Validation of MODIS derived aerosol optical depth over Western
India
Amit Misra,1 A. Jayaraman,1 and Dilip Ganguly1
Received 15 June 2007; revised 16 October 2007; accepted 19 November 2007; published 19 February 2008.
[1] MODIS (Moderate Resolution Imaging Spectroradiometer) derived aerosol optical
depths (AODs) were compared against the ground based observations from Microtops
sunphotometer over Ahmedabad (72.5
E, 23.03
N) in Western India. The region is semi-
arid and poses challenge for the satellite remote sensing of aerosols. Besides comparing
the ground truth with the Collection Version 4 of MODIS aerosol product, the paper
reports the first ever validation of the updated Collection Version 5 of the MODIS aerosol
product over India. The AOD data from Aqua platform is averaged over 0.5

0.5

centered at Ahmedabad and compared with the sunphotometer observation taken within
half an hour to the satellite overpass time. The Version 4 data comparison showed a
large scatter. Further, the comparison for 470 nm and 660 nm behave differently over
different years. Overall, the comparison shows considerable improvement in the
Collection Version 5 aerosol product. Among seasons, Pre-Monsoon (April to May) has
the best correlation and Dry season (December to March) the least. The updated
product has scope for further improvement as the correlations are less than unity, and the
extent of underestimation for 470 nm is more during Dry and Post-Monsoon seasons
whereas that for 660 nm is more during Pre-Monsoon and Monsoon seasons which are
dominated by fine and coarse particles respectively. The results show a better surface
reflectance parameterization by the MODIS Collection Version 5 algorithm as compared
to Version 4 but the aerosol model used in the retrieval algorithm is still not adequate.
Citation: Misra, A., A. Jayaraman, and D. Ganguly (2008), Validation of MODIS derived aerosol optical depth over Western India,
J. Geophys. Res., 113, D04203, doi:10.1029/2007JD009075.
1. Introduction
[2] Aerosols are in the forefront of climate studies since
last two decades owing to the role they play in the Earth-
Atmosphere system by absorbing or scattering the incoming
solar radiation thus warming or cooling the atmosphere
[Bellouin et al., 2005]. Their optical properties which
depend on their chemical composition, size and shape,
determine their radiative behavior which in turn determines
the overall effect they will have on the Earth radiation
budget [Charlson et al., 1992; Chung et al., 2005]. The
major problem in their characterization is on account of
their short lifetime because of which they have high spatial
and temporal variability [Seinfeld and Pandis, 1998]. Be-
sides, the transport processes may bring in aerosols from
other locations and affect the local climate there. These
factors reinforce the necessity of aerosol monitoring on a
larger spatial scale than can be provided by the ground
based measurements [IPCC, 2001].
[3] Satellite based observations can provide detailed
knowledge in this regard on a long timescale covering a
large spatial area [Kaufman et al., 2002a]. They have an
additional advantage, compared to conventional ground
based observations, in that since the same instrument is
making observation globally, the aerosol concentration at
different locations can be compared which will not be
affected by the calibration errors of the instrument. Aerosol
monitoring from space based instruments consists in
extracting the atmospheric contribution from the total signal
measured by the satellite sensor. Aerosol monitoring from
previous sensors was limited to studies over oceans which
have a distinct advantage in that the total measured signal is
not much affected by reflectance from ocean surface away
from sun-glint area [King et al., 1999]. Aerosol retrieval
over oceans is thus more accurate and reliable. Studies over
land are comparatively more challenging because of the
large surface reflectance which may introduce considerable
errors in the retrieved results.
[4] The initial attempts for aerosol retrieval over land
were made with the launch of the POLDER instrument
[Deschamps et al., 1994] which utilized the information
regarding the polarization state of the radiation for the
purpose [Leroy et al., 1997] but had limitations with
onboard calibration. It lasted for only 8 months due to
technical problems in the spacecraft. Aerosol monitoring
over land entered a new era with the launch of the MODIS
instrument [Barnes et al., 1998] onboard the NASA satel-
lites Terra and Aqua in 1999 and 2002 respectively. It
measures the Earth leaving radiances in 36 high resolution
JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 113, D04203, doi:10.1029/2007JD009075, 2008
Click
Here
for
Full
Article
1Physical Research Laboratory, Ahmedabad, India.
Copyright 2008 by the American Geophysical Union.
0148-0227/08/2007JD009075$09.00
D04203 1 of 12

