I’m hoping someone here has some information for me.
The SNAP help information has the following information on possible values in the Layover-Shadow mask generated in SAR Simulation:
However, when the mask is generated, pixel values are not only 0, 1, 2, or 3-- in fact, they range up to 13. Can anyone tell me what pixels should actually be understood as those in layover-shadow? If I take the area for my study region in layover-shadow based on the above pixel value designation, the area is about 700 square kilometers. But, if I take any pixel != 0 in the layover-shadow mask to mean it is in layover and/or shadow, the area is about 1800 square kilometers-- a significant difference.
Thanks in advance,
actually, there is no need to double-post your question. So I will delete your first post.
@cwong or @timmoorhouse can you explain this?
Thank you @marpet,
I reposted because I couldn’t seem to find my original post, and I thought perhaps I’d neglected to publish it.
I appreciate any insight you, @cwong, or @timmoorhouse can offer.
@marpet, curious if you ever found an explanation for this?
I’m sorry to here that you haven’t got an answer. Maybe @lveci can help here?
I have this same issue when generating the layover and shadow mask. Pixel values range up to 9 for one and even 15 for another image.
Has anyone of you been able to find the source and solve this?
Are you sure those values were not created due to interpolation? Nearest neighbour is a safe choice that leaves the original values untouched.
sounds to me like this is rather a bit mask, i.e. 0-3 refer to the bit positions and the result contains bit-combinationed values (and, in contrast to what the explanation seems to indicate, multiple of these categories are set to true). just a guess based on the value range observed…
At least some comment from the developers would be nice
Yes, I agree. @lveci can you have a look at it?
Hi all. Is there an update on this? The values in the mask prior to
Terrain-Correction are fine. So I guess it is a problem of interpolation as @mengdahl pointed out. Although I believe nearest neighbour would not be the best choice for discrete classes. I would prefer mode resampling.
It would be great if one could provide multiple interpolation modes to
Terrain-Correction. Not just one for images and one for the DEM, but different ones for individual bands. Something like this:
@lveci do you think this could be done?