habitat characteristics, density patterns and...
TRANSCRIPT
Habitat Characteristics, Density Patterns and Environmental
Niches of Indo-Pacific Humpback Dolphins (Sousa chinensis)
of the Pearl River Estuary and Eastern Taiwan Strait
A dissertation submitted to the Committee on Graduate Studies
in partial fulfillment of the requirements for the degree of
Doctor of Philosophy
in the Faculty of Arts and Science
TRENT UNIVERSITY
Peterborough, Ontario, Canada
© Copyright by Lauren E. Dares 2019
Environmental and Life Sciences Ph.D. Graduate Program
January 2019
ii
Abstract
Habitat Characteristics, Density Patterns and Environmental Niches of Indo-Pacific
Humpback Dolphins (Sousa chinensis) of the Pearl River Estuary and Eastern
Taiwan Strait
Lauren E. Dares
The purpose of this thesis is to quantify the habitat characteristics, density patterns and
environmental niches of two groups of Indo-Pacific humpback dolphins: Chinese white
dolphins (CWD) of the Pearl River estuary (PRE), and Taiwanese white dolphins (TWD,
=Taiwanese humpback dolphin, THD) found in the eastern Taiwan Strait (ETS). Much
work has already been done on the habitat use of CWDs in parts of the PRE, so the
purpose of my first two chapters was to advance knowledge of the TWD to a comparable
level. Chapter 2 contains the first published description of the relatively shallow, inshore,
estuarine habitat of the TWD. General environmental characteristics and observed group
sizes were consistent with other populations of humpback dolphins, and group sizes were
not correlated with the environmental variables measured during surveys. Chapter 3
investigated density patterns of ETS humpback dolphins, finding spatiotemporal
heterogeneity across the study area. Humpback dolphin densities fluctuated from year to
year, but some parts of the study area were consistently used more than others.
Environmental characteristics again did not influence dolphin densities, though more
dolphins than expected were sighted in waters adjacent to major land reclamations, which
may be related to the location of these areas close to major rivers. In Chapter 4, niches of
the TWD and CWDs found in the PRE were compared using species distribution models,
which indicated significant niche overlap. This may be due to niche conservatism
iii
maintaining similar fundamental niches between the two groups since their historical split
>10,000 years ago, or a result of the intrinsic biotic factors that influence occurrence data
affecting the hypervolume dimensions of each realized niche in similar ways. Geographic
predictions indicate that most of the TWD’s range has likely been surveyed, and that
there may be connectivity between PRE humpback dolphins and at least one
neighbouring putative population due to continuous predicted suitable habitat in waters
that remain poorly surveyed. Overall, my thesis demonstrates that density patterns may
vary over time, but on a broad temporal scale, these two allopatric groups of Indo-Pacific
humpback dolphins have similar habitat requirements in geographically isolated, but
environmentally similar locations.
Keywords:
Sousa chinensis, habitat, density, species distribution model, niche overlap
iv
Preface
I have written my dissertation in manuscript format, as two of my chapters have been
published in the peer-reviewed literature, and the third will be submitted for publication
in the future, all with myself as the first author. Chapter 2 was published in Aquatic
Mammals, and Chapter 3 in Estuarine, Coastal and Shelf Science. Each chapter has been
written as a standalone manuscript, and the published chapters are in the style of the
journal in which they have been published. All of my research has been conducted in
collaboration with others, and I have used the plural, “we”, where appropriate throughout
this thesis. The title page of each chapter contains the names of collaborators. Permission
to reprint articles from copyright holders can be found in Appendix A.
v
Acknowledgements
Before beginning my studies at Trent, never in my wildest dreams would I have
imagined myself pursuing a Ph.D. on dolphins at a land-locked university in central
Ontario (much less a species found halfway around the world!), and there are many
people to thank who made it possible. First of all, I would like to thank my supervisor,
Dr. Bradley White, for his support throughout this journey, for always emphasizing the
big picture, and encouraging me to find the “story” in my work. I would also like to thank
my committee members, Drs. Dennis Murray and Bruce Pond for all of their time,
resources and invaluable advice given over the years. I am especially grateful to Bruce
for stepping up during the last few months of my thesis to ensure I made it past the finish
line.
I was incredibly fortunate to have been able to participate in field work in both
Hong Kong and Taiwan, experiencing the culture in each setting as well as observing
humpback dolphins in their natural habitats. For those opportunities, as well as their
guidance and expertise, I would like to thank Drs. John Wang and Samuel Hung, without
whom none of this work would have been possible. Data collected in Taiwan is part of an
ongoing research project conducted by CetAsia Research Group with financial support
from Wild At Heart Legal Defense Association (Taiwan), the Matsu Fish Conservation
Union, Small Cetacean Fund of the International Whaling Commission, FormosaCetus
Company Ltd., Ocean Park Conservation Foundation Hong Kong, Hong Kong Cetacean
Research Project, and Hong Kong Dolphin Conservation Society. Additional thanks to
Shih-Chu Yang for her work in Taiwan, and to Sarah Dungan and Shannon Montgomery
for their roles in organizing and transcribing survey datasheets. The long-term monitoring
vi
program on humpback dolphins of the PRE is conducted by the Hong Kong Cetacean
Research Project for the Agriculture, Fisheries and Conservation Department of the Hong
Kong Special Administrative Region. I would like to thank the many (past and present)
staff members and interns of the Hong Kong Dolphin Conservation Society I had the
pleasure of meeting while I was in Hong Kong for their assistance in the field and
introducing me to Hong Kong culture, especially Perry Chan, Vincent Ho, Viena Mak,
and Joanne Yuen.
In addition to the dolphin occurrence data collected by the aforementioned
groups, Chapter 4 would not have been possible without the open access remotely sensed
environmental data available from the Ocean Biology Processing Group (OBPG;
oceancolor.gsfc.nasa.gov), Scripps Institute of Oceanography (topex.ucsd.edu) and
Oregon State University (science.oregonstate.edu/ocean.productivity). The facilities of
the Shared Hierarchical Academic Research Computing Network (SHARCNET;
sharcnet.ca) and Compute/Calcul Canada were also indispensable in the completion of
Chapter 4, without access to which I would likely still be running models today. Also
thanks to Aquatic Mammals for granting permission to reprint an article published in their
journal as Chapter 2 of my thesis.
I am also grateful to the many lab members that have come and gone over the
years for their comments and suggestions on my own work, scientific discussion, as well
as for opportunities to collaborate; thanks to Claryana Araújo-Wang, Jordan Hoffman,
Shiva Javdan, Michelle Klein and Katherine Wright. Thanks also to Adrian Borlestean
for coding help, thesis proofreading, and being a sounding board for ideas.
vii
Too many people to name have helped me in countless ways throughout my time
at Trent, so, to be brief, thank you to: the Trent University and Peterborough Rowing
Clubs for being a truly welcoming community; the many graduate students and faculty
members for their support and mutual commiseration when necessary, especially Shiva
Javdan, Adrian Borlestean, Cayla Austin, Natasha Serrao, Katarina Cetinić, and Sasha
Newar; and also to Joanna Barber, Katy Langille, Andrea Steadman, Brittany Stanyon,
Annalise Ferraro, and Kirsten Steadman, for giving me perspective and making sure I
always have something to look forward to. Thanks also to Pippin, my constant
companion, for making sure I never have quite enough room on my desk to get anything
done efficiently.
Last, but certainly not least, I would like to thank my family - my parents, Chris
and Nancy, my grandmother, Joan, and my sisters, Carolyn and Angela - for their
unwavering support and encouragement over the years, and for always knowing I’d get it
done, in the end.
“So long, and thanks for all the fish.” – Douglas Adams
viii
Table of Contents
Abstract ............................................................................................................................... ii
Preface ................................................................................................................................ iv
Acknowledgements ............................................................................................................. v
Table of Contents ............................................................................................................. viii
List of Tables ..................................................................................................................... xi
List of Figures .................................................................................................................. xiii
Chapter 1 – General Introduction ................................................................................... 1
1.1 – Species’ Habitats and Niches ..................................................................................... 1
1.2 – Genus Sousa ............................................................................................................... 3
1.2.1 – Humpback Dolphins of the Pearl River Estuary ........................................... 5
1.2.2 – Humpback Dolphins of the Eastern Taiwan Strait ........................................ 8
1.3 – Thesis Structure ......................................................................................................... 9
1.4 – References ................................................................................................................ 11
1.5 – Figures ...................................................................................................................... 15
Chapter 2 – Habitat Characteristics of the Critically Endangered Taiwanese
Humpback Dolphins (Sousa chinensis) of the Eastern Taiwan Strait ....................... 16
2.1 – Short Note: Habitat Characteristics of the Critically Endangered Taiwanese
Humpback Dolphins (Sousa chinensis) of the Eastern Taiwan Strait .............................. 17
2.2 – Literature Cited ........................................................................................................ 24
2.3 – Tables and Figures ................................................................................................... 28
Chapter 3 – Spatiotemporal Heterogeneity in Densities of the Taiwanese Humpback
Dolphin (Sousa chinensis taiwanensis) .......................................................................... 32
3.0 – Abstract .................................................................................................................... 33
3.1 – Introduction .............................................................................................................. 34
3.2 – Materials and Methods ............................................................................................. 35
3.2.1 – Surveys ........................................................................................................ 35
3.2.2 – Study Design and Data Analyses................................................................. 36
3.2.2.1 – Spatiotemporal Heterogeneity ............................................................. 36
3.2.2.2 – Environmental Influences .................................................................... 37
3.3 – Results ...................................................................................................................... 39
ix
3.3.1 – Even (10km) Blocks .................................................................................... 39
3.3.2 – Reclamation Blocks ..................................................................................... 40
3.3.3 – Environmental Factors ................................................................................. 41
3.4 – Discussion ................................................................................................................ 41
3.5 – References ................................................................................................................ 52
3.6 – Tables and Figures ................................................................................................... 56
Chapter 4 – Niche Similarity Between Indo-Pacific Humpback Dolphins (Sousa
chinensis) of the Pearl River Estuary and Eastern Taiwan Strait ............................. 59
4.0 – Abstract .................................................................................................................... 60
4.1 – Introduction .............................................................................................................. 62
4.2 – Methods .................................................................................................................... 68
4.2.1 – Survey Data ................................................................................................. 68
4.2.1.1 – Pearl River Estuary .............................................................................. 68
4.2.1.2 – Eastern Taiwan Strait ........................................................................... 69
4.2.2 – Environmental Predictor Data ..................................................................... 69
4.2.3 – Modelling .................................................................................................... 70
4.2.3.1 – Data Pre-Processing ............................................................................. 70
4.2.3.2 – Model Fitting and Cross-Validation .................................................... 73
4.2.3.3 – Model Evaluation ................................................................................. 76
4.2.3.4 – Variable Importance ............................................................................ 78
4.2.3.5 – Ensembles and Niche predictions ........................................................ 78
4.3 – Results ...................................................................................................................... 81
4.3.1 – Dataset Summaries ...................................................................................... 81
4.3.2 – Model Selection ........................................................................................... 82
4.3.3 – Variable Importance .................................................................................... 83
4.3.4 – Niche Comparisons ..................................................................................... 83
4.3.5 – Geographical Predictions ............................................................................. 85
4.4 – Discussion ................................................................................................................ 86
4.5 – References ................................................................................................................ 98
4.6 – Tables and Figures ................................................................................................. 110
x
Chapter 5 – General Conclusions ................................................................................ 117
5.1 – Synthesis: Habitat Characteristics, Density Patterns and Environmental Niches of
Humpback Dolphins in the Pearl River Estuary and Eastern Taiwan Strait ................... 117
5.2 – Conservation Implications and Recommendations ................................................ 120
5.2.1 – Eastern Taiwan Strait................................................................................. 120
5.2.2 – Pearl River Estuary .................................................................................... 121
5.3 – Future directions .................................................................................................... 125
5.4 - References............................................................................................................... 128
Appendix A – Permissions for Inclusion of Published Material ............................... 133
Appendix B – Supplementary Analyses for Chapter 3 .............................................. 142
Predictor Correlation ............................................................................................. 142
Power Analyses ..................................................................................................... 142
References ............................................................................................................. 144
Tables and Figures ................................................................................................. 145
Appendix C – Supplementary Materials for Chapter 4 ............................................ 147
R Code ................................................................................................................... 147
Tables and Figures ................................................................................................. 151
xi
List of Tables
Table 2.1 – Summary statistics of mean environmental variables measured during
sightings of Taiwanese humpback dolphins during surveys conducted in the
eastern Taiwan Strait from 2007-2012............................................................28
Table 3.1 – Taiwanese humpback dolphin sighting, dolphin, and mother-calf pair density
(per 1000 km of survey effort) in waters adjacent to major reclamation
projects compared to more natural waters (i.e. where no major reclamation
exists). Reclamation blocks are given in north-south order. Changbin
Industrial Park and Mailiao Industrial Park act as sources of freshwater, while
Taichung Harbour and Budai Harbour have no freshwater influence............58
Table 4.1 – Results of randomization tests for niche similarity. Null distributions were
generated by calculating niche overlap statistics (Schoener’s D and Warren’s
I) between mean predicted probability of presence made in study area A using
the best models fit using data from study area B and predictions made in
study area A by models of the same algorithm type and predictors but fit
using a randomly selected dataset from study area A...................................114
Table B.1 – Partial effect sizes of predictor variables and statistical power of each
regression conducted in Chapter 3. Stated effect sizes are partial eta squared,
which measures the effect of each variable when other coefficients are held
constant. An optimistic calculation of overall power of each model is listed in
the bottom row, calculated using the largest effect size for any variable
above.............................................................................................................146
Table C.1 – Summary statistics for depth data used as predictor variables in Chapter 4.
Full dataset rows contain summary statistics for all grid cells with survey
effort in each study area and season. Subset rows contain summary statistics
for each of the three thinned subsets for each study area, obtained by the
thinning procedure described in Chapter 4...................................................154
Table C.2 – Summary statistics for distance to shore data used as predictor variables in
Chapter 4. Full dataset rows contain summary statistics for all grid cells with
xii
survey effort in each study area and season. Subset rows contain summary
statistics for each of the three thinned subsets for each study area, obtained by
the thinning procedure described in Chapter 4.............................................155
Table C.3 – Summary statistics for productivity data used as predictor variables in
Chapter 4. Full dataset rows contain summary statistics for all grid cells with
survey effort in each study area and season. Subset rows contain summary
statistics for each of the three thinned subsets for each study area, obtained by
the thinning procedure described in Chapter 4.............................................156
Table C.4 – Summary statistics for turbidity data used as predictor variables in Chapter 4.
Full dataset rows contain summary statistics for all grid cells with survey
effort in each study area and season. Subset rows contain summary statistics
for each of the three thinned subsets for each study area, obtained by the
thinning procedure described in Chapter 4...................................................157
Table C.5 – Summary statistics for sea surface temperature data used as predictor
variables in Chapter 4. Full dataset rows contain summary statistics for all
grid cells with survey effort in each study area and season. Subset rows
contain summary statistics for each of the three thinned subsets for each study
area, obtained by the thinning procedure described in Chapter 4.................158
xiii
List of Figures
Figure 1.1 – Sousa chinensis in Chinese waters. Bold names denote putative populations,
points indicate locations of additional sightings of humpback dolphins in
areas that population status has not been confirmed. Blue areas approximate
study areas for putative populations of S. c. chinensis in Chinese waters.
Red area approximates study area for S. c. taiwanensis...............................15
Figure 2.1 – Study area for the Taiwanese humpback dolphin (Sousa chinensis). The
distribution of preset waypoints (black, even numbered, and open, odd
numbered squares representing offshore and inshore waypoints,
respectively), and the zig-zag tracklines between waypoints are shown.
Sandbars exposed at the lowest high tide of the year are approximated by
dark grey polygons, and thin broken lines mark the 30 m isobaths.............29
Figure 2.2 – Number of sightings of Taiwanese humpback dolphins made in the eastern
Taiwan Strait from 2007-2012 at different distances from shore (a), and
water depths (b), where shore is defined as any land mass exposed at the
lowest high tide of the year (following Ross et al., 2010)............................30
Figure 2.3 – Frequency distribution of the best estimates of the number of Taiwanese
humpback dolphins observed in the eastern Taiwan Strait during sightings
made from 2007-2012...................................................................................31
Figure 3.1 – Study area and distribution of the Taiwanese humpback dolphin. Linear
survey effort in 1 km2 grid cells completed in BSS ≤ 2 is shown, along with
the ten major rivers of western Taiwan. Major reclamation blocks are
outlined, from north to south: Taichung Harbour, Changbin Industrial Park
to the Wang-gong agricultural area, Mailiao Industrial Park (Formosa
Petrochemical Corporation), Budai Harbour. Changbin Industrial Park and
Mailiao Industrial Park both act as sources of freshwater due to their
construction incorporating pre-existing rivers, while Taichung and Budai
Harbours do not contribute freshwater to the study area..............................56
xiv
Figure 3.2 – Variation in Taiwanese humpback dolphin density by 10 km latitudinal
survey block, in (a) sightings per unit of effort (SPUE), (b) dolphins per unit
of effort (DPUE) and (c) mother-calf pairs per unit of effort (MCPUE).
Spatial variation in density is shown in maps, with darker shades indicating
higher density. Boxplots show the variation in density for each survey block
for the study period.......................................................................................57
Figure 4.1 – Location of putative populations (bold text, study areas approximated in blue
for S. c. chinensis and red for S. c. taiwanensis) and additional sightings
(regular text) of Sousa chinensis in Chinese waters, and geographical
context of the two study areas (a). Inset maps show survey effort in square
kilometres covered in each 1km2 grid between 2008-2015 in the Pearl River
Estuary during the wet (b) and dry (c) seasons, and the wet season in the
eastern Taiwan Strait (d). The area outlined in black in the PRE indicates
the political border of Hong Kong, within which the majority of research
efforts in the PRE are focused....................................................................110
Figure 4.2 – Box and whisker plots of environmental variables in each study area, divided
by whether dolphin presences or absences were recorded in the same grid.
Summary statistics for each dataset are given in Appendix C....................111
Figure 4.3 – Mean (± SE) area under the ROC curve (AUC; partial AUC for ETS data;
panel a) and true skill statistic (TSS; panel b) of models included in the
bounding box ensemble for each study area, by algorithm and across
thinned subsets. Models were selected if they met the criteria of AUC > 0.7
and TSS > 0.5. Numbers above each bar indicate the total number of models
of each algorithm selected for each study area (BRT = boosted regression
tree, GAM = generalized additive models, GLM = generalized linear
models, RF = random forest)......................................................................112
Figure 4.4 – Mean (± SE) of predictor variable importance for turbidity, depth, survey
effort, net primary productivity, distance to shore and sea surface
temperature. Predictors were ranked from most (6) to least (1) important for
predicting humpback dolphin presence from models included in ensembles
xv
for each dataset. Predictors that did not appear in a model were given a rank
of 0. Error bars show the standard error of the variable’s mean rank across
models for that dataset, and the numbers above each bar indicate the number
of models for each dataset that included the predictor variable.................113
Figure 4.5 – Niche volumes of the PRE in the wet season (blue), PRE in the dry season
(light blue) and ETS in the wet season (red) in environmental space. Niche
ellipsoids in the turbidity, depth and productivity dimensions are shown in
(a), and in the sea surface temperature, distance to shore and productivity
dimensions in (b) (N.B. productivity was repeated in the second figure to
allow visualization of only five environmental predictor variables in three
dimensions).................................................................................................115
Figure 4.6 – Bounding box ensemble predictions in geographic space for PRE in the dry
season (a,b), PRE in the wet season (c,d) and ETS in the wet season (e,f).
Left-hand side shows the proportion of models out of the total number of
best-predicting models (nPREdry=621, nPREwet=115, nETS=47) that predicted a
presence in each grid cell in PRE in the dry season (a), PRE in the wet
season (c) and the ETS in the wet season (e). Right-hand side shows these
predictions converted to bounding box presence or absence based on the
threshold number of predicted presences that maximized the sum of
sensitivity and specificity for each study area (PRE dry season = 303, PRE
wet season= 85, ETS = 26).........................................................................116
Figure B.1 – Correlation matrix for environmental predictors used in GLMs in Chapter 3,
plotted using the corrplot package in R (Wei and Simko, 2017). Pearson’s
correlation coefficients are shown, and statistically significant correlations
(p < 0.05) are indicated with an asterisk (*)...............................................145
Figure C.1 – Three-dimensional scatterplots of full PRE dry season dataset (a,e, n=2500),
and the three thinned subsets (n=140, prevalence = 0.5) used for species
distribution modelling in Chapter 4. The top four panels (a-d) illustrate
results of the thinning procedure on the depth, turbidity and productivity
(NPP) axes, and the bottom four panels (e-h) are on the distance to shore,
xvi
sea surface temperature (SST) and productivity (NPP) axes (N.B.
productivity was repeated in both sets of plots to facilitate three-dimensional
plotting). Thinned subsets were selected by the procedure described in
Chapter 4, with the original subset (i.e. 140 most-distant points in
environmental space) comprising the first subset (b,f), and the subsequent
two subsets (c,g and d,h) selected by removing the 10 and 20 most-distant
points, respectively, from the original subset and repeating the ranking
procedure....................................................................................................151
Figure C.2 – Three-dimensional scatterplots of full PRE wet season dataset (a,e, n=2814),
and the three thinned subsets (n=180, prevalence = 0.5) used for species
distribution modelling in Chapter 4. The top four panels (a-d) illustrate
results of the thinning procedure on the depth, turbidity and productivity
(NPP) axes, and the bottom four panels (e-h) are on the distance to shore,
sea surface temperature (SST) and productivity (NPP) axes (N.B.
productivity was repeated in both sets of plots to facilitate three-dimensional
plotting). Thinned subsets were selected by the procedure described in
Chapter 4, with the original subset (i.e. 180 most-distant points in
environmental space) comprising the first subset (b,f), and the subsequent
two subsets (c,g and d,h) selected by removing the 10 and 20 most-distant
points, respectively, from the original subset and repeating the ranking
procedure....................................................................................................152
Figure C.3 – Three-dimensional scatterplots of full ETS dataset (a,e, n=673), and the
three thinned subsets (n=80, prevalence = 0.5) used for species distribution
modelling in Chapter 4. The top four panels (a-d) illustrate results of the
thinning procedure on the depth, turbidity and productivity (NPP) axes, and
the bottom four panels (e-h) are on the distance to shore, sea surface
temperature (SST) and productivity (NPP) axes (N.B. productivity was
repeated in both sets of plots to facilitate three-dimensional plotting).
Thinned subsets were selected by the procedure described in Chapter 4, with
the original subset (i.e. 80 most-distant points in environmental space)
xvii
comprising the first subset (b,f), and the subsequent two subsets (c,g and
d,h) selected by removing the 10 and 20 most-distant points, respectively,
from the original subset and repeating the ranking procedure...................153
1
Chapter 1 – General Introduction
1.1 – Species’ Habitats and Niches
The study of species’ relationships with their habitats is a foundation of ecology.
Humans have been describing the environments in which plants and animals are found
since before Darwin had conceived the notion of “species” (Nelson, 1978), and
qualitative descriptions of the environmental characteristics of habitats are often the first
step in the investigation of species-habitat relationships. Efforts towards quantifying
patterns of habitat use led to the development of the niche concept in the early 20th
century, first popularized by Grinnell (1917) with his description of the “scenopoetic”
niche – defined by the abiotic conditions of a focal species’ habitat (sensu Soberón, 2007)
– of the California thrasher. Hutchinson (1957) later extended the definition of a niche to
include both biotic and abiotic factors. He defined a species’ fundamental niche as the “n-
dimensional hypervolume”, at every point within which environmental conditions would
permit the species to persist indefinitely, while its realized niche was understood to be a
subset of this hypervolume, limited by biotic factors such as competition and predation
that preclude the species from utilizing its full fundamental niche (Hutchinson, 1957).
Quantifying a species’ niche in abstract environmental space generally begins with
observations of the habitats occupied in geographical space. Studies of a species’
relationships with its habitat can indicate how it responds to environmental gradients, and
which environmental characteristics define “suitable habitat”. Niche characteristics also
affect local abundance and geographical distributions, as outlined by Brown (1984):
species with broad niches (i.e. generalists, which are found in a variety of environmental
2
conditions) tend to be widespread, and are also more locally abundant than species with
narrow niches (i.e. specialists), which are restricted to very specific habitats; and species
tend to be more abundant at the centre of their range, declining gradually towards the
boundaries (though this pattern has been challenged by some, e.g. Sagarin et al., 2006;
Santini et al., 2018).
Determining the extent of a species’ geographic range can be challenging when the
species is widespread, especially when the range spans multiple political borders –
boundaries that restrict the distribution of funding and movement of researchers to one
particular jurisdiction but have no meaning to the organisms they study. In these cases, it
is necessary to go beyond observed habitat characteristics and density patterns in
geographic space and investigate species’ niches in abstract environmental space. This
requires the use of models to generalize species-habitat relationships. Species distribution
models (SDM; also known as ecological niche models or habitat suitability models) are
used to quantify these relationships based on species occurrences and environmental data
and can then be used to project the species’ niche into geographic space to predict its
distribution. Modelling a species’ niche in environmental space allows the comparison of
niches among species, or populations of the same species, that are found in separate,
often disparate, geographic areas. Maps of species density and distribution are useful in
several conservation applications including establishing possible geographical boundaries
between putative populations to define management units, directing survey effort in
potentially suitable but undersurveyed areas, and identifying important areas for
protection. In this thesis, I have focused on the study of habitat characteristics, density
patterns and niches of two populations of humpback dolphin, genus Sousa.
3
1.2 – Genus Sousa
Genus Sousa is generally accepted to comprise four species of humpback dolphin:
Sousa teuszii, the Atlantic humpback dolphin, is found in the waters of west Africa;
Sousa plumbea, the Indian Ocean humpback dolphin, ranges from the waters of South
Africa to the waters of Myanmar; Sousa chinensis, the Indo-Pacific humpback dolphin, is
found in the Indian and Pacific Oceans; and Sousa sahulensis, found from the Australian
shelf to New Guinea (Jefferson and Rosenbaum, 2014). Mendez et al. (2013) proposed
that allopatry is the most parsimonious explanation for the divergence of S. teuszii and S.
sahulensis, as both species are geographically isolated from the other two humpback
dolphin species. S. plumbea and S. chinensis are sympatric in parts of their range, thus
divergence may be due to reproductive isolation, possibly related to behavioural or
morphological differences (Mendez et al., 2013).
The association of humpback dolphins with shallow, estuarine waters provides
opportunities for distinct populations to occur where areas of suitable habitat are
discontinuous. Mendez et al. (2011) described molecular genetic differences among
putative populations of S. plumbea in South Africa, Mozambique, Tanzania and Oman
that coincided with areas of environmental heterogeneity. Population structure cannot be
quantified without sufficient molecular genetic (which can be difficult to obtain for
marine mammal species as carcasses are often not detected at sea and stranding sites may
be remote) or morphological data (either in the form of high-quality photographic data or
carcasses for physical measurements, the latter of which are subject to similar limitations
as obtaining genetic samples). Thus, the population status of groups of humpback
dolphins remains putative but not confirmed in many parts of their range, and it should be
4
noted that the use of the word “population” throughout this thesis should be understood to
denote putative populations of humpback dolphins rather than populations that have been
confirmed separate by genetic or other analyses.
Such is the case for several populations of S. chinensis found in Chinese waters,
locally known as the Chinese white dolphin (CWD; Fig. 1.1). Jefferson and Hung (2004)
first estimated that there were eight populations of CWD in Chinese waters, but further
surveys by Chen et al. (2009) reduced that number to five: dolphins occurring in the
Jiulong River estuary (JRE), the Pearl River estuary (PRE), Leizhou Bay, the Beibu Gulf,
and the eastern Taiwan Strait (ETS; recently designated as a separate subspecies,
S.c.taiwanensis; Wang et al., 2015). Additional records of CWDs southwest of Hainan
(Li et al., 2016) and near Ningde (Chen et al., 2012) have also extended the species’
known range in Chinese waters, and sightings have occasionally been made in the waters
of Shantou, located between the PRE and JRE. The known populations of CWD are
separated by distances of at least a hundred kilometres, which may prohibit exchange of
individuals between populations, though humpback dolphins in South Africa
(Karczmarski, 2000) and the PRE (Hung and Jefferson, 2004) may have home ranges of
up to a few hundred square kilometres in some cases. The most comprehensive
molecular analysis of Chinese humpback dolphin populations to date did not find
evidence of population genetic structure among the JRE, PRE, and Beibu Gulf
populations, though noted that the small sample size, despite re-analysis of other studies’
data, and selection of molecular markers reduced their ability to accurately measure
patterns of gene flow (Lin et al., 2012). The authors suggested that the apparent lack of
population genetic structure could be due to gene flow among these populations, with
5
unsurveyed areas between known humpback dolphin habitats also possibly constituting
suitable habitats or serving as travelling corridors between large river estuaries (Lin et al.,
2012), though their results may also be due to insufficient time since separation of these
populations to detect any genetic structure. The uncertain population status of humpback
dolphins in Chinese waters and potentially suitable but unsurveyed areas between well-
established humpback dolphin habitats invites the question of what constitutes “suitable”
habitats for humpback dolphins, and whether “suitability” differs for each of the putative
populations found in Chinese waters. To make these comparisons, I focused on two
populations in particular: the Taiwanese subspecies, S. c. taiwanensis, also known as the
Taiwanese white dolphin (TWD; =Taiwanese humpback dolphin, THD), and the PRE
population as a representative of the nominate subspecies, S. c. chinensis, which currently
encompasses all populations of S. chinensis excluding those found in the ETS.
1.2.1 – Humpback Dolphins of the Pearl River Estuary
Numbering approximately 2,500 individuals (Chen et al., 2010), the Pearl River
estuary population is the largest known, and arguably best-studied population of
humpback dolphin in the species’ range. Although CWDs are found throughout the
estuary, political borders, availability of research funding among mainland China and the
Special Administrative Regions of Hong Kong and Macau, and the sheer size of the
estuary make surveying the entire habitat difficult. The most intensive research efforts
have taken place in the waters of Hong Kong, where a long-term monitoring program
funded by the Hong Kong government has been ongoing since 1995. Frequent line-
6
transect surveys have produced spatially and temporally fine-scale estimates of
abundance and density patterns of CWDs in this part of their habitat.
Hung (2008) conducted the first quantitative study of humpback dolphin habitat
use in the PRE, primarily using data collected within Hong Kong’s waters. Hung (2008)
found that prey availability was the most important determinant of humpback dolphin
distribution, with monthly fluctuations in dolphin densities associated with hydrological
parameters including temperature, salinity and water quality, and a clear preference for
deeper waters and steeper benthic slopes. Recent studies have also highlighted the
importance of Hong Kong’s waters for foraging (Or, 2017; Wong, 2017). Annual
government reports have tracked changes in humpback dolphin densities in response to
anthropogenic activities in Hong Kong’s waters in recent years, including land
reclamation, marine construction projects and vessel traffic (e.g. Hung, 2012, 2013, 2014,
2015, 2016). A distinct shift in habitat use from the waters around the heavily-impacted
region north of Lantau Island to western and southwestern Hong Kong’s waters has been
reported based on long-term monitoring work (Hung 2016, 2017), and other studies have
noted the negative effects of these activities on foraging activity and habitat use (Or,
2017; Wong, 2017).
There has been comparatively less research effort in the rest of the PRE in the last
two decades, with only a few surveys in the westernmost waters past Macau. Chen et al.
(2010) conducted a series of line transect surveys in the PRE outside of Hong Kong’s
waters from 2005-2006 and 2007-2008 and produced a preliminary estimate of the
abundance and distribution of the PRE population. Further surveys conducted in
subsequent years included areas as far west as the Moyangjiang River estuary and around
7
Hailing Island (SCSFRI, 2013), and compared encounter rates and abundance estimates
for western waters across the various survey periods (SCSFRI and HKCRP, 2011;
SCSFRI, 2013). In all of these studies, dolphins were found throughout the PRE in both
the wet and dry seasons, but seasonal variations in distribution were evident (Chen et al.
2010; SCSFRI 2013). Whereas humpback dolphins in Hong Kong’s waters have been
observed using more southern, offshore waters in higher densities in the wet season,
presumably shifting in response to greater freshwater influx with increased rainfall
(Hung, 2008; Parsons, 1998), in the western PRE, dolphins were more often sighted in
inshore waters during the wet season, moving to offshore areas during the dry season
(Chen et al., 2010; SCSFRI 2013; SCSFRI and HKCRP, 2011). These shifts are likely in
response to prey distribution, which was supported by fisheries sampling in the western
PRE that indicated offshore movement of humpback dolphin prey species during the dry
season (SCSFRI, 2013). Recent work by Or (2017) found no seasonal differences in
dolphin foraging activity in Hong Kong’s waters, but foraging decreased during the wet
season in Lingding Bay in areas outside of Hong Kong’s waters, suggesting that areas
within Hong Kong’s border are important year-round foraging grounds for dolphins.
Foraging activity also decreased as dolphins moved further from shore, and offshore
areas were unlikely foraging grounds according to that study, suggesting that the central
waters of the PRE may act as a “transit area” for dolphins moving between important
foraging areas closer to shore on either side of the estuary (Or, 2017), but a
comprehensive analysis of habitat use and density patterns across the entire estuary
including the western PRE has not been conducted.
