Atmospheric Correction of Satellite Ocean
Color Imagery: The Black Pixel Assumption
David A. Siegel1,2, Menghua Wang3,
Stéphane Maritorena1 and Wayne Robinson4
1Institute
for Computational Earth System Science,
University of California, Santa Barbara, Santa Barbara, CA, 93106-3060, U.S.A.
2Also,
Department of Geography and
Donald Bren School of Environmental Science and Management,
University of California, Santa Barbara
3University
of Maryland Baltimore County
NASA Goddard Space Flight Center, Greenbelt, MD
4NASA
Goddard Space Flight Center, Greenbelt, MD
To be submitted to Applied Optics
Draft date = September 2, 1999
Acknowledgments
We thank Andrea Magnuson and Larry Harding for providing SeaWiFS imagery and field data
from Chesapeake Bay and Brian Schieber for the SIMBIOS matchup data. Bob Arnone and Rick
Stumpf first questioned the validity of the black pixel assumption and their encouragement is appreciated. Discussions and encouragement from Chuck McClain, Howard Gordon and Andre
Morel are gratefully acknowledged. We acknowledge the support of NASA as part of the
SIMBIOS and SeaWiFS science teams. Chlorophyll observations reported are from the Chesapeake Bay Program which is jointly funded by the EPA and the state governments bordering
Chesapeake Bay. The SeaWiFS satellite mission is a joint venture of the Orbital Sciences Corporation and NASA.
ABSTRACT
The correction of satellite ocean color imagery generally assumes that the water leaving radiance is negligible for the near-infrared (NIR) wavebands. This assumption is referred to as the
black pixel assumption. The black pixel assumption enables NIR aerosol radiances to be determined from a measure of satellite radiance and an estimate of the Rayleigh component alone. In
this contribution, we examine the role of the black pixel assumption on the correction of satellite
ocean color imagery theoretically and using satellite imagery from the Sea-viewing Wide-Field
of view Sensor (SeaWiFS). A simple bio-optical model for the NIR water-leaving reflectance,
wN(NIR), is developed. For reasonably high values of the chlorophyll concentration (Chl > 5 mg
m-3), estimates of wN(NIR) are several orders of magnitude larger than are expected for pure
seawater demonstrating that the ocean is not necessarily black for the NIR wavebands. The misapplication of the black pixel assumption results in an over-correction of the water-leaving reflectance in the visible which is most pronounced in the violet and blue wavebands of the SeaWiFS sensor. The over-correction increases dramatically with the chlorophyll concentration
reaching ~75% of the true water-leaving radiance when Chl is equal to 5 mg m-3. Accounting for
wN(NIR) in the processing of SeaWiFS imagery results in a significant increase in the waterleaving reflectance for the violet and blue bands and a reduction of chlorophyll retrievals when
values of Chl are in excess of roughly 2 mg m-3. Substantial and important improvements are
found for natural waters where Chl values are in excess of 5 mg m-3 and examples are provided.
In general, non-zero values of wN(NIR) influence atmospheric correction results for reasonably
high values of chlorophyll and hence, this process needs to be included in corrections schemes
for global satellite missions.
1.
Introduction
It is well recognized that in the visible more than 90% of the optical signal measured
by an ocean color satellite sensor is due to the confounding influence of the atmosphere.
The atmospheric and ocean surface effects must be removed before ocean radiance signals may be analyzed for the purposes of understanding the ocean biosphere. This step
in the processing of satellite ocean color imagery is usually termed as atmospheric correction. An excellent review of the state of the art in atmospheric correction for satellite
ocean color imagery may be found in Gordon (1997).
The satellite sensed radiance, Lt(), or equivalently reflectance, t() (=
Lt()/(Fo()o) where Fo() is the extraterrestrial solar irradiance and 0 is the cosine of
the solar zenith angle; Gordon and Wang, 1994b) can be partitioned into components corresponding to distinct physical process, or
t   r    a    ra   T g    t wc  t  w  ,
(1)
where the first three terms represent the contributions from atmospheric scattering due to
air molecules (Rayleigh), aerosols, and Rayleigh-aerosol interactions, respectively, T
and t are the direct and diffuse transmittances of the atmospheric column, respectively,
g represents the effects of sun glitter off the sea surface, wc is the reflectance of
ocean whitecaps and w is the water-leaving reflectance. The desired quantity in ocean
color remote sensing is w. In equation (1), the Rayleigh scattering term, r, and
the transmittances, T and t, can be accurately calculated (Gordon et al., 1988a; Gordon and Wang, 1992; Wang, 1999a; Yang and Gordon, 1997), the ocean whitecap contributions can be estimated from the surface wind speed (Gordon and Wang, 1994a; Frouin
et al. 1996), and sun glitter contaminated observations are not used. This leaves two unknowns in equation (1): (i) contributions from multiple scattering of aerosols and Rayleigh-aerosol interactions a    ra   and (ii) signals from ocean water w. The
solution to this problem is made by assuming that the water leaving radiance is zero for
1
2
two of the red or near-infrared wavelengths and solving for the aerosol scattering terms
for these bands. Estimates of a    ra   for the other (visible) bands is made by extrapolating the near-infrared aerosol signals into the visible using appropriate aerosol
models (Gordon and Wang, 1994b). The assumption that the ocean is optically black
(w(NIR) =0) in the red and near-infrared was initially made for clear ocean waters (Gordon and Clark, 1981) and is often referred to as the black pixel assumption.
To relate the derived water-leaving reflectance to the oceanic inherent optical properties, the residual atmospheric effects in w must be eliminated. The normalized waterleaving reflectance, w  N , can be defined from Gordon and Clark (Gordon and Clark,
1981) through the definition of the normalized water-leaving radiance LwN , i.e.,
L 
w
N
 Lw   0 t0  
(2)
and
  
