fesom-jax CORE2 one-year setup: mesh, initial state, and JRA55-do forcing (1958)

Everything needed to run the fesom-jax ocean model on the CORE2 mesh (global, ~1°, 126,858 nodes, 48 levels) for one year. fesom-jax is a differentiable port of the FESOM2 ocean circulation model to JAX. Three archives: core2_mesh_ic.zip (~370 MB) — the mesh (as a dense .npy bundle and in standard FESOM2 text format), the cached PHC 3.0 initial temperature/salinity, the PHC source file, the sea-surface-salinity restoring field, river runoff, and the chlorophyll climatology. core2_partitions.zip (~180 MB) — the dist_N domain decompositions (dist_2 … dist_864). You need these to run on more than one device. Pick the one matching your device count; a single-device run needs none of them. core2_forcing_1958.zip (~10.4 GB) — the eight JRA55-do v1.4.0 surface fields for 1958 (uas, vas, tas, huss, rsds, rlds, prra, prsn), 3-hourly, on the native JRA55 grid. Usage: python scripts/fetch_data.py --dest ~/fesom-data --record <THIS RECORD ID>, then point the model at it with the printed environment variables. See docs/DATA.md in the repository. Each archive carries a README and a CHECKSUMS.sha256. Note on the CORE2 mesh version The CORE2 vertical level files (nlvls.out, elvls.out) were regenerated upstream on 2026-07-03. This package ships the earlier version, because that is what all fesom-jax results were produced with. The two differ at exactly 2 nodes and 4 elements, all in the Ross Sea (~154°W, 77°S), where the version here is deeper (580 m vs 280 m at the largest difference). Both are structurally valid. To match a current FESOM2 run instead, take the level files from the upstream mesh (gitlab.awi.de/fesom/core2) and rebuild the dense bundle with scripts/prepare_mesh.py. Please also cite the underlying model and datasets FESOM2 — Danilov et al. (2017), doi:10.5194/gmd-10-765-2017 CORE2 mesh — Wang et al. (2014), doi:10.5194/gmd-7-663-2014 JRA55-do forcing — Tsujino et al. (2018), doi:10.1016/j.ocemod.2018.07.002