The 2005 Coral-bleaching Event
Roatan (Honduras): Use of Pseudoinvariant Features (PIFs) in Satellite
Assessments
G.P. Rowlands
S.J. Purkis
B.M. Riegl
INTRODUCTION
Radiometric normalisation is a necessary
precursor to detecting coral-bleaching from a time
series of imagery. Using IKONOS and QuickBird,
the pseudo-invariant-feature (PIF) approach was
assessed, and found that small subsets and landbased PIFs lead to erroneous normalisation. In
comparison, favourable results were achieved
by using benthic sand and deep water subsets,
coupled to a sun-deglint process. Furthermore,
it was found the more traditional strategy of
‘difference imaging’ to be compromised by slight
spatial alignment errors between image sets. Our
alternative approach, based on spectral radiance,
ably discerned significant brightening of areas of
seafloor populated by dense stands of Acropora,
corroborating the occurrence of a documented
bleaching event.
First recorded in the Caribbean region as early
as 1911 (Mayer, 1914), coral reef bleaching
has been frequently reported over recent
decades (Aronson et al., 2000; Berkelmans et
al., 2004; Andréfouët et al., 2002; Goreau and
Hayes, 2005). Bleaching is a stress response
whereby corals dissociate from their symbiotic
algae (zooxanthellae), leading to a whitened
appearance (Goreau and Hayes, 2005); this
weakens or kills corals. Bleaching is primarily
associated with sea surface temperature (SST)
anomalies (Berkelmans and Willis, 1999), though
salinity, water opacity, UV radiation and nutrient
levels may also play a role. Recovery depends on
an event’s severity and duration (Grimsditch and
Salm, 2005). Acropora cervicornis is an important
Caribbean coral species (Lamarck, 1816), that
has shown widespread population collapse, and
is considered endangered in many regions (Precht
et al., 2004). Though factors other than bleaching
are commonly cited as the agents of decline,
it is among the most bleaching-prone genera
(Berkelmans et al., 2004) and since it tends to
cover large areas in mono- or paucispecific stands,
its bleaching has the potential to generate a clear
optical signature (Yamano and Tamura, 2004).
G.P. Rowlands
S.J. Purkis
B.M. Riegl
National Coral Reef Institute, Oceanographic
Center
Nova Southeastern University
8000 N. Ocean Drive
Dania, FL 33004, USA
[email protected]
SPATIAL SCIENCE
Remote sensing has proven to be a cost
effective approach to monitoring and change
detection (Mumby et al., 1997; Hochberg et
al., 2003; Palandro et al., 2003b). Using highresolution satellite imagery Purkis and Riegl
(2005) were able to determine spatial and
temporal dynamics of Arabian Gulf coral
99
Vol. 53, No. 1, June 2008
mortality caused by bleaching. Central to image
based analysis is the discrimination of substrate
types based on their unique spectral reflectance
characteristics (Lubin et al., 2001). While sand
is relatively easily distinguished from healthy
coral, determining different reef components is
harder due to similarity of spectral reflectance
characteristics. However, reflectance may
increase two-fold from unbleached to bleached
states (Holden and LeDrew 1998; Clark et al.,
2000) and thus suggests that bleaching should
indeed be detectable. Identification of bleached
coral has however proven difficult as the
coral skeleton has similar spectral reflectance
properties to carbonate rock and sand in all but
the most heavily bleached states.
Elvidge et al., (2004) provided the first
satisfactory account of satellite-based bleaching
detection, using a pseudo-invariant features (PIF)
based approach to normalise a bleached image
to a reference image. There has however been a
notable lack of validation of this technique. This
research addresses this imbalance through an
analysis of another bleaching event, in a different
geographical area and reef regime. Multispectral
QuickBird (2.4m pixel) and IKONOS (4m pixel)
imagery are used in concert, highlighting issues
that may arise when moving between such
sensors, and examine some of the different
methods available when normalising image
radiometry to facilitate bleaching detection.
Moreover, through an analysis of bleaching
at a site of high importance for A. cervicornis
(Keck et al., 2005), this study has important
implications for management of this species and
the wider ecological debate.
Keywords: Coral, bleaching, bleaching detection,
remote sensing, Quickbird, IKONOS.
METHODS
Study Area
The island of Roatan is situated ~55km north of
mainland Honduras, in the Western Caribbean
(16°N, 86°W) (Figure 1). Roatan has a hot and
humid climate, with most rainfall falling in the
months of October-January (D’Croz et al., 1998).
Easterly trade winds (24-42mph) predominate in
summer, while strong winds are common from
the north/west in winter. Direct hurricane hits
occur on a decadal basis, with Hurricane Mitch
(1998) and Fifi (1974) notable in recent times.
