IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 35, NO.

3, MAY 1997 675

Second Simulation of the Satellite Signal

in the Solar Spectrum, 6S: An Overview
Eric F. Vermote, Member, IEEE, Didier Tanré, Jean Luc Deuzé, Maurice Herman, and Jean-Jacques Morcrette

Abstract—Remote sensing from satellite or airborne platforms II. BACKGROUND

of land or sea surfaces in the visible and near infrared is strongly
affected by the presence of the atmosphere along the path from Two atmospheric processes modify the solar radiance re-
Sun to Target (surface) to Sensor. This paper presents 6S (Second flected by a target when viewed from space: absorption by
Simulation of the Satellite Signal in the Solar Spectrum), a the gases (when observation band are overlapping gaseous
computer code which can accurately simulate the above prob- absorption bands) and scattering by the aerosols and the
lems. The 6S code is an improved version of 5S (Simulation of molecules. If the gaseous absorption can be de-coupled from
the Satellite Signal in the Solar Spectrum), developed by the
Laboratoire d’Optique Atmospherique ten years ago. The new scattering as if the absorbants were located above the scattering
version now permits calculations of near-nadir (down-looking) layers, as assumed in the 5S code, the equation of transfer for
aircraft observations, accounting for target elevation, non lam- a lambertian homogeneous target of reflectance at sea level
bertian surface conditions, and new absorbing species (CH4 , altitude viewed by a satellite sensor (under zenith angle of
N2 O, CO). The computational accuracy for Rayleigh and aerosol
view azimuth angle of view ) and illuminated by sun
scattering effects has been improved by the use of state-of-the-
art approximations and implementation of the successive order ( , ) is (see [1] for a complete description)
of scattering (SOS) algorithm. The step size (resolution) used
for spectral integration has been improved to 2.5 nm. The goal
of this paper is not to provide a complete description of the (1)
methods used as that information is detailed in the 6S manual,
but rather to illustrate the impact of the improvements between The various quantities are expressed in terms of equivalent
5S and 6S by examining some typical remote sensing situations.
Nevertheless, the 6S code has still limitations. It cannot handle reflectance defined as where is the
spherical atmosphere and as a result, it cannot be used for measured radiance, is the solar flux at the top of the
limb observations. In addition, the decoupling we are using for atmosphere, and where is the solar zenith
absorption and scattering effects does not allow to use the code angle.
in presence of strong absorption bands. In (1), corresponds to the intrinsic reflectance of the
molecule+aerosol layer, , [respectively ] to the
I. INTRODUCTION total transmission of the atmosphere on the path between the
sun and the surface, (respectively between the surface and
T HE 5S CODE enables to simulate the signal observed
by a satellite sensor for a lambertian target at sea level
altitude. An effort has been done to refine the field of applica-
the sensor). is the spherical albedo of the atmosphere i.e.,
the normalized irradiance backscattered by the atmosphere
when the input irradiance at the bottom is isotropic. is the
tion and the accuracy of the code. After a section presenting
gaseous transmission, for the solar radiation H O, CO , O ,
background materials on the modeling of the remote sensing
and O are the principal absorbing gases. The transmission
signal (Section II), the main section describe the modifications
done to 5S improve the accuracy and capability of the code is a nonlinear function of the effective amount of absorptive
(Section III). A specific sub-section is devoted to the inclusion matters in the atmosphere, and depends also on pressure and
of BRDF (Bidirectional Reflectance Distribution Function) temperature profile. In 5S this term is computed by a band
as well as the problem of variable altitude for both sensor absorption model.
and target. In a last part, Section IV, we present the new The 5S code was written ten years ago and the computer
possibilities for simulation available with the 6S database with resources at this time did not allow the use of accurate methods
a special emphasis on the surface BRDF models, we also give for solving the equation of transfer due to the large com-
some illustration of the application of atmospheric correction putation burden involved. The reflectances and transmissions
to airborne data. were computed using linear single scattering approximation
for molecules and the Sobolev approximation for aerosol
Manuscript received June 6, 1995; revised November 15, 1996. multiple scattering. For molecules, the linear single scattering
E. F. Vermote was with the Laboratoire d’Optique Atmosphérique, Univer- approximation is problematic when Rayleigh optical depth
sity of Lille, France. He is now with the Department of Geography, University
of Maryland, College Park, MD 20742 USA. or view and solar zenith angle become large. Typically at
D. Tanré, J. L. Deuzé, and M. Herman are with the Laboratoire d’Optique wavelength shorter than 0.55 m and for view/solar zenith
Atmospherique, University of Lille, France. angles greater than 30 the error can reach several time
J.-J. Morcrette is with the European Center for Medium Range Weather
Forecast, Reading, U.K. [reflectance units]. This is problematic when looking
Publisher Item Identifier S 0196-2892(97)03464-5. at dark dense vegetated or oceanic surfaces where the target re-
0196–2892/97$10.00  1997 IEEE
676 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 35, NO. 3, MAY 1997

