Chunksize changes output result

[priit111]

I processed multiple S1 datasets with different chunk sizes, and the resulting GTC is very similar, but not identical.

chunks=1024
sarsen.terrain_correction(ds, dem, interp_method="linear", chunks=chunks,  output_urlpath="gtc_{}.tif".format(chunks))

chunks = 1024

chunks = 512

A-B image

The overall difference is very small(the image about about 0.03 max), however the issue is identical for all datasets I tested, that some parts of the image are not identical with different chunk sizes. The overall issue may be a very small floating rounding/shifting issue that is not an issue in practice, however I think it is best to document and potentially take a look at this.

[alexamici]

@priit111 you are right, the main difference should come from the azimuth time when a DEM point is observed, it is controlled by the zero_doppler_distance argument, that unfortunately cannot be controlled by the CLI.

The default value is 1m, that is: once the accuracy of the the azmuth time for the zero Dopper estimation is below 1m for all points in the chunk the geocoding is considered good enough. You can lower the zero_doppler_distance in the code to reach the precision you need (e.g. 1m is not good enough for interferometry for example).

I'll leave the issue open until I'll turn zero_doppler_distance into a command line option.