427 0.270 Galeocerdo cuvier 0.376 0.331 Carcharhinus limbatus/tilstoni 1 0.964 Rhizoprionodon acutus 1 0.736
Comparison of length distributions
There were no significant differences in the shape of the length distributions sampled
with both methods for the family Carcharhinidae and, at the species level, for C.
plumbeus, Carcharhinus spp* and G. cuvier. However, there were significant
differences in the location (i.e. mean length) of the length distributions for the
Carcharhinidae and, at the species level, for C. plumbeus. For these taxa, longline
surveys were more selective of larger individuals (Table 4.3, Figure 4.7). Standard
error bands are wide for those species with small sample sizes; therefore the
interpretation of the results should be undertaken with caution.
4.5 Discussion
We demonstrated that pelagic stereo-BRUVs provide an alternative non-lethal
method of sampling sharks that can be calibrated with standard methods such as
scientific longline surveys. The proportion of Carcharhinidae species sampled by
pelagic stereo-BRUVs and scientific longline surveys was comparable across the
75
study. Pelagic stereo-BRUVs provided a comparable estimate of Carcharhinidae
species that is proportional to longline surveys, whilst also providing abundance
information on other teleost species that were not targeted or captured by longlines
due to the selectivity of the hooks. Longlines sampled a greater proportion of benthic
shark species due to the deployment design that set hooks adjacent to the ballast in
close proximity to the benthos.
Table 4.3 Summary of the lengths of target species measured on pelagic stereo-BRUVs and caught on scientific longline surveys. Maximum (Max), minimum (Min) and mean fork length (FL) are shown in millimetres.
Species Max FL (mm) Min FL (mm) Mean FL (mm)
BRUVs Longline BRUVs Longline BRUVs Longline
Family Carcharhinidae 2937 2870 587 702 1348 1367 Carcharhinus plumbeus 1386 1600 587 730 1089 1280 Carcharhinus spp* 1855 2110 1157 1534 1534 1627 Galeocerdo cuvier 2937 2870 1285 930 2028 1823
The species composition of the Carcharhinidae between the two methods was also
consistent across a broad latitudinal gradient. These findings support previous
studies that define baited video techniques as a suitable, standardised and non-
extractive approach to study the distribution of mobile species across broad spatial
scales (Langlois et al., 2012a; White et al., 2013). In the current study, C. plumbeus
was the most abundant requiem shark species captured by both sampling methods. It
was recorded throughout the study area, but in greater numbers at the northern sites.
Galeocerdo cuvier and the Carcharhinus spp* complex also occurred throughout the
study area, and showed no significant pattern along the latitudinal gradient for either
method. Carcharhinus limbatus/tilstoni and R.acutus were only recorded in the most
northern sites.
Each pelagic stereo-BRUV system yielded equivalent relative abundance estimates
for requiem sharks to that of 5 to 30 hooks in scientific longlines. Effort equivalence
between techniques peaked at 12 hooks for requiem sharks and, at the species level,
effort equivalence peaked at 10, 21 and 24 hooks for C. plumbeus, G. cuvier and C.
spp*, respectively. Due to logistic constraints we could only deploy one camera
system for every longline (50 hooks). Although the target species composition and
relative abundance derived from these techniques were comparable, in absolute
terms longlines caught a greater number of individuals of the target species (159)
76
than those recorded at MaxN on BRUVs (36). In addition, it should be noted that the
methods differed in the area covered and in the amount of bait used. Longline shots,
with a length of 500 m each and more than 10 times the amount of bait in the water,
have a greater ability to attract or encounter fish than a single baited camera system
(Brooks et al., 2011). Thus, increasing sampling effort of the non-extractive pelagic
stereo-BRUVs to approximately one camera deployment for every 10 to 24 hooks is
recommended to exert a sampling effort equivalent to the commonly used scientific
longline surveys.
Precision estimates for both techniques were similar at family level and for the most
abundant target species. Precision is most affected by sampling effort; thus
increasing replication would rapidly enhance sampling precision (Andrew and
Mapstone, 1987). In this study, the number of deployments of pelagic stereo-BRUVs
per site was limited by the complexity of using two methods at once. However,
future studies using this technique could deploy more camera systems
simultaneously, which would rapidly boost replication without added field-time cost
(Santana-Garcon et al., 2014a) and, in turn, it would rapidly enhance their sampling
precision.
Figure 4.7 Comparison of kernel density estimate (KDE) probability density functions for the length distributions of requiem shark species caught by pelagic stereo-BRUVs and scientific longline surveys. Grey bands represent one standard error either side of the null model of no difference between the KDEs for each method.
77
Stereo-BRUVs remove biases due to gear selectivity, such as hook size, that are an
undesired by-product of conventional fishing methods (Cappo et al., 2003). In the
present study, longline surveys were selective towards larger individuals of the
family Carcharhinidae in comparison to pelagic stereo-BRUVs. At the species level,
size selectivity was only significant for C. plumbeus, but KDE tests for other shark
species might lack power to detect differences between methods due to the small
sample sizes available (N < 50) (Bowman and Azzalini, 1997). Mean fork length of
C. plumbeus (1280 mm) and C. spp* (1627 mm) recorded from longline surveys
were larger than those recorded from stereo-BRUVs (1089 and 1534 mm).
Conversely, tiger sharks were on average larger on stereo-BRUVs (2028 mm) in
comparison to longlines (1823 mm). Previous studies have shown species-specific
differences in the length distributions of fish sampled with stereo-BRUVs and line
fishing (Langlois et al., 2012b), or traps (Harvey et al., 2012c) but the differences
reported were not biologically significant. In the present study, low replication was a
major limitation in the analysis of length distributions; hence, further research is
needed to continue exploring the differences between size selectivity of longlines
and pelagic stereo-BRUVs.
Video techniques have proven to be non-intrusive, causing no physical trauma or
physiological stress to the individuals recorded (Brooks et al., 2011). Despite
attempts to minimise the impact on sharks caught in longline surveys, mortality does
occur and the level of post-release mortality is not known (Skomal, 2007). The non-
destructive nature of stereo-BRUVs allows for deployment in fragile and protected
areas and reduces the negative effects of extractive gears when targeting rare and
threatened species (White et al., 2013). Additionally, remote video techniques
provide a permanent record of species behaviour in their natural environment
(Santana-Garcon et al., 2014c; Zintzen et al., 2011). A recurrent behaviour across
shark species observed during video analysis was that individuals were first observed
far from the camera system but remain in the area patrolling the bait source. They
approach the bait in a cautious manner over time. This behaviour suggests that
longer soak times facilitate the recognition of individual features including species,
sex or external markings. Nonetheless, this territorial behaviour could also prevent
other individuals of the same or other species from approaching the cameras, which
78
could affect estimates of species composition and relative abundance (Klages et al.,
2014).
Many species of the family Carcharhinidae are externally similar and visual
identification can be difficult. Identification of individuals to species level from
video alone is the main limitation of pelagic stereo-BRUVs to study requiem sharks
(Santana-Garcon et al., 2014a). Species identification can also be restricted in
fishery-dependent methods, species may be misidentified or pooled under general
categories (Walker, 1998). Identifying features of requiem sharks are often subtle
and the most important of these are tooth shape and numbers, position of the dorsal
fins, colour, and the presence or absence of an interdorsal ridge. These features can
be difficult to assess during the rapid processing of sharks caught on longlines and
this is exacerbated when using remote video techniques. Although most species
could be distinguished on video when individuals come close to the cameras,
identification of some species across all replicate videos may not be possible.
Another constraint of video techniques, although not assessed in this study, are
limitations to identifying the sex of individuals. Claspers of mature males were often
visible, but identification of females and young males with uncalcified claspers was
more challenging. The lack of this information could limit the use of video
techniques in studies of intra-species demographics (Brooks et al., 2011). However,
advances in high-definition digital video and automation of the identification of key
morphological characteristics could improve the rates of identification of species,
sex and even discrimination between individuals (Harvey et al., 2010; Shortis et al.,
2013).
This assessment of the novel pelagic stereo-BRUVs and its comparison to the
commonly used scientific longline surveys provides a better understanding of the
strengths and limitations of each technique. The two methods produce comparable
estimates of relative abundance and species composition for requiem sharks, and the
choice of sampling technique in future should depend on the specific aims of the
study. Scientific longline surveys continue to be a more appropriate approach for
research targeting species that could not be confidently identified on video, or
studies on population biology that require finer intra-specific information such as sex
ratio or reproduction information, the collection of tissue samples (e.g. genetic and
isotopic analyses), or the implantation of conventional or electronic tags (McAuley
79
et al., 2007b). Stereo-BRUVs, however, provide a suitable sampling method that can
be calibrated to standard techniques for studies with broad spatial and temporal
scales, directed at questions of species composition, behaviour, relative abundance
and size distribution of fish assemblages (Langlois et al., 2012a; Santana-Garcon et
al., 2014c; Watson and Harvey, 2009), including highly mobile species (Brooks et
al., 2011; Santana-Garcon et al., 2014a; White et al., 2013). Furthermore, studies
conducted on rare or threatened species, and in areas that are closed to fishing might
require a non-intrusive approach like baited video techniques (White et al., 2013).
Our study demonstrated that pelagic stereo-BRUVs can provide comparable
information to longline surveys on the relative abundance and size composition of
requiem sharks, and determined the required sampling effort to calibrate both
methods.
81
CHAPTER 5 - Effects of a spatial closure on highly mobile fish species: an assessment using pelagic stereo-BRUVs
5.1 Abstract
The effects of a spatial area closure on pelagic fish assemblages within the Houtman
Abrolhos Islands were assessed using mid-water pelagic stereo-BRUVs. The spatial
area closure within the Easter Group of the Houtman Abrolhos Islands was found to
have no significant effect on the species composition and relative abundance of
pelagic fish assemblages. The most abundant demersal target species recorded was
pink snapper (Chrysophrys auratus) and, individuals measured within the spatial
closure were significantly larger than those sampled in the area open to fishing. This
spatial area closure is of moderate size (22.3 km2), but the spatial management of
highly mobile species may require larger area closures than those for reef-associated
species. The monitoring of pelagic species both in large and small spatial area
closures is required in order to better understand how mobile species respond to this
management strategy. Some species were only recorded in relatively low numbers
using pelagic stereo-BRUVs. Moreover, some of the highly mobile pelagic fish
species, such as tunas, mackerel and some shark species proved difficult to measure,
as these species were observed furthest from the camera systems. While pelagic
stereo-BRUVs are an effective fishery-independent approach to monitor spatial
fishing closures, improving the power of replicates by pooling individual
deployments, and increasing the attraction rate of pelagic fish to the stereo-cameras,
will enhance their performance in future studies.
5.2 Introduction
Spatial area closures such as marine reserves, fish habitat protection areas and other
forms of spatial fishing closures have been widely implemented around the world in
coastal areas to protect a broad range of species from exploitation and for
biodiversity conservation (Dichmont et al., 2013; Edgar et al., 2014; Russ and
Alcala, 2011). However, species with different life histories and ecological traits are
82
likely to respond differently to protection (Claudet et al., 2010). There has been
extensive work undertaken on assessing the effects of spatial area closures on
demersal or sedentary species (Babcock et al., 2010; Ballantine, 2014; Claudet et al.,
2008). However, the methods used have often missed or not targeted the mid-water
and, consequently, the response of mobile pelagic species to spatial closures is the
subject of ongoing debate (Claudet et al., 2010; Davies et al., 2012; Game et al.,
2009; Gruess et al., 2011; Kaplan et al., 2010). Highly mobile species are not
expected to respond positively to discrete spatial fishing closures because these are
often of limited size and theoretical evidence suggests that effective protection
requires the areas closed to fishing to be larger than the home range of the
individuals targeted (Gruess et al., 2011; Palumbi, 2004; Walters et al., 2007).
However, recent studies provide evidence suggesting that mobile species may
benefit from spatial area closures (Bond et al., 2012; Claudet et al., 2010; Goetze and
Fullwood, 2013; Jensen et al., 2010; Knip et al., 2012; Pichegru et al., 2010).
Spatial management of pelagic environments in the open ocean, such as closing
areas of the high seas to fishing, is being increasingly discussed (de Juan and
Lleonart, 2010; Game et al., 2009; 2010; Grantham et al., 2011; Hyrenbach et al.,
2000; Kaplan et al., 2010; Mills and Carlton, 1998; Norse, 2005; Sumaila et al.,
2007; White and Costello, 2014). Simultaneously, large offshore marine spatial area
closures are also increasingly being established around the world (Davies et al.,
2012; Kaplan et al., 2014; Notarbartolo-Di-Sciara et al., 2008; Sheppard, 2010). The
main challenges for the implementation of pelagic spatial area closures include: (1)
the need for large closures in order to cover the broad home ranges of pelagic species
(Kaplan et al., 2010); and (2) difficulties associated with compliance because the
majority of pelagic environments (64%) occur outside national jurisdictions making
enforcement difficult and expensive (Sumaila et al., 2007; White and Costello,
2014). Moreover, pelagic fish hotspots can be difficult to identify and are likely to
vary in space and time (Hyrenbach et al., 2000; Norse, 2005). Research on pelagic
ecosystems is often data poor and there is a lack of demographic and baseline data
for many pelagic species (Claudet et al., 2010; Freon and Misund, 1999; Worm et
al., 2005). Research on pelagic fish presents difficulties for the collection and
interpretation of survey data, and often relies exclusively on fisheries catch data that
83
can lead to sampling biases and is not suitable for sampling in areas that are closed to
fishing (Heagney et al., 2007; Ward and Myers, 2007).
Spatial monitoring, both inside and outside spatial area closures, is needed to
quantify the effect of protection on fish assemblages (Murphy and Jenkins, 2010). It
is essential to improve our capacity to monitor changes in marine communities, and
particularly in pelagic ecosystems, using non-destructive and fishery-independent
techniques where applicable (Claudet et al., 2010). Combining traditional sampling
methods and emerging technologies like remote sensing, echo-sounder transects and
Baited Remote Underwater stereo-Video systems (stereo-BRUVs) has been
discussed as being the most effective way to improve the quality of data for spatial
management (Murphy and Jenkins, 2010). For instance, pelagic stereo-BRUVs may
have the potential to monitor the effects of spatial area closures on pelagic fish
(Claudet et al., 2010; Heagney et al., 2007; Santana-Garcon et al., 2014a; Santana-
Garcon et al., 2014d), in the same way that stereo-BRUVs have been widely used to
study such effects on demersal fishes (Goetze et al., 2011; Harvey et al., 2012b;
McLean et al., 2010; Watson et al., 2009).
Baited video techniques are increasingly used to obtain estimates of biodiversity,
relative abundance, behaviour and, when stereo-cameras are used, size and biomass
of a range of marine species (Bailey et al., 2007; Harvey et al., 2013; Harvey et al.,
2012c; Langlois et al., In press; Mallet and Pelletier, 2014; Santana-Garcon et al.,
2014c; White et al., 2013). This method uses bait to attract individuals to the field of
view of the cameras so that individuals can be counted, identified and accurately
measured (Dorman et al., 2012; Hardinge et al., 2013). BRUVs have proven to be a
robust method for assessing fish community structure in deep water (Bailey et al.,
2007; Zintzen et al., 2012), estuaries (Gladstone et al., 2012), tropical or temperate
reefs (Harvey et al., 2012b; Harvey et al., 2012c; Langlois et al., 2010; Unsworth et
al., 2014) and the pelagic environment (Heagney et al., 2007; Santana-Garcon et al.,
2014a; Santana-Garcon et al., 2014d). As opposed to the commonly used benthic
deployments set on the seafloor, pelagic stereo-BRUVs set the camera systems at a
predetermined mid-water depth, which allows the study of pelagic and mobile fish
assemblages that inhabit the water column (Santana-Garcon et al., 2014a).
84
Monitoring the response of mobile species to spatial management in coastal areas
can contribute to understanding the effects of protection in larger offshore spatial
area closures and improve the management of pelagic species in the future. Studying
these effects in nearshore spatial area closures is logistically and economically more
feasible than monitoring large offshore fishing closures (Edgar et al., 2014; Sumaila
et al., 2007). For instance, the Houtman Abrolhos Islands, located off the mid-west
coast of Western Australia, comprise a series of well studied spatial area closures
established to protect valuable and vulnerable reef fish species (Dorman et al., 2012;
Fitzpatrick et al., 2013; Harvey et al., 2012b; McLean et al., 2010; Nardi et al., 2004;
Sumner, 2008; Watson et al., 2007; Watson et al., 2009; Watson et al., 2010). The
Abrolhos are comprised of 4 island groups (North Island, Wallabi group, Easter
group, and Pelsaert group) and each group has one area closed to fishing (these
spatial area closures are termed Reef Observation Areas; ROAs). These areas have
been closed to scalefish fishing since 1994, thus catching fish by line, spear or any
other method is prohibited, but lobster pots are permitted. Periodic monitoring of the
effects of protection on the fish assemblages and, particularly, on the demersal target
species has shown a variety of responses over the years including an increase in
abundance (Harvey et al., 2012b; Nardi et al., 2004; Shedrawi et al., 2014; Watson et
al., 2007), increase in average fish size or biomass due to protection (McLean et al.,
2010; Watson et al., 2009), or no significant difference between the assemblages
inside and outside the fishing closures (Dorman et al., 2012). However, none of the
studies in the area have targeted their sampling to the pelagic environment, thus the
effect of these spatial area closures on highly mobile species remains largely
unknown.
We sampled fish assemblages in the water column in areas open and closed to
fishing using pelagic stereo-BRUVs. The aims of this study were to (1) explore the
effects of protection on pelagic species in the Houtman Abrolhos Islands; (2) assess
the potential of pelagic stereo-BRUVs as a monitoring technique for spatial area
closures; and (3) contribute to the ongoing debate on the use of spatial area closures
for the conservation and management of highly mobile species both in coastal waters
and the high seas.
85
5.3 Methods
5.3.1 Study area
The Houtman Abrolhos Islands are located 60 km offshore from the mid-west coast
of Western Australia between 28° 15’ S and 29° 00’ S. In this study, pelagic stereo-
BRUVs were used to survey fish assemblages in the mid-water at 6 sites in the
Easter group (Figure 5.1). Sites were selected haphazardly, off the reef edge (in the
pelagic environment above the base of the reef slope) between 30 and 35 m deep.
Three sites were within the area closed to fishing and three in areas where fishing is
permitted. The Easter group ROA covers 22.29 km2 and is the second largest spatial
area closure at the Houtman Abrolhos Islands.
Figure 5.1 Location of (A) the Houtman Abrolhos Islands and (B) the study sites inside and outside the spatial area closure at Easter Group. The area shaded in light grey shows the Reef Observation Area closed to fishing since 1994.
5.3.2 Sampling technique
The pelagic stereo-BRUVs used in this study were designed to be deployed,
anchored and to remain at ~10 m depth in the water column (Figure 5.2). The bait
arm acts as a rudder and keeps the camera system facing downstream of the current.
86
The use of ballast and sub-surface floats effectively reduces movement from surface
waves and allows for control over the deployment depth. A full description of the
deployment method is provided in Santana-Garcon et al. (2014a). The camera
systems consisted of two Sony CX12 high definition digital cameras mounted 0.7 m
apart on a steel frame and converged inwards at 8 degrees to allow the measurement
of fish length (Harvey et al., 2010). The bait consisted of 800 g of crushed pilchards
(Sardinops sagax) in a wire mesh basket suspended 1.2 m in front of the cameras.
