|
CHAPTER 13 SECTIONS > Geometric Performance | Radiometric Performance
13.1
Geometric Performance
The geometric performance of the ETM+ is judged against three key requirements
placed on the Landsat 7 system. They are:
Absolute Geodetic Accuracy
- Geometrically corrected products shall be accurate to 250 meters
(1 sigma), excluding terrain effects, without ground control.
- Limited by spacecraft/instrument geometric model accuracy (e.g.,
ephemeris, attitude, alignment knowledge).
Band-to-Band Registration
- Geometrically corrected products shall have the multispectral bands
registered to 0.17 pixels (1 sigma)
- Limited by focal plane alignment and stability
Image-to-Image Registration
- Geometrically corrected images from multiple dates shall be capable
of being registered to an accuracy of 7.3 meters (1 sigma)
- Limited by high frequency distortions within images (e.g., uncompensated
attitude jitter, scan mirror instability)
 Figure 13.1 - Across and Along Track Mean Geodetic Accuracy
of ETM+ Products Since Launch Geodetic accuracy is monitored using calibration scenes containing ground
control points. Scenes are first radiometrically and geometrically corrected.
Control point locations are then measured on the processed imagery and
compared to known ground locations. Any terrain effects are removed analytically
in the comparison. The product's geodetic accuracy depends on the accuracy
of four data inputs. These are:
- Ephemeris data - spacecraft position and velocity
- Attitude data - spacecraft orientation (roll, pitch, yaw)
- Spacecraft clock - links image data to ephemeris and attitude
- ETM+ alignment - orientation of payload to the spacecraft
The ephemeris is estimated post-pass using tracking data. Attitude is
measured by on-board star trackers and gyros. The clock performance is
monitored by the Landsat 7 mission operations center and the ETM+ alignment
is determined by ground processing calibration.
 Figure 13.2 - Across and Along Track Mean Geodetic Accuracy
of ETM+ Products Since Launch and Alignment Calibration The ETM+ alignment is determined by measuring the orientation of the
ETM+ payload relative to the L7 spacecraft attitude control system reference.
Multiple scenes with ground control are used to measure the systematic
biases attributable to ETM+ alignment. During the initial on-orbit calibration
during first ninety days, seven different calibration scenes were used.
Periodic geodetic accuracy testing showed a slow build-up of along-track
bias (Figure 13.1) after July, 1999. A sensor alignment calibration update was performed
in June, of 2000. Twenty-four scenes acquired since July, 1999 (~1 per
cycle) were used to perform the calibration. A separate independent set
of eighteen scenes covering same time span were used to verify the results.
The ETM+ alignment shows time-varying behavior and will continue to be
monitored during the course of the mission. Current trending reveals post-calibration (Figure 13.2)
geodetic accuracy of systematic ETM+ products is approximately 80 meters
(1 sigma) which is much better than 250 meter specification placed on
the system.
 Figure 13.3 - Primary Focal Plane Assembly Across Scan Band
Offsets Since Launch
Download large image (15 KB, GIF) Band-to-band registration assessment is performed periodically throughout
the mission's life. The purpose of this assessment is to measure the relative
alignment of the eight ETM+ spectral bands after processing to Level 1Gs
for verification that the 0.17 pixel band-to-band registration requirement
is being met. If not, the IAS remedies the band alignment by deriving
new band center locations via band-to-band registration calibration and
updating the CPF.
 Figure 13.4 - Cold Focal Plane Assembly Across Band Offsets
Since Launch
Download large image (15 KB, GIF) Band registration is monitored using desert scenes as they provide the
best cross-band correlation performance. The band center locations measured
prelaunch were evaluated using on-orbit data were updated using calibration
scenes during the in-orbit checkout period (first 90 days). The measured
band registration accuracy was 0.06-0.08 pixels. However, the registration
accuracy degraded after July, 1999. Measurements revealed that registration
between the primary and cold focal planes in the line direction increased
to 0.08-0.10 pixels.
