ChapterPDF Available

Predictive mapping of coral reef fish species and communities

Authors:

Figures

Content may be subject to copyright.
219
chapter twelve
Predictive mapping of coral reef
sh species and communities
Simon J. Pittman and Anders Knudby
Introduction
At the spatial scales relevant to the routine movements of many coral reef–associated
shes, habitats exist as a spatially heterogeneous terrain varying in structural attributes,
biotic communities, and seascape context (Grober-Dunsmore etal. 2009, Boström etal.
2011). Seascape studies, focusing on the ecological consequences of spatial patterning,
have revealed that both the patch structure and the terrain morphology inuence the geo-
graphical patterns of sh distributions and diversity at a range of spatial scales (Pittman
etal. 2007a,b, Boström etal. 2011). In the past few decades, however, it has become evident
that the physical structural complexity of shallow tropical seascapes is declining (Pandol
etal. 2005). For instance, in the Caribbean Sea, multiple interacting stressors are associated
with a decline in the abundance of live coral, particularly the large and architecturally
complex branching species, and a measurable attening of the topographic complexity
over the past 60 years (Gardner etal. 2003, Pandol etal. 2005, Alvarez-Filip etal. 2009,
2011). In many locations, the biotic assemblage composition has also changed in what is
often referred to as a “phase shift” from coral to algal dominance (Done 1992, Mumby
2009), with reduced-diversity coral communities comprised of stress-tolerant species
(Green etal. 2008). These changes, combined with shing pressure, have led to a region-
wide decline in reef sh density (Paddack etal. 2009) and size composition, with the larg-
est shes becoming increasingly rare (Stallings 2009). A decline in habitat suitability for
Contents
Introduction ................................................................................................................................. 219
History of predictive mapping in coral reef ecosystems ....................................................... 221
Why a multiscale approach? .....................................................................................................222
Spatial representations of seascape structure .........................................................................222
Machine-learning algorithms .................................................................................................... 224
A multiscale and multialgorithm comparative approach .....................................................225
Case studies of predictive mapping using sh–seascape relationships..............................226
Mapping habitat suitability for a harvested grouper species .......................................... 226
Mapping large-bodied shes to identify essential sh habitat ........................................227
Mapping indicators of coral reef resilience ........................................................................228
Forecasting the impact of declining reef complexity on sh distributions ....................228
Future research ............................................................................................................................ 231
Acknowledgments ...................................................................................................................... 231
References .....................................................................................................................................231
220 Interrelationships between corals and sheries
shes of shery value has negative effects on the sustainability of reef sheries and human
livelihoods (Souter and Linden 2000, Hughes etal. 2005). At broader spatial scales, the
problem of declining quality of coral reefs is compounded by a loss of seagrasses (Orth
etal. 2006) and mangroves (Polidoro etal. 2010) from the surrounding seascape. For mul-
tihabitat shes, many of which are important food sh, adjacent vegetated habitat types
provide complementary and supplementary resources that act synergistically to elevate
local sh productivity and diversity in tropical seascapes (Parrish 1989, Nagelkerken etal.
2002, Mumby etal. 2004, Pittman etal. 2007b, Jones etal. 2010).
Degradation of sh habitat is geographically widespread and unevenly distributed,
creating a signicant logistical challenge for the prioritization of management actions.
Much recent attention has been focused on developing spatial prioritization strategies
to identify a subset of functionally important places, habitats, communities, and species
that can be preferentially protected to ensure the sustainability of key ecosystem function
and services (Roberts etal. 2003, Lourie and Vincent 2004, Klein etal. 2010). Coral reef
ecosystems that are robust or resilient to stressors have also emerged as a criterion for
prioritizing conservation investments (Maynard etal. 2010, McClanahan etal. 2012). As
well as identifying areas of special concern, such as essential sh habitat, biodiversity and
productivity hotspots, and robust and resilient reefs, managers implementing an ecosys-
tem-based approach are increasingly interested in identifying key ecological relationships
to better anticipate the consequences of environmental change (Crowder and Norse 2008,
Foley etal. 2010). Place-based management strategies such as the setting up of marine
protected areas (MPAs) and more comprehensive regional marine spatial management
are increasingly applied worldwide, including in areas with tropical coral reef ecosystems
(Mora etal. 2006, Agardy etal. 2011).
To be effective, prioritization strategies require reliable spatial data, with sufcient
ecological detail to be meaningful for ecosystem-based management, but with sufcient
geographical coverage to be operationally relevant for decision making. This is problematic
because for many regions biological survey data are sparse and unevenly distributed and
often highly clustered. Field sampling is typically conducted using sampling techniques
with relatively small (from 10 cm2 to 100 m2) sample unit areas (line transects, quadrats,
cores, etc.) providing useful detail, but at spatial scales that are too ne to support geo-
graphically comprehensive evaluations in spatial management.
Predictive mapping of biological distributions, sometimes referred to as species
distribution modeling and ecological niche modeling, is now widely recognized as
an effective analytical tool to address spatial information gaps in support of manage-
ment (Guisan and Thuiller 2005, Elith and Leathwick 2009). Crucially important for the
implementation of predictive mapping is the fact that most eld samples now have an
accurate and precise reference in time (time sample was collected) and space (geoloca-
tion of sample) that is becoming increasingly reliable with the widespread use of global
positioning system (GPS) receivers. In addition, technological advances in high-resolution
remote sensing and the increased availability of quantitative tools, such as spatial pat-
tern metrics, allow us to measure and quantify the detailed patterning of coral reef
ecosystems across multiple spatial and temporal scales (Knudby etal. 2007, Goodman
etal. 2013). New, reliable, and cost-effective techniques and analytical tools, such as
geostatistical modeling and machine-learning algorithms combined with geographi-
cal information systems (GIS), are being integrated to model and predictively map
environmental features and biological distributions across broad geographical areas
(Leathwick etal. 2008, Knudby etal. 2010a, Pittman and Brown 2011). These sophisti-
cated modeling tools are capable of more than just providing missing data since they
Downloaded by [Simon Pittman] at 16:45 11 August 2014
221Chapter twelve: Predictive mapping of coral reef sh species and communities
also have the potential to provide new ecological insights through analyses of complex
macroecological relationships.
Here, we review the development of the predictive mapping of tropical marine sh
distributions and then outline the rationale for a multiscale seascape ecology approach
that draws on concepts and analytical tools associated with landscape ecology. We pro-
vide a short overview of the latest generation of machine-learning algorithms for predic-
tive modeling and offer an operational framework for a statistically robust approach. We
highlight the utility of the techniques for addressing a range of spatial information gaps
in coral reef ecosystems with examples that focus on sh–seascape relationships of impor-
tance to the sustainability of reef sheries.
History of predictive mapping in coral reef ecosystems
Early depictions of marine sh species distributions and habitats were typically drawn
on paper maps guided by expert knowledge. It was not until the early 1990s, following
the development of habitat suitability models in terrestrial systems by the U.S. Fish and
Wildlife Service, that computer-based statistical modeling for delineating marine sh dis-
tributions emerged (e.g., Rubec and O’Hop 1996, Christensen etal. 1997). In the United
States, the Magnuson-Stevens Fishery Conservation and Management Act (NMFS 1996)
was amended to require the description and identication of essential sh habitat, repre-
senting societal recognition that habitat quantity and quality are critical to the health and
productivity of sh populations. The Magnuson-Stevens Act therefore became an impor-
tant driver of progress in mapping sh–habitat relationships at a time when the mapping
process was being revolutionized by GIS software, GPS receivers, and a general increase in
the availability of marine spatial data. One of the rst applications of predictive mapping
for tropical marine species was the Florida Estuarine Living Marine Resources System
(FELMER) project, which statistically linked information on the habitat requirements of
marine species (at different life stages) to maps describing the habitat characteristics of the
seaoor and water column to predict species distributions for key shery species (Rubec
etal. 1998a,b). Habitat preferences for each species’ life stage were determined by tting
polynomial regressions to mean catch-per-unit-effort (CPUE) across environmental gra-
dients, with higher mean CPUEs indicating higher abundance. Habitat Suitability Index
(HSI) maps created by linking the geometric mean of the index values to each habitat
layer, were used to create seasonal maps (spring, summer, fall, winter) depicting the spa-
tial distribution of each species’ life stage (Rubec etal. 1998a,b, Rubec etal. 1999). Hotspots
of suitability were then used to identify the most important habitats for each species with
potential to be designated Habitat Areas of Particular Concern. This HSI approach was
also used to develop models that could be applied to unsurveyed areas in order to cost-
effectively ll spatial information gaps.
Many of the early efforts (e.g., Christensen etal. 2003) used linear models with many
statistical assumptions and limited ability to model nonlinear patterns including thresh-
old effects and complex multiscale interactions between environmental variables, all
typical of ecological relationships. Relationships between organisms and environmental
variability were modeled at single and sometimes arbitrary spatial scales, and much of
the ecologically meaningful spatial heterogeneity in seaoor structure was not explicitly
incorporated. More recently, a conceptual and statistical leap has been made in predictive
modeling by incorporating environmental predictors quantied at multiple spatial scales
using spatial pattern metrics from landscape ecology and geomorphology and the lat-
est generation of machine-learning algorithms to model complex nonlinear relationships.
