Spatially-Explicit Predictive Modeling of Coral Species

SPATIALLY-EXPLICIT PREDICTIVE MODELING OF CORAL SPECIES

DISTRIBUTIONS IN THE HAWAIIAN ISLANDS

A DISSERTATION SUBMITTED TO THE GRADUATE DIVISION OF THE UNIVERSITY OF HAWAI‘I AT MĀNOA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF

DOCTOR OF PHILOSOPHY

ZOOLOGY

DECEMBER 2012

Erik C. Franklin

Dissertation Committee:

Paul L. Jokiel, Chairperson Megan J. Donahue Ruth D. Gates Robert J. Toonen Christopher A. Lepczyk

Keywords: coral, boosted regression trees, ensemble model approach, Hawaii, marine protected areas, species distribution model

ACKNOWLEDGMENTS

My advisor, Paul Jokiel, served as teacher, collaborator, and friend. He embodies

the term “mentor”. I deeply appreciate his unwavering support and guidance throughout

the development of my dissertation. His willingness to sponsor a non-traditional student

kept me from serving as a “tech” for the rest of my career. Thanks, Paul.

My committee members Megan Donahue, Ruth Gates, Rob Toonen, and Chris

Lepczyk expanded my scientific horizons in ways that I could never have anticipated at

the beginning of this process. I greatly appreciate the opportunities, projects, and

scientific insights that I have experienced from this talented group of scientists. I also

look forward to many years of productive future collaborations. Thank you, colleagues.

This work was primarily supported by a US Environmental Protection Agency

Science To Achieve Results (STAR) PhD Fellowship for graduate environmental study in Ecoinformatics (FP-91709601-0). Thank you, funders.

Finally, I could never realize this achievement without the loving encouragement and support of my family. My parents, Charles and Angela, allowed me the freedom to

“follow my bliss” and taught me the persistence required to achieve it. My daughter,

Eliza, has given the gift of perspective – that not all life is work, nor should it be. My wife, Giselle, has been a constant source of encouragement throughout the long process of crafting this work. She deserves co-authorship but instead I acknowledge what a lucky man I am to have her as my wife. Thank you, I love you all.

iii

ABSTRACT

Coral reefs are an ecosystem in transition. Scleractinian corals are the foundation species of tropical and subtropical reef ecosystems, yet information about their status is woefully inadequate. In order to appreciate the changes that reefs are undergoing, we need to explore methods that better explain the current conditions of coral species populations.

Using species distribution models, this dissertation examined the physical and biological factors that influence the distribution of six dominant scleractinian species, Montipora capitata, Montipora flabellata, Montipora patula, Pocillopora meandrina, Porites compressa, and Porites lobata in the main Hawaiian Islands. The primary objectives of the dissertation included: (i) compilation of a database of quantitative field observations of the six coral species from reef surveys and spatial environmental covariate data in the main Hawaiian Islands during 2000-2009, (ii) identification of models and environmental factors that were most informative for predicting the distributions of the six coral species in the main Hawaiian Islands using the field observations for model training and validation, (iii) utilization of the model outputs to map spatially-explicit presence and abundances of coral species for near-shore, shallow reefs (~30 m depth) of the main

Hawaiian Islands, and (iv) comparison of the populations of coral species in a network of

MPAs to unprotected reefs using data from the spatial abundance maps.

The results demonstrated that species distribution modeling approaches are an effective means to characterize the distribution, presence, and abundance of corals in the

Hawaiian Islands. Mean significant wave height and max significant wave height were the most influential variables explaining coral presence and abundance (as benthic cover) in the Hawaiian Islands. Models also identified relationships between coral cover with

island, depth, downwelled irradiance, rugosity, slope, and aspect. The rank order of coral

abundance (from highest to lowest) for the MHI was P. lobata, M. patula, P. meandrina,

M. capitata, P. compressa, and M. flabellata. Abundances of the two Porites species

were higher in MPAs than open areas. The three Montipora species and Pocillopora

meandrina had lower abundances in most MPAs compared to open areas. Manele-

Hulopoe and Molokini Marine Life Conservation Districts (MLCD) had higher

abundances for four of the six coral species compared to unprotected areas while Waikiki

MLCD had lower abundances than open areas for all corals. The Hawaiian Islands

Humpback Whales National Marine Sanctuary (HIHWNMS) encompassed coral populations with higher abundances than areas outside the boundaries especially for the four corals (Montipora spp. and P. meandrina) underrepresented in the current MPA network.

It can be concluded that species distribution modeling delivers a methodological approach to spatially-explicit marine population assessments at a macroecological scale that was not previously possible. The utility of SDMs to provide species abundances at a high map resolution across the entire geographic domain represents a significant improvement in our ability to describe the condition of these coral populations. The information on coral species is critically important as baseline data for population connectivity modeling, marine spatial planning, and especially, climate studies. Coral reefs are undergoing rapid change but species responses to environmental drivers are heterogeneous. This work will serve as the framework for future investigations to better assess the conditions of species populations and understand the changes that Hawaiian reefs are experiencing.

TABLE OF CONTENTS

ACKNOWLEDGEMENTS ...... iii ABSTRACT ...... iv LIST OF TABLES ...... viii LIST OF FIGURES…………………………………………………………………….... .x LIST OF ABBREVIATIONS ...... xviii

Chapter 1. General Introduction ...... 1 References… ……………………………………………………………………....5

Chapter 2. An ensemble approach to species distribution modeling of corals on Hawaiian reefs………… ...... ………………………...7

Abstract ……………………………………………………………………………8 Introduction .……………………………………………………………………….9 Materials and Methods…………………………………………………………...13 Results …………………………………………………………………………... 19 Discussion ...... 23 References……………………………………………………………………...... 26

Chapter 3. Predictive modeling of coral distribution and abundance in the Hawaiian Islands ...... 31

Abstract ...... 32 Introduction… …………………………………………………………………....33 Materials and Methods… ………………………………………………………...35 Results ……………………………………………………………………………42 Discussion ………………………………………………………………………..48 References………………………………………………………………………..52

Chapter 4. A niche model analysis of corals in Hawaii: are populations more abundant in marine protected areas?...... 57

Abstract………… ………………………………………………………………..58 Introduction……… ……………………………………………………………....59 Materials and Methods…… ……………………………………………………...61 Results…………………… ………………………………………………………65 Discussion……………… ………………………………………………………..70 References……………… ………………………………………………………..73

Chapter 5. Summary and conclusions ...... …………………….77 Implications and Applications ...... 78 Future Research ...... 80 Conclusions ...... 81 References ...... 81

Appendix A. Figures of environmental covariate data layers for Oahu ...... 82

Appendix B. Table of relative variable importance for environmental covariates in coral species distribution models for Oahu ...... 86

Appendix C. Comparative response plots of environmental variables from eight model methods for four Hawaiian coral species around Oahu, Hawaii ...... 88

Appendix D. Figures of benthic cover observations for six coral species in the main Hawaiian Islands, 2000-2009...... 105

Appendix E. Figures of environmental covariate data layers for the main Hawaiian Islands ...... 109

Appendix F. Figures of geographic model predictions of benthic cover for six coral species in the main Hawaiian Islands, 2000-2009 ...... 121

Appendix G. Table of model results for sensitivity analysis of boosted regression trees for benthic cover of six coral species ...... 133

Appendix H. Figures of response plots of environmental variables in boosted regressions tree models for benthic cover of six coral species ...... 137

Appendix I. Table of interactions between environmental variables and benthic cover of six coral species ...... 141

vii

LIST OF TABLES

Table 2.2 Summary statistics of environmental covariates around Oahu ...... 18

Table 3.2 Summary statistics of environmental covariates around Oahu ...... 40

viii

= mean residual deviance, train corr = training data correlation, cv corr = cross-validation correlation ...... 133

Table I.1 Interactions between environmental variables and coral species cover ...... 143

LIST OF FIGURES

Figure 2.1 Location of Hawaiian Archipelago in the central north Pacific Ocean and Oahu among the main Hawaiian Islands. The study area (dark gray) extends throughout the shallow, coastal waters of Oahu. Prominent geographic features are labeled ...... 12

Figure 2.2 Presence and absence field observations for (a) Montipora capitata, (b) Pocillopora meandrina, (c) Porites compressa, and (d) Porites lobata from 2000-2009 ...... 15

Figure 2.3 Geographic predictions of probability of occurrence around Oahu for (a) Montipora capitata, (b) Pocillopora meandrina, (c) Porites compressa, and (d) Porites lobata. The probability of occurrence (red to blue scale) results from the ensemble consensus model for each species weighted by ranked predictive performance of the various modeling methods ...... 21

Figure 2.4 Comparison of response curve plots for coral species probability of occurrence on maximum significant wave height, max Hs, from the best performing model approach (generalized boosted regression). High probability of occurrence shifts from P. compressa to P. meandrina and P. lobata at max Hs between 1-2 m suggesting a wave height threshold for transitions in coral community dominance. M. capitata did not respond strongly to max Hs ...... 22

Figure 3.1 Geographic map of the Hawaiian Archipelago in the central north Pacific Ocean and eight main Hawaiian Islands with benthic cover field observations for coral species (open circles) compiled from 2000-2009. The study area (darker gray) extends throughout the shallow, coastal waters (0 – 30 m depth). Figures of field observations for the six coral species are in Appendix D ...... 36

Figure 3.2 Geographic maps of model predicted coral (%) for Montipora capitata (a), M. flabellata (b), and M. patula (c), Pocillopora meandrina (d), Porites compressa (e), Porites lobata (f) around the main Hawaiian Islands. Detailed figures of predicted coral cover for species are in Appendix F ...... 43-44

Figure 3.3 Mean benthic cover (%) ± SE of six coral species predicted from final BRT models for each Hawaiian island and the entire main Hawaiian Islands (MHI) ...... 46

Figure 3.4 Geographic map of summed total cover for six coral species predicted from BRT models for the main Hawaiian Islands ...... 47

Figure 4.1 Map of the main Hawaiian Islands with marine protected areas. The Hawaiian Archipelago is in the central north Pacific Ocean. The study area (dark gray) extends throughout the shallow, coastal waters of the main Hawaiian Islands. Marine protected areas include Marine Life Conservation Districts (MLCD), the Hawaii Marine

Laboratory Refuge (HMLR), a Natural Area Reserve (NAR), and the Hawaiian Islands Humpack Whale National Marine Sanctuary (HIHWNMS) ...... 62

Figure 4.2 Benthic cover (± 1SEM) of six coral species in marine protected areas (MPAs; red bars) and open areas (blue bars) of waters around Kauai (includes Niihau), Oahu, Maui (includes Molokai, Lanai, Kahoolawe), and Hawaii predicted from geographic projection of species distribution model results...... 67

Figure 4.3 Benthic cover (± 1SEM) of six coral species in twelve Hawaiian marine protected areas (MPAs) calculated from optimal BRT species models. Abundances in MPAs that exceed coral cover in open areas (orange bars) are contrast with coral covers in MPAs less than open areas (gray bars) ...... 68

Figure 4.4 Benthic cover of six coral species in the Hawaiian Islands Humpback Whale National Marine Sanctuary (red bars) and open areas (blue bars) of waters around Kauai (includes Niihau), Oahu, Maui (includes Molokai, Lanai, Kahoolawe), and Hawaii predicted from geographic projection of species distribution model results ...... 69

Figure A.1 Shallow bathymetry (0 to 30 m depth) of the waters around Oahu ...... 82

Figure A.2 Bathymetric slope (in degrees) of the sea floor around Oahu ...... 82

Figure A.3 Bathymetric aspect (in degrees) of the seafloor around Oahu ...... 83

Figure A.4 Bathymetric rugosity (surface area / planar area) of the sea floor around Oahu .

...... 83

Figure A.5 Sand habitat (proportion) of the sea floor around Oahu ...... 84

Figure A.6 Maximum significant wave heights for the waters around Oahu ...... 84

Figure A.7 Mean significant wave heights for waters the around Oahu ...... 85

Figure A.8 Downwelled irradiance at the sea floor relative to that just below the sea surface around Oahu ...... 85

Figure C.1 Response plots of Montipora capitata probability of occurrence to depth from eight model methods ...... 88

Figure C.2 Response plots of Montipora capitata probability of occurrence to slope from eight model methods ...... 89

Figure C.3 Response plots of Montipora capitata probability of occurrence to aspect from eight model methods ...... 89

Figure C.4 Response plots of Montipora capitata probability of occurrence to rugosity from eight model methods ...... 90

Figure C.5 Response plots of Montipora capitata probability of occurrence to sandbottom from eight model methods ...... 90

Figure C.6 Response plots of Montipora capitata probability of occurrence to maximum significant wave height from eight model methods ...... 91

Figure C.7 Response plots of Montipora capitata probability of occurrence to mean significant wave height from eight model methods ...... 91

Figure C.8 Response plots of Montipora capitata probability of occurrence to downwelled irradiance from eight model methods ...... 92

Figure C.9 Response plots of Pocillopora meandrina probability of occurrence to depth from eight model methods ...... 92

Figure C.10 Response plots of Pocillopora meandrina probability of occurrence to slope from eight model methods ...... 93

Figure C.11 Response plots of Pocillopora meandrina probability of occurrence to aspect from eight model methods ...... 93

Figure C.12 Response plots of Pocillopora meandrina probability of occurrence to rugosity from eight model methods ...... 94

Figure C.13 Response plots of Pocillopora meandrina probability of occurrence to sandbottom from eight model methods...... 94

Figure C.14 Response plots of Pocillopora meandrina probability of occurrence to maximum significant wave height from eight model methods ...... 95

Figure C.15 Response plots of Pocillopora meandrina probability of occurrence to mean significant wave height from eight model methods ...... 95

Figure C.16 Response plots of Pocillopora meandrina probability of occurrence to downwelled irradiance from eight model methods ...... 96

Figure C.17 Response plots of Porites compressa probability of occurrence to depth from eight model methods ...... 96

Figure C.18 Response plots of Porites compressa probability of occurrence to slope from eight model methods ...... 97

xii

Figure C.19 Response plots of Porites compressa probability of occurrence to aspect from eight model methods ...... 97

Figure C.20 Response plots of Porites compressa probability of occurrence to rugosity from eight model methods ...... 98

Figure C.21 Response plots of Porites compressa probability of occurrence to sandbottom from eight model methods...... 98

Figure C.22 Response plots of Porites compressa probability of occurrence to maximum significant wave height from eight model methods ...... 99

Figure C.23 Response plots of Porites compressa probability of occurrence to mean significant wave height from eight model methods ...... 99

Figure C.24 Response plots of Porites compressa probability of occurrence to downwelled irradiance from eight model methods ...... 100

Figure C.25 Response plots of Porites lobata probability of occurrence to depth from eight model methods ...... 100

Figure C.26 Response plots of Porites lobata probability of occurrence to slope from eight model methods ...... 101

Figure C.27 Response plots of Porites lobata probability of occurrence to aspect from eight model methods ...... 101

Figure C.28 Response plots of Porites lobata probability of occurrence to rugosity from eight model methods ...... 102

Figure C.29 Response plots of Porites lobata probability of occurrence to sandbottom from eight model methods ...... 102

Figure C.30 Response plots of Porites lobata probability of occurrence to maximum significant wave height from eight model methods ...... 103

Figure C.31 Response plots of Porites lobata probability of occurrence to mean significant wave height from eight model methods ...... 103

Figure C.32 Response plots of Porites lobata probability of occurrence to downwelled irradiance from eight model methods ...... 104

Figure D.1 Benthic cover field observations for Montipora capitata cover (%) in the main Hawaiian Islands from 2000-2009. See Fig. 3.1 for accurate depiction of geographic scale and position for Hawaiian Islands ...... 105

xiii

Figure D.2 Benthic cover field observations for Montipora flabellata cover (%) in the main Hawaiian Islands from 2000-2009. See Fig. 3.1 for accurate depiction of geographic scale and position for Hawaiian Islands...... 106

Figure D.3 Benthic cover field observations for Montipora patula cover (%) in the main Hawaiian Islands from 2000-2009. See Fig. 3.1 for accurate depiction of geographic scale and position for Hawaiian Islands ...... 106

Figure D.4 Benthic cover field observations for Pocillopora meandrina cover (%) in the main Hawaiian Islands from 2000-2009. See Fig. 3.1 for accurate depiction of geographic scale and position for Hawaiian Islands...... 107

Figure D.5 Benthic cover field observations for Porites compressa cover (%) in the main Hawaiian Islands from 2000-2009. See Fig. 3.1 for accurate depiction of geographic scale and position for Hawaiian Islands ...... 107

Figure D.6 Benthic cover field observations for Porites lobata cover (%) in the main Hawaiian Islands from 2000-2009. See Fig. 3.1 for accurate depiction of geographic scale and position for Hawaiian Islands ...... 108

Figure E.1 Shallow bathymetry (0 to 30 m depth) of the waters around Kauai and Niihau...... 109

Figure E.2 Shallow bathymetry (0 to 30 m depth) of the waters around Maui Nui ...... 109

Figure E.3 Shallow bathymetry (0 to 30 m depth) of the waters around Hawaii ...... 110

Figure E.4 Bathymetric slope (in degrees) of the sea floor around Kauai and Niihau ....110

Figure E.5 Bathymetric slope (in degrees) of the sea floor around Maui Nui ...... 111

Figure E.6 Bathymetric slope (in degrees) of the sea floor around Hawaii ...... 111

Figure E.7 Bathymetric aspect (in degrees) of the sea floor around Kauai and Niihau ..112

Figure E.8 Bathymetric aspect (in degrees) of the sea floor around Maui Nui ...... 112

Figure E.9 Bathymetric aspect (in degrees) of the sea floor around Hawaii ...... 113

Figure E.10 Bathymetric rugosity (surface area / planar area) of the sea floor around Kauai and Niihau ...... 113

Figure E.11 Bathymetric rugosity (surface area / planar area) of the sea floor around Maui Nui ...... 114

xiv

Figure E.12 Bathymetric rugosity (surface area / planar area) of the sea floor around Hawaii ...... 114

Figure E.13 Sand habitat (proportion) of the sea floor around Kauai and Niihau ...... 115

Figure E.14 Sand habitat (proportion) of the sea floor around Maui Nui ...... 115

Figure E.15 Sand habitat (proportion) of the sea floor around Hawaii ...... 116

Figure E.16 Maximum significant wave heights for the waters around Kauai and Niihau ...... 116

Figure E.17 Maximum significant wave heights for the waters around Maui Nui...... 117

Figure E.18 Maximum significant wave heights for the waters around Hawaii ...... 117

Figure E.19 Mean significant wave heights for the waters around Kauai and Niihau ....118

Figure E.20 Mean significant wave heights for the waters around Maui Nui ...... 118

Figure E.21 Mean significant wave heights for the waters around Hawaii ...... 119

Figure E.22 Downwelled irradiance at the sea floor relative to that just below the sea surface around Kauai and Niihau ...... 119

Figure E.23 Downwelled irradiance at the sea floor relative to that just below the sea surface around Maui Nui ...... 120

Figure E.24 Downwelled irradiance at the sea floor relative to that just below the sea surface around Hawaii ...... 120

Figure F.1 Geographic distribution of model prediction for Montipora capitata cover (%) around Kauai and Niihau ...... 121

Figure F.2 Geographic distribution of model prediction for Montipora capitata cover (%) around Oahu ...... 121

Figure F.3 Geographic distribution of model prediction for Montipora capitata cover (%) around Maui Nui ...... 122

Figure F.4 Geographic distribution of model prediction for Montipora capitata cover (%) around Hawaii ...... 122

Figure F.5 Geographic distribution of model prediction for Montipora flabellata cover (%) around Kauai and Niihau ...... 123

Figure F.6 Geographic distribution of model prediction for Montipora flabellata cover (%) around Oahu ...... 123

Figure F.7 Geographic distribution of model prediction for Montipora flabellata cover (%) around Maui Nui ...... 124

Figure F.8 Geographic distribution of model prediction for Montipora flabellata cover (%) around Hawaii ...... 124

Figure F.9 Geographic distribution of model prediction for Montipora patula cover (%) around Kauai and Niihau ...... 125

Figure F.10 Geographic distribution of model prediction for Montipora patula cover (%) around Oahu ...... 125

Figure F.11 Geographic distribution of model prediction for Montipora patula cover (%) around Maui Nui ...... 126

Figure F.12 Geographic distribution of model prediction for Montipora patula cover (%) around Hawaii ...... 126

Figure F.13 Geographic distribution of model prediction for Pocillopora meandrina cover (%) around Kauai and Niihau ...... 127

Figure F.14 Geographic distribution of model prediction for Pocillopora meandrina cover (%) around Oahu ...... 127

Figure F.15 Geographic distribution of model prediction for Pocillopora meandrina cover (%) around Maui Nui ...... 128

Figure F.16 Geographic distribution of model prediction for Pocillopora meandrina cover (%) around Hawaii ...... 128

Figure F.17 Geographic distribution of model prediction for Porites compressa cover (%) around Kauai and Niihau ...... 129

Figure F.18 Geographic distribution of model prediction for Porites compressa cover (%) around Oahu ...... 129

Figure F.19 Geographic distribution of model prediction for Porites compressa cover (%) around Maui Nui ...... 130

Figure F.20 Geographic distribution of model prediction for Porites compressa cover (%) around Hawaii ...... 130

xvi

Figure F.21 Geographic distribution of model prediction for Porites lobata cover (%) around Kauai and Niihau ...... 131

Figure F.22 Geographic distribution of model prediction for Porites lobata cover (%) around Oahu ...... 131

Figure F.23 Geographic distribution of model prediction for Porites lobata cover (%) around Maui Nui ...... 132

Figure F.24 Geographic distribution of model prediction for Porites lobata cover (%) around Hawaii ...... 132

Figure H.1 Partial dependence response plots for environmental variables in model for Montipora capitata cover. Relative percent contribution of each variable, scaled to 100, is in parenthesis of x-axis label. Rug plots along inside top of plots show distribution of sites across that variable, in deciles ...... 138

Figure H.2 Partial dependence response plots for environmental variables in model for Montipora flabelata cover. Relative percent contribution of each variable, scaled to 100, is in parenthesis of x-axis label. Rug plots along inside top of plots show distribution of sites across that variable, in deciles ...... 139

Figure H.3 Partial dependence response plots for environmental variables in model for Montipora patula cover. Relative percent contribution of each variable, scaled to 100, is in parenthesis of x-axis label. Rug plots along inside top of plots show distribution of sites across that variable, in deciles ...... 139

Figure H.4 Partial dependence response plots for environmental variables in model for Pocillopora meandrina cover. Relative percent contribution of each variable, scaled to 100, is in parenthesis of x-axis label. Rug plots along inside top of plots show distribution of sites across that variable, in deciles ...... 140

Figure H.5 Partial dependence response plots for environmental variables in model for Porites compressa cover. Relative percent contribution of each variable, scaled to 100, is in parenthesis of x-axis label. Rug plots along inside top of plots show distribution of sites across that variable, in deciles…………………………………………………….140

Figure H.6 Partial dependence response plots for environmental variables in model for Porites lobata cover. Relative percent contribution of each variable, scaled to 100, is in parenthesis of x-axis label. Rug plots along inside top of plots show distribution of sites across that variable, in deciles ...... 141

xvii

LIST OF ABBREVIATIONS

ANN Artificial neural networks

AUC Area under the curve of the receiver-operator characteristic plot

BACIPS Before-After, Control-Impact Paired Series

BRT Boosted regression tree

CTA Classification tree analysis

CS Customary stewardship

DLNR Department of Land and Natural Resources, State of Hawaii

FDA Flexible discriminant analysis

GAM Generalized additive model

GBM Generalized boosted regression model

GLM Generalized linear model

HIHWNMS Hawaiian Islands Humpback Whale National Marine Sanctuary

HMLR Hawaii Marine Laboratory Refuge (Coconut Island)

IUCN International Union for Conservation of Nature

MARS Multivariate adaptive regression spline

MHI Main Hawaiian Islands

MLCD Marine life conservation district

MPA Marine protected area

NAR Natural Area Reserve

NDBC National Data Buoy Center

NMPAC National Marine Protected Area Center

xviii

NOAA National Oceanic and Atmospheric Administration

NODC National Oceanographic Data Center

NT No-take

PP Partial protection

RF Random forests

ROC Receiver-operator characteristic

SDM Species distribution models/modeling

SRE Surface response envelope

WCPA World Commission on Protected Areas

xix

CHAPTER 1

GENERAL INTRODUCTION

What is the impact of climate change on coral reefs? What might the reefs of the future

look like? Can we identify resilient reef communities? These questions are at the

forefront of contemporary coral reef science and management issues. Yet in order to

properly address these questions, we first need to improve our understanding of the

present conditions of coral populations that comprise the foundation of reef ecosystems

as a baseline to compare against future conditions. Furthermore, these examinations need

to be conducted at broad enough spatial and temporal extents to adequately represent the population structure and dynamics of each species under study.

Scleractinian corals are the foundation species of tropical and subtropical reefs, yet information about their status is woefully inadequate. For example, only 5 of 845 coral species had sufficient species-specific population trend data to recently evaluate their extinction risk using the associated IUCN Red List criteria (Carpenter et al. 2008).

Remote sensing technology has enabled global and regional-scale mapping of shallow

coral reefs (Mumby et al. 2004, Mora et al. 2006) yet the sensors only allow interpretation of habitat-level (e.g., patch reef, fore reef, etc.) or functional group-level

(e.g., coral, algae, sand) information and cannot differentiate between individual species

(Mumby et al. 2004, Goodman and Ustin 2007). Field surveys can provide information at

a species level but are often limited to a small set of geographic locations. As a method to integrate the strengths of the different approaches to improve the biological

characterization of reefs, species distribution models (SDMs) can incorporate field

observations and environmental covariates from observational, remotely sensed, or model

data into statistical models that predict macroecological-scale, spatially-continuous

distributions of coral species (Guisan and Thuiller 2005, Austin 2007, Elith and

Leathwick 2009).

Widely utilized for modeling species in many ecosystems (Elith and Leathwick

2009, Ready et al. 2010, Robinson et al. 2011), SDMs have less frequently been applied to coral reef species but have been used to predict distributions of biological functional

groups and habitat types (Garza-Pérez et al. 2004, Guinette et al. 2006, Chollett and

Mumby 2012), and coral reef community metrics (Harborne et al. 2006, Pittman et al.

2009, Knudby et al. 2010). SDMs are constructed by building a representation of the

realized species niche and extrapolating the niche requirements into geographical space

(Guisan et al. 2007, Elith and Leathwick 2009, Peterson et al. 2011). Comparative analysis of population condition and geographic distribution across a range of temporal

and spatial scales are possible with SDMs (Guisan et al. 2007, Elith and Leathwick 2009,

Peterson et al. 2011). Using species distribution models, this dissertation examines the

physical and biological factors that influence the distribution of six dominant

scleractinian species, Montipora capitata, Montipora flabellata, Montipora patula,

Pocillopora meandrina, Porites compressa, and Porites lobata in the main Hawaiian

Islands (Jokiel et al. 2004). The primary objectives of the dissertation are to

 compile a database of quantitative field observations of the six coral species from

reef surveys and spatial environmental covariate data in the main Hawaiian

Islands during 2000-2009,

 identify models and environmental factors that are most informative for predicting

the distributions of the six coral species in the main Hawaiian Islands using the

field observations for model training and validation,

 use the model outputs to map spatially-explicit presence and abundances of each

coral species for near-shore, shallow reefs (~30 m depth), and

 compare the abundance of each coral species in a network of MPAs to

unprotected reefs from the spatial abundance maps.