w
N


L N  w   t 0 ,
F0   w
(3)
where t 0   is the atmospheric diffuse transmittance in the solar direction. Ocean optical
properties of the near-surface layer can then be determined using either an empirical band
ratio algorithm with w  N or LwN (e.g., Gordon and Morel, 1983; Gordon et al.,
1988b; Morel, 1988; O'Reilly et al., 1998) or an analytical model relating the water leaving reflectance spectrum to the ocean optical properties (e.g., Roesler and Perry, 1996;
Garver and Siegel, 1997; Carder et al. 1999).
The Sea-viewing Wide Field-of-view Sensor (SeaWiFS) has provided the oceanographic community an unprecedented opportunity to assess biological and biogeochemical properties over global scales with a spatial resolution of 1.1 km (at nadir) and a temporal sampling schedule of nearly once per day (McClain et al. 1998). As with all satellite ocean color sensors, an atmospheric correction of the satellite-sensed radiance needs
to be performed before useful ocean signals can be obtained. The SeaWiFS atmospheric
3
correction assumes that the bands 7 and 8 (center at 765 and 865 nm, respectively) are
“black pixels” and used to estimate aerosol radiances and select the appropriate aerosol
optical model to be applied (Gordon and Wang, 1994b). For the clear ocean (generally
open oceans), the atmospheric correction algorithm is usually accurate to within 5% in
the blue. Menghua, We need a good reference here. Are there “real” observations to
support this? Or, just model results?
Unfortunately, the SeaWiFS determinations of water leaving radiance in the violet
and blue (SeaWiFS bands 1 and 2) generally underestimate dramatically field data for
ocean regions with high chlorophyll concentrations (low ocean LwN()). This can be seen
in a comparison of nearly simultaneous match-ups of SeaWiFS and field observations of
LwN() at 412, 443 and 490 nm and chlorophyll a concentration (figure 1). These data are
provided from the Sensor Intercomparison and Merger for Biological and Interdisciplinary Oceanic Studies (SIMBIOS) project of NASA Goddard Space Flight Center. Details
of the procedures used in developing these satellite-field data comparisons may be found
Schieber et al. (Tech memo ref???). Obviously, the present version of SeaWiFS processing (version 2) underestimates determinations of LwN(412) when values of LwN(412)
are small (less than ~0.3 W cm-2 nm-1 sr-1; figure 1a). A similar though less pronounced
underestimation is found for LwN(443). However, the comparison of the field and satellite
chlorophyll estimates is good with no real bias from the one-to-one line (figure 1d). With
the exception of the lowest and the highest chlorophyll retrievals, the present version of
the SeaWiFS data provides excellent chlorophyll concentration retrievals. Overall, 81%
of the observed variance is explained in the measured vs. SeaWiFS chlorophyll concentration comparison (figure 1d). Very low or negative Lwn() at wavelengths under 500
nm can also be seen in SeaWiFS imagery for some coastal and semi-enclosed water
masses. These low retrievals lead to large, seemingly unreasonable, values of chlorophyll
concentration.
4
Underestimation of the water-leaving radiance within the blue to green wavebands results from an incomplete atmospheric correction. Several factors can be responsible for
these failures, such as the presence of absorbing aerosols or the choice of an inadequate
atmospheric model. However, the fact that these low SeaWiFS water-leaving radiances
occur in areas characterized by highly productive waters and generally small influence by
atmospheric dust suggests that the origin of many of the low radiometric retrievals is due
to the ocean itself. We believe that the often inappropriate application of the black pixel
assumption is responsible for some of the excessively high chlorophyll observations made
from SeaWiFS and for some of the observed underestimates of water-leaving radiance for
the blue wavebands. This realization was made first by Robert Arnone and colleagues for
U.S. coastal waters (Arnone et al in Ocean Optics abstracts) . Here, we will demonstrate
the implications of the black pixel assumption and will develop and apply a parameterization to correct for its effects. First, we will use a recent ocean optics climatology (SeaBAM) to address the magnitude and variability of wN(NIR) and to develop a wN(NIR)
algorithm for satellite atmospheric correction. We will assess theoretically the implications of non-blackness of the NIR channels on estimates of water-leaving radiance in the
visible and provide a correction scheme for SeaWiFS imagery. Last, we will demonstrate
the effects of relaxing the black pixel assumption on local to global scales using SeaWiFS
data.
2.
Estimation of Ocean Contributions at the NIR Bands
A robust literature of in situ radiometric data has developed over the past twenty-five
years due to the need for calibration and validation of satellite ocean color data (e.g.,
Gordon and Clark, 1981; Gordon and Morel, 1983; O’Reilly et al. 1998). Unfortunately,
data were almost exclusively collected in the visible spectrum and, to the best of our
knowledge, there are no direct estimates of wN(NIR) available. This is because it is extremely difficult to make reliable radiometric measurements in the near-infrared (e.g.,
5
Gordon and Ding, 1992). Hence, the signal in the NIR wavebands can only be estimated
using optical models of water-leaving radiance based upon the inherent optical properties
of the medium.
Following Gordon et al. (1988b), values of [w()]N may be modeled as a function of
the absorption (a()) and backscattering spectra (bb()) coefficients of the upper layers,
or
2
[w()]N = (t/n)2

i=1
 bb ( ) i

gi 
bb ( )  a() 
(4)
where (t/n)2 accounts for the transmission of upwelling radiance and downwelling irradiance across the sea surface (Austin, 1974, Gordon et al., 1988b, Morel and Antoine,
1994) and the constants g1 and g2 are given as 0.0949 sr-1 and 0.0794 sr-1, respectively
(Gordon et al. 1988b). This formulation assumes that neither non-linear scattering processes (Raman scattering, fluorescence, etc.) nor the bidirectional reflectance distribution
function (BRDF) from a Lambertian surface are important in determining the values of
[w()]N.
In the near-infrared region, the absorption by seawater strongly dominates over other
factors (Hale and Querry, 1973; Palmer and Williams, 1974; Smith and Baker, 1981;
Mobley, 1994). Hence, a(NIR) may be approximated by the value of the absorption coefficient for pure water, aw(NIR). This assumption will make a negligible error in the derived values of [w(NIR)]N. In the case of an atmospheric correction scheme that would
use SeaWiFS band 6 and 8 (670 and 865 nm, respectively) instead of the current band 7
and 8, an accounting for particulate-induced absorption in required and several good algorithms are available (e.g., Bricaud et al. 1998). On the other hand, the modeling of the
backscattering coefficient is more problematic. In the red and near-infrared spectral region, values of bb(NIR) due to particulates is much larger than those due to pure seawater
6
(e.g., Gordon and Morel, 1983). Hence, a detailed knowledge of particulate backscattering coefficient, bbp(NIR), in this spectral range is required.
Provided a model for
bbp(NIR), estimates of NIR normalized water-leaving reflectance may be expressed as
2
[w(NIR)]N (t/n)2

