MACROECOLOGY OF EXPLOITED MARINE SYSTEMS: HUMAN IMPACTS AND THE EFFECTS OF SCALE
by
Derek Paul Tittensor
Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy
at
Dalhousie University Halifax, Nova Scotia November 2007
© Copyright by Derek Paul Tittensor, 2007 Library and Bibliotheque et 1*1 Archives Canada Archives Canada Published Heritage Direction du Branch Patrimoine de I'edition
395 Wellington Street 395, rue Wellington Ottawa ON K1A0N4 Ottawa ON K1A0N4 Canada Canada
Your file Votre reference ISBN: 978-0-494-39078-8 Our file Notre reference ISBN: 978-0-494-39078-8
NOTICE: AVIS: The author has granted a non L'auteur a accorde une licence non exclusive exclusive license allowing Library permettant a la Bibliotheque et Archives and Archives Canada to reproduce, Canada de reproduire, publier, archiver, publish, archive, preserve, conserve, sauvegarder, conserver, transmettre au public communicate to the public by par telecommunication ou par Plntemet, prefer, telecommunication or on the Internet, distribuer et vendre des theses partout dans loan, distribute and sell theses le monde, a des fins commerciales ou autres, worldwide, for commercial or non sur support microforme, papier, electronique commercial purposes, in microform, et/ou autres formats. paper, electronic and/or any other formats.
The author retains copyright L'auteur conserve la propriete du droit d'auteur ownership and moral rights in et des droits moraux qui protege cette these. this thesis. Neither the thesis Ni la these ni des extraits substantiels de nor substantial extracts from it celle-ci ne doivent etre imprimes ou autrement may be printed or otherwise reproduits sans son autorisation. reproduced without the author's permission.
In compliance with the Canadian Conformement a la loi canadienne Privacy Act some supporting sur la protection de la vie privee, forms may have been removed quelques formulaires secondaires from this thesis. ont ete enleves de cette these.
While these forms may be included Bien que ces formulaires in the document page count, aient inclus dans la pagination, their removal does not represent il n'y aura aucun contenu manquant. any loss of content from the thesis. Canada DALHOUSIE UNIVERSITY
To comply with the Canadian Privacy Act the National Library of Canada has requested that the following pages be removed from this copy of the thesis:
Preliminary Pages Examiners Signature Page (pii) Dalhousie Library Copyright Agreement (piii)
Appendices Copyright Releases (if applicable) This thesis is dedicated to the memory of Ransom A. Myers
IV Table of Contents
List of Tables x
List of Figures xi
Abstract xiii
List of Abbreviations and Symbols Used xiv
Acknowledgements xviii
Chapter 1. Introduction 1
1.1 General Introduction 1
1.2 Thesis Structure 2
1.3 Miscellanea 4
Chapter 2. Macroecological Changes in Exploited Marine Systems 6
2.1 Introduction 6
2.2 Changes in Life-History Parameters 10
2.3 Loss of Populations and Species 13
2.3.1 Directed Fisheries 14
2.3.2 Bycatch Effects 15
2.3.3 Range Contraction 16
v 2.3.4 The Allee Effect 17
2.3.5 Loss of Genetic Diversity 19
2.3.6 Historical and Prehistorical Exploration 21
2.4 Changes in Community and Ecosystem Structure, Biodiversity, and Habitat 22
2.4.1 Changes in Community Structure and Ecosystem Dynamics 22
2.4.2 Habitat Modification 33
2.4.3 Emerging Fisheries 34
2.4.4 Destabilisation of Ecosystems 35
2.5 Conclusions 35
2.6 Acknowledgements 38
Chapter 3. Human Impacts on the Species-Area Relationship in Reef Fish Assemblages 39
3.1 Abstract 39
3.2 Introduction 40
3.3 Theory 43
3.4 Materials And Methods 46
3.4.1 Fish Data Collection 46
3.4.2 Habitat Data Collection 49
3.4.3 Calculating Species-Area Relationships 49
3.4.4 Mixed Effects Models 51
3.5 Results 53
3.6 Discussion 62
3.7 Conclusion 67
vi 3.8 Acknowledgements 68
3.9 Appendix 1 - Supplementary Material 69
3.9.1 Testing Functional Forms for the Species-Area Model 69
3.9.2 Tests of Statistical Robustness 72
3.9.3 Mixed-Effects Model Equations 74
3.10 Appendix 2 - Theoretical Model of The Effects of Exploitation on the Power-Law SAR 81
Chapter 4. Predicting Global Habitat Suitability for Stony Corals on Seamounts 91
4.1 Abstract 91
4.1.1 Aim 91
4.1.2 Location 92
4.1.3 Methods 92
4.1.4 Results 92
4.1.5 Main Conclusions 93
4.2 Introduction 93
4.3 Data 97
4.3.1 Coral Data 97
4.3.2 SeamountData 99
4.3.3 Environmental Data 99
4.4. Models 101
4.4.1 Maximum Entropy Modelling 103
4.4.2 Environmental Niche Factor Analysis 105
vii 4.4.3 Model Evaluation 106
4.5 Results 108
4.5.1 Model Evaluation 108
4.5.2 Maxent Results 108
4.5.3 ENFA Results 112
4.5.4 Habitat Suitability for Seamount Summits 112
4.6 Discussion 116
4.7 Conclusions 122
4.8 Acknowledgements 123
Chapter 5. Impact of Anthropogenic Ocean Acidification on Global Cold-Water Stony Coral Habitat 124
5.1 Abstract 124
5.2 Introduction 125
5.3 Methods 127
5.3.1 Coral Database 127
5.3.2 Environmental Layers 129
5.3.3 Habitat Suitability Model 131
5.4 Results 134
5.5 Discussion 141
5.6 Conclusions 144
5.7 Acknowledgements 145
vm Chapter 6. Endemism at Low Sampling Effort: Real or Artefact? 146
6.1 Introduction 146
6.2 Models of Endemism and Sampling 147
6.3 Application of the Method To Data 155
6.3.1 Data 155
6.3.2 The Heuristic Approach 158
6.3.3 Generating Estimates of Confidence 158
6.4 Conclusions and Further Work 162
Chapter 7. Conclusions 164
7.1 Exploitation, Scale and Sampling: Thesis Results 164
7.2 The Macroecological Approach: Strengths and Weaknesses 167
7.3 Management Implications 168
7.4 Where Next? Considerations for Future Marine Macroecological Research 170
References 173
Appendix A: Species List for Species-Area Reef Study 193
Appendix B: Copyright Permissions 202
ix List of Tables
Table 3.1: Results for minimal adequate mixed-effects Model (Power Law) 55
Table 3.2 (Part 1): Indices of diversity for protected and exploited sites 59
Table 3.2 (Part 2): Indices of diversity for protected and exploited sites 60
Table 3.3: Mean adjusted r-squared values for the seven least-squares regression models fit to species-area data 71
Table 3.4: Results for minimal adeqaute mixed-effects models (exponential model) 73
Table 3.5: Spatial combinations of transects used to calculate SARs in the Pacific 78
Table 3.6: GPS coordinates for Pacific and Indian sites 79
Table 3.7: Family names for Figure 3.5 80
Table 4.1: Environmental parameters used to predict habitat suitability 100
Table 4.2: AUC Values for all model runs 109
Table 4.3: Correlation between environmental parameters Ill Table 4.4: Variance explained by the first eight ecological factors in the ENFA model 113
Table 5.1: Sources of environmental data used to predict habitat suitability 130
Table 5.2: AUC values for all cross-validation model runs 140
Table Al: Species list for all study locations in the species-area relationship reef study (Chapter 3) 193
x List of Figures
Figure 2.1: Trends in community biomass of large predatory fishes in open ocean (a-i) and shelf (j-m) ecosystems 8
Figure 2.2: Proportional changes under exploitation in mean age and length at Maturity for marine fishes from the north-temperate regions of the Atlantic and Pacific Oceans 11
Figure 2.3: Changes in the northern limits of fur seal breeding range in New Zealand 18
Figure 2.4: Temporal changes in New Zealand snapper (Pagrus auratus) populations for two locations 20
Figure 2.5: a) Fish biomass plotted against body mass (circles) for the fish community in the North Sea in 2001 and the fitted size spectrum (steep bold line) 24
Figure 2.6: Linear correlations between cod and shrimp biomass time series from nine North Atlantic locations are suggestive of top-down control 29
Figure 2.7: Changes in species abundance, mean body mass, and community composition in the tropical Pacific between the 1950s and the 1990s 31
Figure 3.1: Conceptual diagram depicting an example of changes in the SAR (a) with (b) a decrease in species richness and (c) an increase in mean species patch occupancy for a set of nested sampling units 45
Figure 3.2: Mean fitted power-law species-area relationships for each region, from minimal adequate mixed-effects models except Atlantic (includes depth) and Mediterranean (includes island) 54
Figure 3.3: The effects of fishing on the species-area slope of the power-law model 56
Figure 3.4: Patterns of rarity under fishing pressure 61
Figure 3.5: Normalised change in family abundance relative to protected areas 77
Figure 4.1: Potential large seamount (greater than 1000 m elevation) locations (-14,000) predicted from an analysis of global digital elevation data generated by Kitchingman & Lai (2004) 95
Figure 4.2: Locations of Scleractinia coral samples from seamounts 98
Figure 4.3: Predicted habitat suitability for seamount Scleractinia, using a maximum entropy model 110
Figure 4.4: ENFA predicted habitat suitability for seamount Scleractinia 114
XI Figure 4.5: Predicted habitat suitability for Scleractinia (on a scale from 0 To 100) on the summits of potential large seamounts 115
Figure 5.1: Locations of scleractinian coral records in our database 128
Figure 5.2: Predicted present-day habitat suitability for scleractinians 136
Figure 5.3: Predicted 2099 habitat suitability for Scleractinians given ocean chemistry changes modelled following the IS92a 'business-as-usual' scenario 137 Figure 5.4: Global and regional changes in cold-water stony coral habitat suitability 138
Figure 6.1: False endemics predicted from sampling two identical communities as described in the text 151
Figure 6.2: Percentage of false endemics for differing scenarios of evenness, species richness, and sampling intensity 153
Figure 6.3: Ranked species abundance pooled from abyssal sites (blue bars) and predicted by the TNBD (red dots) 157
Figure 6.4: Mean estimates of false endemics (top) and number of species sampled (bottom) from two identical simulated communities 159
Figure 6.5: Mean false endemics (black) and margin of error (red) for simulated communities derived from TNBD fitted to abyssal data 161
xn Abstract
Macroecology is the study of emergent statistical patterns in ecology. This field has produced numerous interesting results in the terrestrial realm, but has only recently begun to be applied in the marine environment. In this thesis, I use the macroecological approach to examine the effects of human impacts, and the scale of analysis, on the oceans. I begin with an overview of the macroecology of exploited marine systems. Next, I compare species-area relationship data gathered from four different reef systems, and find that there is a consistent impact of exploitation on the scaling of reef fish biodiversity with area. Following this, I use two presence-only habitat suitability models to predict global habitat for cold-water stony corals on seamounts. These corals are often associated with elevated levels of biodiversity. I find suitable habitat on seamounts is located in the North Atlantic down to -2,500 m in depth, and in a circum-global strip in the mid-latitude southern hemisphere down to -1,500 m depth. Patterns are qualitatively similar for both models. I then use the more discriminative model to fit an enlarged data set consisting of cold-water stony corals from all deep-sea environments, and down to 5,500 m in depth. The patterns of global habitat suitability are similar to those for corals on seamounts, suggesting similar niche requirements at this scale. I apply model results to a future climate scenario (IPCC IS92a 'business-as-usual') to examine the likely effects of anthropogenic-induced changes in ocean chemistry. I find that habitat suitability is negatively impacted at low to moderate levels worldwide by 2099, except for in the North Atlantic where the effects are more severe. The final part of my thesis considers whether estimates of endemism in under- sampled environments are real or an artefact of incomplete sampling. I create two identical simulated populations, and sample from them to examine the effects of species richness, evenness, and sampling intensity. Large numbers of false endemics can be predicted at low sampling effort. I then apply these results to data from the abyssal Pacific, to determine how much sampling would be required to have confidence in estimates of endemism.
Xlll List of Abbreviations and Symbols Used
a Significance level a (Chapter 3) Mixed-effects model parameter
P (Chapter 3) Mixed-effects model parameter
AC03 Aragonite saturation state
S (Chapter 3) Mixed-effects model parameter
8 Error term r Gamma function y Shape parameter for the truncat< distribution
C (Chapter 3) Mixed-effects model parameter
X (Chapter 3) Mixed-effects model parameter ij (Chapter 3) Mixed-effects model parameter a Variance
9 Scale parameter for the truncated negative binomial distribution
6 (Chapter 3) Mixed-effects model parameter co (Chapter 3) Mixed-effects model parameter
A Area
AIC Akaike Information Criterion
AUC Area under the curve
xiv Fraction of grid cells within a species' distribution czcs Coastal Zone Color Scanner A fitted constant in the power-law species area relationship c Exponential of the entropy of the raw distribution, Maxent c (Chapter 5) model
DTI United Kingdom Department of Trade and Industry d The intercept of the exponential model
ECHO Pacific sampling site for abyssal polychaetes
/ The slope of the exponential model in log-linear space
GLODAP Global Ocean Data Analysis Project
Gg Gigagram g An unknown probability distribution
8 Approximation to an unknown probability distribution
H Information-theoretic entropy h Location
IPCC Intergovernmental Panel on Climate Change
IS92a Reference ('business as usual') scenario of the IPCC i Protection (fishing) status
J' Pielou's evenness j (Chapter 6) A value of Pielou's evenness j (Chapter 3) Transect or observation circle number k Spatial level with a transect or observation circle
I Depth
Maxent Maximum entropy method
xv m (Chapter 3) Island m (Chapter 6) Number of false endemics
N (Chapter 6) Total number of individuals in a community n (Chapter 6) Number of individuals in a species
PRA Pacific sampling site for abyssal polychaetes p (Chapter 6) Proportion of species that are true endemics
P' Maximum likelihood estimator ofp, the proportion of species that are true endemics p(x) (Chapter 5) Raw probability from Maxent model for a grid cell
ROC Receiver operating characteristic
S Number of species
SAR Species-area relationship
SODA Simple Ocean Data Analysis s (Chapter 3) Number of species s (Chapter 6) Number of species in each simulated community sn probability function for number of species with n individuals
TNBD Truncated Negative Binomial Distribution
UNEP United Nations Environment Programme
VGPM Vertically Generalized Production Model
WOA World Ocean Atlas
WOCE World Ocean Circulation Experiment
WCMC World Conservation Monitoring Centre
X A finite set x (Chapter 5) A grid cell
xvi y (Chapter 6) - Number of species sample z (Chapter 3) - The slope of the power-law species area relationship in log log space z (Chapter 6) - A normal distribution with mean 0 and standard deviation 1
xvn Acknowledgements
As with any large project, a thesis is the product of far more than one person, but rather an accretion of help and inspiration from an enormous range of sources. I am grateful to each and every person who has helped me to get this far.
I am deeply indebted to Ransom Myers and Boris Worm, my PhD supervisors.
They have both had a big effect in shaping my intellectual evolution and growth. In particular, I want to thank RAM for agreeing to take a chance on me when I was somewhat bereft of direction, and for providing me with this opportunity. It is my deep regret that he is not here to witness the end product, but his mark is indelible. I want to thank Boris for his amazing support throughout this process, his unwavering commitment to his students, and his ability to brainstorm for fieldwork ideas with a beer at the Grad.
House. I could not have hoped for better supervisors.
My committee members, Andrew Edwards and Ed Susko, have been helpful, supportive, and engaging. They were always ready to give me guidance when I needed it, and for this I thank them. I also wish to thank those people with whom I have co-authored a paper; acknowledgements appear in the appropriate chapter. Craig Smith and Thomas
Schlacher were an integral part of the endemism project. Many, many other teachers and researchers, from my primary and secondary school days, undergraduate years, and in the
Department of Biology at Dalhousie and at conferences all over the world have inspired me.
I wish to thank John Lindley for facilitating all of the diving that I needed to do to collect field data. He often made impossible things happen at short notice.
xviii I want to thank my friends for being drunken fun-loving people who have been a big part of my life. They include Daniel Jubainville, Gervase Topp, and Andrew 'Warlord of the Northern Wastes' Trider. Someday I would like to get you all together and hoist a glass or two. I would furthermore like to thank Andrea Moe for all the right reasons.
I wish to acknowledge the three inescapable elements that have accompanied me during this long process:
- toothache
obnoxious music
civilization
I wish to thank the English national football team for always threatening great things and always failing to deliver. My astonishing optimism belies your amazingly consistent underachievement.
Finally I wish to thank my family. Ruth, Andy, Ralph and Rosemary - you are all very special people. I would like to spend more time with you.
xix Chapter 1. Introduction
1.1 General Introduction
Macroecology is a relatively new branch of ecology, at least in name, and of considerable and expanding interest (Nee 2002). The term 'macroecology' was originally coined by Brown & Maurer (1989), and in his book on the subject, Brown (1995) defined it as:
'.. .a way of studying relationships between organisms and their environment that involves characterizing and explaining statistical patterns of abundance, distribution, and diversity.'
Although the nomenclature is relatively recent, studies that may be considered macroecological in nature have been carried out for almost a century, including - among many others - the work of Arrhenius (1921) on species-area relationships, Hutchinson and
MacArthur (1959) on the body size distributions of animals, and MacArthur & Wilson
(1967) on the theory of island biogeography (Brown et al. 2003). Though macroecology - given the prefix - has typically concentrated on large-scale studies (Brown & Maurer
1989 ; Brown 1995), over time its definition has been broadened considerably. Witness:
'Macroecology aims to identify general principles or natural laws underlying the structure and function of ecological systems, which are apparent in the patterns of
1 distribution and abundance of entities composing these systems, no matter what the scale of the analysis.'
(Marquet 2002). Thus the discipline has become, or perhaps more accurately enveloped, the statistical analysis of emergent ecological patterns of diversity, distribution, abundance and size. Although originally macroecological analyses tended to be defined by their scale, they often now include a range of scales, thus subsuming that parameter into a mode that is ripe for analysis. It is within this broader framework that this thesis has been undertaken.
Most early studies classified as macroecological were terrestrial (Brown 1995), and indeed the majority of macroecological understanding still lies within this realm.
Interest in applying the tools of macroecology to marine systems, however, is growing
(Witman & Roy, in press). The timeliness of this line of research is clear, given growing anthropogenic impacts upon the oceans (Worm et al. 2006). In this thesis I examine macroecological patterns in a range of marine environments, at a range of scales, and consider the interactions between ecological processes, human exploitation, and the effects of scale and sampling.
1.2 Thesis Structure
Chapter 2 (Macroecology of exploited marine systems) provides an overview of the current state of research in applying macroecological tools to exploited marine environments. What emerges is a need to synthesise information at all scales, from the genetic to the global, in order to be able to better understand and integrate the many effects of exploitation. 2 Chapter 3 {Human impacts on the species-area relationship in reef fish assemblages) examines the effects of fishing on a well-known macroecological pattern, the species-area relationship. This fundamental scaling law of diversity with area, one of the oldest known general laws in ecology (Rosenzweig 1995), is affected in a consistent manner by exploitation, resulting in a reduction in the rate of increase of diversity with area. The effects of this on reef systems - and the opportunities it offers for monitoring reef diversity- are explored.
Chapter 4 {Predicting global habitat suitability for stony corals on seamounts) explores the use of habitat suitability models for predicting the distribution of stony corals on seamounts. Seamounts (undersea mountains), in comparison to the surrounding deep- sea, can be biologically diverse environments with a distinct fauna, of which cold-water stony corals are often a significant part (Clark et al. 2006). These corals provide complex habitat structure for other deep-sea organisms (Roberts et al. 2006), yet the inaccessibility of this environment for sampling has led to a paucity of data for this taxonomic group.
This presents a challenge to most habitat suitability modelling methods, which require presence/absence data; proving absence of corals from seamounts with any reasonable level of confidence is unfeasible at current levels of sampling. I use two recently developed techniques, Environmental Niche Factor Analysis (Hirzel et al. 2002), and
Maximum Entropy Modelling (Phillips et al. 2006), designed specifically to work with presence-only data, to predict the global distribution of these organisms on seamounts. I evaluate the comparative performance of these models, and make habitat suitability predictions for seamounts both known and unknown.
Chapter 5 {Impacts of anthropogenic ocean acidification on global cold-water stony coral habitat) uses a similar approach to Chapter 4 but for a broader range of deep- 3 sea environments. A database of deep-sea stony coral records is used to parameterise a maximum entropy model to predict habitat suitability for the global seafloor and seamounts. There is evidence that potential changes in the availability of calcium carbonate due to oceanic uptake of carbon dioxide may negatively impact calcifying organisms such as cold-water stony corals (Kleypas et al. 1999 ; Orr et al, 2005). I use the habitat suitability model to predict the impact of these anticipated changes in ocean chemistry on cold-water stony coral habitat.
Chapter 6 {Disentangling endemism from inadequate sampling) considers the effects of sampling effort on estimates of endemism. In many poorly sampled environments (such as on seamounts or the deep sea benthos) estimates of endemism can be high. I use simulations to explore whether true endemism is distinguishable from artefacts induced by incomplete sampling. I then apply these models to abyssal data to assess whether estimates of endemism can be properly verified for a given level of sampling effort.
Chapter 7 {Conclusions) considers the overall framework within which these studies are integrated, and further expansions both of this work and of the possibilities opening up within marine macroecology as a whole.
1.3 Miscellanea
Three of these chapters (2, 3 and 4) have been published or submitted. Details are provided on the first page of each of these chapters. Appropriate permissions to reproduce this work have been sought from publishers, and will be included in the appendices. The references from each individual chapter have been compiled in a complete reference list
4 immediately posterior to Chapter 7. Chapter 3 includes an appendix (Appendix 2) directly related to the study, but in preparation as another manuscript. Chapter 2. Macroecological Changes in Exploited Marine Systems
2.1 Introduction
The oceans have been used as an important source of food and materials for much of human history and prehistory, and until relatively recently were viewed as inexhaustible. In 1883 T.H. Huxley, addressing the International Fisheries Exhibition in
London, famously declared that 'Any tendency to over-fishing will meet with its natural check in the diminution of the supply... this check will always come into operation long before anything like permanent exhaustion has occurred.' Yet with the advent of industrialised fishing such predictions have been comprehensively invalidated, with fish stocks crashing and catches decreasing drastically in many parts of the world (e.g. Myers et al. 1997; Pauly et al. 2002), bycatch adversely affecting non-targeted species (e.g.
Lewison et al. 2004a) and marine resource management conflicts being placed squarely on the agenda of many governments. Indeed, exploitation may be the major driver of recent extinctions in the oceans, having an effect greater than that of habitat loss, climate change, pollution, disease, and species invasions (Dulvy et al. 2003).
As an example, in the Canadian waters around eastern Labrador and
Newfoundland, the cod (Gadus morhua) fishery underwent heavy exploitation, resulting in the species becoming commercially extinct and a fishery moratorium being announced in 1992 (Hutchings and Myers 1994) in what was once the largest cod fishery in the
In press as: Tittensor, D. P., Worm, B., and Myers, R. A. 2008. Macroecological changes in exploited marine systems. In: Marine Macroecology (eds. J. D. Wirman and K. Roy). University of Chicago Press.
6 world (McGrath 1911; Thompson 1943). Virtual population analysis suggests that spawner biomass underwent a 99% decline between its maximum and the year in which the moratorium was imposed. Similar patterns have been observed in many other heavily fished regions (Myers et al. 1997; Christensen et al. 2003; Myers and Worm 2005). The magnitude and rapidity of these declines in fish populations can be severe. Figure 2.1 shows community biomass in thirteen open ocean and shelf ecosystems under exploitation pressure (Myers and Worm 2003). Industrialised fisheries typically reduce community biomass by 80% within 15 years. Large predatory fish biomass is estimated to be around 10% of pre-industrial levels (Myers and Worm 2003).
Our perception of what is the 'natural' abundance of a species or community may have changed over the course of the past half century of heavy fishing - the 'shifting baseline' effect (Pauly 1995; Dayton et al. 1998; Baum and Myers 2004). In addition, there can be a time lag between the onset of fishing and the detection of its effects on populations and communities, such as the period between the last reported sighting of a species and its extinction (Dulvy et al. 2003). Analysis of historical data suggests that the lag between overfishing and subsequent ecosystem changes could range from decades to centuries (Jackson et al. 2001; Springer et al. 2003). Significant fishing effort and large impacts on species and communities may also have occurred far earlier than originally thought (e.g. Barrett et al. 2004).
Fishing has an effect at many levels in addition to that of the population and the species. Life-history changes (e.g. Hutchings and Baum 2005), loss of genetic diversity
(e.g. Hauser et al. 2002), habitat alteration (e.g. Cranfield et al. 1999) and changes in
7 Figure 2.1: Trends in community biomass of large predatory fishes in open ocean (a-i) and shelf (j-m) ecosystems. Points denote relative biomass estimates from the beginning of industrialized fishing. Solid curves are individual maximum-likelihood fits. Dashed curves are empirical Bayes predictions derived from fitting a mixed-model. Redrawn from Myers and Worm (2003).
c
5 - Temperate Atlantic
4 V* 3 -
2 -
1 • •*s**~t~. 0 -
1960 1970 1980 1990 2000 1960 1970 19B0 1990 2000 1970 1980 1990 2000 e
1980 2000 1960 1980 2000 9
8 \ Tropical Pacific
•\ 6 A*"" 4 X* 2 X^A,* 0 1
1980 2000 1960 1980 2000 J 100 .£. *-V,. Gulf of Thailand J 80 V
~ 60 i V 2 X* S 40 \** g & 20 \r •?**"-»-'-*-«-,-. 0
1965 1970 1975 1980 1970 1975 1980 1985 1990 1960 1970 1980 1990
1950 1960 1970 1980 1990 8 community structure (e.g. Witman and Sebens 1992; Tegner and Dayton 1999; Dulvy et al. 2004; Worm et al. 2005a) may also result, and these can all lead to detectable changes in the macroecology of the marine environment. In this chapter we take a broad view of macroecology as being the effect of local and small-scale processes upon large-scale patterns in the marine environment, and the analysis and utilisation of these large-scale statistical patterns to infer ecological change from local to global scales. The new tools and analytical processes that a macroecological viewpoint provides enable us to view the interactions and synergies between biological and ecological processes in multiple dimensions. While we consider the four major macroecological patterns, namely species richness, range size, abundance, and body size (Gaston and Blackburn 2000), industrial fishing affects many of the parameters and processes of piscatorial communities and habitats discussed in this chapter. We structure this chapter according to the scale of these various effects, moving from individuals to species to whole ecosystems.
Section 2.2 considers changes at the level of individual life-histories; Section 2.3 discusses the loss of stocks, populations and species. Changes in ecosystem structure, biodiversity, and habitat are covered in Section 2.4. In Section 2.5 we conclude by considering future research needs for further understanding and mitigation of the macroecological effects of fishing. This chapter is not intended to be an exhaustive survey of the effects of fishing on marine environments and communities, but rather an examination of some of the commonly encountered macroecological perturbations that are expressed under severe exploitation pressure. We also limit our discussion to the effects of exploitation through fishing, as opposed to other human impacts on the oceans.
9 2.2 Changes in Life-History Parameters
The life-history traits of many marine species, such as slow growth rates, and aggregating behaviour, leave them vulnerable to exploitation, and even highly fecund species - exactly those predicted by Lamarck during the first half of the 19th century to be safe from threat - are at risk (Sadovy and Cheung 2003). Exploited fish populations often manifest changes in life history traits such as age and size at maturity, and growth rate
(Hutchings and Baum 2005). These changes can have cascading effects on survival until maturity, reproductive lifespan, and subsequently lifetime fecundity (Ernande et al.
2004), and are the subject of this section.
Figure 2.2 shows changes in age and length at maturity of individuals for a number of different fish stocks and species under exploitation pressure; the general trend towards a reduction in both age and length, across a number of regions and species, is clear (Hutchings and Baum 2005). The decline in age at maturity is substantial, averaging
21%, with an average decline in length of 13% across all species (Hutchings and Baum
2005). The earlier maturation time and smaller size at maturity is echoed in a reduction of the mean age and weight of spawning individuals in most of these populations. Such changes in the life-history of individuals could have consequences for the vulnerability of the species to extinction (Dulvy et al. 2003). Smaller individuals may be at greater risk of predation (Munday and Jones 1998), and experimental work suggests that genetic changes in life-history may reduce the capacity for population recovery after overharvesting (Walsh et al. 2006).
Much of the debate on the life-history effects of fishing focuses on whether the observed changes are genetic or phenotypic (Hutchings 2004). The case for genetic
10 Figure 2.2: Proportional changes under exploitation in mean age and length at maturity for marine fishes from the north-temperate regions of the Atlantic and Pacific Oceans.
The period of time each point represents differs among populations. Open triangles represent pelagic species, closed triangles demersal species. Redrawn from Hutchings and
Baum (2005).
NEA Herring(NorSpring) NEA Hem'ng(Norway) • NWA Herring(4Rfall) NWA Herring(White/NDBays) NWA Herring(4Rspring) NWA Herring(Bvista/TrinBays) NEA Cod(Baltic) NWA SilverHate(4VWX) NEA Cod(NEArctic) NEACod(Baltic) NWA Cod(2J) NEA Cod (NEArctic) NWA Cod(2J) NWA Cod(3K) NWA Cod(3K) NWACod(3L) NWACod(3L) NWA Cod(3NO) NWA Cod(4Vs) NWA Cod(3Ps) NWA Cod(4W) NWA Cod(4Vs) NMA Cod(GeorgesBank) NWACod(4W) NMA Cod(GulfMaine) NMA Cod(GeorgesBank) NWA Haddock(4TVW) NMA Cod(GulfMaine) NEA Plaice(NSea) NWAHaddock(4TVW) NWA AmPlaice(2J3K) NEA Plaice(NSea) NWAAmPlaice(3LNO) NWA AmPlaice (2J3K) NWA AmPlaice(3Ps) NWA AmPlaice(3LNO) NWAAmPlaice(4VW) NWA AmPlaice(3Ps) NWAYellowtail(3LNO) NWA Yellowtail(3LNO) J NWAYellowtail(4VW) J I I i I -0.4 -0.3 -0.2 -0.1 0 -0.3 -0.2 -0.1 0 Change in Age at Maturity Change in Length at Maturity
11 change (e.g. Conover and Munch 2002; Grift et al. 2003; Olsen et al. 2004) depends upon the fact that individuals genetically determined for late maturation have a much higher chance of being harvested before reproduction. This leads to more favourable conditions for those individuals that mature early (and hence have a smaller body mass) and a subsequent rapid evolutionary effect. Rapid evolution in response to human-induced selection pressures has been observed in many natural systems (Thompson 1998; Palumbi
2001), and conservation efforts should take the implications of this 'contemporary evolution' into account (Stockwell et al. 2003). An analysis of maturation patterns for cod
(Gadus morhua) stocks from Georges Bank and the Gulf of Maine suggests that evolutionary forces are at least partially responsible for shifts in life-history parameters
(Barot 2004).
The converse of this argument is that phenotypic plasticity can result in a higher growth rate and earlier maturity in response to declines in population density caused by fishing (e.g. Roff 2002). Such changes seem to be largely responsible for life-history changes in Norwegian spring-spawning herring (Clupea harengus), with evolutionary responses playing at most a minor role (Engelhard and Heino 2004). Disentangling the relative roles of these processes is tricky, and some combination of both is likely in effect in many instances (Law 2000; Stokes and Law 2000).
Although there has been some exploration of the relationship between body-size
(another life-history trait) and other frequently examined macroecological parameters
(e.g. range size, abundance patterns, species richness) in marine fishes (e.g. Munday and
Jones 1998; Goodwin et al. 2005), in general these patterns are poorly studied
(Macpherson et al. in press). The observed relationships can be weak or inconclusive, such as the correlation between body size and abundance (Macpherson et al. in press), or 12 variable between taxonomic groups, such as the patterns of maximum body size and mean depth range in pelagic fishes (Smith and Brown 2002). It is likely that fisheries exploitation has a smearing effect on observed relationships, especially given that individuals are often targeted by body size (Bianchi et al. 2000); this deserves further study. To understand underlying macroecological processes, the effects of exploitation should be removed by examining pristine or near-pristine systems (Macpherson et al. in press), although this is rarely possible. Instead, macroecological methods can be used to reconstruct the historical state of a system (e.g. Jennings and Blanchard 2004).
Changes at the scale of the individual can influence the macroecological structure of entire populations, species and communities, the interplay of processes at multiple scales becoming progressively more transparent as greater numbers of systems are combined in meta-analyses. We consider examples from the population and species level in the next section.
2.3 Loss of Populations and Species
The rapid and dramatic decline of marine fish populations and species in areas of intensive fishing activity has been well documented in a number of recent studies (e.g.
Jackson et al. 2001; Baum et al. 2003; Myers and Worm 2003; Jennings and Blanchard
2004; Myers and Worm 2005; Worm et al. 2005b). Fisheries affect both species that are directly targeted, and those that are unintentionally caught as bycatch or affected through indirect means. We consider all of these various effects.
13 2.3.1 Directed Fisheries
It is the extent of the decline in economically important, actively-targeted food fish such as cod (Gadus morhua) that commands much of our attention and leads to some of the most intense debate on effective management strategies (e.g. Hutchings and Myers
1994). In many targeted species, less than 10% of most populations are still extant (e.g.
Myers et al. 1997). As is evident from Figure 2.1, such declines appear in populations throughout the world's oceans, and are often present irrespective of the methodology used for calculations (Myers and Worm 2005). The loss of populations can be so rapid that fisheries lose their viability shortly after opening. For instance, the marbled rockcod
(Notothenia rossii) off South Georgia Island in the South Atlantic was fished for three years with a 99% decline in catch, and then abandoned in the fourth year (Kock 1992).
Once species suffer such drastic declines, the timescale for recovery may be very long. A meta-analysis of many heavily fished species showed that for non-clupeid fishes, there is little evidence of recovery since overfishing occurred (Hutchings 2000).
Species need not disappear entirely ('global extinction') to be dramatically impacted by overexploitation; local extinctions can result in severe disruption of community and habitat structure (e.g. Estes et al. 1989; see also Section 2.4), and species can also be reduced to such low densities that they are functionally extinct in the marine environment ('ecological extinction'). As shown in the influential work of Ricker (1958), it is also possible for subpopulations to go extinct in an economically sustainable fishery, if one stock has a greater catchability than the other.
14 2.3.2 By catch Effects
Losses are not restricted to targeted fish; exploitation can also affect non-target species through direct effects such as gear-induced mortality, and indirectly such as through the alteration of food supplies. Bycatch (the capture of species other than those that are directly targeted) can affect such diverse groups as dolphins (Tudela et al. 2005), leatherback turtles (Lewison et al. 2004b; James et al. 2005), seabirds (Tasker et al.
2000), sharks (Baum et al. 2003), and whales (Springer et al. 2003; Roman and Palumbi
2003). Tudela et al. (2005) studied the impact of fishing upon non-target species by the driftnet fleet operating in the Alboran Sea at the western end of the Mediterranean, targeting swordfish (Xiphias gladius). Annual catch rates for the dolphin species
Delphinus delphis and Stenella coeruleoalba were estimated to be greater than 10% of their population sizes in the Alboran Sea. Shark species suffered the heaviest bycatch rates, with blue shark (Prionace glauca), shortfin mako (Isurus oxyrinchus) and thresher shark (Alopias vulpinus) combined reaching half the total numerical captures of the target species Xiphias gladius (Tudela et al. 2005).
The Pacific leatherback turtle is another species at risk as bycatch. During the
1990s, it was estimated that 1,500 female leatherback turtles were caught per year in the
Pacific Ocean by trawling, longlining, and gillnet fisheries directed for other species
(Spotila et al. 2000). This corresponds to an annual mortality rate of between 23 and 33%.
This is an astonishingly high figure, and clearly has critical consequences for the leatherback, now IUCN listed as critically endangered, and facing extinction in the
Pacific (Spotila et al. 2000; James et al. 2005). Such deleterious effects of bycatch pressure may be especially prominent in long-lived, low-fecundity species such as sharks and whales (Musick 1999). 15 2.3.3 Range Contraction
A positive abundance-range size (abundance-distribution) relationship is an almost universal macroecological pattern in animal assemblages (Gaston and Blackburn
2000). With reduction in population and species abundance comes concomitant range contraction for many exploited marine species. We consider single species here; interspecific changes in the abundance-distribution relationship are discussed in Section
2.4.1.
Even large changes in range size can go undetected without analysis; an example is the barndoor skate (Raja levis), which disappeared from most of its range - likely due to being caught as bycatch in other fisheries - without notice until nearly a half-century of trawl data was analysed (Casey and Myers 1998). Density-dependent differential changes in habitat use between marginal and prime habitats have also been predicted in North Sea cod (Myers and Stokes 1989); such changes are obviously of note given the pressures and abundance declines caused by exploitation.
