A new climatology for atmospheric correction ... - Wiley Online Library

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 112, D14208, doi:10.1029/2006JD007496, 2007

A new climatology for atmospheric correction based on the aerosol inherent optical properties Francis Zagolski,1 Richard Santer,1 and Ouahid Aznay1 Received 10 May 2006; revised 6 February 2007; accepted 1 May 2007; published 25 July 2007.

[1] In the aerosol remote sensing algorithms over ocean, the aerosol models are deduced

from the spectral dependence of the path radiances in the near-infrared region. In the classical scheme of atmospheric corrections, the key point is the derivation of atmospheric signals in the visible part of the solar spectrum using these aerosol models. The aerosol climatology generally includes the standard aerosol models. Nevertheless, the latter may not be representative enough of the variety of aerosols encountered over coastal areas. One way to extend this climatology is to generate additional aerosol models. The latters will be described by their micro-physical properties and their inherent optical properties (IOPs) can then be computed with the Mie’s theory. Coastal and inland stations of CIMEL sun-photometers from the AERONET (AErosol RObotic NETwork) offer an extensive and representative data set of aerosol characteristics (i.e., the apparent optical properties from which can be derived the IOPs) in the marine environment through solar extinction and sky radiance measurements. An iterative method, developed at the Laboratoire Interdisciplinaire des Sciences de l’Environnement (LISE, Wimereux-France), allows to extract the aerosol phase function from these ground-based measurements. CIMEL data, collected over 25 stations during several years, were processed to build-up a database with more than 7000 sequences including single scattering albedos and phase functions at two wavelengths (675 nm and 870 nm). Statistical methods have been applied on this data set to discard the wrong sequences and to suggest a classification in aerosol models through their IOPs. Citation: Zagolski, F., R. Santer, and O. Aznay (2007), A new climatology for atmospheric correction based on the aerosol inherent optical properties, J. Geophys. Res., 112, D14208, doi:10.1029/2006JD007496.

1. Introduction [2] The approach developed here to build-up a new aerosol climatology focuses on the needs of the MERIS (MEdium Resolution Imaging Spectrometer) level-2 ground-segment algorithms, but can be easily generalized to any other sensors working in the solar spectral domain for atmospheric corrections over both land and water surfaces. The main objective is to bring recommendations for generating a new set of look-up tables (LUTs) to be used in the current MERIS algorithms both for aerosol remote sensing and atmospheric corrections over ocean. 1.1. Atmospheric Corrections Over Ocean [3] MERIS [Rast et al., 1999], on ENVISAT (launched on March 1st, 2002), is a programmable medium-spectral resolution imaging spectrometer operating in the solar reflective spectral range [400 nm – 900 nm]. Fifteen spectral bands can be selected by ground command, with a programmable width and spectral location [Merheim-Kealy et al., 1999]. The scene is simultaneously registrated across 1 Universite´ du Littoral Coˆte d’Opale (ULCO), Maison de la Recherche en Environnement Naturel (MREN), Wimereux, France.

Copyright 2007 by the American Geophysical Union. 0148-0227/07/2006JD007496

the entire spectral range through a dispersing system onto a charge-coupled devices (CCDs) array. [4] The top of atmosphere (TOA) radiance (L) measured by MERIS and delivered as the level-1 product needs to be transformed before being employed in the level-2 processing. In a first step, the radiances are normalized to an extraterrestrial solar irradiance equal to p as, Ln ¼ L

p:d 2 ; Es

ð1Þ

in which Es is the solar irradiance, for the central wavelength of each spectral band, corrected for the SunEarth distance d (in AU). This conversion is critical because of the misalignment of the CCDs and the spectrometers in the cameras which results in the variation of Es in the MERIS field of view in any band. In equation (1), Es is known for each pixel. Also, a spectral adjustment is performed to produce a normalized radiance for the nominal central wavelength of each MERIS band. Let us give an example: the central wavelength for the MERIS band-2 is 443 nm. A given pixel is characterized by its specific wavelength at 443.1 nm and two measurements at 412.8 nm and 489.9 nm are employed to transform the 443.1 nm into an equivalent measurement at 443 nm.

D14208

1 of 16

D14208

ZAGOLSKI ET AL.: A NEW AEROSOL CLIMATOLOGY BASED ON IOPS

[5] A second step consists in removing the gaseous absorption from normalized radiances at TOA. These absorbing effects mainly result from the stratospheric ozone (the ozone content is attached in the auxiliary data file) and from the water vapour. For the first correction, the ozone layer being located above the aerosols, the absorption and scattering processes can be separately treated due to the fact that the coupling between Rayleigh scattering and gaseous absorption can be neglected at this pressure level. The ozone transmissivity follows a decreasing exponential law which depends on the ozone optical thickness computed in each spectral band for a standard value of the ozone content (0.32 cm-atm) and weighted by the actual value of the ozone provided by the European Centre for Medium range Weather Forecast (ECMWF). For the water vapour (mostly located in the lower troposphere) case, the correction within a slightly contaminated spectral band is achieved by estimating the water vapour content with a polynomial fit of the ratio of the radiances at 900 nm (with absorption) and at 885 nm (without absorption) MERIS wavelengths [Santer et al., 1999]. In this approach, the water vapour transmittance will also account for the coupling between scattering and gaseous absorption. [6] Finally, we convert the normalized TOA radiance (Ln) into TOA reflectance (r*) corrected for the gaseous absorption as, r* ¼

Ln ; ms Tg

ð2Þ

where Tg is the total gaseous transmittance and ms the cosine of the solar zenith angle. [7] In ocean colour remote sensing, the TOA reflectance (r*) corrected for gaseous absorption can be linearized as, r* ¼ ratm þ T

rw þ rG eM t ; 1 rw S

ð3Þ

aerosol type and the aerosol optical thickness (AOT) at 865 nm (see Antoine and Morel [1999] for more details). 1.2. MERIS Standard Aerosol Models (SAMs) [10] The aerosol types used to generate the level-2 LUTs must be provided by a climatology. The SeaWiFS (Seaviewing Wide Field-of-view Sensor) climatology is built-up with 12 models of Shettle and Fenn [1979]: four types of aerosols (maritime, coastal, tropospheric and urban) with three relative humidities (70%, 90% and 98%). In the following sections, we will use the first letter of the model type and the relative humidity (expressed as percent) to refer to a given aerosol model. As an example M70 stands for the maritime model with a relative humidity of 70%. Among all these models, only the urban types present a significant absorption. These models have also been used in the first MERIS processing [Antoine and Morel, 1999]. [11] An updated set has been employed for the second MERIS processing, which kept the same maritime and coastal models but for four relative humidities (50%, 70%, 90% and 99%) and introduced rural aerosols slightly less absorbing than the previous urban ones. Another major point is the introduction of the so-called blue aerosol models (referred as BL hereafter). The latter was included to account for the strong spectral dependence of the AOT as reported by several authors [Behnert et al., 2003; Santer and Lemire, 2004]. [12] Over land, the aerosol models from Santer et al. [1999] were employed. Their size distribution n(r) follows a Junge power law, nðrÞ ra3 ;

