Geolocation accuracy improvements

[priit111]

Evaluate improvements to geolocaiton accuracy.

The primary thing to improve is using an external orbit state vector file.

ENVISAT format already contains orbit file, however it is only 5 points and no extra padding after and before the data acquisition.

The file in ENVISAT lvl1 metadata is likely the most accurate orbit file available and should be used.

[priit111]

Tested improvements to geolocation accuracy with external orbit file.

Envisat format only contains 5 data points. S1 data format contains more points, likely need to interpolate external Envisat data to match how sarsen works with S1.

Internal metadata for slant range times and azimuth times seems good.

Red points - internal
Blue points - external orbit file

[priit111]

DOR_VOR_AX is the precise orbit file for Envisat and when downloading a Envisat file from the current ESA service, it is always included.

However there seems to be multiple processed DOR_VOR_AX files.

Snap uses files processed in 2012.

First test indicate 2012 vs 2020 DOR_VOR_AX produces minimal differences.

[priit111]

osv_interp.zip

This graph compares 4 interpolations orbit state vectors of a Envisat dataset, for each data row in the dataset a x position is interpolated and the results are compared.

1. Sarsen with internal orbit data
2. Sarsen with external orbit data
   3-4. Lagrange interpolation with internal & external data.

Lagrange orbit interpolation is used for demonstration, as snap uses this(https://github.com/senbox-org/s1tbx/blob/1daae60d572e3ad0ee98c0ce3538c61ad71d7fa1/s1tbx-commons/src/main/java/org/esa/s1tbx/commons/OrbitStateVectors.java#L65)

Envisat file format specifies only 5 points of data.

This is not good enough for the sarsen orbit interpolation method(https://github.com/bopen/sarsen/blob/main/sarsen/orbit.py#L63C32-L63C33 - polynomial fitting).

However adding more orbit state vector points from attains relative stability compare to external orbit file Lagrange orbit interpolation.

Further tests should be done and compared to the Envisat CFI Software interpolation results, however at the moment it seems the external orbit solution is good enough for sarsen.

[priit111]

During testing in the final phases of the ard4asar project in became apparent that there is a small geocoding shift in relation to SNAP.

I reviewed the code and literature and it seems the issue is “Bistatic” Azimuth Bias Correction, which is not needed for Sentinel1, however it is necessary for Envisat.

The problem and the fix is best described in "Guide to S-1 Geocoding" chapter 4.7.2 “Bistatic” Azimuth Bias Correction. Unfortunately we did not have the time to test this during the ARD4ASAR project.

I have added preliminary test code here