Comparative evaluation of airborne LiDAR and ship-based multibeam SoNAR
bathymetry and intensity for mapping coral reef ecosystems
⁎, T.A. Battista
, S.J. Pittman
NOAA/NOS/CCMA Biogeography Branch, 1305 East–West Highway, Silver Spring, MD, 20910, United States
Marine Science Center, University of the Virgin Islands, 2 John Brewer's Bay, St. Thomas, VI 00802, US Virgin Islands
Received 29 October 2008
Received in revised form 29 January 2009
Accepted 31 January 2009
Coral reef ecosystems
Benthic habitat mapping
Large areas of the world's coastal marine environments remain poorly characterized because they have not
been mapped with sufﬁcient accuracy and at spatial resolutions high enough to support a wide range of
societal needs. Expediting the rate of seaﬂoor mapping requires the collection of multi-use datasets that
concurrently address hydrographic charting needs and support decision-making in ecosystem-based
management. While active optical and acoustic sensors have previously been compared for the purpose of
hydrographic charting, few studies have evaluated the performance and cost effectiveness of these systems
for providing benthic habitat maps. Bathymetric and intensity data were collected in shallow water (b50 m
depth) coral reef ecosystems using two conventional remote sensing technologies: (1) airborne Light
Detection and Ranging (LiDAR), and (2) ship-based multibeam (MBES) Sound Navigation and Ranging
(SoNAR). A comparative assessment using a suite of twelve metrics demonstrated that LiDAR and MBES were
equally capable of discriminating seaﬂoor topography (r=N0.9), although LiDAR depths were found to be
consistently shallower than MBES depths. The intensity datasets were not signiﬁcantly correlated at a broad
4×5 km spatial scale (r=−0.11), but were moderately correlated in ﬂat areas at a ﬁne 4 × 500 m spatial
scale (r=0.51), indicating that the LiDAR intensity algorithm needs to be improved before LiDAR intensity
surfaces can be used for habitat mapping. LiDAR cost 6.6% less than MBES and required 40 fewer hours to
map the same study area. MBES provided more detail about the seaﬂoor by fully ensonifying high-relief
features, by differentiating between ﬁne and coarse sediments and by collecting data with higher spatial
resolutions. Surface fractal dimensions and fast Fourier transformations emerged as useful methods for
detecting artifacts in the datasets. Overall, LiDAR provided a more cost effective alternative to MBES for
mapping and monitoring shallow water coral reef ecosystems (b50 m depth), although the unique
advantages of MBES may make it a more appropriate choice for answering certain ecological or geological
questions requiring very high resolution data.
© 2009 Elsevier Inc. All rights reserved.
Benthic habitat maps of coral reef ecosystems support multiple
resource managementobjectives, including understanding and predict-
ing the spatial distribution of resources, detecting environmental
change, supporting spatially-explicit decision making for designing
sampling strategies and for zoning and delineating marine protected
areas (Mumby & Harborne,1999; Ward et al., 1999; Kendall et al., 2004,
Pittman et al., 2007). The majority of digital maps of coral reef eco-
systems have been derived through visual interpretation of seaﬂoor
features in aerial photography or multispectral satellite imagery
resulting in thematic habitat maps, with discrete class boundaries
(Kendall et al., 2001; Battista et al., 2007). While these passive optical
techniques have been effective for mapping coral reef ecosystems, they
do not usually provide accurate and continuous topographic informa-
tion. Topography is ecologically important because it inﬂuences the
spatial distribution of marine organisms and also can be important for
monitoring changes in seaﬂoor morphology (Pittman et al., 2007;
Wilson et al., 2007; Wedding et al., 2008).
In contrast, MBES and LiDAR are active remote sensing systems that
measure the topography and physical characteristics of the seaﬂoor by
respectively pulsing sound or laserlight. The returning pulses can then be
analyzed to provide spatially continuous, high resolution bathymetric
and intensity surfaces (Brock et al., 2004; Dartnell & Gardner, 2004;
Wilson et al., 2007). Ship-based multibeam SoNAR, also known as
multibeam echosounders (MBES), are widely used for bathymetric
mapping of the seaﬂoor for the purpose of updating nautical charts
(Intelmann, 2006) and for mapping both shallow and deep-water marine
ecosystems (Lundblad et al., 2006; Wilson et al., 2007). Increasingly,
Remote Sensing of Environment 113 (2009) 1082–110 0
⁎Corresponding author. Biogeography Branch, 1305 East–West Highway, Silver
Spring, MD, United States. Tel.: +1 301 713 3028; fax: +1 301713 4384.
E-mail address: email@example.com (B.M. Costa).
0034-4257/$ –see front matter © 2009 Elsevier Inc. All rights reserved.
Contents lists available at ScienceDirect
Remote Sensing of Environment
journal homepage: www.elsevier.com/locate/rse
Fig. 1. Overview map. The dashed line denotes the boundary of the Abrir La Sierra Conservation District and the area mapped using MBES. The solid black polygon denotes the area mapped using LiDAR. The hatched polygon denotes the area
where the MBES and LiDAR datasets overlapped. The data in this area was used to compare the sensors.
1083B.M. Costa et al. / Remote Sensing of Environment 113 (2009) 1082–1100
airborne LiDAR systems are also being utilized for updating nautical
charts (McKenzie et al., 2001; Intelmann, 2006) and for a wide range of
coastal applications including, improving tsunami and storm-surge
inundation modeling (Venturato et al., 2007), and developing spatially-
explicit seaﬂoor complexity and biodiversity models (Kuffner et al., 2007;
Wedding et al., 2008; Pittman et al., in press). Measures of topographic
complexity derived from LiDAR bathymetry have also been successfully
used to map (Brock et al., 2004; Brock et al., 2006)andtoquantify
(Storlazzi et al., 2003) the morphology of shallow-water coral reefs.
The relatively limited commercial availability of bathymetric LiDAR
systems and the previous absence of seaﬂoor intensity (i.e., reﬂectance)
have prevented more widespread use of this technology. The ability of
LiDAR systems to collect both seaﬂoor intensity, in addition to seaﬂoor
bathymetry, offers a novel alternative to MBES for mapping coral reef
ecosystems in relatively shallow water (b50 m). There are, however,
advantages and disadvantages associated with both systems, which may
inﬂuence their suitability for mapping certain areas and under certain
environmental conditions. Ship-based MBES systems, while proven to be
exceptionally useful in meeting a wide range of objectives, have several
limitations when collecting data in shallow-water environments. These
limitations include: (1) navigation dangers and challenges posed to ves-
sels performing survey work; (2) the inability to collect data in water
shallower than approximately 15 m (although the exact limiting depth
depends on the type of MBES sensor and survey vessel); (3) the inability
to create seamless, coastal topographic–bathymetric (i.e., land–sea) sur-
faces; and (4) reduced efﬁciencies due to the proportional relationship
between water depth and bottom coverage. In contrast, LiDAR systems
appear better suited for mapping shallow-water environments due to:
(1) negligible navigational risks when conducting shallow-water sur-
veys; (2) greater survey efﬁciencies in shallow-waters, as LiDAR swath
width is nearly independent of depth, (3) the ability to collect seamless,
coastal topographic–bathymetric (i.e., land and sea) datasets, and (4) the
capacity of LiDAR to collect bathymetric and intensity datasets. The ability
to seamlessly map topography and bathymetry is especially valuable for
studies examining or modeling coastal patterns and processes, since
many important marine habitats are emergent at low tides (e.g., intertidal
seagrasses, salt marshes, mangroves and some coral reefs).
Given the advantages outlined above, airborne LiDAR surveys may
theoretically provide an alternative to MBES surveys for collecting
datasets that simultaneously address benthic habitat mapping and
nautical charting requirements. However, little empirical research has
been conducted to quantitatively compare and contrast the operating
costs, the time requirements and thedata products of these two systems
for use in comprehensive synoptic mapping. As LiDAR seaﬂoor intensity
is a relatively new product and has not yetbeen widely used forbenthic
habitat mapping, it is particularly important to better understand how
the LiDAR and MBES intensity surfaces compare. This study seeks to ﬁll
this knowledge gap by comparing spatially-coincident airborne LiDAR
and ship-based MBES datasets to evaluate the efﬁciency and efﬁcacy of
these systems to map the topographies and intensities of the same
shallow-water (b50 m) coral reef ecosystem in southwest Puerto Rico.
The objective of this study was not to derive benthic habitat maps from
these datasets, but rather, to understand whether comparable benthic
habitat maps could be developed and integrated using these different
sensors. The following research questions were addressed using multi-
ple quantitative and qualitative metrics at two spatial scales: (1) the
broad spatial scale (4× 5 km) of the study area, and (2) the ﬁne spatial
scale (500× 4 m) of ﬁve independent transects.
(1) Which technology (LiDAR or MBES) is less costly and less time
consuming for mapping shallow-water coral reef ecosystems?
(2) What are the errors and discrepancies between the LiDAR and
MBES bathymetric and intensity surfaces?
(3) Are LiDAR and MBES able to identify the same seaﬂoor features?
This comparison is of critical importance because it will allow
scientists, resource managers, and surveyors to objectively assess the
strengths and weakness of each sensor for mapping structurally
complex shallow-water environments, and to determine which
system is best suited to meet regional coastal conditions, research
goals and management needs.
2. Data and methods
2.1. Description of the study site
The study site, Abrir La Sierra Conservation District, is a marine
protected area (MPA) that is located off the western coast of Puerto
Rico within the U.S. Exclusive Economic Zone (Fig. 1). This 16.7 km
region is bounded by the parallels 18° 6.5' N and 18° 3.5' N, and by the
meridians 67° 26.9' W and 67° 23.9' W. Abrir La Sierra was designated
as an MPA in 1996 due to concerns over declining reef ﬁsh populations
(Waddell & Clark eds., 2008) and because the economically important
red hind (Epinephelus guttatus) uses Abrir La Sierra as a spawning
aggregation site (Beets & Friedlander, 2004; Nemeth, 2005). Given the
economic and biologic importance of aggregation sites, a concerted
effort has been made to better understand them by mapping and
monitoring changes in their associated benthic habitats.
