remote sensing
Article
A Novel Atmospheric Correction for Turbid Water
Remote Sensing
Dian Wang 1 , Xiangyu Xiang 1 , Ronghua Ma 2,3, * , Yongqin Guo 1 , Wangyuan Zhu 1,4 and Zhihao Wu 1
Citation: Wang, D.; Xiang, X.; Ma, R.;
Guo, Y.; Zhu, W.; Wu, Z. A Novel
Atmospheric Correction for Turbid
Water Remote Sensing. Remote Sens.
2023, 15, 2091. https://doi.org/
10.3390/rs15082091
Academic Editors: Andrew Clive
Banks, Chong Fang, Zhidan Wen and
Shaohua Lei
Received: 28 February 2023
Revised: 7 April 2023
Accepted: 13 April 2023
Published: 15 April 2023
Copyright: © 2023 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
Remote Sens. 2023, 15, 2091. https://doi.org/10.3390/rs15082091
1 Marine Science and Technology College, Zhejiang Ocean University, Zhoushan 316022, China;
wangdian@zjou.edu.cn (D.W.)
2 Nanjing Institute of Geography & Limnology Chinese Academy of Sciences, Nanjing 210008, China
3 University of Chinese Academy of Sciences, Beijing 100191, China
4 Zhejiang Haida Marine Survey, Planning and Design Co., Ltd., Zhoushan 316022, China
* Correspondence: rhma@niglas.ac.cn
Abstract: For the remote sensing of turbid waters, the atmospheric correction (AC) is a key issue.
The “black pixel” assumption helps to solve the AC for turbid waters. It has proved to be inaccurate
to regard all water pixels in the SWIR (Short Wave Infrared) band as black pixels. It is necessary to
perform atmospheric correction in the visible bands after removing the radiation contributions of
water in the SWIR band. Here, the modified ACZI (m-ACZI) algorithm was developed. The m-ACZI
assumes the spatial homogeneity of aerosol types and employs the BPI (Black Pixel Index) and PIFs
(Pseudo-Invariant Features) to identify the “black pixel”. Then, the radiation contributions of waters
in the SWIR band are removed to complete the atmospheric correction for turbid waters. The results
showed that the m-ACZI had better performance than the SeaDAS (SeaWiFS Data Analysis System)
-SWIR and the EXP (exponential extrapolation) algorithm in the visible band (sMAPE < 30.71%,
RMSE < 0.0111 sr−1 ) and is similar to the DSF (Dark Spectrum Fitting) algorithm in floating algae
waters. The m-ACZI algorithm is suitable for turbid inland waters.
Keywords: atmospheric correction; turbid water; ocean color; black pixel
1. Introduction
Atmospheric correction (AC) is a key step of ocean color data processing. The “black
pixel” assumption is commonly applied to AC for waters (including open ocean, coastal,
and inland waters) [1–7]. The radiance of the “black pixel” is assumed to be dominated by
atmospheric radiance, and the contribution from water can be neglected; hence, the AC of
satellite images can be completed using the “black pixel” assumption [6,8]. This assumption
depends on the fact that the water radiation is dominated by the strong absorption of pure
water. Early on, waters in near-infrared (NIR) bands were determined to fit this assumption,
and the water-leaving reflectance (ρw ) (or the remote sensing reflectance (Rrs )) in the visible
bands was derived by removing the aerosol scattering through extrapolation from the NIR
bands [6,8,9].
However, follow-up studies showed that the contribution of turbid water in NIR
bands could not be neglected [10,11]. The short-wave infrared (SWIR) bands with a longer
wavelength and stronger absorption were considered to fit the “black pixel” assumption for
turbid waters [9,12–14]. Wang et al. (2019) found that the ρw of algae waters was strong in
SWIR bands, meaning these bloom waters no longer met the “black pixel” assumption [15].
In Lake Taihu, significant water signals in SWIR bands were detected when algal blooms
occurred (Figure 1). Therefore, the contribution of waters in SWIR bands should be removed
before the estimation of aerosol scattering in visible bands. Wang et al. (2019) developed
the BPI (black pixel index) to obtain the “black pixel” from turbid waters and proposed
the ACZI algorithm (atmospheric correction algorithm based on a zero assumption for the
short-wave infrared band) for inland turbid waters. The water pixels in the 2201 nm band
https://www.mdpi.com/journal/remotesensing
Remote Sens. 2023, 15, 2091 2 of 19

Remote Sens. 2023, 15, 2091 proposed the ACZI algorithm (atmospheric correction algorithm based on a zero assump- 2 of 19
tion for the short-wave infrared band) for inland turbid waters. The water pixels in the
2201 nm band were considered black pixels by the ACZI algorithm for Landsat 8 images,
andwere aerosol scattering
considered blackinpixels
the visible
by thebands
ACZI was derived
algorithm forfrom
Landsatthe 8assumption
images, andofaerosol
the zero
contribution
scattering inof thewaters
visiblein the was
bands 2201derived
nm band from[12].
the Figure 1 shows
assumption of thethat
zerothe Rayleigh-cor-
contribution of
waters
rected in the 2201
reflectance (ρnm
rc) inband
SWIR [12]. Figure
bands 1 shows
is very that in
different thewater
Rayleigh-corrected reflectance
areas without algal blooms;
it (ρ ) in SWIRthat
isrcincorrect bands
all iswater
very pixels
different
in in water
the SWIR areas
bandwithout algal blooms;
are treated as blackitpixels.
is incorrect that
It is neces-
all water pixels in the SWIR band are treated as black pixels.
sary to obtain the “black pixel” with a more accurate identification method. It is necessary to obtain the
“black pixel” with a more accurate identification method.
Here, we modified the identification method of the “black pixel” in the ACZI algo-
Here,and
rithm [12], wethe
modified
m-ACZI thealgorithm
identification
was method of the
developed. “black
In the m-ACZIpixel”algorithm,
in the ACZI thealgo-
water
rithm [12], and the m-ACZI algorithm was developed. In the m-ACZI algorithm, the water
contribution in SWIR bands was removed using the combination of BPI [12] and PIFs
contribution in SWIR bands was removed using the combination of BPI [12] and PIFs
(pseudo-invariant features) [16,17], and accurate aerosol scattering in SWIR bands was
(pseudo-invariant features) [16,17], and accurate aerosol scattering in SWIR bands was
derived. This scheme was evaluated in turbid inland waters, with its performance also
derived. This scheme was evaluated in turbid inland waters, with its performance also
compared
comparedwith withother
otherAC ACalgorithms.
algorithms.

Figure
Figure1.1.(a)
(a)False
Falsecolor
colorimage
image (R:
(R: SWIR band, G:
SWIR band, G: NIR
NIRband,
band,B:B:Blue
Blueband),
band),(b)(b) Algal
Algal bloom
bloom map
map
derived by FAI (floating algae index) [18,19], and the Rayleigh-corrected reflectance (ρrc) in SWIR
derived by FAI (floating algae index) [18,19], and the Rayleigh-corrected reflectance (ρrc ) in SWIR
bands (c,d) and its histograms from Lake Taihu with algal bloom on 30 January 2021. (e) Average ρrc
bands (c,d) and its histograms from Lake Taihu with algal bloom on 30 January 2021. (e) Average ρrc
from three boxes (Z1, Z2, and Z3) without algal blooms.
from three boxes (Z1, Z2, and Z3) without algal blooms.

2. Dataset and Methods

2. Dataset and Methods
2.1.
2.1.Satellite
SatelliteData
Data
The
Thecloudless
cloudlessLandsat8-OLI
Landsat8-OLI (Operational LandImager)
(Operational Land Imager)Level-1
Level-1data datawere
were from
from thethe
USGS
USGS(United
(UnitedState
StateGeological
GeologicalSurvey)
Survey)andandwere
wereprocessed
processed to
to the
the RRrsrsdata
datavia
viathe
theAC
ACal-
gorithms (details
algorithms in Section
(details 2.3.2).
in Section In In
2.3.2). this work,
this three
work, OLI
three OLIimages
imagesofofLakeLakeTaihu
Taihu(dates:
(dates:11
May 2017,
11 May 27 May
2017, 27 May2017, and
2017, and2121December
December2017)
2017)were
were matched withthe
matched with theininsitu
situ data
data to to
evaluate the performance
evaluate the performance of theof the AC algorithms. The time window of data matching
algorithms. The time window of data matching was was
set as ±3 h of the Landsat8 overpass. The AC-driven Rrs data were obtained by averaging
Remote Sens. 2023, 15, 2091 3 of 19
Remote Sens. 2023, 15, 2091 3 of 19

set as ±3 h of the Landsat8 overpass. The AC-driven Rrs data were obtained by averaging
a 3 ××33pixel
pixel(the
(thecoefficient
coefficientof
ofvariation
variationfor
forthese
these valid
valid pixels
pixels was
was <<10%)
10%)area
area surrounding
surrounding
the sample
sample point
point location
location[20–23].
[20–23].AAtotal
totalofof71
71sampling
samplingpoints
points were
were matched
matched (11(11 match-
match-up
up points
points fromfrom 11 May
11 May 2017;2017; 22 match-up
22 match-up points points
from from
27 May272017;
May and
2017;38and 38 match-up
match-up points
from 21from
points December 2017) (Figure
21 December 2).
2017) (Figure 2).

