remote sensing
Article
Analyzing Performances of Different Atmospheric
Correction Techniques for Landsat 8: Application for
Coastal Remote Sensing
Christopher O. Ilori 1,*, Nima Pahlevan 2,3 and Anders Knudby 4
1 Simon Fraser University, 8888 University Drive, Burnaby, BC V5A 1S6, Canada
2 NASA Goddard Space Flight Center, 8800 Greenbelt Road, Greenbelt, MD 20771, USA;
nima.pahlevan@nasa.gov
3 Science Systems and Applications, Inc., 10210 Greenbelt Road, Suite 600 Lanham, MD 20706, USA
4 University of Ottawa, 60 University Private, Ottawa, ON K1N 6N5, Canada; aknudby@uottawa.ca
* Correspondence: cilori@sfu.ca; Tel.: +1-778-929-5350
Received: 28 January 2019; Accepted: 18 February 2019; Published: 25 February 2019





Abstract: Ocean colour (OC) remote sensing is important for monitoring marine ecosystems.
However, inverting the OC signal from the top-of-atmosphere (TOA) radiance measured by satellite
sensors remains a challenge as the retrieval accuracy is highly dependent on the performance of
the atmospheric correction as well as sensor calibration. In this study, the performances of four
atmospheric correction (AC) algorithms, the Atmospheric and Radiometric Correction of Satellite
Imagery (ARCSI), Atmospheric Correction for OLI ‘lite’ (ACOLITE), Landsat 8 Surface Reflectance
(LSR) Climate Data Record (Landsat CDR), herein referred to as LaSRC (Landsat 8 Surface Reflectance
Code), and the Sea-Viewing Wide Field-of-View Sensor (SeaWiFS) Data Analysis System (SeaDAS),
implemented for Landsat 8 Operational Land Imager (OLI) data, were evaluated. The OLI-derived
remote sensing reflectance (Rrs) products (also known as Level-2 products) were tested against
near-simultaneous in-situ data acquired from the OC component of the Aerosol Robotic Network
(AERONET-OC). Analyses of the match-ups revealed that generic atmospheric correction methods
(i.e., ARCSI and LaSRC), which perform reasonably well over land, provide inaccurate Level-2
products over coastal waters, in particular, in the blue bands. Between water-specific AC methods
(i.e., SeaDAS and ACOLITE), SeaDAS was found to perform better over complex waters with
root-mean-square error (RMSE) varying from 0.0013 to 0.0005 sr−1
for the 443 and 655 nm channels,
respectively. An assessment of the effects of dominant environmental variables revealed AC retrieval
errors were influenced by the solar zenith angle and wind speed for ACOLITE and SeaDAS in
the 443 and 482 nm channels. Recognizing that the AERONET-OC sites are not representative of
inland waters, extensive research and analyses are required to further evaluate the performance of
various AC methods for high-resolution imagers like Landsat 8 and Sentinel-2 under a broad range
of aquatic/atmospheric conditions.
Keywords: atmospheric correction; remote sensing reflectance; Landsat 8; band adjustment;
validation; AERONET-OC
1. Introduction
Ocean colour (OC) remote sensing provides information on in-water optical properties indicating
the concentrations of water constituents such as chlorophyll-a. In optically shallow waters, depth and
seafloor spectral reflectance may also be estimated using remotely sensed images. Information about
near-surface, in-water optical properties, in the form of water quality maps, can provide advance
warning of algal bloom development [1] and potentially lead to early mitigation efforts to reduce
Remote Sens. 2019, 11, 469; doi:10.3390/rs11040469 www.mdpi.com/journal/remotesensing
Remote Sens. 2019, 11, 469 2 of 20
health risks and financial losses. Bathymetric maps, derived from water depth estimates, can be used
to produce or update navigational charts [2], reducing the risk of ship groundings. Benthic habitat
maps, inferred from seafloor spectral reflectance, can be used to track changes in the distribution of
seafloor habitats [3,4]. However, extracting ocean colour products such as chlorophyll-a, water depth,
and bottom types from remotely sensed images is difficult because, over blue ocean waters, only ~10%
of the total signal that reaches the TOA (Top of Atmosphere) typically comes from within the water
column [5]. In addition to the radiation leaving the water column (Lw), the TOA radiance measured by
satellite sensors includes contributions from scattering and absorption in the atmosphere and reflection
at the sea surface [6]. It is important to estimate and account for the contribution from these other
sources in order to estimate Lw, which is readily normalized by the total downwelling irradiance just
above the sea surface to yield the remote sensing reflectance (Rrs).
Atmospheric effects are removed through atmospheric correction (AC) [7], but residual errors in
AC can introduce large uncertainties in Rrs estimates, resulting in erroneous retrieval of OC products
such as apparent optical properties [8]. In open ocean waters, where phytoplankton governs the optical
regime, it can be conveniently assumed that there is no water-leaving radiance in the near-infrared
(NIR) region, such that any measured TOA radiance in this spectral band is attributed to atmospheric
path radiance and reflectance from the water surface. While this assumption is valid for open-ocean
waters, in shallow or optically complex waters that are, in general, characterized by a combination of
constituents, such as phytoplankton, coloured dissolved organic matters, and suspended particulate
matters, Rrs(NIR) may be significantly greater than zero because of bottom reflectance (which can
come from highly reflective near-surface vegetation such as kelp and seagrasses) and backscattering
by suspended materials [9]. This can lead to over-correction of atmospheric and surface reflectance
effects, leading to underestimation and even negative Lw estimates within the shorter wavelengths
used to derive OC products [9,10]. To account for the non-negligible Rrs(NIR), algorithms that
work for Case 2 waters have been developed (e.g., [10–12]) and tested (e.g., [13,14]). With these
efforts, it is now possible to retrieve Lw over coastal waters. However, the low spatial resolution of
traditional ocean colour sensors inhibits the detection of detailed features that are not easily discernible
in coarse-resolution satellite images. The recent availability of higher-resolution satellite sensors,
e.g., Landsat 8 Operational Land Imager (OLI) and Sentinel-2 Multispectral Instrument (MSI), with
adequate spectral and radiometric characteristics for ocean colour applications, has the potential to
greatly improve coastal ocean colour applications, including water quality [15,16], bathymetry, and
seafloor habitat mapping [17].
Optimizing the utility of the OLI sensor for aquatic science and applications requires validating
Rrs products to better understand their potential and limitations. A few recent studies have started
investigating the quality of OLI-derived Rrs for coastal applications. For example, Pahlevan et al. [15]
used the AC scheme in SeaDAS software package to determine the best Landsat 8 band combinations
that can minimize error in Rrs retrieval over different coastal water types at select AERONET-Ocean
Colour (OC) sites. Using newly computed calibration gains, they revealed that OLI-derived Rrs
estimates are as good as those from other ocean colour sensors. Likewise, the work of Franz et al.
(2015) [18] illustrated the potential of OLI data for water-related studies. By employing the AC process
in the SeaDAS Level-2 processing algorithm (l2gen), they assessed the quality of OLI-derived Rrs and
subsequently retrieved the chlorophyll-a concentration over the Chesapeake Bay, USA. In agreement
with Pahlevan et al. (2017a) [15], they found that with a precise AC procedure, the high radiometric
quality and improved imaging capabilities of OLI hold great promise for satellite-based coastal
monitoring. Doxani et al. (2018) [19] tested a wide range of AC algorithms over different land cover
types, highlighting the strengths and limitations of each algorithm. More recently, Wei et al. (2018) [20]
also assessed four AC algorithms with a focus over water bodies, revealing that the NIR-short-wave
infrared (SWIR) approach implemented in SeaDAS produced the most robust Rrs estimates from
Landsat 8. However, none of these studies examined the effects of environmental variables on the Rrs
retrieval accuracy of different AC algorithms.
Remote Sens. 2019, 11, 469 3 of 20
With the increase in usage of OC products among the science community, and the need for robust
Rrs products, it is important to understand the potential of AC algorithms. Most of the ocean colour
community has for years been using water-based AC methods for a wide range of applications from
coastal to inland waters, so it is important that the effects of relevant environmental variables on the
Rrs retrieval accuracy of these AC algorithms is examined. Such knowledge may assist in the choice
of AC algorithm for a given set of environmental conditions, and/or improved Rrs retrieval under a
wider range of conditions. Equally important is that other users interested in studying inland waters
(e.g., biogeochemists, aquatic biologists) fully understand the accuracy of generic AC processors,
in particular the land surface reflectance product, which is commonly used.
Here, we pursued an approach similar to Pahlevan et al. (2017a) [15], but expanded it by
evaluating the performances of four different AC algorithms to determine which method produces
the most robust Rrs products in shallow coastal waters. Also, like Doxani et al. (2018) [19], we tested
both land-based and water-based algorithms at multiple sites, but covered more sites over a longer
time period to better capture space-time dynamics related to water optical properties. Using 54 in-situ
measurements from 14 AERONET-OC sites, we (1) tested the following algorithms for atmospheric
correction of Landsat 8 images: (a) Atmospheric and Radiometric Correction of Satellite Imagery
(ARCSI) [21], (b) the Atmospheric Correction for OLI ‘lite’ (ACOLITE) [22], (c) SeaDAS [8], and (d) the
United States Geological Survey’s standard land-based AC used to produce the Landsat 8 Surface
Reflectance (LSR) Climate Data Record (Landsat CDR), herein referred to as LaSRC [23]; (2) analysed
the differences in spectral bands between satellite and in-situ measurements; and (3) examined
the effects of three key environmental variables on the Rrs retrieval accuracy of water-based AC
algorithms. To our knowledge, this is the first inter-comparison exercise that tested AC algorithms
using representative data from many coastal sites with varying atmospheric conditions and optical
properties, combining the three approaches mentioned above in a single study.