ð4Þ

with r the particle radius and a the Angstro¨m exponent which describes the wavelength dependence of the AOT (t a(l)), aðl; l0 Þ ¼

where ratm stands for the intrinsic atmospheric contribution (multiple scattering contributions from the molecules, the aerosols and the Rayleigh-aerosol coupling, as well as from the coupling between atmospheric scattering and Fresnel surface reflection), T the total atmospheric transmittance (Rayleigh + aerosol) exclusive of gaseous absorption, S the spherical albedo relating to the molecules and the aerosols, t the total optical thickness (Rayleigh + aerosol), M the air mass, and rw and rG, respectively the water-leaving radiance reflectance and the direct sun glint reflectance. The Fresnel reflection is accounted for in the computation of the coupling terms between Fresnel reflection and atmospheric scattering. [8] From space, the scattering angular domain is defined in [90 – 180] deg. For estimating the atmospheric path radiance, the aerosol phase function (Pa(Q)) has to be defined in this angular range. The total atmospheric transmittance is associated with the forward scattering region: Q 2 [0 – 90] deg. [9] Over case-1 waters as defined in the MERIS ocean product, rw is considered to be null both at 778.75 nm and 865 nm. Outside of the sun glint area, the signal at TOA corresponds to the intrinsic atmospheric scattering only. Consequently, from the two atmospheric path radiances (acquired at 778.75 nm and 865 nm) we can retrieve the

D14208

logðt a ðlÞ=t a ðl0 ÞÞ : logðl=l0 Þ

ð5Þ

Four values of a have been implemented in the MERIS level-2 processing: 1.5, 2, 2.5, 3. The aerosol refractive index m was set up to 1.44 as the default value. [13] For each aerosol model, the Mie’s theory is used to compute in 13 MERIS spectral bands (the two strong O2 and H2O absorption bands, respectively at 761.875 nm and 900 nm, have been discarded) the following optical properties: the extinction coefficient (sa) and the scattering coefficient or the single scattering albedo (wa); the phase function (Pa(Q)) as function of the scattering angle (Q), which is completed using a Gauss quadrature with 80 angles. [14] Actually, to generate the LUTs required for the atmospheric corrections over ocean, we calculated for each aerosol type and each of the 13 MERIS wavelengths: the ratio of extinction coefficient at l and a reference wavelength selected at 865 nm: sa(l)/sa(865); the single scattering albedo times the phase function: wa(l).Pa(Q,l). Table 1 summarizes the a values for the SAMs employed in the second MERIS level-2 processing. 1.3. Validation of the Aerosol Product and of the Water-Leaving Radiances [15] The MERIS aerosol product consists in the AOT at 865 nm and the Angstro¨m exponent (a). Ground-based

2 of 16


D14208

Table 1. Spectral Dependence of the MERIS Aerosol Models in the NIR Region Model

M50

M70

M90

M99

C50

C70

C90

C99

a(665,865)

0.51

0.41

0.20

0.08

0.78

0.66

0.40

0.22

Model

R50

R70

R90

R99

BL1

BL2

BL3

a(665,865)

1.67

1.66

1.51

1.30

1.50

2.00

2.50

extinction measurements allow to derive these two optical quantities. Thanks to the availability of a large number of CIMEL measurements in the AERONET (AErosol RObotic NETwork) database [Holben et al., 1998], an extensive set of CIMEL data could be used for the validation of the MERIS products. [16] The procedure is as follows: we first extract from the TOA signal the normalized aerosol path radiance (L(1) a ) at 865 nm and 778.75 nm in the like primary scattering approximation. (L(1) a ) is expressed as, Lða1Þ ðm; lÞ ¼

t a ðlÞ wa ðlÞ Pa ðQ; lÞ: 4m

ð6Þ

Therefore, to get the AOT, we need to know the single scattering albedo times the phase function (wa.Pa(Q)). The latter is computed thanks to the use of the SAMs and a specific identification of this aerosol model through the wavelength dependence of (L(1) a ). Equation (6) is introduced here to illustrate the use of the phase function. Actually, the inversion scheme is based on the radiative transfer computations with the same requirement to include wa(l) Pa(Q, l). [17] If the maritime model rules the open ocean, it directly implies that the AOT can be accurately derived from L(1) a . Ignatov [1997] reported that the maritime aerosol model is suitable to retrieve the AOTs from the AVHRR (Advanced Very High Resolution Radiometer) imagery. Moreover, in the case of MODIS (MODerate Resolution Imaging Spectroradiometer), Remer et al. [2002] stressed that the aerosol product over ocean is quite well retrieved, at least with the expected accuracy. For people from the aerosol remote sensing community, they are always pleased by these performances obtained over case-1 waters (which represent the easiest case), whereas the oceanographers generally do not complain of the atmospheric corrections over ocean. Several reasons can explain this satisfaction in the aerosols retrieval over ocean: a good occurrence of the well known maritime aerosol model, a good meteorological visibility, and a self consistency of the radiometric calibration and the atmospheric correction (i.e., the SeaWiFS radiometric calibration is consistent with the formulation of the signal employed in the atmospheric correction scheme and with the use of the maritime aerosol model). [18] Over coastal waters, the validation of the AOTs derived from MERIS and SeaWiFS data is less favorable. Several previous publications clearly stressed that MERIS overestimates the AOT at 865 nm [Zagolski et al., 2005; Zibordi et al., 2002]. The aerosol retrieval is more difficult because of the increase of the aerosol load as well as the larger complexity of the aerosol models. In order to improve the analysis of SeaWiFS imagery, Sturm and Zibordi [2002] included their own aerosol phase function derived from sky radiance measurements. The question of the ability for the

D14208

SAMs to well describe the phase function was raised. A first evaluation of the MERIS SAMs was discussed in Zagolski et al. [2005]. The approach was to identify one SAM from the measured spectral dependence of the AOT, use this SAM to predict the sky radiance in the principal plane at 870 nm, and to compare it with the CIMEL measurement. This study confirmed that the SAMs could cause the large discrepancies observed in the AOT retrieval from the MERIS data. [19] Of course, the systematic over-estimation of the AOT at 865 nm with MERIS directly impacts on the retrieval of the water-leaving radiances which are quite often negative in the blue region over coastal areas. [20] Instead to rely on the micro-physical properties of the aerosol models and to compute their optical properties, the idea is then to build-up a climatology based on the inherent optical properties (IOPs) of aerosols, rather than to rely on the micro-physical properties of the aerosol models to compute their optical properties.

