Aerosol size distribution retrieved from optical depth

Revue des Energies Renouvelables Vol. 15 N°2 (2012) 207 - 218
Aerosol size distribution retrieved from optical depth
measurements in Tamanrasset and Blida
O. Aïssani* and A. Mokhnache
Laboratoire de Physique Energétique, Département de Physique
Faculté des Sciences, Université des Frères Mentouri
Route El Bey, Constantine, Algérie
(reçu le 30 Mars 2009 - accepté le 30 Mars 2012)
Abstract - Since we are often faced with limited, insufficient or contaminated
observations in remote sensing. We will employ a theoretical technique for the retrieval
of the particle size distribution. In this study, the columnar aerosol size distributions
have been inferred by numerically inverting particulate optical depth measurements
obtained in Blida and Tamanrasset from AERONET database at wavelengths region
between 0.340 - 1.640 µm. Our results are in good agreement with those found by other
authors and the correlation between the aerosol optical thickness measurements and the
reproduced AOTs by the inverted columnar aerosol size distribution is very good with a
correlation coefficient more than 0.97.
Résumé - Car nous sommes souvent confrontés à des observations limitées, insuffisantes
ou contaminées dans la télédétection. Nous allons employer une technique théorique
pour l’extraction de la distribution granulométrique des particules. Dans cette étude, les
distributions en taille des aérosols intégrés sur la colonne atmosphérique ont été
obtenues par l’inversion numérique des mesures de l’épaisseur optique des particules
établis à Blida et Tamanrasset à partir des données d’AERONET dans la région
spectrale entre 0.340 - 1.640 µm. Nos résultats sont en bon accord avec ceux trouvés par
d’autres auteurs et la corrélation entre les mesures d’épaisseur optique des aérosols et
l’AOTs reproduises par la distribution en taille inversé est très bonne avec un coefficient
de corrélation plus de 0.97.
Keywords: Aerosol properties - AERONET - Size distribution - Inversion methods.
Atmospheric aerosols are suspensions of small solid or liquid particles in the
atmosphere, which plays an important role in atmospheric and environmental research
since they take part in many physical and chemical processes in the atmosphere.
Because of the wide variety of sources, the properties of atmospheric aerosol
particles, such as size, shape, chemical composition, and optical thickness, may be
heterogeneous, and their temporal and spatial variation can be very large [1-5].
Knowledge of both aerosol optical properties (extinction, scattering cross section,
phase function, and single-scattering albedo) and microphysical aerosol properties
(particle size distribution function and the complex refractive index) is essential for the
determination of the effect of atmospheric aerosols on the climate and for the control of
the air quality [6].
[email protected]
O. Aïssani et al.
Meanwhile, aerosol particle size distribution and aerosol optical thickness are two
important properties for characterizing the aerosol particles and the correction of the
atmosphere [7].
While aerosol size distributions can be measured in-situ, in different size ranges,
using a variety of instruments [8], they can also be derived from measured aerosol
optical properties, such as the spectral aerosol optical depth [9], or backscattering [10],
or a combination of angular scattering intensities and spectral optical depth [11, 12].
Various techniques have been developed for recovering the particle size distribution
function n ( r ) , for example direct regularization methods [7, 13-17] and iterative
methods [18-23].
This study deals only with the inversion of spectral aerosol optical thickness
measurements, and its aim is to determine aerosol size distributions on Blida (36N, 2E)
and Tamanrasset (22.78S, 5.53E) northern and southern Algeria respectively during
2006. To accomplish this, we use the inversion algorithm of King et al. [9].
Since the relationship between the size of atmospheric aerosol particles and the
wavelength dependence of the extinction coefficient was first suggested by Angström in
1929, the size distribution began to be retrieved by extinction measurements.
First, Ångström inferred that the parameters of a Junge size distribution could be
obtained from the aerosol optical thickness (AOT) at multiple wavelengths, and he
obtained the useful Ångström empirical formula of Junge size distribution
τae = β × λ−α , where τae is the measured AOT, β is the turbidity coefficient, and α
is the Ångström exponent reflecting the aerosol size distribution [24].
Assuming that aerosols are homogenous spheres, the relationship between the
aerosol size distribution and AOT τae can be written as Fredholm integral equations of
the first kind,
τae ( λ ) =
∫ πr
2 ×Q
ext ( r, λ, m ) × n c ( r ) × dr
where r is the particle radius, λ the wavelength of incident illumination, m the
complex refractive index of the aerosol particles, Qext ( r, λ, m ) , the extinction
efficiency factor from Mie theory, and n c ( r ) the columnar aerosol size distribution,
that is, the number of particles per unit area per unit radius interval in a vertical column
through the atmosphere.
