s. The
e from
that the
n ó and
igations
lectance
nd 0 rss
sing the
y pixel,
ym the
ditions;
spheric
seen as
of soil
ngly on
TC GN
'scribed
ocedure
ased on
the real
easured
and soil
TE for
tigation
yield
nboard
zetation
from
10isture
simple
ven for
specific
ity; we
surface
\CABA
CABA,
TC and
proach
ethods,
here or
a large
(4)
where L,,; is the measured radiance in the i-th band, p, is the
Lambertian reflectance factor of the given pixel in the i-th band,
E,; (6; 05) is the global irradiation at the surface in the i-th
band, ó, is the normal optical depth in the i-th band, 6, is the
Sun zenith angle and L, , is the path radiance in the i-th band.
The 5S and the successor 6S code proved to be one of the best
RTC for remote sensing applications. However, the required
computational time is very huge ( corrections a full TM image
using 5S code requires weeks in CPU time, using 6S requires
even ten times more), thus in practical applications they are
inapplicable. Rahman and Dedieu (1994) developed a method
called SMAC based on approximations of 5S code runs for a
given atmosphere and aerosol model and for interval of
atmospheric conditions and in the case of TM for bands 1-5.
We developed a similar but computationally faster method to
speed up ACABA based on 6S code.
In Eq. (3.1) L,,; is measured by the satellite, 6, is computed by
6S from 5,,,, (given by ATC and the chosen atmospheric model
in 6S), E, , and L, , are also computed from 9 and 6, by 6S, thus
Eq. (3.1) can be solved for p,.
VERIFICATION OF ACABA
The verification of an atmospheric correction algorithm can be
done by using reliable data measured simultaneously at the
ground and onboard a satellite. The reflectance factor measured
at the ground and the corrected reflectance factor measured
onboard the satellite have to be compared in order to determine
the effectivity of an atmospheric correction algorithm.
Unfortunately, there is a lack of such data for Landsat TM.
However, Moran et al. (1992) published a data set from a
measurement campaign at the Maricopa Agricultural Center
(MAC, Phoenix, Arizona). They conducted measurements
during seven Landsat-5 passes with an Exotech 100 BX
spectroradiometer mounted on an aircraft flying at an altitude
of 150 m. The surfaces were different bare soils and vegetated
surfaces (cotton, winter wheat). They also made a comparison
between the main representatives of the atmospheric correction
methods for Landsat TM data. We completed their analysis with
the application of ACABA to their measurements. This way, we
could verify our method in absolute and relative senses.
The uncorrected (i.e. apparent) reflectance factor, the
reflectance factor corrected by DOS, the reflectance factor from
5S code with DOS input data and the reflectance factor from
LOWTRANT7 code (Kneizys et al. 1988) with DOS input data
are shown in Figs. 9a-d, respectively. The corrected reflectance
factors are shown versus reflectance factor measured by
EXOTECH 100 BX. To compare these results with ours, we
have to be aware of the following facts: MAC is in a desert
area, with very low humidity and stable weather conditions
(small spatial and temporal variations), what is not characteristic
of Europe.
The corrected reflectance factor from ACABA can be seen in
Fig. 10. Because we use the midlatitude summer atmosphere
model and the continental aerosol distribution for the
computation of L, in 6S code (i.e. in ACABA), the latter
necessarily overestimates the water absorption in the 4th band
and hence forward overestimates the reflectance factor too.
Taking this also into account, the ACABA proved to be as
accurate as the 5S code with DOS input in Moran et al. (1992),
which was the most accurate according to their analysis. The 5S
code with DOS input probably should produce a less accurate
result over Hungary and Europe, either because of worse
circumstances or because of the problem of "dark object". We
have to stress again, that only ACABA is independent of other
parts of the given image, i.e. of the location of the given pixel
in the image.
APPLICATION OF ACABA FOR REAL TM DATA
We applied ACABA to two TM images. The first one was a
TM image from South-East Hungary (Gyoma and Endród and
their environs with Berettyó and Sebes-Kórós rivers) of 2 May
1992.
Before we present the results obtained for this subimage, we
note the following: first, the ACABA is optimized for vegetated
surfaces and produces less accurate (for water) or false (for
concrete or asphalt) results on other surfaces. Second, as it can
be seen in Fig. 6, the é,.ss and HN are connected with a very
good linear relationship in the mathematical sense, but this line
has a "width", thus through HN,, BR; will also appear in §,,.ss
for the given pixel, i.e. the brightness variations of the surface
are a little transparent in the 0,5; image. Further improvement
in ACABA is required to eliminate this effect.
The original Band 1 and the corrected Band 1 image can be
seen in Figure 12a and b respectively (the colour Figures 12
and 13 can be found at the end of this volume, in the colour
section). Because the 'presence' of the atmosphere is the
strongest in the Band 1, the effect of the atmospheric correction
can be seen in this image clearly: the correction eliminates not
only the haze sheet, but partly the cloud as well.
Another example can be seen in Figure 13. A TM image from
Western Hungary containing light to heavy hazy areas was
corrected by ACABA. Comparing the original Band 1 image
(Fig. 13a) and the corrected one (Fig. 13b) the power of
ACABA is demonstrated. The enhancemant can also be seen
clearly in the Tasseled Cap Greenness images. The standard TC
Gn (Fig. 13c) shows less details under the haze part of the
image than the ACABA Gn image (Fig. 13d). The ACABA can
also be used for detecting or enhancing air pollution (smoke,
smog and forest fires etc.) by using 9,,,, images in environment
protection applications.
Because of the close similarity between the Landsat TM the
Indian IRS LISS bands, the ACABA can easily be ported for
IRS LISS data. Actually, the IRSIC LISS porting has beeen
done.
CONCLUSIONS
The new procedure for atmospheric correction of Landsat TM
images over vegetated surfaces is based on ATC, a Tasseled
Cap transformation with heuristic choice of axes, and on further
nonlinear transformations, and produces a true atmosphere
correction using only the pixel data themselves.
The full ACABA procedure completed with the use of the 6S
code can retrieve accurate reflectance factors for
multisensor/temporal data and with a the SMAC-like method is
able to perform the atmospheric correction of full images within
hours, and hence to serve a significant improvement even for
classification. The effectiveness of ACABA is demonstrated on
simultaneous ground based and satellite data measured over the
same area, and on real images as well.
Intemational Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 789
