Loading
Life sciences · Preprint
arXiv · September 4, 2026
Early or partial results. Treat as a signal, not a conclusion.
This preprint reports that prescribing cyclone-track channels to a U-Net ocean emulator for Bay of Bengal forecasting degrades skill compared to an ocean-only baseline, because the cyclone signal is out-of-distribution during training (active on only 7.9% of days). The error is largest within the prescribed storm footprint, and replacing the real cyclone map with a no-storm map improves held-out forecast skill by 7.5–16.4%.
Computational comparison study; controlled ablation across three random seeds. Withheld cyclones from GLORYS12 reanalysis in the Bay of Bengal, spanning wind speeds 65 to 150 kt.. Intervention: U-Net neural network with prescribed cyclone-track channels (four channels). Compared with: Identical U-Net without prescribed cyclone-track channels (ocean-only model); also compared to persistence baseline. Bay of Bengal.
Storm-conditioned U-Net lost to persistence in every run across three seeds, with disjoint skill ranges (p = 3.1e-5, paired across storms) Ocean-only U-Net beat persistence in every run across three seeds Cyclone-track channels were non-zero on only 7.9% of training days
Safety was not reported in the material analysed. Check the source before drawing any conclusion about harm.
The source did not state who this applies to in practice.
An unreviewed computational study comparing two neural network architectures on a single regional dataset, demonstrating a methodological pitfall in ocean emulator design but lacking validation across independent datasets or peer review.
As stated by the source record.
Quoted from the source exactly as published.
Graded across the dimensions that decide whether you should act, each from what the source actually supports. There is no single score, and where a dimension was not assessed it says so.
Neural ocean emulators are being proposed for regional forecasting in cyclone-exposed coastal seas, and a natural design choice is to hand the network the cyclone as a prescribed input. We test that choice in the Bay of Bengal and find it harmful. We withhold 15 whole cyclones spanning 65 to 150 kt from GLORYS12 reanalysis and compare two U-Nets that are identical except for four prescribed cyclone-track channels. Across three seeds the ocean-only model beats persistence in every run and the storm-conditioned model loses to it in every run, with the two skill ranges disjoint (p = 3.1e-5, paired across storms). The cause is exposure frequency rather than signal content: the channels are non-zero on only 7.9% of training days, so they are out of distribution the moment they activate. The extra error falls inside the prescribed storm footprint, and replacing the real cyclone map with a no-storm map at inference improves held-out storm forecasts by 7.5 to 16.4% in every seed. The conditioned network has learned a response to a rare signal that is confidently wrong.
Taken from the source record, never inferred. Follow any of these and new work involving them reaches your briefing.