8
1.2.2 – Humpback Dolphins of the Eastern Taiwan Strait
The Taiwanese humpback dolphin is a subspecies of humpback dolphin endemic
to the inshore waters of central western Taiwan in the eastern Taiwan Strait (Wang et al.,
2015). This subspecies is listed as “Critically Endangered” on the IUCN Red List (Wang
and Araújo-Wang, 2018) due to its small population size (estimated at fewer than 80
individuals; Wang et al., 2012) and the numerous anthropogenic threats (Ross et al.,
2010; Slooten et al., 2013; Wang et al., 2004a, 2004b, 2007a) posed by the proximity of
their habitat to the particularly industrialized areas of Taiwan. First scientifically
discovered in 2002 (Wang et al., 2004a), regular surveys have been conducted for this
subspecies since 2007, though mostly restricted to the summer months due to poor
weather for surveys during the rest of the year (Wang and Yang, 2011).
Studies on the habitat use of the TWD have only been conducted in recent years.
The first surveys for humpback dolphins in this area found their main distribution to be
~515 km2 of waters in central western Taiwan (Wang et al., 2007b). Early publications
and workshop reports on the conservation of TWDs produced general descriptions of
their habitat (Wang et al., 2004a, 2004b, 2007a, 2007b) and, based on these descriptions
and knowledge of other humpback dolphin habitats, Ross et al. (2010) recommended that
waters within 3 km of shore or less than 30 m in depth (whichever criterion includes a
larger area) be considered “priority habitat” for the conservation of this population. Wang
and Yang (2011) confirmed that ETS humpback dolphins are year-round residents of the
eastern Taiwan Strait, and remain in the waters of central western Taiwan during the
winter. In at least one of the many small estuaries along the west coast of Taiwan,
frequency of TWD vocalizations indicated that occurrence and habitat utilization patterns
9
shift during different tidal phases (Lin et al., 2013) and with seasonal variation in rainfall
patterns (Lin et al., 2015), suggesting environmental influences on the fine-scale habitat
use patterns or acoustic behaviour of TWDs, though no quantification of large-scale
patterns had been conducted.
1.3 – Thesis Structure
The purpose of this thesis is to further investigate the relationships between
humpback dolphins and their habitats. Previous research on populations of Sousa spp.
have characterized the habitat of these animals as shallow, generally brackish waters. My
objective with this thesis is to expand on this description with population-specific
quantification of habitat and density patterns of the TWD to bring knowledge up to a
comparable level with what is known about other populations of humpback dolphins, and
perform a quantitative comparison of the environmental niches of TWDs and CWDs in
the PRE. Chapter 2 compares the previously unquantified habitat and group size patterns
of ETS humpback dolphins with published descriptions of the habitat of humpback
dolphin populations in other parts of the world. Chapter 3 focuses on describing the
density patterns of ETS humpback dolphins, and investigating the relationship between
these patterns with environmental factors and anthropogenic activities in the study area.
Chapter 4 expands upon the previous two chapters by taking species-environment
relationships beyond in situ data, incorporating remotely sensed environmental data into
species distribution models and allowing a quantitative comparison between the
environmental niches of humpback dolphins in the PRE and ETS. Chapter 5 is a
10
synthesis of the work done in this thesis, and expands on the conservation applications of
my findings and future directions for research on habitat use, distribution patterns and
environmental niches of humpback dolphins.
11
1.4 – References Brown, J.H. 1984. On the relationship between abundance and distribution of species.
Am. Nat. 124, 255–279.
Chen, B., Zheng, D., Yang, G., Xu, X., Zhou, K. 2009. Distribution and conservation of
the Indo-Pacific humpback dolphin in China. Integr. Zool. 4, 240–247.
DOI:10.1111/j.1749-4877.2009.00160.x
Chen, T., Hung, S.K.-Y., Qiu, Y., Jia, X., Jefferson, T.A. 2010. Distribution, abundance,
and individual movements of Indo-Pacific humpback dolphins (Sousa chinensis) in
the Pearl River Estuary, China. Mammalia 74, 117–125.
DOI:10.1515/MAMM.2010.024
Chen, B., Zheng, D., Wang, L., Xu, X., Yang, G. 2012. The northernmost distribution of
Indo-Pacific humpback dolphin (Sousa chinensis) in the world: evidence from
preliminary survey in Ningde, China. Pakistan J. Zool.. 44:1209–1214.
Grinnell, J. 1917. The niche-relationships of the California thrasher. Auk 34, 427–433.
Hung, S.K.-Y. 2008. Habitat use of Indo-pacific humpback dolphins (Sousa chinensis) in
Hong Kong. Ph.D. Thesis, University of Hong Kong. p. 1-253.
DOI:10.5353/th_b4088776
Hung, S.K.-Y. 2012. Monitoring of marine mammals in Hong Kong waters (2011-12).
Final report submitted to the Agriculture, Fisheries and Conservation Department
of the Hong Kong SAR government.
Hung, S.K.-Y. 2013. Monitoring of marine mammals in Hong Kong waters (2012-13).
Final report submitted to the Agriculture, Fisheries and Conservation Department
of the Hong Kong SAR government.
Hung, S.K.-Y. 2014. Monitoring of marine mammals in Hong Kong waters (2013-14).
Final report submitted to the Agriculture, Fisheries and Conservation Department
of the Hong Kong SAR government.
Hung, S.K.-Y. 2015. Monitoring of marine mammals in Hong Kong waters (2014-15).
Final report submitted to the Agriculture, Fisheries and Conservation Department
of the Hong Kong SAR government.
Hung, S.K.-Y. 2016. Monitoring of marine mammals in Hong Kong waters (2015-16).
Final report submitted to the Agriculture, Fisheries and Conservation Department
of the Hong Kong SAR government.
Hung, S.K.-Y. 2017. Monitoring of marine mammals in Hong Kong waters (2016-17).
Final report submitted to the Agriculture, Fisheries and Conservation Department
of the Hong Kong SAR government.
12
Hung, S.K.-Y., Jefferson, T.A. 2004. Ranging patterns of Indo-Pacific humpback
dolphins (Sousa chinensis) in the Pearl River Estuary, People’s Republic of China.
Aquat. Mamm. 30, 159–174. DOI:10.1578/AM.30.1.2004.159
Hutchinson, G.E. 1957. Concluding remarks. Cold Spring Harb. Symp. Quant. Biol. 22,
415–427. DOI:10.1101/SQB.1957.022.01.039
Jefferson, T.A., Hung, S.K.-Y. 2004. A review of the status of the Indo-Pacific humpback
dolphin (Sousa chinensis) in Chinese waters. Aquat. Mamm. 30, 149–158.
DOI:10.1578/AM.30.1.2004.149
Jefferson, T.A., Rosenbaum, H.C. 2014. Taxonomic revision of the humpback dolphins
(Sousa spp.), and description of a new species from Australia. Mar. Mammal Sci.
30, 1494–1541. DOI:10.1111/mms.12152
Karczmarski, L. 2000. Conservation and management of humpback dolphins: the South
African perspective. Oryx 34, 207–216.
Li, S., Lin, M., Xu, X., Xing, L., Zhang, P., Gozlan, R.E., Huang, S.-L., Wang, D. 2016.
First record of the Indo-Pacific humpback dolphins (Sousa chinensis) southwest of
Hainan Island, China. Mar. Biodivers. Rec. 9, 1–6. DOI:10.1186/s41200-016-0005-
x
Lin, W., Chang, L., Frère, C.H., Zhou, R., Chen, J., Chen, X., Wu, Y. 2012.
Differentiated or not? An assessment of current knowledge of genetic structure of
Sousa chinensis in China. J. Exp. Mar. Bio. Ecol. 416, 17–20.
DOI:10.1016/j.jembe.2012.02.002
Lin, T.H., Akamatsu, T., Chou, L.S. 2013. Tidal influences on the habitat use of Indo-
Pacific humpback dolphins in an estuary. Mar. Biol. 160, 1353–1363.
DOI:10.1007/s00227-013-2187-7
Lin, T.H., Akamatsu, T., Chou, L.S. 2015. Seasonal distribution of Indo-Pacific
humpback dolphins at an estuarine habitat: Influences of upstream rainfall.
Estuaries and Coasts. 38, 1376–1384. DOI:10.1007/s12237-014-9886-2
Mendez, M., Subramaniam, A., Collins, T., Minton, G., Baldwin, R., Berggren, P.,
Särnblad, A., Amir, O.A., Peddemors, V.M., Karczmarski, L., Guissamulo, A.,
Rosenbaum, H.C. 2011. Molecular ecology meets remote sensing: environmental
drivers to population structure of humpback dolphins in the Western Indian Ocean.
Heredity. 107, 349–61. DOI:10.1038/hdy.2011.21
Mendez, M., Jefferson, T.A., Kolokotronis, S.-O., Krützen, M., Parra, G.J., Collins, T.,
Minton, G., Baldwin, R., Berggren, P., Särnblad, A., Amir, O.A., Peddemors, V.M.,
Karczmarski, L., Guissamulo, A., Smith, B., Sutaria, D., Amato, G., Rosenbaum,
H.C. 2013. Integrating multiple lines of evidence to better understand the
evolutionary divergence of humpback dolphins along their entire distribution range:
a new dolphin species in Australian waters? Mol. Ecol. DOI:10.1111/mec.12535
13
Nelson, G. 1978. From Candolle to Croizat: Comments on the history of biogeography. J.
Hist. Biol. 11, 269–305.
Or, K.M. 2017. Socio-spatial ecology of Indo-Pacific humpback dolphins (Sousa
chinensis) in Hong Kong and the Pearl River Estuary. Ph.D. Thesis, University of
Hong Kong. p. 1-226.
Parsons, E.C.M. 1998. The behaviour of Hong Kong’s resident cetaceans. Aquat. Mamm.
24, 91–110.
Ross, P.S., Dungan, S.Z., Hung, S.K.-Y., Jefferson, T.A., Macfarquhar, C., Perrin, W.F.,
Riehl, K.N., Slooten, E., Tsai, J., Wang, J.Y., White, B.N., Würsig, B., Yang, S.-C.,
Reeves, R.R. 2010. Averting the baiji syndrome: conserving habitat for critically
endangered dolphins in Eastern Taiwan Strait. Aquat. Conserv. Mar. Freshw.
Ecosyst. 20, 685–694. DOI:10.1002/aqc.1141
Sagarin, R.D., Gaines, S.D., Gaylord, B. 2006. Moving beyond assumptions to
understand abundance distributions across the range of species. Trends Ecol. Evol.
21, 524-530. DOI: 10.1016/j.tree.2006.06.008
Santini, L., Pironon, S., Maiorano, L., Thuiller, W. 2018. Addressing common pitfalls
does not provide more support to geographical and ecological abundant-centre
hypotheses. Ecography. 42, 1-10. DOI: 10.1111/ecog.04027
SCSFRI (South China Sea Fisheries Research Institute). 2013. Study on Indo-Pacific
humpback dolphin population dynamic and their prey availability in western Pearl
River Estuary. Report submitted to WWF Hong Kong. p. 1-75.
SCSFRI (South China Sea Fisheries Research Institute) and HKCRP (Hong Kong
Cetacean Research Project). 2011. Study on the population biology of Chinese
white dolphins (Sousa chinensis) in the western Pearl River estuary, with
exploratory surveys in the Moyangjiang River estuary. Report submitted to WWF
Hong Kong. p. 1-39.
Slooten, E., Wang, J.Y., Dungan, S.Z., Forney, K.A., Hung, S.K.-Y., Jefferson, T.A.,
Riehl, K.N., Rojas-Bracho, L., Ross, P.S., Wee, A., Winkler, R., Yang, S.-C., Chen,
C.A. 2013. Impacts of fisheries on the Critically Endangered humpback dolphin
Sousa chinensis population in the eastern Taiwan Strait. Endanger. Species Res. 22,
99–114. DOI:10.3354/esr00518
Wang, J.Y., Hung, S.K.-Y., Yang, S.-C. 2004a. Records of Indo-Pacific humpback
dolphins, Sousa chinensis (Osbeck, 1765), from the waters of western Taiwan.
Aquat. Mamm. 30, 189–196. DOI:10.1578/AM.30.1.2004.189
Wang, J.Y., Yang, S.-C., Reeves, R.R. 2004b. Report of the first workshop on
conservation and research needs of Indo-Pacific humpback dolphins, Sousa
chinensis, in the waters of Taiwan. Checheng, Pingtung County, Taiwan.
14
Wang, J.Y., Yang, S.-C., Hung, S.K.-Y., Jefferson, T.A. 2007a. Distribution, abundance
and conservation status of the eastern Taiwan Strait population of Indo-Pacific
humpback dolphins, Sousa chinensis. Mammalia 71, 157–165.
DOI:10.1515/MAMM.2007.032
Wang, J.Y., Yang, S.C., Reeves, R.R. 2007b. Report of the Second International
Workshop on the Conservation and Research Needs of the Eastern Taiwan Strait
Population of Indo-Pacific Humpback Dolphins, Sousa chinensis. Checheng,
Pingtung County, Taiwan.
Wang, J.Y., Yang, S.C. 2011. Evidence for year-round occurrence of the eastern Taiwan
Strait Indo-Pacific humpback dolphins (Sousa chinensis) in the waters of western
Taiwan. Mar. Mammal Sci. 27, 652–658. DOI:10.1111/j.1748-7692.2010.00422.x
Wang, J.Y., Yang, S.-C., Fruet, P.F., Daura-Jorge, F.G., Secchi, E.R. 2012. Mark-
recapture analysis of the critically endangered eastern Taiwan Strait population of
Indo-Pacific humpback dolphins (Sousa chinensis): Implications for conservation.
Bull. Mar. Sci. 88, 885–902.
Wang, J.Y., Yang, S.-C., Hung, S.K.-Y. 2015. Diagnosability and description of a new
subspecies of Indo-Pacific humpback dolphin, Sousa chinensis (Osbeck, 1765),
from the Taiwan strait. Zool. Stud. 54, 1–15. DOI:10.1186/s40555-015-0115-x
Wang, J.Y., Araújo-Wang, C. 2018. Sousa chinensis ssp. taiwanensis (amended version
of 2017 assessment). The IUCN Red List of Threatened Species 2018:
e.T133710A122515524. DOI:10.2305/IUCN.UK.2017-
3.RLTS.T133710A122515524.en
15
1.5 – Figures
Figure 1.1 – Sousa chinensis in Chinese waters. Bold names denote putative populations,
points indicate locations of additional sightings of humpback dolphins in areas that
population status has not been confirmed. Blue areas approximate study areas for putative
populations of S. c. chinensis in Chinese waters. Red area approximates study area for S.
c. taiwanensis.
16
Chapter 2 – Habitat Characteristics of the Critically
Endangered Taiwanese Humpback Dolphins (Sousa chinensis)
of the Eastern Taiwan Strait
Lauren E. Dares, Jordan M. Hoffman, Shih-Chu Yang and John Y. Wang
A version of this chapter is published as a Short Note in Aquatic Mammals. Permission to
include this article as granted by the Editor of Aquatic Mammals, along with the
published ADA PDF at the journal editor’s request, are available in Appendix A.
Dares, L.E., Hoffman, J.M., Yang, S.-C., Wang, J.Y. 2014. Habitat characteristics of the
critically endangered Taiwanese humpback dolphin (Sousa chinensis) of the eastern
Taiwan Strait. Aquatic Mammals. 40(4): 368-374.
Contributions:
LED designed and conducted data analyses. JMH and JYW conceived of the study. LED,
JMH and JYW contributed to writing the manuscript, and all authors conducted surveys
to collect data for this study.
17
2.1 – Short Note: Habitat Characteristics of the Critically
Endangered Taiwanese Humpback Dolphins (Sousa chinensis)
of the Eastern Taiwan Strait
The eastern Taiwan Strait population of Indo-Pacific humpback dolphins is also
known as the Taiwanese humpback dolphin, Sousa chinensis (Osbeck 1765) and
comprises only about 74 individuals (95% CI = 68-80 individuals; Wang et al., 2012). An
assessment of this population against the criteria of the IUCN Red List of Threatened
Species resulted in its classification as “Critically Endangered” (Reeves et al., 2008).
Concerns are driven by the presence of a large number of anthropogenic threats from
high human densities along the coastal regions adjacent to the waters inhabited by this
population. These threats include: habitat degradation due to land reclamation, reduction
of freshwater flow to estuaries, harmful fishing interactions, air and water pollution, and
underwater noise (Wang et al., 2004b, 2007a; Ross et al., 2010; Dungan et al., 2011;
Slooten et al., 2013). In general, the Taiwanese population appears to be found year-
round in the shallow coastal waters off central western Taiwan (Wang & Yang, 2011),
seems to be restricted to waters that are within 3 km of shore and usually in or near areas
influenced by rivers or other sources of freshwater (Wang et al., 2004a, 2007a,b). Ross et
al. (2010) defined “priority habitat” for this population as including waters <30 m deep
(measured relative to the lowest low tide of the year) or within 3 km of “shore”, which
was defined as any land that remained dry at the lowest high tide of the year. However,
descriptive characteristics of the waters inhabited by this population have not been
examined in detail. Here we outline distance from shore, water depth, sea surface
18
temperature (SST) and salinity as descriptors of suitable habitat for the Taiwanese
humpback dolphins based on sightings of these animals. We also present the first report
on group sizes for this population, investigate correlations between group sizes and
environmental variables, and compare the group size distribution of Taiwanese humpback
dolphins with other conspecific populations.
A total of 121 surveys were conducted in the eastern Taiwan Strait from 2007-
2012. The majority of the surveys were conducted during the summer months (June-
August) with the exception of three surveys in early spring (April 2010) and nine in early
autumn (September 2008). Surveys were undertaken using a 4.5 m inflatable vessel
travelling between 18-25 km/h with two observers searching 90 degrees on either side of
the bow of the vessel with their naked eyes. Two sets of wide, overlapping zig-zag
transect lines were followed, each alternating between inshore and offshore waypoints
(Figure 2.1). Distance from shore was measured following the definition of “shore” as
outlined in Ross et al. (2010). The main study area covered a linear distance of
approximately 110 km from just north of Tongshiao, Miaoli County (N24o30’) to south
of Taixi, Yunlin County (N23o30’) (Figure 2.1).
When dolphins were sighted, the following set of standard information was
recorded: sea surface temperature (SST) and water depth (Lowrance HDS-5 GPS/sonar
unit), percent salinity (Rixen SM-10), date, time, geographic position (Garmin
GPSMAP76 or the Lowrance HDS-5 GPS), and the observers’ best estimate of the
number of individuals and number of mother-calf pairs present. Locations of dolphin
sightings were plotted in Google Earth 6.0.3 (Google, Inc.) and distances to shore of the
sightings were measured to the nearest metre using the ruler function of this software.
19
Sightings were made at a mean distance from shore of 824±471 m (SD), with no
dolphins sighted further than 2,721 m from shore (Table 2.1; Figure 2.2a). Eight sightings
were recorded within 1 m of shore due to the definition of shore as exposed land at the
lowest high tide of the year – these sightings were made during tidal phases when
sandbars were submerged but are exposed at lower high tides. Sightings were made
predominantly within 1 km of the shoreline, which is consistent with the distributions of
other humpback dolphin populations: sightings of South African populations varied
widely in distance from shore, but were usually within 400 m in Algoa Bay (Karczmarski
et al., 2000) to within 2 km in Richards Bay (Atkins et al., 2004); in the Pearl River
estuary, dolphins can occur 10-15 km from the nearest shore, but this is an enclosed
inshore area with several islands and the water depth rarely exceeds 30 m (see Jefferson,
2000; Hung 2008). In the eastern Taiwan Strait, inshore obstructions such as piers and
extensive oyster mariculture in some regions may create barriers extending from shore, or
natural barriers such as sandbars likely prevent dolphins from accessing some shallower
and intertidal waters even closer to shore.
Although related in most cases, it seems that the distribution of humpback
dolphins is more likely directly related to water depth and accessibility of shallower
waters than proximity to the shoreline itself. Mean depth measured during humpback
dolphin sightings was 6.9±3.4 m with more than half (58%) of sightings being made in
waters less than 7 m deep (Q1 = 4.5 m, Q3 = 8.1 m; Table 2.1; Figure 2.2b). More than
96% of sightings of Taiwanese humpback dolphins occurred in waters less than 20 m
deep, similar to other reports of shallow water preferences by humpback dolphins
(Durham, 1994; Corkeron et al., 1997; Karczmarski et al., 1999; Jefferson, 2000; Atkins
20
et al., 2004). These findings are similar to those of Corkeron et al. (1997), where
humpback dolphins inhabiting the Great Barrier Reef region (Australia) were sighted a
mean distance of 6.7 km from land, but the mean distance to the nearest shallow area (<2
m deep at low tide) was only 2.7 km. Taiwanese humpback dolphins have been recorded
passing through waters deeper than 30m while being followed during photographic
identification monitoring studies, but these infrequent events have only occurred in well-
dredged channels of ports servicing large commercial ships and not offshore (J.Y. Wang,
unpublished data).
Sea surface temperatures measured during sightings varied from 23.2oC in early
April to 31.8oC in late July, and salinities varied from 0.4 - 4.1% (Table 1), indicating
that humpback dolphins are found in a wide range of generally warm, brackish
conditions. Habitat utilization of the Pearl River estuary population of humpback
dolphins in Hong Kong’s waters was linked to prey distributions (Hung, 2008), and the
habitat utilization for Taiwanese humpback dolphins may similarly reflect prey
distribution rather than physiological or other requirements of the dolphins (Lin et al.,
2013). Water depth, sea surface temperature and salinity may affect the distribution of
humpback dolphins directly or indirectly by influencing the distributions of their prey,
which has been observed with other small cetaceans (Heithaus & Dill, 2000; Jaquet &
Gendron, 2002; Bräger et al., 2003). Bathymetry (Hooker et al., 2002; MacLeod & Zuur,
2005) and nutrient availability (Stockin et al., 2008) are also important in determining
prey distribution, but data on these other parameters were not available for analysis in the
present study. Estuaries are some of the most productive habitats in the world (Correll,
1978), and shallower waters (Keller, 1989; Cloern, 1999) and warmer waters (Keller,
21
1989) often have increased productivity, which is frequently used as a proxy for prey
abundance in studies of cetacean habitat characteristics (Scott et al., 2010; Redfern et al.,
2006). Taiwanese humpback dolphins have been sighted in inshore habitats of the eastern
Taiwan Strait year-round (Wang & Yang, 2011) during surveys conducted in months
with recorded minimum temperatures of <16oC (Jan-March 2009; Central Weather
Bureau, 2009), and average sea surface temperatures as low as 14oC have been recorded
in these waters during the winter (Tzeng et al., 2002), thus it is unlikely that this
population is restricted by sea surface temperature alone. Furthermore, fish species
identified as prey for other humpback dolphins populations are frequently found in large
shoals in areas where fresh and salt water mix (Barros et al., 2004; Parra & Jedensjö,
2013), and humpback dolphins of other populations have often been observed feeding at
the freshwater-saltwater interface (Jefferson, 2000; Parsons, 1998, 2004).
A total of 221 humpback dolphin sightings were made during the study period.
The frequency distribution of the number of dolphins in each sighting was right-skewed
(Fig. 2.3) with a mean of 6.2±5.9 (SD) and a median of 4. Singletons were the most
frequently observed (21%) while groups of two or more dolphins comprised 79% of the
sightings (Fig. 2.3). Excluding singletons, group size varied from the most frequently
observed two dolphins up to an estimated maximum of 31 (note: based on photo-
identification of individuals, there were at least 41 individuals in this largest group – JY
Wang, unpublished data), with a mean of 7.6±5.9 (SD) and median of 6. Mother-calf
pairs were present in more than half (60%) of the sightings with two or more individuals.
Investigating the relationships between group size and depth, SST, salinity and distance
to shore (Spearman’s rank correlation coefficient) showed no significant correlations (p >
22
0.05 for all tests). The frequency distribution of group size for the Taiwanese population
appears similar in shape to reports from other populations, including the high frequency
of small group sizes. The frequency of encountering singletons of the Taiwanese
population was similar to observations made in South African waters (20% in Durham,
1994; 15.4% in Karczmarski et al., 1999) but was lower than in Hong Kong’s waters
(~30% in Jefferson, 2000). The mean group size of Taiwanese humpback dolphins was
also consistent with that reported for humpback dolphins (Jefferson & Rosenbaum, 2014)
found in other regions (e.g., Durham, 1994; Jefferson, 2000). However, the largest group
of Taiwanese humpback dolphins observed (best field estimate of 31 individuals, which
was actually at least 41 individuals based on individual identification using photographs)
was larger than reported for most other populations (Durham, 1994; Karczmarski et al.,
1999; Chen et al., 2010; Zhou et al., 2007). Frequent observations of small group size
may be related to highly fluid social relationships within the Taiwanese population
(Dungan, 2011) as individuals do not appear to form stable, long-term groups. Larger
groupings may be a result of occasional gatherings that occur during periods of peak
abundance of prey species (Finley & Gibb, 1982) or for non-feeding reasons such as
socializing or mating (Baird & Dill, 1996). Group sizes did not appear to be affected by
the environmental variables measured in this study, although fewer sightings of
individual dolphins in the eastern Taiwan Strait and South African waters when
compared to Hong Kong may be a result of habitat characteristics. Despite taxonomic
differences between humpback dolphins in Chinese and South African waters (Mendez et
al., 2013; Jefferson & Rosenbaum, 2014), areas inhabited by the Taiwanese, Algoa Bay
and Natal Coast populations are relatively linear (see Karczmarski et al., 1999; Durham,
23
1994) compared to the large Pearl River estuary, which contains a number of small
islands and extends further inland, possibly contributing to the differences in group size.
In conclusion, the distribution of this population in relatively shallow, inshore,
estuarine waters and the group sizes observed were consistent with most other
populations of humpback dolphins. Group sizes were not correlated with the
environmental variables measured in this study, but the spatial distribution of the
Taiwanese population is most likely related to the abundance of prey species, as estuarine
waters are known for high productivity. Humpback dolphin occurrence may also be
driven at least in part by certain habitat features that may be preferred while resting,
mating or socializing, and by other physical variables not investigated directly in this
study, such as bathymetric features (e.g. sea floor slope), productivity, sea floor substrate
and anthropogenic influence. Future studies will investigate behavioural habitat
preferences for this population in addition to using remote sensing data and habitat
models to describe relationships between humpback dolphin abundance, distribution, and
a wider variety of environmental and anthropogenic factors that could not be addressed
presently. Better understanding of the factors correlated with the distribution of the
Taiwanese humpback dolphins will remain an important area for research to further
improve our knowledge about this critically endangered population and to provide
guidance and focus efforts to reduce human impacts.
24
2.2 – Literature Cited
Atkins, A., Pillay, N., Peddemors, V.M. (2004). Spatial distribution of Indo-Pacific
humpback dolphins (Sousa chinensis) at Richards Bay, South Africa:
environmental influences and behavioural patterns. Aquatic Mammals, 30(1), 84-9.
DOI:10.1578/AM.30.1.2004.84
Baird, R.W., Dill, L.M. (1996). Ecological and social determinants of group size in
transient killer whales. Behavioral Ecology, 7(4), 408-416.
Barros, N.B., Jefferson, T.A., Parsons, E.C.M. (2004). Feeding habits of Indo-Pacific
humpback dolphins (Sousa chinensis) stranded in Hong Kong. Aquatic Mammals,
30(1), 179–188. DOI:10.1578/AM.30.1.2004.179
Bräger, S., Harraway, J.A., Manly, B.F.J. (2003). Habitat selection in a coastal dolphin
species (Cephalorhynchus hectori). Marine Biology, 143, 233-244.
DOI:10.1007/s00227-003-1068-x
Central Weather Bureau. (2009). Monthly Climate Data. Retrieved 28 May 2014 from
http://www.cwb.gov.tw/V7e/climate/monthlyData/mD.htm.
Chen, T., Hung, S.K., Qiu, Y., Jia, X., Jefferson, T.A. (2010). Distribution, abundance,
and individual movements of Indo-Pacific humpback dolphins (Sousa chinensis) in
the Pearl River Estuary, China. Mammalia, 74, 117-125.
DOI:10.1515/MAMM.2010.024
Cloern, J.E. (1999). The relative importance of light and nutrient limitation of
phytoplankton growth: a simple index of coastal ecosystem sensitivity to nutrient
enrichment. Aquatic Ecology, 33, 3-16.
Corkeron, P.J., Morisette, N.M., Porter, L., Marsh, H. (1997). Distribution and status of
hump-backed dolphins, Sousa chinensis, in Australian waters. Asian Marine
Biology, 14, 49-59.
Correll, D.L. (1978). Estuarine productivity. BioScience, 28(10), 646-650.
Dungan, S.Z. (2011). Comparing the social structures of Indo-Pacific humpback dolphins
(Sousa chinensis) from the Pearl River Estuary and Eastern Taiwan Strait.
(Master’s thesis, Trent University, Peterborough, Canada).
Dungan, S.Z., Riehl, K.N., Wee, A., Wang, J.Y. (2011). A review of the impacts of
anthropogenic activities on the critically endangered eastern Taiwan Strait Indo-
Pacific humpback dolphins (Sousa chinensis). Journal of Marine Animals and
Their Ecology, 4, 3-9.
Durham, B. (1994) The distribution and abundance of the humpback dolphin (Sousa
chinensis) along the Natal coast, South Africa. (Master’s thesis, University of
Natal, Durban, South Africa).
Finley, K.J., Gibb, E.J. (1982). Summer diet of the narwhal (Monodon monoceros) in
Pond Inlet, northern Baffin Island. Canadian Journal of Zoology, 60(12), 3353-
3363. DOI:10.1139/z82-424
25
Heithaus, M.R., Dill, L.M. (2002) Food availability and tiger shark predation risk
influence bottlenose dolphin habitat use. Ecology, 83(2), 480-491.
Hooker, S.K., Whitehead, H., Gowans, S., Baird, R.W. (2002). Fluctuations in
distribution and patterns of individual range use of northern bottlenose whales.
Marine Ecology Progress Series, 225, 287-297.
Hung, S.K. (2008). Habitat use of Indo-Pacific humpback dolphins (Sousa chinensis) in
Hong Kong. (Doctoral dissertation, University of Hong Kong, Hong Kong).
Jaquet, N., Gendron, D. (2002). Distribution and relative abundance of sperm whales in
relation to key environmental features, squid landings and the distribution of other
cetacean species in the Gulf of California, Mexico. Marine Biology, 141, 591-601.
DOI:10.1578/AM.30.1.2004.179
Jefferson, T.A. (2000). Population biology of the Indo-Pacific hump-backed dolphin in
Hong Kong waters. Wildlife Monographs, 144, 1-65.
Jefferson, T.A., Rosenbaum, H.C. (2014). Taxonomic revision of the humpback dolphins
(Sousa spp.), and description of a new species from Australia. Marine Mammal
Science. DOI:10.1111/mms.12152
Karczmarski, L., Cockcroft, V.G., McLachlan, A. (1999). Group size and seasonal
pattern of occurrence of humpback dolphins Sousa chinensis in Algoa Bay, South
Africa. South African Journal of Marine Science, 21(1), 89-97.
DOI:10.2989/025776199784126024
Karczmarski L., Cockcroft, V.G., McLachlan, A. (2000). Habitat use and preferences of
Indo-Pacific humpback dolphins Sousa chinensis in Algoa Bay, South Africa.
Marine Mammal Science, 16(1), 65-79.
Keller, A.A. (1989). Modeling the effects of temperature, light, and nutrients on primary
productivity: An empirical and a mechanistic approach compared. Limnology and
Oceanography, 34(1), 82-95.
Lin, T.H., Akamatsu, T., Chou, L.S. 2013. Tidal influences on the habitat use of Indo-
Pacific humpback dolphins in an estuary. Marine Biology, 160(6): 1353-1363.
DOI:10.1007/s00227-013-2187-7
MacLeod, C.D., Zuur, A.F. (2005). Habitat utilization by Blainville’s beaked whales off
Great Abaco, northern Bahamas, in relation to seabed topography. Marine Biology,
147, 1-11. DOI:10.1007/s00227-004-1546-9.
Mendez, M., Jefferson, T.J., Kolokotronis, S.O., Krützen, M., Parra, G.J., Collins, T., …
Rosenbaum, H.C. (2013). Integrating multiple lines of evidence to better
understand the evolutionary divergence of humpback dolphins along their entire
distribution range: a new dolphin species in Australian waters?. Molecular Ecology,
22(23), 5936-5948. DOI:10.1111/mec.12535
Parra, G.J., Jedensjö, M. (2013). Stomach contents of Australian snubfin (Orcaella
hensohni) and Indo-Pacific humpback dolphins (Sousa chinensis). Marine Mammal
Science, 30(3), 1184-1198. DOI:10.1111/mms.12088
26
Parsons, E.C.M. (1998). The behaviour of Hong Kong’s resident cetaceans: the Indo-
Pacific humpbacked dolphin and the finless porpoise. Aquatic Mammals, 24, 91-
110.
Parsons, E.C.M. (2004). The behaviour and ecology of the Indo-Pacific humpback
dolphin (Sousa chinensis). Aquatic Mammals, 30(1), 38-55.
DOI:10.1578/AM.30.1.2004.38
Redfern, J.V., Ferguson, M.C., Becker, E.A., Hyrenback, K.D., Good, C., Barlow, J., ...
Werner, F. (2006) Techniques for cetacean-habitat modeling. Marine Ecology
Progress Series, 310, 271-295.
Reeves, R.R., Dalebout, M.L., Jefferson, T.A., Karczmarski, L., Laidre, K., O’Corry-
Crowe, G., … Zhou, K. (2008) Sousa chinensis (eastern Taiwan Strait
subpopulation). In: IUCN 2011 Red List of threatened species. Retrieved 30 Aug
2012 from http://www.iucnredlist.org.
Ross, P.S., Dungan, S.Z., Hung, S.K., Jefferson, T.A., MacFarquhar, C., Perrin, W.F., …
Reeves, R.R. (2010). Averting the baiji syndrome: conserving habitat for critically
endangered dolphins in eastern Taiwan Strait. Aquatic Conservation: Marine and
Freshwater Ecosystems, 20, 685-694. DOI:10.1002/aqc.1141
Scott, B.E., Sharples, J., Ross, O.N., Wang, J., Pierce, G.J., Camphuysen, C.J. (2010).
