RETRIEVING AEROSOL CHARACTERISTICS FROM

SATELLITE OCEAN COLOR MULTI-SPECTRAL SENSORS

USING A NEURAL-VARIATIONAL METHOD

D. Diouf

, S. Thiria

, A. Niang

, J. Brajard

and M. Crepon

Ecole Supérieure Polytechnique, Université Cheikh Anta Diop de Dakar, BP 5085, Dakar Fann, Sénégal

IPSL/LOCEAN, Université Paris 6, 75252, Paris, France

Keywords: Multi-layer perceptrons, Atmospheric correction, Variational inversion.

Abstract: We present a new algorithm suitable for retrieving and monitoring Saharan dusts from satellite ocean-color

multi-spectral observations. This algorithm comprises two steps. The first step consists in classifying the

TOA spectra using a neuronal classifier, which provides the aerosol type and a first guess value of the

aerosol parameters. The second step retrieves accurate aerosol parameters by using a variational

optimization method. We have analyzed 13 years of SeaWiFS images (September 1997-December 2009) in

an Atlantic Ocean area off the coast of West Africa. As the method takes into account Saharan dusts, the

number of pixels processed is an order of magnitude higher than that processed by the standard SeaWiFS

algorithm. We note a strong seasonal variability. The Saharan dust concentration is maximal in summer

during the rainy season and minimal in autumn when the vegetation bloom due to the rainy season prevents

soil erosion by the wind.

1 INTRODUCTION

Aerosols are an important component of the Earth

climate system. They reflect the downwelling solar

radiations and thus contribute to cooling the

atmosphere on the one hand and on the other hand,

they may also absorb infrared radiation emitted by

Earth, thus contributing to warming the atmosphere

depending on their quality. A good knowledge of

aerosol properties is therefore necessary for

understanding climate variability and modeling it.

The mass concentration of aerosols is closely related

to the optical thickness

, which is a measure of the

light attenuation. Aerosols are also characterized by

their type (dust, maritime, soot ...).

A major source of aerosols is the Sahara desert,

which seeds the tropical Atlantic atmosphere with

Saharan dusts, which are absorbing aerosols.

These aerosols cross the Atlantic Ocean transported

by the trades winds and may be detected as far away

as the Caribbean Island and South America (Moulin

et al, 1997).

During the last 15 years, several satellites

carrying multi-spectral radiometers dedicated to

ocean-color observation have been launched. They

provide a daily global coverage of Earth at a scale of

some kilometers. These ocean-color radiometers

also provide information about aerosol parameters,

since the atmosphere is located between the ocean

and the satellite. Ocean color radiometer signals

have been intensively used to monitor aerosol

parameters over the ocean (Gordon and Wang,

1994); (Tanré et al., 1997) and to retrieve their most

significant parameters.

The standard aerosol products provided by Space

Agencies such as the SeaWiFS products distributed

by NASA are limited to a quite low optical thickness

(less than 0.35). Moreover, the algorithms used for

SeaWiFS products are not able to deal with

absorbing aerosols nor to retrieve the aerosol

typology.

This paper presents a new method for deriving

aerosol characteristics including those of absorbing

aerosols, from satellite ocean-color data.

2 DATA SETS

2.1 The SeaWiFS Data Set

For this study we use daily luminance measurements

made by the SeaWiFS sensor off the West Africa

297

Diouf D., Thiria S., Niang A., Brajard J. and Crepon M..

RETRIEVING AEROSOL CHARACTERISTICS FROM SATELLITE OCEAN COLOR MULTI-SPECTRAL SENSORS USING A NEURAL-VARIATIONAL

METHOD.

DOI: 10.5220/0003638802970303

In Proceedings of the International Conference on Neural Computation Theory and Applications (NCTA-2011), pages 297-303

ISBN: 978-989-8425-84-3

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

coast in an area between 8°-24°N and 14°-30°W.

These measures extend the period of 1997-2009.

