Issue displacement time-series

[Malvas]

Dear Yunjun,

I have processed my stack of interferograms with ISCE (stackSentinel.py). In the routine workflow of MintPy I chose the bridging method to correct unwrapping errors and pyaps to correct tropospheric delay. I also removed quadratic phase ramps. I am not sure about the results looking at the three marked cases in the displacement time-series, which I attached to this post (they do not seem very realistic). How can I improve my results?

I would appreciate your suggestions/help!

Best regards,

Alejandra

[Zhang Yunjun]

Hi Alejandra,

These patterns look like tropospheric/ionospheric turbulence to me, unless there is other information that could confirm it's surface deformation. This is not surprising as the atmospheric correction is an active area of research and far from perfect.

If you would like to focus on the geophysical interpretation, you could simply treat them as outliers and exclude them from your analysis using mintpy.residualRMS.* options.

Yunjun

[Malvas]

Hi Yunjun,

thank you for your reply. Well, indeed there was a volcanic explosion on 2019.20.09 (the second one marked in the figure) with an ash cloud height of ~3.4 km and another one on 2020.08.08 (the last one marked in the figure) with an elevation of the ash cloud of ~5 km. These events could explain the tropospheric turbulence, as you suggest. However, there is no information related to surface deformation for the 2019.28.06 (the first one marked in the figure).

Regarding the residual RMS, this displacement time series already includes this setting:

########## Phase Residual for Noise Evaluation
## Phase Residual Root Mean Square
mintpy.residualRMS.maskFile  = maskTempCoh.h5
mintpy.residualRMS.deramp    = linear
mintpy.residualRMS.cutoff    = 3

Do you maybe have another suggestion about this setting?

Best regards,

Alejandra

[Zhang Yunjun]

Hi Alejandra,

Very cool. It's interesting to know there are activities on 2 of those dates, it's will be good to confirm the if the SAR image at 20 Sep 2019 is acquired before or after the event. The SAR acquisition time in saved as CENTER_LINE_UTC attribute.

Since these pattern are now possibly real surface deformation, how to treat / interpret them is subjective to your geophysical judgement. A lot of volcano geodesy papers have discussed related issues.

On the software/processing side, you simply change the mintpy.residualRMS.cutoff value to exclude the noisy dates, which of course would change the confidence interval in terms of outlier detection as described in my paper.

Yunjun

[Malvas]

Hi Yunjun,

Thanks for your comments. Indeed! The volcanic explosions on those dates (2019-09-20 and 2020-08-08) occurred before the SAR images were acquired. That explains the possible tropospheric-turbulence patterns. Nevertheless, there is a strong signal on one date (2019-06-28), but no information related to ground deformation (either edifice-wide nor volcanic-wide) for that date. There is also no deformation evidence for the 2019-08-15.   

I followed your suggestion about the residualRMS.cutoff value. I tried it with different cutoff values: 2, 1.5 and 1. Unfortunately, it does not seem to improve the results (see figure 2). Additionally, I made other changes like these:

Before: mintpy.unwrapError.method = bridging  

After: mintpy.unwrapError.method = bridging+phase_closure

Before: mintpy.networkInversion.waterMaskFile = no

After: mintpy.networkInversion.waterMaskFile = waterMas.h5

Before: mintpy.networkInversion.minTempCoh = auto

After: mintpy.networkInversion.minTempCoh  = 0.6

Before: mintpy.residualRMS.deramp  = linear

After: mintpy.residualRMS.deramp  = quadratic

Before: mintpy.residualRMS.cutoff  = 3

After: mintpy.residualRMS.cutoff  = 1

A general problem that I have with these SAR images is the low coherence in spite of "good conditions" of perpendicular baseline and temporal resolution (see figure 3). However, the vegetation of the zone is quite dense. For the ISCE-processing, I used the stack Sentinel processor, with a number of connections between each date of 5, a filter strength of 0.5 and 9 looks in range and 3 looks in azimuth

I would appreciate, if you took a look at this case and suggested to me please how to improve the results.

Thanks a lot!

Best regards,

Alejandra

[Zhang Yunjun]

Hi Alejandra,

The residualRMS option helps the velocity estimation, but won't affect the time-series result. However, it does provide a very subjective "justification" to discard some of the acquisitions for data interpretation.

I would suggest trying the ISCE-2 stack processing with more looks and stronger filtering strength if the resulting low spatial resolution and potential signal distortion is acceptable to you. 

In areas with very dense vegetation, "there is no magic" as my friend says. Until the day the free-and-open L-band NISAR mission is up and running, our lives will be less affected by decorrelation.

Yunjun

[Malvas]

HI Yunjun,

Thanks for your comments. Then I misunderstood the residual phase RMS, because -as far as I understood- the RMS calculates the residual phase time-series (phi-resid). For that reason, I expected a change in the displacement time-series. Thanks for the clarification.

More looks could introduce more distortion in this case. That means I can't do much with my results.

By the way Yunjun, your work with this package is really great! Thanks a lot!

Best regards,

Alejandra