The dissertation follows the general formatting guidelines for the University of Hawaii at

Manoa Biology Department. The introductory section provides an overview of the research activities with the structure of each chapter as a scientific article. The content of the three research chapters follow in more detail.

The second chapter, “An ensemble model approach to species distribution modeling of corals on Hawaiian Reefs”, details the initial application of the SDM approach to coral species in Hawaii. Using an ensemble suite of methods including machine learning, regression and discriminant analysis methods, the environmental variables that influenced the presence of the four dominant coral species (Porites compressa, P. lobata, Pocillopora damicornis, and Montipora capitata) were modeled

for the reefs around the island of Oahu. Environmental variables included bathymetry,

rugosity, downwelled irradiance, max significant wave height, mean significant wave

height, sandbottom, aspect, and slope. Using a dataset of compiled reef observations for

2000-2009, the models results demonstrate the different environmental niches and

geographic distribution of each coral species. Model performances were good to excellent

3 and provided a foundation to proceed with an expanded study for the main Hawaiian

Islands.

In Chapter 3, “Predictive modeling of coral distribution and abundance in the

Hawaiian Islands”, the SDM approach was expanded to include the six most dominant coral species (Porites compressa, P. lobata, Pocillopora damicornis, Montipora capitata,

M. flabellata, and M. patula) for the eight main Hawaiian islands. Coral abundance (as benthic cover) for each species was modeled as the response variable against the same set of environmental variables from Chapter 2 in addition to an “island” variable that accounted for potential interisland geographic differences. Using boosted regression trees, the models of coral species abundances were produced and projected to the geography of the shallow coral reefs of the Hawaiian Islands. From the geographic distributions of abundance, rank orders of abundance were calculated for each species for the entire main Hawaiian islands and individually for each island. From a summation of the abundance of the six coral species, hot spots of high coral cover were identified throughout the MHI. These results provide a detailed quantitative and geographic perspective on the distribution and abundance of coral species in Hawaii that has not previously been reported.

The fourth chapter, “A niche model analysis of corals in Hawaii: are populations more abundant in marine protected areas?”, utilizes the results of the SDM modeling for the six species from Chapter 3 to compare the coral populations within and outside of

Hawaiian MPAs. The MPA network was evaluated as effective if the coral species abundance inside the MPA was equivalent to or greater than the coral abundance in non- protected reef areas. Using this criteria, the coral abundances within MPAs by island

groups as well as within each of 12 MPAs (e.g., 9 Marine Life Conservation Districts, a

Natural Area Reserve, an Island Reserve, and a Marine Life Refuge) were compared to

those outside of the protected areas. The coral abundances were also evaluated within the

boundaries of the Hawaiian Islands Humpback Whales National Marine Sanctuary to

support their ongoing management review process. This chapter provides information to

support potential future marine resource spatial management actions for coral species in

the Hawaiian archipelago.

The dissertation concludes with a summary chapter that synthesizes the content of

the research, suggests future investigations to follow up on this work, and provides a brief

conclusion section for the scientific study.

References

Austin M (2007) Species distribution models and ecological theory: a critical assessment and some possible new approaches Ecological Modelling 200:1-19 Carpenter KE, Abrar M, Aeby G, Aronson RB, Banks S, Bruckner A, Chiriboga A, Cortes J, Delbeek JC, DeVantier L, Edgar GJ, Edwards AJ, Fenner D, Guzmán H, Hoeksema BW, Hodgson G, Johan O, Licuanan WY, Livingstone SR, Lovell ER, Moore JA, Obura DO, Ochabvillo D, Polidoro BA, Precht WF, Quibilan MC, Reboton C, Richards ZT, Rogers AD, Sanciangco J, Sheppard A, Sheppard C, Smith J, Stuart S, Turak E, Veron JEN, Wallace C, Weil E, Wood E (2008) One- third of reef-building corals face elevated extinction risk from climate change and local impacts. Science 321:560-563 Chollett I, Mumby PJ (2012) Predicting the distribution of Montastrea reefs using wave exposure. Coral Reefs 31:493-503 Elith J, Leathwick JR (2009) Species distribution models: ecological explanation and prediction across space and time. Ann Rev Ecol Evol Sys 40:677-697 Garza-Pérez JR, Lehmann A, Arias-González JE (2004) Spatial prediction of coral reef habitats: integrating ecology with spatial modeling and remote sensing. Mar Ecol Prog Ser 269:141-152 Goodman J, Ustin SL (2007) Classification of benthic composition in a coral reef environment using spectral unmixing. J Appl Remote Sens 1:011501 Guinette JM, Bartley JD, Iqbal A, Fautin DG, Buddemeier RW (2006) Modeling habitat

distribution from organism occurrences and environmental data: case study using anemonefishes and their sea anemone hosts. Mar Ecol Prog Ser 316:269-283 Guisan A, Thuiller W (2005) Predicting species distribution: offering more than simple habitat models. Ecology Letters 8:993-1009 Guisan A, Zimmermann NE, Elith J, Graham CH, Phillips S, Peterson AT (2007) What matters for predicting the occurrences of trees: techniques, data, or species characteristics? Ecological Monographs 77:615-630 Harborne AR, Mumby PJ, Zychaluk K, Hedley JD, Blackwell PG (2006) Modeling the beta diversity of coral reefs. Ecology 87:2871-2881 Jokiel PL, Brown EK, Friedlander A, Rodgers SK, Smith WR (2004) Hawaii coral reef assessment and monitoring program: spatial patterns and temporal dynamics in reef coral communties. Pacific Science 58:159-174 Knudby A, LeDrew E, Brenning A (2010) Predictive mapping of reef fish species richness, diversity and biomass in Zanzibar using IKONOS imagery and machine- learning techniques. Remote Sensing of Environment 114: 1230–1241 Mora C, Andréfouët S, Kranenburg S, Rollo A, Costello M, Veron J, Gaston KJ, Myers RA (2006) How protected are coral reefs? Science 314:757-760 Mumby PJ, Skirving W, Strong AE, Hardy JT, LeDrew EE, Hochberg EJ, Stumpf RP, David LT (2004) Remote sensing of coral reefs and their physical environment. Mar Pollut Bull 48:219-228 Peterson AT, Soberón J, Pearson RG, Anderson RP, Martinez-Meyer E, Nakamura M, Araújo MB (2011) Ecological niches and geographic distributions. Princeton University Press Pittman SJ, Costa BM, Battista TA (2009) Using lidar bathymetry and boosted regression trees to predict the diversity and abundance of fish and corals. Journal of Coastal Research 25: 27–38 Ready J, Kaschner K, South AB, Eastwood PD, Rees T, Rius J, Agbayani E, Kullander S, Froese R (2010) Predicting the distribution of marine organisms at the global scale. Ecol Modelling 221:467-478 Robinson LM, Elith J, Hobday AJ, Pearson RG, Kendall BE, Possingham HP, Richardson AJ (2011) Pushing the limits in marine species distribution modelling: lessons from the land present challenges and opportunities. Global Ecol Biogeo 20:789-802

CHAPTER 2

AN ENSEMBLE MODEL APPROACH TO SPECIES DISTRIBUTION MODELING

OF CORALS ON HAWAIIAN REEFS

EC Franklin,

PL Jokiel, MJ Donahue

Hawaii Institute of Marine Biology, School of Ocean and Earth Science and Technology,

University of Hawaii, Kaneohe, Hawaii 96744 USA

Submitted to CORAL REEFS as a REPORT

Abstract

This study used an ensemble model approach to develop species distribution models

(SDMs) of the four dominant Hawaiian coral species (Montipora capitata, Porites

compressa, Porites lobata, and Pocillopora meandrina) around the island of Oahu,

Hawaii, USA. We incorporated a diverse suite of model approaches including regression-

based, machine learning, and discriminant analysis methods to integrate in situ coral

surveys with environmental data of wave heights, benthic geomorphology, and

downwelled irradiance to predict species distributions. Models were fitted and evaluated

using a standard split-sample strategy (70% train/ 30% test) of the coral species

observation dataset with the training-validation process repeated 100 times for each

species and model. For each species, a consensus model was constructed with the mean

weighted probability of occurrence from all model runs ranked by predictive performance

of each modeling method from the ensemble. Model accuracy was evaluated using the

area under the curve (AUC) of a receiver-operator characteristic (ROC) plot. Mean

significant wave height, max significant wave height, depth, downwelled irradiance,

bathymetric slope and aspect were the most important variables influencing the predicted

species distributions. High probability of occurrence shifts from P. compressa to P.

meandrina and P. lobata at maximum significant wave heights between 1-2 m. Model

accuracy on evaluation datasets ranged from good to excellent predictive ability (AUC

0.8-0.9). The application of SDMs through an ensemble model approach has significant

potential to address a critical need for realistic and accurate species distribution

information for the conservation and management of coral reefs.

Introduction

Understanding the distribution of a species is a fundamental aspect of marine conservation and ecology. A recent shift in marine resource management toward ecosystem-based approaches and marine spatial planning drives a growing requirement for accurate depictions of spatially-explicit and biologically-realistic species distributions

(Arkema et al. 2006, Crowder and Norse 2008, Klein et al. 2010). For coral reefs, the need to assess species distributions is critical given the unique role of particular corals as habitat, prey, or host for reef associated organisms (Munday et al. 1997, Stat et al. 2008,

Graham et al. 2011, Wilson et al. 2011). Advances in remote sensing have facilitated regional-scale mapping of shallow coral reefs (Mumby et al. 2004) yet the technology can only detect biological information at a habitat (e.g., patch reef, fore reef, etc.) or functional group level (e.g., coral, algae, sand) but not for individual species (Mumby et al. 2004, Goodman and Ustin 2007). Field survey data can provide information at a species level but is often collected only from a small set of geographic locations. To overcome these limitations, species distribution modeling (SDMs) offer an approach that can incorporate biological field observations and environmental covariates from observational, remotely sensed, or model data into statistical models that predict spatially continuous distributions of species (Guisan and Thuiller 2005, Austin 2007, Elith and

Leathwick 2009).

SDM is the process of building a representation of the realized niche requirements for a species and extrapolating these requirements into a geographical region (Guisan et al. 2007, Elith and Leathwick 2009, Peterson et al. 2011). The realized niche describes

9 the intersection of a multi-dimensional array of environmental conditions that are suitable for a species to persist constrained by biotic interactions and disturbances that limit the species full occupancy of its fundamental niche (Hutchinson 1957, Leibold 1995, Chase and Leibold 2003, Elith and Leathwick 2009). Ecological-niche models of species’ distributions provide a framework for comparative analysis of populations and ecosystem processes across a range of temporal and spatial scales (Guisan et al. 2007, Elith and

Leathwick 2009, Peterson et al. 2011). Widely utilized for modeling species in many ecosystems (Elith and Leathwick 2009, Ready et al. 2010, Robinson et al. 2011), SDMs have less frequently been applied to coral reef species although they have been used to predict distributions of biological functional groups and habitat types (Garza-Pérez et al.

2004, Guinette et al. 2006, Chollett and Mumby 2012) as well as coral and reef fish community metrics (Harborne et al. 2006, Pittman et al. 2009, Knudby et al. 2010,

Pittman and Brown 2011). In order to construct SDMs, a set of relevant, spatially and temporally coincident environmental data layers need to be available.

Coral species distributions are influenced by a number of environmental factors such as wave energy, benthic geomorphology, and turbidity. In the Hawaiian Islands, disturbance from waves is the primary factor that structures coral communities (Dollar

1982, Grigg 1983, Jokiel et al. 2004, Engels et al. 2004, Storlazzi et al. 2005). Dollar

(1982) found that wave energy and storm frequency strongly influenced the vertical zonation of coral species dominance at a site on the Kailua-Kona coast of Hawaii. He identified four distinct reef zones from shallow inshore to deeper offshore areas:

Pocillopora meandrina boulder zone, Porites lobata reef bench zone, Porites compressa slope zone, and P. lobata rubble zone (Dollar 1982). Grigg (1983) described coral

10 community composition for the islands and atolls of the Hawaiian archipelago and identified P. lobata, P. compressa, Montipora capitata (identified as M. verrucosa), and

P. meandrina as the ecologically-dominant coral species by rank order of abundance from offshore southwest-facing reefs. From a multivariate analysis of extensive monitoring survey data for main Hawaiian Island reefs, Jokiel et al. (2004) identified wave height and direction as well as depth, rugosity, and organic sediment content (an indicator of turbid, low light environments) as major factors that structure Hawaiian coral communities. Engels et al. (2004) expanded upon Dollar’s (1982) classification model of coral zonation by using wave-induced near bed shear stress and water depth to explain P. meandrina, P. compressa, P. lobata, and Montipora sp. benthic cover, species dominance, and coral morphologies for the south shore of Molokai, an island in the main

Hawaiian Islands. Using wave-induced near bed shear stress and depth, Storlazzi et al.

(2005) developed a quantitative model of wave control on coral breakage and species distribution for M. capitata, P. compressa, P. lobata, and P. meandrina around Molokai, north Lanai, and northwest Maui (Storlazzi et al. 2005). The model was used to identify geographic areas of refuge from the particular wave energy threshold that would induce breakage in each coral species (Storlazzi et al. 2005). These works identified the dominant coral species in Hawaii and a set of environmental drivers that structure reef communities.

For this research, we employ an ensemble model approach to develop species distribution models of the four dominant Hawaiian coral species (M. capitata, P. compressa, P. lobata, and P. meandrina) around the island of Oahu. Using the BIOMOD package in R (Thuiller et al. 2009), we incorporate a diverse suite of model approaches

including regression-based, machine learning, and discriminant analysis methods to

integrate in situ coral surveys with environmental data of wave exposure, benthic

geomorphology, and downwelled irradiance to predict species distributions. Using an

ensemble approach, we construct a consensus model for each species from the set of

model runs that explicitly incorporates the uncertainty inherent in choosing between

modeling methods (Guisan and Thuiller 2005, Thuiller et al. 2009). We discuss the

geographic distributions of the coral species and the set of environmental factors most

prominently used by the models to construct the distributions. We conclude with a

discussion of applications for the distribution data in marine conservation planning and

future research directions in coral SDM research.

Materials and methods

Study Area

The Hawaiian Islands are a volcanic chain of islands and atolls in the central north

Pacific Ocean that extend in a northwest – southwest axis over approximately 2,500 km

(Fig. 2.1, Fletcher et al. 2008). The eight main Hawaiian Islands have a human population of approximately 1.4 million with 70% of people concentrated in the state capital, Honolulu, on the island of Oahu (US Census Bureau 2010). The geography of

Oahu is characterized by prominent coastal capes and headlands such as Kaena Point,

Mokapu Peninsula, Makapuu Point, and Barbers Point that demarcate coastal exposures to different climate and ocean conditions (Fig. 2.1). Oahu’s north coast is exposed to large northern hemisphere winter swells (≥ 7 m), while the south shore experiences wave activity in summer (Fletcher et al. 2008). The eastern or windward side of the island experiences consistent easterly tradewinds (10-20 kn) that generate steady wind waves

(Fletcher et al. 2008). Pearl Harbor and Kaneohe Bay are the only large, natural semi- enclosed waters bodies in the main Hawaiian Islands (Fig. 2.1). Pearl Harbor and

Kaneohe Bay experience minimal wave activity but typically sustain turbid conditions from wind-driven benthic sediment resuspension and nearshore inputs such as adjacent watershed runoff from storms (Hunter and Evans 1995, Coles et al. 1997, Jokiel 2006).

Coral reefs are found around the entire island with the majority of reefs found in Kaneohe

Bay and along the north and east coasts (Battista et al. 2007).

Coral species observations

We compiled a species occurrence database of four Hawaiian coral species (M. capitata,

P. meandrina, P. compressa, P. lobata) from scientific monitoring programs (Brown et

al. 2004, Brown et al. 2007, NOAA 2005, NOAA 2011) and research project data

archived in the National Oceanographic Data Center (NODC 2011). The compilation

included 4,675 total observations recorded during 2000-2009 on coral reefs between 0-30

m depth around the island of Oahu (Fig. 2.2). Survey methods included in-situ diver

observations and interpreted photo-quadrats for survey areas ranging from 0.25 m2 to 25 m2 that also provided coordinates of latitude and longitude for each observation. Using

the location information, we mapped the presence and absence observations for each coral species as vector point features and converted those to raster grids georectified to

the 50 m resolution base analysis grid. No significant correlation was observed between

sampled area within a grid cell and species presence. Presence and absence of each coral

species were determined for grid cells with at least one survey. Cells were assigned a

species presence if any of the surveys contained within the cell included a presence observation of the coral species (Table 2.1). Full model results presented here did not

differ substantially from preliminary model runs with randomized subset selections with

a prevalence of 0.5 for presence and absence cells (Jiménez-Valverde et al. 2009). Data

manipulation was performed using base functions in R (R Development Core Team

2010), ArcGIS (v. 9.3.1, ESRI 2009), and geoprocessed using scripts in Python

(http://www.python.org).

(a) (b)

Figure 2.2 Presence and absence field observations for (a) Montipora capitata, (b) Pocillopora meandrina, (c) Porites compressa, and (d) Porites lobata from 2000-2009.

Table 2.1 Number of presence and absence observations and model grid cells for four coral species around Oahu. Cells were assigned a species presence if any of the surveys contained within the cell included a presence observation of the coral species. Field Observations (#) Model Grid Cells (#) Species Presence Absence Presence Absence Montipora capitata 458 366 147 220 Pocillopora meandrina 243 581 134 233 Porites compressa 363 1840 111 905 Porites lobata 302 522 158 209

Environmental Data Layers

Eight environmental covariate data layers including depth, bathymetric aspect,

bathymetric rugosity, bathymetric slope, proportion of sand bottom (in relation to

hardbottom), maximum significant wave height, mean significant wave height, and downwelled irradiance were derived from empirical observations or model output (Table

2.2). Digital files for all environmental data layers were georectified to a base analysis grid of 139,465 cells (approximately 348 km2) that covered the extent of the study

domain in ArcGIS (v. 9.3.1, ESRI 2009) and geoprocessed using scripts in Python

(http://www.python.org). Detailed figures of the environmental data layers for Oahu are

in Appendix A.

Depth for the study domain around Oahu was determined from a bathymetry

synthesis for the main Hawaiian Islands (Hawaii Mapping Research Group 2011). The

horizontal resolution of the bathymetry synthesis was approximately 50 m (0.0005

degrees) and the extent covered most of the study domain. For cells that contained no

bathymetric data, depths recorded from NOAA National Geodetic Survey soundings and

coral reef survey observations were used to fill gaps where possible. After gap filling

with empirical depth observations, we used an iterative nearest neighbor method, in an 8-

cell neighborhood, to calculate depth for no data cells using the average depth of the

neighborhood to create a no gaps bathymetry file. This method was used for

approximately 0.6% of study grid cells.

Three measures of benthic geomorphology (slope, aspect, and rugosity) were

derived from the bathymetry data layer. Bathymetric slope was the steepest angle,

measured in degrees, of a plane defined for a depth grid cell and its surrounding eight

neighbors. Bathymetric aspect was the steepest downslope direction, measured in

compass degrees (0 o - 360o) of a plane defined by the slope grid cell and its eight

surrounding neighbors. Bathymetric rugosity was the ratio between the surface area and

the planimetric area of the depth grid cell and its eight surrounding neighbors.

Hardbottom and sandbottom habitat areas delineated from interpreted satellite imagery (Battista et al. 2007) were converted from digital polygon features to 5 m resolution raster grids. Sandbottom areas included sand, mud, and silt habitats. Bottom habitat raster cells (at 5 m resolution) were then summed within the cells of the basemap grid (at ~50 m resolution) to derive a proportion of sand cover layer.

Spectral wave data from WAVEWATCH III (WW3 v3.14, Tolman 2009) for every 6 hours during January 2000-December 2009 was used to force a SWAN hindcast model (v 40.51, SWAN Team, 2006) to obtain parametric wave data for Oahu. Maximum significant wave height, max Hs, and mean significant wave height, mean Hs, were

estimated for the 10-year period at a grid resolution of 0.005 degrees which was

resampled to 0.0005 degrees using an 8-cell nearest neighbor smoothing algorithm on

mean values. Results were validated from a comparison of computed and measured Hs values at NOAA/NDBC Buoys 51201 and 51202 which demonstrated good overall correlation (r =0.9) with a slight underestimate in modeled Hs values (Arinaga and

Cheung 2012).

Downwelled irradiance was modeled using the Beer-Lambert law in the form:

- -KdZ Ed(Z) = Ed(0 )e where Ed(Z) is the downwelled irradiance at depth Z determined from

- the bathymetry data layer, Ed(0 ) is the irradiance just below the sea surface, and Kd is the diffuse attenuation coefficient (Kirk 1994). A diffuse attenuation coefficient (Kd) for

PAR (photosynthetically active radiation, 400-700 nm) of 0.054 was used for coastal waters greater than 10 m depth, 0.212 for coastal waters shallower than 10 m, and 0.273

for lagoonal waters including Kaneohe Bay, Pearl Harbor, and Keehi Lagoon (Connolly

et al. 1999; Isoun et al. 2003; Jacobson 2005). A digital file of downwelled irradiance,

- Ed(Z) / Ed(0 ), was calculated as the proportion of downwelled irradiance at depth Z from

the bathymetry data file to the irradiance just below the surface.

Table 2.2 Summary statistics of environmental covariates around Oahu. Variable Mean SD Range Unit Aspect 170.9 104.8 0.0 – 360.0 ° Depth -10.4 8.1 -30.0 – 0.0 m Max significant wave height 2.5 1.4 0.00 – 7.1 m Mean significant wave height 1.0 0.4 0.00 - 2.1 m Downwelled irradiance 0.4 0.2 0.0 – 1.0 proportion Rugosity 1.0006 0.0018 1.0 – 1.0556 ratio Sandbottom 0.24 0.40 0.0 – 1.0 proportion Slope 1.2 1.3 0 – 14.4 °

Statistical Modeling

Models were fitted and evaluated using a standard split-sample strategy of the coral

species observation dataset. We used the BIOMOD package in R (Thuiller et al. 2009),

fitting for each species artificial neural networks (ANN, Ripley 1996), classification tree

analysis (CTA, Breiman et al. 1984), flexible discriminant analysis (FDA), generalized

additive models (GAM, Hastie and Tibshirani 1990), generalized boosted regression models (GBM, Ridgeway 1999), generalized linear models (GLM, McCullagh and

Nelder 1989), multivariate adaptive splines (MARS, Friedman 1991), random forests

(RF, Breiman 2001), and surface response envelopes (SRE, Busby 1991). Each model

run was trained using randomly-selected 70% subsets of occurrence data and validated

with the remaining 30% test data. The training-validation process was repeated 100 times for each species and model. We tested the predictive power of the models using the area under the curve (AUC) of a receiver-operator characteristic (ROC) plot (Fielding and

Bell 1997, Elith et al. 2006). Predictions were considered no better than random at an

AUC of 0.5, poor between the values of 0.5–0.7, acceptable from 0.7-0.8, good from 0.8-

0.9, and excellent above 0.9 with a maximum value of 1.0 representing a perfect classification (after Hosmer and Lemeshow 2000). The relative importance of model variables was determined by a randomization procedure that evaluated correlation coefficients between models with and without each environmental variable individually randomized 100 times. A strong positive correlation indicated a lack of importance for the variable to the predictive capability of the model (Thuiller et al. 2009). For each species, a consensus model was constructed with the mean weighted probability of occurrence from all model runs ranked by predictive performance of each modeling method from the ensemble. The probabilities of occurrence were then projected to geographic space using the model relationships on the set of environmental covariate maps.

Results

The models predicted the geographic distributions of the four coral species (Fig. 2.3) with better model accuracies for P. compressa and P. lobata than M. capitata and P. meandrina (Table 2.3). Mean Hs, max Hs, depth, light, slope, and aspect were the most

important variables to explain the distributions of the four species. A table of relative

variable importance for each environmental predictor in all species distribution models is

in Appendix B. M. capitata was predicted to occur along the north- and east-facing

coasts, at Maile Point and Barbers Point, and in wave-sheltered Kaneohe Bay and Pearl

Harbor (Fig. 2.3a). Mean Hs, depth, and downwelled irradiance were the most important

variables explaining the distribution of M. capitata. P. meandrina was predicted along all

coasts of Oahu especially in areas of high maximum Hs and steep slopes such as Kaena

Point, Makapuu Point, and Mokapu peninsula (Fig. 2.3b). Models predicted the highest

probability of occurrence for P. compressa in areas with low max Hs, mean Hs, and

downwelled irradiance such as Kaneohe Bay and Pearl Harbor (Fig, 2.3c). P. lobata was

predicted to occur along all coasts of Oahu, especially along north- and east-facing coasts

that experience relatively high max Hs and mean Hs with depth also an important variable

(Fig. 2.3d). Using max Hs alone, a strong transition in species dominance was identified

for max Hs from P. compressa at max Hs less than 1 m to P. meandrina and P. lobata at

greater than 2 m max Hs (Fig. 2.4). Response plots for all models and coral species are available in Appendix C.

Model accuracy on evaluation datasets ranged from good to excellent predictive ability (Table 2.3). P. compressa and P. lobata had generally good/excellent AUC values

(~0.85-0.91) while accuracy for M. capitata and P. meandrina was good (~0.81). For all

the species, generalized boosted regression (GBM) was the best performing method and

classification tree analysis (CTA) was the worst method (Table 2.3). Ensemble models fit

to the entire dataset exhibited improved model accuracies for P. lobata (AUC: 0.86 –

1.00), P. compressa (0.78 – 0.99), P. meandrina (0.86 – 1.00), and M. capitata (0.87 –

1.00).

(a) (b)

Table 2.3 Model accuracy on evaluation datasets using AUC with standard deviations (±) that illustrate the variability over 100 iterations. Model methods are artificial neural networks (ANN), classification tree analysis (CTA), generalized additive models (GAM), generalized boosted regression models (GBM), generalized linear models (GLM), multivariate adaptive regression splines (MARS), flexible discriminant analysis (FDA), and random forests (RF). M. capitata P. meandrina P. compressa P. lobata ANN 0.76 ± 0.05 0.80 ± 0.05 0.88 ± 0.05 0.86 ± 0.03 CTA 0.73 ± 0.05 0.79 ± 0.05 0.79 ± 0.05 0.82 ± 0.04 FDA 0.83 ± 0.04 0.81 ± 0.04 0.90 ± 0.03 0.86 ± 0.03 GAM 0.82 ± 0.04 0.82 ± 0.04 0.91 ± 0.02 0.86 ± 0.03 GBM 0.84 ± 0.04 0.83 ± 0.04 0.93 ± 0.02 0.88 ± 0.03 GLM 0.82 ± 0.04 0.82 ± 0.04 0.91 ± 0.02 0.85 ± 0.03 MARS 0.81 ± 0.05 0.79 ± 0.04 0.91 ± 0.03 0.86 ± 0.03 RF 0.83 ± 0.04 0.82 ± 0.03 0.92 ± 0.02 0.88 ± 0.02

Montipora capitata Pocillopora meandrina Porites compressa Porites lobata Probability occurrence of 0.0 0.2 0.4 0.6 0.8 1.0

012345

Maximum Significant Wave Height (m) Figure 2.4 Comparison of response curve plots for coral species probability of occurrence on maximum significant wave height, max Hs, from the best performing model approach (generalized boosted regression). High probability of occurrence shifts from P. compressa to P. meandrina and P. lobata at max Hs between 1-2 m suggesting a wave height threshold for transitions in coral community dominance. M. capitata did not respond strongly to max Hs.