Luminances are at wavelengths 412nm, 443nm,

490nm, 510nm, 555nm, 670nm.765nm and 865nm.

For each wavelength

, the TOA reflectance

computed.

According to Gordon and Wang (1994) the Top

Of the Atmosphere (TOA) reflectance

is the sum

of several components that can be computed

separately: the Rayleigh multiple scattering (air

molecules) in the absence of aerosols, can be

accurately computed by using the atmospheric

pressure, and the whitecap contribution by taking

into account the wind speed. We removed pixels

contaminated by the sun glitter, using a geometrical

mask. The signal that was finally used in our

classification method was therefore:

used

(1)

where

is the reflectance resulting from multiple

scattering of aerosols in the absence of the air,

the interaction term between molecular and aerosol

scattering,

, is the contribution of the water and t

is the transmittance of the atmosphere at a given

wavelength (

λ)

In equation (1),

is small in the red and near-

infrared, so that

used

mainly depends on the

aerosol term

raa

at 670, 765 and 865 nm. For

the other visible bands, it is expected that the aerosol

term remains large enough in most situations to

allow us to retrieve pertinent information (in

particular absorption capability) about the particles

at these wavelengths.

We used satellite data sets comprising ten

dimensional vectors, whose components are eight

wavelengths measured by the radiometer and two

viewing angles since the reflectance spectra depend

on the geometry of the measurement. These angles

are the sun zenith angle

and the scattering angle

defined as:

()

ΔΦ+−= cossinsincoscosarccos

svsv

(2)

where

ΔΦ

is the azimuth angle difference

between the satellite and the sun, and

is the

viewing zenith angle.

Each vector, whose components correspond to

the SeaWiFS wavelengths, represents a

used

spectrum.

2.2 The Learning Data Set

The learning data set consists of observed

obs

used

ex-

tracted from pixels of SeaWiFS images off the West

Africa coast during the year 2003 and two associated

viewing angles (i.e., the sun zenith angle

and the

scattering angle

). All the available daily SeaWiFS

images were homogeneously sampled (one pixel-

line over 10) providing 426,117 clear-sky spectra

obs

used

. The learning dataset Data

obs

is thus

composed of ten component vectors i.e. the eight

wavelengths measured by the radiometer and the

two viewing angles.

2.3 The Labeling Data Set

The second data set, Data

expert

consists of the

used

expert

computed at eight wavelengths with a 2-layer

radiative transfer model (Gordon & Wang, 1994) for

various optical thickness values, chlorophyll content

and geometry of the measurement and for five

aerosol models. Each Data

expert

vector comprises

eight spectral components (

used

expert

) and two

geometry components which are the sun zenith angle

and the scattering angle

. To these ten

components which were used for the labeling the

referent vectors provided by the unsupervised

classification, we added the aerosol type and the

optical thickness

at 865 nm. Data

expert

comprises

6,000,000 simulated vectors using four aerosol

models and one absorbing aerosol (Moulin et al,

2001). The five aerosol models were computed at

four different relative humidity (70%, 80%, 90%,

99%). Data

expert

was used in order to introduce the

expertise and to retrieve the aerosol type and the

optical thickness values.

3 THE METHOD

In this study, we used two successive statistical

models for analyzing the Data

obs

images; the Self

Organizing Map (SOM, Kohonen, 2001) model and

the NeuroVaria method (Jamet et al., 2005); (Brajard

et al., 2006). We first processed the images with a

SOM model, which is well suited for visualizing and

clustering a high-dimensional data set. We denoted

this topological map as SOM-A-S (SOM-Angle-

Spectrum). In the light of the results obtained by

Niang et al., (2006), we chose a similar architecture

for SOM-A-S: a two-dimensional array with a large

number of neurons (20 x 30 = 600). SOM-A-S was

learned on the Data

obs

of the year 2003. The vectors

of the learning data set were thus clustered into 600

groups, allowing a highly discriminative

NCTA 2011 - International Conference on Neural Computation Theory and Applications

