Possible ESD sensitivity scaling inconsistency in TOPS Stack

[WU-Pengzhan]

Hi everyone,

I’m processing Sentinel-1 data with ISCE2 2.6.5 and noticed a clear burst-boundary phase jump in the 2025-03-17 / 2025-03-29 interferogram, even after ESD correction. Neither acquisition is the stack reference.

While investigating, I noticed that overlapSpectralSeparation() in

overlap_withDEM.py

adds topStart/botStart before applying the azimuth offsets. My understanding is that these offsets already map overlap-local coordinates to full source-burst coordinates. Could the overlap origin be counted twice here?

I’m also unsure why the frequency separations from both acquisitions are summed, since the downstream estimate represents their relative azimuth displacement. Am I missing a convention or compensation elsewhere?

For this pair, the stored sensitivity was about 211.5 rad/pixel, but a small offset perturbation using the official resampler gave about 53.0 rad/pixel. After adjusting the coordinate/sensitivity calculation and reprocessing, the circular-mean phase difference in the examined overlap dropped from −1.080 to −0.022 rad. I kept the inputs, ESD looks (5 azimuth × 15 range), and threshold (0.85) unchanged.

This is only a local test, so I’d appreciate your thoughts on whether I’m interpreting the coordinates and scaling correctly. I can share more details and a before/after comparison if helpful.

Here are the specific lines I’m referring to:

• [Overlap coordinates and offsets](

  isce2/contrib/stack/topsStack/overlap_withDEM.py

  Lines 85 to 98 in b7702cf

  	y = np.arange(topStart, topStart+overlapLen)[:,None] * np.ones((overlapLen, topBurstIfg.numberOfSamples))
  	x = np.ones((overlapLen, topBurstIfg.numberOfSamples)) * np.arange(topBurstIfg.numberOfSamples)[None,:]
  	if os.path.exists(azMasTop) and os.path.exists(rgMasTop):
  	yy = np.memmap( azMasTop, dtype=np.float32, mode='r',
  	shape=(topBurstIfg.numberOfLines, topBurstIfg.numberOfSamples))
  	xx = np.memmap( rgMasTop, dtype=np.float32, mode='r',
  	shape=(topBurstIfg.numberOfLines, topBurstIfg.numberOfSamples))
  	else:
  	yy = 0.0
  	xx = 0.0
  	azi = y + yy

  ): y already includes topStart, and the azimuth offset is then added.
• [How geo2rdr defines the offsets](

  isce2/components/zerodop/geo2rdr/src/geo2rdr.f90

  Lines 374 to 375 in b7702cf

  	rgoff(pixel) = ((rngpix - rngstart)/dmrg) - 1.0d0*(pixel-1)
  	azoff(pixel) = ((tline - tstart)/dtaz) - 1.0d0*(line-1)

  ): this seems to imply that the source coordinate should be local_y + offset, without adding the crop origin again.
• [Combined frequency separation](

  isce2/contrib/stack/topsStack/overlap_withDEM.py

  Line 205 in b7702cf

  frequencySeparation = -Ktm2*tm2 + Ktm1*tm1 + Kts1*ts1 - Kts2*ts2 + fm2 - fm1 + fs1 -fs2

  ): this is the expression whose scaling I’m unsure about.

Thanks!

[WU-Pengzhan]

For reference, these were my original processing settings:

• ISCE2 2.6.5, TOPS Stack with NESD coregistration.
• Sentinel-1A, ascending track 171, IW1/IW2, VV.
• Stack reference: 2025-01-04.
• ESD coherence threshold: 0.85.
• ESD multilooking: 5 azimuth × 15 range looks (script defaults, not overridden).
• ESD network connections: 3, giving 87 pair observations across 31 acquisitions.
• Final interferogram looks: 1 azimuth × 1 range.