Why does the pre-event image is so dark with such a huge quantity of pixels with low values of backscattering? I am supposing this image is not rappresenting any flooding event or maybe this consideration is wrong?
Two images have same relative orbit, both ascending. I’d use them for change detectiong too but I got back really weird behaviors.
Have you used Sentinel-1 products from the same orbit? Maybe it depends on difference in acqusition geometry?
I noticed at your second histogram (September 21st 2019) that Max value is quite different then September 3rd 2019. In my opinion 5 dB is huge difference.
I was exploring a sentinel-2 image: actually dark areas correspond to bare soil.
I was thinking: bare soil appear dark when it is very dry and with higher values of backscattering when it is wet. This is because of differences of dielettric costant, is it right?
Is this behavior affecting the output of log ratio for flooding identification? I mean , these area could be classified as false positive
Yes that is correct. Also if the soil is rather smooth the return will be darker.
BTW you are not very limited in your pre-event image selection so you could for example download 3-5 scenes and average them for a lower-speckle high-resolution reference scene.