Major currents flow west-north-west, though
change locally on a seasonal basis. Reef-based
tourism is a major contributor to the island’s
economy. Rising visitor numbers coupled to the
Figure 1. Location of the offshore Smith Bank and Cordelia Shoal study site relative to Roatan Island, The Bay Islands, and
mainland Honduras.
Vol. 53, No. 1, June 2008
100
SPATIAL SCIENCE
Figure 2. a) 2004 distribution of Acropora cervicornis (Purkis et al., 2006), along with position of pre bleaching ground
survey overlaying IKONOS 2003 image; b) 2005 bleaching event at Smith Bank overlaying QuickBird image, showing
position of photographic survey points and high density A. cervicornis pixel subsets used for spectral discrimination analysis;
c) Cumulative percentage cover of different benthic classes, based on 10 points per photographic record (n=400) showing
also the incidence of bleaching.
SPATIAL SCIENCE
101
Vol. 53, No. 1, June 2008
expansion of tourism in the north has lead to the
widespread degradation of fringing reef systems
(Keck, 2004). In contrast, Smith Bank and Cordelia
Shoal (Figure 1) are held in high regard for their
reef health and A. cervicornis stands of unrivalled
in size in the Caribbean (Keck et al., 2005). Rising
out of deep water (>150m) ~1.5km off the
south-west tip of the main island the banks are
unsheltered and inaccessible for much of the year.
They receive minimal direct anthropogenic impact
from either diving or fishing.
Field Recordings
Field recordings are detailed in Figures 2
and 3. From the extensive 2004 survey and
methodology of Purkis et al., (2006), the
distribution of A. cervicornis prior to the
2005 bleaching event was identified. Part of
a Caribbean wide phenomenon, bleaching
occurred on Roatan from September through
early November 2005. A photographic survey
was carried out in the bleaching aftermath
on the 26 th November. Due to inclement
weather conditions at the time of this survey,
boat access to the homogenous A. cervicornis
patch at the centre of the bank was prohibited.
Survey concentrated on the mixed coral
assemblage around the periphery, allowing a
good cross species bleaching assessment to
be made. Mortality due to bleaching could
be identified from areas of bare skeleton with
recent filamentous algae growth. Sampling point
position was recorded using a global positioning
system (GPS), and depth recorded using a dive
computer. Coral species cover and bleaching
extent were assessed by point count. As shown
in Figure 2c, bleaching was largely restricted to
A. cervicornis. Small isolated disease and focal
feeding scars were also noted in a few images.
Bleaching was recorded at depths ranging from
2m to 8m and a mean depth of 4.87m (stdv
1.61).
Image acquisition and processing
Pre-bleaching IKONOS imagery (referenced
hereafter by year: 2000, 2001, 2003) were
acquired with standard radiometric processing,
along with a QuickBird image (2005) taken during
the period of maximum mainland bleaching
(Table 1). Images were collected under favourable
low cloud cover and water turbidity conditions,
though mild sun glint was apparent, particularly
in the 2001 image. The QuickBird image was
converted from digital number (DN) into units of
Figure 3. a) 2004 pre-bleaching images showing the vast, dense, homogeneous, and healthy Acropora cervicornis habitat
at Smith Bank and Cordelia Shoal; b) Images taken one month after the 2005 A. cervicornis bleaching at Smith Bank. A
short time had elapsed since the height of the bleaching event; mortality from bleaching was identified as non-living tissue
with recent filamentous algae growth.
Vol. 53, No. 1, June 2008
102
SPATIAL SCIENCE
Image
2000
2001
2003
2005
Sensor type
Cloud cover
IKONOS
10/03/2000
15:56
<1.0%
IKONOS
24/03/2001
16:19
2.0%
IKONOS
29/12/2003
16:27
2.0%
QuickBird
14/09/2005
16:37
<1.0%
Pixel size (m)
4.0
4.0
4.0
2.4
Date
Time (GMT*)
*local time = -6hrs GMT
Table 1. Study image attributes.
at-sensor-radiance L (μW cm-2 nm-1 sr-1), and the
radiometric conversion formula of Taylor (2006)
and coefficients of Peterson (2001) adapted to
correct IKONOS DN to at-sensor radiance.
Image registration
Comparison of spectral radiance/reflectance
values between images requires pixels in the same
spatial resolution and geographical alignment.
The 2003 IKONOS image was used as the
master image since it suffered least from sunglint, and was in the middle of the temporal
series. All images were registered using 1st degree
polynomial warping between stable, clearly
identified ground control points (GCPs) (e.g. rock
and reef features). Nearest neighbour re-sampling
was used to register pixels and convert QuickBird
imagery to the 4m resolution. In choosing GCPs,
a trade-off occurs between accurate position,
and good across-scene coverage to prevent
model bias. The best models used 105, 100,
and 97 GCPs for the 2000, 2001 and 2005
images respectively, with a model root mean
square (RMS) error of 1.09, 0.96 and 1.06 pixels
(3.84-4.36m). Individual point RMS errors ranged
from 0-2.54 pixels (0-10m). Co-registration was
also assessed through visual overlay.