Figure 2. In situ sampling points in Lake Taihu.

Figure 2. In situ sampling points in Lake Taihu.

2.2.
2.2. Field
Field Spectral
Spectral Data
Data
The Rrsrs was
was measured
measured using
using the
the spectrometer
spectrometer (ASD (ASD FieldSpec
FieldSpec 4) following NASA
(National Aeronautics and Space Administration) Ocean optics protocols [24] (Figure (Figure 3):
3):
∙ ∙
𝑅

− · (1)
Lw L t σ L ∙ sky · ρ p
Rrs = = (1)
Ed
where Lt is the total water-leaving radiance; · L preflectance; Lg is the radiance of the
ρp isπthe
reference
where Lt ispanel; Lsky water-leaving
the total is the sky radiance; Lw is
radiance; ρp the water-leaving
is the reflectance; Lradiance; and Ed is the
g is the radiance of the
downwelling plane irradiance. Additionally, the water surface reflectance
reference panel; Lsky is the sky radiance; Lw is the water-leaving radiance; and factor
Ed σiswas
the
assumed to be 0.028 [25].
downwelling plane irradiance. Additionally, the water surface reflectance factor σ was
assumed to be 0.028 [25].
emote Sens. 2023, 15, 2091 4

Remote Sens. 2023, 15, 2091 4 of 19

Figure
Figure 3. Measured
3. Measured remotesensing
remote sensing reflectance
reflectance (R(R
rs )rs)from
fromLake Taihu
Lake (dates:
Taihu 11 May
(dates: 2017,2017, 27
11 May
27 May 2017, and 21 December 2017). The line color represents different sampling dates. The
2017, and 21 December 2017). The line color represents different sampling dates. The color co
color column represents the wavelength range of each band of OLI (Deep blue: 443 nm band, Light
represents the wavelength range of each band of OLI (Deep blue: 443 nm band, Light blue: 48
blue: 483 nm band, Green: 561 nm band, Red: 655 nm band, Yellow: 865 nm band).
band, Green: 561 nm band, Red: 655 nm band, Yellow: 865 nm band).
2.3. Methods
2.3.2.3.1. The m-ACZI Algorithm
Methods
The black pixel of the ACZI algorithm was identified using the combination of BPI
2.3.1. The m-ACZI Algorithm
(Equation (2)) [12] and FAI (the floating algae index) [18,19], and this algorithm assumed
The
that blackpixels
all water pixelinof
thethe
2201ACZI
nm bandalgorithm
accord withwas identified
the “black pixel”using the combination
hypothesis. However, of
the
(Equation
ρ in SWIR bands was significant and non-negligible when algal blooms
w (2)) [12] and FAI (the floating algae index) [18,19], and this algorithm assu occurred
(Figure 1). In this study, the m-ACZI algorithm changed ACZI’s black pixel identification
that all water pixels in the 2201 nm band accord with the “black pixel” hypothesis. H
method of the black pixels of ACZI to the BPI-PIF combination, and these black pixels are
ever, the ρto
applied w in SWIR bands was significant and non-negligible when algal blooms occu
remove the water contribution in SWIR bands. After this stage, the following
(Figure
process isIn
1). this
the study,
same as thethe m-ACZI
ACZI algorithm algorithm
[12]. Thechanged ACZI’s black
BPI was developed pixel
by Wang identifica
et al.
method
(2019a)of[12]
thetoblack pixels
identify of ACZI
the black pixeltofrom
the turbid
BPI-PIF combination,
waters and Schott
(Equation (2)). these et
black
al. pixel
(1988) [17] considered that urban construction land has pseudo-invariant
applied to remove the water contribution in SWIR bands. After this stage, the follow features (PIFs)
and proposed the PIF mask method. The BPI and PIF mask followed the methods described
process is the same as the ACZI algorithm [12]. The BPI was developed by Wang e
by Wang et al. (2019a) [12] and Schott et al. (1988) [17], respectively.
(2019a) [12] to identify the black pixel from turbid waters (Equation (2)). Schott et al. (1
[17] considered that urban construction |ρrc (655 ) − has
land ρrc (561 )|
pseudo-invariant features (PIFs) and
BPI = (2)
ρrc (655) − ρrc (865)
posed the PIF mask method. The BPI and PIF mask followed the methods describe
Wang etThe
al.PIFs
(2019a) [12]features
are with and Schott
that doetnot
al.change
(1988)their
[17],reflectivity
respectively.
properties drastically
over time in a scene [17]. The surface reflectance of PIFs is similar for the two dates. Hence,
| |
𝐵𝑃𝐼
the difference between the Rayleigh-corrected reflectance (ρrc ) of PIFs from two dates can
be considered as the difference between two values of aerosol reflectance (ρa ) of PIF. The
black
Thepixels
PIFs are
arepixels
withconsisting
features only
thatofdoatmospheric
not change radiance (Rayleigh scattering
their reflectivity and
properties
drasti
over time in a scene [17]. The surface reflectance of PIFs is similar for the two dates. He
the difference between the Rayleigh-corrected reflectance (ρrc) of PIFs from two dates
be considered as the difference between two values of aerosol reflectance (ρa) of PIF.
 Remote Sens. 2023, 15, 2091 5 of 19

Remote Sens. 2023, 15, 2091
aerosol scattering). If a pixel is a true “black pixel”, the difference between its ρrc for two
dates is the same as the difference between its ρa for the same two dates (Equations (5)–(7)).
Furthermore, theoretically, if the difference in the ρrc of a water pixel between two dates
in the SWIR band is equal or approximate to the difference in the ρrc of PIFs between two
dates, this water pixel can be taken as a “black pixel”. Therefore, the characteristics of PIFs
can be used to more accurately obtain the BPIs for atmospheric correction. The flowchart of
the m-ACZI algorithm is shown in Figure 4.

R.Schott et al. 1988

OLI image
NIR/Red band 6SV-LUT
ratio 2201nm band
(threshold: 0.05)
(threshold: 1.3)
ρrc(λ), t(λ)
Construction land pixel
BPI mask
(threshold: 0~0.1)
PIF mask
(Wang et al. 2019a)

ρrc images from two dates

YES DBPI ≈ DPIF(10% variation) NO

Black pixels water pixels

Aerosol
scattering
ρa(SWIR) ratio Rrs
Figure
Figure4. Processing procedure
4. Processing of the m-ACZI
procedure algorithm.
of the m-ACZI Note: These thresholds
algorithm. Note: areThese
for Lake Taihu
thresholds ar
and may need to be adjusted for other waters [12,17].
and may need to be adjusted for other waters [12,17].
(1) The ρrc and transmittance (t) are obtained from a LUT (lookup table) generated
for all OLI bands using
Described 6SV (the
below vector
is the versionof
process of developing
the Second Simulation
the m-ACZIof the Satellite
algorithm.
Signal in the Solar Spectrum) according to sun and sensor geometry [26,27]. The ρrc can
(1)derived
be We from
extracted the PIFofmask
the reflectance according
the atmosphere (ρt ).to Schott
The et al.,reflectance
sea surface 1988 [17] and Con
from
2016
whitecaps and[16]. Forwere
sunglint Lake Taihu,
ignored the 2201 nm band with a threshold of 0.0
[1,2,28,29].
obtain the water mask, and ρrc = the
ρt − ρthreshold
r
of the NIR/RED band
(3) ratio
for obtaining the vegetation mask. The previous two masks were co
For land pixels:
logic. AND. gate, creating a “PIF mask” that rejects the vegetation
ρrc = ρ a + tρs (4)
and w
accepts the construction land features.
where ρs is the surface reflectance.
(2) ForIna water
this step,
pixel: we calculated the ρrc and t using the 6SV-LUT from ACO
pheric correction for OLI = ρ a + tρw
ρrc lite). (5)

(3) Based on step 2, we adopted the BPI [12] to obtain the BPI mask, and
of the BPI was set to 0~0.1.
(4) Following steps 1, 2, and 3, concerning t values, two images with si
were selected to obtain their PIF mask and BPI mask, respectively. The
were applied to the ρrc (SWIR) images, respectively, to obtain the PIF an
Remote Sens. 2023, 15, 2091 6 of 19

(2) Two masks are made by the BPI for two date images (T1 and T2) [4,12], and the
overlapped pixels in two masks (BPI mask) are retained. The difference in ρrc (SWIR) for
two date images (DBPI ) is calculated and masked by the BPI mask.