2. Materials and Methods
2.1. Landsat 8 OLI Data
OLI measures TOA radiance in the visible, NIR, and short-wave infrared (SWIR) bands, at a spatial
resolution of 30 m. Compared to its predecessors, OLI includes a new coastal/aerosol band
(435–451 nm) in addition to the traditional blue (452–512 nm), green (533–590 nm), and red (636–673 nm)
bands. The addition of a new band, together with enhanced spectral coverage and radiometric
resolution, enables improved observation of water bodies from space and the ability to estimate
the concentration of atmospheric aerosols for AC [16,24,25] (note that aerosol estimation for AC by
the coastal band is done over land). Compared to data from existing global ocean colour missions,
the higher spatial resolution has the potential to make important contributions to ocean colour remote
sensing, such as separating and mapping in-water constituents in coastal waters [16,25]. Although OLI
signal-to-noise ratios (SNRs) (Table 1) are generally lower than those of heritage ocean colour sensors
such as SeaWiFS or the Moderate Resolution Imaging Spectroradiometer onboard Aqua (MODISA),
its enhanced SNR across all bands compared to the past Landsat missions improves OLI’s ability to
measure subtle variability in near-surface conditions and ultimately make OLI products a valuable
source of data for ocean colour studies [16,18].
Table 1. Comparison of the band centres and the signal-to-noise ratios of MODIS and Landsat 8
Operational Land Imager (OLI) at specified levels of typical spectral radiance.
Band centres (nm)
MODIS 443 488 555 645 858 1640 2130
SeaWiFS 443 490 555 670 865 NA NA
OLI 443 482 56 655 865 1609 2201
Remote Sens. 2019, 11, 469 4 of 20
Table 1. Cont.
Signal-to-noise ratio (SNR)
MODIS 838 802 228 128 201 275 110
SeaWiFS 950 1000 850 500 350 NA NA
OLI 344 478 279 144 67 30 14
Ltyp (w m−2 µ
−1
sr−)
MODIS 4.9 32.1 29 21.8 24.7 7.3 1.0
SeaWiFS 70.2 53.1 33.9 8.3 4.5 NA NA
OLI 69.8 55.3 27.5 13.4 4.06 0.353 0.0467
2.2. AERONET-OC Data
To validate the performance of the AC processors applied to the OLI data, we acquired
122 cloud-screened and fully quality-controlled Level 2.0 AERONET-OC in-situ measurements
of normalized water-leaving radiance (nLw) for 14 AERONET-OC sites, including 12 coastal sites
(i.e., Galata Platform, Gloria, GOT Seaprism, Gustav Dalen Tower, Helsinki Lighthouse, Long Island
Sound Coastal Observatory (LISCO), Martha’s Vineyard Coastal Observatory (MVCO), Thornton
C-Power, USC Seaprism, Venise, WaveCIS Site CSI-6, Zeebrugge-MOW1) and two lake sites (Lake
Erie, Palgrunden) (Figure 1). AERONET-OC, managed by NASA’s Goddard Space Flight Center
(GSFC) [26], is a sub-network of the AERONET federated instrument [27,28]. Although OLI and
AERONET-OC have somewhat different spectral bands, a set of comparable bands centred at 441 nm,
491 nm, 551 nm, and 667 nm can be used for cross-comparison purposes. Thus, AERONET-OC data
were collected in four spectral bands centred at 441, 491, 551, and 667 nm, for comparison with OLI’s
four visible bands, centred at 443, 483, 561 and 655 nm. Note that there is a slight difference, i.e., ±1 to
±3 nm, in all four bands among measurements at different sites. As Rrs is not directly available from
the AERONET-OC sites, the normalized water-leaving radiances (nLw, W m−2
sr−1
) were divided by
the top-of-the atmosphere (TOA) solar irradiance (F0) [29] to obtain Rrs.
Remote Sens. 2018, 10, x FOR PEER REVIEW 4 of 20
2.2. AERONET-OC Data
To validate the performance of the AC processors applied to the OLI data, we acquired 122
cloud-screened and fully quality-controlled Level 2.0 AERONET-OC in-situ measurements of
normalized water-leaving radiance (nLw) for 14 AERONET-OC sites, including 12 coastal sites (i.e.,
Galata Platform, Gloria, GOT Seaprism, Gustav Dalen Tower, Helsinki Lighthouse, Long Island
Sound Coastal Observatory (LISCO), Martha's Vineyard Coastal Observatory (MVCO) , Thornton CPower, USC Seaprism, Venise, WaveCIS Site CSI-6, Zeebrugge-MOW1) and two lake sites (Lake Erie,
Palgrunden) (Figure 1). AERONET-OC, managed by NASA’s Goddard Space Flight Center (GSFC)
[26], is a sub-network of the AERONET federated instrument [27,28]. Although OLI and AERONETOC have somewhat different spectral bands, a set of comparable bands centred at 441 nm, 491 nm,
551 nm, and 667 nm can be used for cross-comparison purposes. Thus, AERONET-OC data were
collected in four spectral bands centred at 441, 491, 551, and 667 nm, for comparison with OLI’s four
visible bands, centred at 443, 483, 561 and 655 nm. Note that there is a slight difference, i.e., ±1 to ±3
nm, in all four bands among measurements at different sites. As Rrs is not directly available from the
AERONET-OC sites, the normalized water-leaving radiances (nLw, W m-2 sr-1) were divided by the
top-of-the atmosphere (TOA) solar irradiance (F0) [29] to obtain Rrs.
Figure 1. Map showing the 14 validation sites from the ocean colour (OC) component of the Aerosol
Robotic Network (AERONET-OC) station. 1: Galata, 2: Gloria, 3: GOT Seaprism, 4: Gustav Dalen
Tower, 5: Helsinki, 6: Lake Erie, 7: Long Island Sound Coastal Observatory (LISCO), 8: Martha's
Vineyard Coastal Observatory (MVCO), 9: Palgrunden, 10: Thornton C-Power; 11: USC Seaprism, 12:
Venise, 13: WaveCIS Site CSI-6, 14: Zeebrugge-MOW1.
2.3. Match-Up Exercise
To obtain the in-situ Rrs data needed to test AC procedures for OLI, we performed a match-up
exercise between the AERONET-OC measurements and OLI data as follows: (i) Using the OLI
metadata database file provided by the United States Geological Survey (USGS), python code was
created to automatically retrieve all Landsat 8 OLI scenes and the contemporaneous AERONET-OC
data (from the AERONET-OC website) that were within a ± 30-minute time window of Landsat 8
overpass times, for the April 2013 to May 2017 timeframe (note that a strict time window of ± 30
minutes, which reduces the number of match-up pairs, was used to limit the error introduced by
water movement between satellite and AERONET-OC observations, and ensure the quality of matchups). This yielded a text file containing a total of 122 match-ups with coincident satellite and in-situ
data for 14 AERONET-OC sites (Figure 2), as well as information on aerosol optical thickness, solar
zenith angles (SZA), and wind speed for each match-up. All corresponding OLI scenes were
subsequently bulk-downloaded using Landsat-util, a tool to automatically find and download
multiple Landsat 8 scenes. Some of the 122 OLI scenes visibly contained a non-negligible amount of
specular reflection off the sea surface (sunglint) in the area of the AERONOET-OC site. As not all AC
algorithms have the capacity for sunglint correction, to obtain realistic and comparable Rrs across all
Figure 1. Map showing the 14 validation sites from the ocean colour (OC) component of the Aerosol
Robotic Network (AERONET-OC) station. 1: Galata, 2: Gloria, 3: GOT Seaprism, 4: Gustav Dalen
Tower, 5: Helsinki, 6: Lake Erie, 7: Long Island Sound Coastal Observatory (LISCO), 8: Martha’s
Vineyard Coastal Observatory (MVCO), 9: Palgrunden, 10: Thornton C-Power; 11: USC Seaprism,
12: Venise, 13: WaveCIS Site CSI-6, 14: Zeebrugge-MOW1.