Band calibration analysis showed a systematic shift in the band 5, 6,
7 locations after July, 1999. The primary
focal plane band centers (Figure 13.3) are very stable but the cold
focal plane band centers (Figure 13.4) more variable with a 3-4 microradian mean
shift. The cold focal plane offsets coincide with change in ETM+ operating
temperature range which is hotter than during the 90-day checkout. The
band center calibration was updated for data acquired after July, 1999
although analysis revealed that band center estimates from individual
scenes are still highly correlated with temperature telemetry. The current
calibration is time-dependent pending development of a temperature dependent
model. Nonetheless, registration performance (Figure 13.5)
was well within 0.17 pixel specification.
The multi-temporal registration accuracy was determined using cloud-free
scenes of the geometric calibration "super-sites". The term "geometric
super-site" is used to describe those pre-selected WRS path-row locations
for which ground control, digital terrain data, and reference imagery
have been collected. This supporting data set makes it possible to produce
precision and terrain corrected ETM+ images, and to compare them to accurately
geo-registered reference images.
The required ground control, terrain models, and reference images were
derived from digital orthophoto (DOQ) data. The one meter resolution DOQs
were mosaicked and reduced to 15 meter resolution. Five cloud-free images
of two separate calibration sites were used to measure registration accuracy.
The image registration assessment was performed in two ways. First, the
ETM+ images were compared against the DOQ reference images. This provides
a measure of image distortion relative to an absolute reference. Second,
two ETM+ scenes were cross correlated. This provides a measure of image
distortion that changes scene-to-scene although systematic calibration
distortions may cancel out.
 Figure 13.6 - Pre and Post Scan Mirror Calibration Registration
Accuracy
Download large image (13 KB, GIF) Coupled with the image registration analysis is the need to measure the
ETM+ scan mirror performance to ensure the pre-launch profile is correct.
The geometric calibration super-site scenes were also used for this purpose.
A DOQ reference image was constructed to provide full-width coverage of
a Landsat 7 scene so that measurements at all scan angles could be obtained.
Mirror deviations as a function of scan angle were obtained by ??? These were
cross-correlating the ETM+ scene to the reference image.
No apparent problem was observed with the along scan mirror profile.
A minor adjustment to the cross scan profiles was made to model slightly
non-linear scan line corrector behavior. Also, an adjustment to the prelaunch
scan angle monitor start/stop angles was made to improve image-to-image
registration accuracy. The scan mirror calibration update was made to
the CPF in the fourth quarter of 1999.
Image registration accuracy (Figure 13.6) was
measured before and after the scan mirror calibration. Results revealed
that the required image registration accuracy was achieved using the baseline
prelaunch scan mirror calibration parameters. Specifically,
- All-scene average registration to DOQ:
- 5.8 meters along scan and 4.7 meters across scan (1 sigma)
- All-scene average ETM+ to ETM+ registration:
- 3.9 meters along scan and 1.8 meters across scan (1 sigma)
- Two individual scenes fell outside specification versus DOQ.
The image registration also improved using the updated scan mirror calibration
parameters. Analysis yielded the following results.
- Registration to DOQ:
- 4.3 meters along scan and 4.2 meters across scan (1 sigma)
- ETM+ to ETM+ registration:
- 3.2 meters along scan and 1.9 meters across scan (1 sigma)
- All scenes were within specification.
Regular monitoring has revealed the ETM+ scan period is increasing with
time due to growth in turnaround time, probably caused by bumper wear.
However, the active scan time is showing increased variability, especially
in the forward scans. The rate of growth (Figure 13.7)
of turnaround time, however, appears to be stable. The impact to ETM+
data is that scan gaps will gradually increase with time. Also, the scan
mirror could, theoretically, lose synchronization with the calibration
flag if scan time gets too long. This would effectively end the Landsat
7 mission. However, this mirror behavior is similar to that observed in
Landsat 5 TM.
Figure 13.9 - ETM+ GXA Data Anomaly The gimballed X-band antenna (GXA), when maneuvered in an across-track
slew, sometimes disturbs the ETM+ scan mirror. This occurs when GXA the
stepper motor frequencies correspond to scan mirror harmonics. The impact
to ETM+ data is a wider than normal (Figure 13.9) variation
in scan line length. Most extreme examples exceed the maximum allowable
scan length leading which leads to dropped scans. This phenomenon was
not observed on earlier missions as pointable X-band antennae are new
on Landsat 7.