298

representation of Data

obs

. The second dataset,

Data

expert

, representing the expertise, was used to

decode the SeaWiFS images. The principle of the

method is to compare the ten-component vectors of

Data

expert

whose associated parameters are known,

with those of the neurons of SOM-A-S according to

a distance. At the end of the labeling, each neuron of

SOM-A-S map has captured a set of

ertexp

and

takes a label, which is extracted from that set

according to the procedure described in Niang et al.,

(2006). The only difference between the two

versions being that the old one uses a first map to

determine 10 different classes of angles, each one

giving rise to a dedicated SOM map for the

classification of the reflectance spectra, while SOM-

A-S uses a unique map doing a data fusion between

the viewing angles and the spectra. By using a

unique map, we avoided the threshold effect that is

induced by the two steps classification (angle and

then reflectance) and the eleven SOM maps

described in Niang et al., (2003).

Each neuron is therefore associated with an

atmospheric and ocean physical parameters (

τ,

and an aerosol type. The SOM-A-S map being

labelled, we are able to analyze a satellite image by

projecting the ten component vector (reflectances

and viewing angles) associated with each pixel on

the SOM-A-S map. Pixels captured by a neuron are

assigned to the aerosol type and optical thickness

associated with this neuron. For monthly

climatology images, the aerosol type is estimated as

the median of the types of the images considered.

The second statistical model improves the

retrieval of the optical thickness. We used a neuro-

variational algorithm, called NeuroVaria, that is able

to provide accurate atmospheric corrections for

inverting satellite ocean-colour measurements. The

algorithm minimizes a weighted quadratic cost

function, J, by adjusting control parameters

(atmospheric and oceanic) such as

and C (Brajard

et al., 2008). J describes the difference between the

satellite measurement

obs

and a simulated

reflectance

sim

computed using radiative transfer

codes modelled by supervised neural networks (the

so called Multi-Layer-Perceptrons, MLP). The

minimization implies the computation of the

gradient of J with respect to the control parameters

and consequently of the derivatives of the MLPs,

which is done by the classical gradient back-

propagation algorithm (Bishop, 1995). The novelty

of the version of NeuroVaria developed in this work

is that the MLPs modelling the radiative transfer

codes were specially designed to take African dusts

into account. Moreover we used the atmospheric

parameter values given by SOM-A-S and validated

using in situ data (see section 4), as first guesses of

the NeuroVaria algorithm minimization. Since the

efficiency of a minimizing procedure depends on the

first guesses of the control parameters, we expect to

improve the accuracy of the retrieved parameters.

Using these two statistical models sequentially is

indeed a mixed neuro variational method. We

denoted it in the following by SOM-NV.

4 VALIDATIONS OF THE

AEROSOL PARAMETERS

USING SOM-A-S

As SOM-A-S takes into account Saharan dusts, the

number of pixels processed is an order of magnitude

higher than that processed by the standard SeaWiFS

algorithm. As an example, on October 07 2003,

SOM-A-S processed 29,083 pixels while SeaWiFS

processed 16,193 pixels only; on October 12 SOM-

A-S processed 30,300 pixels and SeaWiFS 3,338

only. Besides a statistical comparison between the

SOM-A-S and SeaWiFS algorithms was made for

values of

<0.35. The Mean Relative Error (MRE)

remains low (22.88% for October 07 2003 and

16.16% for October 12 2003) and the Root Mean

Square Error (RMSE) was less than 0.04 for both

days. As a preliminary conclusion, the values

retrieved by SOM-A-S seem consistent with and

very close to those retrieved by the classical

algorithm of SeaWiFS for

<0.35.