2.3. Match-Up Exercise
To obtain the in-situ Rrs data needed to test AC procedures for OLI, we performed a match-up
exercise between the AERONET-OC measurements and OLI data as follows: (i) Using the OLI metadata
Remote Sens. 2019, 11, 469 5 of 20
database file provided by the United States Geological Survey (USGS), python code was created to
automatically retrieve all Landsat 8 OLI scenes and the contemporaneous AERONET-OC data (from
the AERONET-OC website) that were within a ±30-min time window of Landsat 8 overpass times,
for the April 2013 to May 2017 timeframe (note that a strict time window of ±30 min, which reduces
the number of match-up pairs, was used to limit the error introduced by water movement between
satellite and AERONET-OC observations, and ensure the quality of match-ups). This yielded a text file
containing a total of 122 match-ups with coincident satellite and in-situ data for 14 AERONET-OC sites
(Figure 2), as well as information on aerosol optical thickness, solar zenith angles (SZA), and wind
speed for each match-up. All corresponding OLI scenes were subsequently bulk-downloaded using
Landsat-util, a tool to automatically find and download multiple Landsat 8 scenes. Some of the 122 OLI
scenes visibly contained a non-negligible amount of specular reflection off the sea surface (sunglint)
in the area of the AERONOET-OC site. As not all AC algorithms have the capacity for sunglint
correction, to obtain realistic and comparable Rrs across all AC methods, scenes with visible specular
reflection were excluded, leaving 69 of the original 122 scenes; (ii) Following the approach of Bailey and
Werdell (2006) [30], a regional subset of Landsat 8 data was generated for each scene by (1) extracting
a 7 × 7 pixel window centred on the location of the AERONET-OC site, and (2) removing the centre
3 × 3 pixels from that window to limit the effect of noise from the site’s superstructures and shadows
cast by it; (iii) For SeaDAS, remaining low-quality pixels were then removed by employing the internal
SeaDAS exclusion flags, which include flags for land, clouds, cloud-shadow, ice, stray light, low nLw
(555), high viewing zenith angle (>60◦
), high sunglint, and high TOA radiance. Scenes without any
unflagged pixels were eliminated from the match-up exercise, leaving 54 scenes for comparison with
AERONET-OC data. Similarly, internal ACOLITE exclusion flags were used to remove low-quality
pixels for ACOLITE, which left 56 scenes for AERONET-OC comparison. The 54 scenes remaining
after applying the SeaDAS exclusion flags also passed ACOLITE’s exclusion flags, and were therefore
used for further analysis. To ensure an unbiased inter-comparison, we included pixels with negative
(Table A3) and zero Rrs retrievals from all methods; (iv) We then obtained the per-band median Rrs
values of the unflagged pixels from each Landsat regional subset for final comparison with in-situ
data, and used the median AERONET-OC measurements collected within the ±30-min window of the
Landsat 8 overpass to represent in-situ match-ups.
Remote Sens. 2018, 10, x FOR PEER REVIEW 5 of 20
AC methods, scenes with visible specular reflection were excluded, leaving 69 of the original 122
scenes; (ii) Following the approach of Bailey and Werdell (2006) [30], a regional subset of Landsat 8
data was generated for each scene by (1) extracting a 7 × 7 pixel window centred on the location of
the AERONET-OC site, and (2) removing the centre 3 × 3 pixels from that window to limit the effect
of noise from the site’s superstructures and shadows cast by it; (iii) For SeaDAS, remaining lowquality pixels were then removed by employing the internal SeaDAS exclusion flags, which include
flags for land, clouds, cloud-shadow, ice, stray light, low nLw (555), high viewing zenith angle (> 60o),
high sunglint, and high TOA radiance. Scenes without any unflagged pixels were eliminated from
the match-up exercise, leaving 54 scenes for comparison with AERONET-OC data. Similarly, internal
ACOLITE exclusion flags were used to remove low-quality pixels for ACOLITE, which left 56 scenes
for AERONET-OC comparison. The 54 scenes remaining after applying the SeaDAS exclusion flags
also passed ACOLITE’s exclusion flags, and were therefore used for further analysis. To ensure an
unbiased inter-comparison, we included pixels with negative (Table A3) and zero Rrs retrievals from
all methods; (iv) We then obtained the per-band median Rrs values of the unflagged pixels from each
Landsat regional subset for final comparison with in-situ data, and used the median AERONET-OC
measurements collected within the ± 30-minute window of the Landsat 8 overpass to represent insitu match-ups.
Figure 2. Number of match-ups between Landsat 8 OLI scenes and AERONET-OC site measurements
within ± 30-minute window of the Landsat 8 overpass (GAL: Galata, GLO: Gloria, GOT: Got Seaprism,
GUS: Gustav Dalen Tower, HEL: Helsinki, ERIE: Lake Erie, LIS: LISCO, MVC: MVCO, PAL:
Palgrunden, THO: Thornton C-Power, USC: USC Seaprism, VEN: Venise, WAV: WaveCIS Site CSI-6,
ZEB: Zeebrugge-MOW1). Dark blue represents the total number of initial match-ups within a ± 30-
minute time window of the Landsat 8 overpass times for each site. Light blue represents the total
number of final match-ups used for analysis after excluding scenes with sunglint and performing the
match-up exercise.
2.4. Data Processing
2.4.1. Description of Atmospheric Correction Algorithms
Atmospheric correction of the OLI data was carried out using four algorithms: ARCSI, ACOLITE
(version 20170113.0), SeaDAS (version 7.4), and LaSRC. Note that LaSRC is a product that has already
been processed for surface reflectance by the United States Geological Survey (USGS). Both ACOLITE
and SeaDAS have been specifically designed for AC over water surfaces, whereas ARCSI and LaSRC
have not; we therefore refer to the latter two as land-based methods. The output of the water-based
methods is Rrs, which is directly comparable to the in-situ data from AERONET (after conversion: Rrs
= nLw/ F0), while the output of the land-based methods is in units of above-surface diffuse reflectance
R(0+), which we converted to Rrs using:
ܴ௥௦ = ܴ(0+)/π (1)
Figure 2. Number of match-ups between Landsat 8 OLI scenes and AERONET-OC site measurements
within ±30-min window of the Landsat 8 overpass (GAL: Galata, GLO: Gloria, GOT: Got Seaprism,
GUS: Gustav Dalen Tower, HEL: Helsinki, ERIE: Lake Erie, LIS: LISCO, MVC: MVCO, PAL: Palgrunden,
THO: Thornton C-Power, USC: USC Seaprism, VEN: Venise, WAV: WaveCIS Site CSI-6, ZEB:
Zeebrugge-MOW1). Dark blue represents the total number of initial match-ups within a ±30-min time
window of the Landsat 8 overpass times for each site. Light blue represents the total number of final
match-ups used for analysis after excluding scenes with sunglint and performing the match-up exercise.
Remote Sens. 2019, 11, 469 6 of 20
2.4. Data Processing
2.4.1. Description of Atmospheric Correction Algorithms
Atmospheric correction of the OLI data was carried out using four algorithms: ARCSI, ACOLITE
(version 20170113.0), SeaDAS (version 7.4), and LaSRC. Note that LaSRC is a product that has already
been processed for surface reflectance by the United States Geological Survey (USGS). Both ACOLITE
and SeaDAS have been specifically designed for AC over water surfaces, whereas ARCSI and LaSRC
have not; we therefore refer to the latter two as land-based methods. The output of the water-based
methods is Rrs, which is directly comparable to the in-situ data from AERONET (after conversion:
Rrs = nLw/ F0), while the output of the land-based methods is in units of above-surface diffuse
reflectance R(0+
), which we converted to Rrs using:
Rrs = R

0
+


/π (1)
ARCSI is an open-source software program developed at Aberystwyth University [21]. It is a
relatively new AC algorithm with functionalities to process multispectral images from both commercial
and publicly available sensors and also to obtain processed data for direct use in remote sensing
analyses [31]. It is a command line tool where Py6S [32] can be implemented to correct multispectral
images to above-surface diffuse reflectance using the 6S model [33], which simulates ground and
atmospheric radiation under a variety of conditions. Within the 6S method, input parameters such
as the Aerosol Optical Thickness (AOT), vertical column water vapour, and ozone concentration are
automatically used by the 6S method to characterize the state of the atmosphere.
ACOLITE is a binary distribution of Landsat 8 OLI and Sentinel-2 MSI processing software
developed by the Royal Belgian Institute of Natural Sciences [22,34]. It is an image-based AC algorithm
that estimates Lw by correcting for molecular and aerosol scattering in the atmosphere using the Gordon
and Wang (1994a) approach [5]. Molecular reflectance correction, based on viewing and illumination
geometries, is performed with a 6SV-based look-up table [33]. Unlike SeaDAS, which uses 80 aerosol
models for aerosol estimation [5,35], aerosol reflectance is estimated by determining aerosol types
from the ratio of reflectances in two SWIR bands over water pixels where reflectance can be assumed
zero, an approach similar to Ruddick et al. (2000) [36]. Based on this assumption, it also retrieves
water-leaving reflectances in both the visible and NIR bands together with other parameters of interest
in marine and inland waters. ACOLITE is primarily designed for processing Landsat 8 OLI data for
aquatic remote sensing applications, but has recently been modified and updated to include processing
of Sentinel-2 MSI data [37].