Discussion

Using an ensemble approach to species distribution modeling, this study created continuous spatial distribution maps of the probability of occurrence for the dominant four Hawaiian coral species around Oahu (Fig. 2.3). These models identified the most important sets of environmental variables for each species (Fig 2.4, Appendix B). The distribution maps can be used for spatially-explicit ecological studies and marine conservation planning activities.

Wave heights were consistently one of the most important variables to explain the distribution of the four dominant coral species in Hawaii. This finding supports prior

studies that identified wave exposure or wave energy as the primary factor influencing

the distribution and composition of Hawaiian coral reefs (Dollar 1982; Grigg 1983; Jokiel

2004; Engels et al. 2004; Storlazzi et al. 2005). P. meandrina, P. compressa, and P.

lobata were strongly influenced by maximum Hs with a species dominance transition

observed between 1-2 m max Hs (Fig. 2.4). Although significant wave height is a surface

observation, it appears to have a predictive capacity for coral species similar to other

wave-related metrics such as near-bottom shear stress (Storlazzi et al. 2005) or wave

exposure (Chollett and Mumby 2012). Shallow Hawaiian benthic communities in very

high wave environments are dominated by crustose coralline algae (Engels et al. 2004).

Predicted occurrences of P. meandrina were lower along the shallow, high-wave

environment of the north shore (aspect of 300-350o) than other coastal areas around Oahu

perhaps reflecting the dominance of coralline algae in those environments. In general, P.

meandrina was predicted to occur in sites with shallow depth, steep slope and high wave energy (max Hs), characteristic of many shallow, basalt boulder habitats around Oahu.

Maximum Hs interacted with depth as a strong predictor for P. lobata with occurrences

becoming more likely deeper than 5 m; a similar vertical transition from the P.

meandrina boulder zone to the deeper P. lobata bench zone was identified by Dollar

(1982). In low wave energy environments typical of Kaneohe Bay or Pearl Harbor, low

light levels corresponded to higher predicted occurrences for P. compressa and M.

capitata (Fig. 2.3a, c). Both environments experience high turbidity from sediment

resuspension and adjacent watershed runoff (Hunter and Evans 1995, Coles et al. 1997,

Jokiel 2006).

Similar modeling approaches incorporating reef species observations and

environmental data for waves, benthic habitats, and geomorphology have been used to map coral reef beta diversity (Harbourne et al. 2006), Caribbean Montastrea spp. forereef habitats (Chollett and Mumby 2012), and Puerto Rican reef fish communities (Pittman et al. 2009, Pittman and Brown 2011). High model accuracies from this study and others

(Harbourne et al. 2006, Pittman et al. 2009, Pittman and Brown 2011, Chollett and

Mumby 2012) strongly suggest that SDMs offer a promising way to generate spatially continuous distributions for coral reef species.

Although the model ensemble provided strong results for the four coral species, we have several suggestions regarding model performance, appropriateness of

environmental variables, and geographic distribution of field samples that should lead to

improved model performance and more biologically accurate distributions. Different

modeling approaches sometimes produce widely divergent models of species’ response to particular environmental variables. For example, the probability of occurrence of M.

capitata with increasing proportion of sandbottom was variously predicted to be constant

and high (CTA), constant and low (GBM, RF), monotonically decreasing (ANN),

increasing then sharply decreasing (MARS), or moderate and unimodal (GLM, GAM).

Weighting models with better predictive performance using a decay function (Thuiller et

al. 2009), as opposed to a committee average (i.e., equal weights), achieved higher model

accuracies for the final ensemble models.

While predictive, the models in this study were limited by the available set of

environmental variables, which represented the benthic geomorphology, wave

environment, and downwelled irradiance around Oahu. Jokiel et al. (2004) found that

bottom complexity (i.e., rugosity) influenced the composition of Hawaiian coral

communities but this study did not find a strong response to rugosity for any coral

species. Most likely, this result reflected the mismatch in spatial scale between rugosity

calculated for the models (~50 m) and rugosity measured from field surveys (~2 cm,

Jokiel et al. 2004). Although fine grain bathymetry data (< 10 m cell size) could improve

the predictive power of modeled bottom complexity and are available for Oahu, it is not

available for the entire study domain. Wave heights are surface observations, but wave

environments at the sea floor are more accurately characterized by near bed shear stress

or water velocity (Storlazzi et al. 2005), two potential variables for future SDM

development. Light is an important environmental factor for corals due to their

photosynthetic intracellular symbionts. The downwelled irradiance variable represented

proportional light levels at depth compared to the surface; this assumed similar surface

irradiance levels throughout the study domain. A better representation of light levels at

depth, including nearshore hydrological influences could provide improved model

response in nearshore environments for P. compressa and M. capitata.

Species observations compiled for this study were collected around all coasts of

Oahu as well as Kaneohe Bay and Pearl Harbor (Fig. 2.2). Although most areas were well sampled, several locations had high sample clusters (Fig. 2.2) which may bias model results towards the characteristics of those areas. While this issue should be addressed by focusing future coral surveys on undersampled areas on the west, southwest, and northeast coasts of Oahu, there are also modeling approaches that may mitigate this bias.

Future studies could include a spatially-explicit term in the models that addresses spatial

25 autocorrelation (Latimer et al. 2006) or could subset the existing dataset by geographic location to achieve equivalent prevalence between areas (Jiménez-Valverde et al. 2009).

Marine spatial planning and ecosystem-based strategies require accurate information on the geographical distribution of species and the set of environmental variables that most influence those distributions (Crowder and Norse 2008, Klein et al.

2010). For example, coral SDMs can be used to inform spatially-explicit threat assessment for coral reefs (Selkoe et al. 2009) or be coupled with spatial optimization approaches for marine conservation planning (Leathwick et al. 2008). Information at a species level is critical since differential responses have been observed between coral species to thermal stressors (Guest et al. 2012) and disease (Aeby et al. 2011). Despite the uncertainties inherent in our findings, we demonstrated an ensemble model approach to develop accurate SDMs and identify primary environmental drivers for the spatially- explicit distributions of four coral species in Hawaii. The application of SDMs has significant potential to address a critical need for realistic and accurate species distribution information for the conservation and management of coral reefs.

References

Aeby GS, Williams GJ, Franklin EC, Kenyon J, Cox EF, Coles S, Work TM (2011) Patterns of coral disease across the Hawaiian Archipelago: relating disease to environment. PLoS ONE 6(5): e20370 Arkema KK, Abramson SC, Dewsbury BM (2006) Marine ecosystem-based management: from characterization to implementation. Frontiers Ecol Environ 10:525-532 Arinaga RA, Cheung KF (2012) Atlas of global wave energy from 10 years of reanalysis and hindcast data. Renewable Energy 39:49-64 Austin M (2007) Species distribution models and ecological theory: a critical assessment and some possible new approaches Ecol Modelling 200:1-19 Battista TA, Costa BM, Anderson SM (2007) Shallow-water benthic habitats of the main

eight Hawaiian Islands (DVD). NOAA Technical Memorandum NOS NCCOS 61, Biogeography Branch. Silver Spring, MD Breiman L, Friedman J, Stone CJ, Olshen RA (1984) Classiﬁcation and regression trees. Chapman and Hall Breiman L (2001) Random forests. Mach Learn 45: 5-32 Brown E, Cox E, Jokiel P, Rodgers K, Smith W, Tissot B, Coles SL, Hultquist J (2004) Development of benthic sampling methods for the coral reef assessment and monitoring program (CRAMP) in Hawai‘i. Pac Sci 58:145-158 Brown E, Minton D, Daniel R, Klasner F, Basch L, Snyder A, Craig P, Dicus G, DeVerse K, Jones T (2007) Benthic marine community monitoring protocol – Pacific island network. Natural Resource Report NPS/PACN/NRTR—2007/002. National Park Service, Fort Collins, Colorado, USA Busby JR (1991) BIOCLIM a bioclimate analysis and prediction system. In: Margules CR, Austin MP (eds), Nature conservation: cost effective biological surveys and data analysis. CSIRO, pp. 6468. Chase JM, Leibold MA (2003) Ecological niches: linking classic and contemporary approaches. University of Chicago Press Chollett I, Mumby PJ (2012) Predicting the distribution of Montastrea reefs using wave exposure. Coral Reefs 31:493-503 Coles SL, DeFelice RC, Eldredge LG, Carlton JT, Pyle RL, Suzumoto A (1997) Biodiversity of marine communities in Pearl Harbor, Oahu, Hawaii with observations on introduced exotic species. Bishop Museum Tech Report 10. Honolulu, Hawaii Connolly JP, Blumberg AF, Quadrini JD (1999) Modeling fate of pathogenic organisms in coastal waters of Oahu, Hawaii. J Environmental Engineering 125:398-406 Crowder L, Norse E (2008) Essential ecological insights for marine ecosystem-based management and marine spatial planning. Mar Policy 32:772-778 Dollar SJ (1982) Wave stress and coral community structure in Hawaii. Coral Reefs 1:71- 81 Engels MS, Fletcher CH, Field ME, Storlazzi CD, Grossman EE, Rooney JJB, Conger CL, Glenn C (2004) Holocene reef accretion: southwest Molokai, Hawaii, U.S.A. J Sed Res 74:255-269 Elith J, Graham CH, Anderson RP, Dudík M, Ferrier S, Guisan A, Hijmans RJ, Huettmann F, Leathwick JR, Lehmann A, Li J, Lohmann LG, Loiselle BA, Manion G, Moritz C, Nakamura M, Nakazawa Y, Overton JM, Peterson AT, Phillips SJ, Richardson K, Scachetti-Pereira R, Schapire RE, Soberón J, Williams S, Wisz MS, Zimmermann NE (2006) Novel methods improve prediction of species’ distributions from occurrence data. Ecography 29, 129–151 Elith J, Leathwick JR (2009) Species distribution models: ecological explanation and prediction across space and time. Ann Rev Ecol Evol Sys 40:677-697 Fielding AH, Bell JF (1997) A review of methods for the assessment of prediction errors in conservation presence/absence models. Env Conserv 24:38-49. Fletcher CH, Bochicchio C, Conger CL, Engels MS, Feirstein EJ, Frazer N, Glenn CR, Grigg RW, Grossman EE, Harney JN, Isoun E, Murray-Wallace CV, Rooney JJ, Rubin KH, Sherman CE, Vitousek S (2008) Geology of Hawaii reefs. In Riegl BM, Dodge RE. Coral reefs of the USA. New York: Springer

Friedman J (1991) Multivariate adaptive regression splines. Ann Stat 19:1141 Garza-Pérez JR, Lehmann A, Arias-González JE (2004) Spatial prediction of coral reef habitats: integrating ecology with spatial modeling and remote sensing. Mar Ecol Prog Ser 269:141-152 Goodman J, Ustin SL (2007) Classification of benthic composition in a coral reef environment using spectral unmixing. J Appl Remote Sens 1:011501 Graham NAJ, Chabanet P, Evans RD, Jennings S, Letourneur Y, MacNeil MA, McClanahan TR, Ohman MC, Polunin NVC, Wilson SK (2011) Extinction vulnerability of coral reef fishes. Ecology Letters 14: 341-348 Grigg RW (1983) Community structure, succession, and development of coral reefs in Hawaii. Mar Ecol Prog Ser 11:1-14 Guest JR, Baird AH, Maynard JA, Muttaqin E, Edwards AJ, Campbell SJ, Yewdall K, Affendi YA, Chou LM (2012) Contrasting patterns of coral bleaching susceptibility in 2010 suggest an adaptive response to thermal stress. PLoS ONE 7: e33353 Guinette JM, Bartley JD, Iqbal A, Fautin DG, Buddemeier RW (2006) Modeling habitat distribution from organism occurrences and environmental data: case study using anemonefishes and their sea anemone hosts. Mar Ecol Prog Ser 316:269-283 Guisan A, Thuiller W (2005) Predicting species distribution: offering more than simple habitat models. Ecology Letters 8:993-1009 Guisan A, Zimmermann NE, Elith J, Graham CH, Phillips S, Peterson AT (2007) What matters for predicting the occurrences of trees: techniques, data, or species characteristics? Ecological Monographs 77:615-630 Harborne AR, Mumby PJ, Zychaluk K, Hedley JD, Blackwell PG (2006) Modeling the beta diversity of coral reefs. Ecology 87:2871-2881 Hastie TJ, Tibshirani R (1990) Generalized additive models. Chapman and Hall Hawaii Mapping Research Group (2011) Main Hawaiian Islands Multibeam Bathymetry Synthesis. http://www.soest.hawaii.edu/hmrg/Multibeam/grids.php Hosmer DW, Lemeshow S (2000) Applied Logistic Regression, second edition. John Wiley & Sons Hunter CL, Evans CW (1995) Coral reefs in Kaneohe Bay, Hawaii: two centuries of western influence and two decades of data. Bull Mar Sci 57:501-515 Hutchinson GE (1957) Concluding remarks. Cold Spring Harbor Symposia on Quantitative Biology 22:415–427 Isoun E, Fletcher C, Frazier N, Gradie J (2003) Multi-spectral mapping of reef bathymetry and coral cover; Kailua Bay, Hawaii. Coral Reefs 22:68-82 Jacobson EC (2005) Light attenuation in a nearshore coral reef ecosystem. MS thesis, University of Hawaii Jiménez-Valverde A, Lobo JM, Hortal J (2009) The effect of prevalence and its interaction with sample size on the reliability of species distribution models. Community Ecology 10:196-205 Jokiel PL, Brown EK, Friedlander A, Rodgers SK, Smith WR (2004) Hawai’i coral reef assessment and monitoring program: spatial patterns and temporal dynamics in reef coral communities. Pac Sci 58:159-174 Jokiel PL (2006) Impact of storm waves and storm floods on Hawaiian reefs. Proc 10th ICRS 1:390-398

Klein CJ, Steinback C, Watts M, Scholz AJ, Possingham HP (2010) Spatial marine zoning for fisheries and conservation. Fron Ecol Env 8:349-353 Kirk JTO (1994) Light and photosynthesis in aquatic ecosystems. Cambridge University Press, Cambridge, UK Knudby A, LeDrew E, Brenning A (2010) Predictive mapping of reef fish species richness, diversity and biomass in Zanzibar using IKONOS imagery and machine- learning techniques. Remote Sensing of Environment 114: 1230–1241 Latimer AM, Wu S, Gelfand AE, Silander JA (2006) Building statistical models to analyze species distributions. Ecol Applications 16:33-50 Leathwick J, Moilanen A, Francis M, Elith J, Taylor P, Julian K, Hastie T, Duffy C (2008) Novel methods for the design and evaluation of marine protected areas in offshore waters. Conservation Letters 1: 91–102 Leibold MA (1995) The niche concept revisited: mechanistic model and community context. Ecology 76:1371-1382 McCullagh P, Nelder JA (1989) Generalized linear models. Chapman and Hall Mumby PJ, Skirving W, Strong AE, Hardy JT, LeDrew EE, Hochberg EJ, Stumpf RP, David LT (2004) Remote sensing of coral reefs and their physical environment. Mar Pollut Bull 48:219-228 Munday PL, Jones GP, Caley MJ (1997) Habitat specialization and the distribution and abundance of coral-dwelling gobies. Mar Ecol Prog Ser 152:227-239 NOAA (2005) Fish habitat utilization patterns and evaluation of the efficacy of marine protected areas in Hawaii: integration of NOAA digital benthic habitat mapping and coral reef ecological studies. NOAA Technical Memorandum NOS NCCOS 23 NOAA (2011) Coral Reef Ecosystem Division, Pacific Islands Fisheries Science Center. \ http://www.pifsc.noaa.gov/cred NODC (2011) National Oceanographic Data Center. http://www.nodc.noaa.gov/ Peterson AT, Soberón J, Pearson RG, Anderson RP, Martinez-Meyer E, Nakamura M, Araújo MB (2011) Ecological niches and geographic distributions. Princeton University Press Pittman SJ, Brown KA (2011) Multi-scale approach for predicting fish species distributions across coral reef seascapes. PLoS ONE 6(5): e20583 Pittman SJ, Costa BM, Battista TA (2009) Using lidar bathymetry and boosted regression trees to predict the diversity and abundance of fish and corals. Journal of Coastal Research 25: 27–38 Ready J, Kaschner K, South AB, Eastwood PD, Rees T, Rius J, Agbayani E, Kullander S, Froese R (2010) Predicting the distribution of marine organisms at the global scale. Ecol Modelling 221:467-478 Ripley BD (1996) Pattern recognition and neural networks. Cambridge University Press Ridgeway G (1999) The state of boosting. Comput Sci Stat 31: 172-181 Robinson LM, Elith J, Hobday AJ, Pearson RG, Kendall BE, Possingham HP, Richardson AJ (2011) Pushing the limits in marine species distribution modelling: lessons from the land present challenges and opportunities. Global Ecol Biogeo 20:789-802 Selkoe KA, Halpern BS, Ebert CM, Franklin EC, Selig ER, Casey KS, Bruno J, Toonen

RJ (2009) A map of human impacts to a “pristine” coral reef ecosystem, the Papahānaumokuākea Marine National Monument. Coral Reefs 28:635-650 Storlazzi CD, Brown EK, Field ME, Rodgers K, Jokiel PL (2005) A model for wave control on coral breakage and species distribution in the Hawaiian islands. Coral Reefs 24:43-55 Stat M, Morris E, Gates RD (2008) Functional diversity in coral-dinoflagellate symbiosis. PNAS 105:9256-9261 SWAN Team (2006) SWAN User Manual: SWAN Cycle III version 40.51. Delft University of Technology Tolman HL (2009) User manual and system documentation of WAVEWATCH III version 3.14. NOAA / NWS / NCEP / MMAB Technical Note 276 Thuiller W, Lafourcade B, Engler R, Araújo R (2009) BIOMOD - a platform for ensemble forecasting of species distributions. Ecography 32:369-373 US Census Bureau (2010) 2010 Population finder. http://www.census.gov/popfinder/ Wilson SK, Depczynski M, Fisher R, Holmes TH, O'Leary RA, Tinkler P (2010) Habitat associations of juvenile fish at Ningaloo Reef, Western Australia: the importance of coral and algae. PLoS ONE 5:e15185. doi:10.1371/journal.pone.0015185

CHAPTER 3

PREDICTIVE MODELING OF CORAL DISTRIBUTION AND ABUNDANCE IN

THE HAWAIIAN ISLANDS

E. C. Franklin1,

P. L. Jokiel1, M. J. Donahue1

1Hawaii Institute of Marine Biology, School of Ocean and Earth Science and

Technology, University of Hawaii, Kaneohe, Hawaii 96744 USA

Submitted to MARINE ECOLOGY PROGESS SERIES as a RESEARCH ARTICLE

Abstract

This study developed species distribution models (SDMs) of the six dominant Hawaiian coral species (Montipora capitata, M. flabellata, M. patula, Pocillopora meandrina,

Porites compressa, and P. lobata) around the main Hawaiian Islands (MHI). To construct

the SDMs, we used boosted regression tree (BRT) models to investigate relationships

between the abundance (i.e., benthic cover) for each species with a set of environmental

variables for the time period 2000-2009. Mean significant wave height and max

significant wave height were the most influential variables explaining coral abundance in

the Hawaiian Islands. The BRT models also identified relationships between coral cover

and island, depth, downwelled irradiance, rugosity, slope, and aspect. The rank order of

coral abundance (from highest to lowest) for the MHI was P. lobata, M. patula, P.

meandrina, M. capitata, P. compressa, and M. flabellata. Mean coral cover predicted for each species was relatively low (≤ 5%) at each island and for the entire main Hawaiian

Islands except for P. lobata around Hawaii (11%). The areas of highest predicted coral cover summed for the six species were Kaneohe Bay on Oahu, the wave-sheltered reefs of Molokai, Lanai, Maui and Kahoolawe, and the Kohala coast of Hawaii. Regional-scale characterizations of coral species from these SDMs provide the framework for spatially- explicit population modeling and marine spatial planning of Hawaiian coral reefs.

Introduction

Coral reefs are an ecosystem in transition (Dubinsky and Stambler 2011). As reefs

transform over the next century, the ability to understand the magnitude and composition

of coral community changes depends upon the accuracy and resolution of the biological

characterization of reef ecosystems. Scleractinian corals are the foundation species of

tropical and subtropical reefs, yet information about their status is woefully inadequate.

For example, only 5 of 845 coral species had sufficient species-specific population trend

data to recently evaluate their extinction risk using the associated IUCN Red List criteria

(Carpenter et al. 2008). Remote sensing technology has enabled global and regional-scale

mapping of shallow coral reefs (Mumby et al. 2004, Mora et al. 2006) yet the sensors

only allow interpretation of habitat-level (e.g., patch reef, fore reef, etc.) or functional

group-level (e.g., coral, algae, sand) information and cannot differentiate between

individual species (Mumby et al. 2004, Goodman and Ustin 2007). Field surveys can

provide information at a species level but are often limited to a small set of geographic

locations. As a method to integrate the strengths of the different approaches to improve

the biological characterization of reefs, species distribution models (SDMs) can

incorporate field observations and environmental covariates from observational, remotely

sensed, or model data into statistical models that predict macroecological-scale, spatially-

continuous distributions of coral species (Guisan and Thuiller 2005, Austin 2007, Elith

and Leathwick 2009).

Widely utilized for modeling species in many ecosystems (Elith and Leathwick

2009, Ready et al. 2010, Robinson et al. 2011), SDMs have less frequently been applied

to coral reef species but have been used to predict distributions of biological functional

groups and habitat types (Garza-Pérez et al. 2004, Guinette et al. 2006, Chollett and

Mumby 2012), and coral reef community metrics (Harborne et al. 2006, Pittman et al.

2009, Knudby et al. 2010). SDMs are constructed by building a representation of the

realized species niche and extrapolating the niche requirements into geographical space

(Guisan et al. 2007, Elith and Leathwick 2009, Peterson et al. 2011). Comparative analysis of population condition and geographic distribution across a range of temporal

and spatial scales are possible with SDMs (Guisan et al. 2007, Elith and Leathwick 2009,

Peterson et al. 2011). In order to construct SDMs, spatially and temporally coincident

biological and environmental data layers need to be available for use in a modeling technique.

1982, Grigg 1983, Jokiel et al. 2004, Engels et al. 2004, Storlazzi et al. 2005). Dollar

(1982) found a vertical zonation of coral species dominance, from shallow to deep, of a

Pocillopora meandrina boulder zone, Porites lobata reef bench zone, and Porites

compressa structured by wave energy and storm frequency. In addition to wave height

and direction, Jokiel et al. (2004) identified depth, rugosity, island age, and organic

sediment content (an indicator of turbid, low light environments) as significant factors

that structure Hawaiian coral communities. The ecologically-dominant coral species by

rank order of abundance in the Hawaiian Islands were Porites lobata, P. compressa,

Montipora capitata, Pocillipora meandrina, M. patula, and M. flabellata (Grigg 1983,

Jokiel et al. 2004). These works identified the dominant coral species in Hawaii and a set of environmental drivers that structure reef communities.

For this study, we develop species distribution models for the benthic cover of six

Hawaiian coral species (M. capitata, M. flabellata, M. patula, P. compressa, P. lobata, P.

meandrina) around the main Hawaiian Islands. To construct the SDMs, we use boosted regression trees (BRT), an efficient, ensemble method for fitting statistical models of

species response variables (e.g., benthic cover) from environmental predictor variables

(Elith et al. 2006, Elith et al. 2008, Chapter 2). We integrate field surveys for corals with

environmental data of wave exposure, benthic geomorphology, and downwelled

irradiance from 2000-2009 to predict species distribution and abundance. Using the BRT

models, we identify optimal models for each species from a set of model runs that

explore the best model parameters to minimize predictive deviance (Elith et al. 2008).

We discuss the geographic distributions and benthic cover patterns of the coral species

and the set of environmental factors most prominently used by the models to construct

the distributions.

Materials and Methods

Study Area

The study area included the shallow seafloor (≥ -30 m) around the eight main Hawaiian

Islands. The Hawaiian archipelago encompasses a group of volcanic islands and atolls

that span 2,500 km in the central north Pacific Ocean (Fig. 3.1, Fletcher et al. 2008). The

geography of these volcanic islands is characterized by prominent coastal capes and

headlands that demarcate coastal exposures to different climate and ocean conditions

(Fig. 3.1). The north coast of Kauai, Oahu, and Maui are exposed to large northern

hemisphere winter swells (≥ 7 m), while southern hemisphere storms produce waves along Hawaiian south shores in summer (Fletcher et al. 2008). The eastern or windward

side of the islands experience consistent easterly tradewinds (10-20 kn) that generate

steady wind waves (Fletcher et al. 2008). There are only two large, natural semi-enclosed

waters bodies in the main Hawaiian Islands, Pearl Harbor and Kaneohe Bay on Oahu

(Fig. 3.1). Coral reefs are found around the coasts and embayments of all islands (Battista

et al. 2007).

((( ((((((((((( ((( ( ((( Hawaiian Archipelago (((((((((( (( ((((( ( (( ( Niihau ((( (( ( (( ((( ( (( (((((((( (( ((( ( (( (((( ( ((( ( (( (( ( ((( ( (( (((( ( (( (((((( (( (( (( ((( (((((((((((( (( ((( (( ((( ( ( ((

((((( Kauai ((( (((( ((( (((( (((((((((((( ((( ( ((( ( ( ((((( ( ((((((( ((( (((((( (( (((( (( ( (((( (( (( ( (( (((( ( ( ((((( ( ( ( ((( ( ((((((( ( ((((( (((((((((( Molokai ((((( ( ((((((( ( (((((((( ( ((((((((((((((( (( Oahu ( ( ( ( ( (((( (((((( (((( (( Maui ( ((((( ((((( ( ( ( ( ((((( (( ( (( ((( ( (((( ( ( (( ( ((((( ( ( ( ( ( ( ( ( ((((( ( (((((((((((((((((((( ( (( (( Lanai ( ((( ((( (( (((((( ( (( ( (((((( (((( (((( ( (((((((( ( Kahoolawe (( ( ( (( ( ( ( ((((((( ( (((( (((( (( ((( ( (((( (( (( Hawaii ((( ( ( ( ( ( ( (( (((((( ( (( ((( (( ((((( (( (( (( (( (( ((( ( Legend ((( (( (( ( ( ( ( (((( (((((( ( (( ( ( ((((( Coral cover observation from field survey ( ( ( ( (((( ((( ( ( ( ((( ( ( 025 50 100 Kilometers ( ( ((

Coral species benthic cover observations

We compiled a benthic cover database of six Hawaiian coral species (M. capitata, M.

flabellata, M. patula, P. meandrina, P. compressa, P. lobata) from scientific monitoring programs (Brown et al. 2004, Brown et al. 2007, NOAA 2005, NOAA 2011) and research project data archived in the National Oceanographic Data Center (NODC 2011).

