results from cultivated and woody cover run with pipeline showed discrepancy on tiles randomly from run to run, after investigations and experiments, confirmed that it's caused by numexpr package, the relevant issue was raised https://github.com/pydata/numexpr/issues/494
input bands missing due to unavailable inputs from even upper streams, hence allow the graceful exist.
Note:
I have put into a hot (means very hacky) fix for the numexpr package, which is remarked in the docker file. We must follow up on this issue and address in the future.
Addressed issues:
numexprpackage, the relevant issue was raised https://github.com/pydata/numexpr/issues/494Note: I have put into a hot (means very hacky) fix for the
numexprpackage, which is remarked in the docker file. We must follow up on this issue and address in the future.