Figure 5.2 Deployment design of pelagic stereo-BRUVs to sample fish assemblages in the water column.
5.3.3 Experimental design
The experimental design consisted of 2 factors: Status (2 levels, fixed: Open and
Closed to fishing) and Site (3 levels, nested in Status, random). Twelve replicate 2-
hour deployments were conducted at each site. This sampling effort was considered
suitable following the results obtained from Santana-Garcon et al. (2014a) that
estimated the optimal soak time and replication required for sampling using pelagic
stereo-BRUVs. Over 4 days of sampling, 72 mid-water stereo-camera systems were
deployed allowing a minimum distance of 500 m between replicates sampled
87
simultaneously, thus minimizing the potential for overlap of bait plumes and to
reduce the likelihood of fish moving between replicates. Further research on bait
plume dispersion in the mid-water is needed in order to confirm the minimum
distance that should be allowed between deployments and to estimate the sampling
area (Heagney et al., 2007; Rizzari et al., 2014). Samples were collected during
daylight hours, allowing at least 1 h between sampling and sunrise or sunset, to
avoid possible crepuscular variation in fish assemblages (Birt et al., 2012). Sampling
was interspersed so that all 6 sites were being sampled at the same time in order to
avoid differences between sites being confounded by the temporal variability of
pelagic fish assemblages.
5.3.4 Image analysis
Stereo-camera pairs were calibrated before and after fieldwork using the software
CAL (SeaGIS Pty Ltd). Calibration followed the procedure outlined in Harvey and
Shortis (1998) to allow for accurate length and distance measurements of the fish
sampled. The video images were analysed using the software ‘EventMeasure
(Stereo)’ (SeaGIS Pty Ltd). All fish observed were quantified, identified to the
lowest taxonomic level possible and measured. A conservative measure of relative
abundance, MaxN, was recorded as the maximum number of individuals of the same
species appearing in a frame at the same time. MaxN avoids repeat counts of
individual fish re-entering the field of view (Cappo et al., 2003; Priede et al., 1994).
Schools of small pelagic fish (total length < 200 mm), commonly referred to as
baitfish, were excluded from the analysis.
Range (distance of the fish to the system) and fork length (FL) were measured from
the stereo-video imagery for each individual recorded at the time of MaxN. In order
to obtain quantitative measurements of fish using stereo-video, individuals must be
observed simultaneously in the field of view of the two stereo-cameras. Moreover, to
ensure precision and accuracy of length measurements, fish must be within a
predefined distance and orientation to the camera system (Harvey et al., 2010). In
this study, fish were measured within a range of 8 m and when oriented at more than
45° to the stereo-video system.
88
5.3.5 Data analysis
Multivariate assemblage data were analysed using non-parametric permutational
analysis of variance (PERMANOVA), in the PRIMER-E statistical package (version
6.1.15, Anderson et al., 2008). A two-way PERMANOVA was conducted to test for
differences in fish assemblages sampled in the mid-water at sites open and closed to
fishing. A log5 Modified Gower dissimilarity matrix was chosen for the multivariate
analysis to exclude joint absences and to place emphasis on detecting changes in the
species composition and relative abundance (Anderson, 2006; Anderson et al.,
2011a). Log 5 was considered suitable given that the largest MaxN value in the
dataset was just 65. A total of 9,999 permutations of the data were computed to
obtain P significance values. If the number of unique permutations was less than
100, a Monte Carlo P value was used to interpret the result (Anderson et al., 2008).
Univariate PERMANOVAs were used to test for differences in species richness and
relative abundance of some target species groups between areas open and closed to
fishing. All univariate analyses were conducted using the Euclidean distance
dissimilarity measure on the raw data (Anderson et al., 2011a). The target species
groups chosen for analysis were: mackerels (Grammatorcynus bicarinatus and
species of the genus Scomberomorus), jacks (species of the family Carangidae),
tunas (species of the genus Thunnus), sharks (Galeocerdo cuvier and species of the
genus Carcharhinus) and demersal fishes targeted by commercial and recreational
fishing (Chrysophrys auratus, Glaucosoma hebraicum, Lethrinus nebulosus,
Choerodon rubescens and Plectropomus leopardus). Mackerels, jacks and tunas are
considered to be the pelagic fish most highly targeted by fishers at the Houtman
Abrolhos Islands. The demersal target species group includes reef-associated species
that were observed in the mid-water and are those highly targeted by fishers (Watson
et al., 2009).
The length distributions of species in the areas open and closed to fishing were
compared using kernel density estimates (KDEs). Only species that had length
measurements of at least 10 fish both inside and outside the spatial area closure were
included in this analysis (N>20). The KDE method is sensitive to differences in both
the shape and location of length distributions (Sheather and Jones, 1991). KDE
analyses were conducted using the packages ‘KernSmooth’ (Wand, 2013) and ‘sm’
89
(Bowman and Azzalini, 2013) in R software version 3.0.2 (R Core Team, 2014)
following the method described by Langlois et al. (In press; 2012b). For each
species, the statistical analysis between the pairs of length distributions was based on
the null model of no difference and a resulting permutation test. The statistical test
compared the area between the KDEs for each method to that resulting from
permutations of the data into random pairs. To construct the test, the geometric mean
between the bandwidths for measurements inside and outside the spatial area closure
was calculated (Bowman and Azzalini, 1997). If the data from both status have the
same distribution, the KDEs should only differ in minor ways due to within
population variance (Langlois et al., 2012b). The ‘sm.density.compare’ function in
the ‘sm’ package was used to plot the length distributions where the resulting band
shows the null model of no difference between the pair of KDEs. This shaded band
is centred on the mean KDE and extends one standard error above and below,
thereby indicating which regions of the length-frequency distribution are likely to be
causing any significant differences (Bowman and Azzalini, 2013).
5.4 Results
There was no significant difference in the species composition and relative
abundance of pelagic fish assemblages recorded in areas open and closed to fishing
at the Houtman Abrolhos Islands. The two-way PERMANOVA indicated that
neither status nor site had a significant effect on community structure (P < 0.05;
Table 5.1). From the 72 pelagic stereo-BRUVs deployed, a total of 350 individuals
of 24 taxa were identified representing 10 families. Twenty-eight deployments
(39%) did not record any fish (11 inside and 17 outside the spatial area closure). In
total, there were more individuals recorded within the area ‘closed’ to fishing (212
individuals of 20 taxa), than at the ‘open’ sites where fishing is permitted (138 of 14
taxa). All taxa were confidently identified to the species level except for two taxa
that were grouped at the genus level (Scomberomorus spp and Pseudocaranx spp)
and one taxon that was defined as a species complex (Carcharhinus spp). The latter
combines four shark species of the genus Carcharhinus (bronze whaler
Carcharhinus brachyurus, dusky C. obscurus, spinner C. brevipinna and spot-tail C.
sorrah sharks) that cannot be confidently distinguished from video imagery alone
(Last and Stevens, 2009).
90
Table 5.1 PERMANOVA based on log 5 Modified Gower dissimilarities to test the effects of spatial area closures (Status) on the composition and relative abundance of fish assemblages sampled using pelagic stereo-BRUVs.
Source df MS Pseudo-F P (perm)
Status 1 0.847 0.828 0.528 Site (Status) 4 1.023 1. 079 0.372 Residual 66 0.947 Total 71
Figure 5.3 Total abundance (sum of MaxN) of fish taxa sampled using pelagic stereo-BRUVs in areas open and closed to fishing at the Houtman Abrolhos Islands.
The data recorded using pelagic stereo-BRUVs was sparse and many species were
only recorded in low numbers (between 1 and 88 individuals per species; Figure
5.3). The frequency of occurrence of most species was also low with species
recorded in 1 to 14 deployments throughout the study (Table 5.2). Amberjack
Seriola dumerili was the species with the highest number of individuals sighted and
was recorded in 3 deployments, only within the area closed to fishing. Pink snapper
Chrysophrys auratus (previously known as Pagrus auratus; Gomon et al., 2008) was
the most abundant of the demersal species observed in the mid-water; they occurred
in higher numbers inside the spatial area closure but were also recorded in the area
91
open to fishing (observed in 10 and 4 deployments respectively). All other demersal
species observed (spangled emperor Lethrinus nebulosus, baldchin groper
Choerodon rubescens, dhufish Glaucosoma hebraicum and coral trout Plectropomus
leopardus) were only recorded within the area closed to fishing. Other species that
were only detected in the ‘closed’ sites included yellowtail kingfish Seriola lalandi,
gold-spotted trevally Carangoides fulvoguttatus, skipjack trevally Pseudocaranx spp
and blacktip shark Carcharhinus limbatus/tisltoni. On the other hand, there were a
few species like wahoo Acanthocybium solandri and yellowfin tuna Thunnus
albacares that were only recorded in the areas open to fishing.
Table 5.2 Number of deployments where species were sampled using pelagic stereo-BRUVs inside (Closed) and outside (Open) a spatial fishing closure at the Houtman Abrolhos Islands.
Number of
deployments sighted
Species Open Closed
Acanthocybium solandri 1 0 Carangoides fulvoguttatus 0 2 Carcharhinus limbatus 0 1 Carcharhinus plumbeus 1 1 Carcharhinus spp 5 5 Choerodon rubescens 0 1 Chrysophrys auratus 4 10 Echeneis naucrates 0 2 Eubalichthys caeruleoguttatus 1 3 Galeocerdo cuvier 1 1 Glaucosoma hebraicum 0 1 Gnathanodon speciosus 1 0 Grammatorcynus bicarinatus 2 4 Lethrinus nebulosus 0 5 Plectropomus leopardus 0 1 Pseudocaranx spp 0 1 Remora remora 1 1 Scomberomorus spp 7 7 Selaroides leptolepis 1 0 Seriola dumerili 0 3 Seriola lalandi 0 4 Thunnus alalunga 2 2 Thunnus albacares 1 0 Thunnus tonggol 3 2
Total species richness was higher in the sites closed to fishing than in the fished area
(Figure 5.4A). The univariate statistical analysis found significant differences in
species richness between sites, but differences for status were not significant (Table
92
5.3). The mean relative abundance of mackerels, tunas and sharks was greater in the
fished area while jacks and the demersal target species were more abundant within
the ROA (Figure 5.4B). However, the univariate PERMANOVAs for these species
groups showed that these differences were not statistically significant for status
given the variability between sites.
Figure 5.4. The mean species richness per site (A) and the mean relative abundance per site of the main target groups (B) in areas open and closed to fishing at the Houtman Abrolhos Islands. Error bars show standard error.
Table 5.3 Summary of univariate PERMANOVA tests for species richness and relative abundance of the main target fish groups sampled using pelagic stereo-BRUVs. Analyses were based on Euclidean distance dissimilarities of the raw data to test for differences in Status (areas open and closed to fishing) and Sites. Fields in bold-faced show significant P-values.
Source Species
Richness Mackerels Jacks Tunas Sharks
Demersal target species
Status 0.263 0.581 0.157 0.407 0.409 0.439 Site (Status) P < 0.05 0.446 0.849 P < 0.05 0.804 P < 0.05
Highly mobile pelagic fish like tunas, mackerel and some shark species proved
difficult to measure lengths, as they were the species observed furthest from the
pelagic stereo-BRUVs (Figure 5.5). In contrast, demersal target species and some
reef associated pelagic fish, like the genera Pseudocaranx and Seriola, were
recorded much closer to the camera systems. The mean fork length and mean ranges
were estimated at the genus level (Table 5.4). Fork length was measured for only 131
individuals (37% of the total), 49 of these individuals were measured in the area
open to fishing and 82 within the fishing closure. Additionally, measures of range
93
were estimated for 200 fish (57 %), 65 individuals within the ROA and 135 outside
the spatial area closure. Only pink snapper C. auratus fitted in the criteria to be
included in the KDE analysis (i.e. length measurements available for at least 10 fish
inside and outside the spatial closure). Although the shape of the length distribution
of C. auratus sampled in the areas open and closed to fishing did not differ
significantly (P = 0.99); the location of the length distributions (i.e. the mean length)
were significantly different (P < 0.01). On average, larger pink snapper were
recorded within the spatial area closure (Figure 5.6). However, the low number of
individuals measured both inside and outside the spatial fishing closure for all other
species prevented us from doing further statistical analysis on their length structure
or estimated biomass. Distance of individuals to the camera system was one of the
main factors preventing accurate length estimates.
Table 5.4 Summary of length and range measurements at genus level combining areas open and closed to fishing. The mean length (cm) represents fork length and the mean range (m) is the average distance of individuals to the stereo-camera system. The number of individuals measured [N] is shown in brackets and (*) shows the genus of demersal target species.
Genus Mean Length
(cm) [N] Mean Range
(m) [N] Acanthocybium - 11.10 [1] Carangoides 122.2 [3] 5.05 [4] Carcharhinus 180.4 [5] 9.95 [15] Choerodon* 69.9 [1] 2.21 [1] Chrysophrys* 64.5 [48] 2.35 [56] Galeocerdo 324.5 [1] 6.17 [1] Glaucosoma* - 1.96 [1] Grammatorcynus 72.8 [20] 7.49 [32] Lethrinus* 68.7 [13] 2.27 [14] Plectropomus* - 2.29 [1] Pseudocaranx 35.2 [1] 1.36 [1] Scomberomorus 98.2 [6] 9.80 [25] Seriola 101.4 [26] 3.17 [30] Thunnus 92.7 [4] 10.48 [12]
94
Figure 5.5 The mean range, distance of the individuals to the stereo-camera system, for fish genus recorded using pelagic stereo-BRUVs. Grey silhouettes represent demersal species and black silhouettes are pelagic and highly mobile fish. The shaded area shows the field of view of the cameras where fish length can be measured.
Figure 5.6 Comparison of kernel density estimate (KDE) probability density functions for the length distributions of pink snapper Chrysophrys auratus in areas open and closed to fishing. Grey bands represent one standard error either side of the null model of no difference between the KDEs for each method.
95
5.5 Discussion
We found no significant difference between the species composition and relative
abundance of fish assemblages sampled in the water column inside and outside the
Easter group spatial area closure at the Houtman Abrolhos Islands. Sharks and
targeted pelagic species like mackerels and tuna were recorded in higher numbers
outside the spatial area closure but the effect of the fishing closure was not
statistically significant due to variation between sites. Additionally, jacks mostly of
the genus Seriola, were almost exclusively recorded within the spatial area closure.
Jacks are less mobile than other pelagic fish such as tunas and are known to have
responded positively to spatial protection (Edgar et al., 2014; Russ et al., 2004).
However, we found that the small proportion of deployments where schools of jacks
were recorded prevented this difference from being statistically significant and thus
conclusive.
The lack of significant effect on pelagic fish assemblages could be an indication that
the spatial area closure we studied, which is not representative of all habitat types in
the region and was originally designed to protect reef associated species, might not
be adequate, or large enough, to have an effect on highly mobile species like
mackerels, tunas and sharks. Our results contrast with other studies that have found
that mobile species could benefit from protection, whatever their home-range and
irrespective of the size of the spatial area closure (Claudet et al., 2010). However, it
has been suggested that small area closures could in fact benefit only a fraction of
individuals of the population that may be ‘less mobile’ as a result of intra-specific
differences in movement behaviour (Davies et al., 2012), but such an effect would
not be detected in this study. Moreover, similarities in species composition and
abundance of pelagic fish assemblages inside and outside this spatial area closure
may be attributable to the relatively low fishing pressure targeting pelagic fish
species in the study area in comparison to demersal species. Monitoring pelagic
species in other spatial area closures of a range of sizes and exploitation levels would
improve the interpretation of our results. Furthermore, recording physical parameters
such as current speed and bathymetry, throughout the study area would allow testing
for confounding factors that could be driving the differences in fish assemblages
between sites.
96
Although the differences were again not statistically significant, demersal target
species were, in total, recorded in higher numbers within the spatial area closure.
These results are in line with recent studies in this area that found no significant
difference in the abundance of demersal target species inside and outside the fishing
closure using benthic BRUVs (Dorman et al., 2012). However, pelagic stereo-
BRUVs are not designed to target demersal species and only sampled a fraction of
the demersal species population, mainly large individuals that venture off the reef
into the mid-water attracted to the bait plume. This behaviour might differ between
fish in the areas open and closed to fishing as target species within spatial area
closures can become bolder and respond stronger to bait or diver surveys (Davidson,
2001; Willis et al., 2000). Moreover, pink snapper (C. auratus) measured within the
spatial area closure were significantly larger than those sampled in the area open to
fishing.
From our results, we can infer that pelagic species might not show a positive
response to spatial area closures even if these have proved to be effective for
demersal target species. Hence, future monitoring and management plans should
incorporate sampling techniques that target the pelagic environment when assessing
any change in fish community structure. Most methods used in the pelagic
environment are destructive, target specific species as a result of gear selectivity (e.g.
hook size, bait type) and look at very specific questions or at very broad scales (i.e.
ocean wide). Pelagic stereo-BRUVs can help explore patterns occurring at the
assemblage or community level (Santana-Garcon et al., 2014d). This study provides
baseline data on the pelagic fish diversity and relative abundance at the Houtman
Abrolhos Islands that can be compared to future surveys in the area. Pelagic stereo-
BRUVs have proven to be a robust, non-destructive and fishery-independent method
that can be standardised to study fish assemblages across spatial areas and over time
(Santana-Garcon et al., 2014a; Santana-Garcon et al., 2014d). This method provides
a permanent record of the species in the area and allows for the observation of fish
behaviour in their natural environment (Santana-Garcon et al., 2014c). Moreover,
remote video techniques can be used in areas and at depths beyond the limits of
diver-based studies (Cappo et al., 2006), and sampling can take place both during the
day or at night using lights (Fitzpatrick et al., 2013; Santana-Garcon et al., 2014d).
97
However, the identification of taxa to the species level from video alone has proven
challenging and some species can only be grouped at higher taxonomic levels. This
is a common limitation across video techniques (Harvey et al., 2014; Mallet and
Pelletier, 2014), but it is exacerbated for pelagic species because many taxa share
similar morphological traits (i.e. coloration, streamlined fusiform body, pointy snout
and deeply forked tails), and they have adapted to appear almost invisible in the
pelagic environment (Santana-Garcon et al., 2014a). We also observed that larger
pelagic species often remain further from the camera systems than demersal species,
which complicates both their identification and measurement. We were able to
measure only a small fraction of the individuals recorded, 37 % were measured for
fork length and 57 % for range. The percentage of fish measured in this study using
pelagic stereo-BRUVs is lower than that of other studies using similar stereo-
BRUVs deployed on the seafloor (Harvey et al., 2012c; Langlois et al., 2012b;
Watson et al., 2009). Given the low measurement rate for pelagic species, length
distributions of species inside and outside the spatial area closure could only be
compared for one demersal species, pink snapper (C. auratus). The larger pelagic
species remained further from the stereo-cameras than demersal species, which often
approached the systems and fed on the bait bag (Figure 5.5). This difference in
behaviour between demersal target species and some pelagic species can be in part
attributed to the fact that large pelagic fish are largely visual predators that need to
visually locate their prey. Some shark species (e.g. genus Carcharhinus) are known
to remain far from the camera system initially, but as they “patrol” the area they
come closer to the bait source if the cameras are left in the water for longer soak
times (e.g. soak times of 2 to 6 h, Santana-Garcon et al., 2014d). Therefore,
increasing the attraction rate of pelagic fish to the cameras and adapting the method
for calibration of stereo-video to allow for accurate measurements at a greater
distance from the stereo-cameras, would improve the data in future studies using
pelagic stereo-BRUVs.
