revised RESTART options; fix string matching for EASE tile file

[weiyuan-jiang]

Revised and cleaned up GEOSldas RESTART options:

• Clarified scope and constraints of RESTART=1 and RESTART=2.
• For RESTART=1, the tile file and bcs files are now pulled from the GEOSldas experiment in RESTART_PATH, with a basic check for bcs consistency between the restart experiment and the new experiment.
• Introduced RESTART=3 for restarting from restarts produced by the GCM.
• Cleaned up stray code from RESTART=G, which had been removed some time ago.

Fixed string matching for EASE tile file to accommodate "EASE*-Pfafstetter" tile file that is added into bcs directory for the benefit of runoff routing (GEOS-ESM/GEOSgcm_GridComp#1143)

Testing:

• Successfully zero-diff tested by @gmao-rreichle on 18 March 2026; zero-diff result required replacing inconsistent bcs files (greenness, NIRDF, VISDF) in the RESTART_PATH in Matt's test setup. These inconsistent bcs files were not used in the old version of the code but are used after the changes implemented in the PR.
• RESTART='2' and RESTART='3' are not used in nightly zero-diff tests and were tested separately by @biljanaorescanin on 19 March 2026 revised RESTART options; fix string matching for EASE tile file  #160 (comment)

[gmao-rreichle]

@weiyuan-jiang : Thanks for adding this bug fix. One question: Do we also need the same protection here (for 'RESTART'==2)?

GEOSldas_GridComp/GEOSldas_App/ldas.py

Line 376 in 68d08a1

in_tilefiles_ = glob.glob(inpdir+'/*.nc4')

It's a bit unclear to me why we handle the TILING_FILE for 'RESTART'==1 in a different place in the code from where we handle the TILING_FILE for 'RESTART'==2. In both cases, we have restart files (and thus a tiling file) from an existing GEOSldas experiment. Shouldn't the logic for obtaining the TILING_FILE (from the ./rc_out directory) be the same (or nearly the same) in both cases? (For 'RESTART'==2, I think we perface the tiling file name with 'MAPL_', but that could be done with an if-condition that applies to just the file name within the larger logic of assembling the path and deciding if we use the txt or nc4 file.

Also, I wonder if the code involved in the bug fix here also is related to the bug that Devon reported. It may make sense to fix both bugs in this PR (if Devon's problem is in fact a bug in GEOSldas setup).

[weiyuan-jiang]

@weiyuan-jiang : Thanks for adding this bug fix. One question: Do we also need the same protection here (for 'RESTART'==2)?

GEOSldas_GridComp/GEOSldas_App/ldas.py

Line 376 in 68d08a1

in_tilefiles_ = glob.glob(inpdir+'/*.nc4')

It's a bit unclear to me why we handle the TILING_FILE for 'RESTART'==1 in a different place in the code from where we handle the TILING_FILE for 'RESTART'==2. In both cases, we have restart files (and thus a tiling file) from an existing GEOSldas experiment. Shouldn't the logic for obtaining the TILING_FILE (from the ./rc_out directory) be the same (or nearly the same) in both cases? (For 'RESTART'==2, I think we perface the tiling file name with 'MAPL_', but that could be done with an if-condition that applies to just the file name within the larger logic of assembling the path and deciding if we use the txt or nc4 file.

Also, I wonder if the code involved in the bug fix here also is related to the bug that Devon reported. It may make sense to fix both bugs in this PR (if Devon's problem is in fact a bug in GEOSldas setup).

This is not a bug for Devon because she did not setup run on EASE grid. I will make changes on the route PR when we need the tile2pfaf file in the rc_out directory. But I agree it is quite messy here. I will do some cleanup in the other PR

[gmao-rreichle]

This is not a bug for Devon because she did not setup run on EASE grid. I will make changes on the route PR when we need the tile2pfaf file in the rc_out directory. But I agree it is quite messy here. I will do some cleanup in the other PR

My comment wasn't meant to be specific to the EASE grid. I think the code in question executes independent of the grid/tile space. Is there a reason for the separation of what seems to be (nearly) the same action into a block for RESTART==1 and another block elsewhere for RESTART==2?

Re. Devon, I was just wondering if the TILING file is processed correctly. The bug fix here touches the lines that are dealing with the nc4 tiling file, and we suspected that Devon's problem may have to do with a mismatched TILING file.

In any case, the bug fix that the PR intends to fix is for EASE and touches LDAS setup. My suggestion was to add the bug fix for Devon (if there is indeed a bug) into this PR -- assuming the fix also touches LDAS setup. (We can rename the PR to something like "LDAS setup bug fixes...")

[weiyuan-jiang]

In the first Restart == 1, it actually tries to figure out where are the BC files for the outputs. It there is no 'domain', all BC files are from BC. In the second Restart == 2, it gets input tile file for remapping later on

[gmao-rreichle]

@weiyuan-jiang : I added a couple of commits and the following comments. Please take a look and let me know if you see anything wrong or have any questions, thanks

[gmao-rreichle]

@weiyuan-jiang : See comments below. After we address them, the PR would be ready for testing by @biljanaorescanin

[biljanaorescanin]

Testing summary:

1. Nightly tests pass.
2. Option RESTART2 works.
3. Option RESTART3 works
4. Option M doesn't work because M2 is not on archive it changed path.