Valid pixels in water areas

[wenzong95]

Hi Scott,

I've obtained some results for a section of the Melbourne area using the CSLC data processed with Dolphin. However, I have a few questions regarding the derived images in the "timeseries" folder.

I noticed that, in the deformation images from this folder, there are valid pixels appearing in the ocean area. The same situation is present in the velocity.tif file.

My questions are:

1. Why are there valid pixels in the ocean area?
2. How is velocity.tif generated? Specifically, what logic is used to mask out pixels in this image?
3. How can I generate velocity and deformation images with pixels of low temporal coherence masked out?

The image below shows an example deformation image from the "timeseries" folder, opened in SNAP. The left-up area as marked is the ocean area.

I'm also confused as to why some pixels are masked out when viewed in SNAP, but all pixels are displayed when using QGIS.

Thank you for your help.

[scottstanie]

Hi @wenzong95 ,

Why are there valid pixels in the ocean area?

Just to make sure I understand - You create/specify a water mask during processing and specify it as the mask_file, and you were hoping to have the outputs to show nodata in the areas with 0s?

So far I have been slow to add processing steps which "zero-out" or set portions of the outputs to nan since it's not reversible.
But this has been asked about/requested before (#540 (comment)), so I'm thinking about a few options:

1. Finish this PR: [New Feature]: Add an option to create + apply a quality mask after displacement #593 which could be a separate "post-process masking" step that would allow users to specify not just one water mask, but add thresholds on the temporal coherence, on the phase similarity, etc.
2. Add a configuration option like zero_masked_pixels, which sets the pixels you requested to be ignored to zero in the final timeseries folder.

Let me know if you have other ideas that make sense.
If you'd like to fix the results now before the features are in, I'd do something like I suggested here: #540 (comment)

[wenzong95]

Hi Scott,

Thank you for your response.

Previously, I used SNAP to open the Dolphin-derived results. I noticed that SNAP always displays a nodata_band_1 mask (as shown in the figure below). I initially assumed this mask was an inherent layer in the .tif file and was generated by Dolphin. That’s why I raised my earlier questions.

From what I understand now, Dolphin does not apply any masking to the results in the “timeseries” folder. Is that correct?
Do you think it would be a good idea to mask out low-coherence pixels using the temporal_coherence_average_***.tif file as a mask?

Thank you very much for your help.

[scottstanie]

Oh and

I'm also confused as to why some pixels are masked out when viewed in SNAP, but all pixels are displayed when using QGIS.

The QGIS sounds right. I have not used SNAP before... is there somewhere that describes how it displays pixels? That looks rather strange to me haha.