Real-time in situ prediction of ocean currents

Journal: Ocean Engineering (Q1)

Abstract

Deep autoregressive networks (LSTM and Transformer) can predict ocean currents in real-time at any location in the world.

Data

Ocean current data at 831 sites (222 stations have data length greater than 2 months) from NOAA. Sampling interval: 6 or 10 min

Conclusion

LSTM and Transformer can predict ocean currents in real-time at any location in the world and even better than traditional Harmonic Method.

e.g., Comparison of speed Normalized Root Mean Squared Error (NRMSE) (lower is better)

Station ID	Depth (ft)	LSTM	TF	HM
hb0401	15.7	0.14	0.12	0.12
jx0302	29.5	0.1	0.09	0.33
jx0701	15	0.09	0.10	0.12
cb0701	13.9	0.18	0.19	0.2