See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/358204367

Automated high-resolution satellite-derived coastal bathymetry mapping

Article in International Journal of Applied Earth Observation and Geoinformation · March 2022
DOI: 10.1016/j.jag.2022.102693

CITATIONS READS

13 259

4 authors, including:

Matthew Mccarthy David Hughes

Oak Ridge National Laboratory Oak Ridge National Laboratory
19 PUBLICATIONS 478 CITATIONS 9 PUBLICATIONS 36 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Matthew Mccarthy on 07 February 2022.

The user has requested enhancement of the downloaded file.

 International Journal of Applied Earth Observations and Geoinformation 107 (2022) 102693

Contents lists available at ScienceDirect

International Journal of Applied Earth

Observations and Geoinformation
journal homepage: www.elsevier.com/locate/jag

Automated high-resolution satellite-derived coastal bathymetry mapping

Matthew J. McCarthy a, b, *, Daniel B. Otis b, David Hughes a, Frank E. Muller-Karger b
a
Oak Ridge National Laboratory, Geospatial Science and Human Security Division, Remote Sensing Group, 1 Bethel Valley Road, Oak Ridge, TN 37830, USA
b
Institute for Marine Remote Sensing, College of Marine Science, University of South Florida, 140 7th Ave. South, Saint Petersburg, FL 33701, USA

A R T I C L E I N F O A B S T R A C T

Keywords: Accurate and up-to-date maps of coastal bathymetry are fundamental for coastal resource management, com­
WorldView mercial and military navigation, and aquaculture, among many applications. Existing methods for bathymetry
Chlorophyll mapping require intensive and costly field surveys or targeted aerial captures, neither of which are easily or
Turbidity
affordably replicated for repeat mapping and change monitoring. Satellite-derived bathymetry (SDB), however,
High-performance computing
offers the potential to map shallow water bodies repeatedly and efficiently with high spatial and temporal
resolution (i.e., daily-weekly at 5 m or better). One challenge to large-scale implementation of SDB lies in the
automated derivation of inherent water column properties such that they may be accurately compensated for
across a variety of depth and substrate conditions. Here we present an algorithm that leverages WorldView
(Maxar/Digital GlobeTM) satellite imagery to map the entire 3700 km2 Florida Keys (USA) island chain at 2-meter
resolution without the need for any in-situ data collections. Preprocessing included radiometric calibration,
atmospheric correction, automated deglinting, and automated detection of optically deep water, which was then
used to estimate chlorophyll-a concentration assuming that the study area is primarily comprised of Case-I water
(i.e., those where the optical signal is dominated by water, Chlorophyll-a in phytoplankton, and properties that
vary in proportion to Chlorophyll-a concentration). Estimating Chlorophyll-a concentration allowed us to
calculate the appropriate tuning coefficients used in a spectral band ratio equation for estimating bathymetry.
The entire process was fully automated from ingestion of Level-1B image to bathymetry raster output. Mapping
the Florida Keys from Key Largo to Key West required 34 WorldView images and was completed in approxi­
mately 27 min for an average processing time of 47 s per image using a single GPU core (i.e., supercomputing
resources were not needed). After combining the products (mosaicking) in ArcMap, the wall-to-wall bathymetry
map was validated against a LiDAR-derived bathymetry model with over 600,000 points; results show an RMSE
of 1.95 m over depths from 0 to 15 m.

1. Introduction least the 1970s to estimate water depth in clear, shallow water. The
process is based on the observation of water-leaving radiance, knowl­
Accurate and up-to-date maps of coastal bathymetry are critical for edge of light attenuation with depth in clear water, and a model cali­
multiple sectors and applications, from vessel navigation to port con­ brated using in-situ depth measurements (Stumpf et al., 2003). More
struction, coastal resource management and urban planning amid complex approaches use radiative transfer modeling of the water body’s
accelerating sea-level rise (Mayer et al., 2018; Stocker et al., 2013). inherent optical properties (IOPs) (Hedley et al., 2005; Hughes et al.,
Traditional approaches to mapping bathymetry rely on ship soundings 2001; Kutser et al., 2020; Sandidge and Holyer, 1998). The remote
or LiDAR measurements from aerial vehicles that are costly, labor- sensing of aquatic properties includes many challenges, such as detect­
intensive, and time-consuming (Li et al., 2019; Stumpf et al., 2003). ing and correcting for sun glint, wave action, high suspended sediment,
As a result, approximately 70–80% of the global coastal zone lacks ac­ or type of bottom (Kay et al., 2009; Martin et al., 2016).
curate bathymetry maps, let alone those of high spatio-temporal reso­ Many SDB efforts choose water bodies that are Case 1 (i.e.,
lution (Caballero and Stumpf, 2020; Mayer et al., 2018). phytoplankton-dominated) such that Chlorophyll-a (Chl-a) is the pri­
Satellite-derived bathymetry (SDB) has existed in practice since at mary driver of light attenuation because the effect of phytoplankton on