Various methods that have been proposed for solving this equation include those in
which an assumption about the analytical form of the size distribution to be retrieved is
made [25, 26], as well as methods that make no assumption about the size distribution
shape [9, 27].
In this article, the constrained linear inversion is used to solving the equation (1);
this method was introduced by Twomey [1] and developed by King et al. [9], to
Aerosol size distribution retrieved from optical depth measurements in Tamanrasset…
determine n c ( r ) from spectral measurements of τae ( λ ) , let n c ( r ) = h ( r ) × f ( r ) ,
where h ( r ) is rapidly varying function of r , while f ( r ) is more slowly varying. So,
the equation (1) becomes:
τae ( λ ) =
∫ πr
q rj + 1
j =1 rj
× Qext ( r, λ, m ) × h ( r ) × f ( r ) × d r
π r × Qext ( r, λ, m ) × h ( r ) × f ( r ) × d r
Here r1 = ra and rb = rq +1 and r1 , r2 ,..., rq +1 are the q + 1 boundaries of q coarse
intervals of integration. If f ( r ) is constant within each coarse interval. Thus the
Fredholm equation is replaced by simultaneous equations
ˆ f + εr
g = A
g i = τ ae ( λ i )
rj + 1
Ai j =
i = 1, 2,..., p
π r 2 × Q ext ( r, λ i , m ) × h ( r ) × d r
j = 1, 2,..., p
f j = f ( rj )
The elements ε i of the unknown vector ε represent the deviations between
measurement ( g i ) and theoretical estimate ( ∑ A i j × f j ). These deviations arise from
measurement and quadrature (integration) errors and due to uncertainties of the kernel
function π r 2 × Q ext ( r, λ, m ) .
r j are the midpoints of coarse intervals. Note, the solution of equation (3) is
obtained in logarithmic scale with respect to particle radius r .
If h ( r ) takes the form of a Junge size distribution [28],
h ( r ) = r − ( ν* +1)
With this substitution, King et al. [9] apply a quadrature method and a minimization
procedure leading to the solution vector
f = A T × C −1 × A + γ H
)−1 × A T × C−1 × g
where C is the measurement covariance matrix; γ is a nonnegative Lagrange
multiplier (choice of the Lagrange multiplier is discussed by King [29]); H is a
smoothing matrix; A is the matrix representation of the kernel function.
The smoothing matrix H is defined by Twomey [30] as:
O. Aïssani et al.
⎡ 1 −2 1
⎢− 2 5 − 4 1
⎢ 1 −4 6 −4 1
1 −4 6 −4
H = ⎢
− 4 5 − 2⎥
1 − 2 1 ⎦⎥
The measurement covariance matrix C is diagonal with elements given by
Ci j = σ 2τ × ( λi ) × δi j , where δi j is Kronecker delta function [31].
Two important advantages can result from the assumption n c ( r ) = h ( r ) × f ( r ) .
Firstly, when h ( r ) represents size distribution n c ( r ) exactly, the solution vector f will
have all components equal to one. Secondly, the smoothing constraint guaranties
minimum curvature of f on a linear scale and it works better when f j = f ( rj ) are
almost constants.
The columnar size distribution n c ( r ) varies over some orders of magnitude and has
an explicitly large curvature, thus it is difficult to constraint.
We want to retrieve columnar aerosol size distribution n c ( r ) over the radii interval
[ ra , rb ] . This is a typical inverseproblem and we can solve it using an iteration process.
This means we have to start with some zero-approximation h (0) ( r ) of the rapidly
varying multiplier in n c ( r ) = h ( r ) × f ( r ) and use it to evaluate a first-order
approximation of solution vector f (1) with aid of (5).
Then we have to utilize f (1) and obtain some reasonable first order approximation
h (1) ( r ) , which better represents the size distribution than the initially assumed
weighting function. The first-order weighting function is then substituted back into (4)
from which a second-order f (2) is obtained through (5). This iterative procedure is
continued until a stable result is obtained [32].
In most cases the columnar size distributions can be classified in terms of three
different types of distributions, although gradations between two different types are
occasionally observed making this classification somewhat arbitrary [9, 29].
The Figure 1 illustrate the classification of the measured AOTs and the
corresponding expected solution types I, II and III of distributions in the same figures
1.-a-, -b- and -c- respectively.