2.2. Data acquisition: LiDAR and MBES system speciﬁcations
Abrir La Sierra was mapped with LiDAR between April 7th and May
15th, 2006 and with MBES between April 17th and April 21st,2007. The
seaﬂoor's physical structure (i.e., its geomorphology) was measured by
calculating the intensity of an individual pulse of light or ping of sound
scattered from the seaﬂoor. The intensity of light or sound scattered by
the seaﬂoor is indicative of sediment hardness and roughness
characteristics. These individual measurements were used to create
continuous images of the seaﬂoor's physical structure, which will be
referred to hereafter as “intensity.”Additionally, seaﬂoor depth was
measured by determining the time required for an individual pulse of
light or sound to travel from the sensor to the seaﬂoor and back again.
These individual measurements were used to create seamless images of
the seaﬂoor depth, which will be referred to hereafter as “bathymetry.”
Bathymetry from the LiDAR and MBES sensors were collected and
processed to meet the International Hydrographic Organization's
(IHO) Order 1 horizontal and vertical accuracy standards (IHO, 2008).
The horizontal accuracy of both datasets is better than ± 5 m+ 5% of
the depth, and the vertical accuracy or maximum total vertical
uncertainty (TVU) of both datasets is better than ± 0.82 m (Table 1).
Maximum TVU (at a 95% conﬁdence level) was calculated by
accounting for the ﬁxed and variable vertical uncertainties associated
with increasing depths. These uncertainties are described by following
equation (Eq. (1)).
a=0.5 m = the portion of uncertainty that does not vary with depth
b=0.013 m= the portion of uncertainty that varies with depth
b⁎d=the portion of uncertainty that varies with depth.
Vertical accuracy or maximum TVU (total vertical uncertainty) of LiDAR and MBES
bathymetric datasets based on IHO Order 1 accuracy requirements.
Uncertainty Depth (m)
10 20 30 40 50
Fixed uncertainty (m) 0.25 0.25 0.25 0.25 0.25
Variable uncertainty (m) 0.02 0.07 0.15 0.27 0.42
Maximum TVU (± m) 0.52 0.56 0.63 0.72 0.82
1084 B.M. Costa et al. / Remote Sensing of Environment 113 (2009) 1082–1100
The IHO order 1 standard ensured that the LiDAR and MBES data-
sets had the same horizontal and vertical accuracies. These compar-
able accuracies made it possible to directly and quantitatively
compare the LiDAR and MBES datasets and to determine the utility
of LiDAR hydrographic charting data for habitat mapping.
2.2.1. The LiDAR system
LiDAR data were acquired for elevations between 50 m above sea
level and 70 m below sea level using a Laser Airborne Depth Sounder
(LADS) Mk II Airborne System. This airborne system uses a 900 Hz Nd:
YAG (neodymium-doped yttrium aluminum garnet) laser, which is
split by an optical coupler into an infrared (1064 nm) beam and a green
(532 nm) beam. The infrared beam measures the datum height at
nadir and the green beam oscillates across-track to measure depths
and/or elevations. The DeHavilland Dash 8-200 aircraft ﬂew the
hydrographic survey at altitudes between 1200 and 2200 ft and at
ground speeds between 140 and 175 kn. The LiDAR survey was
conducted at different altitudes so that survey activities could continue
below low cloud ceilings. Altitudes varied among survey lines, but they
did not vary while on a single survey line. The airborne survey achieved
200% seabed coverage with 4 × 4 m spot spacings. Raw data were
logged using the Tenix LADS Airborne System and were converted
using the LADS Mk II Ground System. Soundings were positioned
relative to the NAD83 UTM 19 N horizontal coordinate system and to
the Mean Lower Low Water (MLLW) vertical tidal coordinate system
(Stephenson & Sinclair, 2006).
2.2.2. The MBES system
MBES data were acquired for depths from 10 to 55 m using a Seabat
Reson 8124 (200 kHz) multibeam echosounder mountedon the NOAA
ship Nancy Foster. This hull-mounted system measured water depths
across a 120° swath made up of 80 individual beams of sound with
1.75°× 1.5° widths. It achieved seaﬂoor coverage of 3.5× water depth to
approximately 75 m. The beams to the port and starboard of nadir
overlapped adjacent survey lines by ≤10%. The vessel survey speed
was 4 to 6 kn. Its positioning and orientation were determined by the
Applanix POS/MV 320 V4. A GPS aided Inertial Motion Unit (IMU)
provided measurements of roll, pitch and heading. The POS/MV
obtained its positions from two dual frequency Trimble Zephyr GPS
antennae. An auxiliary Trimble DSM 132 DGPS system provided an
RTCM differential data stream from the U.S. Coast Guard Continually
Operating Reference Station at Port Isabel, Puerto Rico. CTD (con-
ductivity, temperature and depth) measurements were taken approxi-
mately every 4 h using a Seabird Electronics SBE-911 or SBE-19 CTD to
correct for the changing sound velocities in the water column. Raw
data were logged using Triton ISIS
7.1 software, and were referenced
to the NAD83 UTM 19 N horizontal coordinate system and tothe Mean
Lower Low Water (MLLW) vertical tidal coordinate system (Battista &
2.2.3. Footprint sizes and swath widths of the systems
Before comparing LiDAR and MBES, it is important to understand
how the footprint sizes and swath widths of these sensors change with
depth (Fig. 2). The footprint size of a sensor is the area on the seaﬂoor
ensoniﬁed by each individual laser pulse or ping of sound. The swath
width of a sensor is the across-track distance that is mapped while on a
surveyline. For MBES, the system's beam footprint size and swath width
are depth dependent because pings of sound have more time to spread
out in the water column as the slant range increases (i.e., the distance a
beam travels from the sensor to the seaﬂoor). The morethe ping spreads
out, the larger the area ensoniﬁed on the seaﬂoor and vice versa.
Fig. 2. Diagram illustrating the different acquisition geometries of MBES and LiDAR. For MBES (left), swath width and beam footprint size depend on depth. As the seaﬂoor becomes
), MBES swath width becomes wider (SW
) and the sizes of the beam footprints become larger (B
). For LiDAR (right),
swath width and beam footprint size depend primarily on the scan angle (θ
). The size of the swath width and beam footprint sizes can be preserved at different altitudes because
the scan angle can be adjusted accordingly line by line. In the diagram, the scan angle (θ
) of the sortie ﬂying at a lower altitude (Altitude
) was widened to maintain a consistent
swath width (SW
) and laser footprint size (B
). Changing seaﬂoor depths (Seaﬂoor
) do not greatly impact the size of the LiDAR swath (SW
) and laser footprints (B
) because light travels differently than sound in water.
1085B.M. Costa et al. / Remote Sensing of Environment 113 (2009) 1082–1100
For LiDAR, the system's laser footprint and swath width are nearly
independent of depth and are primarily inﬂuenced by scan angle of the
system (i.e., the maximum angle from nadir that is scanned by the
laser). Scan angles can be modiﬁed between survey lines to maintain
consistent footprint sizes at different sortie altitudes (Stephenson &
Sinclair, 2006). If the same altitude is maintained however, larger scan
angles would increase the laser pulse footprint sizes at the outer edges
of the swath because these pulses would ensonify the seaﬂoor at larger
incident angles. The incident angle is the angle (perpendicular to the
seaﬂoor) that is formed when a beam strikes the seaﬂoor. Conversely,
smaller scan angles would decrease the footprint size (again, of laser
pulses at the outer edges of the swath) because these pulses would
ensonify the seaﬂoor at smaller incident angles. Depth can change the
footprint size of a laser slightly (because light is refracted by water and
scattered by suspended particles in the water column), but not nearly
to the same degree as for MBES systems. This allows the swath width of
the LiDAR system to remain fairly constant, despite changing depths.
For this survey, the footprint of the laser was held constant at
approximately 2.5× 2.5 m at the sea surface and 4 × 4 m at depth
2.3. Data processing: Creating bathymetry and intensity surfaces
2.3.1. LiDAR and MBES bathymetry and intensity
The full resolution LiDAR and MBES bathymetric and intensity
datasets were preprocessed, so that the analogous bathymetry and
intensity surfaces would have identical radiometric and spatial resolu-
tions (Fig. 3). The LiDAR bathymetric data was corrected for aircraft
height and heading, offsets between sensors, latency, mirror and
platform angles, sea surface model errors, refraction of the laser beam
at the sea surface, the effects of scattering of the beam in the water
column and the inﬂuence of tides. It was then processed to create a 16-
bit, 4× 4 m raster surface. The LiDAR intensity data was processed to
create an 8-bit, 5× 5 m raster surface. This surface was derived using a
proprietary Tenix LADS algorithm, which calculated relative return
intensity as a ratio between the transmitted and returned energy for
each laser pulse (Stephenson, 2007). The returned energy was normal-
ized for changes in gain and losses through the water column. Energy
lost in the water column was estimated using measures of water clarity
and the path length (not slant range) of a laser pulse through the water
column. Laser pulse incident angles were considered to be constant.