Sub-surface hotspots in shallow seas: fine-scale limited locations of predator
foraging habitat indicated by tidal mixing and sub-surface chlorophyll. Marine
Ecology Progress Series, 408, 207-226. DOI:10.3354/meps08552
Slooten, E., Wang, J.Y., Dungan, S.Z., Forney, K.A., Hung, S.K., Jefferson, T.A., …
Chen, C.A. 2013. Impacts of fisheries on the Critically Endangered humpback
dolphin Sousa chinensis population in the eastern Taiwan Strait. Endangered
Species Research, 22, 99-114. DOI:10.3354/esr00518
Stockin, K.A., Pierce, G.J., Binedell, V., Wiseman, N., Orams, M.B. (2008). Factors
affecting the occurrence and demographics of Common dolphins (Delphinus sp.) in
the Hauraki Gulf, New Zealand. Aquatic Mammals, 34(2), 200-211.
DOI:10.1578/AM.34.2.2008.200
Tzeng, W.N., Wang, Y.T., Chang, C.W. (2002). Spatial and temporal variations of the
estuarine larval fish community on the west coast of Taiwan. Marine and
Freshwater Research, 53, 419-430.
Wang, J.Y., Yang, S.C. (2011). Evidence for year-round occurrence of the eastern
Taiwan Strait Indo-Pacific humpback dolphins (Sousa chinensis) in the waters of
western Taiwan. Marine Mammal Science, 27(3), 652-658. DOI:10.1111/j.1748-
7692.2010.00422.x
Wang, J.Y., Hung, S.K., Yang, S.C. (2004a). Records of Indo-Pacific humpback
dolphins, Sousa chinensis (Osbeck, 1765), from the waters of western Taiwan.
Aquatic Mammals, 30(1), 87-194. DOI:10.1578/AM.30.1.2004.189
27
Wang, J.Y., Yang, S.C., Reeves, R.R. (2004b, February). Report of the first workshop on
the conservation and research of Indo-Pacific humpback dolphins, Sousa chinensis,
of the waters of Taiwan. National Museum of Marine Biology and Aquarium,
Checheng, Taiwan.
Wang, J.Y., Yang, S.C., Reeves, R.R. (2007a). Report of the second international
workshop on conservation and research needs of the eastern Taiwan Strait
population of Indo-Pacific humpback dolphins, Sousa chinensis. National Museum
of Marine Biology and Aquarium, Checheng, Taiwan
Wang, J.Y., Yang, S.C., Hung, S.K., Jefferson, T.A. (2007b). Distribution, abundance
and conservation status of the eastern Taiwan Strait population of Indo-Pacific
humpback dolphins, Sousa chinensis. Mammalia, 71, 157-165.
DOI:10.1515/MAMM.2007.032
Wang, J.Y., Yang, S.C., Fruet, P.F., Daura-Jorge, F.G., Secchi, E.R. (2012). Mark-
recapture analysis of the critically endangered eastern Taiwan Strait population of
Indo-Pacific humpback dolphins (Sousa chinensis): implications for conservation.
Bulletin of Marine Science, 88(4), 885-902.
Zhou, K., Xu, X., Tian, C. (2007). Distribution and abundance of Indo-Pacific humpback
dolphins in Leizhou Bay, China. New Zealand Journal of Zoology, 34(1), 35-42.
DOI:10.1080/03014220709510061
28
2.3 – Tables and Figures
Table 2.1 – Summary statistics of mean environmental variables measured during
sightings of Taiwanese humpback dolphins during surveys conducted in the eastern
Taiwan Strait from 2007-2012.
Depth SST (oC) Salinity (%) Distance to Shore (m)
Mean±SD 6.9±3.4 29.8±1.2 3.0±0.6 824±471
Minimum 1.9 23.2 0.4 1
Maximum 27.1 31.8 4.1 2721
Median 6.2 29.9 3.1 735
29
Figure 2.1 – Study area for the Taiwanese humpback dolphin (Sousa chinensis). The
distribution of preset waypoints (black, even numbered, and open, odd numbered squares
representing offshore and inshore waypoints, respectively), and the zig-zag tracklines
between waypoints are shown. Sandbars exposed at the lowest high tide of the year are
approximated by dark grey polygons, and thin broken lines mark the 30 m isobaths.
30
Figure 2.2 – Number of sightings of Taiwanese humpback dolphins made in the eastern
Taiwan Strait from 2007-2012 at different distances from shore (a), and water depths (b),
where shore is defined as any land mass exposed at the lowest high tide of the year
(following Ross et al., 2010).
0 10
0
20
0
30
0
40
0
50
0
60
0
70
0
80
0
90
0
10
00
11
00
12
00
13
00
14
00
15
00
16
00
17
00
18
00
19
00
20
00
21
00
22
00
23
00
24
00
25
00
26
00
27
00
28
00
29
00
0
5
10
15
20
25
30
35
Distance from shore (m)
Freq
uen
cy
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28
0
5
10
15
20
25
30
35
Depth (m)
Freq
uen
cy
a)
b)
31
Figure 2.3 – Frequency distribution of the best estimates of the number of Taiwanese
humpback dolphins observed in the eastern Taiwan Strait during sightings made from
2007-2012.
0
5
10
15
20
25
30
35
40
45
50
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
Sigh
tin
g fr
equ
ency
Field estimate of number of dolphins
32
Chapter 3 – Spatiotemporal Heterogeneity in Densities of the
Taiwanese Humpback Dolphin (Sousa chinensis taiwanensis)
Lauren E. Dares, Claryana C. Araújo-Wang, Shih-Chu Yang, and John Y. Wang.
A version of this chapter was published in Estuarine, Coastal and Shelf Science.
Permission to include this article can be found in Appendix A.
Dares, L.E., Araújo-Wang, C., Yang, S.-C., Wang, J.Y. 2017. Spatiotemporal
heterogeneity in densities of the Taiwanese humpback dolphin (Sousa chinensis
taiwanensis). Estuarine, Coastal and Shelf Science. DOI: 10.1016/j.ecss.2016.12.020
Contributions:
LED, CAW and JYW contributed to study design and writing of the manuscript. LED
processed data and conducted all statistical analyses. All authors participated in surveys
that produced data included in this study.
33
3.0 – Abstract
The inshore, estuarine distribution of Taiwanese humpback dolphins (THD) along
the west coast of Taiwan puts them in direct conflict with many anthropogenic activities.
We investigated the influence of environmental factors (depth, sea surface temperature
(SST), salinity and distance to the nearest freshwater source) and coastal developments
on THD density. Clear heterogeneity in density was found across the range of the THD,
with significant spatial and temporal variation in mean densities. Density was not directly
related to any environmental factors examined, which may be due to temporal variability
and hydrological and oceanographic conditions that create, at coarse temporal resolution,
a continuous river delta along the central west coast of Taiwan rather than isolated,
separate river estuaries. High number of dolphins per unit of survey effort (DPUE) and
mother-calf pairs per unit of survey effort (MCPUE) were found in waters adjacent to
reclamation areas, but neither distance to reclamation site nor distance to nearest river
were found to be significant predictors of density. Most reclamation projects in THD
habitat are situated near the mouths of major rivers or result in the creation of artificial
confluences of smaller rivers, streams and other freshwater outlets, such as waste
outflows. Thus, dolphins appear to use these areas in the absence of high quality natural
habitat that has been lost to large-scale coastal reclamation throughout their range.
34
3.1 – Introduction
Variation in population density across species’ ranges has frequently been linked
to environmental features, and may indicate differences in habitat use (Ferrer and
Donazar, 1996; Morris, 1987). Although habitat use does not necessarily represent habitat
quality (Van Horne, 1983), some habitat characteristics may be important for foraging,
predator avoidance, socializing or reproduction (e.g. Azevedo et al., 2007; Hastie et al.,
2004; Heithaus and Dill, 2002). However, it is often difficult to elucidate why particular
areas are important to or used by marine mammal populations, because the majority of
the life of a marine mammal is unavailable for observation. Therefore, a description of
the spatial distribution of population density may bring initial insights on the factors
driving habitat use and, consequently, for identifying areas for further research to better
understand the reasons for a species’ occurrence.
Understanding population densities hinges on quantifying the rates at which
animals are detected in different parts of their range. This is often complicated by the
inability of researchers to cover the entire study area equally due to logistical,
environmental or weather issues (e.g. Cañadas et al., 2005; Embling et al., 2010). Thus,
uneven survey effort in the study area must be considered when calculating densities,
otherwise estimates will be inflated in areas where greater survey effort is allocated.
For some species of marine mammals, environmental factors such as water depth
and distance from shore are important drivers of spatial distribution (e.g. Baumgartner,
1997; de Godoy et al., 2015; de Stephanis et al., 2008; Marubini et al., 2009). For coastal
species such as humpback dolphins (Sousa spp.), river estuaries have been particularly
linked to species’ occurrence (Jefferson and Karczmarski, 2001), especially for
35
populations in Chinese waters (Jefferson, 2000). Differences in habitat conditions such as
freshwater input may create variation in productivity between estuaries, in turn
influencing prey availability (Mallin and Paerl, 1994) for dolphins, while the presence of
anthropogenic disturbance may cause animals to avoid such areas (Weilgart, 2007).
The habitat of the Taiwanese humpback dolphin (THD; Sousa chinensis
taiwanensis) in the eastern Taiwan Strait is characterized by shallow, estuarine waters
within 3 km of shore (Dares et al., 2014; Ross et al., 2010) and is located in a highly
industrialized region, which puts the dolphins in constant conflict with human activities
(Dungan et al., 2011; Wang et al., 2004; Wang et al., 2007). Here we hypothesize that
THD density patterns are determined by the dolphins’ attraction to major freshwater
sources and avoidance of anthropogenic developments in the form of land reclamation
(i.e. disturbed or degraded) sites. We predict that dolphin densities will be higher in close
proximity to major rivers, and lower in waters adjacent to land reclamation sites in
comparison to more natural coastlines. We investigate these effects by comparing
differences in number of sightings and the numbers of dolphins and mother-calf pairs per
unit of survey effort.
3.2 – Materials and Methods
3.2.1 – Surveys
Line-transect surveys were conducted in the eastern Taiwan Strait from 120o43’51
N to 120o4’33 N between April 2007 and October 2015 using a 4.5m inflatable vessel
(for detailed survey methods, see Dares et al. 2014; Wang et al., 2012). Surveys could not
36
be conducted consistently from November-March due to inclement weather during these
months (Wang and Yang, 2011). Due to the low observer height of the survey platform,
the ability of observers to detect animals could be compromised by even moderately
windy or choppy conditions. Thus, only on-effort survey segments in which the Beaufort
Sea State (BSS) was recorded as ≤ 2 were included in the dataset to reduce the number of
animals being missed in poor marine conditions. Survey tracks recorded using a Garmin
GPSMAP 76 were imported into ArcMap 10.3.1 (ESRI, 2015) and then converted to
lines to calculate the linear distance covered during on-effort search for dolphins.
3.2.2 – Study Design and Data Analyses
3.2.2.1 – Spatiotemporal Heterogeneity
Due to the linearity of the habitat (animals found within 3 km of shore along a
<200 km stretch of coastal waters; Dares et al., 2014; Ross et al., 2010), the survey area
was divided into 18 approximately equal blocks of ~10 km latitudinal distance to
investigate the spatial heterogeneity of THD density throughout its known distribution.
Total and yearly survey effort (km) were calculated within each block, as well as yearly
number of sightings, number of dolphins and number of mother-calf pairs to calculate
SPUE (sightings per 1000 km survey effort), DPUE (dolphins per 1000 km of effort), and
MCPUE (mother-calf pairs per 1000 km of effort) following the formula:
SPUE = # of sightings in block
km of survey effort in block × 1000 (1)
For the spatial density analyses, any 10 km block that contained less than 200 km of
survey effort (~3% of total survey effort) was excluded from further analyses to prevent
inflation of densities due to low survey effort (the small denominator effect), leaving 14
37
blocks for comparison. For the yearly analyses, any blocks with less than 3% of the total
survey effort in that year were excluded. Chi-squared tests were used to determine
whether the distributions of sightings, dolphins and mother-calf pairs in each survey
block differed significantly from an expected uniform distribution based on the
proportion of the total survey effort conducted in each block according to the formula:
expected sightings = km of survey effort in block
total km of survey effort × total # of sightings (2)
Two-way analysis of variance tests were performed in R (R Core Team, 2015) to
compare variation in the yearly SPUE, DPUE and MCPUE for each survey block (note
that density calculations used in these tests were not multiplied by a factor of 1000).
Tukey’s honestly significant difference (HSD) was used to determine which survey
blocks and which years differed significantly in pairwise comparisons.
3.2.2.2 – Environmental Influences
The influence of artificial coastline on THD density was analyzed by outlining
four major areas of land reclamation and creating blocks that extended offshore of these
areas in ArcMap 10.3.1 (ESRI, 2015; Fig. 3.1). Total survey effort and total number of
sightings, dolphins and mother-calf pairs in all reclaimed blocks were calculated, and chi-
squared tests were used to determine whether the observed number of sightings, dolphins
and mother-calf pairs differed significantly from expected (calculated as in (2)) between
reclamation blocks and the rest of the survey area. In addition, another chi-squared test
was performed using expected number of sightings, dolphins, and mother-calf pairs in
each of the four reclamation blocks to test for heterogeneity in distribution among these
blocks.
38
Regression models were also used to determine the effects of environmental
conditions on THD density. The survey area was divided into 1 km2 grid cells, and the
number of sightings, dolphins and mother-calf pairs were used along with the survey
effort to calculate SPUE, DPUE and MCPUE for each grid cell as in (1). Grid cells with
less than 15 km of survey effort were removed from further analyses to avoid inflation of
density calculations due to the small denominator effect. The distance of each 1 km2 grid
cell to the mouth of the nearest of eight major rivers (Houlung, Dajia, Daan, Dadu,
Juoshuei, Peikang, Potzu and Chishui rivers; Williams and Chang, 2008) an outflow from
a major industrial complex (Luenwei Channel of the Changbin Industrial Park), or one
natural river that also has input from industrial sources (Xin Huwei River) (Fig. 3.1) was
calculated in ArcGIS 10.3.1 (ESRI, 2015). The distance of each grid cell to the nearest
point along the outline of the four major reclamation projects (as outlined in Google
Earth (2015) and imported into ArcMap 10.3; Fig. 3.1) was also calculated. To quantify
the influence of each river, measurements of salinity, sea surface temperature (SST) and
water depth taken during sightings of THDs (as in Dares et al., 2014) were used to
calculate mean values of each variable in each grid cell with sightings. Environmental
variables were tested for collinearity (see Appendix B) and linear regression was
performed in R (R Core Team, 2015) to establish the relationship between environmental
variables and the distance from the nearest river mouth. Poisson generalized linear
models (GLMs; quasipoisson in the presence of overdispersion, as in DPUE and
MCPUE) were also used to determine the relationships between SPUE, DPUE and
MCPUE and these environmental variables. The length of survey effort conducted in
each grid cell over the course of the study period was used as an offset, and the distance
39
to the nearest river and distance to the nearest reclamation project included as an
interaction term to test whether relationships between THD density and distance to
reclamation varied with the proximity to a freshwater source. Models were compared
using Akaike’s information criteria, corrected for small sample size (AICc; or QAICc for
quasipoisson models, which accounts for a dispersion parameter > 1; Burnham and
Anderson, 2002) among models containing all possible combinations of candidate
predictor variables, including an intercept-only model. Models with ΔAICc (or ΔQAICc)
<2 units larger than the model with the lowest AIC (or QAIC value) were considered to
have equal support (Burnham and Anderson, 2002).
3.3 – Results
3.3.1 – Even (10km) Blocks
From 2007-2015, 8586 km were surveyed and 209 sightings were made in BSS
≤2, which corresponded to 1345 dolphins and 201 mother-calf pairs recorded in the 14
blocks (Fig. 3.2). The highest SPUE and DPUE were found in block 6, followed by
blocks 5 and 11. Block 6 also had the highest MCPUE, and blocks 10 and 11 had higher
MCPUE than block 5 despite lower SPUE and DPUE in these blocks (Fig. 3.2). Chi-
squared tests indicated significant heterogeneity in observed numbers of sightings,
dolphins and mother-calf pairs (p < 0.001 for all tests) during the entire study period.
Tests for heterogeneity of THD density by year indicated that the number of observed
dolphins in survey blocks differed from expected in all years combined (p < 0.001 for all
tests), but the observed number of sightings and mother-calf pairs differed from
expectation only in some years.
40
Two-way analysis of variance tests indicated significant differences in mean
SPUE, DPUE and MCPUE between at least two survey blocks and at least two years (p <
0.05 for all tests). Post-hoc Tukey’s HSD tests indicated that differences in mean SPUE
and mean MCPUE were due to blocks 6 and 9, which had the highest and lowest values,
respectively, and the difference in DPUE was due to blocks 5 and 9 – again the blocks
with the highest and lowest DPUE values. Mean SPUE and DPUE differed significantly
in 2008 and 2011, and mean MCPUE differed significantly in 2011 and 2015.
3.3.2 – Reclamation Blocks
A total of 4845 km of survey effort was completed in waters adjoining
reclamation areas, with 121 sightings of 780 dolphins and 124 mother-calf pairs,
compared to 3982 km in waters with more natural coastlines, where 88 sightings of 565
dolphins and 77 mother-calf pairs were recorded (Fig. 3.1). Observed number of dolphins
(p < 0.001) and number of mother-calf pairs (p < 0.025) were significantly higher than
expected based on the amounts of survey effort in the reclaimed and more natural areas,
but the difference in observed and expected number of sightings was not significant (p >
0.1) between reclaimed and natural areas. When each reclamation block was considered
separately, the number of dolphin sightings in each reclaimed block again did not differ
from expected (p > 0.1), but significant heterogeneity was found among individual
reclaimed areas and the natural areas when comparing observed and expected number of
dolphins and mother-calf pairs (p < 0.001 for both tests). More dolphins than expected
were observed in waters adjacent to both Mailiao Industrial Park and Changbin Industrial
Area, while Budai Harbour and Taichung Harbour had fewer observed dolphins than
expected based on the amount of survey effort in those blocks.
41
The highest SPUE were observed adjacent to Changbin Industrial Park (28
sightings/1000 km), closely followed by the area adjacent to Mailiao Industrial Parks (26
sightings/1000 km), which were slightly higher than the SPUE for the more natural areas
(22 sightings/1000 km; Table 3.1). The highest DPUE and MCPUE was observed in the
Mailiao Industrial Park block (212 dolphins/1000 km and 32 pairs/1000 km), followed by
Changbin Industrial Park (184 dolphins/1000 km and 32 pairs /1000 km), both of which
were higher than in the more natural blocks (142 dolphins/1000 km and 19 pairs/1000
km; Table 3.1).
3.3.3 – Environmental Factors
The mean distance to the closest river mouth of a 1 km2 grid with at least one
sighting was 5.0 km (s.d. ± 3.2 km). Although the furthest distance was 14.3 km, 92% of
1 km2 grid cells with at least one sighting were within 10 km of a river. Multiple linear
regression indicated that no models containing any combination of environmental
predictors had greater support than the intercept-only model in predicting the distance of
the grid to a river (ΔAICc < 2; see Appendix B for further analyses). Comparison of
GLMs for relationships between SPUE, DPUE or MCPUE and any of the environmental
variables, including distance to river and distance to reclamation projects indicated that
no models containing any combination of predictor variables had more support than an
intercept-only model (ΔAICc <2; see Appendix B for further analyses).
3.4 – Discussion
The, inshore distribution of the THD is consistent with what is known about the
distributions of other populations of humpback dolphins, and is likely driven by the
42
distribution of the THD’s estuarine prey species (Dares et al., 2014). Differences in
SPUE, DPUE and MCPUE among survey blocks indicated significant spatial
heterogeneity in THD density across their range. The differences in mean SPUE, DPUE
and MCPUE were observed only between the highest- and lowest-density blocks. This
suggests that densities vary significantly across blocks from year to year, but are
consistently higher in some blocks than others, likely due to variations in environmental
conditions.
Estuarine productivity has been linked to fish biomass in estuaries (Mallin and
Paerl, 1994; Wilson, 2002), which could contribute to the temporal variability observed
in the DPUE and MCPUE of THDs as larger groups of animals and mother-calf pairs
may change their habitat use slightly from year-to-year to follow major shifts in prey
distribution. The eastern Taiwan Strait is a dynamic hydrogeographical region that is
influenced by freshwater flow from the many rivers that drain here, mixing with the
warm waters of the Kuroshio Current that enter the strait from the South China Sea (Huh
et al. 2011; Wang and Chern, 1988). Annual variability in the Kuroshio Current (Rudnick
et al., 2011) and differences in freshwater effluent among rivers can cause variation in
productivity as the nutrient and sediment discharge from rivers and degree of mixing
between fresh and salt waters may vary between estuaries (Gillanders and Kingsford,
2002).
There are 10 major freshwater effluent sources within the survey area, therefore
there are few areas that are not within ~10 km of any one of these ten major sources of
freshwater. There are also a number of small rivers that empty into the study area that
were not included in our analyses, which may also attract some dolphins to their
43
estuaries. The lack of any significant relationship between environmental variables and
the distance to the nearest river may suggest that relatively estuarine conditions are
present throughout most of the study area where THDs were sighted, regardless of the
proximity to a freshwater source, at least during the wet season when the surveys in this
study were conducted. However, since environmental data were only collected when
sightings were made, as few as one or two measurements of depth, SST, and salinity may
have been used to derive the mean values used as predictors in our regression models.
Thus, our already small dataset may not have been representative of the conditions in grid
cells with few THD sightings. (Also see Appendix B for further discussion)
Furthermore, the temporal variability in freshwater input from all of these sources
may complicate the relationship between THD density and proximity to river mouths. For
example, differences in echolocation click rates were interpreted to represent the usage of
the waters of one estuary (the Xin Huwei River, adjacent to the Mailiao Industrial Park)
by THDs, which differed depending on seasonal and tidal patterns (Lin et al., 2013; Lin
et al., 2014). THDs were found to venture closer to the river mouth during the winter
season and at higher tides, and move further offshore in summer when upstream rainfall
increases and during the ebb tide (Lin et al., 2013; Lin et al., 2014). Similar patterns may
also exist for other rivers that were not examined by the previous studies. Therefore,
further data collection and analyses on a finer temporal and geographic scale would help
to better elucidate the relationship between THD population density and the influence of
rivers.
Also, although THDs are present in the study area during the winter months
(Wang and Yang, 2011), our analyses only include data collected during the wet season,
44
when rainfall and thus freshwater outflow from rivers is at higher levels than other times
during the year (Lin et al., 2014). As a result, the patterns of spatiotemporal heterogeneity
observed in this study may not reflect habitat patterns in the dry season when there is less
freshwater influence in the survey area, and THDs may be more restricted to areas closer
to river mouths to find estuarine prey species. Our surveys were also only conducted
during daylight hours, and thus can only represent THD density during the day.
Differences in THD distribution related to the tidal cycle has been described in the Xin
Huwei river estuary (Lin et al., 2013), and other diurnal or tidal differences may also
exist in other estuaries along the west coast of Taiwan. Total outflow volume of each
river is likely also an important factor driving the heterogeneity in THD densities, and
may explain the lack of an apparent relationship between dolphin densities and distance
to a river. For example, the survey block with the lowest SPUE and DPUE, and no
mother-calf sightings was directly adjacent to the Juoshuei River, which is the largest
river in western Taiwan (Williams and Chang, 2008). However, the outflow volume and
influence of the Juoshuei River on the surrounding waters, and thus its estuarine function,
may be significantly less than its physical size suggests because this river has been
significantly dammed and diverted for agricultural, residential, and industrial use in
recent decades (Taiwan Water Resources Agency, 2016; Wang et al., 2004; Williams and
Chang, 2008). There are no dolphin density data prior to increased water extraction levels
so it cannot be determined if the reduced water flow to the Juoshuei River estuary, and/or
the land creation in the estuary for the Mailiao Industrial Park itself resulted in the present
low dolphin densities. However, some older local fishermen once called the waters on top
of where the Mailiao Industrial Park now sits (at the southern part of the Juoshuei River
45
estuary), the “dolphin pool”, which suggests that dolphin densities were likely much
higher, and this may have been important dolphin habitat in the past. The association of
humpback dolphins in Chinese waters with freshwater input is well-reported (Jefferson,
2000; Parsons, 2004), so the placement of reclamation projects on or in close proximity
to these areas destroys important habitat and may expose the THD to other anthropogenic
risks such as vessel traffic and pollution (Wang et al., 2004).
The natural coastline fronting THD habitat has declined by more than 20% in the
past 20 years (Karczmarski et al., 2016b; Wang et al., 2007), with three of the largest
reclamation projects being situated at the mouths of the two largest rivers (Dadu and
Juoshuei). Although the distance to the nearest reclamation project did not significantly
influence THD density, we found relatively higher numbers of dolphins and mother-calf
pairs in waters adjacent to reclaimed areas than expected. This suggests that the
immediate proximity of artificial coastline did not impact relative density, because the
waters adjacent to reclamation projects are still used by THDs, even though the quantity
and quality of the habitat has been reduced and degraded. Dolphin densities in these areas
may have always been relatively higher than other parts of the THD range, and the
apparent non-effect of the reclamation projects on dolphin density may be related to the
location of each large reclamation project being near the mouths of major rivers. For
instance, the Changbin Industrial Park was constructed in the 1970s (Bristow, 2010),
south of the Dadu river, and contains a channel that collects industrial, agricultural, and
residential outflows as well as the effluents from several small streams. This aggregation
of several smaller freshwater sources may artificially enhance productivity and attract
prey species of THDs, thereby also drawing dolphins to these waters while further
46
exposing them to industrial effluent, human waste and other pollutants (Wang et al.,
2004; Wang et al., 2007). Additionally, it has been over 40 years since construction of
this reclamation was completed, so enough time has likely elapsed that any initial
disturbance due to construction activities has passed.
Similarly, construction of the Mailiao Industrial Park began in 1992 and was
completed in the early 2000s (Formosa Petrochemical Corporation, 2014). The high
MCPUE in the survey block immediately adjacent to this reclamation, as well as within
the reclamation block itself, indicates that mother-calf pairs are still frequenting this area
despite the extensive development that has already occurred. There is no information on
THD density in this area before or during construction, but humpback dolphins have been
observed to avoid areas during construction periods and return after construction has
ended, at least in the case of percussive pile driving (Jefferson et al., 2009; Würsig et al.,
2000). Large-scale land reclamations, such as those that have occurred in the study area,
involve filling shallow coastal areas with rocks and sediment, resulting in complete loss
of that habitat from the marine ecosystem. Although the immediate construction impacts
of these reclamations are no longer a factor, the long-term effects of this removal of
habitat on cetacean distribution is difficult to quantify. Humpback dolphins in Hong
Kong and South Africa can live to at least 35 years of age (Jefferson et al., 2012), so it is
highly likely that some older THDs remember this area from before construction
degraded this potentially important habitat, and continue to return in the absence of better
quality habitat.
A recent study by Karczmarski et al. (2016) on the distribution of THDs
concluded that differences in SPUE between three sectors of the study area are due to
47
varying levels of habitat degradation (expressed as a habitat integrity index; HII) in each
sector rather than “natural patchiness of their environment”. The conclusions of that study
were made by looking at the study area at a very low spatial resolution and in the absence
of analyses of any oceanographic variables that might explain some of the differences in
distribution of sightings across their sectors. The division of their study area into three
arbitrary and unequally-sized sectors greatly influences their results, as moving the
borders of the central sector to include the Dadu River should significantly increase the
SPUE values observed in this part of the study area, thus affecting their conclusions that
the low SPUE in the central sector is due exclusively to the reclamation activities that
occurred there. Furthermore, the report from which the data used in Karczmarski et al.
(2016) was obtained (Chou and Lee, 2010; Chou et al., 2011) did not adequately sample
the inshore waters of the Dadu River estuary, where dolphins are often observed and in
which we, with more thorough sampling of this estuary, found high SPUE values.
Furthermore, the reports did not include track lines and how survey effort was allocated
to various study areas. Information on DPUE was also not included, and thus any factors
that might affect the number of dolphins using an area (as opposed to simply encounters)
in their sectors, was excluded. Our analyses of reclamation blocks in the study area did
not show significant differences in SPUE, indicating that including analyses of the
number of dolphins, rather than just the number of sightings, is an important facet for
investigating patterns in THD distribution. Finally, Karczmarski et al (2016) asserted that
reclamation was entirely responsible for the distribution of THDs based on a regression
model that showed a relationship between HII and SPUE. However, when sectors were
added to the regression as random effects, this relationship was no longer significant.
48
Using sector as a random effect only indicates that there is no relationship between HII
and SPUE when the data are separated by sector. This does not exclude “natural
patchiness” from affecting THD distribution based on this evidence alone, especially
when considering that HII was calculated based on the amount of habitat lost within each
sector.
Extensive habitat destruction and degradation in most of the THD’s restricted
range may mean that better quality habitat may not be available. In other examples where
marine construction has influenced the habitat use of a cetacean species, individuals may
relocate to less impacted areas in search of higher quality habitats following degradation
due to construction activities (e.g. harbour porpoises leaving an area affected by offshore
windfarm construction; Carstensen et al., 2006). In contrast, THDs do not have this
option, as they are found in a relatively small amount of suitable habitat located within
the study area. When high quality habitats are degraded or lost to reclamation, THDs
cannot simply relocate as their isolation from other estuary systems by the deep waters of
the Taiwan Strait means that they are already using all of the highest-quality habitat
available to them. Many of the dolphins that once frequented the area in and around the
Juoshuei/Xin Huwei river estuaries continue to use the area. However, these dolphins are
using the much-reduced southern portion (the Xin Huwei estuary) following the large
decrease in freshwater input from the Juoshuei River and large loss of this habitat due to
the construction of the Mailiao Industrial Park. Dolphins are sighted more often in the
southern waters, where the Xin Huwei River continues to drain into the sea, than those
west and north of the industrial park where there is little freshwater input. Despite the
continued degradation of the habitat, dolphins may still return to the general area, which
49
may be both culturally and biologically important to the dolphins, even in close proximity
to a large petrochemical plant and the associated shipping traffic.
In contrast, neither of the two reclamation blocks with considerably lower dolphin
densities, Taichung Harbour and Budai Harbour, have any significant freshwater outflow;
each serves as a port for large shipping vessels and fishing boats, respectively. The
Taichung Harbour reclamation began in 1976 (Hsu et al., 2007), putting it on a
comparable timeline to the Changbin Industrial area, thus any disturbance due to
construction activities are likely well in the past. The harbour’s south border is located on
the northern shore of the Dadu estuary, while the Daan and Dajia rivers are to the north of
the harbour. Although there are many sightings of dolphins within these estuaries, there
are very few sightings in the waters adjacent to Taichung Harbour itself. There is also a
dredge-maintained channel leading into the port, and although THDs have been observed
to cross this deep channel travelling at high swimming speeds (Dares et al., 2014), areas
with frequent dredging activities are typically avoided by other small cetaceans (e.g.
Pirotta et al., 2013). The more recent construction at Budai Harbour (begun in the 1990s
with a recent expansion completed in 2009; Chiayi County Government, 2009), is more
likely to have had a negative effect on dolphin densities as quantified in this study,
although other populations of humpback dolphins have been reported to return to areas
disturbed by construction in less than two years (Würsig et al., 2000). Surveys for this
study were not consistently conducted in the waters adjacent to Budai Harbour prior to
2010, thus we were unable to compare densities before, during and after the expansion. In
addition, disturbance and displacement by expansive oyster culturing and beam-trawling
in this area may also deter dolphins from using these waters more frequently.
50
In conclusion, we found considerable spatial heterogeneity in population density
of the THD. These dolphins appear to be less tied to any individual river than other
populations of humpback dolphins and can be almost anywhere within the study area
while staying in a generally estuarine habitat. We were unable to establish any significant
relationship between environmental variables and the distance to a river, which may be
due to sparse data collection in some areas that did not allow a representative sample of
the highly variable habitat conditions, as well as a small overall sample size for statistical
tests. Further quantification of river influence and oceanographic features should be
considered with a more robust dataset. Analyses at a finer temporal and spatial resolution
may be necessary to fully elucidate the importance of freshwater influences. The
influence of freshwater sources may have confounded the relationship between THD
density and the distance to major reclamation sites as dolphins appeared to be associated
with waters adjacent to reclaimed areas. However, rather than a real preference for waters
adjacent to reclaimed coastlines, the location of these large construction sites at the two
largest estuaries are likely the reasons for the observed pattern. This is further supported
by much lower dolphin densities near large reclamation projects from which there were
no large freshwater inputs. Specific estuaries may also be of significant cultural
importance for individuals, and unavailability of better quality habitat throughout the
restricted range of the THD means that dolphins must continue to use degraded habitats
near coastal developments. Given the Critically Endangered status of this subspecies and
its reliance on shallow estuarine habitats, future coastal reclamation projects should be
avoided at all costs. Further alteration and degradation of the THD remaining habitat,
51
together with the plethora of existing threats, including fisheries and pollution, could
trigger a population decline beyond recovery.
52
3.5 – References Azevedo, A.F., Oliveira, A.M., Viana, S.C., Van Sluys, M., 2007. Habitat use by marine
tucuxis (Sotalia guianensis) (Cetacea, Delphinidae) in Guanabara Bay, south-
eastern Brazil. J. Mar. Biol. Assoc. U.K. 87, 201-205.
Baumgartner, M.F., 1997. The distribution of Risso’s dolphin (Grampus griseus) with
respect to the physiography of the northern Gulf of Mexico. Mar. Mam. Sci. 13,
614-638.