5S enables to directly use pre-defined spectral bands for

several instruments in flight ten years ago. It also gives
the possibility to enter the spectral response with a 5-nm
resolution. New instruments have now finer spectral bands and
it becomes difficult to enter a representative spectral response
at a 5-nm resolution. Moreover the database of 5S instrument
spectral responses have to be actualized to reflect sensors
currently used.
In 5S, the ground target could be uniform or non uniform
but has to be lambertian. None of the natural targets are
lambertian, but the proper treatment of nonlambertian target
needs to compute the coupling of the atmospheric directional
downward radiation field with Bi-directional Reflectance Dis-
Fig. 1. Contribution of multiple scattering to the intrinsic aerosol reflectance tribution Function (BRDF) of the target that is computationally
as a function of the view zenith angle for several aerosol optical depth in
case of a continental aerosol model for AVHRR channel 1. Computations are expensive. Once again, the authors of 5S had to sacrifice the
performed for two relative azimuths ( = 0 for v < 0,  = 180 for reality versus the practical use of a simulation code.
v > 0). In 5S, in order to simplify the radiative problem the target
has to be at sea level and the sensor out of the atmosphere. In
flectance is of the order or [reflectance units]. order to be closer to real case situation one has to consider
In addition, the intrinsic atmospheric reflectance observed over the elevation of the target which influences the molecular
black target, was written in 5S as , the sum of
scattering as well as gaseous absorption. Also, in most cases,
aerosols and Rayleigh contributions which is not fully valid
field campaign involving airborne prototype instruments re-
at short wavelengths of the solar spectrum ( m)
quires the possibility to treat within the atmosphere sensor
and/or large sun and view zenith angles [2]. For aerosol, the
cases. The “exact” solution to the radiative problem is once
contribution of multiple scattering can reach rapidly half of
again not obvious and requires the possibility to compute the
the total aerosol contribution (Fig. 1) and therefore an accurate
vertical distribution of the radiation. The successive order of
method for the computation of multiple scattering contribution
scattering (SOS) method [5] enables to solve the equation of
is essential.
transfer for in-homogenous atmosphere for a discrete number
The molecular scattering properties, its phase function,
of atmospheric layer and is an adequate solution to the above
spectral dependence of the optical depth and depolarization
factor are well known and stable. Conversely, the aerosol mentioned problem.
scattering properties are variable. For a given aerosol model, III. RADIATIVE TRANSFER MODELING IN 6S
the phase function, extinction and absorption coefficients have
to be computed as a function of the wavelength. The optical A. Atmospheric Terms
properties of each individual spherical particle (real and imag-
inary index) as well as the size distribution of the particles 1) Rayleigh: Rayleigh scattering has been extensively stud-
are the parameters used in the Mie computations. In 5S, ied and the values of the three atmospheric functions , , and
these computations for the basic aerosol components (soot, (1) have been tabulated by Chandrasekhar [6] among others.
oceanic, dust like, water soluble) defined by the International Since tables are not convenient, we developed analytical
Radiation Commission were performed at a fixed step in radius expressions [7] giving a sufficient accuracy. An accuracy of
constrained by the available computing resources. 0.001 (reflectance unit) is achieved for the reflectance, the
The gaseous transmission is computed in 5S using random transmission function is estimated within a relative accuracy
exponential band models [3]. These models permit computa- better than 0.7% when the differences between the exact
tion of transmission within relatively large spectral bands (20 computations of the spherical albedo and our expression are
cm ) suitable for the purpose of simulating the absorption around 0.003 for which corresponds to the most
within radiometer bands of several thousands of cm (as unfavorable conditions.
used in the Advance Very High Resolution Radiometer bands, 2) Aerosol: In 5S, the scattering properties of the aerosol
AVHRR). However, this may be problematic when trying to layer were computed using the Sobolev approximation for
simulate the absorption for spectrometers of higher resolution the reflectance [8], the Zdunkowski approach [9] for the
with bands covering only a few hundreds cm . The random transmission, and a semi-empirical formula for the spherical
exponential band models are computed using AFGL atmo- albedo. The goal was to provide the user having limited
spheric absorption line parameters published in 1982 [4]. Also, computing resources with a fast approximation. The drawbacks
in the 0.25–4.0 m range, some gases absorb radiation that are to using these approximations were that the accuracy of
not taken into account in 5S computation, namely Methane the computations could be off by a few time
(CH ), nitrous oxide (N O), and carbon monoxide (CO). The [reflectance units] especially at large view and sun angles
6S code is now able to compute the atmospheric transmission or high optical thicknesses. In addition, these approximations
due to absorbing gases in the solar reflective spectral range could be completely insufficient to handle the integration
with a good accuracy when scattering effects are neglected. of the downward radiance field with nonlambertian ground
Again, it has limitations when both effects are considered. conditions (cf. Fig. 7), a problem in simulating BRDF. The
VERMOTE et al.: SECOND SIMULATION OF THE SATELLITE SIGNAL 677