D BPI BPI ( SW IR ) − ρ BPI ( SW IR )

= ρrc rc T2
 T1    (6)
BPI
= ρ aT1 (SW IR) + t T1 ρw BPI ( SW IR ) − ρ BPI ( SW IR ) + t ρ BPI ( SW IR )
T1 a T2 T2 w T2

(3) The PIF mask is extracted according to Schott et al. (2016) [16]. The construction
land features are considered as PIFs, and this extraction is applied over the construction
area within the image (Figure 5). For PIFs, the ρs is invariable, and the difference in ρrc
(SWIR) for two date images (DPIF ) is masked by the PIF mask.

D PIF = ρrc
PIF ( SW IR ) − ρ PIF ( SW IR )
T1 rc T2

Remote Sens. 2023, 15, 2091

= (ρ aPIF
T1
(SW IR) + t T1 ρsPIF
T1
(SW IR)) − (ρ aPIF
T2
(SW IR) + t T2 ρsPIF
T2
(SW IR)) (7)
7 of 19
= ρ aPIF
T1
(SW IR) − ρ aPIF
T2
(SW IR) + ∆tρsPIF (SW IR)

Figure 5. Example of the PIF mask determination. (a) False color image (R: SWIR band, G: NIR band,
Figure 5. Example of the PIF mask determination. (a) False color image (R: SWIR band, G: NIR
B: Blue band) with vegetation in green; (b) Construction land and water; and (c) PIF mask over
band, B:near
airports BlueLake
band)Taihu,
with vegetation
China. in green; (b) Construction land and water; and (c) PIF mask over
airports near Lake Taihu, China.

∆t will make ρS PIF residual in DPIF and affect the matching of DPIF and DBPI ; for this,
we analyze and discuss the impact of ∆t on DPIF (1609) (details in Section 4.4). The results
showed that when ∆t was less than 0.0107, the DPIF change caused by ∆t was about 2.70%,
which is within the allowable error range.
(4) The ρa BPI and ρa PIF are similar in one OLI image based on the assumption of the
spatial uniformity distribution of aerosol types. The DBPI matches the mode number of DPIF
(with ±10% variation) to obtain the final BPI-PIF mask. The black pixels were identified
using the BPI-PIF mask (Example: Figure 6). The aerosol scattering from black pixels is
used to remove the water contribution for nonblack pixels.

D BPI ∼
= D PIF (8)

(5) The atmospheric correction for all water pixels is completed using the same steps
as the ACZI algorithm [12], and Rrs (or ρw ) data are obtained.
Described below is the process of developing the m-ACZI algorithm.
 Figure 5. Example of the PIF mask determination. (a) False color image (R: SWIR band, G: NIR band,
B:2023,
Remote Sens. Blue 15, band)
with vegetation in green; (b) Construction land and water; and (c) PIF mask over
2091 7 of 19
airports near Lake Taihu, China.

Figure 6. Histogram of DBPI (blue bar) and DPIF (yellow bar) derived by two date images (Dates:
Figure 6. Histogram of DBPI (blue bar) and DPIF (yellow bar) derived by two date images (Dates: 11
11 May 2017 and 27 May 2017) in Lake Taihu, China.
May 2017 and 27 May 2017) in Lake Taihu, China.
(1)
We extracted the PIF mask according to Schott et al., 1988 [17] and Concha and Schott
2016Correction
2.3.2. Other Atmospheric [16]. For Lake Taihu, the 2201 nm band with a threshold of 0.05 was used to
Algorithms
obtain the water mask, and the threshold of the NIR/RED band ratio was set to 1.3
for obtaining the vegetation mask. The previous two masks were combined using
The SeaDAS-SWIR Algorithm
logic. AND. gate, creating a “PIF mask” that rejects the vegetation and water and only
For turbid waters, the “black
accepts pixel” assumption
the construction land features. in NIR bands is invalid because of
(2) In this step, we calculated
the contribution of phytoplankton, detritus, and rc and t using
the ρsuspended the 6SV-LUT
sediment from ACOLITE
to water backscatter (atmo-
spheric correction for OLI lite).
[2,10]. Research from Hale et al. (1973) indicated that the water in SWIR bands has stronger
(3) Based on step 2, we adopted the BPI [12] to obtain the BPI mask, and the threshold of
absorption than that intheNIR bands
BPI was [30],
set to and the turbid water in the former meets the re-
0~0.1.
quirement of the(4) “black pixel”
Following assumption
steps [13]. Therefore,
1, 2, and 3, concerning Wang
t values, two and
images withShi (2005)
similar [10]
t values were
selected to obtain their PIF mask and BPI mask, respectively. Then, these masks were
applied to the ρrc (SWIR) images, respectively, to obtain the PIF and BPI images. We
subtracted two of each image type to obtain the DPIF and DBPI images.
(5) We counted the average value of DPIF and used the average DPIF with 10% variation
as the judgment value for DBPI , i.e., DBPI pixels greater than the average DPIF with
10% variation were discarded.
(6) Following the ACZI process [12], we counted the median value from the final retained
DBPI pixels, and this median value was taken as ρa on two SWIR bands. The aerosol
scattering ratio (ε), consisting of these two ρa (SWIR) values, was applied to all water
pixels, and aerosol scattering in the visible and near-infrared bands was obtained
using an exponential extrapolation method, thus completing the final atmospheric
correction process to obtain Rrs .
Remote Sens. 2023, 15, 2091 8 of 19

2.3.2. Other Atmospheric Correction Algorithms

The SeaDAS-SWIR Algorithm
For turbid waters, the “black pixel” assumption in NIR bands is invalid because of the
contribution of phytoplankton, detritus, and suspended sediment to water backscatter [2,10].
Research from Hale et al. (1973) indicated that the water in SWIR bands has stronger
absorption than that in NIR bands [30], and the turbid water in the former meets the
requirement of the “black pixel” assumption [13]. Therefore, Wang and Shi (2005) [10]
developed an AC algorithm for turbid water based on extrapolation from SWIR bands and
integrated this AC algorithm into SeaDAS (SeaWiFS Data Analysis System, referred to as
SeaDAS-SWIR in this paper; more details can be found in Wang and Shi (2005) [10]).
The ACOLITE (atmospheric correction for OLI lite; this study worked with version
20211112) tool was developed for coastal and inland waters by the Remote Sensing and
Ecosystem Modelling (REMSEM) team. It performs this process using dark spectrum fitting
(DSF) and exponential extrapolation (EXP) [14,26,31–33].

The EXP Algorithm

The idea of the EXP algorithm is consistent with the SeaDAS-SWIR algorithm, both of
which are based on exponential extrapolations of the ratio of multiple aerosol scattering in
SWIR bands, but the EXP algorithm does not deal with aerosol LUTs (lookup tables) [14].
The EXP algorithm only uses an exponential extrapolation method to estimate aerosol
scattering in visible bands and NIR bands [28].

The DSF Algorithm

The DSF algorithm is presented for meter-scale-resolution optical satellite imagery [26,31].
Similar to other AC algorithms, this algorithm was also developed based on the “black
pixel” concept [26]. However, the difference is that it defines black pixels as pixels with zero
surface reflection in one scene image, including land pixels and water pixels. These black
pixels are employed to estimate the atmospheric radiance for other pixels. It can be seen
that the DSF depends on two assumptions: (1) the spatial distribution of the atmosphere is
uniform in an image; (2) at least one pixel whose surface reflection is about zero exists in
one scene image [26,31].

2.3.3. Algorithm Accuracy Analysis

The root-mean-square error (RMSE), symmetric mean absolute percentage error
(sMAPE), and Bias were used to determine the differences between the in situ data and the
AC-driven data: s

2
∑in=1 Rm − Ri
RMSE = (9)
n

| R m − Ri |
∑in=1
| R m + Ri |
sMAPE = × 2 × 100% (10)
n
 m i 
(R −R )
∑in=1 Ri
Bias = × 100% (11)
n
where Ri is the in situ Rrs data, Rm is the OLI-Rrs data via atmospheric correction, and n is
the number of match-up pairs.