i=1

i
bbp ( )  bbw ( )

gi 
 b ( )  b ( )  a ( ) 
bp
bw
w


(5)
where the necessary parameters are presented in Table 1.
The difficult issue is to model bbp(NIR) as there has been little significant research.
Here, we will compare two basic approaches. The first uses estimates of the chlorophyll
a concentration and a bio-optical assumption to determine bbp(NIR) (e.g., Gordon and
Morel, 1983; Morel, 1988; Morel and Maritorena, in prep.) while the second uses determinations of water leaving radiance and the assumption of optical closure (Carder et al.,
1999). Both are empirical relationships derived from field data within the visible spectral
region and are extrapolated into the NIR for present purposes.
The bio-optical modeling of bbp(NIR) assumes that its variability is driven by the
chlorophyll a content of the water (e.g., Gordon and Morel, 1983; Morel, 1988). For the
present purposes, bio-optical estimates of bbpBO(NIR) are made using
bbpBO(NIR) = 0.416 Chl0.766 (0.002 + (550/NIR) (0.02 (0.5 - 0.25 log10(Chl))))
(6)
where Chl is the chlorophyll concentration (in mg m-3) and NIR is the center NIR wavelength of interest. The left term of equation 6 gives the particulate scattering coefficient
at 550 as recently updates in Loisel and Morel (1998) while the right term in parentheses
gives the spectral dependance and the magnitude of the backscattered fraction (Morel,
1988). This formulation assumes that the spectral dependence for bbpBO(NIR) goes as -1
throughout the entire spectral range. Other similar bio-optical algorithms are available
7
for bbpBO(NIR) (e.g., Gordon et al., 1988; Morel, 1987; Mobley, 1994; Morel and Maritorena, in prep.) which give broadly similar results (within a factor of 5 for [w(NIR)]N).
The optical closure backscatter model, bbpOC(NIR), is derived from field-derived estimates of bbp() (e.g., Carder et al. 1999). This parameterization assumes that the magnitude of spectral backscatter is a linear function of the remote sensing reflectance at 551
nm, Rrs(551), while the spectral slope of particulate backscatter is a function of the ratio
of Rrs(443) to Rrs(488), or
551 Y0 Y1 R rs (443) R rs (488)

bbpOC(NIR) = (X0 + X1 Rrs(551))
  
(7)
where X0 = -0.00182, X1 = 2.058, Y0 = -1.13, and Y1 = 2.57 (Carder et al., 1999). The ratio of Rrs(443) to Rrs(488) is large in blue waters and smaller in green and turbid waters.
Hence, the spectral slope for bbp() varies from 0 for turbid waters to greater than 2 for
clear, oligotrophic waters (e.g., Morel et al. 1989; Stramski and Kiefer, 1991; Morel and
Maritorena, in prep.). The intensity of the backscatter is controlled by the value of
Rrs(551). This spectral band is typically in spectral "hole" or “hinge point” in the water
leaving radiance spectrum because of the generally low absorption by the various components in the waters at this wavelength. Consequently, to first order, the effects of
backscattering regulate ocean color variations in this portion of the spectrum (e.g., Gordon and Morel, 1983). Similar results were found using the semi-analytical, reflectance
inversion model of Garver and Siegel [1997].
The two alternative estimates of bbp() are compared with each other using the SeaBAM data set (O'Reilly et al 1998) and are plotted against the in situ chlorophyll a concentration in figure 2a for bbp(443) and in figure 2b for bbp(865). Statistical estimates of
mean and standard deviation for the red and NIR wavebands are also presented in Table
2. Compared to the bio-optical determinations of bbp(443) and bbp(865), the closurebased estimates appear to be a much weaker function of Chl, paricularly at 443 nm. For
the observations represented in the SeaBAM data set, mean values of bbp(443) and
8
bbp(865) are smaller for the bio-optical estimates than with the other method. The closure-based backscatter formulation predicts a large degree of variability about its mean
relationship with chlorophyll likely to be partly caused by the inherent noise in the in situ
data. Some of these estimates approach values of the backscattering coefficient for pure
seawater (figure 2a and 2b).
The two methods for determining bbp() are used to provide scaling estimates for
[w(NIR)]N using the SeaBAM data set. As expected, an increasing trend in [w(NIR)]N is
found with increasing chlorophyll a concentration (fig. 2cd). Values of [w(NIR)]N increase with chlorophyll concentration from 10-5 for Chl  0.5 mg m-3 to about 10-4 for
Chl concentrations greater than 1 mg m-3. Both estimates of [w(NIR)]N are an order of
magnitude greater than expected for pure seawater (where bbp(NIR) = 0; the horizontal
line in figures 2c and 2d). Estimates of [w(NIR)]N found using the bio-optical bbp() algorithm are generally greater than the optical closure model although there is a great deal
of scatter about that relationship (fig. 2 cd).
The choice of the appropriate [w(NIR)]N parameterization is not-straight-forward as
there are few, if any, observations from which to base a comprehensive parameterization
upon. Obviously, this is an area for future research. The bio-optical model approach has
already been applied extensively within the ocean optics community (e.g., Morel, 1994;
Bricaud et al, 1998, Ciotti et al.1999) whereas the closure model has been recently introduced and has not been widely applied. The optical closure model relies on reflectance
measurements that are often “noisy” at high chlorophyll concentrations. Further, the
large degree of variability in the closure model, especially those low estimates which approach values of bbw(), raise questions of its applicability as a global model. Hence, we
will use the bio-optical model (eq. 6) for further analyses remembering that there are
many uncertainties inherent in the present [w(NIR)]N estimates. Some of these will be
discussed in the later in this manuscript. Detailed observations and analyses are required
before many of these uncertainties will be better constrained.
9
3.
Effects of w  N at the NIR Bands on the Atmospheric Corrections
In this section, we address the importance of relaxing the black pixel assumption by
evaluating its role in determinations of w  N and on two-band ratio determinations of
chlorophyll. Finally, we provide an iteration scheme to correct the effects of the ocean
NIR w  N for implementation in the processing of SeaWiFS ocean color imagery.
3A. Errors in w  N Retrievals
We evaluate the role of non-black NIR reflectance using the SeaWiFS atmospheric
correction algorithm (Gordon and Wang, 1994b). A maritime aerosol model with a relative humidity (RH) of 80% (M80) is used for the present evaluation. The effects of sun
glitter and white cap reflectance are not considered.
The primary experiments are performed by comparing retrievals with a fully black
ocean (w  N = 0 for all ) to retrievals made for a black ocean in the visible spectral
region (w  N = 0 for 400 <  < 700 nm) but whose reflectance in the NIR,
[w(NIR)]N, is a known function of Chl (given by equations 5 and 6). For the two cases,
top of the atmosphere (TOA) reflectance spectra, t  , are determined and then used
with the present version of the SeaWiFS atmospheric correction algorithm to provide retrievals of w  N with and without the effects of a non-black ocean in the NIR. Differences between the two retrieved w  N spectra,  w  N , may be due to inherent
errors in the algorithm itself and the effects of non-zero NIR ocean reflectance. Inherent
errors in the SeaWiFS atmospheric correction algorithm are small (typically 0.001 in
normalized reflectance units; Gordon, 1997; Gordon and Wang, 1994b; Wang, 1999b).
Inherent errors are calculated by propagating a zero w  N spectrum to the top of the
atmosphere and then estimating w  N using the standard SeaWiFS atmospheric correction procedure. The difference between the true and retrieved w  N determines the
inherent error. This small factor is subtracted from the results of the NIR experiments to