* Corresponding author at: Oak Ridge National Laboratory, Geospatial Science and Human Security Division, Remote Sensing Group, 1 Bethel Valley Road, Oak
Ridge, TN 37830, USA.
E-mail addresses: [email protected] (M.J. McCarthy), [email protected] (D.B. Otis), [email protected] (D. Hughes), [email protected] (F.E. Muller-Karger).

https://doi.org/10.1016/j.jag.2022.102693
Received 8 November 2021; Received in revised form 5 January 2022; Accepted 18 January 2022
0303-2434/© 2022 Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
M.J. McCarthy et al. International Journal of Applied Earth Observation and Geoinformation 107 (2022) 102693

light attenuation is relatively well-constrained by decades of ocean color 2. Materials and methods
research. Therefore, recent SDB methods have focused on accounting for
attenuation due to Chl-a concentration in an image using empirical re­ 2.1. Study area
lationships (Li et al., 2019; Ma et al., 2021), assumptions about the
concentration (Kerr and Purkis, 2018), or by selecting “clean-water” We applied our automated algorithm to the Florida Keys, USA
images with negligible turbidity (Li et al., 2021). With the Chl-a con­ (Fig. 1), which spans ~183 km in length. We selected an area of 3700
centration for each image measured or assumed, a band-ratio approach km2 for mapping a target bathymetry range of 0–15 m over diverse
is then applied using coefficients tuned to account for Chlorophyll- benthic substrates (e.g. sand, coral, seagrasses, hard bottom). Much of
induced attenuation. However, “clean-water” images do not always the area has relatively clear “Case 1” waters (i.e., water column con­
exist, arbitrary Chl-a concentration assumptions can produce substantial stituents are phytoplankton and co-varying variables), and was mapped
errors, and confounding factors such as sunglint can preclude accurate with LiDAR in the wake of Hurricane Irma (NOAA Hurricane Irma,
estimates of Chl-a concentration from the spectral data. We developed Florida Keys: Supplemental LiDAR & Shoreline Mapping, 2020). We
an algorithm that fully automates the preprocessing, deglinting (as matched the area covered by the LiDAR bathymetry map between Key
needed), Chl-a concentration estimation, and bathymetry mapping of 2- Largo and Key West for comparison with our satellite-derived bathym­
meter resolution satellite imagery. etry products.
Recent advances in image processing and high-performance
computing have paved the way for efficient processing of large vol­ 2.2. WorldView imagery
umes of imagery and other data (McCarthy et al., 2018; Shelestov et al.,
2017). Simultaneous advances in site-independent SDB algorithms, We used 34 WorldView (WV) images collected by the WorldView-2
including forward physical-modeling of light attenuation with depth, (WV2) and WorldView-3 (WV3) satellite sensors owned and operated
and adaptive bathymetry estimation, no longer require field calibration by Maxar/Digital Globe™ (Table 1). WV2 and WV3 were launched in
(Kerr and Purkis, 2018; Li et al., 2019). In this study, we derived and 2009 and 2014, respectively, as polar-orbiting, push-broom multispec­
applied site-agnostic algorithms to map 34 satellite images covering the tral sensors that record data in eight channels spanning the visible to
Florida Keys from Key Largo to Key West (183 km) in a fully automated near-infrared spectral range (WV3 offers additional wavelengths not
protocol. included in this study) with spatial resolution in the 1-2-meter range
depending on viewing angle. Images were ordered in Level-1B format as
GeoTIFFs.

Fig. 1. The Florida Keys study area. The map was restricted to the region bounded in red to coincide with the extent of LiDAR data and available WorldView satellite
imagery (Imagery is from ArcGIS “Imagery” basemap). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of
this article.)

