Tuning SARvey time-series & velocity for bridge monitoring and asc/desc V–H decomposition

[emanueliwanow]

Hi,

I'm using SARvey on top of an ISCE + Miaplpy workflow to monitor bridges (civil assets) with Sentinel-1 data. My current pipeline is:

• ISCE: Sentinel-1 TOPS stack (SLCs/IFGs)
• Miaplpy: stack loading + phase linking only
• SARvey: LOS time-series and velocity estimation from the phase-linked stack
• Custom scripts (outside SARvey):
  • Combine ascending and descending LOS results into vertical and horizontal displacements
  • Compute Hotelling T² statistics from these time-series for anomaly detection.

So SARvey is used for LOS time-series and velocity (not for Hotelling T²).

Context

• Input to SARvey: phase-linked stack from Miaplpy (with use_phase_linking_results = true)
• AOIs: very small, individual bridges (~500 m), with typically a few hundred to a few thousand points
• Using coherence-based masks (e.g. coh ≥ 0.7) and a limited number of interferograms
• Processing both ascending and descending tracks and then combining them outside SARvey into vertical and horizontal components

Issues I’m seeing

• LOS time-series and velocities over the bridge deck look noisier or less spatially consistent than expected.
• When I combine asc/desc into vertical and horizontal components (outside SARvey), the results are sometimes unstable or sensitive to configuration choices in SARvey.
• Some zones or areas have relatively few points, which may be affecting the robustness of the time-series/velocity estimates.

Questions

1. Configuration when using external phase linking

   When use_phase_linking_results = true and phase linking is done by Miaplpy:

   • Are there recommended values or good starting points for parameters such as:
     • num_siblings
     • use_ps
     • coherence thresholds and masks
   • Are there particular assumptions that SARvey makes about the phase-linked input that I should be careful to match?

2. Tuning for small AOIs (single bridge)

   For a single bridge or very small asset:

   • What do you recommend for:
     • Coherence threshold (e.g. 0.7 vs 0.8+)
     • Minimum number of points per zone or per asset
     • Minimum number of interferograms / time-span for stable LOS time-series and velocities
   • Are there adjustments to defaults that you would specifically suggest when the number of points is relatively low and tightly clustered?

3. Preparing data for asc/desc vertical & horizontal decomposition

   Since my goal is to use SARvey LOS outputs from both ascending and descending tracks to derive vertical and horizontal displacements outside of SARvey:

   • Are there any recommended settings or best practices to ensure SARvey’s LOS time-series are well-suited for this decomposition?
   • Any notes on:
     • handling reference points / reference areas,
     • ensuring consistency between how asc and desc are processed in SARvey,
     • or exporting SARvey outputs in a way that keeps all necessary geometry information?

4. Diagnostics and quality control

   Before I use SARvey outputs in my external vertical/horizontal decomposition and Hotelling T² analysis:

   • Which SARvey plots/outputs do you consider essential to check (e.g., residuals, spatial patterns of velocity, coherence distributions)?
   • Any recommendations on additional filtering (e.g. removing obvious outliers) at the SARvey stage to improve downstream robustness?

5. Example configs / case studies

   If you have:

   • Example SARvey configurations tailored for bridge or infrastructure monitoring,
   • Or papers/tutorials showing SARvey applied to small AOIs where the LOS is later decomposed into vertical and horizontal components,

   I’d be very grateful for any references.

I can share:

• A sample SARvey configuration JSON (including the phase_linking and preparation sections),
• Example LOS velocity / time-series plots for asc and desc,
• And logs, if that helps diagnose potential misconfigurations.

Thanks a lot for your work on SARvey and for any advice on using it for small civil structures with external V–H decomposition and anomaly detection.

[Andreas-Piter]

Hi @emanueliwanow,
Your questions are quite general, but I will try to answer your questions as far as possible. The choice of the processing parameters depend on the study area and the expected deformation behaviour.

1. In MiaplPy, you have chosen a number of siblings for phase linking (default: 10). As this value is not stored in the phase_series.h5 file and, hence, cannot be read from there, you need to insert this value in the SARvey config. It is used to differentiate PS candidates from DS candidates. I would recommend to leave use_ps to false as the selection of PS in Miaplpy might lack robustness for the task. PS pixels will anyway be detected with SARvey.

2. The reliability of the pixel selection usually increases with increasing stack size. Use a lower threshold for datasets with more images, but keep the coherence threshold high for few images in the stack.
   Your use case (observing small bridges with Sentinel-1) might be challenging particularly due to the spatial resolution of the sensor.

3. We do not have any suggestions or best practices for the decomposition of the signal. If you gain experience with it, we are happy to hear your suggestions. Contributions are welcome!

4. Please always check your displacement results for plausibility based on the expected displacement behaviour. In SARvey, a temporal filtering is not applied to keep the original signal which also represents the noise level of the scatterers. If your application requires filtering the noise, please apply filter on the exported results. The definition of outliers might also be specific to your application.

5. We don't have tutorials for small scale monitoring yet, but plan to work on tutorials in the future.

I can offer to setup a meeting to discuss dataset-specific questions. If you are interested, please send me an email.

Best, Andreas