Range contractions are not limited to fish populations; reef modification and disturbance from 130 years of fishing in the Foveaux Strait, New Zealand, has led to the reduction of oyster density to such low values in some areas that fishers have abandoned them (Cranfield et al. 1999). Meta-analysis of oyster fisheries on three continents shows that this pattern of fishery collapse is commonplace (Kirby 2004).
Such effects are not confined to the recent past. An example of the effects of pre- industrialised exploitation on species ranges concerns pinnipeds in New Zealand (Smith
2005). The New Zealand fur seal (Ar otocephalus forsteri), one of eight species of southern fur seals, was hunted as a means of subsistence during the prehistoric period by the Maori, who colonized the region at around 750 years B.P. It was then killed for its 16 skin during the historic period when Europeans colonized, beginning in 1769 A.D.
During these time periods, seal colonies underwent a substantial range reduction. Figure
2.3 shows the estimated northern limits for the fur seal breeding ranges, and their successive retreat with time, from the beginnings of Maori settlement up to early colonization by Europeans. Historically, exploitation is known to be the major contributing factor responsible for reductions in fur seal abundance and distribution, given the large number of individuals taken (Smith 2005). Prehistorically, however, it is more difficult to determine whether exploitation, or another factor such as habitat degradation, climatic change, or introduced predators, was the underlying cause.
However, analysis by Smith (2005) appears to rule out everything but exploitation as having a significant effect on range reduction. Thus human-induced impacts upon the range of this species have been occurring for centuries. The effects of exploitation changed according to the time period and the intensity of effort, but the combined long- term impact was that the fur seal's range was reduced to a fraction of its original span.
2.3.4 The Allee Effect
In the wake of population declines and range contractions, the Allee effect, also known as depensation, has been hypothesised to be a contributing factor to the slow rate of recovery of many overexploited marine species (Frank and Brickman 2000; Hutchings and Reynolds 2004), although evidence for this is not always readily apparent (Myers et al. 1995). When population sizes fall below a certain threshold level of abundance, the lower density of individuals may result in reduced population growth per capita. There are several mechanisms (for example, reduced probability of encountering potential spawning partners) that may play a role in this effect. Such complications need to be 17 Figure 2.3: Changes in the northern limits of fur seal breeding range in New Zealand.
Estimates are for the beginning (1250), middle (1500) and end (1790) of local prehistory, and after historic sealing (1850). Redrawn from Smith (2005).
18 considered when modelling observed macroecological patterns in exploited systems. If the Allee effect is not incorporated in population dynamic models, this is an oversight that ignores what may be an effective reduction of biomass (Liermann and Hilborn 1997).
2.3.5 Loss of Genetic Diversity
There are further losses under exploitation, namely those of genetic variability in sub-populations, which should be of great concern due to the vast timescale on which such variability has evolved, and the correspondingly large timescale necessary for it to be recovered. Though studies on the loss of genetic diversity in exploited marine species have only begun to be conducted recently and are few in number, in at least one case, namely the New Zealand snapper (Pagrus auratus), there appears to have been a reduction in genetic variability due to exploitation (Hauser et al. 2002); Figure 2.4 shows the change in genetic structure of these two snapper stocks. The Hauraki Gulf stock was first exploited in the 1800's and overfished by the mid-1980's, with a corresponding decline in standing stock biomass during this period. The Tasman Bay stock followed a different exploitation pattern, with the fishery only commencing in the middle of the 20th century. In Figure 2.4, therefore, the Tasman Bay time-series starts with the stocks essentially at a natural level, while the Hauraki Gulf time-series depicts a stock that has already been exploited. Two measures of genetic diversity (number of alleles per locus, and mean expected heterozygosity) show significant decreases in the Tasman Bay region during the 50-year period, but only random fluctuations in Hauraki Gulf. It is possible that much of the genetic diversity had already been lost from the Hauraki Gulf stock by
1950. Such losses can drastically reduce the effective population size relative to the censused population size, and should this be commonplace it may have a large-scale, long 19 Figure 2.4: Temporal changes in New Zealand snapper (Pagrus auratus) populations for two locations. From top: Annual catch (AC), spawning stock numbers (N) and biomass
(SSB: dashed line), genetic diversity (mean number of alleles per locus (Na), and mean expected heterozygosity (He); both means + 95% confidence limits of 30 individuals).
Redrawn from Hauser et al. (2002).
Tasman Bay Hauraki Gulf 15 o o 10 p o' 5 < 0
0.76" i i 0.76 - 0.68- -i 0.68" 1 T 1940 1960 1980 2000 1940 1960 1980 2000 Year 20 term effect on population persistence and adaptability (Hauser et al. 2002). Should loss of diversity commonly occur during the early phases of exploitation, then studies initiated after the commencement of fisheries will likely detect no significant changes in diversity. Such patterns, however, are not consistent. A study of North Sea cod {Gadus morhud) (Hutchinson et al. 2003) shows an apparent decrease then recovery of genetic diversity under heavy exploitation, while a study of Newfoundland cod (Gadus morhud) showed no loss in genetic diversity despite drastic changes in population abundances (Ruzzante et al. 2001). The diverse range of results within such a slim set of studies, coupled with the potential importance of loss of genetic diversity, marks this out as an important area for future research. 2.3.6 Historical and Prehistorical Exploitation As noted in Section 2.3.3, the impacts of exploitation upon populations and species are not limited to modern-day industrialised fishing. Sea otters {Enhydra lutris), once active across the Pacific Rim, were driven to numerous local extinctions by the fur trade prior to the 20 century (Estes et al. 1989). Even before this, the hunting of otters by aboriginal Aleuts caused substantial changes in otter densities. These prehistoric changes in otter populations have been implicated as the probable cause of a shift between two alternate nearshore community stable states in the Aleutian islands (Simenstad et al. 1978; also see Section 2.4.1). The impacts of single-species exploitation can thus propagate and become visible at macroecological scales involving multiple species. Transformations such as these, of whole communities and ecosystems, are the subject of the next section. 21 2.4 Changes in Community and Ecosystem Structure, Biodiversity, and Habitat The changes effected by fisheries at the level of the ecosystem are extensive and challenging to interpret. Multiple anthropogenic impacts upon the marine environment produce complex yet recognisable symptoms of degradation (Tegner and Dayton 1999; Jackson et al. 2001; Lotze and Milewski 2004), and in some cases appear to have been affecting entire ecosystems for centuries (e.g. Pandolfi et al. 2003). The macroecological makeup of large regions, visible in community structure, as well as diversity, range size, abundance, life-history, and evenness of constituent species, can be substantially altered by industrial fishing practices. Such effects are difficult to disentangle from those caused by a broad spectrum of other processes, e.g. climate change (Hughes et al. 2003; Worm 2005). Yet reports of habitat and ecosystem-scale shifts caused by overfishing are surfacing with increasing regularity for a variety of marine ecosystems. 2.4.1 Changes in Community Structure and Ecosystem Dynamics The worldwide phenomenon known as 'fishing down the food web' (Pauly et al. 1998; Pauly et al. 2001) involves targeting progressively lower trophic levels for fishing as catches begin to stagnate or decline. Initially, this results in an ephemeral increase in catch, but ultimately it leads to cascading changes in relative species abundance and community structure, and as a result a reduction in the mean trophic level of fisheries landings. A different, but related, shift was observed by Bianchi et al. (2000) in a meta analysis of fishing data from many regions. The size composition of exploited demersal communities appeared to change in high-latitude regions, with a relative decline of larger- sized fish. 22 It is Ysry difficult to a§§e§§ ehange where baseline details of the pre-impaoted system are not available. Recently, macroecological theory has been applied in order to estimate original, unexploited stock levels and to provide a theoretical comparison to current conditions (Jennings and Blanchard 2004). The biomass and abundance of fishes that would be expected in an unexploited North Sea community were calculated using a theoretical abundance-body mass relationship (size spectrum) based on measurements of primary production and predator-prey body mass ratios. These estimates were then compared to 2001 species-size-abundance data from trawl surveys. The biomass of the fish community as a whole in the 2001 surveys was less than half that expected for an unexploited ecosystem. Large fish were particularly severely affected, showing a reduction of approximately one hundred-fold in biomass; see Figure 2.5(a). This approach highlighted the differential vulnerability of large fish to exploitation. (Jennings and Blanchard 2004). The abundance-distribution relationship, a correlation between local abundance and geographical distribution (e.g. Brown 1984; also see Section 2.3.3), is another macroecological pattern for which changes have been observed within an exploited marine community. The abundance-distribution relationships for 24 common marine fishes on the Scotian Shelf and in the Bay of Fundy, Canada, were examined by Fisher and Frank (2004) over a period of 31 years. In addition to intraspecific temporal trends for some stocks (i.e., correlated changes in abundance and distribution within a species), a trend was observed in the slope of the interspecific abundance-distribution relationship (i.e. that of the multi-species system). The value of the slope almost doubled between 1970 and 2001 (Fisher and Frank 2004). The authors proposed that direct fishery effects, 23 Figure 2.5: a) Fish biomass plotted against body mass (circles) for the fish community in the North Sea in 2001 and the fitted size spectrum (steep bold line). Remaining lines indicate predicted slopes and locations of unexploited size spectra. Three size spectra are presented, corresponding to transfer efficiencies (TE) of 0.100, 0.125 and 0.150 (for all predictions primary production is 1956 g WW year"1; predator prey mass ratio 390:1). Redrawn from Jennings and Blanchard (2004). b) Schematic representation of 2 forces interacting to produce different interspecific abundance-distribution relationships based on same assemblage of species at different points in time. Redrawn from Fisher and Frank (2004). c) Fish biomass density and community structure in the northwestern Hawaiian islands (NWHI) and the main Hawaiian islands (MHI). White bars represent apex predators (APP), grey bars herbivores (HRB), and black bars lower level carnivores (LLC). Redrawn from Friedlander and DeMartini (2002). d) Fishing intensity and average density of predatory fishes (filled circles), and crown-of-thorns starfish (empty circles) for thirteen Fijian islands. Islands without starfish show an empty diamond for starfish density. Redrawn from Dulvy et al. (2004). 24 Trophic level 4.2 4.4 4.6 4.8 5 5.2 TE=0.150 "g Shift in fisheries' prey to TE=0.125 § include additional species with Size-selective TE=0.100 = restricted predatlon by fisheries distributions 3 targets large, abundant, widespread species - allows more small Individuals per unit area 2 3 4 Distribution Body mass (log10 g) • LLC 1 HRB I I APP 53.9% | 3.0% ^^ 55.2% 18.1% ^M^ H 41.8% NWHI MHI -1 Location Fishing pressure (people km reef) 25 or fishery-driven second-order trophic effects, probably drove this change. Among the factors suggested were increased abundance due to decreased body size (and therefore resource requirements) of targeted species, commercial fisheries switching exploited species, and density compensation through increases in prey and competitor species. A schematic of the interspecific abundance-distribution shift is presented in Figure 2.5(b). This is a remarkable example of an observed temporal change in a macroecological pattern; such manipulations cannot easily be intentionally (experimentally) implemented at the large scales at which these processes operate. Thus marine systems that have a long history of exploitation and observation provide us with a valuable opportunity to examine the causal factors that underlie macroecological processes. Another macroecological pattern that has been observed to change under exploitation in the marine environment is the species-area relationship (SAR). An analysis of the SAR for fish assemblages on reefs in the Atlantic, Indian, Pacific, and Mediterranean found that fishing pressure consistently resulted in a lower slope of the SAR, with the effect being proportional to the intensity of the fishing (Tittensor et al. 2007). This is an alteration of the fundamental rate of scaling of diversity with area, one of the oldest known general laws in ecology. It may prove possible to use changes in observed macroecological patterns as assessment tools and ecological indicators of the effects of exploitation. Field programs to assess species-area relationships could potentially be used to quantify the impact of fishing on reef fish diversity, for example (Tittensor et al. 2007; Chapter 3). Another means of assessing impacts is to compare similar geographic regions with differing exploitation histories. Using this approach, Friedlander and DeMartini (2002) found striking differences in community structure between lightly and heavily exploited 26 reef systems of the Hawaiian islands. The grand mean standing stock of the northwestern Hawaiian islands, a remote and lightly fished area, was 260% that of the main Hawaiian islands, an urbanized and heavily fished area. The community composition was also remarkably different, as depicted in Figure 2.5(c), with the biomass of the northwestern Hawaiian islands being dominated by apex predators (54%), whereas in the main Hawaiian islands herbivores accounted for 55%, small-bodied lower-level carnivores 42%, and apex predators merely 3%. Grand mean weight per individual was also higher in the northwestern Hawaiian islands (Friedlander and DeMartini 2002). Clearly, the Hawaiian example shows the dramatic effects that may result under heavy fishing. Yet even relatively small takes can result in cascading effects on community structure. In their 1996 study of the effects of fishing on the structure of Fijian reef fish communities, Jennings and Polunin found that a 5% annual removal offish biomass can cause drastic community changes (Jennings and Polunin 1996). An example of indirect effects induced by artisanal exploitation of large predatory fishes, once more from Fiji, is the study of Dulvy et al. (2004), shown in Figure 2.5(d), in which loss of predators coincided with an increase in crown-of-thorns starfish density. The increased starfish population resulted in higher predation upon reef-building corals and coralline algae, and their replacement by non-reef building taxa - a habitat shift caused, albeit indirectly, by exploitation. The scale and consequences of this effect, in a region of non- industrialised fishing, should perhaps give cause for concern. Given the potential consequences of ecosystem change, it is important to understand the forces that interact to effect these changes. A meta-analytic study of the coupled predator-prey pair Atlantic cod (Gadus morhua) and northern shrimp {Pandalus borealis) attempted to uncover some of these interactions (Worm and Myers 2003). 27 Biomass time-series trends were used to determine whether control mechanisms were "top-down" or "bottom-up". Eight of nine regions showed negative correlations between cod and shrimp, a pattern consistent with top-down control (see Figure 2.6). This finding implies that overexploitation of a predator can result in cascading effects at lower trophic levels, confirming an earlier study of the community-level effects of overfishing in the northwest Atlantic (Witman and Sebens 1992). The highly diverse communities of cold- water kelp ecosystems show a similar dynamic, in which overexploitation of sea urchin predators facilitates destructive grazing by this species. This leads to cascading effects for other species in the habitat (Tegner and Dayton 2000; Steneck et al. 2002). Frank et al. (2005) provide evidence for a trophic cascade in the formerly cod-dominated Scotian Shelf ecosystem off eastern Canada. Systems such as these, with top-down control, should be managed under a multi-species framework, with species interactions taken into consideration (Worm and Myers 2003). Large-scale changes in community composition caused by exploitation are also thought to have occurred in the Bering Sea and the Grand Banks food webs, driven by whaling and cod fishing respectively (Worm et al. 2005a). Similarly, shifts of this scale have been observed in the North Pacific Ocean, where there has been a sequential collapse in the populations of seals, sea lions, and sea otters (Springer et al. 2003). Both studies present evidence that top-down forcing mechanisms are responsible for these community phase shifts, with industrial whaling inducing a series of successive community changes over time. Attempts have also been made to document and understand the driving forces behind community change on a global scale. In an analysis of Japanese longlining data 28 Figure 2.6: Linear correlations between cod and shrimp biomass time series from nine North Atlantic locations are suggestive of top-down control. Gg = gigagram, equivalent to 1000 metric tons. Redrawn from Worm and Myers (2003). Labrador Northern Newfoundland Flemish Cap O 4.5- <*b o .^_ 4.0- °o o 3.5- o 1 2 3 4 5 6 7 2 3 4 5 6 2.5 3.0 3.5 4.0 4.5 Northern Gulf of St. Lawrence Eastern Scotian Shelf Gulf of Maine o£o o ° °6># # ^r V u 0 o°° o i • • 2.0 3.0 4.0 5.0 Skagerrak 5.4 5.8 6.2 6.6 7.0 7.4 7.8 3.4 3.6 3.8 4.0 4.2 Log cod biomass (Gg) 29 from 1952-2000, Worm et al. (2005b) uncovered a decline of tuna and billfish diversity of between 10% and 50% in all oceans studied, a pattern linked to both climate and fishing. This analysis suggests that the long-term, low-frequency variation in community composition has been driven by the effects of exploitation, whereas the higher-frequency year-to-year variation was climate induced, modifying decadal trends only when lasting regime shifts such as the Pacific Decadal Oscillation occurred (Worm et al. 2005b). New attempts are being made to document the impacts of fishing at many structural levels and hence reach a more complete understanding of the dynamics involved. In this way, simultaneous changes in community composition and individual- level traits have been observed under fishing pressure. Ward and Myers (2005) used fisheries observer and scientific survey data from the 1950s and 1990s to detail changes in size composition, species abundance, and biomass of communities in the tropical Pacific targeted by longline operations. Figure 2.7 shows simultaneous changes in structure across three biological dimensions that represent three hierarchical levels: body mass, species abundance, and community structure. The summed overall abundance of species captured per 1000 hooks was drastically reduced, with large predators showing the greatest declines in abundance, along with sizeable decreases in mean body mass. There are concomitant changes in the community structure and relative abundance of species. The available evidence implicates fishing as the most likely cause of these changes (Ward and Myers 2005). This example shows the linkages between macroecological processes at multiple biological scales in exploited marine systems, as described in the introduction to this chapter. While we have largely focussed thus far in this section on the impacts of modern fishing, community shifts are not an entirely recent phenomenon. Historical 30 Figure 2.7: Changes in species abundance, mean body mass, and community composition in the tropical Pacific between the 1950s and the 1990s. The four most abundant species in the 1950s are labelled. Redrawn from Ward and Myers (2005). 1950s (6232 kg) -120 Blue shark bund a > -100 ne e ( n - 80 Yellowfin tuna - 60 p 100 0 h o 1990s (828 kg) - 40 oks ) Silky shark - 20 Bigeye tuna J™J fa* - 0 I 1 1 1 1 i i i 100 50 0 50 100 50 0 50 Mean body mass (kg) 31 overexploitation of sea otters (Enhydra lutris) as a food source has been indicted as a probable cause for the shift between two alternate stable-state community structures in the Aleutian islands (Simenstad et al. 1978), one dominated by macroalgae (and sea otters) the other by epibenthic herbivores such as sea urchins. When a substantial sea otter population exists, it keeps the sea urchins and other herbivorous epibenthic macroinvertebrates in check through predation, allowing the macroalgae to persist. When the otter populations are at meagre levels, herbivorous invertebrates are present in much greater densities, and subsequently overgraze and virtually exclude macroalgae. Overexploitation of sea otters by aboriginal Aleuts may well have disrupted the system into this second state (Simenstad et al. 1978). Indeed, we are starting to discover that large-scale removals and losses occurred throughout the coastal oceans prior to the 20th century, resulting in vast anthropogenically induced changes (Jackson 2001). Analyses of historic and prehistoric patterns are important and useful tools, not only to piece together changes in community structure that occurred before the functional patterns visible in contemporary ecosystems, but also to predict the shifts that may occur with further exploitation. It is likely that further examples of trophic cascades caused by fisheries exploitation, both past and present, will be unearthed as increasing numbers of systems are targeted for study (Pinnegar et al. 2000). The importance of these structural shifts in communities is gaining international recognition, as evidenced by the fact that the trophic integrity of ecosystems is under consideration as one of a suite of global biodiversity indicators by the United Nations Environmental Programme (UNEP 2004). More information on these transitions is needed in order to gain a greater understanding of when and where they are likely to occur in the future. 32 2.4.2 Habitat Modification Habitat modification is another side effect of exploitation that can indirectly affect the macroecology of target and non-target species. Although habitat damage can occur through destructive fishing practices such as dynamiting and cyanide fishing, here we concentrate on the effects of trawling and dredging. For instance, in Foveaux Strait, New Zealand, estimates derived from von-Bertalanffy growth models suggest that a reduction in habitat complexity and an increase in disturbance caused by dredging from the oyster fishery have impeded the growth of juvenile blue cod {Parapercis colias) in the region (Carbines et al. 2004). A comprehensive survey of the multifarious effects of the many different types of fishing on the marine habitat is beyond the scope of this chapter, but as an example, bottom-trawling or dredging-induced habitat degradation has been observed in regions including Ireland & Norway (Hall-Spencer et al. 2002), Georges Bank (Auster 1996; Collie et al. 1997), the Gulf of Maine (Auster 1996), offshore California (Friedlander et al. 1999), and the North Sea (Jennings et al. 2001; Schratzberger and Jennings 2002; Schratzberger et al. 2002). Benthic trawling is well known to 'scour' the seafloor, weighty steel trawl doors gouging bottom habitat and leaving marks that are readily detectable with sidescan-sonar (Friedlander et al. 1999). Such damage, and its resultant effects on habitat complexity (Auster 1996) and disturbance (Schratzberger et al. 2002), often reduces species richness and biomass, and can change community structure and species composition (e.g. Collie et al. 1997). Long-term studies of bottom trawling in the North Sea (Jennings et al. 2001) indicated that infaunal and epifaunal biomass showed significant decreases from trawling disturbance, but there appeared to be a minimal effect on the trophic structure of the assemblage. 33 Experimental studies on the effects of trawling depict greatly differing estimates of mortality and destruction (Moran and Stephenson 2000), a disparity that may be attributable to differences in the nature of the gear used. Such results give hope that enforcement of fishing practices with reduced effects on habitat and non-target species can be used to mitigate the effects of exploitation. Monitoring and enforcement of fishing practices, however, remains problematic (Moran and Stephenson 2000; Tudela et al. 2005), particularly as the fishing practices that are most effective at maximising catch of target species are often the most destructive. 2.4.3 Emerging Fisheries Given the deleterious effects of exploitation on ecosystems and communities, there are urgent concerns about emerging deep-sea fisheries (Roberts 2002), especially upon seamounts (Clark et al. 2006), commonly areas of high biodiversity and endemism (Richer de Forges et al. 2000). Fisheries are often displaced onto these fragile habitats after shelf and slope fisheries have collapsed or had moratoriums imposed. Deepwater fish species are often long-lived, late maturing, slow growing, and have low fecundity, characteristics that render them extremely vulnerable to the effects of overexploitation, especially as many species aggregate on seamounts (Koslow et al. 2000). These life- history features also cause long-lasting changes in community structure after fishing, with a correspondingly slow recovery time. There is a common "boom-and-bust" cycle for seamount fisheries, in which a highly productive resource is exploited, yielding high catches, and then collapses rapidly, at which point another seamount is targeted with the same effect (Clark et al. 2006). Such concerns have caused several governments to prepare legislation and carry out scientific surveys of these features within their waters. 34 2.4.4 Destabilisation of Ecosystems Changes in the biodiversity of ecosystems through the loss of species may also have further unexpected and deleterious effects. Theoretical work has shown that complex food-webs may maintain stability through many weak trophic interactions (McCann et al. 1998), or the fluctuating selection of trophic links (Kondoh 2003). Worm et al. (2006), using meta-analyses across a broad range of scales, found that the loss of biodiversity in marine systems has a destabilising effect, resulting in increased resource collapse and decreased recovery potential, together with negative impacts on ocean ecosystem services. Taken together, the diversity of studies outlined in this section indicate that multiple concurrent shifts in many aspects of biological communities can occur under many levels of exploitation pressure in a large variety of marine environments. The evidence that is accumulating could be incorporated into future management strategies. More research and elucidation of these large-scale changes is also vital. These subjects are discussed in the final section. 2.5 Conclusions The new tools of macroecology allow us to synthesise data, sift for patterns, and hypothesise processes. Recent studies have provided evidence for changes of breathtaking extent under exploitation (Baum et al. 2003; Worm and Myers 2003; Jennings and Blanchard 2004; Ward and Myers 2005; Worm 2005b). The macroecological patterns of marine species, communities, and ecosystems are changing in fundamental ways. Although not all such changes may be immediately apparent, their long-term 35 consequences can be profound. Among the patterns discussed in this chapter, shifts in life-history, population ranges, species abundances and densities, and community composition have all been observed. The array of modifications and impacts that humans have caused in the oceans is impressively - or perhaps depressingly - large (Peterson and Estes 2001). Achieving fisheries that are sustainable in the long-term is an outcome that is typically sought after by all stakeholders - fishermen, their communities, scientists, fisheries managers, and politicians. Yet the work that remains in order to inform attempts to mitigate or reverse our impacts is vast. There is very little research addressing how exploitation changes the relationships between macroecological variables in marine systems. We believe that more studies are needed to further interpret the linkages between commonly examined macroecological parameters (such as species richness, species range, body size, and abundance), and the effect of exploitation on these couplings (Fisher and Frank 2004; Jennings and Blanchard 2004). Historical studies are necessary to quantify pre-exploitation abundances, and meta-analyses are needed now more than ever to synthesise the vast quantities of data at regional, basin, and global scales (Christensen et al. 2003). The vulnerability of marine species to extinction is an area of current research and concern (Dulvy et al. 2003; Hutchings and Reynolds 2004; Myers & Ottensmeyer, 2005). It is now recognised that species could be at greater risk of extinction than previously thought (e.g. Roberts and Hawkins 1999). Predicting the spectrum of future scenarios for marine species is a research area in which work is needed to guide and optimize management strategies (Myers and Worm 2005) for sustainability, as is the use of macroecological parameters as tools to assess the extent of changes 36 wrought upon marine ecosystems (Fisher and Frank, 2004; Jennings and Blanehard 2004; Tittensor et al. 2007). Effective management can mitigate impacts and lead to recovery, while lack of action can lead to full-scale population collapse and extinction (Myers and Worm 2005). As discussed within this chapter, future management strategies need to take into consideration the following: potential effects of both genotypic and phenotypic changes in the life-history of fishes (Conover and Munch 2002), multi-species interactions, community- and ecosystem-level dynamical shifts, the impacts of habitat alteration on the macroecology of marine assemblages, and the optimal utilization of tools such as marine reserves (Roberts et al. 2005). Superimposed upon this challenge is the difficulty that each new generation of fisheries scientists experiences an altered set of initial conditions in the systems they observe. This can result in the 'shifting baseline' effect, in which the new abundances that serve as the baseline for the system are radically different from those of the natural environment, having suffered from another generation's worth of anthropogenic disturbance and exploitation (Pauly 1995). Due to the enormous declines and shifts that have been shown to occur across a wide range of species, scales, and systems, fundamental changes in our approach to the management of fisheries (such as ecosystem based management, for example) are necessary if we are to implement sustainability. Fishing appears to be both the earliest and greatest form of anthropogenic disturbance in many marine ecosystems (Jackson et al. 2001; Dulvy et al. 2003). Macroecological studies have shown that these disturbances often result in simultaneous shifts in dynamics at multiple scales. Given the rapidity of species collapse and ecosystem disturbance commonly observed under exploitation, it is important we recognise the time 37 constraints under which we operate, in order to help guide us in our attempts to sustain fisheries and alleviate our long-term impacts upon the macroecology of these vast environments. 2.6 Acknowledgements We wish to thank J. K. Baum, C. A. Ottensmeyer, R. M. Tittensor, and J. Witman for helpful and constructive comments. We acknowledge funding from the Sloan Census of Marine Life, the Pew Charitable Trust, and the Natural Sciences and Engineering Research Council of Canada. 38 Chapter 3. Human Impacts on the Species-Area Relationship in Reef Fish Assemblages 3.1 Abstract The relationship between species richness and area is one of the oldest, most recognised patterns in ecology. Here we provide empirical evidence for strong impacts of fisheries exploitation on the slope of the species-area relationship. Using comparative field surveys of fish on protected and exploited reefs in three oceans and the Mediterranean Sea, we show that exploitation consistently depresses the slope of the species-area relationship for both power-law and exponential models. The magnitude of change appears to be proportional to fishing intensity. Results are independent of taxonomic resolution and robust across coral and rocky reefs, sampling protocols, and statistical methods. Changes in species richness, relative abundance, and patch occupancy all appear to contribute to this pattern. We conclude that exploitation pressure impacts the fundamental scaling of biodiversity as well as the species richness and spatial distribution patterns of reef fish. We propose that species-area curves can be sensitive indicators of community-level changes in biodiversity, and may be useful in quantifying the human imprint on reef biodiversity, and potentially elsewhere. Published as: Tittensor, D. P., Micheli, F., Nystrom, M., and Worm, B. 2007. Human impacts on the species-area relationship in reef fish assemblages. Ecol. Lett., 10: 760 - 772. 39 3.2 Introduction The relationship between number of species (or higher taxa) and area is one of the most well-known macroecological patterns, and has been recognised since the middle of the nineteenth century (Rosenzweig 1995). Recently it has become the focus of much attention (e.g. Brose et al. 2004; Drakare et al. 2006; Martin & Goldenfeld 2006) as ecologists continue to develop underlying theory and synthesise empirical knowledge. Species-area relationships (SARs) are commonly (Drakare et al. 2006) described by the power function: S=cAz (3.1) where S is the number of species, A the area, c a fitted constant (the number of species in an area of size A = 1), and z represents the slope in log-log space and hence the rate of accumulation of diversity with area. Although a number of other functional forms have been fitted to species-area data (Tjorve 2003), the power-law is the most frequently applied, and the slope parameter z has been applied in terrestrial conservation to estimate extinction rates due to habitat loss (Pimm & Raven 2000) or climate change (Thomas et al. 2004). Here we provide evidence that the SAR in reef fish assemblages is sensitive to fisheries exploitation, and that the slope parameter may capture and quantify complex changes in community structure. The effects of fishing on marine ecosystems worldwide have been well documented in terms of impacts upon populations and communities (e.g. Pauly et al. 2002; Myers & Worm 2003; Jennings & Blanchard 2004), yet general macroecological patterns in exploited marine ecosystems remain relatively unexplored (Fisher & Frank 2004). These 40 population and community level effects of exploitation may be captured in different ways by the numerous and various indices of biodiversity. Two of the most common measures of diversity are species richness and relative species abundance (Magurran 2004). Many other diversity indices represent some combination of these (e.g. Simpson's index, Shannon's index). However, such measures tend to be insensitive to changes in spatial heterogeneity and patchiness that may occur independently of changes in richness and abundance. The species-area relationship, however, is inherently affected by the spatial distribution of individuals and species (e.g. He & Legendre 2002). We examine the impacts of exploitation on SARs of reef fish assemblages. Potential human impacts on the parameters of the SAR have, to the best of our knowledge, been little explored in the scientific literature. A recent meta-analysis of SARs does not include any papers that specifically examine this topic (Drakare et al. 2006). McClanahan (1994), Chittibabu & Parthasarathy (2000) and Reddy & Parthasarathy (2003) have presented data on changes in species richness with sampling area for protected (disturbed) and unprotected (less disturbed) sites - Kenyan coral reefs (also see McClanahan et al. 1999), tropical evergreen forest trees, and tropical evergreen forest lianas, respectively. However, as no model was fitted in any of these instances it is difficult to infer the magnitude, consistency, and significance of any potential changes in the parameters of the SAR. Nonetheless, these studies raise important concerns about the potential for human impacts to affect the SAR. Flather (1996) fitted models to avian species-accumulation data and found that intensively-used landscapes accumulated species less rapidly than landscapes with a greater proportion of natural habitats. Although it is not clear how the sampling scheme (particularly the temporal component of sample accumulation) would be converted to a classical species-area curve (Scheiner 41 2004; Adler et al. 2005), the results are again suggestive. Death (2000) fitted species-area models to benthic invertebrate data collected from two stream sites, one of which was more affected by anthropogenic disturbance resulting from changes in land-use. There was a significant difference in the power-law intercept log(c) between the two sites, but no change in the slope. As far as we are aware, therefore, there have been no studies that demonstrate a direct human impact on the slope parameter of the SAR. Here we provide evidence for a consistent impact of exploitation on this scaling rate for multiple reef fish assemblages, and test for underlying changes in components of diversity that maybe responsible. We used a series of comparative field surveys to examine SARs for coral reef fishes at sites located in three oceans (Glover's Atoll, Atlantic; Zanzibar Island, Indian; Line Islands, Pacific Ocean) and for rocky reef fishes in the Tuscan Archipelago (Mediterranean Sea). Using standardized protocols we compared protected to unprotected areas in order to observe the impact of fisheries exploitation. 'Protected areas' in the context of this paper are well enforced no-take zones where fishing is excluded; we cannot at present examine the effects of other disturbances that may occur. Mixed-effects models were fitted to test the hypothesis that fishing, through the removal of particular species and changes in the abundance and spatial heterogeneity of those remaining, may affect the slope of the species-area relationship. The SAR appears to consistently capture changes in these factors, and in some instances appears more sensitive than either species richness or relative species abundance as a measure of the effects of exploitation. 