GAMMA Questions

Hello,

I would look very carefully at the distribution of errors and the quality of the
geocoding.
For example, does the DEM have the similar resolution as the radar image?  I
would first look at how accurately the image was terrain geocoded, for example
looking at the residuals of the offset fit between the image and the simulated
radar image from the DEM. Once you are reasonable sure that the radar image and
the simulated image are coregistered well then you can proceed to using
rdc_trans to generate an initial lookup table to coregister the two images,
resample the secondary image, measure the offsets, and then update the lookup table.

There are 2 scripts in the gamma software, one is called SLC_coreg.py, and the
other is called SLC_resamp_lt_all, and this second script includes an additional
step for updating the entries in the lookup-table and also permits iterative
resampling.

Input to SLC_resamp_lt_all is a list of the SLCs, the reference SLC. All the
SLCs in the SLC_tab are then coregistered to the reference scene. There could
only be one SLC in the SLC_tab if you only have one pair. Also required is a DEM
resampled with the same resolution as MLI radar image.  The DEM is in
slant-range radar coordinates. The program pixel_area can be used to resample
the DEM from map coordinates (e.g. lat/lon or what ever projection you have) to
radar slant-range coordinates. SLC_resamp_lt_all is a script written in Perl and
you can follow all the steps. Similarly, SLC_coreg.py is written in python and
can be examined.

If the interferometric correlation between the scenes is low, then you will have
higher residuals.  The program dis2SLC vismph_pwr.py, or viscpx.py can be very
useful for visually examining the offset between different SLC images. I
strongly recommend examine only a part of the scene using the options to select
just a segment of the scene.

Best regards,
Charles

On 17.04.2024 14:50, zw Liu wrote:

GAMMA Questions - GAMMA Processing: Co-registration of large-baseline
imterferometric pair <https://dep1doc.gfz-potsdam.de/boards/9/topics/654>

/zw Liu/

Dear all,

When conducting co-registration for a large-baseline interferometric pair,
matching the parameters outlined in the GAMMA manual for "China LT1 (LuTan-1)
SAR support within GAMMA Software," and with an effective baseline exceeding
2000 meters, despite introducing an external DEM for co-registration, the
range error persists significantly, with a standard deviation exceeding 0.4
pixels. Do you have any suggestions to mitigate this range fitting error?

Furthermore, in the previously mentioned manual, the use of
"offset_pwr_tracking" is recommended to mitigate residual coregistration
errors. Are there any examples available for reference?
Thank you in advance for any guidance or suggestion.

---

You have received this notification because you have either subscribed to it,
or are involved in it.
To change your notification preferences, please click here:
http://dep1doc.gfz-potsdam.de <http://dep1doc.gfz-potsdam.de>

Dear Charles,

Thank you for the prompt response.

I'm utilizing TanDEM-X 90m DEM to assist in processing the high-resolution LT-1 data (1.6m and 2.1m). The coregistration between the master SAR image and external DEM seems to be successful, with final model fit STD of 0.1906 in range and 0.1243 in azimuth.

Despite attempting SLC_coreg.py and SLC_resamp_lt_all for co-registration as suggested, there hasn't been any noticeable improvement. All range coregistration errors persist significantly, surpassing 0.4 pixels. As a last resort, I turned to "offset_pwr_tracking" to estimate the offset field between the master image and the resampled slave image. However, this method resulted in higher residuals, around 0.8 pixels in the range direction.

Do you have any suggestions to address this issue?

Best regards,
Liu

Dear Liu,

You have to look also at the thresholds you set for accepting offset
measurements and how the interferometric correlation looks. Have you made
interferograms? Is the correlation high in areas that you expect to have high
correlation?
How many offsets were accepted out of the total number tested?

using offset_pwr_tracking is only necessary if ithere is an offset field that is
not described by the orbit data
How large are the initial offsets without refinement?
Having residuals like this is not necessarily an indication of a problem.  For
example, much of you scene many have low correlation.

Best regards,
Charles

On 18.04.2024 15:02, zw Liu wrote:

GAMMA Questions - GAMMA Processing: RE: Co-registration of large-baseline
imterferometric pair
<https://dep1doc.gfz-potsdam.de/boards/9/topics/654?r=656#message-656>

/zw Liu/

Dear Charles,

Thank you for the prompt response.

I'm utilizing TanDEM-X 90m DEM to assist in processing the high-resolution
LT-1 data (1.6m and 2.1m). The coregistration between the master SAR image and
external DEM seems to be successful, with final model fit STD of 0.1906 in
range and 0.1243 in azimuth.

Despite attempting SLC_coreg.py and SLC_resamp_lt_all for co-registration as
suggested, there hasn't been any noticeable improvement. All range
coregistration errors persist significantly, surpassing 0.4 pixels. As a last
resort, I turned to "offset_pwr_tracking" to estimate the offset field between
the master image and the resampled slave image. However, this method resulted
in higher residuals, around 0.8 pixels in the range direction.

Do you have any suggestions to address this issue?

Best regards,
Liu

---

You have received this notification because you have either subscribed to it,
or are involved in it.
To change your notification preferences, please click here:
http://dep1doc.gfz-potsdam.de <http://dep1doc.gfz-potsdam.de>

Dear Charles,

I selected a threshold of 0.8 (nr and naz: 64, rwin and azwin: 16) and accumulated more than
1000 points to fit the registration polynomial.

Moreover, the test area is predominantly flat, with only a few patches of rugged terrain.
Even when employing a 7*7 window for coherence estimation, the coherence is very low in whether flat or rugged area.

Best regards,
Liu

Hello Liu,

When the coherence is low, and the terrain is flat, use larger windows
(256x256). Is the coherence low due to ground cover? Without refinement, are the
coregistered images close to one another (pixel level offsets?) That can be
verified visually.

If the coregistration of the scene is acceptable (just a few pixels), and the
scene has few features, you may consider using refinement from only one point,
using the program offset_pwr_init to select a feature in the scene and use that
one patch with recognizable features.

Were the data collected from the same orbit?

Best regards,
Charles

On 19.04.2024 16:08, zw Liu wrote:

GAMMA Questions - GAMMA Processing: RE: Co-registration of large-baseline
imterferometric pair
<https://dep1doc.gfz-potsdam.de/boards/9/topics/654?r=658#message-658>

/zw Liu/

Dear Charles,

I selected a threshold of 0.8 (nr and naz: 64, rwin and azwin: 16) and
accumulated more than
1000 points to fit the registration polynomial.

Moreover, the test area is predominantly flat, with only a few patches of
rugged terrain.
Even when employing a 7*7 window for coherence estimation, the coherence is
very low in whether flat or rugged area.

Best regards,
Liu

---

You have received this notification because you have either subscribed to it,
or are involved in it.
To change your notification preferences, please click here:
http://dep1doc.gfz-potsdam.de <http://dep1doc.gfz-potsdam.de>

Hello Charles,

Yes, the data were collected from the same orbit (bistatic acquisition), and there is minimal ground cover.
I've experimented with both larger and smaller windows for feature selection, but there's no noticeable difference in the results.

A peculiar observation is that when estimating coherence from the original interferogram, it's very low,
but coherence estimation from the flattened interferogram indicates higher coherence in flat areas.

Best regards,
Liu