 Figure 13.10 - ETM+ GXA Data Anomaly The scan mirror may lose synchronization or, in extreme cases, restart
during imaging. Such an occurrence is correlated with regions of high
electron flux (Figure 13.8) at 705 km orbital altitude particularly in polar regions
and within the South Atlantic anomaly. The correlation was confirmed by
a July, 2000 solar flare which resulted in 14 anomalies in a single day.
The impact to ETM+ Data is dropped scans and calibration
flag incursions (Figure 13.10) into the Earth imaging area. The scan mirror controller
sees an "extra" timing pulse and thus loses synchronization with calibration
flag. This phenomenon may have occurred on Landsat 5.
On May 31, 2003 the scan line corrector (SLC),
which compensates for the forward motion of the satellite, failed at
approximately 21:45.
Subsequent
efforts
to recover
the
SLC
have not
been successful, and the problem appears to be permanent.
Landsat 7 Enhanced is still capable
of acquiring useful image data with the SLC turned off, particularly
within the central portion of any given scene. Various interpolation
and compositing schemes are currently be investigated to expand the coverage
of useful data. An interpolation example can be seen in Figure 13.13.
Landsat 7 ETM+ will therefore continue to acquire image data in the "SLC-off" mode
and has so since July 14, 2003. More information on the SLC-off anomaly
can be found in Chapter 11.
Figure 13.11 - Top image: Pre-SLC anomaly scene, middle
of image.
Middle image: Scene after SLC anomaly.
Bottom image:
Scene after SLC anomaly, with interpolation.
13.1.7 Modulation Transfer Function (MTF) Characterization |
Back to Top |
The MTF of the ETM+ is regularly measured for comparison to the pre-launch
test results and for monitoring long-term instrument performance. The
MTF characterization methodology involves the analysis of cloud
free scenes (Figure 13.12) over the Lake Ponchartrain
bridge (Figure 13.13) in Louisiana. The bridge is long, straight, double-spanned
(two 10m spans, separated by 24.4m) is approximately aligned with the
Landsat ground track (<1°), and offers high signal contrast to the
waters below.
Figure 13.12 - ETM+ Image of the MTF Characterization Site  Figure 13.13 - ETM+ Images of Lake Ponchartrain Bridge  Figure 13.14 -
Lake Ponchartrain Bridge
Crossover The MTF characterization is performed on level 1R data. The 1R data is
in 16-bit scaled radiance form with image artifacts removed. The data
undergoes no geometric resampling. A 256x2048 image window covering the
bridge is manually identifed and extracted. The image line numbers containing
bridge crossovers (Figure 13.14) are identified and
removed from the extracted image. The corresponding windows for the 30-meter
bands are then extracted.
A bridge profile is then built by extracting the bridge segments from
each data line. Each segment consists of 16 pixels centered about a 3-point
moving average. The forward and reverse scans are separated and classified
according to bridge segment phase. Phase is determined by correlating
each bridge segment to bridge templates (Figure 13.15)
templates shifted at 0.125 pixel increments from -1 to +1 pixels (17 total).
Each segment is assigned to a phase "bin" based on the offset of the best
matching template. The range of 8 consecutive bins with the most segments
is then identified. The segments within this bin range are then averaged
resulting in 8 mean bin segments containing 16 samples each. The samples
from the mean bins are interleaved and the end result is a 128 pixel oversampled
bridge profile.
An analytical bridge model was constructed in the frequency domain using
the bridge span width, gap between spans and the intensities of the two
spans and background water as model parameters. An optical transfer function
(OTF) that borrows from previous Landsat 5 TM work was developed for ETM+.
The OTF is a mathamatical statement that describes the relationship between
the input and output of an imaging system. It provides a complete measure
of system performance in that it includes the phase relationship as well
as the amplitude degradation of the bridge as the frequency changes.
The OTF is then used to compare the bridge models with the actual data.
The forward and reverse scan bridge models are multiplied by the OTF model
in the frequency domain. An inverse FFT is applied to convert the models
to the space domain. The root sum square (RSS) differences between the
models and the data for both forward and reverse scans are then computed.
To minimize the RSS difference the bridge intensities were adjusted while
keeping the span width and separation fixed. The RSS differences were
also minimized by adjusting the variable OTF parameters while keeping
the detector model fixed. Numerous iterations yielded a final OTF model.
|