Aerosol size distribution retrieved from optical depth measurements in Tamanrasset…
Fig. 1: Measured AOTs and corresponding columnar size distributions
-a- Type I; -b- Type II; -c- Type III [9]
The method for determining the columnar aerosol size distribution n c ( r ) described
in the section 3 has been carried out at Tamanrasset and Blida since 2006 from aerosol
optical depth measurements by AERONET (AErosol RObotic NETwork) at five and
eight different wavelengths ranging between 0.340 and 1.640 µm.
In our study, all inversions were performed assuming the complex refractive index
of the aerosol particles was wavelength and size independent and given by
m = 1.45 − 0.00 i as in King et al. [9].
O. Aïssani et al.
In lieu of n c ( r ) or, equivalently, d N d r , the size distribution results are
presented in terms of d N d log r , representing the number of particles per unit area per
unit log radius interval in a vertical column through the atmosphere.
5.1 In Tamanrasset
From the inspection of results for 192 different days at Tamanrasset, we derive three
-cFig. 2: Measured and retrieved AOT and corresponding
columnar size distribution for 05 Mars 2006, (-a-, -b- and –c-)
Fig. 2 illustrates the spectral optical depth measurements for 05 Mars 2006,
corresponding size distribution and the linear fit between optical depth measurements
Aerosol size distribution retrieved from optical depth measurements in Tamanrasset…
and calculations, this day represents an example of the first type (Type I) for wich the
measured AOTs nearly follow Ångström’s formula given by equation,
τM ( λ ) = β × λ− α
The observed Mie optical depths and corresponding standard deviations are shown
in the left portion of the figure while the size distributions obtained by inverting these
data is shown in the right portion.
τc ( λ ) in the left portion indicates how the inverted size distributions are able to
reproduce the AOTs measurements (i.e., the direct problem g = A × f ). The aerosol size
distributions illustrated in the same figure can be best described as constructing Junge or
two slopes type of distributions.
On the same figure, we consider the linear fit between τM ( λ ) and τc ( λ ) , we see
that the correlation is good with a correlation coefficient R = 0.9887 .
For the data of 21 June, when measured AOTs exhibit small negative curvature
(Fig.3) the solutions tend to be monodisperse (Type II). This is not unexpected because
the tendency for negative curvature suggests an absence of both small and large
The optical depth measurements which produce this second class of distributions are
typically very large. An important modification is the case when experimental data
constitute two overlapping monodisperse distributions (we can see this case on Blida
results presented below).
For the correlation between the observed Mie optical depths τM ( λ ) and the
reproduced AOTs τc ( λ ) is always good and R = 0.9946 .
O. Aïssani et al.
-cFig. 3: Measured and retrieved AOT and corresponding
columnar size distribution for 21 June 2006, (-a-, -b- and –c-)
The most interesting distribution type is one for which the AOTs intermediate
between these of type I and type II. In this case τM ( λ ) tends generally to have positive
An example of this type (Type III) is illustrated in Fig. 4 for the data of 19 May. The
solution is usually a bi-modal distribution and the correlation coefficient between the
observed and the reproduced AOTs is R = 0.9700 .
Aerosol size distribution retrieved from optical depth measurements in Tamanrasset…
-cFig. 4: Measured and retrieved AOT and corresponding
columnar size distribution for 19 May 2006, (-a-, -b- and –c-)
5.2 In Blida
The inspection of plots of log τ ( λ ) vs. log τ M ( λ ) (Fig. 5) reveals the presence of
at least two kinds of aerosols with possible bi-modal and Junge or tow slopes
distributions because plots on May to November have positive curvature and all other
curves nearly follow Ångström’s formula. The retrieved columnar number distributions
(Fig. 5) are quite interesting.
Fig. 5: Measured and retrieved AOT and corresponding
columnar size distribution during 2006 in Blida
O. Aïssani et al.
For 15 February, 11 March, 12 April and 30 December are constituted by aerosols
with quite similar properties and distributions are Junge-type. The 18 May, 02 June, 09
july, 16 october and 12 november aerosols ensemble has all bi-modal distributions with
a Large amount of small particles, deep minimum at 0.3870 µm and maximum at
0.9718 µm.
Distributions of 03 August have similar but smoother shape – minimum at 0.3870
µm is much shallower and about six orders more particles are present at largest radii
compared with 09 July.
The data case of τ (λ ) M for 18 May has small negative curvature at the four longest
wavelengths, and the corresponding size distribution could equally well be categorized
as a type II (monodisperse) distribution composed by two monodisperse which could be
fine ash and water droplets, thus making classification according to three distinct types
somewhat arbitrary [9].
In this paper, we investigate linear optimization method for the solution of the
atmospheric aerosol particle size distribution function retrieval problem from spectral
measurements of the particulate (Mie) optical depth at wavelengths region between
0.340 and 1.640 µm.