LiDAR intensity values that were ±3 standard deviations fromthe mean
were ﬁltered and removed from the ﬁnal 5× 5 m surface. The LiDAR
intensity surface was gridded at a coarser resolution to minimize the
amount of noise and reduce the number of data gaps seen in the
intensity image. The 8-bit LiDAR intensity surface was not radio-
metrically rescaled(after it was clippedto the same geographicextent as
the MBES intensity surface) because adding more quantization values
would have introduced artiﬁcialvalues into the image (i.e.,since higher
resolution data cannot be generated from lower resolution data without
interpolation). Comparing interpolated values with empirical values
would not have been a fair and direct comparison of LiDAR and MBES
The MBES bathymetric data was corrected for sensor offsets, latency,
roll, pitch, yaw, staticand dynamic draft, thechanging speed of sound in
the water column and the inﬂuence of tides. It was then processed to
create a 16-bit, 4× 4 m raster surface. The MBES intensity data were
received as raw Reson 8124 .xtf (Extended Triton Format) ﬁles, which
recorded the uncorrected intensity value from each snippet (i.e., from
each beam of sound for each ping of sound). The .xtf ﬁles were
geometrically and radiometrically corrected using Geocoder 3.0 (Fon-
seca & Calder, 2005). In particular, the intensity surface was geome-
trically corrected for navigation attitude, transducer attitude and slant
range distortion using the MBES bathymetric surface. It was radio-
metrically corrected for changes in acquisition gains,powerlevels, pulse
widths, incidence angles and ensoniﬁcation areas. All snippets were
preserved during these corrections, allowing the full resolution data to
be used to create the ﬁnal mosaic. This mosaic was then spatially
resampled from 1× 1 to 5×5 m using a nearest neighbor algorithm, and
rescaled from 32-bits to 8-bits using the following algorithm (Eq. (2)):
8bit MBES Intensity Surface
=ð32bit MBES Intensity Surface½
28.81=32-bit MBES intensity surface minimum (dBs)
49.46 = 32-bit MBES intensity surface maximum (dBs)
255=Largest integer supported by 8-bit image
Conversion from a 32 to 8-bit surface did not reduce the
radiometric resolution of the MBES intensity surface because all
Fig. 3. Flowchart describing the methods used to create bathymetry and intensity surfaces from the full resolution LiDAR and MBES datasets.
1086 B.M. Costa et al. / Remote Sensing of Environment 113 (2009) 1082–1100
213 unique values (with 3 signiﬁcant digits) were transferred from
the 32-bit image to the resulting 8-bit image.
2.4. Which sensor was less costly & time consuming to deploy in Abrir La
Eight metrics were calculated to describe the spatial resolutions of,
and time required for, the two systems to map the same geographic
area. These metrics included: (1) total area mapped, (2) spatial
resolutions of the bathymetry and intensity surfaces, (3) total number
of soundings, (4) estimated total time on survey lines, (5) total
number of survey lines, (6) total length of survey lines, (7) average
swath width of survey lines, and (8) average vessel speed on survey
One metric was calculated to describe the monetary costs of the
two sensors to map the same geographic area. This metric (called
estimated total cost) was derived by calculating the percent difference
between the total cost of acquiring and processing the LiDAR datasets
versus the MBES datasets. These total cost estimates included the
following activities for both sensors: (1) mobilization and demobili-
zation; (2) data acquisition, processing and delivery of tidal data; and
(3) data acquisition, processing and delivery of the bathymetric and
intensity data. Mobilization and demobilization speciﬁcally included
the costs associated with employee travel, per diem and day rate, as
well as equipment transport to the site. Data acquisition, processing
and delivery of tidal data speciﬁcally included the costs associated
with retrieving and processing the ﬁnal, veriﬁed 6-minute tide heights
and zones. In the case of the LiDAR survey, it also included setting up,
monitoring and dismantling an auxiliary tide gauge at Punta
Guanajibo, Puerto Rico. Data acquisition, processing and delivery of
the bathymetric and intensity data included the costs associated with
operating and maintaining the vessel while on survey (i.e., the money
needed to pay for both contractor and federal government employee
time and materials). This metric also included the costs associated
with processing and cleaning the bathymetric and intensity datasets,
as well as deriving ﬁnal products for delivery. It is also important to
note that the LiDAR survey was conducted by a contractor at the
request of NOAA's Ofﬁce of Coast Survey, while the MBES survey was
conducted by a mix of in-house contractors and federal government
employees at NOAA's Center for Coastal Monitoring and Assessment.
2.5. Explanation of spatial scales
The LiDAR and MBES datasets were compared at both a broad
spatial scale (4 × 5 km) and at a ﬁne spatial scale (4 × 500 m). The
dimensions of the broad spatial scale were chosen because of its
relevance to marine resource protection and management, given that
the majority (77%) of U.S. federal marine protected areas are over
in size (NOAA MPA Federal Programs, 2008). The dimensions
of the ﬁne spatial scale were chosen because of its relevance to ﬁsh-
habitat relationships. Namely, Pittman et al. (in press) found that
bathymetric complexity metrics at this approximate spatial scale were
the most important predictors for 12 out of 17 ﬁsh species richness,
biomass and abundance metrics.
2.6. What are the errors and discrepancies between the LiDAR and MBES
bathymetric and intensity surfaces?
The two datasets were compared at a broad spatial scale using
qualitative and quantitative methods, including surface subtraction,
fractal dimensions (D) and fast Fourier transformations (FFTs). All
analyses were performed in ArcGIS 9.2, LandSerf 2.2 (Wood, 2005)
and ENVI 4.5 software packages.
2.6.1. Comparisons at a broad spatial scale: Surface subtractions
To identify and quantify the differences between the two sensors at
a broad spatial scale, each MBES surface was subtracted from its
analogous LiDAR surface. The subtractions were performed using
ArcGIS Spatial Analyst's (SA) Raster Calculator. The resulting raster
was reclassiﬁed to denote pixels that had depth differences greater
than the maximum allowable vertical error designated by IHO Order 1
2.6.2. Comparisons at a broad spatial scale: Fractal dimensions
The fractal dimension (D)(Mandelbrot, 1983)ofasurfaceisascale-
independent measure of terrain variability, denoting a continuum from
aﬂat surfaceto a space-ﬁlling rough surface (Wilson et al., 2007). Fractal
dimension surfaces were generated from the LiDAR and MBES
bathymetric data using the variogram method (Shih et al., 1999)and
calculated in a 9× 9 cell window withinLandserf v2.2 (Wood, 2005). The
fractal surfaces were visually inspected for linear features of elevated
roughness parallel or perpendicular to the survey lines. Linear features
are indicative of errors associated with uncalibrated survey equipment
(e.g., due to changing sound velocities) or unfavorable environmental
conditions (e.g., due to changing turbidity conditions). Areas surround-
ing these linear features were spatially queried to estimate the
magnitude of the vertical offsets between the two depth surfaces.
2.6.3. Comparisons at a broad spatial scale: Fast Fourier transformations
FFTs produce images depicting the individual spatial frequency
components of a dataset (Mather, 2004; Lillesand & Kiefer, 2000). The
average brightness value of the image (i.e., its zero frequency com-
ponent) is located in the center of the transformed image. Pixels
farther away from the image center (i.e., moving concentrically
outward) represent the increasing spatial frequency components of
the image. Subsets (620×620 pixels) of the LiDAR and MBES intensity
and bathymetric surfaces were transformed into the frequency domain
using the FFT method available in ENVI 4.5. This analysis technique was
employed because errors that are periodic (but singular) have higher
frequencies than the average brightness value in an image. As a result,
they tend to cluster in an FFT image, making them easier to visually
identify in frequency space than in geographic space.
2.7. Do LiDAR and MBES identify the same seaﬂoor features?
The LiDAR and MBES datasets were compared a broad spatial scale
using quantitative methods, including Spearman's Rank correlations,
Ordinary Least Squares regression and empirical semi-variograms.
These same datasets were also compared at a ﬁne spatial scale using
qualitative and quantitative methods, including proﬁle views, Spear-
man's Rank correlations, Ordinary Least Squares regression, empirical
semi-variograms and semi-variogram modeling. All analyses were
performed in ArcGIS 9.2, JMP and S-Plus software packages.
2.7.1. Comparisons at a broad spatial scale: Uncorrelated grid locations
Correlations between LiDAR and MBES depths and LiDAR and MBES
intensities were calculated using Spearman's Rank correlation. Tests
for spatial autocorrelation and directionality were conducted to ensure
each point was a statistically independent sample. Spatial autocorrela-
tion thresholds were determined by creating and analyzing eight
regularly-spaced point grids between 50 and 1000 m using a global
Moran's I test in ArcGIS's SA Extension. These uniformly distributed
point locations were found to be spatially autocorrelated up to 770 m
(p≥0.1; z≥1.86). Anisotropic variograms for the north (0°) and
southeast (149°) azimuths were also created to ensure that the mean
orientation of the survey lines (149° for MBES and 0° for LiDAR) did not
introduce a directional bias into the datasets.
Thirty-three and thirty-oneindependent points were identiﬁed on a
regularly-spaced grid of 770 m for the bathymetric and intensity
surfaces, respectively. Spearman's ρrank correlation was used to
1087B.M. Costa et al. / Remote Sensing of Environment 113 (2009) 1082–1100
compare: (1) the corresponding values of each of these point locations
and (2) the LiDAR intensity versus bathymetry and MBES intensity
versus bathymetry at coincident sample locations (n=205) randomly
generated throughout the study area. The MBES intensity and bathy-
metric surfaces were compared to be consistent, even though previous
investigations have established a signiﬁcant correlation between
acoustic return strength and sediment grain size (Davis et al., 1996;
Goff et al., 2000; Kostylev et al., 2001; Collier & Brown, 2005).
2.7.2. Comparisons at a ﬁne spatial scale: Creating transects and
Five topographically and structurally diverse areas were used to
evaluate the ability of both sensors to identify the same seaﬂoor fea-
tures at ﬁne spatial scales. One transect (500 m in length) was drawn in
each area. The LiDAR and MBES bathymetry and intensity values were
extracted, graphed, visually compared and tested using the Spearman's
ρrank. In addition, the LiDAR intensity versus bathymetry and MBES
intensity versus bathymetry were also compared using Spearman's
ρrank test at randomly generated coincident sample locations
(n=125) around each of the ﬁve transects.