Phase 1 Normalisation
In order to assess temporal change, radiance
or reflectance values of like features should be
comparable. In line with Elvidge et al. (2004),
analysis and normalisation was restricted
to multispectral bands. Despite conversion to
commensurate units of radiance, factors including
varying atmospheric, water surface and water
column properties, as well as solar angle still
resulted in scene differences. Two approaches
were initially assessed:
SPATIAL SCIENCE
• Images were corrected for atmospheric effects
using the Fast Line-of-Site Atmospheric Analysis
of Spectral Hypercubes (FLAASH) software
package in ENVI. FLAASH incorporates
MODTRAN4 radiative transfer code, and
corrects for atmospheric absorption, scattering
and adjacency effects. As data of atmospheric
variables was not recorded at time of image
capture, a 40-mile horizontal visibility and
tropical maritime atmosphere was used as
this was considered the most relevant for the
study region. FLAASH provides outputs in
units of scaled irradiance reflectance relative to
a Lambertian surface.
• As used by Elvidge et al. (2004), the pseudoinvarient features (PIF) correction technique
is an empirical line conversion based on the
assumption of a linear relationship between
image bands across time, for features that
are spatially well defined and spectrally
radiometrically stable (Song et al., 2002). Such
techniques have been used for a number
of other reef and terrestrial applications
(Andréfouët et al., 2001; Karpouzli and
Malthus, 2003). Following the methodology of
Schott et al. (1988), five potential PIF features
were identified. Balancing light and dark
subsets, these were combined into five distinct
groups for analysis (Figure 4). PIF spectra of
each reference image band were regressed
against equivalent spectra from the 2005
image, yielding gain and offset values. These
were then applied to reference imagery. Since
only a small area of runway paint was available,
PIFs were restricted to 15 pixel subsets to
reduce the risk of sub-pixel mixing effects.
Care was also taken to avoid sun-glinted pixels
over marine features where possible. The most
successful PIF approach was also applied to
FLAASH-Modtran corrected for comparison.
103
Vol. 53, No. 1, June 2008
Figure 4. a) Phase 1 PIF normalisation of at-sensor radiance data; (upper left) location of PIF features and their fifteen
pixel subsets; (lower left) the five PIF groups analysed; (right) example regression using PIF group 3 data. b) Phase 2 PIF
normalisation following sun-deglint, (upper left) location of the fifty pixel subsets; (lower left) the single PIF group analysed;
(right) example regression for PIF group 5.
To test normalisation, three 5x5 pixel benthic
sand subsets were used. For independent
assessment these were taken from Smith Bank
at a different depth and geographic location
from PIF subsets (Figures 3 and 4). As an invariant
feature, sand is not susceptible to community
shifts, which might confound assessment over
biological substrate. Mann-Whitney U tests and
Vol. 53, No. 1, June 2008
visual assessment were used to compare the
success of normalisation relative to the 2005
image.
Phase 2 Normalisation
Following development of the above technique,
three concerns were apparent 1) the low sample
size of phase 1 PIF subsets, 2) correction over
104
SPATIAL SCIENCE
Figure 5. Phase 1 normalisation. Reference sand spectra relative to the bleaching scene (2005) i) before and ii) after PIF
group 3 normalisation a) radiance correction and b) following FLAASH-Modtran atmospheric correction. Plots: boxes =
interquartile range, median = central horizontal line, O = outliers, and * = extremes. n = 75.
benthic sand might not imply correction over
coral substrate, and 3) sun glinted pixels could
confound correction using benthic features. A
further round of PIF correction using larger (50
pixel) subsets of deep water and benthic sand
was therefore applied to the best-corrected
imagery from the above process. New test
subsets were created over benthic sand and
mixed coral assemblage areas (Figure 2). Finally,
using Matlab the sun-glint removal algorithm
of Hedley et al. (2005), as well as a dark pixel
SPATIAL SCIENCE
relative atmospheric correction, was applied to all
images.
