Up to Transformations and interpolations
Hello Oasis team, I'm currently reworking the OpenIFS-FESOM2 (AWI-CM3) coupling interface. One of a number of proposed changes is to switch the atm->oce flux coupling from GAUSWGT + GSSPOS conservative postprocessing to SCRIPR CONSERV FRACNNEI + GSSPOS. The other is to calculate the SCRIPR CONSERV FRACNNEI weights in the direction oce -> atm and to use them as basis for stochastic selection of coupled points. I have a implemented a prototype stochastic coupling in MCT that works with GAUSWGT weights. In also introduced the oasis_write_corner routine in FESOM2 to write the cla and clo fields into grids.nc. I have two issues with my conservative rmp_ files. Firstly, they take a long time to compute. Between our low res 126k oce and 40k atm surface point grids I can compute the GAUSWGT serially in ~6min. For CONSERV weights I used the offline tool with 40 MPI * 48 OMP and it took ~4 min. Assuming the typically linear scaling of the parallel SCRIPR implementation, it would have taken ~7680 min = 5.3 days to calculate these serially. Is this normal or an indication of something going wrong? I tried 1, 40 and 500 bins, and found it was expensive regardless, and in contrast to [Piacentini 2018]. If I want to scale this up the high resolution, it would be slow even with the parallel scripr. Secondly: The offline remapping tool does a unit test and writes the result in error_interp.nc. The maximum error is 3515433.4 %. Obviously using this remapping weight results in model blowup (I tested it and indeed got a model blowup). I assume I did something wrong with the corner lat/lon? I followed your advice from a few years ago, and always use the maximum number of corners on our unstructured mesh, repeating the last one a few times where the actual number of corners is smaller than the maximum one. I have uploaded the namcouples, the grids, areas, masks, and the computed remapping weight files on the dkrz swift server: https://swiftbrowser.dkrz.de/public/dkrz_b94ae58d-50e8-4ad6-b453-afff299daabf/oasis_remappings/ Cheers, Jan [Piacentini 2018]: https://oasis.cerfacs.fr/wp-content/uploads/sites/114/2021/08/GLOBC_WN_Piacentini_Parallel_SCRIP_cmgc_18_34_2018.pdf
