ding
ions
eing
ition
with
jally
the
and
din
tion
sing
acts
ere
and
ring
ited
lute
256.
hen
nof
gle
nto
rely
ns,
ose
ted,
tion
ing
will
ISOr
tion
nity
the
67
As an adjunct to this equalization process, new processing techniques involving the use of dark shutter information were
investigated for their potential to minimize remaining image artifacts. One technique based upon characteristics of the DC
restore function and short term time variations in detector response was proposed as the most effective candidate.
In order to test the hypotheses of whether such equalization techniques would prove effective in removing streaking and
striping, it was decided that, since no data is yet available from the ETM+, some simulated version of the calibrator panel
equalization would be used to calculate correction coefficients which could be applied to real TM reflective band imagery. This
simulation was created by locating an extremely uniform region within a Landsat 5 TM image, assuming that it was perfectly
uniform, calculating the correction coefficients, applying them to the entire image from which the extraction was made, and to
other images, both of the same scene as well as others collected at different times, and evaluating the results both
quantitatively and qualitatively. A similar procedure was investigated as an alternative to using pre-launch measured detector
gains for Landsat 4 (Fischel, 1984). The details of the process follow.
2. CHARACTERIZATION OF THEMATIC MAPPER RESPONSE
At the end of every scan of the TM, a shutter passes in front of the focal plane. The shutter provides both a dark source
(blocking light from the aperture) and a light source for reflective band calibration. The light intensity is varied by turningon and
off, in different combinations, three different calibration lamps providing seven different light levels. Each level is maintained
for approximately 40 scans. Therefore, when provided radiance estimates for the light levels, eight different calibration points
are available for establishing an estimate of each detector's response curve. Since the detectors are assumed to be linear, a
linear least squares curve fit is calculated for all of the points providing a gain and bias value for each detector.
Inthe ETM+ the configuration is very similar, except that only one lamp will be used, thus providing a single light intensity level
instead of seven non-zero levels.
3. ESTABLISHMENT OF A UNIFORM EQUALIZATION TARGET
Creation of an equalization target required two steps. First, a Level 0 (raw, uncalibrated detector outputs) image of a known
relatively uniform scene was selected. In this case it was fortunate that Santa Barbara Research Center, the developer of the
TM, was kind enough to lend us images of the Libyan desert collected by Landsat 5 from 1984 to 1991, which they had used for
detector response drift studies in recent years. The August 27, 1991 image was selected as the source for equalization.
Second, an automated search of the image from spectral band 1 was conducted to find the most uniform area withinthe scene.
A sliding window, 512 by 512 pixels in size, was placed over the image, beginning in the upper left hand corner. The standard
deviation of the pixels within the window was calculated and stored with the window location. The window was then repeatedly
shifted maintaining a 64 pixel overlap with the previous window, and the accompanying standard deviation calculated for each
area until the entire image had been covered. That window of pixels with the smallest standard deviation was assumed to be
the most uniform area within the scene. Inthe case ofthis image, this was in the location starting at sample 449 and line 2049
(lines are designated as running in the cross track direction). Statistics for this region were:
[12115.1 DN
-2.51 DN
Min=104 DN
Max=149 DN
The standard deviation was actually small enough to be comparable to the estimated system noise level.
4. EXTERNAL EQUALIZATION PROCESSING
The average Entrance Aperture Radiance (EAR) values were calculated from the uniform source image as follows:
For each reflective band, m, and each detector, k:
i=1 j=1
hn T (nscan) (nsamp) (1)
nscan NSaMP (ea, source, ;-bias; ;)
Ee
IAPRS, Vol. 30, Part 5W1, ISPRS Intercommission Workshop “From Pixels to Sequences’, Zurich, March 22-24 1995