We first described this method of solution developed by King in 1978 and classified
experimental data and expected solution type. Then, we applied the proposed method on
Tamanrasset site southern Algeria for 192 days during 2006.
When, the columnar aerosol size distributions have been determined from data
delivered by AERONET network at five wavelengths 0.440, 0.500, 0.675, 0.870 and
1.020 µm. The optical depth measurements and corresponding aerosol size distributions
are illustrated in Fig. 2-4 for a few of these days. Our results are in agreements with
those found by King in Tucson, Arizona.
His results generally indicate (at least for r ≥ 0.1µm ) that the aerosol size
distribution on a particular day can be represented either as a Junge distribution (type I),
a relatively monodisperse distribution such as a log-normal or gamma distribution (type
II), or as a two-component system consisting of a combination of both of these types
(type III).
Finally, we want to apply the method on Blida site northern Algeria from spectral
data at eight wavelengths 0.340, 0.380, 0.440, 0.500, 0.675, 0.870, 1.020, 1.640 µm,
when the retrieved columnar number distributions are quite interesting, what is
represented on Fig. 5.
The presence of at least two kinds of aerosols is very clear with bi-modal (Type III)
and Junge-type (Type I) distributions. The type I distributions occur mainly in the
winter and early spring months and the type III occur throughout the late spring to late
summer months.
[1] S. Twomey, ‘Atmospheric Aerosols’, Developments in Atmospheric Science, Vol. 7, Elsevier
Scientific Pub. Co., 302 pages, 1977.
[2] C.E. Junge, ‘Air Chemistry and Radioactivity’, Academic Press, 382 pages, 1963.
Aerosol size distribution retrieved from optical depth measurements in Tamanrasset…
[3] G.F. Bohren and D.R. Huffman, ‘Absorption and Scattering of Light by Small Particles’,
Wiley, 1983.
[4] C.N. Davies, ‘Size Distribution of Atmospheric Particles’, Journal of Aerosol Science, Vol. 5,
N°3, pp. 293 – 300, 1974.
[5] Y.F. Wang and C.C. Yang, ‘Regularizing Active Set Method for Retrieval Of Atmospheric
Aerosol Particle Size Distribution Function’, Journal of Optical Society of America A, Vol.
25, pp. 348 - 356, 2008.
[6] G. Tonna, T Nakajima and R. Rao, ‘Aerosol Features Retrieved from Aureole Data: A
Simulation Study Concerning a Turbid Atmosphere’, Applied Optics, Vol. 34, N°21, pp. 4486 4499, 1995.
[7] Y.F. Wang, S. Fan, X. Feng, G.J. Yan and Y.N. Guan, ‘Regularized inversion method for
retrieval of aerosol particle size distribution function in W1,2 space’, Applied Optics, Vol. 45,
N°28, pp. 7456 - 7467, 2006.
[8] P.H. McMurry, ‘A Review of Atmospheric Aerosol Measurements’, Atmospheric
Environment, Vol. 34, pp. 1959 - 1999, 2000.
[9] M.D. King, D.M. Byrne, B.M. Herman and J.A. Reagan, ‘Aerosol Size Distributions Obtained
by Inversion of Spectral Optical Depth Measurements’, Journal of the Atmospheric Sciences,
Vol. 35, N°11, pp. 2153 - 2167, 1978.
[10] A. Ben-David, B.M. Herman and J.A. Reagan, ‘Inverse Problem and the Pseudoempirical
Orthogonal Function Method of Solution. 2. Use’, Applied Optics, Vol. 27, N°7, pp. 1243 1254, 1988.
[11] T. Nakajima, G. Tonna, R. Rao, P. Boi, Y. Kaufman and B. Holben, ‘Use of Sky Brightness
Measurements from Ground for Remote Sensing of Particulate Polydispersions’, Applied
Optics, Vol. 35, N°15, pp. 2672 - 2686, 1996.
[12] O.V. Dubovik, T.V. Lapyonok and S.L. Oshchepkov, ‘Improved Technique for Data
Inversion: Optical Sizing of Multicomponent Aerosols’, Applied Optics, Vol. 34, N°36, pp.
8422 - 8436, 1995.
[13] D.L. Phillips, ‘A Technique for the Numerical Solution of Certain Integral Equations of the
First Kind’, Journal of the Association for Computing Machinery, Vol. 9, N°1, pp. 84 – 97,
[14] S. Twomey, ‘On the Numerical Solution of Fredholm Integral Equations of the First Kind by
the Inversion of the Linear System Produced by Quadrature’, Journal of the Association for
Computing Machinery, Vol. 10, N°1, pp. 97 – 101, 1963.