Bleaching Detection
Here, bleaching is defined as a general whitening in
the appearance of the coral, recognising that the
degree of change in spectral radiance/reflectance
may not be proportional to the severity of a
bleaching event (Enríquez et al., 2005). Radiance
spectra were extracted for pixels containing
ground control points. Past study suggests that
the reflected spectra of many classes may be
105
Vol. 53, No. 1, June 2008
Figure 6. Phase 2 normalisation. a) PIF group 5 (sand and deep-water) correction, sand and mixed coral radiance spectra
shown; b) PIF group 5 and sun-deglint corrected, sand and mixed coral radiance spectra shown. Plots: boxes = interquartile
range, median = central horizontal line, + = outliers, n = 50 per plot.
mixed within a single pixel and restrict bleaching
detection to large and homogenous stands
(Yamano and Tamura, 2004). Since such large,
monospecific stands of A. cervicornis existed,
and primarily this species bleached, the decision
was made to place two 5x5 pixel subsets within
dense homogenous stands of this species (Figure
2b). Bleaching areas were classified based on the
boundaries of the 2005 bleaching distribution in
these test subsets relative to reference imagery.
The difference imaging (DI) approach of Elvidge et
al. (2004) was also applied, subtracting the best
aligned reference image from the 2005 image.
Following band rearrangement and selection of
the appropriate stretch, areas of greatest change
take on a gold appearance.
RESULTS
Normalisation Procedures
At wavelengths greater than ~600nm, light is
rapidly absorbed by the water column. Analysis
was therefore restricted to bands 1 (blue) and 2
(green), which have most applicability for benthic
habitat discrimination. Comparison over sand
test areas showed that images had significantly
different radiometry prior to PIF or FLAASHModtran correction, (Figure 5ai) (p= < 0.001).
Following FLAASH-Modtran conversion to
reflectance, only band 1 of the 2003 reference
image was spectrally inseparable from the 2005
Vol. 53, No. 1, June 2008
image (p= 0.807) (Figure 5bi). Goodness of fit of
PIF group data within band-to-band regressions
was generally high (e.g Figure 4a) (R 2 range =
0.725-0.988; mean R 2= 0.916, 0.959, 0.95,
0.842, 0.867 n= 75, 60, 45, 45, 30 for PIF groups
1-5 respectively). Despite high R 2, few bands
were statistically inseparable at the 0.05 level. PIF
group 3 was the exception to this rule. At-sensor
radiance in band 2 of each reference image was
comparable to the 2005 image (p = ≥ 0.3) (Figure
5aii), while similar radiance values were also
comparable in band 1 2003 (p = 0.906). Though
slightly brighter, other images within this band
showed marked improvement relative to the 2005
image. PIF group 3 radiance normalisation was
better than FLAASH-Modtran correction alone.
When applied to FLAASH-Modtran data (Figure
5aii) the spectral-reflectance relationships in band
2 improved for the 2001 (p = 0.546) and 2000
(p = 0.517) reference images relative to 2005,
the 2003 image and all band 1 spectra remained
different. PIF group 3, applied to at-sensor
radiance corrected data, was considered the best
image set for further analysis.
Goodness of fit for phase 2 PIF group 5
data remained high (e.g Figure 4b) (R2 range =
0.856-0.95; mean R2 = 0.903 and 0.916, n = 100,
before and after sun-deglint). Visual assessment
following phase 2 normalisation shows marked
improvement (Figure 6a). Sand spectra were
inseparable (p = ≥ 0.11), however spectra over
106
SPATIAL SCIENCE
coral were split into two groups 2005-2000 and
2001-2003 (p = ≥0.21). Dark pixel correction
resulted in reduced radiance overall in sundeglinted imagery, however the process clearly
helped normalisation (Figure 6b). Following sundeglint and PIF correction, only band 1 of the
2000 image varied significantly (p = < 0.05),
while only band 2 of the 2001 image varied (p
=< 0.03). At both ground control points and high
A. cervicornis cover areas that were tested for
bleaching, the 2000 and 2003 reference images
were inseparable (p = > 0.201), though the 2001
image still varied. The relative pattern between
the three reference images was constant
between mixed coral and A. cervicornis subsets.
Thus normalisation was most effective for the
2000 and 2003 images.
Spectral Analysis and Change
Detection
At the ground control points, band 1 and 2
radiance in the 2005 image was slightly higher
than the 2000 and 2003 (z-range = -4.183
to -2.285, d.f. = 1, p = < 0.022), but not the
2001 images (z-range = -1.74, -1.36, d.f. =1, p
= 0.081, 0.173; bands 1 and 2, respectively).