Bristow, R. (Ed.)., 2010. Planning in Taiwan: Spatial Planning in the Twenty-First
Century. Routledge, London.
Burnham, K.P., Anderson, D.R., 2002. Model Selection and Multimodel Inference: A
Practical Information-Theoretic Approach (Second Edition). Springer, New York.
Cañadas, A., Sagarminaga, R., de Stephanis, R. Urquiola, E., Hammond, P.S., 2005.
Habitat preference modelling as a conservation tool, proposals for marine protected
areas for cetaceans in southern Spanish waters. Aquat. Conserv. 15, 495-521. DOI:
10.1002/aqc.689
Carstensen, J., Henriksen, O.D., Teilman, J. 2006. Impacts of offshore windfarm
construction on harbour porpoises: acoustic monitoring of echolocation activity
using porpoise detectors (T-PODs). Mar. Ecol. Prog. Ser. 321, 295-308.
Chiayi County Government., 2009. “Expansion of the three Budai harbors legislation
procedures is complete”.
<http,//www.cyhg.gov.tw/wSite/ct?xItem=1731&ctNode=14871&mp=12>.
Accessed 18 Feb 2016.
Chou, L.S., Lee, J.D, 2010. Habitat Hotspots of Humpback Dolphins, Sousa chinensis,
and Master Planning for Conservation Management (in Chinese). Taipei: Forestry
Bureau, Council of Agriculture, Executive Yuan 65pp.
Chou, L.S., Lee, J.D., Kao, C.C., Chuang, C.T., Chuang, C.T., Chen, C.F., Yang, W.C.,
Lee, P.F., Shao, K.T., Chen, M.H., Wei, R.C., Tsai, H.C., 2011. Population
Ecology, Critical Habitat and Master Planning for Marine Mammal Protected Area
of Indo-Pacific Humpback Dolphin, Sousa chinensis (in Chinese). Taipei: Forestry
Bureau, Council of Agriculture, Executive Yuan 202pp.
Dares, L.E., Hoffman, J.H., Yang, S.C., Wang, J.Y., 2014. Habitat characteristics of the
Critically Endangered Taiwanese Humpback dolphins (Sousa chinensis) of the
Eastern Taiwan Strait. Aquat. Mam. 40, 368-374. DOI: 10.1578/AM.40.4.2014.368
de Godoy, D.F., Andriolo, A., de Fatima Filla, G., 2015. The influence of environmental
variables on estuarine dolphins (Sotalia guianensis) spatial distribution and habitat
used in the Estuarine Lagunar Complex of Cananéia, southeastern Brazil. Ocean.
Coast. Manage. 106, 68-76.
de Stephanis, R., Cornulier, T., Verborgh, P., Sierra, J.S., Gimeno, N.P., Guinet, C.,
2008. Summer spatial distribution of cetaceans in the Strait of Gibraltar in relation
to the oceanographic context. Mar. Ecol. Prog. Ser. 353, 275-288.
53
Dungan, S.Z., Riehl, K.N., Wee, A., Wang, J.Y., 2011. A review of the impacts of
anthropogenic activities on the critically endangered eastern Taiwan Strait Indo-
Pacific humpback dolphins (Sousa chinensis). J. Mar. Anim. Ecol. 4, 3-9.
Embling, C.B., Gillibrand, P.A., Gordon, J., Shrimpton, J., Stevick, P.T., Hammond, P.S.,
2010. Using habitat models to identify suitable sites for marine protected areas.
Biol. Conserv. 143, 267-279. DOI: 10.1016/j.biocon.2009.09.005
ESRI, 2015. ArcGIS 10.3.1 for Desktop. Redlands, CA.
Ferrer, M., Donazar, J.A., 1996. Density-dependent fecundity by habitat heterogeneity in
an increasing population of Spanish imperial eagles. Ecology. 77(1), 69-74.
Formosa Petrochemical Corporation, 2014. “Our History”.
<http,//www.fpcc.com.tw/en/history.html>. Accessed Feb 9, 2016.
Gillanders, B.M., Kingsford, M.J., 2002. Impact of changes in flow of freshwater on
estuarine and open coastal habitats and the associated organisms. Oceanogr. Mar.
Biol. 40, 233-309.
Google Earth V 7.1.5.1557, December 13, 2015. Taiwan. 23.754221oN 120.396004oE,
eye alt 196.54 km. SIO, NOAA, U.S. Navy, NGA, GEBCO. Landsat 2016. <
http://www.earth.google.ca>. Accessed Jan 21, 2016.
Hastie, G.D., Wilson, B., Wilson, L.J., Parsons, K.M., Thompson, P.M., 2004. Functional
mechanisms underlying cetacean distribution pattern hotspots for bottlenose
dolphins are linked to foraging. Mar. Biol. 144, 397-403. DOI: 10.1007/s00227-
003-1195-4
Heithaus, M.R., Dill, L.M., 2002. Food availability and tiger shark predation risk
influence bottlenose dolphin habitat use. Ecology. 83, 480-491.
Hsu, T.W., Lin, T.Y., Tseng, I.F., 2007. Human impact on coastal erosion in Taiwan. J.
Coastal Res. 23, 961-973.
Huh, C.A., Chen, W., Hsu, F.H., Su, C.C., Chiu, J.K., Lin, S., Liu, C.S., Huang, B.J.,
2011. Modern (<100 years) sedimentation in the Taiwan Strait: Rates and source-
to-sink pathways elucidated from radionuclides and particle size distribution. Cont.
Shelf Res. 31, 47-63.
Jefferson, T.A., 2000. Population biology of the Indo-Pacific hump-backed dolphin in
Hong Kong waters. Wildlife Monogr. 144, 1-65.
Jefferson, T.A., Hung, S.K., Robertson, K.M., Archer, F.I., 2012. Life history of the
Indo-Pacific humpback dolphin in the Pearl River Estuary, southern China. Mar.
Mam. Sci. 28, 84-104. DOI: 10.1111/j.1748-7692.2010.00462.x
Jefferson, T.A., Hung, S.K., Würsig, B., 2009. Protecting small cetaceans from coastal
development: Impact assessment and mitigation experience in Hong Kong. Mar.
Policy. 33, 305-311. DOI: 10.1016/j.marpol.2008.07.011
Jefferson, T.A., Karczmarski, L., 2001. Sousa chinensis. Mamm. Species. 655, 1-9.
Karczmarski, L., Huang, S.L., Wong, W.H., Chang, W.L., Chan, S.C., Keith, M., 2016.
Distribution of a coastal delphinid under the impact of long-term habitat loss: Indo-
54
Pacific humpback dolphins off Taiwan’s west coast. Estuaries Coasts. DOI:
10.1007/s12237-016-0146-5
Lin, T.J., Akumatsu, T., Chou, L.S., 2013. Tidal influences on the habitat use of Indo-
Pacific humpback dolphins in an estuary. Mar. Biol. 160, 1353-1363. DOI:
10.1007/s00227-013-2187-7
Lin, T.H., Akamatsu, T., Chou, L.S., 2014. Seasonal distribution of Indo-Pacific
humpback dolphins at an estuarine habitat: Influences of upstream rainfall. Estuar.
Coast. 38, 1376-1384.
Mallin, M.A., Paerl, H.W., 1994. Planktonic trophic transfer in an estuary, Seasonal, diel,
and community structure effects. Ecology. 75, 2168-2184.
Marubini, F., Gimona, A., Evans, P.G., Wright, P.J., Pierce, G.J., 2009. Habitat
preference and interannual variability in occurrence of the harbour porpoise
Phocoena phocoena off northwest Scotland. Mar. Ecol. Prog. Ser. 381, 297-310.
DOI: 10.3354/meps07893
Morris, D.W., 1987. Tests of density-dependent habitat selection in a patchy
environment. Ecol. Monog. 57(4), 269-281.
Parsons, E.C.M., 2004. The behavior and ecology of the Indo-Pacific humpback dolphin
(Sousa chinensis). Aquat. Mam. 30(1), 38-55. DOI: 10.1578/AM.30.1.2004.38
Pirotta, E., Laesser, B.E., Hardaker, A., Riddoch, N., Marcoux, M., Lusseau, D., 2013.
Dredging displaces bottlenose dolphins from an urbanised foraging patch. Mar.
Pollut. Bull. 74, 396-402. DOI: 10.1016/j.marpolbul.2013.06.020
R Core Team (2015). R: A language and environment for statistical computing. R Founda
tion for Statistical Computing, Vienna, Austria. <https://www.R-project.org/.>
Ross, P.S., Dungan, S.Z., Hung, S.K., Jefferson, T.A., MacFarquhar, C., Perrin, W.F.,
Riehl, K.N., Slooten, E., Tsai, J., Wang, J.Y., White, B.N., Würsig, B., Yang, S.C.,
Reeves, R.R., 2010. Averting the baiji syndrome, conserving habitat for critically
endangered dolphins in Eastern Taiwan Strait. Aquat. Conserv. 20, 685-694. DOI:
10.1002/aqc.1141
Rudnick, D.L., Jan, S., Centurioni, L., Lee, C.M., Lien, R.C., Wang, J., Lee, D.K., Tseng,
R.S., Kim, Y.Y., Chern, C.S., 2011. Seasonal and mesoscale variability of the
Kuroshio near its origin. Oceanography. 24, 52-63. DOI:
dx.doi.org/10.5670/oceanog.2011.94
Taiwan Water Resources Agency, 2015. Reservoirs, Dams and Weirs of Taiwan.
<http,//eng.wra.gov.tw/ct.asp?xItem=48253&CtNode=7677>. Accessed Feb 18,
2016.
Van Horne, B., 1983. Density as a misleading indicator of habitat quality. J. Wildlife
Manage. 47, 893-901.
Wang, J., Chern, C.S., 1988. On the Kuroshio Branch in the Taiwan Strait during
wintertime. Prog. Oceanogr. 21, 469-491.
55
Wang, J.Y., Yang, S.C., Reeves, R.R., 2004. Report of the first workshop on the
conservation and research of Indo-Pacific humpback dolphins, Sousa chinensis, of
the waters of Taiwan. Natl. Mus. of Mar. Biol. Aquar., Checheng, Taiwan
Wang, J.Y., Yang, S.C., Hung, S.K., Jefferson, T.A., 2007. Distribution, abundance and
conservation status of the eastern Taiwan Strait population of Indo-Pacific
humpback dolphins, Sousa chinensis. Mammalia. 71, 157-165. DOI:
10.1515/MAMM.2007.032
Wang, J.Y., Yang, S.C., 2011. Evidence for year-round occurrence of the eastern Taiwan
Strait Indo-Pacific humpback dolphins. (Sousa chinensis) in the waters of western
Taiwan. Mar. Mam. Sci. 27, 65-658. DOI: 10.1111/j.1748-7692.2010.00422.x
Wang, J.Y., Yang, S.C., Fruet, P.F., Daura-Jorge, F.G., Secchi, E.R., 2012. Mark-
recapture analysis of the critically endangered eastern Taiwan Strait population of
Indo-Pacific humpback dolphins (Sousa chinensis): Implications for conservation.
Bull. Mar. Sci. 88, 885-902. DOI: dx.doi.org/10.5343/bms.2010.1097.
Weilgart, L.S., 2007. The impacts of anthropogenic noise on cetaceans and implications
for management. Can. J. Zool. 85, 1091-1116. DOI: 10.1139/Z07-101
Williams, J., Chang, C.D., 2008. Taiwan’s Environmental Struggle, Toward a Green
Silicon Island. Routledge, London.
Wilson, J.G., 2002. Productivity, fisheries and aquaculture in temperature estuaries.
Estuar. Coast. Shelf Sci. 55, 953-967. DOI: 10.1006/ecss.2002.1038
Würsig, B., Greene, C.R., Jefferson, T.A., 2000. Development of an air bubble curtain to
reduce underwater noise of percussive piling. Mar. Environ. Res. 49, 79-93.
56
3.6 – Tables and Figures
Figure 3.1 – Study area and distribution of the Taiwanese humpback dolphin. Linear
survey effort in 1 km2 grid cells completed in BSS ≤ 2 is shown, along with the ten major
rivers of western Taiwan. Major reclamation blocks are outlined, from north to south:
Taichung Harbour, Changbin Industrial Park to the Wang-gong agricultural area, Mailiao
Industrial Park (Formosa Petrochemical Corporation), Budai Harbour. Changbin
Industrial Park and Mailiao Industrial Park both act as sources of freshwater due to their
construction incorporating pre-existing rivers, while Taichung and Budai Harbours do not
contribute freshwater to the study area.
57
Figure 3.2 – Variation in Taiwanese humpback dolphin density by 10 km latitudinal
survey block, in (a) sightings per unit of effort (SPUE), (b) dolphins per unit of effort
(DPUE) and (c) mother-calf pairs per unit of effort (MCPUE). Spatial variation in density
is shown in maps, with darker shades indicating higher density. Boxplots show the
variation in density for each survey block for the study period.
58
Table 3.1 – Taiwanese humpback dolphin sighting, dolphin, and mother-calf pair density
(per 1000 km of survey effort) in waters adjacent to major reclamation projects compared
to more natural waters (i.e. where no major reclamation exists). Reclamation blocks are
given in north-south order. Changbin Industrial Park and Mailiao Industrial Park act as
sources of freshwater, while Taichung Harbour and Budai Harbour have no freshwater
influence.
Reclamation block
Sightings per 1000 km
survey effort
(SPUE)
Dolphins per 1000 km
survey effort
(DPUE)
Mother-calf pairs per
1000 km survey effort
(MCPUE)
Taichung Harbour 18 79 5
Changbin Industrial Park 28 185 32
Mailiao Industrial Park 26 212 32
Budai Harbour 8 30 4
All reclaimed areas 25 161 26
Natural areas 22 142 19
59
Chapter 4 – Niche Similarity Between Indo-Pacific Humpback
Dolphins (Sousa chinensis) of the Pearl River Estuary and
Eastern Taiwan Strait
Lauren E. Dares, Samuel K. Hung, John Y. Wang, Bradley N. White
Contributions:
The broad concept of this chapter, and editing of the final product, was a collaborative
effort of all authors. LED was responsible for study design, execution and writing. LED,
JYW and SKH participated in surveys that produced data used in this chapter.
60
4.0 – Abstract
Geographic separation of species can lead to differentiation, speciation, and
divergence of habitat requirements when separation persists over considerable timescales.
In this chapter, I compared the environmental niches of representative populations of the
two current subspecies of Indo-Pacific humpback dolphins (Sousa chinensis), which are
separated by more than 600 km and the relatively deep waters of the Taiwan Strait.
Survey data collected from 2008-2015 in the Pearl River Estuary (PRE) in southern
China and the eastern Taiwan Strait (ETS) were used along with remotely sensed depth,
turbidity, net primary productivity, sea surface temperature and distance to shore to
produce species distribution models (SDM) that linked the probability of humpback
dolphin presence in each area with these environmental characteristics during the wet
season (April-September) in both the PRE and ETS, and the dry season (October-March)
in the PRE. Area under the receiver operating characteristic curve (AUC) and true skill
statistic (TSS) values calculated from a spatially-buffered leave-one-out cross validation
procedure indicated that there were several best-performing models, which were
combined into a bounding box ensemble for each study area. Survey effort was an
important predictor in all best-performing models, while relative importance of
environmental predictors varied across models with no single environmental predictor
emerging as the most important. Predictions in environmental space indicated a slight
shift in niche dimensions in the PRE between wet and dry seasons as different predictors
are prioritized in each season, and comparisons of niche overlap indicated that ETS and
PRE niches are more similar than expected by chance despite taxonomic differences
between these two groups of humpback dolphins, and hydrogeographical differences
61
between the two study areas. Geographical predictions indicated that most of the
distribution of ETS humpback dolphins has likely been surveyed, and important areas are
located in the waters of central western Taiwan. Predictions in the PRE indicated the need
for further surveys to the west of the PRE, as there is predicted suitable habitat beyond
surveyed waters and possible connectivity between the PRE population of humpback
dolphins and others along the coast of southern China.
62
4.1 – Introduction
Indo-Pacific humpback dolphins (Sousa chinensis) inhabit estuarine and coastal
areas throughout the Indian and Pacific Oceans, from the waters of eastern India to
southeastern China (Mendez et al., 2013). Much of the research effort on Indo-Pacific
humpback dolphins has occurred in Chinese waters where five putative populations are
found, mostly distributed along the southeast coast of China (i.e. in the Jiulong River
Estuary, JRE, Chen et al., 2011; eastern Taiwan Strait, ETS, Wang et al., 2004; Pearl
River estuary, PRE, Jefferson, 2000; Leizhou Bay, Xu et al., 2012; and the Beibu Gulf;
Chen et al., 2009; 2016) with additional sighting records to the southwest of Hainan
Island (Li et al., 2016), between the PRE and JRE in the waters of Shantou (Wu, 2010),
and further north, near Ningde (Chen et al., 2012; Fig. 4.1). Once thought to inhabit a
continuous distribution along the majority of the Chinese coastline today the distribution
of humpback dolphins (locally called the Chinese white dolphin, CWD) is fragmented
(Jefferson and Smith, 2016), and putative populations are separated by at least a hundred
kilometres. Previous studies have compared morphological (Chen et al., 2018; Wang et
al., 2008, 2015) and genetic (Chen et al., 2010a; Lin et al. 2012) differences among
populations, but of the putative populations of humpback dolphin in Chinese waters, only
those found in the eastern Taiwan Strait (ETS) are considered taxonomically distinct
(Mendez et al., 2013). Separated from other populations of humpback dolphins by the
depths of the Taiwan Strait, this group comprises the entire subspecies S. c. taiwanensis
(locally known as the Taiwanese white dolphin, TWD; Wang et al., 2015). Although
there is some evidence of morphological differentiation among populations of CWD in
China (Chen et al., 2018), the remaining populations currently fall under the umbrella of
63
the nominate subspecies, S. c. chinensis. Taiwan was connected to the mainland of China
by a land bridge during the last glacial maximum (see Voris, 2000), and humpback
dolphins likely travelled to Taiwan from the waters of mainland China via the shallow
waters associated with this land bridge around 17,000-18,000 years ago (Wang et al.,
2015). Rising sea levels over the post-glacial period isolated the west coast of Taiwan
from mainland China, resulting in the geographic separation of the two subspecies of
Indo-Pacific humpback dolphin during this interval. Distinguishing morphological
differences in spotting patterns have emerged between TWDs and CWDs over time
(Wang et al., 2008, 2015), however a quantitative comparison of the environments
inhabited by different groups of Indo-Pacific humpback dolphins has not yet been
conducted.
Qualitative contrasts have been made between the general habitat characteristics
of the TWD and other populations of humpback dolphins (Dares et al., 2014), however
the geographic separation of the two S. chinensis subspecies and paucity of fine-scaled
occurrence data for many populations of humpback dolphins in Chinese waters make
quantitative comparisons difficult. A long-term monitoring program in the waters of
Hong Kong has made the PRE population arguably the best-studied, producing detailed
records of CWD abundance and distribution patterns within Hong Kong’s marine border
on the eastern side of the estuary. The rest of the estuary, however, constitutes a much
larger area of dolphin habitat than what is contained within Hong Kong’s borders, and is
surveyed much less frequently (see Chen et al., 2010b; SCSFRI and HKCRP, 2011;
SCSFRI, 2013). Hung (2008) first investigated the relationships between dolphin density
patterns and environmental characteristics in Hong Kong’s waters, reporting that
64
measures of dolphin density were highly correlated with several hydrological (i.e.
temperature, salinity, Secchi disk depth, turbidity, chlorophyll-a and total nitrogen) and
bathymetric (i.e. depth, benthic slope and shoreline type) parameters. Based on
similarities in habitat characteristics between the PRE and ETS (Dares et al., 2014), it
was expected that similar environmental variables (i.e. water depth, distance to shore,
turbidity, net primary productivity, and sea surface temperature (SST)) would also be
important predictors of the occurrence of TWDs in the eastern Taiwan Strait.
Despite qualitative similarities in hydrogeological habitat characteristics between
TWDs and CWDs, differences in habitat configuration between the PRE (i.e. contained,
funnel-shaped estuary with anti-clockwise circulation within the estuary and several
small islands; Mao et al., 2004; Zong et al., 2004) and ETS (i.e. relatively linear, subject
to a strong offshore branch of the Kuroshio current; Huh et al., 2011; Joe and Chern,
1988) raises the possibility that the relationships between environmental characteristics
and the occurrence of humpback dolphins in each area differ. When two species are
sympatric, examination of habitat characteristics may be sufficient to establish that
resource partitioning occurs, as they can be directly compared based on observed
environmental conditions. However, comparisons between allopatric species (or
subspecies, as in the case of Indo-Pacific humpback dolphins) are more challenging due
to the geographic distance between populations of interest, differences in data collection
between study areas, and differing climatic or biotic influences in separate habitats. To
overcome many of these issues, species-habitat relationships can be generalized and
contrasted in environmental space by modelling the species’ niches.
65
Hutchinson (1957) conceptualized a species’ niche as the n-dimensional
hypervolume at every point within which the species can persist indefinitely. In this
representation, the hypervolume exists in an abstract space in which each of the n
dimensions is a factor that influences fitness, and the entire hypervolume represents the
species’ fundamental niche, comprising the conditions that are suitable for a population to
thrive. Soberón (2007) distinguished between the types of variables that influence the
volume of the Hutchinsonian niche: Grinnellian factors (named for Joseph Grinnell, who
first described the scenopoetic niche of the California thrasher; Grinnell, 1917) include all
abiotic variables, while Eltonian factors (for Charles Elton, whose definition of niche
concerned a species’ “place” in an ecosystem relative to predators, prey and competitors;
Elton, 1927) are the biotic interactions that may reduce the volume of a species’
fundamental niche through competition for resources or increased predation risk. The
overlap of the biotic and abiotic factors that positively affect a species’ fitness is called
the realized niche, and limiting the realized niche to areas in geographic space that are
physically accessible to the species gives the occupied distribution (see the BAM diagram
of Soberón and Peterson, 2005, which outlines the biotic, abiotic and movement factors
that affect species’ niches and distributions). Species’ distributions exist in both
geographic and environmental (i.e. Hutchinson’s abstract n-dimensional) space (Peterson
and Soberón, 2012), so for clarity in this chapter distributions in geographic space are
referred to as “suitable habitat” and distributions in environmental space are referred to as
“niches”.
Models of species’ niches can be constructed using occurrence records and
associated measurements of environmental characteristics to establish relationships
66
between the species and its habitat (Guisan and Thuiller, 2005). These relationships are
the basis of species distribution modelling (SDM; also called ecological niche modelling
or habitat suitability modelling in some contexts; Soberón and Nakamura, 2009), and
involve two transitions between geographic and environmental space (Peterson and
Soberón, 2012): initial observations of occurrence and habitat characteristics are made in
geographic space, and the first transition occurs when relationships among these
observations are modelled in environmental space; the second transition involves using
these models to make predictions in geographic space, classifying previously unsurveyed
areas as suitable or unsuitable habitat for a species (Elith and Leathwick, 2009; Peterson
and Soberón, 2012). Since occurrence records are inherently subject to biotic factors (e.g.
competition, predation) in the environment, it is largely accepted that SDMs model the
realized niche of a species, and depending on the type of occurrence data used (presence-
only vs. presence-absence) and spatial extent of the prediction area, predictions made in
geographic space may represent either the species’ occupied or potential distribution
(Araújo and Peterson, 2012; Kearney, 2006; Peterson and Soberón, 2012).
As in the divergence of genetic and physical characteristics, species’ niches can
shift over evolutionary timescales (Peterson, 2011) such as the time period that has
elapsed since the separation of the ETS from the rest of China. Based on the taxonomic
divergence and differences in their respective habitats, I predicted that TWDs of the ETS
and CWDs found in the PRE occupy different realized niches in environmental space,
and the importance of environmental predictors in determining habitat suitability differs
for each group of humpback dolphins. To test these predictions, I constructed SDMs
using occurrence data collected in each study area, and compared modelled niches of
67
each group of humpback dolphins. Occurrence data were obtained from surveys
conducted in each study area, for which the prime objective is abundance estimation of
each population, albeit via different methodologies: line transect data are analyzed for
distance sampling abundance estimation in the PRE, especially within Hong Kong’s
waters (Chen et al., 2010b; e.g. Hung, 2016), while mark-recapture using photo-
identification data is conducted in the ETS (Wang et al., 2012). Due to these methods of
data collection, the priority placed on research efforts in certain parts of the habitat (i.e.
due to funding allocation for surveys in Hong Kong’s waters), and the sheer size of the
study areas in question, survey effort is not evenly distributed across the entire habitat,
and there are potential areas of suitable habitat that have never been surveyed. The
western boundary of the PRE population has yet to be definitively established as only
exploratory surveys have occurred west of Xiachuan Island (SCSFRI and HKCRP,
2011), and the northern and southern limits of the distribution of the TWD have not been
determined due to low survey effort outside of the areas with the highest dolphin
densities (see Dares et al., 2017). Assuming sufficiently large population size and
equilibrium with the climate (i.e. dispersal to new geographic areas is not in progress),
species generally inhabit the extent of the suitable habitat area that is accessible to them.
Thus, I predicted that suitable habitats for the CWD in the PRE and the TWD in the ETS
extend beyond currently surveyed waters based on the existence of additional estuarine
habitat and previous records of humpback dolphins in unsurveyed parts of both study
areas. SDMs were used to make predictions in geographic space beyond known
humpback dolphin habitat in each study area to determine the extent of these suitable
habitats in each study area.
68
4.2 – Methods
4.2.1 – Survey Data
4.2.1.1 – Pearl River Estuary
Line transect surveys in the PRE were conducted by the Hong Kong Cetacean
Research Project from 2008-2015, in which predetermined track lines placed
systematically throughout the study area were followed in search of humpback dolphins.
In waters within Hong Kong’s borders, track lines are located approximately 1 km apart
and traversed regularly as part of a long-term monitoring program (see Hung, 2008 for
detailed survey protocol). For the rest of the PRE, dedicated surveys were conducted less
frequently in the waters of mainland China and Macau during the study period (see Chen
et al., 2010b). Survey effort was recorded as the number of times a grid cell was traversed
in good conditions (Beaufort Sea State <5), as line-transect estimates of effective strip
width indicate that travelling through a grid cell once is sufficient for observers to cover
the entire grid cell (i.e. one unit of survey effort is equal to 1km2 surveyed).
The study area was overlaid with a grid consisting of 1km2 grid cells, and the total
survey effort for each grid cell was calculated for wet (April-September) and dry
(October-March) seasons across the study period (Fig. 4.1 b,c). Division of the year into
wet and dry season was based on the months which account for 80% of the annual
rainfall in the PRE (Zhang et al., 2009). Each grid cell was also designated as a presence
or absence grid cell based on whether at least one dolphin had ever been sighted in that
grid cell in each season during the study period.
69
4.2.1.2 – Eastern Taiwan Strait
Line transect surveys were conducted in the eastern Taiwan Strait during the wet
season (April-September; also months that account for the majority of the rainfall in
western Taiwan; Chen et al., 1999) from 2008-2015 using a 4.5m inflatable Zodiac
following zig-zag transects alternating between preset inshore and offshore waypoints
(see Dares et al., 2014 for detailed survey protocol). As with the PRE survey data, 1 km2
grid cells were overlaid onto the study area and presence or absence status recorded for
each grid cell. Due to the lower observer platform, effective strip width for ETS surveys
was estimated using a detection function calculated on the perpendicular distances of
sightings from the transect line (Buckland et al., 2001), and this distance was used to
create a buffer around transect lines in ArcMap 10.4 (ESRI, 2016). The area of this buffer
intersecting with each grid cell was used to calculate the survey effort in that grid (Fig.
4.1d).
4.2.2 – Environmental Predictor Data
In situ environmental data were not collected during every survey in either study area,
so remote sensing data were used to ensure full spatial coverage of environmental
variables, and to facilitate predictions of humpback dolphin probability of presence in
unsurveyed areas. Monthly mean sea surface temperature (SST) in degrees Celsius and
diffuse attenuation coefficient at 490 nm per metre (a proxy for turbidity; Stumpf and
Pennock, 1991), calculated from ocean colour data recorded by NASA’s MODIS Aqua
satellite at a spatial resolution of 4 km x 4 km were obtained from online databases
(OBPG, 2015a,b) and high-quality pixels corresponding to the study areas extracted using
SeaDAS 7.3 (Baith et al., 2001). Net primary ocean productivity (NPP) data, calculated
70
from a temperature-dependent chlorophyll-a model (Behrenfeld and Falkowski, 1997),
indicating the monthly mean milligrams of carbon produced per pixel per day, were
obtained from Oregon State University (OSU, 2017) in grid cells of 17 km x 18 km.
Water depth was assumed to be static over time since the study period spanned survey
seasons across eight years, and was estimated from satellite altimetry data recorded at 1-
minute spatial resolution (Smith and Sandwell, 1997; Scripps Institute of Oceanography,
2015). Monthly data were divided into wet (April-September) and dry (October-March)
seasons based on the months with above-average rainfall in both study areas (Zhang et
al., 2009; Chen et al., 1999), and seasonal means taken for each pixel across the study
period. Dry season data encompassed the end of one year and beginning of the next to
ensure continuity. Seasonal means for each dynamic environmental variable, and static
depth measurements were then interpolated via empirical Bayesian kriging (ArcMap
10.4; ESRI 2016) to create continuous surfaces of each environmental variable with pixel
sizes corresponding to the 1 km2 grid system created for each study area. Distance from
the centre of each grid cell to the nearest point on the shoreline was calculated in ArcMap
10.4 (ESRI, 2016) using shapefiles for each study area to represent land. As with depth,
the distance to the shoreline was considered a static variable despite tidal differences in
each study area due to the length of the study period.
4.2.3 – Modelling
4.2.3.1 – Data Pre-Processing
Preliminary model fits indicated significant spatial autocorrelation (SAC) in
model residuals. SAC is a case of pseudoreplication that occurs when samples are taken
71
in close proximity to one another and statistical models do not take the spatial structure of
the data into account (Legendre, 1993). The inherent spatial structure of many
environmental variables means that collecting additional data does not necessarily
increase the amount of information contained within the dataset, as many samples in
close proximity often have the same or similar environmental conditions, and thus the
effective sample size of the dataset is smaller than the true sample size (Dale and Fortin,
2002; Griffith, 2005). When models incorporate spatial patterns in fitting the data (i.e.
spatial patterns in the response data are explained by environmental variables), model
residuals do not show spatial dependency and SAC is not an issue. However if a model
does not fit these patterns well, then SAC will be evident in model residuals, violating the
assumption of independence, which affects model outputs (Dormann, 2007).
Various statistical methods to address spatial autocorrelation have been evaluated
elsewhere (see Dormann et al., 2007), however comparable spatial methods have not
been developed for any of the machine learning algorithms that have recently appeared in
the SDM literature (e.g. see Elith et al., 2006; Qiao et al., 2015). To facilitate the use of
multiple algorithms in this chapter, data were pre-processed to reduce spatial
autocorrelation to avoid the need for statistical interventions within model structure.
Aggregating occurrence and environmental data into grid cells is one frequently used
method of reducing spatial autocorrelation and biased sampling of occurrence data
(Fourcade et al., 2014; Hijmans and Elith, 2017), and was accomplished by the
organization of the occurrence and environmental data in this chapter. Additional
rarefaction was required, however, and data were further “thinned” by selecting the grid
cells that were the most different in their environmental characteristics. Thinning
72
(=filtering) datasets reduces spatial autocorrelation and improves SDM performance
compared to modelling unthinned datasets (Boria et al. 2014), and thinning in
environmental space has been found to be more effective than filtering in geographic
space (i.e. where one or more occurrences are omitted if they are within some geographic
distance of other occurrences; Aiello-Lammens et al., 2015; de Oliveira et al., 2014;
Varela et al., 2014). I developed a custom algorithm (see Appendix C for R code) to thin
my datasets in environmental space, similar to the procedure outlined by de Oliveira et al.
(2014). For each dataset, presence grid cells were plotted in environmental space, with
each of the candidate predictor variables (depth, turbidity, productivity, SST and distance
to shore) as a dimension. Mahalanobis distances between every pair of points in the
presence dataset were calculated, and an algorithm in R (R Core Team, 2016) was
designed to systematically eliminate the most proximate point, then re-calculate
Mahalanobis distances among the remaining points in the dataset. The algorithm
continued until no points remained, resulting in a list of least- to most-distant presence
points in environmental space. This procedure was repeated for absence data points from
each dataset (i.e. surveyed grid cells where no dolphins had been sighted during the study
period). Preliminary models of each algorithm type and combination of predictor
variables were then fit using candidate dataset sizes varying from 40 to 200 data points,
with equal numbers of presences and absences to give a prevalence (proportion of
presences and absences) of 0.5 in each dataset. A dataset containing all presences and an
equal number of most-distant absences for that study area was also included for
comparison. Global Moran’s I tests were performed using the ape package in R (Paradis
et al., 2004) on the residuals of each model, and the largest candidate dataset size that
73
contained no significant positive spatial autocorrelation (observed Moran’s I value < 0
and/or p > 0.05) was chosen for each dataset to be used in model evaluation. Thinned
datasets were also tested for collinearity of environmental predictors using the corrplot
package in R (Wei and Simko, 2017) by ensuring that the absolute value of the Pearson
correlation coefficient (r) between any two environmental predictors was < |0.7|
(Dormann et al., 2013).
To ensure final model predictions reflected the full range of environmental
variation in each dataset, an alternative subset was selected by removing the 10 most-
distant presences and 10 most-distant absences in environmental space, then re-running
the abovementioned Mahalanobis distance ranking algorithm. This resulted in a new,
differently-ordered list of least- to most-distant points, from which an alternative most-
distant subset of the same size as the original subset was selected. This process was
repeated, removing most-distant points in increments of 10 from the original subset, until
all of the points in the original subset were removed, so that the final alternative subset
contained no points that were present in the original subset (see Appendix C for
illustration and summary statistics of final thinned datasets used for modelling).