Fig. 3. Comparison between two versions the Successive Order of scattering

computations of the atmospheric intrinsic reflectance, one includes polariza-
tion effect and the other just performs scalar computation as it is done in 6S.
Fig. 2. Comparison of 5S, 6S (13 layers, 24 angles) with the Successive The atmospheric parameters are the same as for Fig. 2.
Order of scattering computation of the atmospheric intrinsic reflectance for
a turbid atmosphere. The coupling between absorption by the water vapor and
scattering from the aerosol could be important, because the
new scheme used to compute the scattering properties of the aerosol and water vapor can be located at the same level in the
aerosol+Rayleigh system relies on the successive order of atmosphere (2–3 km). The 6S code cannot handle the problem
scattering (SOS) method in its scalar version. The accuracy of in an exact way since the absorption should be computed along
such a scheme is better than a few [reflectance units] every path after each scattering event. Therefore, we consider
[10]. It also enables exact simulations of airborne observations. three extreme cases in 6S, the water vapor above the aerosol
In addition, since the downward radiation field is computed for layer [maximum absorption, see (2) for ], the water
a quadrature of 13 gaussian emerging angles by 13 azimuths, it vapor under the aerosol layer [minimum absorption, see (2) for
will provide the necessary inputs for considering BRDF at the ], and an average case where we consider that half of the
boundary conditions (see Section III-B2). Fig. 2 shows for a water vapor present in the atmosphere absorbs the aerosol path
turbid atmosphere, i.e., an atmosphere containing aerosol, the radiance [see (2) for ]. In that respect, (1) is modified
comparison between the successive order of scattering method in 6S as:
results (with 24 gauss angles and 26 layers) and 6S (using
12 gauss angles and 13 layers) at 550 nm. The difference
is negligible but enables to gain a factor 4 in computation
time. If the computation time is not a requirement the user
can increase the number of layers up to 26 since it is a
parameter of the code, then will get a very good accuracy
even for optical thickness larger than 0.5. Fig. 2 also shows the
results obtained by 5S. Although the accuracy of the molecular (2)
intrinsic reflectance is still acceptable at this wavelength, the where refers to the gaseous absorption for other
aerosol contribution is biased by the simplified scheme used gases than water vapor, refers to H O absorption,
in 5S especially in the forward scattering direction. Finally, and is an estimate of the aerosol intrinsic
Fig. 3 shows the difference between the scalar SOS results reflectance. Equation (2) is still clearly an approximation
(which does not take account for polarization as it is done since we assume both effects occur separately, but cases
in 6S) versus the vectorial SOS results (taking into account 1 and 3 are maximizing/minimizing the effect when case 2
polarization by molecules and aerosols) the error is small is statistically meaningfull. For each cases, we compute the
compared to the signal observed (less than 1.6% in relative) top of the atmosphere reflectance, so the uncertainty due
which justifies the use of the scalar code when molecules to the variable vertical distribution of aerosol versus water
and aerosols are mixed. Let us remind that, when rayleigh vapor can be considered. We did not considered in these
scattering only is considered, polarization is taken into account computation any coupling between molecular scattering
through empirically adjusted coefficients. and water vapor, because water vapor absorption bands are
3) Coupling Scattering-Absorption: For the transmission located mostly at wavelengths where Rayleigh scattering is
resulting from gaseous absorption, we do make exact weak, therefore it results in errors of around 0.001 or less.
computations, as long as statistical models are considered For others gases, the scheme for decoupling absorption and
as exact. Then, for transmissions and reflectances resulting scattering selected in 5S has been conserved.
from molecules or aerosols scatterings, we perform accurate
computations by the use of the SOS method. When both B. Surface Contribution
effects occur simultaneously, we separate them. If it is quite 1) Nonuniform Target: In case of nonuniform surface, let
appropriate for ozone, it is not true for water vapor. us first assume that the target is small enough that the photons
678 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 35, NO. 3, MAY 1997

it reflects are not scattered on the surface-satellite path. Let

us note the target reflectance and the reflectance of its
surroundings assumed to be uniform. The signal at the top of
the atmosphere with no gaseous absorption [11] can be written

(3)

If the environment is not uniform, (3) is still valid if the

reflectance is correctly defined. The definition of has
to consider that the weights of the pixels are depending upon
their distance to the target. Let us assume that the surface
reflectance at a point has polar coordinate ( ). In the (a)
simple case of a nadir observation, by defining a function
we call in the next the environment function, the reflectance
is given by

(4)

is the probability that a photon which would be directly

transmitted to the target with the atmosphere, is in reality
scattered and impacts the surface within a circle whose the
origin is the target and the radius. By reciprocity, it corre-
sponds also to photons coming from the environment within
a circle and reaching the sensor after being scattered. The
limit conditions are and . If the target
of reflectance is not infinitely small but is a circle of radius
surrounded by a uniform background of reflectance , we
define the reflectance of the environment by (b)
Fig. 4. (a) Environment function for a pure molecular atmosphere (lines)
(5) for different view zenith angle (v ) compared to approximation used in 6S
(symbols) as a function of the distance to the imaged pixel (r ) and (b) same
and (3) simply becomes as (a) but for aerosol.

and aerosol. As it can be observed on Fig. 4(a) and (b), there

(6)
is a dependence of the function on the view direction for
view zenith angle larger than 30 . In order to account for this
where the gaseous absorption has been considered above the
effect, we choose to fit the environment function at the desired
scattering layer.
view angle solely as a function of the environment function
The function depends upon the molecules and aerosols
computed for a nadir view as it is suggested by Fig. 4(a) and
phase function, their optical thicknesses and vertical distribu-
(b). The results presented on Fig. 4(a) and (b) [symbols] show
tions. Molecules and aerosols contributions to can be
that a simple polynomial function of nadir view environment
simply separated by the use of
function whose coefficients depend on the logarithm of the
cosine of view angle is adequate. For molecules, the
(7)
function is fitted by the simple expression

with [resp. ] is the upward diffuse transmission

function of the molecules (resp. aerosols), and is the (8a)
total upward diffuse transmission (i.e., the sum of diffuse for aerosol, a polynomial of a higher degree is needed, that is
upward transmission due to molecules and aerosols). In (7)
the functions can be computed from a Monte Carlo code
by assuming standard vertical distribution of both constituents
and a continental aerosol model. For 6S, we look at the
dependence of these environment function as a function of the
view zenith angle. Fig. 4(a) and (b) shows for several value
of the view zenith angle the environment function of Rayleigh (8b)
VERMOTE et al.: SECOND SIMULATION OF THE SATELLITE SIGNAL 679

(a)

(b)
Fig. 5. (a) Isolines of the pixel background contribution to the signal at the top of the atmosphere for a pure molecular case. The energy source is 104
W and each pixel is considered to have a lambertian reflectance of 1. The contribution of background is the number of Watt coming from each cell
(201 2 201 cells). The plain line are for nadir viewing, the broken line are for a view angle of 70 . (b) Same as (a) but for a atmosphere composed
exclusively of aerosol particles (Continental model).