2
M.J. McCarthy et al. International Journal of Applied Earth Observation and Geoinformation 107 (2022) 102693

Table 1
Image name, location, and date for satellite imagery used.
Filename Lat (UL) Lat (LR) Lon (UL) Lon (LR) Year Month Day

10AUG13161421-M1BS-013431031010 25.11683 24.96801 − 80.4575 − 80.2885 2010 8 13

12NOV19163316-M1BS-500064643130 24.66029 24.52404 − 81.8459 − 81.6838 2012 11 19
12NOV19163317-M1BS-500064643130 24.55014 24.43911 − 81.8458 − 81.6838 2012 11 19
15APR18160013-M1BS-013455264010 24.79311 24.69694 − 81.504 − 81.3491 2015 4 18
15APR18160014-M1BS-013455264010 24.72845 24.63232 − 81.5038 − 81.3493 2015 4 18
15APR18160015-M1BS-013455264010 24.66385 24.59085 − 81.5036 − 81.3494 2015 4 18
15APR18160025-M1BS-013442579010 24.67771 24.58895 − 81.3653 − 81.2162 2015 4 18
15APR18160026-M1BS-013442579010 24.74106 24.64123 − 81.3653 − 81.2162 2015 4 18
15APR18160028-M1BS-013442579010 24.86829 24.76559 − 81.3654 − 81.2161 2015 4 18
15APR18160100-M1BS-013442580010 24.7481 24.62952 − 81.2334 − 81.08 2015 4 18
15APR18160101-M1BS-013442580010 24.68356 24.56978 − 81.2332 − 81.0802 2015 4 18
15APR24155535-M1BS-013442582010 24.80838 24.69489 − 81.1017 − 80.9302 2015 4 24
15APR24155536-M1BS-013442582010 24.74368 24.63041 − 81.1015 − 80.9304 2015 4 24
15APR24155537-M1BS-013442582010 24.67905 24.56788 − 81.1012 − 80.9306 2015 4 24
15NOV12161727-M1BS-013442583010 25.15469 25.0018 − 80.5146 − 80.3206 2015 11 12
16JAN16162157-M1BS-013442581010 24.69494 24.59645 − 81.6951 − 81.5602 2016 1 16
16JAN16162158-M1BS-013442581010 24.63002 24.53146 − 81.695 − 81.5602 2016 1 16
17FEB15160921-M1BS-013442584010 24.72282 24.58756 − 80.8991 − 80.7355 2017 2 15
17FEB15160923-M1BS-013442584010 24.83317 24.68696 − 80.8992 − 80.7356 2017 2 15
17OCT19162454-M1BS-013199222010 24.83919 24.7422 − 80.9713 − 80.8332 2017 9 19
17OCT19162509-M1BS-013199223010 24.69588 24.59331 − 81.0893 − 80.953 2017 9 19
17OCT19162539-M1BS-013199221010 24.83692 24.72405 − 81.3384 − 81.1915 2017 9 19
17OCT19162555-M1BS-013199224010 24.76347 24.64781 − 81.4713 − 81.3158 2017 9 19
17OCT19162556-M1BS-013199224010 24.82932 24.71104 − 81.4721 − 81.3151 2017 9 19
18AUG22162606-M1BS-013442587010 25.21654 25.11778 − 80.3729 − 80.2305 2018 8 22
18AUG22162607-M1BS-013442587010 25.15188 25.05334 − 80.3729 − 80.2306 2018 8 22
18AUG22162608-M1BS-013442587010 25.08727 24.98899 − 80.3728 − 80.2306 2018 8 22
18FEB16162712-M1BS-013442578010 24.99815 24.97844 − 80.6663 − 80.4573 2018 2 16
18NOV10161106-M1BS-013442588010 25.01037 24.8762 − 80.6638 − 80.4962 2018 11 10
18NOV10161107-M1BS-013442588010 24.91082 24.76616 − 80.6637 − 80.4964 2018 11 10
18NOV10161120-M1BS-013442589010 24.83467 24.68783 − 80.8138 − 80.6468 2018 11 10
18NOV10161121-M1BS-013442589010 24.94515 24.79716 − 80.8139 − 80.6467 2018 11 10
18NOV10161216-M1BS-013442586010 24.70899 24.56988 − 81.6048 − 81.4157 2018 11 10
19APR28155605-M1BS-013442590010 25.0306 24.87505 − 80.5873 − 80.3639 2019 4 28

Fig. 2. Automated bathymetry mapping pipeline.