2. Aerosol IOPs Derived From CIMEL Measurements [21] The CIMEL E-318 sun-photometer is fitted with four filters at 1020 nm, 870 nm, 675 nm and 440 nm for measuring atmospheric aerosol optical properties. This instrument presents three scanning protocols: (1) sun, (2) almucantar, and (3) principal plane (see Holben et al. [1998] for more details). The AOTs are derived from the extinction measurements. Sky radiances are collected in the principal plane (PPL). 2.1. CIMEL Database [22] Twenty-five AERONET sites were selected over the world (see map on Figure 1) from years 1998 to 2003. The table in Figure 1 indicates for each site the principal investigators (PIs), the period of data acquisition, the number of data (AOTs and sky radiances) sequences we downloaded from the AERONET web-site (N0) and analyzed (N1). N0 corresponds to the number of PPL acquisitions (larger than 90000 measurements) which yields to a number of AOTs larger than 500000 for the same period. N1 is the number of sequences for which the solar zenith angle (qs) is larger than 65 deg., and the AOTs and the sky radiances are correctly measured and recorded at 440 nm, 675 nm and 870 nm. The total of N1 sequences is 7109. [23] We firstly derived the AOTs and the spectral dependence of the extinction coefficient. Then, we followed the inverse method of the sky radiance proposed by Santer and Martiny [2003] to derive the aerosol single scattering albedo times the phase function (wa.Pa(Q)). This method has been applied for several days on the in-flight calibration of MERIS [Martiny et al., 2005] and of SeaWiFS [Martiny et al., 2006] in the red and near-infrared spectral domains. The increasing role of the Rayleigh scattering emphasizes the influence of the aerosol vertical distribution and this parameter has been introduced in the inversion scheme. [24] What we used here is an updated version of this algorithm [Zagolski and Santer, 2005], which first brings improvements in the radiative transfer computations based on the use of the successive orders (SO) of the scattering code [Deuze´ et al., 1989]. Second, the vertical distribution

3 of 16

D14208


D14208

Figure 1. Map of the 25 selected AERONET sites. Name of the site, principal investigators (PIs), period of CIMEL measurements, number of CIMEL data sequences (AOTs and sky radiances) collected for each site (N0) and processed (N1) for retrieving aerosol optical properties are detailed in table on the right side. of the aerosols is introduced using an exponential law with a vertical scale height (Ha). [25] The inputs for the nominal inversion are the followings: an aerosol scale height (Ha) of 3 km; a dark water surface and a Fresnel reflection driven by the Cox and Munk [1954] wave slope distribution, induced by a windspeed of 7.2 m/s; an iterative process initiated with a Junge phase function corresponding to a = 1, and a refractive index m = 1.44 without any absorption. [26] These sky radiances (N1 = 7109) have been processed in order to derive the aerosol single scattering albedo times the phase function (wa.Pa(Q)). 2.2. Outputs From the Data Processing [27] What we derived from the CIMEL measurements are listed hereafter: (1) the AOTs (t a) at 440 nm, 675 nm and 870 nm, from which we can get sa(l)/sa(865) using the Angstro¨m exponent (a) determined either between 440 nm and 675 nm, or 675 nm and 870 nm; (2) the aerosol single scattering albedo times the phase function (wa.Pa(Q)) at 675 nm and 870 nm for 80 scattering angles. Because the phase function is normalized, its angular integration yields to the single scattering albedo. [28] A data filtering has been applied on the initial set of 7109 sequences. Table 2 displays the number of CIMEL sequences we selected after applying the following filters: [29] 1. The sky radiance at the scattering angle of 6 deg. is measured twice with two different gains. Actually, the sun collimator is selected for Q lower than 6 deg. whereas the sky collimator is used for Q larger than 6 deg. If the two radiances differ by more than 10%, then the sequence is

rejected. If not, only the sky collimator measurement is considered. Thus, the validation of this angular smoothness on wa.Pa(Q) allowed to discard a substancial number of sequences. [30] 2. A threshold is applied on the Angstro¨m exponent a(440, 870) computed between 440 nm and 870 nm. The latter is ranged between 2.5 (corresponding to the small particle sizes) and 0.5 (for the large particle sizes). [31] 3. The aerosol single scattering albedo at 675 nm (wa(675)) is bounded between 0.6 and 1.05, which allows to account for the errors resulting from the calibration uncertainties. [32] 4. The aerosol single scattering albedo at 870 nm (wa(870)) is included within 0.6 and 1.05. [33] 5. Both wa(675) and wa(870) are ranged within 0.6 and 1.05.

3. First Analysis of the CIMEL Database 3.1. Spectral Dependence of the AOT [34] The spectral dependence of the AOT between 675 nm and 870 nm allows us to select an aerosol model. The AOT Table 2. Number of CIMEL Phase Sequences Selected Among an Initial Set of 7109, After Applying Several Steps of Data Filteringa Case

Smoothness

a(440,870)

wo(675)

wo(870)

wo(675) + wo(870)

N

3157

2685

2388

2268

2171

a

(1) An angular smoothness on wa.Pa(Q), (2) 2.5 a(440,870) 0.5, (3) 0.6 wo(675) 1.05, (4) 0.6 wo(870) 1.05, and (5) 0.6 wo(675) 1.05 and 0.6 wo(870) 1.05.

4 of 16


D14208

D14208

in which, the incident direction is defined by (m, f) and the diffused direction by (m0, f0), df corresponding to the difference in azimuth between f and f0. Because equation (8) is usually employed in most atmospheric correction algorithms, the derivation of Fa from the sky radiance measurements is of concern. [37] Figure 4a illustrates the angular dependence of Fa(m, 675). Figures 4b and 4c display Fa(m, 675) as function of a, respectively for a zenith angle (q) of 4 deg. and 60 deg. The spectral dependence of Fa(m, l) versus a is depicted on Figure 4d.

4. A New Aerosol Climatology Based on the IOPs

Figure 2. Spectral dependence between a(440, 870) and a(675, 870). at 870 nm has to be then extrapolated to the blue region. Figure 2 relates the spectral dependence between a(440, 870) and a(675, 870). As expected, for small values the plot is scattered (near zero division). As mean, we obtain, að440; 870Þ=að675; 870Þ ¼ 0:95

ð7Þ

Equation (7) suggests that we need to introduce a dynamic Angstro¨m exponent to derive the AOT in the visible range from the AOT at 870 nm. 3.2. Spectral Dependence of the Single Scattering Albedo [35] We check here if the outputs of our inversion scheme fall down in the expected geophysical range. Figure 3 displays the single scattering albedo values derived from the analysis of all sky radiance measurements at 870 nm. The distribution of wa(870) versus a(440, 870) is quite scattered because of the poor accuracy on this determination. No trend appears with a. Both for the two wavelengths at 675 nm and 870 nm, the average value of a is around to 1.1 and the mean values of wa are 0.899 and 0.883 respectively.