with , , , and uniform lambertian reflectance of 1 for the whole scene. The
contribution of background is the number of Watts coming
However, it has to be pointed out that if the approximations from each cell (201 201 cells). For molecules in case of
(8a), (8b) enable to take into account adjacency effect for an ar- a view angle of 70 [Fig. 5(a)] which represents an extreme
bitrary view angle, they implied uniformity of the background case, the contribution of the background in the direction of
as a function of azimuth. Fig. 5(a) [resp. Fig. 5(b)] presents the observer is greater. It results that the isolines which were
for molecules (resp. aerosols) isolines of the pixel background circles in case of nadir view are now ellipses shifted toward the
contributions considering a energy source of 10 W and a observer. For aerosols [Fig. 5(b)], the background contribution
680 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 35, NO. 3, MAY 1997

is less important and the deformation due to the tilted view

(70 ) is also less important due to the fact that aerosols forward
scattering predominates.
One direct conclusion of Fig. 5(a) and (b) is that contri-
butions of the adjacent pixels for a large view angle don’t
comply to the symmetry in azimuth. Therefore, the 6S results,
in case of large view angles, have to be interpreted more like
a sensitivity test to the problem of adjacency effect rather than
an actual way to perform adjacency effect correction.
2) BRDF: In 6S, the coupling between the BRDF (Bidirec-
tional Reflectance Distribution Function) and the downward
radiance at the surface level is taken into account according
to the scheme presented in Tanre et al. [12]. The contribution Fig. 6. Comparison of the sum of the coupling terms atmosphere-BRDF:
of the target to the signal at the top of the atmosphere is T OA 0 (R+A + e0 m s ) computed by 6S with the same quantity
decomposed as the sum of four terms: a) the photons directly computed by the SOS code for different atmospheric conditions (clear,
average, turbid). The ground BRDF is from Kimes measurements over a
transmitted from the sun to the target and directly reflected plowed field fitted with Hapke BRDF model.
back to the sensor; b) the photons scattered by the atmosphere
then reflected by the target and directly transmitted to the and , and also by the fact that multiple scattering
sensor; c) the photons directly transmitted to the target but tends to be isotropic. The form is convenient because the exact
scattered by the atmosphere on their way to the sensor; and computation will require a double integration. Thus, the signal
finally d) the photons having at least two interactions with at the top of the atmosphere is written as (without gaseous
the atmosphere and one with the target. One can compute absorption for clarity purpose only)
exactly each contributions a)–d) according to the following set
of (9a)–(9d) as already shown in Tanre et al. [12] by defining
(9a)

(10)

Therefore, the only approximation in the computing scheme

is the estimation of multiple interaction between target and
atmosphere. We present Fig. 6 the comparison for a typical
BRDF signature, the plowed field measured by Kimes [13],
of the 6S results with independent computations performed
by decomposing the BRDF in Fourier series and include it
(9b) as ground boundary condition in an SOS code as done in the
model published by Deuzé et al. [14]. We only compare the
where is the total downward diffuse transmission sum of the four last terms of (10) that involve a coupling
between the atmosphere and the surface. As it can be seen the
approximation done in the computation of multiple interaction
(9c)
between ground and atmosphere is valid and only brings sub-
where is the total upward diffuse transmission
stantial error (still lower than ) for turbid atmosphere
(9d) and large view zenith angle. We also presented, Fig. 7, the
error done for the plowed field case again, by neglecting any
In 6S, the first three contributions are computed exactly using coupling between BRDF and atmosphere and use (1) only
the downward radiation field given by the SOS method for valid for a lambertian target to perform atmospheric correction,
several directions, which allows us to perform the integrations by assuming the atmospheric properties known.
in (9b) and (9c). The contribution which involves at least
two interaction between the atmosphere and the BRDF (9d) is C. Elevated Target Simulation
approximated by taking equal to the hemispherical albedo
For a target not at sea level, (1)1 is modified as follows:
of the target

(9e)
(11)

1 For clarity purposes we illustrate the approach taken in 6S on the basis of

This approximation is justified by the limited impact on the (1), even if in the code coupling molecules-aerosol and aerosol-water vapor
total signal of this last contribution, weighted by the product are taken into account.
VERMOTE et al.: SECOND SIMULATION OF THE SATELLITE SIGNAL 681

(a)
Fig. 7. Limitations of the lambertian assumption for atmospheric correction
[use of (1) instead of (8)] for a gentle atmosphere. In that case, correction
using directional information (- - - -) is conducted by fitting the results of
the correction using (1) (+) with the Hapke BRDF model and reinjecting the
BRDF to compute coupling terms of (9a)–(9e).

The amount of scattering particles above the target

(molecules and aerosols) and the amount of gaseous absorbants
are related to the target altitude. In the 5S code, the amount
and types of aerosol are input parameters, thus the aerosol
characteristics implicitly depend on target altitude because
they are measured/estimated at target location. In 6S, the
target altitude can be an input: after selection of the
atmospheric profile, the target altitude is used to compute
a new atmospheric profile by stripping out the atmospheric (b)
level under target altitude and interpolating if necessary. This
Fig. 8. (a)–(b): Influence of the altitude of the target on the gaseous
way, an exact computation of the atmospheric functions is transmission function computed in the case of AVHRR channel 1(a) and
performed that account for coupled pressure-temperature 2(b). The atmospheric profile is the tropical atmosphere and the view and
effect on absorption. sun zenith angles are equal to 30 .

