Running this graph works nicely in Europe and Africa, but for all the extends I tried in South America and East Asia, it returns an image that is almost pure nodata: coh_2images_GeoTiff.xml
Is this a common issue?
Sometimes the spatial extent seems to change between the source bursts and the resulting coherence image. And the memory usage can pass 30Gb in those cases.
If needed, I can make a minimal example that runs in a docker container.
Tested with SNAP 11 and 12
This issue comes from the ClouDInSAR project, which brings interferometric capabiities in CDSE via openEO. @mengdahl
The source issue on the project GitHub is DEM Errors in Andes region · Issue #101 · cloudinsar/s1-workflows · GitHub
After my tests, I think it’s an issue related to OOM, since on my VM with 64GB of ram the results are valid.
Can you share some test results and images on the forum?
The wrong CRS was used. This commit fixed the issue: set AUTO UTM for coherence proj · cloudinsar/s1-workflows@9861acc · GitHub
@mengdahl starting from a graph computing the coherence, I opened it in SNAP GUI graph builder and checked the various steps.
I realized that it if I set the auto-UTM in the Terrain Correction step:
finally, I found the auto-UTM code in this topic: Auto UTM zone in the graph file? - #9 by dzatta
and replacing the previous in the graph made it to work fine:
<mapProjection>AUTO:42001</mapProjection>
The workflow can easily be reproduced following the instructions here (the docker is still based on SNAP 12):
docker build -t openeo_insar . -f OpenEO_Dockerfile
docker run -it -v ./:/src/ -e AWS_ACCESS_KEY_ID=$AWS_ACCESS_KEY_ID -e AWS_SECRET_ACCESS_KEY=$AWS_SECRET_ACCESS_KEY --rm openeo_insar /src/sar/sar_coherence.py /src/sar/example_inputs/input_dict_japan_new.json
Is the bug that auto UTM does not always work? Please someone check how the situation is in SNAP13.
Can you share some test results and images on the forum?
This issue is resolved in SNAP 13. Please update the software and install all the available updates.
Thanks for the feedback, we can’t update to SNAP 13 yet due to the subset issue mentioned here: SNAP13 : Subsetting before Terrain Correction fails or gives wrong results - #5 by na-ea