Preliminary SAC tests were then conducted on these alternative subsets to determine
which should proceed to modelling based again on the proportion of preliminary models
that showed negative or insignificant SAC in residuals according to global Moran’s I.
4.2.3.2 – Model Fitting and Cross-Validation
Probability of humpback dolphin presence in each study area was modelled with a
suite of algorithms frequently used in SDM (Elith et al., 2006; Qiao et al., 2015),
implemented using various R packages. Generalized linear models (GLM; stats package,
74
R Core Team, 2016) relate the response variable to the linear combination of predictors
using a link function. Logistic regression GLMs were used with observed dolphin
presences and absences as the response variable and second degree polynomials of each
environmental variable and survey effort as predictors. Generalized additive models
(GAM; mgcv package, Wood, 2011) are semi-parametric extensions of GLMs that also
make use of a logit link function to model binomial data, but use smoothing functions to
fit relationships between response and predictor variables, and thus are more flexible than
GLMs (Guisan et al., 2002). To reduce significant overfitting of training datasets, GAMs
were implemented with four degrees of freedom (as per Elith et al., 2006)
Random forest (RF) is a machine learning algorithm that averages a large number of
regression or classification trees to model probability of presence from environmental
data (Breiman, 2001). A random subset of the possible environmental predictors are used
at each split so that trees are less correlated with one another than trees that consider
every predictor at every node, which can improve predictive performance (Breiman,
2001; James et al., 2015). Random forests were implemented using the randomForest
package in R (Liaw and Wiener, 2002), using ensembles of regression trees as
recommended by Hijmans and Elith (2017).
Boosted regression trees (BRT, also known as gradient boosting machines) also take
the average of a large number of regression trees, but make use of boosting – a machine
learning concept that uses several simple rules to make predictions rather than attempting
to find one highly accurate rule (Schapire, 2003). A large number of small regression
trees are fit in a stagewise fashion, where the first tree reduces loss of predictive
performance as much as possible, and each additional tree focuses on the residuals to
75
further reduce the loss function (Elith et al., 2008). Since boosted regression trees are
prone to overfitting the data when too many trees are added (Hastie et al., 2001), a cross-
validation procedure (gbm.step in package dismo; Hijmans et al., 2017) was used to
determine the optimal number of trees during the model selection procedure.
Several candidate models of each algorithm were fit so that all possible combinations
of environmental variables were considered in model evaluation. Survey effort was also
included in each candidate model to account for the differing amounts of effort expended
in certain parts of each study area to avoid biasing models towards conditions in highly-
surveyed areas.
A spatially buffered leave-one-out cross-validation procedure (SBLOOCV; as
described by Roberts et al., 2017) was implemented to ensure that training and test
datasets were spatially independent (see Appendix C for R code). The major range (i.e.
furthest distance at which SAC was evident in the semivariogram) of the SAC in the
residuals was assessed using directional semivariograms (gstat package; Pebesma, 2004)
calculated from the residuals of a preliminary GLM performed on each full dataset (i.e.
not thinned to reduce SAC), with each model parameterized with only an intercept term
as a predictor. The major range was then used as a buffer distance around the left-out test
grid cell during the SBLOOCV procedure, so that all data points within this buffer
distance of the test grid cell were omitted from the training dataset for that iteration of
cross-validation. Each model was then fit using the training data and used to predict the
probability of dolphin presence in the test grid cell based on the environmental
characteristics of that grid cell. Any training datasets that contained fewer than 50 data
76
points or which produced models with convergence issues or other errors during the
model fitting procedure were omitted from model evaluation.
4.2.3.3 – Model Evaluation
Model predictions from the SBLOOCV procedure were assessed using two model
evaluation criteria that are commonly employed in SDM: area under the receiver
operating characteristic curve (AUC) and the true skill statistic (TSS). The AUC is a
threshold-independent measure of model predictive performance that is calculated from a
receiver operating characteristic curve (ROC). A ROC is a graphical representation of the
trade-off between true positive rate (TPR; proportion of correctly predicted presences,
also known as sensitivity) and the false positive rate (FPR; proportion of absences
incorrectly predicted as presences, often given as 1-specificity, with specificity being the
proportion of correctly predicted absences) at all possible thresholds to convert
probabilities of presence to binary presence/absence values (Fielding and Bell, 1997;
Pearce and Ferrier, 2000). The area under this curve (AUC) is a frequently-used measure
of a model’s predictive ability (Elith et al., 2006; Elith and Leathwick, 2009; Manel et al.,
2001; Thuiller et al., 2009).
TSS is a threshold-dependent measure of model accuracy which accounts for the
accuracy expected to occur by chance and is not dependent on the prevalence of the data
(Allouche et al., 2006). TSS is calculated by the formula:
TSS = Sensitivity + Specificity − 1 (1)
and ranges from -1 (predictions no better than random) to +1 (perfect prediction;
Allouche et al., 2006).
77
Additional considerations were made during model evaluation due to the high
likelihood of false absences (i.e. absences recorded despite dolphins being present in a
grid cell) in each dataset. Due to the low survey effort in most of the PRE study area
outside of Hong Kong’s waters, grid cells with absences located outside of Hong Kong’s
administrative border (Fig. 4.1a,b) are potentially false absences, in which dolphins may
not have been observed due to insufficient survey effort rather the grid cell being
unsuitable. However, the high survey effort around Hong Kong means that absences
located within Hong Kong’s waters can be considered true absences, thus these absences,
together with all presences observed across the entire PRE, were used to evaluate PRE
wet season and dry season models. There are few, if any, reliable absences in the ETS,
which poses a problem for the use of AUC in evaluation of TWD models. Recent studies
have questioned the utility of AUC in SDM for various reasons, including its equal
weighting of false positive and false negative predictions when these predictions errors
have both unequal likelihood and unequal consequences in ecological data (Liu et al.,
2011; Lobo et al., 2008). In the case of the ETS dataset, where reliable absences were not
available, partial AUC (pAUC, where the area under only a portion of the curve is
calculated) was used to emphasize the importance of true positive rate in model
evaluation due to the likely high number of false absences. This was accomplished by
calculating the area under the curve across all values of FPR when the TPR was > 0.95
(i.e. fewer than 5 TWD presences in the ETS dataset were incorrectly predicted as
absences; McClish, 1989). High-performing models for each dataset were selected based
on an AUC (or pAUC for TWD models) ≥ 0.7 and TSS ≥ 0.5.
78
4.2.3.4 – Variable Importance
To compare the relative importance of predictor variables among datasets,
predictors for each high-performing model were scored from 1 (least important) to 6
(most important), with excluded predictors receiving a score of 0. For GLMs, scores were
based on slope parameter estimates of the mean-standardized predictors, while the mean
of the smoothing parameter coefficients was used for GAMs, and relative importance of
predictor variables was used in the case of BRTs and RFs (see Elith et al., 2008;
Friedman, 2001). The mean and standard error of these scores were then taken for each
predictor to compare the relative importance of each environmental variable in predicting
probability of dolphin presence for each dataset.
4.2.3.5 – Ensembles and Niche predictions
All best-performing models for each dataset were selected for inclusion in ensemble
predictions based on their AUC (or pAUC) and TSS scores. Models meeting these
criteria were combined in a “bounding box” ensemble (Araújo and New, 2007), which
was used to visualize niches in environmental space and project these niches into
geographic space to visualize suitable humpback dolphin habitat in each study area and
season. Probabilities of dolphin presence in each grid cell were converted to
presence/absence based on the probability threshold that maximized the sum of
sensitivity and specificity for that model, based on a ROC constructed using predicted
probabilities and observed presences or absences in surveyed grid cells (Jiménez-
Valverde and Lobo, 2007; Manel et al., 2001). The sum of these values was taken across
models to calculate a bounding box ensemble score for each grid cell indicating how
many models predicted a presence in that grid cell. Scores were then converted to
79
bounding box presence or absence based on the threshold number of models that again
maximized the sum of sensitivity and specificity.
Predictions were made in environmental space on a dataset extending beyond the
ranges of environmental variables observed in each study area to explore the boundaries
of the predicted niches outside of observed environmental conditions. Survey effort was
included in this dataset as a constant for each study area, using the mean survey effort
across grid cells with dolphin presences for each full dataset. Best-performing models for
each study area were then used to predict probability of dolphin presence at each position
in environmental space, with combinations of the five environmental variables used as
predictors (depth, distance to shore, productivity, turbidity and sea surface temperature)
acting as coordinates in 5-dimensional space. Probabilities were then converted to model
presences and absences and bounding box presences and absences according to the
thresholds mentioned above. Niches were visualized in environmental space as 3-
dimensional ellipsoids encompassing the volumes of predicted dolphin presence using the
rgl package (Adler et al., 2018) to represent the n-dimensional hypervolume.
Overlap of CWD and TWD niches was quantified using randomization tests for niche
similarity described by Warren et al. (2008). This tests the hypothesis that the best
models for study area A predict dolphin presences in study area B better than expected by
chance, and vice versa. First, two measures of niche overlap were calculated (using the
modOverlap function in the fuzzySim R package; Barbosa, 2015) for predictions made in
each study area (A and B) using models for species A and B: Schoener’s D (Schoener,
1968) is a frequently-used statistic for niche overlap in ecology that ranges from 0 (no
overlap) to 1 (niches are identical), calculated by
80
𝐷(𝑝A, 𝑝B) = 1 −1
2∑ |𝑝A,𝑖 − 𝑝B,𝑖|𝑖 (2)
where 𝑝A and 𝑝B are the distributions of probabilities predicted by models for datasets A
and B, and 𝑝A,𝑖 and 𝑝B,𝑖 are the predicted probabilities for each study area in grid cell i;
Warren’s I (Warren et al., 2008) is a similarity statistic developed to allow the
comparison of Hellinger distance, a mathematical measure of the distance between two
probability distributions,
𝐻(𝑝A, 𝑝B) = √∑ (√𝑝A,𝑖 − √𝑝B,𝑖)2𝑖 (3)
with more conventional ecological measures of niche overlap. Similar to Schoener’s D,
Warren’s I,
𝐼(𝑝A, 𝑝B) = 1 −1
2𝐻(𝑝A, 𝑝B)2 (4)
ranges from 0 to 1 (Warren et al., 2008). For study area A, these niche similarity metrics
were then compared to null distributions comprising 100 values of D and I calculated
between the mean probability of dolphin presence predicted in study area A using the
best models for study area B and the mean predicted probability of dolphin presence in
study area A from random models. Random models were of the same algorithm types as
the best models from study area B (i.e. GLM, GAM, BRT or RF), with the same
candidate predictors and number of presences and absences as used to generate the
models for study area B, but calibrated on a random selection of grid cells drawn from the
study area A dataset and randomly assigned as presences or absences. I and D values
indicating the degree of similarity between best study area A and study area B model
predictions on the study area A dataset were then compared to the null distribution using
81
a two-tailed z-test to determine whether the predictions made in study area A by study
area B models were more similar or more different than expected by chance. This
procedure was then repeated in the opposite direction, and with each possible pairing of
PRE wet season, PRE dry season and ETS datasets and models.
Predictions of humpback dolphin distribution in geographic space beyond surveyed
waters were made using the best-performing models for each study area to predict
probability of dolphin presence on interpolated remotely sensed data. Probability values
for each 1 km2 grid cell were converted to presence/absence based on the threshold that
maximized the sum of sensitivity and specificity, and converted to bounding box
ensemble presences and absences as described above.
4.3 – Results
4.3.1 – Dataset Summaries
Sizes of the available datasets varied due to the large estuarine area of the PRE
compared to the much smaller system of estuaries along the west coast of Taiwan (Fig.
4.1). There were 673 1 km2 grid cells with survey effort in the ETS, 102 of which had a
dolphin presence recorded at least once during the wet seasons of the eight year study
period. The PRE wet season dataset comprised 2814 grid cells with effort, 324 with
presences, while the PRE dry season dataset had 2500 grid cells with effort, including 315
with presences.
Ranges of some environmental variables differed among datasets. Depth and
distance to shore were considered static variables across the survey period, thus there
82
were few differences between the PRE wet and PRE dry datasets (Fig. 4.2a,b), which can
be attributed to variability in which grid cells outside of Hong Kong’s waters were
surveyed at different times of the year. Depth of grid cells in the ETS were comparable
with the PRE, though surveyed grid cells were generally closer to shore and in shallower
waters than either PRE dataset, with the exception of some outliers. The ETS was
characterized by lower turbidity (Fig. 4.2d) and net primary productivity (Fig. 4.2c) than
the PRE in both wet and dry seasons. The largest variation was in sea surface
temperature, with visible separation among datasets (Fig. 4.2e). The highest observed
temperatures were recorded in the ETS, followed by the PRE in the wet season.
Temperatures in the dry season in the PRE were, as expected, lower than the other two
datasets, as the dry season coincides with the cooler autumn and winter months.
Thinning each dataset in environmental space to reduce SAC resulted in three
subsets each for ETS, PRE in the wet season and PRE in the dry season, with subsets
comprising 80 grid cells, 180 grid cells and 140 grid cells, respectively. Subsets showed
no significant positive SAC (Moran’s I ≤ 0 or p > 0.05) in preliminary tests for the
majority of the candidate models across algorithm families and candidate predictor
variables. The only exceptions were RFs trained on the thinned ETS dataset, which
exhibited significant SAC (Moran’s I p < 0.05) with subsets as small as 40 data points.
As a result, ETS RFs were excluded from further analyses.
4.3.2 – Model Selection
Bounding box ensembles comprised 621 models for the PRE in the dry season,
115 for the PRE in the wet season, and 47 for the ETS. The majority of the best-
performing models across all datasets were GLMs, while some GAMs were also selected
83
for each study area. Some RFs were selected for the PRE in both the wet and dry seasons,
but were omitted from analyses on ETS data due to preliminary Moran’s I tests. No BRTs
were selected for PRE datasets, but several performed well for ETS (Fig. 4.3).
4.3.3 – Variable Importance
Survey effort was the most important predictor of dolphin presence for all
datasets. The relative importance of environmental predictors varied across datasets,
including across subsets for each study area (Fig. 4.4). No single environmental variable
emerged as the most important in predicting the probability of dolphin presence in either
study area relative to the other predictors included in the analyses.
4.3.4 – Niche Comparisons
Comparisons of model overlap based on Schoener’s D and Warren’s I differed
based on the direction of predictions. Predictions made in the PRE during the wet season
using the best ETS models indicated significantly less model overlap than expected by
chance according to both model overlap statistics (Table 4.1). In contrast, predictions
made in the ETS using the PRE wet season models showed more model overlap than
expected by chance (Table 4.1), indicating that ETS models predicted dolphin presences
in the PRE worse than models constructed from randomly selected data, though the
opposite was true for PRE wet model predictions made in the ETS: PRE wet models
predicted ETS dolphin presences significantly better than random data models.
Conditions in the ETS make it difficult to survey during the winter months (Wang
and Yang, 2011), so no direct comparison between PRE and ETS could be made for the
dry season. Instead, model overlap was compared between PRE dry season models and
84
ETS models from the wet season. Similar to the comparison between ETS and PRE wet
season models, predictions of PRE dry season models on the available ETS wet season
data also indicated higher model overlap than expected by chance according to both D
and I, but ETS models predicted PRE dry season presences worse than random models
(Table 4.1). Comparisons between PRE wet season and PRE dry season models indicated
higher model overlap than expected by chance in both directions, except for the
calculation of Warren’s I when predicting on PRE dry season data using PRE wet season
models (Table 4.1), which indicated less model overlap than expected by chance,
suggesting that dry season models, at least, may be capable of producing reliable
predictions of humpback dolphin presence regardless of the season in which they are
applied.
Projecting niches into environmental space with covariance ellipsoids showed
complete separation of niches along only the SST axis, which appears to reflect the
differences in SST between datasets observed in the data used to fit models (see Fig. 4.2).
The TWD niche was limited to temperatures > 29oC, while the CWD wet season niche
encompassed a temperature range from 27.5-28.5oC, and CWD dry season niche
occupied a lower range of temperatures, from 22-23oC (Fig. 4.5). Niche ellipsoids
overlapped along all other environmental axes, with CWD wet and dry season ellipsoids
showing similar dimensions in depth, distance to shore and productivity, but differing
slightly along the turbidity axis as the CWD wet season niche spanned a wider range of
turbidity values (Fig. 4.5). The TWD niche was limited to shallower waters much closer
to shore than either CWD niche, but showed more generality along the productivity and
turbidity axes, spanning the ranges included in the environmental space prediction dataset
85
of both variables (Fig. 4.5). Niches for the CWD in the wet and dry seasons do not truly
represent separate niches, rather that the niche for CWDs in the PRE occupies the entire
environmental space between the two visualized niches and a gradual shift in
environmental space likely occurs from the wet season to dry season, which may also be
the case for TWDs in the ETS.
4.3.5 – Geographical Predictions
Model predictions in the PRE varied somewhat between wet and dry seasons (Fig.
4.6a-d). Bounding box ensembles for both seasons predict continuous suitable habitat for
CWDs stretching from the waters of central and western Hong Kong in the east to at least
Hailing Island in the west. Both models also predicted pockets of suitable habitat to the
east of the PRE, along the coast of China east of Hong Kong, and also further west, to
both the south and northeast of Leizhou Bay. Suitable habitat in Hong Kong’s waters
extends slightly further south in the wet season compared to the dry season, while this
pattern is reversed in the western part of the study area. Bounding box predictions also
indicated that suitable habitat for CWDs is continuous along the coastline of southern
China between Hong Kong and Leizhou Bay during the dry season, whereas wet season
predictions show a ~75 km gap in suitable habitat to the west of Hailing Island.
ETS bounding box ensembles predicted suitable habitat for TWDs primarily in
the waters of central western Taiwan, stretching from just south of the Daan River estuary
to the southwestern tip of a large sandbar in nearshore waters, Waisanding Zhou, with an
additional small area of predicted suitable habitat located further south (Fig. 4.6e,f).
Suitable habitat was predicted to be discontinuous within this range, with gaps occurring
in the Juoshuei River estuary and north of the northeastern tip of Waisanding Zhou.
86
4.4 – Discussion
Humpback dolphins of the ETS and PRE occupy similar environmental niches,
and all of the environmental variables included in these analyses were important
predictors of humpback dolphin presence in some models. Predictions of CWD presence
indicated that suitable habitat extends beyond surveyed waters of the PRE, with the
westernmost limit of suitable habitat predictions varying between wet and dry seasons.
Suitable habitat for TWDs is located in the central waters of the ETS, with some
discontinuity between suitable areas, and at least some models predicting TWD presence
to the north and south of currently surveyed waters.
PRE wet and dry season models overlap with ETS models more than would be
expected due to regional similarities between the two habitats when making predictions
in the ETS. This suggests that, despite some differing environmental conditions between
the study areas, humpback dolphins found in the ETS and PRE occupy similar niches in
environmental space. There was less niche overlap when ETS models were applied to
PRE data, which may be due to overfitting of models to the small ETS dataset, resulting
in lower predictive performance when these models were applied to the larger study area
and more varied environmental conditions in the PRE. Species distribution models
require a minimum number of presences to adequately model the relationships between a
species and its environment, and smaller sample sizes are associated with decreased
predictive performance (Stockwell and Peterson, 2002; Wisz et al., 2008). To reduce
spatial autocorrelation in model residuals, the ETS dataset was thinned to a relatively
87
small sample size (n=80, prevalence=0.5), and effects of modelling small sample sizes
may have been exacerbated by the spatially-buffered cross-validation procedure
employed in this study. Data points within the buffer distance of the test point were
excluded from model fitting for each fold of cross-validation, which may have
additionally decreased the sample size of the training dataset, though folds with very
small training datasets (n < 50) were omitted from the calculation of model evaluation
criteria in an effort to minimize the small sample size effects. Specialist species found
with narrow geographic ranges are more easily-modelled than wide-ranging generalists
(Hernández et al., 2006; van Proosdij et al., 2016), and surveying areas within and outside
of the species range, providing true absences where the species does not occur improves
model performance, especially for wide-ranging species (Brotons et al., 2004).
Humpback dolphins are considered generalist predators due to their varied diet of
estuarine fish species (Parra and Jedensjö, 2014), and the focus of surveys in both the
ETS and PRE on areas of known dolphin occurrence to facilitate mark-recapture and line
transect abundance estimates are not the ideal circumstances of data collection for species
distribution modelling. Areas of true dolphin absence are infrequently surveyed if they
are visited at all, especially in the ETS, while the long-term monitoring program in Hong
Kong’s waters also includes other cetacean species found in the region, so waters where
humpback dolphins do not regularly occur are still surveyed frequently, thus providing
some reliable absences that were used in model selection.
Environmental predictors were considered to be scenopoetic (i.e. abiotic factors
that are not themselves affected by the presence or absence of dolphins, sensu Soberón,
2007) at the spatial and temporal scales examined in this chapter. However, occurrence
88
data are inherently influenced by biotic factors in the habitat (e.g. competition, predation;
Kearney, 2006; Pearman et al., 2007), and thus the comparison here is between realized
niches of humpback dolphins. Overlap of realized niches may indicate that the
fundamental niches of ETS and PRE humpback dolphins are also very similar, but may
also be due to analogous biotic factors in each study area limiting the realized niches of
ETS and PRE humpback dolphins to similar subvolumes of environmental space, but not
reflecting differentiation of the rest of the fundamental niche.
The observed overlap of realized niches may be due to niche conservatism if this
overlap extends to similarities in fundamental niches between TWDs and CWDs of the
PRE. Niche conservatism is the tendency of related species to maintain similar ecological
niches, with differentiation becoming evident over evolutionary timescales – generally
tens to hundreds of thousands of years (Peterson, 2011; Peterson et al., 1999; Wiens and
Graham, 2005). It appears that, if the observed niche overlap does in fact reflect their
fundamental niches, sufficient time has not elapsed for significant niche differentiation to
occur between TWDs and CWDs. The absence of selection pressures of sufficient
magnitude to shift the dimensions or location of the fundamental niches of each group of
humpback dolphins in environmental space is likely also a factor. Although some
differences in environmental characteristics between the PRE and ETS were observed,
the functional processes measured in estuaries even in very different parts of the world
are often remarkably similar (Baird and Ulanowicz, 1993). Environmental conditions
between the PRE and ETS differ most noticeably in productivity, temperature and
turbidity, but niches constructed in environmental space indicate that suitable conditions
for humpback dolphins in each study area extend beyond the conditions where dolphins
89
have been observed. Visualizing niches in environmental space provides a useful
illustration of niche overlap across a wide and continuous range of environmental
variables, however it should be noted that some of the combinations of environmental
predictors included in an n-dimensional hyperspace may not exist in the real-world
environment in which species are found (Warren et al., 2008). Predictions made in these
parts of environmental space are extrapolations that may not reflect the true relationships
between a species and its environment, and could account for the differences in
visualized niche dimensions between ETS and PRE humpback dolphins. However, there
was still overlap between niche ellipsoids, suggesting that the environmental conditions
examined in this chapter between the two study areas are not dissimilar enough for the
niches to diverge significantly, or that the biotic factors inherently affecting the dolphin
occurrence data have limited their realized niches to overlapping dimensions in
environmental space.
Biotic factors that affect species distributions include predation and competition
for resources (Hutchinson, 1958; Soberón, 2007). Indo-Pacific humpback dolphins have
few marine predators in both the PRE and ETS. Non-human predators of humpback
dolphins and other small cetaceans in other parts of the world include some large shark
species and marine mammal-eating cetaceans such as killer whales (Orcinus orca;
Ballance, 2009; Cockcroft, 1991; Corkeron, 1990). Some of these species have been
recorded in Chinese waters in the past (Zhou et al., 1995; Parsons, 1998), and injuries that
may have been a result of attempted predation have been observed on TWDs (Wang et
al., 2017) and on finless porpoises (Neophocaena phocaenoides), another small cetacean
species found in the PRE (Parsons and Jefferson, 2000), but detailed information on
90
predator distribution or abundance is not available in either study area. Direct takes of
humpback dolphins in either study area are rare, however they are frequently injured or
killed as fisheries bycatch in both study areas (Karczmarski et al., 2016a; Wang et al.,
2017). Humpback dolphins also compete with humans for prey resources (i.e. fisheries)
and space in their habitats, as extensive land reclamation and other marine construction
has taken place in both study areas (Wong, 2017; Karczmarski et al., 2016b). There are
no records of other potential marine competitors in the inshore waters of the ETS study
area, though they may be present further offshore (Zhou et al., 1995). At least two other
cetacean species have been observed in the PRE: finless porpoises are considered
residents in Hong Kong’s southeastern, and more oceanic, waters (Parsons, 1998), and
bottlenose dolphins (Tursiops truncatus) are occasionally sighted in offshore waters
across the estuary (SCSFRI, 2013; Zhou et al., 1995). Either of these species may be
considered competitors of CWDs as their presence may limit humpback dolphins’
realized niche in the PRE.
Apparently similar biotic factors exist in both study areas that might influence the
realized niche of each group of humpback dolphins, however spatial data to quantify
these factors are insufficient or non-existent in most cases. Biotic factors influence
species’ distributions on finer spatial and temporal scales than abiotic factors (Pearson
and Dawson, 2003; Soberón, 2007), and, according to the Eltonian niche hypotheses, the
effects of biotic interactions may be captured by coarse-resolution climatic variables such
as those used as predictors in this chapter, or may not affect distributions at all at large
extents (Soberón and Nakamura, 2009).
91
A final potential caveat to the observation of significant niche overlap of
humpback dolphins in the PRE and ETS is that these results are limited to the
environmental variables included as predictors in this chapter. These variables have been
identified as important predictors of humpback dolphin density in other studies (Hung,
2008; Or, 2017; Wong, 2017), and are frequently-used predictors of marine mammal
distribution in general (Redfern et al., 2006). Other potentially useful environmental
factors exist (e.g. seabed slope, salinity) that might be important predictors of humpback
dolphin presence, but could not be included as data on the same spatial and temporal
scales as the occurrence data used were not available. Conclusions of niche divergence or
conservatism depend on the selection of predictor variables to some degree, as niche
separation is more likely in high-dimensional environmental space, and less likely in
fewer dimensions (Peterson, 2011). Thus, it is possible that inclusion of additional
predictor variables could separate the compared niches, though there is no definitive rule
on the number of appropriate predictors to use to strike a balance in dimensionality of the
niche comparison (Peterson, 2011).
All of the environmental variables included in these models were important
predictors of probability of dolphin presence in some models for each study area, with no
single variable emerging as a definitive “most important” predictor. The emphasis in this
chapter is on predictive performance of models rather than explanatory power, thus less
parsimonious models which best predicted dolphin presences were favoured in model
selection to facilitate making predictions in unsurveyed waters in each study area, likely
contributing to mixed variable importance. Predictor variables were also ranked by
importance in each model and the mean taken for each variable across models for each
92
dataset, so the mean variable importance does not necessarily reflect the magnitude of
each variable’s importance
Each of the environmental predictors included in these models has been linked to
cetacean distributions in other studies, and are frequently used as predictors in models of
cetacean distribution (Kaschner et al., 2006; Redfern et al., 2006). Net primary
productivity was included as a proxy for prey abundance, however depth, distance to
shore, turbidity and sea surface temperature may also indirectly affect prey abundance
and distribution and the ability of dolphins to catch their prey. High net primary
productivity of estuaries is often reasoned to be due to the deposition of nutrient-rich
sediment from rivers into the mixing area between fresh- and oceanic waters.
Eutrophication from terrestrial inputs may not be directly responsible for high estuarine
biomass (Nixon et al., 1986), however positive correlations between primary productivity
and fish abundance have been reported in several estuarine ecosystems (e.g. Chassot et
al., 2007; Mallin and Paerl, 1994; Yáñez-Arancibia et al., 1993). Humpback dolphins are
generalist predators that feed upon benthic and pelagic fish species that spend at least part
of their life cycle within estuaries (Barros et al., 2004), thus dolphins may benefit from
increased energy availability at lower trophic levels in high productivity waters due to
their varied prey.
The importance of depth and distance to shore as predictors of humpback dolphin
probability of presence are likely also related to prey distributions, as many humpback
dolphin prey species are generally found in shallow, coastal waters (Barros et al., 2004)
The distribution and abundance of many pinniped (Burns et al., 2004; Mizuno et al.,
2002) and cetacean species (Cañadas et al., 2002; Hooker et al., 2002) have been linked
93
to depth, distance to shore and other bathymetric features, with most studies concluding
these patterns are related to prey availability. Hung (2008) found that CWDs in Hong
Kong’s waters preferred deeper waters and steeper benthic slopes, while recent work by
Or (2017) linked distance to shore with foraging activity across the PRE, finding that
foraging decreased in offshore waters.
Turbidity may serve as a proxy for salinity and could again be related to foraging.
Salinity could not be included as a predictor in this study due to limited availability of
remotely sensed salinity data for the duration of the study period. However, turbidity
maxima are often associated with freshwater-saltwater interfaces as oceanic and fresh
water mix and circulate, resulting in more turbid waters where this process occurs (Festa
and Hansen, 1978; Uncles and Stephens, 1993). The abundance of some species of
mesozooplankton were more affected by environmental characteristics in these stratified
layers in the PRE (Tan et al., 2004), with higher zooplankton abundance in the wet
season (Huang et al., 2004), providing prey for lower-trophic level species in the
estuarine food web, which are in turn preyed upon by piscivorous fish and dolphins (Pan
et al., 2016). Hung (2008) also found that turbidity was significantly correlated with
monthly CWD densities in Hong Kong’s waters. In the larger context of the PRE, Hong
Kong’s waters have relatively low turbidity due to the anti-clockwise circulation of water
through the estuary, which causes the eastern side (including the waters of western Hong
Kong) to be more influenced by clearer, saline oceanic waters, while the western side of
the estuary is influenced by the turbid freshwater flow from the river (Chen et al., 2004).
Monthly fluctuations in turbidity and other hydrological characteristics as investigated by
Hung (2008) are likely reflected in the response of CWDs to finer-scale shifts in prey
94
movement, while the seasonal means taken across multiple years in this chapter likely
show more general patterns of prey distribution.
The coarse temporal resolution likely also affected the importance of sea surface
temperature on probability of humpback dolphin presence. Monthly sea surface
temperature influences prey abundance and biodiversity in at least one estuary in Taiwan
(Tzeng and Wang, 1992), and although distribution and abundance data for fish species
were unavailable for the PRE, temperatures have been observed to influence zooplankton
abundance and biomass (Li et al., 2006; Tan et al., 2004), thus our data may not reflect
the finer-scale patterns that affect prey distributions, and thus dolphin distribution. Other
ephemeral weather phenomena like eddies and temperature fronts, which have been
linked to marine mammal distribution patterns (Doniol-Valcroze et al., 2007; Mendes et
al., 2002; Praca et al., 2009; Selzer and Payne, 1988), are also likely not reflected in the
environmental data used in these analyses.
Based on model predictions, CWDs likely range further west of the PRE than the
area that has currently been surveyed. Chen et al. (2010b) surveyed the PRE outside of
Hong Kong’s waters as far west as Shangchuan Island, and although the number of CWD
sightings decreased as surveys approached Shangchuan Island, the authors of that study
mentioned anecdotal reports of dolphins sighted near Hailing Island, ~60 km west of
surveyed waters. A later study extended the survey area in the western PRE to include
waters around Xiachuan Island, sighting dolphins on the eastern side of Hailing Island
during exploratory surveys (SCSFRI and HKCRP, 2011). This area is approximately the
boundary of the predicted suitable habitat for PRE humpback dolphins in the wet season,
and suggests humpback dolphins occur at least this far west of the PRE, potentially as
95
one contiguous population. Predictions in the dry season suggest continuous suitable
habitat spans the gap from Hailing Island to Leizhou Bay, which is home to the second-
largest known population of S. c. chinensis (Xu et al., 2015). Differences in seasonal
predictions of suitable habitat may be due to lower rainfall and subsequent decrease in
freshwater influx from the outlets of the Pearl River during the dry season. Seasonal
shifts in density have been observed in Hong Kong’s waters, with CWDs in this part of
the habitat moving further offshore during the wet season, presumably in response to
increased freshwater flow and subsequent shift of prey species distribution (Hung, 2008).
Studies to the west of the estuary have recorded the inverse of this pattern – in the wet
season, CWDs are found further inshore, moving further offshore during the dry season
(SCSFRI and HKCRP, 2011), and have quantified similar shifts in in prey density
(SCSFRI, 2013). Decreased freshwater influence during the dry season may result in
more uniform environmental conditions across the entire study area in comparison to the
wet season. CWDs may therefore travel farther afield to forage during the dry season,
while in the wet season they may take advantage of the highly productive waters of
Lingding Bay and western Hong Kong for foraging (Or, 2017). Hung and Jefferson
(2004) found seasonal shifts in home ranges of some humpback dolphins in Hong Kong’s
waters, and Chen et al. (2010b) suggested that dolphins first observed to the eastern and
western extremes of the habitat mostly intermingle during the height of the wet season.
However, analysis of more individuals and across the entire PRE are necessary to further
investigate how individual patterns of habitat use differ between wet and dry seasons.
Continuous areas of suitable habitat may exist between the PRE and south of Leizhou
Bay. Xu et al. (2015) reported (citing their own unpublished data) that humpback
96
dolphins have not been sighted north of Jianjiang River, however further surveys between
Shangchuan Island and Jianjiang River and exchange of photo-identification catalogues
between researchers in both areas should be conducted to determine whether there is
connectivity and exchange of individuals between these two CWD populations.