 NIR
determine the error due to NIR effects alone,  w
.
10

 NIR
Example  w
 spectra clearly illustrate the effects of non-zero NIR water re-
flectance on retrieved normalized water-leaving reflectances (figure 3). This example is
determined using the M80 aerosol model, an aerosol optical thickness at 865 nm of 0.1
and chlorophyll concentrations ranging from 0.1 to 5 mg m-3. Solar and viewing geometries of  0 = 20°,  = 20°,   = 90° (figure 3a and 3c) and  0 = 40°,  = 40°,   = 90°

 NIR
(figure 3b and 3d) were used. The magnitude of the error term,  w
N, is shown
in figures 3a and 3b for the two geometries, while these error estimates are normalized in
figures 3c and 3d using the bio-optical model of Gordon et. al. (1988b). Similar results
were obtained for other aerosol models, for various aerosol optical thicknesses, and for
other solar and viewing geometries (not shown).

 NIR
In general, values of  w
 increase dramatically with the chlorophyll concen-
tration and this effect is more important for the blue wavebands (figure 3). Ignoring the
NIR ocean contributions leads to an over-correction of aerosol reflectance biasing
   retrievals low in the visible. The effects become important (>10% of the retrieved    ) for Chl > 0.5 mg m-3. Hence, for clear ocean regions (where Chl < 0.5
w
N
w
N
mg m-3), the SeaWiFS correction algorithm performs well. However, errors are quite
large for ocean regions with high chlorophyll concentrations (Chl > 0.5 mg m-3) and nonblack NIR reflectance effects must be considered.
3B. Errors in Two-Band Ratio Chlorophyll Retrievals
Many algorithms for determining ocean chlorophyll concentrations typically use ratios of normalized water-leaving reflectance (e.g., Gordon and Morel, 1983; O’Reilly et al.,
1998). For example, the SeaWiFS chlorophyll algorithm (OC2v2) uses the ratio of SeaWiFS bands 3 and 5 (R(3,5) = [w(490)]N / [w(555)]N) in a polynomial relationship
(O’Reilly et al., 1998 and later update). Other band ratios are often used. Here, we calculate the error in the retrieved band ratio estimates due to the black pixel assumption. We
 NIR
define the error for any arbitrary band ratio, R
i, j  , as
11
   b gbg

bg
b g     b g
di
di
w i
R bNIR gi , j 
w
j
N
N
NIR
w
i
NIR
w
j
N
N

bg
 d
 i
w i
w
j
N
(8)
N
where i and j are the SeaWiFS band numbers. In the following, estimates of R NIRi, j 
are determined for a given chlorophyll concentration using the previous determinations of




 wNIR  and values of w i N and w  j 
N
estimated from Gordon et al.
[1988b].
Typical values of R NIRi, j  in the retrieved ratio values between SeaWiFS bands 2
and 5, R(2,5), and bands 3 and 5, R(3,5), are shown in Table 4. As before, the M80 aerosol model with aerosol optical thickness of 0.1 at 865 nm and the two solar and viewing
geometries are used.
The present results show that for chlorophyll concentrations less
than 1 mg m-3, the two-band ratio values are accurate to within 1-2% even if the NIR
ocean contribution is ignored. For higher chlorophyll environments (Chl > 2 mg m-3), the
band ratio error magnitude increases dramatically (Table 4). Further, the magnitude of
the band ratio under-estimation for high chlorophyll concentrations is much greater for
R(2,5) than for R(3,5). However for clear ocean conditions, empirical models based upon
R(2,5) may be better, from the point of view of atmospheric correction, than R(3,5) models as the NIR correction will have less of an effect on the band ratio estimation. This is
because although the normalized errors are similar, the magnitude of R(2,5) is significantly larger than R(3,5) in clear natural waters.
We compare the effects of the black pixel assumption using the standard SeaWiFS algorithm and another polynomial band ratio algorithm using R(2, 5) (Morel-3 algorithm in
O’Reilly et al., 1998). The errors in chlorophyll retrievals for a non-black, NIR waterleaving radiance are not large (< 5%) for low chlorophyll containing waters (< 0.5 mg m3;
Table 5).
However, for chlorophyll concentrations greater than 2 mg m-3, the errors
are extremely large reaching more than 100%. The errors due to the NIR ocean contribution are greater for the Morel-3 algorithm than for the OC2v2 algorithm as Morel-3 rela-
12
tionship uses the R(2,5) ratio. In addition, the errors in the Morel-3 empirical relationship
become important at significantly lower chlorophyll concentrations than is found for the
OC2v2 algorithm (Table 5). This supports the notion that the NIR induced errors will be
greater for the R(2,5) based chlorophyll retrievals than with R(3,5) estimates (such as is
employed in OC2v2). Similar results were obtained with different aerosol models, aerosol optical thicknesses and solar and viewing geometries. We conclude that the NIR
ocean contribution must be included in the atmospheric correction schemes for high chlorophyll conditions and that its effects are not important for clear natural waters.