3
M.J. McCarthy et al. International Journal of Applied Earth Observation and Geoinformation 107 (2022) 102693

2.3. Image preprocessing comprises similar attenuation conditions throughout.

Our approach is to further automate the Li et al. (2019) method by
Image processing followed the pipeline shown in Fig. 2 to stan­ automatically identifying optically deep water, using it to estimate the
dardize all images for accurate comparative analysis. Preprocessing image’s attenuation properties (i.e. Chl-a), and calculating the tuning
included transforming digital number to at-sensor radiances using the coefficients from a revised version of the Li et al. (2019) Eqs. (10) and
calibration factors supplied with each image’s metadata. Further pro­ (11).
cessing followed the atmospheric- and air-water-interface-correction Water was identified as pixels with a reflectance value < 0.2 in the
protocol of Kerr and Perkis (2018 equations 2–6 and references NIR1 band (i.e., band 7). Optically deep water (ODW) was defined as
therein), which account for date, time, Earth-Sun distance, latitude, water pixels containing NIR1 values within the 5th to 10th percentile of
longitude, and Sun and satellite geometries. Rayleigh path radiance was all water pixels. This range was selected based on Overstreet and
calculated using the protocol of Dash et al. (2012 and references Legleiter (2017), and the observation that the darkest NIR1 pixels (i.e. 0-
therein). The output generated was subsurface remote sensing reflec­ 5th percentile) were anomalous dark bodies (e.g. cloud shadow or dark
tance (rrs). freshwater bodies) that did not represent the coastal-water attenuation
properties targeted for this work.
Chlorophyll-a concentration (Chla) was estimated based on:
2.4. Deglinting
w = rrs (Green) − 0.46*rrs (Red) − 0.54*rrs (Coastal) (4)
Sun glint is a common problem in aquatic satellite images caused by
the specular reflection of sunlight on the surface of the water, which Chla = 10(− 0.4909+191.659*w)
(5)
often saturates a sensor. Glinted pixels must either be corrected or dis­
where w was calculated for every ODW pixel before taking the me­
carded from further processing. Correcting for sun glint in high-
dian of those w values less than 0 (Hu et al., 2012; Werdell and Bailey,
resolution images can be done in a variety of ways (Hochberg et al.,
2005). We replaced the Blue band with the Coastal band in Eq. (4)
2003; Kay et al., 2009), but typically requires manual identification of
because the original equation by Hu et al. (2012) used wavelength 443
glinted samples, which precludes automation. To incorporate glint
nm, which is closer to the center wavelength of the Coastal Band, and
correction into our automated protocol we first predicted the likelihood
following the Kerr and Perkis (2018) finding that, in Case 1 waters, the
of glint presence in the scene. Sun glint depends on the sun-target-sensor
Coastal band has a lower attenuation rate than the Blue band. Tuning
geometry, and is most likely to occur when sun and sensor azimuths are
coefficients were calculated as:
approximately 180 degrees apart, and with higher zenith angles (Jack­
son and Alpers, 2010; Zhou et al., 2017). To predict glint likelihood, we B1 = 52.083*e(1.77*Chla) (6)
derived the following equations to quantify relative sun and sensor az­
imuth and zenith angles, normalized to their maximum likely values: B0 = 50.156*e(1.70*Chla) (7)
AZ rel = (|AZsun − AZ sat − 180|)/180 (1) The exponential scalars from Li et al. (2019) were established for the
Planet Dove satellites, so we calculated ours by setting B1 and B0 equal
ZN rel = (ZNsun + ZN sat )/120 (2) to the field-based values calculated by Kerr and Perkis (2018) for the
where AZ refers to azimuth angles, and ZN refers to zenith angles. By Florida Keys WorldView image (68.3 and 65.1, respectively), and using
summing AZrel and ZNrel we would conservatively predict that sun glint our Chla value estimated from the same image. Kerr and Perkis (2018)
is likely for values less than or equal to 0.65. This threshold was deter­ reported that setting their Chla estimate for this image to 0.2 mg m− 3
mined by manually evaluating an ancillary dataset containing glinted achieved the most accurate results. Our estimate for this image using
and glint-free images. equations 4–5 was 0.15 mg m− 3.
If the scene was likely to contain sun glint, glint correction was
performed as described in McCarthy et al. (2020) to automatically 2.6. Implementation
identify the glinted-pixel spectral profile and correct it for every glinted
image prior to bathymetry estimation. We ran this algorithm on the 34 WV images with scripts written in
Python and processed using a single core of the 1 Tesla K80 GPU running
2.5. Bathymetry algorithm on a Dell C4130 server node at Oak Ridge National Laboratory. Images
were run one at a time, but this method could be parallelized for future
Bathymetry was calculated with the Stumpf et al. (2003) band ratio implementation. Advanced computational resources (e.g., high-
method as performance computing clusters) were not necessary for implementing
( ) the algorithm, but could be leveraged for enhanced performance.
ln Mrrs,i
z = B1 ( ) + B0 (3) Output maps from each image were then mosaicked into a single
ln Mrrs,j bathymetry map in ArcMap. Overlap between scenes was handled by
averaging coinciding pixels. As in Li et al. (2019), we did not correct the
where i and j correspond to the blue and green bands, respectively, M
tidal effect of the WV-derived depth because the reference DTMs were
is a scaling factor of 1000, and B1 and B0 are tuning coefficients. This
collected over a period of three months at varying times of day and were
model is based on the exponential attenuation of light with depth, the
not individually time-stamped.
rate of which varies with wavelength and water-column optical prop­
erties. The tuning coefficients, B1 and B0, were derived by Stumpf et al.
(2003) through linear regression of a band ratio from a single satellite 2.7. Validation
image against observed depths from 0 to 12 m. Kerr and Perkis (2018)
developed a forward-modeled algorithm to estimate the coefficients for Bathymetry was validated by calculating RMSE using the NOAA
multiple satellite sensors in different study areas with specific benthic LiDAR-derived digital topobathymetric models (DTMs) as reference data
substrate types and assuming a Chl-a concentration of 0.2 mg m− 3. Li (1-meter horizontal resolution). DTMs were created from LiDAR data
et al. (2019) developed an adaptive bathymetry estimation algorithm acquired by aerial survey between November 2018 and March 2019.
that relies on the assumptions that (1) optically deep water in an image Bathymetry DTM vertical accuracy is reported at 11 cm (NOAA Hurri­
can be used to estimate the inherent and apparent water column prop­ cane Irma, Florida Keys: Supplemental LiDAR & Shoreline Mapping, 2020).
erties, and (2) that the relatively small coverage of a single image Validation points were created in ArcMap using the Create Random