LaSRC is a Level-2 data set produced and released as a provisional product by the USGS since
January 2015, primarily to support terrestrial remote sensing applications. Unlike the precursor
algorithm, i.e., Landsat Ecosystem Disturbance Adaptive Processing System (LEDAPS), used for
previous Landsat satellites such as Landsat 4-5 Thematic Mapper (TM) and Landsat 7 Enhanced
Thematic Mapper Plus (ETM+), which used the 6S model, LaSRC is generated using a dedicated
Landsat Surface Reflectance code [24]. Data are available as standalone climate data records (CDRs)
that represent specific geophysical and biophysical properties of the land surface [23]. AC is mainly
based on the MODIS collection 6-AC algorithm, which uses a radiative transfer model for the inversion
of atmospheric parameters such as aerosol and water vapour [24]. It should be noted that surface
reflectance is provided in seven spectral bands (the first seven OLI bands) only for scenes with a solar
zenith angle less than 76◦
, and that the 443 nm and 482 nm bands are not suitable for analysis as they
are ‘consumed’ for aerosol inversion tests within the LaSRC. Although still provisional and under
continuous improvement, LaSRC has been validated and assessed for land applications [38–41], and
has a dedicated aerosol retrieval algorithm for pixels over water [23].
SeaDAS, which uses an NIR and a SWIR band for aerosol estimation for processing OLI imagery,
contains an AC scheme originally designed for open ocean water based on the assumption of negligibl
Remote Sens. 2019, 11, 469 7 of 20
Lw in the NIR bands. This approach, which is NASA’s operational AC algorithm [5], includes an
l2gen (Level 2 generator) to retrieve Rrs and other optical and geophysical water and atmospheric
properties. Following some improvements to estimation of aerosol contributions (e.g., [12,42,43]) the
l2gen processor in SeaDAS can now be used for deriving Rrs in moderately turbid coastal waters [14].
2.4.2. Atmospheric Correction Procedure and Validation
To derive Rrs, all four AC algorithms were parameterized using their default processing options.
In addition, the following processing was implemented: (1) In SeaDAS, out-of-band correction options
were set to zero (outband_opt = 0), i.e. Rrs was reported at full bandpass, without correction to the
nominal band centre, and no sunglint correction was implemented (glint_opt = 0) since there was
no such correction option available for other algorithms; (2) In ARCSI, the ‘clear water’ option was
used for the reflectance of a ground target as processing requires an option from ‘green vegetation’,
‘clear water’, ‘sand’, or ‘lake water’. Also, AOT and other atmospheric parameters were automatically
identified and estimated by ARCSI during batch processing. To derive AOT for each scene, realistic
minimum and maximum values of 0.001 and 0.9, respectively, were manually specified; (3) Finally, to
allow for consistency among all methods, we assumed a perfect sensor calibration by applying unity
gains for vicarious calibration across all bands for all processors. For SeaDAS, aerosol correction was
implemented following the Gordon and Wang (1994a) approach [5], with the NIR/SWIR correction
option (865–1609 nm band combination) as suggested in Pahlevan et al. (2017a) [15] and Mobley et al.
(2016) [7]. ACOLITE aerosol correction was implemented using the default SWIR option (1609 and
2201 nm band combination) which computes Rayleigh-corrected reflectance from the SWIR bands for
moderate and turbid waters. Each AC algorithm was applied to the final 54 Landsat 8 OLI scenes,
representing a wide variety of coastal and atmospheric conditions (Table A1). When comparing their
performance, we considered AERONET-OC data as the reference with negligible uncertainties. Note
that uncertainties in the AERONET-OC in-situ measurements are ~5% in the blue to green bands and
~8% in the red band [26]. As noted in Pahlevan et al. (2017b) [44] and Mélin and Sclep (2015) [45],
compensating for discrepancies arising from the differences in nominal band centre wavelengths
is crucial for obtaining a robust match-up analysis across all spectral bands (in particular for OLI’s
relatively broad spectral bands). To this end, we carried out a spectral band adjustment using the deep
neural network approach as described in Pahlevan et al. (2017b) [44] and compared Rrs derived with
and without band adjustments.
The algorithm performance was compared using six metrics including:
Root-Mean-Square Error (RMSE) = s
1
n
n
∑
i=1
(xmea − x
est)
2
(2)
mean bias =
n
∑
i=1

x
mea − x
est

n
(3)
Spectral Angle = cos−1


n
∑
i=1
x
mea
.x
est
s
n
∑
i=1
xmea2
s
n
∑
i=1
x
est2


(4)
as well as the coefficient of determination (R2
), slope, and intercept of the line fitted using least-squares
regression between in-situ and satellite Rrs estimates. The xmea and xest are AERONET-OC and
satellite-derived Rrs data, respectively. The Spectral Angle (SA), which is insensitive to spectral
amplitude, is used to quantify the similarity between satellite and in-situ Rrs spectra. Values close to 0
indicate high similarity
Remote Sens. 2019, 11, 469 8 of 20
3. Results and Discussion
3.1. Validation of AC Algorithms
Scatter plots showing the estimated (OLI) and observed (AERONET-OC) Rrs values for each
match-up are presented in Figure 3, and summary statistics are tabulated in Table 2. There are clear
differences between the water-based and land-based AC algorithms, with SeaDAS and ACOLITE
outperforming ARCSI and LaSRC in all metrics for all bands, with only one exception (slope for
Rrs(482)). Between the two water-based methods, SeaDAS outperforms ACOLITE in every metric
for all bands, with the exception of slopes for Rrs(482), Rrs(561), and Rrs(655). SeaDAS had RMSEs
close to zero between 0.0013 and 0.0005 1/sr across all four wavelengths) demonstrating a high
degree of similarity between in-situ and OLI-estimated Rrs with OLI data processed through SeaDAS
(Table 2). A comparison of RMSE results of the per-band difference with and without band adjustments
(Figure A1) shows that SeaDAS was the AC method most sensitive to spectral band differences, with
the largest improvement of band adjustment occurring in the 655 nm channel. Spectral Angle values
obtained for all algorithms showed that SeaDAS and ACOLITE have the highest similarity with in-situ
RrsRemote Sens.
spectra (ARCSI: 0.46, ACOLITE: 0.27, LaSRC: 0.53 and SeaDAS: 0.20).
2018, 10, x FOR PEER REVIEW 8 of 20
Figure 3. Scatterplots of the relationship between in-situ measurements (x-axis) and OLI estimates (yaxis) for each OLI band acquired over 14 AERONET-OC sites. Regression lines are shown in colours,
while the thick dotted black lines are 1:1 lines. (a) 443 nm; (b) 482 nm; (c) 561 nm; (d) 651 nm.
Table 2. Statistical results for the retrieved remote sensing reflectance (Rrs) obtained for all processors
with and without band adjustment (values in parenthesis represent results without band adjustment).
Best metrics are highlighted in bold letters. After-band-adjustment linear fit, which was employed to
reveal the relationship between in-situ and modelled Rrs, improves with increasing wavelength for
both Atmospheric Correction (ACOLITE and SeaDAS), with R2 values of 0.70/0.84, 0.85/0.92, 0.92/0.95,
and 0.93/0.97 for bands 1 through 4 for ACOLITE/SeaDAS, respectively. A similar trend is seen for
ARCSI and LaSRC, for the first three bands.
R2 Slope RMSE (1/sr) Intercept p-values
Rrs 443
ARCSI 0.43 (0.41) 0.91 (0.89) 0.0085 (0.0085) 0.0080 (0.0084) 8.92e-08
ACOLITE 0.70 (0.68) 0.97 (0.97) 0.0039 (0.0039) 0.0036 (0.0037) 4.16e-15
LaSRC 0.05 (0.05) 0.23 (0.25) 0.0042 (0.0042) 0.0050 (0.0050) 0.11
SeaDAS 0.84 (0.84) 1.08 (1.08) 0.0013 (0.0013) -0.0006 (-0.0006) 2.36e-22
Rrs 482
ARCSI 0.68 (0.63) 1.01 (0.92) 0.0065 (0.0063) 0.0060 (0.0061) 2.00e-13
ACOLITE 0.85 (0.79) 1.03 (0.94) 0.0032 (0.0031) 0.0027 (0.0029) 1.99e-14
LaSRC 0.44 (0.43) 0.60 (0.56) 0.0035 (0.0035) 0.0041 (0.0041) 3.77e-08
SeaDAS 0.92 (0.87) 1.09 (1.00) 0.0012 (0.0015) -0.0002 (0.00009) 5.44e-30
Rrs 561
ARCSI 0.77 (0.77) 0.95 (0.97) 0.0051 (0.0048) 0.0046 (0.0042) 5.27e-18
ACOLITE 0.92 (0.87) 1.00 (0.98) 0.0016 (0.0019) 0.0005 (0.0002) 1.38e-29
LaSRC 0.80 (0.78) 0.83 (0.83) 0.0030 (0.0029) 0.0027 (0.0025) 9.48e-20
SeaDAS 0.95 (0.92) 1.03 (1.21) 0.0012 (0.0011) 0.00005 (-0.0003) 1.13e-34
Rrs 665
ARCSI 0.64 (0.63) 0.91 (1.06) 0.0033 (0.0034) 0.0028 (0.0026) 4.49e-13
ACOLITE 0.93 (0.89) 0.98 (1.13) 0.0010 (0.0013) 0.0006 (0.0005) 1.91e-31
LaSRC 0.52 (0.50) 0.65 (0.75) 0.0022 (0.0021) 0.0011 (0.0010) 8.39e-10
SeaDAS 0.97 (0.92) 1.01 (1.21) 0.0005 (0.0011) -0.0001 (-0.0003) 4.00e-40
Figure 3. Scatterplots of the relationship between in-situ measurements (x-axis) and OLI estimates
(y-axis) for each OLI band acquired over 14 AERONET-OC sites. Regression lines are shown in colours,
while the thick dotted black lines are 1:1 lines. (a) 443 nm; (b) 482 nm; (c) 561 nm; (d) 651 nm.