Downloaded by [Simon Pittman] at 16:45 11 August 2014
222 Interrelationships between corals and sheries
Comparative multimodel analysis has demonstrated that this spatial ecoinformatics
approach to ecological modeling can boost predictive performance and provide new
insights into potentially important drivers of ecological patterns (Elith etal. 2006, Pittman
etal. 2007a, Knudby etal. 2010a, Pittman and Brown 2011).
Why a multiscale approach?
Ecological theory tells us that interactions across scales are important (Allen and Starr
1982, Kotliar and Wiens 1990, Holling 1992, Levin 1992). From the landscape ecology per-
spective, which incorporates a hierarchical conceptual framework, we recognize that spe-
cies and communities respond to their environment at a range of spatial scales (Schneider
2001, Pittman and McAlpine 2003). Even within a single species, individuals at different
life stages or of different sexes may respond to the environment in different ways, and at
different scales, although some generalities will exist. To add to the already complex scale
effects, the organism’s response will also be modied by locational differences, such as the
local distribution of habitat structure, predators, and prey.
However, for most marine species, even those that have been well studied, the relevant
scale range is either estimated or, in most cases, remains unknown. Movement patterns
can guide scale selection, but rarely are the relevant spatial and temporal dimensions of
movement patterns known. Lacking knowledge of a specic focal spatial scale, we recom-
mend adoption of an exploratory multiscale approach. This not only is judicious as a bet-
hedging strategy when faced with scale uncertainty, but also provides a technique with
which to begin incorporating a wider range of structural heterogeneity into the analyses
and an opportunity to examine scale-dependent responses.
A multiscale seascape approach is particularly relevant to reef sh ecology since many
shes use multiple habitat types through routine daily home range movements, ontoge-
netic habitat shifts, and seasonal or spawning migrations (Nagelkerken etal. 2002, Pittman
and McAlpine 2003). For example, in the context of locating and measuring habitat suit-
ability, a princess parrotsh (Scarus taeniopterus) may only persist in seascapes where diur-
nal foraging habitat is sufciently close to suitable nocturnal resting habitat providing
adequate protection from predators. If suitable resources are not located within a distance
traversable by an individual sh, then the seascape will likely offer low habitat suitabil-
ity. If similar habitat preference occurs among multiple herbivorous shes, then the low
habitat suitability areas will support less herbivory than high-suitability areas. This habi-
tat may consequently be of lower value to sheries. Complex sh–seascape interrelation-
ships with implications for sheries and reef resilience can be modeled and made spatially
explicit with the predictive mapping techniques.
Spatial representations of seascape structure
The spatial patterning of seascapes can be represented in two-dimensional (i.e., benthic
habitat maps) and three-dimensional (i.e., digital elevation) models. From these models, a
wide range of derivative environmental predictors can be created for predictive mapping
of sh–seascape relationships. Predictors can be classied as relatively stable or relatively
dynamic along a continuum of dynamism (Hyrenbach etal. 2000). Relatively stable pre-
dictors include seaoor features such as geology (geomorphological features/zones and
bathymetry) and benthic communities or habitat types (seagrasses, colonized hardbottom,
mangroves, etc.). Relatively dynamic predictors include water surface and water column
properties such as salinity, temperature, ocean fronts, and phytoplankton concentration.
Downloaded by [Simon Pittman] at 16:45 11 August 2014
223Chapter twelve: Predictive mapping of coral reef sh species and communities
The majority of coral reef ecosystem applications have focused on the stable seaoor
features as reliable representations of seascape structure with known ecological relevance
to reef-associated shes. Seaoor structure is usually represented as either thematic ben-
thic habitat maps or three-dimensional digital terrain models. Benthic habitat maps are
typically two-dimensional maps of internally homogeneous patches (habitat or biotope
classes), with discrete boundaries represented as polygons. Benthic habitat maps can also
be represented as a grid of cells (i.e., raster data) in which a habitat type is assigned a
unique value. Predictors can be based on individual habitat types or on patch context
by quantifying the seascape composition (amount and variety of habitat types) and con-
guration (spatial arrangement of habitat patches) in the surrounding seascape (Wedding
etal. 2011). Advances in high resolution multispectral and hyperspectral remote sensing
now make it more likely that habitat maps can incorporate reliable information on the
biotic composition of patches such as the amount of live coral and macroalgae (Hedley
2013, Wozencraft and Park 2013). This could further improve model performance.
In addition to discrete patchiness, the marine environment exhibits spatial variability
in the form of continuous, multidimensional gradients. For example, hydrodynamic inter-
actions with coastline geomorphology, riverine inuences, and seaoor topography create
a wide range of spatial gradients in marine environmental conditions (e.g., salinity, wave
action, depth, temperature, nutrients). The continuum model recognizes that a range of
ecological processes may affect habitat suitability for different species through time, in a
spatially continuous and potentially complex way. The premise is that individuals within
a species are likely to respond to spatial gradients in resources such as food and refuge or
other environmental conditions (Austin and Smith 1989, Fischer and Lindenmayer 2006).
An additional benet of the gradient model is that it retains the captured heterogeneity
and avoids the subjectivity associated with boundary and thematic designations associ-
ated with habitat maps. High resolution bathymetric data and associate backscatter col-
lected using techniques such as airborne laser altimetry (e.g., light detection and ranging
or LiDAR) and sonar (e.g., multibeam or side-scan sonar) can be used to construct continu-
ously varying surfaces exhibiting complex vertical, horizontal, and compositional struc-
ture (Wedding etal. 2008, Pittman etal. 2009).
In the absence of continuous sampling, interpolated or predictive surfaces can be
modeled using point data from intensive georeferenced eld samples. A cost-effective
way to create predictor variables is to identify useful surrogate or proxy variables which
describe heterogeneous ecological patterns that are difcult to map directly, such as bio-
logical diversity, community composition, and some biotic distributions. Useful surrogate
variables have been found that can be either biotic or abiotic features providing they can
be reliably mapped (Ward etal. 1999, Beger and Possingham 2008, Harborne etal. 2008,
McArthur et al. 2010, Mellin et al. 2011). Seaoor maps that depict ecosystem structure,
such as benthic habitat maps, are spatial representations that typically integrate both biotic
and abiotic features of the seaoor (Brown etal. 2011), although some maps now also have
incorporated both benthic and pelagic features into seascape classes (e.g., Roff etal. 2003).
Bathymetry is probably the single most useful predictor of marine biotic distributions
due to its importance for marine ecological patterns and processes. Although not always
available at sufcient spatial resolution for coral reef ecosystems, where bathymetry data
have been examined, the three-dimensional arrangement of structural features over the
seaoor surface (i.e., topographic complexity) has been an important driver of sh distribu-
tions and diversity (Pittman etal. 2009, Pittman and Brown 2011). Topographic complexity
at a range of spatial scales inuences a wide range of the biological, chemical, and physical
aspects of a coral reef system such as hydrodynamics, nutrient pathways, predator–prey
Downloaded by [Simon Pittman] at 16:45 11 August 2014
224 Interrelationships between corals and sheries
interactions, and larval settlement (Zawada 2011). In predictive models of habitat suitabil-
ity for reef sh, higher species richness is associated with higher topographic complexity
(Pittman et al. 2007a, 2009). Interestingly, the highest species richness is not associated
with the highest topographic complexity, suggesting that the relationship is nonlinear or
the relationship is affected by other variables (Pittman and Brown 2011). Awide range of
metrics are available for quantifying complex structure on continuously varying surfaces
(McGarigal and Cushman 2005) and we refer to these as terrain morphometrics because
they measure the surface morphology (e.g., slope, aspect, curvature, rugosity, etc.) from a
digital terrain model of bathymetry.
Geographical position metrics such as distance to shore, distance to shelf edge, and
distance to river mouth have proved useful through interaction with other predictor vari-
ables because they can capture unobserved, unknown, or unmapped patterns such as
inshore–offshore gradients in physical and chemical conditions and proximity to ecologi-
cally relevant features (e.g., nearshore nursery areas, shelf-edge spawning sites, freshwater
outow). Although rarely quantied, distance to shore has made a major contribution to
models of reef sh distributions in examples from Australia (Mellin et al. 2010) and the
Caribbean (Pittman and Brown 2011).
Machine-learning algorithms
Traditional statistical methods such as linear, logarithmic, and logistic regression have
commonly been used to examine complex ecological relationships, leading to errone-
ous and misleading interpretations of ecological relationships when the assumptions
underpinning the chosen regression model are not justied by the data (Jones and Syms
1998, Breiman 2001). An alternative and more exible approach is provided by machine-
learning or statistical-learning algorithms, a family of nonparametric models where the
structure of the relationships between variables is not dened a priori but rather derived
from iterative training and testing using random subsets of the available data (Hastie etal.
2009). Machine-learning algorithms have been shown to outperform parametric models in
many multimodel comparisons in ecology, and have become the default tools for today’s
ecological modeling of all but the simplest ecosystems (Elith et al. 2006, Phillips et al.
2006, Pittman etal. 2007a, Knudby etal. 2010b). For example, Elith etal. (2006) compared
16 predictive modeling techniques, including both conventional and machine-learning
algorithms using presence-only data for 226 species from six regions, and showed that
machine-learning algorithms consistently outperformed other algorithms. For modeling
coral reef ecosystems, Knudby etal. (2010b) showed that machine-learning algorithms pro-
vided signicant increases in performance over more conventional modeling techniques
such as generalized additive models and linear regression. Examples gaining popular-
ity in ecology are articial neural networks, support vector machines, and various forms
of decision tree ensembles such as boosted regression trees (BRT) and random forest.