4
M.J. McCarthy et al. International Journal of Applied Earth Observation and Geoinformation 107 (2022) 102693

Points tool by digitizing a bounding box for the study area and casting Table 2
points within it that were separated from each other by at least 10 m to RMSE and sample size per depth range.
avoid spatial autocorrelation. Quality control included eliminating Depth (m) RMSE (m) n
points that intersected cloud cover by digitizing clouds and removing
<15 1.95 635,081
intersected points. Additionally, DTMs featured data gaps surrounded by <12 1.90 629,169
sparse, pixelated bathymetry data, which we assumed to correspond to <10 1.85 620,387
poor LiDAR returns. We eliminated points that intersected data gaps and <8 1.73 583,292
data-sparse regions. Bathymetry values were extracted for each point <6 1.52 500,988
1.28 397,549
from our map and the DTM. Points that intersected land or contained No
<4
<2 1.16 173,431
Data from either dataset were omitted from validation. A total of
635,081 validation points were used for estimating the accuracy of the
resulting bathymetry map (Fig. 3). collections are displayed in Fig. 4. Our SDB map contains spatial pat­
terns consistent with the DEM from the shallow Florida Bay region north
3. Results of the Keys to areas > 15 m. Artifacts, however, are noticeable where
images overlap. The averaging of overlapping pixels resulted in
3.1. Processing time smoother transitions than simply using one image’s pixel values, but
imperfections in preprocessing, atmospheric correction, air-water
Image processing was completed for all 34 images in 27 min, for an interface correction, and water column correction are likely to create
average processing time from ingestion of Level-1B images to output of unavoidable artifacts. Additionally, inconsistencies may be expected
mapped bathymetry of 47 s. This equated to mapping 138 km2 per where overlapping images were collected at substantially different times
minute for this study area. (e.g. over several years) and bathymetry changed.