2.7.3. Comparisons at a ﬁne spatial scale: Removing broad-scale spatial
variation from transects
Visual assessment of variograms indicated that both LiDAR and
MBES bathymetry and intensity surfaces were non-stationary (i.e., the
mean of each dataset varied in space). In order to accurately describe
how covariance changed across space, the mean of each dataset was
removed by plotting the corresponding LiDAR and MBES surfaces
against one another (at each transect), and ﬁtting a linear model using
ordinary least-squares regression in JMP. The resulting residuals were
exported to text ﬁles and imported into S-Plus statistical software for
2.7.4. Comparisons at a ﬁne spatial scale: Understanding ﬁne-scale
spatial variation in transects
Theoretical variogram models were used to quantitatively describe
how both distance and direction affected the behavior of the underlying
spatial processes (Bailey & Gatrell, 1995) present in the MBES and LiDAR
bathymetry and intensity datasets. The ﬁrst step in this process was to
determine the distributions of the residuals and to create square-root
difference variogram clouds to look for outliers. Outlying values were
retained in this study because they denoted locations where the LiDAR
and MBES surfaces were the most different. The classical variogram
method (Matheron, 1963) was used for transects with normally
distributed residuals, including transects 1 and 2 for the bathymetric
surfaces and transects 1, 2 and 5 for the intensity surfaces. The robust
variogram method (Cressie & Hawkins, 1980) was used to create
variograms for transects with residual QQ plots that had heavier than
normal distributions at the tails, including transects 3, 4 and 5 for the
bathymetric surfaces and transects 3 and4 for the intensity surfaces. The
intensity residuals for transects 1 and 4 were log transformed so that
their distributions approached normality. Isotropic variograms were
created since each transect had a single orientation. Several theoretical
models were ﬁt to these empirical variograms using S-Plus to provide a
continuous description of the covariance structure of the LiDAR and
3.1. Which sensor was less costly and time consuming to deploy in Abrir
Several metrics were calculated to describe the time efﬁciency and
one metric was calculated to compare the cost efﬁciency of the two
systems when mapping the same geographic area (Table 2). These
comparisons determined that LiDAR data cost 6.6% less to acquire than
MBES data. LiDAR was less expensive to collect because the airborne
bathymetry system required 146 fewer survey lines, 362 fewer kilo-
meters and 40 fewer hours to map the same study area. The MBES
system did, however,collect approximately 172 times more soundings,
and acquired bathymetric and intensity surfaces at spatial resolutions
that were two and ﬁve times higher (respectively) than LiDAR
datasets. The cost and efﬁciency of these two sensors were able to be
directly compared (despite these different spatial resolutions) because
both surveys were designed to satisfy the same NOAA speciﬁcations
and IHO Order 1 standards.
3.2. What are the errors and discrepancies between the LiDAR and MBES
bathymetric and intensity surfaces?
3.2.1. Comparisons at a broad spatial scale: Surface subtraction
The largest, negative depth difference between the two bathy-
metric surfaces was −6.53 m and the largest, positive depth difference
was 2.74 m (i.e., the LiDAR surface was 6.53 m shallower and 2.74 m
Results summarizing the cost and efﬁciency of operating LiDAR and MBES systems in shallow-water environments.
LiDAR MBES Deﬁnition Calculation method
20.22 20.22 Total area mapped using LiDAR and MBES “Calculate Geometry”function in ArcMap
4×4 Resampled to 4× 4
(from 2× 2)
Area on the ground that each pixel represents in the bathymetric and intensity
Queried raster properties in ArcMap
5×5 Resampled to 5× 5
(from 1× 1)
1,254,955 216,880,411 Total number valid, individual pulses of light or sound collected in the entire
Queried raw LiDAR/MBES ﬁles in CARIS
2 42 Total time that was needed to map the entire study area (excluding turns,
transit time or CD casts). The MBES survey vessel operated 16 h a day and did
not return to port at night. The LiDAR survey vessel returned to San Juan at
the end of each 8 hour shift.
Calculated by dividing total length of survey lines by average
vessel speed. These numbers were veriﬁed by querying raw
ﬁles in CARIS.
55 201 Total number of transects is needed to survey the entire study area (excluding
Queried shapeﬁle table in ArcMap
258 621 Distance traveled by the survey vessel while mapping the entire survey area “Calculate Geometry”function in ArcMap
76 31 Average across-track width of the area ensoniﬁed during a survey transect
(at 200% coverage for LiDAR; b10% coverage for MBES)
Calculated in ArcMap by: (1) generating six lines perpendicular
to the survey tracklines; (2) intersecting these lines with the
survey tracklines; and (3) averaging the distances between
these intersections using Hawths Tools (Beyer, 2004).
259 15 Average speed of the survey vessel while on a trackline Queried raw LiDAR/MBES ﬁles in CARIS
−6.6 N/A Estimated total cost of acquiring and processing the LiDAR and MBES datasets.
The exact dollar amount are not included due to proprietary restrictions. The
reported percentage reﬂects whether LiDAR data acquisition was more or less
expensive then MBES data acquisition.
The total coast estimate for LiDAR and MBES includes the
following activities: (1) mobilization and demobilization; (2)
data acquisition, procesing and delivery of tidal, bathymetric data
and intensity data.
1088 B.M. Costa et al. / Remote Sensing of Environment 113 (2009) 1082–1100
deeper than the MBES surface at these locations). The majority of the
negative depth differences occurred along topographic features with
high relief, including the shelf edge. The majority of the positive depth
differences were oriented in the NW–SE direction, parallel to the MBES
ship tracklines (Fig. 4). Of 1,263,521 pixels in the study area, 72,318 of
them (~5.7%) had depth differences greater than the maximum
allowable vertical error for IHO Order 1 mapping. Of this 5.7%, slightly
more than half (53%) were less than ±10 cm from meeting the
accuracy standard. The pixels with the largest depth differences (that
were greater than the maximum allowable vertical error) occurred
primarily along the shelf edge in water N35 m deep. The pixels with the
smallest depth differences (that were greater than the maximum
allowable vertical error) occurred primarily on the shelf ﬂat in b35 m
The maximum negative and positive intensity differences between
the two intensity surfaces were −135 and 211, respectively. The
majority of the negative intensity differences occurred in areas
dominated by ﬁne sandy sediments, and the majority of the positive
differences occurred in the deeper (N30 m) and more topographically
complex parts of the study area (Fig. 5).
3.2.2. Comparisons at a broad spatial scale: Fractal dimensions
Fractal surfaces conﬁrmed the presence of along-track artifacts in
both the LiDAR and MBES bathymetric surfaces (Fig. 6). Artifacts in the
LiDAR bathymetric surface were parallel to the orientation of the
aerial tracklines in the N–S direction. They denoted small vertical
offsets of approximately 0.1–0.3 m between parallel, overlapping
ﬂight lines. Artifacts in the MBES bathymetric surface were oriented in
the NW–SE direction, parallel to the orientation of the ship tracklines.
They denoted small vertical offsets of approximately 0.1–0.3 m
between parallel, overlapping ship transects. All of these errors were
less than the maximum, vertical uncertainty allowed by IHO Order 1
3.2.3. Comparisons at a broad spatial scale: Fast Fourier transformations
The FFT bathymetric images revealed that LiDAR and MBES
bathymetric surfaces contained similar vertical frequency components,
as denoted by the bright vertical stripe in the FFT image (Fig. 7a and b).
The two FFT images, however, differed in their horizontal frequencies, as
denoted by the wider horizontal striping and parallel lines present in the
LiDAR FFT image. The horizontal stripes denoted stronger N–S
directionality in the LiDAR bathymetric surface, which corresponded
to the orientation of the ﬂight lines. The horizontal lines also conﬁrmed
the presence of small, vertical differences between adjacent and
overlapping ﬂight lines.
The LiDAR and MBES intensity surfaces had noticeably different
horizontal, vertical and diagonal frequencycomponents (Fig. 8a–c). In
particular, the LiDAR intensity FFT image was dominatedby horizontal
patterning (Fig. 8b), while the MBES intensity FFT image was
dominated by diagonal patterning (Fig. 8c). Diagonal patterns in the
Fig. 4. The LiDAR (left) and MBES (right) bathymetry surfaces collected in Abrir La Sierra. The LiDAR and MBES bathymetric surfaces were subtracted from each other. The resulting
surface is located in the lower left corner. The inset table (on top in the lower right) contains summary statistics for each layer, and the inset table (on the bottom in the lower right)
contains summary information about pixel depth differences that were greater than the maximum allowable vertical error speciﬁed IHO Order 1 vertical accuracy standards.
1089B.M. Costa et al. / Remote Sensing of Environment 113 (2009) 1082–1100
MBES FFT image corresponded to the NW–SE orientation of the ship
transect lines. Similarly, horizontal patterns in the LiDAR FFT image
corresponded to the N–S orientation of the ﬂight lines.
The LiDAR and MBES intensity surfaces also had different high and low
frequency components. In particular, speckling (i.e., random noise) is
noticeable in the MBES FFT image, and is denoted by the even
Fig. 6. The fractal dimension surfaces created using the LiDAR (left) and MBES (right) bathymetry surfaces. The insets for each map provide a closer look at the along-track artifacts
highlighted by fractal dimension surfaces.
Fig. 5. The LiDAR (left) and MBES (right) intensity surfaces collected in Abrir La Sierra. The LiDAR and MBES intensity surfaces were subtracted from each other. The resulting surface
is located in the lower left corner. The inset table in the lower right contains summary statistics for each layer. Please note that the intensity values do not have measurement units
because they are relative (not absolute) numbers.
1090 B.M. Costa et al. / Remote Sensing of Environment 113 (2009) 1082–1100
distribution of bright pixels moving away from the image's center
(Fig. 8c). The other patternof note is the bright patches in each quadrant
of the LiDAR FFT image (Fig. 8b). These patches were ﬁltered out and the
original and inverse FFT images were subtracted from each other in
order to determine their origin (Fig. 8d). This analysis revealed that these
pixels were the result of broad-scale noise in the LiDAR dataset, which
was notmade visible by thesurface subtraction analysis. Thelast pattern
of note isthe similarity between the LiDAR FFT bathymetricand intensity
images (Figs. 7aand8b). Similarities in their horizontal and vertical
frequency components suggest that the LiDAR intensity surface is
strongly correlated with the LiDAR bathymetric surface.