This is perhaps representative of the fact that
only 20 percent of total cover bleached in
these areas (Figure 2c). Within high-density A.
cervicornis areas, the 2005-bleaching image had
noticeably higher at-sensor radiance values than
the reference imagery (z-range = -3.39 to -8.54,
d.f. = 1, p = < 0.001) (Figure 7). The distributional
spread of ground control point pixels was greater
with more outliers. Figure 7b shows that with
the exception of the tail of the 2001 image
distribution, where normalisation appears less
accurate, more than 75 percent of the 2005
radiance-spectral distribution are separable from
reference data. This distribution was therefore
used as the basis to identify pixels with a high
likelihood of containing bleached coral. Habitats,
most importantly the distribution of A. cervicornis,
were well known. Constraining classification to
A. cervicornis habitat, pixels were classified as
bleached if band 1 and 2 radiance was greater
than or equal to the 25 th percentile and less
than or equal to the maximum score (outliers
excluded) of the bleached distribution (Figure
8a and b). Bleached coral was predicted across
much of Smith Bank, but was also prevalent at
Cordelia Shoal. Applying the DI technique (Figure
8c) of Elvidge et al. (2004), the greatest increase
in brightness showed along the interface between
sand and mixed coral assemblage areas. An area
of lighter shading corresponded closely with the
distribution described in Figure 8b.
Figure 7. Bleaching assessment. a) Radiance spectra corresponding to the 2005 survey photopoints, (n = 40 per plot); b)
Radiance spectra corresponding to homogeneous Acropora cervicornis areas (n = 50 per plot). Band 1 (blue) and 2 (green)
plotted for bleaching (2005) and reference images (2000, 2001, 2003). The 25th percentile and maximum values of the
bleaching distribution used for spectral detection are indicated. Plots: boxes = interquartile range, median = central horizontal
SPATIAL SCIENCE
107
Vol. 53, No. 1, June 2008
DISCUSSION
Image Normalisation
Central to our application of coral bleaching from
high-resolution satellite images is their normalisation
to a level spectral plane. Radiometric differences in
terms of at-sensor radiance or surface reflectance
values at specific geographical locations can
then be equated to compositional change within
the pixels. Relative radiometric correction is
likely to remain an important approach. The PIF
normalisation approach has widespread support
across a variety of systems (Coppin and Bauer,
1994; Lenney et al., 1996; Elvidge et al., 2004).
We achieved a good fit to linearity using PIF group
3. Examination of the portion of the regression
line corresponding to the region of the spectrum
occupied by healthy and bleached coral (inset
Figure 4) shows that many sand pixels in this subset
fell above the trend. Inclusion of airport-roof pixels
was necessary to correct the overall model. Failure
of correction using only sand and deep water in
the first instance, suggests that the choice of the
sand subset over represented the bright end of the
sand spectrum and that sample size was originally
too small to account for image registration error
and pixel bleeding. Inclusion of benthic features
increased effectiveness of normalisation. While
there is strong support for the need for more
objective and systematic approaches (Du et al.,
2002), skilled user selection and a degree of trial
and error will remain important. Application of the
sun-deglint algorithm was beneficial in all cases
and examining the radiance spectra of mixed coral
in addition to sand suggests normalisation was
effective across much of the relevant spectrum.
Bleaching Detection
As the images in this and other studies were
not normalised against any universal standard,
spectral radiance values are not directly
comparable. The patterns of increased radiance
associated with bleached coral in bands 1 and 2
in this study support findings elsewhere (Elvidge
et al., 2004; Holden and Le Drew, 1998). Though
a slight increase in radiance was noted at the
photographic control points, this did not result in
the clear separation necessary for classification.
While the control points clearly showed
bleaching, some sand pixels were included in
this analysis as a result of registration error. This
explains the increase in spread and number of
Vol. 53, No. 1, June 2008
outliers in photographic point data compared to
values taken from mixed coral and homogeneous
A. cervicornis areas. Such outlier and mismatch
effects are better accounted for when using
subsets greater in size than projected RMS error,
and positioned well within the bounds of different
benthic classes. It is known that the relationship
between pixel size, colony size and bleaching
extent affects detection (Andréfouët et al., 2002).
With higher resolution satellite imagery, in more
homogenous environments, the probability of
bleached coral detection increases (Yamano and
Tamura, 2004). In this study we had a large area
covered uniformly by A. cervicornis available.
Furthermore, the clear statistical and distributional
separation of radiance values suggests that this
area indeed contained significant quantities of
bleached coral. The range of radiance values
described within these subsets is therefore
considered representative of the range values
expected for pixels containing a high level of
bleached A. cervicornis.