In contrast, the extent of predicted suitable habitat for the TWD has been
surveyed. Huang et al. (2018) produced similar maps of current humpback dolphin
distribution using maximum entropy models with a limited number of dolphin sightings
from a restricted part of the survey area, and predicted that even the “likely habitat
maximum” for TWDs in recent years does not extend further south than waters that were
surveyed and densities described by Dares et al. (2017). Dolphin sightings have
occasionally been made as far north as the Jhonggiang River estuary (Wang et al., 2016)
which was not classified as suitable habitat in this chapter based on overall ensemble
predictions. However, some models did predict dolphin presences in these areas, as far
north as the Danshui River as well as further south of the Waisanding Zhou sandbar.
Predicted suitable habitat for TWDs was discontinuous in some areas of known
dolphin distribution, which may have a methodological or biological explanation. When
constructing ensemble predictions, a threshold number of models was chosen to classify
habitat as suitable or unsuitable; this threshold was selected to maximize the sum of
sensitivity and specificity, and was expected to result in a small number of
misclassifications. Also, the temporal resolution of the environmental data used in this
chapter reflects the average conditions in each grid cell, and thus the mean probability of
dolphin occurrence in that grid cell. Conditions in the ETS are ephemeral at finer
temporal resolution, with spatial patterns in environmental variables changing drastically
97
from day to day in the wet season due to the large influx of freshwater from heavy
rainfall. Thus, dolphins may occasionally be sighted in areas predicted to be unsuitable
based on the average conditions in a grid cell when extreme rainfall or other factors
create temporarily suitable conditions. Also, occurrence records in predicted unsuitable
habitat could be a result of dolphins using less ideal areas in the absence of high quality
habitat in their restricted range (Dares et al., 2017). Hutchinson’s niche definition
specifies that all points within the niche are those which the species in question can
persist indefinitely (Hutchinson, 1957), thus the predicted “absence” areas may constitute
portions of the habitat where conditions have deteriorated to the point that dolphins
would be unable to survive if exclusively using those areas, but are still sighted (albeit at
low densities) as they travel through these areas from one portion of suitable habitat to
another.
In conclusion, CWDs of the PRE and TWDs occupy similar environmental
niches, despite their taxonomic differences. Conducting similar analyses for other
putative populations of CWD in China could be used to compare habitat requirements
among these groups, and predictions could identify contiguous areas of suitable habitat
between them. In the ETS, a larger dataset that spans a wider range of environmental
characteristics and includes some occurrence records north of the Daan River as well as
areas of true absence of humpback dolphins in Taiwan could improve geographic
predictions and more precisely model the relationships between TWDs and their habitats.
Also, incorporation of biotic factors, including anthropogenic impacts, in both areas
could potentially improve model predictions and facilitate the application of SDMs to
conservation and marine spatial planning in both the PRE and ETS.
98
4.5 – References Adler, D., Murdoch, D., Nenadic, O., Urbanek, S., Chen, M., Gebhardt, A., Bolker, B.,
Csardi, G., Strzelecki, A., Senger, A., Eddelbuettel, D. 2018. rgl: 3D Visualization
using openGL. R package version 0.99.9. <https://CRAN.R-
project.org/package=rgl>
Aiello-Lammens, M.E., Boria, R.A., Radosavljevic, A., Vilela, B., Anderson, R.P. 2015.
spThin: An R package for spatial thinning of species occurrence records for use in
ecological niche models. Ecography. 38, 541-545. DOI:10.1111/ecog.01132
Allouche, O., Tsoar, A., Kadmon, R. 2006. Assessing the accuracy of species distribution
models: Prevalence, kappa and the true skill statistic (TSS). J. Appl. Ecol. 43,
1223–1232. DOI:10.1111/j.1365-2664.2006.01214.x
Araújo, M.B., New, M. 2007. Ensemble forecasting of species distributions. Trends Ecol.
Evol. 22, 42–47. DOI:10.1016/j.tree.2006.09.010
Araújo, M.B., Peterson, A.T. 2012. Uses and misuses of bioclimatic envelope modeling.
Ecology. 93, 1527-1539.
Baird, D., Ulanowicz, R.E. 1993. Comparative study on the trophic structure, cycling and
ecosystem properties of four estuaries. Mar. Ecol. Prog. Ser.
DOI:10.3354/meps099221
Baith, K., Lindsay, R., Fu, G., McClain, C.R. 2001. SeaDAS, a data analysis system for
ocean-color satellite sensors. EOS Trans. AGU, 82. pg 202
Ballance, L. 2009. Cetacean Ecology. In: Perrin, W.F., Würsig, B., Thewissen, J.G.M.
(Eds.), Encyclopedia of Marine Mammals, 2nd Edition. Academic Press,
Burlington MA. pp. 196-201.
Barbosa, A.M. 2015. fuzzySim: Applying fuzzy logic to binary similarity indices in
ecology. Meth. Ecol. Evol. 6, 853-858. DOI: 10.1111/2041-210X.12372
Barros, N.B., Jefferson, T.A., Parsons, E.C.M. 2004. Feeding habits of Indo-Pacific
humpback dolphins (Sousa chinensis) stranded in Hong Kong. Aquat. Mamm. 30,
179–188. DOI:10.1578/AM.30.1.2004.179
Behrenfeld, M.J., Falkowski, P.G. 1997. Photosynthetic rates derived from satellite-based
chlorophyll concentration. Limnol. Oceanogr. 42, 1–20.
DOI:10.4319/lo.1997.42.1.0001
Boria, R.A., Olson, L.E., Goodman, S.M., Anderson, R.P. 2014. Spatial filtering to
reduce sampling bias can improve the performance of ecological niche models.
Ecol. Modell. 275, 73-77
Breiman, L. 2001. Random forests. Mach. Learn. 45, 5–32.
99
Brotons, L., Thuiller, W., Araujo, M.B., Hirzel, A.H. 2004. Presence-absence versus
presence-only modelling methods for predicting bird habitat suitability. Ecography.
27, 437–448.
Buckland, S.T., Anderson, D.R., Burnham, K.P., Laake, J.L., Borchers, D.L., Thomas, L.
2001. Introduction to distance sampling: Estimating abundance of biological
populations. Oxford University Press, NewYork.
Burns, J.M., Costa, D.P., Fedak, M.A., Hindell, M.A., Bradshaw, C.J.A., Gales, N.J.,
McDonald, B., Trumble, S.J., Crocker, D.E. 2004. Winter habitat use and foraging
behavior of crabeater seals along the western Antarctic Peninsula. Deep. Res. Part
II Top. Stud. Oceanogr. 51, 2279–2303. DOI:10.1016/j.dsr2.2004.07.021
Cañadas, A., Sagarminaga, R., García-Tiscar, S. 2002. Cetacean distribution related with
depth and slope in the Mediterranean waters off southern Spain. Deep. Res. Part I
Oceanogr. Res. Pap. 49, 2053–2073. DOI:10.1016/S0967-0637(02)00123-1
Chassot, E., Mélin, F., Le Pape, O., Gascuel, D. 2007. Bottom-up control regulates
fisheries production at the scale of eco-regions in European seas. Mar. Ecol. Prog.
Ser. 343, 45–55. DOI:10.3354/meps06919
Chen, T.C., Yen, M.C., Hsieh, J.C., Arritt, R.W. 1999. Diurnal and seasonal variations of
the rainfall measured by the automatic rainfall telemetry system in Taiwan. Bull.
Am. Meteorol. Soc. 80, 2299–2312.
Chen, Z., Li, Y., Pan, J. 2004. Distributions of colored dissolved organic matter and
dissolved organic carbon in the Pearl River Estuary, China. Cont. Shelf Res. 24,
1845–1856. DOI:10.1016/j.csr.2004.06.011
Chen, B., Zheng, D., Yang, G., Xu, X., Zhou, K. 2009. Distribution and conservation of
the Indo-Pacific humpback dolphin in China. Integr. Zool. 4, 240-247.
DOI:10.1111/j.1749-4877.2009.00160.x
Chen, L., Caballero, S., Zhou, K., Yang, G. 2010a. Molecular phylogenetics and
population structure of Sousa chinensis in Chinese waters inferred from
mitochondrial control region sequences. Biochem. Syst. Ecol. 38, 897-905. DOI:
10.1016/j.bse.2010.09.009
Chen, T., Hung, S.K.-Y., Qiu, Y., Jia, X., Jefferson, T.A. 2010b. Distribution, abundance,
and individual movements of Indo-Pacific humpback dolphins (Sousa chinensis) in
the Pearl River Estuary, China. Mammalia 74, 117–125.
DOI:10.1515/MAMM.2010.024
Chen, B., D. Zheng, J. Ju, X. Xu, K. Zhou, and G. Yang. 2011. Range patterns of resident
Indo-Pacific Humpback dolphins (Sousa chinensis, Osbeck 1765) in Xiamen,
China: implications for conservation and management. Zool. Stud. 50:751–762.
100
Chen, B., Zheng, D., Wang, L., Xu, X., Yang, G. 2012. The northernmost distribution of
Indo-Pacific humpback dolphin (Sousa chinensis) in the world: evidence from
preliminary survey in Ningde, China. Pakistan J. Zool.. 44:1209–1214.
Chen, B., Xu, X., Jefferson, T.A., Olson, P.A., Qin, Q., Zhang, H., Yang, G. 2016.
Conservation status of the Indo-Pacific humpback dolphin (Sousa chinensis) in the
northern Beibu Gulf, China. Adv. Mar. Biol. 73:119–139.
DOI:10.1016/bs.amb.2015.10.001
Chen, B., Jefferson, T.A., Wang, L., Gao, H., Zhang, H., Zhou, Y., Xu, X., Yang, G.
2018. Geographic variation in pigmentation patterns of Indo-Pacific humpback
dolphins (Sousa chinensis) in Chinese waters. J. Mammal. 99, 915-922.
DOI:10.1093/jmammal/gyy068
Cockcroft, C.G. 1991. Incidence of shark bites on Indian Ocean humpback dolphins
(Sousa plumbea) off Natal, South Africa. United Nations Environment Program,
Marine Mammal Technical Report. 3, 277-282.
Corkeron, P.J. 1990. Aspects of the behavioural ecology of inshore dolphins Tursiops
truncatus and Sousa chinensis in Moreton Bay, Australia. In: Leatherwood, S. and
Reeves, R.R. (eds.), The Bottlenose Dolphins. pp. 285-293. Academic Press,
London, United Kingdom.
Dale, M.R.T., Fortin, M.J. 2002. Spatial autocorrelation and statistical tests in ecology.
Ecoscience. 9, 162–167. DOI:10.1080/11956860.2002.11682702
Dares, L.E., Hoffman, J.M., Yang, S.-C., Wang, J.Y. 2014. Habitat characteristics of the
critically endangered Taiwanese Humpback Dolphins (Sousa chinensis) of the
Eastern Taiwan Strait. Aquat. Mamm. 40. DOI:10.1578/AM.40.4.2014.368
Dares, L.E., Araújo-Wang, C., Yang, S.-C., Wang, J.Y. 2017. Spatiotemporal
heterogeneity in densities of the Taiwanese humpback dolphin (Sousa chinensis
taiwanensis). Estuar. Coast. Shelf Sci. 187. DOI:10.1016/j.ecss.2016.12.020
de Oliveira, G., Rangel, T.F., Lima-Ribeiro, M.S., Terribile, L.C., Diniz-Filho, J.A.F.
2014. Evaluating, partitioning, and mapping the spatial autocorrelation component
in ecological niche modeling: A new approach based on environmentally
equidistant records. Ecography. 37, 637–647. DOI:10.1111/j.1600-
0587.2013.00564.x
Doniol-Valcroze, T., Berteaux, D., Larouche, P., Sears, R. 2007. Influence of thermal
fronts on habitat selection by four rorqual whale species in the Gulf of St.
Lawrence. Mar. Ecol. Prog. Ser. 335, 207–216. DOI:10.3354/meps335207
Dormann, C.F. 2007. Effects of incorporating spatial autocorrelation into the analysis of
species distribution data. Glob. Ecol. Biogeogr. 16, 129–138. DOI:10.1111/j.1466-
8238.2006.00279.x
101
Dormann, C.F., McPherson, J.M., Araújo, M.B., Bivand, R., Bolliger, J., Carl, G.,
Davies, R.G., Hirzel, A., Jetz, W., Kissling, W.D., Kühn, I., Ohlemüller, R., Peres-
Neto, P.R., Reineking, B., Schurr, F.M., Wilson, R. 2007. Methods to account for
spatial autocorrelation in the analysis of species distributional data: A review.
Ecography. 30: 609-628. DOI: 10.1111/j.2007.0709-7590.05171.x
Dormann, C.F., Elith, J., Bacher, S., Buchmann, C., Carl, G., Carré, G., Marquéz, J.R.G.,
Gruber, B., Lafourcade, B., Leitão, P.J., Münkemüller, T., Mcclean, C., Osborne,
P.E., Reineking, B., Schröder, B., Skidmore, A.K., Zurell, D., Lautenbach, S. 2013.
Collinearity: A review of methods to deal with it and a simulation study evaluating
their performance. Ecography. 36, 027–046. DOI:10.1111/j.1600-
0587.2012.07348.x
Elith, J., Graham, C.H., Anderson, R.P., Dudík, M., Ferrier, S., Guisan, A., Hijmans, R.,
Huettmann, F., Leathwick, J., Lehmann, A., Li, J., Lohmann, L., Loiselle, B.,
Manion, G., Moritz, C., Nakamura, M., Nakazawa, Y., Overton, J.M., Peterson,
A.T., Phillips, S., Richardson, K., Scachetti-Pereira, R., Schapire, R., Soberón, J.,
Williams, S., Wisz, M., E. Zimmermann, N. 2006. Novel methods improve
prediction of species’ distributions from occurrence data. Ecography. 29, 129–151.
DOI:10.1111/j.2006.0906-7590.04596.x
Elith, J., Leathwick, J.R., Hastie, T. 2008. A working guide to boosted regression trees. J.
Anim. Ecol. 77, 802–813. DOI:10.1111/j.1365-2656.2008.01390.x
Elith, J., Leathwick, J.R. 2009. Species distribution models: Ecological explanation and
prediction across space and time. Annu. Rev. Ecol. Evol. Syst. 40, 677–697.
DOI:10.1146/annurev.ecolsys.110308.120159
Elton, C.S. 1927. Animal Ecology. The Macmillan Company, New York.
ESRI. 2016. ArcGIS Desktop: Release 10.4. Redlands, CA.
Festa, J.F., Hansen, D. V. 1978. Turbidity maxima in partially mixed estuaries: A two-
dimensional numerical model. Estuar. Coast. Mar. Sci. 7, 347–359.
DOI:10.1016/0302-3524(78)90087-7
Fielding, A.H., Bell, J.F. 1997. A review of methods for the assessment of prediction
errors in conservation presence/absence models. Environ. Conserv. 24, 38–49.
DOI:10.1017/S0376892997000088
Fourcade, Y., Engler, J.O., Rödder, D., Secondi, J. 2014. Mapping species distributions
with MAXENT using a geographically biased sample of presence data: A
performance assessment of methods for correcting sampling bias. PLoS ONE. 9,
e97122. DOI:10.1371/journal/pone.0097122
Friedman, J. 2001. Greedy function approximation: A gradient boosting machine. Ann.
Stat. 29, 1189–1232. DOI:10.1214/009053606000000795
102
Griffith, D.A. 2013. Establishing qualitative geographic sample size in the presence of
spatial Autocorrelation. Ann. Assoc. Am. Geogr. 103, 1107–1122.
DOI:10.1080/00045608.2013.776884
Grinnell, J. 1917. The niche-relationships of the California thrasher. Auk 34, 427–433.
Guisan, A., Edwards, T.C., Hastie, T. 2002. Generalized linear and generalized additive
models in studies of species distributions: Setting the scene. Ecol. Modell. 157, 89–
100. DOI:10.1016/S0304-3800(02)00204-1
Guisan, A., Thuiller, W. 2005. Predicting species distribution: offering more than simple
habitat models. Ecol. Lett. 8, 993–1009. DOI:10.1111/j.1461-0248.2005.00792.x
Hastie, T., Tibshirani, R., Friedman, J. 2001. The elements of statistical learning: Data
mining, inference and prediction. Springer-Verlag, New York.
Hernández, P.A., Graham, C.H., Master, L.L., Albert, D.L. 2006. The effect of sample
size and species characteristics on performance of different species distribution
modeling methods. Ecography. 29, 773–785. DOI:DOI 10.1111/j.0906-
7590.2006.04700.x
Hijmans, R.J., Elith, J. 2017. Species distribution modeling with R. R Vignette.
DOI:10.1016/S0550-3213(02)00216-X
Hijmans, R.J., Phillips, S., Leathwick, J., Elith, J. 2017. dismo: Species Distribution
Modeling. R package version 1.1-4. <http://CRAN.R-project.org/package=dismo>
Hooker, S.K., Whitehead, H., Gowans, S., Baird, R.W. 2002. Fluctuations in distribution
and patterns of individual range use of northern bottlenose whales. Mar. Ecol. Prog.
Ser. 225, 287–297. DOI:10.3354/meps225287
Huang, L., Jian, W., Song, X., Huang, X., Liu, S., Qian, P., Yin, K., Wu, M. 2004.
Species diversity and distribution for phytoplankton of the Pearl River estuary
during rainy and dry seasons. Mar. Pollut. Bull. 49, 588–96.
DOI:10.1016/j.marpolbul.2004.03.015
Huang, S.-L., Wang, C.-C., Yao, C.-J. 2018. Habitat protection actions for the Indo-
Pacific humpback dolphin: Baseline gaps, scopes, and resolutions for the
Taiwanese subspecies. Aquat. Conserv. Mar. Freshw. Ecosyst. 1–11.
DOI:10.1002/aqc.2875
Huh, C.-A., Chen, W., Hsu, F.-H., Su, C.-C., Chiu, J.-K., Lin, S., Liu, C.-S., Huang, B.-J.
2011. Modern (<100 years) sedimentation in the Taiwan Strait: Rates and source-
to-sink pathways elucidated from radionuclides and particle size distribution. Cont.
Shelf. Res. 31, 47-63. DOI:10.1016/j.csr.2010.11.002
Hung, S.K.-Y. 2008. Habitat use of Indo-pacific humpback dolphins (Sousa chinensis) in
Hong Kong. Ph.D. Thesis, University of Hong Kong p. 1-253.
DOI:10.5353/th_b4088776
103
Hung, S.K.-Y., Jefferson, T.A. 2004. Ranging patterns of Indo-Pacific humpback
dolphins (Sousa chinensis) in the Pearl River estuary, People’s Republic of China.
Aquat. Mamm. 30, 159–174. DOI:10.1578/AM.30.1.2004.159
Hutchinson, G.E. 1957. Concluding remarks. Cold Spring Harb. Symp. Quant. Biol. 22,
415–427. DOI:10.1101/SQB.1957.022.01.039
James, G., Witten, D., Hastie, T., Tibshirani, R. 2015. An introduction to statistical
learning with applications in R. Springer Texts in Statistics.
DOI:10.1016/j.peva.2007.06.006
Jefferson, T.A. 2000. Population biology of the Indo-Pacific humpback dolphin in Hong
Kong waters. Wildl. Monogr. 144, 1–65.
Jefferson, T.A., Smith, B.D. 2016. Re-assessment of the conservation status of the Indo-
Pacific humpback dolphin (Sousa chinensis) using the IUCN criteria. In: Jefferson,
T.A.; Curry, B.E. (Eds.). Humpback dolphins (Sousa spp.): Current status and
conservation, Part 2. Adv. Mar. Biol. 73, 1-26.
Jiménez-Valverde, A., Lobo, J.M. 2007. Threshold criteria for conversion of probability
of species presence to either–or presence–absence. Acta Oecologica 31, 361–369.
DOI:10.1016/j.actao.2007.02.001
Joe, W., Chern, C.-S. 1988. On the Kuroshio branch in the Taiwan Strait during
wintertime. Prog. Oceanog. 21, 469-491. DOI:10.1016/0079-6611(88)90022-5
Karczmarski, L., Huang, S.-L., Or, C.K.M., Gui, D., Chan, S.C.Y., Lin, W., Porter, L.,
Wong, W.-H., Zheng, R., Ho, Y.-W., Chui, S.Y.S., Tiongson, A.J.C., Mo, Y.,
Chang, W.-L., Kwok, J.H.W., Tang, R.W.K., Lee, A.T.L., Yiu, S.-W., Keith, M.,
Gailey, G., Wu, Y. 2016a. Humpback dolphins in Hong Kong and the Pearl River
Delta: Status, threats and conservation challenges. In: Jefferson, T.A.; Curry, B.E.
(Eds.). Humpback dolphins (Sousa spp.): Current status and conservation, Part 2.
Adv. Mar. Biol. 73, 27-64. DOI:10.1016/bs.amb.2015.09.003
Karczmarski, L., Huang, S.-L., Wong, W.-H., Chang, W.-L., Chan, S.C.Y., Keith, M.
2016b. Distribution of a coastal delphinid under the impact of long-term habitat
loss: Indo-Pacific humpback dolphins off Taiwan's west coast. Estuaries Coast. 40,
594-603. DOI:10.1007/s12237-016-0146-5
Kaschner, K., Watson, R., Trites, A., Pauly, D. 2006. Mapping world-wide distributions
of marine mammal species using a relative environmental suitability (RES) model.
Mar. Ecol. Prog. Ser. 316, 285–310.
Kearney, M. 2006. Habitat, environment and niche: what are we modelling? Oikos. 115,
186-191. DOI:10.1111/j.2006.0040-1299-14908.x
Legendre, P. 1993. Spatial autocorrelation: Trouble or New Paradigm? Ecology 74,
1659–1673.
104
Li, K.Z., Yin, J.Q., Huang, L.M., Tan, Y.H. 2006. Spatial and temporal variations of
mesozooplankton in the Pearl River estuary, China. Estuar. Coast. Shelf Sci. 67,
543–552. DOI:10.1016/j.ecss.2005.12.008
Li, S., Lin, M., Xu, X., Xing, L., Zhang, P., Gozlan, R.E., Huang, S.L., Wang, D. 2016.
First record of the Indo-Pacific humpback dolphins (Sousa chinensis) southwest of
Hainan Island, China. Mar. Biodivers. Rec. 9, 1-6. DOI:10.1186/s41200-016-0005-
x
Liaw, A., Wiener, M. 2002. Classification and regression by randomForest. R News. 2,
18-22.
Lin, W., Chang, L., Frère, C.H., Zhou, R., Chen, J., Chen, X., Wu, Y. 2012.
Differentiated or not? An assessment of current knowledge of genetic structure of
Sousa chinensis in China. J. Exp. Mar. Biol. Ecol. 416, 17-20.
DOI:10.1016/j.jembe.2012.02.002
Liu, C., White, M., Newell, G. 2011. Measuring and comparing the accuracy of species
distribution models with presence-absence data. Ecography. 34, 232–243.
DOI:10.1111/j.1600-0587.2010.06354.x
Lobo, J.M., Jiménez-Valverde, A., Real, R. 2008. AUC: A misleading measure of the
performance of predictive distribution models. Glob. Ecol. Biogeogr. 17, 145–151.
DOI:10.1111/j.1466-8238.2007.00358.x
Mallin, M.A., Paerl, H.W. 1994. Planktonic trophic transfer in an estuary: Seasonal, diel,
and community structure effects. Ecology 75, 2168–2184.
Manel, S., Williams, H.C., Ormerod, S.J. 2001. Evaluating presence-absence models in
ecology: The need to account for prevalence. J. Appl. Ecol. 38, 921–931.
Mao, Q., Shi, P., Yin, K., Gan, J., Qi, Y. 2004. Tides and tidal currents in the Pearl River
estuary. Cont. Shelf Res. 1797-1808. DOI:10.1016/j.csr.2004.06.008
McClish, D.K. 1989. Analyzing a portion of the ROC curve. Med. Decis. Mak. 9, 190–
195.
Mendes, S., Turrell, W., Lütkebohle, T., Thompson, P. 2002. Influence of the tidal cycle
and a tidal intrusion front on the spatio-temporal distribution of coastal bottlenose
dolphins. Mar. Ecol. Prog. Ser. 239, 221–229. DOI:10.3354/meps239221
Mendez, M., Jefferson, T.A., Kolokotronis, S.-O., Krützen, M., Parra, G.J., Collins, T.,
Minton, G., Baldwin, R., Berggren, P., Särnblad, A., Amir, O.A., Peddemors, V.M.,
Karczmarski, L., Guissamulo, A., Smith, B., Sutaria, D., Amato, G., Rosenbaum,
H.C. 2013. Integrating multiple lines of evidence to better understand the
evolutionary divergence of humpback dolphins along their entire distribution range:
A new dolphin species in Australian waters? Mol. Ecol. DOI:10.1111/mec.12535
105
Mizuno, A.W., Wada, A., Ishinazaka, T., Hattori, K., Watanabe, Y., Ohtaishi, N. 2002.
Distribution and abundance of spotted seals Phoca largha and ribbon seals Phoca
fasciata in the southern Sea of Okhotsk. Ecol. Res. 17, 79–96.
Nixon, S.W., Oviatt, C.A., Frithsen, J., Sullivan, B. 1986. Nutrients and the productivity
of estuarine and coastal marine ecosystems. J. Limnol. Soc. Sth. Afr 12, 43–71.
OBPG (Ocean Biology Processing Group). 2015a. MODIS-Aqua: Sea surface
temperature Monthly data, 4 km resolution. < https://oceancolor.gsfc.nasa.gov/l3/>
OBPG (Ocean Biology Processing Group). 2015b. MODIS-Aqua: Diffuse attenuation
coefficient at 490 nm Monthly data, 4 km resolution. <
https://oceancolor.gsfc.nasa.gov/l3/>
OSU. 2015. Ocean productivity, standard products, monthly data 1020 x 2160.
<http://www.science.oregonstate.edu/ocean.productivity/index.php>
Or, K.M. 2017. Socio-spatial ecology of Indo-Pacific humpback dolphins (Sousa
chinensis) in Hong Kong and the Pearl River estuary. Ph.D. Thesis, University of
Hong Kong. p. 1-226.
Pan, C.W., Chen, M.H., Chou, L.S., Lin, H.J.L. 2016. The trophic significance of the
Indo-Pacific humpback dolphin, Sousa chinensis, in Western Taiwan. PLoS One
11, 1–19. DOI:10.1371/journal.pone.0165283
Paradis, E., Claude, J., Strimmer, K. 2004. APE: Analysis of phylogenetics and evolution
in R language. Bioinformatics. 20, 289-290.
Parra, G.J., Jedensjö, M. 2014. Stomach contents of Australian snubfin (Orcaella
heinsohni) and Indo-Pacific humpback dolphins (Sousa chinensis). Mar. Mammal
Sci. 30, 1184–1198. DOI:10.1111/mms.12088
Parsons, E.C.M. 1998. The behaviour of Hong Kong's resident cetaceans: the Indo-
Pacific humpbacked dolphins and the finless porpoise. Aquat. Mamm. 24, 91-110.
Parsons, E.C.M., Jefferson, T.A. 2000. Post-mortem investigations on stranded dolphins
and porpoises from Hong Kong waters. J. Wildl. Dis. 36, 342-356.
Pearce, J., Ferrier, S. 2000. Evaluating the predictive performance of habitat models
developed using logistic regression. Ecol. Modell. 133, 225–245.
DOI:10.1016/S0304-3800(00)00322-7
Pearman, P.B., Guisan, A., Broennimann, O., Randin, C.F. 2007. Niche dynamics in
space and time. Trends in Ecology and Evolution. 23, 149-158.
DOI:10.1016/j.tree.2007.11.005
Pearson, R.G., Dawson, T.P. 2003. Predicting the impacts of climate change on the
distribution of species: Are bioclimate envelope models useful? Glob. Ecol.
Biogeog. 12, 361-371.
106
Pebesma, E.J. 2004. Multivariable geostatistics in S: the gstat package. Comput. Geosci.
30, 683-691.
Peterson, A.T. 2011. Ecological niche conservatism: A time-structured review of
evidence. J. Biogeogr. 38, 817–827. DOI:10.1111/j.1365-2699.2010.02456.x
Peterson, A.T., Soberón, J.M., Sanchez-Cordero, V. 1999. Conservatism of ecological
niches in evolutionary time. Science. 285, 1265–1268.
Peterson, A.T., Soberón, J. 2012. Species distribution modeling and ecological niche
modeling: Getting the concepts right. Braz. J. Nat. Con.. 10, 102-107.
DOI:10.4322/natcon.2012.019
Praca, E., Gannier, A., Das, K., Laran, S. 2009. Modelling the habitat suitability of
cetaceans: Example of the sperm whale in the northwestern Mediterranean Sea.
Deep. Res. Part I Oceanogr. Res. Pap. 56, 648–657. DOI:10.1016/j.dsr.2008.11.001
Qiao, H., Soberón, J., Peterson, A.T. 2015. No silver bullets in correlative ecological
niche modelling: Insights from testing among many potential algorithms for niche
estimation. Methods Ecol. Evol. 6, 1126–1136. DOI:10.1111/2041-210X.12397
R Core Team. 2016. R: A language and environment for statistical computing. R
Foundation for Statistical Computing, Vienna, Austria. <http://www.R-
project.org/>
Redfern, J., Ferguson, M., Becker, E. 2006. Techniques for cetacean–habitat modeling.
Mar. Ecol. Prog. Ser. 310, 271–295.
Roberts, D.R., Bahn, V., Ciuti, S., Boyce, M.S., Elith, J., Guillera-Arroita, G.,
Hauenstein, S., Lahoz-Monfort, J.J., Schröder, B., Thuiller, W., Warton, D.I.,
Wintle, B.A., Hartig, F., Dormann, C.F. 2017. Cross-validation strategies for data
with temporal, spatial, hierarchical, or phylogenetic structure. Ecography. 40, 913–
929. DOI:10.1111/ecog.02881
Schapire, R.E. 2003. The boosting approach to machine learning: An overview.
Nonlinear Estim. Classif. 171, 149–171. DOI:10.1.1.24.5565
Schoener, T.W. 1968. The anolis lizards of Bimini: Resource partitioning in a complex
fauna. Ecology. 49, 704-726.
Scripps Institute of Oceanography. 2015. Global topography data.
<https://topex.ucsd.edu/>
SCSFRI (South China Sea Fisheries Research Institute). 2013. Study on Indo-Pacific
humpback dolphin population dynamic and their prey availability in western Pearl
River Estuary. Report submitted to WWF Hong Kong. p. 1-75.
SCSFRI (South China Sea Fisheries Research Institute) and HKCRP (Hong Kong
Cetacean Research Project). 2011. Study on the population biology of Chinese
107
white dolphins (Sousa chinensis) in the western Pearl River estuary, with
exploratory surveys in the Moyangjiang River estuary. Report submitted to WWF
Hong Kong. p. 1-39.
Selzer, L.A., Payne, P.M. 1988. The distribution of white-sided (Lagenorhynchus acutus)
and common dolphins (Delphinus delphis) vs. environmental features of the
continental shelf of the Northeastern United States. Mar. Mammal Sci. 4, 141–153.
Smith, W.H.F., Sandwell, D.T. 1997. Global sea floor topography from satellite altimetry
and ship depth soundings. Science. 26, 1956-1962.
DOI:10.1126/science.277.5334.1956
Soberón, J. 2007. Grinnellian and Eltonian niches and geographic distributions of species.
Ecology Letters. 10, 1115-1123.
Soberón, J. 2010. Niche and area of distribution modelling: A population ecology
perspective. Ecography. 33: 159-167. DOI:10.1111/j.1600.0587.2009.06074.x
Soberón, J. Nakamura, M. 2009. Niches and distributional areas: Concepts, methods and
assumptions. Proc. Nat. Acad. Sci. U.S.A. 106, 19644-19650.
DOI:10.1073/pnas.0901637106
Soberón, J., Peterson, A.T. 2005. Interpretation of models of fundamental ecological
niches and species’ distributional areas. Biodiversity Informatics. 2, 1-10.
Stockwell, D.R.B., Peterson, A.T. 2002. Effects of sample size on accuracy of species
distribution models. Ecol. Modell. 148, 1–13.
Stumpf, R.P., Pennock, J.R. 1991. Remote estimation of the diffuse attenuation
coefficient in a moderately turbid estuary. Remote Sens. Environ. 38, 183–191.
DOI:10.1016/0034-4257(91)90088-N
Tan, Y., Huang, L., Chen, Q., Huang, X. 2004. Seasonal variation in zooplankton
composition and grazing impact on phytoplankton standing stock in the Pearl River
Estuary, China. Cont. Shelf Res. 24, 1949–1968. DOI:10.1016/j.csr.2004.06.018
Thuiller, W., Lafourcade, B., Engler, R., Araújo, M.B. 2009. BIOMOD - A platform for
ensemble forecasting of species distributions. Ecography. 32, 369–373.
DOI:10.1111/j.1600-0587.2008.05742.x
Tzeng, W., Wang, Y. 1992. Structure, composition and seasonal dynamics of the larval
and juvenile fish community in the mangrove estuary of Tanshui River, Taiwan.
Mar. Biol. 113, 481–490.
Uncles, R.J., Stephens, J.A. 1993. The freshwater-saltwater interface and its relationship
to the turbidity maximum in the Tamar estuary, United Kingdom. Estuaries. 16,
126–141. DOI:10.2307/1352770
108
van Proosdij, A.S.J., Sosef, M.S.M., Wieringa, J.J., Raes, N. 2016. Minimum required
number of specimen records to develop accurate species distribution models.
Ecography. 39, 542–552. DOI:10.1111/ecog.01509
Varela, S., Anderson, R.P., García-Valdés, R., Fernández-González, F. 2014.
Environmental filters reduce the effect of sampling bias and improve predictions of
ecological models. Ecography. 37, 1084-1091. DOI:10.1111/j.1600-
0587.2013.00441.x
Voris, H.K. 2000. Maps of Pleistocene sea levels in Southeast Asia: Shorelines, river
systems and time durations. J. Biogeogr. 27, 1153–1167.