Furthermore, much evidence suggests that predictions based on an ensemble of models of
varying structure outperform those based on a single model type (Aruajo and New 2007,
Marmion etal. 2009), although both the models and the method used to combine their pre-
dictions inuence performance (Džeroski and Ženko 2004). Implementation through free
software such as R (R Core Development Team 2012) has improved accessibility and some
of the commonly perceived “black box” mechanics have now been explained to ecologists
(De’ath 2007, Elith etal. 2008, Olden etal. 2008). For example, measures of relative variable
importance and partial dependence plots can provide much insight into the model struc-
ture. Nevertheless, interpretation of model structure remains complicated when the value
Downloaded by [Simon Pittman] at 16:45 11 August 2014
225Chapter twelve: Predictive mapping of coral reef sh species and communities
of one predictor variable inuences the relationship between the response variable and
another predictor variable (e.g., shelter space may correlate positively with sh biomass
at shallow depths but not at deeper depths). Similarly, complications occur when multiple
collinear predictors are used to train a predictive model, as is the case when a multiscale
approach is used and in exploratory research where different but related independent
variables (e.g., turbidity and distance from shore) are included (Knudby etal. 2010b).
A multiscale and multialgorithm comparative approach
We advocate a multiscale and multialgorithm comparative approach (Figure 12.1) in situ-
ations where there are no specic reasons to select a single scale or single modeling algo-
rithm. This exploratory spatial ecoinformatics approach incorporates lessons learned from
data mining applications in a wide range of disciplines. Comparative studies indicate that
model results and subsequent interpretation of predictor relevance are often dependent
on the type of model algorithm used (Knudby etal. 2010b). Furthermore, the evaluation
of model performance can be biased by the choice of metrics and the spatial scale used to
measure model accuracy (Aguirre-Gutiérrez et al. 2013). We recommend using multiple
metrics because each metric quanties different aspects of model performance and the
literature includes reasonable arguments both for and against most of the commonly used
model performance metrics (Fielding and Bell 1997, Pearce and Ferrier 2000). A multial-
gorithm and multimetric approach to model evaluation and selection therefore provides
an opportunity to examine cross-model agreement (and disagreement) that should result
in a more robust analysis with more complete quantication of errors leading to reduced
uncertainty and greater realism in the ecological interpretation. It is popularly quoted
Map accuracy
(e.g., % accuracy,
AUC)
Model selection
(e.g., cross-validation,
AIC, RSME, AUC)
Evaluate input data
(e.g., % accuracy, RSME,
autocorrelation)
Multiscale predictors
and target variable
Multiple
algorithms
Communication
of spatial uncertainty
Mapped
predictions
Map of errors
(e.g., mapping residuals,
interpolation of errors)
Figure 12.1 Multiscale and multimodel approach including three steps of model validation:
(1)assessment of errors in the input data, (2) model validation and selection of best model or model
ensemble, and (3) map accuracy assessment using independent data. AIC, Akaike’s information
criterion; RSME, root mean square error.
Downloaded by [Simon Pittman] at 16:45 11 August 2014
226 Interrelationships between corals and sheries
that “all maps are lies” and “all models are wrong,” but we argue that by taking careful
account of all sources of model bias and error throughout the process, we can produce
predictive models that are of great value to marine ecologists and managers. Although
rarely covered in the modeling literature, greater attention is needed so that sources of
error can be accounted for and explained and techniques can be developed to enhance
the communication of data caveats, errors, and spatial uncertainty (Hunsaker etal. 2001,
Wedding etal. 2011).
Case studies of predictive mapping using
sh–seascape relationships
Mapping habitat suitability for a harvested grouper species
Knowing the distribution of key species of ecological and economic concern across spa-
tially complex seascapes supports management decision making, including siting of
MPAs, evaluation of threats, zoning, and delineation of essential sh habitat. Pittman
and Brown (2011) predicted suitable habitat for selected coral reef-associated sh spe-
cies recorded in underwater visual surveys across the insular shelf of southwestern
Puerto Rico. Species presence/absence association coefcients were calculated for a
range of sh species including Cephalopholis fulva (coney), a common grouper species
targeted by local sheries. During the sh survey period, the bathymetry of the study
area was mapped with airborne LiDAR and processed into a 4 × 4 m spatial resolu-
tion digital terrain model (<1 to 70 m depth). Seventy-nine spatial predictor variables
were developed, including 11 derivatives of seaoor bathymetry quantied at seven
different spatial scales (5, 15, 25, 50, 100, 200, 300 m radii) using a “moving window”
technique within a GIS (Pittman and Brown 2011). Geographical setting was quantied
by creating a grid of distance to shelf edge and distance to coastline measurement.
Species distributions were modeled with BRT and MaxEnt to compare model perfor-
mance. A high probability of occurrence indicated that the grid cell had suitable envi-
ronmental conditions for the species. BRT provided “outstanding” model predictions
(area under the curve [AUC] > 0.9) for three of ve species and MaxEnt for two of ve
species, including C. fulva models that were the highest performing according to stan-
dard interpretation of the AUC metric. Distance to shelf edge was the most inuential
single predictor variable. Of the bathymetric derivatives, slope of the slope, a measure
of topographic complexity, contributed most to explaining the response variables. For
C. fulva, a threshold effect was evident at approximately 2000 m from the shelf edge,
where species occurrence abruptly increased. For C. fulva, the predictive map from the
MaxEnt model was more accurate than the BRT map based on assessment with inde-
pendent species sightings data. Mapped predictions of habitat suitability showed that
although almost all C. fulva were observed close to the shelf edge, not all shelf-edge
areas offered high suitability due to variability in the spatial heterogeneity of the sea-
oor. With regard to scale effects, the best model indicated that topographic complex-
ity (slope of the slope) in the surrounding 100 m radius seascape was most inuential
in dening suitable habitat when interacting with distance to shelf edge. The study
demonstrated that useful distribution models of important reef shery species can be
developed using a wide range of multiscale explanatory variables derived from a single
airborne sensor. Also demonstrated is the utility of applying multimodel algorithms
that are capable of modeling nonlinear relationships with interactions among explana-
tory variables. Little is known, however, about the inuence of spatial heterogeneity
Downloaded by [Simon Pittman] at 16:45 11 August 2014
227Chapter twelve: Predictive mapping of coral reef sh species and communities
in shing effort on the performance of models to accurately predict species distribu-
tions. Comparison between models developed for grouper in shed versus neighboring
unshed reefs could be informative in determining the proportion of suitable habitat
that is actually occupied by the species.
Mapping large-bodied shes to identify essential sh habitat
Identifying and quantifying areas that concentrate biomass in coral reef ecosystems, such
as sh spawning aggregation sites and other essential sh habitat, is a high priority for
sheries management. Costa etal. (2013) used a ship-based, splitbeam echosounder to map
sh position in the water column and their body size along the shelf edge of the U.S. Virgin
Islands simultaneously with a multibeam echosounder to map the seaoor bathymetry.
Multiscale sh–seascape relationships were then modeled using BRT to identify important
environmental drivers of sh occurrence and density and to predictively map locations of
high sh productivity (Figure 12.2). Models for large sh performed well enough to provide
useful maps of distributions (80.4%–86.1% accuracies), and highlighted areas with low sh
density. Water depth, topographic complexity, and proximity to the shelf edge were the
most inuential explanatory variables. The integration of acoustic sensors and multiscale
predictive mapping techniques offered a novel application of geospatial technologies to
rapidly identify sh aggregations and associated habitats to support spatial prioritization
in marine management. The study demonstrated the utility of integrating data from two
different acoustic sensors capturing sh–seascape relationships in four dimensions (three
in space and one in time). These data were collected at night and from water depths where
ecological observations are rarely conducted due to low visibility. A limitation of the tech-
nique, however, is that species cannot be easily identied without additional underwater
observations (divers/submersibles). Future research is now required to develop pattern
recognition software that will automate the identication of sh species in the acoustic
data based on body size combined with behavioral characteristics such as schooling, swim
speed, and position in the water column.
(1) Collect data
MBES
MBES
SBES Large
fish
Large
fish
Large
fish
Depth
Depth
Depth
(SD)
Depth
(SD)
Curvature
(plan)
Curvature
(plan)
Distance
to shelf
Distance
to shelf
Rugosity
Rugosity
Slope of
slope
Slope of
slope
Occurrence
Density
Large
fish
Medium
fish
Medium
fish
Medium
fish
Medium
fish
Small
fish
Randomly chose 50% of data and
removed spatially autocorrelated points
Small
fish
Small
fish
= 3 spatial
predictions
= 3 spatial
predictions
= 3 spatial
predictions
= 3 spatial
predictions
Small
fishCross validation and accuracy
assessment statistics
Cross validation statistics
and receiver operating
characteristic curves
2 × 2 m
2 × 2 m
r = 25 m
r = 25 m
r = 100 m
r = 100 m
r = 300 m
r = 300 m
r = 50 m
r = 50 m
SBES
Tampo B
ank (validation site)
St. John Wedge (training site)
(2) Create predictors/response variables(3) Create ensemble models and spatial predictions
(4) Test model performance
and accuracy
r
r
(n = 10) (n = 10) (n = 10)
(n = 10) (n = 10) (n = 10)
(n = 10) (n = 10) (n = 10)
(n = 10) (n = 10) (n = 10)
Occurrence
Density
Figure 12.2 (See color insert) Example of analytical steps and data types in the multiscale seascape
ecology approach to spatial predictive modeling. MBES, multibeam echosounder for mapping the
seaoor; SBES, singlebeam echosounder for mapping location and body size of sh. (Adapted from
Costa, B., Taylor, C.J., Kracker, L., Battista, T., and Pittman, S.J., PLoS ONE, 9(1), e85555, p1–7, 2013.)
