Test whether the conditioning effect generalizes beyond GLORYS12

Investigate whether the observed degradation caused by rare prescribed cyclone-track conditioning is a property of the GLORYS12 reanalysis or persists when the experiment is repeated using the ORAS5 reanalysis.

Background

The study evaluates ocean-emulator behavior using GLORYS12V1 reanalysis data from a single basin, architecture, and reanalysis product. Its principal finding is that prescribed cyclone-track channels, active on only a small fraction of training days, can cause a storm-conditioned U-Net to underperform an ocean-only model and persistence on held-out Bay of Bengal cyclones.

The authors explicitly limit the generality of this result because the experiment does not establish whether the degradation is specific to GLORYS12 or reflects a broader property of rare conditioning signals in ocean emulation. Repeating the analysis with ORAS5 would provide a direct test of that unresolved question.