The influence of target altitude on has been evaluated

for the case of the AVHRR channel 1 (0.500–0.740 m)
[Fig. 8(a)] and channel 2 (0.690–1.080 m) [Fig. 8(b)]. Be-
cause the ozone layer is located in the upper levels of the
atmosphere, the O amount is not depending on target altitude
and transmission is not affected. Conversely, the target altitude
has an important effect on the absorption by H O because
most of the water vapor is located in the lower atmosphere.
However, the exact sensitivity of the target’s altitude on water
vapor absorption cannot be generalized because the H O
amount is directly connected to the water vapor profile which
is very variable. We can also notice the small impact on O
absorption since there is almost no O absorption lines in these
particular bands.
The effect of target altitude on molecular optical thickness Fig. 9. Influence of the altitude of the target on the molecular optical
is exactly accounted for in 6S. Operationally, because it thickness. A simple valid approximation is to consider that the molecular
is not very efficient to compute integral of the Rayleigh optical depth is proportional to the barometric pressure as shown.

extinction as function of pressure, one may consider that is

proportional to the pressure at target level. Fig. 9 shows that,
for a midlatitude summer pressure profile, and for AVHRR AVHRR channel 1 for the whole globe, using the 1/12 of
channel 1, the difference between exact computations and this degree resolution elevation map. For each pixel a 30 off nadir
approximation is negligible. observation map has been computed for solar zenith angle of
Fig. 10 illustrates the influence of target’s altitude, in term 30 (back scattering). The Digital Elevation Model used in
of variation of the Rayleigh reflectance (in absolute) for this simulation is ETOPO5 [15].
682 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 35, NO. 3, MAY 1997

Fig. 10. Influence of the altitude of target on the molecular intrinsic reflectance, the simulation shows for a typical observation geometry in the backscattering
direction (sun zenith = 30 , view zenith = 30 , relative azimuth = 0 ). The color code gives the absolute difference in unit of reflectance between the
altitude dependent computation and the computation assuming the target is at sea level. The elevation is given by the ETOPO5 model.

D. Airborne Sensor Simulation vapor content for the portion of the atmosphere located under
In case of sensor inside the atmosphere (airborne sensor), the plane.
(1)2 is modified as the following: 1) Atmospheric Reflectance and Transmission: In 6S, the
computation is performed exactly by defining one of the
multiple layers used in the SOS at the altitude of the sensor.
This enables exact computation of both reflectance and
transmission term of a realistic mixing between aerosol and
(12) molecules.
Gaseous absorption is computed with a technique similar 2) Nonuniform Target: In case of airborne observation, (6)
to the one used in the case of a target not at sea level is modified as follows:
except that it only modify the upward path. Practically, the
atmospheric layers above the sensor are stripped from the
profile, so the gaseous transmission is integrated till the sensor
altitude (interpolation of the atmospheric profile is conducted
if necessary). Fig. 11(a) and (b) illustrates the effect of the (13)
sensor altitude on gaseous transmission computation, for
30 , 30 . In the particular case presented, that is where is the optical thickness of the layer under the
AVHRR visible and near infrared channels, we must point out plane, and is the “effective” diffuse upward transmission
that O absorption on the path target-sensor is not depending equal to the total transmission computed exactly
on the sensor altitude because these molecules are located high as described in the previous section minus the modified direct
in the atmosphere. For H O, the absorption is very dependent upward transmission . The term is computed
of the altitude up to 4 km, due to the fact the water vapor as in (5) but instead of using the environment function for
is located is the first 4 km of the atmospheric column. So, the whole atmosphere, a function that takes account
if the observed channel is sensitive to water vapor absorption for the altitude of the sensor, , has been defined:
(as it is the case of AVHRR channel 2) we recommend that
additional measurements of water vapor are taken from the (14)
aircraft (except for stratospheric aircraft). In addition to the
option which allows the user to enter his own total atmospheric The diffuse transmission term related to molecules and
profile, it is now possible to enter aerosol, ozone, and water aerosols scattering effect are computed by the same approach
2 The equation is deliberately simplified in order to make the writing easier used for computing . To compute the and
but in the code all coupling terms are taken into account. functions, Monte Carlo have been performed for
VERMOTE et al.: SECOND SIMULATION OF THE SATELLITE SIGNAL 683

(a)
(a)

(b)
Fig. 11. Variation of the gaseous transmission (T g ) as a function of the (b)
observer altitude, z , (airborne case), for AVHRR channel 1(a) and 2(b) spectral Fig. 12. (a) Variation of the molecular environment function, (FR ) for
responses. The sun zenith angle is 30 and the view zenith angle is 30 , different altitude of the sensor. (b) Variation of the aerosol environment
Tropical atmosphere was used. function, (FA ) for different altitude of the sensor.

different altitudes of the sensor ( km) and by the International Radiation Commission [16] have been
included in 6S as a database. In the code, the closest simulated computed with a finer step in particles radius in 6S than the one
altitudes are used to interpolate the environment function at used in 5S. In addition, several aerosol models (stratospheric
the input sensor altitude. Fig. 12(a) and (b) illustrates some of [17], desertic [18], and aerosol resulting from biomass burning
the simulations performed. As expected, for both molecules [19]) difficult to reproduce using a mixing of basic components
and aerosols, the influence of the environment is decreasing are now available for use in 6S.
[higher ] when the sensor altitude is decreasing. The 2) Spectroscopic Data: The computation scheme has not
respective vertical distribution of molecules (scale height of 8 been changed but improvement have been done concerning the
km) and aerosol (scale height of 4 km) influence the variation resolution and the accuracy of the spectroscopic database. The
of the environment function with the altitude of the sensor. band absorption parameters of 6S has been computed using the
For aerosol the variation of is faster than for the HITRAN database at 10 cm resolution. Important gases in
molecules when the sensor altitude varies from 0.5 to 5 km. the 0.25–4.0 m region, namely CH , CO, NO , are now taken
into account in the computation of the gaseous transmission.
IV. IMPROVEMENT OF THE 5S DATABASE Fig. 13(a)–(c) shows the comparison between MODTRAN (5
The spectral resolution of the code has been improved cm resolution) and 6S for the computation of the transmis-
by refining the step of 5 nm used in 5S to 2.5 nm in sion function in the case of the typical mid-latitude summer
6S. Whenever it was possible, the spectral databases have atmosphere. For clarity purpose, the spectrum has been divided
been updated using original data at 2.5-nm resolution or in three intervals: 0.25–1.20 m [Fig. 13(a)], 1.20–2.40 m
interpolated when the original data were at a coarser resolution. [Fig. 13(b)], and 2.40–4.00 m [Fig. 13(c)]. With this new
In addition to increase the accuracy of the spectral integration, spectral resolution it is now possible to conduct atmospheric
this improvement provides better handle of the problem of fine correction for small bandwidth sensor, as illustrated by Fig. 14,
spectral bands radiometer or spectrometer for which the 5-nm where Advanced Solid state Array Spectrometer [20] (ASAS)
resolution of 5S was too coarse. data over the Konza prairie (Kansas) were corrected for
1) Aerosol Models: The radiative properties of the basic atmospheric effect using 6S. For each of the 29 spectral bands,
components (soot, oceanic, dust like, water soluble) defined the radiance value measured by the instrument were input in
684 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 35, NO. 3, MAY 1997