Our database included 37,710 total benthic cover observations from 2000 to 2009 of the six coral species around the main Hawaiian Islands (Fig. 3.1; Table 3.1). Survey methods included in-situ diver observations and interpreted photo-quadrats for survey areas ranging from 0.25 m2 to 25 m2. Using the location information provided with each

survey, we mapped benthic cover observations for each coral species as vector point

features. Vector points were converted to raster grids using a survey area-weighted mean

of coral cover for each grid cell in ArcGIS (v. 9.3.1, ESRI 2009). The number of grid

cells with data on benthic cover ranged from 1,190 to 5,611 (Table 3.1). Each grid was

georectified to and matched the extent of a 50 m resolution base analysis grid. No significant correlations were observed between sampled area within a grid cell and species cover.

Table 3.1 Number and range of benthic cover (%) from observations and model grid cells for six coral species around the main Hawaiian Islands. Grid cell cover values were computed from a survey area-weighted mean average of observations within that cell. Field Observations Model Grid Cells Species N Cover Range (%) N Cover Range (%) Montipora capitata 4,580 0 – 80.7 1,190 0 – 58.0 Montipora flabellata 4,555 0 – 54.2 1,192 0 – 29.9 Montipora patula 4,555 0 – 72.0 1,193 0 – 72.0 Pocillopora meandrina 4,580 0 – 39.7 1,190 0 – 32.9 Porites compressa 14,860 0 – 100.0 5,611 0 – 94.0 Porites lobata 4,580 0 – 85.7 1,190 0 – 73.6

Environmental Data Layers

We utilized nine environmental covariate data layers for the statistical modeling (Table

3.2). Digital files for all environmental data layers were georectified to a base analysis

grid of 477,795 cells (approximately 1,194 km2) that covered the extent of the study

domain in ArcGIS (v. 9.3.1, ESRI 2009) and geoprocessed using scripts in Python

(http://www.python.org). Data manipulation for coral surveys and environmental

variables was performed using base functions in R (v. 2.14, R Development Core Team

2010). Detailed figures of the environmental data layers for the Hawaiian Islands are in

Appendices A (for Oahu) and E (for Kauai, Niihau, Maui, Molokai, Lanai, Kahoolawe, and Hawaii).

A bathymetry synthesis for the main Hawaiian Islands (Hawaii Mapping Research

Group 2011) provided depth data for the majority of the study domain. The horizontal

resolution of the bathymetry synthesis was approximately 50 m (0.0005 degrees). For

cells that contained no bathymetric data, depths recorded from NOAA National Geodetic

Survey soundings and coral reef survey observations were used to fill gaps where

possible. After gap filling with empirical depth observations, we used an iterative nearest neighbor method, in an 8-cell neighborhood, to calculate depth for no data cells using the average depth of the neighborhood to create a no gaps bathymetry file. This method was

used for approximately 2.7% of study grid cells.

We derived three measures of benthic geomorphology (slope, aspect, and

rugosity) from the bathymetry data layer. Bathymetric slope was the steepest angle,

measured in degrees, of a plane defined for a depth grid cell and its surrounding eight

neighbors. Bathymetric aspect was the steepest downslope direction, measured in

compass degrees (0 o - 360o) of a plane defined by the slope grid cell and its eight

surrounding neighbors. Bathymetric rugosity was the ratio between the surface area and

the planimetric area of the depth grid cell and its eight surrounding neighbors.

Sandbottom habitat areas were converted from digital polygon features delineated from interpreted satellite imagery (Battista et al. 2007) to 5 m resolution raster grids.

Sandbottom included sand, mud, and silt habitats. Sandbottom habitat raster cells (at 5 m

resolution) were summed within the cells of the basemap grid (at ~50 m resolution) to

derive a sandbottom proportion (scale of 0-1) data layer. Sandbottom cells with a value of

1.0 were not included in the analysis.

We forced a SWAN hindcast model (v 40.51, SWAN Team, 2006) with spectral

wave data from WAVEWATCH III (WW3 v3.14, Tolman 2009) for every 6 hours

during January 2000-December 2009 to obtain parametric wave data for the Hawaiian

Islands. Maximum significant wave height, max Hs, and mean significant wave height,

mean Hs, were estimated for the 10-year period at a grid resolution of 0.005 degrees (for

Oahu and Kauai) or 0.01 degrees (for Maui Nui and Hawaii) which was resampled to

0.0005 degrees using an 8-cell nearest neighbor smoothing algorithm on mean values.

Results were validated from a comparison of computed and measured Hs values at

NOAA/NDBC Buoys 51201 and 51202 which demonstrated good overall correlation (r

=0.9) with a slight underestimate in modeled Hs values (Arinaga and Cheung 2012).

Downwelled irradiance was modeled using the Beer-Lambert law in the form:

- -KdZ Ed(Z) = Ed(0 )e where Ed(Z) is the downwelled irradiance at depth Z determined from

- the bathymetry data layer, Ed(0 ) is the irradiance just below the sea surface, and Kd is the

diffuse attenuation coefficient (Kirk 1994). A diffuse attenuation coefficient (Kd) for

PAR (photosynthetically active radiation, 400-700 nm) of 0.054 was used for coastal waters greater than 10 m depth, 0.212 for coastal waters shallower than 10 m, and 0.273 for waters of semi-enclosed embayments including Kaneohe Bay, Pearl Harbor, and

Keehi Lagoon on Oahu (Connolly et al. 1999; Isoun et al. 2003; Jacobson 2005). A

- digital file of downwelled irradiance, Ed(Z) / Ed(0 ), was calculated as the proportion of

downwelled irradiance at depth Z from the bathymetry data file to the irradiance just

below the surface.

We assigned a categorical variable called “island” that represented the closest

island for each grid cell. Four island categories (Kauai, Oahu, Maui, and Hawaii) were

used to represent the eight islands. The Kauai category included Kauai and Niihau while

the Maui category represented Maui, Molokai, Lanai, and Kahoolawe. Oahu and Hawaii

categories did not include additional islands. Significant geographic barriers to biological

connectivity between the four island categories have been found in genetic studies

(Toonen et al. 2011) suggesting spatially distinct population dynamics within island

categories. For convenience, the Maui category is referred to as “Maui Nui” reflecting the

shared geological origin of the group of four islands (Fletcher et al. 2008).

Table 3.2 Summary statistics of environmental covariates around Oahu. Variable Mean SD Range Unit Aspect 182.4 107.0 0.0 – 360.0 ° Depth -13.3 8.8 -30.0 – 0.0 m Island n/a n/a Kauai, Oahu, Maui Nui, n/a Hawaii Max significant wave height 3.0 1.5 0.00 – 8.3 m Mean significant wave height 1.1 0.4 0.00 - 4.3 m Downwelled irradiance 0.4 0.2 0.0 – 1.0 proportion Rugosity 1.002 0.013 1.0 – 2.5 ratio Sandbottom 0.21 0.38 0.0 – 1.0 proportion Slope 1.7 1.7 0 – 14.4 °

Statistical Modeling

Boosted regression tree models were constructed for each coral species cover using the

routines gbm version 1.6–32 (Ridgeway 2012) and gbm.step (Elith et al. 2008) in the R

statistical program version 2.14 (R Development Core Team, http://www.r-project.org).

BRT models combine regression trees that fit environmental predictors to response

variables with a boosting algorithm that assembles an ensemble of trees in an additive,

stage-wise fashion (Hastie et al. 2001, Elith et al. 2008). Within the BRT models, three

terms were used to optimize predictive performance: tree complexity, learning rate, and

bag-fraction. Tree complexity (tc) determined the number of nodes in a tree that should

reflect the true interaction order on the response being modeled, although this is often unknown, and learning rate (lr) was used to shrink the contribution of each tree as it is

added to the model (Elith et al. 2008). The bag-fraction determined the proportion of data

to be selected at each step and therefore the model stochasticity (Elith et al. 2008). We

determined optimal settings for these parameters by examining the cv deviance over tc

values 1–5, lr values of 0.05, 0.01 and 0.005, and bag fractions of 0.5 and 0.75. All possible combinations were run, with the optimal number of trees in each case being

determined by gbm.step (Elith et al. 2008). Each model run included 10-fold cross- validation using 90% subsets of training data and validated with the remaining 10% test data. The combination of the three parameter settings with the lowest cv deviance was then selected to produce the final BRT for each species (Elith et al. 2008). All models were run with arc-sine transformed measures of cover which were treated as a Gaussian

(normal) response distribution (Read et al. 2008, 2011). For the final BRT models, the relative contribution of each predictor was based on the number of times the variable was

selected for splitting, weighted by the squared improvement to the model as a result of

each split, and averaged over all trees (Friedman and Meulman 2003, Elith et al. 2008).

Partial dependency plots were used for interpretation and to quantify the relationship between each predictor variable and response variable, after accounting for the average

effect of all other predictor variables in the model. We used gbm.interactions (Elith et al.

2008) to quantify interaction effects between predictors. The relative strength of

interaction fitted by BRT was quantified by the residual variance from a linear model,

and the value indicates the relative degree of departure from a purely additive effect, with

zero indicating no interaction effects fitted (Elith et al. 2008). We defined a threshold

interaction value and reported the interactions with values ≥ 0.1.

Results

The final BRT models predicted the geographic distributions of benthic cover of the six

dominant coral species for the main Hawaiian Islands (Fig 3.2). A full exploration of

model settings determined the optimal BRT model for each species (Appendix G). Model

settings for the final BRT models ranged from 1350-4500 trees, a tree complexity of 4 or

5, a learning rate of 0.01 or 0.005, and a bag fraction of 0.75 (Table 3.3). Model cross-

validation deviances ranged from 0.009 – 0.021 (Table 3.3). Max Hs and mean Hs were

consistently the most important variables to explain the benthic cover of the six dominant

coral species in Hawaii with varying levels of secondary contributions from island,

aspect, depth, rugosity, slope, and downwelled irradiance to particular species (Table

3.4). 42

. M endix F. pp (b), and ecies are in A M. flabellata p (a), Montipora capitata redicted coral cover for s redicted p ures of g predicted coral cover (%) for around the main Hawaiian Islands. Detailed fi ) c ( atula Figure 3.2 Geographic maps of model p

Porites compressa (d), Pocillopora meandrina l cover (%) for odel predicted cora Hawaiian Islands. (f) around the main the main around (f) P.lobata

(e), and Figure 3.2 (cont) Geographic maps of m

Table 3.3 Model settings and cross-validation deviance of final boosted regression tree (BRT) models for benthic cover of Montipora capitata, M. flabellata, M. patula, Pocillopora meandrina, Porites compressa, and P. lobata around the main Hawaiian Islands. Model settings include the number of trees (nt), tree complexity (tc), learning rate (lr), and bag fraction (bag), and cross-validation deviance (cv dev) with standard error (se). Response variable nt tc lr bag cv dev (se) Montipora capitata cover 2550 5 0.01 0.75 0.011 (0.001) M. flabellata cover 1450 4 0.01 0.75 0.002 (0) M. patula cover 4050 5 0.005 0.75 0.011 (0.001) Pocillopora meandrina cover 3400 4 0.005 0.75 0.009 (0.001) Porites compressa cover 2550 5 0.01 0.75 0.010 (0.001) P. lobata cover 1350 5 0.01 0.75 0.021 (0.001)

Table 3.4 Relative contribution of environmental variables to boosted regression tree (BRT) models of Montipora capitata (Mcap), M. flabellata (Mfla), M. patula (Mpat), Pocillopora meandrina (Pmea), Porites compressa (Pcom), and P. lobata (Plob). Mcap Mfla Mpat Pmea Pcom Plob Mean significant wave height (m) 25.5 26.7 18.7 21.2 33.5 14.5 Max significant wave height (m) 17.4 22.2 16.4 17.8 20.7 14.7 Island (Kauai, Oahu, Maui Nui, Hawaii) 2.7 3.8 11.3 3.5 6.8 28.6 Aspect (°) 14.7 17.1 14.4 11.5 3.2 9.0 Depth (m) 9.8 8.6 10.3 8.6 10.8 9.6 Rugosity (Surface/Planar Area) 6.9 11.4 7.4 13.7 8.5 6.1 Slope (°) 7.0 4.3 8.2 11.4 4.0 8.0 Downwelled irradiance (Ed(Z)/Ed(0-)) 10.4 4.8 9.6 3.1 6.5 6.9 Sandbottom (proportion) 5.5 1.2 3.7 3.1 6.1 2.6

Benthic cover of the three Montipora species was most influenced by max Hs,

mean Hs and bathymetric aspect (Table 3.4) with M. flabellata found in the highest wave

energy environments, M. patula in intermediate wave environments, and M. capitata in

low wave environments along north-east coastlines (Fig 3.2a-c). P. meandrina cover was

predicted to be highest in areas of high maximum Hs, mean Hs, and steep slopes (Fig

3.2d, Table 3.4). Models predicted the highest benthic cover for P. compressa in areas with low max Hs and mean Hs with an interaction of 1.27 between max Hs and mean Hs

(Fig 3.2e, Table 3.4). P. lobata cover was most strongly influenced by the island variable

with highest cover at Hawaii and declining toward Kauai. Max Hs and mean Hs were 45 strong secondary predictors for P. lobata cover (Fig 3.2f, Table 3.4). Generally, highest coral abundances were predicted around Hawaii with a gradually declining gradient of cover toward the northwest as well as a shift from Porites spp. to Montipora spp. community dominance (Fig 3.3).

12 P. lobata M. patula 10 P. meandrina M. capitata 8 P. compressa M. flabellata 6

4 Coral cover (%) Coral cover

Figure 3.3 Mean benthic cover (%) ± SE of six coral species predicted from final BRT models for each Hawaiian island and the entire main Hawaiian Islands (MHI).

Coral cover predicted for each species was relatively low (under 5%) at each island and for the entire main Hawaiian Islands except for P. lobata around Hawaii (Fig

3.3). P. lobata had the highest predicted coral cover at Niihau (2.2%), Molokai (3.3%),

Maui (4.1%), Lanai (5.5%), Kahoolawe (5.2%), and Hawaii (11.1%) while M. patula had the highest cover around Kauai (3.3%) and Oahu (3.0%) (Fig. 3.3). The rank order of abundance (from highest to lowest with coral cover in parenthesis) for the entire main

Hawaiian Islands was P. lobata (4.4%), M. patula (2.2%), P. meandrina (1.44%), M. capitata (1.40%), P. compressa (0.8%), and M. flabellata (0.3%). No island had the same

rank order of abundance as the overall MHI order but Molokai and Maui were most similar to the MHI. Notable divergences from the MHI rank order were P. compressa

with the 2nd and 3rd highest cover at Lanai and Hawaii, respectively, and P. meandrina

with the second highest cover around Maui and Hawaii (Fig. 3.3).

Predicted total coral cover from a summation of the six species varied between a

low of 7.5% (Niihau and Molokai) to 18.4% with an overall mean of 10.5% for the MHI.

Kaneohe Bay on Oahu, the wave-sheltered area of Maui Nui that includes reefs of

Molokai, Lanai, Maui and Kahoolawe, and the Kohala coast of Hawaii were predicted as

areas of the highest total coral cover summed for the six species (Fig. 3.4). Mean island

coral cover ranged between 2-26% (Fig. 3.4) with the highest cover around Lanai and

Hawaii and the lowest at Niihau and Kauai (Fig. 3.4).

Figure 3.4 Geographic map of summed total cover for six coral species predicted from BRT models for the main Hawaiian Islands.

Discussion

Using BRT models, we developed continuous spatial distribution maps of the benthic

cover for the dominant six Hawaiian coral species around the main Hawaiian Islands

(Fig. 3.2, Appendix F). The BRT models identified the most important sets of

environmental variables for each species (Table 3.4, Appendices H, I). Mean coral cover

for each species and species rank abundance by island and the entire MHI were

calculated from final BRT model predicted coral covers (Figs. 3.3, 3.4).

Wave exposure or wave energy has been identified as the primary factor

influencing the distribution, zonation, and composition of Hawaiian coral reefs (Dollar

1982; Grigg 1983; Jokiel 2004; Engels et al. 2004; Storlazzi et al. 2005, Chapter 2). In

this study, both max significant wave height (max Hs) and mean significant wave height

(mean Hs) were consistently the most important variables to explain the benthic cover of

Hawaiian coral species (Table 3.4). This result suggests a synergistic effect between the

typical, daily wave conditions and periodic high energy wave events from storms in

structuring coral communities in the Hawaiian Islands (Dollar 1982, Grigg 1983). Other

environmental variables contributing greater than 10% relative importance to BRT models were island, aspect, depth, rugosity, slope, and downwelled irradiance,

relationships which were also reported by Jokiel et al. (2004).

Predicted cover for P. meandrina was highest in shallow, high-wave energy

environments along the north coasts and headlands of Kauai, Oahu, and Maui and the

entire coastline of Hawaii (Fig 3.2f). These areas are commonly characterized by

shallow, basalt boulder habitats with steep slope, high rugosity, and high-wave energy

(Jokiel et al. 2004, Battista et al. 2007). Although significant wave height is a surface

observation, its predictive capacity for coral species occurrence has been previously

demonstrated (Chapter 2), and appears to perform similarly to other wave-related metrics

such as near-bottom shear stress (Storlazzi et al. 2005) or wave exposure (Chollett and

Mumby 2012).

Of the three Montipora species, M. capitata appears the most broadly adaptable to a range of habitats since it is predicted to occur both along fore reefs as well as in quiescent environments such as Kaneohe Bay on Oahu (Fig 3.2a). The ability of M. capitata to occupy such diverse habitats may be possible due to the extreme phenotypic plasticity that the species exhibits (Forsman et al. 2010). For example, a single colony of

M. capitata may possess sections that are laminar, encrusting, or branching (Forsman et al. 2010). M. patula and M. flabellata were predicted to occur in higher wave environments than M. capitata. Highest predicted cover for these species was along the east coasts of Kauai and Oahu and wave-sheltered areas of Maui Nui (Fig 3.2b,c). Both species are endemic to Hawaii and currently under review for listing as threatened or endangered species under the US Endangered Species Act (Brainard et al. 2011). This work represents the most comprehensive distribution and abundance information available for these two species and should be used to inform future population surveys or conservation efforts for these species.

P. compressa cover dominated low wave-energy environments that are typically shallow, nearshore habitats with turbid waters from sediment resuspension and watershed inputs (Hunter and Evans 1995, Coles et al. 1997, Jokiel 2006). High P. compressa cover was also predicted for wave-sheltered coasts of Maui Nui and deeper waters (> -10 m)

around Hawaii, an observation reported by Dollar (1982, Fig 3.3d). At intermediate wave

energies (1-3 m max Hs), P. lobata was predicted to be the dominant coral (Chapter 2,

Fig. 3.2e) and was the most abundant coral species in the main Hawaiian Islands (Fig.

3.3).

In general, coral cover was predicted to be highest in primarily wave-sheltered

coastlines and embayments. High coral cover locations were predicted throughout the islands but reefs with highest cover were concentrated in Kaneohe Bay on Oahu, the wave-sheltered areas of Maui Nui, and along the west coast of Hawaii (Fig. 3.4). These areas have varying benthic geomorphology and levels of downwelled irradiance but share similar low wave energy characteristics. Previously, Engels et al. (2004) model of modern coral zonation for the Hawaiian Islands related the occurrence of highest total coral cover to low wave-energy environments. Storlazzi et al. (2005) also observed the highest total coral cover along sections of the Molokai coastline with the lowest wave-

induced near bed shear stress. These studies correspond well with the predicted results from the BRT models.

Rank order of coral species abundance differed slightly from prior studies. We found the rank order (from highest to lowest) to be P. lobata, M. patula, P. meandrina,

M. capitata, P. compressa, and M. flabellata. From a survey of the southwest coasts of

Kauai, Oahu, Maui, and Hawaii, Grigg (1983) found the rank order of abundance for the six coral species in this study as P. lobata, P. compressa, M. capitata, P. meandrina, M. patula, and M. flabellata. Jokiel et al. (2004) surveyed sixty monitoring locations throughout the main Hawaiian Islands and observed a similar rank order of abundance

(and coral cover) of P. lobata (6.1%), P. compressa (4.5%), M. capitata (3.9%), M.

patula (2.7%), P. meandrina (2.4%), M. flabellata (0.7%). Compared to Grigg (1983)

and Jokiel et al. (2004), this study found P. compressa to be relatively less abundant

overall and M. patula more abundant than M. capitata.

The BRT models provided strong results for the predicted cover of the six coral species (Table 3.2, Appendix G) but the inclusion of additional environmental variables and a more comprehensive geographic distribution of field samples should lead to better model performance and more accurate cover predictions. For example, finer grain bathymetry data (< 10 m cell size) could improve the predictive power of modeled bottom complexity (such as found in Jokiel et al. 2004) but were not available for the entire study domain. Significant wave heights are a surface measurement. Wave environments at the sea floor are more accurately characterized by bottom water velocity

(Lowe et al. 2009) or near bed shear stress (Storlazzi et al. 2005), which may be two potential variables for future study. A better representation of downwelled irradiance would incorporate the absolute surface irradiance instead of a relative metric which assumes similar surface levels throughout the study domain.

Species observations compiled for this study were collected throughout the main

Hawaiian Islands (Fig. 3.1). Although most areas were well sampled, several locations had high sample clusters (Fig. 3.1) which may bias model results towards the characteristics of those areas. In addition, Kahoolawe, south Molokai, north Lanai, and northeast Oahu were undersampled and we suggest future coral surveys to focus on these areas. In the absence of acquiring additional samples, future studies could include a spatially-explicit term to address potential spatial autocorrelation in the models (Latimer et al. 2006) or select subsets of the existing datasets by geographic location to achieve

area proportional sampling between islands (Jiménez-Valverde et al. 2009) to mitigate

spatial sampling bias.

Regional-scale characterizations of coral species from SDMs provide the

framework for spatially-explicit ecosystem modeling and marine spatial planning of coral

reefs (Crowder and Norse 2008, Klein et al. 2010). SDMs of coral species are critically

useful since species respond differentially to thermal stressors (Guest et al. 2012) and coral diseases (Aeby et al. 2011, Williams et al. 2010), while studies of total coral cover alone overlook changes in reef composition and species dominance. Data from coral

SDMs can be incorporated into spatial optimization exercises for marine conservation

(Leathwick et al. 2008) or for geographically-explicit threat assessments to reefs (Selkoe et al. 2009, Burke et al. 2011). We developed the first accurate, regional-scale coral

SDMs and identified primary environmental drivers for the spatially-explicit distributions of the benthic cover of the six dominant species and total coral cover in the Hawaiian

Islands. The geographic characterization of coral reefs would benefit greatly from the improved coral distribution and abundance information generated from coral SDMs.

References

Aeby GS, Williams GJ, Franklin EC, Kenyon J, Cox EF, Coles S, Work TM (2011) Patterns of coral disease across the Hawaiian Archipelago: relating disease to environment. PLoS ONE 6(5): e20370 Arinaga RA, Cheung KF (2012) Atlas of global wave energy from 10 years of reanalysis and hindcast data. Renewable Energy 39:49-64 Austin M (2007) Species distribution models and ecological theory: a critical assessment and some possible new approaches Ecol Modelling 200:1-19 Battista TA, Costa BM, Anderson SM (2007) Shallow-water benthic habitats of the main eight Hawaiian Islands (DVD). NOAA Technical Memorandum NOS NCCOS 61, Biogeography Branch. Silver Spring, MD Brainard RE, Birkeland C, Eakin CM, McElhany P, Miller MW, Patterson M, Piniak GA

(2011) Status review report of 82 candidate coral species petitioned under the US Endangered Species Act. US DOC, NOAA-TM-NMFS-PIFSC-27 Brown E, Cox E, Jokiel P, Rodgers K, Smith W, Tissot B, Coles SL, Hultquist J (2004) Development of benthic sampling methods for the coral reef assessment and monitoring program (CRAMP) in Hawai‘i. Pac Sci 58:145-158 Brown E, Minton D, Daniel R, Klasner F, Basch L, Snyder A, Craig P, Dicus G, DeVerse K, Jones T (2007) Benthic marine community monitoring protocol – Pacific island network. Natural Resource Report NPS/PACN/NRTR—2007/002. National Park Service, Fort Collins, Colorado, USA Burke L, Reytar K, Spalding M, Perry A (2011) Reefs at risk revisited. World Resources Institute, Wash DC Carpenter KE, Abrar M, Aeby G, Aronson RB, Banks S, Bruckner A, Chiriboga A, Cortes J, Delbeek JC, DeVantier L, Edgar GJ, Edwards AJ, Fenner D, Guzmán H, Hoeksema BW, Hodgson G, Johan O, Licuanan WY, Livingstone SR, Lovell ER, Moore JA, Obura DO, Ochabvillo D, Polidoro BA, Precht WF, Quibilan MC, Reboton C, Richards ZT, Rogers AD, Sanciangco J, Sheppard A, Sheppard C, Smith J, Stuart S, Turak E, Veron JEN, Wallace C, Weil E, Wood E (2008) One- third of reef-building corals face elevated extinction risk from climate change and local impacts. Science 321:560-563 Chollett I, Mumby PJ (2012) Predicting the distribution of Montastrea reefs using wave exposure. Coral Reefs 31:493-503 Coles SL, DeFelice RC, Eldredge LG, Carlton JT, Pyle RL, Suzumoto A (1997) Biodiversity of marine communities in Pearl Harbor, Oahu, Hawaii with observations on introduced exotic species. Bishop Museum Tech Report 10. Honolulu, Hawaii Connolly JP, Blumberg AF, Quadrini JD (1999) Modeling fate of pathogenic organisms in coastal waters of Oahu, Hawaii. J Environmental Engineering 125:398-406 Crowder L, Norse E (2008) Essential ecological insights for marine ecosystem-based management and marine spatial planning. Mar Policy 32:772-778 Dollar SJ (1982) Wave stress and coral community structure in Hawaii. Coral Reefs 1:71- 81 Dubzinsky Z, Stambler N (2011) Coral reefs: an ecosystem in transition. Springer, NY Elith J, Graham CH, Anderson RP, Dudik M, Ferrier S, Guisan A, Hijmans RJ, Huettmann F, Leathwick JR, Lehmann A, Li J, Lohmann LG, Loiselle BA, Manion G, Moritz C, Nakamura M, Nakazawa Y, Overton JM, Peterson AT, Phillips SJ, Richardson K, Scachetti-Pereira R, Schapire RE, Soberón J, Williams S, Wisz MS, Zimmerman NE (2006) Novel methods improve prediction of species’ distributions from occurrence data. Ecography 29:129-151 Elith J, Leathwick JR, Hastie T (2008) A working guide to boosted regression trees. J Anim Ecol 77:802-813 Elith J, Leathwick JR (2009) Species distribution models: ecological explanation and prediction across space and time. Ann Rev Ecol Evol Sys 40:677-697 Engels MS, Fletcher CH, Field ME, Storlazzi CD, Grossman EE, Rooney JJB, Conger CL, Glenn C (2004) Holocene reef accretion: southwest Molokai, Hawaii, U.S.A. J Sed Res 74:255-269 Fletcher CH, Bochicchio C, Conger CL, Engels MS, Feirstein EJ, Frazer N, Glenn CR,