Downloaded by [Simon Pittman] at 16:45 11 August 2014
228 Interrelationships between corals and sheries
Mapping indicators of coral reef resilience
The resilience of an ecosystem can be dened as the ecosystem’s ability to maintain and
restore structure and function in the face of disturbance (Pimm 1984, West and Salm 2003).
Resilient areas are increasingly a focus for inclusion in MPAs and MPA networks, yet little
is known about which areas are the most resilient to disturbances. Indicators quantifying
ecosystem properties thought to confer resilience (McClanahan etal. 2012) can be used as
proxies to guide identication of resilient reefs. Commonly accepted indicators of coral
reef resilience include (but are not limited to) high coral cover and diversity, and low nutri-
ent and sediment levels. For sh communities, high herbivore biomass and diversity is
associated with resilience because herbivory will maintain algal biomass at lower levels,
thus facilitating coral recovery following a disturbance. These indicators can be mapped
with predictive modeling, but the work is in its infancy and map accuracies for many
biological indicators are relatively low when predicted using satellite-based predictors
(Rowlands etal. 2012, Knudby etal. 2013). Knudby etal. (2013) predicted the distribution
of functional richness in herbivorous sh across coral reef ecosystems in the Kubulau tra-
ditional sheries management area of Fiji. The authors found that random forest models,
a machine-learning ensemble technique, were more accurate than geostatistical interpola-
tion for mapping distributions of sh herbivore biomass and functional group richness
(Figure 12.3).
Important future challenges include mapping and modeling connectivity, as well as
the intelligent combination of resilience indicators and auxiliary information into more
direct measures of resilience. Validation of resilience maps using time series of ecological
observations will be the ultimate test of the success of predictive mapping efforts.
Forecasting the impact of declining reef complexity on sh distributions
High resolution digital terrain models of seaoor complexity provide a novel cost-effective
tool for forecasting (and hindcasting) impacts on sh from changes to the topographic
complexity of coral reef ecosystems. Coral reef rugosity, a measure of topographic com-
plexity, is estimated to have decreased by more than 50% since the 1960s (Alvarez-Filip
etal. 2009). Predictive mapping has recently been used as a proof of concept to develop
impact scenarios by mapping the effect of reducing terrain topographic complexity on sh
habitat suitability in southwestern Puerto Rico (Pittman etal. 2011b).
Fish species occurrence, from underwater visual surveys, was statistically linked to a
suite of spatial predictors derived from airborne laser–derived bathymetry (LiDAR) using
MaxEnt (Pittman and Brown 2011). Topographic complexity, measured as the slope of the
slope, was used to spatially model and map probabilities of species presence. Slope of the
slope was then uniformly reduced across the entire terrain by 25% to indicate the esti-
mated decadal decline for southwestern Puerto Rican coral reefs, and by 50%, approximat-
ing Caribbean-wide declines since the 1960s. Predictions of high habitat suitability (using
a consistent probability threshold for each scenario) were then remapped for an abun-
dant herbivorous scraper, S. taeniopterus (princess parrotsh) and a biological indicator
species of live coral cover, Stegastes planifrons (threespot damselsh). Mapped predictions
were overlain and examined for differences in spatial patterning. Areas of suitable habitat,
under different reef “attening” scenarios, were quantied and compared to measure the
change. Suitable habitat for S. taeniopterus reduced by 30% with a 25% reduction in terrain
complexity, and reduced by as much as 66% when terrain was reduced by 50%. With a 25%
reduction in topographic complexity, habitat was lost from the edges of large contiguous
Downloaded by [Simon Pittman] at 16:45 11 August 2014
229Chapter twelve: Predictive mapping of coral reef sh species and communities
Number of herbivore functional groups
Observations
Predictions
2
3
0 1.5
N
36 km
4
3.99
1.84
Figure 12.3 (See color insert) Predictive map of richness in herbivorous sh functional groups
across coral reef ecosystems in Fiji as one indicator in a suite of resilience indicators. Predictive
models were developed using random forest on the basis of high-resolution satellite data.
(Adapted from Knudby, A., Jupiter, S., Roelfsema, C., Lyons, M., and Phinn, S., Remote Sens, 5,
1311–1334, 2013.)
Downloaded by [Simon Pittman] at 16:45 11 August 2014
230 Interrelationships between corals and sheries
patches of suitable coral reef where structure had already existed near the lower thresh-
olds of suitability. With a 50% reduction, patches of suitable habitat fragment reduced even
more and few large contiguous patches remain, whereas the number of small patches with
relatively small interiors of habitat increased across the seascape. For S. planifrons, a 56%
loss of suitable habitat occurred with a 25% reduction in terrain complexity (Figure 12.4).
Simulating decline of topographic complexity by applying a spatially uniform reduction
scenario is likely to be an overly simplistic scenario because it ignores the true spatial
heterogeneity of change. In the absence of information to guide more realistic simula-
tions, predictive mapping offers a exible and cost-effective tool to examine a variety of
scenarios that can be rened in the future, for example, to account for the different impacts
of future storms on deep versus shallow and exposed versus sheltered reefs and impacts
of ocean acidication.
The four case studies described here differ in their geographic setting, choice of
algorithm(s), modeled response variable, and both the metrics and spatial scales used to
quantify the environmental predictors. Given the small number of existing studies to date,
such multidimensional variability between studies hinders the development of synthesis,
and few studies have individually provided direct comparisons between aspects of this
variability beyond the choice of modeling algorithm. Replication of comparison studies
across geographic settings will be necessary to provide more informed guidance on the
choice of both algorithm(s) and the metric and spatial scales of environmental predic-
tors. Nevertheless, in combination the case studies illustrate the wide potential for spatial
predictive modeling of sh and sh-related ecosystem variables on coral reefs, as well
as the range of data sources currently in use for that purpose. The use of a multimodel
Suitable habitat after 25% flattening
0 0.5 1km
17°56’30’’N
17°57’0’’N
17°57’30’’N
67°3’30’’W67°4’0’’W67°4’30’’W67°5’0’’W
Existing suitable habitat for S. planifrons
Figure 12.4 Predicted habitat suitability for threespot damselsh (S. planifrons) across a subset of
the southwestern Puerto Rico study area using unaltered LiDAR-derived topographic complex-
ity and numerically attened topographic complexity to simulate 10-year declines for coral reefs
in southwestern Puerto Rico. MaxEnt was used for modeling predictions. (Pittman, S.J., Costa, B.,
Jeffrey, C.F.G., and Caldow C., 2011. Proceedings of the 63rd Gulf and Caribbean Fisheries Institute
1–5 November, 2010 San Juan, Puerto Rico. p. 420–426.)
Downloaded by [Simon Pittman] at 16:45 11 August 2014
231Chapter twelve: Predictive mapping of coral reef sh species and communities
and multiscale approach is implicitly or explicitly adopted in both these and other stud-
ies (Kendall etal. 2003, Knudby etal. 2010a,b, Wedding et al. 2008, 2011, Pittman et al.
2009). This is a consequence of both our limited knowledge of optimal (focal) spatial scales
at which specic ecological processes operate and the increasing realization that species
respond to their environment at multiple scales.
Future research
In modern marine management, spatial modeling plays a key role in decision support and
is central to the planning and implementation of ecosystem-based management (Pittman
etal. 2011a). Predictive mapping is rapidly emerging as a viable tool for rapid, reliable, and
cost-effective spatial ecological data analysis for marine management, particularly for site
prioritization. Modeling techniques are increasingly adept at untangling large and complex
ecological data and knowledge is growing about the errors and biases impacting the various
stages of model development (samples size and distribution, scale of analysis, etc.). Future
research is needed to apply the multiscale seascape approach to a wider range of marine
ecosystems. Although model parsimony is desirable, it is essential to develop and test a wide
range of remote sensing data to identify useful predictors. The techniques presented in this
chapter can also be applied equally to map human use patterns and threats to ecosystem
health. Little work has focused on predictive mapping of ecosystem function and services
that could provide a cost-effective method to determine ecosystem value across broad spa-
tial scales. For coral reef ecosystems, there is great potential for using models to link spatial
patterns to ecological processes and the emerging work on mapping resilience indicators is
beginning to make the linkage between ecological patterns and process. For some regions of
the earth, it is now feasible to map the geography of herbivory by mapping individual herbi-
vore species and groups of herbivores. With an increase in the number of multiscale ecologi-
cal studies being conducted, it is conceivable that, at some point in the near future, there will
be sufcient knowledge to begin the search for generalities that could lead to the formulation
of new ecological laws explaining the distribution and biodiversity patterns in the seas.
Acknowledgments
We thank the organizers of the workshop “Inter-relationships between coral reef and sh-
eries” held in Tampa, Florida, for inviting this contribution. We are grateful to all of our
scientic divers for many years of data collection in the Caribbean region. Funding for
research conducted by SJP was provided by NOAAs Coral Reef Conservation Program
and NOAA’s Biogeography Branch. AK was supported by the Wildlife Conservation
Society and Simon Fraser University, Canada.