Wang, J.Y., Hung, S.K., Yang, S.-C. 2004. Records of Indo-Pacific humpback dolphins,
Sousa chinensis (Osbeck, 1765), from the waters of western Taiwan. Aquat.
Mamm. 30, 189-196. DOI:10.1578/AM.30.1.2004.189
Wang, J.Y., Hung, S.K.-Y., Yang, S.-C., Jefferson, T.A., Secchi, E.R. 2008. Population
differences in the pigmentation of Indo-Pacific humpback dolphins, Sousa
chinensis, in Chinese waters. Mammalia 72, 302–308.
DOI:10.1515/MAMM.2008.030
Wang, J.Y., Yang, S.-C., Fruet, P.F., Daura-Jorge, F.G., Secchi, E.R. 2012. Mark-
recapture analysis of the critically endangered eastern Taiwan Strait population of
Indo-Pacific humpback dolphins (Sousa chinensis): Implications for conservation.
Bull. Mar. Sci. 88, 885–902.
Wang, J.Y., Yang, S.-C., Hung, S.K.-Y. 2015. Diagnosability and description of a new
subspecies of Indo-Pacific humpback dolphin, Sousa chinensis (Osbeck, 1765),
from the Taiwan strait. Zool. Stud. 54, 1–15. DOI:10.1186/s40555-015-0115-x
Wang, J.Y., Riehl, K.N., Klein, M.N., Javdan, S., Hoffman, J.M., Dungan, S.Z., Dares,
L.E., Araújo-Wang, C. 2016. Biology and conservation of the Taiwanese humpback
dolphins, Sousa chinensis taiwanensis. In: Jefferson, T.A.; Curry, B.E. (Eds.).
Humpback dolphins (Sousa spp.): Current status and conservation, Part 2. Adv.
Mar. Biol. 73, 91-117.
Wang, J.Y., Riehl, K.N.; Yang, S. -C., Araújo-Wang, C. 2017. Unsustainable human-
induced injuries to the Critically Endangered Taiwanese humpback dolphins (Sousa
chinensis taiwanensis). Mar. Pollut. Bull. 116, 167-174.
DOI:10.1016/j.marpolbul.2016.12.080
Wang, J.Y., Yang, S.C. 2011. Evidence for year-round occurrence of the eastern Taiwan
Strait Indo-Pacific humpback dolphins (Sousa chinensis) in the waters of western
Taiwan. Mar. Mammal Sci. 27, 652–658. DOI:10.1111/j.1748-7692.2010.00422.x
Warren, D.L., Glor, R.E., Turelli, M. 2008. Environmental niche equivalency versus
conservatism: Quantitative approaches to niche evolution. Evolution. 62, 2868–
2883. DOI:10.1111/j.1558-5646.2008.00482.x
109
Wei, T., Simko, V. 2017. R package "corrplot": Visualization of a correlation matrix
(Version 0.84). <http://github.com/taiyun/corrplot>
Wiens, J.J., Graham, C.H. 2005. Niche conservatism: Integrating evolution, ecology, and
conservation biology. Annu. Rev. Ecol. Evol. Syst. 36, 519–539.
DOI:10.1146/annurev.ecolsys.36.102803.095431
Wisz, M.S., Hijmans, R.J., Li, J., Peterson, A.T., Graham, C.H., Guisan, A., Elith, J.,
Dudík, M., Ferrier, S., Huettmann, F., Leathwick, J.R., Lehmann, A., Lohmann, L.,
Loiselle, B.A., Manion, G., Moritz, C., Nakamura, M., Nakazawa, Y., Overton,
J.M., Phillips, S.J., Richardson, K.S., Scachetti-Pereira, R., Schapire, R.E.,
Soberón, J.M., Williams, S.E., Zimmermann, N.E. 2008. Effects of sample size on
the performance of species distribution models. Divers. Distrib. 14, 763–773.
DOI:10.1111/j.1472-4642.2008.00482.x
Wong, W.H. 2017. Macro- and micro-scale anthropogenic impacts on Chinese white
dolphins in Hong Kong: Quantifying impacts of habitat loss and coastal tourism.
Ph.D. Thesis, University of Hong Kong. p. 1-253.
Wood, S.N. 2011. Fast stable restricted maximum likelihood and marginal likelihood
estimation of semiparametric generalized linear models. J. R. Stat. Soc. B. 73, 3-36.
Wu, Y. 2010. The survival status of Chinese white dolphin in Guangdong Province,
China. Ph.D. Thesis, Shantou University, Shantou, China.
Xu, X., Song, J., Zhang, Z., Li, P., Yang, G., Zhou, K. 2015. The world’s second largest
population of humpback dolphins in the waters of Zhanjiang deserves the highest
conservation priority. Sci. Rep. 5, 8147. DOI:10.1038/srep08147
Yáñez-Arancibia, A., Lara-Domínguez, A.L., Day, J.W. 1993. Interactions between
mangrove and seagrass habitats mediated by estuarine nekton assemblages:
coupling of primary and secondary production. Hydrobiologia 264, 1–12.
DOI:10.1007/BF00014659
Zhang, Q., Xu, C.Y., Zhang, Z. 2009. Observed changes of drought/wetness episodes in
the Pearl River basin, China, using the standardized precipitation index and aridity
index. Theor. Appl. Climatol. 98, 89–99. DOI:10.1007/s00704-008-0095-4
Zhou, K., Leatherwood, S.; Jefferson, T.A. 1995. Records of small cetaceans in Chinese
waters: A review. Asian Mar. Biol. 12, 119-139.
Zong, Y., Huang, G., Switzer, A.D., Yu, F., Yim, W.W.-S. An evolutionary model for the
Holocene formation of the Pearl River delta, China. Holocene. 19, 129-142.
DOI:10.1177/0959683608098957
110
4.6 – Tables and Figures
Figure 4.1 – Location of putative populations (bold text, study areas approximated in blue
for S. c. chinensis and red for S. c. taiwanensis) and additional sightings (regular text) of
S. chinensis in Chinese waters, and geographical context of the two study areas (a). Inset
maps show survey effort in square kilometres covered in each 1km2 grid between 2008-
2015 in the Pearl River Estuary during the wet (b) and dry (c) seasons, and the wet season
in the eastern Taiwan Strait (d). The area outlined in black in the PRE indicates the
political border of Hong Kong, within which the majority of research efforts in the PRE
are focused.
111
Figure 4.2 – Box and whisker plots of environmental variables in each study area, divided
by whether dolphin presences or absences were recorded in the same grid cell over the
study period. Summary statistics for each dataset are given in Appendix C.
112
Figure 4.3 – Mean (± SE) area under the ROC curve (AUC; partial AUC for ETS data;
panel a) and true skill statistic (TSS; panel b) of models included in the bounding box
ensemble for each study area, by algorithm and across thinned subsets. Models were
selected if they met the criteria of AUC > 0.7 and TSS > 0.5. Numbers above each bar
indicate the total number of models of each algorithm selected for each study area (BRT
= boosted regression tree, GAM = generalized additive models, GLM = generalized
linear models, RF = random forest).
113
Figure 4.4 – Mean (± SE) of predictor variable importance for turbidity, depth, survey
effort, net primary productivity, distance to shore and sea surface temperature. Predictors
were ranked from most (6) to least (1) important for predicting humpback dolphin
presence from models included in ensembles for each dataset. Predictors that did not
appear in a model were given a rank of 0. Error bars show the standard error of the
variable’s mean rank across models for that dataset, and the numbers above each bar
indicate the number of models for each dataset that included the predictor variable.
114
Table 4.1 – Results of randomization tests for niche similarity. Null distributions were generated by calculating niche overlap statistics
(Schoener’s D and Warren’s I) between mean predicted probability of presence made in study area A using the best models fit using
data from study area B and predictions made in study area A by models of the same algorithm type and predictors but fit using a
randomly selected dataset from study area A.
Prediction Data
(Study Area A)
Prediction Models
(Study Area B) D
Null distribution mean ± SD
(95% C.I.) I
Null distribution mean ± SD
(95% C.I.)
PRE
(dry season)
ETS
(wet season)
0.612** 0.865 ± 0.032
(0.859-0.871)
0.861** 0.985 ± 0.007
(0.983-0.986)
PRE
(wet season)
0.791ǂǂ 0.756 ± 0.072
(0.742-0.770)
0.950** 0.954 ± 0.023
(0.949-0.958)
PRE
(wet season)
ETS
(wet season)
0.607** 0.782 ± 0.036
(0.774-0.789)
0.887** 0.963 ± 0.012
(0.961-0.966)
PRE
(dry season)
0.645ǂǂ 0.505 ± 0.016
(0.483-0.528)
0.907ǂǂ 0.813 ± 0.005
(0.806-0.821)
ETS
(wet season)
PRE
(wet season)
0.713ǂ 0.705 ± 0.053
(0.645-0.715)
0.940ǂǂ 0.931 ± 0.025
(0.927-0.936)
PRE
(dry season)
0.650ǂǂ 0.547±0.084
(0.546-0.548)
0.914ǂǂ 0.846 ± 0.046
(0.845-0.846)
** Less model overlap than expected by chance (p << 0.0001) ǂ Greater model overlap than expected by chance (p < 0.005) ǂǂ Greater model overlap than expected by chance (p << 0.0001)
115
Figure 4.5 – Niche volumes of the PRE in the wet season (blue), PRE in the dry season
(light blue) and ETS in the wet season (red) in environmental space. Niche ellipsoids in
the turbidity, depth and productivity dimensions are shown in (a), and in the sea surface
temperature, distance to shore and productivity dimensions in (b) (N.B. productivity was
repeated in the second figure to allow visualization of only five environmental predictor
variables in three dimensions).
116
Figure 4.6 – Bounding box ensemble predictions in geographic space for PRE in the dry
season (a,b) PRE in the wet season (c,d) and ETS in the wet season (e,f). Left-hand side
shows the proportion of models out of the total number of best-predicting models
(nPREdry=621, nPREwet=115, nETS=47) that predicted a presence in each grid cell in PRE in
the dry season (a), PRE in the wet season (c) and ETS in the wet season (e). Right-hand
side shows these predictions converted to bounding box presence or absence based on the
threshold number of predicted presences that maximized the sum of sensitivity and
specificity for each study area (nPREdry=303, nPREwet=85, nETS=26).
117
Chapter 5 – General Conclusions
5.1 – Synthesis: Habitat Characteristics, Density Patterns and
Environmental Niches of Humpback Dolphins in the Pearl
River Estuary and Eastern Taiwan Strait
Defining the characteristics of a species’ habitat, quantifying a species’
relationship to its environment, and mapping density patterns can make important
contributions to the conservation of species, especially for coastal and estuarine animals
whose habitats are often in close proximity to, and frequently affected by, anthropogenic
impacts. The government-funded long-term monitoring program for CWDs in the waters
of Hong Kong has produced a wealth of detailed information on the life history,
abundance, habitat use, and density patterns of these animals, which researchers studying
other populations of humpback dolphins often draw upon when data specific to their
populations are unavailable. Comparatively less information was available for ETS
humpback dolphins, and the goals of the first two chapters of my thesis were to
characterize the habitat where these animals are found, comparing them to other
populations of humpback dolphins, and quantify their density patterns within the habitat.
In Chapter 2, we produced the first description of the habitat characteristics in the
ETS, reporting that Taiwanese humpback dolphins are found in nearshore, generally
brackish waters near estuaries, similar to other humpback dolphin populations in
Australia, South Africa and China, including the PRE. Observed group sizes were not
correlated with environmental data collected during sightings, but were similar to other
118
humpback dolphin populations examined. Singletons were less likely to be observed in
the ETS population in comparison to the PRE population, which may be due to the
considerably larger population size and habitat area of the PRE, or differences in social
structure between CWDs of the PRE and TWDs, as described by Dungan et al. (2016)
and Dungan et al. (2012), respectively. We noted that the habitat configuration of the
PRE differs somewhat from the linear distribution of the TWD and some other
populations of humpback dolphins, though all of the environmental characteristics
examined in Chapter 2 were comparable to published information on the PRE and other
humpback dolphin populations.
In Chapter 3, we quantified density patterns of the TWD, finding significant
spatial and temporal variation across the habitat. Densities varied across the study area
from year-to-year but were consistently higher in some parts of the habitat over time.
There were no significant relationships between density patterns and the environmental
variables assessed, but dolphin densities were higher than expected in waters adjacent to
large-scale land reclamation projects. This was most likely due to the placement of these
projects in close proximity to river mouths or creating artificial confluences of freshwater
from smaller rivers and streams in areas that were likely important to dolphins prior to
construction.
Having described the habitat characteristics and density patterns in Chapters 2 and
3, my objective for Chapter 4 was to conduct a large-scale, quantitative comparison of the
relationships between humpback dolphins in the PRE and ETS and their respective
habitats, and predict suitable habitat beyond currently surveyed areas. Focus of surveys in
each area on abundance estimation (via distance sampling in Hong Kong’s waters and
119
mark-recapture of photo identification data in Taiwan) rather than even survey coverage
of the entire range of each study area contributed to spatial autocorrelation in the data
available for modelling, necessitating the use of a method to address non-independence in
model residuals. To facilitate the use of a variety of different algorithms in Chapter 4, I
focused on data management rather than statistical accommodations for spatially
autocorrelated data. Several studies have thinned (=filtered) their occurrence data to
reduce spatial autocorrelation by removing occurrences that are within some distance of
one another (e.g. Anderson and Raza, 2010; Carroll, 2010; Veloz 2009), and others have
demonstrated the effectiveness of spatial filtering in improving model predictions
(Aiello-Lammens et al., 2015; Boria et al., 2014). However, filtering data in
environmental space has been shown more effective than spatial filtering (de Oliveira et
al., 2014; Varela et al., 2014), so I adapted the method described by de Oliveira et al.
(2014) to thin each dataset based on distances in environmental space to a subset of
points that have demonstrated no spatial autocorrelation in the residuals when modelled.
With the use of species distribution models (SDM), I found that, despite their
separation by the relatively deep waters of the Taiwan Strait for thousands of years, and
the recent taxonomic designation of the TWD as a separate subspecies of humpback
dolphin, these two groups occupied very similar niches. All candidate environmental
variables were important predictors of humpback dolphin probability of presence in
several best-performing models in each study area, possibly due to their indirect influence
on the distribution of the estuarine prey species of humpback dolphins, but the focus of
model selection on predictive ability rather than parsimony was also likely a factor.
Predictions for each study area in geographic space indicated that most of the distribution
120
of the TWD has likely been surveyed, and especially important areas in the waters of
central western Taiwan were highlighted as suitable habitat by ensemble model
predictions. In contrast, suitable habitat for CWDs in the PRE is predicted to extend
further west, beyond waters that have been surveyed to date, possibly as far as Leizhou
Bay, where another large population of CWD is located (Xu et al., 2015). Differences in
model predictions in wet and dry seasons in the PRE do not imply that two discrete
niches exist for this population depending on the season, but should rather be interpreted
as a gradual shift in important environmental variables, primarily sea surface temperature,
predicting dolphin presence as the seasons change.
5.2 – Conservation Implications and Recommendations
As coastal, estuarine cetaceans with habitats in close proximity to areas of very
high human population density, CWDs of the PRE and TWDs have several conservation
issues in common, including habitat loss and degradation, fisheries interactions,
pollution, anthropogenic noise and vessel traffic (e.g. Hung, 2008, 2017; Jefferson et al.,
2009; Wang et al., 2004, 2007, 2016). Despite the similarities between TWDs and CWDs
of the PRE discussed throughout this thesis and the anthropogenic impacts they both face,
conservation action for each population should be tailored to its specific geographic
context.
5.2.1 – Eastern Taiwan Strait
For the TWD, results in Chapter 4 indicate that the majority of the dolphins’
distribution has been surveyed, and both Chapters 3 and 4 highlight especially important
areas for these animals, indicated by areas of high dolphin density in Chapter 3, and
121
predicted presences in Chapter 4. Despite the assertions of Huang et al. (2018) that there
are insufficient baseline data available on habitat use and other threats to direct
conservation efforts for TWDs, Chapter 3 of this thesis (previously published as Dares et
al. 2017), as well as several other recent publications (e.g. Araújo et al., 2014; Ross et al.,
2010; Slooten et al., 2013; Wang et al., 2004, 2007, 2017; Wang and Araújo-Wang,
2017) have outlined habitat use patterns and the many threats facing TWDs, which must
be addressed in the conservation of this subspecies. The already small habitat area along
the coast of central western Taiwan has been heavily developed over the past half
century, and even degraded areas are likely being used by TWDs in the absence of better-
quality habitat (Chapter 3; Karczmarski et al., 2017), thus protection of the entire existing
habitat from further degradation and loss must be a priority for conservation. Previous
studies (Araújo et al., 2014; Slooten et al., 2013; Wang et al., 2017) have also shown the
impact of fisheries on the TWD, emphasizing that reduction of fisheries in dolphin
habitat in Taiwan is crucial to their survival. Other, less visible, impacts also remain to be
addressed, including pollution and habitat degradation from anthropogenic modifications
(e.g. artificial coastline to reduce erosion, diversion of freshwater from rivers for
residential and industrial use, etc.; Wang et al., 2016)
5.2.2 – Pearl River Estuary
Research efforts in Hong Kong’s waters have produced a wealth of information
on habitat use in this area, however it is only a small portion of the range of the PRE
population. Habitat loss and degradation, construction impacts, and vessel traffic are
pressing issues in Hong Kong due to the intensity of human activities concentrated in
these waters and their impacts on dolphin habitat use in this area, but my results in
122
Chapter 4 emphasize the need for an holistic approach to the conservation of the entire
PRE population. The lack of physical boundaries impeding dolphin movements between
Hong Kong’s waters and the waters of mainland China and Macau has been mentioned in
several studies (e.g. Chen et al., 2010; Hung and Jefferson, 2004; Jefferson et al., 2009;
Or, 2017), and analyses that include data collected in Lingding Bay and further west
indicate that habitat use patterns on the western side of the estuary are, in some cases, the
opposite of what has been observed in Hong Kong (e.g. CWD habitat use shifts further
south and offshore in Hong Kong’s waters during the wet season, while shifting closer to
shore in the western part of the PRE; see Chapter 4, Chen et al. 2010; SCSFRI and
HKCRP, 2011; SCSFRI 2013). Conclusions and recommendations based on studies
conducted in a small fraction of the habitat may therefore not reflect the reality and needs
of the entire population. More survey effort in Chinese waters, especially west of
Shangchuan Island is needed, as well as more regular surveys in Lingding Bay and other
parts of the range of PRE humpback dolphins outside of Hong Kong’s waters in order to
monitor the habitat use patterns of these dolphins across their full range.
The availability of photo-identification (photo-ID) data also allows the possibility
of individual-specific investigation of habitat use patterns for humpback dolphins.
General habitat use data (i.e. descriptions of population density patterns) over time can
provide insights into areas that are heavily used, but we are limited in our ability to break
down the patterns in these data into habitat use by particular demographic or other
subsets of the population when simply using field counts of dolphins observed during
surveys. Most Indo-Pacific humpback dolphins can be visually differentiated by spotting
patterns on their dorsal fins and other parts of the body in addition to nicks, notches and
123
scars (Jefferson, 2000; Wang et al., 2012), so researchers can identify individuals when
photographs are taken during sightings. Generating and maintaining a photo-ID database
is an enormous undertaking for a population of ~2500 individuals such as that of the
PRE, but photo-ID data have the potential to grant invaluable insights into habitat use of
this population. Some important applications include investigations into home ranges of
individuals found across the entire PRE study area, seasonal shifts in habitat use by
individuals in the western part of the PRE, and whether some areas are of particular
importance to vulnerable members of the population, such as nursing mothers and
juveniles. Further, exchange of photo-ID catalogues between researchers focusing on
populations of humpback dolphins in other parts of China (e.g. Leizhou Bay, Shantou,
JRE), could facilitate insights into whether these populations are truly separated from one
another in the absence of sufficient molecular data to make such conclusions.
There has been much debate over the efficacy of existing marine protected areas
(MPAs) in the PRE (Karczmarski and Or, 2016; Or, 2017; Wong, 2017). Or (2017) gave
a comprehensive description of existing MPAs in the eastern PRE: There are two existing
marine protected areas in Hong Kong aimed at protecting CWDs, both located in
northwestern waters; and two more proposed MPAs in the southwestern waters, which
are to be implemented by the end of 2018. Nearly the entirety of the northwestern waters
of Hong Kong are to be designated as an MPA in 2023, which will incorporate the two
existing marine parks, however protection measures will not be put into place until the
completion of a third runway at the Hong Kong International Airport, which is being
constructed within the proposed boundaries of this MPA. While these MPAs do
encompass some of the important areas recorded for humpback dolphins in Hong Kong’s
124
waters in the past (Hung 2008, 2013; Hung and Jefferson, 2004), shifts in dolphin habitat
use in recent years – hypothesized to be in response to construction activities in north
Lantau – have resulted in few sightings being recorded in these areas (e.g. Hung 2013,
2014, 2015, 2016, 2017). Further, protection measures and enforcement in these areas
may not be sufficient to ensure the conservation of humpback dolphins in Hong Kong’s
waters. The Guangdong Pearl River Estuary Chinese White Dolphin National Nature
Reserve is the only MPA for humpback dolphins in the PRE, which is divided into three
zones with varying levels of protection. Or (2017) proposed that offshore areas such as
those in the middle of Lingding Bay where this park is located are more likely used as
transit areas by dolphins as they move from foraging grounds on one side of the PRE to
the other, suggesting that protected areas might be better located in areas closer to shore
which are more likely to be used for foraging purposes.
The temporal resolution of my results in Chapter 4 may limit their utility in
designating MPAs in the PRE as there is little discrimination in habitat suitability (i.e.
number of models predicting dolphin presences) across the study area. Grid cells
predicted to be suitable for CWDs showed similar numbers of models predicting dolphin
presences, so it is difficult to rank certain areas by suitability, and likely not precise
enough to direct marine spatial planning in as complicated a situation as the PRE, which
is traversed by multiple shipping lanes, a bridge that includes an underground tunnel,
high-speed ferry routes and fishing vessels, and has been subject to several reclamation
projects and shoreline modification. Anthropogenic impacts such as vessel traffic, marine
construction and reclamation were not included in my analyses due to insufficient spatial
data for these impacts outside of Hong Kong’s waters so the predictions of suitable
125
habitat use in Chapter 4 are based entirely on environmental suitability. This may
overestimate humpback dolphin distribution in light of extensive anthropogenic impacts
in the PRE. What can be drawn from Chapter 4, however, is more support for the need
for cross-boundary conservation efforts and protection measures for CWDs of the PRE.
Suitable habitat for CWDs in the PRE spans the political borders of Hong Kong,
mainland China and Macau, and it is unlikely that the areas contained by the smaller
Special Administrative Regions of Hong Kong and Macau are sufficient to sustain more
than a very small percentage of the population. Thus, the conservation of PRE humpback
dolphins requires a concerted effort by stakeholders in all three jurisdictions to ensure
adequate protection across their entire range.
5.3 – Future directions
There are several avenues of methodological inquiry that arise from my work in this
thesis. SDM has exploded in popularity over the past decade, and several statistical
packages and courses are now available which have made these analyses more accessible
to a wider range of researchers (e.g. BIOMOD2, Thuiller et al., 2009; dismo, Hijmans et
al., 2017; etc.). Previous studies have demonstrated the importance of addressing spatial
autocorrelation in SDMs (Crase et al., 2014; Dormann, 2007; Segurado et al., 2006), and
still others have introduced new methods to account for it (Crase et al., 2012; Dormann et
al., 2007; Kissling and Carl, 2008; Radersma and Sheldon, 2015). However, many of
these methods, may be difficult to implement for non-statisticians, and are not currently
compatible with all SDM algorithms, so are largely unavailable in widely-used SDM
126
packages. The thinning procedure I introduced in Chapter 4, however, is widely
applicable as it addresses spatial autocorrelation via data management and selection
rather than compensating by statistical means. While this procedure is based on the
concepts introduced by de Oliveira et al. (2014), thinning of the dataset, rather than
selecting a single “effective sample size”, makes this method more flexible in situations
where an effective sample size based on equidistance of points in environmental space
might be too small to be modelled with many SDM algorithms. Instead, thinning the
dataset allows exploration of a variety of sample sizes as well as alternative subsets so
that there are sufficient data to be modelled, spatial autocorrelation is insignificant in
model residuals, and variation across subsets can be investigated.
In addition to the previously mentioned gaps in our knowledge still to be addressed
for the conservation of CWDs in the PRE and TWDs, there are several other paths of
inquiry arising from my work. Further investigation of the niches occupied by putative
populations of S. chinensis using similar methods to those implemented in Chapter 4
could reveal further insights into the species-habitat relationships of these estuarine small
cetaceans. Though well-studied in part of their range, PRE humpback dolphins are but
one population of the nominate subspecies, which encompasses a large number of
geographically widespread populations and whose taxonomy is, ultimately, still likely
unresolved (Mendez et al., 2013). Niche comparisons of other populations of S. c.
chinensis in Chinese waters could address the question of whether the PRE population is
an appropriate representative of the entire nominate subspecies. Comparisons with more
geographically distant populations such as those found in estuaries of the eastern Indian
Ocean that, by default, fall under the umbrella of S. c. chinensis but whose membership in
127
that group may be disputed (Mendez et al., 2013) could be conducted to investigate
geographic variation in habitat requirements. Further, comparisons with other species of
genus Sousa, such as the Atlantic or Australian species would indicate whether
interspecific niches differ enough to be quantifiable, or if all humpback dolphins have
relatively similar habitat requirements and speciation is due to geographic separation over
time. Applying SDMs to predicting distributions of humpback dolphins across larger
geographic areas (e.g. from coastal areas in northern China to India or Malaysia) would
also indicate discontinuous areas of suitable habitat, and could reveal areas in unsurveyed
waters that may be inhabited by previously undiscovered populations of humpback
dolphins, potentially further expanding the range of the species.
Overall, my thesis demonstrates that, although density patterns may be variable from
year to year, on a broad temporal scale, the two allopatric groups of Indo-Pacific
humpback dolphins included in my analyses have similar habitat requirements. General
descriptions of their habitats indicated that both CWDs of the PRE and the TWDs of the
ETS inhabit areas that are environmentally similar. As with previously published reports
of CWD density and habitat use in the PRE, TWD density patterns are spatiotemporally
heterogeneous as dolphins shift their use of the habitat from year to year in response to
ephemeral environmental conditions in the ETS. Finally, niche comparisons indicate that
these two groups of humpback dolphin occupy similar realized niches, likely due to the
similarity in biotic and abiotic conditions in their geographically separate ranges,
however geographic variation in these patterns across other populations and species of
humpback dolphins is still to be investigated.
128
5.4 - References Aiello-Lammens, M.E., Boria, R.A., Radosavljevic, A., Vilela, B., Anderson, R.P. 2015.
spThin: an R package for spatial thinning of species occurrence records for use in
ecological niche models. Ecography. 38, 541-545. DOI: 10.1111/ecog.01132
Anderson, R.P., Raza, A. 2010. The effect of the extent of the study region on GIS
models of species geographic distributions and estimates of niche evolution:
preliminary tests with montane rodents (genus Nephelomys) in Venezuela. J.
Biogeo. 37, 1378-1393.
Araújo, C.C., Wang, J.Y., Hung, S.K.-Y., White, B.N., Brito, D. 2014. Viability of the
Critically Endangered eastern Taiwan Strait population of Indo-Pacific humpback
dolphins Sousa chinensis. Endanger. Species Res. 24, 263–271.
DOI:10.3354/esr00605
Boria, R.A., Olson, L.E., Goodman, S.M., Anderson, R.P. 2014. Spatial filtering to
reduce sampling bias can improve the performance of ecological niche models.
Ecol. Modell. 275, 73-77
Carroll, C. 2010. Role of climatic niche models in focal-species-based conservation
planning: Assessing potential effects of climate change on Northern Spotted Owl in
the Pacific Northwest, USA. Biol. Cons. 143, 1432-1437.
DOI:10.1016/j.biocon.2010.03018
Chen, T., Hung, S.K.-Y., Qiu, Y., Jia, X., Jefferson, T.A. 2010. Distribution, abundance,
and individual movements of Indo-Pacific humpback dolphins (Sousa chinensis) in
the Pearl River Estuary, China. Mammalia 74, 117–125.
DOI:10.1515/MAMM.2010.024
Crase, B., Liedloff, A.C., Wintle, B.A., 2012. A new method for dealing with residual
spatial autocorrelation in species distribution models. Ecography. 35, 879–888.
DOI:10.1111/j.1600-0587.2011.07138.x
Crase, B., Liedloff, A., Vesk, P.A., Fukuda, Y., Wintle, B.A., 2014. Incorporating spatial
autocorrelation into species distribution models alters forecasts of climate-mediated
range shifts. Glob. Chang. Biol. 20, 2566–2579. DOI:10.1111/gcb.12598
de Oliveira, G., Rangel, T.F., Lima-Ribeiro, M.S., Terribile, L.C., Diniz-Filho, J.A.F.
2014. Evaluating, partitioning, and mapping the spatial autocorrelation component
in ecological niche modeling: A new approach based on environmentally
equidistant records. Ecography. 37, 637–647. DOI:10.1111/j.1600-
0587.2013.00564.x
Dormann, C.F. 2007. Effects of incorporating spatial autocorrelation into the analysis of
species distribution data. Glob. Ecol. Biogeogr. 16, 129–138. DOI:10.1111/j.1466-
8238.2006.00279.x
129
Dormann, C.F., McPherson, J.M., Araújo, M.B., Bivand, R., Bolliger, J., Carl, G.,
Davies, R.G., Hirzel, A.H., Jetz, W., Daniel Kissling, W., Kühn, I., Ohlemüller, R.,
Peres-Neto, P.R., Reineking, B., Schröder, B., Schurr, F.M., Wilson, R. 2007.
Methods to account for spatial autocorrelation in the analysis of species
distributional data: A review. Ecography. 30, 609–628. DOI:10.1111/j.2007.0906-
7590.05171.x
Dungan, S.Z., Hung, S.K.-Y., Wang, J.Y., White, B.N. 2012. Two social communities in
the Pearl River Estuary population of Indo-Pacific humpback dolphins (Sousa
chinensis). Can. J. Zool. 90, 1031–1043. DOI:10.1139/Z2012-071
Dungan, S.Z., Wang, J.Y., Araújo, C.C., Yang, S.C., White, B.N. 2016. Social structure
in a critically endangered Indo-Pacific humpback dolphin (Sousa chinensis)
population. Aquat. Conserv. Mar. Freshw. Ecosyst. 26, 517–529.
DOI:10.1002/aqc.2562
Hijmans, R.J., Phillips, S., Leathwick, J., Elith, J. 2017. dismo: Species Distribution
Modeling. R package version 1.1-4. <http://CRAN.R-project.org/package=dismo>
Huang, S.-L., Wang, C.C., Yao, C.-J. 2018. Habitat protection actions for the Indo-
Pacific humpback dolphin: Baseline gaps, scopes, and resolutions for the
Taiwanese subspecies. Aquat. Conserv. Mar. Freshw. Ecosyst. 1–11.
DOI:10.1002/aqc.2875
Hung, S.K.-Y. 2008. Habitat use of Indo-pacific humpback dolphins (Sousa chinensis) in
Hong Kong. Ph.D. Thesis, University of Hong Kong p.1-253.
DOI:10.5353/th_b4088776
Hung, S.K.-Y. 2013. Monitoring of Marine Mammals in Hong Kong Waters (2012-13).
Final Report submitted to the Agriculture, Fisheries and Conservation Department
of the Hong Kong SAR Government.
Hung, S.K.-Y. 2014. Monitoring of Marine Mammals in Hong Kong Waters (2013-14).
Final Report submitted to the Agriculture, Fisheries and Conservation Department
of the Hong Kong SAR Government.
Hung, S.K.-Y. 2015. Monitoring of Marine Mammals in Hong Kong Waters (2014-15).
Final Report submitted to the Agriculture, Fisheries and Conservation Department
of the Hong Kong SAR Government.
Hung, S.K.-Y. 2016. Monitoring of Marine Mammals in Hong Kong waters (2015-16).
Hong Kong.
Hung, S.K.-Y. 2017. Monitoring of Marine Mammals in Hong Kong Waters (2016-17).
Final Report submitted to the Agriculture, Fisheries and Conservation Department
of the Hong Kong SAR Government.
130
Hung, S.K.-Y., Jefferson, T.A. 2004. Ranging Patterns of Indo-Pacific Humpback
Dolphins (Sousa chinensis) in the Pearl River Estuary, People’s Republic of China.
Aquat. Mamm. 30, 159–174. DOI:10.1578/AM.30.1.2004.159
Jefferson, T.A. 2000. Population biology of the Indo-Pacific hump-backed dolphin in
Hong Kong waters. Wildl. Monogr. 144, 1–65.
Jefferson, T.A., Hung, S.K.-Y., Würsig, B. 2009. Protecting small cetaceans from coastal
development: Impact assessment and mitigation experience in Hong Kong. Mar.
Policy 33, 305–311. DOI:10.1016/j.marpol.2008.07.011
Karczmarski, L., Huang, S., Wong, W.H., Chang, W.L., Chan, S.C.Y., Keith, M. 2017.
Distribution of a coastal delphinid under the impact of long-term habitat loss: Indo-
Pacific humpback dolphins off Taiwan’s west coast. Estuaries and Coasts 40, 594–
603. DOI:10.1007/s12237-016-0146-5
Karczmarski, L., Or, K.M. 2016. Habitat and areas for the conservation of Chinese white
dolphins in Hong Kong Report submitted to WWF-Hong Kong.
