DEM assisted coregistration for GRD S1

Microwave Toolbox
#1

I tried to use the DEMAssistedCoregistraetion graph to coregister two GRD Sentinel-1 in ground range geometry and it works pretty well (seams to reach the same accuracy as S1-TOPS-coregistration for SLC and better than the automatic coregistration).
Why is this function “hidden” as internal graph? Could it be used as S1-TOPS-GRD-coregistration or I miss some important parts?

Thanks for any hints

1 Like
#2

We’re still working on it.

#3

That is great! I try the operator and it seems to work well (at least with SRTM DEM). Are there any limitation using this function with the current implementation? Will it be available within the next realise?

Thanks!

#4

With that implementation, I don’t think you could get better than the automatic for GRDs.

#5

I tried it out and the results are way much better than the automatic corregistration. The level of accuracy is comparable with the TOPS coregistration. The configuration was as follows:

GRDH image with precise orbits applied. SRTM bicubic interpolation for both DEM and image resemplig. I used SRTM 1sec. Area of application are Alps.

If it implements an algorithm that use the DEM and the (precise) orbits this I expect a better accuracy than an amplitude coregistration based on a first guess on the orbits.

May I miss something but to me it seems a good algorithm

#6

How do you measure the accuracy?

#7

Well there is not so much to measure: if I use automatic coregistration I obtain all NaN. If I use the stack tool with precise orbits I get ~5pixel shift between master and slave. If I use DEMAssistedCoregistraetion I obtain a subpixel accuracy (Visually it is impossible to see any shifts).

I used the following configuration:

GRDH image with precise orbits applied.
STACK: Initial offset method orbits for the stack. bicubic interpolation.
DEM assisted cor: SRTM 1sec for DEM. bicubic interpolation for both DEM and image resemplig

Area of application are Alp with the following images

S1A_IW_GRDH_1SDV_20151125T170628_20151125T170653_008764_00C7CC_3254
S1A_IW_GRDH_1SDV_20151207T170628_20151207T170653_008939_00CCCA_B69C

#8

The standard way to process would be to skip co-registration and directly terrain-correct, and then generate a stack from the terrain-corrected images. I tried this out with your images and they align perfectly.