habitat characteristics, density patterns and...

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.
















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,


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-


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-



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


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,



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


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






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>


chinensis) in Hong Kong and the Pearl River estuary. Ph.D. Thesis, University of


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/>






107



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




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


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,



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


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. 2016. Monitoring of Marine Mammals in Hong Kong waters (2015-16).

Hong Kong.




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






A new dolphin species in Australian waters? Mol. Ecol. DOI:10.1111/mec.12535


chinensis) in Hong Kong and the Pearl River Estuary. Ph.D. Thesis, University of


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




131





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.




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

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

https://www.elsevier.com/about/policies/sharing

http://www.stm-assoc.org/2015_06_08_Voluntary_principles_for_article_sharing_on_scholarly_collaboration_networks.pdf

http://www.responsiblesharing.org/

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


Productivity (mg C m-2 day-1)

Presences Absences


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


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


Sea Surface Temperature (oC)

Presences Absences


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

habitat characteristics, density patterns and...

Documents