3.3. Do LiDAR and MBES identify the same seaﬂoor features?
3.3.1. Comparisons at a broad spatial scale: Uncorrelated grid locations
Comparing the LiDAR and MBES surfaces at spatially uncorrelated
locations (with 770 m uniform grid spacing) revealed patterns in the
variation between the technologies. The MBES bathymetric surface was
strongly correlated with the LiDAR bathymetric surface (p≤0.0001;
r=0.999). The MBES intensity surface was not correlated with the
LiDAR intensity surface (p≤0.5439; r=−0.113). The LiDAR intensity
surface, however, was strongly correlated with the LiDAR bathymetric
surface (p≤0.001; r=0.841). The MBES intensity surface was weakly
correlated with the MBES bathymetric surface (p≤0.001; r=0.273).
3.3.2. Comparisons at a ﬁne spatial scale: Correlations along transects
The ﬁve subsetted locations were topographically and struct urally
different from each other (Table 3). Overall, the MBES bathymetric
surface was strongly and signiﬁcantly correlated with the LiDAR
bathymetric surface along transects 1–5(p≤0.0001; r≥0.906;
Table 4). However, the depths measured by the LiDAR sensor were
consistently shallower than those measured by the MBES sensor
along transects 1–5(Figs. 9–13). The MBES intensity surface was
weakly correlated with the LiDAR intensity surface along transects
2–5(p≤0.02; 0.283≤r≥−0.432; Table 4), and the LiDAR intensity
surface was strongly correlated with the LiDAR bathymetric surface
along transects 2–5(p≤0.0001; r≥0.753; Table 4). The MBES intensity
surface was moderately correlated with the LiDAR intensity surface
along transect 1 (p≤0.02; r=0.5103; Tabl e 4), and the LiDAR intensity
surface was not correlated with the LiDAR bathymetry surface along
transect 1 (p≤0.32; r=0.095; Tabl e 4). The MBES intensitysurface was
not correlated with the MBES bathymetry surface along transects 1–5
(p≥0.074; 0.295≤r≥−0.372; Table 4).
3.3.3. Comparisons at a ﬁne spatial scale: Spatial dependence along
Theoretical variogram models were ﬁt to the empiricalvariograms of
the bathymetric and intensity residuals in order to understand how
points along transects 1–5 changed together over space. The bathy-
metric residualsfor transects 1–5exhibited strongspatial dependence at
both ﬁne and coarse scales up to distances of 125 m (Table 5). These
scalar relationships of spatial dependency were best described by
spherical variogram models, as determined by non-linear least-squares
ﬁts. The fact that all of the transects were described by the same
theoretical variogram models indicated that spatial dependency
remained the same over the entire study area. Unchanging spatial
dependencies suggest that the spatial processes inﬂuencing the depth
returns on the shelf edge were the same as those inﬂuencing the depth
returns on the shelf ﬂat.
The intensity residuals for transects 1, 2, 3 and 5 exhibited little or
no spatial dependence at ﬁne scales (between 0 and 10 m) and
stronger spatial dependence at coarser scales (between 23 and 127 m)
(Tabl e 5). These scalar relationships of spatial dependency were best
described by Gaussian variogram models, as determined by non-linear
least-squares ﬁts (Table 5). The intensity residuals for transect 4 did
not exhibit the same pattern as transects 1, 2, 3 and 5. Instead, they
were best described by a spherical variogram model, and exhibited
weaker spatial dependence (than the other transects) at both ﬁne and
coarse scales up to distances of 50 m. The fact that all of the transects
were not described by the same theoretical variogram models
indicated that second order effects in the intensity datasets were
non-stationary and that spatial dependency changed across the study
area (i.e., at Transect 4). Changing spatial dependencies suggest that
different spatial processes inﬂuenced the intensity of the returns on
the shelf edge versus those returns on the shelf ﬂat.
Recently, there has been tremendous growth in the development
and application of hydrographic data for seaﬂoor characterization and
mapping. In the U.S., there has long been a strong legal mandate and
ﬁnancial support for collecting hydrographic data to support safe
Fig. 8. Fast Fourier transformations. The transformed LiDAR & MBES intensity surfaces were spatially-coincident and were both 620× 620 pixels. a) the original, subsetted LiDAR
intensity image; b) FFT of the LiDAR intensity surface; and c) FFTof the MBES intensity surface. The bright objects in the 4 corners of b) were removed through ﬁltering and the entire
image was inversely transformed back into geographic space. d) is the result of subtracting the original, subsetted LiDAR surface from the ﬁltered image.
Fig. 7. Fast Fourier transformations of: a) LiDAR bathymetricsurface (left); and b) MBES
bathymetric surface (right). The transformed LiDAR & MBES bathymetric surfaces were
spatially-coincident and were both 620 ×620 pixels.
1091B.M. Costa et al. / Remote Sensing of Environment 113 (2009) 1082–1100
maritime navigation (HSIA, 1998), only relatively recently has there
been a legislative mandate for understanding the distribution, health,
and abundance of marine seaﬂoor environments, particularly in tropical
marine coral reef ecosystems (CRCA, 2000). The impetus being to
improve conservation measures of marine ﬂora and biota, while
simultaneously trying to reduce the adverse impacts of anthropogenic
changes (e.g., coastal development, overﬁshing). However, the ability of
coastal managers to evaluate the efﬁcacy of management measures is
predicated on the ability to provide meaningful spatial characterizations
of seaﬂoorhabitats and associated marine life. Thus, scientists have been
exploring the application and adaptation of traditional hydrographic
data collection systems to fulﬁll these diverse data needs.
4.1. Which sensor was less costly and time consuming to deploy in Abrir
Several metrics were calculated to compare time efﬁciency and one
metric was calculated to compare the cost efﬁciency associated with
operating LiDAR and MBES systems in shallow-water environments.
Overall, the LiDAR system was determined to be more cost and time
efﬁcient than the MBES system, although the MBES system collected
bathymetric and intensity data at higher spatial resolutions.
In particular, the MBES system collected bathymetric and intensity
data that were two and ﬁve times higher (respectively) than the
analogous LiDAR datasets. These higher spatial resolutions were the
result of the MBES system acquiring 173 times more soundings than
the LiDAR system. The MBES system acquired more soundings because
its average in situ swath width was half as wide, and its average
acquisition speed was 17 times slower than the average swath width
and acquisition speed for the LiDAR system. This combination of
narrower swath widths and slower survey speeds reduced the size of
the beam footprints, thereby increasing the spatial resolution at which
the data was acquired.
The LiDAR system, on the other hand, was more cost efﬁcient and
time efﬁcient than the MBES system for mapping the same shallow-
water region to the same IHO standard. These higher efﬁciencies
were due to the LiDAR system's distinct acquisition geometry, wider
swath widths and faster survey speeds. In particular, the average
acquisition speed was much faster for LiDAR systems, approximately
140 kn, while the average speed of the su rvey ship was approximately
8 kn. Also for LiDAR, swath width is primarily determined by scan
angle and is nearly independent of depth and aircraft altitude
(Stephenson & Sinclair, 2006). Conversely, the relationship between
swath width and water depth is proportional for the MBES systems
(i.e., the shallower the water, the narrower the swath and the less
area mapped on a single survey line). This difference was highlighted
by the fact that the LiDAR system's average in situ swath width was
twice as wide as the average swath width of the MBES system. Wider
swath widths allowed the LiDAR system to map the same area with
146 fewer tracklines than the MBES system. It should also be noted,
however, that the 120° swath of the Reson 8124 MBES unit is smaller
than some other MBES systems with similar frequencies (e.g., Reson
8101 240 kHz system has a 150° swath). It is also much smaller than
MBES systems with dual SoNAR heads (e.g., Simrad EM 3002) and
interferometric SoNARs, which are an emerging and promising
technology. The combination of broader swaths, fewer tracklines
and faster acquisition speeds collectively explain the increased time
efﬁciency and hence lower costs associated with LiDAR data
Spearman's p correlation coefﬁcients showing the strength of associations am ong
analogous LiDAR and MBES b athymetric and intensity surfaces as well as betwee n
LiDAR or MBES bat hymetric and inte nsity surface s by transec t (Nos. 1–5). Statistical
signiﬁcance of p≤0.02 is denoted by ⁎. Statistical signiﬁcance of p≤0.0 001 is denoted
0.906⁎⁎ MBES intensity
0.987⁎⁎ MBES intensity
0.992⁎⁎ MBES intensity
0.998⁎⁎ MBES intensity
0.997⁎⁎ MBES intensity
Descriptions of habitats and LiDAR and MBES surfaces along transects 1–5. *These characterizations were made using underwater ground truthing video and visual interpretations of
the MBES imagery.
Transect Qualitative description of transect* Descriptive
(relative 8-bit value)
(relative 8-bit value)
#1 Traversed a ﬂat location, dominated by uncolonized
sand and pavement colonized by turf.
Mean 22.15 22.58 173.71 137.18
Standard deviation 0.27 0.27 5.40 16.16
Minimum 21.33 21.70 163.66 111.35
Maximum 22.48 23.05 180.60 171.58
Range 1.15 1.35 16.94 60.23
#2 Traversed a topographically complex location, dominated
by pavement colonized by turf, soft coral, and macroalgae.
Mean 14.51 14.87 149.30 103.62
Standard deviation 1.94 1.96 8.49 8.78
Minimum 11.6 6 11.95 135.00 77.99
Maximum 19.37 19.82 170.46 126.24
Range 7.71 7.87 35.46 48.25
#3 Traversed a topographically homogenous location,
dominated by sand colonized by turf and pavement
colonized by soft corals, turf and macroalgae.
Mean 18.07 18.53 173.55 141.60
Standard deviation 1.09 1.11 9.02 26.00
Minimum 15.14 15.51 145.00 86.21
Maximum 19.38 19.82 180.40 186.85
Range 4.24 4.31 35.40 110.64
#4 Traversed a sloping pavement feature, dominated by
soft corals, coral sponges, macroalgae and turf.
Mean 27.41 27.97 184.14 110.78
Standard deviation 7.02 6.98 20.37 13.93
Minimum 14.46 14.85 144.81 87.35
Maximum 35.36 35.74 207.07 152.18
Range 20.90 20.89 62.26 64.83
#5 Traversed topographically complex location, dominated
by pavement colonized by soft corals, sponges,
microalgae and turf.
