Establish whether wind-speed input can overcome the univariate context barrier

Determine whether integrating even a single wind-speed channel into significant-wave-height forecasting lifts the approximately six-hour context barrier identified for univariate Hs prediction.

Background

The study deliberately uses only past significant wave height as input and interprets the resulting performance ceiling as an information deficit caused by the absence of atmospheric forcing. Wind, pressure, and fetch could provide information about the external forcing that drives changes in the wave field, particularly during storm conditions.

The authors specifically identify the effect of adding one wind-related covariate as unresolved. This is a concrete empirical question distinct from the paper’s completed comparison of univariate architectures.

References

It remains to be demonstrated whether the integration of even a single wind-speed channel lifts the 6-hour barrier identified here.

— On the Limits of Univariate Deep Learning for Significant Wave Height Forecasting  (2609.30688 - Zhai et al., 25 Sep 2026) in Section 4.3, “Limitations and generalisability”

The linear-inertial and two-component hypotheses are not mutually exclusive---both predict architecture independence and skill saturation---and the present experiment cannot distinguish between them. Disentangling these mechanisms would require explicit decomposition of $H_s$ into wind-sea and swell components, which is left to future work.

— On the Limits of Univariate Deep Learning for Significant Wave Height Forecasting  (2609.30688 - Zhai et al., 25 Sep 2026) in Section 4.1, “Why the five families are of no practical consequence”