[15] G.E. Shaw, ‘Inversion of Optical Scattering and Spectral Extinction Measurements to
Recover Aerosol Size Spectra’, Applied Optics, Vol. 18, N°7, pp. 988 – 993, 1979.
[16] K.S. Shifrin and L.G. Zolotov, ‘Spectral Attenuation and Aerosol Particle Size Distribution’,
Applied Optics, Vol. 35, N°12, pp. 2114 – 2124, 1996.
[17] C. Böckmann, ‘Hybrid Regularization Method for the Ill-posed Inversion of
Multiwavelength Lidar Data in the Retrieval of Aerosol Size Distributions’, Applied Optics,
Vol. 40, N°9, pp. 1329 – 1342, 2001.
[18] S. Twomey, ‘Comparison of Constrained Linear Inversion and An Iterative Nonlinear
Algorithm Applied to the Indirect Estimation of Particle Size Distributions’, Journal of
Computational Physics, Vol. 18, N°2, pp. 188 – 200, 1975.
[19] Y.F. Wang, ‘Computational Methods for Inverse Problems and Their Applications’, Higher
Education Press, Beijing, 2007.
[20] M.T. Chahine, ‘Inverse Problems in Radiation Transfer: Determination of Atmospheric
Parameters’, Journal of the Atmospheric Sciences, Vol. 27, N°6, pp. 960 – 967, 1970.
O. Aïssani et al.
[21]. F. Ferri, A. Bassini, and E. Paganini, ‘Modified Version of the Chahine Algorithm to Invert
Spectral Extinction Data for Particle Sizing’, Applied Optics, Vol. 34, N°25, pp. 5829 –
5839, 1995.
[22] C. Böckmann and A. Kirsche, ‘Iterative Regularization Method for Lidar Remote Sensing’,
Computer Physics Communications, Vol. 174, N°8, pp. 607 – 615, 2006.
[23] Y.F. Wang, S.F. Fan and X. Feng, ‘Retrieval of the Aerosol Particle Size Distribution
Function by Incorporating a Priori Information’, Journal of Aerosol Science, Vol. 38, N°8,
pp. 885 – 901, 2007.
[24] A. Angstrom, ‘On the Atmospheric Transmission of Sun Radiation and on Dust in the Air’,
Geografiska Annaler, Vol. 11, pp. 156 - 166, 1929.
[25] P.H. Wang, M.P. McCormick, T.J. Swissler, M.T. Osborn, W.H. Fuller and G.K. Yue,
‘Inference of Stratospheric Aerosol Composition and Size Distribution from SAGE II
Satellite Measurements’, Journal of Geophysical Research, Vol. 94, N°D6, pp. 8435 - 8446,
[26] G.K. Yue, M.P. McCormick and W.P. Chu, ‘Retrieval of Composition and Size Distribution of
Stratospheric Aerosols with the SAGE II Satellite Experiment’, Journal of Atmospheric and
Oceanic Technology, Vol. 3, N°3, pp. 371 - 380, 1986.
[27] G.P. Box, K.M. Sealey and M.A. Box, ‘Inversion of Mie Extinction Measurements Using
Analytic Eigenfunction Theory’, Journal of the Atmospheric Sciences, Vol. 49, N°22, pp.
2074 - 2081, 1992.
[28] C.E. Junge, ‘The Size Distribution and Aging of Natural Aerosols as Determined From
Electrical and Optical Data on the Atmosphere’, Journal of Meteorology, Vol. 12, N°1, pp.
13 - 25, 1955.
[29] M.D. King, ‘Sensitivity of Constrained Linear Inversion to the Selection of the Lagrange
Multiplier’, Journal of the Atmospheric Sciences, Vol. 39, N°6, pp. 1356 - 1369, 1982.
[30] S. Twomey, ‘On the Numerical Solution of the Fredholm Integral Equations of the First Kind
by the Inversion of the Linear System Produced by Quadrature’, Journal of the Association
Computing Machinery, Vol. 10, N°1, pp. 97 - 101, 1963.
[31] M.D. King and D.M. Byrne, ‘A Method for Inferring Total Ozone Content from the Spectral
Variation of Total Optical Depth Obtained with a Solar Radiometer’, Journal of the
Atmospheric Sciences, Vol. 33, N°11, pp. 2242 – 2251, 1976.
[32] B.M. Herman, S.R. Browning and J.A. Reagan, ‘Determination of Aerosol Size Distributions
from Lidar Measurements’, Journal of the Atmospheric Sciences, Vol. 28, N°5, pp. 763 771, 1971.