At first glance, the differential imaging (DI)
approach (Elvidge et al., 2004) indicated highest
bleaching at the interface between sand and
mixed coral assemblage (Figure 8c). However,
cross-reference between radiance values
in these areas in the 2005 and 2003 images
suggested that most pixels fell well within the
bounds identified for largely healthy coral
(Figures 6 and 7). Image overlay revealed a slight
registration error, and a mismatch of sand and
coral consistent with the areas in which the
change was detected. By stretching the image
over the bright portion of the image scene, the
Elvidge et al.(2004) method precludes viewing
areas of greatest negative change (i.e. darker
relative to reference image). Close examination
of imagery under different stretches however
revealed darker areas consistent with the above
interpretation. As previously discussed, the
error associated with image registration could
cause this mismatch. Moreover, re-sampling
during registration could have the effect of
subtly moving boundaries, particularly in the
case of the QuickBird imagery, which was
resampled from a 2.4m to 4m pixel. Using a
spectral basis to classification, our method is
not reliant on direct pixel-to-pixel comparison
where spatial mismatch confou nds good
identification. Only spectral statistics from
108
SPATIAL SCIENCE
Figure 8. Bleached Acropora
cervicornis at the study
site. a) Bleached areas are
selected from the within A.
cervicornis habitat using the
25th percentile and maximum
scores of band 1 (blue) and
2 (green) 2005 bleached
coral spectrum; b) Spectral
classification for Smith
Bank; c) Difference Imaging
technique of Elvidge et al.
(2004) computed for the
same area
the 2005 image were ultimately used to
describe bleaching distribution. By using the
known spatial distribution of homogeneous A.
cervicornis habitat to constrain selection we 1)
avoided the sand/coral interface in preference
for areas where bleaching was more likely, and
2) prevented misclassifying bleached coral as
sand, a potential constraint noted elsewhere
(Clark et al., 2000). The area described using
the technique in this study (Figure 8b), is broadly
similar to the whitened area of the difference
image (Figure 8c), suggesting this should be
SPATIAL SCIENCE
interpreted as bleaching in the DI case. DI is
a less process-intensive procedure, however
expectations should clearly be tempered with
knowledge of image processing accuracy,
habitat distribution and composition. Confusion
in identifying bleached areas with DI as
described here is likely to be most evident when
focusing on small reef areas. With adequate
habitat knowledge, a spectral-based detection
method has value in its own right, though it is
clearly useful to compare with DI.
109
Vol. 53, No. 1, June 2008
In a study of the effect of spatial resolution
on bleaching detection, Andréfouët et al., (2002)
found that a change in resolution between
0.1 and 1m resulted in a loss of detection of
up to 50 percent. Detection loss is probably
less acute in our case, as change between the
QuickBird and IKONOS is of a lesser magnitude
and we detect coral-bleaching over dense and
homogenous. Application of our classification
approach to the whole study area suggested
that bleaching was widespread within such
habitat. This is expected, since the distance
between Smith and Cordelia Bank is <2km, and
reefs tend to bleach (or not) in spatial clusters at
scales of ~10s km (Berkelmans et al., 2004).
Implications for Management
While the 2005 bleaching event had serious
impacts in the eastern Caribbean, and was well
studied there (Rogers et al., 2008; Ballantine
et al., 2008), less information is available from
the western Caribbean. Our study provides a
demonstration of how remote sensing can be
used to assess the health of remote and rarelyvisited reefs, such as the banks off Roatan. An
analysis of the HadISST1 dataset at Roatan did not
indicate 2005 as an anomalous year, particularly
when compared to the high temperatures of
1998 associated with severe bleaching on Roatan
and elsewhere (Rayner et al., 2006). Since no
regular monitoring occurs on the Roatan banks,
and the hindcast temperature did not appear
unusually high, the bleaching event demonstrated
here could have been easily overlooked. While
bleaching was extensive across the banks, it did
not result in the extremes of mortality observed
for other events locally and elsewhere (Keck,
2004; Berkelmans et al., 2004). Understanding
the ecological importance of bleaching frequency,
severity and distribution is vital to understanding
the future trajectories of coral reefs. Highresolution image sensors such as IKONOS and
QuickBird represent significantly improved means
over older sensors (Palandro et al., 2003a,b)
for monitoring coral populations and detecting
bleaching, as we have demonstrated in this study.
CONCLUSION
In line with previous studies, the suitability of this
approach is governed by degree of whitening,
seascape composition and sensor resolution
Vol. 53, No. 1, June 2008
(Andréfouët et al., 2002; Elvidge et al., 2004).
This study has several important conclusions:
• The possibility of bleaching detection is greatly
heightened if a priori the lateral distribution of
corals, particularly bleaching prone species, is
known.
• When the area of interest is extremely shallow,
detection of such bleaching events through
remote sensing using imagery of a suitable
resolution is feasible.
• As a tool, the application of this approach
is particularly suited to species such as A.
cervicornis, which commonly inhabit shallow
environments and aggregate to stands
extending across multiple pixels.
ACKNOWLEDGEMENTS
This publication is a result of funding from the
National Oceanic and Atmospheric Administration
(NOAA), Center for Sponsored Coastal Ocean
Science, under award NA04NOS4260065 to
Nova Southeastern University for the National
Coral Reef Institute (NCRI). This is NCRI
contribution 94. Thanks are also extended to
Jennifer Keck for assistance in field data collection.