References
Agardy T, di Sciara GN, Christie P. 2011. Mind the gap: Addressing the shortcomings of marine pro-
tected areas through large scale marine spatial planning. Mar Policy. 35:226–232.
Aguirre-Gutiérrez J, Carvalheiro LG, Polce C, van Loon EE, Raes N, Reemer M, Biesmeijer JC. 2013.
Fit-for-purpose: Species distribution model performance depends on evaluation criteria—
Dutch hoveries as a case study. PLoS ONE. 8(5):e63708.
Allen TFH, Starr TB. 1982. Hierarchy: Perspectives for ecological complexity. Chicago: University of
Chicago Press.
Alvarez-Filip L, Dulvy NK, Gill JA, Cote IM, Watkinson AR. 2009. Flattening of Caribbean coral reefs:
Region-wide declines in architectural complexity. Proc Roy Soc Lond B Bio. 276:3019–3025.
Downloaded by [Simon Pittman] at 16:45 11 August 2014
232 Interrelationships between corals and sheries
Alvarez-Filip L, Gill JA, Dulvy NK, Perry AL, Watkinson AR, Côté IM. 2011. Drivers of region-wide
declines in architectural complexity on Caribbean reefs. Coral Reefs. 30(4):1051–1060. Available
from: http://dx.doi.org/10.1007/s00338-011-0795-6 via the Internet.
Aruajo MB, New M. 2007. Ensemble forecasting of species distributions. Trends Ecol Evol. 22:42–47.
Austin MP, Smith TM. 1989. A new model for the continuum concept. Vegetatio. 83:35–47.
Beger M, Possingham HP. 2008. Environmental factors that inuence the distribution of coral reef
shes: Modeling occurrence data for broad-scale conservation and management. Mar Ecol Prog
Ser. 361:1–13.
Boström C, Pittman SJ, Simenstad C, Kneib RT. 2011. Seascape ecology of coastal biogenic habitats:
Advances, gaps and challenges. Mar Ecol Prog Ser. 427:191–217.
Breiman L. 2001. Statistical modeling: The two cultures. Stat Sci. 16(3):199–215.
Brown CJ, Smith SJ, Lawton P, Anderson JT. 2011. Benthic habitat mapping: A review of progress
towards improved understanding of the spatial ecology of the seaoor using acoustic tech-
niques. Estuar Coast Shelf Sci. 92:502–520.
Christensen JD, Battista TA, Monaco ME, Klein CJ. 1997. Habitat suitability index modeling and GIS
technology to support habitat management: Pensacola Bay, Florida, case study. Report sub-
mitted to the Gulf of Mexico Program by National Oceanic and Atmospheric Administration,
Strategic Environmental Assessments Division, Silver Spring, MD.
Christensen JD, Jeffrey C, Monaco ME, Caldow C, Kendall MS, Appledoorn RS. 2003. Cross-shelf
habitat utilization patterns of reef shes in southwestern Puerto Rico. Gulf Carib Res. 14:9–27.
Costa B, Taylor CJ, Kracker L, Battista T, Pittman SJ. 2013. Mapping reef sh and the seascape: Using
acoustics and spatial modeling to guide coastal management. PLoS ONE, 9(1): e85555, p1–7.
Crowder L, Norse E. 2008. Essential ecological insights for marine ecosystem-based management
and marine spatial planning. Mar Policy. 32:772–778.
De’ath G. 2007. Boosted trees for ecological modeling and prediction. Ecology. 81:3178–3192.
Done TJ. 1992. Phase-shifts in coral-reef communities and their ecological signicance. Hydrobiologia.
247:121–132.
Džeroski S, Ženko B. 2004. Is combining classiers better than selecting the best one? Mach Learn.
54:255–273.
Elith J, Graham CH, Anderson RP, Dudik M, Ferrier S, Guisan A, Hijmans RJ, Huettmann F, Leathwick
JR, Lehmann A, etal. 2006. Novel methods improve prediction of species’ distributions from
occurrence data. Ecography. 29:129–151.
Elith J, Leathwick JR. 2009. Species distribution models: Ecological explanation and prediction across
space and time. Annu Rev Ecol Evol Syst. 40:677–697.
Elith J, Leathwick JR, Hastie T. 2008. A working guide to boosted regression trees. J Anim Ecol.
77:802–813.
Fielding AH, Bell JF. 1997. A review of methods for the assessment of prediction errors in conserva-
tion presence/absence models. Environ Conserv. 24:38–49.
Fischer J, Lindenmayer DB. 2006. Beyond fragmentation: The continuum model for fauna research
and conservation in human-modied landscapes. Oikos. 112:473–480.
Foley MM, Halpern BS, Micheli F, Armsby MH, Caldwell MR, Crain CM, Prahler E, Rohr N, SivasD,
Beck MW, etal. 2010. Guiding ecological principles for marine spatial planning. Mar Policy.
34:955–966.
Gardner TA, Cote IM, Gill JA, Grant A, Watkinson AR. 2003. Long-term region-wide declines in
Caribbean corals. Science. 301:958–960.
Goodman JA, Purkis SJ, Phinn SR. 2013. Coral reef remote sensing: A guide for mapping, monitoring
and management. Springer, Dordrecht, The Netherlands.
Green DH, Edmunds PJ, Carpenter RC. 2008. Increasing relative abundance of Porites astreoides on
Caribbean reefs mediated by an overall decline in coral cover. Mar Ecol Prog Ser. 359:1–10.
Grober-Dunsmore R, Pittman SJ, Caldow C, Kendall MA, Fraser T. 2009. Chapter 14: A landscape
ecology approach for the study of ecological connectivity across tropical marine seascapes. In:
Nagelkerken I, editor. Ecological connectivity of coral reef ecosystems. Springer, Dordrecht,
The Netherlands. p. 493–530.
Guisan A, Thuiller W. 2005. Predicting species distribution: Offering more than simple habitat mod-
els. Ecol Lett. 8:993–1009.
Downloaded by [Simon Pittman] at 16:45 11 August 2014
233Chapter twelve: Predictive mapping of coral reef sh species and communities
Harborne AR, Mumby PJ, Kappel CV, Dahlgren CP, Micheli F, Holmes KE, Brumbaugh DR. 2008.
Tropical coastal habitats as surrogates of sh community structure, grazing, and sheries
value. Ecol Appl. 18:1689–1701.
Hastie T, Tibshirani R, Friedman J. 2009. The elements of statistical learning. 2nd ed. Dordrecht:
Springer.
Hedley JD. 2013. Chapter 4: Hyperspectral applications. In: Goodman JA, Purkis SJ, Phinn SR,
editors. Remote sensing of coral reefs: A guide for mapping, monitoring and management.
Springer, Dordrecht, The Netherlands. p. 79–112.
Holling CS. 1992. Cross-scale morphology, geometry, and dynamics of ecosystems. Ecol Monogr.
62(4):447–502.
Hughes TP, Bellwood DR, Folke C, Steneck RS, Wilson J. 2005. New paradigms for supporting the
resilience of marine ecosystems. Trends Ecol Evol. 20:380–386.
Hunsaker CT, Friedl MA, Goodchild MF, Case TJ. 2001. Spatial uncertainty in ecology: Implications
for remote sensing and GIS applications. New York: Springer.
Hyrenbach KD, Forney KA, Dayton PK. 2000. Marine protected areas and ocean basin management.
Aquat Conserv. 10:437–458.
Jones GP, Syms C. 1998. Disturbance, habitat structure and the ecology of shes on coral reefs. Aust
J Ecol. 23:287–297.
Jones DL, Walter JF, Brooks EN, Serafy JE. 2010. Connectivity through ontogeny: Fish population
linkages among mangrove and coral reef habitats. Mar Ecol Prog Ser. 401:245–258.
Kendall MS, Christensen JD, Hillis-Starr Z. 2003. Multi-scale data used to analyze the spatial distri-
bution of French grunts, Haemulon avolineatum, relative to hard and soft bottom in a benthic
landscape. Environ Biol Fish. 66:19–26.
Klein CJ, Ban NC, Halpern BS, Beger M, Game ET, Grantham HS, Green A, Klein TJ, Kininmonth S,
Treml E, et al. 2010. Prioritizing land and sea conservation investments to protect coral reefs.
PLoS ONE. 5(8):e12431. Available from http://dx.doi.org/10.1371/journal.pone.0012431 via the
Internet.
Knudby A, Brenning A, LeDrew E. 2010a. Predictive mapping of reef sh species richness, diversity
and biomass in Zanzibar using IKONOS imagery and machine-learning techniques. Remote
Sens Environ. 114:1230–1241.
Knudby A, Brenning A, LeDrew E. 2010b. New approaches to modelling sh-habitat relationships.
Ecol Mod. 221:503–511.
Knudby A, Jupiter S, Roelfsema C, Lyons M, Phinn S. 2013. Mapping coral reef resilience indicators
using eld and remotely sensed data. Remote Sens. 5:1311–1334.
Knudby A, LeDrew E, Newman C. 2007. Progress in the use of remote sensing for coral reef biodiver-
sity studies. Prog Phys Geog. 31:421–434.
