Reprojecting after C2RCC->OLCI changes analysis

snap
#1

Hi

I’ve got a graph process that does the following:

  • Reads a Sentinel 3 OLCI product

  • Import a Vector

  • Subsets to the bounding box of the Vector

  • Masks all bands to the Vector

  • Optical->Thematic Water Processing -> C2RCC Processors -> OLCI

  • at this point the Vector is no longer accessible, so it is imported again, then

  • Masks 1 specific band to the Vector

  • Writes a BEAM DIMAP file

This seems to work and writes the file, however upon inspection of the file, and after having run it manually in SNAP, the resulting RGB image after reprojecting to WGS-84 is different than the original. Furthermore, the C2RCC.OLCI process seems to produce an image that is somewhat skewed (hence the reprojection to WGS-84 to attempt to correct this).

Before Reprojection:

image

After Reprojection:

image

So there’s really 2 questions here:

  1. What happens to the projection after the C2RCC.OLCI processing? Why is it skewed (specifically, when overlaid on a map it is slightly narrower than the polygon, and slightly taller than the polygon).
  2. Why would the analysis results themselves change after reprojection?
#2

Reprojection always causes a distortion to the images.
You can choose to preserve the areas of pixels or the angles line cross. I suggest reading:
Map projection - Wikipedia

By reprojecting to WGS84 you can actually overlay the data on maps we are commonly used to. It should fit in its orientation and its location. You can also export your data as KMZ format and import it into Google Earth. Actually it should give a good match.
In the SNAP help there is a page which tries to explain the reprojection algorithm including orthorectification.

The results do actually not change after reprojection. But the algorithm tries to find the best matching pixel in the source for a pixel in the target product. And sometimes one pixel in the source is in the range of two target pixels.

#3

Thanks Marco. I’ll look into that. I think part of the issue is the 300m resolution we’re restricted to with the S3 imagerey - particularly when overlaying it onto relatively small area polygons.