Pelagic ecosystems are highly variable both temporary and spatially, so high
replication is required to detect community structure. Moreover, sampling should be
conducted at different times of the year to account for seasonal variability and, when
possible, monitoring should include a temporal component to detect changes of the
fish assemblages over time (e.g. annually or every 5 years). In this study we used 12
98
replicates per site following the recommendations from Santana-Garcon et al.
(2014a). However, the data obtained was sparse and was dominated by zeros, which
could lead to difficulties detecting patterns in the community structure (Clarke et al.,
2006a). We defined a replicate as each single 2-hour deployment of the camera
systems, yet future studies could pool various deployments to obtain more powerful
replicates (e.g. reducing the number of samples with no species recorded). For
instance, if the number of sites was increased, the deployments within sites could be
treated as sub-samples in order to improve the distribution of dissimilarities for
multivariate analyses (this is further developed in a subsequent publication).
Moreover, increasing the attraction of pelagic fish to the cameras would not only
improve species identification and length measurements as suggested, but also
increase the power of each camera deployment. Research on visual, sound and odour
attractants using mid-water baited video techniques has shown that a combination of
attractants increases the number of fish in the field of view and brings individuals
closer to the camera (Rees et al., 2014). Moreover, pelagic species often school
around floating objects and other structures in the mid-water, these are commonly
named fish aggregation devices (FADs) (Dempster, 2005). Small pelagic fish have
been reported to aggregate around pelagic stereo-BRUVs as if attracted to their
physical structure (Santana-Garcon et al., 2014a), thus camera systems could be
further adapted to act as FADs in order to increase the number of fish present in the
area that enter the field of view of the stereo-cameras.
Following the increase in spatial area closures in the open ocean (Davies et al., 2012;
Kaplan et al., 2014) and the push for spatial management of the high seas (Game et
al., 2009; White and Costello, 2014), it is essential that sampling methods adapt to
monitoring in this highly dynamic environment in order to understand the effects of
spatial management on pelagic species. Pelagic stereo-BRUVs have not been tested
in open ocean conditions and further development of the technique would be
required to effectively deploy baited cameras in the high seas. Recent technology
developments including inexpensive camera systems, small and robust designs that
can be attached to commercial fishing gear, automatic data recording when systems
are deployed in the water and a capacity for long deployment times (e.g. months)
(Pember et al., 2014) could all be incorporated into pelagic stereo-BRUV systems.
Adaptation of video techniques to open ocean monitoring could include the
99
integration of such systems to satellite-tracked structures, like the FADs used by
commercial fishermen to target tuna (Davies et al., 2014). There is also progress in
the automation of image analysis including software to detect when fish are in the
field of view, and automatic species identification and measurement of individuals
(Shortis et al., 2013). These developments will facilitate the use of pelagic stereo-
BRUVs in broad temporal and spatial scale studies in order to assess the response of
pelagic species to large oceanic spatial area closures. Moreover, a combination of
fishery-independent sampling methods, including emerging technologies like remote
sensing and echo-sounder transects, would improve the quality of data on pelagic
species (Murphy and Jenkins, 2010).
In conclusion, we found no significant effect of a small spatial area closure on the
mid-water fish assemblages at the Houtman Abrolhos Islands, which supports the
theory that spatial management of highly mobile species may require larger area
closures than those targeting reef-associated species (Kaplan et al., 2010). Our study
provides baseline data for future studies and highlights the need to include
monitoring of pelagic species both in large and small spatial area closures in order to
understand how mobile species respond to these management strategies. We found
that pelagic stereo-BRUVs are an effective fishery-independent approach to monitor
spatial area closures. However, improving the power of replicates by pooling
individual deployments, and increasing attraction of pelagic fish to the stereo-
cameras, will enhance their performance in future studies. Further development and
trial of pelagic stereo-BRUVs will be required to implement this technique in
oceanic environments, which will be critical for monitoring the effects of the
increasing spatial area closures in the high seas. The use of fishery-independent
methods to understand the natural variability of pelagic fish assemblages, which has
been poorly studied in past monitoring programmes (Claudet et al., 2010), will allow
for a better management and conservation of these highly mobile species both in
coastal and open ocean waters.
101
CHAPTER 6 - Presettlement schooling behaviour of a priacanthid, the purplespotted bigeye Priacanthus tayenus (Priacanthidae: Teleostei)
6.1 Abstract
We report in situ behavioural observations of presettlement schooling in Priacanthus
tayenus off Coral Bay, Western Australia collected using pelagic Baited Remote
Underwater Stereo-Video systems. Two groups of fish (8 and 9 individuals) were
observed that aggregated into a single school. Mean total length was 24.1 mm (12.5
– 30.2 mm). The fish swam at a mean speed of 8.5 cm s-1 in a group spacing
themselves more or less evenly at a distance of around one body length from the
nearest neighbour within the school. P. tayenus appeared to be sometimes associated
with juveniles of other species. The results presented here add to the limited, but
growing body of literature on the schooling behaviour of the early pelagic stages of
demersal fishes.
6.2 Introduction
Most marine teleost fish species have a pelagic larval phase during their early life-
history. The processes that occur during that period determine their dispersal and
recruitment, and may influence the distribution of adult populations (Mora and Sale,
2002). Extensive research has focused on the biology of the pelagic phase of
demersal fishes, but direct observations and quantitative data on their behaviour in
the pelagic environment is limited and only available for a small number of species
(Leis, 2010; Leis and Carson-Ewart, 1998; Masuda, 2009; Masuda et al., 2003). Leis
(2010) reviewed the limited information available on the ontogeny of schooling in
the pelagic larvae of marine demersal fishes. Schooling may provide protection from
predators, accelerate learning and enhance swimming and orientation abilities
(Masuda, 2009; Shaw, 1978).
102
Presettlement schooling has been described in a few reef fish families including
Gobiidae (Breitburg, 1991), Mugilidae (Kingsford and Tricklebank, 1991), Mullidae
(Leis and Carson-Ewart, 1998; McCormick and Milicich, 1993), Lutjanidae,
Microdesmidae and Pomacentridae (Leis and Carson-Ewart, 1998). However, the
difficulties associated with studying these small and inconspicuous fish in situ has
resulted in most behavioural studies relying on estimates from fish collected in
plankton nets, purse seines or light traps (McCormick and Milicich, 1993; Ohman et
al., 1998) and on observations of reared larvae or laboratory experiments (Leis and
Carson-Ewart, 1998; Leis et al., 1996; Masuda, 2009).
We report in situ observations of presettlement schooling behaviour in priacanthids
off Coral Bay, Western Australia. The behaviour of juvenile Priacanthus tayenus
(Richardson, 1846) was recorded using pelagic Baited Remote Underwater Stereo-
Video systems (stereo-BRUVs). This technique was being trialled to investigate the
distribution of pelagic fish assemblages as it potentially provides a standardised,
non-destructive and fishery independent approach to estimate abundance, diversity
and length of fish in the pelagic environment (Harvey et al., 2004; Heagney et al.,
2007). Presettlement reef fishes were observed in all 48 video deployments; these
observations present a unique opportunity to describe their behaviour in situ.
Presettlement priacanthids were observed in 5 of the 48 videos, and schools of two
or more fish were recorded on 3 of those 5 deployments. This observation provides
the first quantitative data on their presettlement schooling behaviour.
Priacanthids, commonly called bigeyes, comprise a small circum-tropical family of
marine fishes that are characterised generally by extremely large and bright eyes, a
deep body, rough scales and red coloration (Starnes, 1988). Some species are
commercially valuable in South East Asian trawl fisheries and in reef-based fisheries
throughout the tropics (Lester and Watson, 1985; Seah et al., 2011; Senta, 1978).
The purplespotted bigeye, Priacanthus tayenus, is a small species (maximum total
length 29 cm) characterised by deep purple to inky black spots in the pelvic fins
(Starnes, 1988). Adults are epibenthic, can occur in large schools and inhabit rocky
reefs and open bottom areas at depths between 20 and 200 m. Larvae and the early
juvenile stages are pelagic (Starnes, 1988).
103
Most aspects of the biology of the early life-history stages of priacanthids are largely
unknown, in particular, behaviour during the pelagic dispersal phase. Schooling
behaviour has not been reported for juvenile priacanthids; nonetheless, large
aggregations have been reported around surface lights at night in the West Indies
(Caldwell and Bullis, 1971), the western North Atlantic (Caldwell, 1962a) and from
trawl net catches in the Philippines (Senta, 1978).
This study aims to contribute to the knowledge of presettlement schooling behaviour
in the early stages of Priacanthus tayenus. Accurate length measurements,
swimming speed and schooling parameters are described. Stereo-video techniques
offer unique abilities to study the ontogeny of fish behaviour, and this study presents
the first in situ quantitative data on schooling behaviour of the pelagic early-life
history stage of a demersal fish species.
6.3 Material and methods
The data presented here are based on observations of Priacanthus tayenus in March
2012 at Coral Bay (23° 1’ S; 113° 44’ E), Ningaloo Marine Park. Ningaloo Reef is a
fringing coral reef which stretches for approximately 270 km adjacent to the semi-
arid north-west cape of Western Australia. The pelagic stereo-BRUVs consisted of
two SONY HDR CX12 video cameras mounted 0.7 m apart on a base bar inwardly
converged at 8 degrees to gain an optimised field of view of 7 m (Harvey and
Shortis, 1995). The cameras recorded video imagery at 25 frames per second in
MPEG Transport Stream format (MTS), which was converted to high-definition
MPEG format (Harvey et al., 2010). The system was deployed offshore of the reef
slope in a depth of 35 m, and left moored 3 hours in mid-water, 15 m above the
bottom (Figure 6.1). The bait consisted of ~800 g of pilchards (Sardinops sagax) in a
wire mesh basket suspended 1.2 m in front of the two cameras. The video images
were analysed using the software ‘EventMeasure (Stereo)’ (SeaGIS Pty Ltd, see
http://www.seagis.com.au/event.html). EventMeasure Stereo is a purpose built event
logger which an operator can use to manually record the identification and number
fish and their behaviour at a particularly time on a video sequence. By using a
mouse, an operator can determine the location of a point in three-dimensional space
(x, y and z) relative to the centre of the two cameras. As the time sequence is
104
recorded it is possible to calculate the swimming speeds of a fish. It is also possible
to measure the distance to a fish (Harvey and Shortis, 1995; Harvey et al., 2004) and
make accurate and precise measurements of fish length (Harvey et al., 2001a; b;
2002). Species identification was confirmed by Wayne Starnes (personal
communication) from still images and video clips (Figure 6.2).
Figure 6.1 Deployment method (a) and details (b) of the pelagic stereo Baited Remote Underwater Video system (stereo-BRUVs).
In order to obtain quantitative measurements of fish using stereo-video, individuals
must be observed simultaneously in the field of view of the two stereo cameras.
Furthermore, to ensure precision and accuracy of measurements, fish must be within
a predefined distance and orientation to the camera system (Harvey et al., 2010).
Given the small size of the fish targeted in the present study, individuals were only
measured within 2 metres of the cameras and, whenever possible, when the head and
tail of the fish aligned at more than 50° perpendicular to the stereo-video system.
Only individuals from 3 of the 5 videos that recorded priacanthids met the above
criteria. Consequently, the quantitative data presented is from one video that
recorded presettlement schooling behaviour and two videos that showed solitary fish.
Figure 6.2 Still photos of presettlement Priacanthus tayenus (mean total length 24.1 mm) captured from video recorded in Coral Bay, Western Australia.
105
From the image analysis, the behaviour of presettlement priacanthids was described
and their total length (TL) measured from the stereo-video imagery. Total length for
each individual was determined as the average of the measurements at 5 different
frames and the mean TL for each group was calculated from the replicate
measurements of all the individuals in the school. The present study defines
schooling to be fish of the same species swimming in the same direction and in close
proximity, approximately 1 to 2 body lengths apart (Shaw, 1978). The term ‘group’
is used here to facilitate the description of schools that enter the field of view at
different times. Two groups of presettlement fish (A and B) aggregated into a single
school, group (AB). Nearest Neighbour Distance (NND), a quantitative measurement
of schooling behaviour, is the average distance between each fish in the group and its
nearest neighbour (Masuda, 2009; Masuda et al., 2003). From the stereo-video
imagery, NND was sampled from 10 different frames for group A and B and from 5
different frames for group AB. The difference in replicates is due to the short time
that group AB was visible from both cameras. Replicate measurements were at least
25 frames apart (1 second). The mean NND was divided by the average TL of fish in
the group (NND/TL) to facilitate comparison among aggregations of individuals of
different sizes.
Swimming speed was estimated by measuring the distance travelled by an individual
over 25 video frames (equivalent to 1 second) averaged across 3 consecutive
seconds. Mean swimming speed of presettlement schooling priacanthids was
obtained from the speed estimated for 10 individuals at various times of the video.
The swimming speed of solitary fish could only be measured for a single individual
in one video; therefore it was discarded from the analysis.
6.4 Results
In the deployment that recorded schooling, the first priacanthid individuals were
observed after 40.5 minutes. At that time, a school of four P. tayenus entered the
field of view for only one second, and no measurements were made because the
group was only visible from one of the cameras of the stereo system. Presettlement
priacanthids entered the field of view a second time at 60.8 minutes for a period of 2
minutes and the school at that time had 8 individuals (group A). The mean total
106
length of the priacanthids in this group was 24.4 ± 0.3 mm (± Standard Error) and
the individuals ranged from 23.0 ± 0.9 to 27.0 ± 0.4 mm. The NND of group A was
40.0 ± 3.6 mm and the NND/TL averaged 1.6.
A second school of 9 P. tayenus, group B, entered the field of view nearly two
minutes after the first group (at 62 min) and was visible for 30 seconds until both
schools merged into a single group (AB). Individuals in group B were estimated to
average a total length of 23.9 ± 2.2 mm (mean ± S.E.) and ranged from 12.5 ± 0.6 to
30.2 ± 0.4 mm. The mean NND for group B was 22.3 ± 2.4 mm and the NND/TL
was 0.9.
The combined group AB consisted of 17 P. tayenus and was visible for only 18
seconds. The mean TL of individuals in the school was 24.1 ± 0.5 mm and the NND
was 32.3 ± 1.8 mm. The NND/TL was 1.3. There was no clear relationship between
the size of the groups and their NND. Figure 6.3 provides a summary of the
quantitative measurements estimated for the three groups of presettlement P.
tayenus.
Further observations included a school of 7 priacanthids recorded 20 minutes later
(at 81 min), but no measurements were possible because the group was visible from
only one of the cameras of the stereo video system. Later in the video (110.2 min) a
school of 25 fish was recorded. One juvenile carangid, 4 nomeids and 20
presettlement priacanthids could be distinguished in that school. However, the small
size of the individuals and their distance from the camera prevented identification of
all individuals, estimation of their TL or confirmation of that group as AB. Mean
swimming speed of the observed presettlement P. tayenus was 8.5 ± 1.5 cm s-1 and
the measurements for 10 individuals ranged between 4.8 and 18.2 cm s-1. These
estimates are equivalent to a mean swimming speed of 3.6 ± 0.7 TL s-1 and ranged
between 2.0 and 7.5 TL s-1.
In the two videos analysed with solitary presettlement priacanthids, TL of the
individual was also determined as the average of the measurements in 5 different
frames. One solitary individual was 17.3 ± 0.2 mm and the other was 24 ± 0.8 mm.
Priacanthids and monacanthids were the only identifiable demersal fishes in their
pelagic stage observed during this study. Monacanthid fishes, observed in most
107
video deployments, were very structure-orientated, attracted to the structure of the
camera system, and did not join the schools of priacanthids.
Figure 6.3 Quantitative parameters for three groups of presettlement Priacanthus tayenus recorded using pelagic stereo cameras in Coral Bay, Western Australia (group A, 8 individuals; group B, 9 individuals and group AB, 17 individuals). (a) Mean Total Length (TL; mm); (b) mean Nearest Neighbour Distance (NND; mm) and (c) mean NND divided by the average TL. Error bars represent standard error.
6.5 Discussion
These results provide the first observations and quantitative data on the presettlement
schooling behaviour of any priacanthid during its pelagic phase. The presettlement
P. tayenus observed schooling were on average 24.1 mm TL (12.5 – 30.2 mm), and
swam in a group spacing themselves more or less evenly at around one body length
from the nearest neighbour within the school (Figure 6.3). P. tayenus appeared to be
sometimes associated with juvenile carangids or nomeids. The individuals observed
alone were within the same size range as those observed schooling. Due to the small
sample size, the results presented here may not reflect all the natural behaviours of
this species. However, this study provides the first look at in situ presettlement
behaviour using a video technique and may stimulate further work in this field.
Many fish species school during their pelagic phase regardless of their behaviour
after settlement or as adults (Shaw, 1978). NND is the parameter most commonly
used in studies of fish schooling behaviour and NND/TL can be used to compare
schooling across different size classes and species (Masuda, 2009; Masuda et al.,
2003; Soria et al., 2007; Torisawa et al., 2011). The values of NND calculated for
the observed presettlement P. tayenus (0.9 – 1.6 NND/TL) are comparable to the
estimates for other species. Schools of the carangid Pseudocaranx dentex at 25 mm
108
TL have a NND/TL of 0.5 to 1.2 (Masuda, 2009) and the scombrid Scomberomorus
niphonius, which is considered to form loose schools in comparison to most pelagic
fishes, has a NND/TL of 1.2 to 1.5 at a standard length of 19 to 23 mm (Masuda et
al., 2003). However, these studies on the ontogeny of schooling behaviour focus on
pelagic fish species and are limited to laboratory-based experiments using hatchery
reared larvae.
Generally, marine fish in their pelagic phase swim at speeds between 3 – 15 body
lengths per second (Leis, 2010). The in situ swimming speed estimated for the
observed priacanthids, on average 3.6 TL s-1 (8.5 cm s-1), fits in the slower end of
this general estimate. An average swimming speed greater than 3 cm s-1, is
considered to be influential for the dispersal of marine fishes (Leis, 2006; Leis,
2010). However, the measurements of swimming speed presented here should be
treated with caution as speeds are expected to vary and estimates may be influenced
by physical processes such as current and surge.
The limited information available on the behaviour of priacanthids at early stages
makes difficult the comparison and interpretation of the data presented here. Senta
(1978) noted that maturing juveniles of P. tayenus are recruited into schools at about
90-100 mm standard length (SL) and reach 200 mm SL one year after recruiting. It is
not known if the individuals observed in this study, 12.5 – 30.2 mm, were near
settlement. However, Pristigenys alta, another priacanthid species that has similar
maximum size to P. tayenus, is reported to transform from pelagic to demersal phase
at 35 to 55 mm SL (Caldwell, 1962b; Starnes, 1988).
Pelagic stereo-video systems may have the potential to overcome some of the
difficulties associated with studying the behaviour of juvenile fish in the pelagic
environment. Juveniles of various pelagic and demersal species were attracted to
these mid-water structures. Video techniques could be used for in situ behavioural
research at depths beyond the limits of diver-based studies and also for night
observation using a range of light wavelengths or image intensification technology
(Harvey et al., 2012a), and thus could contribute to overcoming a major gap in
knowledge about diel effects on the behaviour of fish larvae (Leis, 2010). Despite
the difficulties of identifying juvenile fish to the species level from video alone, this
technique could be used in combination with other methods, such as purse seine or
109
light traps, to facilitate identification and allow for further morphological and genetic
studies. Further work is needed to understand the strengths and limitations of pelagic
stereo video techniques for studying juvenile fish in their pelagic phase.