4.1. Classification of Phase Functions [38] The influence of the AOT on the aerosol phase function retrieval is illustrated in Figure 5. Results are depicted for a = 1.5 ± 0.1 both at 675 nm and 870 nm and for four scattering angles (90 deg., 110 deg., 130 deg. and 150 deg.). We can note that no trend appears on the phase function versus the AOT when the plots are scattered. Consequently, the robustness of the inversion scheme is checked within the range of the encountered aerosol loads. This analysis has been deeply investigated for the full range of a and the results confirmed what we obtained for a = 1.5. [39] Figure 6 displays variation of the aerosol phase function at two scattering angles (90 deg. and 130 deg.) with a. Even if the plots are scattered, these results indicate expected trends. As an example, the phase function at 90 deg. normally increases when the particle size become smaller as expected by the Rayleigh theory [Lenoble, 1993]. [40] The selection of one aerosol type is based on the a value. As the IOPs of the aerosols appear to be continuous versus a(440, 870), we decided to characterize them by classes of a. The latters have been defined as follows: a is within [2.5; 0] and selected by step of 0.1 (except for the last value which is fixed to 0.05) with a Da of 0.2 (resp., 0.1 for the last value of a) centred on the nominal value. For these 25 classes, in the Q range of [0 150] deg., the

3.3. Aerosol Transmittance [36] According to Gordon and Wang [1994], the aerosol transmittance Ta(m, l) is expressed as, t a ðlÞ Ta ðm; lÞ ¼ exp ð1 wa ðlÞ Fa ðm; lÞÞ ; m

ð8Þ

where Fa(m, l) represents the aerosol forward scattering proportion. The latter is computed using the retrieved aerosol phase function Pa(Q, l) as,

Fa ðm; lÞ ¼

1 4p

Z1 Z2p 0

0

Pa ðQ; lÞ dm0 df;

ð9Þ

Figure 3. Single scattering albedo (wa) at 870 nm versus the Angstro¨m exponent a(675, 870). 5 of 16

D14208


Figure 4. (a) Aerosol forward scattering proportion Fa(qs, 675) versus qs for three Angstro¨m exponents (a = 0.5, 1.0 and 1.5). Spectral dependence of Fa(qs, 675) for two solar zenith angles: (b) qs = 4° and (c) qs = 60°. Spectral dependence of Fa(qs, 675)/Fa(qs, 870) for two solar zenith angles: (d) qs = 4° and (e) qs = 60°.

6 of 16

D14208

D14208


Figure 5. Influence of the AOT on the retrieved aerosol phase function wa.Pa(Q). Results are reported for the class corresponding to hai = 1.5 ± 0.1, at 675 nm (left side) and 870 nm (right side) for three scattering angles: Q = 90° (a, d), Q = 110° (b, e), and Q = 130° (c, f).

7 of 16

D14208

D14208


D14208

Figure 6. Aerosol phase function wa.Pa(Q) at 675 nm (left side) and 870 nm (right side) expressed as function of a(440, 870), for two scattering angles: Q = 90° (a, c), and Q = 130° (b, d). retrieved aerosol phase functions at 675 nm and 870 nm have been averaged by class of a. Tables 3 and 4 give the number of phase functions (N0) per class of a. Moreover, a filtering technique needed to be applied on these phase functions for each class. For that, we computed the root mean square error (rms, denoted as s0) on the phase function for each Gaussian angle. A filtering has then been applied for each scattering angle at one s0, which reduced the number of phase functions to N1 for each class of a. This process was repeated twice. In Tables 3 and 4 are displayed the mean value of rms for scattering angles ranged within [100 –150] deg. After the second round, the rms value (s1) indicates the mean absolute accuracy on the retrieved aerosol phase function. 4.2. Accuracy of the IOPs [41] The initial work from Santer and Martiny [2003] reported some outputs for the sensitivity study based on

theoritical predictions. The phase function was retrieved at 870 nm only, for an homogeneous atmosphere. The salient point was that the relative error on the radiometric calibration results in the same amount of the relative error on the retrieved phase function. What we conducted here is a sensitivity study on the inversion of the full CIMEL database to the following individual parameters: CIMEL sky radiance measurements were multiplied by 1.03 to account for a systematic error of 3 percents in the downwelling radiance calibration; an aerosol scale height (Ha) of 1 km instead of 3 km; a wave slope distribution induced by a windspeed of 3 m/s instead of 7.2 m/s; an initial phase function computed with the Junge size distribution corresponding to a = 1, and a refractive index m = 1.33 instead of 1.44 (without any absorption). [42] Figure 7 displays the impact of the error introduced in the data calibration on the retrieval of the aerosol phase function (wa.Pa(Q)) at 675 nm and 870 nm for a scattering

8 of 16


D14208

D14208

Table 3. General Features of the Aerosol Phase Function Averaging at 675 nma hai

0.05

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

1.1

1.2

N0 N1 hs0i hs1i

70 65 0.18 0.10

182 169 0.17 0.08

215 200 0.17 0.08

188 178 0.15 0.07

184 173 0.10 0.06

167 157 0.10 0.06

146 136 0.12 0.06

153 140 0.13 0.06

139 130 0.11 0.05

154 144 0.08 0.05

190 178 0.06 0.04

209 195 0.07 0.04

217 204 0.09 0.05

hai

1.3

1.4

1.5

1.6

1.7

1.8

1.9

2.0

2.1

2.2

2.3

2.4

N0 N1 hs0i hs1i

231 218 0.08 0.04

247 233 0.09 0.05

253 238 0.09 0.05

262 248 0.08 0.04

263 249 0.07 0.04

254 239 0.06 0.04

200 188 0.06 0.04

132 126 0.06 0.04

84 79 0.05 0.04

41 37 0.04 0.03

30 28 0.06 0.04

12 11 0.08 0.03

a Here hai is the mean value of a, with Da = 0.2, N0 the number of phase functions per class of hai, and N1 the number of phase functions per class of hai after applying a filter at 2s on these phase functions.hs0i is the mean value of the rms computed on the phase function for each Q within [100 – 150] deg., and hs1i the mean of the rms determined on the filtered set of phase functions.

angle (Q) of 120 deg. For each of the two wavelengths, each individual wa.Pa(Q = 120, 3%) derived from downwelling radiance including the systematic error of 3 percents is plotted as function of its associated nominal one (i.e., without introducing an error). As expected, the results show that the plots are little scattered and remain very close to the bissector with a slight over-estimation in the case of the introduction of the systematic error. Table 5 reports the influence of this error in the data calibration on Fa(m, l) at two solar zenith angles (30 deg and 60 deg) and on wa.Pa(Q) for a scattering angle of 90 deg., 110 deg., 130 deg., and 150 deg. We first give the mean of the ratio R defined as, RðQ; l; calib:Þ ¼ Pa ðQ; l; 3%Þ=Pa ðQ; l; 0%Þ;