The overall performance of SeaDAS reveals that the NIR-SWIR aerosol correction option can yield
satisfactory results in low-to-moderately turbid waters. A possible reason for this is that the aerosol
correction scheme, constructed following Ahmad et al. (2010) [42], was based on aerosol data obtained
mainly from AERONET-OC sites [46]. Comparison of R2 values among all methods shows that the
lowest and most diverse values are in the 443 nm wavelength, with values between 0.05 and 0.84. For
LaSRC in particular, the regression line for the comparison in the 443 nm wavelength deviates very
much from the 1:1 line, yielding a poor R2 and slope (Figure 3). The poor correlation and low RMSE
(R2
: 0.05, slope: 0.23) are mostly a result of the large discrepancies between the observed and estimated
Rrs for the Zeebrugge-MOW1 site, where mean Rrs was underestimated by ~70%. This is the largest
underestimation by any method, across all sites. For this site, which is one of the most turbid sites in
Remote Sens. 2019, 11, 469 9 of 20
the AERONET-OC network, mean observed in-situ Rrs in the red band is 0.0155 sr−1
, making it the
only site with Rrs(655) one order of magnitude greater than the mean value of ~0.001 sr−1 observed
for all 14 AERONET-OC sites. This level of turbidity is common for this site, which is located only
~3.65 km from the coastline and receives sediment-rich water inputs from nearby rivers, as also noted
by Vanhellemont and Ruddick (2015) [47] and was clearly visible in additional scenes excluded during
the match-up exercise. Note that the TOA radiance data were used ‘as is’ without optimizing the
vicarious calibration gains (as computed by Pahlevan et al. (2017a) [15]), which might further improve
Rrs retrievals. Similarly, none of the AC methods with different configuration capabilities that might
improve performance were optimized as there was no optimal setting that would work for all cases
considered in this paper.
Table 2. Statistical results for the retrieved remote sensing reflectance (Rrs) obtained for all processors
with and without band adjustment (values in parenthesis represent results without band adjustment).
Best metrics are highlighted in bold letters. After-band-adjustment linear fit, which was employed
to reveal the relationship between in-situ and modelled Rrs, improves with increasing wavelength
for both Atmospheric Correction (ACOLITE and SeaDAS), with R2 values of 0.70/0.84, 0.85/0.92,
0.92/0.95, and 0.93/0.97 for bands 1 through 4 for ACOLITE/SeaDAS, respectively. A similar trend is
seen for ARCSI and LaSRC, for the first three bands.
R
2 Slope RMSE (1/sr) Intercept p-Values
Rrs 443
ARCSI 0.43 (0.41) 0.91 (0.89) 0.0085 (0.0085) 0.0080 (0.0084) 8.92e-08
ACOLITE 0.70 (0.68) 0.97 (0.97) 0.0039 (0.0039) 0.0036 (0.0037) 4.16e-15
LaSRC 0.05 (0.05) 0.23 (0.25) 0.0042 (0.0042) 0.0050 (0.0050) 0.11
SeaDAS 0.84 (0.84) 1.08 (1.08) 0.0013 (0.0013) −0.0006 (−0.0006) 2.36e-22
Rrs 482
ARCSI 0.68 (0.63) 1.01 (0.92) 0.0065 (0.0063) 0.0060 (0.0061) 2.00e-13
ACOLITE 0.85 (0.79) 1.03 (0.94) 0.0032 (0.0031) 0.0027 (0.0029) 1.99e-14
LaSRC 0.44 (0.43) 0.60 (0.56) 0.0035 (0.0035) 0.0041 (0.0041) 3.77e-08
SeaDAS 0.92 (0.87) 1.09 (1.00) 0.0012 (0.0015) −0.0002 (0.00009) 5.44e-30
Rrs 561
ARCSI 0.77 (0.77) 0.95 (0.97) 0.0051 (0.0048) 0.0046 (0.0042) 5.27e-18
ACOLITE 0.92 (0.87) 1.00 (0.98) 0.0016 (0.0019) 0.0005 (0.0002) 1.38e-29
LaSRC 0.80 (0.78) 0.83 (0.83) 0.0030 (0.0029) 0.0027 (0.0025) 9.48e-20
SeaDAS 0.95 (0.92) 1.03 (1.21) 0.0012 (0.0011) 0.00005 (−0.0003) 1.13e-34
Rrs 665
ARCSI 0.64 (0.63) 0.91 (1.06) 0.0033 (0.0034) 0.0028 (0.0026) 4.49e-13
ACOLITE 0.93 (0.89) 0.98 (1.13) 0.0010 (0.0013) 0.0006 (0.0005) 1.91e-31
LaSRC 0.52 (0.50) 0.65 (0.75) 0.0022 (0.0021) 0.0011 (0.0010) 8.39e-10
SeaDAS 0.97 (0.92) 1.01 (1.21) 0.0005 (0.0011) −0.0001 (−0.0003) 4.00e-40
3.2. Inter-Comparison of Reflectance Spectra at Each Site
Comparison of mean estimated and observed Rrs at each AERONET-OC site (Figure A2) shows
that all algorithms except SeaDAS generally overestimate Rrs across all wavelengths, with the largest
and smallest overestimation occurring in the 443 and 665 nm wavelengths, respectively. This is further
supported by the RMSE and bias results (Figure 4). The largest errors (RMSE) and overestimations
(bias) are observed in the 443 nm wavelength, probably due to the strong atmospheric scattering in
this band. ARCSI has the largest overall positive bias in this wavelength, and indeed the highest
overestimation at each site. However, its Rrs results across all wavelengths at the Zeebrugge-MOW1
site compare well with Rrs estimates from SeaDAS and ACOLITE. This suggests that ARCSI has a low
sensitivity to the high concentrations of suspended sediments that dominate this site, as reported by
De Maerschalck and Vanlede (2013) [48]. This may also serve as an indication that ARCSI can better
deal with turbid conditions than can LaSRC, which underestimated Rrs by ~45% at this site. The failure
of LaSRC for this band is likely due to the fact that it is part of the bands used for the aerosol inversion
scheme [23].
Remote Sens. 2019, 11, 469 10 of 20
The best performance from LaSRC across all bands is at Lake Erie (with two match-ups) where all
other AC algorithms except ARCSI also have the best match with in-situ Rrs. LaSRC outperforms other
algorithms for the first three bands. Percentage difference values are 4.5%, −3.2%, and −0.3% for the
first three bands, respectively. For ACOLITE and SeaDAS, the corresponding values are 28.8/−20.7%,
13.6/−12.8%, and −0.4/−6.4%. Indeed, at this site LaSRC has the best Rrs estimate of all methods
in the 561 nm channel, while ACOLITE has the best Rrs estimate in the 655 nm channel, with a
percentage difference of −1.3%, whereas SeaDAS and LaSRC are −18.2% and −30%, respectively.
Similar to the estimated Rrs by LaSRC in Lake Erie in the 561 nm channel, LASRC-derived Rrs(561)
also agrees well with that of in-situ at Zeebrugge-MOW1 site; the percentage difference here is −0.25%.
For SeaDAS and ACOLITE, these values are 6.7% and 4.1%, respectively. LaSRC also outperforms
other AC algorithms in the 443, 482, and 655 nm channels at the GOT-Seaprism site, with only one
match-up. Percentage differences between estimated and in-situ Rrs are 8.8%, 11.1%, and −40.8%,
respectively. For ACOLITE and SeaDAS, the corresponding values are 87.7/−79.3%, 60.9/−46.3%,
and 96.4/−127.6%. This is the only site where SeaDAS uncharacteristically underestimates Rrs across
all four bands. While any conclusion is tentative as GOT Seaprism (2014-026) only has one match-up,
the poor performance of SeaDAS here is as a result of algorithm failure (very low Rrs in 443, 482, and
561 nm wavelengths, and negative Rrs in 655 nm wavelength), which can be attributed to conditions
such as residual effects from cloud shadow or overcorrection for aerosol contribution in one or more
visible band. Overcorrection typically occurs when water-leaving radiance is non-negligible in the
band used to estimate the aerosol contribution [49]. Other instances of failure (as defined above) from
one or more algorithms are: ACOLITE (Gloria 2014-358: band 4, USC Seaprism 2016-222: bands 3
and 4), SeaDAS (GOT Seaprism 2014-026: band 4, Helsinki 2013-235: band 1, Palgrunden 2013-156:
band 1, USC Seaprism 2016-222: band 4, USC Seaprism 2016-334: band 4, Venise 2015-221: band 4) and
LaSRC (WaveCIS: 2013-221: bands 1 and 4). The low or negative Rrs retrievals from these algorithms
indicate a limitation of these algorithms in dealing with the atmospheric and water quality conditions
present at those match-ups.