3. Results
3.1. Assessment of the m-ACZI Algorithm
As Figures 7 and 8 show, the m-ACZI raised the Rrs estimation compared to the
ACZI algorithm, so the histograms of Rrs generally moved and concentrated to high
value areas (Figure 8). Notably, the m-ACZI algorithm did not increase the Rrs values
Remote Sens. 2023, 15, 2091
Remote Sens. 2023, 15, 2091
9 of 19
for all water pixels, but mainly for algal bloom pixels (green pixels in the RGB true-color
images) (Figures 2 and 7). It can also be observed that the m-ACZI algorithm reduced
9 of 19 the
Rrs estimation of the non-algal bloom pixels in the 865 nm band (Figure 7), which were
consistent with the characteristics of the in situ Rrs (Figure 3).
Figure 7. Comparison of the ACZI and the m-ACZI algorithms for OLI Rrs over Lake Taihu (dates:
Figure 7. Comparison of the ACZI and the m-ACZI algorithms for OLI Rrs over Lake Taihu (dates:
11 May 2017, 27 May 2017, and 21 December 2017).
11 May 2017, 27 May 2017, and 21 December 2017).
Remote
RemoteSens. 2023,15,
Sens.2023, 15,2091
2091 10 of
of 19
19

Figure 8. Frequency of Rrs retrieved using the ACZI (blue) and the m-ACZI (orange) algorithm in
Figure 8. Frequency of Rrs retrieved using the ACZI (blue) and the m-ACZI (orange) algorithm in
Lake Taihu (Dates: 11 May 2017, 27 May 2017, and 21 December 2017).
Lake Taihu (Dates: 11 May 2017, 27 May 2017, and 21 December 2017).

Mostofofthe
Most thescatterplots
scatterplots ofof
thethe
ACZIACZIwerewere below
below thethe 1:1 line,
1:1 line, and and
part part
of theofscatterplots
the scatter-
plots derived in the 865 nm band were above the 1:1 line (Figure
derived in the 865 nm band were above the 1:1 line (Figure 9a). The scatterplots of the 9a). The scatterplots of
the m-ACZI were distributed around the 1:1 line, with several significant
m-ACZI were distributed around the 1:1 line, with several significant off-group points in off-group points
in the
the 865 865
nm nmband band (Figure
(Figure 9b). 9b). In general,
In general, compared
compared withwith the scatterplots
the scatterplots of theofACZI,
the ACZI,
the
the scatterplots of the m-ACZI were more concentrated near
scatterplots of the m-ACZI were more concentrated near the 1:1 line. The ACZI-driventhe 1:1 line. The ACZI-driven
RRrsrs was
was underestimated
underestimated in in all
all visible
visiblebands.
bands.The TheRrsRrs
derived
derived bybythethe
m-ACZI
m-ACZI waswasalsoalso
un-
derestimated in the 561 nm and 655 nm bands, but overestimated
underestimated in the 561 nm and 655 nm bands, but overestimated in two BLUE bands in two BLUE bands
(Figure99and
(Figure andTable
Table1).1).The
Theminimum
minimumsMAPE sMAPEwas wasobtained
obtainedby bythe
them-ACZI
m-ACZIin inthe
the655
655nm nm
band,followed
band, followedby bythe
them-ACZI
m-ACZIin inthe
the561
561nm nmband.
band.Similarly,
Similarly,the theminimum
minimumRMSE RMSEwas was
providedby
provided bythe
them-ACZI
m-ACZIininthe the483483nmnmband,
band,followed
followedby bythe
them-ACZI
m-ACZIin inthe
the443
443 nmnm band.
band.
However,the
However, theminimum
minimum bias
bias waswas obtained
obtained fromfrom
thethe
datadata derived
derived by ACZI
by the the ACZIin thein443
thenm 443
nm band (Table 1). Overall, from RMSE and sMAPE, the
band (Table 1). Overall, from RMSE and sMAPE, the m-ACZI was superior to the ACZI m-ACZI was superior to the
ACZI in all visible bands, especially in the 561 nm and 655 nm
in all visible bands, especially in the 561 nm and 655 nm bands; the performance of thebands; the performance of
the m-ACZI
m-ACZI algorithm
algorithm was was greatly
greatly improved.
improved. However,
However, thesethese 2 algorithms
2 algorithms still failed
still failed in the in
the NIR
NIR bandband
(RMSE (RMSE
> 0.0203 sr−1 ;srsMAPE
> 0.0203 −1; sMAPE > 48.31%;
> 48.31%; biasbias > 88.83%).
> 88.83%). FromFromthe the scatterplots
scatterplots of
of the
the 865 865
nm nmband band in Figure
in Figure 9, these
9, these 2 AC2algorithms
AC algorithms havehave overestimations.
overestimations. The assess-
The assessment
results indicate
ment results that thethat
indicate m-ACZI algorithm
the m-ACZI is generally
algorithm better than
is generally thethan
better ACZI algorithm.
the ACZI algo-
rithm.
Table 1. Statistics of the accuracy measures for the ACZI and the m-ACZI algorithms using in situ
Rrs measurements. The BLUE number represents the minimal statistical value.

Method 443 483 561 655 865

RMSE(sr−1 ) 0.0063 0.0062 0.0111 0.0082 0.0228
m-ACZI
sMAPE(%) 30.71 23.01 21.41 21.27 48.31
Bias(%) 52.07 7.59 −16.48 −11.65 88.83
ACZI RMSE(sr−1 ) 0.0070 0.0076 0.0146 0.0102 0.0203
sMAPE(%) 35.88 34.07 29.16 29.10 57.48
Bias(%) −8.91 −12.32 −22.70 −16.18 94.16
Remote Sens. 2023, 15, 2091 11 of 19
Remote Sens. 2023, 15, 2091 11 of 19

Figure
Figure 9.
9. Scatterplots of RRrsrs derived
Scatterplots of derivedby
bythe
theACZI
ACZI(a)
(a)and
andthe
the m-ACZI
m-ACZI (b)(b) algorithms
algorithms versus
versus in situ
in situ Rrs .
Rrs.
3.2. Comparison with other AC Algorithms
TableOnly
1. Statistics of the accuracy
54 match-up samplemeasures for the
points were ACZI
used and the m-ACZI
to evaluate algorithms using
the SeaDAS-SWIR in situ
algorithm
rs measurements. The BLUE number represents the minimal statistical value.
Rbecause some water pixels were masked by this algorithm (the “proc_ocean” option of
SeaDAS
Methodwas set to “2-force all pixels 443 to be processed
483 as ocean”).
561 In Figure
655 10, the scatterplots
865
of the m-ACZI algorithm were more concentrated, while those of other algorithms were
RMSE(sr ) −1 0.0063 0.0062 0.0111 0.0082 0.0228
m-ACZIespecially in the 443 nm and 483 nm bands. The scatterplot distributions of the
discrete,
sMAPE(%) 30.71 23.01 21.41 21.27 48.31
m-ACZI and the DSF were very similar in the 561 nm and 655 nm bands and were closer to
Bias(%) 52.07 7.59 −16.48 −11.65 88.83
the 1:1 line than others (Figure 10). The statistical results of the accuracy also supported
ACZI RMSE(sr −1) 0.0070 0.0076 0.0146 0.0102
these findings (Table 2). The assessment results from the m-ACZI and the DSF were very 0.0203
similar: the RMSE sMAPE(%)difference35.88between them 34.07was less 29.16
than 0.0008 29.10 57.48 and
in visible bands,
both were more Bias(%)
accurate than −8.91
the SeaDAS −12.32
and the −22.70 −16.18It should94.16
EXP algorithms. be noted
that all the AC algorithms failed in the NIR band (RMSE > 0.0134 sr ; sMAPE > 49.31%; − 1
3.2.
BiasComparison
> 88.83%), with otherm-ACZI
and the AC Algorithms
algorithm showed the best comprehensive performance.
Only 54 match-up sample points were used to evaluate the SeaDAS-SWIR algorithm
Table 2. Statistics
because some waterof thepixels
accuracy measures
were maskedfor the
by m-ACZI, SeaDAS,(the
this algorithm DSF,“proc_ocean”
and EXP algorithms
optionusing
of
in situ R measurements. The BLUE number represents the minimal statistical
SeaDAS was set to “2-force all pixels to be processed as ocean”). In Figure 10, the scatter-
rs value.
plots of the m-ACZI algorithm were more concentrated, while those of other algorithms
Method 443 483 561 655 865
were discrete, especially in the 443 nm and 483 nm bands. The scatterplot distributions of
m-ACZI −1 ) 0.0063
the m-ACZI and the DSF RMSE(sr
were very similar in 0.0062
the 561 nm0.0111and 655 nm 0.0082
bands and 0.0228
were
sMAPE(%) 30.71 23.01 21.41 21.27 48.31
closer to the 1:1 line than others (Figure 10). The statistical results of the accuracy also
Bias(%) 52.07 7.59 −16.48 −11.65 88.83
supported these findings
SeaDAS-SWIR (Table
RMSE(sr −12).
) The assessment
0.0113 0.0100 results from
0.0108the m-ACZI
0.0104 and 0.0134
the DSF
were very(n = similar:
54) thesMAPE(%)
RMSE difference 45.11 between 32.70them was less than27.33
20.65 0.0008 in76.35
visible
bands, and both were more accurate 33.32
Bias(%) than the SeaDAS
13.36 and − the
1.38EXP algorithms.
4.37 It126.68
should
−1 )
be noted EXP that all the AC RMSE(sr
algorithms 0.0060in the0.0070
failed NIR band 0.0150 0.0105sr−1; sMAPE
(RMSE > 0.0134 0.0217 >
sMAPE(%) 34.12 27.76 32.79
49.31%; Bias > 88.83%), and the m-ACZI algorithm showed the best comprehensive 28.29 55.56
per-
Bias(%) 2.50 −10.31 −27.00 −19.72 90.38
formance. −1
DSF RMSE(sr ) 0.0058 0.0061 0.0098 0.0080 0.0244
sMAPE(%) 32.27 24.34 17.68 21.80 63.00
Bias(%) 34.22 4.94 −4.46 0.01 113.55
Remote Sens. 2023, 15, 2091 12 of 19
Remote Sens. 2023, 15, 2091 12 of 19