the atmospheric correction field, as well as the 6S parameters.

In particular, the following values were adopted for the target
altitude (0.443 km) and sensor altitude (5.66 km), the aerosol
optical depth. at 0.55 m under the plane (0.222) and total
(0.255), and the water vapor amount under the sensor (3
cm), and total (3.5 cm). The corrected vegetation spectrum is
smoother than the raw signal as expected. The artifacts present
in the raw spectrum around 0.765 m (oxygen absorption
band) and 0.830 m (water vapor absorption) are well removed
by the correction procedure. When integration over a filter is
performed, the spectral resolution is decreased up to 2.5 nm.
3) Surface Data: 6S can handle the simulation of top of the
atmosphere signal for non lambertian target. In that respect,
the user can entered the BRDF characteristics in two ways,
one is to enter a discrete field of directional measurements,
(a)
the other is to enter the parameters of a BRDF model. In
the last case, several models can be chosen, that range from
simple empirical model with limited number of parameters
to more comprehensive description of the BRDF associated
with physical parameters. The models are shortly described
hereafter, the selection in the code is driven by one parameter,
then the user has to specify the inputs of the model he selected.
Hapke’s model is based on fundamental principles of radia-
tive transfer theory, the original version is described in Hapke
[21]. The implementation in 6S of Hapke is similar to the one
suggested by Pinty and Verstraete [22]. The single scattering
part of the signal is computed exactly whereas the multiple
scattering is evaluated using a two-stream approximation. The
model includes a hot-spot component. The four parameters
of the model are the average single scattering albedo of the
scatterers ( ), the asymmetry factor of the Heyney–Greenstein
phase function ( ), the amplitude of the hot-spot [ ], and (b)
the width of the hot spot ( ).
Pinty and Verstraete model [22] is a development of the
Hapke approach applicable to fully developed canopy. The
decomposition of the signal is similar to the previous model
(single scattering, multiple scattering approximation, hot spot).
The differences lie in a better modelization of the hot spot
term, a full parameterization of the scattering properties of the
canopy (leaf orientation, leaf area density, and radii) and the
possibility to choose between a collection of phase functions.
Because there is a lot of input parameters, we will not give
the full list here but encourage the reader to refer to 6S
documentation [23].
Iaquinta and Pinty model [24] is a later improvement of
the original Pinty and Verstraete model [22]. The canopy
is now fully parameterized in term of geometry (leaf angle
distribution, leaf area index) and scattering properties at the (c)
leaf level [reflection ( ) and transmission ( )], the hot spot Fig. 13. (a) Comparison of gaseous transmission between 0.20–1.20 m
is computed using a single input parameter. The influence of computed by MODTRAN with the results of 6S for a typical mid-latitude
summer atmosphere. (b) Same as (a) but between 1.20–2.40 m. (c) Same
soil underneath the canopy has been added and the multiple as (a) but between 2.40–4.0 m.
scattering is now computed using a discrete ordinate method
routine embedded in the code.
Roujean et al. [25] use a simple semi-empirical model with resolution. This model is taking into account both multiple
only three parameters based on simplification of the physical scattering processes and hot spot.
processes of reflection in heterogeneous medium. It has the The next two models are fully empirical. The Minnaert
advantage of being linear and therefore easily inversible and equation for surface BRDF [26] has two adjustable parameters,
in that the parameters remain meaningfull at any spatial the albedo of the surface ( ) and a shape parameter ( ), which
VERMOTE et al.: SECOND SIMULATION OF THE SATELLITE SIGNAL 685

transition 5S–6S. The computation time remains reasonable

considering nowadays computing facilities (2–3 s on a 70MIPS
workstation). The code, a window based interface and a
200 pages manual, are available from anonymous ftp on
kratmos.gsfc.nasa.gov (128.183.112.125)3 or on loasil.univ-
lille1.fr (134.206.50.4)4.