In conclusion, the observations on priacanthids presented here add to the limited, but
growing body of literature on the schooling behaviour of larval and juvenile
demersal fishes. P. tayenus swam in small groups and merged with other
presettlement conspecific schools to form larger groups. Sometimes these
presettlement schools swam in association with other species. Presettlement
schooling behaviour has also been observed in at least six other families of demersal
fish and seems to represent an important behaviour in the early life stages of many
demersal fish species (Leis and Carson-Ewart, 1998).
The observations and measurements of the behaviour of presettlement fishes
presented demonstrate the potential of non destructive stereo-video techniques for
collecting additional information which is complementary to the existing techniques.
Rather than using bait, artificial or natural materials could be placed in front of the
camera system to form a fish aggregation device for juvenile fishes.
111
CHAPTER 7 - General discussion
My research shows that pelagic stereo-BRUVs are an effective, non-destructive and
fishery-independent sampling approach to study pelagic ecosystems. This method
can be calibrated to commonly utilised standard techniques (chapter 4), is suitable
for studies over broad spatial and temporal scales, and can address questions on
species composition, behaviour, relative abundance and size distribution of fish
assemblages in the pelagic environment. In this thesis, I show that pelagic fish
assemblages are stratified by depth (chapter 2), which suggests that pelagic stereo-
BRUVs have the potential to explore the vertical distribution and diel migrations of
fish assemblages in the pelagic environment. My results also indicate that small
spatial area closures may not be adequate for the conservation and management of
highly mobile fish, even if they have proved effective for reef-associated species
(chapter 5). One of the common themes throughout my study was the high temporal
and spatial variability of fish assemblages in the pelagic environment. I investigated
the soak time and the level of replication required to maximise the precision of
sampling with pelagic stereo-BRUVs (chapters 2 and 3). In chapter 6, I demonstrate
how stereo-video can be used to make behavioural observations of juvenile fish in
the pelagic environment.
This thesis critically assesses the use of pelagic stereo-BRUVs and collates all the
information needed to implement, adopt and develop this method in future studies.
The aim of this discussion is to summarise the main findings of the project and
evaluate the strengths and limitations of pelagic stereo-BRUVs as a sampling
method for pelagic ecosystems.
Deployment method
The deployment method presented in chapter 2 and used throughout this project was
developed following the advice of experienced commercial fishermen. This method
allows for pelagic stereo-BRUVs to be deployed from both large research vessels or
from small boats. It also allows multiple systems to be deployed simultaneously in
order to maximise replication without adding field-time costs. The deployment
method presented here is moored to the seafloor and was used at depths from 5 to
112
120 metres, in coastal waters and at exposed offshore sites. The design of the system
makes the bait arm act as a rudder so that the cameras face downstream of the
current, improving the observation of individuals that approach the camera following
the bait plume. The ballast and subsurface floats minimise the vertical movement of
the cameras that may be driven by wave motion. This has proven to be a major
advantage over other systems tested in the mid-water that failed to stabilise the
cameras (e.g. Letessier et al., 2013). The associated costs of building pelagic stereo-
BRUVs and the on-board logistics of using them are similar to those for the
commonly used benthic BRUVs.
Further development and trials are needed to expand the use of pelagic stereo-
BRUVs on drifting systems, and to adapt this technique to open ocean sampling. The
main concerns to take into account would be to: (1) establish the means to track the
position of the systems, to enable retrieval at some pre-determined time (boats
should not remain in the area as this would confound effects of sampling); (2)
explore the differences in sampling area and bait dispersal between drifting and
moored systems (this is important as drifting systems have the potential to cover
large areas based on current speeds); and (3) minimise the effects of both the drifting
action and wave motion to ensure stability of the camera systems underwater.
Attaching stereo-video systems to drifting Fish Attraction Devices (FADs) could
potentially enhance the use of video techniques on the high seas.
Optimisation of sampling effort
I assessed the performance of pelagic stereo-BRUVs under different sampling
regimes in order to optimise sampling (chapter 2 and 3). I recommend standardising
the sample unit size (soak time) to facilitate data comparison across studies and
maximising replication given the resources available. In this thesis, I establish that a
soak time of 120 minutes is optimal when sampling in tropical or warm-temperate
waters, because it provides suitable precision at the assemblage level for univariate
(chapter 2) and multivariate analyses (chapter 3). Soak time has been shown to have
an effect on the number of species and the relative abundance observed (e.g. species
accumulation curve; chapter 2), and differences in behaviour over longer soak times
were also reported (e.g. sharks getting closer to the bait after patrolling the area;
chapter 4). Importantly, longer soak times increase the sampling costs because
113
longer deployments limit the number of replicates that can be undertaken per day in
the field, and the associated time required for video analysis increases (Gladstone et
al., 2012; Santana-Garcon et al., 2014a). However, different soak times may be
required in specific cases (e.g. longer soak times were used to compare pelagic
stereo-BRUVs to scientific longline surveys; chapter 4), thus soak time may be
determined by the objectives of the research program, the specific research question
being addressed and the species that are being targeted.
The sample size required to provide precise data on the assemblage composition and
abundance was determined to be at least 8 replicates per treatment when using
pelagic stereo-BRUVs (chapters 2 and 3). However, the data obtained from the
recommended sampling regime was often sparse and contained many zeros, which
can lead to difficulties in detecting patterns in the community structure (Clarke et al.,
2006a). In dissimilarity-based multivariate analysis, replicates that contain no
species lead to undefined dissimilarity values and, when many pairs of samples have
no (or very few) species in common, it yields uninformative dissimilarity values
(Anderson and Santana-Garcon, In review; Clarke et al., 2006a). Throughout this
project, I defined a ‘sample’ or ‘replicate’ as the data obtained from each BRUV
system deployment. However, given the nature of the data obtained with pelagic
stereo-BRUVs, I would recommend considering each deployment as a sub-sample,
and pooling deployments to obtain more powerful replicates. This is further explored
in recent research by Anderson and Santana-Garcon (In review; Appendix 1), where
data collected with pelagic stereo-BRUVs at the Houtman Abrolhos Islands (data
presented in chapter 5) is analysed to show that the distribution of dissimilarity
measures improves substantially when data from different deployments is pooled
(e.g. no undefined dissimilarities remained once the number of sub-samples
exceeded 2). Further research will allow the establishment of a general
recommendation of the number of deployments that should be pooled as replicates in
studies using pelagic stereo-BRUVs.
Fish attraction to pelagic stereo-BRUVs
Large pelagic fish species remained, on average, further from the camera systems
than demersal species (chapter 5). Greater range (distance from the cameras to the
fish recorded) complicates the detection, identification and accurate measurement of
114
fish lengths (Harvey et al., 2010). I observed that pelagic fish often approached the
cameras and, some species, showed interest in the bait. However, overall, the
proportion of individuals in the area approaching the cameras was lower than that
observed during video analysis of BRUVs deployed on the seafloor. These
observations raise the questions: What is attracting fish into the field of view of
cameras in the mid-water? What cues are attracting individuals to approach the bait
or cameras? Understanding these drivers will allow researchers to improve attraction
and enhance the sampling ability of pelagic stereo-BRUVs.
In this study I recorded fish from a broad range of sizes and trophic levels (i.e. from
small juveniles and planktivorous fish, to large apex predators like tunas and sharks).
Some species were attracted to the physical structure of the cameras and floats in the
mid-water (e.g. juveniles and small pelagics aggregated around the camera system),
while other species responded specifically to the bait source (e.g. biting the bait
basket). Species such as tuna were observed swimming within the field of view, but
did not show specific interest for either the camera system or the bait. These species
may have been observed swimming within the field of view by chance or, the bait
and structure associated with the cameras attracted them to the area, but not
specifically to the bait source. Thus, I believe there is room for improvement in the
attraction cues used in pelagic stereo-BRUVs. Reflective materials, such as flashes
and lures, are commonly used for fishing and these were tested attached to the
camera frame in early trials, but no specifically evaluated. Moreover, recent research
has suggested that a combination of visual, odour and acoustic cues is the most
effective attractant for pelagic fish (Rees et al., 2014). Nonetheless, any bait method
added to the camera systems should be standardised across all deployments (Dorman
et al., 2012; Hardinge et al., 2013).
Strengths and limitations of pelagic stereo-BRUVs
In this thesis I show that pelagic stereo-BRUVs can overcome some of the
difficulties associated with studying pelagic fish assemblages. The main advantage
of pelagic stereo-BRUVs over extractive fishery-independent techniques is their
non-destructive nature, which causes no physical trauma or stress to the individuals
studied. Therefore, this method is suitable to target threatened or protected species
and to be used in areas closed to fishing (Murphy and Jenkins, 2010). Baited video
115
methods have also proven to be effective in removing the gear selectivity bias of
fishing-based methods such as hook-size selectivity in longline surveys (chapter 4;
Harvey et al., 2012c; Langlois et al., In press; Langlois et al., 2012b; Willis et al.,
2000).
Pelagic stereo-BRUVs can be used in areas and at depths beyond the limits of diver-
based studies, and sampling can take place both during the day or using lights at
night (chapter 4; Bailey et al., 2007; Fitzpatrick et al., 2013; Harvey et al., 2012b;
Zintzen et al., 2012). Video techniques provide a permanent record of fish in the
study area, allowing for video analysis to be revisited and for clips to be shared
among researchers of different expertise (Cappo et al., 2003). Stereo-video
recordings also provide the opportunity to record fish behaviour in their natural
environment, including observations of presettlement fish that were previously
limited to catch-release and lab experiments (chapter 6; Leis and Carson-Ewart,
1998; Masuda, 2009; Santana-Garcon et al., 2014c).
All sampling techniques have some inherent biases and limitations (Murphy and
Jenkins, 2010) and, as assessed and discussed throughout this thesis, pelagic stereo-
BRUVs are no exception. The main limitation of pelagic stereo-BRUVs is the
identification of all individuals to species level. This is a common shortcoming for
all video techniques (Mallet and Pelletier, 2014), but it is exacerbated for pelagic
species because many taxa share similar morphological traits (i.e. coloration,
streamlined fusiform body, pointy snout and deeply forked tails), and they have
adapted to appear almost invisible in the pelagic environment (Pepperell, 2010).
Difficulties associated with species identification are not unique to video techniques,
but are also reported for other commonly used fishery-independent methods like
hydroacoustic techniques (Lawson et al., 2001). Fishery-dependent data is not
immune to species identification issues as fishers may also group species at higher
taxonomic levels or under-report the catch of low-value species (Lewison et al.,
2004). The time required for video analysis can be a major cost and limitation of
video techniques, but advances in video processing and the automation of some
components of the analysis appear promising in reducing analysis-time in the near
future (Harvey et al., 2014; Shortis et al., 2013). Understanding the limitations and
biases of a sampling technique allows for a better use of the method in forthcoming
116
studies and provides direction to future research so that such challenges can be
overcome.
Future research direction: opportunities and challenges of using
pelagic stereo-BRUVs
I believe there is potential for pelagic stereo-video to be attached to FADs, or other
commercial fishing gear to facilitate the collection of data in the open ocean.
Similarly, pelagic stereo-BRUVs will need further development to be deployed as
drifting systems. Implementation of stereo-video systems in the vast oceanic systems
will require high replication and involve processing extensive hours of video.
Fortunately, underwater camera equipment, image storage and processing
technology are advancing rapidly (Churnside et al., 2012; Mallet and Pelletier,
2014). Automation of photogrammetric techniques to detect, identify and measure
fish are also developing rapidly (Harvey et al., 2014; Shortis et al., 2009; Shortis et
al., 2013), following the increase in the use of video techniques for fishery
applications (Churnside et al., 2012; Graham et al., 2004; Harvey et al., 2003;
Pember et al., 2014; Rosen et al., 2013; Ruff et al., 1995).
Rapid industry developments in the resolution of video imagery and further research
on alternative calibration methods will assist with reducing the challenge of
identifying and measuring fish that remain far from the camera systems (Harvey et
al., 2014). Currently, the strategy most commonly used to calibrate underwater
stereo-video systems uses a two-layered cube that fills the entire field of view of the
camera (see Harvey and Shortis, 1998; Shortis and Harvey, 1998). Depending on the
size of the cube and the configuration of the camera system, most calibrations occur
at a distance between 2 and 4 metres. This is the optimal distance for measurement.
However, the use of a multi-distance calibration (i.e. calibrations at distances of
perhaps 3, 6 and 10 metres) and incorporating all the data into one bundle adjustment
will, theoretically, produce a more robust set of camera calibration parameters across
a greater range of distances. Multi-distance calibration strategies may make it
possible to accurately measure fish at distances greater than 8 to 10 metres.
Further research is also needed to understand the role of bait in mid-water
deployments and the dynamics of bait plume dispersal. This will help, among other
117
things, to understand the area of attraction and to determine the minimum distance
that should be allowed between simultaneous deployments (Heagney et al., 2007;
Rizzari et al., 2014). Monitoring physical and oceanographic parameters such as tide,
current, wave motion and temperature were not included in my study, but I recognise
that they are critical to model bait dispersal and to characterise pelagic habitats
(Grantham et al., 2011; Gray, 1997). I hope the body of research presented in this
thesis will encourage and facilitate other researchers to implement pelagic stereo-
BRUVs for monitoring and assessing fish assemblages in pelagic ecosystems.
119
Literature cited
Anderson, M.J., 2001. A new method for non-parametric multivariate analysis of variance. Austral Ecology 26(1), 32-46.
Anderson, M.J., 2006. Distance�Based Tests for Homogeneity of Multivariate Dispersions. Biometrics 62(1), 245-253.
Anderson, M.J., Santana-Garcon, J., 2015. Measures of precision for dissimilarity-based multivariate analysis of ecological communities. Ecology Letters 18(1), 66-73.
Anderson, M.J., Gorley, R.N., Clarke, K.R., 2008. PERMANOVA + for PRIMER: Guide to Software and Statistical Methods. PRIMER-E: Plymouth, UK.
Anderson, M.J., Crist, T.O., Chase, J.M., Vellend, M., Inouye, B.D., Freestone, A.L., Sanders, N.J., Cornell, H.V., Comita, L.S., Davies, K.F., 2011a. Navigating the multiple meanings of β diversity: a roadmap for the practicing ecologist. Ecology letters 14(1), 19-28.
Anderson, M.J., Crist, T.O., Chase, J.M., Vellend, M., Inouye, B.D., Freestone, A.L., Sanders, N.J., Cornell, H.V., Comita, L.S., Davies, K.F., Harrison, S.P., Kraft, N.J.B., Stegen, J.C., Swenson, N.G., 2011b. Navigating the multiple meanings of beta diversity: a roadmap for the practicing ecologist. Ecology Letters 14(1), 19-28.
Andrew, N.L., Mapstone, B.D., 1987. Sampling and the description of spatial pattern in marine ecology. Oceanography and Marine Biology 25, 39-90.
Andrew, N.L., Underwood, A.J., 1989. Patterns of abundance of the sea-urchin Centrostephanus-rodgersii (Agassiz) on the central coast of New South Wales, Australia. Journal of Experimental Marine Biology and Ecology 131(1), 61-80.
Angel, M.V., 1993. Biodiversity of the pelagic ocean. Conservation Biology 7(4), 760-772.
Assis, J., Narváez, K., Haroun, R., 2007. Underwater towed video: a useful tool to rapidly assess elasmobranch populations in large marine protected areas. Journal of Coastal Conservation 11(3), 153-157.
Axenrot, T., Didrikas, T., Danielsson, C., Hansson, S., 2004. Diel patterns in pelagic fish behaviour and distribution observed from a stationary, bottom-mounted, and upward-facing transducer. Ices Journal of Marine Science 61(7), 1100-1104.
Babcock, R., Shears, N., Alcala, A., Barrett, N., Edgar, G., Lafferty, K., McClanahan, T., Russ, G., 2010. Decadal trends in marine reserves reveal differential rates of change in direct and indirect effects. Proceedings of the National Academy of Sciences 107(43), 18256-18261.
Bailey, D.M., King, N.J., Priede, I.G., 2007. Cameras and carcasses: historical and current methods for using artificial food falls to study deep-water animals. Marine Ecology-Progress Series 350, 179-191.
Bakun, A., 1996. Patterns in the Ocean. Ocean processes and marine population dynamics.California Sea Grant College System, National Oceanic and Atmospheric Administration (NOAA).
120
Ballantine, B., 2014. Fifty years on: Lessons from marine reserves in New Zealand and principles for a worldwide network. Biological Conservation, http://dx.doi.org/10.1016/j.biocon.2014.1001.1014.
Bartsch, L.A., Richardson, W.B., Naimo, T.J., 1998. Sampling benthic macroinvertebrates in a large flood-plain river: Considerations of study design, sample size, and cost. Environmental Monitoring and Assessment 52(3), 425-439.
Bertrand, A., Josse, E., Bach, P., Gros, P., Dagorn, L., 2002. Hydrological and trophic characteristics of tuna habitat: consequences on tuna distribution and longline catchability. Canadian Journal of Fisheries and Aquatic Sciences 59(6), 1002-1013.
Bianchi, C.N., Morri, C., 2000. Marine biodiversity of the Mediterranean Sea: Situation, problems and prospects for future research. Marine Pollution Bulletin 40(5), 367-376.
Birt, M.J., Harvey, E.S., Langlois, T.J., 2012. Within and between day variability in temperate reef fish assemblages: Learned response to baited video. Journal of Experimental Marine Biology and Ecology 416, 92-100.
Bishop, J., 2006. Standardizing fishery-dependent catch and effort data in complex fisheries with technology change. Reviews in Fish Biology and Fisheries 16(1), 21-38.
Bond, M.E., Babcock, E.A., Pikitch, E.K., Abercrombie, D.L., Lamb, N.F., Chapman, D.D., 2012. Reef Sharks Exhibit Site-Fidelity and Higher Relative Abundance in Marine Reserves on the Mesoamerican Barrier Reef. Plos One 7(3).
Bowman, A.W., Azzalini, A., 1997. Applied smoothing techniques for data analysis: The Kernel approach with S-plus illustrations.Oxford University Press, Oxford.
Bowman, A.W., Azzalini, A., 2013. R package 'sm': nonparametric smoothing methods. R package version 2.2-5.
Breitburg, D.L., 1991. Settlement patterns and presettlement behaviour of the naked goby, Gobiosoma bosci, a temperate oyster reef fish. Marine Biology 109(2), 213-221.
Brooks, E., Sloman, K., Sims, W., Danyichuk, A., 2011. Validating the use of baited remote underwater video surveys for assessing the diversity, distribution and abundance of sharks in the Bahamas. Endangered Species Research 13, 231-243.
Caldwell, D.K., 1962a. Development and distribution of the short bigeye Pseudopriacanthus altus (Gill) in the western North Atlantic. U.S. Department Interior Fishery Bulletin 203, 1-150.
Caldwell, D.K., 1962b. Western Atlantic Fishes of the Family Priacanthidae. Copeia 1962(2), 417-424.
Caldwell, D.K., Bullis, H.R., 1971. Unusually large aggregation of prejuvenile bigeyes, Priacanthus arenatus, in West Indies. Copeia(1), 176-&.
Cappo, M., Speare, P., De'ath, G., 2004. Comparison of baited remote underwater video stations (BRUVS) and prawn (shrimp) trawls for assessments of fish biodiversity in inter-reefal areas of the Great Barrier Reef Marine Park. Journal of Experimental Marine Biology and Ecology 302(2), 123-152.
