Height underestimation InSAR with S-1

Microwave Toolbox Interferometry
#1

Hi! I’ve performed interferometric formation for DSM generation across two different locations containing boreal forests (Vindelfjäll nature reserve, Sweden & Malla strict nature reserve, Finland). I’ve used two image pairs with a tbaseline of 6 days and a pbaseline >150 taken during dry days and gotten quite good results. The only problem I have is that the elevation retrieved across Malla strict nature reserve is showing some height underestimation (up to -40m) in levelled terrain and height overestimation in mountainous regions (very unexpected) while the one in Vindelfjäll nature reserve is showing an overestimation in the levelled terrain (+50m) and underestimation in the mountainous areas (expected). Both DEMs used are measuring from the ellipsoid so I know its not that, I’ve also checked if slope degree and aspect could explain these differences but I couldn’t find any correlation. Lastly I checked if local incidence angle could explain it but can only find a very low positive correlation of 0.2 (1 = strong positive correlation).

Does anyone have any clue what might be the problem? I’ve tried with several image pairs over this location and seem to get similar results every time.

#2

Forests are very difficult for InSAR approaches. How does the interfeogram look like?

#3

Since ive tried many acquisitions ive had a few which looked pretty bad and some that actually had a good and smooth result. But the thing that confuses me is that the areas with high coherence such as mountains with bare ground have the height overestimation when they should, according to previous research have an underestimation, and This is regardless of which DEM i use, which subswath, and so on.

#4

This always depends on the steepness of the terrain and the noise inside the fringes, because both can have drastic impacts on the unwrapping. Did you test different unwrapping techniques (MST, MCF; tile size, overlap…) already?

#5

I did the same for both location But i could give that a try and see of there is any improvement

1 Like
#6

I tried to change the statistical cost method and it actually generated worse results with a larger underestimation in the levelled forested terrain and a higher overestimation in the extreme terrain.