REFERENCES
[1] D. Tanré, M. Herman, P. Y. Deschamps, and A. de Leffe, “Atmospheric
modeling for space measurements of ground reflectances, including
bidirectional properties,” Appl. Opt., vol. 18, no. 21, pp. 3587–3594,
1979.
[2] P. Y. Deschamps, M. Herman, and D. Tanre, “Modeling of the atmo-
spheric effects and its application to the remote sensing of ocean color,”
Appl. Opt., vol. 22, no. 23, pp. 3751–3758, 1983.
[3] R. M. Goody, Atmospheric Radiation 1, Theoritical Basis. Oxford,
U.K.: Oxford Univ. Press, 1964, p. 436.
[4] L. S. Rothman, R. R. Gamacge, A. Barbe, A. Goldman, J. R. Gillis,
Fig. 14. Example of ASAS atmospherically corrected vegetation spectrum L. R. Bown, R. A. Toth, J. M. Flaud, and C. Camy-Peyret, “AFGL
using the 6S code. The intrinsic atmospheric reflectance in the visible and the atmospheric absorption line parameters compilation: 1982 edition,”
gaseous absorption feature by water vapor and oxygen in the near infrared Appl. Opt., vol. 22, pp. 2247–2256, 1983.
are clearly noticeable before correction. [5] J. Lenoble, Ed., Radiative Transfer in Scattering and Absorbing Atmo-
spheres: Standard Computional Procedures. Hampton, VA: A. Deepak,
1985, vol. 300.
affect the BRDF function [ ] as [6] S. Chandrasekhar, Radiative Transfer. New York: Dover, 1960.
[7] E. F. Vermote and D. Tanré, “Analytical expressions for radiative
properties of planar Rayleigh scattering media including polarization
(15) contribution,” JQSRT, vol. 47, no. 4, pp. 305–314, 1992.
[8] V. V. Sobolev, Light scattering in Planetry Atmospheres. New York:
The Walthall model [27] is based on the fact that 2-D contours Pergamon, 1975.
[9] W. G. Zdunkowski, R. M. Welch, and G. Korb, “An investigation of
of BRDF for several surfaces appear to be similar to the the structure of typical two-stream-methods for the calculation of solar
shape of the limacon of Pascal and that (18) fits most BRDF fluxes and heating rates in clouds,” Beitr. Phys. Atmos., vol. 53, no. 2,
measurements used in the paper [27]. We modify slightly the pp. 147–165, 1980.
[10] J. Lenoble and C. Brogniez, “A comparative review of radiation aerosol
original equation found in [27] to make it complied with the models,” Beitr. Phys. Atmos., vol. 57, no. 1, pp. 1–20, 1984.
reciprocity principle. The equation implemented is 6S has four [11] D. Tanré, M. Herman, and P. Y. Deschamps, “Influence of the back-
adjustable parameters ( ) and is written as ground contribution upon space measurements of ground reflectance,
Appl. Opt., vol. 20, pp. 3673–3684, 1981.
[12] , “Influence of the atmosphere on space measurements of direc-
tional properties,” Appl. Opt., vol. 21, pp. 733–741, 1983.
(16) [13] D. S. Kimes, N. W. Newcomb, C. J. Tucker, I. S. Zonneveld, W. Van
Wijngaarden, J. De Leeuw, and G. F. Epema, “Directional reflectance
A spectral model for open ocean for clear water (case I as factor distributions for cover types of northern Africa,” Remote Sens.
Environ., vol. 17, pp. 1–19, 1985.
defined by Morel [28]) is included in 6S. The three parameters [14] J. L. Deuzé, M. Herman, and R. Santer, “Fourier series expansion of
are the concentration in phytoplankton, the wind speed, and the the transfer equation in the atmosphere-ocean system,” JQSRT, vol. 41,
wind direction. The model computes the just above the surface no. 6, pp. 483–494, 1989.
[15] NGDC, ETOPO5 5 Minutes Gridded Elevations/Bathymetry for the
reflectance according to the scheme presented by Morel [28] World. 1993.
and uses the equations developed by Austin [29] to compute [16] “Radiation commission of IAMAP meeting of experts on aerosols and
their climatic effects,” WCP 55, World Meteorological Organization
the reflectance just above the surface. The model takes also (CAS), 1983.
account for the effect of foam [30] and roughness of the [17] M. King, D. Harshvardhan, and A. Arking, “A model of the radiative
ocean (that influences the glint pattern) both for isotropic and properties of the El Chichon Stratospheric Aerosol layer,” J. Clim. Appl.
Meteor., vol. 23, no. 7, pp. 1121–1137, 1984.
directional wind [31]–[33]. Provided the fact that computations [18] E. P. Shettle, “Optical and radiative properties of a desert aerosol
in the current version of 6S do not include polarization model,” in Proc. Symp. Radiation in the Atmosphere, G. Fiocco Ed.
effect, critical for ocean biophysical parameters inversion that Perugia, Italy: A. Deepak, 1984, pp. 74–77.
[19] L. A. Remer, Y. J. Kaufman, and B. N. Holben, “The size distribution
requires high accuracy in the atmospheric effect simulation of ambient aerosol particles: Smoke versus urban/industrial aerosol,”
( at short wavelength), we recommend use of the accepted to Global Biomass Burning, Cambridge, MA: MIT Press,
model for sensitivity studies only. 1996.
[20] J. R. Irons, K. J. Ranson, D. L. Williams, R. R. Irish, and F. G.
Huegel, “An off-nadir-pointing imaging spectroradiometer for terrestrial
V. CONCLUSION ecosystem studies,” IEEE Trans. Geosci. Remote Sensing, vol. 29, no.
1, pp. 66–74, 1991.
The 5S code has been improved both in accuracy and [21] B. Hapke, “Bidirectional reflectance spectroscopy: 4. The extinction
coefficient and the opposition effect,” ICARUS, vol. 67, pp. 264–280,
application field to a new version 6S. Aircraft observations, 1986.
accounting for elevation target, non-Lambertian surface con-
ditions and new absorbing species are now included. The
3 USA
input parameters and the structure of 6S remain globally site.
similar to 5S, enabling existing users to smoothly make the 4 French site.
686 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 35, NO. 3, MAY 1997

