Array NEST

What are the residual values? Try using a much larger window size for the cross correlation in the GCP selection.
Also because these are detected images, try with and without resampling in the create stack step. (resampling here is not recommended for complex products)

Dear Marcus,

thanks a lot for this valuable comment! Now it seems more clear to me why I do not have any chance to co-register these two (ASAR and TSX) images no matter what I try to tell NEST.

However, since these scenes are from the Antarctic I wonder whether a "good enough" DEM does exist at all.

What other possibility do I have with NEST to accurately overlay these two images ... or also ALOS PALSAR and ASAR imagery. I mean, we do not live in a 1-frequency world, right?

Kind regards and thanks a lot again for your helpful comment.

Stefan Kern

Hi,

You are trying to coregister TSX and ASAR image of the same area, and acquired in the same orbit (ascending or descending)?

Quickly few notes and pointers you might find useful:

---
DEM over Antarctica:

you can try this one, The Radarsat-1 DEM (RAMP), http://nsidc.org/data/nsidc-0082.html . Btw: be aware that the projection of this DEM is polar polar stereographic, since neither lat/lon nor UTM makes sense there. I attached the subset of DEM to this thread.

Coregistration:
---
as long as you do not want to interferometrically combine coregistered images from different sensors/modes you should be able to coregister and do something with coregistered data. Still it depends on the application, and what exactly you want to do with the data after the coregistration.

Approach Marcus indicated is a straightforward and simplest way of getting all your images in the same reference geometry (UTM or whatever). Note that you also have to have good orbits here, and that the timing and GCP annotations would influence the overall accuracy of the end result.

Correlation optimization approach: this one could be tricky. If you try to coregister TSX vs ASAR (directly with coregistration implementation in NEST) it will less likely give optimal results. Since you are coregistering apples and oranges. Completely different resolution, scattering mechanisms, etc.... Targets that could be associated to one object in TSX image, are having completely different radiometric characteristics then in ASAR image, etc.... Perhaps you can work out something using features, like roads, but, less likely.

Moreover(!), eg 32x32 window in TSX covers X times smaller area on the ground then in ASAR. etc, etc, etc....

What you can do, and I think it should work: to either multilook TSX image to the sampling of ASAR, or oversample TSX image to resolution of ASAR and try the coregistration then. You mentioned sub-sampling, but it is not clear you tried this.

You can perhaps also geocode TSX, and ASAR, and then coregister (with correlation optimization) ASAR.geocoded to TSX.geocoded. Since TSX has much better accuracy of all the annotations.

Still, approach and success in doing the coregistration of mixed data very much depends on the application.

Hope this somehow helps.

Good luck,
Petar

EDIT: forgot the attachment.

Attachments:

RAMP_dem_subset.png (253.5k)

Dear Petar,

I guess I've found the problem: The tie point grids (latitude, longitude, incidence angle, etc.) which are provided with the TSX data have the same size as the TSX image, yes. And these should be the latitude and longitude of each TSX-grid cell ... but obviously NEST treats these grids as if these coordinates are the coordinates of the entire shown areas, i.e. the TSX image plus the NaN values around.

In other words, if I plot the TSX-image in NEST it shows up exactly as it should as can be seen in the preview image, rotated towards the vertical with areas covered by NaNs in the 4 corners of the shown rectangle area.

If I plot the latitude, however, I get a full coverage, no NaN values; same is true with longitudes.

If I overlay the TSX-image, the latitude image and the longitude image with a graticule overlay, then this fits with the latitude and longitude image but not with the TSX-image because the latter has a different spatial coverage ... how nice ... so I have to convince NEST now that the given lat / lon values are only for the area with "good" TSX-image values ... let's see....

Kind regards,

Stefan Kern