Marine Pollution Bulletin 58 (2009) 1119

Contents lists available at ScienceDirect

Marine Pollution Bulletin

journal homepage: www.elsevier.com/locate/marpolbul

Review

Remote sensing and shhabitat relationships in coral reef ecosystems: Review

and pathways for systematic multi-scale hierarchical research
Camille Mellin a,*, Serge Andrfout a, Michel Kulbicki b, Mayeul Dalleau a, Laurent Vigliola c
a

Institut de Recherche Pour le Dveloppement, UR 128 CoRUs, BP A5, 98848 Nouma cedex, New Caledonia
Institut de Recherche Pour le Dveloppement, UR 128 CoRUs, CBETM, Universit de Perpignan, 66860 Perpignan cedex, France
c
Institut de Recherche Pour le Dveloppement, UR 070, BP 70, 29280 Plouzan, France
b

a r t i c l e
Keywords:
Coral reef
Diversity
Fish
Habitat
Remote sensing
Spatial scale

i n f o

a b s t r a c t
Understanding spatial variations in a, b, and c coral reef sh diversity, as well as both local community
and regional metacommunity structures, is critical for science and conservation of coral reef ecosystems.
This quest implies that shhabitat relationships are characterized across different spatial scales. Remote
sensing allows now for a routine description of habitats from global-regional to detailed reef scales, thus
theoretically offering access to hierarchical spatial analysis at multiple scales. To judge the progress in
using remotely sensed habitat variables for reef sh study, existing peer-reviewed papers on the subject
are reviewed. We tabulated the signicant shhabitat relationships given the different study sites, sh
and habitat variables, statistical analysis, sampling efforts and scales. Studies generally do not corroborate each other. Instead, the exercise provides a diversity of thematic results from which lessons remain
equivocal. It is thus justied to recommend more systematic and hierarchical remote-sensing based
research in the future. We advocate the use of remote-sensing early in the design of the sh study, as part
of a coherent conceptual scheme spanning all spatial scales.
2008 Elsevier Ltd. All rights reserved.

1. Introduction
Understanding how species assemblages are structured in space
is a central goal in ecology and provides the guarantee that conservation efforts are usefully assigned (Gaston 2000; Fisher and Lindenmayer 2008). In coral reef ecosystems, the relationships
between sh assemblages and their habitats depend on the spatial
and ecological scale considered, ranking from coral colonies up to
entire reef complexes (e.g., Chittaro, 2004). The Holy Grail thematic
goal that has led to characterize sh-habitat relationships is to unravel the processes that shape the spatial variation in a, b, and c
diversity. The total pool of species within a region, namely c diversity, is an important determinant of the diversity observed on a
local scale (a diversity) as indicated by both theory (e.g., Hillebrand
and Blenckner, 2002; Harrison and Cornell, 2008) and observation
(e.g., Findley and Findley, 2001; Kulbicki et al., 2005). However, a
diversity is modulated by bottom-up local natural and anthropogenic factors, including those related to habitat. Therefore, the
spatial variation in a diversity (or b diversity) also constrains c
diversity. It is thus necessary to consider the entire continuum of
scales given the top-down and bottom-up inuences, and to assess
the scales and factors (including habitat features) that inuence

* Corresponding author. Tel.: +61 8 8303 6711; fax: +61 8 8303 4347.
E-mail address: camille.mellin@gmail.com (C. Mellin).
0025-326X/$ - see front matter 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.marpolbul.2008.10.010

communities and meta-communities. We deeply acknowledge

that there is no one best scale for ecological research and that
overemphasis on either small-scale research in reductionist ecology or on patterns and processes at the largest scale is dangerous
(Kent, 2005, 2007). However, shhabitat relationships have
mostly been studied at a <100-m spatial scale using in situ habitat
descriptions that varied widely (Grober-Dunsmore et al., 2008).
This is insufcient. Our scientic and conservation issues now require considering the full continuum of spatial scales (Knudby
et al., 2007; Grober-Dunsmore et al., 2008; Hinchey et al., 2008).
During the last decade, the performance of remote sensing techniques has substantially improved. This tool provides now routine,
standardized, high-resolution and multi-scale products to map reef
geomorphology and benthic habitats, and provides new spatial
statistics as well (e.g., Andrfout et al., 2003; Andrfout and
Guzman, 2005; Purkis et al., 2005; Andrfout et al., 2008). Although
this technology has been used for years in studies of terrestrial specieshabitat relationships, particularly in landscape ecology, its
application to reef sh ecology remains in its infancy. As in mid2008, a dozen of studies attempted to couple remotely-sensed habitat characteristics and faunistic data to determine the spatial scale
and habitat features that shape reef sh assemblages. The number
of studies is limited, but the lessons that can be learnt are already
interesting, not really because they provide denitive answers to
the ecological questions asked, but rather because they span a
large variety of sampling designs, statistical analyses, and sites.

12

C. Mellin et al. / Marine Pollution Bulletin 58 (2009) 1119

This review aims to (1) provide a background context with a

short synthesis of major sh-habitat relationships and associated
scales identied thus far in coral reef ecosystems, (2) assess how
the recent development of remote sensing techniques has contributed to the study of reef shhabitat relationships and (3) identify
pathways for optimizing the integration of remote sensing techniques in reef sh ecology, especially in a multi-scale hierarchical
context.

2. Coral reef shhabitat relationships at different spatial scales

Numerous studies have focussed on how environmental factors
and habitats at different spatial scales could inuence the diversity,
the composition and/or the abundance of coral reef sh assemblages. However, sampling reef sh assemblages can only be performed at a small scale, the size of the sampling units being
rarely larger than 1000 m2 and in most cases between 100 and
500 m2. Thus we point out that studies conducted at >1 km scale
are all based on replicates of local observations at <100 m scale.
Its the design of a comparative approach with adequate number
of samples and stratied sampling strategy that eventually denes
the scale of the experiment and patterns, and the type of metacommunities that have been sampled.
At a >10 km scale, while it is also necessary to keep in mind
much debated effects such as Rapoports rule (see for a general
discussion Gaston, 2003) or the mid-domain effect (see Connolly
et al., 2003 for a case study on reef shes and corals), and other
large scale variables such as the distance to biodiversity centre
(e.g. Bellwood and Hughes, 2001; Bellwood et al., 2005), latitudinal
and longitudinal gradient (e.g., Mora and Robertson, 2005), the
supercies of available habitat appeared as one of the most significant factor determining both the c and a diversities of reef shes
(Bellwood and Hughes, 2001). This is in agreement with the theory
of island biogeography. The number of species observed is indeed
expected to increase with island or reef size, as a result of increased
gene ow and decreased probability of extinction (MacArthur and
Wilson, 1963, 1967; Karlson and Cornell, 2002).
At a 0.110 km scale, exposure to ocean and land, as well as
hydrodynamic forcing, are major forcing variables. Reef sh assemblages are known to be structured according to a coastocean gradient around continents and islands. This gradient is visible from
narrow reefs (<1 km, Galzin, 1987) to larger complexes (>10 km,
Williams and Hatcher, 1983; Grimaud and Kulbicki, 1998;
Letourneur et al., 2000). Topographic variations (e.g. passes and
points) maximizing current, dispersal and food availability are
important features, especially for events structuring early-life
stage assemblages like spawning (Heyman et al., 2007). Distance
to land and hydrodynamic exposure are co-variables of larger scale
weather and climate variables sometimes easier to characterize
synoptically. For instance, exposure to dominant winds inuence
the structure and composition of larval (Doherty, 1991; Cowen,
2002; Sponaugle et al., 2005), juvenile (Mellin et al., 2007) and
adult (Kulbicki, 1997) sh assemblages.
Still at 0.110 km scale, sh assemblage diversity, composition
and abundance directly depends on the type of biotopes/benthic
habitats present on the study site (e.g., seagrass beds, mangroves,
and patch reefs; Kendall et al., 2004; Chittaro et al., 2005) and on
the adjacent ones (Dorenbosch et al., 2004, 2005a, 2005b). Different habitats offer different rates of predation, migration and
recruitment success and thus modulate sh assemblages. The
diversity of habitat types and their spatial conguration (or landscape) determines habitat connectivity. Connectivity between habitats has an inuence on the abundance of juvenile, adults or of
theirs preys (Nakamura and Sano, 2004). The presence of a particular mosaic of benthic habitats is itself largely a function of dis-