Figure
Figure10.
10.Scatterplots ofAC-driven
Scatterplots of AC-drivenR Rrs versus
versus in in
situsitu
R .Rrs.
rs rs

3.3. Assessment
Table 2. StatisticsofofAC
theAlgorithms for Different
accuracy measures forWater Types SeaDAS, DSF, and EXP algorithms us-
the m-ACZI,
ing in According
situ Rrs measurements. The BLUE number represents
to the water classification method of Sun et al. the(2012),
minimal statistical
waters value.
were classified
into two types: floating bloom (FB) and un-floating bloom (uFB) [15,34]. Figure 11 shows
Method of R between in situ data and
the scatterplots 443 the AC 483
algorithm in 561 655 types,865
different water
rs
with m-ACZI RMSE(srin)Tables0.0063
detailed statistics presented −1 0.0062
3 and 4. All 0.0111had more
AC algorithms 0.0082 0.0228
discrete
scatterplots in FB than in sMAPE(%)
uFB, especially 30.71
in the two 23.01
blue bands, 21.41
where the21.27bias of the48.31
algorithm increased significantly
Bias(%)(Tables 3 and 4). Table 7.59
52.07 3 shows that the assessment
−16.48 −11.65results88.83
of the m-ACZI and the DSF algorithms were superior to other algorithms in the uFB. In
SeaDAS-SWIR RMSE(sr−1) 0.0113 0.0100 0.0108 0.0104 0.0134
particular, the m-ACZI and the DSF algorithms showed satisfactory performance in the 561
(n = 54) sMAPE(%) 45.11 32.70 20.65
nm and 655 nm bands (RMSE < 0.0087 sr−1 ; sMAPE < 19.31%), which was sufficient to sup-
27.33 76.35
port the quantitative remote Bias(%) 33.32
sensing research 13.36
for turbid waters [5].−1.38
Compared4.37 with those126.68
EXPthe accuracy RMSE(sr
in the uFB, −1 ) AC 0.0060
results of the algorithms 0.0070
in FB were 0.0150 0.0105
reduced (Tables 0.0217
3 and 4).
The DSF algorithm had sMAPE(%)
the best performance34.12in the FB waters,
27.76 followed
32.79 by the m-ACZI
28.29 55.56
algorithm. The SeaDAS-SWIR algorithm masked
Bias(%) 2.50 most
−10.31of the FB pixels, with
−27.00 only 6 in90.38
−19.72
DSF RMSE(sr−1) 0.0058 0.0061 0.0098 0.0080 0.0244
sMAPE(%) 32.27 24.34 17.68 21.80 63.00
Bias(%) 34.22 4.94 −4.46 0.01 113.55
 in the 561 nm and 655 nm bands (RMSE < 0.0087 sr−1; sMAPE < 19.31%), which was suffi
cient to support the quantitative remote sensing research for turbid waters [5]. Compared
with those in the uFB, the accuracy results of the AC algorithms in FB were reduced (Ta
bles 3 and 4). The DSF algorithm had the best performance in the FB waters, followed by
Remote Sens. 2023, 15, 2091 13 of 19
the m-ACZI algorithm. The SeaDAS-SWIR algorithm masked most of the FB pixels, with
only 6 in situ points (18 in situ points in total) participating in the evaluation. The perfor
mance of these
situ points (18AC algorithms
in situ was participating
points in total) not satisfactory,
in the especially
evaluation. in
Thethe two blue of
performance bands, a
wasthese
thatACof algorithms
the m-ACZI algorithm.
was not These
satisfactory, AC algorithms
especially (except
in the two blue bands,the ACZI
as was thatalgorithm
of
overestimated the water-leaving
the m-ACZI algorithm. reflection (except
These AC algorithms in the 443 nm and
the ACZI 483 nm
algorithm) bands (Table 4). A
overestimated
the water-leaving
possible reason was reflection
that thein m-ACZI
the 443 nmalgorithm
and 483 nmfor bands
the(Table 4). A possible reason
FB underestimated the aeroso
was that the m-ACZI algorithm for the FB underestimated the aerosol scattering in the
scattering in the SWIR band, which was further exacerbated by exponential extrapolation
SWIR band, which was further exacerbated by exponential extrapolation. This finally
Thisresulted
finallyinresulted in a serious underestimation of the aerosol scattering and an overes
a serious underestimation of the aerosol scattering and an overestimation of the
timation of the water-leaving
water-leaving reflection in the reflection
443 nm andin the
483 nm443 nm and 483 nm bands.
bands.

Figure 11.11.
Figure Scatterplots
Scatterplotsof
of AC-driven
AC-driven RR
rs rsofof different
different water
water typestypes
versusversus
in situ in
Rrssitu
. TheRred
rs. The red point
points
represent water
represent with
water withfloating blooms,and
floating blooms, and thethe
graygray points
points represent
represent un-floating
un-floating blooms. blooms.

Table 3. Statistics of the accuracy measures for atmospheric correction algorithms for un-floating
blooms using in situ Rrs measurements. The BLUE number represents the minimal statistical value.

Method 443 483 561 655 865

m-ACZI RMSE(sr−1 ) 0.0058 0.0058 0.0087 0.0072 0.0127
(n = 53) sMAPE(%) 25.63 20.49 18.97 17.86 48.62
Bias(%) 21.32 3.93 −14.76 −10.45 91.30
ACZI RMSE(sr−1 ) 0.0071 0.0073 0.0109 0.0085 0.0112
(n = 53) sMAPE(%) 31.77 25.94 23.13 22.59 59.54
Bias(%) −6.01 −6.81 −18.20 −10.54 102.18
Remote Sens. 2023, 15, 2091 14 of 19

Table 3. Cont.

Method 443 483 561 655 865

SeaDAS-SWIR RMSE(sr−1 ) 0.0113 0.0101 0.0107 0.0103 0.0123
(n = 48) sMAPE(%) 44.17 31.74 20.02 26.57 74.39
Bias(%) 30.91 11.20 −2.26 2.61 119.49
EXP RMSE(sr−1 ) 0.0058 0.0068 0.0126 0.0093 0.0123
(n = 53) sMAPE(%) 26.63 24.18 30.28 24.10 55.98
Bias(%) −3.30 −10.77 −25.36 −17.42 93.74
DSF RMSE(sr−1 ) 0.0055 0.0058 0.0075 0.0068 0.0144
(n = 53) sMAPE(%) 26.92 21.81 15.16 19.31 65.51
Bias(%) 8.72 3.16 −1.85 1.95 117.79

Table 4. Statistics of the accuracy measures for atmospheric correction algorithms for floating bloom
using in situ Rrs measurements. The BLUE number represents the minimal statistical value.