4. Discussion
3.2. Bathymetry
This study derives and implements a fully automated algorithm for
3.2.1. Algorithm validation
mapping coastal bathymetry from 2-meter resolution satellite imagery
RMSE was calculated for the entire map using all 635,081 validation
without ground calibration data. The resulting wall-to-wall maps pro­
points, as well as for 2-meter depth increments (Table 2). The overall
vide accurate bathymetry down to 15 m depth, and account for atmo­
RMSE was 1.95 m (0–15 m depth), and shallow depths saw consistent
spheric attenuation, turbidity, and sun glint. We build on recent
decreased error down to the 0–2 m increment with an RMSE of 1.16 m.
advances in the estimation of Chl-a concentration from optical imagery
to account for inherent optical properties by refining algorithms to
3.2.2. Spatial patterns
automate the identification of optically deep water. Through this
Bathymetry maps from this study and previous LiDAR DEM

Fig. 3. Points used to evaluate map accuracy were created randomly throughout the map and separated from each other by 10 m. Inset shows land (i.e., positive
elevation) and gaps in the validation DEM as white space that were excluded from validation.

5
M.J. McCarthy et al. International Journal of Applied Earth Observation and Geoinformation 107 (2022) 102693

1.22–1.86 m (Li et al., 2019), and 3.829 m (Zhu et al., 2020). Further,
Fig. 5 demonstrates the consistent estimation of bathymetry patterns
between our map and the validation data from shallow (i.e., <2 m) to
deep (15 m) water along a 15 km transect in the middle of the study
area. Two areas of erroneously shallow values appear to be caused by a
sediment plume from the adjacent channel (center of plot), and a signal-
to-noise weakness of the sensor to accurately detect deeper depths (right
side of plot) as noted by Kerr and Perkis (Kerr and Purkis, 2018). Future
work will include correcting for tides, which was not done here because
the maps produced were validated against DTMs that were not tidally
corrected.

Data availability statement

WorldView satellite imagery are commercial and are not available

for public dissemination but may be ordered through the provider,
DigitalGlobe. Validation DTMs may be downloaded through the NOAA
DigitalCoast portal (https://coast.noaa.gov/digitalcoast/tools/dav.
html).
Fig. 4. Our WorldView-based map compared with the DEM validation map.
Inset highlights bathymetric features along Key West.
Funding

refinement one is able to calculate used in the robust band ratio method This work was supported by the United States Department of Energy
(Stumpf et al., 2003). [proposal number 0000-Z350-20]; NASA, United States [grant numbers
Recent efforts to advance large-scale SDB mapping successfully NNX14AP62A and 80NSSC20K0017]; NSF, United States [grant number
developed IOP-estimation algorithms and demonstrated accurate deri­ 1762493]; and NOAA United States Integrated Ocean Observing System
vation of tuning coefficients, but relied on single scenes for a given area [grant number NA19NOS0120199].
(Kerr and Purkis, 2018; Zhu et al., 2020), or those manually selected for
pristine viewing conditions that were then mosaicked during pre­ CRediT authorship contribution statement
processing to ensure consistent radiometry (Li et al., 2019). Our
approach accounts for turbid water (assuming Chl-a-dominated Matthew J. McCarthy: Conceptualization, Methodology, Software,
turbidity), automatically identifies and corrects for any apparent sun Validation, Formal analysis, Writing – original draft. Daniel B. Otis:
glint, and applies robust radiometric calibration and atmospheric Conceptualization, Methodology, Validation. David Hughes: Concep­
correction. These advances are significant in that they are more widely tualization, Resources, Supervision, Writing – review & editing, Project
applicable to a variety of scene conditions (i.e., glinted or glint-free, administration. Frank E. Muller-Karger: Conceptualization, Resources,
moderately turbid or clear, collected across multiple years and all sea­ Supervision, Writing – review & editing, Funding acquisition, Project
sons), and require no manual preprocessing. As a result, this algorithm administration.
completed mapping from ingestion of Level-1B imagery to bathymetry-
map output in less than one minute per image. Accuracy based on RMSE
was 1.95 m, compared with 0.89–2.62 m (Kerr and Purkis, 2018),

Fig. 5. Depth along a transect: WorldView map versus DEM validation map. The gap in values corresponds to land.