Mean 15.92 16.19 154.06 100.94
Standard deviation 2.65 2.66 9.30 16.34
Minimum 11.96 12.06 138.14 57.81
Maximum 21.79 22.19 169.23 171.79
Range 9.83 10.13 31.09 113.98
1092 B.M. Costa et al. / Remote Sensing of Environment 113 (2009) 1082–1100
acquisition. It is likely that the costs associated with acquiring LiDAR
data could be reduced markedly (Rohmann & Monaco, 2005)inthe
future by expanding the technical capacity for and operation of LiDAR
sensors by the government.
4.2. What are the errors and discrepancies between the LiDAR and MBES
bathymetric and intensity surfaces?
Quantitatively, the LiDAR and MBES bathymetric surfaces were
highly correlated at a broad spatial scale (r= 0.9992; p≤0.0001).
Qualitatively, the two bathymetric surfaces were visibly similar, as
denoted by the subtraction surfaces, fractal dimension surfaces and
the FFT images. There were, however, three noticeable differences
between the LiDAR and MBES bathymetric surfaces.
The ﬁrst noticeable difference occurred in areas with high relief.
There, the largest differences between the LiDAR and MBES depths (i.e.,
where LiDAR depths were shallower than MBES depths) were found
primarily along features with slopes of 25% or greater. LiDAR and MBES
depths were dissimilar in these high relief areas because different along
and across track acquisition geometries caused the sensors to have
different footprint sizes. The LiDAR laser had a larger footprint (4×4 m)
Fig. 9. (Top and Middle) LiDAR and MBES bathymetry and intensity surfaces for transect 1. (Bottom) Proﬁle views of the LiDAR and MBES bathymetry and intensity surfaces along
1093B.M. Costa et al. / Remote Sensing of Environment 113 (2009) 1082–1100
than did a single MBESbeam (b2 × 2m). Consequently, the depth ranges
within the larger, LiDAR footprints varied more widely than did the
depth ranges within the smaller, MBES footprints (especially in areas of
high relief). Additionally, the shallowest depth returned within a LiDAR
laser footprint was the value assigned to the entire footprint. This
combination of a larger footprint and a shallow-biased footprint caused
LiDAR depth values to be consistently shallower in areas of high relief.
Another important difference to note between the performance of
LiDAR and MBES systems in high relief areas is that LiDAR systems are
limited to ﬁxed, nearly vertical acquisition geometries (NOAA CSC,
1999). This is not the case for all MBES systems. Some MBES systems
(e.g., Kongsberg Simrad units) can be angledso that the beams ensonify
the entire side of a high-relief feature, such as a shelf edge or a vertical
wall (Kongsberg, 2007). Beam steering can also decrease the footprint
size of an acoustic beam by decreasing its incident angle. These two
advantages allow some MBES systems to collect higher resolution and
better quality data in areas with high topographic relief.
The second noticeable difference between the two bathymetric
surfaces occurred along the shelf in water deeper than approximately
35 m. These areas were where the majority of pixels (N60%) were
located that had depth differences greater than the maximum allowable
vertical error. This pattern suggests that the LADS laser did not
Fig. 10. (Top and Middle) LiDAR bathymetryand LiDAR and MBES intensity surfaces for transect 2. (Bottom) Proﬁle views of the LiDAR and MBES bathymetry and intensity surfaces
along transect 2.
1094 B.M. Costa et al. / Remote Sensing of Environment 113 (2009) 1082–1100
consistently penetrate the water column (and scatter off the seaﬂoor) in
areas deeper than 35 m. The laser may have not consistently penetrated
to the seaﬂoor because the water may have been too turbid. Under
exceptionally clear conditions, the LADS LiDAR system is capable of
penetrating 70 m into the water column (Stephenson & Sinclair, 2006).
Such conditions are rare if not unprecedented, however, in the coastal
waters of Puerto Rico. Water column turbidity poses one of the most
signiﬁcant challenges to the acquisition of LiDAR in shallow water areas.
It does so because suspended sediments scatter the laser pulses, and
inhibit or prevent them from hitting the seaﬂoor. This problem can be
mitigated by monitoring water clarity conditions and planning survey
operations accordingly. The LiDAR survey of Abrir La Sierra, for example,
occurred between April and May because the rivers in Puerto Rico had
their smallest ﬂow rates during those months (Stephenson & Sinclair,
2006). If, however, water clarity conditions remain consistently poor in
an area, the only alternative maybe to map it using MBES or another
The third and ﬁnal noticeable difference between the two bathy-
metric surfaces was denoted by the FFT images, wherein the LiDAR
bathymetric surface had three visible horizontal frequencies and the
MBES bathymetric surface had one visible horizontal frequency.
Although there is no conﬁrmed explanation for the presence of two
Fig. 11. (Top and Middle) LiDAR and MBES bathymetry and intensity surfaces for transect 3. (Bottom) Proﬁle views of the LiDAR and MBES bathymetry and intensity surfaces along
1095B.M. Costa et al. / Remote Sensing of Environment 113 (2009) 1082–1100
additional frequency components in the horizontal domain, they likely
denote systematic along-track artifacts in the LiDAR acquisition array as
the artifacts were parallel to the LiDAR N–Sﬂight lines (Stephenson &
Sinclair, 2006; Battista & Stecher, 2007). These artifacts may be a
function of sensitivities due to LiDAR swath width being more than
double that of MBES system.
In comparing the two technologies at a broad spatial scale, the
intensity surfaces were qualitatively and quantitatively different.
Quantitatively, the LiDAR and MBES intensity surfaces were not
correlated (r=−0.1133; p≤0.5439). Qualitatively, the two intensity
surfaces were dissimilar, as denoted by visual interpretation of the
images, the subtraction surfaces and the FFT images. Visually, the MBES
intensity surface appeared to be brighter and more highly resolved than
the LiDAR intensity image. These differences are due to different
acquisition characteristics of the two sensors. Speciﬁcally, the MBES
intensity image is brighter because the MBES system experiences more
noise directly below the sensor head than does the LiDAR system. This
noise is caused by wavelength-scale roughness on the water surface
Fig. 12. (Top and Middle) LiDAR bathymetry and LiDAR and MBES intensity surfaces for transect 4. (Bottom) Proﬁle views of the LiDAR and MBES bathymetry and intensity surfaces
along transect 4.
1096 B.M. Costa et al. / Remote Sensing of Environment 113 (2009) 1082–1100
(Fonseca & Calder, 2005; Makris, 1996). The MBES intensity surface
appears more highly resolved (even though it was resampled to the same
spatial resolution as the LiDAR intensity surface) because it was created
from a dataset with much higher sounding densities. The reasons for
these higher sounding densities were described previously inSection 4.1.
Secondly, the subtraction surfaces denoted large differences
between the two intensity surfaces along the shelf edge. One possible
reason for these differences is the confounding inﬂuence of depth on
LiDAR intensity, suggesting that Tenix LADS intensity algorithm has
not sufﬁciently decorrelated the geometric and radiometric inﬂuence
of LiDAR targets. This explanation was further supported by the fact
that the LiDAR intensity surface was highly correlated with the LiDAR
bathymetric surface at a broad spatial scale (r=0.841; p≤0.001), and
is explained in further detail in Section 4.3.2.
Thirdly, the FFT images highlighted differences in the orientation of
the LiDAR and MBES survey's tracklines (as with the bathymetry
surfaces). In particular, along-track artifacts were noticeable in the
N–S direction in the LiDAR intensity surface and in the NW–SE
Fig. 13. (Top and Middle) LiDAR bathymetryand LiDAR and MBES intensity surfaces for transect 5. (Bottom) Proﬁle views of the LiDAR and MBES bathymetry and intensity surfaces
along transect 5.
1097B.M. Costa et al. / Remote Sensing of Environment 113 (2009) 1082–110 0
direction in the MBES intensity surface. The FFT images also high-
lighted speckling in the MBES intensity image (this phenomenon is
also captured by the semi-variogram nuggets). This speckled pattern is
most likely the result of random, background noise recorded by the
MBES transducer, as the signal sampled at the transducer head is
subject to noise caused by wavelength-scale roughness on the surface
(Makris, 1996; Fonseca & Calder, 2005). Bright, high-frequency objects
were also easily distinguished in the LiDAR FFT intensity image. These
bright, high-frequency patches most likely denote additional noise
fractions highlighted by the algorithm used to calculated seaﬂoor
intensity. When these bright patches were subtracted from the original
image, their spatial distribution appeared to be arbitrary. This pattern
suggests that these noise fractions are the result of less energy being
returned when a laser is pulsed off-nadir. At present, the Tenix LADS
intensity algorithm does not compensate for this reduced return by
correcting for the distance different laser pulses travel from the sensor
to the seaﬂoor. This distance increases as a laser is pulsed at larger
angles to the port and starboard of nadir.
4.3. Do LiDAR and MBES identify the same seaﬂoor features?
At a ﬁne spatial scale, the MBES and LIDAR systems measured the
same local relief, but not the same water depths for all of the transects.
These two systems also did not measure the same intensities in areas of
high relief at a ﬁne spatial scale. The MBES and LIDAR systems did,
however, measure the same intensities in areas that had low relief
(i.e., in areas where the depths changed b1 m vertically over 500 m
horizontally). LiDAR systems, therefore, have the capacity to detect the
same seaﬂoor features as MBES, but LiDAR intensity surfaces need to be
normalized for changes in depth before they are used for habitat
mapping. This conclusion is supported by both qualitative and quan-
titative information, which is outlined below.
In comparing the two technologies at a ﬁne spatial scale, the
bathymetric surfaces were qualitatively and quantitatively similar.
Qualitatively, the two bathymetric surfaces were visually similar, as
denoted by the transectproﬁle graphs. Thesetransect proﬁlegraphs also
showed that the LiDAR bathymetric surface was consistently shallower
(by 0.42 m± 0.007) than the MBES bathymetric surface. This bias was
most likely associated with LiDAR acquisition, and not with MBES
acquisition, because soundings at high frequencies (i.e., N200 kHz)
penetrate only a few centimeters into the seabed (Kagesten, 2008). The
exact penetration depth depends on the physical properties of the
sediment (Hamilton et al., 1956) as well as the power, pulse length
(NOAA CSC, 2008) and grazing angle (Chotiros, 1995) of the signal.