tance to coast, hydrodynamic exposure and topography. Thus, all

these variables are likely to covary in statistical analysis. However,
working explicitly with habitat types, variables and their spatial
structures, offers the possibility to integrate patterns at very small
scales (<100 m, see hereafter) while still providing a functional link
to large scale forcing variables found at 110 km scale.
At a 10100 m scale, habitat rugosity and structural complexity
have a signicant inuence on the number of species present in
sh assemblages (e.g., Beukers and Jones, 1997; Friedlander and
Parrish, 1998; Almany, 2004; Adams et al., 2004; Sale et al.,
2005). These factors likely inuence refuge carrying capacity which
determine the maximum number of species present (see the niche
partitioning theory; MacArthur and Levins, 1964). Other important
documented habitat variables at 10100 m scale include
depth (Friedlander and Parrish, 1998; Nagelkerken et al., 2001;
Srinivasan, 2003), live coral cover (Chabanet et al., 1997; Friedlander and Parrish, 1998) and microhabitat availability (Hixon
and Beets, 1993; Friedlander and Parrish, 1998).

3. Contribution of remote sensing to reef shhabitat studies

In theory, optical remote sensing allows to map consistently
large areas of shallow (025 m depth) reefs using complex geomorphology and habitat typology schemes, sometimes at very high (2
10 m) spatial resolution (Andrfout, 2008). Remote sensing thus
provides in principle a performing tool to identify habitat variables
that most inuence sh community patterns. The high resolution
data directly allows targeting the 10 m10 km scale range. Then,
multiplying the study sites within a comparative study should help
understanding more precisely the inuence of larger scale factors,
such as the importance of habitat supercies, previously grossly
estimated for large scale work (Bellwood and Hughes, 2001).
Thus far, to our knowledge, only a few published studies attempted to explore relationships between sh assemblages and remotely-sensed habitat characteristics. It is however likely that
pioneer work left unpublished or available only through non-peer
reviewed grey literature difcult to locate have investigated the
old thematic questions with the new technology. For instance,
see the work by Kulbicki (1995) for Ouvea atoll. Other type of work
not reviewed here include using satellite imagery to stratify the
sampling and select stations, but without further analysis of the
community using remote sensing-derived variables. As an example, Andrfout and Wantiez (unpubl. data) used Landsat and
Quickbird imagery to stratify the surveys assessing reef sh communities in support of the UNESCO World Heritage application
for New Caledonia reefs. Landsat TM data was also used in a original way by Heyman et al. (2007) to help collecting bathymetry data
to map spawning aggregation sites in Belize. Finally, we do not include here the few studies on sh and remote sensing in freshwater and non coral reef tropical areas (e.g., Herold et al., 2007 for
Lake Tahoe, California). One exception though is the study by
Pittman et al. (2004) that targeted mangroves and tidal ats which
are often adjacent to coral communities, and because of their
explicit landscape ecology approach.
Atlantic Ocean reefs were the most frequently targeted reefs in
our compilation, in US Virgin Islands (USVI) (Kendall, 2005; Pittman et al., 2007; Grober-Dunsmore et al., 2007, 2008) and Florida
(Kuffner et al., 2007) as a logical follow-up of the systematic NOAA
coral reef habitat mapping program. To our knowledge, only three
studies were conducted in the Pacic Ocean (Australia: Pittman
et al., 2004; New Caledonia: Mellin et al., 2007; Hawaii: Friedlander et al., 2007) and one in the Indian Ocean (Chagos Archipelago:
Purkis et al., 2008). We hereafter further review and compare the
specics of each study. Table 1 systematically summarizes all the
relevant aspects for easier comparison.

Table 1
Synthesis of relationships between remotely-sensed habitat characteristics and reef sh assemblages for each considered study. S: species richness, or the total number of species, N: total abundance, B: Biomass, relat.: relationship,
regr.: regression, std dev: standard deviation.
Source/location

Reef type/spatial scale

Image/resolution

Mapping/habitat complexity

Habitat variables

Fish
variables

Data analysis

Friedlander et al.
(2007)
11 Marine Life
Conservation
Districts
+1 Marine Lab
Refuge,
Hawaii

Hard bottom, macroalgae

and sand
Stratied sampling,4 habitat
types  4 management
regimes (total: 500 sh
surveys)
1 station = 2 transects
25 m  5 m (2  125 m2)

IKONOS
1:60 000

Hierarchical classication and

visual interpretation of 4 major
habitat classes and n subclasses
Mapping unit = 0.004 km2

Percent cover of
substrate categories
Area protected
Management regime
Benthic cover (video)
and rugosity (chaintransect method)

S
B
Diversity






Grober-Dunsmore et al.
(2007)
St John, US Virgin Islands

22 patch reefs 12 ha, >1 km

from island coast
1 station. 1500 m-2 reef,
total: 107 stations
1 station = 7.5-m radius area
for point-count sh survey
(177 m2)

Aerial photograph
Altitude 5000m

Visual interpretation of 26 habitat classes condensed into 7

Mapping unit = 1 ac

Rugosity (in situ)

Reef size (GIS)
Perimeter to Area ratio
for patch reefs (GIS)
Seagrass cover within
100, 250, 500 m AND
1kM (Gis)

S
N
B
Diet
mobility
taxonomy
life stage

 Stepwise multiple
regress
 ANOSIM and MDS
according to the
amount of adjacent
seagrass

 Signicant relat. between S and reef

size (P < 0.001)
 S and N explained by cover of adjacent seagrass R2 = 0.320.58
 S best predicted at 250500 m, N at
1 km

Grober-Dunsmore et al.
(2008)
St John, US Virgin Islands

20 patch reefs 12 ha
820 stations per reef,
depending on reef size
1 station = 5 or 7.5 m radius
area for point-count sh surveys (80 or 177 m2)

Aerial photograph
1: 48 000

Rugosity (measured
in situ for each patch
reef)
Patch metrics (n = 3)
and landscape metric
(n = 11) (estimated by
GIS)

S
N
x
diet
mobility
taxonomy
life stage
(total 30
variables)

 Variable reduction
using PCA
 Stepwise multiple
regr. with principal
components and
with individual
habitat features

 Signicant relat. between N and

reef P:A for transient species
(P < 0.001, R = 0.74)
 Reef context correlated with 13 of
30 possible sh variables
 Habitat diversity and rugosity not
corelated with sh variables

Kendall (2005)
Buck Island Reef
Nat.
Monument St
Croix,
US Virgin
Islands

Shallow-water habitats
including patch reefs
(1 ha), seagrass beds etc.
Stratied sampling, hard vs.
soft bottoms (total: 461 sh
surveys)
1 station = transect
25 m  4 m (100 m2)

Aerial photograph

Hierarchical classication and

visual interpretation of 26 habitat classes
Mapping unit = 100 m2

Area of reef around sand

survey sites
Area of sand around reef
survey sites
within 25, 50, 100, 200,
400 and 800 m radii

S
Diversity
(Shannon
index)