Grigg RW, Grossman EE, Harney JN, Isoun E, Murray-Wallace CV, Rooney JJ, Rubin KH, Sherman CE, Vitousek S (2008) Geology of Hawaii reefs. In Riegl BM, Dodge RE. Coral reefs of the USA. New York: Springer Forsman ZH, Concepcion GT, Haverkort RD, Shaw RW, Maragos JE, Toonen RJ (2010) Ecomorph or endangered coral? DNA and microstructure reveal Hawaiian species complexes: Montipora dilatata/flabellata/turgescens & M. patula/verrilli. PLoS ONE 5 e15021 Friedman JH, Meulman JJ (2003) Multiple additive regression trees with application in epidemiology. Statistics in Medicine 22: 1365–1381 Garza-Pérez JR, Lehmann A, Arias-González JE (2004) Spatial prediction of coral reef habitats: integrating ecology with spatial modeling and remote sensing. Mar Ecol Prog Ser 269:141-152 Goodman J, Ustin SL (2007) Classification of benthic composition in a coral reef environment using spectral unmixing. J Appl Remote Sens 1:011501 Grigg RW (1983) Community structure, succession, and development of coral reefs in Hawaii. Mar Ecol Prog Ser 11:1-14 Guest JR, Baird AH, Maynard JA, Muttaqin E, Edwards AJ, Campbell SJ, Yewdall K, Affendi YA, Chou LM (2012) Contrasting patterns of coral bleaching susceptibility in 2010 suggest an adaptive response to thermal stress. PLoS ONE 7: e33353 Guinette JM, Bartley JD, Iqbal A, Fautin DG, Buddemeier RW (2006) Modeling habitat distribution from organism occurrences and environmental data: case study using anemonefishes and their sea anemone hosts. Mar Ecol Prog Ser 316:269-283 Guisan A, Thuiller W (2005) Predicting species distribution: offering more than simple habitat models. Ecology Letters 8:993-1009 Guisan A, Zimmermann NE, Elith J, Graham CH, Phillips S, Peterson AT (2007) What matters for predicting the occurrences of trees: techniques, data, or species characteristics? Ecological Monographs 77:615-630 Harborne AR, Mumby PJ, Zychaluk K, Hedley JD, Blackwell PG (2006) Modeling the beta diversity of coral reefs. Ecology 87:2871-2881 Hastie T, Tibshirani R, Friedman JH (2001) The elements of statistical learning: data mining, inference and prediction. Springer-Verlag, NY Hawaii Mapping Research Group (2011) Main Hawaiian Islands Multibeam Bathymetry Synthesis. http://www.soest.hawaii.edu/hmrg/Multibeam/grids.php Hunter CL, Evans CW (1995) Coral reefs in Kaneohe Bay, Hawaii: two centuries of western influence and two decades of data. Bull Mar Sci 57:501-515 Isoun E, Fletcher C, Frazier N, Gradie J (2003) Multi-spectral mapping of reef bathymetry and coral cover; Kailua Bay, Hawaii. Coral Reefs 22:68-82 Jacobson EC (2005) Light attenuation in a nearshore coral reef ecosystem. MS thesis, University of Hawaii Jiménez-Valverde A, Lobo JM, Hortal J (2009) The effect of prevalence and its interaction with sample size on the reliability of species distribution models. Community Ecology 10:196-205 Jokiel PL, Brown EK, Friedlander A, Rodgers SK, Smith WR (2004) Hawai’i coral reef assessment and monitoring program: spatial patterns and temporal dynamics in reef coral communities. Pac Sci 58:159-174

Jokiel PL (2006) Impact of storm waves and storm floods on Hawaiian reefs. Proc 10th ICRS 1:390-398 Klein CJ, Steinback C, Watts M, Scholz AJ, Possingham HP (2010) Spatial marine zoning for fisheries and conservation. Fron Ecol Env 8:349-353 Kirk JTO (1994) Light and photosynthesis in aquatic ecosystems. Cambridge University Press, Cambridge, UK Knudby A, LeDrew E, Brenning A (2010) Predictive mapping of reef fish species richness, diversity and biomass in Zanzibar using IKONOS imagery and machine- learning techniques. Remote Sensing of Environment 114: 1230–1241 Latimer AM, Wu S, Gelfand AE, Silander JA (2006) Building statistical models to analyze species distributions. Ecol Applications 16:33-50 Leathwick J, Moilanen A, Francis M, Elith J, Taylor P, Julian K, Hastie T, Duffy C (2008) Novel methods for the design and evaluation of marine protected areas in offshore waters. Conservation Letters 1: 91–102 Lowe RJ, Falter JL, Monismith SG, Atkinson MJ (2009) Wave-driven circulation of a coastal reef-lagoon system. J Phys Oceanography, 39:869-889 Mora C, Andréfouët S, Kranenburg S, Rollo A, Costello M, Veron J, Gaston KJ, Myers RA (2006) How protected are coral reefs? Science 314:757-760 Mumby PJ, Skirving W, Strong AE, Hardy JT, LeDrew EE, Hochberg EJ, Stumpf RP, David LT (2004) Remote sensing of coral reefs and their physical environment. Mar Pollut Bull 48:219-228 NOAA (2005) Fish habitat utilization patterns and evaluation of the efficacy of marine protected areas in Hawaii: integration of NOAA digital benthic habitat mapping and coral reef ecological studies. NOAA Technical Memorandum NOS NCCOS 23 NOAA (2011) Coral Reef Ecosystem Division, Pacific Islands Fisheries Science Center. \ http://www.pifsc.noaa.gov/cred NODC (2011) National Oceanographic Data Center. http://www.nodc.noaa.gov/ Peterson AT, Soberón J, Pearson RG, Anderson RP, Martinez-Meyer E, Nakamura M, Araújo MB (2011) Ecological niches and geographic distributions. Princeton University Press Pittman SJ, Costa BM, Battista TA (2009) Using lidar bathymetry and boosted regression trees to predict the diversity and abundance of fish and corals. Journal of Coastal Research 25: 27–38 Ready J, Kaschner K, South AB, Eastwood PD, Rees T, Rius J, Agbayani E, Kullander S, Froese R (2010) Predicting the distribution of marine organisms at the global scale. Ecol Modelling 221:467-478 Read CF, Duncan DH, Vesk PA, Elith J (2008) Biological soil crust distribution is related to patterns of fragmentation and landuse in a dryland agricultural landscape of southern Australia. Landscape Ecol 23:1093-1105 Read CF, Duncan DH, Vesk PA, Elith J (2011) Surprisingly fast recovery of biological soil crusts following livestock removal in southern Australia. J Vegetation Sci 22:905-916 Ridgeway G (2012) Generalized boosted regression models. Documentation on the R package “gbm”, version 1.6–32 Robinson LM, Elith J, Hobday AJ, Pearson RG, Kendall BE, Possingham HP,

Richardson AJ (2011) Pushing the limits in marine species distribution modelling: lessons from the land present challenges and opportunities. Global Ecol Biogeo 20:789-802 Selkoe KA, Halpern BS, Ebert CM, Franklin EC, Selig ER, Casey KS, Bruno J, Toonen RJ (2009) A map of human impacts to a “pristine” coral reef ecosystem, the Papahānaumokuākea Marine National Monument. Coral Reefs 28:635-650 Storlazzi CD, Brown EK, Field ME, Rodgers K, Jokiel PL (2005) A model for wave control on coral breakage and species distribution in the Hawaiian islands. Coral Reefs 24:43-55 SWAN Team (2006) SWAN User Manual: SWAN Cycle III version 40.51. Delft University of Technology Tolman HL (2009) User manual and system documentation of WAVEWATCH III version 3.14. NOAA / NWS / NCEP / MMAB Technical Note 276 Toonen RJ, Andrews KR, Baums IB, Bird CE, Concepcion GT, Daly-Engel TS, Eble JA, Faucci A, Gaither MR, Iacchei M, Puritz JB, Schultz JK, Skillings DJ, Timmers MA, Bowen BW (2011) Defining boundaries for ecosystem-based management: a multispecies case study of marine connectivity across the Hawaiian Archipelago. J Mar Biol 2011:460173. doi:10.1155/2011/460173 Williams GJ, Aeby GS, Cowie ROM, Davy SK (2010) Predictive modeling of coral disease distribution within a reef system. PLoS ONE 5: e9264.

CHAPTER 4

A NICHE MODEL ANALYSIS OF CORALS IN HAWAII: ARE POPULATIONS

MORE ABUNDANT IN MARINE PROTECTED AREAS?

EC Franklin,

PL Jokiel, MJ Donahue

Hawaii Institute of Marine Biology, School of Ocean and Earth Science and Technology,

University of Hawaii, Kaneohe, Hawaii 96744 USA

Submitted to DIVERSITY & DISTRIBUTIONS as a REPORT

Abstract

We use spatially-explicit maps of abundance derived from species distribution models to evaluate coral populations in an marine protected area (MPA) network in Hawaii. Using

continuous spatial distribution maps of niche model-derived abundance, we analyzed the efficacy of an MPA network for the six dominant coral species (Porites compressa, P.

lobata, Pocillopora meandrina, Montipora capitata, M. flabellata, and M. patula) of the eight main Hawaiian Islands from 2000 to 2009. For each species, we modeled abundance on shallow reef habitats (≤ 30 m depth) using boosted regression trees and a suite of environmental covariates. We projected model results to the geographic domain and calculated mean coral species abundances within and outside 12 Hawaiian marine protected areas (MPAs) and the Hawaiian Islands Humpback Whales National Marine

Sanctuary. MPA efficacy was determined by coral abundances within MPAs being equivalent to or greater than abundances of unprotected areas by island group and for each MPA individually. Abundances of the two Porites species were higher in MPAs than open areas. The three Montipora species and Pocillopora meandrina had lower abundances in most MPAs compared to open areas. Manele-Hulopoe and Molokini

Marine Life Conservation Districts (MLCD) had higher abundances for four of the six coral species compared to unprotected areas while Waikiki MLCD had lower abundances for all corals. The HIHWNMS encompasses coral populations with higher abundances than areas outside the boundaries especially for the four corals (Montipora spp. and P. meandrina) underrepresented in the current MPA network. The study provides baseline coral population information for current and future marine protected area planning and evaluation activities within the Hawaiian Islands that was not previously available.

Introduction

Marine protected areas (MPAs) provide a fundamental management tool for the conservation of marine organisms on coral reefs. MPAs can provide numerous ecosystem benefits and services such as enhanced species diversity and community biomass, protected critical nursery or spawning sites, and focal areas for educational or economic activities (Halpern and Warner 2002, Russ 2002, McLeod et al. 2009, Halpern et al.

2010, Cinner et al. 2012). The level of protection offered by an MPA can range from no protection within MPA boundaries (i.e., a “paper park”) to a fully no-take, no entry area

(WCPA/IUCN 2007). Since MPAs are often established opportunistically, a network of

MPAs may have been assembled in an ad hoc fashion (Mora et al. 2006, WCPA/IUCN

2007). Nonetheless, a network of MPAs should reasonably represent the biotic characteristics of the broader reef ecosystems for which they protect (Roberts et al. 2003,

Mora et al. 2006, Laffoley 2008). Scleractinian corals are the foundation species that provide biohermic habitat for the extremely diverse and productive coral reef ecosystems

(Sebens 1994, Knowlton et al. 2010) and their sustained maintenance or enhancement is an implicit goal of coral-reef focused MPAs (Mora et al. 2006, Selig and Bruno 2010).

The success of MPAs in the enhancement of fish and non-coral invertebrate populations has led to theoretical speculation of positive secondary effects on coral populations

(Hughes et al. 2003, Bellwood et al. 2004).

In the main Hawaiian Islands, a network of coral reef MPAs has been established over the last four decades to meet a variety of resource management and conservation objectives (NMPAC 2012). The objectives of the various Hawaiian MPAs include

enhanced recreational opportunities, biodiversity conservation, and sustainable fisheries

with a range of regulatory status including no-take, customary stewardship, partial

protection. (NMPAC 2012). Evaluation of the efficacy of Hawaiian MPAs has

demonstrated positive effects on reef fish species numbers, richness, and biomass

(Friedlander et al. 2003, 2007, 2010) and the productivity of the marine aquarium fishery

(Tissot and Hallacher 2003, Tissot et al. 2004). Prior studies have shown positive

differences within and outside the Hawaiian MPA network in benthic functional groups

(e.g., coral, algae, etc.) or coral morphological categories (lobate, branching, plating)

(Friedlander et al. 2003, 2010) but none have assessed the status of particular coral

species populations. However, protection within MPAs may not provide enhanced coral abundance, especially if regional or global factors drive fluctuations in coral populations

(Graham et al. 2008).

Despite the critical role of coral species in reef ecosystems, no prior analyses have broadly examined the role of MPAs in sustaining coral species in Hawaii. Coral abundance, measured as benthic cover or the proportion of the hard substratum covered by live coral tissue, is a fundamental measure of coral ecosystem health. Here we present a geographically comprehensive, species-level analysis of abundance for the six most common corals within and outside a network of MPAs in the Hawaiian Islands. We use spatially-explicit datasets from species distribution models (SDMs) (Chapter 3) to compare coral populations amongst a set of existing Hawaiian MPAs that possess at least a level of partial protection and support an ongoing management plan review for the

Hawaiian Islands Humpback Whales National Marine Sanctuary that is considering the protection of coral species within their MPA boundaries.

Methods

Study Area

The study domain was the shallow seafloor (≥ -30 m) surrounding the eight main

Hawaiian Islands, a group of volcanic high islands in the central north Pacific Ocean

(Fig. 4.1, Fletcher et al. 2008). Within this domain, we analyzed coral populations within

and outside of twelve marine protected areas including the Pupukea Marine Life

Conservation District (MLCD), Hanauma Bay MLCD, Waikiki MLCD, and Coconut

Island Hawaii Marine Laboratory Refuge (HMLR) of Oahu; Manele-Hulopoe MLCD of

Lanai; Honolua-Mokuleia Bay MLCD, Molokini MLCD, and Ahihi-Kinau Natural Area

Reserve (NAR) of Maui; Kahoolawe Island Reserve; and Lapakahi MLCD, Waialea Bay

MLCD, and Kealakekua Bay MLCD of Hawaii (Table 4.1). The MPAs have a range of regulatory protections for resource extraction (collecting, fishing, etc.) including partial

protection (PP), customary stewardship (CS), and no-take (NT). Populations were also compared inside and outside of the Hawaiian Islands Humpback Whale National Marine

Sanctuary (HIHWNMS) boundaries (Fig. 4.1) which encompasses five discrete areas around Kauai (with area of the 0-30 m depth range of 36 km2), north Oahu (57 km2), southeast Oahu (31 km2), Lanai (235 km2), and Hawaii (40 km2). Although their current

activities focus only on education, research, and resource protection activities related to humpback whales (Megaptera novaeangliae), the HIHWNMS is undergoing a management plan review in 2012-2013 that may expand their regulatory protections to include additional taxa such as scleractinian corals and should benefit from this analysis.

Niihau Coconut Pupukea Island Hanauma Bay MLCD HMLR Kauai MLCD

Honolua-Mokuleia Bay Oahu Molokai MLCD Waikiki MLCD Maui Lanai Hawaiian Archipelago Manele-Hulopoe Ahihi-Kinau NAR MLCD Kahoolawe Molokini Lapakahi MLCD Island MLCD Reserve

Waialea Bay MLCD

Kealakekua Bay MLCD Hawaii

Hawaiian Islands Humpback Whale National Marine Sanctuary

05010025 Kilometers

Table 4.1 Summary of marine protected areas (MPAs) in the main Hawaiian Islands for this study. Regulatory status includes partially protected (PP), no-take (NT), and customary stewardship (CS) MPAs. Area Island Marine protected area Regulatory (km2) Established Oahu Pupukea MLCD PP 0.71 1983* Coconut Island HMLR NT 0.30 1967 Hanauma Bay MLCD NT 0.41 1967 Waikiki MLCD NT 0.31 1988 Lanai Manele-Hulopoe MLCD PP 1.25 1976 Kahoolawe Kahoolawe Island Reserve CS 26.39** 1994 Maui Honolua-Mokuleia Bay MLCD NT 0.18 1978 Ahihi-Kinau NAR CS 3.27 1970 Molokini MLCD NT 0.31 1977 Hawaii Lapakahi MLCD PP 0.59 1979 Waialea Bay MLCD PP 0.14 1985 Kealakekua Bay MLCD PP 1.28 1969 *Pupukea MLCD boundary expanded and rules modified in 2003 **Kahoolawe Island Reserve area only includes seafloor of 0-30 m depth for study

The collection or take of stony corals is regulated in Hawaiian state waters

(Hawaii Administrative Rule 13-95) so there is a baseline level of protection throughout the entire study domain. Although there are 61 MPAs in the state of Hawaii, we selected the 13 MPAs due to their current (or potential, for the HIHWNMS) enhanced level of marine resource protection for coral reefs. The other 48 MPAs were not included in the study since they protect only species captured by the marine aquaria reef fish fishery or deepwater bottomfish fishery, are outside the MHI, or have very limited regulatory protections (NMPAC 2012).

Geospatial Data of Coral Abundance

We used geospatial data of model-derived abundance (as % benthic cover) for six

Hawaiian coral species (M. capitata, M. flabellata, M. patula, P. meandrina, P. compressa, P. lobata) from Chapter 3 to estimate population abundances within and outside MPAs in the MHI. These six coral species represent approximately 98% of the total coral cover found at an array of monitoring sites in Hawaiian waters (Jokiel et al.

2004). Optimal boosted regression tree (BRT) models of abundance were constructed for each coral species using the routines gbm version 1.6–32 (Ridgeway 2012) and gbm.step

(Elith et al. 2008) in the R statistical program version 2.14 (R Development Core Team, http://www.r-project.org). The optimal models for all species utilized nine environmental covariate data layers: depth, bathymetric aspect, bathymetric rugosity, bathymetric slope, proportion of sand bottom (in relation to hardbottom), maximum significant wave height, mean significant wave height, downwelled irradiance, and island. Digital files for all

coral species and environmental data layers were georectified to a base analysis grid of

~50 m (0.00005 degree) resolution with a total of 477,795 cells (approximately 1,194

km2) that covered the extent of the study domain in ArcGIS (v. 9.3.1, ESRI 2009) and geoprocessed using scripts in Python (http://www.python.org). Coral abundance (as %

benthic cover) data from the final BRT models were projected to the geography of the

study domain to create geospatial data files used in the statistical analysis. For details of

the model and mapping procedures, see Methods in Chapter 3.

Statistical Analysis

For each coral species, analyses were performed using the cover (%) values of grid cells

from the niche models as individual observations. Permutational univariate one-way

analysis of variance was used to test for differences in coral cover for each species within

island groups between MPAs and unprotected reef areas. For the six coral species, the

abundance in each grid was used as a replicate for the analysis. Significant population

genetic structure exists for a broad range of marine taxa in the MHI such that geographic

barriers to biological connectivity have been suggested between four island groups: 1)

Kauai and Niihau, 2) Oahu, 3) Maui Nui (Maui, Molokai, Lanai, Kahoolawe), and 4)

Hawaii (Toonen et al. 2011). Thus, coral populations amongst the Hawaiian MPA

network were evaluated within these biologically-relevant groups that reflect the spatially

distinct population dynamics of each group (Gaines et al. 2010). Since the coral species

distribution model data inputs were independently derived for each species, tests for

differences in abundance between MPAs and open areas were conducted separately for

each species and island group. Tests were performed for coral abundances pooled in all

MPAs by island area as well as within each individual MPA compared to open areas for

the island groups. Abundance data (% cover) were arcsine square root transformed before

analysis, and all analyses were completed in the R statistical program version 2.14 (R

Development Core Team, http://www.r-project.org). Only grid cells with hard-bottom

were included in the analyses (i.e., proportion of sand-bottom < 1.0). Data were

converted back to abundance (% cover) after analysis for summary statistics and figures.

Results

For the main Hawaiian Islands, abundances of Montipora capitata (2.5%), Porites compressa (1.4%), and P. lobata (5.4%) were higher inside MPAs than unprotected areas

(1.4%, 0.8%, 4.4%) while MPA abundances of M. flabellata (0.04%), M. patula (1.9%),

and Pocillopora meandrina (0.8%) were lower than non-MPA reefs (0.3%, 2.2%, 1.5%).

Comparing within the more biologically-relevant island groups, abundances of the two

Porites species were higher in MPAs than open areas. The three Montipora species and

Pocillopora meandrina had lower abundances in most MPAs compared to open areas

(Fig. 4.2). The abundances of Porites compressa were higher in MPAs of Oahu (1.2%),

Maui Nui (1.4%), and Hawaii (2.3%) than open areas of those islands (0.6%, 1.0%, and

2.0%) while P. lobata had higher MPA abundances in Maui Nui (5.3%) and Hawaii

(11.6%) than open areas (3.8% and 11.0%). In contrast, Montipora capitata, M.

flabellata, M. patula, and Pocillopora meandrina had lower MPA abundances for most

islands except Maui Nui where M. capitata (2.6%) and M. patula (2.0%) had higher

MPA abundances than open areas (1.4% and 1.5%). Abundances for the open area of

Kauai where no coral MPAs currently exist were P. compressa (0.1%), P. lobata (2.7%),

M. capitata (1.6%), M. flabellata (0.3%), M. patula (2.9%), and P. meandrina (1.3%).

All ANOVA test results were statistically significant (p << 0.01) primarily due to the

large sample sizes analyzed (i.e., n > 10,000).

Differences of coral abundance between individual MPAs and open areas were

similar to the pooled MPA results with Porites species having higher abundances in more

MPAs than Montipora species and Pocillopora meandrina (Fig 4.3). The Montipora

species were less abundant in MPAs with Montipora patula cover higher in 4 of 12

MPAs, M. capitata greater in 3 of 12 MPAs, and M. flabellata in 2 of 12 MPAs (Fig

4.3a-c). Pocillopora meandrina abundance was greater than open areas in only one MPA,

the Manele-Hulopoe MLCD (Fig 4.3d). Porites lobata cover (%) was greater in 7 of 12

MPAs than open areas and P. compressa abundance was higher in 6 of 12 MPAs (Fig

4.3e-f). Three MPAs contained higher abundances of 4 of the 6 coral species than open

areas: Manele-Hulopoe MLCD, Kahoolawe Island Reserve, and Molokini MLCD.

Abundances in two MPAs (Waikiki MLCD and Waialea Bay MLCD) were not greater

than open areas for any coral species.

Coral abundances were higher in the HIHWNMS than open areas for Montipora

species, Pocillopora meandrina, and Porites species for several island areas (Fig 4.4).

Around Kauai, cover in the HIHWNMS of M. flabellata (0.5%), M. patula (3.2%), and

Pocillopora meandrina (2.2%) was higher than open areas (0.3%, 2.8%, and 1.2%).

Abundances of the two Porites species were higher in the HIHWNMS for all islands

(a) Montipora capitata 5.0 Open MPA (%)

Cover 0.0 Kauai Oahu Maui Hawaii (b) Montipora flabellata 0.5 (%)

Cover 0.0 Kauai Oahu Maui Hawaii

(d) Pocillopora meandrina 4.0 (%) 2.0

Cover 0.0 Kauai Oahu Maui Hawaii

(e) Porites compressa 4.0 (%) 2.0

Cover 0.0 Kauai Oahu Maui Hawaii

(f) Porites lobata 15.0 (%)

10.0 5.0 Cover 0.0 Kauai Oahu Maui Hawaii

(a) Montipora capitata 10.0 (%) 5.0 Cover 0.0

(b) Montipora flabellata 0.5 (%)

Cover 0.0

(d) Pocillopora meandrina 2.0 (%)

Cover 0.0 (e) Porites compressa 10.0 (%)

Cover 0.0 (f) Porites lobata 15.0

(%) 10.0

5.0 Cover 0.0

(a) Montipora capitata 2.0 Open HIHWNMS (%)

Cover 0.0 Kauai Oahu Maui Hawaii (b) Montipora flabellata 0.6 (%)

0.4 0.2 Cover 0.0 Kauai Oahu Maui Hawaii

Cover 0.0 Kauai Oahu Maui Hawaii

(d) Pocillopora meandrina 4.0 (%) 2.0 Cover 0.0 Kauai Oahu Maui Hawaii

(e) Porites compressa 10.0 (%) 5.0 0.0 Cover Kauai Oahu Maui Hawaii

(f) Porites lobata 20.0 (%) 10.0 Cover 0.0 Kauai Oahu Maui Hawaii

except P. compressa around Oahu. Cover of M. capitata was higher in the HIHWNMS

(1.7%) than open areas only around Maui (1.1%). The highest coral species cover in any

area and island was P. lobata (14.5%) in the HIHWNMS around Hawaii.

Discussion

One of the primary objectives of MPAs is to maintain abundances of important species at levels higher than adjacent unprotected areas (Agardy 1997; NRC 2001; Edgar et al.

2007). For coral reef ecosystems, the foundation species are scleractinian corals; thus, coral species are the most critical biological constituents to conserve in coral reef-focused

MPAs. In contrast to the positive findings of a global study of coral abundances in and adjacent to MPAs (Selig and Bruno 2010), abundances in a network of Hawaiian MPAs were consistently lower than unprotected areas for four of the six dominant coral species.

The period of time since establishment for the Hawaiian MPAs (16-43 years ago) should be sufficient for coral abundances within the MPAs to increase relative to unprotected areas (Selig and Bruno 2010). Of the coral species abundances in Hawaiian MPAs, over half were at least 50% lower than those in unprotected areas (Fig. 4.2). The current network of MPAs in the main Hawaiian Islands does not appear to encompass areas with sufficiently high coral abundances to exceed those of the broader ecosystem. To address this deficit, future MPAs could be sited in areas of high abundances of the three

Montiporid species and Pocillopora meandrina especially in the Kauai, Oahu and Hawaii island groups.

An evaluation of coral abundances inside and outside the network of Hawaiian

MPAs suggests a bias toward protection of reefs dominated by Porites compressa and P.

lobata. In the Hawaiian Islands, reefs dominated by P. compressa are typically located in

areas with low wave energy such as sheltered embayments like Kaneohe Bay or

Hanauma Bay (Jokiel et al. 2004, Engels et al. 2004, Chapter 3). Since a primary goal of

MLCDs is to provide recreational opportunities (A. Clark, DLNR, pers. com.) then the

disproportionate establishment of MPAs in low wave energy areas that support high

abundances of P. compressa is not surprising. These areas are typically ideal for

snorkeling and diving. In contrast, corals associated with high wave energy environments

such as Pocillopora meandrina and Montipora flabellata have lower abundances in

MPAs within all island groups. High wave energy locations may present more dangerous

ocean conditions and thus lower levels of ocean activities by resource users. Potential

MPAs in those areas may require different criteria from recreational use for their

establishment but should be considered since Pocillopora meandrina abundance was

higher in only a single existing MPA (Manele-Hulopoe MLCD) than unprotected areas.

The most common Hawaiian coral species, Porites lobata, is found in a variety of

lagoonal and fore reef habitats with moderate wave energy that are also popular for

human recreation (Dollar 1982, Jokiel et al. 2004, Chapter 3) and is disproportionately

abundant in the majority of MPAs in the study.