ð10Þ

and its associated rms value. [43] The same approach has been applied for the aerosol scale height, the wind-speed above sea level, and the refractive index of the aerosol model. All the results are reported in Table 6. The conclusions of this sensitivity study at 675 nm and 870 nm are mostly based on the analysis of R: [44] 1. The calibration of the sun-photometer is a key issue. Any multiplicative bias is reported in proportion on the aerosol phase function if we are close to the single scattering regime and if the Rayleigh scattering can be neglected which is more or less the case at 870 nm. At 675 nm, the situation is more complex to analyze but the multiplicative bias is still of the order of the calibration bias. [45] 2. It was expected to observe no impact from the initial guess of the aerosol phase function in the iterative

process which quickly converges. This is especially the case in the forward scattering on the Fa factor because first, the scattering regime is close to the primary one and second, the diffraction peak does not depend upon the particle refractive index. In the backscattering region the influence of the refractive index observed on the retrieval results from the non-knowledge of the phase function for scattering angles larger than 150 deg. The backscattering domain is sensitive to the refractive index and the result clearly illustrates the uncertainty due to the non-access to the forward scattering. Nevertheless, this effect remains quite small. [46] 3. The effect of the vertical distribution is negligible in the forward scattering region where the aerosol scattering dominates. This is still small in the backscattering region even at 675 nm where the influence of the Rayleigh scattering increases. On the other hand, this impact can be cancelled if the same aerosol vertical distribution is assumed both for this phase function retrieval and in the atmospheric correction algorithm. [47] 4. The larger the wind-speed, the larger the amount of reflected light is due to the fact that the direct sun glint invokes more grazing incident angles with high Fresnel reflection coefficient. The F factor is not much sensitive to the wind-speed because of the strong forward scattering. On the aerosol phase function, the errors of a few percents invite to account for the wind-speed in our analysis. 4.3. Spectral Dependence of the IOPs [48] To conduct this study, it is mandatory to well know the aerosol phase functions in the different spectral bands. Over ocean, the aerosol remote sensing algorithm is based on the use of the aerosol path radiances for the doublet

Table 4. Same as Table 3 but at 870 nm hai

0.05

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

1.1

1.2

N0 N1 hs0i hs1i

70 65 0.19 0.09

182 168 0.18 0.08

215 201 0.19 0.08

188 176 0.17 0.07

184 172 0.11 0.05

167 158 0.10 0.05

146 136 0.14 0.06

153 142 0.17 0.07

139 130 0.14 0.05

154 144 0.09 0.05

190 178 0.08 0.05

209 196 0.09 0.05

217 205 0.10 0.05

hai

1.3

1.4

1.5

1.6

1.7

1.8

1.9

2.0

2.1

2.2

2.3

2.4

N0 N1 hs0i hs1i

231 221 0.09 0.05

247 231 0.10 0.05

253 236 0.10 0.05

262 247 0.10 0.05

263 250 0.11 0.05

254 240 0.08 0.05

200 190 0.08 0.05

132 124 0.07 0.05

84 79 0.07 0.05

41 39 0.07 0.06

30 29 0.10 0.08

12 11 0.13 0.08

9 of 16


D14208

D14208

(670 nm and 870 nm) or (770 nm and 865 nm). CIMEL measurements are available at 675 nm and 870 nm. Assuming that the spectral dependence on the measurements remains quite smooth in this near-infrared (NIR) region, the radiance at 770 nm can then be easily obtained by applying a spectral interpolation. Figure 8 depicts the ratio of wa.Pa(Q) computed between 675 nm and 870 nm. Firstly, we can note that in the case where all the individual measurements are considered (Figure 8a), these ratios are well peaked around the one value, but nevertheless they remain quite scattered. Secondly, by averaging wa.Pa(Q) by class of a (Figure 8b) this dispersion is strongly reduced. Then, we accounted for the particle size, through a, by suggesting a spectral dependence on wa.Pa(Q) for different scattering angles. This is illustrated in Figure 8c which clearly stresses a trend on the ratio of wa.Pa(Q) with a. [49] When dealing with the atmospheric corrections for ocean colour, we need to know the phase function in the blue region. As reported by R. Santer et al. (Iterative process to derive the phase function from CIMEL measurements, submitted to International Journal of Remote Sensing, 2007), the retrieval of the aerosol phase function (wa.Pa(Q)) is more challenging and some limitations clearly occur for large AOTs and low solar elevations. Despites this, we launched the inversion scheme in the blue region on the full data set to finally retrieve wa.Pa(Q) at 440 nm. The same classification on a was achieved and the results are reported in Table 7. The spectral dependence on wa.Pa(Q) between 440 nm and 675 nm is illustrated on Figures 8d– 8f). The significance of these results is more questionable because of the high dispersion (some points are out of scale). Consequently we suggest to extrapolate the results from 675 nm and 870 nm for estimating wa.Pa(Q) at 440 nm.

Figure 7. Impact induced by an error of 3 percents introduced in the CIMEL data calibration on the retrieval of the aerosol phase function wa.Pa(120°) at 675 nm (a) and 870 nm (b).

4.4. Comparison Between Simulated and Measured Phase Functions [50] Two sets of aerosol phase functions were available: the first one was computed for the SAMs and the second one resulted from the CIMEL sky radiance analysis. Figure 9 displays the comparison between these two sets of phase functions at 870 nm for some samples of aerosol models

Table 5. Outputs From the Sensitivity Study at 675 nm on wa.Pa(Q) for Several Scattering Angles (Q) and on Fa(qs, l) for Two Solar Zenith Angles (qs)a d calib.(%) hRcalib.i scalib. d m(%) hRmi sm d Ha(%) hRHai sHa d ws(%) hRwsi sws

wa.Pa(90°)

wa.Pa(110°)

wa.Pa(130°)

wa.Pa(150°)

Fa(30°)

Fa(60°)

9.0 1.035 0.024 3.5 0.998 0.003 2.0 1.011 0.011 4.4 0.981 0.012

16.5 1.063 0.038 9.1 0.983 0.013 5.9 1.015 0.015 7.0 0.974 0.016

21.5 1.048 0.037 11.4 0.983 0.008 7.6 1.063 0.016 12.8 0.965 0.018

10.8 1.030 0.030 8.1 1.005 0.012 5.9 0.968 0.015 8.0 0.966 0.013

10.0 1.032 0.018 3.5 1.001 0.005 0.7 1.000 0.006 1.4 0.992 0.005

18.4 1.037 0.023 4.8 1.003 0.008 1.1 1.003 0.008 1.5 0.991 0.006

a Here d is the relative error between wa.Pa(Q) retrieved from downwelling radiances including error induced by either the data calibration (calib.), or the aerosol refractive index (m), or the aerosol scale height (Ha), or the wind-speed ASL (ws), and its nominal value. hRi represents the mean value of the ratio between retrieved and nominal phase functions, and s its associated rms error.