3C. Accounting for wNIR

N
in the SeaWiFS Atmospheric Correction Procedure

An accounting of NIR wNIR
N
effects in atmospheric correction procedures for
satellite ocean color imagery is critical, particularly for high chlorophyll concentration
conditions. However, this requires a retrieval of the chlorophyll concentration that may
be in error due to NIR effects. This means that a NIR correction procedure must be iterative. A NIR correction procedure needs to make a first guess for Chl, calculate an esti-


mate for wNIR
N
and subtract it from equation (1), apply the existing SeaWiFS at-
mospheric correction procedure to retrieve a new Chl and iterate through this procedure
until a converged Chl value is obtained. The initial Chl value, Chl0, is set to 0.2 mg m-3.


The NIR wNIR
N
correction procedure can be summarized schematically as follows:
bg
bg
NIR g
VIS g
.Corr .
min e
Chl 0 Deter

  b
 Atmos

 b
 & Chl  repeat
w
w



N
N



Initial
Iterations
(10)
Obviously, the key to effectively correct for the NIR contribution is to accurately assess

 NIR
w
N
values given only estimate of the bio-optical state of the ocean (cf., chloro-
phyll concentrations) and the characteristics of the atmosphere. The iterative procedure
suggested here is a good first step towards accounting for this effect.
13
4.
Application to SeaWiFS imagery
To assess the importance of the black pixel assumption, we apply the present NIR
correction scheme to SeaWiFS imagery on both local and global scales. First, we use a
SeaWiFS local area coverage (LAC) image from the Chesapeake Bay region to show that
the black pixel assumption leads to large errors in highly productive and turbid waters.
Next, we assess changes in the SIMBIOS global field-satellite match-up data set after parameterizing for the effects of the black pixel assumption. Last, we evaluate the effects of
the present NIR parameterization on global SeaWiFS imagery.
4A. An Example of SeaWiFS Imagery from Chesapeake Bay
As mentioned in the introduction, the SeaWiFS chlorophyll retrievals often
show large overestimates of surface chlorophyll concentration in productive waters. A SeaWiFS LAC chlorophyll image for the Chesapeake Bay region (east
coast of North America) from May 19, 1998 is shown in figure 4a. Most of the
chlorophyll retrievals throughout the bay have retrieved values in excess of 64
mg m-3 (figure 5a). The median observed chlorophyll value is 64 mg m-3 which is
the maximum value digitized in the SeaWiFS processing scheme. However, field
observations from this time period show no values in excess of 15 mg m -3 (figure
5c). A re-analysis of this SeaWiFS image using the present NIR parameterization
shows substantial improvements (figure 4b) compared with the standard processing results (figure 4a). In particular, nearly all of the excessively large chlorophyll retrievals (> 30 mg m-3) are removed and the median chlorophyll concentration is not 64 mg m-3 (figure 5c). The SeaWiFS chlorophyll retrievals within the
main portion of Chesapeake Bay are now consistent with the field observations from the
same period (figure 5b and 5c). The improvement for this particular scene is striking, indicating the importance of the black pixel assumption in highly productive waters.
14
4B. Global in situ match-up analyses
Despite the limited number of observations at high chlorophyll concentration, the
global matchup data set reprocessed with the present NIR algorithm shows some improvement compared with the original analysis (figure 6). In particular, the regression
slopes between the two LwN(412) retrievals are closer to the 1:1 line for the NIR corrected
data set, although the value of the slope remains clearly less than one. Similarly, root
mean square (rms) deviations between the satellite and field observations are less for the
NIR processed SeaWiFS retrievals for all bands of water-leaving radiance and for chlorophyll (figure 6).
Consistent with what is seen with the example from Chesapeake Bay (figures 4 and
5), there is a significant improvement in the correspondence between the SeaWiFS and
field estimates of chlorophyll at high concentrations (figure 7). At concentrations above 1
mg.m-3 the rms differences in chlorophyll retrievals are smaller for the NIR processing
(1.5 mg m-3) compared with the standard processing (2.1 mg m-3) and the effects of outlier match-up points is much smaller.
4C. Global Imagery Analysis
Analysis of global imagery enables the importance of the NIR correction to be put in a
proper context for the ocean color remote sensing of the global ocean. Figure 8 shows
two SeaWiFS 8-day composite scenes for summer (Jul. 12 - 19, 1998) and winter (Jan. 17
- 24, 1998) conditions. The effects of the bio-optical NIR algorithm on global chlorophyll retrievals is only important (> 10%) when the baseline chlorophyll is more than 2
mg m-3. However, these conditions occur for only 2.1 and 1.3% of total number of good
retrievals for the summer and winter composites, respectively. However, there are conditions where the effects are quite important reaching nearly 60% of the baseline value (figure 8).
15
The effects of the black pixel assumption on retrievals of water-leaving radiance can
also be addressed (figure 9). As seen before, only for the highest baseline chlorophyll
concentration categories shown (10-20, 5-10 and 2-5 mg m-3) will the misapplication of
the black pixel assumption have a large influence (> 20%) on the shape of the retrieved
LwN() spectrum (figure 9). Significant effects (~10%) are also observed for the 1-2 mg
m-3 category. It should be noted that the present global analysis will be an underestimate
due to zero replacements done for negative LwN() retrievals in the composite making
procedure in the present version of SeaWiFS processing (Ref???).
5.
Discussion and Future Directions