The successful process of MPA siting requires integrated ecological and

socioeconomic planning activities (Halpern et al. 2010, Cinner et al. 2012) so we do not

identify particular locations of high coral abundance as candidate sites here. Instead, the

coral abundance information will be used to support the ongoing HIHWNMS

management plan review process to evaluate potential actions that may protect these

coral species (M. Chow, HIHWNMS, pers. com.). Within the current boundaries of the

HIHWNMS, species abundances for Montipora flabellata, M. patula, and Pocillopora

meandrina exceed unprotected areas around the Kauai and Oahu island groups. Using species abundance as a planning criterion would provide an ecological foundation for the

evaluation of potential MPA sites in the Hawaiian Islands and provide a more effective approach than using surrogates for species-level information, such as habitat or

environment, that often do not correlate well with intended conservation targets (Beger et al. 2007, Caro 2010, Dalleau et al. 2011).

Although the study represents a significant contribution to understanding the spatial population structure of common Hawaiian reef corals within the MPA network there are a few caveats to address. Our study utilizes the best available data for the temporal and spatial domain of the analysis (see Chapter 3 for methodological details).

Several of the test comparisons reflected small absolute differences in coral abundance (<

1%) which may not represent biologically significant differences between MPAs and unprotected areas although totaled across the seascape even the small differences can contribute to large variances in absolute area of coral cover. Although we compare population abundances between MPAs and unprotected areas, the detection of an “MPA effect” is best accomplished through a rigorous BACIPS design (Stewart-Oaten et al.

1986, Underwood 1994, Russ 2002). Furthermore, our seascape approach is not a traditional marine landscape ecology analysis in that process is inferred from the pattern of the marine landscape or seascape (Turner et al. 2001, Franklin 2010) but rather describes the all-encompassing spatial scope of the study.

Using continuous spatial distribution maps of niche model-derived abundance, we introduce a novel approach to the seascape-scale analysis of the MPA network for the shallow reefs of the main Hawaiian Islands. Prior work in Hawaii evaluated reef fish populations inside and adjacent to MPAs using a spatially-explicit “seascape” approach

(Friedlander et al. 2007, Wedding and Friedlander 2008, Friedlander et al. 2010) yet these studies were geographically limited to a subset of MPAs and nearby unprotected areas that did not encompass the entire domain of shallow coral reefs (i.e., the seascape).

We expand upon their seascape approach to provide comprehensive environmental and abundance data across an array of non-overlapping, geographic grid units conducive to

MPA evaluation, population connectivity studies (Rivera et al. 2011), marine spatial planning activities (Crowder and Norse 2008), and multispecies survey sampling designs

(Smith et al. 2011). By estimating abundances across all reef areas, this study provides the first comprehensive species-level dataset for spatially-explicit population analyses of corals in a Hawaiian MPA network.

References

Agardy TS (1997) Marine protected areas and ocean conservation. Academic Press, San Diego, CA, USA. Anderson MJ (2001) A new method for non-parametric multivariate analysis of variance. Austral. Ecol. 26:32-46. Battista TA, Costa BM, Anderson SM (2007) Shallow-water benthic habitats of the main eight Hawaiian Islands (DVD). NOAA Tech Memo NOS NCCOS 61. Silver Spring, MD, USA. Beger M, McKenna SA, Possingham HP (2007) Effectiveness of surrogate taxa in the design of coral reef reserve systems in the Indo-Pacific. Conservation Biology 21: 1584−1593. Bellwood DR, Hughes TP, Folke C, Nystrom M (2004) Confronting the coral reef crisis. Nature 429: 827–833.

Caro TM (2010) Conservation by proxies: indicator, umbrella, keystone, flagship, and other surrogate species. Island Press, Washington, DC, USA Cinner JE, McClanahan TR, MacNeil MA, Graham NAJ, Daw TM, Mukminin, A, Feary DA, Rabearisoa AL, Wamukota A, Jiddawi N, Campbell SJ, Baird AH, Januchowski-Hartley, FA, Hamed S, Lahari R, Morove T, Kuange J (2012) Comanagement of coral reef social-ecological systems. PNAS 109: 5219-5222. Crowder L, Norse E (2008) Essential ecological insights for marine ecosystem-based management and marine spatial planning. Mar Policy 32:772-778 Dalleau M, Andréfouët S, Wabnitz CCC, Payri C and others (2010) Use of habitats as surrogates of biodiversity for efficient coral reef conservation planning in Pacific Ocean Islands. Conservation Biology 24: 541−552. Dollar SJ (1982) Wave stress and coral community structure in Hawaii. Coral Reefs 1:71- 81. Edgar GJ, Russ GR, Babcock RC (2007) Marine protected areas. In Connell SD, Gillanders BM. Marine Ecology. Oxford University Press, South Melbourne, Australia. Pp. 533-555. Engels MS, Fletcher CH, Field ME, Storlazzi CD, Grossman EE, Rooney JJB, Conger CL, Glenn C (2004) Holocene reef accretion: southwest Molokai, Hawaii, U.S.A. J Sed Res 74:255-269 Franklin EC (2010) Marine landscape ecology emerges from developments in seafloor mapping, GPS, and GIS. In Breman J. Ocean Globe. ESRI Press, Redlands, CA. Friedlander AM, Brown EK, Jokiel PL, Smith WR, Rodgers KS (2003) Effects of habitat, wave exposure, and marine protected area status on coral reef fish assemblages in the Hawaiian archipelago. Coral Reefs 22:291-305. Friedlander AM, Brown EK, Monaco ME (2007) Coupling ecology and GIS to evaluate efficacy of marine protected areas in Hawaii. Ecological Applications 17:715- 730. Friedlander AM, Wedding LM, Brown E, Monaco ME (2010) Monitoring Hawaii’s marine protected areas: examining spatial and temporal trends using a seascape approach. NOAA Tech Memo NOS NCCOS 117. Silver Spring, MD. Gaines SD, White C, Carr MH, Palumbi SR (2010) Designing marine reserve networks for both conservation and fisheries management. PNAS 107:18286-18293. Graham NAJ, McClanahan TR, MacNeil MA, Wilson SK, Polunin NVC, et al. (2008) Climate warming, marine protected areas and the ocean-scale integrity of coral reef ecosystems. PLoS ONE 3: e3039. Halpern BS, Warner RR (2002) Marine reserves have rapid and lasting effects. Ecology Letters 5: 361-66. Halpern BS, Lester SE, McLeod KL (2010) Placing marine protected areas onto the ecosystem-based management seascape. Proceedings of the National Academy of Sciences of the USA 107:18312-18317. Hughes TP, Baird AH, Bellwood DR, Card M, Connolly SR, et al. (2003) Climate change, human impacts, and the resilience of coral reefs. Science 301:929–933. Jokiel PL, Brown EK, Friedlander A, Rodgers SK, Smith WR (2004) Hawai’i coral reef assessment and monitoring program: spatial patterns and temporal dynamics in reef coral communities. Pac Sci 58:159-174. Knowlton N, Brainard RE, Fisher R, Moews M, Plaisance L, Caley MJ (2010) Coral reef

biodiversity. In McIntyre AD. Life in the world’s oceans: diversty, distribution, and abundance. Wiley-Blackwell. West Sussex, UK. Laffoley (2008) Towards networks of marine protected areas. The MPA plan of action for IUCN’s World Commission on Protected Areas. IUCN WCPA, Gland, Switzerland. Mora C, Andréfouët S, Costello M, Kranenburg S, Rollo A, Veron J, Gaston KJ, Myers RA (2006) Coral reefs and the global network of Marine Protected Areas. Science 312, 1750-1751. NOAA National Marine Protected Areas Center [NMPAC]. 2012. Marine Protected Area Inventory [Data file]. http://www.mpa.gov/dataanalysis/mpainventory/. National Research Council [NRC] (2001) Marine protected areas: tools for sustaining ocean ecosystems. National Academy Press, Washington, DC, USA. Rice J, Houston K (2011) Representativity and networks of marine protected areas. Aquatic Conservation Marine and Freshwater Ecosystems 21:649-657. Rivera MAJ, Andrews KR, Kobayashi DR, Wren JLK, Kelley C, Roderick GK, Toonen RJ (2011) Genetic analyses and simulations of larval dispersal reveal distinct populations and directional connectivity across the range of the Hawaiian grouper (Epinephelus quernus). Journal of Marine Biology 2011:765353 doi:10.1155/2011/765353 Roberts C, Andelman S, Branch G, Bustamante R, Castilla JC, Dugan J, Halpern B, Leslie H, Lafferty K, Lubchenco J, McArdle D, Possingham H, Ruckleshaus M, Warner R (2003) Ecological criteria for evaluating candidate sites for marine reserves. Ecological Applications 13: S199-S214. Russ, GR (2002) Yet another review of marine reserves as reef fishery management tools. In: Sale P (ed.) Coral Reef Fishes: dynamic and diversity in a complex ecosystem. Elsevier, San Diego, CA, USA, pp. 421-443. Sebens KP (1994) Biodiversity of coral-reefs - what are we losing and why. American Zoologist 34: 115–133. Selig ER, Bruno JF (2010) A global analysis of the effectiveness of marine protected areas in preventing coral loss. PLoS ONE 5:e9278. Smith SG, Swanson DW, Chiappone M, Miller SL, Ault JS (2011) Probability sampling of stony coral populations in the Florida Keys. Environmental Monitoring and Assessment 183:121-138. Stewart-Oaten A, Murdoch WW, Parker KR (1986) Environmental impact assessment: "pseudoreplication" in time? Ecology 67: 929-940. Tissot BN, Hallacher LE (2003) Effects of aquarium collectors on coral reef fishes in Kona, Hawaii. Conservation Biology 17:1759-1768. Tissot BN, Walsh WJ, Hallacher LE (2004) Evaluating effectiveness of a marine protected area network in west Hawai’I to increase productivity of an aquarium fishery. Pacific Science 58:175-188 Toonen RJ, Andrews KR, Baums IB, Bird CE, Concepcion GT, Daly-Engel TS, Eble JA, Faucci A, Gaither MR, Iacchei M, Puritz JB, Schultz JK, Skillings DJ, Timmers MA, Bowen BW (2011) Defining boundaries for ecosystem-based management: a multispecies case study of marine connectivity across the Hawaiian archipelago. Journal of Marine Biology 2011:460173. doi:10.1155/2011/460173. Turner MG, Gardner RH, O’Neill RV (2001) Landscape ecology in theory and practice.

Springer-Verlag, New York, NY, USA Underwood A] (1994) On beyond BACI: Sampling designs that might reliably detect environmental disturbances. Ecological Applications 4: 3-15. WCPA/IUCN [World Commission on Protected Areas/International Union for the Conservation of Nature] (2007) Establishing networks of marine protected areas: a guide for developing national and regional capacity for building MPA networks. Gland, Switzerland: WCPA/IUCN. Wedding LM, Friedlander AM (2008) Determining the influence of seascape structure on coral reef fishes in Hawaii using a geospatial approach. Marine Geodesy 31:246- 266.

CHAPTER 5

SUMMARY AND CONCLUSIONS

The aim of this dissertation was to develop spatially-explicit predictive distributions of

coral species in the main Hawaiian Islands. This effort included compiling a database of

over 15,000 observations for Montipora capitata, M. flabellata, M. patula, Pocillopora

meandrina, Porites compressa, and P. lobata from shallow reefs (less than 30 m depth)

during 2000-2009 and environmental covariate data layers of significant wave height,

depth, geomorphology, and light. Species distribution models were constructed: (a) using

an ensemble model approach for the presence of four coral species around Oahu, (b)

using a fully explored single method (boosted regression trees) for the abundance (as %

benthic cover) of the six most common coral species for the main Hawaiian Islands, and

Islands with unprotected areas as well as the inside and outside the HIHWNMS. Main findings of the work were:

 Species distribution modeling approaches are an effective means to characterize

the distribution and abundance of corals in the Hawaiian Islands.

 Mean significant wave height and max significant wave height were the most

influential variables explaining coral abundance (as benthic cover) in the

Hawaiian Islands.

 Models also identified relationships between coral cover and island, depth,

downwelled irradiance, rugosity, slope, and aspect.

 The rank order of coral abundance (from highest to lowest) for the MHI was P.

lobata, M. patula, P. meandrina, M. capitata, P. compressa, and M. flabellata.

 Abundances of the two Porites species were higher in MPAs than open areas.

 The three Montipora species and Pocillopora meandrina had lower abundances in

most MPAs compared to open areas.

 Manele-Hulopoe and Molokini Marine Life Conservation Districts (MLCD) had

higher abundances for four of the six coral species compared to unprotected areas

while Waikiki MLCD had lower abundances than open areas for all corals.

 The HIHWNMS encompassed coral populations with higher abundances than

areas outside the boundaries especially for the four corals (Montipora spp. and P.

meandrina) underrepresented in the current MPA network.

Implications and Applications

This work strongly demonstrates that species distribution modeling approaches are an effective means to characterize the distribution and abundance of corals in the Hawaiian

Islands. In Hawaii, regional scale studies of corals have typically utilized coarse habitat descriptors (such as fore reef, patch reef, etc.) or community descriptors (e.g., diversity, richness, total coral cover) as metrics of reef condition (Grigg 1983, Grigg 1998,

Friedlander et al. 2003, Jokiel et al. 2004). This approach of aggregating information for corals at a habitat or community level may neglect or overlook critical dynamics that are occurring at the species level (Van Woesik 2001, Carpenter et al. 2008, Clark et al.

2009). For example, coral species have exhibited differential responses to climate perturbations such as thermal stressors (Baird and Marshall 2002) so that accurate predictions of potential changes in coral ecosystems require knowledge of the reef community species composition (Guest et al. 2012). Furthermore, benthic cover of lobate corals (e.g., Porites lobata) and branching corals (e.g., Pocillopora meandrina) are predictors of coral reef fish species richness and number of individuals on Hawaiian reefs

(Friedlander et al. 2003). The utility of SDMs to provide coral species abundances at a high map resolution across the entire geographic domain of the main Hawaiian Islands represents a significant improvement in our ability to describe the condition of these marine populations.

Another important outcome of this research is the establishment of a methodological system to perform distributional analysis for other marine species in

Hawaii (reef fish, invertebrates, etc.). The SDM approach can also be applied directly to a contemporary or future evaluation of the efficacy of the Hawaiian MPA network for a broad suite of modeled species. Accurate knowledge of the extent and condition of individual reef-associated species is critically important in climate science, ecosystem based management, and marine spatial planning for a variety of future research projects.

Future Research

Several interesting conclusions have been presented here that naturally lead to future research topics to be addressed. A few of the more promising are:

 Apply SDMs in Hawaii to other coral and non-coral species. This work

provides an initial template for SDM modeling of coral reef associated

organisms in Hawaii but there are hundreds to thousands of additional

organisms that could be modeled such as reef reef fish, marine algae, and

echinoderms.

 Explore additional environmental covariates. Although model performances

were good to excellent, there are several explanatory processes which could

be further explored including the influence of nearshore inputs (sediment,

nutrients, and turbidity), empirical measures of light and ocean temperatures

at depth, and biological variables (presence or abundance of predators,

competitors, facilitators, etc.).

 Compare SDM approaches with models that incorporate a term for location.

The SDMs used here are spatially implicit, in that the models are constructed

in a statistical framework that does not consider location explicitly. It would

be fruitful to compare the results of these models with those from spatially-

explicit modeling approaches such as hierarchical Bayesian spatial models.

Conclusion

The central focus of this research was to create spatially-realistic populations of

coral species in Hawaiian waters. By achieving this objective, I’ve established a

foundation for region-wide studies of population connectivity and climate change impacts

in coral reef ecosystems as well as marine conservation zoning activities. Coral reefs are

undergoing rapid change but species responses to environmental drivers are heterogeneous. This work will serve as the framework for future investigations to better assess coral species conditions and understand the change that reefs are experiencing.

References

Baird AH, Marshall PA (2002) Mortality, growth and reproduction in scleractinian corals following bleaching on the Great Barrier Reef. Marine Ecology Progress Series 237:133-141 Carpenter KE, Abrar M, Aeby G, Aronson RB, Banks S, Bruckner A, et al. (2008) One- third of reef-building corals face elevated extinction risk from climate change and local impacts. Science 321:560-563 Clark JS (2009) Beyond neutral science. Trends in Ecology & Evolution 24:8-15 Friedlander AM, Brown EK, Jokiel PL, Smith WR, Rodgers KS (2003) Effects of habitat, wave exposure, and marine protected area status on coral reef fish assemblages in the Hawaiian archipelago. Coral Reefs 22:291-305 Grigg RW (1983) Community structure, succession, and development of coral reefs in Hawaii. Mar Ecol Prog Ser 11:1-14 Grigg RW (1998) Holocene coral reef accretion in Hawaii: a function of wave exposure and sea level history. Coral Reefs 17:263-272 Guest JR, Baird AH, Maynard JA, Muttaqin E, Edwards AJ, et al. (2012) Contrasting patterns of coral bleaching susceptibility in 2010 suggest an adaptive response to thermal stress. PLoS ONE 7(3): e33353. doi:10.1371/journal.pone.0033353 Jokiel PL, Brown EK, Friedlander A, Rodgers SK, Smith WR (2004) Hawai’i coral reef assessment and monitoring program: spatial patterns and temporal dynamics in reef coral communities. Pac Sci 58:159-174 van Woesik R (2001) Coral bleaching: transcending spatial and temporal scales. Trends in Ecology & Evolution 16:119-121

APPENDIX A

FIGURES OF ENVIRONMENTAL COVARIATE DATA LAYERS FOR OAHU

. Figure A.1 Shallow bathymetry (0 to 30 m depth) of the waters around Oahu.

Figure A.2 Bathymetric slope (in degrees) of the sea floor around Oahu.

Figure A.3 Bathymetric aspect (in degrees) of the seafloor around Oahu.

Figure A.4 Bathymetric rugosity (surface area / planar area) of the sea floor around Oahu.

Figure A.5 Sand habitat (proportion) of the sea floor around Oahu.

Figure A.6 Maximum significant wave heights for the waters around Oahu.

Figure A.7 Mean significant wave heights for waters the around Oahu.

Figure A.8 Downwelled irradiance at the sea floor relative to that just below the sea surface around Oahu.

APPENDIX B

TABLE OF RELATIVE VARIABLE IMPORTANCE FOR ENVIRONMENTAL

COVARIATES IN CORAL SPECIES DISTRIBUTION MODELS FOR OAHU

Table B.1 Variable importance for environmental covariates for each model approach and coral species. Variables are aspect (Asp), mean significant wave height (Hs_av), rugosity (Rug), depth (Dpth), downwelled irradiance (Irrad), slope (Slope), and sand (Sand). Montipora capitata Environmental Variables Models Asp Hs_av Rug Hs_mx Dpth Irrad Slope Sand ANN 0.337 0.031 0.000 0.176 0.476 0.209 0.175 0.081 CTA 0.254 0.206 0.052 0.300 0.563 0.000 0.141 0.000 GAM 0.064 0.390 0.000 0.000 0.383 0.000 0.000 0.045 GBM 0.046 0.352 0.012 0.012 0.129 0.083 0.005 0.002 GLM 0.074 0.501 0.000 0.000 0.044 0.143 0.000 0.064 MARS 0.000 0.856 0.000 0.136 0.000 0.260 0.000 0.078 FDA 0.023 0.656 0.000 0.104 0.000 0.364 0.000 0.056 RF 0.050 0.308 0.020 0.049 0.124 0.111 0.013 0.018 SRE 0.007 0.036 0.126 0.000 0.029 0.043 0.066 0.015

Pocillopora meandrina Environmental Variables Models Asp Hs_av Rug Hs_mx Dpth Irrad Slope Sand ANN 0.237 0.176 0.000 0.665 0.192 0.023 0.384 0.072 CTA 0.068 0.000 0.356 0.783 0.000 0.171 0.000 0.000 GAM 0.128 0.000 0.000 0.383 0.000 0.145 0.195 0.083 GBM 0.094 0.015 0.183 0.504 0.007 0.061 0.004 0.011 GLM 0.108 0.000 0.000 0.408 0.000 0.130 0.245 0.083 MARS 0.067 0.396 0.000 0.766 0.231 0.000 0.386 0.000 FDA 0.084 0.000 0.000 0.249 0.000 0.233 0.332 0.000 RF 0.065 0.034 0.160 0.268 0.028 0.078 0.043 0.021 SRE 0.051 0.033 0.038 0.046 0.020 0.019 0.007 0.013

Porites compressa Environmental Variables Models Asp Hs_av Rug Hs_mx Dpth Irrad Slope Sand ANN 0.147 0.125 0.027 0.992 0.028 0.111 0.152 0.012 CTA 0.000 0.067 0.153 0.991 0.248 0.000 0.000 0.000 GAM 0.036 0.239 0.000 1.070 0.000 0.137 0.029 0.036 GBM 0.044 0.023 0.031 0.776 0.030 0.057 0.007 0.003 GLM 0.098 0.225 0.000 1.044 0.000 0.150 0.070 0.038 MARS 0.023 0.536 0.000 0.802 0.099 0.286 0.129 0.028 FDA 0.055 0.212 0.016 0.972 0.091 0.131 0.035 0.013 RF 0.075 0.151 0.043 0.501 0.048 0.096 0.022 0.012 SRE 0.018 0.036 0.018 0.185 0.039 0.076 0.068 0.101

Table B.1 (CONT.) Porites lobata Environmental Variables Models Asp Hs_av Rug Hs_mx Dpth Irrad Slope Sand ANN 0.025 0.125 0.000 0.560 0.327 0.053 0.191 0.059 CTA 0.000 0.111 0.000 0.544 0.368 0.000 0.000 0.000 GAM 0.029 0.362 0.149 0.000 0.270 0.000 0.000 0.080 GBM 0.009 0.101 0.006 0.354 0.142 0.002 0.004 0.003 GLM 0.055 0.261 0.264 0.000 0.275 0.000 0.196 0.073 MARS 0.000 0.189 0.000 0.197 0.747 0.245 0.000 0.000 FDA 0.000 0.342 0.000 0.715 0.339 0.053 0.000 0.000 RF 0.017 0.131 0.025 0.254 0.160 0.019 0.017 0.014 SRE 0.019 0.026 0.026 0.058 0.026 0.019 0.032 0.026 RF 0.017 0.131 0.025 0.254 0.160 0.019 0.017 0.014 SRE 0.019 0.026 0.026 0.058 0.026 0.019 0.032 0.026

APPENDIX C

COMPARATIVE RESPONSE PLOTS OF ENVIRONMENTAL VARIABLES FROM

EIGHT MODEL METHODS FOR FOUR HAWAIIAN CORAL SPECIES AROUND

OAHU, HAWAII

The response plots are an adaption of the evaluation strip method proposed by Elith et al.

(Ecol Modelling 186:280-289). Plots use data from the full model for each coral species.

Model methods are ANN = artificial neural networks, CTA = classification tree analysis,

FDA = flexible discriminant analysis, GAM = generalized additive models, GBM = generalized boosted regression models, GLM = generalized linear models, MARS = multivariate adaptive regression splines, RF = random forests.

ANN CTA FDA GAM GBM GLM MARS RF Probabilityoccurrence of 0.0 0.2 0.4 0.6 0.8 1.0

-20 -15 -10 -5 0

Depth (m)

Figure C.1 Response plots of Montipora capitata probability of occurrence to depth from eight model methods.

ANN CTA FDA GAM GBM GLM MARS RF Probabilityoccurrence of 0.0 0.2 0.4 0.6 0.8 1.0

02468

Slope (Degrees)

Figure C.2 Response plots of Montipora capitata probability of occurrence to slope from eight model methods.

ANN CTA FDA GAM GBM GLM MARS RF Probabilityoccurrence of 0.0 0.2 0.4 0.6 0.8 1.0

0 50 100 150 200 250 300 350

Aspect (Compass Degrees)

Figure C.3 Response plots of Montipora capitata probability of occurrence to aspect from eight model methods.

ANN CTA FDA GAM GBM GLM MARS RF Probabilityoccurrence of 0.0 0.2 0.4 0.6 0.8 1.0

1.000 1.005 1.010 1.015 1.020

Rugosity (Surface:Planar Area)

Figure C.4 Response plots of Montipora capitata probability of occurrence to rugosity from eight model methods.

ANN CTA FDA GAM GBM GLM MARS RF Probabilityoccurrence of 0.0 0.2 0.4 0.6 0.8 1.0

0.00.20.40.60.81.0

Sandbottom (Proportion)

Figure C.5 Response plots of Montipora capitata probability of occurrence to sandbottom from eight model methods.

ANN CTA FDA GAM GBM GLM MARS RF Probabilityoccurrence of 0.0 0.2 0.4 0.6 0.8 1.0

012345

Maximum Significant Wave Height (m)

Figure C.6 Response plots of Montipora capitata probability of occurrence to maximum significant wave height from eight model methods.

ANN CTA FDA

Probabilityoccurrence of GAM GBM GLM MARS RF 0.0 0.2 0.4 0.6 0.8 1.0

0.0 0.5 1.0 1.5

Mean Significant Wave Height (m)

Figure C.7 Response plots of Montipora capitata probability of occurrence to mean significant wave height from eight model methods.

ANN CTA FDA GAM GBM GLM MARS RF Probabilityoccurrence of 0.0 0.2 0.4 0.6 0.8 1.0

0.0 0.2 0.4 0.6 0.8 1.0

Downwelled Irradiance (Light at Surface/Light at Depth)

Figure C.8 Response plots of Montipora capitata probability of occurrence to downwelled irradiance from eight model methods.

ANN CTA FDA

Probability occurrence of GAM GBM GLM MARS RF 0.0 0.2 0.4 0.6 0.8 1.0

-20 -15 -10 -5 0

Depth (m)

Figure C.9 Response plots of Pocillopora meandrina probability of occurrence to depth from eight model methods.

ANN CTA FDA

Probabilityoccurrence of GAM GBM GLM MARS RF 0.0 0.2 0.4 0.6 0.8 1.0

02468

Slope (Degrees)

Figure C.10 Response plots of Pocillopora meandrina probability of occurrence to slope from eight model methods.

ANN CTA FDA

Probabilityoccurrence of GAM GBM GLM MARS RF 0.0 0.2 0.4 0.6 0.8 1.0

0 50 100 150 200 250 300 350

Aspect (Compass Degrees)

Figure C.11 Response plots of Pocillopora meandrina probability of occurrence to aspect from eight model methods.

ANN CTA FDA

Probabilityoccurrence of GAM GBM GLM MARS RF 0.0 0.2 0.4 0.6 0.8 1.0

1.000 1.005 1.010 1.015 1.020

Rugosity (Surface:Planar Area)

Figure C.12 Response plots of Pocillopora meandrina probability of occurrence to rugosity from eight model methods.

ANN CTA FDA

Probabilityoccurrence of GAM GBM GLM MARS RF 0.0 0.2 0.4 0.6 0.8 1.0

0.0 0.2 0.4 0.6 0.8 1.0

Sandbottom (Proportion)

Figure C.13 Response plots of Pocillopora meandrina probability of occurrence to sandbottom from eight model methods.

ANN CTA FDA

Probabilityoccurrence of GAM GBM GLM MARS RF 0.0 0.2 0.4 0.6 0.8 1.0

012345

Maximum Significant Wave Height (m)

Figure C.14 Response plots of Pocillopora meandrina probability of occurrence to maximum significant wave height from eight model methods.

ANN CTA FDA

Probability occurrence of GAM GBM GLM MARS RF 0.0 0.2 0.4 0.6 0.8 1.0

0.0 0.5 1.0 1.5

Mean Significant Wave Height (m)

Figure C.15 Response plots of Pocillopora meandrina probability of occurrence to mean significant wave height from eight model methods.