10 of 16


D14208

D14208

Table 6. Same as Table 5 but at 870 nm d calib.(%) hRcalib.i scalib. d m(%) hRmi sm d Ha(%) hRHai sHa d ws(%) hRwsi sws

wa.Pa(90°)

wa.Pa(110°)

wa.Pa(130°)

wa.Pa(150°)

Fa(30°)

Fa(60°)

14.9 1.036 0.026 10.2 0.988 0.009 1.6 1.006 0.008 3.5 0.983 0.010

8.3 1.027 0.022 5.7 1.007 0.014 1.8 0.988 0.008 4.8 0.978 0.012

4.8 1.028 0.019 1.4 0.999 0.003 0.5 0.997 0.003 8.7 0.972 0.013

4.8 1.027 0.021 1.5 0.999 0.003 0.5 0.997 0.003 5.3 0.967 0.011

5.3 1.027 0.013 1.8 1.000 0.004 0.3 0.999 0.001 1.3 0.993 0.004

5.3 1.027 0.013 1.9 0.999 0.004 0.3 0.999 0.001 1.4 0.992 0.004

(maritime, rural, coastal and land). We added the relative errors between the measured and computed phase functions. The range of scattering angle observed with MERIS is comprised within [100 – 150] deg. What we can observe is that the predicted aerosol phase function may underestimate by a factor of two compared with the measured one. Only the Junge size distribution which is employed for simulating the so-called blue aerosols, well compares. Chomko and Gordon [1998] proposed to use Junge size distributions for the atmospheric corrections over the ocean, which appears here to be a reasonable alternative to the MERIS SAM. 4.5. Consequences on the MERIS Aerosol Product [51] In the framework of MERIS activities, Zagolski et al. [2005] reported some validations of the aerosol product over ocean based on the use of CIMEL data. Results are summarized in Table 8. The database consisted of nearly simultaneous sky radiance measurements and MERIS images acquired over five test sites from AERONET at the closest scattering angles for 14 validation days [Martiny et al., 2005]. From ground-based measurements, we could then directly compare the AOTs at 870 nm with those derived from the MERIS data at 865 nm. In fact, for each of these validation days, a(675, 870) is computed with the CIMEL sequence nearest to the time of MERIS overpass and then compared with a(665, 865) derived from the MERIS acquisitions. [52] According to the linear regression displayed on Figure 10, good agreements can be observed on a for the second MERIS processing. This can be explained by the fact that we do not require any assumption on the SAMs as far as the spectral dependence of wa.Pa(Q) can be neglected. This excellent agreement reflects the ability to properly extract the aerosol path radiance over dark ocean surfaces. Note that with a deeper analysis, except for Venice site (August 13th, 2003), the MERIS aerosols appear whiter than they should be. Poor results obtained with the first MERIS processing are surely due to the fact that the initial family of SAMs is not representative of the full range of observed Angstro¨m exponents. Consequently, the algorithm forces to retrieve bad SAMs. Table 8 stresses that a is not actually correlated with the AOT at 865 nm. The retrieval of AOTs from MERIS measurements remains quite poor both for the two processings. MERIS systematically overestimates the AOT at 865 nm by a factor around to 2.

[53] As a first attempt to introduce our new aerosol climatology in the MERIS atmospheric correction algorithm, we assumed that the latter correctly extracts from the TOA signal the normalized aerosol path radiance at 865 nm in the primary scattering approximation. [54] In order to use our IOPs to generate the new corrected AOTs at 865 nm, we processed with these following steps: [55] 1. The Angstro¨m exponent (aACRI) derived from the MERIS level-2 (provided by ACRI) allows us to select two boundering SAMs using the MERIS aerosol models (see Table 1). [56] 2. The optical properties of the two boundered aerosol models (wa.Pa(Q))SAM1(2) employed in the MERIS level-2 are obtained using aACRI. To determine the optical properties of the corresponding MERIS level-2 SAM, an extrapolation is performed as, ½wa :Pa ðQÞSAM ¼

aACRI aSAM1 aSAM 2 aSAM1 ½wa :Pa ðQÞSAM2 ½wa :Pa ðQÞSAM 1 þ ½wa :Pa ðQÞSAM 1 :

ð11Þ

[57] 3. aACRI allows us to select our IOPs data (wa.Pa(Q))IOP at the same MERIS level-2 scattering angle (QACRI). [58] 4. According to equation (6), we get a new value of the AOT at 865 nm which will be consistent with the new aerosol IOPs, ½t a ðlÞIOP ¼

½t a ðlÞ wa ðlÞ Pa ðQ; lÞSAM ½wa ðlÞ Pa ðQACRI ; lÞIOP

ð12Þ

Figure 11 compares the AOT at 665 nm derived from MERIS in the two ways (i.e., with the SAMs and with the IOPs) and from the CIMEL measurements at 675 nm over the Lampedusa site for several validation days. A linear regression applied on each of the two data sets indicates that the MERIS AOT estimate is slightly improved with the IOPs. However, this improvement was not observed in the NIR region over the same site, for which the MERIS standard aerosol product (AOT at 865 nm) is much more overestimated than the one at 665 nm. Although these results are not conclusive (this study needs to be applied on a more extensive data set for the validation), this

11 of 16

D14208


Figure 8. Histogram of ratio of wa.Pa(Q) computed between 675 nm and 870 nm (left side), and 440 nm and 675 nm (right side) for four scattering angles (Q = 90°, 110°, 130°, and 150°). Results depicted in figures (a, d) derive from all individual CIMEL measurements, whereas figures (b, e) are for the averaging of wa.Pa(Q) by classes of a. Figures (c, f) display the spectral dependence of this ratio.

12 of 16

D14208


D14208

D14208

Table 7. Same as Table 3 but at 440 nm hai

0.05

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

1.1

1.2

N0 N1 hs0i hs1i

-

27 25 0.14 0.07

46 42 0.11 0.08

57 53 0.12 0.09

65 60 0.13 0.10

69 65 0.11 0.09

69 65 0.11 0.09

74 69 0.12 0.09

71 67 0.12 0.09

83 79 0.09 0.07

116 109 0.09 0.07

136 127 0.10 0.07

151 140 0.10 0.07

hai

1.3

1.4

1.5

1.6

1.7

1.8

1.9

2.0

2.1

2.2

2.3

2.4

N0 N1 hs0i hs1i

178 168 0.09 0.06

204 195 0.09 0.06

197 187 0.09 0.06

193 183 0.09 0.06

196 185 0.10 0.06

190 178 0.09 0.06

157 150 0.09 0.07

86 81 0.08 0.06

40 37 0.06 0.04

14 13 0.06 0.04

-

-

methodology suggested for validating the impact of the introduction of new aerosol IOPs on the aerosol product easily applies.