The Angström exponent

(500,870) provided by

AERONET, allows us to attempt to validate the dust

aerosol type provided by SOM-A-S. Since the sun

photometer does not give the aerosol type, it is

thought possible to validate the dusts by studying the

behavior of

(500,870). The low

(500,870)

values (

<0.5) result from the presence of large

particles typical of desert dusts (Nobileau et al.,

2005). In Figure 1 we show the distribution of the

(500,870) of the Dakar-M’Bour AERONET

measurements for the dusty and the non-dusty days

determined by SOM-A-S on the SeaWiFs collocated

pixels. The confidence interval of the average value

(500,870) calculated by SOM-A-S from

SeaWiFs measurements for the dusty days was

between 0.40 and 0.47, whereas it was between

0.61-0.79 for non-dusty days. This means that the

dust classification provided by SOM-A-S is in

agreement with the AERONET measurements,

which permits us to distinguish the dust absorbing-

aerosols from the non-absorbing ones by processing

RETRIEVING AEROSOL CHARACTERISTICS FROM SATELLITE OCEAN COLOR MULTI-SPECTRAL SENSORS

USING A NEURAL-VARIATIONAL METHOD

299

the SeaWiFs observations with SOM-A-S.

Figure 1: Comparison of the

(500, 870) values

measured

at the Dakar AERONET station for dusty days (left) and

for non-dusty days (right). For dusty days, most of the

(500, 870) values are less than 0.5.

5 IMPROVING ATMOSPHERIC

RESTITUTION WITH SOM-NV

We processed the 13 year data set of SeaWiFS

imagery with SOM-NV. This data set presents a

well-marked seasonal variability. Figure 2 shows, as

an example, the monthly situations during winter

(January), spring (March), summer (July) and

autumn (November) of the year 2006 decoded with

SOM-NV for the aerosol optical thickness and with

SOM-A-S for the aerosol type. In winter (January),

the northern part of the studied domain was free of

dusts and the optical thickness was almost zero; in

the southern part, we observed the presence of

Saharan dusts with a small concentration (small

optical thickness). In spring, the Saharan dusts

moved northward and their concentration (optical

thickness) increased. In summer, the Saharan dusts

invaded the entire domain and their concentration

was maximal. In autumn, the Saharan dusts

disappeared but we noted a low optical thickness in

the whole domain due to the presence of non-

absorbing aerosols. The year 2006 represents a

typical year of the data set concerning the seasonal

variability, which is observed every year.

During winter, spring and summer, the presence

of Saharan dusts is linked to the westward wind,

eroding the Sahara ground and transporting dusts

over the Atlantic, as seen in Figure 2. The extent of

Saharan dust is maximal in summer when the Inter

Tropical Convergence Zone (ITCZ) is at its

maximum latitude. The situation in autumn is

puzzling. The wind is still blowing westward in the

southern part of the domain but we do not detect any

Saharan dust. A possible explanation might be due

to the fact that in autumn, the vegetation has

developed following the summer rain, (summer is

the rainy season). The vegetation and soil humidity

inhibit the erosion of the ground in the southern

region of the Sahara, which might explain the

absence of dust south of 20°N in autumn.

A validation can therefore be made by

comparing the optical thickness values retrieved by

SOM-NV and the SeaWiFS algorithm and those

measured at the AERONET stations of Dakar and

Cabo Verde, respectively denoted

SOM-NV

SeaWiFS

and

AERO

We determine

, by taking the mean value of

the five SeaWiFS measurements surrounding the

AERONET ground stations.

We ended up, at the two ground stations, with

1,288 measurements collocated for SOM-NV

retrievals and 623 measurements collocated for the

standard SeaWiFS algorithm (fewer because of the

dust mask). We compared the RMSE and the MRE

SOM-NV

and

SeaWiFS

with respect to the observed

AERO

The results for the AERONET measurements

collocated with those of SeaWiFS and the SOM-NV

for which the SeaWiFS optical thickness value was

less than 0.35 (SeaWiFS critical value) are presented

in Table 1. Table 2 shows comparisons for the

AERONET measurements collocated with those of

the available SOM-NV, which only include

measurements for which the optical thickness value

was higher than 0.35.

