I am processing a dataset of ERS-2 images acquired after the gyroscope failure and I increased the threshold for DC in order to have enough images for APS processing. Are there any processing options in the software I should use to deal with these images besides estimating the azimuth position?
I strongly advise against using ERS-2 images after gyroscope failure. The reason is:
It is not as simple as you change the threshold for DC, the threshold are there for a reason: it is usually set between ±0.5*(doppler centroid frequency) to discard those images that might be aliased during resampling process in coresgistration process because their doppler centroid frequency goes out of the nyquist sampling frequency. The result is that if you use these images there will be serious alias occurs in your images (as we’ve already observed for some test area) and every step next wards will thus become meaningless.
There should a solution out of this by bringing all the images demodulated to zero doppler, but at current time sarproz does not support this specific action for ERS-2 data. Usually I would just try to use other data including Envisat and ALOS which do not have this kind of problem…
What Yuxiao says is partly not correct.
If the Doppler polynomial is precise, Sarproz does the job of resampling the image in the right way.
The problem is that those images highly probably do not have a reliable Doppler polynomial.
If this is the case, you will get lots of problems.
If on the contrary the Doppler polynomial is correct, you can get something out of those images.
The threshold on the DC is not for sampling purposes as Yuxiao says but it’s having to do with the interferometric decorrelation.
So, to perform classic InSAR (with distributed scatterers) you need low variability of DC.
But if you have corner reflectors (small targets, Persistent Scatterers), you can still process coherently the data and, as you say, there will be a phase term depending on the azimuth subcell position (check the papers of my phd thesis).
The problem here is having a precise and reliable Doppler polynomial.
if the precise orbits are totally missing, then you can use the coarse xcorr to roughly align the images (coregistration parameters window), then the usual coregistration will search for finer matches
however, even if you can coregister them, probably it will be difficult to carry out a good interferometric analysis.
If you have coherence, you will be able to remove possible residual fringes with the corresponding function.