International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
will not stop, but, error pixels or lines could be included in the
data set.
2.3 Radiometric Performance
As shown in the Figure 3, data acquisition for the radiometric
calibration has been periodically performed, firstly in 33 days,
then 17 days, after having confirmed the stability of the sensor
responsibity. The result of the calibration has been listed in the
form of the radiometric calibration table and reflected to the
Level 1 Processing. In summary, 3 band data in VNIR have
been degraded 15 to 20 % for these 4 year’s operation. This
degradation is strongest in the Band 1 and weaker in the longer
wavelength. This degradation in SWIR is relatively weaker
compared with VNIR. On the other hand, this change rate in
TIR is significant. tin The detail of the radiometric
performance should be referred to the literature. The calibration
table is updated every time the change exceeds the
predetermined criteria. By this calibration, the DN value in the
Level 1B data can be converted to the radiance value, by using
a simple formula (DN-1)*conversion factor. The interval of the
change of the radiometric calibration table is shortest in TIR as
expected. And, the change of the responsibility is shown in the
figure 5, Tonooka, 2003, where the discrepancy before and after
correction is significant when the interval of the change of the
calibration table is too long. In the worst case, this difference
can be larger than 2 degrees in Band 12 in terms of temperature.
To compensate this discrepancy, a formula of 3™ order
polynomial in function of the data acquisition time is proposed
by Tonooka, 2003.
Band 12
Y AARAU TUT U TT |
ee en? Sen 99888, gasgpgé 6e, Ren.
= + X X X X X 1
> x x x
: ^ x
> 1% M ,
5 % x
B 2 Mg L|
2 d FALLI ene X DM. ¥ vas Se es en ra ee TR i
S Ky h
X Original É i
© Recaiibrated "n
2 dL Ld aa i Ea
0 200 400 600 800 1000 1200 1400
Days since the launch
Fig. 5 Error in TIR and recalibrated values (Tonooka, 2003)
In the mean time, a phenomenon called “cross talk” was found
in 2002 in the decorrelation stretch image by SWIR. The cross
talk is the phenomena that a strong light in the Band 4 is
diffused to the adjacent bands, especially in band 9, which has
relatively weaker light. Since each band of the SWIR has
different look direction, sometimes, the light in Band 4 appears
in the Band 9 image, in a different location in along track
direction. This effect is most significant in the shoreline in cross
track direction. Again, a formula to compensate the cross talk
was proposed by Iwasaki, A, 2003 and can be applied to the
Level IB processing when requested. But, it should be noted
that the cross talk is so subtle that it cannot be seen in a simple
3 SWIR band image in most cases.
In addition, in March/April of 2003, the Deep Space and Lunar
Calibration were conducted by making a pitch maneuver of
686
Terra Satellite. The purpose of the Deep Space Calibration was
to check the Deep Space where there is no radiance, and that for
the Lunar Calibration was to check the radiometric calibration
without any atmospheric effect. The investigation is still
underway. But, some unexpected ghosts similar to the cross talk
were observed not only in SWIR bands.
2.4 Geometric Performance
As mentioned in the beginning of this manuscript, ASTER
sensor has three different telescopes, VNIR, SWIR and TIR,
where the first two are push bloom types and the last, whisk
bloom type. In addition, in SWIR, each band looks different
direction to obtain sufficient radiance value. As a result, the
intra- and inter- telescope band to band registration becomes
very important. All the processing of the registration is
described in the literature by Fujisada[.and Watanabe, 1994.
Target of the registration accuracy was to keep in 0.2 pixel in
intra-telescope registration and 0.3 pixel of the coarser
telescope in inter-telescope registration. Since the band to band
registration is a key factor to the success of the ASTER sensor,
a continuous effort to check the result of the band to band
registration has been performed. And all the band is registered
to the VNIR Band 2. The result of the effort shows that the
band to band registration satisfies the above mentioned target.
On the other hand, geolocation information is one of the
significant characteristics of ASTER sensor. In the level 1 S/W,
the geolocation is calculated as a cross point of the look vector
of each pixel and reference ellipsoid. Therefore, there are
following contributing factors to get the exact geolocation :
location and attitude knowledge of the space craft ( obtained by
TONS : TDRS Orbit Navigation System and Star Tracker
/Rategyro ), stability of the attitude, accuracy of the pointing
angle and its knowledge. To calculate the look vector,
geometric configuration of the sensors was carefully calibrated
during the Initial Checkout period. And the gcolocation has
been periodically checked. The result shows the geolocation in
VNIR of the LIB product is about 50 m in 3 sigma in the low
altitude area. This location is more accurate in L3A ( Ortho-
rectified ) data even in the high mountain, since geometric
distortion caused by the altitude is corrected. This amount of
geometric distortion could be more than hundred meters if the
pointing angle is large in the mountain. All of the geolocation
information is obtained without any use of geometric control
point. This geolocation accuracy is achieved in mid-latitude
region, and could be larger especially in high latitude region.
Again, a formula to correct this accuracy is proposed in the
Web site[###]. The height information in the L3A or L4
product is about 15 m except for the sharp edge of the mountain
or the landform, taking into account the height means here the
height is against the reference ellipsoid and not against the
geoid surface. This is because of the spatial correlation with 9
by 9 windows. Since all the processing is automatic, there are
some mismatching, especially in case of the very flat and
monotonous arca like playa. But in most area, the above
mentioned accuracy is well kept.
2.5 Data Distribution
All the processed ASTER data is available from ASTER GDS
in ERSDAC, Japan and EROS data in LP DAAC, in USA. The
data is separated as a scene. In most cases, data is acquired in
full mode. The product used in general cases is processed in
Level IB. A significant characteristics is that data set in full
mode is associated with backward looking data covering
Internati
approxim
approxin
Higher |
includes
reflectan
DEM/Or
version u
ASTER
Prod
Level 1.
Level 1
73A
2A03 V
2A038
2B01 V
2B0I S.
2B01 T
2B05 V
2B05 S
2B03
2B04
3A01
4A01Z
Once pre
distribute
of the À
CCD, bu
network,
One scer
240 MB.
This nety
America,
South-Ea
data use,
Data seat
the archi
efficient
of Terra
and also
between
applied a
specific z
in operati
Once dat
step. AS
CEOS. [i
But, rece
or GIS S,
capability
1) By ch
the «
follow
were |
accur:
sched
2) There
SWIR
is 60
172 ki