Method 443 483 561 655 865

m-ACZI RMSE(sr−1 ) 0.0078 0.0072 0.0162 0.0107 0.0396
(n = 18) sMAPE(%) 45.68 30.42 28.61 31.29 47.40
Bias(%) 142.60 18.39 −21.53 −15.18 81.53
ACZI RMSE(sr−1 ) 0.0067 0.0086 0.0222 0.0140 0.0355
(n = 18) sMAPE(%) 51.47 58.00 46.92 48.25 51.41
Bias(%) −19.90 −28.52 −35.93 −32.81 70.54
SeaDAS-SWIR RMSE(sr−1 ) 0.0111 0.0097 0.0115 0.0110 0.0204
(n = 6) sMAPE(%) 52.71 40.43 25.69 33.33 92.05
Bias(%) 52.55 30.56 5.67 18.46 184.26
EXP RMSE(sr−1 ) 0.0067 0.0075 0.0205 0.0134 0.0376
(n = 18) sMAPE(%) 56.19 38.30 40.20 40.63 54.32
Bias(%) 19.58 −8.93 −31.81 −26.47 80.47
DSF RMSE(sr−1 ) 0.0067 0.0070 0.0147 0.0107 0.0416
(n = 18) sMAPE(%) 48.00 31.78 25.10 29.13 55.60
Bias(%) 109.31 10.17 −12.16 −5.71 101.07

4. Discussion
The m-ACZI algorithm depends on two prerequisite assumptions: (1) the spatial
distribution of aerosol types is uniform; (2) the black pixels from two date images overlap.
These are the determining conditions for the success of the m-ACZI algorithm.

4.1. The Assumption of the Spatially Uniformity Distribution of Aerosol Types

Solving the aerosol scattering of nonblack pixels relies on the assumption of the
spatially uniform distribution of aerosol types for the m-ACZI algorithm. This assumption
has been recognized by a great amount of research [1,4,14,26,28]. For example, the DSF
and EXP adopted this assumption for the atmospheric correction process. Taking the Lake
Taihu image without algal blooms (obtained on 2 February 2016) as an example (remove
cloud pixels) (Figure 12), the spatial distributions of ρrc (1609) and ρrc (2201) were uniform,
and the standard deviations of ρrc (1609) and ρrc (2201) were 0.0026 (average: 0.0218) and
0.0016 (average: 0.0137), respectively. The standard deviation of ρrc (1609)/ρrc (2201) was
0.0546 (average: 1.5944) (Figure 12). This suggests that Lake Taihu is consistent with the
assumption of the spatially homogeneous distribution of aerosol types.
Remote Sens. 2023, 15, 2091 0.0546 (average: 1.5944) (Figure 12). This suggests that Lake Taihu is consistent
15 with
of 19 th
assumption of the spatially homogeneous distribution of aerosol types.

0.0016 (average: 0.0137), respectively. The standard deviation of ρrc (1609)/ρrc (2201) was
Remote Sens. 2023, 15, 2091 0.0546 (average: 1.5944) (Figure 12). This suggests that Lake Taihu is consistent with
15 of 19 the
assumption of the spatially homogeneous distribution of aerosol types.

Figure 12. Rayleigh-corrected reflectance (ρrc) in SWIR bands and its ratio (ρrc (1609)/ρrc (2201)) fro
Lake Taihu on 2 February 2016.

4.2.Figure
The 12. Rayleigh-corrected
Black
Figure 12. Pixels from Tworeflectance
Rayleigh-corrected Date Images
reflectance
(ρrc) in SWIR bands and its ratio (ρrc (1609)/ρrc (2201)) from
(ρrc ) inOverlap
SWIR bands and its ratio (ρrc (1609)/ρrc (2201)) from
Lake Taihu on 2 February 2016.
Lakem-ACZI
The Taihu on 2 February
algorithm 2016.searches and obtains the “black pixel” based on the combin

tion4.2.
of4.2.
BPI
TheThe and
Black
Black PIF for
Pixels
Pixels two
from
from date
Two
Two Dateimages,
Date Images which requires the BPI and PIF pixels of the tw
ImagesOverlap
Overlap
periodsThe toThe
overlap.
m-ACZI The
m-ACZIalgorithm PIF pixels
algorithm searches
searches areandfrom
and urban
obtains
obtains the
the construction
“black
“blackpixel” based
pixel” land, which
on the
based on combina-is invariab
the combina-
in ationtion
year. of BPI and
Therefore, PIF for two
thetwo date
PIFdate images,
pixels which
of the requires
tworequires the
periods BPI and PIF pixels of the two
of BPI and PIF for images, which theshould
BPI andoverlap.
PIF pixels The occurren
of the two
periods to overlap. The PIF pixels are from urban construction land, which is invariable
probability
periods of BPI
to overlap. pixels
Thetheis
PIFcalculated
pixels in Lake Taihu based on cloudless images
is invariable20
from
in a year. Therefore, PIF pixelsare from
of the twourban construction
periods land, The
should overlap. which
occurrence
to 2018 (a total
in aprobability
year. of 36 images)
Therefore,
of BPI theisPIF
pixels (Figure
pixels in
calculated 13),
ofLake showing
the more
two periods
Taihu based than 30%
onshould
cloudless in a large
overlap.
images areatoof centr
The 2013
from occurrence
Lakeprobability
Taihu. of BPI
This
2018 (a total of 36pixels
means images) is calculated
that there is
(Figure ainvery
13), Lake
showing Taihu
high based
probability
more on cloudless
than 30% ofaobtaining
in images
large area from 2013 B
ofoverlapping
central
to 2018
Lake (a total
Taihu. of
This36 images)
means that (Figure
there is 13),
a veryshowing
high more
probability
pixels from three images, which is sufficient to support the m-ACZI algorithm. than
of 30% in
obtaining a large area
overlapping ofBPI
central
Landsat
Lakepixels from
Taihu. three
This images,
means that which
there is sufficient
is a very to
highsupport the
probability m-ACZI
of algorithm.
obtaining Landsat
overlapping BPI
OLI has accumulated a large number of images to complete the m-ACZI algorithm.
8-OLI
pixels hasthree
from accumulated
images,awhich
large number of images
is sufficient to complete
to support the m-ACZI
the m-ACZI algorithm.
algorithm. Landsat 8-
OLI has accumulated a large number of images to complete the m-ACZI algorithm.

Figure 13.13.
Figure Occurrence
Figure Occurrence
13. probability
Occurrence probabilityof
probability of BPI pixels
BPIpixels
BPI pixels (a)
(a)
(a) and
and
and itshistogram
itsits histogram
histogram (b) (b)(b) in Lake
in Lake
in Lake Taihu
Taihu Taihu (2013–2018
(2013–2018).
(2013–2018).

4.3. The m-ACZI Algorithm Depends on Pure Pixels

The m-ACZI algorithm needs PIF pixels to identify the “black pixel” because they do
not drastically change their reflectivity properties over time. The reflectivity properties of
PIFs are derived from construction land, and mixing from other variable land types (such
as water, forest land, and agricultural land) weakens the invariant reflectivity properties of
PIFs. The theoretical basis of the m-ACZI algorithm requires minimizing the uncertainty
caused by a mixture of pixels, and the high spatial resolution can reduce the uncertainty
 4.4. The Impact of ∆t on DPIF
To analyze the influence of the ∆t on DPIF, the t was calculated using 6S
ACOLITE software (version 20211112)) based on cloudless images from 20
Remote Sens. 2023, 15, 2091 total of 36 images). In the 1609 nm band, the variation range of t was 16 of 190.9981

average t was 0.9632, the standard deviation (SD) of t was 0.0111 (Figure 14)
∆t was 0.04375 to 0.00002, the average ∆t was 0.00971, and the SD of ∆t was 0
of a mixture of pixels [35]. From this point of view, the m-ACZI algorithm depends on
2201 nm band, the variation range of t was 0.9663 to 0.8982, the average t w
pure pixels.
SD of t was 0.0165 (Figure 14), the range of ∆t was 0.06810 to 0.00003, the av
4.4. The Impact of ∆t on DPIF
0.01656, and the SD ∆t was 0.01699.PIFThe variation in t and ∆t in the 1609 nm b
To analyze the influence of the ∆t on D , the t was calculated using 6SV-LUT (from
than
ACOLITE in 2201 nm;(version
software this band wasbased
20211112)) selected as theimages
on cloudless judgmentfrom 2013 band for(aD
to 2018 total ≈ D .
PIF BPI