ANN CTA FDA GAM GBM GLM MARS RF Probability of occurrence of Probability 0.0 0.2 0.4 0.6 0.8 1.0

0.0 0.2 0.4 0.6 0.8 1.0

Downwelled Irradiance (Light at Surface/Light at Depth)

Figure C.16 Response plots of Pocillopora meandrina probability of occurrence to downwelled irradiance from eight model methods.

ANN CTA FDA GAM GBM GLM MARS RF Probability occurrence of 0.0 0.2 0.4 0.6 0.8 1.0

-20 -15 -10 -5 0

Depth(m)

Figure C.17 Response plots of Porites compressa probability of occurrence to depth from eight model methods.

ANN CTA FDA GAM GBM GLM MARS RF Probability of occurrence of Probability 0.0 0.2 0.4 0.6 0.8 1.0

02468

Slope (Degrees)

Figure C.18 Response plots of Porites compressa probability of occurrence to slope from eight model methods.

ANN CTA FDA GAM GBM GLM MARS RF Probability occurrence of 0.0 0.2 0.4 0.6 0.8 1.0

0 50 100 150 200 250 300 350

Aspect (Compass Degrees)

Figure C.19 Response plots of Porites compressa probability of occurrence to aspect from eight model methods.

ANN CTA FDA GAM GBM GLM MARS RF Probabilityoccurrence of 0.0 0.2 0.4 0.6 0.8 1.0

1.000 1.005 1.010 1.015 1.020

Rugosity (Surface:Planar Area)

Figure C.20 Response plots of Porites compressa probability of occurrence to rugosity from eight model methods.

ANN CTA FDA GAM GBM GLM MARS RF Probabilityoccurrence of 0.0 0.2 0.4 0.6 0.8 1.0

0.0 0.2 0.4 0.6 0.8 1.0

Sandbottom (Proportion)

Figure C.21 Response plots of Porites compressa probability of occurrence to sandbottom from eight model methods.

ANN CTA FDA GAM GBM GLM MARS RF Probabilityoccurrence of 0.0 0.2 0.4 0.6 0.8 1.0

012345

Maximum Significant Wave Height (m)

Figure C.22 Response plots of Porites compressa probability of occurrence to maximum significant wave height from eight model methods.

ANN CTA FDA

Probabilityoccurrence of GAM GBM GLM MARS RF 0.0 0.2 0.4 0.6 0.8 1.0

0.0 0.5 1.0 1.5

Mean Significant Wave Height (m)

Figure C.23 Response plots of Porites compressa probability of occurrence to mean significant wave height from eight model methods.

ANN CTA FDA GAM GBM GLM MARS RF Probabilityoccurrence of 0.0 0.2 0.4 0.6 0.8 1.0

0.0 0.2 0.4 0.6 0.8 1.0

Downwelled Irradiance (Light at Surface/Light at Depth)

Figure C.24 Response plots of Porites compressa probability of occurrence to downwelled irradiance from eight model methods.

ANN CTA FDA

Probabilityoccurrence of GAM GBM GLM MARS RF 0.0 0.2 0.4 0.6 0.8 1.0

-20 -15 -10 -5 0

Depth(m) Figure C.25 Response plots of Porites lobata probability of occurrence to depth from eight model methods.

100

ANN CTA FDA

Probabilityoccurrence of GAM GBM GLM MARS RF 0.0 0.2 0.4 0.6 0.8 1.0

02468

Slope (Degrees) Figure C.26 Response plots of Porites lobata probability of occurrence to slope from eight model methods.

ANN CTA FDA

Probabilityoccurrence of GAM GBM GLM MARS RF 0.0 0.2 0.4 0.6 0.8 1.0

0 50 100 150 200 250 300 350

Aspect (Compass Degrees)

Figure C.27 Response plots of Porites lobata probability of occurrence to aspect from eight model methods.

101

ANN CTA FDA

Probabilityoccurrence of GAM GBM GLM MARS RF 0.0 0.2 0.4 0.6 0.8 1.0

1.000 1.005 1.010 1.015 1.020

Rugosity (Surface:Planar Area)

Figure C.28 Response plots of Porites lobata probability of occurrence to rugosity from eight model methods.

ANN CTA FDA

Probability occurrence of GAM GBM GLM MARS RF 0.0 0.2 0.4 0.6 0.8 1.0

0.0 0.2 0.4 0.6 0.8 1.0

Sandbottom (Proportion)

Figure C.29 Response plots of Porites lobata probability of occurrence to sandbottom from eight model methods.

102

ANN CTA FDA GAM GBM GLM MARS RF Probability of occurrence Probability of 0.00.20.40.60.81.0

012345

Maximum Significant Wave Height (m)

Figure C.30 Response plots of Porites lobata probability of occurrence to maximum significant wave height from eight model methods.

ANN CTA FDA

Probabilityoccurrence of GAM GBM GLM MARS RF 0.0 0.2 0.4 0.6 0.8 1.0

0.0 0.5 1.0 1.5

Mean Significant Wave Height (m)

Figure C.31 Response plots of Porites lobata probability of occurrence to mean significant wave height from eight model methods.

103

ANN CTA FDA

Probabilityoccurrence of GAM GBM GLM MARS RF 0.0 0.2 0.4 0.6 0.8 1.0

0.0 0.2 0.4 0.6 0.8 1.0

Downwelled Irradiance (Light at Surface/Light at Depth)

Figure C.32 Response plots of Porites lobata probability of occurrence to downwelled irradiance from eight model methods.

104

APPENDIX D

FIGURES OF BENTHIC COVER OBSERVATIONS FOR SIX CORAL SPECIES IN

THE MAIN HAWAIIAN ISLANDS, 2000-2009.

!! !! ! ! !! ! ! ! !! !!! !! ! !! ! ! ! !!!! !! !!! ! ! !

Kauai ! !!! ! !! ! ! !! !! !!!!! Oahu !!!!! !! !!!!!!! ! ! !!!!!! ! ! !!! ! !!!!!! !!!!!! !! ! !!!!! ! !! !!!! ! ! !!!!!! ! !!!!!! ! !!!!! ! ! ! ! !! ! ! ! ! ! ! ! !! ! ! !! Niihau ! ! ! !! ! ! ! !! ! !! ! ! ! ! ! ! !!! ! !!!!! !!!!! !! !!!!!!!! ! !!!!!!!!! !!!!! !!!!! !! ! !!!!!! !!!!!!!

!!!!!!! ! !!! !! ! !!! !! !! Molokai ! !!!!! !! ! !!!!!!!!!!! ! ! ! !! !! ! ! ! ! ! ! ! ! !! ! ! !! ! !! !! !!!!! !! !!!!!! ! !!!! ! M. capitata cover (%) ! ! ! ! ! ! ! ! ! ! ! 75 - 100 ! ! ! ! ! ! ! ! ! ! ! 50 - 75 ! !!! !!! ! ! !!!! ! !!!! ! ! ! ! ! !! Hawaii 40 - 50 ! !! !!! ! ! !!!!!!!!!!!!!!!!!!! Maui !! ! ! !!! !! 30 - 40 !! ! !!!!!! ! !! ! ! ! Lanai !! !! ! ! ! !! !! 20 - 30 !! ! ! ! !!! ! ! !! ! !!!!!! ! !! !! ! ! !!! ! ! 10 - 20 ! ! ! ! 0 - 10 ! ! ! ! 0 Kahoolawe ! ! ! ! ! ! Figure D.1 Benthic cover field observations for Montipora capitata cover (%) in the main Hawaiian Islands from 2000-2009. See Fig. 3.1 for accurate depiction of geographic scale and position for Hawaiian Islands.

105

!! !! ! ! !! ! ! ! !! !!! !! ! !!! ! ! ! !!!! !! !!! ! ! !

Kauai ! !!! ! !! ! ! !! !! !!!!! Oahu !!!!! !! !!!!!!!! ! ! !!!!!! ! ! ! ! ! !!!!!! !!!!!! !! ! !!!!!! ! !! !!!! ! ! !!!!!! !! !!!!!! ! !!!!! ! ! ! ! Niihau ! ! ! ! ! ! ! ! ! ! !!! ! !!!!! !!!! !! !!!!!!!!!! ! !!!!!!!!!! !!!!! !!!!! !! ! !!!!!!! !!!!!!!

!!!!!!! ! !!! !! ! !!! !! !! Molokai ! !!!!!! !! ! !!!!!!!!!!! ! ! ! !! ! ! ! ! ! ! ! ! ! !! ! ! !! ! !! !! !!!!! !! !!!!!! ! !!!! ! M. flabellata cover (%) ! ! ! ! ! ! ! ! ! ! ! 75 - 100 ! ! ! ! ! ! ! ! ! ! ! 50 - 75 ! !!! !!! ! ! !!!! ! !!!! ! ! ! ! ! !! Hawaii 40 - 50 ! !! !!! ! ! !!!!!!!!!!!!!!!!!! Maui !! ! ! !!! !! 30 - 40 !! ! !!!!!! ! !! ! ! ! Lanai !! !!! ! ! ! !! !! 20 - 30 !! ! ! ! !!! ! ! !! ! !!!!!! ! !! !! ! ! 10 - 20 !!! ! ! ! ! ! ! 0 - 10 ! ! ! ! 0 Kahoolawe ! ! ! ! ! !

!! !! ! ! !! ! ! ! !! !!! !! ! !! ! ! ! !!!! !! !!! ! ! !

Kauai ! !!! ! !! ! ! !! !! !!!!! Oahu !!!!! !! !!!!!!! ! ! !!!!!! ! ! !!! ! !!!!!! !!!!!! !! ! !!!!! ! !! !!!! ! ! !!!!! !! !!!!!! ! !!!!! ! ! ! ! Niihau ! ! ! ! ! ! ! ! ! ! !!! ! !!!!! !!!! !! !!!!!!!!!! ! !!!!!!!!! !!!!! !!!!! !! ! !!!!!! !!!!!!!

!!!!!!! ! !!! !! ! !!! !! !! Molokai ! !!!!!! !! ! !!!!!!!!!!! ! ! ! !! !! ! ! ! ! ! ! ! ! !! ! ! !! ! !! !! !!!!! !! !!!!!! ! !!!! ! M. patula cover (%) ! ! ! ! ! ! ! ! ! ! ! 75 - 100 ! ! ! ! ! ! ! ! ! ! ! 50 - 75 ! !!! !! ! ! !!!! ! !!!! ! ! ! ! ! !! Hawaii 40 - 50 ! !! !!! ! ! !!!!!!!!!!!!!!!!!!!! Maui !! ! ! !!! !! !! ! 30 - 40 !!!!!!! ! !! ! ! ! Lanai !! !!! ! ! ! !! ! 20 - 30 !! ! ! ! !! ! ! !! ! !!!!!! ! !! !! ! ! !!! ! ! 10 - 20 ! ! ! ! 0 - 10 ! ! ! ! 0 Kahoolawe ! ! ! ! ! ! Figure D.3 Benthic cover field observations for Montipora patula cover (%) in the main Hawaiian Islands from 2000-2009. See Fig. 3.1 for accurate depiction of geographic scale and position for Hawaiian Islands. 106

!! !! ! ! !! ! ! ! !! !!! !! ! !!! ! ! ! !!!! !! !!! ! ! !

Kauai ! !!! ! !! ! ! !! !! !!!!! Oahu !!!!! !! !!!!!!! ! ! !!!!!!! ! !!! ! !!!!!! !!!!!! !! ! !!!!!! ! !! !!!! ! ! !!!!! !! !!!!!! ! !!!!! ! ! ! ! !! ! ! ! ! ! ! ! !! ! ! !! Niihau ! ! ! !! ! ! ! !! ! !! ! ! ! ! ! ! !!! ! !!!!! !!!! !! !!!!!!!! ! !!!!!!!!! !!!!! !!!!! !! ! !!!!!! !!!!!!!

!!!!! ! !!! !! ! !!! !! !! Molokai ! !!!!!! !! ! !!!!!!!!!!! ! ! ! !! !! ! ! ! ! ! ! ! ! !! ! ! !! ! !! !! !!!!! !! !!!!!! ! !!!! ! P. meandrina cover (%) ! ! ! ! ! ! ! ! ! ! ! 75 - 100 ! ! ! ! ! ! ! ! ! ! ! 50 - 75 ! !!! !! ! ! !!!! !!!!! ! ! ! ! ! !! Hawaii 40 - 50 ! !! !!! ! ! !!!!!!!!!!!!!!!!! Maui !! ! ! !!! !! 30 - 40 !! ! !!!!!! ! !! ! ! ! Lanai !! !!! ! ! ! !!! !! 20 - 30 !! ! ! ! !!! ! ! !! ! !!!!!! ! !! !! ! ! !!! ! ! 10 - 20 ! ! ! ! 0 - 10 ! ! ! ! 0 Kahoolawe ! ! ! ! ! ! Figure D.4 Benthic cover field observations for Pocillopora meandrina cover (%) in the main Hawaiian Islands from 2000-2009. See Fig. 3.1 for accurate depiction of geographic scale and position for Hawaiian Islands.

!!!!!!!!!!! ! !! !!!!! ! !!!!! !!!! !!!!! ! ! ! !!!!!!! !!! ! ! !! ! !!! !!! ! !!!!!! !!!! !!!! ! !!! !!! !! !!!!!!!!!! !!!! !!!! ! ! !! !!!! ! !!! !!! !!! !!!!! !!! !!! ! !! !! !!!!!! ! ! !! !! !! ! !!!!!!!!! !!!!!!! !! !!! !!!!!!!!!! !!! !!! !!! !!! !!!!!!! ! ! !! !!!! Kauai !!!!!!! !!!! !!!! ! !!!!!!! ! !!!!! !!! ! !!!! !!!!!!! ! !!!!!!!!!! !! ! !!! !!! ! !!!! !! !!! !!! ! !!! ! Oahu !!!!! ! !!!!! !!! ! !! ! !!!!!!!!! !! !! ! !!!! !!! !!! !!!!!! !!! !! !! ! ! !!! !!! !!!!!! !! !! !! ! !!! !!! !!!!!!!! ! !!! !!!!! !! ! !!!!!! !! !!!! !! !!! ! !!!!! ! ! !!!!!! !!!!!! ! ! !!!!!!!! ! !!!! ! !!!!!!!! !!!!!!!!!! !!!!!!!! !!!!!!! !!!! !!!!!!!!!! !! !!!!! !!! ! ! ! ! !!!! ! !! ! ! ! !!! ! ! ! !! ! !! ! ! !! !! Niihau ! !! ! !! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! !!! ! !!!!!! !!!!! !! !!!!!!!!!! ! !!!!!!!!!!! !!!!! !!!!! !! ! !!!!!!! !!!!!!!

!! !!!!!!!! ! ! !!! !! !!!! !!! !! !! Molokai ! ! !!!!!! !! ! !!!!!!!!!!!!!!!!! !!!!!!!! ! ! ! !! !! ! !!!!!! !! !!!! ! !!!!! ! ! ! ! ! !!!!! !! !!!! ! ! ! !!! ! !!!!!! !!!!! !! !!!!! !!!! !! !!!!!! !!!! !!!! !!! P. compressa cover (%) ! !!! ! !!! !! !! ! !!!! !!! !!!!!! ! ! ! ! !!!! ! ! ! !!! !! 75 - 100 !!!!! !! ! !! !!!! ! !!! ! !!!!!!!!! !!!!!! ! ! ! ! !! ! !!! ! !! 50 - 75 !!! ! !!!!!!! !!! ! !!! ! ! ! !!!! ! !!!!!! ! ! ! ! ! !! Hawaii 40 - 50 !!!!! ! !! ! !!! ! ! ! !!!!!!!!!!!!!!!!!!!!!!!!!!! Maui !! !!!!! !!! !! !! ! ! ! !!! ! !! 30 - 40 ! !! !!!!! !! !!!! !!! !! ! ! ! ! Lanai !! !!!!! !!! ! !! !! !!!!! ! ! 20 - 30 !! ! !!!!!!!!!!! !! !!! ! ! ! !!!!! !!!! !!! ! !!!!! ! !!!!!!! ! !!!!!!!!!! !!!! !! ! !!!!!! 10 - 20 ! !! ! ! ! 0 - 10 ! !!!!! !!! ! !! 0 Kahoolawe ! ! ! !!!! ! ! !! ! ! !! Figure D.5 Benthic cover field observations for Porites compressa cover (%) in the main Hawaiian Islands from 2000-2009. See Fig. 3.1 for accurate depiction of geographic scale and position for Hawaiian Islands.

107

!! !! ! ! !! ! ! ! !! !!! !! ! !! ! ! ! !!! !! !!! ! ! !

Kauai ! !! ! !! ! ! !! !! !!!!! Oahu !!!!! !! !!!!!!! ! ! !!!!!! ! ! ! ! ! !!!!!! !!!!! !! ! !!!!! ! !! !!!! ! ! !!!!! !! !!!!!! ! !!!!! ! ! ! ! !! ! ! ! ! ! ! ! !! ! ! !! Niihau ! ! ! !! ! ! ! !! ! !! ! ! ! ! ! ! !!! ! !!!!! !!!! !! !!!!!!!! ! !!!!!!!!!! !!!!! !!!!! !! ! !!!!!! !!!!!!!

!!!!! ! !!! !! ! !!! !! !! Molokai ! !!!!! !! ! !!!!!!!!!!! ! ! ! !! !! ! ! ! ! ! ! ! ! !! ! ! !! P. lobata (%) ! ! !! !!!!! !! !!!!!! ! !!!! ! Plob_cov ! ! ! ! ! ! ! ! ! ! ! 75 - 100 ! ! ! ! ! ! ! ! ! ! ! 50 - 75 ! !!! !! ! ! !!!! ! !!!! ! ! ! ! ! !! Hawaii 40 - 50 ! !! !!! ! ! !!!!!!!!!!!!!!! Maui !! ! ! !!! !! 30 - 40 !! ! !!!!!! ! !! ! ! ! Lanai !! !! ! ! ! !!! ! 20 - 30 !! ! ! ! !! ! ! !! ! !!!!!! ! !! !! ! ! !!! ! ! 10 - 20 ! ! ! ! 0 - 10 ! ! ! ! 0 Kahoolawe ! ! ! ! ! ! Figure D.6 Benthic cover field observations for Porites lobata cover (%) in the main Hawaiian Islands from 2000-2009. See Fig. 3.1 for accurate depiction of geographic scale and position for Hawaiian Islands.

108

APPENDIX E

FIGURES OF ENVIRONMENTAL COVARIATE DATA LAYERS FOR THE MAIN

HAWAIIAN ISLANDS

Depth (m)

-30

Oahu Kauai

Niihau

Figure E.1 Shallow bathymetry (0 to 30 m depth) of the waters around Kauai and Niihau.

Molokai

Lanai Maui

Depth (m)

0 Kahoolawe -30

Figure E.2 Shallow bathymetry (0 to 30 m depth) of the waters around Maui Nui.

109

Hawaii

Depth (m)

-30

Figure E.3 Shallow bathymetry (0 to 30 m depth) of the waters around Hawaii.

Slope (°)

High : 14.4

Low : 0.0

Oahu Kauai

Niihau

Figure E.4 Bathymetric slope (in degrees) of the sea floor around Kauai and Niihau.

110

Molokai

Lanai Maui

Slope (°)

High : 14.4 Kahoolawe Low : 0.0

Figure E.5 Bathymetric slope (in degrees) of the sea floor around Maui Nui.

Hawaii

Slope (°)

High : 14.4

Low : 0.0

Figure E.6 Bathymetric slope (in degrees) of the sea floor around Hawaii.

111

Aspect (°)

360

Oahu Kauai

Niihau

Figure E.7 Bathymetric aspect (in degrees) of the sea floor around Kauai and Niihau.

Molokai

Lanai Maui

Aspect (°)

360 Kahoolawe 1

Figure E.8 Bathymetric aspect (in degrees) of the sea floor around Maui Nui.

112

Hawaii

Aspect (°)

360

Figure E.9 Bathymetric aspect (in degrees) of the sea floor around Hawaii.

Rugosity (Surface/Planar Area)

High : 2.5

Low : 1.0

Oahu Kauai

Niihau

Figure E.10 Bathymetric rugosity (surface area / planar area) of the sea floor around Kauai and Niihau.

113

Molokai

Lanai Maui Rugosity (Surface/Planar Area)

High : 2.5 Kahoolawe Low : 1.0

Figure E.11 Bathymetric rugosity (surface area / planar area) of the sea floor around Maui Nui.

Hawaii

Rugosity (Surface/Planar Area)

High : 2.5

Low : 1.0

Figure E.12 Bathymetric rugosity (surface area / planar area) of the sea floor around Hawaii.

114

Sandbottom (proportion)

High : 1.0

Low : 0.0

Oahu Kauai

Niihau

Figure E.13 Sand habitat (proportion) of the sea floor around Kauai and Niihau.

Molokai

Lanai Maui Sandbottom (proportion)

High : 1.0 Kahoolawe Low : 0.0

Figure E.14 Sand habitat (proportion) of the sea floor around Maui Nui.

115

Hawaii

Sandbottom (proportion)

High : 1.0

Low : 0.0

Figure E.15 Sand habitat (proportion) of the sea floor around Hawaii.

Max Significant Wave Height (m)

High : 8.3

Low : 0.0

Oahu Kauai

Niihau

Figure E.16 Maximum significant wave heights for the waters around Kauai and Niihau.

116

Molokai

Lanai Maui Max Significant Wave Height (m)

High : 8.3 Kahoolawe Low : 0.0

Figure E.17 Maximum significant wave heights for the waters around Maui Nui.

Hawaii

Max Significant Wave Height (m)

High : 8.3

Low : 0.0

Figure E.18 Maximum significant wave heights for the waters around Hawaii.

117

Mean Significant Wave Height (m)

High : 4.3

Low : 0.0

Oahu Kauai

Niihau

Figure E.19 Mean significant wave heights for the waters around Kauai and Niihau.

Molokai

Lanai Maui Mean Significant Wave Height (m)

High : 4.3 Kahoolawe Low : 0.0

Figure E.20 Mean significant wave heights for the waters around Maui Nui.

118

Hawaii

Mean Significant Wave Height (m)

High : 4.3

Low : 0.0

Figure E.21 Mean significant wave heights for the waters around Hawaii.

Downwelled Irradiance (proportional to surface)

High : 1.0

Low : 0.0

Oahu Kauai

Niihau

Figure E.22 Downwelled irradiance at the sea floor relative to that just below the sea surface around Kauai and Niihau.

119

Molokai

Lanai Maui Downwelled Irradiance (Ed(Z) / Ed(0-))

High : 1.0 Kahoolawe Low : 0.0

Figure E.23 Downwelled irradiance at the sea floor relative to that just below the sea surface around Maui Nui.

Hawaii Downwelled Irradiance (Ed(Z) / Ed(0-))

High : 1.0

Low : 0.0

Figure E.24 Downwelled irradiance at the sea floor relative to that just below the sea surface around Hawaii.

120

APPENDIX F

FIGURES OF GEOGRAPHIC MODEL PREDICTIONS OF BENTHJIC COVER FOR

SIX CORAL SPECIES IN THE MAIN HAWAIIAN ISLANDS, 2000-2009.

Montipora capitata cover (%)

High : 43

Low : 0 Oahu Kauai

Niihau

Figure F.1 Geographic distribution of model prediction for Montipora capitata cover (%) around Kauai and Niihau.

Montipora capitata cover (%)

High : 43

Low : 0

Figure F.2 Geographic distribution of model prediction for Montipora capitata cover (%) around Oahu.

121

Molokai

Lanai Maui Montipora capitata cover (%)

High : 43

Low : 0 Kahoolawe

Figure F.3 Geographic distribution of model prediction for Montipora capitata cover (%) around Maui Nui.

Hawaii

Montipora capitata cover (%)

High : 43

Low : 0

Figure F.4 Geographic distribution of model prediction for Montipora capitata cover (%) around Hawaii.

122

Montipora flabellata cover (%)

High : 20

Low : 0 Oahu Kauai

Niihau

Figure F.5 Geographic distribution of model prediction for Montipora flabellata cover (%) around Kauai and Niihau.

Montipora flabellata cover (%)

High : 20

Low : 0

Figure F.6 Geographic distribution of model prediction for Montipora flabellata cover (%) around Oahu.

123

Molokai

Lanai Montipora flabellata Maui cover (%)

High : 20

Low : 0 Kahoolawe

Figure F.7 Geographic distribution of model prediction for Montipora flabellata cover (%) around Maui Nui.

Hawaii

Montipora flabellata cover (%)

High : 20

Low : 0

Figure F.8 Geographic distribution of model prediction for Montipora flabellata cover (%) around Hawaii. 124

Montipora patula cover (%)

High: 48

Low: 0 Oahu Kauai

Niihau

Figure F.9 Geographic distribution of model prediction for Montipora patula cover (%) around Kauai and Niihau.

Montipora patula cover (%)

High: 48

Low: 0

Figure F.10 Geographic distribution of model prediction for Montipora patula cover (%) around Oahu.

125

Molokai

Lanai Maui Montipora patula cover (%)

High: 48

Low: 0 Kahoolawe

Figure F.11 Geographic distribution of model prediction for Montipora patula cover (%) around Maui Nui.

Hawaii

Montipora patula cover (%)

High: 48

Low: 0

Figure F.12 Geographic distribution of model prediction for Montipora patula cover (%) around Hawaii.

126

Pocillopora meandrina cover (%)

High : 23

Low : 0 Oahu Kauai

Niihau

Figure F.13 Geographic distribution of model prediction for Pocillopora meandrina cover (%) around Kauai and Niihau.

Pocillopora meandrina cover (%)

High : 23

Low : 0

Figure F.14 Geographic distribution of model prediction for Pocillopora meandrina cover (%) around Oahu.

127

Molokai

Lanai

Pocillopora meandrina Maui cover (%)

High : 23

Low : 0 Kahoolawe

Figure F.15 Geographic distribution of model prediction for Pocillopora meandrina cover (%) around Maui Nui.

Hawaii

Pocillopora meandrina cover (%)

High : 23

Low : 0

Figure F.16 Geographic distribution of model prediction for Pocillopora meandrina cover (%) around Hawaii.

128

Porites compressa cover (%)

High : 60

Low : 0 Oahu Kauai

Niihau

Figure F.17 Geographic distribution of model prediction for Porites compressa cover (%) around Kauai and Niihau.

Porites compressa cover (%)

High : 60

Low : 0

Figure F.18 Geographic distribution of model prediction for Porites compressa cover (%) around Oahu.

129

Molokai

Lanai Porites compressa cover (%) Maui Value High : 60

Low : 0 Kahoolawe

Figure F.19 Geographic distribution of model prediction for Porites compressa cover (%) around Maui Nui.

Hawaii

Porites compressa cover (%)

High : 60

Low : 0

Figure F.20 Geographic distribution of model prediction for Porites compressa cover (%) around Hawaii.

130

Porites lobata cover (%)

High : 51

Low : 0 Oahu Kauai

Niihau

Figure F.21 Geographic distribution of model prediction for Porites lobata cover (%) around Kauai and Niihau.

Porites lobata cover (%)

High : 51

Low : 0

Figure F.22 Geographic distribution of model prediction for Porites lobata cover (%) around Oahu.

131

Molokai

Lanai Porites lobata cover (%) Maui Value High : 51

Low : 0 Kahoolawe

Figure F.23 Geographic distribution of model prediction for Porites lobata cover (%) around Maui Nui.