[22] B. Pinty and M. M. Verstraete, “Extracting information on surface Jean Luc Deuzé, received the M.S. degree in
properties from directional reflectance measurements,” J. Geophys. Res., physics in 1970 and the Ph.D. degree in atmospheric
vol. 96, no. D8, pp. 2865–2874, 1991. optics in 1974 from the University of Lille, France.
[23] E. F. Vermote, D. Tanré, J. L. Deuzé, M. Herman, and J. J. Morcrette, He is currently a Professor of Physics at
“Second simulation of the satellite signal in the solar spectrum: User the University of Lille and a researcher at the
manual,” University of Maryland/Laboratoire d’Optique Atmospherique, Laboratoire d’Optique Atmospherique. His research
1994. activities focus on radiative transfer modelling
[24] J. Iaquinta and B. Pinty, “Adaptation of a bidirectional reflectance within the atmosphere and the inversion of sky
model including the hot-spot to an optically thin canopy,” in Proc. radiance measurements for aerosol monitoring. He
Spectral Signature in Remote Sensing Conf., Val d’Isére, France, 1994, is also involved in field campaigns to monitor
pp. 683–690. aerosols (Airborne POLDER flights) and to estimate
[25] J.-L. Roujean, M. Leroy, and P.-Y. Deschamps, “A bidirectional re- the radiance at the sensor level (SPOT calibration, atmospheric corrections).
flectance model of the earth’s surface for the correction of remote One of his main interests is in exploiting polarization data to retrieve aerosol
sensing data,” J. Geophys. Res., vol. 97, no. D18, pp. 20,445–20,468, properties for climate investigations. Within the POLDER project, he is in
1992. charge of the inversion schemes for getting aerosol characteristics over both
[26] M. Minnaert, “The reciprocity principle in lunar photometry,” Astrophy. ocean and land surfaces.
J., vol. 93, pp. 403–410, 1941.
[27] C. L. Walthall, J. M. Norman, J. M. Welles, G. Campbell, and B. L.
Blad, “Simple equation to approximate the bidirectional reflectance from
vegetative canopies and bare soil surface,” Appl. Opt., vol. 24, no. 3,
pp. 383–387, 1985. Maurice Herman received the degree in physics in
[28] A. Morel, “Optical modeling of the upper ocean in relation to its 1962, and the Ph.D. degree in physics in 1968, both
biogenous matter content (Case I Waters),” J. Geophys. Res., vol. 93, from the University of Lille, France.
no. C9, pp. 10,479–10,768, 1988. He is currently Professor and Director of the Lab-
[29] R. W. Austin, “The remote sensing of spectral radiance from below the oratoire d’Optique Atmospherique at Universite des
ocean surface,” in Optical Aspects of Oceanography, N. G. Jerlov and Sciences et Technologies of Lille. His experiment
E. S. Nielsen, Eds. San Diego, CA: Academic, 1974. is principally in the area of aerosol remote sensing.
[30] P. Koepke, “Effective reflectance of oceanic white caps,” Appl. Opt., His research activities concern radiative transfer and
vol. 23, no. 24, pp. 1816–1824, 1984. atmospheric signal modeling, planetary atmosphere
[31] C. Cox and W. Munk, “Measurement of the roughness of the sea surface studies from analysis of telescopic observations, and
from photographs of the Sun’s glitter,” J. Opt. Soc. Am., vol. 44, no. ground-based and balloon-borne remote sensing of
11, pp. 838–850, 1954. aerosols, with special emphasis on polarization.
[32] , “Slopes of the sea surface deduced from photographs of sun
glitter,” Bull. Scripps Inst. Oceanogr., vol. 6, pp. 401–488, 1965.
[33] , “Some problems in optical oceanography,” J. Marine Res., vol.
14, pp. 198–227, 1955.
Jean-Jacques Morcrette received the “These de
3eme cycle” and “Doctorat d’Etat” in atmospheric
physics in 1977 and 1984, respectively, from the
Universite des Sciences et Techniques de Lille,
Eric F. Vermote (M’95) received the Engineer France.
degree in computer science in 1987 from Ecoles des He was a visiting fellow with the National Re-
Hautes Etudes Industrielles (H.E.I.), Lille, France, search Council of Canada in 1978 to 1979, working
and the Ph.D. degree in atmospheric optics from on satellite retrieval of surface temperature for the
the University of Lille in 1990. Great Lakes and on the radiation transfer for the
He is currently assistant research scientist in the Canadian Climate Center GCM. He has been an
Department of Geography, University of Maryland, atmospheric scientist with CNRS (Centre National
with work performed at NASA Goddard Space de la Recherche Scientifique) since 1980. After a post-doctoral fellowship at
Flight Center, Greenbelt, MD. He is EOS/MODIS the National Center for Atmospheric Research from mid-1984 to mid-1986,
Science Team Member responsible for the atmo- he was detached from CNRS to ECMWF (European Centre for Medium-range
spheric correction over land surfaces in the visible Weather Forecasts) where he is in charge of the development and validation
to middle infrared. His research interests cover radiative transfer modeling, of the radiation transfer parameterization for the forecast model.
vicarious calibration, atmospheric correction and aerosol retrieval.

Didier Tanré, received the M.Sc. degree in physics

in 1975 and the “These de 3eme Cycle” and
“Doctarat d’Etat” in atmospheric physics in 1977
and 1982, respectively, all from the Universite des
Sciences et Techniques de Lille, France.
He is an atmospheric scientist in the CNRS
(Centre National de la Recherche Scientifique),
collocated with the Laboratoire d’Optique At-
mospherique at the Universite des Sciences et
Techniques de Lille, Villeneuve d’Ascq, France,
where he has been a research scientist since 1982.
He has been a member of the MODIS/EOS Science team since 1989. In
addition to these positions, he was an NRC Resident Research Associate at
NASA Goddard Space Flight Center, Greenbelt, MD, from 1989 through
1991. His research is presently focused on aerosol remote sensing from
satellite as well as on ground-based measurements.