A Spatiotemporal Machine Learning Framework for the Prediction of Metocean Conditions in the Gulf of Mexico: Application to Loop Current and Loop Current Eddy Forecasting DOI
Edward C. C. Steele, Matthew P. Juniper, Ajit C. Pillai

et al.

Offshore Technology Conference, Journal Year: 2025, Volume and Issue: unknown

Published: April 28, 2025

Abstract We are entering an exciting new era of data-driven weather prediction, where forecast models trained on historical data (including observations and reanalyses) offer alternative to directly solving the governing equations fluid dynamics. By capitalizing a vast amount available information – capturing their inherent patterns that not represented explicitly such machine learning-based techniques have potential increase accuracy, augmenting traditional physics-based equivalents. Here, we adapt apply promising learning framework originally proposed by present authors for regional prediction ocean waves operational forecasting Loop Current Eddies (LC/LCEs) in Gulf Mexico (GoM). The approach consists using attention-based long short-term memory recurrent neural network learn temporal from observations, is then combined with random forest based spatial nowcasting model, high-resolution reanalysis data, develop complete spatiotemporal basin. Since approaches typically physics-agnostic, identical developed can be used surface currents, only difference being training datasets which this exposed. This illustrated here period three months October 2022 December 2022, model driven observation sites northern GoM. As such, it unrealistic expect performance unseen week January 2023 equivalent smaller/simpler domains more favorable quantity, quality coverage/distribution input but, despite these severe constraints, ability plausible structure LC/LCE system nonetheless impressive. architecture MaLCOM allows easy interrogation behavior us better unpick explain its characteristics thus providing path inform further enhancements. While still at early stage refinement, extension currents demonstrates encouraging fundamentally different way metocean general, forecasts particular, generated offshore energy sector, leveraging sparse sensor networks as basis predictions (further extending value when collected additional purpose mind). Provided suitable coverage, quantity available, advent very low cost, able run on-demand, in-house, standard laptop or desktop computers herald opportunities improving real-time decision-making support planning workability.

Language: Английский

A Spatiotemporal Machine Learning Framework for the Prediction of Metocean Conditions in the Gulf of Mexico: Application to Loop Current and Loop Current Eddy Forecasting DOI
Edward C. C. Steele, Matthew P. Juniper, Ajit C. Pillai

et al.

Offshore Technology Conference, Journal Year: 2025, Volume and Issue: unknown

Published: April 28, 2025

Abstract We are entering an exciting new era of data-driven weather prediction, where forecast models trained on historical data (including observations and reanalyses) offer alternative to directly solving the governing equations fluid dynamics. By capitalizing a vast amount available information – capturing their inherent patterns that not represented explicitly such machine learning-based techniques have potential increase accuracy, augmenting traditional physics-based equivalents. Here, we adapt apply promising learning framework originally proposed by present authors for regional prediction ocean waves operational forecasting Loop Current Eddies (LC/LCEs) in Gulf Mexico (GoM). The approach consists using attention-based long short-term memory recurrent neural network learn temporal from observations, is then combined with random forest based spatial nowcasting model, high-resolution reanalysis data, develop complete spatiotemporal basin. Since approaches typically physics-agnostic, identical developed can be used surface currents, only difference being training datasets which this exposed. This illustrated here period three months October 2022 December 2022, model driven observation sites northern GoM. As such, it unrealistic expect performance unseen week January 2023 equivalent smaller/simpler domains more favorable quantity, quality coverage/distribution input but, despite these severe constraints, ability plausible structure LC/LCE system nonetheless impressive. architecture MaLCOM allows easy interrogation behavior us better unpick explain its characteristics thus providing path inform further enhancements. While still at early stage refinement, extension currents demonstrates encouraging fundamentally different way metocean general, forecasts particular, generated offshore energy sector, leveraging sparse sensor networks as basis predictions (further extending value when collected additional purpose mind). Provided suitable coverage, quantity available, advent very low cost, able run on-demand, in-house, standard laptop or desktop computers herald opportunities improving real-time decision-making support planning workability.

Language: Английский

Citations

0