42 3.3 Theory Several mechanisms have been put forth to explain the highly robust pattern of species-area relationships (SARs). Particularly prominent are the environmental heterogeneity hypothesis (i.e. larger areas contain more habitats) and the demographic process hypothesis (i.e. a dynamic balance of dispersal, colonization, speciation, and extinction leads to the SAR) (Drakare et al. 2006), though it has also been suggested that the SAR may be a sampling phenomenon (Connor & McCoy 1979). An experimental test of these three causes (Hoyle 2004) demonstrated that for a natural microecosystem the contribution of the metapopulation effect was roughly equivalent to the sum of the habitat-heterogeneity and sampling effects. A number of studies have elucidated the linkages between diversity, range size, aggregation, and SARs. For example, Leitner & Rosenzweig (1997) noted that, under an assumption that range size and abundance are positively correlated and a lognormal distribution of abundance, it was possible to derive a relation connecting point diversity, range size, and provincial species diversity. Ney-Nifle & Mangel (1999) assumed characteristic patterns of geographic range and occupancy and ascertained that the main features of the SAR depended on these characteristics (along with patch census method). He & Legendre (2002) examined how changes in the species abundance distribution and intraspecies spatial aggregation affected the number of species found within a censused region. They derived models combining evenness, aggregation, and area, and determined that an increase in evenness caused a correspondent increase in species richness within a sampling area, and that an increase in intraspecies aggregation caused a decrease in species richness. Changes resulting from one factor could be counterbalanced by changes in the other. Picard et al. (2004) derived a model addressing the effect of the spatial 43 distribution of species on the S AR, and found that it could be as important as the species- abundance distribution in modifying the SAR. Harte et al. (2005) generated a model based on the assumption of equal spatial allocation probabilities for individuals of a species at every scale under consideration, and from this were able to uniquely determine the shape of the SAR from the species abundance distribution. All of the above models demonstrate that changes in species richness, evenness, and spatial distribution can affect the SAR. He & Legendre (2002) also provided a useful two-level conceptual model in which the ultimate drivers of the relationship between species and area (environmental and biotic factors) generated patterns of species abundance and spatial distribution: the proximate, observable factors influencing the shape of SAR. We might expect that exploitation could be added to the list of ultimate drivers of the SAR, since it can have substantial effects on species diversity (e.g. Worm et al. 2005), ecosystem structure (e.g. Roberts 1995), and habitat (e.g. Coleman & Williams 2002). But how might such changes manifest in the parameters of a SAR? We can envision such effects both from the models above and from a simple graphical example to help conceptualise connections between species diversity and the power-law SAR. Figure 3.1(a) shows the power-law SAR that would arise from a particular arrangement of species in a set of nested sampling quadrats. Changes such as a decrease in species richness (Fig. 3.1b) or an increase in average patch occupancy per species (Fig. 3.1c) can affect both parameters of the power-law SAR. Thus we hypothesize that the complex interplay of changes in species richness, spatial distribution, and relative species abundance caused by exploitation may affect the SAR. We used replicated field surveys to test whether such effects are indeed visible, consistent, and whether they can be generalized across different reef habitats. 44 Figure 3.1: Conceptual diagram depicting an example of changes in the SAR (a) with (b) a decrease in species richness and (c) an increase in mean species patch occupancy for a set of nested sampling units. The top row represents the distribution of species within units, with colours representing different species. The bottom row represents the species- area relationship derived from such a distribution. A) B) C) •: • o o •o • • • • • • • ® V) <0 to CO (0 CO O o o 9 n- 9 Log10Area Log10Area Log10Area 45 3.4 Materials and Methods 3.4.1 Fish Data Collection To gather empirical evidence for testing the effects of exploitation on species-area relationships in the field, we conducted replicated underwater surveys on coral reef sites in the Atlantic, Indian and Pacific Oceans, and on rocky reef sites in the Mediterranean Sea. Data were collected using standardized scientific SCUBA surveys, at both protected and exploited sites, using either transect lines or point censuses. Atlantic Ocean: Data were gathered during April 2005 from Glover's Reef, a 260 km2 atoll (16° 42.5' to 16° 56' N, 87° 53.5' to 87° 40.5' W) located approximately 30 km offshore from Belize (McClanahan & Muthiga 1998). No longlining, netting or traps are permitted anywhere within the atoll; fisheries are mainly artisanal. A triangular section in the southern half of the atoll has been designated a fully protected reserve since 1993, with no extractive activities permitted. Eight 100 m transects were surveyed in the protected area and eight within the fished region of the atoll. All sites were located on the forereef slope and very similar in terms of habitat features and exposure. Four transects in each section were surveyed at 5 m depth and four at 10 m. All sites were selected at random within the stratified design. The transects were divided into twenty intervals of 5 m length, and presence/absence for all fish species recorded within 1 m on each side of the transect, for a total area of 200 m2 per transect. After laying the transect and giving fish time to settle, one observer slowly swam the transect and recorded free-swimming pelagic species from 0-15 m ahead. The second observer followed and recorded demersal and benthic species. This order of censusing aimed to reduce error from fish avoiding human presence. 46 Mediterranean Sea: Stationary point counts were conducted at the Mediterranean islands of Capraia and Giannutri, part of the Tuscan Archipelago off the northwestern coast of Italy. Capraia has a coastline of 27 km, Giannutri 11 km. Both islands have rocky reefs with protected and exploited regions. Point counts were conducted in circles of 5 m radius at 10 m depth. 12 sites (6 protected, 6 unprotected) were sampled at Giannutri and 9 (3 protected, 6 unprotected) at Capraia, with sites on both sides of each island. Eight replicate censuses were conducted at each site, and three complete sets of these surveys were conducted. Point counts were conducted by two divers for 5 minutes, followed by an additional minute of searching for benthic species within the 5 m radius circle. Methods are fully described in Micheli et al. (2005). Pacific Ocean: Christmas Island, Fanning Island, and Palmyra Island are three atolls in the central Pacific Line Islands chain. They experience varying degrees of fishing pressure but similar oceanic conditions. Christmas Island has been inhabited for approximately 2,000 years, and currently supports 7,000-8,000 inhabitants engaged in commercial, subsistence, and localized sport fisheries. Fanning Island's population recently increased to approximately 3,500; historically, however, the island sustained a much smaller population. At both of these islands local fishermen use gill nets and hook and line to catch reef sharks for Asian markets. Nets, lines, traps, and spears are used to catch reef fishes for local consumption and for the aquarium trade. Palmyra Atoll has been privately owned for 100 years and was purchased by The Nature Conservancy in 2000. It is now under the jurisdiction of the US Fish and Wildlife Service and protected as a National Wildlife Refuge. With its minimal historical and present population, Palmyra has not had an extensive fishery at any time. Visual fish surveys were conducted in May 2005. Three to seven randomly-placed 4 m x 50 m belt transects were surveyed 47 between 2-10 m depth at sites on outer-lagoon reefs located along the leeward side of each island. Three sites were surveyed at Christmas and Fanning Islands, and four at Palmyra (GPS coordinates in Table 3.6). In total, 15 transects were surveyed at Christmas Island, 14 at Fanning Island* and 20 at Palmyra Atoll. Two observers surveyed each transect: one recorded the abundance of apex predators -jacks (Carangidae), snappers (Lutjanidae), sharks (Carcharhinidae), and groupers (Serranidae) - along the transect, while the other observer estimated the abundance of every other demersal fish larger than 5 cm, by family. Methods are fully described in Stevenson et al. (2007). Indian Ocean: Five sites (Bawe, Changuu, Chumbe, Nyange, and Pange; GPS coordinates in Table 3.6) near the island of Zanzibar, Tanzania were surveyed using transect sampling. All reefs were similar in physical structure; fringing small islands or sandbanks stretching from the water surface down to the bottom at approximately 10 m depth. Four of the sites - Nyange, Changuu, Bawe, and Pange - are open to fisheries, whereas Chumbe Island Coral Park has been a well-protected no-take marine reserve since 1994. Nyange is located 8 km from the shore of Zanzibar and is accessible by larger fishing boats, whereas Changuu (5 km), Bawe (5 km), and Pange (3 km) are within day- range of smaller fishing vessels such as dugout canoes (mtumbwi) and hence experience visibly increased fishing effort. Chumbe is located 3 km from shore. Species diversity, abundance, and size (to the nearest 5 cm) offish were surveyed using 5 m x 50 m belt transects placed randomly between 3 m and 8 m depth parallel to the reef crest (n = 10 per site). Due to the difficulty of identifying very small fishes to species level, only individuals > 5 cm were included in the study. The transects were further subdivided into 1 m, 10 m, and 25 m nested sections from the beginning of each transect, and fish data recorded separately for each of these sections. 48 3.4.2 Habitat Data Collection Habitat differences between fished and unfished regions may have potentially confounding influences upon SARs. Furthermore, fishing may alter habitat as well as affecting fish communities directly. For this reason, we collected habitat data at three of four regions surveyed (Indian Ocean, Pacific Ocean, and Mediterranean Sea) to quantify whether there was a difference between protected and unprotected regions. Habitat data were not collected for the Atlantic Ocean; transects were situated in areas with qualitatively similar percentages of live coral cover at both depths in fished and protected areas. Indian Ocean: Data were collected on percentage live coral cover, algae, rubble and dead coral cover, and 'other' habitat types. The benthic community structure was surveyed using 50 m (n = 10 per reef) transects placed randomly at depths between 3 and 8 m and parallel to the reef crest, using the line-intercept transect method described by English et al. (1997). Pacific Ocean: Habitat data were collected using 33 replicate 0.25 m quadrats at each site. Percentage algal cover and live coral cover were averaged for each site on each island. Mediterranean Sea: Thirty 0.5 m2 quadrats were sampled at each site on each island. Percentage encrusting (coralline) algae, bare rock, and cover by other algae and invertebrates were measured. Replicate quadrats were averaged for each site. 3.4.3 Calculating Species-Area Relationships Though there has been some debate over the most appropriate model for the species-area relationship (Tjorve 2003), the power-law function typically provides the best fit at intermediate spatial scales such as those of this study (He & Legendre 1996). In a previous study on coral reef SARs (Chittaro 2002), the power-law provided the best fit 49 of the models tested in most cases. To ensure that it was indeed an appropriate model, we empirically tested seven functional forms for modelling the data using a least-squares framework: the power-law, exponential model, untransformed model, log(species) vs. area, breakpoint regression (both power-law and exponential) and the cumulative Weibull distribution, a non-linear sigmoidal model (all models described in Tjorve (2003); methods fully described in Appendix 1). Two of the linear models were consistently among the best-fitting models in all regions (the power-law and the exponential model; Table 3.3, Appendix 1), and so we used both functions to fit the data in our mixed-effects model analysis. The power-law function is described in equation (3.1). The exponential model is: S = d+Jlog(A) (3.2) where S is the number of species, A area, d the intercept, and/the slope in log-linear space. For all spatial levels we calculated log(species + 1) for the power-law model to prevent taking logs of zero. In the Pacific we calculated family-area relationships, according to the taxonomic resolution of the data. Species-area relationships were constructed for the different regions as follows. Atlantic Ocean: Data from each transect were accumulated in (spatially consecutive) 10 9 9 9 9 9 m , 20 m , 50 m , 100 m and 200 m sections; in all cases, we used all sections from each transect at every spatial level in the calculations. Mediterranean Sea: Stationary observation circles of 5 m radius were combined using every possible combination for each sampling repetition at each site (consisting of 8 circles), since there was no obvious 50 spatial ordering; thus, each species-area relationship consisted of eight 5 m radius circles, twenty-eight sets of two 5 m circles, seventy sets of four 5 m circles, and one set of eight 5 m circles. Data from each survey was used to construct separate SARs; thus we had three sets of SARs for each site. Pacific Ocean: Each site consisted of 3-7 transects with no obvious spatial ordering. The spatial combinations used to form each species-area relationship depended upon the number of transects in each site (details in Table 3.5). Indian Ocean: Nested belt transect subdivisions of sizes 5 m2, 50 m2, 125 m , and 250 m (one of each level on each transect) were used to construct species-area relationships. There were ten transects for each site. 3.4.4 Mixed Effects Models Primary statistical analyses of the SAR data for all regions were conducted using linear mixed-effects models (Pinheiro & Bates 2000). This approach allowed us to construct the species-area relationship and conduct statistical analyses within a single modelling framework. Moreover, mixed-effects models explicitly incorporate the effects of autocorrelation between spatial scales within each random effect unit (Pinheiro & Bates 2000), an important consideration with nested data. We allowed the intercept of each transect or observation circle to vary as a random effect. The effects of area and protection were included as fixed-effects for all regions; the interaction between these terms is the effect of protection on the slope of the SAR. Region-specific fixed-effects were depth (Atlantic), live coral cover, algal cover, rubble and dead coral, and other habitat (Indian), live coral cover and algal cover (Pacific), and island (Capraia or Giannutri), encrusting (corralline) algae, bare rock, and algal and invertebrate cover 51 (Mediterranean). Full models included all first-order interactions and all linear terms. Two models, one containing coral cover and one algal cover, were used to assess habitat effects in the Pacific as a single model would not converge due to limited degrees of freedom. Equations for the mixed-effects models are given in Appendix S3. Data were converted from abundance to presence/absence where necessary. Habitat percent cover data were arcsine square-root transformed before being incorporated in the mixed-effects models. Mixed-effects models were fitted by the method of restricted maximum likelihood in S-Plus 7 (Insightful Inc., Seattle, WA, USA). Use of likelihood ratio tests for assessing fixed-effects terms is not recommended for mixed-effects models (Pinheiro & Bates 2000); model selection and simplification were therefore carried out using backwards stepwise regression from the full model, removing the least significant fixed-effect term (assessed with marginal F-tests) at each step until only terms with a significance level of p < 0.05 remained (Pinheiro & Bates 2000). We used conditional t-tests to assess the marginal significance of fixed-effects coefficients in the final minimal adequate models (Pinheiro & Bates 2000). Insignificant linear terms were retained when they were involved in a significant interaction effect. Results for the power-law model are presented in the text; full exponential model results are available in the Supplementary Material (Table 3.4). Further tests of statistical robustness were also carried out using more traditional methods such as ANOVA and non-parametric tests (See section 3.9.2). 52 3.5 Results Figure 3.2 depicts the mean fitted power-law SARs for each region, and Figure 3.3 the mean slope parameters z with standard errors. Table 3.1 gives results for the minimal adequate models. In all regions, protected (unfished) locations showed a higher mean species-area slope than exploited (fished) regions. In the Atlantic this effect was significant (P <0.0001) for both depths (shallower and deeper forereef slope). In the Indian Ocean the mean slope decreased proportionally to the accessibility (and thus visible fishing pressure) of the reef (Fig. 3.3). The difference was significant between the protected area (Chumbe) and the most accessible exploited region (Pange, P < 0.0001), as well as one of the two next-most accessible regions (Bawe, P = 0.025). In the Pacific the effect of exploitation on the mean species-areas slope was significant between the protected area, Palmyra, and both unprotected areas, Christmas Island (P = 0.011) and Fanning Island (P = 0.0098). Both full models in the Pacific resulted in the same minimal adequate model. In the Mediterranean the difference in mean slope between fished and unfished sites was significant for both islands (P < 0.0001). We focus on changes in mean slope z as the value of the intercept log(c) is dependent upon the units of areal measurement, and changes in intercepts may only be assessed if the slopes are not significantly different. Habitat effects and their interaction terms were not significant for any of the three regions tested, and none of the minimal adequate models retained habitat as a factor. Thus we found no evidence of significant habitat differences between protected and unprotected areas in these regions, nor of habitat effects on the parameters of the power-law SAR. 53 Figure 3.2: Mean fitted power-law species-area relationships for each region, from minimal adequate mixed-effects models except Atlantic (includes depth) and Mediterranean (includes island). Red lines represent protected regions. Atlantic: solid lines are shallow transects, dashed lines deep. Indian: solid red line is Chumbe, solid black Pange, dash-dot Nyange, dashed Changuu, dotted Bawe. Pacific: solid red line is Palmyra, solid black Fanning, dotted line Christmas. Mediterranean: solid lines are Giannutri, dotted lines Capraia. Note the different scaling on each axis. Atlantic Indian 1.6- Xs § 1.2- 1.1- X 1 s 1 1.5 2 Pacific 1.25 § 1.2 | 1.15 s ^••'' f 1.1 s*^'' 1.05 ^y-'' 2.4 2.6 2.8 3^ 2.2 2.4 2.6 2.8 z Log10Area (m ) 54 Table 3.1: Results for minimal adequate mixed-effects models (power law). Atlantic: Intercept and Log(Area) are intercept and slope for the protected sites. Unprotected is the difference between protected and unprotected site intercepts. Log(Area) * Unprotected is the difference between protected and unprotected site slopes. Indian, Pacific, Mediterranean: results are presented in a similar manner; named sites are exploited. Non significant coefficients retained only when overall term is significant or involved in an interaction term. Parameter Estimate s.e. t P Atlantic Ocean Intercept 0.449 0.036 12.313 O.0001 Log(Area) 0.536 0.018 29.817 <0.0001 Unprotected 0.127 0.052 2.458 0.028 Log(Area) * Unprotected -0.092 0.025 -3.628 O.0001 Indian Ocean Intercept 0.350 0.047 7.492 O.0001 Log(Area) 0.715 0.033 21.615 O.0001 Bawe 0.062 0.066 0.935 0.36 Changuu -0.051 0.066 -0.766 0.45 Nyange 0.057 0.066 0.869 0.39 Pange 0.218 0.066 3.301 0.0020 Log(Area) * Bawe -0.106 0.047 -2.263 0.025 Log(Area) * Changuu -0.069 0.047 -1.470 0.14 Log(Area) * Nyange -0.041 0.047 -0.870 0.39 Log(Area) * Pange -0.227 0.047 -4.857 O.0001 Pacific Ocean Intercept 0.356 0.068 5.239 O.0001 Log(Area) 0.291 0.025 11.643 O.0001 Christmas Island 0.213 0.102 2.091 0.075 Fanning Island 0.243 0.106 2.283 0.056 Log(Area) * Christmas Island -0.095 0.037 -2.547 0.011 Log(Area) * Fanning Island -0.102 0.039 -2.602 0.0098 Mediterranean Sea Intercept 0.331 0.016 20.359 O.0001 Log(Area) 0.383 0.005 83.396 O.0001 Unprotected 0.135 0.022 6.263 O.0001 Log(Area) * Unprotected -0.044 0.006 -7.288 O.0001 55 Figure 3.3: The effects of fishing on the species-area slope of the power-law model. Mean fitted slopes with standard errors are shown as calculated from minimal mixed- effects models except Atlantic (includes depth) and Mediterranean (includes island). Red indicates a protected area, black exploited (with increasing fishing pressure from left to right in the Indian Ocean). Atlantic Indian 0.75- T 0.55- T T • J•_ T • 1 • T a 0.65- _L • T £0.50- 1 -L • CO T 0.55- 0.45- T • T • • 1 0.45- X Unfished Fishe1 d Unfished Fished Chumbe Nyange Changuu Bawe Pange Deep Shallow Pacific Mediterranean 0.30- T • 0.38 g.0.25- 1 2 0.36 CO 0.20 T T • 1 1 0.34 Palmyra Christmas Fanning Unfished Fished Unfished Fished Capraia Giannutri 56 Results for the exponential SAR model were very similar, with statistically significant (P < 0.05) decreases in slope with exploitation for all four regions (Table 3.4). In the Indian Ocean, the effect was significant at an additional site, Changuu (P < 0.0001), roughly the same distance from shore as Bawe. The most visible difference between the power-law and exponential models was that in the Mediterranean the minimal adequate exponential model retained interaction effects between area and encrusting algae, and area and algal/invertebrate cover, indicating that the relationship between species richness and area was sensitive to the amount of algal cover and encrusting algae. There was no evidence, however, for significant differences in these habitat variables between protected and unprotected sites, indicating that habitat differences were not responsible for observed changes. Power-law results were tested for robustness using analysis of variance and non- parametric tests (See section 3.9.2); for all tests in all regions the effects of exploitation were statistically significant at P < 0.05, though in some cases the level of significance changed depending on the model used. The factors responsible for the observed changes in the slope parameter z appeared to depend upon the region. Table 3.2 shows a breakdown of changes in species richness, relative abundance, and mean patch occupancy per species, all of which can contribute to changes in the SAR slope (see Theory section). Mean patch occupancy was lower in protected sites, while the other measures varied in both magnitude and direction between regions. In the Atlantic, the decrease in species-area slope appears to be a function of increased richness at larger sampled scales in protected areas, together with changes in relative species abundance (both significant at P < 0.01), although we cannot rule out habitat effects from our qualitative measures of habitat similarity. In the Indian Ocean, for 57 three of four fished sites there was a significant increase in species richness at larger scales, and significant differences in relative species abundance in comparison to the protected site. Two exploited sites also had significantly higher mean spatial patch occupancy per species (Table 3.2). In the Mediterranean, the only statistically significant difference was increased mean species richness at the smallest spatial scales in protected areas. In the Pacific, none of these indices were statistically significant individually, though the change in SAR slope was. This may indicate either that a combination of changes in species richness, evenness, and spatial patchiness were important, or that unmeasured parameters were responsible for the observed changes. Figure 3.4 shows relative species abundance (derived from presence/absence data in the Atlantic; family abundance for the Pacific) plots for all study locations on a log-log scale, with significance assessed in Table 3.2. In three regions (Atlantic, Indian, Pacific) the protected area distribution showed a longer tail of rare species. In the fourth region (Mediterranean) the species rank-abundance distributions for the protected areas had a tail of rare species that was either the same length as (Giannutri Island) or shorter than (Capraia Island) that for unprotected areas; thus, species richness was actually equal or lower in the protected areas. In exploited regions relative abundance was clearly more uneven at some sites in the Indian Ocean (Fig. 3.4), but the pattern was less clear in other regions (Fig. 3.4; Table 3.2). 58 Table 3.2: Part 1 Indices of diversity for protected and exploited sites. Significant differences from protected sites in bold. Pacific are family data. Ocean Site Mean species Significance Mean Significa Total richness vs. protected^ species nce vs. species smallest richness protected* richness sample unit largest sample unit Atlantic Unprotected 9.3 + 0.9 0.88 34.9 + 0.0017 67 3.1 Protected 9.2 + 2.7 - 41.4 + - 83 3.1 Indian Pange 3.8 +1.6 0.0068 32.9 + 0.0002 84 4.1 Bawe 2.3 + 0.8 0.68 38.7 + 0.0002 94 4.4 Changuu 1.6+1.0 0.22 33.3 + 0.0002 121 2.9 Nyange 2.3 + 1.0 0.66 49.2 + 0.058 133 5.7 Chumbe* 2.1+0.7 - 55.2 + - 154 6.6 Mediterran Capraia 12.0 + 2.8 < 0.0001 23.8 + 0.10 37.51 ean unprotected 4.0 Capraia 10.0 + 2.5 - 22.0 + - 34 protected 2.7 Giannutri 11.4 + 2.7 0.017 22.7 + 0.71 43 unprotected 3.1 Giannutri 10.6 + 2.5 - 23.3 + - 43 protected 3.3 Pacific Christmas 9.5 + 1.5 0.87 12.3 ± 0.17 14* 1.5 Fanning 9.5+1.4 0.83 13.7 + 0.63 16 2.1 Palmyra* 9.7 + 2.0 - 15.0 + - 18.3* 2.2 * Protected. f Significance assessed using Wilcoxon rank-sum test. Changes with significance level P < 0.05 are reported in bold. * Corrected for unequal sampling effort by resampling equal numbers of transects 10,000 times and taking the mean. 59 Table 3.2: Part 2 Indices of diversity for protected and exploited sites. Significant differences from protected sites in bold. Pacific are family data. Ocean Site Relative Patch occupancy Significance species smallest sampling vs. abundance, unit1 protected^ significance vs. protected5 Atlantic Unprotected 0.0001 0.18 + 0.21 0.19 Protected - 0.16 + 0.15 - Indian Pange 0.0001 0.15 + 0.13 0.080 Bawe 0.0001 0.16 + 0.12 0.050 Changuu 0.0001 0.16 + 0.07 0.040 Nyange 0.25 0.14 + 0.07 0.095 Chumbe* 0.11+0.02 - Mediterranean Capraia 0.95 0.40 + 0.25* 0.34 unprotected Capraia protected - 0.37 + 0.28 - Giannutri 0.83 0.38 + 0.24 0.34 unprotected Giannutri - 0.34+0.24 - protected Pacific Christmas 0.50 0.73+0.251 0.20 Fanning 0.97 0.67 + 0.27 0.57 Palmyra* - 0.60 + 0.30* - * Protected. f Significance assessed using Wilcoxon rank-sum test. Changes with significance level P < 0.05 are reported in bold. * Corrected for unequal sampling effort by resampling equal numbers of transects 10,000 times and taking the mean. § Mean relative species abundance per largest sampling unit. Atlantic data presence/absence only; Pacific is relative family abundance. Significance assessed using Kolmogorov-Smirnov test. 11 Within each large sampling unit in which a species had non-zero abundance (i.e. for which species was used to construct SAR). 60 Figure 3.4: Patterns of rarity under fishing pressure. Shown are rank-abundance curves for species (Atlantic, Indian, Mediterranean) and families (Pacific) on a log-log scale. Red lines represent protected areas, black fished areas. Pacific: Solid red line is Palmyra Atoll, solid black line Fanning Island, dotted black line Christmas Island. Indian: Solid red line is Chumbe, solid black line Pange, dotted black line Bawe, dashed line Changuu, dash-dot line Nyange. Mediterranean: dotted line is Capraia Island, solid line Giannutri Island. Data are corrected for sampling effort by subsampling 10,000 times and taking the mean. Atlantic curves based on presence/absence data. Atlantic Indian k>o10Sp6dos rank 61 3.6 Discussion Our results indicate that exploited reefs had systematically lower values of the species-area slope parameters z (power-law) and/(exponential function) in comparison with adjacent sites that experienced less or no fishing. This pattern was surprisingly consistent across all regions studied, irrespective of local species composition, habitat type, depth, or hydrographic factors. It is possible that human impacts other than fishing may affect the scaling of diversity with area; however, we could not test for such effects in this analysis. Changes in species richness, relative species abundance, and mean patch occupancy all appeared to contribute to the observed changes in SAR slope, conforming to theoretical work that has demonstrated the importance of these factors (e.g. He & Legendre 2002; Picard et al. 2004). However, none were statistically significant for all regions. This suggests that the SAR may be a sensitive measure of the effects of exploitation, since it appeared to detect changes when some of the other indices did not. Only patch occupancy showed a consistent direction of change across all regions. It is interesting to note, however, that all protected regions exhibited one or both of an increase in species richness at the largest scale, or a decline in species richness at the smallest scale. We might anticipate either of these to lead to an increase in slope. The effects of relative species abundance on species richness have been analysed by He & Legendre (2002), who found that an increase in species evenness should also correspondingly increase species richness in a sampling area. This relationship appears to hold for the Indian Ocean reefs, in that the three exploited sites with significantly lower evenness than the protected site also have significantly lower species richness at larger scales (Table 3.2). 62 Our data suggest that protected areas tend to have higher overall diversity, and increased spatial heterogeneity at smaller scales (Table 3.2), along with the observed higher values of the SAR slope parameters z and/ Species diversity can therefore be expected to be higher in protected areas at larger spatial scales. Why these changes are generally not reflected at smaller spatial scales is unknown. We speculate that higher diversity may intensify interspecies competition for both physical and niche space, resulting in lower occupancy per unit area for each individual species. Additionally, there are likely more predators in unfished regions (Roberts 1995; Micheli et al. 2004), which could affect the aggregation behaviour of smaller species. Although biodiversity loss can affect ocean ecosystem services (Worm et al 2006), it is unclear how changes in the scaling of reef biodiversity may be interpreted in terms of the 'health' of the ecosystem. Although habitat structure can exert a powerful effect on fish diversity, the inclusion of habitat data in mixed-effects models allowed for the partitioning of habitat and exploitation effects in our analysis. At the locations for which we had such data there was no evidence that systematic habitat changes between exploited and protected areas may have confounded the effects of fishing. The only habitat effects retained in any minimal adequate mixed model were those of algal/invertebrate and encrusting algal cover and their interaction with area in the exponential model for the Mediterranean. This model, however, did not include any terms that suggested a significant difference in habitat between protected and unprotected regions. Though we can exclude habitat effects at the scale of measurement, there are other potential caveats. In particular, we must consider the possibility that observed changes in diversity may be sampling artefacts. Firstly, the behaviour of fish, particularly large predatory fish that tend to be targeted by exploitation, may vary between protected and unprotected areas (Kulbicki 63 1998). This can lead to a change in detectability driven by attraction to or repulsion from divers, biasing estimates of abundance in comparison to unfished systems. We note, however, that of families present in two or more regions, only 15 of 30 (50 %) show a consistent pattern of decline or increase in average abundance under exploitation within each region, dropping to 4 of 30 (13 %) across all exploited sites in all regions (see Figure 3.5 & Table 3.7). Thus fully half of these families, under identical exploitation pressure, showed both declines and increases among exploited sites within at least one region. This suggests that behaviour is likely not of overriding importance, since we would anticipate similar responses within families should this be the case. Secondly, when species are less abundant (as is likely in fished regions) the likelihood of encounter is reduced, thus leading to another potential sampling artefact. However, in the Mediterranean, species richness was higher or equal for exploited areas on both islands, yet still the SAR slope declined. In the Pacific, family-level grouping should reduce the effect of such an artefact upon our results. Thus it seems unlikely that this factor is playing a strong role in observed differences. Thirdly, though typical for similar reef studies, the scale of the sampling units may be too small to capture the full impacts of fishing on the abundances of large, highly mobile species. Nevertheless, the scale and replication of our study was sufficient to detect significant differences in the SAR between exploited and unexploited sites in all regions. We do not suggest that this study is a full census of these reef regions, nor are we trying to extrapolate to larger scales; we detect differences in patterns of spatial distribution, richness, and abundance of species within a finite sampling area. Within each region our sampling and statistical methods were consistent between fished and unfished sites, so any bias in the methodology will affect both treatments equally, and be 64 unlikely to introduce systematic variation. Thus, although we cannot rule out the effects of sampling on our study, given the consistency of results across diverse regions, the most likely explanation remains that of impacts on diversity patterns caused by exploitation. Fourthly, although we included habitat as a variable in our mixed-effects analysis, differences between exploited and protected systems at less than the grain of the habitat measurements would not be captured by our surveys. Reefs can be highly complex habitats, and such fine-scale complexity could potentially have an effect on our results. Although the mixed-effects analysis suggests coarse habitat differences are not apparent between our protected and exploited sites, further investigation of fine-scale differences would be prudent before fully excluding it as a confounding factor. The generality of our results across different reef ecosystems, together with the fact that they capture relatively subtle gradients in fishing pressure, leads us to propose that SAR slopes may be useful as a complement to existing metrics of biodiversity for gauging the impacts of human exploitation upon reefs. There are a number of unresolved questions, however, that need to be addressed before the usefulness of such a method can be fully ascertained. For one, it is possible for changes in one aspect of diversity (e.g. spatial heterogeneity) to offset another (e.g. relative species abundances) in their effect on the parameters of the SAR (He & Legendre 2002). This would reduce the sensitivity of the SAR at detecting a combined set of changes. For this reason, the consistency of the direction of underlying trends in diversity on reefs needs to be confirmed before such a method can be applied. Moreover, the theoretical basis for SARs, while developing rapidly (e.g. He & Legendre 2002; Picard et al. 2004; Harte et al. 2005; Martin & Goldenfeld 2006) is complex, and a more complete model of observed changes under exploitation needs to be integrated within this framework. Limitations of the method include that it is not possible to disentangle the underlying effects responsible for changes in the SAR without the use of other indices (such as those in Table 3.2). Comparisons should only be conducted between geographically adjacent and environmentally comparable reefs, due to regional differences in SAR slopes (as seen in Fig. 3.3). The method may not be easily applied at larger spatial sampling scales, especially given that the slope of the SAR can vary with spatial scale (e.g. Crawley & Harral 2001; Fridley et al. 2005). We caution that the robustness and sensitivity of this pattern across different sampling scales needs to be quantified. We chose to examine a similar spatial extent for all study regions to avoid the confounding effects of scale among studies, and to operate on scales commonly used for the assessment of reef diversity. Finally, this study was carried out on nested sampling units; the results may be different when isolates are used to construct the SAR. These difficulties, however, are offset by a number of advantages. Unlike conventional indices, SARs appear to be sensitive to changes in spatial heterogeneity, relative species abundance, and species richness at multiple scales. As exploitation could potentially alter any or all of these facets of diversity, an index that is responsive to all of these could prove useful, and sensitive to combinations of changes in these parameters in a manner that other indices are not (as in our results). SARs appear to condense the numerous effects of exploitation pressure on fish assemblages into a concise, easily comparable value (the slope parameter; z for the power-law,/for the exponential model, though we caution that the c parameter also affects the slope of the power-law model in linear space). The effects of exploitation on the SAR slope appear robust to changes in survey methodology, statistical analysis, and taxonomic resolution. Only presence/absence data are required to construct SARs, and as they do not require 66 measurement of abundance or body size they are a relatively practical and inexpensive survey tool. The apparent sensitivity at higher taxonomic levels could also be useful when taxonomic expertise is limited (for instance, when identification to species level is not possible). Given these multiple practical considerations, and that SARs can be calculated from rapid and simple SCUBA surveys, we believe that the usefulness of SARs as a method for ascertaining human impacts on coral reefs deserves further investigation. We speculate that it might be possible to use temporal changes in the SAR to monitor recovery rates of newly protected areas and trends in diversity under exploitation over time. We further suggest that evidence of anthropogenic changes upon the SAR could be visible in other marine, freshwater, and perhaps terrestrial environments, though there are likely to be substantial differences between habitats (e.g. Death 2000). 3.7 Conclusion We have provided evidence for fisheries impacts on one of the oldest known general laws in ecology, the species-area relationship. The slope parameter of the SAR, and thus the scaling of diversity on reefs, is consistently altered by exploitation. Our results suggest that the slope parameter of the SAR can in some instances be more sensitive to the effects of exploitation on marine biodiversity than species richness, relative species abundance, or patch occupancy alone. The changes in scaling of diversity on shallow-water reefs we have described are robust irrespective of geographical location, depth, method of analysis, model function fitted, and even taxonomic resolution. Such effects may be driven by changes in species richness, relative species abundance, and average patch occupancy. Our results are generally consistent with theoretical work on 67 the effects of these changes on the slope of the SAR. We propose exploration of the potential use of SARs in addition to other indices for quantifying anthropogenic impacts on reefs. While human impacts on point measures of marine biodiversity such as species richness are well recognized (e.g. Worm et al. 2005), alteration of the underlying scaling laws of diversity hints at a deeper raft of ecosystem changes. Fish species diversity scales closely with the diversity of functional traits (Micheli & Halpern 2005); thus changes in spatial species richness may have functional ecosystem consequences (Nystrom & Folke 2001). That the effects of exploitation on the SAR slope are clearly visible even at the relatively low intensities of artisanal fishing as seen on Glovers Atoll and Zanzibar Island provides further evidence that our impacts are not limited to industrialized fisheries. 