GAMMA Questions

Hello Cynthia,

Phase unwrapping errors are best seen by displaying the phase so that jumps of
2PI are visible. For example with the program disdt_pwr.py use min and max
values -6.28 6.28 with cyclical display (cflg =1). I use the the rmg.cm color
map, but jet.cm are also good alternatives. The visdt_pwr.py  program operates
similarly.

A phase unwrapping error is defined as adding an incorrect multiple of 2PI to
the phase to restore the phase value prior to wrapping. There is always an
unknown absolute phase constant, so some point in the scene must be assigned as
the reference point, where the assigned multiple of 2PI is 0.

Spatial or temporal continuity of the phase is used as a basis for judging if
the phase is correctly unwrapped, but that can be misleading if the physical
process or geometry actually lead to phase jumps. Examples of physical processes
that can lead to phase jumps are earthquakes or fast moving glaciers. Image
geometry such as layover can also lead to jumps of the phase at linear
boundaries.  Geometrically related errors are common.  Removing any known phase
trends prior to unwrapping will reduce errors.

It is not always clear where errors have occurred and it is a topic that you
should discuss with other experienced interferometry users. There have been a
number of attempts to building neural networks to recognize phase unwrapping
errors. There are also techniques to build radar systems that are robust in this
respect. For example using multiple baselines can help recognize errors, Also
creating different interferograms in a time series using different combinations
can identify incorrectly unwrapped interferogram regions.  High temporal and
spatial sampling of the data always help since then the phase increments between
adjacent samples is small, reducing the potential for phase unwrapping errors
due to noise from decorrelation.

The fundamental problem with recognizing incorrectly unwrapped interferogram
regions is that there is no unique answer just from the interferogram itself.
Furthermore, there are often regions with just random phase from decorrelation,
with no deformation information. It can also be that due to geometry or low
spatial resolution, the phase values are insufficiently sampled so that they can
be unwrapped. One of the techniques that can be applied that is very effective
in improving phase unwrapping is to oversample the SLC by a factor of 2 prior to
forming the interferogram. The reason for this is that the spatial bandwidth of
the interferogram is twice as wide as the SLC! The interferogram is formed by
cross-correlation of the SLC images and this effectively convolves the spectra.
Oversampling prevents aliasing of the interferometric phase!

Your interferogram looks actually quite well processed. There is at least one
area that is probably not unwrapped correctly and that is  the colored region at
the bottom of the scene, where there appear to be discontinuities.

What I suggest is that you not mask the scene based on correlation at all, and
only mask the data AFTER unwrapping. Our experience has been that masking areas
can actually generate unwrapping errors, since the algorithm attempts remove
inconsistencies in the phase and masking out noisy areas drives the
inconsistencies apart, when actually they are close together and easily
resolved. Over filtering of the interferogram prior to unwrapping can also lead
to persistent phase unwrapping errors. I suggest that you use the newer filter
programs adf2 that adaptively changes the filtering parameter for filtering
based on the interferogram coherence prior to unwrapping. It applies very little
filtering in areas of high coherence.

Looking at the unwrapped interferogram including the noisy areas, gives better
context and confidence that the unwrapping proceeded correctly.

Having multiple interferograms is also very useful since unwrapping errors are
often revealed when looking at multiple interferograms.

Best regards,

Charles

Hello Cynthia,

Phase unwrapping errors are best seen by displaying the phase so that jumps of
2PI are visible. For example with the program disdt_pwr.py use min and max
values -6.28 6.28 with cyclical display (cflg =1). I use the the rmg.cm color
map, but jet.cm are also good alternatives. The visdt_pwr.py  program operates
similarly.

A phase unwrapping error is defined as adding an incorrect multiple of 2PI to
the phase to restore the phase value prior to wrapping. There is always an
unknown absolute phase constant, so some point in the scene must be assigned as
the reference point, where the assigned multiple of 2PI is 0.

Spatial or temporal continuity of the phase is used as a basis for judging if
the phase is correctly unwrapped, but that can be misleading if the physical
process or geometry actually lead to phase jumps. Examples of physical processes
that can lead to phase jumps are earthquakes or fast moving glaciers. Image
geometry such as layover can also lead to jumps of the phase at linear
boundaries.  Geometrically related errors are common.  Removing any known phase
trends prior to unwrapping will reduce errors.

It is not always clear where errors have occurred and it is a topic that you
should discuss with other experienced interferometry users. There have been a
number of attempts to building neural networks to recognize phase unwrapping
errors. There are also techniques to build radar systems that are robust in this
respect. For example using multiple baselines can help recognize errors, Also
creating different interferograms in a time series using different combinations
can identify incorrectly unwrapped interferogram regions.  High temporal and
spatial sampling of the data always help since then the phase increments between
adjacent samples is small, reducing the potential for phase unwrapping errors
due to noise from decorrelation.

The fundamental problem with recognizing incorrectly unwrapped interferogram
regions is that there is no unique answer just from the interferogram itself.
Furthermore, there are often regions with just random phase from decorrelation,
with no deformation information. It can also be that due to geometry or low
spatial resolution, the phase values are insufficiently sampled so that they can
be unwrapped. One of the techniques that can be applied that is very effective
in improving phase unwrapping is to oversample the SLC by a factor of 2 prior to
forming the interferogram. The reason for this is that the spatial bandwidth of
the interferogram is twice as wide as the SLC! The interferogram is formed by
cross-correlation of the SLC images and this effectively convolves the spectra.
Oversampling prevents aliasing of the interferometric phase!

Your interferogram looks actually quite well processed. There is at least one
area that is probably not unwrapped correctly and that is  the colored region at
the bottom of the scene, where there appear to be discontinuities.

What I suggest is that you not mask the scene based on correlation at all, and
only mask the data AFTER unwrapping. Our experience has been that masking areas
can actually generate unwrapping errors, since the algorithm attempts remove
inconsistencies in the phase and masking out noisy areas drives the
inconsistencies apart, when actually they are close together and easily
resolved. Over filtering of the interferogram prior to unwrapping can also lead
to persistent phase unwrapping errors. I suggest that you use the newer filter
programs adf2 that adaptively changes the filtering parameter for filtering
based on the interferogram coherence prior to unwrapping. It applies very little
filtering in areas of high coherence.

Looking at the unwrapped interferogram including the noisy areas, gives better
context and confidence that the unwrapping proceeded correctly.

Having multiple interferograms is also very useful since unwrapping errors are
often revealed when looking at multiple interferograms.

Best regards,

Charles

Dear Charles,
Thank you so much for your helpful and patient advice.
I will try following your way.
Very appreciate.
Best regards,
Cynthia