 Simple linear
regress. (for S)
 Rank (Spearman)
correlation analysis
for diversity

 Most relevant scales: 100 m for S

400 m for diversity

Kuffner et al.
(2007)
Biscayne
Nat.
Park, FL, USA

12 patch reefs 2-5 ha, separated by 100 m

1013 stations per reef randomly selected
1 station = 7.5-m radius area
for point-count sh survey
(177 m2)

 Nasa EAARL (airborne

laser
1
point.m 2)
 Quickbird for denition of patch reef
contours

Digital Elevation Model (DEM)

built from EEARL and validated
with acoustic data (1-m
resolution)

Rugosity (i.e., topographic complexity)

dened at three scales
(2, 5, 10 m kernel sizes)
 estimated using
DEM
 measured using the
chain-transect method

S
N

 Simple linear
regress.
 2-way ANOVA with
rugosity levels (low,
medium or high)

 Weak relat. between estimated and

measured rugosities (R2 = 0.15,
P < 0.001)
 Signicant relation between estimated rugosity and S at 5-m scale
only (R2 = 0.08, P < 0.001)
 Signicant differences in N
between reefs and rugosity levels

Mellin et al.
(2007)
New Caledonia
lagoonal islets

Shallow-water habitats
including patch reefs, seagrass beds, macroalgae
Stratied sampling (total 12
stations)
1 station = 5-m radius area
for point-count sh survey
(80 m2)

Aerial photograph
(1.5-m resolution)

10100 m scale: depth,

rugosity, heterogeneity,
percent covers of
benthic categories
0.11 km scale:
distance to barrier reef,
distance to coast
Wind speed and direction, temperature

S
N

 GLMs at each scale

for identif. of the
most relevant (juveniles only) predictor
variables
 GLM nal with a
selection of best
predictor variables

 Most relevant scale: 10100 m

 Most important predictor variables:
depth, macro-algal cover, rugosity,
cross-shelf location, wind speed

Mapping unit = 1 ac

Classication into 5 biotopes

(35-m resolution) and in-situ
denition of biotope
characteristics

Fish variables positively correlated

with the level protection
Percent cover of hard bottom positiv. correlated with sh variables

C. Mellin et al. / Marine Pollution Bulletin 58 (2009) 1119

Independent mapping:
 26 habitat classes, condensed
into 9
 9 geomorphologic units

ANOSIM
ANOVA
ANCOVA
MDS

Main results

(continued on next page)

13

14

Table 1 (continued)
Reef type/spatial scale

Image/resolution

Mapping/habitat complexity

Habitat variables

Fish
variables

Data analysis

Main results

Pittman et al. (2004)

Moreton Bay, QD, Australia

Mangroves and tidal ats

(including seagrass beds,
macroalgae)
401 samples
1 sample=16m2 or 50m2
unitdepending on habitat
and on sampling method
(i.e., captures by net)

Aerial photograph
1:10 000
(35 cm  35 cm
resolution)
Landsat (30-m
resolution)

Supervised and unsupervised spectral classicationssynthesis within GIS

11-habitat class map for
habitat type description, and
4- habitat class map for
landscape description

Three levels: (1) landscape mosaic structure at

100 m and 300 m scales,
(2) individual habitat type
and (3) within-patch
structure variables
(in situ and video
assessments)

S
N
N of
abundant
species

 Variable reduction
using PCA
 Cluster dening
Landscape Structural Types
 Path analysis

 Signicant relat. between

landscape structure and
sh variables at 300-m scale
 S, N correlated with
habitat heterogeneity
and vegetation cover

Pittman et al. (2007)

 Puerto Rico (model
calibration)
 US Virgin Islands
(V.I.) (model
application)

Shallow-water habitats
including patch reefs,
seagrass beds, mangroves
Stratied sampling, hard
vs. soft bottoms (total:
2130 sh surveys)
1 station = transect
25 m  4 m (100 m2)

Depth data
acquired from
NOAAs
GEODAS and by
scuba divers
NOAAs benthic
habitat maps

Bathymetric model using

TIN (GIS) 5m  5m
resolution
13 benthic habitat classes
Rugosity (chain-transect
metho) assigned to each
habitat class

Depth Std dev of depth

Rugosity Area of benthic
habitat classes
Spatial scale from 225 m2
to 416 000 m2

 Comparison of multiple linear regr.,

neural networks,
regr. trees
 Test for spatial
auto-correlation
 Best model used to
predict S around US
V.I. (test for model
generality)

Purkis et al. (2008) Diego

Garcia atoll (Chagos
Archipelago)

Norther tip of atoll reef

rim
7 stations within 200 ha
1
station = 4  5-m
radius
areas for point-count sh
survey (320 m2)

IKONOS (4-m
resolution)

9-habitat class map

Bathymetry (DEM) derived
from Ikonos, 4-m resolution,
vaildated using acoustic data

Two denitions of rugosity: (1) triangulated rugosity-index calculated for

each habitat class, (2)
bathymetric
variance within 4400 m
kernel sizes at each
station
Habitat diversity
(evenness index)

S
N
diet
mobility
max. size

Monte Carlo Simulation to test for

linear regr. with
signicantly
greater R2 than
expected

Regress. tree: best

model, 75% accuracy of S
predictions (i.e., high,
medium or low S)
Best predictors [scale]:
(1) rugosity [7000 m2]
(2) depth std dev
[2000 m2 and 200 m2]
(3) depth
 Most relevant scale:
<40 m
 Signicant relation
between rugosity and S, not
with S per functional group
 Signicant relation between
rugosity and N per functional
group

C. Mellin et al. / Marine Pollution Bulletin 58 (2009) 1119

Source/location

C. Mellin et al. / Marine Pollution Bulletin 58 (2009) 1119

3.1. Reef types, spatial scales, images and resolution

The studies focussed on very different reef settings. The results
from a global coral reef mapping project conducted with Landsat
images at 30-m resolution was used to compare consistently the
general geomorphological setting of each study (Andrfout
et al., 2006). Landsat derived maps are not provided here because
of the contrasted extent of each study (full Hawaiian archipelago
in Friedlander et al. , 2007 vs. individual Florida patch reefs in Kuffner et al., 2007) which precluded here a consistent map presentation. However, maps visualized on a GIS system showed that the
studied Atlantic reefs have nothing in common with the studied Indian Ocean reefs (Diego Garcia) and Hawaiian reefs in terms of
geomorphological structures (atoll, cross-shelf reefs, etc), extent
of coral dominated zone (i.e. reef ats, forereefs), degree of exposure to ocean and wind, distance to land and deep waters, and habitatgeomorphology patchiness. This diversity posed a problem to
conduct a rigorous comparison between studies, but was also an
advantage in order to explain how diverse, and equivocal, the thematic conclusions can be in terms of habitatsh relationships.
Several studies have exclusively considered coral reef zones
(e.g., Kuffner et al., 2007; Grober-Dunsmore et al., 2007, 2008)
while others relied on a stratied sampling design including other
shallow-water habitats such as seagrass beds, mangroves or sandy
areas (Kendall, 2005; Kuffner et al., 2007; Pittman et al., 2007;
Grober-Dunsmore et al., 2007). The New Caledonia study also focussed on patch reefs but located within a closed, large, shallow lagoon (Mellin et al., 2007). Indian Ocean case study looked at the tip
and inner portion of an atoll reef rim, extending over an approximate area of 200 ha (Purkis et al., 2008).
High-resolution aerial photographs were used by Pittman et al.
(2004), Kendall (2005), and Mellin et al. (2007). For mangroves and
tidal ats of Moreton Bay (Queensland, Australia), the resolution
reached up to 35 cm (Pittman et al., 2004). The satellite imagerybased study used map derived from either 2.5-m resolution Quickbird images (Kuffner et al., 2007), 4-m resolution IKONOS images
(Friedlander et al., 2007; Purkis et al., 2008) and 30-m resolution
Landsat images (Pittman et al., 2004).
3.2. Habitat characteristics, rugosity and map complexity
The optical systems, either spaceborne or airborne, provided
rst digital multispectral images that needed to be processed and
interpreted by classication or visual interpretation to deliver
maps of habitat categories. Habitat typologies were generally obtained from eld work, but benthic covers and substrate compositions were not always systematically described in each paper
(Pittman et al., 2004; Mellin et al., 2007; Friedlander et al., 2007).
Remotely-sensed images were processed either by supervised classication (i.e., guided by a set of previously identied polygons;
Pittman et al., 2004; Mellin et al., 2007; Purkis et al., 2008) or unsupervised classication (i.e., Pittman et al., 2004; Friedlander et al.,
2007). Post-classication grouped individual classied pixels into
polygons. Generally, no minimum mapping unit area was dened.
However, the USVI habitat classication used by Grober-Dunsmore
et al. (2007, 2008) dened minimum size polygons of 1 ac
(4047 m2).
Output maps generally combined geomorphologic units and
benthic variables to hierarchically dene a number of habitat
classes ranging from 5 (Mellin et al., 2007) to 26 (Kendall, 2005;
Grober-Dunsmore et al., 2007, 2008). The 26-habitat scheme could
then be hierarchically collated into seven (Grober-Dunsmore et al.,
2007) or nine classes (Grober-Dunsmore et al., 2008) to handle
lower complexity products.
Some studies considered mean depth (Pittman et al., 2007; Mellin et al., 2007) when others considered bathymetric variance, or

