Masked reference point

[scottstanie]

Discussed in #670

^{Originally posted by giorgoskouroupis January 8, 2026}
Hello @scottstanie and the whole team :)

I have been running the dolphin and so far it seems quite robust.
thank you for your work. :)

However i came to a small bug(?), i guess from using a (water) mask file.
After an unsuccessful inversion run (all good up until the inversion step) the log reads:

[INFO|timeseries|L153] 2026-01-07T09:38:41+0100: Selecting a reference point for unwrapped interferograms
INFO:dolphin:Selecting a reference point for unwrapped interferograms
[INFO|timeseries|L1313] 2026-01-07T09:38:41+0100: Selecting reference point
INFO:dolphin:Selecting reference point
[INFO|timeseries|L1395] 2026-01-07T09:38:41+0100: Creating intersection of connected components
INFO:dolphin:Creating intersection of connected components

100%|██████████| 156/156 [00:29<00:00, 5.37it/s]

[INFO|timeseries|L1415] 2026-01-07T09:39:11+0100: Found 36530 connected components in intersection
INFO:dolphin:Found 36530 connected components in intersection
[INFO|timeseries|L1363] 2026-01-07T09:39:11+0100: Saving ReferencePoint(row=1456, col=122) to ./reference_point.txt
INFO:dolphin:Saving ReferencePoint(row=1456, col=122) to ./reference_point.txt
[INFO|stitching|L694] 2026-01-07T09:39:11+0100: Using ./water_mask_tile_0.tif, already matches size of ./unwrapped/20200105_20200117.unw.tif
INFO:dolphin:Using ./water_mask_tile_0.tif, already matches size of ./unwrapped/20200105_20200117.unw.tif
[INFO|timeseries|L207] 2026-01-07T09:39:11+0100: Inverting network of 600 unwrapped ifgs
INFO:dolphin:Inverting network of 600 unwrapped ifgs
[INFO|timeseries|L1060] 2026-01-07T09:39:11+0100: Inverting network using L1-norm minimization
INFO:dolphin:Inverting network using L1-norm minimization
[INFO|timeseries|L1095] 2026-01-07T09:39:11+0100: Using unweighted unw inversion
INFO:dolphin:Using unweighted unw inversion
[WARNING|timeseries|L1101] 2026-01-07T09:39:12+0100: Masked data found at 1456, 122.
WARNING:dolphin:Masked data found at 1456, 122.
[WARNING|timeseries|L1102] 2026-01-07T09:39:12+0100: Zeroing out reference pixel. Results may be wrong.
WARNING:dolphin:Zeroing out reference pixel. Results may be wrong.

From the look of the code, the reference point is calculated before any consideration of the the mask in

dolphin/src/dolphin/timeseries.py

Lines 152 to 176 in 4daa481

	if reference_point is None:
	logger.info("Selecting a reference point for unwrapped interferograms")
	if quality_file is None:
	raise ValueError("Must provide quality_file if not reference_point given")
	ref_point = select_reference_point(
	quality_file=quality_file,
	output_dir=output_path,
	candidate_threshold=reference_candidate_threshold,
	ccl_file_list=conncomp_paths,
	)
	else:
	ref_point = ReferencePoint(row=reference_point[0], col=reference_point[1])
	ifg_date_pairs = [get_dates(f, fmt=file_date_fmt) for f in unwrapped_paths]
	sar_dates = sorted(set(flatten(ifg_date_pairs)))
	# Read in the binary mask
	if mask_path is not None:
	matching_mask_path = _get_matching_raster(
	mask_path, output_path, unwrapped_paths[0]
	)
	mask = io.load_gdal(matching_mask_path, masked=True).filled(0)
	bad_pixel_mask = mask == 0
	else:
	bad_pixel_mask = np.zeros((rows, cols), dtype=bool)

If this is indeed a small bug, a naive solution would be to call the mask into the "def select_reference_point(...)"

dolphin/src/dolphin/timeseries.py

Lines 1313 to 1314 in 4daa481

	logger.info("Selecting reference point")
	quality_file_values = io.load_gdal(quality_file, masked=True)

and multiply the quality_file_values with the mask to remove masked areas, but i am not sure if there would be a more appropriate solution.

Any ideas, or thoughts?

Many thanks for your time :)