Cappo, M., Harvey, E., Shortis, M., 2006. Counting and measuring fish with baited video techniques - an overview. Australian society for fish biology workshop proceedings, 101-114.
121
Cappo, M., Harvey, E., Malcom, H., Speare, P., 2003. Potential of video techniques to monitor diversity, abundance and size of fish in studies of Marine Protected Areas. In: Beumer, J., Grant, A., Smith, D. (Eds.), Aquatic Protected Areas - what works best and how do we know? World Congress on Aquatic Protected Areas. Australian Society for Fish Biology, Cairns, Australia, pp. 455-464.
Cappo, M., Speare, P., Wassenberg, T., Harvey, E., Rees, M., Heyward, A., Pitcher, R., 2001. Use of Baited Remote Underwater Video Stations (BRUVS) to survey demersal fish – how deep and meaningful? In: Harvey, E., Cappo, M. (Eds.), Direct sensing of the size frequency and abundance of target and non-target fauna in Australian Fisheries. Fisheries Research and Development Coorperation, Rottnest Island, Western Australia, pp. 63-71, ISBN 61 74052 74057 74052.
Cheung, W.W., Watson, R., Pauly, D., 2013a. Signature of ocean warming in global fisheries catch. Nature 497(7449), 365-368.
Cheung, W.W., Sarmiento, J.L., Dunne, J., Frölicher, T.L., Lam, V.W., Palomares, M.D., Watson, R., Pauly, D., 2013b. Shrinking of fishes exacerbates impacts of global ocean changes on marine ecosystems. Nature Climate Change 3(3), 254-258.
Churnside, J., Jech, M., Tenningen, E., Clement, P.M., Fjälling, A., Jaffe, J.S., Jones, E.G., Lundgren, B., Macaulay, G.J., McElderry, H., 2012. Fishery applications of optical technologiesInternational Council for the Exploration of the Sea.
Clarke, K.R., Warwick, R.M., 2001. Change in marine communities: an approach to statistical analysis and interpretation, 2nd edition. PRIMER-E: Plymouth, UK.
Clarke, K.R., Gorley, R.N., 2006. PRIMER v6: User Manual/Tutorial. PRIMER-E: Plymouth, UK.
Clarke, K.R., Somerfield, P.J., Chapman, M.G., 2006a. On resemblance measures for ecological studies, including taxonomic dissimilarities and a zero-adjusted Bray-Curtis coefficient for denuded assemblages. Journal of Experimental Marine Biology and Ecology 330(1), 55-80.
Clarke, K.R., Chapman, M.G., Somerfield, P.J., Needham, H.R., 2006b. Dispersion-based weighting of species counts in assemblage analyses. Marine Ecology-Progress Series 320, 11-27.
Claudet, J., Osenberg, C.W., Benedetti�Cecchi, L., Domenici, P., García�Charton, J.A., Pérez�Ruzafa, Á., Badalamenti, F., Bayle�Sempere, J., Brito, A., Bulleri, F., 2008. Marine reserves: size and age do matter. Ecology letters 11(5), 481-489.
Claudet, J., Osenberg, C.W., Domenici, P., Badalamenti, F., Milazzo, M., Falcón, J.M., Bertocci, I., Benedetti-Cecchi, L., García-Charton, J.-A., Goñi, R., Borg, J.A., Forcada, A., de Lucia, G.A., Pérez-Ruzafa, Á., Afonso, P., Brito, A., Guala, I., Diréach, L.L., Sanchez-Jerez, P., Somerfield, P.J., Planes, S., 2010. Marine reserves: Fish life history and ecological traits matter. Ecological Applications 20(3), 830-839.
Cochran, W.G., Cox, G.M., 1957. Experimental Designs. 2nd ed. John Wiley & Sons, New York.
Davidson, R., 2001. Changes in population parameters and behaviour of blue cod (Parapercis colias; Pinguipedidae) in Long Island—Kokomohua Marine
122
Reserve, Marlborough Sounds, New Zealand. Aquatic Conservation: Marine and Freshwater Ecosystems 11(6), 417-435.
Davies, T.K., Mees, C.C., Milner-Gulland, E., 2014. The past, present and future use of drifting fish aggregating devices (FADs) in the Indian Ocean. Marine Policy 45, 163-170.
Davies, T.K., Martin, S., Mees, C., Chassot, E., Kaplan, D.M., 2012. A review of the conservation benefits of marine protected areas for pelagic species associated with fisheries. ISSF Technical Report 2012-02. International Seafood Sustainability Foundation, McLean, Virginia.
Davison, A.C., Hinkley, D.V., 1997. Bootstrap methods and their applicationCambridge university press.
de Juan, S., Lleonart, J., 2010. A conceptual framework for the protection of vulnerable habitats impacted by fishing activities in the Mediterranean high seas. Ocean & Coastal Management 53(11), 717-723.
Dempster, T., 2005. Temporal variability of pelagic fish assemblages around fish aggregation devices: biological and physical influences. Journal of Fish Biology 66(5), 1237-1260.
Dichmont, C.M., Ellis, N., Bustamante, R.H., Deng, R., Tickell, S., Pascual, R., Lozano�Montes, H., Griffiths, S., 2013. Evaluating marine spatial closures with conflicting fisheries and conservation objectives. Journal of Applied Ecology 50(4), 1060-1070.
Dorman, S.R., Harvey, E.S., Newman, S.J., 2012. Bait Effects in Sampling Coral Reef Fish Assemblages with Stereo-BRUVs. Plos One 7(7), e41538-e41538.
Downing, J.A., Downing, W.L., 1992. Spatial aggregation, precision, and power in surveys of fresh-water mussel populations. Canadian Journal of Fisheries and Aquatic Sciences 49(5), 985-991.
Dulvy, N.K., Baum, J.K., Clarke, S., Compagno, L.J.V., Cortes, E., Domingo, A., Fordham, S., Fowler, S., Francis, M.P., Gibson, C., Martinez, J., Musick, J.A., Soldo, A., Stevens, J.D., Valenti, S., 2008. You can swim but you can't hide: the global status and conservation of oceanic pelagic sharks and rays. Aquatic Conservation-Marine and Freshwater Ecosystems 18(5), 459-482.
Edgar, G.J., Stuart-Smith, R.D., Willis, T.J., Kininmonth, S., Baker, S.C., Banks, S., Barrett, N.S., Becerro, M.A., Bernard, A.T.F., Berkhout, J., Buxton, C.D., Campbell, S.J., Cooper, A.T., Davey, M., Edgar, S.C., Forsterra, G., Galvan, D.E., Irigoyen, A.J., Kushner, D.J., Moura, R., Parnell, P.E., Shears, N.T., Soler, G., Strain, E.M.A., Thomson, R.J., 2014. Global conservation outcomes depend on marine protected areas with five key features. Nature 506(7487), 216-220.
Ellis, D.M., Demartini, E.E., 1995. Evaluation of a video camera technique for indexing abundances of juvenile pink snapper, Pristipomoides-filamentosus, and other Hawaiian insular fishes. Fishery Bulletin 93(1), 67-77.
Evans, K., Baer, H., Bryant, E., Holland, M., Rupley, T., Wilcox, C., 2011. Resolving estimation of movement in a vertically migrating pelagic fish: Does GPS provide a solution? Journal of Experimental Marine Biology and Ecology 398(1-2), 9-17.
Field, J.G., 2002. Oceans 2020: science, trends, and the challenge of sustainability. Island Press.
Fitzpatrick, C., McLean, D., Harvey, E.S., 2013. Using artificial illumination to survey nocturnal reef fish. Fisheries Research 146, 41-50.
123
Freon, P., Misund, O.A., 1999. Dynamics of pelagic fish distribution and behaviour: effects on fisheries and stock assessment.Blackwell, Fishing News Books, London.
Freon, P., Cury, P., Shannon, L., Roy, C., 2005. Sustainable exploitation of small pelagic fish stocks challenged by environmental and ecosystem changes: A review. Bulletin of Marine Science 76(2), 385-462.
Game, E.T., Grantham, H.S., Hobday, A.J., Pressey, R.L., Lombard, A.T., Beckley, L.E., Gjerde, K., Bustamante, R., Possingham, H.P., Richardson, A.J., 2009. Pelagic protected areas: the missing dimension in ocean conservation. Trends in Ecology & Evolution 24(7), 360-369.
Game, E.T., Grantham, H.S., Hobday, A.J., Pressey, R.L., Lombard, A.T., Beckley, L.E., Gjerde, K., Bustamante, R., Possingham, H.P., Richardson, A.J., 2010. Pelagic MPAs: The devil you know Response. Trends in Ecology & Evolution 25(2), 63-64.
Gladstone, W., Lindfield, S., Coleman, M., Kelaher, B., 2012. Optimisation of baited remote underwater video sampling designs for estuarine fish assemblages. Journal of Experimental Marine Biology and Ecology 429, 28-35.
Goetze, J.S., Fullwood, L.A.F., 2013. Fiji's largest marine reserve benefits reef sharks. Coral Reefs 32(1), 121-125.
Goetze, J.S., Langlois, T.J., Egli, D.P., Harvey, E.S., 2011. Evidence of artisanal fishing impacts and depth refuge in assemblages of Fijian reef fish. Coral Reefs 30(2), 507-517.
Gomon, D.M.F., Bray, D.J., Kuiter, R.H., 2008. Fishes of Australia’s southern coast. New Holland Publishers (Australia).
Graham, N., Jones, E.G., Reid, D.G., 2004. Review of technological advances for the study of fish behaviour in relation to demersal fishing trawls. ICES Journal of Marine Science: Journal du Conseil 61(7), 1036-1043.
Grantham, H.S., Game, E.T., Lombard, A.T., Hobday, A.J., Richardson, A.J., Beckley, L.E., Pressey, R.L., Huggett, J.A., Coetzee, J.C., van der Lingen, C.D., Petersen, S.L., Merkle, D., Possingham, H.P., 2011. Accommodating Dynamic Oceanographic Processes and Pelagic Biodiversity in Marine Conservation Planning. Plos One 6(2).
Gray, J.S., 1997. Marine biodiversity: Patterns, threats and conservation needs. Biodiversity and Conservation 6(1), 153-175.
Gruess, A., Kaplan, D.M., Hart, D.R., 2011. Relative Impacts of Adult Movement, Larval Dispersal and Harvester Movement on the Effectiveness of Reserve Networks. Plos One 6(5).
Habeeb, R.L., Trebilco, J., Wotherspoon, S., Johnson, C.R., 2005. Determining natural scales of ecological systems. Ecological Monographs 75(4), 467-487.
Halpern, B.S., Walbridge, S., Selkoe, K.A., Kappel, C.V., Micheli, F., D'Agrosa, C., Bruno, J.F., Casey, K.S., Ebert, C., Fox, H.E., Fujita, R., Heinemann, D., Lenihan, H.S., Madin, E.M.P., Perry, M.T., Selig, E.R., Spalding, M., Steneck, R., Watson, R., 2008. A global map of human impact on marine ecosystems. Science 319(5865), 948-952.
Hardinge, J., Harvey, E.S., Saunders, B.J., Newman, S.J., 2013. A little bait goes a long way: The influence of bait quantity on a temperate fish assemblage sampled using stereo-BRUVs. Journal of Experimental Marine Biology and Ecology 449(0), 250-260.
Harry, A.V., Morgan, J.A.T., Ovenden, J.R., Tobin, A.J., Welch, D.J., Simpfendorfer, C.A., 2012. Comparison of the reproductive ecology of two
124
sympatric blacktip sharks (Carcharhinus limbatus and Carcharhinus tilstoni) off north-eastern Australia with species identification inferred from vertebral counts. Journal of Fish Biology 81(4), 1225-1233.
Harvey, E., Shortis, M., 1995. A system for stereo-video measurement of sub-tidal organisms. Marine Technology Society Journal 29(4), 10-22.
Harvey, E., Fletcher, D., Shortis, M., 2001a. A comparison of the precision and accuracy of estimates of reef-fish lengths determined visually by divers with estimates produced by a stereo-video system. Fishery Bulletin 99(1), 63-71.
Harvey, E., Fletcher, D., Shortis, M., 2001b. Improving the statistical power of length estimates of reef fish: a comparison of estimates determined visually by divers with estimates produced by a stereo-video system. Fishery Bulletin 99(1), 72-80.
Harvey, E., Fletcher, D., Shortis, M., 2002. Estimation of reef fish length by divers and by stereo-video - A first comparison of the accuracy and precision in the field on living fish under operational conditions. Fisheries Research 57(3), 255-265.
Harvey, E., Fletcher, D., Shortis, M.R., Kendrick, G.A., 2004. A comparison of underwater visual distance estimates made by scuba divers and a stereo-video system: implications for underwater visual census of reef fish abundance. Marine and Freshwater Research 55(6), 573-580.
Harvey, E., Cappo, M., Shortis, M., Robson, S., Buchanan, J., Speare, P., 2003. The accuracy and precision of underwater measurements of length and maximum body depth of southern bluefin tuna (Thunnus maccoyii) with a stereo-video camera system. Fisheries Research 63(3), 315-326.
Harvey, E.S., Shortis, M.R., 1998. Calibration stability of an underwater stereo-video system: Implications for measurement accuracy and precision. Marine Technology Society Journal 32(2), 3-17.
Harvey, E.S., Butler, J.J., McLean, D.L., Shand, J., 2012a. Contrasting habitat use of diurnal and nocturnal fish assemblages in temperate Western Australia. Journal of Experimental Marine Biology and Ecology 426, 78-86.
Harvey, E.S., Cappo, M., Kendrick, G.A., McLean, D.L., 2013. Coastal Fish Assemblages Reflect Geological and Oceanographic Gradients Within An Australian Zootone. PloS one 8(11), e80955.
Harvey, E.S., Cappo, M., Butler, J.J., Hall, N., Kendrick, G.A., 2007. Bait attraction affects the performance of remote underwater video stations in assessment of demersal fish community structure. Marine Ecology-Progress Series 350, 245-254.
Harvey, E.S., Goetze, J., McLaren, B., Langlois, T., Shortis, M.R., 2010. Influence of Range, Angle of View, Image Resolution and Image Compression on Underwater Stereo-Video Measurements: High-Definition and Broadcast-Resolution Video Cameras Compared. Marine Technology Society Journal 44(1), 75-85.
Harvey, E.S., Dorman, S.R., Fitzpatrick, C., Newman, S.J., McLean, D.L., 2012b. Response of diurnal and nocturnal coral reef fish to protection from fishing: an assessment using baited remote underwater video. Coral Reefs 31(4), 939-950.
Harvey, E.S., Newman, S.J., McLean, D.L., Cappo, M., Meeuwig, J.J., Skepper, C.L., 2012c. Comparison of the relative efficiencies of stereo-BRUVs and traps for sampling tropical continental shelf demersal fishes. Fisheries Research 125, 108-120.
125
Harvey, E.S., McLean, D.L., Frusher, S., Haywood, M.D.D., Newman, S.J., Williams, A., 2014. The use of BRUVs as a tool for assessing marine fisheries and ecosystems: a review of the hurdles and potential. FRDC 2010/002, ISBN: 978-971-74052-74265-74056. 74202pp.
Hays, G.C., Richardson, A.J., Robinson, C., 2005. Climate change and marine plankton. Trends in Ecology & Evolution 20(6), 337-344.
Heagney, E.C., Lynch, T.P., Babcock, R.C., Suthers, I.M., 2007. Pelagic fish assemblages assessed using mid-water baited video: standardising fish counts using bait plume size. Marine Ecology-Progress Series 350, 255-266.
Hellmann, J.J., Fowler, G.W., 1999. Bias, precision, and accuracy of four measures of species richness. Ecological Applications 9(3), 824-834.
Holling, C.S., 1992. CROSS-SCALE MORPHOLOGY, GEOMETRY, AND DYNAMICS OF ECOSYSTEMS. Ecological Monographs 62(4), 447-502.
Hyrenbach, K.D., Forney, K.A., Dayton, P.K., 2000. Marine protected areas and ocean basin management. Aquatic Conservation-Marine and Freshwater Ecosystems 10(6), 437-458.
Jackson, D.A., 1997. Compositional data in community ecology: The paradigm or peril of proportions? Ecology 78(3), 929-940.
Jensen, O.P., Ortega-Garcia, S., Martell, S.J., Ahrens, R.N., Domeier, M.L., Walters, C.J., Kitchell, J.F., 2010. Local management of a “highly migratory species”: the effects of long-line closures and recreational catch-and-release for Baja California striped marlin fisheries. Progress in Oceanography 86(1), 176-186.
Jones, E.G., Tselepides, A., Bagley, P.M., Collins, M.A., Priede, I.G., 2003. Bathymetric distribution of some benthic and benthopelagic species attracted to baited cameras and traps in the deep eastern Mediterranean. Marine Ecology-Progress Series 251, 75-86.
Kaplan, D.M., Chassot, E., Gruss, A., Fonteneau, A., 2010. Pelagic MPAs: The devil is in the details. Trends in Ecology & Evolution 25(2), 62-63.
Kaplan, D.M., Chassot, E., Amandé, J.M., Dueri, S., Demarcq, H., Dagorn, L., Fonteneau, A., 2014. Spatial management of Indian Ocean tropical tuna fisheries: potential and perspectives. ICES Journal of Marine Science: Journal du Conseil DOI: 10.1093/icesjms/fst233.
Kingsford, M.J., Tricklebank, K.A., 1991. Ontogeny and behavior of Aldrichetta
forsteri (TELEOSTEI, MUGILIDAE). Copeia(1), 9-16. Klages, J., Broad, A., Kelaher, B.P., Davis, A., 2014. The influence of gummy
sharks, Mustelus antarcticus, on observed fish assemblage structure. Environmental Biology of Fishes 97(2), 215-222.
Knip, D.M., Heupel, M.R., Simpfendorfer, C.A., 2012. Evaluating marine protected areas for the conservation of tropical coastal sharks. Biological Conservation 148(1), 200-209.
Langlois, T.J., Harvey, E.S., Fitzpatrick, B., Meeuwig, J.J., Shedrawi, G., Watson, D.L., 2010. Cost-efficient sampling of fish assemblages: comparison of baited video stations and diver video transects. Aquatic Biology 9(2), 155-168.
Langlois, T.J., Newman, S.J., Cappo, M., Harvey, E.S., Rome, B.M., Skepper, C.L., Wakefield, C.B., In press. Length selectivity of commercial fish traps assessed from in situ comparisons with stereo-videos: is there evidence of sampling bias? Fisheries Research, DOI: 10.1016/j.fishres.2014.1006.1008.
Langlois, T.J., Radford, B.T., Van Niel, K.P., Meeuwig, J.J., Pearce, A.F., Rousseaux, C.S.G., Kendrick, G.A., Harvey, E.S., 2012a. Consistent
126
abundance distributions of marine fishes in an old, climatically buffered, infertile seascape. Global Ecology and Biogeography 21(9), 886-897.
Langlois, T.J., Fitzpatrick, B.R., Fairclough, D.V., Wakefield, C.B., Hesp, A., McLean, D.L., Harvey, E.S., Meeuwig, J.J., 2012b. Similarities between Line Fishing and Baited Stereo-Video Estimations of Length-Frequency: Novel Application of Kernel Density Estimates. Plos One 7(11), 1-9.
Last, P.R., Stevens, J.D., 2009. Sharks and Rays of AustraliaCSIRO PUBLISHING, Australia.
Lawson, G.L., Barange, M., Freon, P., 2001. Species identification of pelagic fish schools on the South African continental shelf using acoustic descriptors and ancillary information. Ices Journal of Marine Science 58(1), 275-287.