15

rugosity (Kuffner et al., 2007; Pittman et al., 2007; Grober-Dunsmore et al., 2007; Purkis et al., 2008). For rugosity, Kuffner et al.
(2007), Pittman et al. (2007) and Purkis et al. (2008) computed
maps at different scales and from different techniques. For 12
mid-shelf patch reefs (25-m depth) in Florida, Kuffner et al.
(2007) pioneered an original approach, by estimating local bathymetric variance at 2, 5 and 10-m spatial scales and associated topographic complexity (rugosity) using a airborne bathymetric laser,
namely the NASA Experimental Advanced Airborne Research Lidar
(EEARL). This technology allows the remote assessment of shallow
subtidal topography by collecting laser elevation soundings at a
spatial density of 1 m-2 for an aircraft altitude of 300 m. Around
Puerto Rico and US Virgin Islands, Pittman et al. (2007) calculated
the standard deviation of depth across seven spatial scales (225 m2
up to 416 025 m2) using a bathymetric model built from a Triangulated Interpolated Network of depth point data. At a ner spatial
scale, rugosity was also estimated in situ and assigned to each
5 m cell. In Diego Garcia (Chagos Archipelago), a bathymetric
DEM of shallow areas (<15 m) could be directly derived from IKONOS satellite imagery and was calibrated/validated using acoustic
bathymetry (Purkis et al., 2008). Rugosity was then computed
using two methods, (1) as a triangulated rugosity-index dened
for each benthic assemblage mapped using satellite data, and (2)
as the variance of bathymetry within a circular kernel expanding
from 4 to 400 m around each sh sampling station.
3.3. Landscape spatio-temporal variables
Landscape ecologys paradigm is that size and conguration of
habitat patches inuence assemblage connectivity and population
dynamics (Turner et al., 2001). This seems especially relevant in
coral reef ecosystems where suitable habitats (e.g., shallow patch
reefs) may be dispersed within deeper soft bottom areas (GroberDunsmore et al., 2008). However, despite the well-known inuence
of reef size and connectivity on sh population dynamics (e.g., Roberts, 1997; Sale, 2004), only a handful of studies used landscape
ecology concepts and remote sensing to study sh-habitat relationship (Pittman et al., 2004; Grober-Dunsmore et al., 2007, 2008). In
these studies, landscape ecology indices were estimated for different patches of a same habitat type using GIS tools. Structural indices
included for example mean patch size and the ratio patch perimeter: area (Grober-Dunsmore et al., 2007, 2008). Friedlander et al.
(2007) estimated also using GIS the area of each habitat type under
different management regime (marine reserve, sheries management area or open access). Ecological indices, such as habitat diversity within a given spatial neighbourhood, were estimated through
an index of dominance similar to ShannonWiener index in Purkis
et al. (2008) and Grober-Dunsmore et al. (2008).
It is worth noting that the spatial scale considered for habitat
description in our compilation ranged from 4 m2 (Kuffner et al.,
2007) up to 416 000 m2 (Pittman et al., 2007). However, most studies examined variables estimated at four or ve spatial scales in order to identify the scale(s) at which shhabitat relationships were
most signicant. For instance, the Atlantic Ocean studies have
quantied the area of the different benthic habitats around sh
sampling stations for different neighbourhood (kernel) sizes (Kendall, 2005; Pittman et al., 2007; Grober-Dunsmore et al., 2007,
2008). The most continuous and complex habitat description included three scales (Pittman et al. 2004): (1) whole landscape
mosaics (10s of hectares; e.g., patch richness, Shannons evenness
index), (2) habitat type (100s m2 to hectares; e.g., total area of all
patches of a same habitat, percentage area occupied by a given
habitat) and (3) within-patch scale (cm2 to m2; e.g., number of seagrass species, leaf length, and shoot density).
Finally, only one study examined the temporal variations of the
sh assemblages, considering dynamic forcing variables, such as

16

C. Mellin et al. / Marine Pollution Bulletin 58 (2009) 1119

hydrodynamics and temperature. This juvenile sh study considered mean monthly wind speed and direction, and mean monthly
temperature in combination with exposure to dominant winds and
distance to coast and to barrier reef (Mellin et al., 2007).
3.4. Fish variables
In parallel to habitat and landscape descriptions, the different
studied sh variables and sampling strategy may largely constrain
the results obtained and need to be considered. The sh sampling
unit (i.e., station) were either a 57.5-m radius circular area (80
177 m2), possibly replicated (e.g., Purkis et al., 2008) when sh
were sampled using the Bohnsack and Bannerot (1986) pointcount method, or one or two replicated 25  4 m or 25  5 m transects (100250 m2). The total number of stations varied substantially, ranging from 7 (Purkis et al., 2008) up to 2 130 stations
(Pittman et al., 2007).
Fish variables were generally much more consistent between
studies than habitat variables. Among the nine major studies reviewed, all considered the species richness (i.e., total number of
sh species), six considered the total abundance, two considered
species diversity (e.g. Shannons indices) and two considered sh
biomass. Only three studies considered the distribution of these
variables among sh functional groups (e.g. according to diet and
size or mobility) and/or among ontogenetic stages (juveniles or
adults).
3.5. Data analysis
The joint examination of sh and habitat variables were carried
out using different statistical multivariates approaches. However,
the analysis was often preceded by the logarithmic transformation
of faunistic response variables in order to normalize their distribution (Clarke and Warwick 2001). A reduction of the number of
studied variables was undertaken according to their multicolinearity (Pittman et al., 2004; Mellin et al., 2007; Grober-Dunsmore
et al., 2008) which was assessed using a Principal Component Analysis (PCA; McGarigal and Marks, 1995).
Spearman rank correlation coefcient was applied when
assumptions for parametric tests were not met (e.g., normal distribution of response variables; Sokal and Rohlf, 1995). Analysis of
similarities (ANOSIM; Clarke and Warwick, 2001) were used to test
for signicant differences in sh assemblages associated with different habitats, as well as non-metric multi-dimensional scaling
(MDS; Clarke and Warwick, 2001) to visualize distance between
sh assemblages (Grober-Dunsmore et al., 2007; Friedlander
et al., 2007).
Parametric methods generally included analysis of variance
(ANOVA; e.g., Kuffner et al., 2007) and analysis of covariance (ANCOVA; e.g., Friedlander et al., 2007), simple regression (Kendall,
2005; Kuffner et al., 2007) and multiple regression (Pittman
et al., 2007; Grober-Dunsmore et al., 2007, 2008). These methods
quantied the proportion of variability in the response variable
(e.g., sh species richness) that could be explained by a single or
combination of habitat characteristics. In addition, principal components dened by PCA could be used as explanatory variables in
multiple regressions instead of habitat characteristics (GroberDunsmore et al., 2008). Diagnostics of regression outputs were performed considering residual plots (Sokal and Rohlf, 1995).
Pittman et al. (2004) used a straightforward extension of multiple regression, named path analysis, which allowed a decomposition of correlations into their causal (direct) and non-causal
(indirect) components (Mitchell, 1992; Shipley, 1997, 2000). Path
analysis generated diagrams representing the relationships between variables at different scales, their causal link and its significance. Purkis et al. (2008) pointed out one limit of linear