Hawaii

Porites lobata cover (%)

High : 51

Low : 0

Figure F.24 Geographic distribution of model prediction for Porites lobata cover (%) around Hawaii.

132

APPENDIX G

TABLE OF MODEL RESULTS FOR SENSITIVITY ANALYSIS OF BOOSTED

REGRESSION TREES FOR BENTHIC COVER OF SIX CORAL SPECIES

Table G.1 Model results for sensitivity analysis of boosted regression trees for benthic cover of six coral species. Rows in bold represent the selected best model for each response variable. Model diagnostics and results are nt = number of trees, tc = tree complexity, lr = learning rate, bf = bag fraction, tot dev = mean total deviance, resid dev = mean residual deviance, train corr = training data correlation, cv corr = cross-validation correlation. Tot Resid Train Species nt tc lr bf dev dev Cv dev (se) corr Cv corr (se) PLOB 900 5 0.01 0.5 0.042 0.013 0.022 (0.002) 0.839 0.693 (0.022) 1250 4 0.01 0.5 0.042 0.013 0.022 (0.002) 0.842 0.694 (0.017) 1600 3 0.01 0.5 0.042 0.014 0.022 (0.001) 0.826 0.689 (0.021) 2400 2 0.01 0.5 0.042 0.015 0.023 (0.001) 0.803 0.676 (0.011) 3000 1 0.01 0.5 0.042 0.021 0.025 (0.001) 0.714 0.641 (0.015) 1350 5 0.01 0.75 0.042 0.01 0.021 (0.001) 0.878 0.702 (0.02) 1150 4 0.01 0.75 0.042 0.013 0.022 (0.001) 0.84 0.689 (0.027) 1150 3 0.01 0.75 0.042 0.015 0.023 (0.001) 0.807 0.678 (0.0174) 2350 2 0.01 0.75 0.042 0.015 0.023 (0.001) 0.805 0.675 (0.011) 3100 1 0.01 0.75 0.042 0.021 0.025 (0.002) 0.711 0.637 (0.018) 1900 5 0.005 0.5 0.042 0.013 0.022 (0.001) 0.845 0.686 (0.018) 2350 4 0.005 0.5 0.042 0.013 0.022 (0.001) 0.836 0.689 (0.017) 3350 3 0.005 0.5 0.042 0.014 0.022 (0.001) 0.83 0.694 (0.011) 3400 2 0.005 0.5 0.042 0.017 0.023 (0.001) 0.78 0.673 (0.022) 3750 1 0.005 0.5 0.042 0.022 0.025 (0.001) 0.694 0.634 (0.017) 1900 5 0.005 0.75 0.042 0.012 0.022 (0.001) 0.851 0.705 (0.018) 2400 4 0.005 0.75 0.042 0.013 0.022 (0.001) 0.844 0.694 (0.011) 2500 3 0.005 0.75 0.042 0.015 0.023 (0.001) 0.813 0.68 (0.016) 2550 2 0.005 0.75 0.042 0.018 0.024 (0.001) 0.763 0.664 (0.014) 5400 1 0.005 0.75 0.042 0.021 0.025 (0.001) 0.705 0.633 (0.014) 200 5 0.05 0.5 0.042 0.013 0.022 (0.001) 0.842 0.696 (0.01) 200 4 0.05 0.5 0.042 0.014 0.022 (0.001) 0.822 0.691 (0.018) 300 3 0.05 0.5 0.042 0.014 0.023 (0.001) 0.818 0.68 (0.019) 400 2 0.05 0.5 0.042 0.016 0.023 (0.001) 0.787 0.667 (0.02) 800 1 0.05 0.5 0.042 0.02 0.025 (0.002) 0.726 0.633 (0.02) 250 5 0.05 0.75 0.042 0.011 0.022 (0.002) 0.871 0.705 (0.019) NULL 4 0.05 0.75 0.042 0.015 0.024 (0.001) 0.815 0.67 (0.025) 300 3 0.05 0.75 0.042 0.014 0.023 (0.001) 0.826 0.677 (0.015) 400 2 0.05 0.75 0.042 0.016 0.024 (0.001) 0.792 0.667 (0.012) 650 1 0.05 0.75 0.042 0.021 0.026 (0.001) 0.713 0.623 (0.025) PCOM NULL 5 0.01 0.5 0.015 2850 4 0.01 0.5 0.015 0.006 0.01 (0) 0.783 0.581 (0.018) 4650 3 0.01 0.5 0.015 0.006 0.01 (0) 0.781 0.577 (0.017) 5850 2 0.01 0.5 0.015 0.007 0.01 (0.001) 0.727 0.552 (0.018) 5300 1 0.01 0.5 0.015 0.01 0.011 (0.001) 0.574 0.476 (0.032) 2550 5 0.01 0.75 0.015 0.005 0.01 (0.001) 0.814 0.588 (0.019) 2450 4 0.01 0.75 0.015 0.006 0.01 (0.001) 0.781 0.571 (0.016)

133

Tot Resid Train Species nt tc lr bf dev dev Cv dev (se) corr Cv corr (se) PCOM 3650 3 0.01 0.75 0.015 0.006 0.01 (0.001) 0.772 0.57 (0.022) 3900 2 0.01 0.75 0.015 0.008 0.01 (0.001) 0.707 0.541 (0.027) 5100 1 0.01 0.75 0.015 0.01 0.011 (0) 0.572 0.495 (0.019) 4100 5 0.005 0.5 0.015 0.006 0.01 (0.001) 0.783 0.574 (0.016) 5750 4 0.005 0.5 0.015 0.006 0.01 (0.001) 0.781 0.587 (0.023) 6300 3 0.005 0.5 0.015 0.007 0.01 (0.001) 0.748 0.567 (0.019) 7000 2 0.005 0.5 0.015 0.008 0.01 (0.001) 0.69 0.549 (0.014) 5500 1 0.005 0.5 0.015 0.011 0.011 (0.001) 0.545 0.482 (0.021) 4250 5 0.005 0.75 0.015 0.006 0.01 (0) 0.797 0.578 (0.02) 4150 4 0.005 0.75 0.015 0.006 0.01 (0.01) 0.765 0.578 (0.025) 5200 3 0.005 0.75 0.015 0.007 0.01 (0) 0.746 0.571 (0.015) 6600 2 0.005 0.75 0.015 0.008 0.01 (0.001) 0.694 0.559 (0.021) 6550 1 0.005 0.75 0.015 0.01 0.011 (0) 0.552 0.485 (0.017) 600 5 0.05 0.5 0.015 0.005 0.01 (0.001) 0.812 0.573 (0.017) 600 4 0.05 0.5 0.015 0.006 0.01 (0.001) 0.78 0.577 (0.017) 1100 3 0.05 0.5 0.015 0.006 0.01 (0) 0.792 0.562 (0.018) 1250 2 0.05 0.5 0.015 0.007 0.01 (0.001) 0.731 0.546 (0.02) 2850 1 0.05 0.5 0.015 0.009 0.011 (0.001) 0.631 0.511 (0.008) 350 5 0.05 0.75 0.015 0.006 0.01 (0) 0.778 0.58 (0.025) 450 4 0.05 0.75 0.015 0.006 0.01 (0.001) 0.771 0.576 (0.022) 950 3 0.05 0.75 0.015 0.006 0.01 (0) 0.793 0.569 (0.023) 1350 2 0.05 0.75 0.015 0.007 0.01 (0) 0.747 0.559 (0.018) 1700 1 0.05 0.75 0.015 0.01 0.011 (0.001) 0.598 0.473 (0.042) MCAP 1700 5 0.01 0.5 0.016 0.004 0.011 (0.001) 0.88 0.534 (0.036) 1350 4 0.01 0.5 0.016 0.006 0.012 (0.001) 0.829 0.521 (0.036) 2200 3 0.01 0.5 0.016 0.005 0.012 (0.001) 0.841 0.505 (0.043) 3500 2 0.01 0.5 0.016 0.006 0.012 (0.001) 0.817 0.506 (0.041) 3450 1 0.01 0.5 0.016 0.01 0.013 (0.001) 0.638 0.425 (0.041) 2550 5 0.01 0.75 0.016 0.003 0.011 (0.001) 0.926 0.575 (0.027) 2000 4 0.01 0.75 0.016 0.004 0.012 (0.001) 0.884 0.507 (0.048) 2500 3 0.01 0.75 0.016 0.005 0.012 (0.001) 0.865 0.516 (0.048) 2050 2 0.01 0.75 0.016 0.007 0.012 (0.001) 0.781 0.5 (0.032) 4700 1 0.01 0.75 0.016 0.01 0.013 (0.002) 0.659 0.431 (0.034) 2850 5 0.005 0.5 0.016 0.005 0.011 (0.001) 0.866 0.566 (0.025) 3400 4 0.005 0.5 0.016 0.005 0.012 (0.001) 0.854 0.522 (0.039) 3000 3 0.005 0.5 0.016 0.006 0.012 (0.001) 0.806 0.529 (0.033) 6300 2 0.005 0.5 0.016 0.006 0.012 (0.001) 0.808 0.513 (0.025) 5250 1 0.005 0.5 0.016 0.01 0.013 (0.002) 0.614 0.412 (0.047) 2450 5 0.005 0.75 0.016 0.005 0.011 (0.001) 0.868 0.536 (0.042) 3150 4 0.005 0.75 0.016 0.005 0.012 (0.002) 0.861 0.511 (0.039) 4150 3 0.005 0.75 0.016 0.005 0.012 (0.001) 0.848 0.511 (0.034) 4200 2 0.005 0.75 0.016 0.007 0.012 (0.001) 0.783 0.507 (0.017) 6050 1 0.005 0.75 0.016 0.01 0.013 (0.001) 0.619 0.439 (0.045) 200 5 0.05 0.5 0.016 0.006 0.012 (0.002) 0.828 0.5 (0.051) 550 4 0.05 0.5 0.016 0.004 0.012 (0.001) 0.89 0.537 (0.036) 400 3 0.05 0.5 0.016 0.006 0.012 (0.001) 0.827 0.511 (0.045) 750 2 0.05 0.5 0.016 0.006 0.012 (0.001) 0.813 0.516 (0.043) 1700 1 0.05 0.5 0.016 0.008 0.012 (0.001) 0.726 0.472 (0.039) 200 5 0.05 0.75 0.016 0.005 0.012 (0.001) 0.848 0.503 (0.038) 300 4 0.05 0.75 0.016 0.005 0.011 (0.001) 0.856 0.514 (0.045) 550 3 0.05 0.75 0.016 0.004 0.011 (0.001) 0.869 0.54 (0.026) 550 2 0.05 0.75 0.016 0.006 0.012 (0.001) 0.809 0.5 (0.045) 600 1 0.05 0.75 0.016 0.01 0.013 (0.001) 0.617 0.409 (0.051)

134

Tot Resid Train Species nt tc lr bf dev dev Cv dev (se) corr Cv corr (se) PMEA 1150 5 0.01 0.5 0.013 0.005 0.01 (0.001) 0.815 0.531 (0.029) 1000 4 0.01 0.5 0.013 0.006 0.01 (0.001) 0.768 0.517 (0.033) 2200 3 0.01 0.5 0.013 0.005 0.01 (0.001) 0.808 0.522 (0.038) 3100 2 0.01 0.5 0.013 0.006 0.01 (0) 0.77 0.516 (0.025) 2550 1 0.01 0.5 0.013 0.009 0.011 (0.001) 0.593 0.459 (0.043) 1050 5 0.01 0.75 0.013 0.005 0.009 (0.001) 0.815 0.561 (0.025) 1650 4 0.01 0.75 0.013 0.005 0.01 (0.001) 0.832 0.53 (0.032) 2450 3 0.01 0.75 0.013 0.005 0.01 (0.001) 0.829 0.529 (0.026) 2800 2 0.01 0.75 0.013 0.006 0.01 (0.001) 0.764 0.527 (0.019) 1 0.01 0.75 2250 5 0.005 0.5 0.013 0.005 0.009 (0.001) 0.816 0.542 (0.031) 3450 4 0.005 0.5 0.013 0.005 0.009 (0.001) 0.827 0.546 (0.028) 3700 3 0.005 0.5 0.013 0.005 0.01 (0.001) 0.791 0.533 (0.022) 4400 2 0.005 0.5 0.013 0.007 0.01 (0.001) 0.735 0.52 (0.036) 2950 1 0.005 0.5 0.013 0.009 0.011 (0.001) 0.561 0.45 (0.028) 2500 5 0.005 0.75 0.013 0.005 0.01 (0.001) 0.836 0.542 (0.026) 3400 4 0.005 0.75 0.013 0.005 0.009 (0.001) 0.836 0.551 (0.03) 3350 3 0.005 0.75 0.013 0.006 0.01 (0.001) 0.788 0.534 (0.032) 4900 2 0.005 0.75 0.013 0.006 0.01 (0.001) 0.749 0.523 (0.039) 2950 1 0.005 0.75 0.013 0.009 0.011 (0.001) 0.559 0.457 (0.034) 200 5 0.05 0.5 0.013 0.005 0.01 (0.001) 0.796 0.531 (0.022) 300 4 0.05 0.5 0.013 0.005 0.01 (0.001) 0.808 0.52 (0.039) 350 3 0.05 0.5 0.013 0.006 0.01 (0) 0.781 0.494 (0.039) 700 2 0.05 0.5 0.013 0.006 0.01 (0.001) 0.779 0.513 (0.043) 1050 1 0.05 0.5 0.013 0.008 0.011 (0) 0.641 0.469 (0.034) NULL 5 0.05 0.75 300 4 0.05 0.75 0.013 0.005 0.01 (0.001) 0.818 0.521 (0.029) 450 3 0.05 0.75 0.013 0.005 0.01 (0.001) 0.818 0.529 (0.039) 600 2 0.05 0.75 0.013 0.006 0.01 (0.001) 0.77 0.504 (0.03) 700 1 0.05 0.75 0.013 0.009 0.011 (0.001) 0.607 0.472 (0.029) MFLA 1500 5 0.01 0.5 0.003 0.001 0.002 (0) 0.902 0.468 (0.063) 1650 4 0.01 0.5 0.003 0.001 0.002 (0) 0.887 0.499 (0.066) 1500 3 0.01 0.5 0.003 0.001 0.003 (0.001) 0.839 0.457 (0.06) 3100 2 0.01 0.5 0.003 0.001 0.002 (0) 0.835 0.493 (0.066) 4700 1 0.01 0.5 0.003 0.002 0.003 (0) 0.689 0.399 (0.027) 1650 5 0.01 0.75 0.003 0.001 0.002 (0) 0.923 0.493 (0.06) 1450 4 0.01 0.75 0.003 0.001 0.002 (0) 0.89 0.532 (0.066) 2350 3 0.01 0.75 0.003 0.001 0.002 (0) 0.893 0.43 (0.077) 2800 2 0.01 0.75 0.003 0.001 0.002 (0) 0.835 0.505 (0.058) 9400 1 0.01 0.75 0.003 0.002 0.003 (0) 0.754 0.45 (0.057) 3900 5 0.005 0.5 0.003 0.001 0.002 (0) 0.922 0.499 (0.058) 4550 4 0.005 0.5 0.003 0.001 0.002 (0) 0.911 0.49 (0.067) 3850 3 0.005 0.5 0.003 0.001 0.002 (0) 0.86 0.497 (0.055) 3300 2 0.005 0.5 0.003 0.001 0.003 (0) 0.779 0.507 (0.044) 5050 1 0.005 0.5 0.003 0.002 0.003 (0.001) 0.628 0.4 (0.053) 2450 5 0.005 0.75 0.003 0.001 0.002 (0) 0.905 0.487 (0.048) 2650 4 0.005 0.75 0.003 0.001 0.002 (0) 0.886 0.492 (0.069) 5100 3 0.005 0.75 0.003 0.001 0.002 (0) 0.899 0.512 (0.07) 3800 2 0.005 0.75 0.003 0.001 0.002 (0) 0.803 0.471 (0.062) 4900 1 0.005 0.75 0.003 0.002 0.003 (0.001) 0.624 0.391 (0.05) 500 5 0.05 0.5 0.003 0 0.003 (0) 0.936 0.439 (0.072) 500 4 0.05 0.5 0.003 0.001 0.002 (0) 0.917 0.503 (0.064) 450 3 0.05 0.5 0.003 0.001 0.003 (0) 0.868 0.474 (0.044)

135

Tot Resid Train Species nt tc lr bf dev dev Cv dev (se) corr Cv corr (se) MFLA NULL 2 0.05 0.5 0.003 900 1 0.05 0.5 0.003 0.002 0.003 (0.001) 0.683 0.409 (0.046) 250 5 0.05 0.75 0.003 0.001 0.002 (0) 0.905 0.51 (0.077) 300 4 0.05 0.75 0.003 0.001 0.002 (0) 0.894 0.476 (0.066) 250 3 0.05 0.75 0.003 0.001 0.002 (0) 0.844 0.509 (0.059) 850 2 0.05 0.75 0.003 0.001 0.002 (0) 0.872 0.502 (0.05) 2800 1 0.05 0.75 0.003 0.001 0.003 (0) 0.79 0.438 (0.05) MPAT 2900 5 0.01 0.5 0.018 0.003 0.011 (0.001) 0.935 0.616 (0.026) 1850 4 0.01 0.5 0.018 0.005 0.012 (0.001) 0.878 0.595 (0.021) 3200 3 0.01 0.5 0.018 0.005 0.012 (0.001) 0.883 0.583 (0.032) 3250 2 0.01 0.5 0.018 0.007 0.013 (0.001) 0.808 0.525 (0.034) 2100 1 0.01 0.5 0.018 0.013 0.015 (0.001) 0.575 0.433 (0.026) 2300 5 0.01 0.75 0.018 0.003 0.012 (0.001) 0.931 0.609 (0.029) 3100 4 0.01 0.75 0.018 0.003 0.012 (0.001) 0.928 0.588 (0.036) 4600 3 0.01 0.75 0.018 0.003 0.012 (0.001) 0.921 0.584 (0.033) 3550 2 0.01 0.75 0.018 0.006 0.013 (0.001) 0.825 0.504 (0.037) 3000 1 0.01 0.75 0.018 0.012 0.015 (0.002) 0.599 0.439 (0.047) 4150 5 0.005 0.5 0.018 0.003 0.012 (0.001) 0.914 0.586 (0.049) 3200 4 0.005 0.5 0.018 0.005 0.013 (0.001) 0.866 0.563 (0.036) 4750 3 0.005 0.5 0.018 0.005 0.012 (0.001) 0.859 0.571 (0.036) 4950 2 0.005 0.5 0.018 0.008 0.014 (0.001) 0.783 0.509 (0.04) 5550 1 0.005 0.5 0.018 0.012 0.015 (0.002) 0.597 0.443 (0.027) 4050 5 0.005 0.75 0.018 0.003 0.011 (0.001) 0.922 0.625 (0.024) 5400 4 0.005 0.75 0.018 0.003 0.011 (0.001) 0.919 0.622 (0.029) 6600 3 0.005 0.75 0.018 0.004 0.012 (0.001) 0.896 0.577 (0.036) 4000 2 0.005 0.75 0.018 0.008 0.013 (0.001) 0.767 0.516 (0.039) 4800 1 0.005 0.75 0.018 0.013 0.015 (0.002) 0.583 0.436 (0.031) 350 5 0.05 0.5 0.018 0.004 0.012 (0.001) 0.898 0.594 (0.02) 750 4 0.05 0.5 0.018 0.003 0.012 (0.001) 0.929 0.593 (0.034) 1300 3 0.05 0.5 0.018 0.003 0.012 (0.002) 0.932 0.579 (0.052) 1150 2 0.05 0.5 0.018 0.005 0.013 (0.001) 0.859 0.544 (0.022) 600 1 0.05 0.5 0.018 0.012 0.015 (0.001) 0.598 0.421 (0.043) 300 5 0.05 0.75 0.018 0.004 0.012 (0.001) 0.901 0.572 (0.045) 450 4 0.05 0.75 0.018 0.004 0.012 (0.001) 0.902 0.598 (0.03) 600 3 0.05 0.75 0.018 0.004 0.013 (0.001) 0.886 0.562 (0.036) 250 2 0.05 0.75 0.018 0.009 0.013 (0.001) 0.727 0.538 (0.027) 450 1 0.05 0.75 0.018 0.013 0.015 (0.001) 0.577 0.432 (0.031) Note: Species codes are P. lobata (PLOB), P. compressa (PCOM), M. capitata (MCAP), P. meandrina (PMEA), M. patula (MPAT), and M. flabellata (MFLA)

136

APPENDIX H

FIGURES OF RESPONSE PLOTS OF ENVIRONMENTAL VARIABLES IN

BOOSTED REGRESSION TREE MODELS FOR BENTHIC COVER OF SIX CORAL

SPECIES

fitted function fitted function fitted function fitted -0.05 0.05 0.15 -0.05 0.05 0.15 -0.05 0.05 0.15 0.00.51.01.52.0 02468 0 50 150 250 350

mean_Hs (22.2%) max_Hs (16.3%) aspect (13.6%) fitted functionfitted functionfitted functionfitted -0.05 0.05 0.15 -0.05 0.05 0.15 -0.05 0.05 0.15 -30 -25 -20 -15 -10 -5 0 024681012 0.0 0.2 0.4 0.6 0.8 1.0

depth (12.6%) slope (9.6%) light (9.5%) fitted function fitted function fitted function fitted -0.05 0.05 0.15 -0.05 0.05 0.15 -0.05 0.05 0.15 1.00 1.04 1.08 1.12 0.0 0.2 0.4 0.6 0.8 1.0 1.0 1.5 2.0 2.5 3.0 3.5 4.0

rugosity (8.4%) sand (6.2%) island (1.6%)

137

fitted function fitted function fitted function fitted -0.02 0.04 -0.02 0.04 -0.02 0.04

0.00.51.01.52.0 02468 0 50 150 250 350

mean_Hs (26.7%) max_Hs (22.2%) aspect (17.1%) fitted functionfitted functionfitted functionfitted -0.02 0.04 -0.02 0.04 -0.02 0.04

1.00 1.04 1.08 1.12 -30 -25 -20 -15 -10 -5 0 0.0 0.2 0.4 0.6 0.8 1.0

rugosity (11.4%) depth (8.6%) light (4.8%) fitted function fitted function fitted function fitted -0.02 0.04 -0.02 0.04 -0.02 0.04

024681012 1.0 1.5 2.0 2.5 3.0 3.5 4.0 0.0 0.2 0.4 0.6 0.8 1.0 slope (4.3%) island (3.8%) sand (1.2%) Figure H.2 Partial dependence response plots for environmental variables in model for Montipora flabelata cover. Relative percent contribution of each variable, scaled to 100, is in parenthesis of x-axis label. Rug plots along inside top of plots show distribution of sites across that variable, in deciles.

fitted function fitted function fitted function fitted -0.06 0.02 -0.06 0.02 -0.06 0.02

0.0 0.5 1.0 1.5 2.0 2.5 3.0 02468 0 50 150 250 350 mean_Hs (18.7%) max_Hs (16.4%) aspect (14.4%) fitted functionfitted functionfitted functionfitted -0.06 0.02 -0.06 0.02 -0.06 0.02 1.01.52.02.53.03.54.0 -30 -25 -20 -15 -10 -5 0 0.0 0.2 0.4 0.6 0.8 1.0 island (11.3%) depth (10.3%) light (9.6%) fitted function fitted function fitted function fitted -0.06 0.02 -0.06 0.02 -0.06 0.02

024681012 1.00 1.02 1.04 1.06 1.08 1.10 0.00.20.40.60.81.0 slope (8.2%) rugosity (7.4%) sand (3.7%)

138

-0.05 0.05 -0.05 0.05 -0.05 0.05 fitted function fitted function fitted function fitted

0.0 0.5 1.0 1.5 2.0 2.5 3.0 02468 1.00 1.04 1.08 1.12 mean_Hs (26.6%) max_Hs (20.2%) rugosity (13.7%) -0.05 0.05 -0.05 0.05 -0.05 0.05 fitted functionfitted functionfitted functionfitted

024681012 0 50 150 250 350 -30 -25 -20 -15 -10 -5 0 slope (12.4%) aspect (8.9%) depth (8.5%) -0.05 0.05 -0.05 0.05 -0.05 0.05 fitted function fitted function fitted function fitted

0.00.20.40.60.81.0 1.0 1.5 2.0 2.5 3.0 3.5 4.0 0.00.20.40.60.81.0 light (5.6%) island (2.3%) sand (1.8%)

fitted function fitted function fitted function fitted -0.05 0.10 -0.05 0.10 -0.05 0.10 0.0 0.5 1.0 1.5 2.0 2.5 3.0 02468 -30 -25 -20 -15 -10 -5 0

mean_Hs (36%) max_Hs (18.8%) depth (9%) fitted functionfitted functionfitted functionfitted -0.05 0.10 -0.05 0.10 -0.05 0.10 1.00 1.10 1.20 1.30 1.0 1.5 2.0 2.5 3.0 3.5 4.0 0.0 0.2 0.4 0.6 0.8 1.0

rugosity (8.3%) island (7.6%) light (6.7%) fitted function fitted function fitted function fitted -0.05 0.10 -0.05 0.10 -0.05 0.10 0 50 150 250 350 024681012 0.0 0.2 0.4 0.6 0.8 1.0 aspect (5.1%) slope (4.4%) sand (4.1%)

139

fitted function fitted function fitted function fitted -0.05 0.10 -0.05 0.10 -0.05 0.10 1.01.52.02.53.03.54.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0 02468

island (32.9%) mean_Hs (16.1%) max_Hs (12.8%) fitted function fitted function fitted function fitted -0.05 0.10 -0.05 0.10 -0.05 0.10 0 50 150 250 350 -30 -25 -20 -15 -10 -5 0 024681012

aspect (9.7%) depth (9.3%) slope (8%) fitted function fitted function fitted function fitted -0.05 0.10 -0.05 0.10 -0.05 0.10 1.00 1.02 1.04 1.06 1.08 1.10 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 rugosity (5.1%) light (4.8%) sand (1.4%)

140

APPENDIX I

TABLE OF INTERACTIONS BETWEEN ENVIRONMENTAL VARIABLES AND

BENTHIC COVER OF SIX CORAL SPECIES

Table I.1 Interactions between environmental variables and coral species cover. Interaction Response Variable Predictor A Predictor B value Montipora capitata cover Mean Hs Aspect 0.29 Max Hs Mean Hs 0.27 Light Max_Hs 0.15 Max Hs Depth 0.13 Max Hs Aspect 0.12 Montipora flabellata cover Max Hs Mean Hs 0.30 Mean Hs Depth 0.17 Montipora patula cover Mean Hs Aspect 0.50 Mean Hs Island 0.43 Light Aspect 0.16 Pocillopora meandrina cover Max Hs Mean Hs 0.35 Porites compressa cover Max Hs Mean Hs 1.69 Mean Hs Depth 0.37 Max Hs Aspect 0.25 Max Hs Island 0.21 Sand Max Hs 0.17 Max Hs Depth 0.16 Rugosity Mean Hs 0.14 Light Mean Hs 0.11 Slope Light 0.11 Porites lobata cover Mean Hs Island 0.60 Max Hs Island 0.20 Depth Aspect 0.12 Max Hs Aspect 0.10

141