The correlation coefficient between

SOM-NV

and

AERO

is higher than that between

SeaWiFS

and

AERO

showing the good performances of the SOM-NV

method. This is confirmed by the scatter plot of

SOM-NV

and

AERO

, and

SeaW

and

AERO

(Figure 3 for

Dakar, Figure 4 for Cabo Verde). But it is also

important to note that the SOM-NV neural decoding

allows the retrieval of high optical thickness values

(i.e., greater than 0.35, above which the SeaWiFS

algorithm does not work) with a good accuracy.

6 CONCLUSIONS

We have developed an original and efficient two-

step method for retrieving optical properties (type

and optical thickness) from TOA reflectance

measured by satellite-borne multi-spectral ocean-

color sensors. The method is based on a combination

of a neural network classification and a variational

optimization. It makes use of the full spectrum and

two viewing angles of measurements to perform the

aerosol identification.

NCTA 2011 - International Conference on Neural Computation Theory and Applications

300

Figure 2: Monthly map of optical thickness (left panels) computed with SOM-NV and extent of Saharan dust (right panels)

for (from top to bottom) January, March, July and November 2006 computed with SOM-A-S.

RETRIEVING AEROSOL CHARACTERISTICS FROM SATELLITE OCEAN COLOR MULTI-SPECTRAL SENSORS

USING A NEURAL-VARIATIONAL METHOD

301

Table 1: Comparison of the performances (RMSE and MRE) obtained on the optical thickness (< 0.35) computed by the

SOM-NV and the SeaWiFS product with respect to the concomitant AERONET measurements at Dakar (M’Bour) and

Cabo Verde (Sal Island) averaged for the 12 years from 1997 to 2009.

Station Number of

Collocated data

RMSE

MRE (%) Correlation coefficient

SOM-NV SeaW. SOM-NV SeaW. SOM-NV SeaW.

Dakar 232 0.023 0.025 33.8 32.6 0.83 0.69

Cap-Vert 391 0.017 0.018 42.2 40.8 0.79 0.70

Table 2: Performances (RMSE and MRE) obtained on the optical thickness computed by the SOM-NV for

>0.35 only,

with respect to AERONET measurements at Dakar (M’Bour) and Cabo Verde averaged for the 12 years.

Station Number of Collocated data RMSE

MRE (%) Correlation coefficient

Dakar 338 0.025 21.9 0.88

Cap-Vert 327 0.024 25.2 0.91

Figure 3: Scatter plot of the optical thickness

measurements computed by SOM-NV (Δ) and the

SeaWiFS product (*) with respect to the AERONET

measurements at Dakar.

This new method allows retrieval of the aerosol

optical properties from the statistical properties of

the data and for the first time the identification of the

aerosol type. Besides, it gives accurate results for

optical thickness values greater than 0.35, which is

not the case for the standard SeaWiFS product. This

allowed us to substantially increase the number of

pixels processed with respect to the standard

SeaWiFS algorithm by an order of magnitude as

shown in Table 1 and Table 2. Moreover, the

method permits detection of absorbing aerosols,

such as Saharan dusts, which is still a challenge.

The monthly mean optical thickness measured by

the SeaWiFS sensor is strongly correlated with the

ground based measurements (AERONET stations),

which validates the pertinence of the method.

Analysis of the 13 years of observation show an

important seasonal variability associated with the

wind direction and intensity. The Saharan dust con-

Figure 4: Scatter plot of the optical thickness

measurements computed by SOM-NV (Δ) and the

SeaWiFS product (*) with respect to the AERONET

measurements at Cabo Verde.

centration is maximal in summer when the ITCZ is

at its maximum latitude and minimum in autumn

when the vegetation bloom reduces the soil erosion

by the wind. This 13 year climatological data set

available at http://www.locean-ipsl.upmc.fr/

~POACC/ may be used to assess the seasonal

variability of the mass of Saharan dust transported

by the wind over the Atlantic Ocean. This new

method can be easily implemented. Climatologists

will get a better estimate of the aerosol concentration

over the ocean and will have access to the aerosol

type, which is important to understand their impact

on climate.