regressions: the probability of type-I error (i.e., reject H0 while it

is true) increases when numerous simple correlations are tested
between habitat and sh variables (e.g., Curtin and Schulz, 1998).
Although a Bonferroni correction (or equivalent) may be another
option, Purkis et al. (2008) used Monte Carlo simulations (MCS)
to generate distributions of the coefcient of determination R2 that
could be obtained by chance, and to test if R2 values obtained from
observed data were signicantly different from a random distribution (Gotelli and Ellison ,2004).
Mellin et al. (2007) proposed a predictive model of juvenile sh
species richness and abundance based on habitat characteristics.
First, General Linear Models (GLMs) were built for each considered
spatial scale to identify the most appropriate and important scale
in order to build a nal model, which accounted for a selection
of all spatial and temporal predictors. GLM yield predictions within
the limits of observed values (Guisan and Zimmermann, 2000; Guisan and Thuiller, 2005). GLM handle modelling variables obeying
different types of distribution, including the quasi-Poisson distribution that often characterizes faunistic data. In the only study
considering temporal variations, mean monthly wind speed was
identied as an important factor explaining juvenile sh assemblages. This result suggests that temporal variations of oceanographic factors and broad scale circulation can play an important
role in determining spatial patterns in sh distribution, at least
for juvenile sh. However, given the paucity of time series available, it is difcult to conclude on their relative importance compared to spatial variations in habitats. This warrants further
investigations.
Finally, Pittman et al. (2007) compared the performances of
three types of models including multiple linear regression, articial
neural networks (Lek and Gugan, 1999) and regression trees
(Breiman et al., 1984). Spatial autocorrelation of the response variable was tested using Morans I Index (Cliff and Ord, 1981); in
addition, the authors examined semi-variograms of species richness covariance in function of increasing distances between stations. Among these three models, calibrated in Puerto Rico
waters, the best model was selected based on model predictive
accuracy. Three coefcients were used as indices of predictive
accuracy: the coefcient of determination R2, the area under the
curve (AUC; Fielding and Bell, 1997) calculated from receiver-operating curves (ROC); and Cohens kappa statistic (Cohen, 1960). The
most accurate Puerto-Rico model was then used to predict the
number of sh species for two US Virgin Islands. Model generality
was tested using the three aforementioned indices calculated for
these geographically-distinct coral reefs.
3.6. Thematic results
The thematic results describe the signicant shhabitat relationships found by each study, given the different reef types, sh
and habitat variables, statistical analysis, sampling efforts and
scales. Reef types can be dened based on their biogeographical
location, geomorphology and habitat structures, but also in terms
of level of disturbance (mostly shing here) and impact on sh
community. The Caribbean reefs may be in less pristine states
due to shing than the remote Diego Garcia site and this is likely
to inuence the detected sh-habitat patterns. A near-healthy sh
population is likely to follow a very different rule of distribution
than one that is missing key functional groups. Keeping this caveat
in mind, the most striking lesson is that studies do not corroborate
each other. Instead, we are left with a diversity of thematic results
from which lessons remain equivocal. This calls for more systematic research framework, as discussed in the nal section.
First the relationships between topographic variables (i.e.,
depth, rugosity) and sh variables showed considerable variations.
Average depth was the most important variable in predictive

C. Mellin et al. / Marine Pollution Bulletin 58 (2009) 1119

models of juvenile richness and abundance (Mellin et al., 2007)

while it only ranked fth in regression trees of total sh species
richness. Depth explained less than 5% of variance in the response
variable (Pittman et al., 2007). Estimated rugosity was generally
poorly correlated to species richness (Grober-Dunsmore et al.,
2008) while it was the most important variable to explain sh species richness at a 7000-m2 scale in (Pittman et al. 2007). In Florida,
this relationship was signicant only for a 5-m kernel (R2 = 0.08;
Kuffner et al. 2007). However the relationship between estimated
and measured rugosity was also very weak (R2 = 0.15) which could
explain the apparent contradiction with previous results that
emphasized rugosity. Indeed when eld measurements of rugosity
were considered, this variable showed signicant correlation with
sh species richness (Pittman et al., 2007; Mellin et al., 2007). Estimated rugosity had a signicant effect on sh species richness and
abundance when it was considered as a discrete (i.e., high, medium
or low) variable only (Kuffner et al., 2007). Finally, a positive relationship was also observed between estimated rugosity and sh
abundance for selected functional groups, such as species mobility
(Purkis et al., 2008).
Regarding landscape and ecological indices, a variety of patterns
emerged. Species richness in Atlantic Reefs was signicantly correlated to patch reef size (Grober-Dunsmore et al., 2007). In Morreton bay, total sh abundance was positively and signicantly
correlated to habitat diversity (Pittman et al., 2004), which was
also related to abundance of species in Diego Garcia (Purkis
et al., 2008). Species richness showed signicant relationships with
hard-bottom percent cover (Kendall, 2005; Mellin et al., 2007) and
seagrass cover (Pittman et al., 2004). Both species richness and
abundances also increased with area of adjacent seagrass beds
(Grober-Dunsmore et al., 2007, 2008). Furthermore, this relationship was particularly signicant for selected trophic groups and
ontogenetic stages such as mobile macrocarnivores and Haemulid
juveniles. Lastly, a number of landscape indices appeared signicantly correlated to the distribution patterns of particular species:
74% of variations in transient sh abundances was explained by the
patch reef perimeter:area ratio and by the number of habitat
patches (Grober-Dunsmore et al., 2008).
The kernel size at which shhabitats relationships were significant considerably varied. Species richness and abundance were
correlated to rugosity at scales <40 m, to percent cover of hard bottom at 100 m scale (Kendall, 2005) and to several habitat characteristics assessed at 10100 m scale for juveniles (Mellin et al.,
2007). By contrast, sh species richness were correlated to habitat
variables at 100500 m scale (100250 m for juveniles, 500 m for
adults) while sh abundance was correlated to habitat variables
around 1 km scale (Grober-Dunsmore et al., 2007). Lastly, sh
diversity was best correlated with seagrass area at a 400-m scale
(Kendall, 2005) and sh richness was correlated with landscape
indices at a 300-m scale (Pittman et al. 2004).

