Forest degradation

Show Room
#1

Hello! I want to make a forest classification and I use radar images , Sentinel 1. I have just VV and VH band for the area. I computed the Radar Forest Degradation Index (RFDI) , VV/VH, mean, median, but I can not see significant differences between a degraded region or a region with dense and healthy forest. I have a real mask with several regions that are sure in which class they are, but the result (RFDI? VV/VH, etc.) in that region is not homogeneous. The mask represents some areas with different levels of degradation. Someone checked it physically, in the are of that forest. I don’t know how I would classify the regions in that forest if I can’t know/see the difference between dense and degraded regions. Do you have any idea how I could find a classification criterion between these regions? Thank you!

#2

there are many reasons why the RFDI doesn’t reflect the field reference:

  • the RFDI was initially designed for quad-pol data which is more senstitive to forest structure than VV/VH of Sentinel-1
  • the forest type or its degradation is not affecting the backscatter intensity (at C-band) in a way you would need it. For example, broad-leaved forests and deciduous forests have different temporal backscatter profiles over the year, so I would also expect them to respond different to degradation
  • the spatial resolution is not sufficient to detect the areas of different degradation levels

Maybe you can show what you have by now.

#3

I analyse the backscatter profile over one year. You’re right, there’s differents responses to degradation. I put here o photo with the mask . mask ( Depending on the color, there are different classes) here with the forest area

foret.PNG
foret.PNG905×626 260 KB
an here with rfdi rfdi. How could I classify the regions in this situation?

#4

would it make sense to mask forest areas in the image enhance the variation inside these areas?

did you calculate the RFDI for two different times and then compare the output?

If so, please make sure that both acquisitions are from the same track because Sentinel-1 backscatter can change quite much depending on the looking direction and angle.

1 Like
#5

I use the mask just to see if the rfdi is homogeneous in those regions. I will try to compute rfdi for 2 different times and then compare them. Thank you!