Additionally, this depth offset was most likely not a datum issue because
one would expect the vertical offset to be fairly uniform along transects
1–5 (considering they traverse such a small area). These vertical offsets
were, however, not uniform within or among transects. This offset was
also not a tidal issue because tide corrections were applied to both
datasets, and because the combined tidal error budget of the two
surveys (0.37 m) was smaller than the average depth difference
between the two sensors (0.44 m ± 0.19). If the vertical depth
differences between the sensors were due to tides alone, one would
expect the average depth difference to be about equal or below the
combined tidal error budgets.
Given that these explanations seem unlikely, the most probable
explanation has to do with howthe seabed is identiﬁed in the raw laser
waveform during Sortie Run Processing (SRP). SRP is the automatic
processing phase whereby sounding depths are calculated on a line-by-
line basis. During this step, the raw waveforms were processed to
identify depths for the two most likely bottom return pulses (Fig. 14).
These two pulses were then classiﬁed using their signal-to-noise ratios,
agreement with their nearest neighbors and a maximum likelihood
estimator.The most likely bottom return pulsewas selected based on the
above classiﬁcation and a shoal weighting function (Stephenson &
Sinclair, 2006). It is this shoal weighting function, in particular, which
may have caused the LiDAR depth soundings to be consistently
shallower than the MBES depth soundings. More research should be
conducted examining how the application or absence of this function
affects the depth of LiDAR soundings.
Quantitatively, the LiDAR and MBES bathymetric surfaces were
highly correlated for all transects (p≤0.05; r≥0.9058). This result is
further supported by the fact that all of the empirical variograms were
ﬁt by bounded (i.e., spherical) theoretical variogram models. The
theoretical variogram models also suggest that there was little
random error or noise associated with the two bathymetric surfaces.
This pattern is supported by the fact that none of the theoretical
models had signiﬁcant nugget effects (i.e., the residual values for each
transect were spatially autocorrelated at ﬁne spatial scales).
In comparing the two technologies at a ﬁne spatial scale, the in-
tensity surfaces were qualitatively and quantitatively dissimilar. These
dissimilarities were most prevalent in areas of high relief, such as in
transects 2, 3, 4 and 5. The MBES and LiDAR intensity surfaces were
weakly correlated for all four of these transects and were negatively
correlated for three of these transects. It is unlikely this difference is due
to changing macroalgal abundances since the LiDAR and MBES datasets
were collected in thesame season. Rather, the weak correlation is more
likely a result of the confounding inﬂuence of depth on LiDAR intensity,
and additional need to decorrelate the geometric and radiometric
Theoretical variogram models ﬁt to the empirical variograms of transects 1–5. These models
denote the degree to and thresholdat which points along a transect are spatially related to one
another. Statistical signiﬁcance of p≤0.05 is denoted by ⁎. Statistical signiﬁcanceof p≤0.005 is
denoted by ⁎⁎.
Transect Surface Theoretical
Range Sill Nugget
#1 Bathymetry Spherical 125.1⁎⁎ 0.0176⁎⁎ −0.0005
Intensity Gaussian 126.5⁎⁎ 0.3785⁎⁎ 0.0394
#2 Bathymetry Spherical 31.94⁎⁎ 0.0216⁎⁎ 0.0019
Intensity Gaussian 31.40⁎⁎ 63.811⁎⁎ 7.0831⁎⁎
#3 Bathymetry Spherical 39.80⁎⁎ 0.0053⁎⁎ −0.0003
Intensity Gaussian 64.12⁎80.644⁎0.2801
#4 Bathymetry Spherical 117.4⁎⁎ 0.0444⁎⁎ −0.0008
Intensity Spherical 49.58⁎0.0 009⁎0.001
#5 Bathymetry Spherical 47.06⁎⁎ 0.0310⁎⁎ −0.0030
Intensity Gaussian 23.08⁎⁎ 48.361⁎⁎ 10.698⁎⁎
Fig. 14. LADS LiDAR sounding waveforms (adapted Stephenson & Sinclair, 2006). (3)
Depth bar showing depth of seabed; (4) Depth bar of alternate seabed; (5) Depth of the
seabed; (6) Depth of alternate seabed; (7) Signal-to-noise ratio of seabed; (8) Signal-to-
noise ratio o alternate seabed.
1098 B.M. Costa et al. / Remote Sensing of Environment 113 (2009) 1082–1100
inﬂuences on the LiDAR signal. This explanation is further supported by
the strong correlations between LiDAR intensity and LiDAR bathymetry
transects 2, 3, 4 and 5.
While the MBES and LiDAR intensity surfaces were dissimilar in
areas of high relief, they were qualitatively and quantitatively similar
in areas that were ﬂat (i.e., transect 1). Flat areas were deﬁned as areas
where depths changed less than 1.15 m, vertically over 500 m,
horizontally. The MBES and LiDAR intensity surfaces for this transect
were moderately and signiﬁcantly correlated (p≤0.05; r=0.5103),
unlike all the other transects. This moderate correlation is most likely
because the LiDAR intensity return was not confounded by changes in
depth. If this was the case, why then were the MBES and LiDAR
intensity surfaces for transect 1 not more strongly correlated? One
possible explanation is the amount of background noise in the MBES
intensity. Speciﬁcally, the acoustic intensity signal sampled at the
transducer head is subject to random noise (Fonseca & Calder, 2005).
These ﬂuctuations produce speckling in the registered intensity
imagery (Fonseca & Calder, 2005). This noise was reduced by anti-
speckling algorithms, but it was not fully eliminated in the imagery.
While transects 1, 2 and 5 had signiﬁcant nugget effects, transects 3
and 4 did not. Transect 3 almost exclusively traverses an area of ﬁne-
grained, unconsolidated sediment. This ﬁne sediment may have ab-
sorbed much of the acoustic energy when it was ensoniﬁed (Shumway,
1960;andHamilton,1972), thereby reducing the amount of background
noise returned to the transducer head. The thickness of this sediment
layer may have also impacted the amount of returned acoustic energy,
although the depth of this sandy sediment layer was not known. In
general, the penetration depth of marine sediments (by sound)depends
on the roughness and density of the sediment (Hamilton et al.,1956), as
well as on the frequency (Hamilton, 1972;Bowles, 1997) and grazing
angle (Chotiros, 1995) of the beam.
Transect 4 was located on the shelf edge, where the quality of the
LiDAR intensity surface degraded due to depth limitations (Stephen-
son & Sinclair, 2006). This degradation of MBES and LiDAR intensity
along the shelf may have introduced errors at broader spatial scales
which obscured the background noise at ﬁner spatial scales. This
broad-scale error may also explain why transect 4 had weaker spatial
autocorrelation between points than the other transects (which is
denoted by the fact that transect 4's variogram was ﬁt best by a
spherical model, while the other transects' variograms were ﬁt best by
Two important ﬁndings originate from this study. First, LiDAR was
found to be more time and cost efﬁcient than MBES, although MBES
collected data at higher spatial resolutions. Wider swath widths and
faster acquisition speeds accelerated the rate at which the study area
could be mapped, and concurrently decreased the spatial resolution of
the data being collected. Conversely, narrower swath widths and
slower acquisition speeds increased the spatial resolution of the data
being collected, while concurrently slowing the rate at which the study
area could be mapped. Both survey efﬁciency and spatial resolution,
therefore, need to be taken into account when choosing the most
appropriate sensor to map an area. Further research is also needed to
determine the optimum spatial resolution of either MBES or LiDAR
hydrographic data for addressing speciﬁc ecological questions (Ken-
dall & Miller, 2008).
Second, LiDAR was found to have the ability to identify similar
seaﬂoor features as MBES systems, although additional improvements
are needed to isolate the LiDAR seaﬂoor intensity measurements from
the depth component of the signal waveform. It is also important to
note that the penetration of the LiDAR laser was inhibited by turbidity
in water N35 m. The turbidity of an area, consequently, needs to be
considered when determining an area's suitability for laser mapping.
LiDAR was also unable to fully ensonify high-relief features and to
differentiate between ﬁne and coarse sediments. These details about
the seaﬂoor maybe important if the goal of a project is to study the
surﬁcial geology of an area, and also need to be taken into account
when choosing the most appropriate sensor to map an area.
Given: (1) the increased efﬁciency of LiDAR and its ability to detect
comparable seaﬂoor features, (2) the recent availability of LiDAR in-
tensity data, (3) the burgeoning operational costs of ships to conduct
hydrographic surveys,(4) the inability of MBES to collect data in water
shallower than approximately 15 m, (5) the navigational risks of
operating vessels in poorly charted shallow-water environments, and
(6) the logistical ease of operating aircraft in remote areas, marine
resource managers and scientists should increasingly consider the use
of LiDAR as an alternative to MBES when workingto inventory shallow-
water tropical marine environments in support of ecosystem-based
The authors would like to thank NOAA's Coral Reef Conservation
Program for their continued support of the CCMA Biogeography
Branch's effort to characterize coral reef ecosystems in the U.S.
Caribbean. We also thank Matthew Kendall and the three anonymous
reviewers for their help in reﬁning this manuscript.
Bailey, T. C., & Gatrell, A. C. (1995). Interactive spatial data analysis (pp. 1−413). Essex,
England: Prentice Hall.
Battista, T. A., Costa, B. M., & Anderson, S. M. (2007). Shallow-water benthic habitats of
the main eight Hawaiian islands. NOAA Technical Memorandum NOS NCCOS 61
Battista,T. A., & Stecher,M. (2007). MBES dataacquisition & processing report:Project NF-07-
06-USVI-HAB, NOAA Data Acquisition & Processing Report NOS NCCOS CCMA (On-line).
Beets, J., & Friedlander, A. (2004). Evaluation of a conservation strategy: a spawning
aggregation closure for red hind, Epinephelus guttatus, in the U.S. Virgin Islands.
Environmental Biology of Fishes,55,91−98.