Kotliar NB, Wiens JA. 1990. Multiple scales of patchiness and patch structure: A hierarchical frame-
work for the study of heterogeneity. Oikos. 59:253–260.
Leathwick J, Moilanen A, Francis M, Elith J, Taylor P, Julian K, Hastie T. 2008. Novel methods for the
design and evaluation of marine protected areas in offshore waters. Conserv Lett. 1:91–102.
Levin SA. 1992. The problem of pattern and scale in ecology. Ecology. 73(6):1943–1967.
Lourie SA, Vincent ACJ. 2004. Using biogeography to help set priorities in marine conservation.
Conserv Biol. 18:1004–1020.
Marmion M, Parviainen M, Luoto M, Heikkinen RK, Thuiller W. 2009. Evaluation of consensus
methods in predictive species distribution modelling. Divers Distrib. 15:59–69.
Maynard JA, Marshall PA, Johnson JE, Harman S. 2010. Building resilience into practical conserva-
tion: Identifying local management responses to global climate change in the Southern Great
Barrier Reef. Coral Reefs. 29:381–391.
McArthur MA, Brooke BP, Przeslawski R, Ryan DA, Lucieer VL, Nichol S, McCallum AW, Mellin C,
Cresswell ID, Radke LC, etal. 2010. On the use of abiotic surrogates to describe marine benthic
biodiversity. Estuar Coast Shelf Sci. 88:21–32.
McClanahan TR, Donner SD, Maynard JA, MacNeil MA, Graham NAJ, Maina J, Baker A, Jahson
BAI, Beger M, Campbell SJ, etal. 2012. Prioritizing key resilience indicators to support coral
reef management in a changing climate. PLoS ONE. 7(8):e42884. Available from http://dx.doi.
org/10.1371/journal.pone.0042884 via the Internet.
Downloaded by [Simon Pittman] at 16:45 11 August 2014
234 Interrelationships between corals and sheries
McGarigal K, Cushman SA. 2005. The gradient concept of landscape structure. In: Wiens J, Moss M,
editors. Issues and perspectives in landscape ecology. Cambridge: Cambridge University Press.
p. 112–119.
Mellin C, Bradshaw CJA, Meekan MG, Caley MJ. 2010. Environmental and spatial predictors of spe-
cies richness and abundance in coral reef shes. Global Ecol Biogeogr. 19:212–222.
Mellin C, Delean S, Caley J, Edgar G, Meekan M, Pitcher R, Przeslawski R, Williams A, BradshawC.
2011. Effectiveness of biological surrogates for predicting patterns of marine biodiversity:
Aglobal meta-analysis. PLoS ONE. 6:e20141. Available from http://dx.doi.org/10.1371/jour-
nal.pone.0020141 via the Internet.
Mellin C, Parrott L, Andrefouet S, Bradshaw CJA, MacNeil MA, Caley MJ. 2012. Multi-scale marine
biodiversity patterns inferred efciently from habitat image processing. Ecol Appl. 22:792–803.
Mora C, Andrefouet S, Costello MJ, Kranenburg C, Rollo A, Veron J, Gaston KJ, Myers RA. 2006.
Coral reefs and the global network of marine protected areas. Science. 312:1750–1751.
Mumby PJ. 2009. Phase shifts and the stability of macroalgal communities on Caribbean coral reefs.
Coral Reefs. 28:761–773.
Mumby PJ, Edwards AJ, Arias-Gonzalez JE, Lindeman KC, Blackwell PG, Gall A, Gorczynska MI,
Harborne AR, Pescod CL, Renken H, etal. 2004. Mangroves enhance the biomass of coral reef
sh communities in the Caribbean. Nature. 427:533–536.
Nagelkerken I, Roberts CM, van der Velde G, Dorenbosch M, van Riel MC, de la Morinere EC,
Nienhuis PH. 2002. How important are mangroves and seagrass beds for coral-reef sh? The
nursery hypothesis tested on an island scale. Mar Ecol Prog Ser. 244:299–305.
NMFS (National Marine Fisheries Service). 1996. Magnuson-Stevens Fishery Conservation and
Management Act: As amended through October 11, 1996. U.S. Department of Commerce,
National Oceanic and Atmospheric Administration, National Marine Fisheries Service, NOAA
Technical Memorandum NMFS-F/SPO-23. Silver Spring, MD: NOAA.
Olden JD, Lawler JJ, Poff NL. 2008. Machine learning methods without tears: A primer for ecologists.
Quart Rev Biol. 83:171–193.
Orth RJ, Carruthers TJB, Dennison WC, Duarte CM, Fourqurean JW, Heck KL, Hughes AR, Kendrick
GA, Kenworthy WJ, Olyarnik S, etal. 2006. A global crisis for seagrass ecosystems. Bioscience.
56:987–996.
Paddack MJ, Reynolds JD, Aguilar C, Appeldoorn RS, Beets J, Burkett EW, Chittaro PM, Clarke K,
Esteves R, Fonseca AC, etal. 2009. Recent region-wide declines in Caribbean reef sh abun-
dance. Curr Biol. 19:590–595.
Pandol JM, Jackson JBC, Baron N, Bradbury RH, Guzman H, Hughes TP, Micheli F, Ogden J,
Possingham H, Kappel CV, et al. 2005. Are U.S. coral reefs on the slippery slope to slime?
Science. 307:1725–1726.
Parrish JD. 1989. Fish communities of interacting shallow water habitats in tropical oceanic regions.
Mar Ecol Prog Ser. 58:143–160.
Pearce J, Ferrier S. 2000. Evaluating the predictive performance of habitat models developed using
logistic regression. Ecol Model. 133:225–245.
Phillips SJ, Anderson RP, Schapire RE. 2006. Maximum entropy modeling of species geographic dis-
tributions. Ecol Model. 190:231–259.
Pimm SL. 1984. The complexity and stability of ecosystems. Nature. 307:321–326.
Pittman SJ, Brown KA. 2011. Multi-scale approach for predicting sh species distributions across
coral reef seascapes. PLoS ONE. 6(5):e20583. Available from http://dx.doi.org/10.1371/jour-
nal.pone.0020583 via the Internet.
Pittman SJ, Christensen J, Caldow C, Menza C, Monaco M. 2007a. Predictive mapping of sh species
richness across shallow-water seascapes of the U.S. Caribbean. Ecol Model. 204:9–21.
Pittman SJ, Hile SD, Caldow C, Monaco ME. 2007b. Using seascape types to explain the spatial pat-
terns of sh using mangroves in Puerto Rico. Mar Ecol Prog Ser. 348:273–284.
Pittman SJ, Costa B, Battista TA. 2009. Using Lidar bathymetry and boosted regression trees to pre-
dict the diversity and abundance of sh and corals. J Coastal Res. S(53):27–38.
Pittman SJ, Connor DW, Radke L, Wright DJ. 2011a. Application of estuarine and coastal classica-
tions in marine spatial management. In: Wolanski E, McLusky DS, editors. Treatise on estuarine
and coastal science. Vol. 1. Waltham, MA: Academic Press. p. 163–205.
Downloaded by [Simon Pittman] at 16:45 11 August 2014
235Chapter twelve: Predictive mapping of coral reef sh species and communities
Pittman SJ, Costa B, Jeffrey CFG, Caldow C. 2011b. Importance of seascape complexity for resilient
sh habitat and sustainable sheries. In: Proceedings of the 63rd Gulf and Caribbean Fisheries
Institute 1–5 November, 2010 San Juan, Puerto Rico. p. 420–426.
Pittman SJ, Costa B, Wedding LM. 2013. Chapter 6: LiDAR applications. In: Goodman JA, Purkis SJ,
Phinn SR, editors. Remote sensing of coral reefs: A guide for mapping, monitoring and man-
agement. Springer, Dordrecht, The Netherlands. p. 145–174.
Pittman SJ, McAlpine CA. 2003. Movements of marine sh and decapod crustaceans: Process, theory
and application. Adv Mar Biol. 44:205–294.
Polidoro BA, Carpenter KE, Lorna C, Duke NC, Ellison AM, Ellison JC, Farnsworth EJ, Fernando
ES, Kathiresan K, Koedam NE, et al. 2010. The loss of species: Mangrove extinction risk
and geographic areas of global concern. PLoS ONE. 5:e10095. Available from http://dx.doi.
org/10.1371/journal.pone.0010095 via the Internet.
R Core Development Team. 2012. R: A language and environment for statistical computing. Vienna,
Austria: R Foundation for Statistical Computing. Available from http://www.R-project.org/
via the Internet.
Roberts CM, Andelman S, Branch G, Bustamante RH, Castilla JC, Dugan J, Halpern BS, Lafferty KD,
Leslie H, Lubchenco J, etal. 2003. Ecological criteria for evaluating candidate sites for marine
reserves. Ecol Appl. 13:S199–S214.
Roff JC, Taylor ME, Laughren J. 2003. Geophysical approaches to the classication, delineation and
monitoring of marine habitats and their communities. Aquatic Conserv. 13:77–90.
Rowlands G, Purkis S, Riegl B, Metsamaa L, Bruckner A, Renaud P. 2012. Satellite imaging coral reef
resilience at regional scale. A case-study from Saudi Arabia. Mar Pollut Bull. 64:1222–1237.
Rubec PJ, Bexley JCW, Norris H, Coyne MS, Monaco ME, Smith SG, Ault JS. 1999. Suitability mod-
eling to delineate habitat essential to sustainable sheries. In: Benaka LR, editor. Fish habitat:
Essential sh habitat and restoration. Am Fish Soc Symp. 22:108–133.