(the
of 36red boxInin
images). theFigure
1609 nm15a) band, near Lake range
the variation Taihu of twas taken
was 0.9981 to as an the
0.9543, example
average to illu
t was 0.9632, the standard deviation (SD) of t was 0.0111
fluence of the ∆t on D (1609). Furthermore, to ensure the independence
PIF (Figure 14), the range of ∆t was of
0.04375 to 0.00002, the average ∆t was 0.00971, and the SD of ∆t was 0.01255. In the 2201
ρnm
s was driven by the ACOLITE. To reduce the error caused by the ACOLI
band, the variation range of t was 0.9663 to 0.8982, the average t was 0.9149, the SD of t
rithm,
was 0.0165 the(Figure
average ρsPIF
14), the rangewas considered
of ∆t was 0.06810 toas the real
0.00003, ρsPIF (Figure
the average 15c). This ρ
∆t was 0.01656,
and the SD ∆t was 0.01699. The variation
from the true value, but it can still be considered in t and ∆t in the 1609 nm band was less than
as a reference for analyzing in
2201 nm; thisPIF band was selected as the judgment band for DPIF ≈ DBPI . Suzhou City (the red
ofbox∆tinon D 15a)
Figure . Since there
near Lake were
Taihu no images
was taken that completely
as an example to illustrate the corresponded
influence of the to th
value,
∆t on Dwe(1609).
PIF selected the corresponding
Furthermore, two images
to ensure the independence of the approximate
of verification, ρs was drivenaverag
by
ysis. the ACOLITE. To reduce the error
The maximum ∆t (∆tPIF= 0.04375) in the caused by the ACOLITE-AC algorithm,
1609 nm band the was
average from 2
PIF PIF
ρs was considered as the real ρs (Figure 15c). This ρs deviated from the true value,
12/29/2014
but it can stilland 05/14/2018,
be considered and the
as a reference for∆t = 0.01065
analyzing (approximate
the influence of ∆t on Daverage
PIF . Since ∆t) in

band wasno
there were from
images2 images from 12/29/2014
that completely corresponded and to the06/15/2018.
average ∆t value, As weshownselectedin Figur
the corresponding two images of the approximate
= 0.04375, the average ∆tρs was 0.0058, the median ∆tρs was 0.0055,
PIF average ∆t for analysis. The maximum
PIF
∆t (∆t = 0.04375) in the 1609 nm band was from 2 images from 12/29/2014 and 05/14/2018,
∆tρ sPIF/DPIF was 14.85%, and the median ∆tρsPIF/DPIF was 10.84%. When ∆t =
and the ∆t = 0.01065 (approximate average ∆t) in the 1609 nm band was from 2 images
average
from 12/29/2014∆tρsPIFand was 0.0014, As
06/15/2018. the median
shown ∆tρ
in Figure 15,sPIF
whenwas ∆t =0.0013,
0.04375, thetheaverage
average ∆t
∆tρs
4.62%,
PIF was
and0.0058, the median
the median ∆tρ∆tρ
PIF
s s /Dwaswas
PIF PIF 0.0055, It can∆tρ
the average
2.68%.
PIF
bes seen
PIF
/D that was 14.85%,
∆t can introd
and the median ∆tρs PIF /DPIF was 10.84%. When ∆t = 0.01065, the average ∆tρs PIF was
into the algorithm, but when ∆t is less than 0.01065 (this was a common occ
0.0014, the median ∆tρs PIF was 0.0013, the average ∆tρs PIF /DPIF was 4.62%, and the median
error
∆tρs PIFit/Dintroduced
PIF was 2.68%. was It can small
be seen and
that ∆twithin the allowable
can introduce an error intoerror range.but
the algorithm, We prop
when ∆t is less than 0.01065 (this was a common occurrence),
two images with similar t values for the m-ACZI algorithm, and this t valu the error it introduced was
small and within the allowable error range. We propose selecting two images with similar t
estimated by the LUT.
values for the m-ACZI algorithm, and this t value can be pre-estimated by the LUT.

Figure 14. Variations in the transmittance (t) in SWIR bands (based on cloudless images from 2013 to
2018, a total of 36 images).
Remote Sens. 2023, 15, 2091 17 of 19

Remote Sens. 2023, 15, 2091 Figure 14. Variations in the transmittance (t) in SWIR bands (based on cloudless
17 ofimages
19

from 2013 to 2018, a total of 36 images).

Figure 15. Impact of ∆t (means different between t from two date images) on DPIF (1609) (taking
Figure 15. Impact of ∆t (means different between t from two date images) on DPIF (1609) (taking
Suzhou City (red box) near Lake Taihu as an example). (b) RGB image of Suzhou City (red box
Suzhou City (red box) PIF near Lake Taihu as an example). (b) RGB image of Suzhou City (red box in
in (a)). The average ρs (1609) (c) was driven by ACOLITE based on cloudless images from 2013
(a)). The average ρsPIF (1609) (c) wasPIF driven by ACOLITE based on cloudless images from 2013 to
to 2018 (total of 36 images). ThePIFD is the difference in ρrc (SWIR) for two dates images; the
2018 (total of 36 images). The D is the difference in ρrc (SWIR) for two dates PIFimages; the ∆tρsPIF
∆tρs PIF represents the residual surface reflectance caused by ∆t in DPIF , and ∆tρ s /DPIF is the
represents thePIF residual surface reflectance caused by ∆tPIF in D , and ∆tρs /D is the ratio of ∆tρsPIF
PIF PIF PIF
ratio of ∆tρs to DPIF , which represents the error of D caused by ∆t. These two ∆t values
to D , which represents the error of D caused by ∆t. These two ∆t values were the maximum ∆t
PIF PIF
were the maximum ∆t (0.0438) and ∆t = 0.0107 (approximate average ∆t) from 36 images (2013 to
(0.0438) and ∆t = 0.0107 (approximate average ∆t) from 36 images (2013 to 2018), respectively.
2018), respectively.

5. 5.
Conclusions
Conclusions
TheThe“black
“blackpixel”
pixel” assumption
assumptionplaysplaysa key
a keyrole in providing
role in providing atmospheric (especially
atmospheric (especially
aerosol) scattering information to solve the atmospheric correction process
aerosol) scattering information to solve the atmospheric correction process for water. for water. We We
modified the identification method for the “black pixel” in the ACZI algorithm as a combi-
modified the identification method for the “black pixel” in the ACZI algorithm as a com-
nation of BPI and PIF and provided the m-ACZI algorithm. The m-ACZI algorithm assumes
bination of BPI and PIF and provided the m-ACZI algorithm. The m-ACZI algorithm as-
the spatial homogeneity of aerosol types and employs BPI and PIF to identify the “black
sumes
pixel”,the spatial
which homogeneity
together of aerosolsuitable
make this algorithm types and employs
for inland waterBPI andasPIF
(such to identify
lakes, rivers, the
“black pixel”, which
and reservoirs). together
Compared make
with this algorithm
the ACZI algorithm,suitable
m-ACZIfor hadinland
better water (such as
performance in lakes,
rivers,
the visible bands, especially in the 561 nm and 655 nm ones. The evaluation results showedperfor-
and reservoirs). Compared with the ACZI algorithm, m-ACZI had better
mance inperformance
that the the visible ofbands, especially
the m-ACZI in the
algorithm 561 nmtoand
is superior the 655 nm ones. ACZI,
SeaDAS-SWIR, The evaluation
and
EXP algorithms
results showed thatin waters with algal blooms.
the performance of theThis algorithm
m-ACZI is a useful
algorithm attempt totoimprove
is superior the SeaDAS-
SWIR, ACZI, and EXP algorithms in waters with algal blooms. This algorithm is a useful
Remote Sens. 2023, 15, 2091 18 of 19

the accuracy of AC for turbid water and provides a new scheme for obtaining black pixels
from turbid inland waters.