3.8 Acknowledgements We thank R. A. Myers for guidance and suggestions, W. Blanchard for statistical advice, and A. Edwards and C. Field for helpful comments. N. Gotelli and the anonymous referees made comments resulting in substantial improvements to the manuscript. Taxonomic and technical help were given by the Department of Fisheries (Belize), R. Carballo, N. Hawthorne, J. Lindley, and C. Wabnitz. We acknowledge funding from The Sloan Foundation (Census of Marine Life, FMAP program) and NSERC. Data from the Line Islands were collected with funding and logistical support from Stanford University and the Sea Education Association, MA, USA. Zanzibar data collected by J. Lokrantz, A. Norstrom, and M. N., with funding support from the Swedish International Development and Cooperation Agency (Sida). Data from Mediterranean rocky reefs collected with support from a National Geographic Society grant to F. M. 68 3.9 Appendix 1 - Supplementary Material 3.9.1 Testing Functional Forms for the Species-Area Model Although the power-law is the function most commonly fit to species-area data, a number of other models have been also used (Tjorve 2003). The next most frequent is the exponential model (Tjorve 2003). More recently, it has been suggested that a piecewise regression (Lomolino & Weiser 2001) or sigmoidal (Lomolino 2000) model may in some instances provide a better fit to such data. Lomolino (2000) contends that a sigmoidal species-area curve may be the result of an asymptoting of species richness at larger spatial scales - although this point is hotly debated (Williamson et al. 2001) - and a 'small-island effect' at smaller spatial scales (Lomolino & Weiser 2001), for which species richness is independent of island size below a certain threshold. We fit four linear models (power, exponential, untransformed, and log(species) versus area, all two parameters), two piecewise regression models (piecewise-power and piecewise-exponential, both three parameters), and one nonlinear model (cumulative Weibull distribution, three parameters) to the data; see Tjorve (2003) for details of all models. Piecewise regression models were fit using the iterative breakpoint search procedure of Lomolino & Weiser (2001) between the minimum and maximum areas for each region, with the breakpoint location considered to be a third parameter. Model diagnostics and statistical comparisons are at present poorly resolved in a mixed-effects framework (e.g. Pinheiro & Bates 2000). Therefore we compare candidate models in a least-squares framework by fitting each transect or site separately in each region and comparing mean adjusted r-squared values. The adjusted r-squared measure, however, suffers from the problem of potentially being affected by spatial dependence in the data (which is explicitly considered in the mixed-effects framework), and 69 transformations of the dependent variable. Nonetheless, with these caveats in mind, the adjusted r-squared values provide a reasonable guide to model fit. The results indicate that no one model provides the best fit at all locations (Table 3.3). The power-law model and exponential model provide consistently good and parsimonious results. The other two linear models give a far less satisfactory fit in all regions. Visual inspection of the data indicates that there is no evidence for a small-island effect in the data, and this is reflected in the fact that the adjusted r-squared values of the piecewise regression models are consistently lower than those for single regression lines of the same functional form. Similarly, there is little evidence for a sigmoidal shape to the data, and in three of four regions the sigmoidal model is outperformed by simpler linear models, despite an extra parameter that allows for far greater flexibility in terms of model fit. We also note that the formula for adjusted r-squared penalises extra parameters relatively lightly when there are a large number of data points, and this is of note when assessing the trade-off between parsimony and fit. Further evidence of lack of fit in the sigmoidal model is given by the fact that in three regions there were problems with model convergence that in some instances could not be resolved; testing the model in a full non linear mixed-effects framework resulted in even greater convergence difficulties. Given the results in Table 3.3 and the issues outlined above, we fit all species-area relationships in the mixed-effects model framework with both the power-law and exponential functions, the two most parsimonious models that perform consistently well across all regions. We focus on the power-law in the main paper is it is the model whose parameters are the most familiar and interpretable to most ecologists. Table 3.4 presents similar results for the minimal adequate exponential model. The effects of exploitation 70 Table 3.3: Mean adjusted r-squared values for the seven least-squares regression models fit to species-area data. Adjusted r-squared values are transformed using Fisher's z- transform prior to taking the mean (Sokal & Rohlf, 1995). The z-transformed average is then back-transformed. Ocean Power- Log Exponential Untransformed Piecewise Piecewise Weibull law species model power exponential Atlantic 0.867 0.350 0.871 0.780 0.863 0.878 0.882 Indian 0.914 0.483 0.930 0.906 0.859 0.910 0.901* Pacific 0.563 0.468 0.524 0.422 0.474 0.425 0.529* Mediterranean 0.698 0.579 0.705 0.658 0.694 0.703 0.704* * Convergence difficulties meant that adjusted r-squared values could not be calculated for all sites. Indian: 29 of 50 calculated. Pacific: 9 of 10 calculated. Mediterranean: 62 of 63 calculated. 71 are consistently significant regardless of whether the power-law or the exponential model is used. 3.9.2 Tests of Statistical Robustness In order to check the robustness of the results to the mixed-effects model framework, the data were analysed using a range of alternative statistical techniques. In the Pacific and Indian oceans, separate fished regions were combined into one set of 'fished' data for all tests. Firstly, we applied a permutation test (requiring no parametric assumptions) by randomly assigning 'exploited' and 'protected' tags to species-area relationships calculated from power-law least-squares regressions. Exploited and protected sites had significantly different slopes at the P < 0.05 level in all regions (i.e. differences between means were outside of the 97.5% quantile). Similar results held for a bootstrapping test (bias-corrected and accelerated method, 1000 repetitions sampling with replacement from power-law species-area slopes; 95% confidence limits for fished and exploited means did not overlap in all regions). Next we conducted a least-squares regression (using a power-law model to fit the species-area curve; habitat data not included in models) followed by an ANOVA, using AIC to determine the minimum adequate model. The effects of exploitation were significant for all regions (Atlantic: P = 0.0067, Indian: P < 0.0001, Pacific: P = 0.04, Mediterranean: P = 0.0013). It is suggested (Milliken & Johnson 1992) that if data do not satisfy the condition of homogeneity of variance at the 0.01 level then ANOVA should not be used; our data do satisfy this criterion for all locations (F test for variance equality; Atlantic: P = 0.013, Indian: P = 0.32, Pacific: P = 0.50, Mediterranean (Capraia): P = 0.92, (Giannutri): P = 0.41). 72 Table 3.4: Results for minimal adequate mixed-effects models (exponential model). Atlantic: Intercept and Log(Area) are intercept and slope for protected sites. Unprotected is the difference between protected and unprotected site intercepts. Log(Area) * Unprotected is the difference between protected and unprotected site slopes. Indian, Pacific, Mediteranean: results presented similarly. Named sites exploited. Non-significant coefficients retained only when overall term is significant or involved in an interaction. Parameter Estimate s.e. t P Atlantic Ocean Intercept -13.554 1.021 -13.281 <.0001 Log(Area) 22.342 0.497 44.925 <.0001 Unprotected 4.816 1.443 3.337 0.0049 Log(Area) * Unprotected -4.391 0.703 -6.243 <.0001 Indian Ocean Intercept -9.810 1.778 -5.517 <0.0001 Log(Area) 29.068 1.228 23.663 O.0001 Bawe 5.274 2.514 2.097 0.042 Changuu 5.639 2.514 2.243 0.030 Nyange 2.781 2.514 1.106 0.28 Pange 7.875 2.514 3.132 0.0031 Log(Area) * Bawe -9.239 1.737 -5.318 <0.0001 Log(Area) * Changuu -11.551 1.737 -6.649 O.0001 Log(Area) * Nyange -2.825 1.737 -1.626 0.11 Log(Area) * Pange -13.000 1.737 -7.483 <0.0001 Pacific Ocean Intercept -17.674 2.395 -7.379 O.0001 Log(Area) 11.190 0.889 12.590 O.0001 Christmas Island 14.627 3.583 4.082 0.0047 Fanning Island 16.048 3.749 4.281 0.0037 Log(Area) * Christmas Island -5.736 1.324 -4.334 O.0001 Log(Area) * Fanning Island -6.173 1.394 -4.429 O.0001 Mediterranean Sea Intercept 11.922 37.310 0.320 0.75 Log(Area) -18.075 10.050 -1.799 0.072 Unprotected 2.529 0.867 2.917 0.0050 Encrusting algae -14.776 18.088 -0.817 0.42 Algal and invertebrate cover -28.390 38.850 -0.731 0.47 Log(Area) * unprotected -0.505 0.234 -2.161 0.031 Log(Area) * Encrusting algae 16.975 4.872 3.484 0.0005 Log(Area) * algal and invertebrate 32.415 10.465 3.097 0.0020 cover 73 Nevertheless, we also analysed data using Welch's test (which does not assume equality of variance); P < 0.02 for all regions (Atlantic: P = 0.012; Indian: P = 0.0017; Pacific: P = 0.011; Mediterranean: P = 0.0009). We repeated the analysis on all data using Wilcoxon's rank-sum test (a non-parametric test that does not assume normality); effects of exploitation are significant at P < 0.02 for all regions (Atlantic: P = 0.007, Indian: P = 0.003, Pacific: P = 0.019, Mediterranean: P = 0.0017). ANOVA, unlike mixed-effects models, does not account for spatial dependence within nested species-area data. We tested the Atlantic data by partitioning it in such a way as to ensure independence at each spatial scale, then conducting a linear regression followed by an ANOVA. We used two different partitioning schemes for each transect (7; 1 x 5m length, 1 x 10m, 1 x 15m, 1 x 20,1 x 50m, 2; 3 x 5m, 2 x 10m, 1 x 15m, 1 x 20m, lx 40m). Results were significant at P = 0.027 (model 1) and P = 0.0094 (model 2). In both cases the minimum adequate model retained the effect of fishing intensity (location) only, not depth. Thus results were significant in all regions independent of whether a mixed-effects or more traditional statistical framework was used. 3.9.3 Mixed-Effects Model Equations Equations depict linear terms only; first-order interaction terms are included in the full models but not shown for clarity: 74 Atlantic log(SpecieSyW) = a + Transect,- + /?log(Area) + /Protection + JDepth + eyu (3.3) where Transect,- ~ N(0, a/ ), Syu ~ N(0, o*) Indian log(Specieshjk) = a + Transect, +/?log( Area) + /Location + <5(Live Coral Cover) + £"(Algal Cover) + //(Rubble and dead coral) + 6>(Other habitat) + shjk (3.4) where Transect,- ~ A^O, a/), £hjk ~ N(0, Pacific log(SpecieShjk) = a + Transect,- + /?log(Area) + /Location + <5(Live coral cover)+ zhjk (3.5) log(Specieshjk) = a + Transect,- + /flog(Area) + /Location + S(Algal cover)+ shjk (3.6) 2 2 where Transect,- ~ N(0, aT ), shjk ~ N{Q, a ) Mediterranean log(Species/,ytm) = a + ObsCirc, + /?log(Area) + /Protection + ^Island + ^(Encrusting (coralline) algae) + //(Bare rock) + co(Algal and invertebrate cover) + eijkm (3.7) 75 where ObsCirc,~iV(0, a0c),sijkm~N(0, (f) and h represents location, i the protection (fishing) status, j the transect or observation circle number, k the spatial level within the transect or observation circle, / the depth, and m the island. The coefficients a, ft, y, d, C tj, d, co, and X represent parameters to be estimated from the data, £ the error terms, and GT and ooc the variance of the random effects. The exponential models followed the same form but with the dependent variable being species, not log(species). Two models were used to assess habitat effects in the Pacific as a single model would not converge due to limited degrees of freedom. 76 Figure 3.5: Normalised change in family abundance relative to protected areas. Abundances are plotted as (F - U) / (F + U), where F is the mean abundance on fished transects and U the mean abundance of non-fished transects. Colours represent functional groups: red are piscivores, blue herbivores or omnivores. Hollow points represent poorly sampled families for which <10 individuals were observed in all transects or point counts. Data are ordered in terms of mean normalised change within a region. Family names are given in Table 3.7. Atlantic Ocean Indian Ocean *~ • e • Pange * Bave ? a- • Changuu (B T Nyange •5 v * T T #* ° T A » O • 1 -0 5 ( 10 15 20 25 5 10 15 20 25 30 35 40 Family number Family number Pacific Ocean Mediterranean Sea • *- 11 * • O 0 • • a • 77 Table 3.5: Spatial combinations of transects used to calculate SARs in the Pacific. Combinations of transects depended upon the number of transects at each site. Every possible combination of appropriate transects was used. No. of Small Medium-small Medium Large transects at site 3 1 transect - 2 transects 3 transects 4 1 transect - 2 transects 4 transects 5 1 transect - 2 transects 5 transects 6 1 transect 2 transects 3 transects 6 transects 7 1 transect 2 transects 3 transects 7 transects 78 Table 3.6: GPS coordinates for Pacific and Indian sites. Island Site Latitude Longitude Christmas 1 1° 56.417'N 157° 29.214'W 2 1° 57.580'N 157° 29.053'W 3 1° 56.424' N 157° 29.357'W Fanning 1 3° 51.786'N 159° 2.170'W 2 3° 54.605' N 159° 23.477'W 3 3° 50.506' N 159° 21.640'W Palmyra 1 5° 52.255'N 162° 06.612* W 2 5° 52.622' N 162° 06.933'W 3 5° 52.170'N 162° 06.901' W 4 5° 52.241'N 162° 02.704' W Bawe 6° 09.183'S 39° 08.567' E Changuu 6° 06.900' S 39° 09.967' E Chumbe 6° 16.767' S 39° 10.533'E Nyange 6° 13.067' S 39° 08.933' E Pange 6° 11.083'S 39° 09.617' E Table 3.7: Family names for Figure 3.5 Family Atlantic Indian Pacific Mediterranean Number 1 Aulostomidae Syngnathidae Zanclidae Congridae 2 Synodontidae Muraenidae Balistidae Carangidae 3 Chaetodontidae Plotosidae Ostraciidae / Atherinidae Tetraodontidae 4 Grammistinae Synodontidae Labridae Gadidae 5 Tetraodontidae Pomacentridae Pomacentridae Gobiidae 6 Serranidae Nemipteridae Carangidae Scorpaenidae 7 Pomacanthidae Labridae Holocentridae / Sciaenidae Priacanthidae 8 Holocentridae Pinguipedidae Pomacanthidae Sparidae 9 Scaridae Mullidae Lutjanidae Blenniidae 10 Pomacentridae Microdesmidae Caesionidae Tripterygiidae 11 Gobiidae Clupeidae Serranidae Apogonidae 12 Labridae Chaetodontidae Acanthuridae Mullidae 13 Ostraciidae Apogonidae Syngnathidae / Mugilidae Belonidae 14 Sciaenidae Blenniidae Chaetodontidae Centracanthidae 15 Haemulidae Serranidae Scaridae Labridae 16 Carangidae Zanclidae Lethrinidae Pomacentridae 17 Chaenopsidae Monacanthidae Mullidae Serranidae 18 Acanthuridae Siganidae Kyphosidae Muraenidae 19 Mullidae Tetraodontidae Mugilidae 20 Scombridae Scorpaenidae Sharks 21 Sparidae Scaridae 22 Lutjanidae Caesionidae 23 Sphyraenidae Pomacanthidae 24 Labrisomidae Ostraciidae 25 Blenniidae Gobiidae 26 Balistidae Lethrinidae 27 Monacanthidae Holocentridae 28 Kyphosidae Echeneidae 29 Muraenidae Lutjanidae 30 Acanthuridae 31 Balistidae 32 Dasyatidae 33 Haemulidae 34 Kyphosidae 35 Priacanthidae 36 Pempheridae 37 Aulostomidae 38 Ephippidae 39 Platycephalidae 40 Cirrhitidae See Appendix at the end of the thesis for Table Al: Species list for all study locations. 80 3.10 Appendix 2 - Theoretical Model of the Effects of Exploitation on the Power- Law SAR This appendix is related to, but was not apart of, the original publication of this paper. It is now being prepared as a separate manuscript. We might expect that exploitation could be added to the list of ultimate drivers of the SAR, as it is well known that it can have substantial effects on diversity (e.g. Worm et al. 2005) and ecosystem structure (e.g. Roberts 1995). But how might such changes manifest themselves in the parameters of a power-law SAR? In order to further explore such questions, we create a simple model focussing on the effects of changes in species richness and spatial patch occupancy. The assumption of a power-law SAR restricts the generality of our model but gives useful insight into what we might expect from an empirical comparison between exploited and unexploited regions. Consider a set of Q > 1 non-nested sampling units, each of which is of area X (Type II or III in the classification scheme of (Scheiner 2003), depending upon whether or not the sampling units are contiguous). Assume that every sampling unit is completely censused. Let there be N species distributed amongst these sampling units (N > 1). It is possible for any species to be present in any patch (i.e. the range of all species is equal to or greater than the total sample area). We make no assumptions about the distribution of relative species abundance. Define fy to be an indicator function such that/y= 1 if species i is present within sampling unity, and/;, = 0 otherwise. Define «, to be the number of 81 species in unit.;'; by definition, n} =^dfiJ - The total species richness in all sampling 1=1 units is: Stot=N (3.8) The mean species richness for a single sampling unit is: s 3 9 = 2>y/g ( - ) 7=1 The simplest possible SAR consists of two spatial levels, namely an individual sampling unit and the sum of all sampling units. At the scale of the individual sampling units, we use mean species richness (Eqn. 3.9) as is standard (Scheiner 2003). We construct this SAR using the power function (Eqn. 3.1), where z is the slope in log-log space and c a fitted parameter representing the number of species in an area of size A = 1 (Drakare et al. 2006). We log transform (using base 10 logs) such that: log 5 = log c+z log A (3.10) 82 For each of the two spatial levels, we can insert equations (3.8) and (3.9) into equation (3.10) respectively, to get: log TV = log c+ z(log Q + log X) (3.11) f Q \ log = logc + zlog/l (3.12) v» J with equation (3.11) deriving from the fact that the total area is QX and the properties of logarithms. These two equations represent the upper and lower points of our SAR. Equation (3.11) - equation (3.12) then gives us: ( Q \ log(iV)-log £/i,./g = zlogg (3.13) Hence: f \ log NQ v >' J J z = logQ (3.14) We can determine the minimum and maximum values of the term in the round brackets on the right hand side of equation (3.14). The minimum occurs when all species are 83 present in every sampling unit (since ^ «. = NQ), and has the value NQ/NQ = 1. The 7=1 maximum is Q (when each species is only present within a single patch, in which case o ^tij = N, since all species must be somewhere). The value of z, therefore, must vary 7=1 between zero and one. This is in agreement with Harte et al. (1999), who found that the power law form of the SAR implies self-similarity and hence that 1 > z > 0. We can substitute the equation for z into equation (3.11) to derive a formula for log c: NQ NQ log A logc = logyV-log -log (3.15) log0 7=1 J NQ log/i log c - log yV- log + 1 (3.16) logQ i—lj=\ 7 We now consider the effects of changes in species richness and patch occupancy upon the values of the power-law parameters z and c. i. Change in Species Richness Suppose that we remove x species from the set of sampling units (through exploitation), where 0 species-area exponent in the exploited system (ze) is derived from equation (3.14) as: 84 (N-x)Q log Q E J (3.17) log(0 N-x where nj(e) - ^fy is the number of species in sampling unity in the exploited system. The difference in slope between the original and the exploited system is then: N 7=1 HKe) log e (N-x) y n (3.18) z- z„ = ioge There is no change in the value of z from removal of species if and only if the term in the round brackets equals 1. Let us define NI (N-x) as diarge and —^ as Bsman. An equivalent expression for the circumstances in which there is no change in z is then: ^/arge "~ "small (3.19) Equation (3.19) compares the ratio of change in species richness at the large spatial scale (Oiarge) to the ratio of change in species richness at the small scale (9smaii)- If the equality is satisfied then there is no change in the value of z with exploitation. This is only true when the proportion of species richness lost at both spatial scales is equal. If the equality does not hold, then the slope of the SAR in the exploited system can either 85 increase or decrease, which we shall illustrate with two examples. Firstly, let us consider the smallest possible change in the mean species richness of an individual sampling unit. This occurs when the x removed species were present in only one sampling unit, while the remainingN-x species are in all Q sampling units. Then: *...„ = (Af-;) + '/g (3-20) N-x which is strictly less than 0iarge- Thus the ratio of change at the smallest spatial scale is smaller than the ratio of change at the largest spatial scale. The right hand side of equation (3.18) is therefore positive and thus ze < z; the slope parameter decreases when total species richness is reduced. Conversely, now consider the largest possible change in the mean species richness of an individual sampling unit by assuming that the x removed species existed in all small sampling units, while the N- x species that remain are found in only one sampling unit. Then: (N-x)/Q + x N + xjQ-l) sma"~ (N-x)IQ " N-x C } and this is strictly larger than 0iarge, since Q>\. Thus the right hand side of equation (3.17) is negative and so ze>z; the slope parameter increases when total species richness is reduced. So a decrease in total species richness can cause an increase, decrease, or no change to the slope parameter z, depending upon the relative change in species richness at large and small scales. 86 As for the fitted constant c, the value for the exploited system ce remains unchanged only if, from equation (10): f logA NQ \N-X)Q^ log c - log ce -\ogN- log(N -x)- + 1 log log logg Q n, V2 V VZ y=i J J) (3.22) which simplifies to: N log^ logc-logce =log +i log (3.23) N-x ioge (TV-*)£%,. log/i /arlargee logc-logce=log(#/argJ- + 1 log * (3.24) logg y is equal to zero. Here, the dependence of c upon the units of areal measurement becomes clear, as the term in the square brackets can take values that are positive or negative, depending upon k, the area of an individual sampling unit. Thus the value of log c may increase, decrease, or remain the same under exploitation, depending upon 0iarge, 6smau, and the units of areal measurement. 87 Applying a comparable procedure to an increase in species richness under disturbance shows that similar relationships hold; z may decline, remain unchanged, or increase, depending upon the relative change at large and small scales in log-log space, and the same holds for c but with the added complication that the units of areal measurement play a role. Thus we conclude that a change in total species richness under exploitation, or indeed any type of perturbation, can cause a predictable change (dependent on the factors outlined above) in the values of the slope parameter z and the fitted parameter c of a power-law S AR. ii. Change in Spatial Patchiness Define o, to be the number of sampling units occupied by species i, i.e. Q n °i - ^fn sucn mat 1- °i - Q- Define o = ^o,• IN to be the mean occupancy across all .7=1 (=1 species. Then, from equation (3.9), the mean species richness for an individual sampling unit is: ~^f.JLfijlQ = YtfvIQ = fJoiIQ = miQ (3.25) ./=1 1=1 1=1 _/=l ,=1 Now assume that mean patch occupancy declines by an amount b, with 0 < b < o. Then the expression for mean species richness at the smallest spatial scale becomes: s = N(o-b)/Q (3.26) 88 Changing patch occupancy has no effect on total species richness, of course, which remains N. Then from equations (3.14), (3.25) and (3.26), the difference in slope between z and ze will be: z_z _ \og{NQ)-log(M?IQ) \og(NQ)-\og{N(o -b)IQ) log(0 log(0 which after tidying becomes: z~ze =r4^[lo8(^(^-^)/0-log(A^/0] (3.28) log(0 Since (o-b) < 6, the expression on the right hand side is negative, and so the slope parameter z strictly increases in an exploited system under reduced average patch occupancy. From equation (3.11) we can derive the change in the parameter c as: logc-logce =-(z-ze )(log Q + log A) (3.29) We already know that z - ze is negative, and hence the value of c decreases unless log Q + log X < 0. This can only occur when Qk < 1; i.e. when the units of measurement are such that the summed area of all sampling units < 1. In log-log space, therefore, log(QX) will be on the left-hand side of the x-axis at 0 (since log(l) = 0 and QX < 1). We can visualise this (in log-log space) as the SAR intercept (log c) being to the right of the total sampling area (QX), so that the change in c outlined above is reversed relative to that 89 when the total sampling area is to the right of the SAR intercept. This serves up warning, again (Rosenzweig 1995), of the confusion associated with the parameter c in power-law SARs: log(c) is the intercept of the SAR in log-log space, and c is the predicted number of species in an area of size A = 1. Hence a decrease in patch occupancy will cause a strict increase in the value of z, and a decrease in c unless the total study area is less than 1, in which case c will increase. A similar line of reasoning shows that an increase in patch occupancy will decrease the value of z. These results are in line with those derived by He & Legendre (2002). If we make the reasonable assumption that increased intraspecies aggregation reduces mean patch occupancy then a reduction in species richness (at the smallest spatial scales) occurs in both our model and theirs with this process. This will lead to an increase in the slope of a power-law SAR if total species richness remains constant. Also in agreement with He & Legendre (2002), we can see that it is possible for net changes in the slope of the SAR caused by one factor (species richness) to be counterbalanced by changes in another (spatial patch occupancy). Thus we have shown that for a simple power-law SAR it is possible for changes in species richness and patch occupancy, caused by exploitation or indeed any form disturbance, to have a predictable effect on both the z and c parameters. The situation is undoubtedly more complex in more realistic systems (i.e. SARs with > 2 spatial levels), but, at least in principle, we might anticipate such effects. 90 Chapter 4. Predicting Global Habitat Suitability for Stony Corals on Seamounts 4.1 Abstract 4.1.1 Aim Our knowledge of the distributions of species in the deep-sea is extremely limited, because of the considerable difficulties of sampling such an inaccessible environment. Data coverage for particular species groups is often sparse and true absence data lacking, providing substantial challenges to conventional methods of statistical modelling. One taxonomic group of considerable interest is the cold-water stony corals (Phylum Cnidaria; Class Anthozoa; Order: Scleractinia). Such corals may provide complex habitat for other benthic organisms. They are frequently found on seamounts (undersea mountains), environments that, relative to the surrounding seafloor, may have an abundance of hard substrata providing suitable habitat for stony corals and their associated fauna. Yet sampling of this taxon on seamounts is spatially heterogeneous and almost non-existent in some regions of the global marine environment, and information on its distribution lacking. Here we use methods specifically designed for such data to model global habitat suitability for cold-water stony corals on seamounts. Subm. as: Tittensor, D. P., Baco, A. R., Brewin, P., Clark, M. R., Consalvey, M., Hall-Spencer, J., Rowden, A., Schlacher, T., Stocks, K. & Rogers, A. D. Predicting global habitat suitability for stony corals on seamounts. 91 4.1.2 Location Seamounts worldwide. Our knowledge of the locations of seamounts is limited (-15,000 likely seamounts identified out of up to -100,000 predicted with an elevation > 1,000 m), so we derive a general pattern of habitat suitability at multiple depths, then map this to the summits of likely seamount locations. 4.1.3 Methods We compiled a database consisting of all accessible records of scleractinian corals on seamounts from the literature, museum records, online databases, and individual scientists. Two modelling approaches were selected for their suitability in modelling such presence-only data: maximum entropy modelling (Maxent) and environmental niche factor analysis (ENFA). Fourteen environmental data layers were included in the models, each at a 1 degree grid cell resolution and in 250 metre depth increments. We generated habitat suitability maps, and used a cross-validation process with a threshold-independent metric to assess and compare models. 4.1.4 Results Both models generated patterns of habitat suitability that are similar to and consistent with current knowledge. Habitat that is relatively highly suitable was located in the North Atlantic at all depths, and within a band between 20 and 50 degrees south at up to 1500 metres depth. The remainder of the global oceans are predicted to be relatively unsuitable, except for intermittent near-surface patches. Factors associated with high predicted habitat suitability include the aragonite saturation state, percent oxygen saturation and dissolved oxygen. Dissolved inorganic carbon, nitrate, and phosphate are 92 among the factors for which low levels are associated with high predicted habitat suitability. 4.1.5 Main Conclusions Suitable habitat for Scleractinia on seamounts at the global scale is predicted to be spatially heterogeneous both in the horizontal and vertical axes. The North Atlantic and a circum-global strip between 20 and 50 degrees south appear to be key areas for this Order. Our analyses indicate factors that appear to be important in constraining large- scale distributions for stony corals. Both models performed well under cross-validation, and gave qualitatively similar results, though the Maxent method consistently outperforms ENFA. Though predictions must be interpreted within the framework and assumptions of the models, these results present a first-order hypothesis of where these taxa are likely to be found that can be tested by further sampling. 4.2 Introduction Despite rapid advances in technology and a growing interest in the world's oceans, the deep sea continues to remain proportionally one of the least biologically sampled ecosystems on the planet (Glover & Smith, 2003). Through remarkable efforts, there have been substantial increases in our recent understanding of this environment, and new research continues to provide unexpected insights (e.g. Brandt et al. 2007). Yet given data limitations, understanding patterns of diversity and distribution in the deep sea poses a special challenge, especially at large spatial scales (Gage, 2004). Statistical modelling techniques provide a useful tool to synthesize available data and give generality to our 93 understanding of the ecology of such remote environments. In this chapter we explore the factors that may limit the distribution of stony corals on seamounts, and make predictions of locations where suitable habitat for this taxonomic group is likely to be found. Seamounts - undersea mountains - are generally conical, of volcanic origin, with a base that may be circular, elliptic, or elongate (Rogers 1994). In comparison to the surrounding deep-sea, seamounts provide a unique environment often characterised by strong local currents and the presence of rocky substrata, and are typically occupied by communities of sessile suspension-feeding organisms including deep-sea corals (Rogers 1994 ; Koslow et al 2001 ; Clark et al. 2006). The depth of a seamount, as elsewhere in the deep sea, plays a large part in structuring the biological communities to be found therein (O'Hara, 2007; Rogers et al. 2007). Pelagic predators may form aggregations to exploit the high availability of food resources sometimes associated with these features (Worm et al. 2003; Tynan et al, 2005). They can also have high apparent levels of diversity and endemism (Richer de Forges et al. 2000), although these assertions have been challenged recently (Samadi et al. 2006 ; O'Hara 2007; Hall-Spencer et al, In press). Because of incomplete mapping of the deep sea floor, the total number of seamounts is unknown, with recent estimates varying by around an order of magnitude, from -14,000 to -100,000 for features with an elevation >1,000 m (Wessel 2001 ; Kitchingman & Lai 2004). Figure 4.1 shows the location of seamounts predicted by the analysis of Kitchingman & Lai (2004), derived from a global bathymetry data set with 2 minute resolution. This analysis certainly does not contain every large seamount, as 94 Figure 4.1: Potential large seamount (greater than 1000 m elevation) locations (-14 000) predicted from an analysis of global digital elevation data generated by Kitchingman & Lai (2004). Colour indicates summit depth in metres. evidenced by the fact that the coral database we compiled has records from seamounts not included in these predictions. Cold-water coral reefs are common features on seamounts and the slopes of continental margins and islands (Buhl-Mortensen & Mortensen 2005 ; Roberts et al. 2006). Some species of Scleractinia (stony corals) can form reefs that, in comparison to shallow-water tropical reefs, may have similar or higher associated diversity of some animal groups (Rogers 1999). This diversity results from the complex habitat structure that such reefs provide (e.g. Freiwald et al, 2002), although the functional relationships between the coral and the many associated species are as yet generally unknown (Clark et al. 2006). Cold-water corals are slow-growing organisms that are vulnerable to a range of human impacts and disturbance, including bottom trawling (e.g. Clark & O'Driscoll 2003), hydrocarbon drilling (on continental margins), seabed mining, and ocean acidification (Roberts et al. 2006). These threats have raised concerns about the potential loss of the habitats formed by deep-water corals - and their associated diversity - by both scientists and policymakers (e.g. UN General Assembly, 2006; Turley et al, 2007). This has led to a pressing need to redress the lack of knowledge of the global distribution of cold-water corals and other habitat-forming species on seamounts (Clark et al 2006). We combined a recently compiled database consisting of samples of stony corals from seamounts with gridded global environmental data to fit two species distribution models. Both modelling techniques have been developed to work with presence-only data; i.e., where true absence data are either unavailable or unreliable, as is the case in much of the deep sea and in other inaccessible environments. Verification of the absence of stony corals from even an individual seamount would in itself be an imposing task and 96 unquestionably beyond the reach of current data. In light of this, we use only known presence data from those seamount locations at which scleractinians have been sampled. We focus our efforts at a broad oceanic scale, whist noting that the factors responsible for distributions may be markedly different at the scale of an individual seamount. 4.3 Data 4.3.1 Coral Data A database containing publicly-accessible geo-referenced records of stony coral samples from seamounts has recently been compiled (Rogers et al. 2007). Data sources included the primary scientific literature, online databases, museum databases, and records held by individual scientists. The sampling of corals from seamounts is very uneven (Figure 4.2), with the majority of samples coming from the southwest Pacific and the North Atlantic, while the Indian Ocean contained very few. In total, there were 1,880 scleractinian records, of which 1,651 contained latitude, longitude and depth information. Records of fossilised corals were further excluded from the analysis. Corals taken from a trawl in which the exact depth is uncertain and given as a range are, for the purposes of this study, assumed to come from the midpoint of that range. We binned scleractinian records to a one degree global grid with 250 m depth intervals down to 2500 m (resolution determined by the environmental factors; see below). Duplicate records were removed, giving each cell a binary record of presence or absence. Below 2500 m seamount stony coral samples are very rare (-10 records globally), so we restrict our models to this maximum depth. On this grid, a total of 404 cells contained seamount coral presence records. 97 Figure 4.2: Locations of Scleractinia coral samples from seamounts. Colour indicates depth in metres. 90 E 135 E 180 E 80 S 500 1000 1500 2000 2500 98 4.3.2 Seamount Data We followed the most commonly accepted definition of a 'large' seamount as having a vertical height >1,000 m above the surrounding sea-floor (Wessel 2001). We used seamount locations generated by the bathymetric analysis of Kitchingman & Lai (2004). Some coral records in our database are from seamounts not detected by this analysis; to resolve this, we model habitat suitability for the whole global ocean <2,500 m depth, but restrict coral presence data to seamounts only (i.e. we do not include coral records from other deep-sea habitats). In this way, we can assess global habitat suitability for those seamounts both known (Kitchingman & Lai 2004) and yet to be discovered (by predicting general habitat suitability for seamounts at each depth layer). 4.3.3 Environmental Data Environmental data layers were selected for their potential importance in driving seamount coral distributions at the grain of our study. All such data (listed in Table 4.1) were compiled onto a one-degree resolution global grid, centred on the midpoint of each degree cell. This resolution was chosen to fit with data availability; both World Ocean Atlas (WOA) and Global Ocean Data Analysis Project (GLODAP) data were available at a one degree resolution. Vertically, data were linearly interpolated into boxes spanning 250m, with the first box going from the surface to 250 m, and the last box from 2250m to 2500m. Physical data and primary productivity model output were all long-term annual means. Where possible, data were selected from the 1990s for maximum congruence with GLODAP environmental data. Environmental layers needed to be in two-dimensional form for input into our models; we achieved this by concatenating depths horizontally and using dummy variables for latitude and longitude. 99 Table 4.1: Environmental parameters used to predict habitat suitability. GLODAP = Global Ocean Data Analysis Project; SODA = Simple Ocean Data Assimilation 1.4.2; VGPM = Vertically Generalized Productivity Model; WO A = World Ocean Atlas 2005. Parameter Units Source Reference(s) Alkalinity (total) umol kg"1 GLODAP (Key et al. 2004) Delta C03 (aragonite umol kg"1 Derived from (Key et al. 2004); saturation state) GLODAP data (Orr et al. 2005); (Zeebe and Wolf-Gladrow 2001) Depth m WOA (Locarnini et al. 2006) Dissolved oxygen mil"1 WOA (Garcia et al. 2006a) Export primary mg C m"2 yr"' VGPM (Behrenfeld and Falkowski productivity 1997) Nitrate umol 1"' WOA (Garcia et al. 2006b) Percent oxygen saturation % WOA (Garcia et al. 2006a) Phosphate umol 1"' (Garcia et al. 2006b) Primary productivity mg C m"2 yr"' VGPM (Behrenfeld and Falkowski (overlying water) 1997) Salinity Pss WOA (Antonov et al. 2006) Silicate Umol 1"' WOA (Garcia et al. 2006b) Total C02 (dissolved umol kg"1 GLODAP (Key et al. 2004) inorganic carbon) Temperature °C WOA (Locarnini et al. 2006) Regional current velocity cms"1 SODA (Carton et al. 2000) 100 The World Ocean Atlas 2005 data (http://www.nodc.noaa.gov/OC5/WOA05/pr_woa05.html) used in our models are composite annual objectively analysed means. Global Data Analysis Project (GLODAP) gridded data (Key et al. 2004) are mostly derived from 1990's WOCE (World Ocean Circulation Experiment) cruises. VGPM (Vertically Generalized Production Model) outputs (Behrenfeld and Falkowski 1997) are depth-integrated surface values corrected for cloudiness, derived from data collected between 1977 and 1982. SODA (Simple Ocean Data Analysis) modelled current velocities (Carton et al. 2000) were the grand mean of the annual means for the period 1990-1999, using the 1.4.2 version of the model. The aragonite saturation state was calculated using GLODAP data and following the 2 A[C03 "]A method of (Orr et al. 2005), with constants as described in (Orr et al. 2005) and equations following Zeebe & Wolf-Gladrow (2001). Positive A[C032"]A values indicate supersaturation, negative values undersaturation. Depth is included as a parameter not because it is important per se, but because it correlates with unmeasured factors (such as pressure, for example). 