4. Pathways for future multi-scale, hierarchical, research

The relationships between coral reef sh and remotely-sensed
habitat characteristics varied widely across regions, reef types
and sampling strategies, without suggesting any clear converging
conclusions or generic rules of sh assemblages across different
scales. Instead, studies remained very specic and unique in terms
of methods, and did not really show attempts at using consistent
protocols from one site to another. However, remote sensing certainly offers the best mean to achieve anywhere on the planet systematic hierarchical analysis across spatial scales to understand
the role of habitats, as performed in MacNeil et al. (in press) for
Tuamotu atolls. As a consequence, the future efforts in using
remote sensing to study sh-habitat relationships should not be

17

isolated, but part of a coherent conceptual scheme spanning all

spatial and ecological scales, across a gradient of contrasted levels
of perturbations on reefs and sh communities.
So how can a common habitat denition be gured out across
a range of different scales? Our review suggests that key components to build such a hierarchy of habitat from remote sensing are:
 Geomorphology (e.g., atolls and their morphometric attributes
like size and aperture, and also barrier reef, fringing reef, patch
reef complex attributes) and reef types (according to detailed
geomorphological units, exposure and depth). These are currently globally and consistently mapped using Landsat images
considering a 800-class typology of reef units (Andrfout
et al., 2006).
 Benthic assemblages. They were not considered in MacNeil et al.
(in press) but central to the papers reviewed here. Indeed,
remote sensing gives access to complex maps of habitats
described as benthic assemblages (see Andrfout, 2008,).
 Rugosity and depth (Kuffner et al., 2007; Purkis et al., 2008),
landscape indices of habitat diversity and spatial structures
at all scales, from geomorphology and reef types (Andrfout
and Guzman, 2005) to benthic assemblages (Pittman et al.,
2004). At the end of the spectrum, detailed benthic information
on cover, architecture and taxonomy at metric and sub-metric
scale are accessible but only from eld work, along transects
for instance.
In summary, habitats can be dened at all scales in a hierarchical and standardized fashion for all reefs worldwide, and systematically mapped with great thematic richness for at least the
three levels described above. This habitat description could then
guide the sh sampling according to multi-scale strategies like
those described in Cornell et al. (2007) for corals, but with a stronger, more objective guidance in terms of selection of representative
sampling sites.
Access to a comprehensive and hierarchical characterisation of
reefs, from benthic habitats and their landscape properties offers
the opportunity to rethink study designs to denitely solve the
spatial variation in a, b, and c diversity according to landscape congurations at different scales. The hierarchical approach initiated
by MacNeil et al. (in press) using results from the TYPATOLL comparative program was a good attempt for one region, unfortunately
the set of atolls studied does not cover the full range of atoll conguration found in the Tuamotu Archipelago (Andrfout et al.,
2005). The very large atolls were not included and the degree of
aperture to oceanic waters was apparently not used. This is
unfortunate since atoll selection for the TYPATOLL program was
explicitly based on aperture (Adjeroud et al., 2000), in addition to
size, lack of signicant population (i.e., no shing pressure) and
distance to each other (Dufour and Harmelin-Vivien, 1997).
Furthermore, the Tuamotu analysis do not include the benthic
assemblage scale, i.e. habitat maps as found in Mellin et al.
(2007), Purkis et al. (2008) and Andrfout (2008). In the MacNeil
et al. (in press) case study, using remote sensing to hierarchically
characterize (1) the complete Tuamotu Archipelago geomorphological diversity (Andrfout et al. 2005), (2) the complete
Tuamotu Archipelago landscape diversity (size, aperture, exposure) (Andrfout et al. 2005), and (3) the complete range of
benthic assemblages for the studied atolls, would have been more
powerful and representative (Andrfout et al. 2004).
A multi-scale hierarchical remote sensing-based habitat approach is necessary to determine the regional causes of local sh
richness, and the environmental inuences on both regional and
local species richness. For such goals, the different approaches described by Harrison and Cornell (2008) offer an attractive scheme
where remote sensing multi-scale habitats products could be

18

C. Mellin et al. / Marine Pollution Bulletin 58 (2009) 1119

directly plugged in. Especially attractive would be the application

of multi-scale equation structural model (Grace, 2006). This approach would allow considering simultaneously richness, density
and biomass, globally, and by taxonomic (families, genius, species)
and functional (e.g., trophic groups) levels that may have, as shown
in some of the papers reviewed here, specic responses to habitat
variables. However statistical issues may arise, like the enhanced
probability of making type I errors when considering multiple tests
of correlation between sh and habitat variables and across multiple scales. As pioneered by Purkis et al. (2008), testing against null
models might be recommended over Bonferroni corrections, which
can rapidly become impractical when the number of repeat nested
tests increases. To determine the factors that inuence both regional and local sh richness, adequate statistical methods and modelling tools are absolutely needed to sort out patterns within a
hierarchical and multifactor approach.
Remote sensing products needs to be considered during the
conceptual development of each ecological study (Andrfout
and Riegl, 2004), and not just be used as an a posteriori, afterthought tool for generalization given existing data. The habitat referential created with remote sensing is likely to help dening in
advance the limits of each study, making meaningful comparisons
within a hierarchical multi-scale framework, elaborating innovative hypotheses and testing them with static and time-series of sh
data and models. This standardized multi-scale framework would
allow meaningful inter-regional investigations and cross-study
meta-analyses at different scales, including the global scale. Finally, predictive models of sh community spatial structures could be
designed and generalized for management with stronger condence. Though such decision-support tools have been proposed
at least once (Mellin et al., 2007), both their local design and the
lack of a global referential limit their application elsewhere.
Acknowledgements
This review was made possible with the support from the Programme dvaluation des Ressources Marines de la Zone conomique Exclusive de Nouvelle-Caldonie (ZoNCo).
References
Adams, A.J., Locascio, J.V., Robbins, B.D., 2004. Microhabitat use by a postsettlement stage estuarine sh: evidence from relative abundance and
predation among habitats. Journal of Experimental Marine Biology and
Ecology 299, 1733.
Adjeroud, M., Andrfout, S., Payri, C., Orempuller, J., 2000. Physical factors of
differentiation in macrobenthic communities between atoll lagoons in the
Central Tuamotu Archipelago (French Polynesia). Marine Ecology Progress
Series 196, 2538.
Almany, G.R., 2004. Does increased habitat complexity reduce predation and
competition in coral reef sh assemblages? Oikos 106, 275284.
Andrfout, S., 2008. Coral reef habitat mapping using remote sensing: a user vs
producer perspective. Implications for research, management and capacity
building.. Journal of Space Science 52, 113129.
Andrfout, S., Guzman, H.M., 2005. Coral reef distribution, status and
geomorphology-biodiversity relationship in Kuna Yala (San Blas) archipelago,
Caribbean Panama. Coral Reefs 24, 3142.
Andrfout, S., Riegl, B., 2004. Remote sensing: a key-tool for interdisciplinary
assessment of coral reef processes. Coral Reefs 23, 14.
Andrfout, S., Kramer, P., Torres-Pulliza, D., Joyce, K.E., Hochberg, E.J., Garza-Perez,
R., Mumby, P.J., Riegl, B., Yamano, H., White, W.H., Zubia, M., Brock, J.C., Phinn,
S.R., Naseer, A., Hatcher, B.G., Muller-Karger, F.E., 2003. Multi-sites evaluation of
IKONOS data for classication of tropical coral reef environments. Remote
Sensing of Environment 88, 128143.
Andrfout, S., Hochberg, E.J., Chevillon, C., Muller-Karger, F.E., Brock, J.C., Hu, C.,
2004. Multi-scale remote sensing of coral reefs. In: Miller, R.L., Castillo, C.E.D.,
McKee, B.A. (Eds.), Remote Sensing of Coastal Aquatic Environments. Kluwer
Academic Publishers, Dordrecht.
Andrfout, S., Chauvin, C., Spraggins, S., Torres-Pulliza, D., Kranenburg, C., 2005.
Atlas des rcifs coralliens de Polynsie franaise. Centre IRD de Nouma, Service
de la Pche de Polynsie, Nouma.
Andrfout, S., Muller-Karger, F.E., Robinson, J.A., Kranenburg, C.J., Torres-Pulliza, D.,
Spraggins, S.A., Murch, B., 2006. Global assessment of modern coral reef extent