5. Conclusions [59] Aerosol remote sensing and atmospheric schemes require to rely on SAMs which enable to know the standard

aerosol IOPs. The AERONET offers a unique opportunity to describe these IOPs. The spectral dependencies of the AOTs are directly available from the extinction measurements. An algorithm to derive the aerosol phase function from the sky radiance measurements was available to us. Our major task was to select and to process a significant amount of CIMEL sky radiances. For this database, we

Figure 9. Comparison between aerosol phase functions wa.Pa(Q) at 870 nm derived from the CIMEL measurements and from the MERIS SAMs: (a) maritime model (RH = 99%), (b) coastal model (RH = 90%), (c) rural model (RH = 90%), and (d) blue aerosol model (Junge). Relative differences (expressed as percent) computed between CIMEL derived phase function and SAM are plotted by crosses. 13 of 16


D14208

D14208

Table 8. AOTs at 865 nm (MERIS)/870 nm (CIMEL), and a Derived From the CIMEL Acquisitions and From the Two MERIS Processings First Processing

CIMEL

Second Processing

Site

Date

AOT

a

AOT

a

AOT

a

Venise Venise Lampedusa Gotland Gotland Helgoland Lampedusa Venise Venise Calcofi Venise Calcofi Venise Venise

10/08/2003 13/08/2003 20/08/2003 07/09/2003 08/09/2003 16/09/2003 21/09/2003 16/10/2003 25/10/2003 28/10/2003 04/11/2003 10/11/2003 10/11/2003 13/11/2003

0.22 0.18 0.24 0.07 0.06 0.22 0.13 0.10 0.12 0.14 0.11 0.14 0.13 0.15

1.111 1.489 0.567 0.673 1.052 0.442 0.665 0.950 1.585 1.497 0.673 0.205 0.629 0.532

0.29 0.22 0.21 0.04 0.06 0.15 0.06 0.05 0.05 0.06 0.05 0.06 0.04 0.11

0.870 1.780 0.990 0.730 0.710 1.150 1.860 0.670 1.200 1.600 1.370 0.360 0.460 1.580

0.20 0.08 0.20 0.09 0.07 0.16 0.07 0.08 0.09 0.07 0.07 0.08 0.16

0.993 2.439 0.587 0.423 0.751 1.252 0.588 0.811 1.653 0.990 0.529 0.778 1.284

proposed a set of aerosol IOPs both for the aerosol remote sensing and the atmospheric correction. [60] Limitations to this work are well identified. We blindly believe in the accurate calibration of the CIMEL instrument. The CIMEL data were processed using a radiative transfer code associated with basic assumptions (a plane parallel atmosphere even if low solar elevations and grazing view angles are involved, and an homogeneous system even if some CIMEL stations are located on the coastline which may imply the so-called adjacency effects in the measurements) and with some standard inputs (i.e., a barometric pressure of 1013.25 hPa, a standard climatologic ozone content as function of date and location to correct for the ozone absorption at 675 nm, and a standard wind-speed of 7.2 m/s above sea level). These inputs can be improved by collecting meteorological data. Biases due to the extreme geometrical conditions or to the eventual presence of adjacency effects could be tackled if a larger database was

Figure 10. MERIS versus CIMEL derived Angstro¨m exponent a(440, 675).

generated in order to conduct a statistical investigation on these parameters. [61] All the analyses have been completed in the principal plane let say by tradition and not by conviction, because the CIMEL measurements in the almucantar geometry should also be used to study specific parameters such as the vertical distribution of particles, the adjacency effects and the geometrical conditions. [62] In addition, the spectral extrapolation of the IOPs from the red to the visible region needs to be validated. The strategy would consist in predicting the sky radiances at 440 nm and in comparing them with the CIMEL measurements. [63] The angular extrapolation of the optical properties to the near backscattering region has to be also checked. This could be achieved by using MERIS data in the angular scattering domain [150– 180] deg. and in the NIR region (over black sea surface). Simultaneous CIMEL data would indicate what will be the new aerosol model, and TOA radiances simulated with this model would be then com-

Figure 11. MERIS AOT at 665 nm (from standard aerosol product and from correction for IOPs) versus CIMEL AOT at 675 nm over the Lampedusa site.

14 of 16

D14208


pared with the MERIS level-1b radiances. As far as we can rely on the MERIS radiometric calibration, this comparison would validate the IOPs at least in the red and NIR spectral regions. It can be envisaged to use the MERIS backscattering measurements combined with the CIMEL sky radiances in order to retrieve the aerosol phase function at the scattering angles which are not accessible from the ground. Other possibilities for estimating the aerosol phase function would be to use simultaneous CIMEL data sequences with measurements from multi-view sensors such as POLDER (POLarization and Directionality of Earth’s Reflectances), MISR (Multi-angle Imaging Spectro-Radiometer) or AATSR (Advanced Along-Track Scanning Radiometer). [64] We conducted a first attempt to evaluate our approach through comparisons between AOTs simultaneously derived from MERIS and CIMEL. Although some preliminary results obtained near the Lampedusa site stress a better estimate of the aerosol product at 665 nm with the IOPs, they need to be deeply investigated on a more extensive data set for the validation. [65] We plan to evaluate the improvements brought by replacing the existing micro-physical aerosols climatology by our IOPs derived models. These new IOPs will be then employed to generate new LUTs to be implemented for testing in the current MERIS algorithms. The latter will allow to fully validate our approach. Of course, the general approach used here over ocean can be duplicated as well over land.

Notation AATSR Advanced Along-Track Scanning Radiometer (http://www.leos.le.ac.uk/aatsr/). ACRI Groupe ACRI – Me´ canique Applique´ e – Observation de la Terre et Sciences de l’Environnement (Sophia-Antipolis, France) (http://www.acri.fr/). AERONET AErosol RObotic NETwork (http://aeronet. gsfc.nasa.gov/). AOT Aerosol Optical Thickness. ASL Above Sea Level. AU Astronomic Unit. AVHRR Advanced Very High Resolution Radiometer (http://edc.usgs.gov/products/satellite/avhrr. html). CCD Charge-Coupled Device. CIMEL Portable tracking sun-photometer designed to conduct atmospheric studies (http://www. cimel. fr/photo/sunph_us.htm). ECMWF European Centre for Medium range Weather Forecast (http://www.ecmwf.int/). ENVISAT ESA satellite (http://envisat.esa.int/). ESA European Space Agency (http://www.esa.int/). IOP Inherent Optical Properties. LISE Laboratoire Interdisciplinaire des Sciences de l’Environnement (Wimereux – France). LUT Look-Up Table. MERIS MEdium Resolution Imaging Spectrometer (http://envisat.esa.int/instruments/meris/). MISR Multi-angle Imaging SpectroRadiometer (http://www-misr.jpl.nasa.gov/).