ACKNOWLEDGEMENTS

We are grateful for the support we received from

CNES and IRD. We thank Dr. H. R. Gordon and C.

NCTA 2011 - International Conference on Neural Computation Theory and Applications

302

Moulin for providing the synthetic database. We

thank the AERONET team and Dr. Tanré from LOA

(Lille) for kindly providing the sun-photometer data

at Dakar and Cabo Verde.

REFERENCES

Bishop, C. (1995). Neural networks for pattern

recognition. Oxford University Press.

Brajard, J., Jamet, C., Moulin, C. and Thiria, S. (2006).

Use of a neuro-variational inversion for retrieving

oceanic and atmospheric constituents from satellite

ocean colour sensor: Application to absorbing

aerosols. Neural Networks 19, 178-185.

Brajard, J., Niang, A., Sawadogo, S., Fell, F., Santer, R

and Thiria, S. (2007). Estimating Aerosol parameters

from MERIS ocean colour sensor observations by

using topological maps. Intern. J. Remote Sensing ,

Vol28, N° 3-4 pp 781-795.

Gordon, H. R. & Wang, M. (1994). Retrieval of water-

leaving radiances and aerosol optical thickness over

the oceans with SeaWiFS: A preliminary algorithm.

Applied Optics, 33(3), 443-453.

Holben, B., Eck, T., Slutsker, I., Tanré, D., Buis, J. P.,

Setzer, A., et al. (1998). AERONET. A federated

instrument network and data archive for aerosol

characterization. Remote Sensing of Environment,

Vol. 66, n

1, p. 1-16.

Husar, R., Stowe, L. and Prospero, J. (1997).

Characterization of tropospheric aerosols over the

oceans with NOAA advanced very high resolution

radiometer optical thickness operational product.

Journal of Geophysical Research, 102(16), 889-909.

Jamet, C., Thiria, S., Moulin, C., and Crepon, M. (2005).

Use of a neuro-variational inversio for retrieving

oceanic and atmospheric constituents from ocean color

imagery. A feasibility study. Journal of Atmospheric

and Oceanic Technology, 22(4), 460-475.

Doi:10.1175/JTECH1688.1

Kohonen, T. (2001). Self organizing maps (3

ed.). Berlin

Heidelberg: Springer Verlag. p. 501. Moulin, C.,

Lambert, C. E., Dulac, F. and Dayan, U. (1997).

Control of atmospheric export of dust from North

Africa by the North Atlantic Oscillation. Nature, 387,

691-694.

Moulin, C., Gordon, H. R., Banzon, V. F., and Evans, R.

H. (2001). Assessment of Saharian dust absorption in

the visible from SeaWiFS imagery. Journal of

Geophysical Research, 106, 18239-18250.

Niang, A., Badran, F., Moulin, C., Crépon, M. & Thiria, S.

(2006). Retrivial of aérosol type and optical thickness

over the Mediterranean from SeaWiFS images using

an automatic neural classification method. Remote

Sensing of Environment, Vol. 100, n

1, p. 82-94.

Nobileau, D. and Antoine, D. (2005). Detection of blue-

absorbing aerosols using near-infrared and visible

(ocean color) remote sensing observations. Remote

Sensing of Environment, Vol. 95, n

3, p. 368-387.

Tanre, D., Deroo, C., Duhaut, P., Herman, M., Morcrette,

J., Perbos, J. and Deschamps, P. Y. (1997).

Description of a computer code to simulate the

satellite signal in the solar spectrum: 5s code.

International Journal of Remote Sensing, 11 : 659-

668.

RETRIEVING AEROSOL CHARACTERISTICS FROM SATELLITE OCEAN COLOR MULTI-SPECTRAL SENSORS

USING A NEURAL-VARIATIONAL METHOD

303