The present study demonstrates that the black pixel assumption in ocean col-
or remote sensing must be reconsidered in productive waters where the nearsurface chlorophyll concentrations are greater than 2 mg m-3. For these waters,
the shape of the retrieved LwN() spectrum is altered which may have an important impact on semi-analytical algorithms of ocean color (e.g., Garver and
Siegel, 1997; Carder et al. 1999; Ciotti et al. 1999). However, many aspects of the
present analyses are still rather poorly understood. These include the assumptions used to relate the NIR water-leaving radiance to NIR inherent optical properties and the modeling of NIR inherent optical properties as a function of upper
layer chlorophyll content. In the following, we address several of the outstanding issues involved the present NIR correction procedure and provide some
thoughts about future research directions.
Many important radiative transfer processes have been neglected in the present estimates of [w(NIR)]N. These include the contributions to the water-leaving
radiance due to Raman scattering, inconsistency with the bidirectional reflectance distribution function (BRDF) for the NIR wavebands and the influence of
changes in the ambient ocean temperature on the present determinations of
16
[w(NIR)]N. We address several of these issues using the Hydrolight (vers. 4.02)
radiative transfer model (Mobley ref = ???). The inclusion of Raman transpectral
scattering processes increases the estimates of NIR water-leaving reflectance,
[w(NIR)]N. We find that for chlorophyll concentrations greater than ~0.5 mg m -3
and reasonable solar zenith angles, the error in estimates of [w(NIR)]N by not including Raman scattering is less than 5% (results not shown). Only for oligotrophic concentrations (Chl = 0.05 mg m-3), do the errors approach 10%.
Another poorly constrained factor is the role of the BRDF on the NIR wavebands. We used the Hydrolight model to assess differences in water-leaving radiance in the plane perpendicular to the solar plane for wavelengths of 443 and
765 nm for different solar illumination geometries and chlorophyll concentrations. The anti-solar plane is used to represent a typical scan line for an ocean
color imager. We find differences in the water-leaving radiance along the scan
line normalized to the nadir looking radiance estimate are consistent between the
443 and 765 nm wavebands (within 10% worst case). Further, there are no large
differences found in these normalized differences along the scan line with changes in solar zenith angle or chlorophyll concentration. Hence, BRDF issues appear
to be similar for the NIR wavebands as they are for the visible wavebands and
we expect this to be a minor issue for the correction of black pixel assumption.
Changes in ambient sea surface temperature (SST) can also affect estimates of
[w(NIR)]N by altering the absorption coefficient for pure seawater. Pegau et al. (1997)
made laboratory determinations of the change in aw(750) per degree of seawater temperature. Using their estimate of 0.0106 m-1 C-1 and assuming clear water conditions (Chl =
0), we expect a 4% decrease in the value of [w(750)]N for a 10 C increase of SST. Thus,
ambient temperature-induced changes in seawater inherent optical properties in the NIR
will not have a significant effect on NIR correction procedures.
17
As described above, the present parameterization for NIR particulate backscatter has
many limitations. The present approach appears to provide an improvement in SeaWiFS
retrievals especially for productive waters. It is not at all satisfactory from a theoretical
point of view. An important limitation is the modeling of bbp(NIR) as a function of the
chlorophyll concentration. Such a parameterization is probably reasonable for oligotrophic and mesotrophic waters. However, it will be questionable at best in most high
Chl conditions where backscatter can originate from other components than chlorophyll
(Sydor and Arnone. 1997; Gould et al. 1999; Toole and Siegel, 1999). These materials
may include detrital biological material and abiotic particulates such as suspended sediments. For that reason, the modeling of bbp(NIR) on a closure-based basis seems preferable. This approach first requires an accurate development data set to tune a NIR closure
algorithm. Such data currently do not exist and this is an obvious area for future work.
Clearly, there are limits on the applicability of the black pixel assumption and for
productive waters NIR effects in the water-leaving radiance must be corrected for. A value of Chl of 2 mg m-3 appears to be a good breakpoint above which NIR effects must be
considered. However, the misapplication of the black pixel assumption is not the only
difficulty creating the over correction of SeaWiFS water-leaving radiance spectra (figures
1 and 6) and more work is required. As described here, there are many short-comings in
the present modeling of [w(NIR)]N and reliable measurements data sets from which to
build models of [w(NIR)]N are required.
18
REFERENCES
Austin, R.W. The remote sensing of spectral radiance from below the ocean surface. In:
Optical Aspects of Oceanography, N.G. Jerlov and E.S. Nielsen, Eds., pp. 317-344, Academic Press, San Diego, Calif., 1974
Bricaud A, Morel A, Babin M, Allali K, Claustre H, 1998, Variations of light absorption
by suspended particles with chlorophyll a concentration in oceanic (case 1) waters: Analysis and implications for bio-optical models. J. Geophys. Res., 103, 31,033-31,044.
Carder K.L., F.R. Chen, Z.P. Lee, S.K. Hawes and D. Kamykowski, 1999, Semianalytic
Moderate-Resolution Imaging Spectrometer algorithms for chlorophyll a and absorption
with bio-optical domains based on nitrate-depletion temperatures. J. Geophys. Res., 104,
5403-5421.
Ciotti, AM; Cullen, JJ; Lewis, MR.(1999) A semi-analytical model of the influence of