Legendre, P., Anderson, M.J., 1999. Distance-based redundancy analysis: testing multispecies responses in multifactorial ecological experiments. Ecological Monographs 69(1), 1-24.
Leis, J.M., 2006. Are larvae of demersal fishes plankton or nekton? In: Southward, A.J., Sims, D.W. (Eds.), Advances in Marine Biology, Vol 51. Academic Press Ltd-Elsevier Science Ltd, London, pp. 57-141.
Leis, J.M., 2010. Ontogeny of behaviour in larvae of marine demersal fishes. Ichthyological Research 57(4), 325-342.
Leis, J.M., Carson-Ewart, B.M., 1998. Complex behaviour by coral-reef fish larvae in open-water and near-reef pelagic environments. Environmental Biology of Fishes 53(3), 259-266.
Leis, J.M., Sweatman, H.P.A., Reader, S.E., 1996. What the pelagic stages of coral reef fishes are doing out in blue water: Daytime field observations of larval behavioural capabilities. Marine and Freshwater Research 47(2), 401-411.
Lester, R.J.G., Watson, R.A., 1985. Growth, mortality, parasitism, and potential yields of two Priacanthus species in the South China Sea. Journal of Fish Biology 27(3), 307-318.
Letessier, T.B., Meeuwig, J.J., Gollock, M., Groves, L., Bouchet, P.J., Chapuis, L., Vianna, G., Kemp, K., Koldewey, H.J., 2013. Assessing pelagic fish populations: The application of demersal video techniques to the mid-water environment. Methods in Oceanography, DOI: 10.1016/j.mio.2013.1011.1003.
Lewison, R.L., Crowder, L.B., Read, A.J., Freeman, S.A., 2004. Understanding impacts of fisheries bycatch on marine megafauna. Trends in Ecology & Evolution 19(11), 598-604.
Mallet, D., Pelletier, D., 2014. Underwater video techniques for observing coastal marine biodiversity: A review of sixty years of publications (1952–2012). Fisheries Research 154(0), 44-62.
Masuda, R., 2009. Behavioral Ontogeny of Marine Pelagic Fishes with the Implications for the Sustainable Management of Fisheries Resources. Aqua-Bio Science Monographs 2(2), 1-56.
Masuda, R., Shoji, J., Nakayama, S., Tanaka, M., 2003. Development of schooling behavior in Spanish mackerel Scomberomorus niphonius during early ontogeny. Fisheries Science 69(4), 772-776.
McAuley, R.B., Simpfendorfer, C.A., Wright, I.W., 2007a. Gillnet mesh selectivity of the sandbar shark (Carcharhinus plumbeus): implications for fisheries management. Ices Journal of Marine Science 64(9), 1702-1709.
McAuley, R.B., Simpfendorfer, C.A., Hyndes, G.A., Lenanton, R.C.J., 2007b. Distribution and reproductive biology of the sandbar shark, Carcharhinus
127
plumbeus (Nardo), inWestern Australian waters. Marine and Freshwater Research 58(1), 116-126.
McCormick, M.I., Milicich, M.J., 1993. Late pelagic stage goatfishes - distribution patterns and inferences on schooling behavior. Journal of Experimental Marine Biology and Ecology 174(1), 15-42.
McLean, D.L., Harvey, E.S., Fairclough, D.V., Newman, S.J., 2010. Large decline in the abundance of a targeted tropical lethrinid in areas open and closed to fishing. Marine Ecology-Progress Series 418, 189-199.
Merritt, D., Donovan, M.K., Kelley, C., Waterhouse, L., Parke, M., Wong, K., Drazen, J.C., 2011. BotCam: a baited camera system for nonextractive monitoring of bottomfish species. Fishery Bulletin 109(1), 56-67.
Meyer, C.G., Papastamatiou, Y.P., Holland, K.N., 2010. A multiple instrument approach to quantifying the movement patterns and habitat use of tiger (Galeocerdo cuvier) and Galapagos sharks (Carcharhinus galapagensis) at French Frigate Shoals, Hawaii. Marine Biology 157(8), 1857-1868.
Mills, C.E., Carlton, J.T., 1998. Rationale for a system of international reserves for the open ocean. Conservation Biology 12(1), 244-247.
Mora, C., Sale, P.F., 2002. Are populations of coral reef fish open or closed? Trends in Ecology & Evolution 17(9), 422-428.
Murphy, H.M., Jenkins, G.P., 2010. Observational methods used in marine spatial monitoring of fishes and associated habitats: a review. Marine and Freshwater Research 61(2), 236-252.
Myers, E., 2013. A short crepuscular changeover period in a temperate reef fish assemblage in Western Australia. The University of Western Australia, Perth, Australia.
Myers, R.A., Worm, B., 2003. Rapid worldwide depletion of predatory fish communities. Nature 423(6937), 280-283.
Myers, R.A., Baum, J.K., Shepherd, T.D., Powers, S.P., Peterson, C.H., 2007. Cascading effects of the loss of apex predatory sharks from a coastal ocean. Science 315(5820), 1846-1850.
Nardi, K., Jones, G., Moran, M., Cheng, Y., 2004. Contrasting effects of marine protected areas on the abundance of two exploited reef fishes at the sub-tropical Houtman Abrolhos Islands, Western Australia. Environmental Conservation 31(02), 160-168.
Norse, E.A., 2005. Pelagic protected areas: the greatest parks challenge of the 21st century. Parks 15, 32-39.
Notarbartolo-Di-Sciara, G., Agardy, T., Hyrenbach, D., Scovazzi, T., Van Klaveren, P., 2008. The pelagos sanctuary for Mediterranean marine mammals. Aquatic Conservation-Marine and Freshwater Ecosystems 18(4), 367-391.
Ohman, M.C., Munday, P.L., Jones, G.P., Caley, M.J., 1998. Settlement strategies and distribution patterns of coral-reef fishes. Journal of Experimental Marine Biology and Ecology 225(2), 219-238.
Oksanen, J., Blanchet, F.G., Kindt, R., Legendre, P., Minchin, P.R., O'Hara, R.B., Simpson, G.L., Solymos, P., Henry, M., Stevens, H., Wagner, H., 2013. vegan: Community Ecology Package. R package version 2.0-9.
Palumbi, S.R., 2004. Marine reserves and ocean neighborhoods: the spatial scale of marine populations and their management. Annu. Rev. Environ. Resour. 29, 31-68.
Pauly, D., 2009. Beyond duplicity and ignorance in global fisheries. Scientia Marina 73(2), 215-224.
128
Pauly, D., Christensen, V., Dalsgaard, J., Froese, R., Torres, F., 1998. Fishing down marine food webs. Science 279(5352), 860-863.
Pauly, D., Christensen, V., Guenette, S., Pitcher, T.J., Sumaila, U.R., Walters, C.J., Watson, R., Zeller, D., 2002. Towards sustainability in world fisheries. Nature 418(6898), 689-695.
Pember, M.B., de Lestang, S., Slawinski, D., 2014. An industry-based approach to regional-scale monitoring of benthic ecosystems, 10th International Temperate Reefs Symposium. The University of Western Australia. Perth, Australia.
Pepperell, J.G., 2010. Fishes of the open ocean. A natural history and illustrated guide.The University Of Chicago Press, Chicago and London., 266 pp.
Pichegru, L., Grémillet, D., Crawford, R., Ryan, P., 2010. Marine no-take zone rapidly benefits endangered penguin. Biology Letters 6(4), 498-501.
Pinsky, M.L., Worm, B., Fogarty, M.J., Sarmiento, J.L., Levin, S.A., 2013. Marine taxa track local climate velocities. Science 341(6151), 1239-1242.
Priede, I.G., Bagley, P.M., Smith, A., Creasey, S., Merrett, N.R., 1994. Scavenging deep demersal fishes of the Porcupine Seabight, Northeast Atlantic - Observations by baited camera, trap and trawl. Journal of the Marine Biological Association of the United Kingdom 74(3), 481-498.
Pringle, J.D., 1984. Efficiency estimates for various quadrat sizes used in benthic sampling. Canadian Journal of Fisheries and Aquatic Sciences 41(10), 1485-1489.
R Core Team, 2012. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria.
R Core Team, 2013. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria.
R Core Team, 2014. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria.
Rees, M.J., Knott, N.A., Davis, A.R., 2014. Comparison of sight, sound and scent attractants for pelagic fishes to remote mid-water video stations, 10th International Temperate Reefs Symposium. The University of Western Australia. Perth, Australia.
Rizzari, J.R., Frisch, A.J., Connolly, S.R., 2014. How robust are estimates of coral reef shark depletion? Biological Conservation 176, 39-47.
Robison, B.H., 2009. Conservation of Deep Pelagic Biodiversity. Conservation Biology 23(4), 847-858.
Rosen, S., Jorgensen, T., Hammersland-White, D., Holst, J.C., 2013. DeepVision: a stereo camera system provides highly accurate counts and lengths of fish passing inside a trawl. Canadian Journal of Fisheries and Aquatic Sciences 70(10), 1456-1467.
Ruff, B.P., Marchant, J.A., Frost, A.R., 1995. Fish sizing and monitoring using a stereo image analysis system applied to fish farming. Aquacultural Engineering 14(2), 155-173.
Russ, G.R., Alcala, A.C., 2011. Enhanced biodiversity beyond marine reserve boundaries: The cup spillith over. Ecological Applications 21(1), 241-250.
Russ, G.R., Alcala, A.C., Maypa, A.P., Calumpong, H.P., White, A.T., 2004. Marine reserve benefits local fisheries. Ecological Applications 14(2), 597-606.
Sainte-Marie, B., Hargrave, B.T., 1987. Estimation of scavenger abundance and distance of attraction to bait. Marine Biology 94(3), 431-443.
129
Santana-Garcon, J., Newman, S.J., Harvey, E.S., 2014a. Development and validation of a mid-water baited stereo-video technique for investigating pelagic fish assemblages. Journal of Experimental Marine Biology and Ecology 452, 82-90.
Santana-Garcon, J., Leis, J.M., Newman, S.J., Harvey, E.S., 2013. Presettlement schooling behaviour of a priacanthid, the Purplespotted Bigeye Priacanthus tayenus (Priacanthidae: Teleostei). Environmental Biology of Fishes, 1-7.
Santana-Garcon, J., Newman, S.J., Langlois, T.J., Harvey, E.S., 2014b. Effects of a spatial closure on highly mobile fish species: an assessment using pelagic stereo-BRUVs. Journal of Experimental Marine Biology and Ecology 460, 153-161.
Santana-Garcon, J., Leis, J.M., Newman, S.J., Harvey, E.S., 2014c. Presettlement schooling behaviour of a priacanthid, the Purplespotted Bigeye Priacanthus tayenus (Priacanthidae: Teleostei). Environmental Biology of Fishes 97(3), 277-283.
Santana-Garcon, J., Braccini, M., Langlois, T.J., Newman, S.J., McAuley, R.B., Harvey, E.S., 2014d. Calibration of pelagic stereo-BRUVs and scientific longline surveys for sampling sharks. Methods in Ecology and Evolution, DOI: 10.1111/2041-1210X.12217.
Santos, M.B., González-Quirós, R., Riveiro, I., Iglesias, M., Louzao, M., Pierce, G.J., 2013a. Characterization of the pelagic fish community of the north-western and northern Spanish shelf watersa. Journal of Fish Biology 83(4), 716-738.
Santos, M.B., Gonzalez-Quiros, R., Riveiro, I., Iglesias, M., Louzao, M., Pierce, G.J., 2013b. Characterization of the pelagic fish community of the north-western and northern Spanish shelf waters. Journal of Fish Biology 83(4), 716-738.
Seah, Y.G., Mazlan, A.G., Abdullah, S., Zaidi, C.C., Usup, G., Mohamed, C.A.R., 2011. Feeding Guild of Dominant Trawl Species in the Southeastern Waters of Peninsular Malaysia. Journal of Biological Sciences 11(2), 221-225.
Senta, T., 1978. High incidence of aggregations of the bigeyes, Priacanthus tayenus and P. macracanthus in the South China Sea. Bulletin of the Faculty of Fisheries, Nagasaki University 45, 1-4.
Shaw, E., 1978. Scooling fishes. American Scientist 66(2), 166-175. Sheather, S.J., Jones, M.C., 1991. A reliable data-based bandwidth selection method
for Kernel Density-Estimation. Journal of the Royal Statistical Society Series B-Methodological 53(3), 683-690.
Shedrawi, G., Harvey, E., McLean, D., Prince, J., Bellchambers, L., Newman, S., 2014. Evaluation of the effect of closed areas on a unique and shallow water coral reef fish assemblage reveals complex responses. Coral Reefs, DOI: 10.1007/s00338-00014-01160-00333.
Sheppard, C., 2010. Marine protected areas and pelagic fishing: The case of the Chagos Archipelago. Marine Pollution Bulletin 60(11), 1899-1901.
Shortis, M., Harvey, E., Abdo, D., 2009. A REVIEW OF UNDERWATER STEREO-IMAGE MEASUREMENT FOR MARINE BIOLOGY AND ECOLOGY APPLICATIONS. In: Gibson, R.N., Atkinson, R.J.A., Gordon, J.D.M. (Eds.), Oceanography and Marine Biology: An Annual Review, Vol 47. Crc Press-Taylor & Francis Group, Boca Raton, pp. 257-292.
130
Shortis, M.R., Harvey, E.S., 1998. Design and calibration of an underwater stereo-video system for the monitoring of marine fauna populations. International Archives of Photogrammetry and Remote Sensing 32, 792-799.
Shortis, M.R., Ravanbakhsh, M., Shafait, F., Harvey, E.S., Mian, A., Seager, J.W., Culverhouse, P.F., Cline, D.E., Edgington, D.R., 2013. A review of techniques for the identification and measurement of fish in underwater stereo-video image sequences, Conference of Videometrics, Range Imaging, and Applications XII; and Automated Visual Inspection. Proceedings of SPIE, Munich, GERMANY.
Siegenthaler, U., Sarmiento, J.L., 1993. ATMOSPHERIC CARBON-DIOXIDE AND THE OCEAN. Nature 365(6442), 119-125.
Simpfendorfer, C.A., Hueter, R.E., Bergman, U., Connett, S.M.H., 2002. Results of a fishery-independent survey for pelagic sharks in the western North Atlantic, 1977-1994. Fisheries Research 55(1-3), 175-192.
Skomal, G.B., 2007. Evaluating the physiological and physical consequences of capture on post-release survivorship in large pelagic fishes. Fisheries Management and Ecology 14(2), 81-89.
Soria, M., Freon, P., Chabanet, P., 2007. Schooling properties of an obligate and a facultative fish species. Journal of Fish Biology 71(5), 1257-1269.
Starnes, W.C., 1988. Revision, phylogeny and biogeographic comments on the cirmtropical marine percoid fish family Priacanthidae. Bulletin of Marine Science 43(2), 117-203.
Stobart, B., Garcia-Charton, J.A., Espejo, C., Rochel, E., Goni, R., Renones, O., Herrero, A., Crec'hriou, R., Polti, S., Marcos, C., Planes, S., Perez-Ruzafa, A., 2007. A baited underwater video technique to assess shallow-water Mediterranean fish assemblages: Methodological evaluation. Journal of Experimental Marine Biology and Ecology 345(2), 158-174.
Stoner, A.W., Laurel, B.J., Hurst, T.P., 2008. Using a baited camera to assess relative abundance of juvenile Pacific cod: Field and laboratory trials. Journal of Experimental Marine Biology and Ecology 354(2), 202-211.
Sumaila, U.R., Zeller, D., Watson, R., Alder, J., Pauly, D., 2007. Potential costs and benefits of marine reserves in the high seas. Marine Ecology Progress Series 345, 305-310.
Sumner, N.R., 2008. An Assessment of the Finfish Catch by Recreational Fishers, Tour Operators, Commercial Lobster Fishers, and Commercial Wetline Fishers from the Houtman Abrolhos Islands During 2006.
Thorson, J.T., Simpfendorfer, C.A., 2009. Gear selectivity and sample size effects on growth curve selection in shark age and growth studies. Fisheries Research 98(1-3), 75-84.
Torisawa, S., Fukuda, H., Suzuki, K., Takagi, T., 2011. Schooling behaviour of juvenile Pacific bluefin tuna Thunnus orientalis depends on their vision development. Journal of Fish Biology 79(5), 1291-1303.
Trenkel, V.M., Berger, L., 2013. A fisheries acoustic multi-frequency indicator to inform on large scale spatial patterns of aquatic pelagic ecosystems. Ecological Indicators 30, 72-79.
Underwood, A.J., 1997. Experiments in Ecology: their logical design and interpretation using analysis of variance.Cambridge University Press, Cambridge.
Unsworth, R.K.F., Peters, J.R., McCloskey, R.M., Hinder, S.L., 2014. Optimising stereo baited underwater video for sampling fish and invertebrates in
131
temperate coastal habitats. Estuarine, Coastal and Shelf Science, DOI: 10.1016/j.ecss.2014.1003.1020.
Von Der Emde, G., Mogdans, J., Kapoor, B.G., 2004. The senses of fish: adaptations for the reception of natural stimuli.Narosa Publishing House, New Delhi, India.
Walker, T.I., 1998. Can shark resources be harvested sustainably? A question revisited with a review, of shark fisheries. Marine and Freshwater Research 49(7), 553-572.
Walters, C.J., Hilborn, R., Parrish, R., 2007. An equilibrium model for predicting the efficacy of marine protected areas in coastal environments. Canadian Journal of Fisheries and Aquatic Sciences 64(7), 1009-1018.
Wand, M., 2013. KernSmooth: Functions for kernel smoothing for Wand & Jones (1995). R package version 2.23-10.
Ward, P., Myers, R.A., 2005a. Inferring the depth distribution of catchability for pelagic fishes and correcting for variations in the depth of longline fishing gear. Canadian Journal of Fisheries and Aquatic Sciences 62(5), 1130-1142.
Ward, P., Myers, R.A., 2005b. Shifts in open-ocean fish communities coinciding with the commencement of commercial fishing. Ecology 86(4), 835-847.
Ward, P., Myers, R.A., 2007. Bait loss and its potential effects on fishing power in pelagic longline fisheries. Fisheries Research 86(1), 69-76.
Ward, P., Myers, R.A., Blanchard, W., 2004. Fish lost at sea: the effect of soak time on pelagic longline catches. Fishery Bulletin 102(1), 179-195.
Ward, P.J., Myers, R.A., 2006. Do habitat models accurately predict the depth distribution of pelagic fishes? Fisheries Oceanography 15(1), 60-66.
Watson, D.L., Harvey, E., 2009. Influence of the Leeuwin Current on the distribution of fishes and the composition of fish assemblages. Journal of the Royal Society of Western Australia 92, 147-154.
Watson, D.L., Harvey, E.S., Kendrick, G.A., Nardi, K., Anderson, M.J., 2007. Protection from fishing alters the species composition of fish assemblages in a temperate-tropical transition zone. Marine Biology 152(5), 1197-1206.
Watson, D.L., Anderson, M.J., Kendrick, G.A., Nardi, K., Harvey, E.S., 2009. Effects of protection from fishing on the lengths of targeted and non-targeted fish species at the Houtman Abrolhos Islands, Western Australia. Marine Ecology-Progress Series 384, 241-249.
Watson, D.L., Harvey, E.S., Fitzpatrick, B.M., Langlois, T.J., Shedrawi, G., 2010. Assessing reef fish assemblage structure: how do different stereo-video techniques compare? Marine Biology 157(6), 1237-1250.