Rubec PJ, Christensen JD, Arnold WS, Norris H, Steele P, Monaco ME. 1998a. GIS and modeling:
Coupling habitats to Florida sheries. J Shellsh Res. 17(5):1451–1457.
Rubec PJ, Coyne MS, McMichael Jr RH, Monaco ME. 1998b. Spatial methods being developed in
Florida to determine essential sh habitat. Fisheries. 23(7):21–25.
Rubec PJ, O’Hop J. 1996. GIS applications for sheries and coastal resources management. Gulf
States Marine Fisheries Commission Publication No. 43. Ocean Springs, MS: Gulf States Marine
Fisheries Commission.
Schneider DC. 2001. The rise of the concept of scale in ecology. BioScience. 51(7):545–553.
Souter DW, Linden O. 2000. The health and future of coral reef ecosystems. Ocean Coast Manage.
43:657–688.
Stallings CD. 2009. Fishery-independent data reveal negative effect of human population density
on Caribbean predatory sh communities. PLoS ONE. 4:e5333. Available from http://dx.doi.
org/10.1371/journal.pone.0005333 via the Internet.
Ward TJ, Vanderklift MA, Nicholls AO, Kenchington RA. 1999. Selecting marine reserves using habi-
tats and species assemblages as surrogates for biological diversity. Ecol Appl. 9:691–698.
Wedding LM, Friedlander AM, McGranaghan M, Yost RS, Monaco ME. 2008. Using bathymetric
lidar to dene nearshore benthic habitat complexity: Implications for management of reef sh
assemblages in Hawaii. Remote Sens Environ. 112:4159–4165.
Wedding L, Lepczyk C, Pittman SJ, Friedlander A, Jorgensen S. 2011 Quantifying seascape structure:
Extending terrestrial spatial pattern metrics to the marine realm. Mar Ecol Prog Ser. 427:219–232.
West J, Salm R. 2003. Resistance and resilience to coral bleaching: Implications for coral reef conser-
vation and management. Conserv Biol. 17:956–967.
Wozencraft JM, Park JY. 2013. Integrated LiDAR and hyperspectral. In: Goodman JA, Purkis SJ,
Phinn SR, editors. Coral reef remote sensing: A guide for mapping, monitoring and manage-
ment. Dordrecht: Springer. p. 175–191.
Zawada DG. 2011. Reef topographic complexity. In: Hopley D, editor. Encyclopedia of modern
coral reefs: Structure, form, and process. Encyclopedia of Earth Sciences. Dordrecht: Springer.
p.902–906.
Downloaded by [Simon Pittman] at 16:45 11 August 2014
Downloaded by [Simon Pittman] at 16:45 11 August 2014
... This level of variability is common in coral reef fisheries and managers are in need of simple, yet meaningful metrics to guide management actions (Nash and Graham 2016). While species differences in terms of life-span and age at maturity will influence timelines of recovery, habitats characteristics which support targeted fish assemblages can be identified and used to select priority areas for management (Pittman and Knudby 2014). In addition to their practicality, a significant advantage of using assemblage level metrics for spatial modeling is the low prevalence of null values which improves model performance (Wisz et al. 2008). ...
... Spatially comprehensive, continuous input and output datasets eliminate the need for the simplifying assumptions common in traditional approaches that can increase uncertainty when results are extrapolated across the area of interest. Predictive mapping fills gaps in survey coverage, allows for testing of management scenarios, and provides spatially comprehensive information for managers including estimates of prediction precision (Pittman and Knudby 2014). Finally, the approach is flexible so that it can be applied anywhere where demersal fish populations are targeted, and any response variable derived from fish survey data can be modeled depending on the research or management question. ...
Article
Full-text available
To design effective marine reserves and support fisheries, more information on fishing patterns and impacts for targeted species is needed, as well as better understanding of their key habitats. However, fishing impacts vary geographically and are difficult to disentangle from other factors that influence targeted fish distributions. We developed a set of fishing effort and habitat layers at high resolution and employed machine learning techniques to create regional-scale seascape models and predictive maps of biomass and body length of targeted reef fishes for the main Hawaiian Islands. Spatial patterns of fishing effort were shown to be highly variable and seascape models indicated a low threshold beyond which targeted fish assemblages were severely impacted. Topographic complexity, exposure, depth, and wave power were identified as key habitat variables which influenced targeted fish distributions and defined productive habitats for reef fisheries. High targeted reef fish biomass and body length were found in areas not easily accessed by humans, while model predictions when fishing effort was set to zero showed these high values to be more widely dispersed among suitable habitats. By comparing current targeted fish distributions with those predicted when fishing effort was removed, areas with high recovery potential on each island were revealed, with average biomass recovery of 517% and mean body length increases of 59% on Oahu, the most heavily fished island. Spatial protection of these areas would aid recovery of nearshore coral reef fisheries. This article is protected by copyright. All rights reserved.
... 1. Important physical and ecological patterns and processes (e.g., basic habitat distributions and critical habitat functions) that occur in the planning area, including their response to changing conditions; 2. Ecological condition and relative ecological importance or values of areas within the planning area, using regionally-developed evaluation and prioritization schemes; 3. Economic and environmental benefits and impacts of ocean, coastal, and Great Lakes uses in the region; 4. Relationships and linkages within and among regional ecosystems, including neighboring regions both within and outside the planning area and the impacts of anticipated human uses on those connections; 5. Spatial distribution of, and conflicts and compatibilities among, current and emerging ocean uses in the area; 6. Important ecosystem services in the area, and their vulnerability or resilience to the effects of human uses, natural hazards, and global climate change; 7. Contributions of existing placed-based management measures and authorities; and 8. Future requirements of existing and emerging ocean, coastal, and Great Lakes uses situations with very sparse data, or where only specific components of the environment are being considered. This is usually achieved through spatial predictive modeling using techniques such as machine-learning algorithms and geostatistical modeling to fill data gaps [74,77]. An important focus of BAF is the quantification of uncertainty arising from data deficiencies and data processing. ...
... A variety of statistical predictive modeling techniques ranging from geostatistical interpolation of data (kriging) to statistical models that use predictor variables (e.g., species distribution modeling, machine-learning) to dynamic ecological models, or a hybrid of these approaches, have proven to be rapid, reliable and cost-effective for filling data gaps [82,18,40,77]. In general, attempts to "scale up", or interpolate, using simple geostatistical or other purely spatial interpolation approaches applied to relatively fine scale and sparse survey data often results in high spatial error [42]. ...
Article
Full-text available
This paper presents the Biogeographic Assessment Framework (BAF), a decision support process for marine spatial planning (MSP), developed through two decades of close collaborations between scientists and marine managers. Spatial planning is a considerable challenge for marine stewardship agencies because of the need to synthesize information on complex socio-ecological patterns across geographically broad spatial scales. This challenge is compounded by relatively short time-frames for implementation and limited financial and technological resources. To address this pragmatically, BAF provides a rapid, flexible and multi-disciplinary approach to integrate geospatial information into formats and visualization tools readily useable for spatial planning. Central to BAF is four sequential components: (1) Planning; (2) Data Evaluation; (3) Ecosystem Characterization; and (4) Management Applications. The framework has been applied to support the development of several marine spatial plans in the United States and Territories. This paper describes the structure of the BAF framework and the associated analytical techniques. Two management applications are provided to demonstrate the utility of BAF in supporting decision making in MSP.
... It is rarely known a priori which is likely to be most ecologically meaningful and the preferred data type is rarely explicitly stated in the research questions. As such, an exploratory multi-scale and multi-model approach may better capture the complex assemblage-seascape associations, especially with diverse assemblages with diverse life-history characteristics and mobility (Pittman & Knudby, 2014). ...
Article
Geographical patterning of fish diversity across coral reef seascapes is driven by many interacting environmental variables operating at multiple spatial scales. Identifying suites of variables that explain spatial patterns of fish diversity is central to ecology and informs prioritization in marine conservation, particularly where protection of the highest biodiversity coral reefs is a primary goal. However, the relative importance of conventional within-patch variables versus the spatial patterning of the surrounding seascape is still unclear in the ecology of fishes on coral reefs. A multi-scale seascape approach derived from landscape ecology was applied to quantify and examine the explanatory roles of a wide range of variables at different spatial scales including: (i) within-patch structural attributes from field data (5 × 1 m2 sample unit area); (ii) geometry of the seascape from sea-floor maps (10–50 m radius seascape units); and wave exposure from a hydrodynamic model (240 m resolution) for 251 coral reef survey sites in the US Virgin Islands. Non-parametric statistical learning techniques using single classification and regression trees (CART) and ensembles of boosted regression trees (TreeNet) were used to: (i) model interactions; and (ii) identify the most influential environmental predictors from multiple data types (diver surveys, terrain models, habitat maps) across multiple spatial scales (1–196,350 m2). Classifying the continuous response variables into a binary category and instead predicting the presence and absence of fish species richness hotspots (top 10% richness) increased the predictive performance of the models. The best CART model predicted fish richness hotspots with 80% accuracy. The statistical interaction between abundance of living scleractinian corals measured by SCUBA divers within 1 m2 quadrats and the topographical complexity of the surrounding sea-floor terrain (150 m radius seascape unit) measured from a high-resolution terrain model best explained geographical patterns in fish richness hotspots. The comparatively poor performance of models predicting continuous variability in fish diversity across the seascape could be a result of a decoupling of the diversity-environment relationship owing to structural degradation leading to a widespread homogenization of coral reef structure.