phytoplankton community structure on the relationship between light attenuation and
ocean color. J. Geophys. Res., 104, 1559-1578.
Frouin, R., M. Schwindling and P.Y. Deschamps, 1996, Spectral reflectance of sea foam
in the visible and near-infrared - In situ measurements and remote sensing implications. J.
Geophys. Res., 101, 14361-14371.
Garver, S.A., and D.A. Siegel, 1997, Inherent optical property inversion of ocean color
spectra and its biogeochemical interpretation: I. Time series from the Sargasso Sea. J.
Geophys. Res., 102, 18,607-18,625.
Gordon, H.R. (1997), Atmospheric correction of ocean color imagery in the Earth Observing System era, J. Geophys. Res., 102: 17,081-17,106.
Gordon, H.R. and A.Y. Morel, 1983, Remote Assessment of Ocean Color for Interpretation of Satellite Visible Imagery: A Review, Springer-Verlag, New York, 114 pp.
Gordon, H.R., Brown, J.W., and Evans, R.H. (1988a), Exact Rayleigh scattering calculations for use with the Nimbus-7 Coastal Zone Color Scanner, Appl. Opt., 27: 862-871.
Gordon, H.R., Brown, O.B., Evans, R.H., Brown, J.W., Smith, R.C., Baker, K.S., and
Clark, D.K. (1988b), A semianalytic radiance model of ocean color, J. Geophys. Res., 93:
10909-10924.
Gordon, H.R., and Clark, D.K. (1981), Clear water radiances for atmospheric correction
of coastal zone color scanner imagery, Appl. Opt., 20: 4175-4180.
Gordon, H.R., and Wang, M. (1992), Surface roughness considerations for atmospheric
correction of ocean color sensors. 1: The Rayleigh scattering component, Appl. Opt., 31:
4247-4260.
19
Gordon, H.R., and D.K. Clark, 1980, Atmospheric effects in the remote-sensing of phytoplankton pigments Bound.-Lay. Meterol., 18, 299-313.
Gordon, H.R., and K.Y. Ding, 1992, Self-shading of in-water instruments. Limnol.
Oceanogr., 37, 491-500.
Gordon, H.R., and Wang, M. (1994a), Influence of oceanic whitecaps on atmospheric
correction of ocean-color sensor, Appl. Opt., 33: 7754-7763.
Gordon, H.R., and Wang, M. (1994b), Retrieval of water-leaving radiance and aerosol
optical thickness over the oceans with SeaWiFS: A preliminary algorithm, Appl. Opt., 33:
443-452.
Gould, R.W., R.A. Arnone and P.M. Martinolich, 1999, Spectral dependence of
the scattering coefficient in case 1 and case 2 waters. Appl. Opt., 38, 2377-2383.
Hale, G.M. and Querry, M.R. (1973). Optical constants of water in the 200-nm to
200-µm wavelength region. Appl. Opt., 12: 555-563.
Loisel H., and A. Morel, 1998, Light scattering and chlorophyll concentration in case 1
waters: A reexamination, Limnol. Oceanogr., 43, 847-858.
McClain, C.R., Cleave, M.L., Feldman, G.C., Gregg, W.W., Hooker, S.B., and Kuring,
N., Science Quality SeaWiFS Data for Global Biosphere Research, in Sea Technology,
pp. 10-16, 1998.
Mobley, C.D., 1994, Light and Water: Radiative Transfer in Natural Waters, Academic
Press, San Diego CA 592 pp.
Mobley, C.D., Hydrolight 4.02 reference
Morel, A. (1988), Optical modeling of the upper ocean in relation to its biogenous matter
content (case 1 waters), J. of Geophys. Res., 93: 10749-10768.
Morel, A., 1987, chlorophyll-specific scattering coefficient of phytoplankton - a simplified theoretical approach. Deep-Sea Res. Part A, 34, 1093-1105.
Morel, A., and Y.H. Ahn, 1990, Optical efficiency factors of free-living marine bacteria influence of bacterioplankton upon the optical-properties and particulate organic carbon
in oceanic waters. J. Mar. Res., 48, 145-175.
Morel and Maritorena, in prep.
Morel, A. and D. Antoine. Heating Rate within the Upper Ocean in Relation to its BioOptical State. J. Phys. Oceanogr., 24 (7) : 1652-1665, 1994.
20
O'Reilly, J.E., Maritorena, S., Mitchell, B.G., Siegel, D.A., Carder, K.L., Garver, S.A.,
Kahru, M., and McClain, C.R. (1998), Ocean color chlorophyll algorithms for SeaWiFS,
J. of Geophys. Res., 103: 24,937-24,953.
Palmer, K.F. and Williams, D.W. (1974). Optical properties of water in the near infrared.
J. Opt. Soc. Amer., 64(8) : 1107-1110.
Pegau, W.S., D. Gray and J.R.V. Zaneveld, 1997, Absorption and attenuation of visible
and near-infrared light in water: dependence on temperature and salinity. Appl. Opt., 36,
6035-6046.
Smith, R.C., and K.S. Baker, 1981, Optical properties of the clearest natural waters. Applied Optics, 20, 177-184.
Stramski, D. and D.A. Kiefer, 1991, Light-scattering by microorganisms in the open
ocean, Prog. Oceanogr., 28, 343-383.
Sydor, M., and R.A. Arnone, 1997, Effect of suspended particulate and dissolved organic
matter on remote sensing of coastal and riverine waters. Appl.Opt., 36, 6905-6912.
Toole, D.A., and D.A. Siegel, 1999: Diagnosis of ocean color variability in the Santa
Barbara Channel. For Journal of Geophysical Research.
Wang, M. (1999a), Atmospheric correction of ocean color sensors: Computing atmospheric diffuse transmittance, Appl. Opt., 38: 451-455.
Wang, M. (1999b), A sensitivity study of SeaWiFS atmospheric correction algorithm:
Effects of spectral band variations, Rem. Sens. Environ., 67: 348-359.
Yang, H., and Gordon, H.R. (1997), Remote sensing of ocean color: assessment of waterleaving radiance bidirectional effects on atmospheric diffuse transmittance, Appl. Opt.,
36: 7887-7897.
21
Figure Captions
Figure 1. The SIMBIOS in situ - SeaWiFS imagery match up comparison for LwN() at
412, 443, 490 and Chl. The SeaWiFS observations are processed using the standard version 2 processing procedures. Procedures explaining the matchup data set procedure are
provided in Schieber et al. (1999).
Figure 2. Comparison of bbp() and [w()]N esimates versus chlorophyll a concentration
using the SeaBAM data set for a) bbp(443), b) bbp(865), c) [w(760)]N and d) [w(865)]N.
The results of the bio-optical algorithm (eq. 4) are shown as the solid line while the points
are from the closure model (eq. 5). The SeaBAM (N=919) data set is used (O'Reilly et
al. 1998).