6
M.J. McCarthy et al. International Journal of Applied Earth Observation and Geoinformation 107 (2022) 102693

Declaration of Competing Interest Kutser, T., Hedley, J., Giardino, C., Roelfsema, C., Brando, V.E., 2020. Remote sensing of
shallow waters – a 50 year retrospective and future directions. Remote Sens.
Environ. 240, 111619 https://doi.org/10.1016/j.rse.2019.111619.
The authors declare that they have no known competing financial Li, J., Knapp, D.E., Schill, S.R., Roelfsema, C., Phinn, S., Silman, M., Mascaro, J.,
interests or personal relationships that could have appeared to influence Asner, G.P., 2019. Adaptive bathymetry estimation for shallow coastal waters using
the work reported in this paper. Planet Dove satellites. Remote Sens. Environ. 232, 111302. https://doi.org/
10.1016/j.rse.2019.111302.
Li, J., Knapp, D.E., Lyons, M., Roelfsema, C., Phinn, S., Schill, S.R., Asner, G.P., 2021.
Acknowledgements Automated global shallow water bathymetry mapping using google earth engine.
Remote Sensing 13, 1469. https://doi.org/10.3390/rs13081469.
Ma, Y., Zhang, H., Li, X., Wang, J., Cao, W., Li, D., Lou, X., Fan, K., 2021. An exponential
Notice: This manuscript was authored by UT-Battelle, LLC, under algorithm for bottom reflectance retrieval in clear optically shallow waters from
contract DE-AC05-00OR22725 with the US Department of Energy multispectral imagery without ground data. Remote Sensing 13, 1169. https://doi.
(DOE). The US government retains and the publisher, by accepting the org/10.3390/rs13061169.
Martin, J., Eugenio, F., Marcello, J., Medina, A., 2016. Automatic sun glint removal of
article for publication, acknowledges that the US government retains a multispectral high-resolution worldview-2 imagery for retrieving coastal shallow
non-exclusive, paid-up, irrevocable, worldwide license to publish or water parameters. Remote Sensing 8, 37. https://doi.org/10.3390/rs8010037.
reproduce the published form of this manuscript, or allow others to do Mayer, L., Jakobsson, M., Allen, G., Dorschel, B., Falconer, R., Ferrini, V., Lamarche, G.,
Snaith, H., Weatherall, P., 2018. The nippon foundation—GEBCO seabed 2030
so, for US government purposes. DOE will provide public access to these project: the quest to see the world’s oceans completely mapped by 2030. Geosciences
results of federally sponsored research in accordance with the DOE 8, 63. https://doi.org/10.3390/geosciences8020063.
Public Access Plan. Disclosures: The authors declare no conflicts of McCarthy, M.J., Radabaugh, K.R., Moyer, R.P., Muller-Karger, F.E., 2018. Enabling
efficient, large-scale high-spatial resolution wetland mapping using satellites.
interest.
Remote Sens. Environ. 208, 189–201.
McCarthy, M.J., Jessen, B., Barry, M.J., Figueroa, M., McIntosh, J., Murray, T.,
References Schmid, J., Muller-Karger, F.E., 2020. Automated high-resolution time series
mapping of mangrove forests damaged by hurricane Irma in Southwest Florida.
Caballero, I., Stumpf, R.P., 2020. Atmospheric correction for satellite-derived Remote Sensing 12, 1740. https://doi.org/10.3390/rs12111740.
bathymetry in the Caribbean waters: from a single image to multi-temporal NOAA Hurricane Irma, Florida Keys: Supplemental LiDAR & Shoreline Mapping
approaches using Sentinel-2A/B. Opt. Express 28 (8), 11742. https://doi.org/ (Technical Data Report No. FL-1806-TB-C), 2020. Quantum Spatial, Corvallis, OR.
10.1364/OE.390316. Overstreet, B.T., Legleiter, C.J., 2017. Removing sun glint from optical remote sensing
Dash, P., Walker, N., Mishra, D., D’Sa, E., Ladner, S., 2012. Atmospheric correction and images of shallow rivers: removing sun glint from river imagery. Earth Surf. Process.