Bowles, F. A. (1997). Observations on attenuation and shear-wave velocity in ﬁne-
grained, marine sediments. The Journal of the Acoustical Society of America,101 (1),
Brock, J. C., Wright, W. C., Clayton, T. D., & Nayegandhi, A. (2004). LiDAR optical rugosity
of coral reefs in Biscayne National Park, Florida. Coral Reefs,23,48−59.
Brock, J. C., Wright, C. W., Kuffner, I. B., Hernandez, R., & Thompson, P. (2006). Airborne
LiDAR sensing of massive stony coral colonies on patch reefs in the northern Florida
reef tract. Remote Sensing of Environment,104,31−42.
Chotiros, N. P. (1995). Biot model of sound propagation in water-saturated sand. The
Journal of the Acoustical Society of America,101,199−214.
Collier, J. S., & Brown, C. J. (2005). Correlation of sidescan backscatter with grain size
distribution of surﬁcial seabed sediments. Marine Geology,214,431−449.
CRCA (Coral Reef Conservation Act 20 00). P.L. 106–562; 16 U.S.C. §§ 6401 et seq;
December 23, 2000.
Cressie, N., & Hawkins, D. M. (1980). Robust estimation of the variogram: I. Mathema-
Dartnell, P., & Gardner, J. V. (2004). Predicting seaﬂoor facies from multibeam bathy-
metry and backscatter data. Photogrammetric Engineering & Remote Sensing,70,
Davis, K. S., Slowey, N. C., Stender, I. H., Fiedler, H., Bryant, W. R., & Fechner, G. (1996).
Acoustic backscatter and sediment textual properties of inner shelf sands, north-
eastern Gulf of Mexico. Geo-Marine Letters,16, 273−278.
Fonseca, L., & Calder, B. (2005). Geocoder: an efﬁcient backscatter map constructor.
Proceedings of the U.S. Hydrographic 2005, San Diego, CA.
Goff, J. A., Olson, H. C., & Duncan, C. S. (200 0). Correlation of side-scan backscatter
intensity with grain-size distribution of shelf sediments, New Jersey margin. Geo-
Hamilton, E. L. (1972). Compressional-wave attenuation in marine sediments. Geo-
Hamilton, E. L., Shumway, G., Menard, H. W., & Shipek, C. J. (1956). Acoustic and other
physical properties of shallow-water sediments off San Diego. The Journal of the
Acoustical Society of America,28,1−15 .
HSIA Hydrographic Services Improvement Act (1998). 33 U.S.C. §§ 892, 892a–892d.
Intelmann, S. S. (2006). Comments on hydrographic and topographic LIDAR acquisition
and merging with multibeam sounding data acquired in the Olympic Coast
National Marine Sanctuary. NOAA Technical Memorandum NMS Conservation Series
IHO (International Hydrographgic Organization) (2008). IHO standards for hydrographic
surveys: Special publication N° 44 (pp. 1−36)., 5th Ed. Monaco, France:
International Hydrographic Bureau.
1099B.M. Costa et al. / Remote Sensing of Environment 113 (2009) 1082–1100
Kagesten, G. (2008). Geological seaﬂoor mapping with backscatter data from a multibeam
echo sounder. M.S. Thesis. Department of Environmental and Aquatic Engineering,
Uppsala University, Sweden, 38pp.
Kendall, M. S., Monaco, M. E., Buja, K. R., Christensen, J. D., Kruer, C. R., Finkbeiner, M., &
Warner, R. A. (2001). Methods used to map the benthic habitats of Puerto Rico and
the U.S. Virgin Islands. NOAA Technical Report (Online).
Kendall, M. S., Kruer, C. R., Buja, K. R., Christensen, J. D., Diaz, E., Warner, R. A., & Monaco,
M. E. (2004). A characterization of the shallow-water coral reefs and associated
habitats of Puerto Rico. Gulf and Caribbean Research,16,177−184.
Kendall, M. S., & Miller, T. (2008). The inﬂuence of thematic and spatial resolution on
maps of a coral reef ecosystem. Marine Geodesy,31,75−102.
Kongsberg (2007). SIS—Seaﬂoor Information System, Operator manual, Document 850-
164709 (pp. 1−550). Horton, Norway: Kongsberg Maritime AS.
Kostylev, V. E., Todd, B. J., Fader, G. B. J., Courtney, R. C., Cameron, G.D. M., & Pickrill, R. A.
(2001). Benthic habitat mapping on the Scotian Shelf based on multibeam bathy-
metry, surﬁcial geology and sea ﬂoor photographs. Marine Ecology Progress Series,
Kuffner, I. B., Brock, J. C., Grober-Dunsmore, R., Bonito, V. E., Hickey, T. D., & Wright, C. W.
(2007). Relationship between ﬁsh communities and remotely sensed measure-
ments in Biscayne National Park, Florida, USA. Environmental Biology of Fishes,78,
Lillesand, T. M., & Kiefer, R. W. (2000). Remote sensing and image interpretation
(pp. 1−724)., 4th ed. New York , U.S.A.: John Wiley & Son s Ltd.
Lundblad, E. R., Wright, D. J., Miller, J., Larkin, E. M., Rinehart, R., Naar, D. F.,Donahue, B. T.,
Anderson, S. M., & Battista, T. (2006). A benthic terrain classiﬁcation scheme for
American Samoa. Marine Geodesy,29,89−111.
Makris, N. C. (1996). Estimating surface orientation from sonar images. Journal of the
Acoustical Society of America,99, 2449.
Mandelbrot, B. (1983). The fractal geometry of nature (pp.1−46 8). New York, U.S.A.: W.
H. Freeman and Company.
Mather, P. M. (2004). Computer processing of remotely-sensed images (pp. 1−324)., 3rd
ed. West Sussex, England: John Wiley & Sons Ltd.
Matheron, G. (1963). Principles of geostatistics. Economic Geology,58, 1246−1266.
McKenzie, C., Gilmour, B., Van Den Ameele, E. J., & Sinclair, M. (2001). Integration of
LiDAR data in CARIS HIPS for NOAA charting. Fugro Pelagos White Papers (pp. 1−16).
San Diego, CA: Fugro Pelagos.
Mumby, P. J., & Harborne, A. R. (1999). Development of a systematic classiﬁcation
scheme of marine habitats to facilitate regional management and mapping of
Caribbean coral reefs. Biological Conservation,88,155−163.
Nemeth, R. S. (2005). Population characteristics of a recovering U.S. Virgin Islands red
hind spawning aggregation following protection. Marine Ecology Progress Series,
NOAA CSC (Coastal Services Center) (1999). South Carolina's coast: a remote sensing
perspective, NOAA Coastal Services Center CD-ROM (On-line).
NOAA CSC (Coastal Services Center) (2008). Benthic habitat mapping: mapping tech-
niques: Acoustics: Sub-bottom proﬁling, NOAA Coastal Services Center (On-line).
NOAA MPA (Marine Protected Area) Federal Programs (2008). The Marine Protected
Areas Inventory. MPA Inventory: GIS Spatial Data (Online).
Pittman, S., Christenson, J., Caldow, C., Menza, C., & Monaco, M. (2007). Predictive
mapping of ﬁsh species richness across shallow-water seascapes in the Caribbean.
Pittman, S. J., Costa, B., & Battista, T. A. (in press). Using LiDAR bathymetry and boosted
regression trees to predict the diversity and abundance of ﬁsh and corals. Journal of
Coastal Research: Special Issue of Coastal Applications of Airborne LiDAR Remote
Rohmann, S. O., & Monaco, M. E. (2005). Mapping southern Florida's shallow-water
coral ecosystems: An implementation plan. NOAA Technical Memorandum NOS
NCCOS 19 (Online).
Shih, T. Y., Hwang, J. T., & Tsai, T. J. (1999). The fractal properties of sea surface topo-
graphy derived from TOPEX/POSEIDON (1992–1996). Computers and Geosciences,
Shumway, G. (1960). Sound speed and absorption studies of marine sediments by a
resonance method. Geophysics,25,451−467.
Stephenson, D. (2007). LiDAR relative reﬂectivity report: Project OPR-I305-KRL-06, NOAA
Descriptive Report NOS NCCOS.
Stephenson, D., & Sinclair, M. (2006). NOAA LiDAR data acquisition & processing report:
Project OPR-I305-KRL-06, NOAA Data Acquisition & Processing Report NOS OCS
Storlazzi, C. D., Logan, J. B., & Field, M. E. (2003). Quantitative morphology of a fringing
reef tract from high-resolution laser bathymetry: Southern Molokai, Hawaii. Geo-
logical Society of America Bulletin,115 , 1344−1355.
Venturato, A. J., Arcas, D., & Kânoğlu, U. (2007). Modeling tsunami inundation from a
Cascadia subduction zone earthquake for LongBeach and Ocean Shores, Washington.
NOAA Technical Memorandum OAR PMEL-137 (Online).
Waddel, J. E, & Clarke, A. M. (2008). The state of coral reef ecosystems of the United
States and Paciﬁc freely associated states. In J. E. Waddell & A. M. Clarke (Eds.),
NOAA Technical Memorandum NOS NCCOS 73 (Online).
Ward, T. J., Vanderklift, M. A., Nicholls, A. O., & Kenchington, R. A. (1999). Selecting
marine reserves using habitats and species assemblages as surrogates for biological
diversity. Ecological Applications,9,691−698.
Wedding, L., Friedlander, A., McGranaghan, M., Yost, R., & Monaco, M. (2008). Using
bathymetric LiDAR to deﬁne nearshore benthic habitat complexity: implications for
management of reef ﬁsh assemblages in Hawaii. Remote Sensing of Environment,
Wilson, M. F. J., O'Connell, B., Brown, C., Guinan, J. C., & Grehan, A. J. (2007). Multiscale
terrain analysis of multibeam bathymetry data for habitat mapping on the
continental slope. Marine Geodesy,30,3−35.
Wood , J. (20 05). Landserf Version 2.2 [GIS Software package]. URL http://www.landserf.org
110 0 B.M. Costa et al. / Remote Sensing of Environment 113 (2009) 1082–1100