4.4 Models For some species, particularly those that are sessile or easily detectable, it is possible to census an environment to such an extent that the presence and the absence of individual species can be near-certain (e.g. trees in measured plots). A wide variety of statistical models are available for these types of data (Guisan & Zimmerman 2000). In many environments, however, it is not feasible to sample in a manner that provides reliable absence data given realistic logistical constraints. Although it is generally preferable to use absence data where available because of the extra information contained therein (Bretons et al. 2004), presence-only statistical models have been developed for when it is not. In this study we use maximum entropy modelling (Phillips et al. 2006) and Environmental Niche Factor Analysis (ENFA) (Hirzel et al. 2002), both of which have been specifically designed for presence-only data. Maxent has been shown to perform well in comparison to traditional methods (Elith et al. 2006), and in simulations ENFA was robust to the quality and quantity of species data (Hirzel et al. 2001). Both of these models are also fairly robust to correlations among environmental data variables, removing some of the uncertainty arising from model selection (Hirzel et al. 2002 ; Phillips et al. 2006), though making it more challenging to determine the underlying factors responsible for distribution patterns. As with any generalized representation of real ecological systems, these models cannot capture the full complexity of underlying processes, and results should be viewed through this cautionary lens. We highlight some limitations here. In particular, presence- only modelling techniques have as yet been unable to include the effects of spatial autocorrelation, and cannot use the variety of methods available for presence-absence data (Dormann et al. 2007). For example, the use of a contagion factor (Segurado and Araujo 2004) would be highly questionable, since the lack of absence data means that cells without a confirmed sample may be wrongly labelled. Endogenous spatial autocorrelation in species data can occur due to biotic processes such as dispersal, spatial competition and inhibition, migration, and mortality (Legendre 1993 ; Fortin & Dale 2005). Detecting and mitigating the effects of spatial autocorrelation in presence-only models is a valuable area of research for the future, but although we cannot rule out any 102 effect, there are a number of reasons why we believe it is unlikely to have a strong influence on our results. Firstly, the horizontal resolution (grain) at which we are working 0(10-100 km) relative to the size and isolation of individual seamounts suggests that it is reasonable to assume that between-cell effects are likely to be minimised. Intra-specific effects will be mitigated by the fact that we are modelling the distribution of the whole Order. We note that coral water corals can persist over geologic time-scales; samples from the Atlantic & Mediterranean have shown continuous growth for up to 50,000 years (Schroder-Ritzraeu et al. 2005). This longevity may reduce the effects of short-term biotic interactions. Non-stochastic distribution models necessarily assume pseudo-equilibrium between observed species locations and the environment (Guisan & Thuiller 2005). The longevity of these corals makes this more likely to be the case. Furthermore, niche-based predictive habitat distribution models typically model the fundamental niche (Guisan & Zimmermann 2000) while biological (e.g. competition) interactions and stochastic events may prevent a taxonomic group from fully occupying this niche. These models will also not capture processes at a spatial scale less than that of the grid, a point which is considered further in the discussion. 4.4.1 Maximum Entropy Modelling Maximum entropy modelling (Phillips et al. 2006) is a technique adopted from statistical mechanics. Assume an unknown probability distribution g over a finite set X (such as the cells of a study area). We approximate g with a probability distribution g. The information-theoretic entropy (Shannon 1948) of our approximate distribution is defined as: 103 #(g) = -£g(*)lng(*) • (1) XEX In a general sense, the maximum entropy principle (Jaynes 1957) suggests that the best approach to approximating an unknown probability distribution is to maximise entropy, subject to constraints representing incomplete information. A distribution with higher entropy is less 'constrained'; the maximum entropy principle therefore ensures that no unnecessary constraints are placed on g. Algorithms guaranteed to converge to the maximum entropy distribution have been developed (Dudik et al. 2004). For further details on the method, see Phillips et al. (2006). We use MaxENT software (Phillips et al. 2006) to fit the Maxent model, using default model parameters (a convergence threshold of 10"5, a maximum iteration value of 1,000, and automatic regularization with a value of 10"4). A jack-knifing procedure was used to examine the importance of each variable, by comparing the model with that variable absent relative to that with it present. Habitat suitability maps were constructed by calculating a raw probability value p(x) for each grid cell x, such that the total of all cell probabilities summed to one. This value was then scaled logistically using the equation cp(x) / (1 + c p(x)), where c is the exponential of the entropy of the raw distribution, resulting in a relative habitat suitability value ranging from zero to one. All coral presence points were used to construct habitat suitability maps. 104 4.4.2 Environmental Niche Factor Analysis Environmental niche factor analysis (ENFA) is a niche-based predictive habitat suitability modelling technique for presence-only data based on multivariate ordination (Hirzel et al. 2002). The factors produced by the model are uncorrelated and have biological significance; the first represents species marginality (the absolute difference between the global mean and the species mean in the multidimensional environmental space) and the remainder specialization (the ratio of variance between the global distribution and species distribution). The factors are ordered by decreasing amounts of variance explained, except for the first factor, which explains all of the marginality and some portion of the specialization. In contrast to other ordination techniques, therefore, the first factor may explain less of the variance than subsequent factors. The larger the value of the marginality, the more different the species niche requirements are from the background mean (note that the value of the marginality will depend on the range of the background cells). A specialization value of greater than one indicates a more specialized niche, again relative to the background environment. For full details of this modelling technique, see Hirzel et al. (2002). We used the geometric mean algorithm to calculate a habitat suitability index for each cell (Hirzel & Arlettaz 2003). The habitat suitability maps were constructed using the isopleth method following (Hirzel et al. 2002), and using all coral data. We used a broken stick distribution to determine the number of factors (nine) that were used to construct habitat suitability maps (for details see Hirzel et al. 2002). Habitat suitability maps were indexed with a range of zero to one hundred, where higher values indicate more suitable habitat. Assumptions of ENFA include that data are multinomial (though Hirzel et al. (2002) suggest that it is robust to deviations from normality), and that occurrence data span the environmental range of the taxonomie pup in que§ti8n. Prior to running the model, environmental data were normalized using the Box-Cox transformation (Sokal & Rohlf 1995). We used Biomapper 3.2 (Hirzel et al. 2002) for the ENFA model. 4.4.3 Model Evaluation Although many methods exist for evaluating the predictive capabilities of presence-absence models (Guisan & Zimmerman, 2000), validating the performance of presence-only models is a subject of considerable and ongoing research (Boyce et al. 2002; Hirzel et al. 2006), because of the problems inherent in discriminating models predicting blanket presences from those that are more selective. Validation indices that have been developed for such situations are often dependent on arbitrarily selected thresholds. Recent developments, however, have shown considerable promise in resolving some of these difficulties (e.g. Hirzel et al. 2006 ; Phillips et al. 2006). We used a threshold independent measure, the AUC (area-under-curve), to assess our models (Zweig and Campbell 1993 ; Fielding and Bell 1997). Although AUC was originally derived (in a species-distribution modelling sense) for presence-absence models, it can also be used for presence-only models. AUC is calculated by summing the area under a receiver operating characteristic (ROC) curve, a plot of sensitivity (fraction of true positives, i.e. the proportion of positive instances that are classified positive) against 1 - specificity (the false positive rate, i.e. the fraction of negative instances that are classified positive) for all possible thresholds of a binary classifier. The value of an AUC index varies between 0 (performance worse than random) and 1 (perfect discrimination), with 0.5 being indistinguishable from random. Defined in 106 this manner, ROC plots require presence/absence data. Plotting sensitivity against a random sample of background locations (i.e. without species presences), however, is equivalent to replacing absences with pseudo-absences, and is sufficient to define an ROC curve (Wiley et al. 2003 ; Phillips et al. 2006). Utilising this method, however, means that in contrast to an ROC curve created using presence/absence data for which the maximum obtainable AUC is 1, the maximum achievable AUC is 1 - a I 2, where a is the fraction of grid cells that the species' distribution covers. This is typically an unknown quantity, so it is not possible to determine how optimal an AUC value is when generated by this procedure. It is, however, possible to determine whether the AUC is statistically distinguishable from a random model (AUC value of 0.5), and to compare the prediction strength of multiple models using the same data. For the first procedure we use a Wilcoxon rank-sum test statistic, and for the second a non-parametric test based on the theory of generalized ^/-statistics (Delong et al. 1988). We used a cross-validation procedure to evaluate the performance of our models, by creating ten random partitions of the occurrence localities, splitting the data in each partition between calibration (70%) and evaluation (30%) data sets. The same ten random partitions were used for both models. AUC values were calculated for the evaluation data. Habitat suitability maps were constructed from the full data set. Cross-validation procedures and statistical comparisons were carried out in Matlab v. 6.5 (http: //www. mathworks. com). 107 4.5 Results 4.5.1 Model Evaluation AUC values indicated that both models were statistically better than random (p < 0.0001 ; Wilcoxon rank-sum test) for all ten partitions of the data (Table 4.2). The Maxent model had a significantly higher AUC value than the ENFA model for every partition (p < 0.0001 ; Delong et al. (1988) non-parametric test). Direct comparison of habitat suitability maps is not possible because of the different scaling used by each model. Instead we consider each habitat suitability model separately. 4.5.2 Maxent Results Predicted habitat suitability from the maximum entropy model for a range of depths is depicted in Figure 4.3. Habitat with high relative suitability was predicted for the North Atlantic at all depths down to 2500 m. A circum-global strip between 20 and 50 degrees south had high relative suitability at above 1500 m depth. Parts of the North Pacific had suitable near-surface habitat, but little below 750 m and none below 1000 m. The Indian Ocean only had suitable habitat in the circum-global band in the south. Habitat suitability at depths below 1750 m (not shown) is similar to that displayed in Fig. 4.3d. Jack-knifing of variables suggested that the aragonite saturation state (ACO3) and total C02 are the variables with the most importance in constraining the distribution of stony corals (results not shown), though substantial correlation among environmental variables means that this must be interpreted with caution (Table 4.3; see discussion for further details). 108 Table 4.2: AUC values for all model runs. Models were calibrated using training data (70% of occurrence points, randomly selected), and AUC values calculated from test data (30% of occurrence points). All model runs fit significantly better than random (p < 0.0001). Maxent AUC values significantly larger than ENFA for all partitions (p < 0.0001). Partition ENFA Maxent 1 0.778 0.876 2 0.759 0.862 3 0.785 0.868 4 0.798 0.901 5 0.777 0.867 6 0.779 0.874 7 0.806 0.875 8 0.782 0.858 9 0.761 0.869 10 0.764 0.873 Mean 0.779 0.872 Standard 0.015 0.012 deviation 109 Figure 4.3: Predicted habitat suitability for seamount Scleractinia, using a maximum entropy model. Top left: 0 m to 250 m. Top right: 500 m to 750 m. Bottom left: 1000 m to 1250 m. Bottom right: 1500 m to 1750 m. Higher values indicate more suitable habitat. 10.1 110 Table 4.3: Correlation between environmental parameters. Parameters are the same on both axes. Alk. ACQ3 Pep. Dis.Q2 Nit. Pc. Q2 sat. Pho. P.Prd. Sal. Sil. TCQ2 Tmp. Vel. Ex. Prd. Alkalinity 1.00 -0.66 0.64 -0.48 0.60 -0.53 0.56 -0.07 0.12 0.86 0.82 -0.44 -0.20 -0.02 Delta -0.66 1.00 -0.86 0.25 -0.63 0.45 -0.62 0.04 0.28 -0.79 -0.74 0.66 0.26 0.02 CO3 Depth 0.64 -0.86 1.00 0.06 0.31 -0.11 0.29 0.00 0.01 0.63 0.50 -0.64 -0.28 O02 Dissolved -0.48 0.25 0.06 1.00 -0.68 0.94 -0.68 -0.08 0.10 -0.47 -0.68 -0.21 -0.01 -0.15 O: Nitrate 0.60 -0.63 0.31 -0.68 1.00 -0.82 0.96 -0.10 -0.47 0.78 0.90 -0.42 -0.12 O00 Percent 02 -0.53 0.45 -0.11 0.94 -0.82 1.00 -0.83 -0.03 0.27 -0.62 -0.83 0.04 0.04 -0.08 Saturation Phosphate 0.56 -0.62 0.29 -0.68 0.96 -0.83 1.00 -0.12 -0.50 0.77 0.88 -0.41 -0.11 -0.02 Primary -0.07 0.04 0.00 -0.08 -0.10 -0.03 -0.12 1.00 0.06 -0.20 -0.07 0.38 0.02 0.83 Productivity Salinity 0.12 0.28 0.01 0.10 -0.47 0.27 -0.50 0.06 1.00 -0.27 -0.30 0.27 0.02 0.08 Silicate 0.86 -0.79 0.63 -0.47 0.78 -0.62 0.77 -0.20 -0.27 1.00 0.91 -0.70 -0.23 -0.20 Total CQ2 0.82 -0.74 0.50 -0.68 0.90 -0.83 0.88 -0.07 -0.30 0.91 1.00 -0.48 -0.17 -0.04 Temperature -0.44 0.66 -0.64 -0.21 -0.42 0.04 -0.41 0.38 0.27 -0.70 -0.48 1.00 0.24 0.42 Velocity -0.20 0.26 -0.28 -0.01 -0.12 0.04 -0.11 0.02 0.02 -0.23 -0.17 0.24 1.00 0.05 Export -0.02 0.02 0.02 -0.15 0.00 -0.08 -0.02 0.83 0.08 -0.20 -0.04 0.42 0.05 1.00 Productivity 4.5.3 ENFA Results The marginality value of the ENFA model was 1.485; the specialisation 1.504. The marginality means that stony corals have environmental requirements substantially different from the background mean (Table 4.4) A specialisation value of greater than one indicated that this Order has a relatively narrow environmental niche. Factors that were associated with high habitat suitability included the aragonite saturation state (ACO3), dissolved oxygen, and percent oxygen saturation. Furthermore, low levels of phosphate, nitrate, silicate, and dissolved inorganic carbon were associated with high habitat suitability. High correlation among factors means that these results must be interpreted with caution (Table 4.3). Predicted suitability for a range of depths is shown in Figure 4.4. Suitable habitat in the ENFA model is spatially distributed in a manner similar to the Maxent model, though with fewer fine-scale features. The most noticeable difference is lower habitat suitability for the near-surface northeastern North Atlantic (Figs. 4.3, 4.4). 4.5.4 Habitat Suitability for Seamount Summits Both models show similar patterns of habitat suitability for likely seamount summits (Fig. 4.5, values not directly comparable, only relative suitability). Seamount summits in the North Atlantic are highly suitable - as a result of the high relative habitat suitability at all depths (Figs. 4.3 & 4.4). Other suitable seamount summits are above 1500 m depth in the circum-global strip in the southern hemisphere; these are particularly clustered around New Zealand, with others patches scattered along this strip. The few predicted seamount summits in the North Pacific with a high relative suitability are those 112 Table 4.4 Variance explained by the first eight ecological factors in the ENFA model. Factor one explains the marginality; the remainder the specialization. The cumulative explained specialization of the first nine factors is 90.6%. Factor 1 explains 100% of the marginality. Factor 1 2 3 4 5 6 7 8 9 Alkalinity (total) -0.242 0.057 0.104 0.147 0.002 0.545 0.131 0.334 0.486 Delta C03 (aragonite 0.317 0.245 0.174 0.682 0.293 0.144 0.190 0.034 0.330 saturation state) Depth -0.233 0.085 0.020 0.181 0.035 0.399 0.012 0.020 0.217 Dissolved oxygen 0.204 0.037 0.565 0.147 0.535 0.057 0.351 0.359 0.408 Export productivity 0.149 0.003 0.000 0.009 0.007 0.005 0.010 0.026 0.034 Nitrate -0.325 0.550 0.139 0.053 0.039 0.002 0.687 0.048 0.304 Percent oxygen 0.271 0.023 0.731 0.104 0.592 0.316 0.028 0.097 0.352 saturation Phosphate -0.328 0.786 0.149 0.265 0.021 0.132 0.066 0.023 0.054 Primary productivity 0.306 0.014 0.003 0.047 0.014 0.076 0.117 0.035 0.149 Salinity 0.179 0.052 0.027 0.088 0.090 0.267 0.114 0.138 0.201 Silicate -0.361 0.017 0.105 0.072 0.161 0.173 0.225 0.600 0.359 Total C02 (DIC) -0.315 0.064 0.156 0.587 0.481 0.543 0.476 0.278 0.041 Temperature 0.283 0.020 0.165 0.123 0.094 0.008 0.207 0.537 0.152 Water velocity -0.042 0.005 0.000 0.009 0.004 0.028 0.007 0.008 0.068 Explained specialisation 0.095 0.218 0.171 0.107 0.092 0.072 0.064 0.050 0.037 Figure 4.4: ENFA predicted habitat suitability for seamount Scleractinia. Top left: 0 m to 250 m. Top right: 500 m to 750 m. Bottom left: 1000 m to 1250 m. Bottom right: 1500 m to 1750 m. Higher values indicate more suitable habitat. .180 W35 VW0 Wt5 W 0 45 E90 B35 B80 E W3S V*0 W45 W 0 45 E90 B35 B80 E 80 S 80 S .180 W05 VCO Wt5 WO 45 E90 H35 B80 E .180 W35 WO W45 W 0 45 E90 B35 B80 E 80 80 S 80 S 114 Figure 4.5: Predicted habitat suitability for Scleractinia (on a scale from 0 to 100) on the summits of potential large seamounts. Top: Results from Maxent model (suitability values multiplied by 100). Bottom: Results from ENFA model. Only seamounts with summit depths less than 2500 m are included. Due do differing scaling algorithms, values cannot be directly compared, only patterns of relative suitability. .1JP°W 136°W 90°W 4S°W 0° 45° E 90° E 135° E 180° E 0 10 20 30 40 50 60 70 80 90 100 115 projecting into shallow waters (see Figure 4.1). The most noticeable difference is the higher predicted habitat suitability for seamount summits in the tropical Atlantic by ENFA than by Maxent. Figs. 4.1, 4.3d & 4.4d suggest that this is due to most predicted seamount summits in this region being below 1,500 m depth, where relative habitat suitability is predicted to be higher by the ENFA model. 4.6 Discussion We used two presence-only modelling methods to construct habitat suitability maps for stony corals on seamounts. Both models show similar spatial patterns of high and low suitability (Figs. 4.3 & 4.4), consistent with current knowledge and with an earlier iteration of the ENFA model (Clark et al. 2006). The Maxent approach has an AUC value significantly higher than that of ENFA, indicating that it more accurately classifies test locations as being suitable relative to background locations. Mapping habitat suitability to the summits of 14,000 predicted seamount locations shows a heterogeneous pattern of habitat suitability for stony corals, with high suitability in the North Atlantic and in a mid-latitude strip in the southern hemisphere. The North Atlantic and patches within the southern hemisphere have previously been identified as an area of high habitat suitability for the deep-water coral Lophelia pertusa based on ENFA analysis of records of occurrence (Davies et al, In press). Disentangling the factors responsible for both driving and constraining the distributions of stony corals on seamounts at the global scale is complex, and ultimately requires background knowledge of the ecosystem and of coral biology. Factors that are strongly correlated with high habitat suitability for these taxa may not in themselves be 116 important, but may rather be correlated with another important variable. Furthermore, there is strong correlation between many of the environmental data layers used in the analyses, and although the models are robust to this, it further complicates the assignation of relative importance to each individual environmental variable. Nonetheless, the relative importance of each factor in the ENFA model (Table 4.4 and the jack-knife results from the Maxent model (not shown) can be used to provide some guidance. Of all the variables in the study, those of least importance were water velocity and export productivity (Table 4.4. High water velocity or turbulent flow is known to be likely very important in determining the distribution of deep-sea corals at the local scale (metres - 10s kilometres) on seamounts and in other geological settings (e.g. Genin et al, 1986; Frederiksen et al, 1992). Studies employing ENFA on deep-water octocorals and on the deep-water reef-forming scleractinian Lophelia pertusa have demonstrated that these corals favoured areas of high current velocity (regional scale) or steep or irregular topography (regional and global scale; Bryan & Metaxas, 2007 ; Davies et al, in press). High current flow can provide transport of nutrients and food to corals (e.g. Thiem et al, 2006) as well as preventing siltation of colonies and removing sediment from the seabed (Rogers 1994). At the global scale, however, the averaging of velocity over the grid cells in the present study failed to capture the fine resolution detail around each individual seamount, and hence the localised variation in flow. Export productivity provides little information to the model, again because it likely failed to capture small scale detail. The factors that were most important in determining habitat suitability on the scale examined were high levels of aragonite saturation, dissolved oxygen and percent oxygen saturation. Levels of aragonite saturation have been shown to closely regulate the calcification rates of zooxanthellate and azooxanthellate stony corals and other organisms 117 (e.g. Leclercq et al, 2000; Kleypas et al, 2006), and have been speculated to constrain the distributions of cold-water stony corals (Guinotte et al. 2006; Turley et al, 2007). The modelling appears to confirm this, with low suitability in the North Pacific, where the aragonite saturation horizon is shallow (~ 50 - 600m) in comparison to the North Atlantic (~ 2500 - 3000 m) (Feely et al. 2004). For cold-water coral reefs, accumulation of the coral framework is a balance between coral growth and destruction of coral skeletons by bioeroding organisms such as sponges and polychaete worms (Rogers, 1999). Thus conditions of low aragonite saturation are likely to be particularly unfavourable for cold- water coral reef formation as the balance between framework building processes and reef destruction are altered, especially as seawater may cause dissolution of calcifying organisms at low levels of saturation (Orr et al., 2005; Kleypas et al, 2006). The strong influence of high oxygen concentrations on the occurrence of scleractinian corals was also found for ENFA analysis of the distribution of Lophelia pertusa where the coral was found to occur in waters of 4.6 - 7.2ml l"1 (Davies et al, In press). This study was consistent with previous observations of the occurrence of Lophelia pertusa in waters of oxygen concentrations between ~ 4 - 5.5 ml 1" (Freiwald 2002, Fig. 9). Ecophysiological studies on Lophelia pertusa suggest that it cannot maintain aerobic metabolism at oxygen concentrations below 3 ml l"1 (Dodds et al, In press). Hypoxia has also been demonstrated to decrease calcification rates in zooxanthellate corals in the dark (Al-Horani et al, 2007). This effect is negated in daylight by the production of oxygen during photosynthesis, a mechanism not available to deep-water corals (Al-Horani et al, 2007). Overall, it would appear that regions of the ocean where conditions of low oxygen exist are unsuitable as habitat for stony corals. 118 Low levels of nitrate, silicate, phosphate and total dissolved inorganic carbon were also determined to be important for high relative habitat suitability, as found for Lophelia pertusa (Davies et al, in press). This confirms that nutrient concentrations, or factors correlated with nutrient concentrations are important in the determination of deep water coral distribution. Nutrient effects on coral growth have been extensively explored for warm-water corals, where nitrogen and phosphorus may increase photosynthesis but can inhibit calcification rates (Marubini and Thake, 1999 ; Renegar and Riegl, 2005; but see Langdon and Atkinson, 2005). Furthermore, the effects of nutrient enrichment may be additive with the effects of low aragonite saturation, rendering the environment particularly unsuitable for the occurrence of scleractinian corals. Increased nutrients and temperature can both also increase the occurrence and severity of coral diseases in shallow-water stony corals (Bruno et al. 2003, 2007 ; Boyett et al, 2007). Studies of the effects of nutrients on calcification rates have not been carried out for cold-water corals, but these azooxanthellate corals do not photosynthesise, and hence we might speculate that only the negative effects of high nutrient concentrations may affect them. The components of the seawater carbonate system (dissolved inorganic carbon, alkalinity, and delta CO3) are all highly correlated; high levels of delta CO3 are negatively correlated with dissolved inorganic carbon and alkalinity. Thus the availability of aragonite corresponds with low alkalinity and dissolved inorganic carbon. All of the above-mentioned environmental variables mentioned are strongly correlated (Table 4.3) Although this suggests that several variables could likely be left out of the models (two of phosphate, nitrate, and silicate, for example) without significant loss of information, both methods are robust to high correlations among environmental parameters; hence all variables are left in so as to retain as much information as possible. 119 This leaves as yet unresolved the question of which factors are the most important in determining the suitability of seamount habitat for stony corals. Experimental and observational studies on calcification rates in cold water corals and responses to physical environmental parameters are required to resolve this question (see Kleypas et al, 2006), as it cannot be determined from correlative studies such as this. Although seasonal changes in the physical and biological environment are likely to impact both shallow- and deep-water stony coral growth, there is considerably less variation in most variables (such as temperature, for example) at depth than in the surface ocean. Nonetheless, using high temporal resolution data, where available, to study seasonal effects on scleractinian growth and distribution could provide important additional information. Furthermore, the limitations of a lxl degree resolution grid could be explored by performing a higher spatial resolution analysis for a region in which finer- scale data were available. The consistency of patterns at differing spatial resolutions could be tested, and the relative importance of fine- and coarse-scale structure compared. A further potential caveat is that the calcification rate for cold-water corals is likely contingent upon environmental factors (see previous paragraph), and it could be that regions with smaller, slower growing corals are not sampled as effectively. If this were the case, then the results would be influenced by correlates of the calcification rate, rather than those of coral presence. The coral data used in this study were collected with numerous differing sampling techniques, so it is difficult to estimate the importance of this bias, but we note that as scleractinians are slow-growing, long-lived organisms, and many are reef-forming, it seems likely that, given reasonably long periods of environmental suitability, they would grow to sufficient size to be consistently captured by most sampling techniques. 120 The patterns predicted by the models are consistent with what is known about the distribution of this taxon (Figs. 4.3 & 4.4). The Atlantic has a relatively deep aragonite saturation horizon and high oxygen availability, thus providing suitable conditions for these corals to grow (e.g. Rogers, 1999; Hall-Spencer et al, In press). The Pacific Ocean has a much shallower aragonite saturation horizon, which could be responsible for the low predicted habitat suitability at depth (Feely et al. 2004). The northern Indian Ocean is a well-known extreme oxygen minimum zone (e.g. Helly and Levin, 2004), and the models predict low habitat suitability at all depths for this region (Garcia et al. 2006a). The maps of predicted habitat suitability on likely seamount summits (Fig. 4.5) suggest that suitability is strongly determined by geographical location, and also by depth, with shallower seamounts more likely to provide good habitat (except for in the North Atlantic, where suitability is high down to at least 2500 m). Habitat suitability on seamount summits likely differs from that on seamount flanks, especially for those regions where habitat suitability varies substantially by depth. The Maxent model outperforms the ENFA model for every cross-validation partition (Table 4.2). This is likely as it is able to fit more complex combinations of variables ('features', in the terminology of the Maxent model), including thresholds, quadratic functions, and products. ENFA, in contrast, is only able to fit linear dependencies of the species niche on environmental variables, unless transformations or nonlinear combinations of variables are also incorporated as layers. We therefore recommend the testing of multiple modelling approaches on deep-sea presence-only data, along with threshold independent metrics of validation, as there may be substantial differences in performance. Nonetheless, both models fit with AUC scores of > 0.75 for 121 all cross-validation partitions, which indicates a good discrimination ability appropriate for many uses (Pearce & Ferrier, 2000). 4.7 Conclusions Stony corals are slow-growing, long-lived, and vulnerable organisms, particularly on seamounts which are often the target of exploitation because of the high concentrations offish present (Clark et al. 2006). The maps derived from our statistical models suggest that global habitat for stony corals on seamounts is highly heterogeneous, being concentrated in the North Atlantic and in a circum-global strip between 20 and 50 degrees south. Such predictions can be verified by further sampling. The modelling that we have carried out presents a testable hypothesis as to the distribution of this Order on seamounts, and one that can only be assessed with a comprehensive and spatially explicit sampling scheme. Higher resolution environmental data or further coral samples would allow for refinement of these models. An understanding of likely differences in coral habitat between seamount summits, seamount flanks, and for different types of seamounts would also greatly increase our understanding. This could come through fine-scale predictive modelling of habitat suitability on individual seamounts, were such data available, and would provide enormous insight into the small-scale factors that are important for these organisms. Local-scale analyses may also reveal further regional differences, and species- specific analyses may identify varied requirements within this Order. Human impacts on cold-water corals can be substantial (Clark et al. 2006). Such effects may grow in the future as a result of climate change, and increased exploitation 122 pressure. The likely shallowing of the aragonite saturation horizon (Orr et al. 2005) could have a negative effect on habitat suitability on seamounts for deep-water stony corals (Guinotte et al. 2006 ; Turley et al, 2007). Our analysis suggests that efforts to protect these fragile organisms on seamounts must take the spatial heterogeneity of suitable habitat into account. This suitable habitat may be mostly limited to the North Atlantic and a circum-global strip in the southern hemisphere. The potential importance of these corals for the associated community matrix suggests that increasing our understanding of their distribution is important in helping to protect seamount biodiversity. Using predictive models to maximise the understanding gleaned from sampling efforts is one way in which to do this. 4.8 Acknowledgements We would like to thank R. A. Myers, J. MacPherson, W. Blanchard and A. Dickson for guidance and advice. This paper is a joint product of the CenSeam and FMAP programs of the Census of Marine Life. The ideas for this paper were conceived at a workshop in Wellington, New Zealand, sponsored by UNEP and the Netherlands Department of Nature, Ministry of Agriculture, Nature and Food Quality. 123 Chapter 5. Impact of Anthropogenic Ocean Acidification on Global Cold-Water Stony Coral Habitat 5.1 Abstract Cold-water stony corals are important ecosystem engineers (Jones et al. 1994) that create habitat for a diverse range of deep-water species. Ecological knowledge of this Order has only recently begun to come to light, yet evidence is already beginning to accumulate that they are under threat from a range of anthropogenic activities (Roberts et al. 2006). Among these, an increase in ocean acidity has been predicted to have a potentially detrimental impact on these corals over the next century (Guinotte et al. 2006). Precise knowledge of the likely extent of this threat is hampered by our lack of understanding of the distribution of these corals, due to limited sampling of the remote environments in which they live. We used a recently compiled database of cold-water stony coral sample locations and a species distribution model to predict global patterns of likely habitat suitability. We then applied these models to the projected state of ocean chemistry in 2099 under the IS92a 'business-as-usual' scenario. We find that predicted coral habitat in our model is distributed very heterogeneously, being concentrated in the northern North Atlantic and around New Zealand. Projected changes in ocean chemistry are predicted to cause a pronounced reduction in habitat suitability in the North Atlantic, In preparation as: Tittensor, D. P., Baco, A., Hall-Spencer, J. H., Myers R. A., Orr, J, Rogers, A. Impact of anthropogenic ocean acidification on global cold-water stony coral habitat. 124 and a low-to-moderate impact elsewhere. Seamount summits and upper flanks are predicted to be less impacted by these changes, likely due to their elevation placing them in waters with a higher aragonite saturation state. Smaller features, and the lower flanks of seamounts, are likely to be impacted in a similar manner to the benthos. These results suggest that anthropogenic-induced changes on ocean chemistry are likely to severely impact cold-water stony corals in the North Atlantic benthos, and have a lesser but still visible impact elsewhere in the global ocean. 5.2 Introduction The deep-sea benthos covers more than half of the Earth's solid surface (Gage & Tyler 1991), and has extreme ambient pressure, low temperature and no sunlight. It is the most remote ecosystem on Earth, yet extraordinarily species-rich (Snelgrove & Smith 2002) with new species continually being discovered (e.g. Brandt et al. 2007). The deep- sea benthos, with the exception of vent communities, is reliant on export production from the surface waters, often arriving as faecal matter or detritus (Snelgrove & Smith 2002). Although the existence of cold-water corals has been known of since the 18th century, they are among the deep-sea taxa for which the scale and abundance of their distribution has been revealed on recently (Roberts et al. 2006). Cold-water corals, as the name implies, are typically associated with the colder conditions found in offshore waters, although, strictly speaking, they are not limited to the deep-sea. Scleractinians, or stony cold-water corals, are azooanthellate suspension feeders, many species of which can form reefs (Roberts et al. 2006). These corals are often associated with high levels of faunal diversity, possibly due to the complex habitat that 125 they provide (Clark et al. 2006). Cold-water corals are intrinsically vulnerable to physical disturbance from human impacts. Bottom trawling for deep-water fish can severely damage cold-water corals and is thought to currently represent the most widespread, and the most damaging disturbance (e.g. Koslow et al. 2001 ; Hall-Spencer et al. 2002). Hydrocarbon drilling and seabed mining could have additional impacts in the future (Roberts et al. 2006). Recent research points to another potentially severe anthropogenic impact, the increase in ocean acidity due to elevated concentrations of atmospheric CO2. (Feely et al. 2004 ; Orr et al. 2005). Oceanic carbonate ion concentrations [CO32"] decrease under uptake of atmospheric CO2 according to the reaction pathway 2 C02 + CO3 " + H20 -* 2HC03" (1) (Orr et al. 2005). This reaction simultaneously decreases the ocean pH (by increasing hydrogen ion concentrations), and reduces the amount of calcite and aragonite, the two polymorphs of calcium carbonate used by marine calcifying organisms. This has a double-impact on calcifying organisms: one, by increasing the rate of dissolution of calcified shells when carbonate is undersaturated, and two, by reducing the availability of carbonate ions with which to form calcerous secretions (Orr et al. 2005). Assuming continually increasing anthropogenic emissions of CO2 into the atmosphere, the Intergovernmental Panel on Climate Change (IPCC) 'business-as-usual' scenario (labelled IS92a) estimates an atmospheric concentration of 788 p.p.m.v. in the year 2100. This is the scenario used to force the future ocean chemistry model of Orr et al. (2005). Here we model the effects of this scenario on cold-water stony coral habitat by 126 emulating present-day habitat suitability; and then extrapolating to future projections of ocean chemistry. 5.3 Methods 5.3.1 Coral Database The coral database was derived from four sources. Firstly, records of stony coral occurrence on seamounts worldwide, gathered through the literature, museum records, and from individual scientists (Rogers et al. 2007). Secondly, a database of Lophelia pertusa records compiled by the UNEP-WCMC (United Nations Environment Programme - World Conservation Monitoring Centre). Thirdly, a database of North-East Atlantic cold-water corals compiled by J. Hall-Spencer (unpubl.). Fourthly, surveys conducted by the United Kingdom Department of Trade and Industry (Hall-Spencer et al. in review). Records without geo-referenced coordinates or depth data were removed, and the remainder combined into a database totalling 6,025 coral presence records. Records containing only fossil corals were removed from the analysis (leaving 5,922 records). Coral records below 3500 m (n = 54) were excluded because some environmental layers were missing information in the grid cells where these corals were located. Figure 5.1 shows the locations of these coral samples. These records were then gridded to a one- degree by one-degree grid, and 250m vertical depth boxes (to match environmental data). In the end, 763 grid cells contained coral presence records, ranging from 65 degrees N to 63 degrees S. 127 Figure 5.1: Locations of scleractinian coral records in our database. Fossil records and those from > 3500 m in depth are excluded. Colour indicates depth in metres. 