REFERENCES
Andréfouët, S., Müller-Karger, F., Hochberg, E.
J., Hu, C. and Carder, K. L. (2001) Change
detection in shallow water reef environments
using Landsat 7 ETM + data. Remote Sensing
of Environment, vol. 79, pp. 150-162.
Andréfouët, S., Berkelmans, R., Odriozola, L.,
Done, T., Oliver, J. and Müller-Karger, F. (2002)
Choosing the appropriate spatial resolution for
monitoring coral-bleaching events using remote
sensing. Coral Reefs, vol. 21, pp. 147-154.
Aronson, R.B. Precht, W. F., Macintyre, I. G. and
Murdoch, T. J. T. (2000) Coral bleach-out in
Belize. Nature, vol. 405, p. 36.
Ballantine, D.L., Appledorn, R.S., Yoshioka, P., Weil,
E., Armstrong, R., Garcia, J.R., Otero, E., Pagan,
F., Sherman, C., Hernandez-Delgado, E.A.,
Bruckner, A., Lilystrom, C., (2008) Aspects of
Puerto Rican coral reef biology. In B.M. Riegl
and R.E. Dodge, eds. Coral Reefs of the USA.
Springer Science, p. 371-402.
Berkelmans, R. De’ath, G. Kininmonth, S. and
Skirving, W. J. (2004) A comparison of the
110
SPATIAL SCIENCE
1998 and 2002 coral-bleaching events on
the Great Barrier Reef: spatial correlation,
patterns and predictions. Coral Reefs, vol. 23,
pp. 74-83.
Berkelmans, R. and Wills, B. L. (1999) Seasonal
and local spatial patterns in the upper thermal
limits of corals in the inshore Central Great
Barrier Reef. Coral Reefs, vol. 18, pp. 219-228.
Clark, C. D., Mumby, P. J., Chisholm, J. R. M.,
Jaubert, J. and Andrefouet, S. (2000) Spectral
discrimination of coral mortality states
following a severe bleaching event. International
Journal of Remote Sensing, vol. 21(11), pp.
2321-2327.
Coppin, P. R. and Bauer, M. E. (1994) Proceessing
of multitemporal Landsat TM imagery to
optimise extraction of forest cover change
features. IEEE Transactions on Geoscience and
Remote Sensing, vol. 32(4), pp. 918-927.
Du, Y., Teillet, P. M. and Cihlar, J. (2002)
Radiometric normalisation of multitemporal
high-resolution satellite images with quality
control for land cover change detection.
Remote Sensing of Envirionment, vol. 82, pp.
123-134.
D’Croz, L. D., Jackson, J. B. C. and Best, M. M. R.
(1998) Siliciclastic-carbonate transitions along
shelf transects through the Cayos Cochinos
Archipelago, Honduras. Revista de Biologia
Tropical, vol. 46(54), pp. 57-66.
Elvidge, C. D., Dietz, J. B., Berkelmans, R.,
Andréfouët, S., Skirving, W., Strong, A. E. and
Tuttle, B. T. (2004) Satellite observation of
Keppel Islands (Great Barrier Reef) 2001 coralbleaching using IKONOS data. Coral Reefs, vol.
23, pp. 123-132.
Enríquez, S., Méndez, E. R. and Inglesia-Prieto, R.
(2005) Multiple scattering on coral skeletons
enhances light absorbtion by symbiotic algae.
Limnology and Oceanography, vol. 50(4), pp.
1025-1032.
Goreau, T. J. F. and Hayes, R. L. (2005) Global coral
reef bleaching and sea surface temperature
trends from satellite-derived hotspot analysis.
World Resource Review, vol. 17(2), pp.
254-293.
Grimsdich, G. D. and Salm, R. V. (2005) Coral reef
resilience and resistance to bleaching. IUCN,
Gland, Switzerland.
SPATIAL SCIENCE
Hedley, J. D., Harborne, A.R. and Mumby, P. J.
(2005) Simple and robust removal of sun
glint for mapping shallow-water benthos.
International Journal of Remote Sensing, vol.
26, pp. 2107-2112.
Hochberg, E. J., Atkinson, M. J. and Andréfouët,
S. (2003) Spectral reflectance of coral reef
bottom-types worldwide and implications for
coral reef remote sensing. Remote Sensing of
Environment, vol. 85, pp. 159-173.
Holden, H. and LeDrew, E. (1998) Spectral
discrimination of healthy and non-healthy
corals based on cluster analysis, principal
c o m p o n e nt s a n a l y s i s, a n d d e r i vat i ve
spectroscopy. Remote Sensing of Environment,
vol. 65, pp. 217-224.