White, C., Costello, C., 2014. Close the High Seas to Fishing? PLoS Biol 12(3), e1001826.
White, J., Simpfendorfer, C.A., Tobin, A.J., Heupel, M.R., 2013. Application of baited remote underwater video surveys to quantify spatial distribution of elasmobranchs at an ecosystem scale. Journal of Experimental Marine Biology and Ecology 448(0), 281-288.
Wickham, H., 2007. Reshaping Data with the {reshape} Package. Journal of Statistical Software 21(12), 1-20.
Wickham, H., 2009. ggplot2: elegant graphics for data analysis. Springer New York. Wickham, H., 2011. The Split-Apply-Combine Strategy for Data Analysis. Journal
of Statistical Software 40(1), 1-29.
132
Willis, T.J., Babcock, R.C., 2000. A baited underwater video system for the determination of relative density of carnivorous reef fish. Marine and Freshwater Research 51(8), 755-763.
Willis, T.J., Millar, R.B., Babcock, R.C., 2000. Detection of spatial variability in relative density of fishes: comparison of visual census, angling, and baited underwater video. Marine Ecology-Progress Series 198, 249-260.
Worm, B., Sandow, M., Oschlies, A., Lotze, H.K., Myers, R.A., 2005. Global patterns of predator diversity in the open oceans. Science 309(5739), 1365-1369.
Worm, B., Barbier, E.B., Beaumont, N., Duffy, J.E., Folke, C., Halpern, B.S., Jackson, J.B.C., Lotze, H.K., Micheli, F., Palumbi, S.R., Sala, E., Selkoe, K.A., Stachowicz, J.J., Watson, R., 2006. Impacts of biodiversity loss on ocean ecosystem services. Science 314(5800), 787-790.
Zar, J.H., 1999. Biostatistical Analysis. 4th Edition ed. Prentice Hall International, Inc., New Jersey, 929 pp.
Zintzen, V., Roberts, C.D., Anderson, M.J., Stewart, A.L., Struthers, C.D., Harvey, E.S., 2011. Hagfish predatory behaviour and slime defence mechanism. Scientific Reports 1(131), 1-6.
Zintzen, V., Anderson, M.J., Roberts, C.D., Harvey, E.S., Stewart, A.L., Struthers, C.D., 2012. Diversity and Composition of Demersal Fishes along a Depth Gradient Assessed by Baited Remote Underwater Stereo-Video. Plos One 7(10), e48522-e48522.
133
Appendix 1 - Measures of precision for dissimilarity-based multivariate analysis of ecological communities
Final version published as: Anderson, M.J., Santana-Garcon, J., 2015. Measures of
precision for dissimilarity-based multivariate analysis of ecological communities.
Ecology Letters 18(1), 66-73.
Authors: Marti J. Anderson, New Zealand Institute for Advanced Study
(NZIAS), Massey University, Albany Campus,
Julia Santana-Garcon, The UWA Oceans Institute (M470) and School
of Plant Biology, The University of Western Australia, Australia,
Running title: Measures of precision for community ecology
Keywords: assemblage data; community ecology; dissimilarities; multivariate
analysis; PERMANOVA; precision; replicates; sampling methods; standard error;
variability
134
Abstract
Ecological studies require key decisions regarding the appropriate scale and
number of sampling units. No methods currently exist to measure precision for
multivariate assemblage data when dissimilarity-based analyses are intended to
follow. We describe tools to assess the appropriate size of a “replicate” and to
measure precision for communities. We propose a pseudo multivariate dissimilarity-
based standard error (MultSE) be used to assess sample-size adequacy and describe a
novel double-resampling method to quantify its uncertainty. These ideas are
extended to complex experimental designs by using the pseudo residual mean square
from a PERMANOVA analysis as a measure of variability, including within-cell
double-resampling methods, for rigorous inference.
Introduction
The measurement of ecological communities requires key decisions
regarding the size and number of individual sampling units that should be used in
order to characterise (and analyse) communities of interest. Two essential questions
are: (i) How big should a sampling unit be in order to adequately capture what we
mean (in a given study) by the word “replicate” and (ii) How many such replicates
are needed in order to get reasonably precise measures of multivariate variability for
purposes of testing relevant null hypotheses regarding community structure?
To measure the density or relative abundance of a single species (univariate),
methods exist to allow researchers to assess (e.g., from a series of replicates obtained
in preliminary pilot investigations) the adequacy of any given choice of sampling-
unit size and/or number (e.g., Andrew & Mapstone 1987; Downing & Downing
1992). In particular, for univariate data one may calculate a value of precision for a
given sample of n units of a given size. Precision measures the degree of
135
concordance among multiple estimates of a given parameter (such as the mean) for
the same population (Cochran & Cox 1957). This is reflected by the variability of an
estimate; precision increases as variance decreases. Although variance is
independent of sample size, precision is measurable from the sampling programme
and increases with the number and size of sampling units.
An appropriate measure of univariate precision can be calculated from a
given sample as the standard error divided by the mean; namely, ySEp /= , where
, s is the sample standard deviation, is the sample mean and n is the
number of sampling units. Note that, so defined, as values of p decrease, precision
improves. Measures of precision can be calculated from pilot data, a plot of p vs. n
can be drawn, and we expect a gradual decrease and levelling-off in the value of p
with increasing n in such a plot. One may then consider that a value of n around this
apparent asymptote in p (i.e., where the slope of the curve becomes very small)
would be reasonable to use for future investigations of the population, as further
increases in n would not result in substantially greater increases in precision.
Alternatively, given a set of pilot data, one may calculate a value of n required to
achieve a desired level of precision set a priori (such as p = 0.2, etc.) as
, rounded up to the nearest integer.
For multivariate data, especially multi-species count data that are gathered
with the intention of characterising ecological communities, there is no obvious
measure of multivariate precision that can be used to assist researchers in decision-
making processes regarding adequate sampling. First, what do we mean by “the
community”? We presume here that the researcher has already defined the
community conceptually initially by reference to a particular spatial and temporal
inferential domain (e.g., trees on Great Barrier Island, invertebrates within a given
depth-range on the Norwegian continental shelf within a particular pre-defined area
specified by latitude and longitude, etc.). This domain is also likely to include
specificity caused by logistic constraints in sampling organisms which further limits
nsSE /= y
2)]/([ ypsn =
136
inference (e.g., only macro-invertebrates large enough to be retained on a 0.5 mm
sieve, only trees greater than a certain diameter at breast height (dbh), etc.).
Although these decisions might be somewhat arbitrary, often based on convenience
or historical legacies (hence, the word “assemblage” is preferable to the word
“community” for multi-species datasets, see in particular Underwood 1986), we
presume the researcher is sampling from the same “universe” with respect to these
definitions within a given study (as in Colwell et al. 2004), and that what is required
to characterise the community, so defined, from a sampling perspective, will follow
directly from this.
Multivariate analyses of ecological communities are very often based on
measures of dissimilarity, such as Bray-Curtis, Hellinger or Jaccard, that
appropriately emphasise changes in the composition of the identities or relative
abundances of individual species (Legendre & Gallagher 2001; Clarke et al. 2006;
De Càceras et al. 2013). Thus, although a more classical Euclidean-based
multivariate analogue of the above univariate precision index might be considered
(e.g., such as the determinant of a variance-covariance matrix on normalized data,
see also Kotz & Lovelace 1998; Pearn & Kotz 2006), these implicitly or explicitly
rely on assumptions of multivariate normality, and are unlikely to shed any light on
ecologically relevant compositional information.
Non-parametric measures of precision for multivariate data have been
developed recently by Šiman (2014). These are based on the calculation of
multivariate quantiles using halfspace depth regions (Rousseeuw & Ruts 1999;
Serfling 2002). Although these ideas might well be applied fruitfully to sample
points in a principal coordinate (PCO) space that preserves a chosen dissimilarity
measure (e.g., Gower 1966; Legendre & Anderson 1999; McArdle & Anderson
2001), there remain important problems. The regions must be convex, the
calculations require a large number of data points, and the computations are so
demanding as to make them unfeasible for any more than 4 or 5 variables (Šiman
137
2014). However, datasets in community ecology typically are high-dimensional with
many species.
One aspect of community structure is richness – the number of species –
which is known simply to increase with sample size (or sample area). A plot of the
cumulative number of species with increases in the number of sampling units (a
species-accumulation curve, Gotelli & Colwell 2001; Magurran 2003) can be drawn
from a set of samples taken from a given area. The point at which this curve begins
to level-off can indicate the number of samples beyond which not many new species
will be encountered. The usual purpose of such plots, however, is generally to
estimate the total number of species in a given (usually large) area (e.g., Ugland et
al. 2003, Colwell et al. 2004; Gray et al. 2004), rather than to infer the number of
sampling units required to characterize variation in community structure within that
area. Furthermore, the size of individual sampling units is usually driven heavily by
logistic constraints (e.g., only 1m2 plots can be examined within a reasonable length
of time in the field while diving under SCUBA, etc.). Yet, if the sizes of individual
sampling units are too small, then many units may contain no species at all (leading
to undefined dissimilarity values) and there also may be many pairs of units that
have no (or very few) species in common, yielding dissimilarity values that are
uninformative (Clarke et al. 2006).
Our purpose here is to describe some simple approaches to aid in the
assessment of appropriate sampling of ecological communities for subsequent
analysis using dissimilarity-based multivariate methods (e.g., Clarke & Gorley 2006;
Anderson et al. 2008). First, we consider the goal of obtaining enough information
about the community at the level of an individual replicate that will yield reasonable
distributions of dissimilarity values – i.e., we address the question: “How many
sampling units, when pooled together (i.e., sub-samples), are needed (or, what size of
individual sampling unit is needed) to provide a reasonable measure of community
structure for comparative analysis?” In other words, what is reasonable to use as a
138
“replicate” sampling unit? We propose the use of simple graphics that show key
features of the distributions of dissimilarity measures with increasing sizes/numbers
of sub-samples.
Second, presuming an appropriate replicate sampling unit (potentially
comprised of a number of pooled sub-samples) has been defined, we then ask: “How
many replicates are needed to sufficiently characterize the community being sampled
with reasonable precision for subsequent dissimilarity-based multivariate analysis?”
We propose the use of a pseudo multivariate dissimilarity-based standard error
(MultSE) – a direct analogue to the univariate standard error – as a useful quantity
for assessing sample-size adequacy for ecological studies. This can be calculated
easily from a single set of pilot data or on residuals after conditioning on some more
complex sampling design or model. We further propose a double-resampling
technique (permutation followed by bootstrapping) in order to quantify uncertainty
in MultSE values with increasing numbers (or sizes) of sampling units.
Methods & Results
The methods proposed are illustrated through ecological examples. R code (R
Core Team 2014) and associated datasets (as *.csv files for input into R) are
provided in Supporting Information (see Appendix S1).
Choosing the number of sub-samples to pool in order to obtain “replicates”
In some cases, multivariate community data may be very sparse; individual
sampling units may contain very few species and arguably do not represent the
larger-scale community of interest very well. Hence, one may combine (pool) data
from several replicates (effectively treating them as sub-samples) before proceeding
with multivariate analyses. Several important dissimilarity measures commonly used
in ecology – namely, Bray-Curtis, Sørenson or Jaccard – lose resolution and can
139
become increasingly sporadic near their boundaries (Clarke et al. 2006). This tends
to occur when individual sampling units contain few or no species, hence many pairs
of units have no species in common. Two sampling units with no species in them are
also simply “undefined” by any of these three measures. While Clarke et al. (2006)
suggested a “feathering-in” of a dummy species (present everywhere) to help combat
this problem, another approach would be to pool information across multiple small-
scale sampling units to obtain a larger-scale replicate. But how many should be
pooled together to achieve an appropriately informative “replicate”?
Santana-Garcon et al. (2014) studied the vertical distribution of pelagic fish
assemblages off Western Australia using pelagic Baited Remote Underwater stereo-
Video systems (BRUVs) and determined that reasonably precise estimates of total
abundance and richness of fish assemblages could be achieved using ca. 8 replicate
deployments of video cameras each of 120 minutes’ duration. Similar gear was used
to study fish assemblages at the Houtman-Abrolhos Islands (28°40' S, 113°50' E).
Here, the sampling design consisted of n = 12 pelagic stereo-BRUV deployments at
a depth of 10 m in the water column from each of 6 sites: 3 were inside and 3 were
outside of an area closed to fishing (see Santana-Garcon et al. in review, for details).
A total of 24 fish species were recorded in this study. Preliminary analyses based on
a variety of resemblance measures (not shown) found no significant effects for any
of the factors, so the study effectively offers a total of N = 72 sampling units that
may be considered as genuine replicate measures of the same “universe” – in this
case, the pelagic fish community at the Abrolhos Islands.
These data are extremely sparse and are dominated by zeros. A species-
accumulation curve does not appear to show any clear “levelling-off” (Fig. 1a). The
distribution of Bray-Curtis dissimilarities among the 72 original replicates after
square-root transformation (Fig. 1b) shows a highly skewed distribution, with 76.4%
of the pair-wise dissimilarities being equal to 1.0 (no species in common) and 14.8%
being undefined, corresponding to pairs of sampling units that had no species in
140
them. When groups of individual sampling units (which we should properly consider
now as “sub-samples”) are pooled together, the distribution of dissimilarities
improves substantially (Fig. 2).
Although semi-parametric (e.g., PERMANOVA, Anderson et al. 2008) and
non-parametric (e.g., ANOSIM, Clarke & Gorley 2006) multivariate analyses based
on dissimilarity measures or their ranks do not assume anything specific regarding
the distribution of dissimilarities used as input, it is nevertheless desirable that the
chosen measure being used has some genuine information content, some reasonable
resolution (allowing distinction between large and small dissimilarities across a
given dataset), as well as having an appropriate contextual meaning for tests of
hypotheses regarding relevant ecological properties (such as composition) for the
communities of interest. For example, if sparseness has arisen because of some
ecological phenomenon under study (such as predation), then perhaps a different
dissimilarity measure which treats joint absences as a sign of similarity should be
considered (such as the Gower measure, e.g., Langlois et al. 2005). Similarly, if
dissimilarities of 100% (no species in common) have arisen because the data
genuinely span very large biogeographic distances, then a taxonomic resemblance
measure (e.g., Clarke et al. 2006) may be appropriate. However, if the reason for the
sparseness is indeed simply that the scale of the sampling units is simply too small to
capture much, then treating these as sub-samples and pooling them together is in
order before proceeding. In particular, methods that fit linear models in the space of
the original resemblance measure (i.e., PERMANOVA, DISLTM, dbRDA or CAP –
see Anderson et al. 2008) rely critically on the information content of the measure
itself.
We propose randomizing the order of the sampling units and plotting the mean
(along with 0.025 and 0.975 quantiles) for (at least) two quantities: (i) the proportion
of dissimilarities that are equal to 1.0 (100% dissimilar), and (ii) the proportion of
dissimilarities that are undefined (i.e., “NaN”), with increasing numbers of pooled
141
sub-samples being used to form the replicate sampling units. Doing this for the
Abrolhos dataset shows no undefined dissimilarities remain once the number of sub-
samples exceeds 2, and no replicates that are 100% different remain after pooling
more than 10 sub-samples (Fig. 1b). This accords with what is shown in Fig. 2 for
the original ordering of sub-samples, where a much more reasonable distribution of
dissimilarities is observed when the number of sub-samples per replicate is greater
than or equal to 8.
A multivariate measure of pseudo standard error
As a direct analogue to the univariate measure of precision, we consider a
multivariate measure of pseudo standard error (MultSE, first described by Anderson
et al. 2001), as follows. Let D be a ( ) matrix of dissimilarities among all
pairs of sampling units, i = 1, ..., n and j = 1, ..., n. Using Huygens’ theorem (e.g.,
Legendre & Anderson 1999; Anderson 2001), the sum of squared inter-point
dissimilarities divided by the number of sampling points (n) is directly interpretable
as a multivariate measure of pseudo sums of squares in the space of the chosen
dissimilarity measure:
(1)
The quantity in equation (1) is also equivalent to the sum-of squared distances from
individual sampling points to their centroid in the space of the chosen resemblance
measure (Anderson 2001). Furthermore, dividing this by (n – 1) yields a quantity
that is directly interpretable as a multivariate measure of pseudo variance in the
space of the chosen dissimilarity measure:
(2)
nn× }{ ijd
ndsn
i
n
ij
ij /)1(
1 1
2∑ ∑
−
= +=
=
ndvn
i
n
ij
ijn/
)1(
1 )1(
2)1(
1 ∑ ∑−
= +=−
=
142
Indeed, in the case of a single variable and Euclidean distance, s is equal to the usual
classical univariate sum-of-squares and v is equal to the usual classical unbiased
measure of the sample variance.
Note that the word “pseudo” has been used throughout here to distinguish s in
(1) above from the classical multivariate sums-of-squares-and-cross-products
(SSCP) matrix and v in (2) above from the classical multivariate sample variance-
covariance matrix. The classical matrices will clearly take into account correlation
structures in the multivariate (Euclidean) space of the original variables, whereas
these pseudo measures do not. To see this, note that in the case of multivariate data
and Euclidean distance, s in (1) is equal to the trace of the classical SSCP matrix and
v in (2) is equal to the trace of the classical sample variance-covariance matrix.
Finally, based on the above, a multivariate pseudo standard error therefore can
be calculated as:
(3)
Although the univariate measure of precision takes the standard error divided by the
sample mean, this is not possible in the present case because the sample mean is
clearly a vector here (being multivariate). However, we consider that m in (3) is
perfectly suitable as a measure of precision within a given study, because the
estimate of the mean (generally obtained as the sample average in univariate
analysis) is unbiased for any sample size. The only consequence of failing to divide
by the mean is to produce a value which is not standardized in any way, hence can
only be used within the context of a given study, and not compared among studies.
Thus, although some chosen value of precision for univariate analysis might be
suggested as a rule of thumb to be applied to studies across a given field (e.g., p =
0.2 or p = 0.5, etc.), the value of m in (3) is dependent on the scale of the
dissimilarity measure chosen, so no such generalization for its value across multiple
studies is apparent. However, as illustrated by the examples below, m will still
nvm /=
143
nevertheless be quite useful for examining relative precision for different sample
sizes within a given study.
We propose that, given a set of pilot multivariate data having a total of N
sampling units, one may draw a random subset (without replacement) of size n = 2,
3, 4, ..., N and for each of these random subsets, a value of m can be calculated,
which may be denoted in each case by , , ... . Repeat this process for
a very large number of random draws (say 10,000), then calculate the mean values
obtained under permutation for each sample size as , , ... . A plot of
vs. k for k = 2, 3, ..., N will then provide the graphic we require to assess
precision with increasing sample size. We propose that these means be calculated
using sampling without replacement (i.e., using a permutation method) specifically
to ensure that the values obtained will be unbiased.
Next, we propose that a bootstrapping method be used to draw error bars on
the plot. It is clear that the number of possible permutations for n = N is 1 (i.e., this is
simply the full set of data) and further that the number of random subsets without
replacement that would yield unique values for (equivalent to simply
leaving out one observation at a time) is also limited to just N possible values.