0 500 1000 1500 2000 2500 3000 3500 5.3.2 Environmental Layers Environmental layers thought to be important for the distribution of cold-water corals at a global scale (Clark et al. 2006 ; Chapter 4) were combined from published sources (Table 5.1). Data were globally gridded at a one degree resolution, the finest grain available for the majority of the data. Vertically, we gridded data into 250 m boxes from 0 m to 5500 m. Physical data and primary productivity model output were all long- term annual means. Where possible, data were selected from the 1990s for maximum congruence with the Global Ocean Data Analysis Project (GLODAP) environmental data. Environmental layers needed to be in two-dimensional form for input into our models; we achieved this by concatenating depths horizontally and using dummy variables for latitude and longitude. The World Ocean Atlas 2005 data (http://www.nodc.noaa.gov/OC5/WOA05/pr_woa05.html) used in our model were composite annual objectively analysed means. GLODAP gridded data (Key et al. 2004) were mostly derived from 1990's WOCE (World Ocean Circulation Experiment) cruises. VGPM (Vertically Generalized Production Model) outputs (Behrenfeld and Falkowski 1997) were depth-integrated surface values of productivity corrected for cloudiness, derived from CZCS (Coastal Zone Color Scanner) pigment data collected between 1977 and 1982. SODA (Simple Ocean Data Analysis) modelled current velocities (Carton et al. 2000) were the grand mean of the annual means for the period 1990-1999, using the 1.4.2 version of the model. Stony corals form their calcite skeletons from aragonite ions; the aragonite saturation state was calculated using GLODAP data and following the A[C032" ]A method of (Orr et al. 2005), with constants as described in (Orr et al. 2005). Positive A[C03 "]A values indicate aragonite supersaturation, negative values indicate 129 Table 5.1: Sources of environmental data used to predict habitat suitability. GLODAP = Global Ocean Data Analysis Project; SODA = Simple Ocean Data Assimilation 1.4.2; VGPM = Vertically Generalized Productivity Model; WOA = World Ocean Atlas 2005. Layer Past (P) / Units Source Reference(s) Future (F) Alkalinity (total) P umol kg"1 GLODAP (Key et al. 2004) Delta C03 (aragonite P umolkg"' Derived from (Orr et al. 2005) saturation state) GLODAP data Depth P M WOA (Locarnini et al. 2006) Dissolved oxygen P mil' WOA (Garcia et al. 2006a) Export primary P mg C m"2 yr"' VGPM (Behrenfeld and productivity Falkowski 1997) Nitrate P umol 1"' WOA (Garcia et al. 2006b) Percent oxygen P % WOA (Garcia et al. 2006a) saturation Phosphate P Umol l"1 WOA (Garcia et al. 2006b) Primary productivity P mg C m'z yr"1 VGPM (Behrenfeld and (overlying water) Falkowski 1997) Salinity P Pss WOA (Antonov et al. 2006) Silicate P umol 1"' (Garcia et al. 2006b) Total C02 (dissolved P umolkg"1 GLODAP (Key et al. 2004) inorganic carbon) Temperature P °C WOA (Locarnini et al. 2006) Regional current P cms"1 SODA (Carton et al. 2000) velocity Alkalinity (total) (2099 F umol kg"1 Derived from (Orr et al. 2005) under is92a) GLODAP data Delta C03 (aragonite F Umolkg"' Derived from (Orr et al. 2005) saturation state) (2099 GLODAP data under is92a) Total C02 (dissolved F umolkg"' Derived from (Orr et al. 2005) inorganic carbon) (2099 GLODAP data under is92a) 130 undersaturation. Future ocean chemistry parameters (Table 5.1) were calculated following Orr et al. (2005), under the IS92a 'business-as-usual' scenario. Modelling interannual variability in surface ocean [CO32] suggested a negligible effect in comparison to anthropogenic changes (Orr et al. 2005). Depth is included as a parameter not because it is important per se, but because it correlates with unmeasured factors (such as pressure, for example). 5.3.3 Habitat Suitability Model There are a range of modelling methods available for predicting habitat suitability and species distributions (Guisan & Thuiller 2005). Habitat suitability models follow different paradigms, but commonly attempt to relate field observations of species to environmental predictors (Guisan & Zimmerman 2000). Most of these models are designed for situations in which both presence and absence data are available. When only presence data are available, such as for cold-water corals, it is necessary to use models specifically designed for these data (Segurado & Araujo 2004). We use the maximum entropy method (Maxent) to model habitat suitability. Maxent is a machine-learning method that approximates an unknown probability distribution by maximising entropy (Shannon 1948) subject to constraints representing incomplete information. In the species distribution context, the unknown probability distribution is the species distribution, and the constraints are that the expected value of each environmental variable, or function of each environmental variable, should match or closely match its empirical average (for full details see Phillips et al. 2006). Maxent was one of two presence-only models tested on a seamount subset of the present coral database (Chapter 4), and was found to consistently provide the best classification of 131 presence records. We extend the previous version of this model (Chapter 4) by incorporating a more comprehensive coral database, with samples from many deep-ocean habitats, and by extending the model vertically down to a depth of 5500m, more than twice the depth of the original model. This is possible due to the increased numbers of records at depths > 2500 m in our full database. Many coral records in the database come from seamounts, ridges, or other elevated features too fine-scale to be detected on a one-degree resolution global bathymetry. In order to properly categorise these coral records in an appropriate vertical stratification scheme, we calculate habitat suitability for every 250 m environmental depth layer, and then map the results to World Ocean Atlas bathymetry. Using this method more accurately locates coral records in the vertical axis, as many come from features well over 1000 m higher than the seafloor. It also allows us to model changes in habitat suitability for predicted seamount summit locations that have been identified from a bathymetric analysis (Kitchingman & Lai 2004). Threshold-independent measures are appropriate for evaluating habitat suitability models (Fielding & Bell 1997), and do not require the selection of an arbitrary threshold for validation. These measures have been only recently been developed for presence-only models (Hirzel et al. 2006 ; Phillips et al. 2006). We use the AUC (area-under-the-curve) statistic to evaluate the fit of our model (Fielding & Bell 1997). This measure is derived from signal detection theory, and is commonly used in medicine (Zweig & Campbell 1993). The AUC value is calculated from a receiver operating characteristic (ROC) curve, a plot of the sensitivity of a binary classifier against (1 - specificity) as the discrimination threshold is varied. The sensitivity is defined as the fraction of true positives, i.e. the proportion of positive instances that are classified positive. 1 - 132 specificity is the false positive rate, i.e. the fraction of negative instances that are classified positive. The habitat suitability model output is used for classification. AUC is calculated by summing the area under an ROC curve for all possible thresholds. The value of an AUC index varies between 0 (performance worse than random) and 1 (perfect discrimination), with 0.5 being indistinguishable from random. Defined in this manner, ROC plots require presence/absence data. Plotting sensitivity against a random sample of background locations (i.e. without species presences), however, is equivalent to replacing absences with pseudo-absences, and is sufficient to define an ROC curve (Wiley et al. 2003 ; Phillips et al. 2006). Utilising this method, however, means that in contrast to an ROC curve created using presence/absence data (for which the maximum obtainable AUC is 1) the maximum achievable AUC is 1 - a 12, where a is the fraction of grid cells that the species' distribution covers. This is typically an unknown quantity, so it is not possible to determine an optimal AUC value using this procedure. It is, however, possible to determine whether the AUC is statistically distinguishable from a random model (AUC value of 0.5) by a Wilcoxon rank-sum test statistic, and to compare the prediction strength of models using a non-parametric test based on the theory of generalized (/-statistics (Delong et al. 1988). .We used a cross-validation procedure to evaluate the performance of our models, by creating ten random partitions of occurrence localities, and splitting the data in each partition between calibration (70%) and evaluation (30%) data sets. AUC values were calculated on the evaluation data. Habitat suitability maps were constructed from the full set of coral presence records. We use MaxEnt software (Phillips et al. 2006) to fit the Maxent model, using default model parameters (a convergence threshold of 10"5, a maximum iteration value of 133 1,000, and automatic regularization with a value of 10"4). A jackknifing procedure was used to examine the importance of each variable, by comparing the model with that variable absent relative to that with it present. Standard methods for analysing the effects of spatial autocorrelation cannot be used with presence-only models, and there is not at present any method suitable for such data (Dormann et al. 2007). However, for reasons covered in detail in Chapter 4, namely the grain of the study, and the longevity of these corals, we anticipate that such effects will not be substantial. Habitat suitability maps were constructed by calculating a raw probability value p(x) for each grid cell x, such that the total of all cell probabilities sum to one. This value is then scaled logistically using the equation cp(x) / (1 + cp(x)), where c is the exponential of the entropy of the raw distribution, resulting in a relative habitat suitability value ranging from zero to one, where larger values indicate higher relative habitat suitability. As we do not require binary predictions, we do not set an arbitrary threshold for species presence. Cross-validation procedures and statistical comparisons were carried out in Matlab v. 6.5 (http://www.mathworks.com). 5.4 Results All ten cross-validation partitions were statistically better at classifying presence records than random, with all AUC values > 0.9 (Table 5.2). From the jackknifing procedure (results not shown), the aragonite saturation state, temperature, and primary productivity were identified as the most important factors in determining the distribution of this Order, though due to strong correlation between many environmental factors (e.g. aragonite saturation state, total CO2, and temperature all 134 correlated at > 0.75) these results should be interpreted with caution. Thus we refrain from further evaluating the relative importance of such highly correlated factors Habitat suitability maps for present day conditions are shown in Figure 5.2. Three distinct seafloor regions appear to have high relative suitability for these corals (Fig.5.2a): the North Atlantic, the area around New Zealand, and much of the shallow water along the continental shelves and slopes. These distributions are consistent with current knowledge about cold-water stony coral distribution (Roberts et al. 2006), although due to missing environmental data we are not able to make predictions for the Caribbean, the Norweigan coast, or the Phillipines, where stony corals have been observed. When habitat suitability for likely seamount summits (Kitchingman & Lai 2004) are superimposed on this map (Fig. 5.2b), a broad array of suitable seamount summits in the North Atlantic, the areas around New Zealand and Hawaii, and a scattering of shallow summits throughout the rest of the oceans become apparent. This pattern is similar to a previous study on global habitat suitability for stony corals on seamounts (Chapter 4), suggesting that niche requirements for corals on seamounts are not dissimilar to those elsewhere in the deep ocean at this resolution. Habitat suitability maps for projected global changes under the IS92a scenario (Fig. 5.3a) on the benthic environment reveal that the effects of changes in ocean chemistry are predicted in many locations, but could be most dramatic in the North Atlantic. In this region, habitat suitability declines substantially for many benthic locations above around thirty degrees north. Changes to continental shelf regions and the area around New Zealand are visible, but of low to moderate impact in comparison to the large decreases in the northern North Atlantic. The effects on seamounts are also visible, 135 Figure 5.2: Predicted present-day habitat suitability for scleractinians. Higher values represent more suitable habitat, a) Benthic b) Benthic with likely predicted large seamount summits (Kitchingman & Lai 2004) superimposed. a) 90 E 135 E 180 E 80 S b) „1 135 W 90 W 45 W 0 45 E 90 E 135 E 180 80 ff w 3 <^%=* """i 60 N ^-i ^ §' 60 S 80 S' 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Figure 5.3: Predicted 2099 habitat suitability for Scleractinians given ocean chemistry changes modelled following the IS92a 'business-as-usual' scenario, a) Benthic b) Benthic with likely predicted large seamount summits (Kitchingman & Lai 2004) superimposed. a) 135 E 180 E 80 S b > J, 135 W 90 W 45 W 0 45 E 90 E 135 E 18C 30 w 3 C£» — 60 N ^•f iLitt^ /^^ V " 60 Si 80 SL 0.1 0.2 0.3 0.4 137 Figure 5.4: Global and regional changes in cold-water stony coral habitat suitability. Blue lines are present day, red predicted 2099 under the IS92a scenario. Lines represent percentage of grid cells/seamount summits with predicted habitat suitability value or higher. A) Global benthos. B) North Atlantic benthos north of 30 degrees north. C) Global likely seamount summits. D) Likely North Atlantic seamount summits north of 30 degrees north. A) C) 100 D) 0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8 Predicted habitat suitability Predicted habitat suitability 138 but less pronounced (Fig. 5.3b). This is likely due to the greater height of seamount summits in comparison to the surrounding bathymetry, placing them into depth strata with increased aragonite saturation (Feely et al. 2004). Plotting changes as percentages (Fig. 5.4) further reveals the geographic heterogeneity of effects. Globally, the cumulative percentage of benthic cells with habitat suitability between -0.1 and -0.9 is marginally reduced (Fig. 5.4a). The majority of global benthic cells have habitat suitability < 0.1 in both present and future predictions. For the North Atlantic at greater than 30 degrees north, however, the cumulative proportion of cells with habitat suitability of between -0.1 and -0.95 is substantially reduced, suggesting a severe localised reduction in suitable habitat (Fig. 5.4b). For seamount summits (Fig. 5.4c, 5.d) the changes are less noticeably localised, with a small reduction in the proportion of seamounts with habitat suitability values of between -0.1 and -0.7 both globally and in the northern North Atlantic. This suggests that acidification impacts on seamount stony corals are predicted to be more spatially homogeneous, and less pronounced. Model validation by AUC suggests very good discrimination, with AUC values being > 0.9 for all cross-validation partitions (Table 5.2). 139 Table 5.2: AUC values for all cross-validation model runs. Models were calibrated using 70% of occurrence points (randomly selected), and AUC values calculated from the remaining 30% of occurrence points. All model runs fit significantly better than random {p< 0.0001). Partition Maxent AUC 1 0.924 2 0.943 3 0.928 4 0.935 5 0.932 6 0.921 7 0.928 8 0.916 9 0.937 10 0.927 Mean 0.929 Standard deviation 0.008 5.5 Discussion We have presented the results of a global habitat suitability model for cold-water stony corals, and predicted changes due to anthropogenic perturbations of the oceanic carbonate system. Present-day highly suitable benthic habitat is located in two major areas, the northern North Atlantic and around New Zealand, with further areas of high suitability in the shallower water overlying continental shelves. Seamount summit habitat suitability is also high in these regions, but high habitat suitability was also found in other scattered locations where likely seamount summits are in shallower water more suitable than the surrounding benthos. Projected changes in ocean carbonate chemistry under the IS92a scenario for 2099 result in a substantial decrease in habitat suitability for seafloor stony corals in the northern half of the North Atlantic (-40 to 60 degrees N), and a low to moderate decrease in habitat suitability elsewhere and on seamount summits. Changes in habitat suitability under the IS92a scenario are due to an overall reduction in aragonite saturation, a shallowing of the aragonite saturation horizon (the interface between supersaturation and undersaturaton), and an increase in ocean acidity (Orr et al. 2005). Experimental and empirical evidence have both suggested that a decrease in the availability of aragonite ions may affect cold-water corals (Orr et al. 2005; Guinotte et al. 2006); our habitat modelling study appears to confirm this, but provides new evidence for the spatial heterogeneity of this effect. The niche requirements for cold-water stony corals from all deep-sea environments appear to be very similar to those for seamounts (Chapter 4), although we emphasise that this may not be true at a more localised scale. It is certainly possible that other environmental parameters not included in our analysis may have an additional 141 effect; for instance, the greater prevalence of hard substrate on seamounts appears to often be coupled with the presence of cold-water stony corals (Clark et al. 2006). When species distribution models are used to project into future scenarios, the results must be interpreted with appropriate caution. The most severe limitation is that models are based on the fundamental rather than the realized species niche, and any changes in future biotic interactions (due, for example, to differing species responses to climatic change) will induce errors into model predictions (Guisan & Thuiller 2005). Prediction errors are likely to be linked to the capacity of a species to occupy its fundamental (as opposed to realized) niche. We speculate that, given the longevity of these organisms (coral samples with continuous growth for -50,000 years) (Roberts et al. 2006), it is reasonable to assume that the distribution would approximate the fundamental niche. It can also be problematic when models make predictions outside of the environmental range of the underlying data; in such cases Maxent 'clamps' distributions to the maximum/minimum of the original range as appropriate. Thus, ranges of environmental variables in future scenarios are restricted to the same ranges as model calibration data. Our model shows virtually no evidence of 'clamping'. In a previous study, the performance of Maxent under scenarios of past and future climates has been shown to overlap well with mechanistic models (Hijmans & Graham 2006). Anthropogenic-induced changes in earth's climate system are likely to affect ocean temperature (e.g. Delworth et al. 2002) and salinity (e.g. Clark et al. 2002), as well as chemistry. It is a challenging prospect to accurately predict such potentially abrupt events; we limit this study to the effects of ocean chemistry, whilst recognising that there 142 may be additive or interactive effects between multiple changing environmental parameters. An AUC value of > 0.9 represents very good model discrimination between presence points and random background points (Pearce & Ferrier 2000). This suggests that the model performs consistently well in determining appropriate habitat (Table 5.2), assuming that the input coral data span the true environmental range for this Order. The results, both present day and future, could have management implications for cold-water stony corals and their associated biodiversity. The two largest patches of suitable present-day habitat are located in two global hotspots around New Zealand and in the northern North Atlantic. The patch around New Zealand is almost entirely within that country's exclusive economic zone (EEZ), whereas that in the northern North Atlantic is about 40 percent in EEZs, with the remainder in the high seas. Most of the suitable habitat elsewhere in the global ocean is within coastal EEZs. Thus it would appear that much can be done by individual countries to protect cold-water stony coral habitat from the effects of present day impacts such as trawling, particularly those in the two key regions. The area of habitat predicted to be most negatively affected by changes in ocean chemistry is located in the high seas of the North Atlantic. Habitat suitability in other regions shows a small to moderate decline. We also speculate that protection of suitable future cold-water coral habitat on the high seas may be most effective through the protection of seamounts and elevated features rather than the benthos, due to the shallowing of the aragonite saturation horizon. However, we add four cautionary notes to this interpretation. One: other factors may affect future habitat suitability, including unmeasured parameters and biotic interactions. Furthermore, changes in the emission scenario (e.g., reduction in the input of CO2 into the atmosphere) would also affect predictions. Two: we model the 143 fundamental niche of this taxonomic group, not the realized niche. Thus it is possible for highly suitable habitat not to contain stony corals due to biological interactions or other mechanisms that have prevented this Order from becoming established. Three: the scaling issues inherent in such a large-scale analysis mean that local variation is not captured by the model, and hence fine resolution detail will be obscured. Four: Our models do not include all known locations of these corals, including areas such as the Caribbean, Phillipines, and the Norweigan Coast (Roberts et al. 2006), due to missing environmental data in these regions. These may potentially be important regions for stony cold-water corals. In light of these cautionary notes, we believe that model results need further confirmation through extensive sampling, particularly for those regions in which we have very little data, before they can be considered appropriate for the basis of any management decisions. Model results should not be interpreted without appropriate consideration of the limitations involved in species distribution modelling (e.g. Pearson & Dawson 2003 ; Guissan & Thuiller 2005). We suggest, however, that the model provides a useful first approximation as to the likely distribution of cold-water stony corals, and of the potential effects of changes in ocean carbonate chemistry. We emphasise that the distribution of other coral groups such as Octocorals is likely to be very different, and also requires study. 5.6 Conclusions Modelling habitat suitability for cold-water stony corals based on the database we compiled gives some insight into the likely distribution of this important taxonomic group. Evaluation of our results suggests that our model has high discriminatory 144 capability, based on the environmental parameters and coral records that we have included. The present-day model predictions could be verified by further sampling of the cold-water marine environment. The impacts of changes in ocean acidity on habitat suitability for these organisms vary depending on region, but are substantial for the largest area of suitable habitat, in the upper half of the North Atlantic. The apparently limited spatial distribution of suitable habitat, coupled with the localised predictions of relative decline in this habitat, suggests that a cautious approach is necessary for conservation purposes. The impacts of other anthropogenic activities such as trawling on these corals can also be substantial. A management approach incorporating the likely trajectories off these multiple human effects may be the most viable method of ensuring that cold-water stony corals remain widely distributed in the oceans of our planet. 5.7 Acknowledgements This paper is a joint project of the FMAP and CenSeam programs of the Census of Marine Life (CoML). We thank UNEP-WCMC and the U.K. Department of Trade and Industry (DTI) for making their coral data available to us. We thank the member of the CenSeam Data Analysis Working Group and UNEP for initiating the workshop that lead to this work. We are grateful to J. McPherson for many helpful comments. 145 Chapter 6. Endemism at Low Sampling Effort: Real or Artefact? 6.1 Introduction An endemic species is defined as having a restricted range, typically small in scale. Such species are considered to be more vulnerable to extinction than widely distributed species. This is likely due to the correlation between abundance and distribution (Brown 1995), with small ranges often correlated with low local abundance, and the greater impact of stochastic events on smaller populations. Endemic species are an important component of decision making for conservation purposes, where they are frequently used to index biodiversity (e.g. UNEP 2004), and taken into consideration when siting protected areas (e.g. Kerr 1997 ; Reid 1998). High rates of apparent endemism have been found in some marine habitats and regions, particularly in the deep-sea. For example, the proportion of potential seamount endemics (i.e. species that have not been found on the open sea-floor) may be 29% to 34% for macro- and megafauna on Tasmanian seamounts (Richer de Forges et al. 2001). Yet these environments are often chronically undersampled, with samples that are small in scale (cm ~ m) drawn from populations that can be separated by tens or even hundreds of kilometres. Such samples have been used to generate the controversial estimates of millions of undiscovered species in the deep sea (Grassle & Maciolek 1992 ; May 1992). The traditional view of seamounts, derived from the theory of island biogeography 146 (MacArthur & Wilson, 1967), is that they act as 'stepping stones' for dispersal. Their isolation may lead to reduced gene-flow and elevated levels of endemism. Some recent studies have begun to question these predictions (e.g. O'Hara 2007). Given the importance of endemics in influencing conservation and management decisions, it seems prudent to gain an understanding of how the level of sampling might affect estimates of endemism. In this chapter I created and sampled from simulated communities, and then applied the approach to a deep-sea data set to further explore these issues. 6.2 Models of Endemism and Sampling An obvious first step to understanding the interactions between species richness, relative species abundance, sampling, and endemism is to generate a system with known properties and examine how variation in each of these parameters affects empirical estimates. I simulated a system in which the true number of endemics is zero for this purpose. Consider the situation in which there are two sites, each containing the same s species, with identical distributions of relative species abundances (evenness). A simplifying assumption is that we are dealing with infinite populations, so sampling is with replacement (i.e. the relative proportions of each species do not change as samples are removed). Although this is obviously not true for real biological systems, it is a reasonable approximation when sample sizes are very small relative to total population sizes. I consider endemics to be species that are present at one site only. Given that each site has the same s species, there are no true endemics in this model system. Yet inadequate sampling might imply that endemic species are present if they are only 147 detected in one location, suggesting that they are restricted to that site only. I call such species 'false endemics'. I proceeded by sampling from these two communities and plotting the number of false endemics present at different sampling intensities. In order to simulate the model community I needed to make an assumption about the underlying species abundance distribution. Unfortunately, there is no commonly agreed-upon model for relative species abundances. The most widely used models are the log-normal and log-series (Magurran 2004), both of which have performed well in fiting a wide variety of species abundance patterns found in the real world. Indeed, in many cases, both of these models can fit species-abundance data equally well, although in systems where the mode of the distribution is not in the smallest class the lognormal provides a better fit. Yet many other models have been suggested to be plausible, and fitted to species abundance data (e.g. Harte et al. 1999 ; Magurran 2004). Given this lack of consensus (Magurran 2004), I proceeded by selecting the most appropriate model according to two criteria: (i) the model must be discrete, and (ii) the model must generate data sets that have monotonic changes in evenness. The lognormal fails the first criterion, and the log-series presents difficulties for the second (He & Legendre 2002). One model with the properties that I required is the truncated negative binomial distribution (TNBD) (He & Legendre 2002). This model is specified by the equation _T( + n)f Y 1 r (6.1) '»" n\T(y) ll + J (1 + ^r-l where T is the gamma function, y a shape parameter, individuals, and sn a probability function for the proportion of species with n individuals (Pielou 1975). Evenness strictly decreases with decreasing y, thus satisfying the second criterion. For a community of JV individuals and S species, the parameters of the TNBD are related by the equation W = ^ . (6.2) \-{\ + (pYy s (He & Legendre 2002). By holding N and S constant and varying y (and in consequence (p), it is possible to unambiguously assess the effects of changing evenness. An important consideration is the most appropriate index with which to measure 'evenness'. As with models of relative species abundance, there are numerous possible indices, both from the biological literature and other scientific disciplines (Magurran 2004). I used Pielou's evenness J' (Pielou 1975). This is the most widely used index of evenness in ecology (He & Legendre 2002), and is the Shannon information (equation 4.1, labelled therein as 'information theoretic entropy'), divided by the maximum information. J' is scaled from zero to one, with zero representing the most uneven community and one a perfectly even community in which all species have the same relative abundance. In practice, the minimum value of J' is dependent upon the number of individuals. I used the TNBD to generate relative species abundances for two identical communities each consisting of 200 species (i.e. S = 200) by holding TV fixed at 10,000, setting the number of species with n individuals to be Ssn, varying y between 0 and 10, 149 and using fixed-point iteration to calculate community (for a total of 2a individuals sampled) and calculated the number of false endemics, m, as the number of species present in one sample only. The maximum number of false endemics is min(2a, 2S), when every species is detected at one site only. I used a Monte Carlo procedure, with fifty iterations of each set of sampling. Results for sampling up to 2a = 50,000 individuals from populations with three different levels of evenness are shown in Figure 6.1. Several interesting aspects emerge from this figure. It is clear that both relative species abundances and the number of individuals sampled have an effect on estimates of endemism. The most even community (Fig. 6.1a) has the highest initial estimates of endemism, but more quickly approaches the true value (zero) with increased sampling. The least even community (Fig. 6.1c) initially has a lower estimate of the number of endemics, but requires more sampling to converge to the true value. These patterns are due to the differing rates at which common and rare species are detected in samples. In a relatively even community, many species are encountered rapidly upon sampling, and hence there is a greater chance of finding species at a single location only, leading to high initial estimates of endemism. Relatively quickly, however, these same species are encountered in both locations and the estimates of endemism decline. With a more uneven distribution, a few species are very common and appear in samples rapidly, thus leading to lower initial estimates of endemism as these common species are soon found at both sites. Many species are rare, however, and sampled only infrequently, so there may be a substantial amount of sampling effort required for 150 Figure 6.1: False endemics predicted from sampling two identical communities as described in the text. Values are means from 50 Monte Carlo iterations, a) J' = 0.98. b) J' = 0.87. c) J' = 0.71. The number of individuals sampled from each community ranges from a = 250 to a = 25,000, in increments of 250. A) 50000 B) I 50000 C) 10000 20000 30000 40000 50000 Number of individuals sampled 151 detecting them at both sites. It therefore takes much more sampling before the false estimates of endemism decrease. Another complicating factor is that the number of endemics for a given amount of individuals sampled will depend on the number of species in the community, not just their relative abundances. I examined this effect by following a similar procedure as above, but using identical communities with 50, 100, and 200 species. I used the TNBD to generate patterns of species abundance with varying degrees of evenness for each scenario. The results are shown in Fig. 6.2, with number of endemics converted to percentage of species sampled so that the graphs are comparable. From Fig. 6.2, it can be seen that at equivalent levels of sampling intensity, the percentage of false endemics is equal or greater in communities with more species, when individuals are distributed identically for both sampling sites. The effects of lower evenness are the same as depicted in Fig. 6.1 (lower initial estimates of endemism, more sampling required to converge to true value), though this is more visible for communities with more species. These results suggest that even for moderate levels of sampling (with replacement) it is possible to detect a large proportion of false endemics - species that are sampled at only one site but in reality are present at both. The proportion detected is dependent upon the level of sampling, the relative species abundances, and the number of species. The question naturally arises as to whether it is possible to apply these results to real data. I attempted this in the following section. 152 Figure 6.2 : Percentage of false endemics for differing scenarios of evenness, species richness, and sampling intensity. Lower values of J' indicate a less even distribution of individuals among species. The simulated community consisted of either A) 50 species, B) 100 species, or C) 200 species. Each panel colour-coded from red (maximum percentage of false endemics) to dark blue (zero false endemics). The number of individuals sampled from each community ranges from a = 250 to a = 25,000, in increments of 250. 153 00 Percentage of false endemics Percentage of false endemics Percentage of false endemics o o o o , s s s s a -J t L. ./ / /.. - ./... 6.3 Application of the Method to Data Using real data, two general approaches can be applied to validate the results presented above. One is to fit the data with a statistical model of abundance (such as the TNBD), and then use the sampling process outlined in Section 6.2 to generate a heuristic estimate of the level of sampling necessary to ensure that estimates of endemism are reasonably robust. This assumes that the sample data are distributed in the same way as the community from which they are taken, and that the fitted statistical abundance model is appropriate. The validity of these assumptions is discussed below. The second approach, which builds upon the first, is to follow the same procedure but then to calculate a measure of confidence in the predictions. In this section I consider both of these approaches. 6.3.1 Data The data that I used to test my method were samples of polychaetes from the abyssal Pacific (Glover et al. 2002). Polychaetes often contribute a large fraction of macrofaunal species richness and abundance at abyssal depths (e.g. Glover et al. 2001), and are thus an appropriate taxon for understanding patterns of diversity in this environment. Samples were collected by remotely operated vehicles and box core samplers from the ECHO and PRA sites described in Glover et al. (2002), both situated between 12°N and 15°N, 126°W and 129°W in the central Pacific. The sites were located at 4500 m and 4800 m depth respectively. Fifteen box cores in total were drawn from the ECHO site and sixteen from the PRA site. Samples were washed and then transferred to 70 to 80% ethanol for permanent storage. Some species were previously unknown to 155 science, and these were assigned unique code numbers to ensure that they could be distinguished. 1,552 individuals were collected, and 171 species of polychaetes identified. In order to use the data, I made a number of simplifying assumptions. Firstly, I assumed that sites were sampled with replacement. As mentioned above, this is not strictly true for any biological community, but reasonable in this instance, where abundances are very high at between 150 and 300 individuals per metre squared. Secondly, I assumed that the samples were distributed in the same way as the community from which they were drawn. This is almost certainly not the case; there are likely species at both locations that were not sampled. Although further work remains to be conducted into the effect this may have on my results, I note here that for a particular value of evennessy = J', an increase in the number of species increases the estimate of false endemics (Fig. 6.2). Additional unsampled species in the community from which the sample is drawn would likely change the value of evenness, as J' is dependent upon species richness. Such changes are likely to decrease the value of/', as species that are not detected in the sample would be rare, and thus would decrease the value of this metric. This once more leads to higher estimates of false-endemism at low to moderate levels of sampling (Fig. 6.2). Thus this assumption appears conservative. A further consideration is fitting an appropriate model. In many circumstances, multiple models of relative species abundance can fit data equally well (Magurran, 2004). Figure 6.3 shows the fit of the TNBD to the pooled (from both sites) abyssal data. The fit appears to be good, and both a Kolmogorov-Smirnov test and a chi-squared test confirm this (null hypothesis that both model and data are drawn from the same community not rejected at significance level 0.05). Thus I considered that the TNBD was a reasonable model for this data, though the effects of fitting different models remain to be explored. 156 Figure 6.3: Ranked species abundance pooled from abyssal sites (blue bars) and predicted by the TNBD (red dots). The x-axis has been restricted to 50 individuals for visual clarity. T I I I I I l i 45- 40- - 35- - i i M W Numbe r o f specie s 15- | - 10- Hi - 5- ill " 1 _ n - mini „i i ii.i i•im'••••••••••••1 10 15 20 25 30 35 40 45 50 Number of individuals 6.3.2 The Heuristic Approach The Pielou index for the TNBD model fit to the data above is 0.77.1 thus used the model outlined in Section 6.2 to sample from two identical communities with 171 species, and J' equal to 0.77. The results are shown in Figure 6.4. Fig 6.4 shows that almost fifty false endemics can be detected from two sites with identical species, at the same level of evenness and number of species as the abyssal data. Estimates of endemism remain high (~4 - 48) until at least 10,000 individuals have been sampled. In fact, not until -15,500 individuals have been sampled does the estimate of number of endemics decline below 0.5. This would suggest that for the two abyssal sites estimates of endemism cannot be distinguished from artefacts, given the level of sampling conducted (1,552 individuals). It is of note that the mean number of species sampled does not reach the true value of 171 until around 5,000 individuals have been sampled; this could point to the community from which the abyssal samples are drawn being differently distributed to the samples. 6.2.3 Generating Estimates of Confidence Is there a way in which to have some measure of confidence that sampling is good enough to ensure that estimates of endemism are real and not artefacts? One approach to answering this question might be to generate a margin of error about an estimator ofp, the proportion of species that are true endemics. A difficulty is that, for my simulations, it is known that/? is zero. Acknowledging this problem, the approach that follows may be considered a first attempt at answering this question. 