Author Contributions: Conceptualization, D.W. and R.M.; methodology, D.W.; validation, D.W.,
X.X. and W.Z.; formal analysis, D.W.; investigation, D.W.; resources, R.M.; writing—original draft
preparation, D.W.; writing—review and editing, D.W., R.M., X.X. and Z.W.; funding acquisition, D.W.,
Y.G. and R.M. All authors have read and agreed to the published version of the manuscript.
Funding: This research was funded by the Project of Education of Zhejiang province (Grant No. Y202250625),
the National Natural Science Foundation of China (Grant No. 42106019), and the Open Foundation
from Marine Sciences in the First-Class Subjects of Zhejiang (Grant No. OFMS002). The work was
partly supported by the National Earth System Science Data Center, the 14th Five-year Network
Security and Informatization Plan of Chinese Academy of Sciences (Grant No. CAS-WX2021SF-0306),
and the Scientific Data Center of Nanjing Institute of Geography & Limnology Chinese Academy of
Sciences (CAS-WX2022SDC-SJ05).
Data Availability Statement: Not Applicable.
Acknowledgments: We thank the study participants from NIGLAS (Zhigang Cao, Ming Shen, Tianci
Qi and Junfeng Xiong, and Xu Fang). Thanks to USGS for the OLI data and REMSEM for the
ACOLITE software. We also acknowledge the data support from Lake-Watershed Science Data
Center, National Earth System Science Data Sharing Infrastructure, National Science & Technology
Infrastructure of China (http://lake.geodata.cn) (accessed in April 2019).
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Hu, C.; Carder, K.L.; Muller-Karger, F.E. Atmospheric Correction of SeaWiFS Imagery over Turbid Coastal Waters. Remote Sens.
Environ. 2000, 74, 195–206. [CrossRef]
2. Ruddick, K.G.; Ovidio, F.; Rijkeboer, M. Atmospheric correction of SeaWiFS imagery for turbid coastal and inland waters. Appl.
Opt. 2000, 39, 897–912. [CrossRef] [PubMed]
3. He, Q.; Chen, C. A new approach for atmospheric correction of MODIS imagery in turbid coastal waters: A case study for the
Pearl River Estuary. Remote Sens. Lett. 2014, 5, 249–257. [CrossRef]
4. Wang, J.; Lee, Z.; Wang, D.; Shang, S.; Wei, J.; Gilerson, A. Atmospheric correction over coastal waters with aerosol properties
constrained by multi-pixel observations. Remote Sens. Environ. 2021, 265, 112633. [CrossRef]
5. IOCCG. Atmospheric Correction for Remotely-Sensed OceanColour Products; International Ocean Colour Coordinating Group:
Dartmouth, NS, Canada, 2010.
6. Wang, M.; Gordon, H.R. A Simple, Moderately Accurate, Atmospheric correction algorithn for SeaWiFS. Remote Sens. Environ.
1994, 50, 231–239. [CrossRef]
7. Gordon, H.R.; Menghua, W. Influence of oceanic whitecaps on atmospheric correction of ocean-color sensors. Appl. Opt. 1994, 33,
7754–7763. [CrossRef]
8. Gordon, H.R.; Clark, D.K. Clear Water Radiance for Atmospheric Correction of Coastal Zone Color Scanner Imegery. Appl. Opt.
1981, 20, 4175–4226. [CrossRef]
9. Wang, M.; Shi, W. The NIR-SWIR combined atmospheric correction approach for MODIS ocean color data processing. Opt.
Express 2007, 24, 15722–15733. [CrossRef]
10. Wang, M.; Shi, W. Estimation of ocean contribution at the MODIS near-infrared wavelengths along the east coast of the U.S.: Two
case studies. Geophys. Res. Lett. 2005, 32, L13606. [CrossRef]
11. Wang, M. Remote sensing of the ocean contributions from ultraviolet to near-infrared using the shortwave infrared bands:
Simulations. Appl. Opt. 2007, 46, 1535–1547. [CrossRef]
12. Wang, D.; Ma, R.; Xue, K.; Li, J. Improved atmospheric correction algorithm for Landsat 8–OLI data in turbid waters: A case
study for the Lake Taihu, China. Opt. Express 2019, 27, A1400–A1418. [CrossRef]
13. Shi, W.; Wang, M. An assessment of the black ocean pixel assumption for MODIS SWIR bands. Remote Sens. Environ. 2009, 113,
1587–1597. [CrossRef]
14. Vanhellemont, Q.; Ruddick, K. Turbid wakes associated with offshore wind turbines observed with Landsat 8. Remote Sens.
Environ. 2014, 145, 105–115. [CrossRef]
15. Wang, D.; Ma, R.; Xue, K.; Loiselle, S. The Assessment of Landsat-8 OLI Atmospheric Correction Algorithms for Inland Waters.
Remote Sens. 2019, 11, 169. [CrossRef]
16. Concha, J.A.; Schott, J.R. Retrieval of color producing agents in Case 2 waters using Landsat 8. Remote Sens. Environ. 2016, 185,
95–107. [CrossRef]
Remote Sens. 2023, 15, 2091 19 of 19

17. Schott, J.R.; Salvaggio, C.; Volchok, W.J. Radiometric scene normalization using pseudoinvariant features. Remote Sens. Environ.
1988, 26, 1–16. [CrossRef]
18. Hu, C.; Lee, Z.; Ma, R.; Yu, K.; Li, D.; Shang, S. Moderate Resolution Imaging Spectroradiometer (MODIS) observations of
cyanobacteria blooms in Taihu Lake, China. J. Geophys. Res. 2010, 115, C04002. [CrossRef]
19. Hu, C. A novel ocean color index to detect floating algae in the global oceans. Remote Sens. Environ. 2009, 113, 2118–2129.
[CrossRef]
20. Shen, M.; Duan, H.; Cao, Z.; Xue, K.; Qi, T.; Ma, J.; Liu, D.; Song, K.; Huang, C.; Song, X. Sentinel-3 OLCI observations of water
clarity in large lakes in eastern China: Implications for SDG 6.3.2 evaluation. Remote Sens. Environ. 2020, 247, 111950. [CrossRef]
21. Xue, K.; Ma, R.; Shen, M.; Li, Y.; Duan, H.; Cao, Z.; Wang, D.; Xiong, J. Variations of suspended particulate concentration and
composition in Chinese lakes observed from Sentinel-3A OLCI images. Sci. Total Environ. 2020, 721, 137774. [CrossRef]
22. Cao, Z.; Duan, H.; Feng, L.; Ma, R.; Xue, K. Climate- and human-induced changes in suspended particulate matter over Lake
Hongze on short and long timescales. Remote Sens. Environ. 2017, 192, 98–113. [CrossRef]
23. Feng, L.; Hu, C.; Chen, X.; Tian, L.; Chen, L. Human induced turbidity changes in Poyang Lake between 2000 and 2010:
Observations from MODIS. J. Geophys. Res. Ocean. 2012, 117, C07006. [CrossRef]
24. Mueller, J.L.; Fargion, G.S.; Charles, R.; McClain, C.R. Biogeochemical and bio-optical measurements and data analysis protocols.
In Ocean Optics Protocols for Satellite Ocean Color Sensor Validation, Revision 5; Tech, N., Ed.; NASA Tech: Washington, DC, USA,
2003; Volume V.
25. Mobley, C.D. Estimation of the remote-sensing reflectance from above-surface measurements. Appl. Opt. 1999, 38, 7442–7455.
26. Vanhellemont, Q.; Ruddick, K. Atmospheric correction of metre-scale optical satellite data for inland and coastal water applica-
tions. Remote Sens. Environ. 2018, 216, 586–597. [CrossRef]
27. Vermote, E.F.; Member, I.; Tan, D.; Deuze, J.L.; Herman, M. Second Simulation of the Satellite Signal in the Solar Spectrum, 6s: An
Overview. IEEE Trans. Geosci. Remote Sens. 1997, 35, 675–687. [CrossRef]
28. Vanhellemont, Q.; Ruddick, K. Advantages of high quality SWIR bands for ocean colour processing: Examples from Landsat-8.
Remote Sens. Environ. 2015, 161, 89–106. [CrossRef]
29. He, X.; Bai, Y.; Pan, D.; Tang, J.; Wang, D. Atmospheric correction of satellite ocean color imagery using the ultraviolet wavelength
for highly turbid waters. Opt. Express 2012, 20, 20754–20770. [CrossRef]
30. Hale, G.M.; Querry, M.R. Optical constants of water in the 200 nm to 200 µm wavelength region. Appl. Opt. 1973, 12, 555–563.
[CrossRef]
31. Vanhellemont, Q. Adaptation of the dark spectrum fitting atmospheric correction for aquatic applications of the Landsat and
Sentinel-2 archives. Remote Sens. Environ. 2019, 225, 175–192. [CrossRef]
32. Vanhellemont, Q.; Ruddick, K. Atmospheric correction of Sentinel-3/OLCI data for mapping of suspended particulate matter
and chlorophyll-a concentration in Belgian turbid coastal waters. Remote Sens. Environ. 2021, 256, 112284. [CrossRef]
33. Vanhellemont, Q. ACOLITE processing for Sentinel-2 and Landsat-8 atmospheric correction and aquatic applications. In
Proceedings of the 2016 Ocean Optics Conference, Victoria, BC, Canada, 23–28 October 2016.
34. Sun, D.; Li, Y.; Wang, Q.; Le, C.; Lv, H.; Huang, C.; Gong, S. Specific inherent optical quantities of complex turbid inland waters,
from the perspective of water classification. Photochem. Photobiol Sci. 2012, 11, 1299–1312. [CrossRef] [PubMed]
35. IOCCG. Uncertainties in Ocean Colour Remote Sensing; International Ocean Colour Coordinating Group: Dartmouth, NS,
Canada, 2019.

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.