I am attempting to geocode several TanDEM-X acquisitions but experience an offset of tens of meters that differ for each acquisition. I have tried ellipsoid correction with geolocation grid as well as range-doppler terrain correction and export to both geographic WGS84 and projected UTM grid. All have the same result.
I am not experiencing this with Sentinel-1 imagery. If anyone has an idea for what may be off or what to try, it would be very appreciated.
some questions to narrow it down:
- where is your data located?
- which DEM is used for terrain correaction?
- what is the product level and acquisition mode of your data?
- are all images acquired from the same track (relative orbit)?
Ellipsoid Correction is only applicable over sea surfaces, because nearly all land areas suffer from foreshortening and layover.
Thank you for your quick reply. The imagery is acquired over coastal Alaska near Utqiagvik. So the scene contains a coastline with terrain at a few tens of meters. This far north I resorted to ASTER for the DEM, but it looks to be very small changes between TC and EC due to the low elevation.
I am using CoSSC. I tried to import the coregistered product as TanDEM-X as well as the TerraSAR-X acquisition as TerraSAR. This does not make a difference for the offset.
Yes, relative orbit = 16 for all images.
I notice that year 2017 is referenced correctly, but not the other years which features a variable offset. However, if I process let’s say two images in 2020, the wrong offset is identical. However, the offsets are not identical between years.
It all seems very strange and inconsistent. Let me know if you have some ideas.
thank you. The fact that the same track results in different offsets really makes me wonder.
Would it make sense to co-register all produdcts against one product which has been correctly geocoded and perform the terrain correction afterwards?
Yes, I am confused as well why this happens. But you idea is good as your suggestion seems to work great. Just lucky that I had one correct one to co-register to. Thanks a lot for the help!
I’m not very familiar CoSSC, would this document explain what is causing the difference? Section 6.2 talks about co-registration issues.
In CoSSC - both images are already co-registered. If you are looking to geocode, you want the metadata corresponding to the active sensor (Tx) for geocoding either image. The passive image (Rx only) in its original geometry is not distributed as part of the CoSSC product as far as I know. The metadata corresponding to passive image is useful to generate the DEM from CoSSC and is not representative of the Range Doppler grid of the coregistered product.
Marcus, thank you for providing this documentation, which is quite insightful.
Piyushrpt, this is great to learn now as I have been puzzled and held back by this for quite a while. Andreas provided a great workaround, but this essentially explains my observations and the best approach for doing this.
I consistently used TSX for all my imagery. However, TDX was predominately the primary satellite in these products. However, TSX was the primary satellite in the 2017 acquisitions, which explains why this year worked out good. When I now geocode the TDX imagery instead in the cases the TSX imagery didn’t work, this works flawlessly.
Thank you all so much for your helpful replies and for getting to the bottom of this!