158 Figure 6.4: Mean estimates of false endemics (top) and number of species sampled (bottom) from two identical simulated communities. Both communities are distributed following the TNBD fitted to pooled abyssal data as described in the text. 5000 10000 15000 20000 180 S 160 <" 140 .120 100 10000 15000 20000 Number of individuals sampled Consider the situation of sampling from two identical communities, as in Section 6.2. Label the number of species that have been sampled m, and the number of those species present at only one site - false-endemics - y. Assume that;; given m follows a binomial distribution. The maximum likelihood estimator of/? is then/?', where/*' is ylm, the proportion of endemics observed in the sample. As discussed earlier, for the simulated community the true value of/? is zero, but/?', the maximum likelihood estimator, is the quantity of interest in this case. Assuming that the sampling distribution of the estimator p' is normally distributed (a reasonable assumption at large sample sizes), a margin of error on/?' can be calculated as p'±zjp'(l-p')m (6.3) where z is the appropriate value from the normal distribution for a confidence interval a. In this instance, only the upper margin of error is of interest, in order to generate some measure of certainty in the upper limit of the likely number of false endemics. An acceptable level has to be defined for this margin of error. One option would be for m times this value to be less than one; i.e., an upper margin of error of less than one endemic species at 95% confidence. Of course,/? \y and m all vary with sampling effort. For the TNBD fitted to the abyssal data, the margin of error under this model is shown in Fig. 6.5. This margin gradually decreases as sampling effort increases, and reaches a value of less than one at 18,500 individuals sampled. Thus, under the assumptions of the model listed above, about 3,000 more individuals need to be sampled over and above the heuristic approach to ensure an acceptable margin of error. Figure 6.5: Mean false endemics (black) and margin of error (red) for simulated communities derived from TNBD fitted to abyssal data. 5000 10000 15000 20000 Individuals sampled 161 The validity of assuming thatj|m is binomially distributed, given that the true proportion of endemics is zero, needs further investigation. 6.4 Conclusions and Further Work In this chapter I have presented simulations of limited sampling on two known and identical communities, to estimate rates of false endemism (i.e. species which appear to be in only one community but are actually present in both). These estimates are strongly dependent on sampling intensity, the number of species in the community, and the relative evenness of these species. I explored, under a number of limiting assumptions, the application of this method to data. I fitted a statistical model of abundance to abyssal polychaete data, and used it to estimate the level of sampling required to distinguish true endemics from potential artefacts of sampling. This work by necessity ignores many complicating factors. In reality, sampling is typically undertaken at more than two sites. Exploring the effect that this has upon simulated estimates of endemism would be a useful next step. Similarly, considering sampling without replacement would relax a key assumption of the model and widen its applicability. It is important to note that estimates of endemism at low levels of sampling are not necessarily incorrect, but it seems unclear as to whether they are real or artefacts. Another route through which this model might be extended is to run simulations on communities with known levels of endemism, and see what effect this has on estimates. I currently do not incorporate the spatial distribution of individuals into the model. If the abundance-occupancy relationship (e.g. Freckleton et al. 2005), a common macroecological pattern, holds for seamount communities, then species with lower 162 abundance are also likely to have a lower patch occupancy. Thus these species will generally be less likely to be sampled than if they were uniformly or randomly distributed. This biases the model, resulting in a greater number of individuals needing to be sampled for estimates of false endemism to decline (and thus the current model is conservative in this regard). Incorporating a spatially explicit distribution of individuals within the model would begin to address this issue. Endemic species are evolutionarily and biologically interesting, and are considered to be vulnerable to extinction. Estimates of endemism can be high for habitats that are typically undersampled, such as seamounts (Richer de Forges et al. 2001), the deep sea, and some terrestrial environments such as rainforests. From the simulations I carried out, and the model that I fitted to abyssal data, it appeared that substantial sampling is necessary to have reasonable confidence in predicted levels of endemism. This level of sampling is currently impractical for seamounts and other inaccessible environments, thus raising doubts over present estimates of endemism. The modelling of endemism at low levels of sampling intensity, and the level of sampling necessary to quantify estimates with reasonable precision, could prove to be a useful tool in biodiversity management. It may also be useful in refining estimates of the world's undiscovered diversity, which critically depends on assumptions about the levels of species richness in the deep sea (Grassle & Maciolek 1992). 163 Chapter 7. Conclusions 7.1 Exploitation, Scale and Sampling: Thesis Results In Chapter 1,1 listed the major themes that would be addressed, through a macroecological approach, in this thesis. They were: exploitation, the effects of scale, and the effects of sampling. I will now consider some overarching conclusions that have emerged from my research on each of these issues. Chapter 2 (Macroecological changes in exploited marine systems) was a review of macroecological approaches that have been used to analyse data from exploited marine ecosystems. The chapter was organised by scale, going from individuals, to populations, to species, to ecosystems. The most striking aspect emerging from this review was that the effects of exploitation are visible at all of these scales, from changes in the age at maturity of individual fish to the destabilisation of entire ecosystems and declining biodiversity on an ocean scale (Chapter 2). The statistical analysis of emergent patterns, which I would argue is the basis of the macroecological approach (as opposed to experimental or purely theoretical approaches), has helped to highlight many of these effects. Indeed, I think it is reasonable to suggest that the scale of human impacts upon the oceans was not fully appreciated until the macroecological approach was applied (see for example Myers & Worm 2003 ; Worm et al. 2006). Chapter 3 (Human impacts on the species-area relationship in reef-fish assemblages) explicitly addressed the effects of exploitation on the scaling of biodiversity 164 with increasing sampling area. I found a very consistent pattern, whereby exploitation reduced the rate of increase of reef fish biodiversity with area. The effect appeared to be proportional to the level of fishing intensity. Thus in an exploited region one is less likely to encounter new species as one increases the area of census in comparison to a similar protected area. This work demonstrated a human impact on a macroecological parameter previously considered to be determined purely by ecological and biological processes. It joins other studies that have shown similar human-induced effects on natural parameters in the marine environment, such as the abundance-body mass relationship (Jennings & Blanchard 2004) and the abundance-distribution relationship (Fisher & Frank 2004), among others. The unveiling of broad ecological changes that might otherwise have gone unnoticed is in my view one of the strengths of the macroecological approach, making it relevant to marine conservation and management. Chapter 4 (Predicting global habitat suitability for stony corals on seamounts) considered sampling effort, and how to use limited sampling data to predict species distributions. I constructed habitat suitability models to predict the global distribution of stony corals on seamounts. This distribution appeared to be very heterogeneous, with widely separated patches of suitable habitat occurring in the North Atlantic and in a mid- latitude circum-global strip in the southern hemisphere. The likely factors that emerged as being important in constraining the distribution of corals on seamounts corresponded with current knowledge of the niche requirements of this Order (Clark et al. 2006 ; Roberts et al. 2006), though this needs to be confirmed by further experiments and sampling. An important aspect of this chapter was the comparison of two different presence-only modelling methods for these data. Both models produced similar patterns of habitat suitability. One must always view habitat suitability models with due consideration of 165 inherent limitations and assumptions, but I would view these models as providing a first- order, testable hypothesis as to the distribution of these corals on seamounts. Chapter 5 (Impact of anthropogenic ocean acidification on global cold-water stony coral habitat) used the maximum entropy approach which was shown to have superior discrimination for seamount coral data in Chapter 4.1 increased the scope of that study to include all cold-water coral habitats, an increased depth range, and then predicted the likely effect of human-induced changes in ocean chemistry. Increased ocean acidification (through uptake of anthropogenic CO2) produced effects on habitat suitability that were uniformly negative, but spatially very heterogeneous, with the strongest impact being visible in the North Atlantic. These changes are related to the level of aragonite saturation, which decreases with depth. Negative effects are thus less pronounced on the summits of seamounts, as they penetrate into shallower water than the surrounding benthos. Predicting these impacts could be useful in informing management decisions for this potentially vulnerable environment; I discuss this further below. Chapter 6 (Endemism at low sampling effort: real or artefact?) considered whether estimates of endemism can be over-inflated due to low levels of sampling. This chapter thus addressed both the effects of scale (as endemics are inherently defined by a given spatial scale), and sampling. I used a truncated negative binomial distribution to generate various ranges of evenness for species abundances, and then sampled from two identical distributions to enumerate false endemics. This issue is particularly important in the deep- sea environment, where estimates of endemism are often very high (e.g. Richer de Forges et al. 2001), but sampling is very limited. I then tested this approach on abyssal polychaete data, to predict the number of individuals that must be sampled to ensure that 166 estimates of endemism are robust. There are many additional steps that are necessary before this method becomes a practical tool; I consider them later in this chapter. 7.2 The Macroecological Approach: Strengths and Weaknesses Although there is some confusion as to what constitutes a macroecological approach (Brown 1995 ; Lawton 1999 ; Marquet 2002), I consider the statistical analysis of broad-scale patterns in data to be the core of macroecological work. Conducting ecological research in this manner is an approach which, like any other, has both strengths and weaknesses. By aggregating finer-scale detail, general patterns can emerge that might otherwise not have been visible. These may then shed light on the underlying ecological processes. Furthermore, results can be compared among environments, across scales and among regions. This is a key strength of macroecology in my view; that it can provide generalisation. In some circumstances, macroecological research is the only possible tool, when experiments are impossible to conduct, and theoretical analyses untenable due to a lack of familiarity with underlying ecological processes. I believe the main limitation of macroecology is that it cannot be used to prove causality, since it is not experimental or manipulative in nature. Rather, analyses are performed on observations or samples. We can strengthen confidence in these predictions through meta-analyses, but experiments are ultimately required to prove causation. Every approach in ecology has its weaknesses. Experiments cannot always be conducted, especially at large scales, due to logistics, ethics, or other impracticalities. 167 Theoretical approaches can provide substantial insight into the underpinnings of ecosystems, but also cannot prove causation and may have limited applicability to the real world. Thus I view the above approaches as being complementary, and a balanced research program should include all of them, for an understanding of ecological processes from multiple viewpoints, each depicting the natural world through their own peculiarly shaped lens. Much of the work in my thesis has been at the community scale, and it is here that I disagree with Lawton (1999), who considers that general 'laws' in ecology are most clearly visible at the population level and at large-spatial scales, but obscured at the community level. Some of this disagreement may arise from differences in terminology, as Lawton (1999) clearly considers macroecological studies to be defined by their large spatial or temporal scale. For example, I believe my species-area work, though conducted at the community level, to be macroecological in nature. It clearly reveals human impacts on emergent patterns at the community level, at spatial scales of 1-1000 m. Thus through a generalisation of data, i.e. through combining fine-scale details, patterns arise that can be statistically tested, whatever the spatial scale (see Chapter 2). This raises the important point that the term 'macroecology' needs to be clearly defined, or at least agreed upon. I suggest that the nomenclature has become somewhat confused since the original definition (Brown & Maurer, 1989) was subsequently expanded. 7.3 Management Implications A number of potential management implications emerge from this thesis. These are discussed in the appropriate chapters, but I summarize them again here. Firstly, the 168 twin effects of exploitation and scale need to be considered in concert when assessing biodiversity (Chapter 3). The optimal size of marine protected areas, for example, is a subject of considerable research (e.g. Walters 2000 ; Halpern 2003), and the alteration of the rate of biodiversity accumulation with increasing area could inform such considerations. Changes in the species-area relationship, furthermore, could potentially be used to assess degradation of reef habitats, as a field tool to complement other metrics of biodiversity. Chapter 4 predicts areas likely to contain suitable habitat for Scleractinians on seamounts. These environments are under considerable and increasing threat from the effects of exploitation, ocean acidification, and potentially seabed mining (Clark et al. 2006). Knowledge of the distribution of this Order may help to identify key areas, especially as reef-building species can be associated with elevated levels of biodiversity (Clark et al. 2006). Chapter 5 provides an assessment of cold-water habitat on the sea-floor, and how this may change given future human impacts on ocean chemistry. Knowledge of likely patterns of change can help to maximise returns from resource allocation, for example in the siting of protected areas. As much of the apparently suitable habitat is within exclusive economic zones rather than on the high-seas, the door is open for individual countries to manage these sensitive habitats in such a way as to increase their resilience to climate change (Hughes et al. 2003). Chapter 6, although not directly related to management decisions, may provide a basis for refining current estimates of endemism, a common metric used as the basis for conservation decisions (e.g. Myers et al. 2000 ; UNEP 2004). Although much remains to 169 be done before this approach can be applied, it could prove particularly useful for environments that are commonly undersampled, such as seamounts and abyssal plains. 7.4 Where Next? Considerations for Future Marine Macroecological Research Many options are open for future macroecological research in exploited marine systems. Several ideas arise from the material considered in this thesis. Firstly, the consistency of macroecological patterns shown to have been anthropogenically altered in the marine environment (such as the species-area relationship, or the abundance-body mass relationship) need to be verified through further research. Secondly, the application of species distribution models to the marine realm presents many opportunities. There are many modelling approaches, and in some circumstances marine presence/absence data could be available which would open up a whole new range of possibilities. Nonetheless, other approaches to presence-only data, such as using pseudo-random absence data in a presence-absence approach, remain (as far as I am aware) to be tested for the oceanic realm. The scope for validating these approaches, and applying them to marine systems, is vast. Substantial differences between these environments, such as differing patterns of dispersal in the oceans and their inherently three-dimensional nature, may produce unexpected results. A body of theory must be built up as to the suitability of different approaches in a variety of situations, much as has been acquired in the terrestrial realm. A key issue with regard to the marine environment is the effects of spatial autocorrelation on presence-only models, as much marine data is likely to be presence-only, and there are as yet no methods to incorporate spatial autocorrelation into the presence-only approach (Dormann et al. 2007). 170 Thirdly, much remains to be explored regarding the modelling of endemism and sampling. Key issues are to explore the applicability of the binomial approximation for the proportion of endemics, to expand the simulations to include n > 2 sites, to attempt an analytical solution to the expected proportion of endemics, and to further test abundance model predictions with real data. This fourth point is particularly pressing, as there is no common consensus as to which model of relative abundance fits most community data (Magurran 2004), and indeed several models often fit data equally well. Another avenue that beckons is the further application of macroecological techniques developed for terrestrial systems to the marine environment. The contrasts between these two components of the biosphere are enormous, but this may engender greater learning, through examination of particular differences. Interesting examples include, for example, the latitudinal gradient of biodiversity, with species richness peaking at the equator and dwindling towards the poles as a well-known and apparently robust terrestrial pattern (Gaston 2000). Yet several marine taxa have been shown to diverge from this pattern, such as zooplankton (Rutherford et al. 1999), tuna and billfish (Worm et al. 2005), and marine mammals (K. Kaschner,pers. comm.) where species richness tends to peak at mid-latitudes (20-40 degrees N or S). Syntheses of these patterns, and the search for answers as to why they should be different, will surely be a stimulating topic for future research. Ultimately, ecology, at its core, revolves around the gathering of field data and observations. Enormous amounts of effort are expended to acquire this information. Modelling provides one way in which to maximise the amount of information gleaned from field data. Macroecological modelling, furthermore, allows us to take a particular set of observations or data, consider the effects of scale, examine emergent statistical 171 patterns, and combine it with other data sets in search 01 generalities. This is ultimately very different from an experimental approach. Yet experiments are not always possible, especially at large scales, often due to the inaccessibility of the marine environment and the logistics involved. Perhaps nowhere is this truer than in the deep-sea (Glover & Smith 2003). In such situations, macroecology can offer insight and understanding into the operation of ecological processes, and into the dynamic patterns of biodiversity to which they give rise. Given the growing scope and scale of human impacts upon our oceans, this understanding is urgently needed. 172 References Adler, P. B., White, E. P., Lauenroth, W. K., Kaufman, D. M., Rassweiler, A. & Rusak, J. A. (2005). Evidence for a general species-time-area relationship. Ecology, 86: 2032-2039. Al-Horani, F. A., Tambutte, E. & D. Allemand. (2007). Dark calcification and the daily rhythm of calcification in the scleractinian coral, Galaxeafascicularis. Coral Reefs 26: 531-538. Antonov, R. A., Locarnini, R. A., Boyer, T. P., Mishonov, A. V., & Garcia, H. E. (2006). World Ocean Atlas 2005, Volume 2: Salinity. Ed: S. Levitus. U. S. Government Printing Office, Washington, D. C, USA. Arrhenius O. (1921). Species and area. J. Ecology, 9: 95-99. Auster, P.J., Malatesta, R. J., Langton, R. W., Watling, L., Valentine, P. C., Donaldson, C. L. S., Langton, E. W. , Shepard, A. N. & Babb., I. G. (1996). The impacts of mobile fishing gear on seafloor habitats in the Gulf of Maine (Northwest Atlantic): Implications for conservation offish populations. Rev. Fish. Sci. 4: 185- 202. Barot, S., Heino, M., O'Brien, L. & Dieckmann, U. (2004). Long-term trend in the maturation reaction norm of two cod stocks. Ecol. Appl. 14: 1257-1271. Barrett, J.H., Locker, A. M. & Roberts, C. M. (2004). The origins of intensive marine fishing in medieval Europe: the English evidence. Proc. R. Soc. Lond. B27\: 2417-2421. Baum, J.K. & Myers, R. A. (2004). Shifting baselines and the decline of pelagic sharks in the Gulf of Mexico. Ecol. Lett. 7: 135-145. Baum, J.K., Myers, R. A., Kehler, D. G., Worm, B., Harley, S. J. & Doherty, P. A.. (2003). Collapse and Conservation of Shark Populations in the Northwest Atlantic. Science 299: 389-392. Behrenfeld, M. J., & Falkowski, P. G. (1997). Photosynthetic rates derived from satellite- based chlorophyll concentration. Limnol. Oceanogr., 42: 1-20. Bianchi, G., Gislason, H., Graham, K., Hill, L., Jin, X., Koranteng, K., Manickchand- Heileman, S., Paya, I., Sainsbury, K., Sanchez, F. & Zwanenburg, K. (2000). Impact of fishing on size composition and diversity of demersal fish communities. ICES J. Mar. Sci. 57: 558-571. 173 Boyce, M. S., Vernier, P. R., Nielsen, S. E. & Schmiegelow, F. K. A. (2002). Evaluating resource selection functions. Ecol. Model., 157: 281-300. Boyett, H. V., Bourne, D. G. & Willis, B. L. (2007). Elevated temperature and light enhance progression and spread of black band disease on staghorn corals of the Great Barrier Reef. Mar. Biol, 151: 1711-1720. Brandt, A., Gooday, A. J., Brandao, S. N., Brix, S., Brokeland, W., Cedhagen, T., Choudhury, M., Cornelius, N., Danis, B., De Mesel, I., Diaz, R. J., Gillan, D. C, Ebbe, B., Howe, J. A., Janussen, D., Kaiser, S., Linse, K., Malyutina, M., Pawlowski, J., Paupach, M. & Vanreusel, A. (2007). First insights into the biodiversity and biogeography of the Southern Ocean deep sea. Nature, 447: 307- 311. Brose U., Ostling A., Harrison K. & Martinez, N.D. (2004). Unified spatial scaling of species and their trophic interactions. Nature, 428: 167-171. Brotons, L., W. Thuiller, Araujo, M. B. & Hirzel, A. H. (2004). Presence-absence versus presence-only modelling methods for predicting bird habitat suitability. Ecography, 27: 437-448. Brown, J.H. (1984). On the relationship between abundance and distribution of species. Am. Nat. 124: 255-279. Brown J. H. (1995). Macroecology. University of Chicago Press, Chicago. Brown J. H., Gillooly, J. F., West, G. B. & Savage, V. M. (2003). The next step in macroecology: from general empirical patterns to universal ecological laws. In: Macroecology - concepts and consequences. Eds. Blackburn T. M. & Gaston K. J. Cambridge University Press, Cambridge, U. K. Brown J. H. & Maurer, B. A. (1989). Macroecology: the division of food and space among species on continents. Science, 243: 1145-1150. Bruno, J. F., Petes, L. E., Harvell, C. D. & Hettinger, A. (2003). Nutrient enrichment can increase the severity of coral diseases. Ecol. Lett., 6: 1056-1061. Bruno, J. F., Selig, E. R., Casey, K. S., Page, C. A., Willis, B. L., Harvell, C. D., Sweatman, H. & Melandy, A. M. (2007). Thermal stress and coral cover as drivers of coral disease outbreaks. PLOSBiol, 5: 1220-1227. Bryan, T. L. & Metaxas, A. (2007). Predicting suitable habitat for deep-water gorgonian corals on the Atlantic and Pacific continental margins of North America. Mar. Ecol. Prog. Ser., 330: 113-126. 174 Buhl-Mortensen, L. & Mortensen, P. B. 92005). Distribution and diversity of species associated with deep-sea gorgonian corals off Atlantic Canada. In Cold-water corals and ecosystems. Eds. A. Freiwald, A. & Roberts, J. M. Springer Publishing, Heidelberg, Germany. Carbines, G., Jiang, W. M., & Beentjes, M. P. (2004). The impact of oyster dredging on the growth of blue cod, Parapercis colias, in Foveaux Strait, New Zealand. Aquat. Cons. Mar. Fresh. Ecosys. 14: 491-504. Carton, J. A., Chepurin, G., Cao, X. H. & Giese, B. (2000). A Simple Ocean Data Assimilation analysis of the global upper ocean 1950-95. Part I: Methodology. J. Phys. Oceanogr., 30:294-309. Casey, J.M. & Myers, R. A. (1998). Near extinction of a large, widely distributed fish. Science 281:690-692. Chittaro P.M. (2002). Species-area relationships for coral reef fish assemblages of St. Croix, US Virgin Islands. Mar. Ecol. Prog. Ser., 233, 253-261. Chittibabu C. V. & Parthasarathy N. (2000). Attenuated tree species diversity in human- impacted tropical evergreen forest sites at Kolli hills, Eastern Ghats, India. Biodivers. Conserv.,9, 1493-1519. Christensen, V., Guenette, S., Heymans, J. J., Walters, C. J., Watson, R., Zeller, D. & Pauly, D. (2003). Hundred-year decline of North Atlantic predatory fishes. Fish Fish. 4: 1-24. Clark, M. R. & O'Driscoll, R. (2003). Deepwater fisheries and aspects of their impact on seamount habitat in New Zealand. J. Northwest Atl. Fish. Sci., 31: 441-458. Clark, M. R., Tittensor D., Rogers, A. D., Brewin, P., Schlacher, T., Rowden, A., Stocks, K., & Consalvey, M. (2006). Seamounts, deep-sea corals and fisheries: vulnerability of deep-sea corals to fishing on seamounts beyond areas of national jurisdiction. UNEP-WCMC, Cambridge, U. K. Clark, P. U., Pisias, N. G., Stacker, T. F. & Weaver, A. J. (2002). The role of the thermohaline circulation in abrupt climate change. Nature, 415: 863-869. Coleman F. C. & Williams S. L. (2002). Overexploiting marine ecosystem engineers: potential consequences for biodiversity. Trends. Ecol. Evol., 17: 40-44. Collie, J.S., Escanero, G. A. & Valentine, P. C. (1997). Effects of bottom fishing on the benthic megafauna of Georges Bank. Mar. Ecol. Prog. Ser. 155: 159-172. Connor E.F. & McCoy E.D. (1979). The statistics and biology of the species-area relationship. Am. Nat, 113, 791-833. Conover, D.O. & Munch, S. B. (2002). Sustaining fisheries yields over evolutionary time scales. Science 297: 94-96. Cranfield, H.J., Michael, K. P. & Doonan, I. J. (1999). Changes in the distribution of epifaunal reefs and oysters during 130 years of dredging for oysters in Foveaux Strait, southern New Zealand. Aquat. Cons. Mar. Fresh. Ecosys. 9: 461-483. Crawley M.J. & Harral J.E. (2001). Scale dependence in plant biodiversity. Science, 291, 864-868. Davies, A. J., Roberts, J. M. and Wisshak, M. In press. Predicting suitable habitat for the cold-water reef framework-forming coral Lophelia pertusa (Scleractinia). Deep- Sea Res. I. Dayton, P.K., Tegner, M. J., Edwards, P. B. & Riser, K. L. (1998). Sliding baselines, ghosts, and reduced expectations in kelp forest communities. Ecol. Appl. 8: 309- 322. Death, R. G. (2000). The effect of land use on species area relationships in benthic stream invertebrates. Verh. Int. Ver. Theor. Aug. Limnol, 27: 2519-2522. Delworth, T. L., Stouffer, R. J., Dixon, K. W., Spelman, M. J., Knutson, T. R., Broccoli, A. J., Kuschner, P. J. & Wetherald, R. T. (2002). Review of simulations of climate variability and change with the GFDL R30 coupled climate model. Clim. Dynam., 19: 555-574. DeLong, E. R., DeLong, D. M., & Clarke-Pearson, D. L. (1988). Comparing the areas under two or more correlated receiver operating characteristic curves: a non- parametric approach. Biometrices, 44: 837-845. Dodds, L. A., Roberts, J. M., Taylor, A. C. & Marubini, F. In press. The cold-water coral Lophelia pertusa (Scleractinia) reveals metabolic tolerance to temperature and dissolved oxygen change. J. Exp. Mar. Biol. Ecol. Dormann, C. F., McPherson, J. M., Araujo, M. B., Bivand, R., Bolliger, J., Gudrun, C, Davies, R. G., Hirzel, A., Jetz, W., Kissling, D., Kiihn, I., Ohlemuller, R., Peres- Neto, P. R., Reineking, B., Schrodere, B., Schurr, F. M. & Wilson, R. In press. Methods to account for spatial autocorrelation in the analysis of species distributional data: a review. Ecography. Drakare S., Lennon J.J. & Hillebrand H. (2006). The imprint of the geographical, evolutionary and ecological context on species-area relationships. Ecol. Lett., 9, 215-227. Dudik, M., Phillips, S. J. & Schapire, R. E. (2004). Performance guarantees for regularized maximum entropy density estimation. In: Proceedings of the 17th Annual Conference on Computational Learning Theory. ACM Press, New York. 176 Dulvy, N.K., Freckleton, R. P. & N.V.C. Polunin. (2004). Coral reef cascades and the indirect effects of predator removal by exploitation. Ecol. Lett. 7: 410-416. Dulvy, N.K., Sadovy, Y. & Reynolds, J. D. (2003). Extinction vulnerability in marine populations. Fish Fish. 4: 25-64. Elith, J., Graham, C. H., Anderson, R. P., Dudik, M., Ferrier, S., Guisan, A., Hijmans, R. J., Huettmann, F., Leathwick, J., Lehmann, A., Li, J., Lohmann, L. G., Loiselle, B. A., Manion, G., Moritz, C.,Nakamura, M., Nakazawa, Y., Overton, J. M., Peterson, A. T., Phillips, S. J., Richardson, K., Scachetti-Pereira, R., Schapire, R., Soberon, J., Williams, S., Wisz, M. S. & Zimmermann, N. E. (2006). Novel methods improve prediction of species' distributions from occurrence data. Ecography, 29: 1-23. Engelhard, G.H. & Heino, M. (2004). Maturity changes in Norwegian spring-spawning herring Clupea harengus: compensatory or evolutionary responses? Mar. Ecol. Prog. Ser. 272: 245-256. English, S., Wilkinson, C. & Baker, V. (1997). Survey manual for Tropical Marine Resources (2nd Edition). Australian Institute of Marine Science, Townsville. Ernande, B., Dieckmann, U. & Heino, M. (2004). Adaptive changes in harvested populations: plasticity and evolution of age and size at maturation. Proc. R. Soc. Lond.B 271: 415-423. Estes, J.A., Duggins, D. O. & Rathbun, G. B. (1989). The ecology of extinctions in kelp forest communities. Conserv. Biol. 3: 252-264. Feely, R. A., Sabine, C. L., Lee, K., Berelson, W., Kleypas, J., Fabry, V. J. & Millero, F. J. (2004). Impact of anthropogenic CO2 on the CaC03 system in the oceans. Science, 305: 362-366. Fielding, A. H. & Bell, J. F. (1997). A review of methods for the assessment of prediction errors in conservation presence/absence models. Environ. Conserv., 24: 38-49. Fisher J. A. D. & Frank K. T. (2004). Abundance-distribution relationships and conservation of exploited marine fishes. Mar. Ecol. Prog. Ser., 279, 201-213. Flather, C. H. (1996). Fitting species-accumulation functions and assessing regional land use impacts on avian diversity. J. Biogeogr., 23, 155-168. Fortin, M.-J. & Dale, M. R. T. (2005). Spatial analysis: a guide for ecologists. Cambridge University Press, Cambridge, U. K. Frank, K.T. & Brickman, D. (2000). Allee effects and compensatory population dynamics within a stock complex. Can. J. Fish. Aquat. Sci., 57: 513-517. Frank, K. T., Petrie, B., Choi, J. S. & Leggett, W. S. (2005). Trophic cascades in a formerly cod-dominated ecosystem. Science, 308: 1621-1623. Frederiksen, R., Jensen, A. & Westerberg, H. (1992). The distribution of the scleractinian coral Lophelia pertusa around the Faroe Islands and the relation to internal tidal mixing. Sarsia, 11: 157-171. Freiwald, A. (2002). Reef-forming cold-water corals. In: Ocean Margin Systems. Eds. Wefer, G., Billett, D., Hebbeln, D., Jorgensen, B. B., Schluter, M. & Van Weering, T. Springer-Verlag, Berlin. Freiwald, A., Huhnerbach, V., Lindberg, B., Wilson, J.B. & Campbell, J. (2002). The Sula Reef Complex, Norwegian Shelf. Fades, 47: 179-200. Friedlander, A.M., Boehlert, G. W., Field, M. E., Mason, J. E., Gardner, J. V. & Dartnell, P. (1999). Sidescan-sonar mapping of benthic trawl marks on the shelf and slope off Eureka, California. Fish. Bull, 97: 786-801. Fridley J. D., Peet R. K., Wentworth T. R. & White, P. S. (2005). Connecting fine- and broad-scale species-area relationships of southeastern U. S. flora. Ecology, 86: 1172-1177. Friedlander, A.M. & DeMartini, E. E. (2002). Contrasts in density, size, and biomass of reef fishes between the northwestern and the main Hawaiian islands: the effects of fishing down apex predators. Mar. Ecol. Prog. Ser., 230: 253-264. Froese, D. & Pauly, D. (2006). Fishbase, version (02/2006) [WWW document]. URL http ://www.fishbase.org Gage, J. D. (2004). Diversity in deep-sea benthic macrofauna: the importance of local ecology, the larger scale, history and the Antarctic. Deep-Sea Res. 11,51: 1689- 1708. Gage, J. D. & Tyler, P. A. (1991). Deep-sea biology: a natural history of organisms at the deep-sea floor. Cambridge University Press, Cambridge, U. K. Garcia, H. E., Locarnini, R. A., Boyer, T. P. & Antonov, J. I., (2006a). World Ocean Atlas 2005, Volume 3: Dissolved oxygen, apparent oxygen utilisation, and oxygen saturation. Ed: S. Levitus. U. S. Government Printing Office, Washington, D. C. Garcia, H. E., Locarnini, R. A., Boyer, T. P. & Antonov, J. I. (2006b). World Ocean Atlas 2005, Volume 4: Nutrients (phosphate, nitrate, silicate). Ed: S. Levitus. U. S. Government Printing Office, Washington, D. C. Gaston, K. J. (2000). Global patterns in biodiversity. Nature, 405: 220-227. 178 Gaston, K.J. & Blackburn, T. M. (2000). Pattern and process in macroecology. Blackwell Science, Oxford, U. K. Genin, A., Dayton, P. K., Lonsdale, P. F. & Spiess, F. N. (1986). Corals on seamount peaks provide evidence of current acceleration over deep-sea topography. Nature, 322:59-61. Glover, A. G. & Smith, C. R. (2003). The deep-sea floor ecosystem: current status and prospects of anthropogenic change by the year 2025. Environ. Conserv., 30: 219- 241. Glover, A. G., Paterson, G. L. J., Bett, B., Gage, J., Sibuet, M., Shaeder, M. & Hawkins, L. (2001). Patterns in polychaete abundance and diversity from the Maderia Abyssal Plain, north-east Atlantic. Deep-Sea Res. I, 48: 217-236. Glover, A. G., Smith, C. R., Paterson, G. L. J., Wilson, G. D. F., Hawkins, L. & Sheader, M. (2002). Polychaete species diversity in the central Pacific abyss: local and regional patterns, and relationships with productivity. Mar. Ecol. Prog. Ser., 240: 157-170. Goodwin, N.B., Dulvy, N. K. & Reynolds, J. D. (2005). Macroecology of live-bearing in fishes: latitudinal and depth range comparisons with egg-laying relatives. Oikos, 110:209-218. Grassle, J. F. & Maciolek, N. J. (1992). Deep-sea species richness: regional and local diversity estimates from quantitative bottom samples. Am. Nat., 139: 313-341. Grift, R.E., Rijnsdorp, A. D., Barot, S., Heino M., & Dieckmann, U. (2003). Fisheries- induced trends in reaction norms for maturation in North Sea plaice. Mar. Ecol. Prog. Ser., 257: 247-257. Guinotte, J. M., Orr, J., Cairns, S., Freiwald, A., Morgan, L. & George, R. (2006). Will human induced changes in seawater chemistry alter the distribution of deep-sea scleractinian corals? Front. Ecol. Environ., 4: 141-146. Guisan, A. & Thuiller, W. (2005). Predicting species distribution: offering more than simple habitat models. Ecol. Lett., 8: 993-1009. Guisan, A. & Zimmerman, N. E. (2000). Predictive habitat distribution models in ecology. Ecol. Model, 135: 147-186. Hall-Spencer, J., Allain, V. & Fossa, J. H. (2002). Trawling damage to Northeast Atlantic ancient coral reefs. Proc. R. Soc. Lond. B, 269: 507-511. Hall-Spencer, J. M., Rogers, A. D., Davies, J. & Foggo, A. In press. Historical deep-sea coral distribution on seamounts, oceanic islands and continental shelf slope habitats in the Northeast Atlantic. Bull. Mar. Sci. 179 Hall-Spencer, J., Tasker, M., Christiansen, S., Rogers, S. & Hoydal, K. Subm. The design of Marine Protected Areas on High Seas and territorial waters of Rockall. Halpern, B.S. (2003). The impact of marine reserves: do reserves work and does reserve size mater? Ecol. App., 13: S117-S137. Harte, J., Kinzig A. & Green, J. (1999). Self-similarity in the distribution and abundance of species. Science, 284: 334 - 336. Harte J., Conlisk E., Ostling A., Green J.L. & Smith A.B. (2005). A theory of spatial structure in ecological communities at multiple spatial scales. Ecol. Monogr., 75: 179-197. Hauser, L., Adcock, G. J., Smith, P. J., Ramirez, J. H. B. & Carvalho, G. R. (2002). Loss of microsatellite diversity and low effective population size in an overexploited population of New Zealand snapper (Pagrus auratus). Proc. Nat. Acad. Sci., 99: 11742-11747. He F.L. & Legendre P. (1996). On species-area relations. Am. Nat., 148: 719-737. He F. L. & Legendre P. (2002). Species diversity patterns derived from species-area models. Ecology, 83: 1185-1198. Helly, J. J. & Levin, L. A. (2004). Global distribution of naturally occurring marine hypoxia on continental margins. Deep-Sea Res. I, 51: 1159-1168. Hijmans, R. J. & Graham, C. H. (2006). The ability of climate envelope models to predict the effect of climate change on species distributions. Glob. Change. Biol., 12: 2272-2281. Hirzel, A. H., & Arlettaz, R. (2003). Modeling habitat suitability for complex species distributions by environmental-distance geometric mean. Environ. Manag., 32: 614-623. Hirzel, A. H., Hausser, J., Chessel, D. & Perrin, N. (2002). Ecological-niche factor analysis: How to compute habitat-suitability maps without absence data? Ecology, 83: 2027-2036. Hirzel, A. H., Heifer, V. & Metral, F. (2001). Assessing habitat-suitability models with a virtual species. Ecol. Model, 145:111-121. Hirzel, A. H., Le Lay, G., Heifer, V., Randin, C. & Guisan, A. (2006). Evaluating the ability of habitat suitability models to predict species presences. Ecol. Model, 199: 142-152. Hoyle M. (2004). Causes of the species-area relationship by trophic level in a field-based microecosystem. Proc. R. Soc. Lond. B, 271: 1159-1164. 