One possible reason for the generally poor performance of ARCSI can be the aerosol contribution
removal which relies on estimates from (1) dense dark vegetated surfaces, based on the assumption
that reflectance of vegetated pixels is sufficiently dark, and a linear relationship between reflectance
in the SWIR and blue bands or (2) dark pixels in the blue band, based on the assumption of an
invariant aerosol concentration over the entire scene. However, these assumptions can easily be
violated as finding a vegetated pixel that satisfies this condition may be difficult in scenes acquired
over coastal waters and AOT variations may be sufficiently large such that adjoining pixels may have
significantly different AOT. For the blue bands in particular, per-scene AOT estimates may lead to
erroneous retrievals. For LaSRC, the generally poor performance may be due to the use of land-based
pixels for aerosol estimation. In addition, for LaSRC, retrieving accurate Rrs estimates over water
requires the presence of a considerably large land area adjoining the water pixels. The majority of the
AERONET-OC sites used in this study only have relatively small nearby land surfaces. This may help
explain the few instances of good performance near land masses (e.g., for the Lake Erie and Zeebrugge
MOW-1 sites).
Remote Sens. 2018, 10, x FOR PEER REVIEW 10 of 20
such as residual effects from cloud shadow or overcorrection for aerosol contribution in one or more
visible band. Overcorrection typically occurs when water-leaving radiance is non-negligible in the
band used to estimate the aerosol contribution [49]. Other instances of failure (as defined above) from
one or more algorithms are: ACOLITE (Gloria 2014-358: band 4, USC Seaprism 2016-222: bands 3 and
4), SeaDAS (GOT Seaprism 2014-026: band 4, Helsinki 2013-235: band 1, Palgrunden 2013-156: band
1, USC Seaprism 2016-222: band 4, USC Seaprism 2016-334: band 4, Venise 2015-221: band 4) and
LaSRC (WaveCIS: 2013-221: bands 1 and 4). The low or negative Rrs retrievals from these algorithms
indicate a limitation of these algorithms in dealing with the atmospheric and water quality conditions
present at those match-ups.
One possible reason for the generally poor performance of ARCSI can be the aerosol contribution
removal which relies on estimates from (1) dense dark vegetated surfaces, based on the assumption
that reflectance of vegetated pixels is sufficiently dark, and a linear relationship between reflectance
in the SWIR and blue bands or (2) dark pixels in the blue band, based on the assumption of an
invariant aerosol concentration over the entire scene. However, these assumptions can easily be
violated as finding a vegetated pixel that satisfies this condition may be difficult in scenes acquired
over coastal waters and AOT variations may be sufficiently large such that adjoining pixels may have
significantly different AOT. For the blue bands in particular, per-scene AOT estimates may lead to
erroneous retrievals. For LaSRC, the generally poor performance may be due to the use of land-based
pixels for aerosol estimation. In addition, for LaSRC, retrieving accurate Rrs estimates over water
requires the presence of a considerably large land area adjoining the water pixels. The majority of the
AERONET-OC sites used in this study only have relatively small nearby land surfaces. This may help
explain the few instances of good performance near land masses (e.g. for the Lake Erie and Zeebrugge
MOW-1 sites).

Figure 4. Overall band-by-band RMSE and mean bias results for all algorithms.
3.3. Influence of Environmental Factors for SeaDAS and ACOLITE
To understand the impact environmental factors may have on Rrs retrieval errors from the waterbased AC methods, we investigated the influence of three variables: AOT(869), SZA, and hourly wind
speed. These three variables are known to influence Rrs retrievals [50,51], e.g. AOT(870) and SZA have
been found to reduce the quality of water-leaving radiance derived from SeaWiFS and MODIS
sensors [52]. Figure 5a–c illustrates the error (xest − xmea) for each match-up point as a function of each
environmental parameter, for each AC method. Negative values imply that an algorithm
underestimated the observed Rrs value, and vice versa. We used tests of the statistical significance
(two-tailed, α = 0.05, critical value = 0.2262) of the individual Pearson correlation coefficients to guide
this analysis (Pearson correlation coefficients were used to show the strength of the relationship as
they have been widely used in similar studies). While ACOLITE consistently overestimated Rrs in the
443 and 482 nm bands, as also noted in [53], errors for both SeaDAS and ACOLITE were not
significantly influenced by AOT (Figure 5a, no statistically significant correlations). However, SZA
was significantly and positively correlated with Rrs retrieval errors from SeaDAS for all four bands
(i.e., r = 0.495486743, 0.483529464, 0.253699366, and 0.427793365, respectively), and from ACOLITE
for the 443 and 482 nm bands (i.e., r = 0.239717715 and 0.228792001, respectively) (Figure 5b). A similar Figure 4. Overall band-by-band RMSE and mean bias results for all algorithms.
Remote Sens. 2019, 11, 469 11 of 20
3.3. Influence of Environmental Factors for SeaDAS and ACOLITE
To understand the impact environmental factors may have on Rrs retrieval errors from the
water-based AC methods, we investigated the influence of three variables: AOT(869), SZA, and hourly
wind speed. These three variables are known to influence Rrs retrievals [50,51], e.g., AOT(870) and
SZA have been found to reduce the quality of water-leaving radiance derived from SeaWiFS and
MODIS sensors [52]. Figure 5a–c illustrates the error (xest − x
mea) for each match-up point as a function
of each environmental parameter, for each AC method. Negative values imply that an algorithm
underestimated the observed Rrs value, and vice versa. We used tests of the statistical significance
(two-tailed, α = 0.05, critical value = 0.2262) of the individual Pearson correlation coefficients to guide
this analysis (Pearson correlation coefficients were used to show the strength of the relationship as they
have been widely used in similar studies). While ACOLITE consistently overestimated Rrs in the 443
and 482 nm bands, as also noted in [53], errors for both SeaDAS and ACOLITE were not significantly
influenced by AOT (Figure 5a, no statistically significant correlations). However, SZA was significantly
and positively correlated with Rrs retrieval errors from SeaDAS for all four bands (i.e., r = 0.495486743,
0.483529464, 0.253699366, and 0.427793365, respectively), and from ACOLITE for the 443 and 482 nm
bands (i.e., r = 0.239717715 and 0.228792001, respectively) (Figure 5b). A similar pattern was evident
for wind speed, which was significantly positively correlated with Rrs retrieval errors from SeaDAS for
all bands except band 3 (for which a positive correlation was present, but not statistically significant)
and from ACOLITE for the 443 and 482 nm bands (Figure 5c). We further examined the significance of
the relationship between each environmental variable and errors across all wavelengths and found
that SZA and AOT(869) were significant for SeaDAS in the first three and first two wavelengths,
respectively, while ACOLITE was only influenced by wind speed in the 443 nm channel. These
patterns, while generally causing only small errors in Rrs retrieval, may guide further developments of
both AC methods to make them more robust across the range of environmental conditions.
Remote Sens. 2018, 10, x FOR PEER REVIEW 11 of 20
pattern was evident for wind speed, which was significantly positively correlated with Rrs retrieval
errors from SeaDAS for all bands except band 3 (for which a positive correlation was present, but not
statistically significant) and from ACOLITE for the 443 and 482 nm bands (Figure 5c). We further
examined the significance of the relationship between each environmental variable and errors across
all wavelengths and found that SZA and AOT(869) were significant for SeaDAS in the first three and
first two wavelengths, respectively, while ACOLITE was only influenced by wind speed in the 443
nm channel. These patterns, while generally causing only small errors in Rrs retrieval, may guide
further developments of both AC methods to make them more robust across the range of
environmental conditions.
Figure 5. Scatterplots of the error (sr-1) showing the dependency of Rrs retrieval accuracy from both
ACOLITE and SeaDAS on (a) AOT(869), (b) SZA, and (c) wind speed. AOT(869) and wind speed Figure 5. Cont.
Remote Sens. 2019, 11, 469 12 of 20 Remote Sens. 2018, 10, x FOR PEER REVIEW 11 of 20
pattern was evident for wind speed, which was significantly positively correlated with Rrs retrieval
errors from SeaDAS for all bands except band 3 (for which a positive correlation was present, but not
statistically significant) and from ACOLITE for the 443 and 482 nm bands (Figure 5c). We further
examined the significance of the relationship between each environmental variable and errors across
all wavelengths and found that SZA and AOT(869) were significant for SeaDAS in the first three and
first two wavelengths, respectively, while ACOLITE was only influenced by wind speed in the 443
nm channel. These patterns, while generally causing only small errors in Rrs retrieval, may guide
further developments of both AC methods to make them more robust across the range of
environmental conditions.
Figure 5. Scatterplots of the error (sr-1) showing the dependency of Rrs retrieval accuracy from both
ACOLITE and SeaDAS on (a) AOT(869), (b) SZA, and (c) wind speed. AOT(869) and wind speed
Figure 5. Scatterplots of the error (sr−1
) showing the dependency of Rrs retrieval accuracy from both
ACOLITE and SeaDAS on (a) AOT(869), (b) SZA, and (c) wind speed. AOT(869) and wind speed were
derived from coincident measurements at each AERONET-OC site used in this study, while SZA was
obtained by subtracting the sun elevation angle provided in the Landsat 8 metadata from 90. Each
circle represents a match-up data point, for a total of 54 data points across the 14 AERONET-OC sites.