and diversity for regional science and management applications: a view from
space. In: Suzuki, Y., Nakamori, T., Hidaka, M., Kayanne, H., Casareto, B.E.,
Nadaoka, K., Yamano, H., Tsuchiya, M., Yamazato, K., (Eds). 10th ICRS. Japanese
Coral Reef Society, Okinawa.
Andrfout, S., Costello, M., Rast, M., Sathyendranath, S., 2008. Preface: Earth
observations for marine and coastal biodiversity and ecosystems. Remote
Sensing of Environment 112, 32973299.
Bellwood, D.R., Hughes, T., 2001. Regional-scale assembly rules and biodiversity of
coral reefs. Science 292, 15321534.
Bellwood, D.R., Hughes, T.P., Connolly, S.R., Tanner, J., 2005. Environmental and
geometric constraints on Indo-Pacic coral reef biodiversity. Ecology Letters 8,
643651.
Beukers, J.S., Jones, G.P., 1997. Habitat complexity modies the impact of piscivores
on a coral reef sh population. Oecologia 114, 5059.
Bohnsack, J.A., Bannerot, S.P., 1986. A stationary visual census technique for
quantitatively assessing community structure of coral reef shes. NOAA
Technical Report NMFS 41, 115.
Breiman, L., Friedman, J.H., Olshen, R.A., Stone, C.J., 1984. Classication and
Regression Trees. Wadsworth & Brooks, Monterey, CA.
Chabanet, P., Ralambondrainy, H., Amanieu, M., Faure, G., Galzin, R., 1997.
Relationships between coral reef substrate and sh. Coral Reefs 16, 93102.
Chittaro, P.M., 2004. Fish-habitat associations across multiple spatial scales. Coral
Reefs 23, 235244.
Chittaro, P.M., Usseglio, P., Sale, P.F., 2005. Variation in sh density, assemblage
composition and relative rates of predation among mangrove, seagrass and
coral reef habitats. Environmental Biology of Fishes 72, 113.
Clarke, K.R., Warwick, R.M., 2001. Change in Marine Communities: An Approach to
Statistical Analysis and Interpretation, 2nd ed. Plymouth Marine Laboratory,
Plymouth, UK.
Cliff, A.D., Ord, J.K., 1981. Spatial Processes: Models and Applications. Pion, London.
Cohen, J., 1960. A coefcient of agreement for nominal scales. Educational and
Psychological Measurement 20, 3746.
Connolly, S.R., Bellwood, D.R., Hughes, T.P., 2003. Indo-Pacic biodiversity of coral
reefs: deviations from a mid-domain model. Ecology 84, 21782190.
Cornell, H.V., Karlson, R.H., Hughes, T.P., 2007. Scale-dependent variation in coral
community similarity across sites, islands, and island groups. Ecology 88, 1707
1715.
Cowen, R.K., 2002. Larval dispersal and retention and consequences for population
connectivity. In: Sale, P.F. (Ed.), Coral Reef Fishes Dynamics and Diversity in a
Complex Ecosystem. Academic Press, London.
Curtin, F., Schulz, P., 1998. Multiple correlations and Bonferronis correction.
Biological Psychometry 44, 775777.
Doherty, P.J., 1991. Spatial and temporal patterns in recruitment. In: Sale, P.F. (Ed.),
The Ecology of Fishes on Coral Reefs. Academic Press, London.
Dorenbosch, M., van Riel, M.C., Nagelkerken, I., van der Velde, G., 2004. The
relationship of reef sh densities to the proximity of mangrove and seagrass
nurseries. Estuarine, Coastal and Shelf Science 60, 3748.
Dorenbosch, M., Grol, M.G.G., Christianen, M.J.A., Nagelkerken, I., van der Velde, G.,
2005a. Indo-Pacic seagrass beds and mangroves contribute to sh density and
diversity on adjacent coral reefs. Marine Ecology Progress Series 302, 6376.
Dorenbosch, M., Grol, M.G.G., Nagelkerken, I., van der Velde, G., 2005b. Different
surrounding landscapes may result in different sh assemblages in East African
seagrass beds. Hydrobiologia 563, 4560.
Dufour, P., Harmelin-Vivien, M., 1997. Research program for typology of atoll
lagoons: strategy and sh results. In: Proceedings of the 8th International
Symposium on Coral Reefs, vol. 1, pp. 843848.
Fielding, A.H., Bell, J.F., 1997. A review of methods for the assessment of prediction
errors in conservation presence/absence models. Environmental Conservation
24, 3849.
Findley, J.S., Findley, M.T., 2001. Global, regional, and local patterns in species
richness and abundance of butteryshes. Ecological Monographs 71, 6991.
Fisher, J., Lindenmayer, D.B., 2008. Beyond fragmentation: the continuum model for
fauna research and conservation in human-modied landscapes. Oikos 112,
473480.
Friedlander, A.M., Parrish, J., 1998. Habitat characteristics affecting sh assemblages
on a Hawaiian coral reef. Journal of Experimental Marine Biology and Ecology
224, 130.
Friedlander, A.M., Brown, E.K., Monaco, M.E., 2007. Coupling ecology and GIS to
evaluate effcacy of marine protected areas in Hawaii. Ecological Applications
17, 715730.
Galzin, R., 1987. Structure of sh communities of French Polynesian coral reefs. I.
Spatial scales. Marine Ecology Progress Series 41, 129136.
Gaston, K.J., 2000. Global patterns in biodiversity. Nature 405, 220227.
Gaston, K.J., 2003. The Structure and Dynamics of Geographic Ranges. Oxford
University Press, Oxford.
Gotelli, N.L., Ellison, A.M., 2004. A Primer of Ecological Statistics. Sinauer Associates,
Sunderland.
Grace, J.B., 2006. Structural Equation Modeling and Natural Systems. Cambridge
University Press, Cambridge.
Grimaud, J., Kulbicki, M., 1998. Inuence de la distance locan sur les
peuplements ichtyologiques des rcifs frangeants de Nouvelle-Caldonie.
Comptes Rendus de lAcadmie des Sciences Series III Sciences de la Vie
321, 923931.
Grober-Dunsmore, R., Frazer, T.K., Lindberg, W.J., 2007. Reef sh and habitat
relationships in a Caribbean seascape: the importance of reef context. Coral
Reefs 26, 201216.