bands from 0.4 to 14.0 microns with a spatial resolution of
250 m, 500 m and 1 km depending on the wavelength. Its
large swath of 2330 km allows the nearly global coverage
within 1 or 2 days. The wide spectral coverage has advan-
tage in deriving aerosol size distribution information and
hence in uncoupling the total aerosol amount into contribu-
tion from natural and anthropogenic parts [Kaufman et al.,
2005]. Parallel measurement in visible and IR channels has
also helped in overcoming the biggest obstacle in the
aerosol retrieval over land viz, the separation of the surface
reflectance part from the total measured signal [King et al.,
1992, 1999, 2003].
[5] In spite of these advantages, the space based measure-
ments from MODIS, like any other sensor, suffer from
various drawbacks. These include possibility of cloud
contamination and errors arising due to various assumptions
in the retrieval algorithm principally about the aerosol type
and surface reflectance. To circumvent the uncertainties
pertaining to these parameters, the procedure relies on the
ground based observations which provide the exhaustive
database of the aerosol microphysical properties [D’ Almeida
et al., 1991; Holben et al., 1998]. Even after the retrieval has
been accomplished, the satellite retrieved data has to be
validated against the ground truth data. Being based on the
measurement of attenuation in direct solar radiation, data
from ground based observations are not limited by the
aerosol type and surface reflectance related constraints.
Hence they represent the benchmark against which to verify
the accuracy and validity of satellite based retrievals. An
extensive validation exercise tests the efficacy of the re-
trieval algorithm, conditions under which it works satisfac-
torily and cases where further improvement is needed.
Further, the validation exercise helps to quantify the
improvements to the algorithm which are made from time
to time. Validation of MODIS aerosol optical depth data
started and is in progress nearly since the data from the
sensor started coming in. Chu et al. [2002, 2003], Ichoku et
al. [2002, 2003], Remer et al. [2002, 2005], Levy et al.
[2005] and several other groups working on the validation
efforts report their results based on the extensive validation
of the aerosol product at its different stages of development.
However, a detailed and extensive validation of the MODIS
Level 2 data over the Indian subcontinent is still lacking.
For example, Tripathi et al. [2005] studied 1 year of Level 2
data whereas Jethva et al. [2005] and Prasad and Singh
[2007] used Level 3 gridded data for their comparisons.
More details on their results will be discussed in the Results
and Discussion section. The present paper discusses the
validation of 4 years of the Level 2 MODIS retrieved
aerosol optical depth product over Ahmedabad, an urban
location in western India. Both versions of the aerosol
product are considered so as to be able to assess the
improvement in accuracy with transition from collection
version 4 to version 5. In the present validation study, the
aerosol product data from only the Aqua platform is used.
To the best of our knowledge, this is the first time that the
validation results from collection version C005 of MODIS
aerosol optical depth data over India are being reported.
[6] Section 2 provides a brief overview of the MODIS
retrieval algorithm for version 4 and upgrades to version 5.
Only the algorithm over land will be discussed. Section 3
discusses the study location and local meteorology which is
important while discussing the differences in the ground
truth values and the MODIS retrieved values of the aerosol
optical depth. Details of data analysis are given in section 4.
Results and discussion of the validation effort are presented
in section 5.
2. Overview of the MODIS Algorithm for Aerosol
Optical Depth Retrieval Over Land
[7] The retrieval philosophy of MODIS is based on the
transparency of aerosols in the mid-IR wavelengths. The
algorithm proceeds with the identification of dark targets in
the satellite image which are essentially the regions of low
surface reflectance as obtained from the mid-IR channel
reflectance. The surface reflectance in visible wavelengths
is obtained from the corresponding values in the mid-IR
using the empirical relations [Kaufman et al., 1997b,
2002b]:
R470 ¼ R2130=4; R660 ¼ R2130=2 ð1Þ
[8] These relations were derived based on the atmospheric
correction of the AVIRIS and Landsat Thematic Mapper
data during SCAR-A experiment and represent the parallel
processes affecting the surface reflectance in visible and
mid-IR wavelengths [Kaufman et al., 1997b, 2002b]. These
values along with the satellite and solar geometries are input
to the radiative transfer equation to derive the value of
aerosol optical depth. The operational procedure for the
retrieval process is the so-called Look-Up Table approach
wherein the satellite measured radiances in the blue and red
channels are matched against the simulated values of the
top-of-atmosphere reflectance. The cloud mask and gaseous
absorption is applied at the pre-processing stage. The
decision about aerosol model is taken based on the ratio
of path radiance in red and blue channels and the geograph-
ical location of the study area. Further details about the
algorithm are discussed by Remer et al. [2005] and Kaufman
et al. [1997a].
[9] The previous MODIS algorithm (Collection version
C004) has recently been updated [Levy et al., 2007b] to
improve its performance after the initial validation results
showed scope for further improvement when compared with
the ground based observations [Chu et al., 2002, 2003;
Ichoku et al., 2002, 2003; Remer et al., 2005; Levy et al.,
2005]. The modifications correspond to the inclusion of
polarization in the radiative transfer calculation [Levy et al.,
2004], the angular dependence of surface reflectance ratios
[Remer et al., 2001; Gatebe et al., 2001] and the update of
the aerosol models based on the aerosol climatology derived
from the worldwide AERONET observations [Dubovik et
al., 2002].
[10] The negligence of polarization in the radiative trans-
fer calculation at the computational stage may lead to
artificial errors in the results. Levy et al. [2004] performed
model studies to find how much effect the neglect of
polarization in the radiative transfer calculation will have
on the overall accuracy of the MODIS retrieved aerosol
optical depth. They found the difference to be positive or
negative depending on the scattering angle whereas the
magnitude of difference was dependent principally on the
sun and satellite geometry. Further, they foresee the differ-
D04203 MISRA ET AL.: MODIS AOD VALIDATION OVER WESTERN INDIA
2 of 12
D04203