vicarious calibration of oceansat-1 Ocean Color Monitor (OCM) data in coastal case 2 Landforms 42 (2), 318–333. https://doi.org/10.1002/esp.4063.
waters. Remote Sensing 4, 1716–1740. https://doi.org/10.3390/rs4061716. Sandidge, J.C., Holyer, R.J., 1998. Coastal bathymetry from hyperspectral observations
Hedley, J.D., Harborne, A.R., Mumby, P.J., 2005. Technical note: simple and robust of water radiance. Remote Sens. Environ. 65 (3), 341–352. https://doi.org/10.1016/
removal of sun glint for mapping shallow-water benthos. Int. J. Remote Sens. 26 S0034-4257(98)00043-1.
(10), 2107–2112. https://doi.org/10.1080/01431160500034086. Shelestov, A., Lavreniuk, M., Kussul, N., Novikov, A., Skakun, S., 2017. Exploring google
Hochberg, E.J., Andrefouet, S., Tyler, M.R., 2003. Sea surface correction of high spatial earth engine platform for big data processing: classification of multi-temporal
resolution ikonos images to improve bottom mapping in near-shore environments. satellite imagery for crop mapping. Front. Earth Sci. 5 https://doi.org/10.3389/
IEEE Trans. Geosci. Remote Sensing 41 (7), 1724–1729. https://doi.org/10.1109/ feart.2017.00017.
TGRS.2003.815408. Stocker, T.F., Qin, D., Plattner, G.-K., Tignor, M., Allen, S.K., Boschung, J., Nauels, A.,
Hu, C., Lee, Z., Franz, B., 2012. Chlorophyll a algorithms for oligotrophic oceans: a novel Xia, Y., Bex, V., Midgley, P.M., 2013. Climate Change 2013: The Physical Science
approach based on three-band reflectance difference: a novel ocean chlorophyll A Basis. Cambridge University Press, Cambridge, UK.
algorithm. J. Geophys. Res. 117 (C1) https://doi.org/10.1029/2011JC007395. Stumpf, R.P., Holderied, K., Sinclair, M., 2003. Determination of water depth with high-
Hughes, D., Holyer, R., Lee, Z., 2001. Remote sensing algorithms by numerical inversion resolution satellite imagery over variable bottom types. Limnol. Oceanogr. 48
of radiative transfer models: neural network and optimization methods compared. (1part2), 547–556. https://doi.org/10.4319/lo.2003.48.1_part_2.0547.
In: Presented at the International Conference on Current Problems in Optics of Werdell, P.J., Bailey, S.W., 2005. An improved in-situ bio-optical data set for ocean color
Natural Waters, D.S. Rozhdestvensky Optical Society, St. Petersburg, Russia. algorithm development and satellite data product validation. Remote Sens. Environ.
Jackson, C.R., Alpers, W., 2010. The role of the critical angle in brightness reversals on 98 (1), 122–140. https://doi.org/10.1016/j.rse.2005.07.001.
sunglint images of the sea surface. J. Geophys. Res. 115 (C9) https://doi.org/ Zhou, G., Xu, W., Niu, C., Zhang, K., Ma, Z., Wang, J., Zhang, Y., 2017. Versatile time-
10.1029/2009JC006037. dependent spatial distribution model of sun glint for satellite-based ocean imaging.
Kay, S., Hedley, J., Lavender, S., 2009. Sun glint correction of high and low spatial J. Appl. Remote Sens 11 (1), 016020. https://doi.org/10.1117/1.JRS.11.016020.
resolution images of aquatic scenes: a review of methods for visible and near- Zhu, J., Hu, P., Zhao, L., Gao, L., Qi, J., Zhang, Y., Wang, R., 2020. Determine the stumpf
infrared wavelengths. Remote Sensing 1, 697–730. https://doi.org/10.3390/ 2003 model parameters for multispectral remote sensing shallow water bathymetry.
rs1040697. J. Coastal Res. 102 https://doi.org/10.2112/SI102-007.1.
Kerr, J.M., Purkis, S., 2018. An algorithm for optically-deriving water depth from
multispectral imagery in coral reef landscapes in the absence of ground-truth data.
Remote Sens. Environ. 210, 307–324. https://doi.org/10.1016/j.rse.2018.03.024.

View publication stats