180 Hughes, T.P., Baird, A. H., Bellwood, D. R., Card, M., Connolly, S. R., Folke, C, Grosberg, R., Hoegh-Guldberg, O., Jackson, J. B. C, Kleypas, J., Lough, J. M., Marshall, P., Nystrom, M., Palumbi, S. R., Pandolfi, J. M., Rosen, B. & Roughgarden, J. (2003). Climate change, human impacts, and the resilience of coral reefs. Science, 301: 929-933. Hutchings, J.A. (2000). Collapse and recovery of marine fishes. Nature, 406: 882-885. Hutchings, J. A. (2004). Evolutionary biology: The cod that got away. Nature, 428: 899- 900. Hutchings, J.A. & Baum, J. K. (2005). Measuring marine fish biodiversity: temporal changes in abundance, life history, and demography. Phil. Trans. Roy. Soc. B., 360:315-338. Hutchings, J.A. & Myers, R. A. (1994). What can be learned from the collapse of a renewable resource? Atlantic cod, Gadus morhua, of Newfoundland and Labrador. Can. J. Fish. Aquat. Set, 51: 2126-2146. Hutchings, J.A. & Reynolds, J. D. (2004). Marine fish population collapses: consequences for recovery and extinction risk. Bioscience, 54: 297-309. Hutchinson G. E. & MacArthur, R. H. (1959). A theoretical ecological model of size distributions among species of animals. Am. Nat., 93: 117-125. Hutchinson, W.F., van Oosterhout, C, Rogers, S. I. & Carvalho, G. R. (2003). Temporal analysis of archived samples indicates marked genetic changes in declining North Sea cod {Gadus morhua). Proc. R. Soc. Lond. B, 270: 2125-2132. Jaynes, E. T. 1957. Information theory and statistical mechanics. Phys. Rev., 106: 620- 630. Key, R. M., Kozyr, A., Sabine, C. L., Lee, K., Wanninkhof, R., Bullister, J. L., Feely, R. A., Millero, F. J., Mordy, C. & Peng, T. H. (2004). A global ocean carbon climatology: Results from Global Data Analysis Project (GLODAP). Glob. Biogeochem. Cycles, 18, doi: 10.1029/2004GB002247 Kitchingman, A. & Lai, S. (2004). Inferences on potential seamount locations from mid- resolution bathymetric data. In: Seamounts: Biodiversity and Fisheries. Eds. Morato, T. Pauly, D., U.B.C. Fisheries Centre., Vancouver, B.C. Kleypas, J. A., Feely, R. A., Fabry, V. J., Langdon, C, Sabine, C. L. & Robbins. L. L. (2006). Impacts of Ocean Acidification on Coral Reefs and Other Marine Calcifiers: A Guide for Future Research, report of a workshop held 18-20 April 2005, St Petersburg, FL. 181 Koslow, J. A., Gowlett-Holmes, K., Lowry, J. K., O'Hara, T., Poore, G. C. B. & Williams, A. (2001). Seamount benthic macrofauna off southern Tasmania: community structure and impacts of trawling. Mar. Ecol. Prog. Ser., 213: 111- 125. Jackson, J.B.C. (2001). What was natural in the coastal oceans? Proc. Natl. Acad. Sci., USA 98: 5411-5418. Jackson, J.B.C, Kirby, M. X., Berger, W. H., Bjorndal, K. A., Botsford, L. W., Bourque, B. J., Bradbury, R. H., Cooke, R., Erlandson, J., Estes, J. A., Hughes, T. P., Kidwell, S., Lange, C. B., Lenihan, H. S., Pandolfi, J. M., Peterson, C. H., Steneck, R. S., Tegner, M. J. & Warner, R. R. (2001). Historical overfishing and the recent collapse of coastal ecosystems. Science, 293: 629-638. James, M.C., Ottensmeyer, C. A. &. Myers, R. A. (2005). Identification of high-use habitat and threats to leatherback sea turtles in northern waters: new directions for conservation. Ecol. Lett, 8: 195-201 Jennings S. & Blanchard J. L. (2004). Fish abundance with no fishing: predictions based on macroecological theory. /. Anim. Ecol, 73: 632-642. Jennings, S., Pinnegar, J. K., Polunin, N. V. C. & Warr, K. J. (2001). Impacts of trawling disturbance on the trophic structure of benthic invertebrate communities. Mar. Ecol. Prog. Ser., 213: 127-142. Jennings, S. & Polunin, J. V. C. (1996). Effects of fishing effort and catch rate upon the structure and biomass of Fijian reef fish communities. J. Appl. Ecol, 33: 400-412. Jones, C. G., Lawton, J. H., Shachak, M. (1994). Organisms as ecosystem engineers. Oikos, 69: 373-386. Kerr, J. T. (1997). Species richness, endemism, and the choice of areas for conservation. Conserv. Biol, 11: 1094-110. Kleypas, J. A., Buddemeier, R. W., Archer, D., Gattuso, J.-P., Langdon, C. & Opdyke, B. N. 1999. Geochemical consequences of increased atmospheric carbon dioxide on coral reefs. Science, 284: 118-120. Kirby, M.X. (2004). Fishing down the coast: Historical expansion and collapse of oyster fisheries along continental margins. Proc. Nat. Acad. Sci. USA, 101: 13096- 13099. Kock, K.-H. (1992). Antarctic fish and fisheries. Cambridge University Press, Cambridge, U. K. Kondoh, M. (2003). Foraging adaptation and the relationship between food-web complexity and stability. Science, 299: 1388-1391. 182 Koslow, J.A., Boehlert, G. W., Gordon, J. D. M., Haedrich, R. L., Lorance, P.& Parin, N. (2000). Continental slope and deep-sea fisheries: implications for a fragile ecosystem. ICES J. Mar. Sci., 57: 548-557. Kulbicki M. (1998). How the acquired behaviour of commercial reef fishes may influence the results obtained from visual censuses. J. Exp. Mar. Biol. Ecol, 222: 11-30. Langdon, C. & Atkinson, M. J. (2005). Effect of elevated pCC>2 on photosynthesis and calcification of corals and interactions with seasonal change in temperature/irradiance and nutrient enrichment. /. Geophys. Res. 110: C09S07, doi: 10.1029/2004JC002576 Law, R. (2000). Fishing, selection, and phenotypic evolution. ICES J. Mar. Set, 57: 659- 668. Lawton, J. H. (1999). Are there general laws in ecology? Oikos, 84: 177-192. Leclercq, N., Gattuso, J.-P. & Jaubert, J. (2000). CO2 partial pressure controls the calcification rate of a coral community. Glob. Change Biol, 6: 329-334. Legendre, P. (1993). Spatial autocorrelation: trouble of new paradigm? Ecology, 1A: 1659-1673. Leitner W.A. & Rosenzweig M.L. (1997). Nested species-area curves and stochastic sampling: a new theory. Oikos, 79: 503-512. Lewison, R.L., Crowder, L. B, Read, A. J. & Freeman. S. A. (2004a). Understanding impacts of fisheries bycatch on marine megafauna. Trends Ecol. Evol., 19: 598- 604. Lewison, R.L., Freeman, S. A. & Crowder, L. B. (2004b). Quantifying the effects of fisheries on threatened species: the impact of pelagic longlines on loggerhead and leatherback sea turtles. Ecol. Lett. ,7:221-231. Liermann, M. & Hilborn, R. (1997). Depensation in fish stocks: a hierarchic Bayesian meta-analysis. Can. J. Fish. Aquat. Sci., 54: 1976-1984. Locarnini, R. A., Mishonov, A. V., Antonov, J. I., Boyer, T. P. & Garcia, H. E. (2006). World Ocean Atlas 2005, Volume 1: Temperature. U. S. Government Printing Office, Washington, D. C. Lomolino M.V. (2000). Ecology's most general, yet protean pattern: the species-area relationship. J. Biogeogr., 27: 17-26. Lomolino M.V. & Weiser M.D. (2001). Towards a more general species-area relationship: diversity on all islands, great and small. J. Biogeogr., 28: 431-445. 183 Lotze, H.K. & Milewski, I. (2004). Two centuries of multiple human impacts and successive changes in a North Atlantic food web. Ecol. Appl, 14: 1428-1447. MacArthur R. H. & Wilson, E. O. (1967). The Theory of Island Biogeography. Princeton University Press, Princeton. Macpherson, E., Hastings, P. A. & Robertson, D. R. In press. Macroecological patterns among marine fishes. In Marine Macroecology. Eds. Witman, J. D. & Roy, K. University of Chicago Press, Chicago. Magurran A.E. (2004). Measuring biological diversity. Blackwell Science, Oxford. Marquet, P. A. (2002). The search for general principles in ecology. Nature, 418: 723. Marubini, F. & Thake, B. (1999). Bicarbonate addition promotes coral growth. Limnol. Oceanogr., 44: 716-720. Martin, H. G. & Goldenfeld, N. (2006). On the origin and robustness of power-law species-area relationships in ecology. Proc. Natl. Acad. Sci. USA, 103: 10310- 10315. May, R. M. (1992). Bottoms up for the oceans. Nature, 357: 278-279. McCann, K., Hastings, A. & Huxel, G. R. (1998). Weak trophic interactions and the balance of nature. Nature, 395: 794-798. McClanahan T.R. (1994). Kenyan coral-reef lagoon fish - effects of fishing, substrate complexity, and sea-urchins. Coral Reefs, 13: 231-241. McClanahan, T. R. & Muthiga, N. A. (1998). An ecological shift in a remote coral atoll of Belize over 25 years. Environ. Conserv., 25: 122-130. McClanahan, T. R., Muthiga, N. A., Kamukuru, A. T., Machano, H. & Kiambo, R. W. (1999). The effects of marine parks and fishing on coral reefs of northern Tanzania. Biol. Conserv., 89: 161-182. McGrath, P.T. (1911). Newfoundland in 1911. Whitehead, Morris & Co., London, U.K. Micheli F., Halpern, B. S., Botsford, L. W. & Warner, R. R. (2004). Trajectories and correlates of community changes in no-take marine reserves. Ecol. Appl., 14: 1709-1723. Micheli F., Benedetti-Cecchi L., Gambaccini S., Bertocci I., Borsini C, Osio G.C. & Roman F. (2005). Cascading human impacts, marine protected areas, and the structure of Mediterranean reef assemblages. Ecol. Monogr., 75: 81-102. 184 Micheli F. & Halpern B.S. (2005). Low functional redundancy in coastal marine assemblages. Ecol. Lett., 8: 391-400. Milliken G.A. & Johnson D.E. (1992). Analysis of Messy Data. Volume 1: Designed Experiments. Chapman & Hall, New York. Moran, M.J. & P. C. Stephenson. (2000). Effects of otter trawling on macrobenthos and management of demersal scalefish fisheries on the continental shelf of north western Australia. ICES J. Mar. Set, 57: 510-516. Munday, P.L. & Jones, G. P. (1998). The ecological implications of small body size among coral reef fishes. Oceanogr. Mar. Biol., 36: 373-411. Musick, J. A. (1999). Life in the slow lane: ecology and conservation of long-lived marine animals. American Fisheries Society, Bethesda, Maryland. Myers, N., Mittermeier, R. A., Mittermeier, C. G., da Fonseca, G. A. B. & Kent, J. (2000). Biodiversity hotspots for conservation priorities. Nature, 403: 853-858. Myers, R.A., Barrowman, N. J., Hutchings, J. A. & Rosenberg, A. A. (1995). Population dynamics of exploited fish stocks at low population levels. Science, 269: 1106- 1108. Myers, R.A., Hutchings, J. A. Barrowman, N. J. (1997). Why do fish stocks collapse? The example of cod in Atlantic Canada. Ecol. Appl, 7: 91-106. Myers, R.A. & Stokes, K. (1989). Density dependent habitat utilization of groundfish and the improvement of research surveys. ICES CM 1989/D 15. Myers R.A. & Worm B. (2003). Rapid worldwide depletion of predatory fish communities. Nature, 423: 280-283. Myers, R. A. & Worm, B. (2005). Extinction, survival or recovery of large predatory fishes. Proc. R. Soc. Lond. B, 360: 13-20. Nee S. (2002). Biodiversity: thinking big in ecology. Nature, 417: 229-230. Ney-Nifle M. & Mangel M. (1999). Species-area curves based on geographic range and occupancy. J. Theor. Biol, 196: 327-342. Nystrom M. & Folke C. (2001). Spatial resilience of coral reefs. Ecosystems, 4: 406-417. O'Hara, T. D. (2007). Seamounts: centres of endemism or species richness for ophiuroids? Glob. Ecol. Biogeogr., doi: 10.1111/j.l466-8238.2007.00329.x 185 Olsen, E.M., Heino, M., Lilly, G. R., Morgan, M. J., Brattey, J., Ernande, B. & Dieckmann, U. (2004). Maturation trends indicative of rapid evolution preceded the collapse of northern cod. Nature, 428: 932-935. Orr, J. C, Fabry, V. J., Aumont, O., Bopp, L., Doney, S. C, Feely, R. A., Gnanadesikan, A., Gruber, N., Ishida, A., Joos, F., Key, R. M., Lindsay, K., Maier-Reimer, E., Matear, R., Monfray, P., Mouchet, A., Najjar, R. G., Plattner, G.-K., Rodgers, K. B., Sabine, C. L., Sarmiento, J. L., Schlitzer, R., Slater, R. D., Totterdell, I. J., Weirig, M.-F., Yamanaka, Y. & Yool, A. (2005). Anthropogenic ocean acidification over the twenty-first century and its impact on calcifying organisms. Nature, 437: 681-686. Palumbi, S.R. (2001). Humans as the world's greatest evolutionary force. Science, 293: 1786-1790. Pandolfi, J.M., Bradbury, R. H., Sala, E., Hughes, T. P., Bjorndal, K. A., Cooke, R. G., McArdle, D. McClenachan, L., Newman, M. J. H., Paredes, G, Warner, R. R. & Jackson, J. B. C. (2003). Global trajectories of the long-term decline of coral reef ecosystems. Science, 301: 955-958. Pauly, D. (1995). Anecdotes and the shifting baseline syndrome of fisheries. Trends Ecol. Evol., 10:430. Pauly, D., Christensen, V., Dalsgaard, J., Froese, R. & Torres, F. (1998). Fishing down marine food webs. Science, 279: 860-863. Pauly D., Christensen V., Guenette S., Pitcher T.J., Sumaila U.R., Walters C.J., Watson R. & Zeller D. (2002). Towards sustainability in world fisheries. Nature, 418: 689-695. Pauly, D., Palomares, M. L., Froese, R, Sa-a, P., Vakily, M., Preikshot, D. & Wallace, S. (2001). Fishing down Canadian aquatic food webs. Can. J. Fish. Aquat. Sci., 58: 51-62. Pearce, J. & Ferrier, S. (2000). Evaluating the predictive performance of habitat models developed using logistic regression. Ecol. Model., 133: 225-245. Peterson, C.H. & Estes, J. A. (2001). Conservation and management of marine communities. In: Marine community ecology. Eds. Bertness, M. D., Gaines, S. D. & Hay, M. E. Sinauer Associates, Sunderland, MA. Phillips, S. J., Anderson, R. P., & Schapire, R. E. (2006). Maximum entropy modeling of species geographic distributions. Ecol. Model, 190: 231-259. Picard N., Karembe M. & Birnbaum P. (2004). Species-area curve and spatial pattern. Ecoscience, 11: 45-54. 186 Pielou, E. C. (1975). Ecological Diversity. Wiley, New York. Pimm S.L. & Raven P. (2000). Biodiversity - Extinction by numbers. Nature, 403: 843- 845. Pinheiro J.C. & Bates D. M. (2000). Mixed-effects models in S and S-Plus. Springer- Verlag, New York. Pinnegar, J.K., Polunin, N. V. C, Francour, P., Badalamenti, F., Chemello, R., Harmelin- Vivien, M. L., Hereu, B., Milazzo, M., Zabala, M., D'Anna, G. & Pipitone, C. (2000). Trophic cascades in benthic marine ecosystems: lessons for fisheries and protected-area management. Environ. Conserv.,21: 179-200. Reddy, M. S. & Parthasarathy N. (2003). Liana diversity and distribution in four tropical dry evergreen forests on the Coromandel coast of south India. Biodivers. Conserv., 12: 1609-1627. Reid, W. V. (1998). Biodiversity hotspots. Trends. Ecol. Evol, 13: 275-280. Renegar, D. A. & Riegl, B. M. (2005). Effect of nutrient enrichment and elevated CO2 partial pressure on growth rate of Atlantic scleractinian coral Acropora cervicornis. Mar. Ecol. Prog. Ser., 293: 69-76. Richer de Forges, B., Koslow, J. A. & Poore, G. C. B. (2000). Diversity and endemism of the benthic seamount fauna in the southwest Pacific. Nature, 405: 944-947. Ricker, W.E. (1958). Maximum sustained yields from fluctuating environments and mixed stocks. J. Fish. Res. Board Can., 15: 991-1006. Roberts CM. (1995). Effects of fishing on the ecosystem structure of coral reefs. Conserv. Biol., 9: 988-995. Roberts, CM. (2002). Deep impact: the rising toll of fishing in the deep sea. Trends Ecol. Evol., 17: 242-245. Roberts, CM. & Hawkins, J. P. (1999). Extinction risk in the sea. Trends Ecol. Evol., 14: 241-246. Roberts, CM., Hawkins, J. P. & Gell, F. R. (2005). The role of marine reserves in achieving sustainable fisheries. Philos. T. Roy. Soc. B., 360: 123-132. Roberts, J. M., Wheeler, A. J. & Freiwald, A. (2006). Reefs of the deep: the biology and geology of cold-water coral ecosystems. Science, 312: 543-547. Roff, D.A. (2002). Life history evolution. Sinauer Associates, Sunderland, MA. Rogers, A. D. (1994). The biology of seamounts. Adv. Mar. Biol, 30: 305-350. 187 Rogers, A. D. (1999). The biology of Lopheliapertusa (Linnaeus 1758) and other deep- water reef-forming corals and impacts from human activities. Int. Rev. Hydrobiol, 84:315-406. Rogers, A. D., Baco-Taylor, A., Griffiths, H. & J. M. Hall-Spencer. In press. Corals on seamounts. In: Seamounts: ecology, fisheries and conservation. Eds. Pitcher, T. J., Hart, P. J. B., Morato, T., Santos, R. & Clark, M. Blackwell Scientific, Oxford U.K. Roman, J. & Palumbi, S. R. (2003). Whales before whaling in the North Atlantic. Science, 301:508-510. Rosenzweig M.L. (1995). Species diversity in space and time. Cambridge University Press, Cambridge, U. K. Rutherford, S., D'Hondt, S. & Prell, W. (1999). Environmental controls on the geographic distribution of zooplankton diversity. Nature, 400: 740-753. Ruzzante, D.E., Taggart, C. T., Doyle, R. W. & Cook, D. (2001). Stability in the historical pattern of genetic structure of Newfoundland cod (Gadus morhua) despite the catastrophic decline in population size from 1964 to 1994. Conserv. Genet., 2: 257-269. Sadovy, Y. & Cheung, W. L. (2003). Near extinction of a highly fecund fish: the one that nearly got away. Fish Fish., 4: 86-99. Samadi, S., Bottan, L., Macpherson, E., Richer De Forges, B., Boisselier, M.-C. (2006). Seamount endemism questioned by the geographic distribution and population genetic structure of marine invertebrates. Mar. Biol., 149: 1463-1475. Scheiner, S. M. (2004) A melange of curves - further dialogue about species-area relationships. Glob. Ecol. Biogeogr., 13: 479-484. Schratzberger, M., Dinmore, T. A. & Jennings, S. (2002). Impacts of trawling on the diversity, biomass and structure of meiofauna assemblages. Mar. Biol. 140: 83-93. Schratzberger, M. & Jennings, S. (2002). Impacts of chronic trawling disturbance on meiofaunal communities. Mar. Biol., 141: 991-1000. Schroder-Ritzrau, A., Freiwald, A. & Mangini, A. (2005). U/Th-dating of deep-water corals from the eastern North Atlantic and the western Mediterranean Sea. In: Cold-water corals and ecosystems. Springer, Berlin. Segurado, P. & Araiijo, M. B. (2004). An evaluation of methods for modelling species distributions. J. Biogeogr., 31: 1555-1568. 188 Shannon, C. E. (1948). A mathematical theory of communication. Bell System Technical Journal, 27: 379-423, 623-656. Simenstad, C.A., Estes, J. A. & Kenyon, K. W. (1978). Aleuts, sea otters, and alternate stable-state communities. Science, 200: 403-411. Smith, I. (2005). Retreat and resilience: fur seals and human settlement in New Zealand. In The exploitation and cultural importance of sea mammals. Ed. Monks, G. G. Oxbow Books, Oxford, U.K. Smith, K.F. & Brown, J. H. (2002). Patterns of diversity, depth range and body size among pelagic fishes along a gradient of depth. Glob. Ecol. Biogeogr., 11: 313- 322. Snelgrove, P. R. & Smith, C. R. (2002). A riot of species in an environmental calm: the paradox of the species-rich deep-sea floor. Oceanogr. Mar. Biol. Ann. Rev. 40: 311-342. Sokal, R. R. & Rohlf, F. J. (1995). Biometry, 3rd edition. W. H. Freeman, New York. Spotila, J.R., Reina, R. D., Steyermark A. C, Plotkin, P. T. & Paladino, F. V. (2000). Pacific leatherback turtles face extinction. Nature, 405: 529-530. Springer, A.M., Estes, J. A. van Vliet, G. B., Williams, T. M., Doak, D. F., Danner, E. M., Forney, K. A. & Pfister, B. (2003). Sequential megafaunal collapse in the North Pacific Ocean: An ongoing legacy of industrial whaling? Proc. Nat. Acad. Sci. USA, 100: 12223-12228. Steneck, R.S., Graham, M. H., Bourque, B. J., Corbett, D., Erlandson, J. M., Estes, J. A. & Tegner, M. J. (2002). Kelp forest ecosystems: biodiversity, stability, resilience and future. Environ. Conserv., 29: 436-459. Stockwell, C.A., Hendry, A. P. & Kinnison, M. T. (2003). Contemporary evolution meets conservation biology. Trends Ecol. Evol, 18: 94-101. Stokes, K. & Law, R. (2000). Fishing as an evolutionary force. Mar. Ecol. Prog. Ser., 208: 307-309. Stevenson C, Katz L. S., Micheli F., Block B., Heiman K. W., Perle C, Weng K., Dunbar R. & Witting J. (2007). High apex predator biomass on remote Pacific islands. Coral Reefs, 26: 47-51. Tasker, M.L., (Kees)Camphuysen, M C. J., Cooper, J., Garthe, S., Montevecchi, W. A. & Blaber, S. M. (2000). The impacts of fishing on marine birds. ICES J. Mar. Sci., 57:531-547. 189 Tegner, MJ. & P.K. Dayton. (1999). Ecosystem effects of fishing. Trends Ecol. Evol, 14: 261-262. Tegner, M. J. & Dayton, P. K. (2000). Ecosystem effects of fishing in kelp forest communities. ICES J. Mar. Sci., 57: 579-589. Thiem, 0., Ravagnan, E., Fossa, J. H. and J. Berntsen. (2006). Food supply mechanisms for cold-water corals along a continental shelf edge. J. Mar. Sys., 60: 207-219. Thomas CD., Cameron A., Green R.E., Bakkenes M., Beaumont L.J., Collingham Y.C., Erasmus B.F.N., de Siqueira M.F., Grainger A., Hannah L., Hughes L., Huntley B., van Jaarsveld A.S., Midgley G.F., Miles L., Ortega-Huerta M.A., Peterson A.T., Phillips O.L. & Williams S.E. (2004). Extinction risk from climate change. Nature, All: 145-148. Thompson, H. (1943). A biological and economic study of cod (Gadus callarias, L.) in the Newfoundland area, including Labrador. Newfoundland Department of Natural Resources Fisheries Bull, 14: 1-160. Thompson, J.N. (1998). Rapid evolution as an ecological process. Trends. Ecol. Evol, 13: 329-332. Tittensor, D.P., Micheli, F. Nystrom, M. & Worm, B. (2007). Human impacts on the species-area relationship in reef fish assemblages. Ecol. Lett., 10: 760-772. Tjorve E. (2003). Shapes and functions of species-area curves: a review of possible models. J. Biogeogr., 30: 827-835. Tudela, S., Kai Kai, A., Maynou, F., El Andalossi, M. & Guglielmi, P. (2005). Driftnet fishing and biodiversity conservation: the case study of the large-scale Moroccan driftnet fleet operating in the Alboran Sea (SW Mediterranean). Biol. Conserv., 121:65-78. Turley, C. M., Roberts, J. M. and J. M. Guinotte. (2007). Corals in deep-water: will the unseen hand of ocean acidification destroy cold-water ecosystems? Coral Reefs, 26: 445-448. Tynan, C. T., Ainley, D. G., Barth, J. A., Cowles, T. J., Pierce, S. D. and L. B Spear. (2005). Cetacean distributions relative to ocean processes in the northern Californian Current System. Deep-Sea Res. II, 52: 145-167. UNEP (United Nations Environmental Programme). (2004). Provisional global indicators for assessing progress towards the 2010 biodiversity target. CBD COP Document UNEP/CBD/COP/7/INF/33. Nairobi, Kenya. 190 UN General Assembly (2006) Resolution adopted by the General Assembly. 60/30. Oceans and the law of the sea. Available at: http://daccessdds.un.org/doc/UNDOC/GEN/N05/489/34/PDF/N0548934.pdf7Ope nElement Walsh, M.R., Munch, S. B., Chiba, S. & Conover, D. O. (2006). Maladaptive changes in multiple traits caused by fishing: impediments to population recovery. Ecol. Lett., 9: 142-148. Walters, C. (2000). Impacts of dispersal, ecological interactions, and fishing effort dynamics on efficacy of marine protected areas: how large should protected areas be? Bull. Mar. Set, 66: 745-757. Ward, P. & Myers., R. A. (2005). Shifts in open-ocean fish communities coinciding with the commencement of commercial fishing. Ecology, 86: 835-847. Wessel, P. (2001). Global distribution of seamounts inferred from gridded Geosat/ERS-1 altimitry. J. Geophys. Res. B, 106:19,431 - 19,441. Williamson M., Gaston K.J. & Lonsdale W.M. (2001). The species-area relationship does not have an asymptote! J. Biogeogr., 28: 827-830. Wiley, E. O., McNyset, K. M., Peterson, A. T., Robins, C. R., Stewart, A. M. (2003). Niche modeling and geographic range predictions in the marine environment using a machine-learning algorithm. Oceanography 16: 120-127. Witman, J. & Roy, K. In press. Marine Macroecology. University of Chicago Press, Chicago. Witman, J.D. & Sebens, K. P. (1992). Regional variation in fish predation intensity: a historical perspective in the Gulf of Maine. Oecologia, 90: 305-315. Worm, B. & Myers, R. A. (2003). Meta-analysis of cod-shrimp interactions reveals top- down control in oceanic food webs. Ecology, 84: 162-173. Worm B., Barbier E.B., Beaumont N., Duffy J.E., Folke C, Halpern B.S., Jackson J.B.C., Lotze H.K., Micheli F., Palumbi S.R., Sala E., Selkoe K., Stachowicz J.J. & Watson R. (2006). Impacts of biodiversity loss on ocean ecosystem services. Science, 314: 787-790. Worm, B., Lotze, H. K. & Myers, R. A. (2005a). Ecosystem effects of fishing and whaling in the North Pacific and Atlantic Ocean. In: Whales, Whaling and Ocean Ecosystems. Eds. Estes, J. A., Brownell, R. L., DeMaster, D. P., Doak, D. F. & Williams, T. M. University of California Press, Berkley, CA. Worm B., Sandow M., Oschlies A., Lotze H.K. & Myers R.A. (2005b). Global patterns of predator diversity in the open oceans. Science, 309: 1365-1369. 191 Zeebe, R. E. & Wolf-Gladrow, D. A. (2001). CO2 in seawater: equilibrium, kinetics, isotopes. Elsevier, New York. Zweig, M. H., and G. Campbell. 1993. Receiver-Operating Characteristic (Roc) Plots - a Fundamental Evaluation Tool in Clinical Medicine. Clinical Chemistry, 39: 561- 577. Appendix A: Species List for Species-Area Reef Study Table Al: Species list for all study locations in the species-area relationship reef study (Chapter 3). Columns labelled '+/-' indicate normalised abundance change in fished region relative to unfished; + indicates increase in species abundance, - indicates decrease, and • indicates no change. Wherever possible, common names for species follow FishBase nomenclature (Froese & Pauly, 2006). Atl. Ind. Pac. Med Family Species Common name +/- +/- +/- +/- Acanthuridae Acanthurus bahianus Ocean surgeonfish Acanthuridae Acanthurus chirurgus Doctorfish Acanthuridae Acanthurus coeruleus Blue tang Acanthuridae Acanthurus leucosternon Powderblue surgeonfish + Acanthuridae Acanthurus lineatus Lined surgeonfish - Acanthuridae Acanthurus mata Elongate surgeonfish - Acanthuridae Acanthurus nigricauda Epaulette surgeonfish - Acanthuridae Acanthurus nigrofuscus Brown surgeonfish + Acanthuridae Ctenochaetus binotatus Twospot surgeonfish - Acanthuridae Ctenochaetus striatus Striated surgeonfish - Acanthuridae Ctenochaetus strigosus Spotted surgeonfish - Acanthuridae Naso annulatus Whitemargin unicornfish Acanthuridae Naso lituratus Orangespine unicornfish Acanthuridae Naso vlamingii Bignose unicornfish - Acanthuridae sp. 1 - + Acanthuridae sp. 2 - + Acanthuridae Zebrasoma desjardinii Desjardin's sailfin tang - Acanthuridae Zebrasoma scopes Twotone tang - Acanthuridae - Apogonidae Apogon imberbis Cardinal fish + Apogonidae Apogon leptacanthus Threadfin cardinalfish + Apogonidae Archamia fucata Orangelined cardinalfish Apogonidae Cheilodipterus arabicus Tiger cardinal + Apogonidae Cheilodipterus artus Wolf cardinalfish + Atl. Ind. Pac. Med Family Species Common name +/- +/- +/- +/- Apogonidae Cheilodipterus macrodon Large toothed cardinalfish + Apogonidae Cheilodipterus Five-lined cardinalfish quinquelineatus + Atherinidae Atherina sp. - + Aulostomidae Aulostomus chinensis Chinese trumpetfish - Aulostomidae Aulostomus maculatus Trumpetfish + Balistidae Balistapus undulatus Orange-lined triggerfish Balistidae Balistes vetula Queen triggerfish Balistidae Balistoides viridescens Titan triggerfish Balistidae Melichthys niger Black triggerfish Balistidae Sufflamen chrysopterus Halfmoon triggerfish Balistidae Belonidae / Syngnathidae Blenniidae Cirripectes sp. Blenniidae Meiacanthus mossambicus Mozambique fangblenny Blenniidae Ophioblennius macclurei Redlip blenny Blenniidae Parablennius rouxi Blenniidae Plagiotremus Bluestriped fangblenny rhinorhynchus Blenniidae sp. 1 Caesionidae Caesio caerulaurea Blue and gold fusilier Caesionidae Caesio lunaris Lunar fusilier Caesionidae Pterocaesio marri Marr's fusilier Caesionidae Pterocaesio pisang Banana fusilier Caesionidae Carangidae Caranx ruber Bar jack Carangidae Seriola dumerili Forkbeard Carangidae Carcharhinidae Centracanthidae Spicara flexuosa Blotched picarel Centracanthidae Spicara maena Blotched picarel Centracanthidae Spicara smarts Picarel Chaenopsidae Emblemariopsis sp. Darkheaded blenny Chaenopsidae Lucayablennius zingaro Arrow blenny Chaetodontidae Chaetodon auriga Threadfin butterflyfish Chaetodontidae Chaetodon bennetti Bluelashed butterflyfish Chaetodontidae Chaetodon falcula Blackwedged butterflyfish Chaetodontidae Chaetodon guttatissimus Peppered butterflyfish Chaetodontidae Chaetodon kleinii Sunburst butterflyfish + Chaetodontidae Chaetodon lineolatus Lined butterflyfish + Chaetodontidae Chaetodon lunula Raccoon butterflyfish + Chaetodontidae Chaetodon melannotus Blackback butterflyfish + Chaetodontidae Chaetodon ocapistratus Foureye butterflyfish + 194 Atl. Ind. Pac. Med Family Species Common name +/- +/- +/- +/- Chaetodontidae Chaetodon ocellatus Spotfm butterflyfish + Chaetodontidae Chaetodon striatus Banded butterfyfish • Chaetodontidae Chaetodon trifascialis Chevron butterflyfish - Chaetodontidae Chaetodon trifasciatus Melon butterflyfish + Chaetodontidae Chaetodon Zanzibar butterflyfish zanzibariensis + Chaetodontidae Heniochus acuminatus Pennant coralfish - Chaetodontidae Heniochus monoceros Masked bannerfish - Chaetodontidae Cirrhitidae Paracirrhites arcatus Arc-eye hawkfish Cirrhitidae Paracirrhites forsteri Blackside hawkfish Clupeidae Herklo tsich thys Bluestripe herring quadrimaculatus Clupeidae Spratelloides delicatulus Delicate round herring + Congridae Conger conger European conger + Dasyatidae Taeniura lymma Bluespotted ribbontail ray Echeneidae Echeneis naucrates Live sharksucker Ephippidae Platax orbicularis Orbicular batfish Gobiidae Unknown goby Gobiidae Amblygobius hectori Hector's goby Gobiidae Coryphopterus Bridled goby glaucofraenum + Gobiidae Coryphopterus lipernes Peppermint goby + Gobiidae Coryphopterus sp. Masked / Glass goby + Gobiidae Cryptocentrus strigilliceps Target shrimp goby Gobiidae Gobiosoma genie Cleaning goby Gobiidae Gobiosoma oceanops Neon goby Gobiidae Gobiosoma prochilos Broadstripe goby Giant goby Gobiidae Gobius cobitis + Red-mouthed goby Gobiidae Gobius cruentatus + Gobiidae Gobius geniporus Slender goby Golden goby Gobiidae Gobius luteus + Black goby Gobiidae Gobius niger + Gobiidae sp. 1 Gobiidae sp. 2 Grammatidae Gramma loreto Fairy basslet Haemulidae Diagramma pictum Painted sweetlips Haemulidae Haemulon album Margate Haemulidae Haemulon flavolineatum French grunt Haemulidae Haemulon plumieri White grunt Haemulidae Haemulon sciurus Bluestriped grunt Haemulidae Plectorhinchus Lemon sweetlip flavomaculatus Haemulidae Plectorhinchus gaterinus Blackspotted rubberlip Haemulidae Plectorhinchus schotaf Minstrel sweetlip Holocentridae Holocentrus adscensionis Squirrelfish Holocentridae Holocentrus rufus Longspine squirrelfish + Holocentridae Myripristis murdjan Pinecone soldierish 195 Atl. Ind. Pac. Med Family Species Common name +/ +/. +/. +/- Holocentridae Myripristis violacea Lattice soldierfish - Holocentridae Neoniphon sammara Sammara squirrelfish - Holocentridae Sargocentron Silverspot squirrelfish caudimaculatum + Holocentridae Sargocentron sp. - - Holocentridae + Kyphosidae Kyphosus cinerascens Blue seachub + Kyphosidae Kyphosus sectator/incisor Bermuda/Yellow sea chub Kyphosidae Kyphosus vaigiensis Brassy chub - Kyphosidae - Labridae Anampses Bluespotted wrasse caeruleopunctatus + Labridae Anampses meleagrides Spotted wrasse - Labridae Anampses twistii Yellowbreated wrasse - Labridae Bodianus axillaris Axilspot hogfish . Labridae Bodianus diana Diana's hogfish . Labridae Bodianus rufus Spanish hogfish Labridae Cheilinus chlorourus Floral wrasse - Labridae Cheilinus fasciatus Redbreast wrasse + Labridae Cheilinus oxycephalus Snooty wrasse + Labridae Cheilinus trilobatus Tripletail wrasse - Labridae Cheilio inermis Cigar wrasse + Labridae Chelinus chlorurus - + Labridae Cirrhilabrus exquisitus Exquisite wrasse + Labridae Clepticus parrae Creole wrasse Labridae Coris cuvieri African coris - Labridae Corisjulis Mediterranean rainbow wrasse Labridae Epibulus insidiator Slingjaw wrasse + Labridae Gomphosus caeruleus Green birdmouth wrasse + Labridae Halichoeres bivittatus Slippery dick Labridae Halichoeres garnoti Yellowhead wrasse Labridae Halichoeres hortulanus Checkerboard wrasse _ Labridae Halichoeres maculipinna Clown wrasse Labridae Halichoeres marginatus Dusky wrasse + Labridae Halichoeres pictus Rainbow wrasse Labridae Halichoeres scapularis Zigzag wrasse + Labridae Hemigymnus fasciatus Barred thicklip + Labridae Hemigymnus melapterus Blackeye thicklip - Labridae Labrichthys unilineatus Tubelip wrasse + Labridae Labroides bicolor Bicolor cleaner wrasse - Labridae Labroides dimidiatus Bluestreak cleaner wrasse + Labridae Labrus merula Brown wrasse + Labridae Labrus viridis - + Labridae Lachnolaimus maximus Hogfish Labridae Macropharyngodon Vermiculate wrasse bipartitus + 196 Atl. Ind. Pac. Med Family Species Common name +/- +/- +/- +/- Labridae Novaculichthys taeniourus Rockmover wrasse + Labridae Oxycheilinus arenatus Speckled maori wrasse + Labridae Oxycheilinus diagrammus Cheeklined wrasse + Labridae Oxycheilinus mentalis Mental wrasse + Labridae Pseudocheilinus evanidus Striated wrasse + Labridae Pseudocheilinus Sixline wrasse hexataenia + Labridae Pteragogus flagellifera Cocktail wrasse + Labridae Pteragogus pelycus Sideburn wrasse - Labridae sp. I - + Labridae sp. 2 - + Labridae Stethojulis albovittata Bluelined wrasse + Labridae Stethojulis bandanensis Red shoulder wrasse - Labridae Stethojulis interrupta Cutribbon wrasse + Labridae Symphodus cinereus Grey wrasse - Labridae Symphodus doderleini - - Labridae Symphodus mediterraneus Axillary wrasse - Labridae Symphodus melanocercus - + Labridae Symphodus ocellatus - + Labridae Symphodus roissali Five-spotted wrasse + Labridae Symphodus rostratus - + Labridae Symphodus tinea East Atlantic peacock wrasse + Labridae Thalassoma Bluntheaded wrasse amblycephalum - Labridae Thalassoma bifasciatum Bluehead + Labridae Thalassoma hardwicke Sixbar wrasse + Labridae Thalassoma hebraicum Goldbar wrasse + Labridae Thalassoma lunare Moon wrasse + Labridae Thalassoma pavo Ornate wrasse . Labridae Labrisomidae Malacoctenus boehlkei Diamond blenny Labrisomidae Malacoctenus triangulatus Sadled blenny Lethrinidae Lethrinus erythracantus Lethrinidae Lethrinus harak Thumbprint emporer Lethrinidae Lethrinus obsoletus Orange-striped emporer Lethrinidae Lethrinus olivaceus Longface emporer Lethrinidae Monotaxis grandoculis Humpnose big-eye bream Lethrinidae sp. 1 Lethrinidae sp. 2 Lethrinidae Lutjanidae Aprion virescens Green jobfish Lutjanidae Lutjanus apodus Schoolmaster Lutjanidae Lutjanus bohar Two-spot red snapper Lutjanidae Lutjanus griseus Gray snapper Lutjanidae Lutjanus mahogoni Mahogany snapper Lutjanidae Lutjanus monostigma Onespot snapper 197 Atl. Ind. Pac. Med Family Species Common name +/- +/- +/- +/- Lutjanidae Macolor niger Black and white snapper - Lutjanidae Ocyurus chrysurus Yellowtail snapper Lutjanidae + Monacanthidae Aluterus scriptus Scrawled filefish + Monacanthidae Amanses scopas Broom filefish + Monacanthidae Cantherhines pardalis Honeycomb filefish - Monacanthidae Oxymonacanthus Harlequin filefish longirostris - Monacanthidae Pervagor janthinosoma Blackbar filefish + Mugilidae Mugil sp. - - Mugilidae - Mullidae Mulloidichthys martinicus Yellow goatfish Mullidae Mullus surmuletus Striped red mullet + Mullidae Parupeneus barberinus Dash-and-dot goatfish + Mullidae Parupeneus cyclostomus Goldsaddle goatfish + Mullidae Parupeneus macronemus Longbarbel goatfish - Mullidae Parupeneus rubescens Rosy goatfish + Mullidae Pseudupeneus maculatus Spotted goatfish + Mullidae Upeneus tragula Freckled goatfish + Mullidae Muraenidae Gymnothorax javanicus Giant moray Muraenidae Gymnothorax miliaris Goldentail moray Muraenidae Muraena Helena Mediterranean moray Nemipteridae Scolopsis ghanam Arabian monocle bream Ostraciidae Lactophrys trigonus Trunkfish Ostraciidae Lactophrys triqueter Smooth trunkfish Ostraciidae Ostracion cubicus Yellow boxfish Ostraciidae Ostracion meleagris Whitespotted boxfish Ostraciidae / Tetraodontidae + Pempheridae Pempheris schwenkii Black-stripe sweeper - Pempheridae Pempheris vanicolensis Vanikoro sweeper Phycidae Phycis phycis Forkbeard + Pinguipedidae Parapercis hexophtalma Speckled sandperch + Platycephalida Papilloculiceps longiceps Tentacled flathead e Plotosidae Plotosus lineatus Striped eel catfish + Pomacanthidae Centropyge bispinosus Twospined angelfish Pomacanthidae Centropyge multispinis Dusky angelfish Pomacanthidae Holacanthus ciliaris Queen angelfish Pomacanthidae Holacanthus tricolor Rock beauty Pomacanthidae Pomacanthus arcuatus Gray angelfish Pomacanthidae Pomacanthus chrysurus Goldtail angelfish Pomacanthidae Pomacanthus imperator Emporer angelfish Pomacanthidae Pomacanthus paru French angelfish Pomacanthidae Pomacanthus Semicircle angelfish semicirculatus 198 Atl. Ind. Pac. Med Family Species Common name +/- +/- +/- +/- Pomacanthidae Pygoplites diacanthus Royal angelfish - Pomacanthidae + Pomacentridae A budefduf saxatilis Sergeant major Pomacentridae A budefduf sexfasciatus Scissortail sergeant + Pomacentridae Abudefdufsp. - + Pomacentridae A budefduf sparo ides False-eye sergeant + Pomacentridae Abudefdufvaigiensis Indo-Pacific sergeant + Pomacentridae Amblyglyphidodon Yellowbelly damselfish leucogaster + Pomacentridae Amphiprion akallopisos Skunk clownfish + Pomacentridae Amphiprion allardi Twobar anemonefish + Pomacentridae Chromis agilis Agile chromis + Pomacentridae Chromis atripectoralis Black-axil chromis + Pomacentridae Chromis chromis Damselfish - Pomacentridae Chromis cyanea Blue chromis + Pomacentridae Chromis dimidiata Chocolatedip chromis _ Pomacentridae Chromis insolata Sunshinefish Pomacentridae Chromis lepidolepis Scaly chromis - Pomacentridae Chromis multilineata Brown chromis Pomacentridae Chromis nigrura Blacktail chromis - Pomacentridae Chromis opercularis Doublebar chromis - Pomacentridae Chromis ternatensis Ternate chromis _ Pomacentridae Chromis viridis Blue green damselfish + Pomacentridae Chromis weberi Weber's chromis + Pomacentridae Dascyllus aruanus Whitetail dascyllus - Pomacentridae Dascyllus carneus Cloudy dascyllus + Pomacentridae Dascyllus trimaculatus Threespot dascyllus . Pomacentridae Microspathodon chrysurus Yellowtail damselfish Pomacentridae Neoglyphidodon melas Bowtie damselfish + Pomacentridae Neopomacentrus azysron Yellow-tail demoiselle + Pomacentridae Neopomacentrus Regal demoisells cyanomos Pomacentridae Plectroglyphidodon dickii Blackbar devil - Pomacentridae Plectroglyphidodon Johnston Island damsel johnstonianus + Pomacentridae Plectroglyphidodon Whitespotted devil lacrymatus + Pomacentridae Pomacentrus caeruleus Caerulean damselfish + Pomacentridae Pomacentrus diencaeus Longfin damselfish Pomacentridae Pomacentrus fuscus Dusky damselfish Pomacentridae Pomacentrus leucostictus Beaugregory Pomacentridae Pomacentrus partitus Bicolor damselfish + Pomacentridae Pomacentrus pavo Sapphire damsel + Pomacentridae Pomacentrus planifrons Threespot damselfish + Pomacentridae Pomacentrus sulfureus Sulphur samsel + Pomacentridae Pomacentrus trichourus Paletail damsel + Pomacentridae Pomacentrus trilineatus Threeline damsel + Pomacentridae Pomacentrus variabilis Cocoa damselfish + Pomacentridae sp. 1 - + 199 Atl. Ind. Pac. Med Family Species Common name +/- +/- +/- +/- Pomacentridae sp. 2 - + Pomacentridae sp. 3 - + Pomacentridae sp. 4 - - Pomacentridae sp. 5 - - Pomacentridae Stegastes nigricans Dusky farmerfish + Pomacentridae + Priacanthidae Priacanthus blochii Paeony bulleye - Priacanthidae Priacanthus hamrur Moontail bullseye - Ptereleotris Ptereleotris evides Blackfin dartfish + Scaridae Calotomus carolinus Carolines parrotfish - Scaridae Cetoscarus bicolor Bicolor parrotfish + Scaridae Chlorurus atrilunula Bluemoon parrotfish + Scaridae Chlorurus sordidus Daisy parrotfish - Scaridae Scams coelestinus Midnight parrotfish Scaridae Scarus frenatus Bridled parrotfish - Scaridae Scarus ghobban Blue-barred parrotfish + Scaridae Scarus inserti Striped parrotfish + Scaridae Scarus niger Dusky parrotfish - Scaridae Scarus psittacus Common parrotfish - Scaridae Scarus scaber Fivesaddle parrotfish - Scaridae Scarus taeniopterus Princess parrotfish Scaridae Scarus tricolor Tricolour parrotfish - Scaridae Scarus vetula Queen parrotfish Scaridae Scarus viridifucatus Roundhead parrotfish - Scaridae Sp. 1 (juv) - + Scaridae Sparisoma aurofrenatum Redband parrotfish Scaridae Sparisoma chrysopterum Redtail parrotfish Scaridae Sparisoma rubripinne Yellowtail parrotfish Scaridae Sparisoma viride Stoplight parrotfish + Scaridae - Sciaenidae Equetus punctatus Spotted drum Sciaenidae Sciaena umbra Brown meagre + Scombridae Scomberomorus regalis Cero Scorpaenidae Pterois antennata Broadbarred firefish - Scorpaenidae Pterois miles Devil firefish + Scorpaenidae Pterois radiata Radial firefish - Scorpaenidae Scorpaena porcus Black scorpionfish + Scorpaenidae Scorpaena scrofa Largescaled scorpionfish Serranidae Aethaloperca rogaa Redmouth grouper - Serranidae Anthias anthias Swallowtail seaperch - Serranidae Anyperodon Slender grouper leucogrammicus Serranidae Cephalopholis argus Peacock hind - Serranidae Cephalopholis boenak Chocolate hind + Serranidae Cephalopholis miniata Coral hind - Serranidae Epinephelus cruentatus Graysby + Serranidae Epinephelus fulvus Coney 200 Atl. Ind. Pac. Med Family Species Common name +/- +/- +/- +/- Serranidae Epinephelus guttatus Red hind + Serranidae Epinephelus marginatus Dusky grouper + Serranidae Epinephelus striatus Nassau grouper - Serranidae Hypoplectrus indigo Indigo hamlet - Serranidae Hypoplectrus nigricans Black hamlet + Serranidae Hypoplectrus puella Barred hamlet + Serranidae Hypoplectrus unicolor Butter hamlet + Serranidae Mycteroperca bonaci Black grouper - Serranidae Mycteroperca tigris Tiger grouper + Serranidae Plectropomus laevis Blacksaddled coralgrouper _ Serranidae Pseudanthias Sea goldie squamipinnis _ Serranidae Rypticus saponaceus Greater soapfish - Serranidae Serranus cabrilla Comber - Serranidae Serranus scriba Painted comber + Serranidae Serranus tigrinus Harlequin bass + Serranidae - Siganidae Siganus argenteus Streamlined spinefoot - Siganidae Siganus stellatus Brownspotted spinefoot + Siganidae Siganus sutor Shoemaker spinefoot + Sparidae Boops boops Bogue + Sparidae Calamus bajonado Jolthead porgy - Sparidae Calamus calamus Saucereye porgy - Sparidae Dentex dentex Common dentex - Sparidae Diplodus annularis Annular seabream + Sparidae Diplodus puntazzo Sharpsnout seabream - Sparidae Diplodus sargus White seabream - Sparidae Oblada melanura Saddled seabream - Sparidae Sarpa salpa Salema + Sparidae Spondyliosoma cantharus Black seabream + Sphyraenidae Sphyraena barracuda Great barracuda - Sphyraenidae Sphyraena sphyraena European barracuda - Syngnathidae Corythoichthys Network pipefish flavofasciatus + Synodontidae Synodus intermedins Sand diver + Synodontidae Synodus variegatus Variegated lizardfish + Tetraodontidae Arothron stellatus Starry toadfish + Tetraodontidae Canthigaster rostrata Sharpnose puffer + Tetraodontidae Canthigaster solandri Spotted sharpnose - Tetraodontidae Canthigaster valentini Valentinni's sharpnose puffer Tripterygiidae Tripterygion sp. - + Unknown - Unknown blenny + Unknown - - - Zanclidae Zanclus cornutus Moorish Idol _ Zanclidae + 201