Recently found a number of scenes with strange non-linear deformations. Such scenes a lot. The difference in the angle of shooting is not more than 4-6 degrees. At a resolution of 10 meters (channel 8) there are strong differences in the binding. The shadow shows that the sun is shining from a different angle. But even under these conditions, unlikely to occur by such shifts, such as roads. What caused it?
The scene took raw to JPEG2000. Internal binding is the same (checked using gdalinfo).
The differences found in the scenes, beginning with the month of December.
Can anyone give the exact answer?
In your Datastrip metadata, there is a field that denotes EPHEMERIS_QUALITY> I wonder what the values in the respective products are?
Also, if there is a FAILED test in the OLQC, it will be hlighted in the “Quality_Control_Checks”/ “Quality_Inspections” fields in the MTD_MSIL1C. The tests are:
SENSOR_QUALITY_FLAG
GEOMETRIC_QUALITY_FLAG
GENERAL_QUALITY_FLAG
FORMAT_CORRECTNESS_FLAG
RADIOMETRIC_QUALITY_FLAG
These flags are either PASSED or FAILED
If a test is FAILED, a report bearing the relevant title will be generated in the DS QI_DATA subfolder.
OK. And what are the values of the EPHEMERIS QUALITY in the DS for these Tiles? Can you find those and put them here, linking the Compact Filename with their EPHEMERIS_QUALITY values, please?
Thanks. That all looks good - the values are all as I would expect them to be. I think we can therefore rule out issues in data quality, so it’s likely to do with variations in orbital dynamics between scenes.
I’ll have to think on it, and do some learning, too
Regarding the first scene: I am not sure we can see a non-linear distortion here. It looks like a relatively standard shift that we can observe between two acquisitions. The typical shift between two successive acquisitions is 1.2 pixels at 95% confidence.
The second scene is very nice and interesting: the images are very well aligned but the bridge position changes. This is most probably a parallax effect coming from the fact that the bridge is not at ground level. Sentinel images are rectified according to a Digital Elevation Model (DEM) sampled at 90 m horizontal resolution. Fine details like the bridge cannot be represented, so the bridge will be projected to the water surface, at a position which depends from the viewing angle. If you have some spare time you could probably compute the altitude of the bridge with these two images.
The third one looks like an open-sky mining site. The apparent shift around the holes could be due to a combination of a) shadowing effects b) on-going mining activity c) inaccuracy of the DEM in this evolving area.
Generally speaking, we can find some local distortions when comparing different repeat orbits in places where the DEM is not accurate. Improvement of the DEM accuracy is in the to-do list of the Sentinel-2 team, but this will require some time to implement.