The 54 match-ups and their corresponding environmental parameter values are tabulated in Table A2.
4. Conclusions
This paper provides an evaluation of four atmospheric correction algorithms (ACOLITE, ARCSI,
LaSRC, and SeaDAS) for estimating Rrs. Fifty-four match-ups were used to test the performance of these
algorithms over mostly coastal sites that form part of the AERONET-OC network. After accounting for
spectral band differences in AERONET-OC and OLI measurements/products, the Rrs products from all
algorithms were compared to AERONET in-situ Rrs data. The generic AC methods (ARCSI and LaSRC)
were less accurate for deriving Rrs in coastal environments than water-based methods (ACOLITE
and SeaDAS). These AC methods were particularly unreliable in the 443 and 482 nm channels, and
performed well at only a few sites located in nearshore and inland waters. SeaDAS produced the best
performance overall, while ACOLITE, though it performed better than the two generic AC methods,
was less accurate than SeaDAS for Rrs retrievals over (mostly) low-to-moderately coastal waters such
as those typical of the AERONET-OC sites. Analyses of differences in spectral bands between satellite
and in-situ measurements revealed that band adjustment minimized differences between sensors with
different spectral bands. A relationship seems to exist between Rrs retrieval accuracy for the two
water-based AC methods and two atmospheric variables: SZA and wind speed. Future studies should
examine these relationships further and consider related improvements to the AC methods. Neither of
the water-based AC methods can currently be used to process images from commercial sensors such as
WorldView-2/3 (which have improved spatial resolution) or previous Landsat missions (though this
capability is available in an in-house version of SeaDAS for future public release). Given the usefulness
of high spatial resolution data and the understanding that can be gained from time-series analysis
for aquatic studies, such improvements would be valuable. Our findings are primarily applicable
to nearshore coastal waters under low aerosol condition (i.e., AOT (869) ≤ 0.2). Further validation
is required over inland waters (e.g., recently established sites in Green Bay, Grizzly Bay and Lake
Okeechobee across the United States) and at stations with few match-up points (e.g., GOT Seaprism
and Lake Erie with one and two match-ups, respectively) to better understand the performance of
each AC method for various science and application areas. In future studies, the authors intend to
evaluate the performance of these AC algorithms over inland waters such as those found over the
GloboLakes sites.
Author Contributions: Conceptualization: C.O.I., N.P., and A.K.; Formal analysis: C.O.I., N.P., and A.K.;
Investigation: C.O.I.; Methodology: C.O.I.; Project administration: N.P. and A.K.; Resources: C.O.I., N.P., and
A.K.; Software: C.O.I.; Supervision: N.P. and A.K.; Validation: C.O.I.; Visualization: C.O.I.; Writing—original
draft: C.O.I.; Writing—review and editing: C.O.I.
Remote Sens. 2019, 11, 469 13 of 20
Funding: Nima Pahlevan was funded under NASA ROSES #NNX16AI16G and the USGS Landsat Science Team
Award #140G0118C0011.
Acknowledgments: The authors wish to thank USGS for the distribution of Landsat 8 Level-1 data products.
We are grateful for the efforts of all the staff, site support people, and the team responsible for the processing
and archiving all the 14 AERONET-OC site-data used in this study. We are particularly thankful to all the
principle investigators: Giuseppe Zibordi, principal investigator of the Galata Platform, Gloria, Gustav Dalen
Tower, Helsinki Lighthouse, and Venise sites; Brent Holben, the principal investigator for the GOT Seaprism;
Tim Moore, Steve Ruberg, and Menghua Wang, the principal investigators of the Lake Erie ste; Sam Ahmed
and Alex Gilerson, the principal investigators of the LISCO site, Hui Feng and Heidi M. Sosik, the principal
investigators of the MVCO site; Susanne Kratzer, the principal investigator of the Palgrunden site, Dimitry Van
der Zande, the principal investigator of the Thornton C-Power and Zeebrugge MOW-1 sites, Burton Jones and
Curtiss Davis, the principal investigators of the USC Seaprism site, Brent Holben, the principal investigator of
the USC Seaprism-2 site, and Alan Weidemann, Bill Gibson, and Robert Arnone, the principal investigators of
the WaveCIS CSI-6 site. We thank Giuseppe Zibordi for answering questions about AERONET-OC data. We are
deeply thankful to Quinten Vanhellemont and Kevin Ruddick for the development and support of ACOLITE, Pete
Bunting and Dan Clewley for the development and support of ARCSI, the NASA Ocean Biology Processing Group
for the development and support of the SeaDAS software, and the USGS Landsat Science Teams for processing
the Landsat 8 to surface reflectance product (LaSRC).
Conflicts of Interest: The authors declare no conflicts of interest.
Appendix A
Remote Sens. 2018, 10, x FOR PEER REVIEW 13 of 20
the USC Seaprism-2 site, and Alan Weidemann, Bill Gibson, and Robert Arnone, the principal investigators of
the WaveCIS CSI-6 site. We thank Giuseppe Zibordi for answering questions about AERONET-OC data. We are
deeply thankful to Quinten Vanhellemont and Kevin Ruddick for the development and support of ACOLITE,
Pete Bunting and Dan Clewley for the development and support of ARCSI, the NASA Ocean Biology Processing
Group for the development and support of the SeaDAS software, and the USGS Landsat Science Teams for
processing the Landsat 8 to surface reflectance product (LaSRC).
Conflicts of Interest: The authors declare no conflicts of interest.
(a) (b)
(c) (d)
Figure A1. The root-mean-square errors showing the impacts of per-band spectral adjustment on
AERONET-OC match-ups. For all AC methods, there is no noticeable effect in the 443 nm channel.
Similarly, for the land-based AC methods, there are no observable differences in the 443 and 482 nm
channels. Band adjustment improves the results for bands 2, 3, and 4 for SeaDAS, decreasing RMSE
values by 16.6, 23.9, and 43.8% in the 482, 561, and 655 nm wavelengths, respectively, and also
improves results for bands 3 and 4 for ACOLITE by 15.6 and 24.2%, respectively. For SeaDAS, the
largest observable difference is in the 655 nm channel. This is by far the largest improvement from
band adjustment across all bands and AC methods. Overall, SeaDAS is the most sensitive method to
spectral band differences, with the largest difference (improvement) in the 655 nm channel. (a) ARCSI;
(b) LaSRC; (c) ACOLITE; (d) SeaDAS.
Table A1. Satellite scenes and their correspondent sites.
Landsat Scene ID Site
['LC81810302014141LGN00' Galata
['LC81810302014253LGN00' Galata
['LC81810302015240LGN00' Galata
['LC81810302015352LGN00' Galata
['LC81800292014086LGN00' Gloria
['LC81800292014358LGN00' Gloria
['LC81800292015041LGN00' Gloria
['LC81800292015361LGN00' Gloria
['LC81280542014026LGN00' GOT_Seaprism
['LC81920192013151LGN00' Gustav_Dalen_Tower
['LC81880182013235LGN00' Helsinki_Lighthouse
['LC81880182014190LGN00' Helsinki_Lighthouse
['LC81880182016180LGN00' Helsinki_Lighthouse
['LC81880182016228LGN00' Helsinki_Lighthouse
['LC81880182016260LGN00' Helsinki_Lighthouse
['LC80200312016219LGN00' Lake_Erie
['LC80200312016235LGN00' Lake_Erie
['LC80130322013273LGN00' LISCO
['LC80130322014004LGN00' LISCO
Figure A1. The root-mean-square errors showing the impacts of per-band spectral adjustment on
AERONET-OC match-ups. For all AC methods, there is no noticeable effect in the 443 nm channel.
Similarly, for the land-based AC methods, there are no observable differences in the 443 and 482 nm
channels. Band adjustment improves the results for bands 2, 3, and 4 for SeaDAS, decreasing RMSE
values by 16.6, 23.9, and 43.8% in the 482, 561, and 655 nm wavelengths, respectively, and also improves
results for bands 3 and 4 for ACOLITE by 15.6 and 24.2%, respectively. For SeaDAS, the largest
observable difference is in the 655 nm channel. This is by far the largest improvement from band
adjustment across all bands and AC methods. Overall, SeaDAS is the most sensitive method to
spectral band differences, with the largest difference (improvement) in the 655 nm channel. (a) ARCSI;
(b) LaSRC; (c) ACOLITE; (d) SeaDAS.
Remote Sens. 2019, 11, 469 14 of 20
Table A1. Satellite scenes and their correspondent sites.