Karpouzli, E. and Malthus, T. (2003) The empirical
line method for the atmospheric correction
of IKONOS imagery. International Journal of
Remote Sensing, vol. 24 (5), pp. 1143-1150.
Keck, J. (2004) Changes in coral populations on
the northwest coast of Roatan, Bay Islands,
Honduras, subsequent to 1998 bleaching event
and hurricane Mitch. MSc thesis, Nova Southeastern University, Ft Lauderdale.
Keck, J., Houston, M., Purkis, S. J. and Riegl, B. M.
(2005) Unexpectedly high cover of Acropora
cervicornis on offshore reefs at Roatan
(Honduras). Coral Reefs, vol. 24(3), p. 509.
Lenney, P., Woodcock, C. E., Collins, J. B. and
Hamdi, H. (1996). The status of agricultural
lands in Egypt: the use of multitemporal NDVI
features derived from Landsat TM. Remote
Sensing of Environment, vol. 56, pp. 8-20.
Lubin, D., Li, W., Dustan, P., Mazel, C. H. and
Stamnes, K. (2001) Spectral signatures of coral
reefs: features from space. Remote Sensing of
Environment, vol. 75, pp. 127-137.
Mayer, A. G. (1914) The effects of temperature
upon marine animals. Carn Institute, Washigton.
Published Papers, Marine Lab Tortugas, vol. 6,
pp. 3-24.
Mumby, P. J., Green, E. P., Edwards, A. J. and Clark
C. D. (1997) Coral reef habitat mapping: how
much detail can remote sensing provide?
Marine Biology, vol. 130, pp. 193-202.
Peterson, B. (2001) IKONOS relative spectral
re s p o n s e a n d rad i o m et r i c ca l i b rat i o n
111
Vol. 53, No. 1, June 2008
coefficients. Document Number SE-REF-016
Rev A. Thornton, CO: Space Imaging Inc.
Precht,W. F., Robbart, M. L. and Aronson, R. B.
(2004) The potential listing of Aropora species
under the US Endangered Species Act. Marine
Pollution Bulletin, vol. 49, pp. 534-536.
Palandro, D., Andréfouët, S., Dustan, P. and
Müller-Karger, F. E. (2003a) Change detection
in coral reef communities using Ikonos satellite
sensor imagery and historic aerial photography.
International Journal of Remote Sensing, vol.
24(4), pp. 873-878.
Taylor, M. (2006) IKONOS planetary reflectance
and mean solar exoatmospheric irradiance.
Ikonos Planetary Reflectance, Quality System
Online Review, vol. 1, pp. 1-4.
Yamano, H. and Tamura, M. (2004) Detection
limits of coral reef bleaching by satellite remote
sensing: Simulation and data analysis. Remote
Sensing of Environment, vol. 90(1), pp. 86-103.
Palandro, D., Andréfouët, S., Müller-Karger, F. E.,
Dustan, P., Hu, C. and Hallock, P. (2003b)
Detection of changes in coral reef communities
using Landsat 5 TM and Landsat 7 ETM+ Data.
Canadian Journal of Remote Sensing, vol.
29(2), pp. 201-209.
Purkis, S. J., Myint, S. and Riegl, B. (2006)
Enhanced detection of Acropora cervicornis
using a contextural classifier. Remote Sensing
of Environment, vol. 101, pp. 82-94.
Purkis, S. J. and Riegl, B. (2005) Spatial and
temporal dynamics of Arabian Gulf coral
assemblages quantified from remote-sensing
and in situ monitoring data. Marine Ecology
Progress Series, vol. 287, pp. 99-113.
Rayner, N.A., Brohan, P., Parker, D. E., Folland,
C. K., Kennedy, J. J., Vanicek, M.. Ansell, T.
and Tett, S. F. B. (2006) Improved analyses
of changes and uncertainties in sea surface
temperature measured in situ since the midnineteenth century: the HadSST2 data set.
Journal of Climate, vol. 19(3), pp. 446-469.
Rogers, C.S. et al. (2008) Ecology of coral reefs
in the US virgin islands. In B.M. Riegl and R.E.
Dodge, eds. Coral Reefs of the USA. Springer
Science, pp. 301-370.
Schott, J. R., Salvaggio, C. and Volchok, W. J.
(1988) Radiometric scene normalisation using
pseudo-invariant features. Remote Sensing of
Environment, vol. 26, pp. 1-16.
Song, C., Woodcock, C. E. Seto, K. C., Lenney, M.
P. and Macomber, S. A. (2001) Classification
and change detection using Landsat TM
data: When and how to correct atmospheric
effects? Remote Sensing of Environment, vol.
75, pp. 230-244.
Vol. 53, No. 1, June 2008
112
SPATIAL SCIENCE