ON THE USE OF DYNAMIC MODE DECOMPOSITION FOR TIME-SERIES FORECASTING OF SHIPS MANEUVERING IN WAVES

Andrea Serani, Paolo Dragone, Frederick Stern, Matteo Diez

correctmedium confidence

Category: Not specified
Journal tier: Specialist/Solid
Processed: Sep 28, 2025, 12:56 AM
arXiv Links: Abstract ↗PDF ↗

Audit review

The paper empirically shows that ADMD (derivatives + time-shifts) outperforms standard DMD for near-term forecasting and identifies the best NOW=4 settings as (5415M) NIW=8, NDE=2, NTS=2 and (KCS) NIW=8, NDE=4, NTS=2, using 1001 random samples; it also notes usefulness up to roughly two encounter periods and that ADMD is consistently better than DMD in the case studies . The model reproduces these settings and adds a coherent weighted least-squares/GLS rationale for why augmentation helps and why one-step fit cannot worsen in noiseless linear dynamics. Minor divergences are interpretive (e.g., the paper’s “best on average” delay of 1/8 Te vs. the model’s emphasis on 1/16 Te), but they do not undermine the main agreement.

Referee report (LaTeX)

\textbf{Recommendation:} minor revisions

\textbf{Journal Tier:} specialist/solid

\textbf{Justification:}

Empirical support for ADMD’s advantage over DMD is comprehensive and well presented for two realistic ship maneuvering datasets, with clear metrics and a full-factorial sweep. The results are practically useful and largely consistent across cases and horizons. The paper would benefit from a short theoretical framing (weighted LS view of augmentation), an explicit discussion of the eigenvalue-stability heuristic, and guidance on when particular delay lengths (1/16 vs. 1/8 Te) are preferable. These are clarifications rather than fundamental issues.