SliceAssembly is missing data


The Sentinel-1 Slice Assembly tool seems to “forget” sometimes(!) polarisations (VV,VH) of individual slices. When running the graph S1_Assembly.xml (2.7 KB)

I expect for each polarisation (VV, VH) of each swath (IW1, IW2, IW3) an assembled longer swath. However, very frequently (but not reproducible even with exactly the same input data) in one or two swaths the upper slice is missing. For example, when running it on the two scenes


in the first try both, IW2_VH and IW3_VV miss the upper slice (IW1_VV is fine).
in the second try, IW1_VV misses the upper slice (but IW2_VH and IW3_VV are fine).

To me, this seems very likely to be a bug (or an effect of insufficient resources? - 32 GB RAM). Side information: quite frequently, the processing stops with the error “GC overhead limit exceeded”.

Context: my general goal is to coregister S1A and S1B for change detection over an area which is covered by multiple swaths and two consecutive S1 slices (north, south). To avoid slice-assembly in a larger graph, I pre-process consecutive slices and use them as input in the larger graph.

A related post be here: SliceAssembly-TOPSAR-Split is missing data

Processing time is ~20 min on an Intel Xeon E3-1270-V2 3.5GHz, 4 Cores, 8 Logical Processors with 32 GB RAM.
It would be much faster to select only the required swaths and bursts. However, after selection (TOPSAR-Split) I cannot recombine multiple swaths into a single product, i.e. the following graph writes only one swath to the disk:
Using separate write operates at the end, works but creates two produces. Processing time is then only 1-2 minutes.

Edit: I could reproduce the above mentioned problem of missing bursts three times using the following graph and data: S1_Assembly-split.xml (4.1 KB)

Every time, IW2_VH misses the upper bursts:

The bursts in the upper slice exist in the source product, since I could (for these scenes) assemble the whole slice without missing bursts using the first shown graph (without TOPSAR-Split).