Issue with computing phase trend (SBAS regression)

[Peau-Rouge]

Hello Alexey!

I'm encountering difficulties computing the trend (systematic phase) component from my SBAS interferograms using the sbas.regression method.

I have successfully unwrapped SBAS interferograms with sbas.unwrap_snaphu(intf_sbas, corr_sbas) and visualized them.

However, the regression fails to produce any output.

I’m attaching the notebook here: https://colab.research.google.com/drive/18hy8-4wIMvU1zc_OMKtXAu-mjWKgpa5J#scrollTo=DJcYUKIS0RZh

Thank you for your time,

All the best,

An

[AlexeyPechnikov]

I don't see any issues with the regression here, it is calculated:

https://colab.research.google.com/drive/1iTqVIOUKw4jQr2H8lYAD1vjB2-WK3Q5f?usp=sharing

In case it does not complete for you, I use 'L4 GPU' runtime (12 vCPU cores). For your selected runtime 'v2-8 TPU’, I have 96 cores, and some NetCDF lock-related issues are very likely on Google Colab storage.

[Peau-Rouge]

Hi Alexey,

Thank you very much for your feedback and for pointing out the GPU setup. In the meantime, I managed to fix the issue on my side, but I believe the problem with the trend calculation was indeed related to the TPU runtime. Your remark about using the L4 GPU makes perfect sense, and I’ll definitely keep this in mind for future runs.

Best regards,

[AlexeyPechnikov]

By the way, there are PyGMTSAR Docker images available at https://hub.docker.com/r/pechnikov/pygmtsar and tested starting from 1 CPU, 2 GB RAM host, so you definitely can run them literally on any hardware. See a table for different configurations and times on the page for reference.

Also, I found you don’t mask your water areas by average stack correlation and the results look noisy. It seems the water areas are shallow here and amplitude-based persistent scatterers metrics do not work for you as expected.

[Peau-Rouge]

Hi Alexey,

Thank you very much for these explanations. I now understand that the noisy results may be linked to water areas not being masked based on average stack correlation, and that shallow water can cause issues with amplitude-based persistent scatterers metrics. I must admit I’m not fully familiar with this aspect, so your remark is very helpful.

In our context, we are working on groundwater dynamics maps, combining monthly groundwater measurements with ground surface deformations inferred from Sentinel-1 DInSAR using the SBAS approach. The evolution of displacements is calculated with respect to the first acquisition, and the results are projected along the vertical direction to quantify soil compaction.

Would you have any practical advice or recommended strategies for better handling shallow water areas in this kind of workflow? For instance, is it mainly about applying stricter correlation thresholds, or do you suggest other approaches?

Thanks again for your guidance,
Best regards,

[AlexeyPechnikov]

Commonly, Goldstein filter helps a lot in the cleanup phase for such cases. Also, you would try to enhance coherence by fixing polarization change (i.e using two polarizations VV and VH to restore VV).