Chapter
Full-text available
A new paradigm has emerged for management of coral reefs in an era of changing climate – managing for resilience. A fundamental need for such management to be effective is our ability to measure and map coral reef resilience. We review the resilience concept and factors that may make a coral reef more or less resilient to climate-driven impacts, and focus on recent advances in a trio of technologies: remote sensing, spatial distribution modeling, and ecosystem simulation; that promise to improve our ability to quantify coral reef resilience across reefs. Remote sensing allows direct mapping of several ecosystem variables that influence reef resilience, including coral and algal cover, as well as a range of coral reef stressors, as exemplified by three case studies. Spatial distribution modeling allows exploitation of statistical relationships between mappable environmental variables and factors that influence resilience but which cannot be mapped directly, such as herbivore biomass. Ecosystem simulation modeling allows predictions to be made for the trajectories of reef ecosystems, given their initial state, interactions between ecosystem components, and a realistic current and future disturbance regime. Together, these technologies have the potential to allow production of coral reef resilience maps. We conclude with a fourth case study that illustrates integration of resilience maps into a multi-objective decision support framework. Implementation of the managing for resilience paradigm is still in its infancy, but the rapidly advancing technologies reviewed here can provide the resilience maps needed for its successful operationalization.
Article
Full-text available
We compare the value of using habitat categories and species assemblages as surrogates for marine biological diversity in the context of choosing a set of representative areas for a marine reserve network. Habitat categories were based on interpretation of aerial photographs and maps, and on local knowledge. Species assemblages were created from comprehensive survey data on 977 taxa (mainly species), derived from an intensive three-year study of a temperate marine embayment, and classified into plant, fish, and invertebrate assemblages. Reserves were selected using a heuristic iterative algorithm to simulate a marine reserve network based on 10-80% representation of each surrogate. The effectiveness of each surrogate was evaluated by comparing the number of taxa that would be coincidentally included in each simulated reserve for the bay. Areas selected to represent 10% or 20% of the surrogates were best chosen using fish or invertebrate assemblages, because by spatial coincidence, they included 60-80% of all available taxa. However, areas selected to represent ≥40% of the surrogates were generally best derived from habitat categories, because they included ≥93% of all available taxa. Plant assemblages were generally poor surrogates for overall species richness. These findings suggest that habitat-level surrogates may be a highly cost-effective method for initial identification of high-priority areas to manage marine diversity of coastal ecosystems.
Article
Full-text available
Mangrove forests are one of the world's most threatened tropical ecosystems with global loss exceeding 35% (ref. 1). Juvenile coral reef fish often inhabit mangroves, but the importance of these nurseries to reef fish population dynamics has not been quantified. Indeed, mangroves might be expected to have negligible influence on reef fish communities: juvenile fish can inhabit alternative habitats and fish populations may be regulated by other limiting factors such as larval supply or fishing. Here we show that mangroves are unexpectedly important, serving as an intermediate nursery habitat that may increase the survivorship of young fish. Mangroves in the Caribbean strongly influence the community structure of fish on neighbouring coral reefs. In addition, the biomass of several commercially important species is more than doubled when adult habitat is connected to mangroves. The largest herbivorous fish in the Atlantic, Scarus guacamaia, has a functional dependency on mangroves and has suffered local extinction after mangrove removal. Current rates of mangrove deforestation are likely to have severe deleterious consequences for the ecosystem function, fisheries productivity and resilience of reefs. Conservation efforts should protect connected corridors of mangroves, seagrass beds and coral reefs.
Article
Full-text available
To fill the need for information relating fishery species to habitat and anthropogenic effects on marine ecosystems, the Florida Marine Research Institute created the Marine Resources Geographic Information System (MRGIS). This comprehensive, coastwide database for research and management of fisheries provides spatial information for siting of clam aquaculture, hard-bottom mapping, and fishery exclusion zones. Fisheries-independent monitoring for Florida estuaries and habitat suitability modeling are also being used to create predicted distribution maps of fish and invertebrate species. Four case histories are summarized, designed to define relationships between habitat and marine fish and invertebrate species in Florida.
Article
Full-text available
Spatial Modeling of Fish Habitat Suitability in Florida Estuaries Abstract Spatial habitat suitability index (HSI) models were developed by a group of collaborating scientists to predict species relative abundance distributions by life stage and season in Tampa Bay and Charlotte Harbor, Florida. Habitat layers and abundance-based suitability index (Si) values were derived from fishery-independent survey data and used with HSI models that employed geographic information systems. These analyses produced habitat suitability maps by life stage and season in the two estuaries for spotted seatrout (Cynoscion nebulosus), bay anchovy (Anchoa mitchilli), and pinfish (Lagodon rhomboides). To verify the reliability of the HSI models, mean catch rates (CPUEs) were plotted across four HSI zones. Analyses showed that fish densities increased from low to optimum zones for the majority of species life stages and seasons examined, particularly for Charlotte Harbor. A reciprocal transfer of Si values between estuaries was conducted to test whether HSI modeling can be used to predict species distributions in estuaries lacking fisheries-independent monitoring. The similarity of Si functions used with the HSI models accounts for the high similarity of predicted seasonal maps for juvenile pinfish and juvenile bay anchovy in each estuary. The dissimilarity of Si functions input into HSI models can account for why other species life stages had dissimilar predicted maps.
Article
Full-text available
In June 2000, the National Ocean Service and University of Puerto Rico initiated a long-term reef-fish-monitoring program in La Parguera, Puerto Rico. Objectives of this ongoing work are to: 1) develop spatially-explicit estimates of reef fish habitat utilization patterns to aid in defining essential habitats, and 2) provide a quantitative and ecologically sound foundation to delineate marine reserve boundaries. Central to this effort are recently completed digital and georeferenced benthic habitat maps for the near-shore waters of Puerto Rico. The GIS-based map served as a framework for development of a spatially stratified reef-fish-monitoring program across the shelf. Simultaneous collections of fish size and abundance data, and micro-scale habitat distribution and quality data were taken along a 25 x 4 m transect for each monitoring station. Sampling included coral reef, mangrove, and seagrass habitats within three cross-shelf zones unique to the insular shelf of La Parguera (inner lagoon, outer lagoon, and bank-shelf). A total of 106 stations were surveyed during the first year of sampling. Over 50,000 fishes, representing 123 species and 36 families were counted. Analyses showed clear patterns of habitat utilization across the seascape, and ontogenetic shifts in habitat selection within some species. Results also indicated that habitat type was more important than cross-shelf location in determining spatial patterns among reef fishes in the study area. Mesoscale spatially-explicit logistic models were developed to estimate distribution and expected density of some species among habitats.
Book
Remote sensing stands as the defining technology in our ability to monitor coral reefs, as well as their biophysical properties and associated processes, at regional to global scales. With overwhelming evidence that much of Earth’s reefs are in decline, our need for large-scale, repeatable assessments of reefs has never been so great. Fortunately, the last two decades have seen a rapid expansion in the ability for remote sensing to map and monitor the coral reef ecosystem, its overlying water column, and surrounding environment. Remote sensing is now a fundamental tool for the mapping, monitoring and management of coral reef ecosystems. Remote sensing offers repeatable, quantitative assessments of habitat and environmental characteristics over spatially extensive areas. As the multi-disciplinary field of coral reef remote sensing continues to mature, results demonstrate that the techniques and capabilities continue to improve. New developments allow reef assessments and mapping to be performed with higher accuracy, across greater spatial areas, and with greater temporal frequency. The increased level of information that remote sensing now makes available also allows more complex scientific questions to be addressed. As defined for this book, remote sensing includes the vast array of geospatial data collected from land, water, ship, airborne and satellite platforms. The book is organized by technology, including: visible and infrared sensing using photographic, multispectral and hyperspectral instruments; active sensing using light detection and ranging (LiDAR); acoustic sensing using ship, autonomous underwater vehicle (AUV) and in-water platforms; and thermal and radar instruments. Emphasis and Audience This book serves multiple roles. It offers an overview of the current state-of-the-art technologies for reef mapping, provides detailed technical information for coral reef remote sensing specialists, imparts insight on the scientific questions that can be tackled using this technology, and also includes a foundation for those new to reef remote sensing. The individual sections of the book include introductory overviews of four main types of remotely sensed data used to study coral reefs, followed by specific examples demonstrating practical applications of the different technologies being discussed. Guidelines for selecting the most appropriate sensor for particular applications are provided, including an overview of how to utilize remote sensing data as an effective tool in science and management. The text is richly illustrated with examples of each sensing technology applied to a range of scientific, monitoring and management questions in reefs around the world. As such, the book is broadly accessible to a general audience, as well as students, managers, remote sensing specialists and anyone else working with coral reef ecosystems.
Chapter
This book is the second of two volumes in a series on terrestrial and marine comparisons, focusing on the temporal complement of the earlier spatial analysis of patchiness and pattern (Levin et al. 1993). The issue of the relationships among pattern, scale, and patchiness has been framed forcefully in John Steele’s writings of two decades (e.g., Steele 1978). There is no pattern without an observational frame. In the words of Nietzsche, “There are no facts… only interpretations.”