ear
fan
ıtter
dow
le 1
ated
ble.
‘the
re3
the
bia.
rent
the
eof
ze it
age
Nas
han
lual
ues
ack
nal,
ing
ally
lual
lual
Is).
nce
fter
71
processing are so small that they are comparable to an along track interference pattern which, it is thought, is caused by
interference from the payload control data (PCD) sampling (= 32 khz). This interference pattern is usually so small that it can
only be detected through the use of Fast Fourier Transforms or autocorrelation functions.
DETECTO BAND 1 BAND 2 BAND 3 BAND 4 BAND 5* BAND 7
1 216.35583 91.01775 146.59180 85.23219 375.10727 587.15314
2 214.82460 90.95529 148.74100 84.86255 371.20142 579.55518
3 215.61977 90.03220 145.61705 84.80483 372.48212 585.76086
4 213.76900 91.11665 146.86333 83.56823 367.03473 585.35052
5 215.66486 90.19747 146.41582 84.66369 375.79434 590.46161
6 212.16615 91.46665 144.38597 84.89708 369.05185 972.74518
7 214.34270 89.85090 144.03462 85.61960 365.07385 580.03027
8 212.26170 91.63770 145.98790 85.10206 379.22787 574.37115
9 213.34251 89.52427 144.33228 83.57074 372.26974 582.75049
10 214.78102 90.43723 146.01863 85.16158 373.74826 573.30853
11 215.06982 90.68965 144.87016 84.80897 371.25192 587.31573
12 214.76381 90.89744 146.63857 84.36308 372.84979 977.11377
13 214.40010 90.07504 145.03952 84.31039 373.26019 578.58093
14 213.00774 89.80899 146.24689 85.02716 371.83533 574.29248
15 215.22348 90.61319 146.39751 84.62737 374.86841 581.24176
16 215.40404 89.33324 147.51375 84.78180 375.70493 581.90546
r
Table 2. Equalized detector gains [ Gl] ri | for Landsat 5 Thematic Mapper in units of DN per (mW/cm?2—st).
One problem observed was the occasional appearance of a single scan droop stripe that would begin at one side of the image
and fade away toward the other side. It was noted that, for those scans, the difference between the prior scan's average ADC
and the current scan's BDC was biased by as much as 0.5 DN over the surrounding scan's values. A validity check was added
to detect such occurrences and replace the non-representative value with the current scan's ADC value.. Because of the
relative consistency of dark shutter values within an image, this procedure proved adequate for eliminating these stripes. No
explanation has ever been postulated for the existence of these spurious values.
Some concern existed over the potential impact of another Landsat 5 TM artifact on this scan by scan compensation.
Sometimes called memory effect or bright target recovery, it causes a detector to remain at full saturation output, sometimes
for several hundred pixels, followed by occasional “ringing” after the detector has been saturated by a bright target (often
clouds). If sucha situation were to occur atthe end of a scan, justbefore the dark shutter and lamp, it was thought that the effect
might spill over to the shutter collection and interfere with the bias compensation. In fact, this memory effect did occur in the
Colorado Springs image, in clouds, at one edge of the image. It produced some highly visible, long, but temporary streaks
(each detector reacts differently). Yet it had no apparent impact on the quantitative window average or on the striping
suppression. :
7. CONCLUSIONS
Certain conclusions seem hard to ignore. First, the gains of individual detectors within a spectral band are extremely stable
relative to one another. Any relative drift would have manifested itself as streaking after the correction was applied, since the
same gain correction was used across all seven years of test data used and, at most, the apparent drift was limited to a fraction
of adigital number. Fromthis itcan be concludedthat he internal calibration lamps are not necessary for the removal of image
artifacts (although they may be necessary for tracking overall array drift), and that the simple application of the constant gain
coefficients in Table 2 can be used for improving all Landsat 5 TM imagery.
If, in the future, the detector gains do start to drift apart, the search procedure described earlier can be used to locate a new
uniform area and to recalculate the gains.
Even more importantis the proper selection and application of the window data within the dark shutter used for determining the
average offsets for thatscan, detector and sample. Most ofthe striping can be eliminated as long as both a window before and
after (but not inside of) DC restore are selected, and a value interpolated from these selections is used to remove the bias at
each image sample location.
IAPRS, Vol. 30, Part 5W1, ISPRS Intercommission Workshop “From Pixels to Sequences’, Zurich, March 22-24 1995