| In-
lite
end
nary
ISOTS
light
/el-2
LC
obal
sing
geta-
>c.24
se of
ducts
Kinds
ram-
itmo-
ducts
The
igher
Cor-
n In-
ed by
, and
ndard
rithm
ail al-
shows
lative
roper
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
3.1 Precise Geometric Correction
Satellite data is often used for validation using ground based mea-
surement. Therefore, it is needed to be precise position to pixel.
The accuracy of geometric correction is dependent on the ac-
curacy of the satellite position and attitude. The GLI precise
geometric correction algorithm enables to determine the precise
satellite position and attitude using ground control points (GCPs).
This algorithm converts satellite position, velocity and attitude
for one segment to navigation data. And then, precise satellite
position and attitude are determined by utilizing GCPs. Conven-
tionally, GCPs are obtained by human operator, but this algorithm
is able to extract automatically by using each 1 path data. Suf-
ficient numbers of GCPs are necessary for exterior orientation.
Two image patches contain coastal lines. One is a template from
the GSHHS fine coastal data. Another is a reference image made
by the binarization of the original image. The template is selected
as a GCP candidate from the GCP library which have to be pre-
pared. In this work, the GCP collection is realized by the template
matching. But it can get parameters to specify each scene. This
algorithm is able to determine precise satellite position and atti-
tude with methodology based on photogrammetry. The scene is
rectified mapped image(Latitude/Longitude coordinates or Polar
Stereo Projection). The rectification of original image is carried
out using the results of the exterior orientation. GTOPO30 is
the digital elevation data used for 250m geometric correction for
terrain elevation. Output pixel value is radiance [W/m?/str/um]
derived from LIB data. lkm precise geometric correction im-
ages are generated from LIB data. However, in 250m precise
geometric correction, the algorithm will start from L1A on nadir
data, which is similar to MODIS 250m processing described in
Hashimoto(2002).
3.2 Composite
After precise geometric correction processing, output value is
TOA radiance derived from L1B. When land users use optical
moderate resolution satellite data, composite is conducted for
clouds removal. And land users often use composited AVHRR-
NDVI products, which generated with the maximum value com-
posite (MVC) technique. MVC selects the maximum NDVI value
on a per pixel basis over a set compositing period. However,
this algorithm has problems associated with satellite sensor con-
ditions. In practice, composite data are complicated owing to the
intrinsic behavior of the sensor, surface bi-directional reflectance
factors with STSG (Sun-Target-Sensor Geometry), and contami-
nation of the various spectral response. Therefore, the constraint
view angle maximum value composite (CVMVC) technique (Cih-
lar et al.,1994a) is applied to generate these composites in GLI
higher level processing. This algorithm produces geometrically
corrected 16-day surface composites, which may select the best
value pixel over a composite period, based on cloudiness and at-
mospheric contamination. Cloud detection and screening algo-
rithm products cloud flags on a pixel basis. This is similar to a
threshold technique to prevent the selection of extreme off-nadir
pixels. The CVMVC technique works very much the same way as
the classical MVC. In Ist day composite, the pixels are selected
by satellite zenith angle, and save as work file 1. In 2nd day com-
posite, the pixels are selected by satellite zenith angle, and save
as work file 2. In 3rd day composite, the pixels are selected by
satellite zenith angle, and save as temporary file, and then maxi-
mum NDVI value is selected on each pixels in work file 1,2, and
temporary file. Therefore, work file 1 is maximum NDVI pixels,
and work file 2 is next large NDVI value pixels. After 4th day
composite, it is same way as 3 day composite. Output value of
this algorithm is TOA reflectance in VNIR and SWIR wavelength
region (Ch.1-29),
813
2e Alsat
pega Fo cos(0,)
where poss is GLI observed reflectance, Fo[W/m?/um] is irradi-
ance based on Thuiller 2002,
Lsac| W/m?/str/um] is GLI observed radiance, and 0,|rad] is so-
lar zenith angle. MTIR region (Ch.30-36) is GLI observed TOA
radiance. The GLI project adopt solar irradiance from Thuil-
lier2002 (Thuillier et al.,2003). GLI calibration team calculates
GLI solar irradiances. These are the weight-integrated Thuil-
lier2002 irradiances over GLI spectral responses. The GLI spec-
tral responses are running averaged +2 samples (data interval is
about Inm) window to reduce a measurement noise, and both
response and solar irradiance datasets are linearly interpolated
to 0.1nm spectral resolution before the integration. Irradiance
data sets are linearly interpolated to 0.1nm spectral resolution be-
fore the integration. Thuillier2002 is new spectral irradiance data
set included SOLSPEC observed data, which is the UV and visi-
ble solar spectrum acquired by various spaceborne sensors flown
during the ATLAS Space Shuttle missions (spectral resolution is
Inm-5nm, range 200nm-2500nm). In case of longer wavelength
than 2500nm, MODTRAN4.0 IR solar irradiance is used. And
the compositing algorithm will operate on projected gridded data
defined by the GLI tile scheme (total of 56 tiles). In this study,
250m composite algorithm is same as 1km algorithm, CVMVC.
But this paper shows the results of test running in central Japan
in current status.
3.3 Atmospheric Correction
Correction for rayleigh scattering and ozone absorption is applied
in GLI land algorithm. Correction for aerosol over land is not
conducted. Much of the computation during atmospheric correc-
tion requires intensive CPU time due to floating point process-
ing. Therefore, this algorithm is adopted method using Look-Up
Tables (LUTs). And this algorithm is applied after composite
processing. GLI land atmospheric correction has some assump-
tions. Ozone layers are above molecular layer, and all molecules
are above aerosols. Moreover, aerosol layer + ground surface is
assumed as Lambertian, and all layers are horizontally homoge-
neous. Rayleigh scattering and ozone absorption are corrected
with the assistance of ancillary data
NOAA/TOVS data set and GTOPO30. GLI observed reflectance
at Top-Of-Atmosphere is described as the following equation:
Pos (Tos; TR, 0s, 0,; Qs-v) =
Tos(T03:05,0/) X
In, (TE; 0, )psTR1 (Tr, 0,)
1 — Sp(TR)Ps
(PR(TR, 05, Ov, Ps—v) + )
where pos, is GLI observed reflectance derived from compos-
ite algorithm, To, is ozone transmittance, pg is path radiance,
Tn, is downward transmittance, Tg, is upward transmittance,
Sg is Spherical Albedo, and ps is Rayliegh/Ozone Corrected Re-
flectance, which is output of this algorithm. The path radiance,
upward and downward transmittances and spherical albedo were
tabulated as functions of optical thickness of rayleigh scattering
(Tr), view and illumination angles. Then, these four values will
be retrieved based on the values of 7g for each pixel at each land
channel based on the pixel elevation, BPF, and STSG. 77 relates
to elevation through standard pressure and temperature (US62),
which can be calculated with GTOPO30. And To, can be de-
rived from NOAA/TOVS data. This algorithm for 250m data will
be also same as 1km algorithm. However, this algorithm is under
development.