C. Mellin et al. / Marine Pollution Bulletin 58 (2009) 1119

Grober-Dunsmore, R., Frazer, T.K., Beets, J., Lindberg, W.J., Zwick, P., Funicelli, N.A.,
2008. Inuence of landscape structure on reef sh assemblages. Landscape
Ecology 23, 3753.
Guisan, A., Thuiller, W., 2005. Predicting species distribution: offering more than
simple habitat models. Ecology Letters 8, 9931009.
Guisan, A., Zimmermann, N.E., 2000. Predictive habitat distribution models in
ecology. Ecological Modelling 135, 147186.
Harrison, S., Cornell, H., 2008. Toward a better understanding of the regional causes
of local community richness. Ecology Letters. doi:10.1111/j.14610248.2008.01210.x.
Herold, M., Metz, J., Romsos, J.S., 2007. Inferring littoral substrates, sh habitats, and
sh dynamics of Lake Tahoe using IKONOS data. Canadian Journal of Remote
Sensing 33, 445456.
Heyman, W.D., Ecochard, J.L.B., Biasi, F.B., 2007. Low-cost bathymetric mapping for
tropical marine conservation a focus on reef sh spawning aggregation sites.
Marine Geodesy 30, 3750.
Hillebrand, H., Blenckner, T., 2002. Regional and local impact on species diversity
from pattern to processes. Oecologia 132, 479491.
Hinchey, E.K., Nicholson, M.C., Zajac, R.N., Irlandi, E.A., 2008. Preface: Marine and
coastal applications in landscape ecology. Landscape Ecology 23, 15.
Hixon, M.A., Beets, J.P., 1993. Predation, prey refuges, and the structure of coral-reef
sh assemblages. Ecological Monographs 63, 77101.
Karlson, R.H., Cornell, J.H., 2002. Species richness of coral assemblages: detecting
regional inuences at local spatial scales. Ecology 83, 452463.
Kendall, M.S., 2005. A method for investigating seascape ecology of reef sh. In:
Proceedings of the 56th Annual Gulf and Caribbean Fisheries Institute, 2005.
R.LeRoy Creswell, Ed. Fort Pierce, FL 56, pp. 355366.
Kendall, M.S., Christensen, J.D., Caldow, C., Coyne, M., Jeffrey, C., Monaco, M.E.,
Morrison, W., Hillis-Starr, Z., 2004. The inuence of bottom type and shelf
position on biodiversity of tropical sh inside a recently enlarged marine
reserve. Aquatic Conservation: Marine and Freshwater Ecosystems 14, 113
132.
Kent, M., 2005. Biogeography and macroecology. Progress in Physical Geography 29,
256264.
Kent, M., 2007. Biogeography and macroecology: now a signicant component of
physical geography. Progress in Physical Geography 31, 643657.
Knudby, A., LeDrew, E., Newman, C., 2007. Progress in the use of remote sensing for
coral reef biodiversity studies. Progress in Physical Geography 31, 421434.
Kuffner, I.B., Brock, J.C., Grober-Dunsmore, R., Bonito, V.E., Hickey, T.D., Wright, C.W.,
2007. Relationships between reef sh communities and remotely sensed
rugosity measurements in Biscayne National Park, Florida, USA.
Environmental Biology of Fishes 78, 7182.
Kulbicki, M., 1995. Etimating demersal lagoonak sh stock in Ouva, an atoll of New
Caledonia. Joint FFA/SPC Workshop on the Management of South Pacic Inshore
Fisheries. South Pacic Commission, Nouma, New Caledonia, Integrated
Coastal Fisheries Management Project Technical Document, vol. 12(2).
Kulbicki, M., 1997. Results of ten years of research (19851997) by ORSTOM on
diversity, density, biomass and community trophic structure of the lagon and
reef shes in New Caledonia. Cybium 21, 4779.
Kulbicki, M., Bozec, Y.M., Green, A., 2005. Implications of biogeography in the use of
butteryshes (Chaetodontidae) as indicators for Western and Central Pacic
areas. Aquatic Conservation: Marine and Freshwater Ecosystems 15, S109
S126.
Lek, S., Gugan, J.F., 1999. Articial neural networks as a tool in ecological modeling:
an introduction. Ecological Modelling 120, 6573.
Letourneur, Y., Labrosse, P., Kulbicki, M., 2000. Distribution spatiale des stocks de
poissons rcifaux dmersaux dintrt commercial et effort de pche en
Province Nord de Nouvelle-Caldonie. Oceanologica Acta 23, 595606.
MacArthur, R., Levins, R., 1964. Competition, habitat selection, and character
displacement in a patchy environment. PNAS 51, 12071210.

19

MacArthur, R.H., Wilson, E.O., 1963. An equilibrium theory of insular zoogeography.

Evolution 17, 373387.
MacArthur, R.H., Wilson, E.O., 1967. The Theory of Island Biogeography. Princeton
University Press, Princeton.
MacNeil, M.A., Graham, N.A.J., Polunin, N.V.C., Kulbicki, M., Galzin, R., HarmelinVivien, M., Rushton, S.P., in press. A confederacy of processes: hierarchical
drivers of reef-sh metacommunity structure. Ecology.
McGarigal, K., Marks, B.J., 1995. Relationships between landscape structure and
breeding birds in the Oregon coast range. Ecological Monographs 65,
235260.
Mellin, C., Andrfout, S., Ponton, D., 2007. Spatial predictability of juvenile sh
species richness and abundance in a coral reef environment. Coral Reefs 26,
895907.
Mitchell, R.J., 1992. Testing evolutionary and ecological hypotheses using path
analysis and structural equation modelling. Functional Ecology 6, 123129.
Mora, C., Robertson, D.R., 2005. Causes of latitudinal gradients in species richness: a
test with shes of the tropical Eastern Pacic. Ecology 86, 17711782.
Nagelkerken, I., van der Velde, G., Cocheret de la Morinire, E., 2001. Fish feeding
guilds along a gradient of bay biotopes and coral reef depth zones. Aquatic
Ecology 35, 7386.
Nakamura, Y., Sano, M., 2004. Is there really lower predation risk for juvenile shes
in a seagrass bed compared with an adjacent coral area? Bulletin of Marine
Science 74, 477482.
Pittman, S.J., McAlpine, C.A., Pittman, K.M., 2004. Linking sh and prawns to their
environment: a hierarchical landscape approach. Marine Ecology Progress
Series 283, 233254.
Pittman, S.J., Christensen, J.D., Caldow, C., Menza, C., Monaco, M.E., 2007. Predictive
mapping of sh species richness across shallow-water seascapes in the
Caribbean. Ecological Modelling 204, 921.
Purkis, S., Riegl, B., Andrfout, S., 2005. Geomorphology and scale-invariant facies
patterns on a modern carbonate ramp (Arabian Gulf, Dubai, U.A.E) described by
an optical-acoustic remote sensing approach. Journal of Sedimentary Research
75, 861876.
Purkis, S.J., Graham, N.A.J., Riegl, B.M., 2008. Predictability of reef sh diversity and
abundance using remote sensing data in Diego Garcia (Chagos Archipelago).
Coral Reefs 27, 167178.
Roberts, C.M., 1997. Connectivity and management of Caribbean coral reefs. Science
278, 14541457.
Sale, P.F., 2004. Connectivity, recruitment variation, and the structure of reef sh
communities. Integrative and Comparative Biology 44, 390399.
Sale, P.F., Danilowicz, B.S., Doherty, P.J., Williams, D.M., 2005. The relation of
microhabitat to variation in recruitment of young-of-year coral reef shes.
Bulletin of Marine Science 76, 123142.
Shipley, B., 1997. Explanatory path analysis with applications in ecology and
evolution. The American Naturalist 149, 11131138.
Shipley, B., 2000. Cause and Correlation in Biology: A Users Guide to Path Analysis,
Structural Equations and Causal Inference. Cambridge University Press,
Cambridge.
Sokal, R.R., Rohlf, F.J., 1995. Biometry. Freeman and Company, San Francisco, CA.
Sponaugle, S., Lee, T., Kourafalou, V., Pinkard, D., 2005. Florida current frontal eddies
and the settlement of coral reef shes. Limnology and Oceanography 50, 1033
1048.
Srinivasan, M., 2003. Depth distributions of coral reef shes: the inuence of microhabitat structure, settlement, and post-settlement processes. Oecologia 137,
7684.
Turner, M.G., Gardner, R.H., ONeill, R.V., 2001. Landscape Ecology in Theory and
Practice: Patterns and Process. Springer Press, New York.
Williams, D.M., Hatcher, A.I., 1983. Structure of sh communities on outer slopes of
inshore, mid-shelf and outer-self reefs of the Great Barrier Reef. Marine Ecology
Progress Series 10, 239250.