Figure 3. The errors  wNIR 
N
in the retrieved w  N by ignoring the NIR ocean
contributions for the SeaWiFS bands 1-5 for the aerosol M80 model with optical thickness of 0.1 at 865 nm, 7 chlorophyll concentration values, and for the solar and viewing
geometries of (a) and (c)  0 = 20°,  = 20°,   = 90° and (b) and (d)  0 = 40°,  = 40°,
  = 90°. Note that (c) and (d) are in relative errors (%). The curves from the top to the
bottom in these figures correspond to Chl concentrations of 0.1, 0.3, 0.5, 1.0, 1.5, 2.0, and
5.0 mg/m3, respectively.
Figure 4. SeaWiFS LAC chlorophyll scene for the Chesapeake Bay and adjacent waters
from May 19, 1998 (S1998139171559.L1A_HNSG) processed using a) the SeaWiFS
standard processing version 2 and b) the present NIR correction procedure. The purple
dots in a) correspond to the location of field observations used in making figure 5c.
Figure 5. Chlorophyll concentration histograms for observations taken from within
Chesapeake Bay from May 19, 1998 using a) the SeaWiFS LAC scene and the standard
processing version 2, b) the SeaWiFS LAC scene and the present NIR correction proce-
22
dure and c) from in situ observations taken between May 18 and 20, 1998. The purple
dots in figure 4a provide the location of the data used in making figure 5c.
Figure 6. The SIMBIOS matchup data set for LwN() at 412, 443, 490 and Chl after the
NIR correction procedure is performed. The format is identical to figure 1.
Figure 7. Relative error in the SIMBIOS match-up data set for chlorophyll a concentration using a) standard processing and b) with the present NIR bio-optical algorithm. The
rms deviation for high chlorophyll conditions (baseline > 1 mg m-3) decreases by a significant amount after the NIR correction is employed (2.04 to 1.36 mg m-3).
Figure 8. Percentage reduction in SeaWiFS chlorophyll retrievals after implementation of
the NIR correction procedure (solid line) as a function of the baseline chlorophyll concentration. Data are shown for two 8-day composite SeaWiFS GAC scenes for a) summer
(Jul 12 - 19, 1998) and b) winter (Jan 17 - 24, 1998) conditions. Also shown are the percentage occurrence of the baseline chlorophyll intervals (dotted).
Figure 9. Percentage improvement in water-leaving radiance spectra retrievals (upper)
and normalized to the estimated spectra (lower) after implementation of the NIR correction procedure (corrected - baseline). Data are shown for two 8-day composite SeaWiFS
GAC scenes for summer (right; Jul 12 - 19, 1998) and winter (left; Jan 17 - 24, 1998)
conditions. The different lines correspond to categories of chlorophyll concentrations
from (to line) 10-20 mg m-3, 5-10 mg m-3, 2-5 mg m-3, 1-2 mg m-3, 0.5-1 mg m-3, etc.
Appreciable changes are found only for Chl levels greater than 1 mg m-3.
23
Table 1: Parameters Used For Determining LwN(NIR)
Center Wavelength
670 nm
760 nm
865 nm
SeaWiFS Band #
6
7
8
aw() (m-1)
0.4346
2.550
4.286
Reference
Smith & Baker (1981) bands 6 & 7
Hale & Query (1973) band 8
bbw()
(m-1)
Fo() (W cm-2 nm-1)
0.00041
0.00024
0.00014 Smith & Baker (1981)
153.41
122.24
98.82
SeaWiFS ref = Sean Bailey ???
Table 2: Ensemble Mean and Standard Deviation (in '( )') Estimates for bbp(443),
bbp(760) and bbp(865) using SeaBAM (N=919)
Model
bbp (443) (m -1)
bbp (760) (m-1)
bbp(865) (m -1)
Bio-Optical (eq. 4)
0.00385 (0.00417)
0.00250 (0.00290)
0.00227 (0.00268)
Carder et al. [1999]
0.00475 (0.00388)
0.00208 (0.00261)
0.00174 (0.00239)
Table 3: Ensemble Mean and Standard Deviation (in '( )') Estimates for LwN(670),
LwN(760) and LwN(865) using SeaBAM (N=919)
Model
LwN(670) (x103)
LwN(760) (x103)
LwN(865) (x103)
Bio-Optical (eq. 4)
52.98 (42.73)
6.78 (7.16)
2.87 (3.19)
Carder et al. [1999]
50.17 (45.73)
5.75 (6.45)
2.23 (2.85)
24
TABLE 4. Error R NIRi, j  (%) in the retrieved ratio of the normalized water-leaving
reflectance between the SeaWiFS bands 2 and 5 and bands 3 and 5 for a Maritime
RH=80% aerosol model with  a 865 = 0.1 and various cholorphyll concentrations and
for the two solar and viewing geometries.
Chl
(mg/m3)
0=20°, =20°, =90°
0=40°, =40°, =90°
R NIR2, 5 (%) R NIR3,5 (%) R NIR2, 5 (%) R NIR3,5 (%)
0.1
1.13
1.02
2.05
1.89
0.3
0.96
1.43
0.90
1.83
0.5
-0.51
1.17
-1.55
1.21
1.0
-6.49
-0.58
-9.98
-1.37
1.5
-13.56
-3.00
-18.02
-4.19
2.0
-20.22
-5.45
-25.63
-7.05
5.0
-66.27
-23.56
-60.48
-21.63
TABLE 5. Error in the retrieved chlorophyll concentration using the Morel-3 and
OC2v2 algorithms. The atmosphere is specified as a Maritime RH=80% aerosol type
with  a 865 = 0.1. Various true chlorophyll concentrations are used for two solar and
viewing geometries.
Chl
0=20°, =20°, =90°
0=40°, =40°, =90°
(mg/m3)
Morel-3 (%)
OC2v2 (%)
Morel-3 (%)
OC2v2 (%)
0.1
-2.0
-2.8
-3.7
-5.1
0.3
-1.5
-3.3
-1.4
-4.2
0.5
0.8
-2.7
2.6
-2.7
1.0
12.0
1.3
18.4
3.15
1.5
26.9
7.0
35.8
9.8
2.0
42.2
12.9
53.4
16.7
5.0
158.2
58.2
144.4
53.4