TerrainCorrection step in bath mode - problem

Microwave Toolbox
#1

Hi all,

I have a question about batch processing of S1 data. I’ve created the graph in GUI mode of S1 tbx v.1.1.1. (I’m using s1tbx v.1.1.1 because snap 3.0/4.0 have problem with RAM memory optimization. There is no way to process interferogram of whole scene (3IW) in SNAP software. It results with “java heap space” error. Sentinel-1 Toolbox v1.1.1 can compute this interferogram without any problem. Used graph for whole scene: read->split->applyOrbitFile->backGeocoding->interferogram->merge->filtering->topoPhaseRemoval-> TerrainCorrection->write). The problem occurs when i try to use this graph in batch mode of s1tbx. Computation become looped and after couple hours when I kill the process I can see 1 burst of the result (about couple % of whole interferogam). And it is independed of computation time. It is allways the same result. I’ve tried to use other images and problem still occur. In the other hand the same processing in GUI mode is complete in about 2 hours without any problem. So there is my question: Is there some problem with TerrainCorrection step in batch mode? I think the problem is in selection of source bands of this step. Is there any chance to repair this issue? And what with RAM optimisation problem? For s1tbx v.1.1.1 11GB RAM memory is enought to process interferogram computation from whole 2 scenes and SNAP 4.0 need at least 28GB RAM for basic interferogram compurtation (without filtering, topoPhaseRemoval and terrainCorrection steps).

#2

SNAP 4 seem to work ok with 16 GB without using file cache option.
Have you tried the file cache option in tools options s1tbx?

#3

thx for relply @YiPing, file cache is not a problem in a snap 4.0. The question is why the same graph can be computed with 11 GB of RAM on Sentinel-1 toolbox v1.1.1. and can not with SNAP.

#4

Have you regenerated the graph in 4.0? Running old graphs on new versions of SNAP can cause unexpected problems…

#5

yes, I made two separate graphs, one on s1tbx v1.1.1 and second in SNAP. Do you know anything about TerrainCorrection step in batch mode?

#6

Perhaps @lveci could help?

#7

The batch processing is intended for a list of single input products producing a single output product per graph.
You can’t do coregistration which needs at least two products.
For this, you would be better off writing a script and using gpt to do many insar chains.

Single or Multi-temporal speckle filter?
#8

@lveci why not? There is no problems with corregistration of 2 S-1 images. The problem is only with TerrainCorrecion step. Interferogram computation (without terrain correction) in batch mode will succeed.