EUVIMEDDescobrir EUVIMED
Crossref

Early Warning for Maritime Storm Formation Using Temporal Autoencoder-Based Anomaly Detection

Snehashish Srivastava, Haiping Xu, Donghui Yan, Ramprasad Balasubramanian

Journal of Engineering Research and Sciences · 2026-05 · 5 (5) · 19-39

DOI: 10.55708/js0505003 ↗

At a glance

Journal article in Journal of Engineering Research and Sciences, 2026-05. Written by Snehashish Srivastava and 3 co-authors. 43 references.

ResumoDados bibliográficosCiteRelated studies

Abstract (original language)

Storms remain a serious hazard at sea, exposing vessels to rapidly changing conditions that endanger human life and result in substantial economic losses. Satellite-based detection methods are widely used but require significant computational resources and depend on land-to-sea communication links, which may become unreliable during severe weather. Machine learning approaches offer strong potential for early detection; however, they typically rely on large labeled datasets that are often unavailable for rare or rapidly developing storms. Moreover, many existing models depend on predefined storm signatures, limiting their ability to adapt to previously unseen conditions in real time. To address these challenges, we propose a sensor-based framework for early detection of non-tropical maritime storms using key meteorological variables, including atmospheric pressure, humidity, wind speed, sea surface temperature, and near-surface air temperature. These variables can be measured directly by onboard sensors, enabling continuous monitoring without reliance on external communication infrastructure. The proposed method employs a temporal autoencoder trained exclusively on storm-free data to learn normal atmospheric patterns and detect anomalies associated with storm development. By identifying deviations from normal temporal behavior, the framework provides early warnings before storms fully develop. Comprehensive case studies using both synthetic and real-world meteorological data demonstrate that the proposed approach can detect developing storms in advance, achieving average lead times exceeding one hour while maintaining strong detection performance. These results highlight the potential of the framework as a practical and reliable solution for enhancing maritime safety.

Cite

APA

Srivastava, S., Xu, H., Yan, D., & Balasubramanian, R. (2026). Early Warning for Maritime Storm Formation Using Temporal Autoencoder-Based Anomaly Detection. Journal of Engineering Research and Sciences, 5(5), 19-39. https://doi.org/10.55708/js0505003

Vancouver

Srivastava S, Xu H, Yan D, Balasubramanian R. Early Warning for Maritime Storm Formation Using Temporal Autoencoder-Based Anomaly Detection. Journal of Engineering Research and Sciences. 2026;5(5):19-39. doi:10.55708/js0505003

BibTeX
@article{Srivastava2026,
  author = {Srivastava, Snehashish and Xu, Haiping and Yan, Donghui and Balasubramanian, Ramprasad},
  title = {Early Warning for Maritime Storm Formation Using Temporal Autoencoder-Based Anomaly Detection},
  journal = {Journal of Engineering Research and Sciences},
  year = {2026},
  volume = {5},
  number = {5},
  pages = {19--39},
  doi = {10.55708/js0505003}
}
Start for free →