Landsat Scene ID Site
[‘LC81810302014141LGN00’ Galata
[‘LC81810302014253LGN00’ Galata
[‘LC81810302015240LGN00’ Galata
[‘LC81810302015352LGN00’ Galata
[‘LC81800292014086LGN00’ Gloria
[‘LC81800292014358LGN00’ Gloria
[‘LC81800292015041LGN00’ Gloria
[‘LC81800292015361LGN00’ Gloria
[‘LC81280542014026LGN00’ GOT_Seaprism
[‘LC81920192013151LGN00’ Gustav_Dalen_Tower
[‘LC81880182013235LGN00’ Helsinki_Lighthouse
[‘LC81880182014190LGN00’ Helsinki_Lighthouse
[‘LC81880182016180LGN00’ Helsinki_Lighthouse
[‘LC81880182016228LGN00’ Helsinki_Lighthouse
[‘LC81880182016260LGN00’ Helsinki_Lighthouse
[‘LC80200312016219LGN00’ Lake_Erie
[‘LC80200312016235LGN00’ Lake_Erie
[‘LC80130322013273LGN00’ LISCO
[‘LC80130322014004LGN00’ LISCO
[‘LC80130322015023LGN00’ LISCO
[‘LC80130322015279LGN00’ LISCO
[‘LC80130322016266LGN00’ LISCO
[‘LC80110312013291LGN00’ MVCO
[‘LC80110312014038LGN00’ MVCO
[‘LC80110312014150LGN00’ MVCO
[‘LC80110312015025LGN00’ MVCO
[‘LC80110312014086LGN00’ MVCO
[‘LC81950192013156LGN00’ Palgrunden
[‘LC81950192016165LGN00’ Palgrunden
[‘LC81990242016129LGN00’ Thornton_C-power
[‘LC81990242016305LGN00’ Thornton_C-power
[‘LC80410372014312LGN00’ USC_SEAPRISM
[‘LC80410372016222LGN00’ USC_SEAPRISM_2
[‘LC80410372016318LGN00’ USC_SEAPRISM_2
[‘LC80410372016334LGN00’ USC_SEAPRISM_2
[‘LC81920292014106LGN00’ Venise
[‘LC81920292015013LGN00’ Venise
[‘LC81920292015221LGN00’ Venise
[‘LC81920292016016LGN00’ Venise
[‘LC81920292016128LGN00’ Venise
[‘LC81920292016192LGN00’ Venise
[‘LC81920292016240LGN00’ Venise
[‘LC80220402013240LGN00’ WaveCIS_Site_CSI
[‘LC80220402013320LGN00’ WaveCIS_Site_CSI
[‘LC80220402014019LGN00’ WaveCIS_Site_CSI
[‘LC80220402014291LGN00’ WaveCIS_Site_CSI
[‘LC80220402014323LGN00’ WaveCIS_Site_CSI
[‘LC80220402015038LGN00’ WaveCIS_Site_CSI
[‘LC80220402015342LGN00’ WaveCIS_Site_CSI
[‘LC80220402016009LGN00’ WaveCIS_Site_CSI
[‘LC80220402016041LGN00’ WaveCIS_Site_CSI
[‘LC80220402016073LGN00’ WaveCIS_Site_CSI
[‘LC81990242014091LGN00’ Zeebrugge-MOW1
[‘LC81990242014219LGN00’ Zeebrugge-MOW1
Remote Sens. 2019, 11, 469 15 of 20
Table A2. Values of environmental parameters for each match-up.
Station Date SZA
(
0
)
AOT 869
(nm)
Wind Speed
(m/s)
Chlorophyll-a
(mg/m3
)
Galata_2014141 27.68254 0.061308 4.109681 1.15
Galata_2014253 41.58995 0.116449 3.284061 1.10
Galata_2015240 37.50297 0.058537 2.129808 0.73
Galata_2015352 68.64532 0.191727 4.643727 0.62
Gloria_2014086 45.39897 0.039736 1.40709 1.03
Gloria_2014358 69.93295 0.009096 13.20025 2.28
Gloria_2015041 62.24466 0.011158 9.488579 1.64
Gloria_2015361 70.11336 0.01644 8.966497 1.31
Got_2014026 39.26405 0.184762 2.348026 0.81
Gustav_2013151 37.76921 0.045492 7.765895 1.44
Helsinki_2013235 49.6655 0.045049 7.813921 4.11
Helsinki_2014190 38.82244 0.052049 5.058227 5.19
Helsinki_2016180 38.00992 0.036343 3.183139 3.87
Helsinki_2016228 47.29317 0.015965 5.356986 3.00
Helsinki_2016260 58.31484 0.014555 7.385065 3.66
LakeErie_2016219 31.06475 0.036835 4.838009 5.32
LakeErie_2016235 35.04874 0.032271 2.098577 5.84
LISCO_2013273 45.86886 0.02143 6.629846 6.12
LISCO_2014004 65.76056 0.009206 3.691909 3.92
LISCO_2015023 63.25687 0.01911 5.097444 5.36
LISCO_2015279 48.07922 0.025828 6.469751 4.84
LISCO_2016266 43.53788 0.03848 4.592692 4.06
MVCO_2013291 53.30351 0.016554 8.056089 3.24
MVCO_2014038 60.57072 0.025061 6.897844 4.52
MVCO_2014086 43.2755 0.042702 8.934463 4.96
MVCO_2014150 25.55208 0.054678 2.590076 1.50
MVCO_2015025 64.20812 0.036832 10.15678 5.03
Palgrunden_2013156 37.03093 0.01894 3.94127 7.58
Palgrunden_2016165 36.71529 0.013707 0.5948 6.87
Thornton_2016129 36.60009 0.070453 7.756932 16.3
Thornton_2016305 66.8652 0.058625 2.756128 3.24
USCSeaPrism_2014312 52.58024 0.028872 4.974118 0.22
USCSeaPrism_2016222 37.88999 0.074677 3.123159 0.63
USCSeaPrism_2016318 54.05641 0.027335 3.450807 0.30
USCSeaPrism_2016334 57.61152 0.026866 3.217656 0.61
Venise_2014106 37.88999 0.023221 6.324373 3.41
Venise_2015013 68.62708 0.039125 3.700884 1.19
Venise_2015221 33.40939 0.125445 3.092528 0.78
Venise_2016016 68.39465 0.011226 6.740557 0.58
Venise_2016128 31.50274 0.03962 1.123216 1.01
Venise_2016192 27.88954 0.085338 1.76539 1.59
Venise_2016240 38.46594 0.033166 1.342931 1.87
WaveCIS_2013240 28.31299 0.080524 3.036319 2.15
WaveCIS_2013320 50.72926 0.069036 6.575934 2.20
WaveCIS_2014019 54.4896 0.03491 7.112117 3.99
WaveCIS_2014291 42.45005 0.016669 3.183118 1.55
WaveCIS_2014323 51.5107 0.016451 2.907233 1.53
WaveCIS_2015038 50.60994 0.022994 2.271182 1.80
WaveCIS_2015342 55.17512 0.033926 1.131623 3.37
WaveCIS_2016009 56.01941 0.072489 4.151914 3.19
WaveCIS_2016041 49.98228 0.008506 5.38026 3.97
WaveCIS_2016073 39.38958 0.052527 5.027627 2.76
Zeebruge_2014091 49.09826 0.093231 2.259445 3.42
Zeebruge_2014219 37.90776 0.13111 3.071374 4.11
Remote Sens. 2019, 11, 469 16 of 20 Remote Sens. 2018, 10, x FOR PEER REVIEW 2 of 20
WaveCIS_2014019 54.4896 0.03491 7.112117 3.99
WaveCIS_2014291 42.45005 0.016669 3.183118 1.55
WaveCIS_2014323 51.5107 0.016451 2.907233 1.53
WaveCIS_2015038 50.60994 0.022994 2.271182 1.80
WaveCIS_2015342 55.17512 0.033926 1.131623 3.37
WaveCIS_2016009 56.01941 0.072489 4.151914 3.19
WaveCIS_2016041 49.98228 0.008506 5.38026 3.97
WaveCIS_2016073 39.38958 0.052527 5.027627 2.76
Zeebruge_2014091 49.09826 0.093231 2.259445 3.42
Zeebruge_2014219 37.90776 0.13111 3.071374 4.11
Remote Sens. 2018, 10, x FOR PEER REVIEW 2 of 20
Figure A2. Line graphs showing the Rrs spectra of each of the 14 AERONET-OC stations (Results were
averaged for each station except GOT Seaprism for which only one match-up is available). Figure A2. Cont.
Remote Sens. 2019, 11, 469 17 of 20 Remote Sens. 2018, 10, x FOR PEER REVIEW 2 of 20
Figure A2. Line graphs showing the Rrs spectra of each of the 14 AERONET-OC stations (Results were
averaged for each station except GOT Seaprism for which only one match-up is available).
Figure A2. Line graphs showing the Rrs spectra of each of the 14 AERONET-OC stations (Results were
averaged for each station except GOT Seaprism for which only one match-up is available).
Table A3. Cases of negative Rrs retrievals from the four AC algorithms.
ACOLITE LaSRC SeaDAS
561 nm USC Seaprism: 2016222 443 nm WaveCIS: 2013320 443 nm Palgrunden: 2013156
655 nm Gloria: 2014358
USC Seaprism: 2016222 655 nm WaveCIS: 2013320
MVCO: 2014150 655 nm
GOT Seaprism: 2014026
USC Seaprism: 2016222
Venise: 2015221