D14208

MODIS MODerate Resolution Imaging Spectroradiometer (http://modis.gsfc.nasa.gov/). NIR Near-InfraRed. POLDER POLarization and Directionality of Earth’s Reflectances (http://smsc.cnes.fr/POLDER/ Fr/). PPL Principal Plane. RH Relative Humidity (expressed as percent). rms Root mean square error. SAM Standard Aerosol Model. SeaWiFS Sea-viewing Wide Field-of-view Sensor (http://oceancolor.gsfc.nasa.gov/SeaWiFS/). SO Successive Orders of the scattering code. TOA Top of Atmosphere. ULCO Universite´ Littoral Coˆte d’Opale (WimereuxFrance) (http://www.univ-littoral.fr/). [66] Acknowledgments. This work has been supported by the European Space Agency (ESA). The authors would like to thank all the AERONET PIs for their assistance in the use of the different CIMEL instruments. The authors are also grateful to Je´roˆme Vidot for his help in the CIMEL data extraction, and to Constant Mazeran from ACRI for providing the database with the comparison between MERIS and CIMEL derived AOTs.

References Antoine, D., and A. Morel (1999), A multiple scattering algorithm for atmospheric correction of remotely sensed ocean color (MERIS instrument): Principle and implementation for atmospheres carrying various aerosols including absorbing ones, Int. J. Remote Sens., 20(9), 1875 – 1916. Behnert, I., R. Doerffer, G. Becu, P.-Y. Deschamps, and N. Fomferra (2003), Aerosol retrieval from MERIS and ground-based radiometers in the German bight, turbid coastal waters, paper presented at ENVISAT/ MERIS User Workshop, Frascati, Italy, 10 – 13 Nov. Chomko, R., and H. R. Gordon (1998), Atmospheric correction of ocean color imagery: Use of the Junge power-law aerosol size distribution with variable refractive index to handle aerosol absorption, Appl. Opt., 37, 5560 – 5572. Cox, C., and W. Munk (1954), Measurements of roughness of the sea surface from photographs of the sun glitter, J. Opt. Soc. Am., 44(11), 838 – 888. Deuze´, J. L., M. Herman, and R. Santer (1989), Fourier series expansion of the transfer equation in the atmosphere-ocean system, J. Quant. Spectrosc. Radiat. Transfer, 41(6), 483 – 494. Gordon, H. R., and M. Wang (1994), Retrieval of water-leaving radiances and aerosol optical thickness over the oceans with SeaWiFS: A preliminary algorithm, Appl. Opt., 33(3), 443 – 452. Holben, B., et al. (1998), AERONET – A federated instrument network and data archive for aerosol characterization, Remote Sens. Environ., 66, 1 – 16. Ignatov, A. (1997), Estimation of the aerosol phase function in backscatter from simultaneous satellite and sun-photometer measurements, J. Appl. Meteorol., 36, 688 – 694. Lenoble, J. (1993), Atmospheric Radiative Transfer, 532 pp., A. Deepak, Hampton, Va. Martiny, N., R. Santer, and I. Smolskaia (2005), Vicarious calibration of MERIS over dark waters in the near-infrared, J. Remote Sens. Environ., 94(4), 475 – 490. Martiny, N., R. Frouin, and R. Santer (2006), Vicarious calibration of SeaWiFS in the near-infrared using the dark ocean, Appl. Opt., 44(36), 7828 – 7844. Merheim-Kealy, P., J. P. Huot, and S. Delwart (1999), The MERIS groundsegment, Int. J. Remote Sens., 20, 1703 – 1712. Rast, M., J. L. Be´zy, and S. Bruzzi (1999), The ESA MEdium Resolution Imaging Spectrometer MERIS: A review of the instrument and its mission, Int. J. Remote Sens., 20, 1681 – 1702. Remer, L. A., et al. (2002), Validation of MODIS aerosol retrieval over ocean, Geophys. Res. Lett., 29(12), 8008, doi:10.1029/2001GL013204. Santer, R., and F. Lemire (2004), Validation of MERIS atmospheric corrections: Towards a new aerosol climatology, paper presented at ENVISAT/ MERIS-AATSR (ESA/ESRIN), Salzburg, Austria, 6 – 10 Sept. Santer, R., and N. Martiny (2003), Sky radiance measurements for ocean colour calibration-validation, Appl. Opt., 42(6), 896 – 907.

15 of 16

D14208


Santer, R., V. Carre`re, P. Dubuisson, and J. C. Roger (1999), Atmospheric corrections over land for MERIS, Int. J. Remote Sens., 20(9), 1819 – 1840. Shettle, E. P., and R. W. Fenn (1979), Models for the aerosols of the lower atmosphere and the effects of humidity variations on their optical properties, Air Force Geophys. Lab. Tech. Rep. AFGL-TR-79-0214, Hanscom Air Force Base, Mass. Sturm, B., and G. Zibordi (2002), SeaWiFS atmospheric correction by an approximate model and vicarious calibration, Int. Remote Sens, J., 23, 489 – 501. Zagolski, F., and R. Santer (2005), Retrieval of aerosol phase functions from sky radiance measurements in the framework of MERIS aerosol remote sensing validation, paper presented at ENVISAT/MERIS-AATSR (ESA/ESRIN), Frascati, Italy, 26 – 30 Sept.

D14208

Zagolski, F., R. Santer, J. Vidot, and F. Thieuleux (2005), Validation of the MERIS atmospheric correction over water using ground-based measurements of the solar extinction and of the sky radiances, paper presented at ENVISAT/MERIS-AATSR (ESA/ESRIN), Frascati, Italy, 26 – 30 Sept. Zibordi, G., F. Me´lin, J.-F. Berthon, and D. van der Linde (2002), Validation of MERIS marine products at the Acqua Alta Oceanographic Tower (AAOT), final report of the ESA-JRC, contract 19682-2002-07 (AO-ID 9020), 17 pp., Joint Res. Cent., Ispra, Italy.

O. Aznay, R. Santer, and F. Zagolski, Universite´ du Littoral Coˆte d’Opale (ULCO), Maison de la Recherche en Environnement Naturel (MREN), 32 Avenue Foch, F-62930 Wimereux, France. ([email protected])

16 of 16