Hence, we propose obtaining a further random subset as a random draw with
replacement (i.e., a bootstrap sample), of size n = 2, 3, 4, ..., N and for each of these
to calculate a value of m, denoted in each case by , , ... . Consider the
0.025 and 0.975 quantiles of the bootstrap distribution for each sample size n = k for
k = 2, 3, ..., N as and , respectively. Unlike the permutation
approach, the bootstrap distribution is known to be biased (e.g., Davison & Hinkley
1997). An empirical estimate of the bias, for each sample size k, is given by:
(4)
Hence, bias-adjusted 0.025 and 0.975 quantiles are provided by
π2=nm π
3=nm πNnm =
π2=nm π
3=nm πNnm =
πknm =
π)1( −= Nnm
β2=nm β
3=nm βNnm =
)025.0(βknq = )975.0(β
knq =
)( βπknknkn mmb === −=
144
and , (5)
respectively. As our proposed method uses permutations to obtain mean values for
the plot, and a separate bias-adjusted bootstrapping approach to obtain appropriate
error bars, we shall refer to this overall process generally as a “double resampling”
method. We note also that clearly other choices of quantiles may easily be calculated
here in a similar fashion (e.g., 0.25, 0.75, etc.), if desired.
Choosing an appropriate size or number of replicates using MultSE
The precision of univariate data has been assessed in order to optimise the
sampling effort of BRUV techniques in estuarine (Gladstone et al. 2012) and pelagic
environments (Santana-Garcon et al. 2014). The parameters potentially affecting the
sampling ability of pelagic stereo-BRUVs include soak time (time that cameras are
left recording underwater) and the number of replicate deployments.
Data were collected in March 2012 near Tantabiddi Passage in Ningaloo
Marine Park (21°53’ S, 113°56’ E), Western Australia. The pelagic stereo-BRUVs
used in this study are described in Santana-Garcon et al. (2014). The systems were
designed to remain in the mid-water at ca. 5 m depth (7.17 ± 0.54 m, Mean ± SE), 30
m above the bottom and were deployed for a soak time of up to 3 hours. There were
12 replicate deployments (6 at each of 2 sites) obtained during day light hours. Video
images were analysed using the software ‘EventMeasure (Stereo)’ (SeaGIS Pty Ltd).
All fish recorded were quantified and identified to the lowest taxonomic level
possible. To avoid repeating counts of individual fish re-entering the field of view, a
conservative measure of relative abundance (MaxN) was recorded as the maximum
number of individuals of the same species appearing at the same time (Priede et al.
1994). MaxN was recorded for each of three different soak times: 60, 120 and 180
minutes.
})025.0({ knkn bq == +β })975.0({ knkn bq == +β
145
No previous assessment has been done to optimize multivariate precision for
dissimilarity-based community analysis of these fish assemblages. We wished to
identify: (i) an appropriate length of time that cameras should be deployed; and (ii)
an appropriate number of replicate deployments beyond which no substantial
increases in MultSE would accrue.
First, we examined the means and quantiles of MultSE with increasing sample
size using either the permutation or the bootstrap method alone for the 60 minute
soak time, with analyses based on square-root transformed abundances and Bray-
Curtis dissimilarities. The permutation method alone yields obvious constraints on
the error bars for larger sample sizes, while the bootstrap method alone yields means
that are biased downwards (Fig. 3a). This motivates the use of our double resampling
approach, which shows that no appreciable precision is gained by deploying the
cameras longer than 120 minutes, and that a levelling-off in MultSE occurs for
sample sizes around n = 7 or 8 (Fig. 3b).
We readily concede that the idea of “levelling-off” must rest in the eyes of the
beholder; the approach we suggest is intended to be a heuristic diagnostic tool only,
and not a prescription. Nevertheless, the analyses of MultSE in this case coincide
well with the results obtained for the precision of either of the univariate measures of
richness or total abundance (Santana-Garcon et al. 2014). This coincidence might
not always occur, however, and will depend on the nature of the data and the
characteristics that are emphasised by the chosen dissimilarity measure.
A further example of the proposed method yields additional insight. We
consider next an analysis of soft-sediment macrofauna described by Ellingsen &
Gray (2002), who studied beta diversity and its relationship with environmental
heterogeneity in benthic marine systems over large spatial scales in the North Sea.
Samples of soft-sediment macrobenthic organisms were obtained from 101 sites
occurring in five large areas along a transect of 15 degrees of latitude. A total of 809
146
taxa were recorded overall, and replicate sampling units here consisted of
abundances pooled across five benthic grabs (sub-samples) obtained at each site. For
more details, see Ellingsen & Gray (2002), Ugland et al. (2003) and Anderson et al.
(2006). Of interest here is to identify the number of replicate sampling units required
to adequately sample the benthic communities for comparative analysis on the basis
of the Jaccard resemblance measure. MultSE was calculated with the double
resampling method for increasing numbers of replicates separately for each of the
five areas. Results show clear heterogeneity in the MultSE values across the 5
different areas, with the greatest variability occurring in area 3, followed by areas 4
and 5, then area 2 and finally area 1 (Fig. 4a). This agrees naturally with the
differences in multivariate dispersions found previously (using PERMDISP;
Anderson 2006) using the full set of data (Anderson et al. 2006; 2008). What is
added to our understanding here by considering values of MultSE with increasing
sample size (including the double-resampling method) is that these differences in
dispersion are already apparent once sample sizes within each area reach about n =
10 to 12 sampling units. Knowledge of what level of sampling may be required to
identify significant heterogeneity can be quite useful. When dealing with large areas,
for example, better estimates of overall richness can be obtained by generating
species accumulation curves for each of several smaller sub-sampled areas, due to
heterogeneity in community structure (turnover) across the larger sampling extent
(Ugland et al. 2003; 2005).
Calculating MultSE for more complex designs
The idea of using MultSE as a measure of precision can be readily extended for
use with more complex sampling designs by noting that the residual mean square
provided from a PERMANOVA analysis (Anderson 2001) is itself a measure of
multivariate pseudo error variance in the space of the chosen dissimilarity measure.
Thus, as described in McArdle & Anderson (2001), let D = { } be an ( )
dissimilarity matrix, let A = { } = { }, and let G = be
ijd nn ×
ija 22
1ijd− )()( 11 11IA11I ′−′−
nn
147
Gower’s (1966) centered matrix, where 1 is a column of 1’s of length n and I is an (
) identity matrix. Next, let matrix X be an ( ) design matrix of full rank
that codes for the full PERMANOVA model, including intercept (for example, in a
one-way ANOVA design, g would be equal the number of groups). Finally, let
matrix H be the usual projection matrix for the design, namely .
The residual mean square for the PERMANOVA model (Anderson et al. 2008)
estimates error variability in the space of the resemblance measure and is calculated
as:
(6)
where “tr” denotes the trace of a matrix. After obtaining v from equation (6), one can
then apply equation (2) to calculate precision and perform the double resampling
method for increasing numbers of replicates, as described above. Note however that
the replicates here occur within the structure of a model framework – a specific
ANOVA design. This means the double resampling method must be applied
separately for increasing numbers of samples within each individual cell in the
design. For example, in the case of a one-way ANOVA model, the double
resampling is done separately within each group.
By way of example, consider data on fish assemblages from the Poor Knights
Islands in New Zealand, described by Willis & Denny (2000) and Anderson &
Willis (2003). Divers counted temperate reef fishes belonging to each of 62 species
in each of nine 25 m × 5 m transects at each site. Data from the transects were
pooled at the site level and a number of sites around the Poor Knights Islands were
sampled at each of three different times: September 1998 (15 sites), March 1999 (21
sites) and September 1999 (20 sites). These span the point in time when the Poor
Knights Islands were classified as a no-take marine reserve (October 1998).
Analyses of MultSE with the double resampling method vs. increasing sample
sizes for each of the three times of sampling (based on log(x+1)-transformed data
nn× gn ×
XXXXH ′′= −1][
}){()(1 GHI −=−
trvgN
148
and Bray-Curtis dissimilarities), was initially done separately for each time and
showed no sign of heterogeneity of dispersions, unlike the Norwegian macrofauna
(cf. Fig. 4b and Fig. 4a). The residual mean square (quantity v from equation (6)
above) can therefore legitimately be calculated as a single measure of error
variability from the residuals of a one-way ANOVA model for the three groups
corresponding to the three times of sampling (Fig. 5). Precision in the residual mean
square levels off and does not improve substantially for these assemblages once the
number of sites sampled within each time reaches around 7 or 8 (Fig. 5).
Discussion
We have provided some tools for assessing an appropriate number and scale of
sampling units for dissimilarity-based multivariate analysis of ecological data. First,
we consider it is wise, before embarking on dissimilarity-based modelling, to
examine the distributions of dissimilarities directly to check if large numbers of
values approach the limits of what the chosen dissimilarity measure is able to
measure. If so, then perhaps the choice of dissimilarity measure should be revisited.
Another possibility is that large numbers of sparse sampling units are present which
should be pooled together to provide enough information regarding the community
of interest prior to subsequent analysis. We have provided a general tool for
assessing the numbers of sub-samples that might be chosen to define a reasonable
“replicate” in this context.
Second, we have proposed a measure of precision for multivariate data
(MultSE) based on inter-point dissimilarities that is a direct analogue of the well-
known univariate standard error. A double resampling method allows unbiased
measures of variation in MultSE to be obtained. We recognize the clear limitations of
this approach. The measure takes no account of the shape of the multivariate data
cloud, only its overall dispersion (size). Future developments should include better
characterization of the actual shape of the data cloud, which might be achieved, for
149
example, through the use of multivariate kernel density estimation (KDE) in the
principal coordinate (PCO) space of the resemblance measure. This will require
enhanced computer power and programming to cope with analyses of high-
dimensional datasets like those typically found in ecology. Future work should focus
also on how such approaches might compare with, say, a more classical approach to
measure precision, such as the determinant of a normalized variance-covariance
matrix in the PCO space.
A further limitation of the approach outlined here is its scale-dependency.
Values of MultSE will necessarily depend on the dissimilarity measure used, so
cannot be easily compared across studies. More work is needed to assess what levels
of precision (as measured by MultSE) might be acceptable for different resemblance
measures that are bounded and have a common scale, such as Bray-Curtis.
Finally, although we have described the analysis of precision using the residual
mean square from a PERMANOVA model only for a one-way case, extensions to
higher-way models are clearly very easy to implement using the formulation in
equation (6) for any model specified fully within matrix X. Some care may be
needed in such cases, however, to ensure that the double resampling is performed
within appropriate combinations of factor levels, given the particular higher-way
design of interest.
Acknowledgements
MJA was supported by a Royal Society of New Zealand Marsden Grant
MAU1005. Logistic support for acquiring the data analysed here from Western
Australia was provided by the University of Western Australia and the Department
of Fisheries, Government of Western Australia.
150
References
Anderson, M.J. (2001). A new method for non-parametric multivariate analysis of
variance. Austral Ecol., 26, 32-46.
Anderson, M.J. (2006). Distance-based tests for homogeneity of multivariate
dispersions. Biometrics, 62, 245-253.
Anderson, M.J., Ellingsen, K.E. & McArdle, B.H. (2006). Multivariate dispersion as
a measure of beta diversity. Ecol. Lett., 9, 683-693.
Anderson, M.J., Gorley, R.N. & Clarke, K.R. (2008). PERMANOVA+ for PRIMER:
Guide to Software and Statistical Methods. PRIMER-E, Plymouth, UK.
Anderson, M.J., Saunders, J.E. & Creese, R.G. (2001). Ecological monitoring of the
Okura Estuary. Report 1: Results of a pilot study. Technical report prepared
by Auckland UniServices for Auckland Regional Council, University of
Auckland, New Zealand, 29 pp.
Anderson, M.J. & Willis, T.J. (2003). Canonical analysis of principal coordinates: a
useful method of constrained ordination for ecology. Ecology, 84, 511-525.
Andrew, N.L. & Mapstone, B.D. (1987). Sampling and the description of spatial
pattern in marine ecology. Oceanogr. Mar. Biol. Annu. Rev., 25, 39-90.
Clarke, K.R. & Gorley, R.N. (2006). PRIMER v6: User manual/Tutorial. PRIMER-
E Ltd, Plymouth.
Clarke, K.R., Somerfield, P.J. & Chapman, M.G. (2006). On resemblance measures
for ecological studies, including taxonomic dissimilarities and a zero-
adjusted Bray-Curtis coefficient for denuded assemblages. J. Exp. Mar. Biol.
Ecol., 330, 55-80.
151
Cochran, W.G. & Cox, G.M. (1957). Experimental designs, 2nd edition. John Wiley
& Sons, New York.
Colwell, R.K., Mao, C.X. & Chang, J. (2004). Interpolating, extrapolating, and
comparing incidence-based species accumulation curves. Ecology, 85, 2717-
2727.
Davison, A.C. & Hinkley, D.V. (1997). Bootstrap methods and their application.
Cambridge University Press, Cambridge.
De Càceras, M., Legendre, P. & He, F. (2013). Dissimilarity measurements and the
size structure of ecological communities. Method. Ecol. Evol., 4, 1167-1177.
Downing, J.A. & Downing, W.L. (1992). Spatial aggregation, precision, and power
in surveys of freshwater mussel populations. Can. J. Fish. Aquat. Sci., 49,
985-991.
Ellingsen, K.E. & Gray, J.S. (2002). Spatial patterns of benthic diversity: is there a
latitudinal gradient along the Norwegian continental shelf? J. Anim. Ecol.,
71, 373-389.
Gladstone,W., Lindfield, S., Coleman, M. & Kelaher, B. (2012). Optimisation of
baited remote underwater video sampling designs for estuarine fish
assemblages. J. Exp. Mar. Biol. Ecol., 429, 28-35.
Gotelli, N.J. & Colwell, R.K. (2001). Quantifying biodiversity: procedures and
pitfalls in the measurement and comparison of species richness. Ecol. Lett., 4,
379-391.
Gower, J.S. (1966). Some distance properties of latent root and vector methods used
in multivariate analysis. Biometrika, 53, 325-338.
152
Gray, J.S., Ugland, K.I. & Lambshead, J. (2004). Species accumulation and species
area curves – a comment on Scheiner (2003). Global Ecol. Biogeogr., DOI:
10.1111/j.1466-822X.2004.00114.x, 1-4.
Kotz, S. & Lovelace, C.R. (1998). Process capability indices in theory and practice.
Oxford University Press, New York.
Langlois, T.J., Anderson, M.J. & Babcock, R.C. (2005). Reef-associated predators
influence adjacent soft-sediment communities. Ecology, 86, 1508-1519.
Legendre, P. & Anderson, M.J. (1999). Distance-based redundancy analysis: testing
multispecies responses in multifactorial ecological experiments. Ecol.
Monogr., 69, 1-24.
Legendre, P. & Gallagher, E.D. (2001). Ecologically meaningful transformations for
ordination of species data. Oecologia, 129, 271-280.
Magurran, A.E. (2003). Measuring biological diversity. Blackwell Publishing,
Oxford.
McArdle, B.H. & Anderson, M.J. (2001). Fitting multivariate models to community
data: a comment on distance-based redundancy analysis. Ecology, 82, 290-
297.
Pearn, W.L. & Kotz, S. (2006). Encyclopedia and handbook of process capability
indices. World Scientific Publishing, Singapore.
Priede, I.G., Bagley, P.M., Smith, A., Creasey, S. & Merrett, N.R. (1994).
Scavenging deep demersal fishes of the Porcupine Seabight, Northeast
Atlantic - Observations by baited camera, trap and trawl. J. Mar. Biol. Assoc.
U. K., 74, 481-498.
153
R Core Team. (2014). R: A language and environment for statistical computing. R
Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0,
URL http://www.R-project.org/ .
Rousseeuw, P.J. & Ruts, I. (1999). The depth function of a population distribution.
Metrika 49: 213-244.
Santana-Garcon, J., Newman, S.J., Langlois, T.J. & Harvey, E.S. (In review). Effects
of a spatial closure on highly mobile fish species: an assessment using
pelagic stereo-BRUVs. J. Exp. Mar. Biol. Ecol.
Santana-Garcon, J., Newman, S.J. & Harvey, E.S. (2014). Development and
validation of a mid-water baited stereo-video technique for investigating
pelagic fish assemblages. J. Exp. Mar. Biol. Ecol., 452, 82-90.
Serfling, R. (2002). Quantile functions for multivariate analysis: approaches and
applications. Stat. Neerl., 56, 214-232.
Šiman, M. (2014). Precision index in the multivariate context. Commun. Stat. Theor.
M., 43, 377-387.
Ugland, K.I., Gray, J.S. & Ellingsen, K.E. (2003). The species-accumulation curve
and estimation of species richness. J. Anim. Ecol., 72, 888-897.
Ugland, K.I., Gray, J.S. & Lambshead, P.J.D. (2005). Species accumulation curves
analysed by a class of null models discovered by Arrhenius. Oikos, 108, 263-
274.
Underwood, A.J. (1986). What is a community? In: Patterns and processes in the
history of life (eds Raup, D.M. & Jablonski, D.). Springer-Verlag, Berlin, pp.
351-367.
Willis, T.J. & Denny, C.M. (2000). Effects of Poor Knights Islands Marine Reserve
on demersal fish populations. Report to the Department of Conservation,
154
Research Grant No. 2519, Leigh Marine Laboratory, University of Auckland,
Auckland, New Zealand.
SUPPORTING INFORMATION
Additional Supporting Information may be downloaded via the online version
of this article at Wiley Online Library (www.ecologyletters.com).
155
FIGURES
Fig. 1. (a) Species accumulation curve vs number of sampling units and (b)
proportion of Bray-Curtis dissimilarities equal to 1.0 (no species in common) or
undefined (i.e. NaN) for the Abrolhos dataset (N = 72, p = 24) with increasing
numbers of sampling units being pooled together to define a replicate. Error bars
indicate the 2.5 and 97.5 percentiles of the distribution of values obtained under
1000 permutations of the order of sampling units.
156
Fig. 2. Frequency distributions for Bray-Curtis dissimilarities for the Abrolhos
dataset (also showing the frequency of undefined (i.e. NaN) dissimilarities) with
increasing numbers of sampling units being pooled together to define a replicate,
from n = 2 (top) to n = 10 (bottom). Pooling of units was done simply in order of
their occurrence in the data file.
157
Fig. 3. Multivariate pseudo standard error as a function of sample size on the basis of
Bray-Curtis dissimilarities calculated on square-root transformed fish abundance
data from Ningaloo (a) for a soak time of 60 minutes only, showing results (means
with 2.5 and 97.5 percentiles as error bars) from 10,000 resamples obtained using
either the permutation or bootstrap approach and (b) for all three soak times, using
the double-resampling method, with permutation-based means and bias-adjusted
bootstrap-based error bars (with 10,000 resamples for each).
158
Fig. 4. Multivariate pseudo standard error as a function of sample size (a) for each of
5 different areas on the basis of Jaccard dissimilarities for macrofaunal communities
from the Norwegian continental shelf and (b) for each of 3 different times on the
basis of Bray-Curtis dissimilarities calculated on log(x+1)-transformed counts of
fishes from the Poor Knights Islands, New Zealand. A double-resampling scheme
was used to generate means for each sample size using 10,000 permutations and
error bars as bias-adjusted 2.5 and 97.5 percentiles from 10,000 bootstrap resamples.
159
Fig. 5. Multivariate pseudo residual mean square from a one-way ANOVA model as
a function of sample size on the basis of Bray-Curtis dissimilarities calculated on
log(x+1)-transformed counts of fishes from the Poor Knights Islands, New Zealand.
A double-resampling scheme was used to generate means for each sample size using
10,000 permutations and error bars as bias-adjusted 2.5 and 97.5 percentiles from
10,000 bootstrap resamples. Resampling was done separately within each of the
three sampling times (i.e., within each of the 3 groups in the one-way ANOVA
model design).
160