DOI:10.13140/RG.2.2.34756.42889
Kissling, W.D., Carl, G. 2008. Spatial autocorrelation and the selection of simultaneous
autoregressive models. Glob. Ecol. Biogeogr. 17, 59–71. DOI:10.1111/j.1466-
8238.2007.00334.x
Mendez, M., Jefferson, T.A., Kolokotronis, S.-O., Krützen, M., Parra, G.J., Collins, T.,
Minton, G., Baldwin, R., Berggren, P., Särnblad, A., Amir, O.A., Peddemors, V.M.,
Karczmarski, L., Guissamulo, A., Smith, B., Sutaria, D., Amato, G., Rosenbaum,
H.C. 2013. Integrating multiple lines of evidence to better understand the
evolutionary divergence of humpback dolphins along their entire distribution range:
A new dolphin species in Australian waters? Mol. Ecol. DOI:10.1111/mec.12535
Or, K.M. 2017. Socio-spatial ecology of Indo-Pacific humpback dolphins (Sousa
chinensis) in Hong Kong and the Pearl River Estuary. Ph.D. Thesis, University of
Hong Kong. p. 1-226.
Radersma, R., Sheldon, B.C. 2015. A new permutation technique to explore and control
for spatial autocorrelation. Methods Ecol. Evol. 6, 1026-1033. DOI:10.1111/2041-
210X.12390
Ross, P.S., Dungan, S.Z., Hung, S.K.-Y., Jefferson, T.A., Macfarquhar, C., Perrin, W.F.,
Riehl, K.N., Slooten, E., Tsai, J., Wang, J.Y., White, B.N., Würsig, B., Yang, S.-C.,
Reeves, R.R. 2010. Averting the baiji syndrome: conserving habitat for critically
endangered dolphins in Eastern Taiwan Strait. Aquat. Conserv. Mar. Freshw.
Ecosyst. 20, 685–694. DOI:10.1002/aqc.1141
SCSFRI (South China Sea Fisheries Research Institute). 2013. Study on Indo-Pacific
humpback dolphin population dynamic and their prey availability in western Pearl
River Estuary. Report submitted to WWF Hong Kong. p. 1-75.
131
SCSFRI (South China Sea Fisheries Research Institute) and HKCRP (Hong Kong
Cetacean Research Project). 2011. Study on the population biology of Chinese
white dolphins (Sousa chinensis) in the western Pearl River estuary, with
exploratory surveys in the Moyangjiang River estuary. Report submitted to WWF
Hong Kong. p. 1-39.
Segurado, P., Araujo, M.B., Kunin, W.E. 2006. Consequences of spatial autocorrelation
for niche-based models. J. Appl. Ecol. 43, 433–444. DOI:10.1111/j.1365-
2664.2006.01162.x
Slooten, E., Wang, J.Y., Dungan, S.Z., Forney, K.A., Hung, S.K.-Y., Jefferson, T.A.,
Riehl, K.N., Rojas-Bracho, L., Ross, P.S., Wee, A., Winkler, R., Yang, S.-C., Chen,
C.A. 2013. Impacts of fisheries on the Critically Endangered humpback dolphin
Sousa chinensis population in the eastern Taiwan Strait. Endanger. Species Res. 22,
99–114. DOI:10.3354/esr00518
Thuiller, W., Lafourcade, B., Engler, R., Araújo, M.B. 2009. BIOMOD - A platform for
ensemble forecasting of species distributions. Ecography. 32, 369–373.
DOI:10.1111/j.1600-0587.2008.05742.x
Varela, S., Anderson, R.P., García-Valdés, R., Fernández-González, F. 2014.
Environmental filters reduce the effect of sampling bias and improve predictions of
ecological models. Ecography. 37, 1084-1091
Veloz, S.D. 2009. Spatially autocorrelated sampling falsely inflates measures of accuracy
for presence-only niche models. J. of Biogeog. 36. 2290-2299. DOI:
10.1111/j.1365-2699.2009.02174.x
Wang, J.Y., Araújo-Wang, C. 2017. Severe mutilation of a Critically Endangered
Taiwanese humpback dolphin Sousa chinensis taiwanensis by fishing gear. Dis.
Aquat. Organ. 123, 257–262. DOI:10.3354/dao03098
Wang, J.Y., Yang, S.-C., Reeves, R.R. 2004. Report of the first workshop on
conservation and research needs of Indo-Pacific humpback dolphins, Sousa
chinensis, in the waters of Taiwan. Checheng, Pingtung County, Taiwan.
Wang, J.Y., Yang, S.C., Reeves, R.R. 2007. Report of the second international workshop
on the conservation and research needs of the eastern Taiwan Strait population of
Indo-Pacific humpback dolphins, Sousa chinensis. Checheng, Pingtung County,
Taiwan.
Wang, J.Y., Yang, S.-C., Fruet, P.F., Daura-Jorge, F.G., Secchi, E.R. 2012. Mark-
recapture analysis of the critically endangered eastern Taiwan Strait population of
Indo-Pacific humpback dolphins (Sousa chinensis): Implications for conservation.
Bull. Mar. Sci. 88, 885–902.
Wang, J.Y., Riehl, K.N., Klein, M.N., Javdan, S., Hoffman, J.M., Dungan, S.Z., Dares,
L.E., Araújo-Wang, C. 2016. Biology and conservation of the Taiwanese humpback
132
dolphin, Sousa chinensis taiwanensis. Advances in Marine Biology. 73, 91-117.
DOI:10.1016/bs.amb.2015.07.005
Wang, J.Y., Riehl, K.N., Yang, S.C., Araújo-Wang, C. 2017. Unsustainable human-
induced injuries to the Critically Endangered Taiwanese humpback dolphins (Sousa
chinensis taiwanensis). Mar. Pollut. Bull. 116, 167–174.
DOI:10.1016/j.marpolbul.2016.12.080
Wong, W.H. 2017. Macro- and micro-scale anthropogenic impacts on Chinese white
dolphins in Hong Kong: Quantifying impacts of habitat loss and coastal tourism.
Ph.D. Thesis, University of Hong Kong. p. 1-253.
Xu, X., Song, J., Zhang, Z., Li, P., Yang, G., Zhou, K. 2015. The world’s second largest
population of humpback dolphins in the waters of Zhanjiang deserves the highest
conservation priority. Sci. Rep. 5, 8147. DOI:10.1038/srep08147
133
Appendix A – Permissions for Inclusion of Published Material
Chapter 2: Permission to reprint article published in Aquatic Mammals
134
135
136
137
138
139
140
141
Chapter 3 – Permission to reprint work published in Estuarine, Coastal and Shelf Science
Retrieved from https://www.elsevier.com/about/policies/sharing, July 9, 2018
Article Sharing
Authors who publish in Elsevier journals can share their research in several ways.
Researchers who have subscribed access to articles published by Elsevier can share
too. There are some simple guidelines to follow, which vary depending on the article version
you wish to share. Elsevier is a signatory to the STM Voluntary Principles for article sharing
on Scholarly Collaboration Networks and a member of the Coalition for Responsible Sharing
Published Journal Articles
• Theses and dissertations which contain embedded PJAs as part of the formal
submission can be posted publicly by the awarding institution with DOI links
back to the formal publications on ScienceDirect
142
Appendix B – Supplementary Analyses for Chapter 3
Predictor Correlation
Prior to conducting the analyses in Chapter 3, predictor variables were tested for
significant multicollinearity that would affect model selection and variable importance in
linear and generalized linear models. Significant correlation (p < 0.05) was only detected
between distance to river and distance to reclamation, however the correlation coefficient
(r=0.29; Fig. B1) was below the threshold value of r=|0.7| that has been shown to affect
model selection (Dormann et al., 2007).
Power Analyses
Post-hoc power analyses were conducted to determine whether sample sizes were
sufficient to be confident in the rejection of the null hypothesis based on observed effect
sizes of the GLMs used to analyze relationships between environmental variables (depth,
sea surface temperature and salinity) and the distance to a river and measures of dolphin
density (SPUE, DPUE, MCPUE). Partial effect sizes of each predictor were calculated
for each model using the modelEffectSizes function in the lmSupport package in R
(Curtin, 2018; Table B.1). Statistical power was calculated using the modelPower
function (Curtin, 2018) using the largest effect size for any predictor in each model, the
sample size (n=73), degrees of freedom and significance level (α=0.05; Table B.1).
The use of the largest effect size in calculations gives an optimistic value of the
power of each model. However, even these optimistic values are low, indicating a high
probability of Type II error. These results indicate that the variation in the data is too
143
large to make a reliable conclusion about the relationships between environmental
characteristics of the ETS and measures of dolphin density or proximity to a river. In
Chapter 4, we reported that there were no significant relationships in any of the GLM
comparisons made, concluding that this result may have been due to sparse sampling of
environmental data in some grid cells so that calculated means may not be truly
representative of the average conditions within a given grid cell. The above power
analyses necessitate the inclusion of an additional caveat, that the lack of observed
relationships may be the result of a highly variable dataset. A larger and more robust
dataset could reveal relationships between environmental factors and dolphin densities, or
at least increase statistical power so that conclusions of no relationship can be made with
confidence. However, availability and efficacy of larger datasets is limited: the remotely
sensed data used in Chapter 4 gives a more robust estimate of average conditions within
each grid cell as seasonal means were calculated from monthly measurements across
several years, but there are only so many grid cells with dolphin sightings, limiting the
sample size of the analysis. Data collected in future surveys could be added to the dataset
and the analyses revisited at a later date, however spatial autocorrelation is likely to be a
factor (as demonstrated in Chapter 4), and inclusion of these data could inflate the
effective sample size of the data and cause inflated Type I error as described by Dale and
Fortin (2002), so there is likely an upper limit on the amount of spatial data that can be
analyzed.
144
References
Dale, M.R.T., Fortin, M.J. 2002. Spatial autocorrelation and statistical tests in ecology.
Ecoscience. 9, 162–167. DOI:10.1080/11956860.2002.11682702
Dormann, C.F., Elith, J., Bacher, S., Buchmann, C., Carl, G., Carré, G., Marquéz, J.R.G.,
Gruber, B., Lafourçade, B., Leitão, P.J., Münkemüller, T., Mcclean, C., Osborne,
P.E., Reineking, B., Schröder, B., Skidmore, A.K., Zurell, D., Lautenbach, S., 2013.
Collinearity: A review of methods to deal with it and a simulation study evaluating
their performance. Ecography. 36, 027–046. DOI:10.1111/j.1600-
0587.2012.07348.x
Curtin, J. 2018. lmSupport: Support for Linear Models. R package version 2.9.13.
<http://CRAN.R-project.org/package=lmSupport>
Wei, T., Simko, V. 2017. R package “corrplot”: Visualization of a Correlation Matrix. R
Package, Version 0.84. <http://github.com/taiyun/corrplot>
145
Tables and Figures
Figure B.1 – Correlation matrix for environmental predictors (mean_sal = mean salinity,
mean_sst = mean sea surface temperature, mean_depth = mean depth, riverdist = distance
from the centre of the grid cell to the nearest river, and recdist = distance from the centre
of the grid cell to the edge of the nearest major land reclamation project) used in GLMs in
Chapter 3, plotted using the corrplot package in R (Wei and Simko, 2017). Pearson’s
correlation coefficients are shown, and statistically significant correlations (p < 0.05) are
indicated with an asterisk (*).
146
Table B.1 – Partial effect sizes of predictor variables and statistical power of each
regression conducted in Chapter 3. Stated effect sizes are partial eta squared, which
measures the effect of each variable when other coefficients are held constant. An
optimistic calculation of overall power of each model is listed in the bottom row,
calculated using the largest effect size for any variable above.
SPUE model DPUE model MCPUE model Distance to
river model
Predictor variable Partial effect Size (ηp2)
Depth 7.4 x 10-3 5.2 x 10-4 6.7 x 10-3 6.7 x 10-4
Sea surface
temperature
2.3 x 10-5 2.0 x 10-4 2.9 x 10-3 6.7 x 10-5
Salinity 2.2 x 10-4 2.0 x 10-6 4.4 x 10-3 1.2 x 10-3
Distance to a river 6.6 x 10-2 6.3 x 10-4 1.7 x 10-5 NA
Distance to a
reclamation site
3.5 x 10-2 5.6 x 10-6 1.6 x 10-2 NA
Model power 0.58 0.05 0.10 0.06
147
Appendix C – Supplementary Materials for Chapter 4
R Code
Custom algorithms written in R and used in Chapter 4 are included below. A brief
description of the purpose of each algorithm is given, as well as descriptions of the
arguments required for each function. Note that these functions have only been tested on
data used in this thesis.
Ranking algorithm
Produces list of rownames of least- to most-distant presences; enter dataset name (data
frame with binary presence/absence column named “pres”) and column numbers
containing environmental variables
rank.pts <- function(dataset, varcols){ dataset1.mdist.mean.sd <- data.frame(rm.point=rep(0,nrow(dataset)), mindist = rep(0,nrow(dataset)), mean = rep(0,nrow(dataset)), max.dist=rep(0,nrow(dataset)), median=rep(0,nrow(dataset)), sd = rep(0,nrow(dataset))) dataset1 <- dataset for (i in 1:nrow(dataset)){ dataset1.list <- split(dataset1, seq(nrow(dataset1))) #Split data frame into list of length nrows dataset1.env.cov <- cov(dataset1[varcols]) #Make covariance matrix dataset1.mdist.list <- lapply(dataset1.list, function(a){ #Calculate distance from each list object to every point in dataset1 tryCatch(mahalanobis(x=dataset1[varcols], center=as.numeric(a[varcols]), cov = dataset1.env.cov), error=function(e) "inv.error") #Makes note of error when inverting matrix without quitting function }) dataset1.mdist.matrix <- data.frame(lapply(dataset1.mdist.list, cbind)) #Make matrix from list object if(dataset1.mdist.matrix[1,1]=="inv.error"){ break #Break loop when matrix can’t be inverted } dataset1.mdist.matrix2 <- as.dist(dataset1.mdist.matrix, diag = FALSE, upper = FALSE) #Convert to distance matrix mindist <- min(dataset1.mdist.matrix2) #Find smallest distance between two points mindist.rows<- (names(which(as.matrix(dataset1.mdist.matrix2) == min(dataset1.mdist.matrix2), arr.ind = TRUE)[, 1])) #Get names of two points that mindist is between extract.mindist.rows <- subset(dataset1.mdist.matrix, rownames(dataset1.mdist.matrix) %in% mindist.rows) #Take rows of mindist points from dist matrix
148
rm.point<- rownames(which(extract.mindist.rows == min(extract.mindist.rows[extract.mindist.rows > mindist]), arr.ind = TRUE)) #Get name of point with next-smallest dist dataset1.mdist.mean.sd[i,] <- c(rm.point, mindist, mean(dataset1.mdist.matrix2), max.dist = max(dataset1.mdist.matrix2), median=median(dataset1.mdist.matrix2), sd(dataset1.mdist.matrix2)) #Write mean and stdev of dataset.mdist.matrix2 to new df to find plateau/smallest sd later dataset1 <- subset(dataset1, !rownames(dataset1) %in% rm.point) #Remove point with two smallest mdists from dataset by name } #Fill in last few points left over after matrix error rm.point.mvec <- dataset1.mdist.mean.sd[dataset1.mdist.mean.sd$rm.point>0,]$rm.point missing.pres.pts <-subset(dataset, !rownames(dataset) %in% rm.point.mvec) missing.pres.pts <- data.frame(rm.point=rownames(missing.pres.pts), mindist = rep(0,nrow(missing.pres.pts)), mean= rep(0,nrow(missing.pres.pts)), max.dist = rep(0,nrow(missing.pres.pts)), median = rep(0,nrow(missing.pres.pts)), sd = rep(0,nrow(missing.pres.pts))) dataset1a.mdists.mean.sd <- rbind(dataset1.mdist.mean.sd[dataset1.mdist.mean.sd$rm.point>0,], missing.pres.pts) rm.point.mvec <- dataset1a.mdists.mean.sd$rm.point return(rm.point.mvec)
}
Subsetting algorithm
Creates multiple thinned subsets using rank.pts algorithm given above, as well as data
management to produce a list of the desired number of thinned subsets. Enter dataset (as
above), columns containing predictor variables (as above), desired sample size for
thinned dataset, number of thinned subsets required, and increment to remove points from
first thinned subset to generate additional subsets.
alt.subsets <- function(dataset, varcols, thinned.n, n.subsets, subset.incr){
pres.data <- dataset[dataset$pres >0,] abs.data <- dataset[dataset$pres==0,] thinned.n <- thinned.n/2 pres.rank <- rank.pts(pres.data, varcols) #Get ranked least- to most-distant presences abs.rank <- rank.pts(abs.data, varcols) #Get ranked least- to most-distant absences subset.vec <- seq(0, n.subsets-1)
149
rm.increments.pres.list <- lapply(subset.vec, function(a){subset(pres.data, rownames(pres.data) %in% pres.rank[c(1:(length(pres.rank)-(subset.incr*a)))])}) rm.increments.abs.list <- lapply(subset.vec, function(a){subset(abs.data, rownames(abs.data) %in% abs.rank[c(1:(length(abs.rank)-(subset.incr*a)))])}) pres.subsets.pts <- lapply(rm.increments.pres.list, function(a){rank.pts(a, varcols)}) #run ranking algorithm on each pres dataset abs.subsets.pts <- lapply(rm.increments.abs.list, function(a){rank.pts(a, varcols)}) #Run ranking algorithm on each abs dataset #Invert lists so that most-distant point is first pres.subsets.pts.inv <- lapply(pres.subsets.pts, rev) abs.subsets.pts.inv <- lapply(abs.subsets.pts, rev) #Trim pts list to desired length (thinned.n) pres.subsets.pts.inv.trim <- lapply(pres.subsets.pts.inv, function(a){a[1:thinned.n]}) abs.subsets.pts.inv.trim <- lapply(abs.subsets.pts.inv, function(a){a[1:thinned.n]}) #Subset pres and abs datsets using inverted, trimmed ranked points list pres.subsets.list <- lapply(pres.subsets.pts.inv.trim, function(a){subset(pres.data, rownames(pres.data) %in% a)}) abs.subsets.list <- lapply(abs.subsets.pts.inv.trim, function(a){subset(abs.data, rownames(abs.data) %in% a)}) thinned.subsets <- list() for(i in 1:n.subsets){ thinned.subsets[[i]] <- rbind(pres.subsets.list[[i]], abs.subsets.list[[i]]) }
return(thinned.subsets) }
Spatially Buffered Leave-One-Out Cross-Validation (SBLOOCV)
Algorithm
Example of SBLOOCV function used in Chapter 4, shown for GAMs. Performs spatially
buffered leave-one-out cross-validation with user-defined buffer distance. Enter data (as
above), model formula, name of spatial data frame, and buffer distance. This algorithm
was written based on the concepts described by Roberts et al. (2017) – see Chapter 4 for
full reference.
#Required packages: require(gstat) #For semivariograms require(MuMIn) #For AICc require(mgcv) #For GAMs require(sp) #For distance calculations
150
#Convert data to spatial data frame to be used for distance calculations for buffer spdata <- dataset coordinates(spdata) <- c("x", "y") #Do visual inspection of semivariogram to determine major range of SAC --- use this value as "buffer" in function below resid <- residuals(glm(pres~1, data = dataset, family = “binomial”) vario.dir = variogram(resid~1, data = spdata, alpha = c(0,30,60,90,120,150)) #Specify separation of points for directional semivariogram plot(vario.dir, as.table = TRUE) gam.sbloocv <- function(dataset, formula, spdata, buffer){
preds <- data.frame(x=rep(0,nrow(dataset)), y=rep(0,nrow(dataset)), obs.pres=rep(0,nrow(dataset)), prob=rep(0,nrow(dataset)), aicc=rep(0,nrow(dataset)), training.n=rep(0,nrow(dataset))) warnings <-data.frame(warnMsg=rep(0,nrow(dataset))) withCallingHandlers({ for (i in 1:nrow(dataset)) { w <- length(warnings()) distances <- vector(mode = "list", length = nrow(dataset)) distances[[i]] <- which(spDists(spdata, spdata[i,]) >= buffer) train <- subset(dataset, rownames(dataset) %in% distances[[i]]) if (nrow(train) < 50) { next } test <- dataset[i,] run.gam <- gam(formula, data = train, family = "binomial") preds[i, ] = c(dataset[i,]$x, dataset[i,]$y, dataset[i,]$pres, exp(predict(run.gam, test))/(1+exp(predict(run.gam,test))), AICc(run.gam), nrow(train)) } }, warning = function(w){ warnings$warnMsg[i] <<-w$message invokeRestart("muffleWarning") }) warnings$warnPres <- as.numeric(warnings$warnMsg) #Captures any warnings encountered at any i of loop preds$warnMsg<- warnings$warnMsg preds$warnPres<- warnings$warnPres return(preds)
}
151
Tables and Figures
Figure C.1 – Three-dimensional scatterplots of full PRE dry dataset (a,e, n=2500), and
the three thinned subsets (n=140, prevalence = 0.5) used for species distribution
modelling in Chapter 4. Lighter colours indicate that the point is closer to the origin (i.e.
the viewer) on the z-axis. The top four panels (a-d) illustrate results of the thinning
procedure on the depth, turbidity and productivity (NPP) axes, and the bottom four panels
(e-h) are on the distance to shore, sea surface temperature (SST) and productivity (NPP)
axes (N.B. productivity was repeated in both sets of plots to facilitate three-dimensional
plotting of five environmental variables). Thinned subsets were selected by the procedure
described in Chapter 4, with the original subset (i.e. 140 most-distant points in
environmental space) comprising the first subset (b,f), and the subsequent two subsets
(c,g and d,h) selected by removing the 10 and 20 most-distant points, respectively, from
the original subset and repeating the ranking procedure.
152
Figure C.2 – Three-dimensional scatterplots of full PRE wet dataset (a,e, n=2814), and
the three thinned subsets (n=180, prevalence = 0.5) used for species distribution
modelling in Chapter 4. Lighter colours indicate that the point is closer to the origin (i.e.
the viewer) on the z-axis. The top four panels (a-d) illustrate results of the thinning
procedure on the depth, turbidity and productivity (NPP) axes, and the bottom four panels
(e-h) are on the distance to shore, sea surface temperature (SST) and productivity (NPP)
axes (N.B. productivity was repeated in both sets of plots to facilitate three-dimensional
plotting). Thinned subsets were selected by the procedure described in Chapter 4, with
the original subset (i.e. 180 most-distant points in environmental space) comprising the
first subset (b,f), and the subsequent two subsets (c,g and d,h) selected by removing the
10 and 20 most-distant points, respectively, from the original subset and repeating the
ranking procedure.
153
Figure C.3 – Three-dimensional scatterplots of full ETS dataset (a,e, n=673), and the
three thinned subsets (n=80, prevalence = 0.5) used for species distribution modelling in
Chapter 4. Lighter colours indicate that the point is closer to the origin (i.e. the viewer) on
the z-axis. The top four panels (a-d) illustrate results of the thinning procedure on the
depth, turbidity and productivity (NPP) axes, and the bottom four panels (e-h) are on the
distance to shore, sea surface temperature (SST) and productivity (NPP) axes (N.B.
productivity was repeated in both sets of plots to facilitate three-dimensional plotting).
Thinned subsets were selected by the procedure described in Chapter 4, with the original
subset (i.e. 80 most-distant points in environmental space) comprising the first subset
(b,f), and the subsequent two subsets (c,g and d,h) selected by removing the 10 and 20
most-distant points, respectively, from the original subset and repeating the ranking
procedure.
154
Table C.1 – Summary statistics for depth data used as predictor variables in Chapter 4.
Full dataset rows contain summary statistics for all grid cells with survey effort in each
study area and season. Subset rows contain summary statistics for each of the three
thinned subsets for each study area, obtained by the thinning procedure described in
Chapter 4.
Depth (m)
Presences Absences
Study Area Dataset Mean ± SD Median Min-Max Mean ± SD Median Min – Max
PRE
(dry season)
Full
(n=2500)
8.3 ± 4.0 7.3 1.9-24.3 12.5 ± 8.4 10.1 0.9-35.3
Subset 1
(n=140)
8.8 ± 4.5 7.3 1.9-23.3 12.6 ± 8.2 11.0 0.9-34.2
Subset 2
(n=140)
8.8 ± 4.7 7.4 1.9-23.3 12.8 ± 8.4 10.8 0.9-35.3
Subset 3
(n=140)
8.8 ± 4.8 7.4 1.9-23.3 12.6 ± 7.9 10.2 2.4-35.3
PRE
(wet season)
Full
(n= 2814)
8.2 ± 3.9 7.4 1.4-24.3 13.6 ± 9.0 10.9 0.9-35.3
Subset 1
(n=180)
9.0 ± 4.9 8.0 1.4-24.3 13.8 ± 8.5 12.0 0.9-35.3
Subset 2
(n=180)
9.0 ± 4.9 8.3 1.4-24.3 13.4 ± 8.6 11.8 1.4-35.3
Subset 3
(n=180)
8.9 ± 5.0 7.7 1.4-24.3 13.5 ± 8.7 11.6 1.4-35.3
ETS
(wet season)
Full
(n=673)
6.3 ± 3.9 5.5 0.1-20.6 8.7 ± 6.9 6.7 0.1-47.9
Subset 1
(n=80)
6.8 ± 4.5 5.6 0.3-20.6 10.2 ± 10.8 6.2 0.1-47.9
Subset 2
(n=80)
7.0 ± 4.1 6.8 0.1-16.9 10.3 ± 10.0 6.8 0.3-47.9
Subset 3
(n=80)
7.1 ± 4.0 6.3 0.1-16.9 9.3 ± 9.4 6.3 0.1-37.7
155
Table C.2 – Summary statistics for distance to shore data used as predictor variables in
Chapter 4. Full dataset rows contain summary statistics for all grid cells with survey
effort in each study area and season. Subset rows contain summary statistics for each of
the three thinned subsets for each study area, obtained by the thinning procedure
described in Chapter 4.
Distance to shore (km)
Presences Absences
Study Area Dataset Mean ± SD Median Min-Max Mean ± SD Median Min-Max
PRE
(dry season)
Full
(n=2500)
3.5 ± 3.3 2.3 0-13.5 6.6 ± 4.2 5.9 0-22.3
Subset 1
(n=140)
5.1 ± 3.8 3.8 0-13.5 6.5 ± 5.4 5.4 0-22.3
Subset 2
(n=140)
4.7 ± 3.9 3.5 0-13.5 6.4 ± 5.1 5.3 0-21.7
Subset 3
(n=140)
4.7 ± 3.8 3.6 0-13.5 6.3 ±5.0 5.6 0-21.7
PRE
(wet season)
Full (n=
2814)
3.5 ± 3.4 2.3 0-14.6 6.9 ± 4.4 6.1 0-22.8
Subset 1
(n=180)
4.5 ± 3.7 3.7 0-13.5 5.8 ± 5.1 4.6 0-22.8
Subset 2
(n=180)
4.5 ± 3.7 3.8 0-14.6 5.8 ± 4.8 4.6 0-21.9
Subset 3
(n=180)
4.6 ± 3.9 3.6 0-14.6 5.7 ± 4.8 5.1 0-21.9
ETS
(wet season)
Full
(n=673)
1.5 ± 1.2 1.2 0-5.0 2.2 ± 1.5 1.9 0-7.0
Subset 1
(n=80)
1.6 ± 1.3 1.4 0-5.0 1.8 ± 1.4 1.6 0-5.6
Subset 2
(n=80)
1.5 ± 1.4 1.2 0-5.0 2.2 ± 1.8 1.8 0-6.6
Subset 3
(n=80)
1.6 ± 1.4 1.2 0-5.0 2.0 ± 1.8 1.8 0-6.9
156
Table C.3 – Summary statistics for productivity data used as predictor variables in
Chapter 4. Full dataset rows contain summary statistics for all grid cells with survey
effort in each study area and season. Subset rows contain summary statistics for each of
the three thinned subsets for each study area, obtained by the thinning procedure
described in Chapter 4.
Productivity (mg C m-2 day-1)
Presences Absences
Study Area Dataset Mean ± SD Median Min-Max Mean ± SD Median Min-Max
PRE
(dry season)
Full
(n=2500)
2830 ± 239 2839 2200-3386 2628 ± 466 2661 1592-3518
Subset 1
(n=140)
2819 ± 291 2826 2200-3386 2640 ± 453 2716 1592-3412
Subset 2
(n=140)
2826 ± 272 2841 2308-3386 2632 ± 426 2674 1592-3412
Subset 3
(n=140)
2615 ± 278 2812 2308-3386 2635 ± 431 2662 1592-3402
PRE
(wet
season)
Full (n=
2814)
3277 ± 241 3341 2126-3897 2766 ± 665 2819 1348-3959
Subset 1
(n=180)
3222 ± 239 3245 2126-3897 2785 ± 614 2893 1409-3916
Subset 2
(n=180)
3229 ± 307 3245 2254-3897 2833 ± 620 2946 1348-3915
Subset 3
(n=180)
3230 ± 299 3254 2254-3897 2796 ± 619 2832 1348-3941
ETS
(wet
season)
Full
(n=673)
1396 ± 97 1371 1201-1708 1339 ± 216 1352 716-1715
Subset 1
(n=80)
1416 ± 126 1359 1201-1701 1245 ± 292 1315 716-1714
Subset 2
(n=80)
1410 ± 106 1374 1306-1708 1269 ± 255 1329 728-1714
Subset 3
(n=80)
1396 ± 99 1362 1306-1708 1279 ± 260 1335 1306-1708
157
Table C.4 – Summary statistics for turbidity data used as predictor variables in Chapter 4.
Full dataset rows contain summary statistics for all grid cells with survey effort in each
study area and season. Subset rows contain summary statistics for each of the three
thinned subsets for each study area, obtained by the thinning procedure described in
Chapter 4.
Turbidity (m-1)
Presences Absences
Study Area Dataset Mean ± SD Median Min-Max Mean ± SD Median Min-Max
PRE
(dry season)
Full
(n=2500)
0.33 ± 0.05 0.33 0.22-0.51 0.34 ± 0.10 0.36 0.18-0.61
Subset 1
(n=140)
0.36 ± 0.07 0.36 0.22-0.51 0.35 ± 0.09 0.37 0.18-0.59
Subset 2
(n=140)
0.35 ± 0.06 0.36 0.22-0.51 0.35 ± 0.10 0.36 0.18-0.61
Subset 3
(n=140)
0.35 ± 0.06 0.35 0.22-0.51 0.35 ± 0.10 0.36 0.18-0.61
PRE
(wet season)
Full
(n= 2814)
0.59 ± 0.14 0.58 0.40-1.05 0.54 ± 0.16 0.51 0.24-1.06
Subset 1
(n=180)
0.63 ± 0.14 0.63 0.40-1.05 0.56 ± 0.18 0.59 0.25-1.06
Subset 2
(n=180)
0.64 ± 0.14 0.63 0.41-1.05 0.57 ± 0.18 0.59 0.24-1.03
Subset 3
(n=180)
0.64 ± 0.15 0.64 0.41-1.05 0.56 ± 0.18 0.56 0.24-1.03
ETS
(wet season)
Full
(n=673)
0.19 ± 0.03 0.18 0.16-0.37 0.21 ± 0.06 0.19 0.12-0.39
Subset 1
(n=80)
0.21 ± 0.05 0.20 0.16-0.37 0.21 ± 0.07 0.20 0.12-0.38
Subset 2
(n=80)
0.20 ± 0.03 0.19 0.16-0.33 0.21 ± 0.06 0.19 0.12-0.35
Subset 3
(n=80)
0.19 ± 0.02 0.18 0.16-0.27 0.21 ± 0.06 0.20 0.12-0.36
158
Table C.5 – Summary statistics for sea surface temperature data used as predictor
variables in Chapter 4. Full dataset rows contain summary statistics for all grid cells with
survey effort in each study area and season. Subset rows contain summary statistics for
each of the three thinned subsets for each study area, obtained by the thinning procedure
described in Chapter 4.
Sea Surface Temperature (oC)
Presences Absences
Study Area Dataset Mean ± SD Median Min-Max Mean ± SD Median Min-Max
PRE
(dry season)
Full
(n=2500)
22.7 ± 0.3 22.7 22.5-23.4 22.7 ± 0.3 22.7 22.1-23.5
Subset 1
(n=140)
22.8 ± 0.2 22.7 22.5-23.4 22.8 ± 0.3 22.7 22.1-23.4
Subset 2
(n=140)
22.8 ± 0.2 22.7 22.5-23.4 22.8 ± 0.3 22.7 22.1-23.4
Subset 3
(n=140)
22.8 ± 0.2 22.7 22.5-23.3 22.8 ± 0.3 22.7 22.1-23.4
PRE
(wet season)
Full
(n= 2814)
27.3 ± 0.3 27.2 26.8-27.9 27.3 ± 0.4 27.3 26.6-27.9
Subset 1
(n=180)
27.4 ± 0.3 27.3 26.8-27.9 27.3 ± 0.3 27.3 26.6-27.9
Subset 2
(n=180)
27.4 ± 0.3 27.3 26.8-27.9 27.3 ± 0.3 27.4 26.6-27.9
Subset 3
(n=180)
27.4 ± 0.3 27.3 26.8-27.9 27.3 ± 0.3 27.4 26.6-27.9
ETS
(wet season)
Full
(n=673)
29.2 ± 0.3 29.2 28.8-30.6 29.4 ± 0.5 29.3 28.5-31.3
Subset 1
(n=80)
29.3 ± 0.4 29.2 28.9-30.6 29.7 ± 0.7 29.7 28.6-31.3
Subset 2
(n=80)
29.2 ± 0.3 29.2 28.9-29.9 29.6 ± 0.6 29.5 28.6-31.1
Subset 3
(n=80)
29.2 ± 0.2 29.2 28.8-29.7 29.6 ± 0.7 29.6 28.7-31.2