Abstract:
Ocean subsurface thermal structure prediction is an area of active research field because of its scientific importance attach to ocean dynamic, air-sea interaction, and c...Show MoreMetadata
Abstract:
Ocean subsurface thermal structure prediction is an area of active research field because of its scientific importance attach to ocean dynamic, air-sea interaction, and climate change, but currently, most of the ocean temperature predictions are oriented to the sea surface temperature (SST) due to the lack of observed profile data inside the ocean, especially in some marginal sea areas, such as the South Yellow Sea Cold Water Mass (SYSCWM). In fact, the prediction for ocean subsurface thermal structure is more important than SST in some ocean fields. In this letter, a dynamic coupling vertical multifeature difference time series prediction model based on bi-long short-term memory (DVMFD-Bi-LSTM) is proposed for the subsurface thermal structure prediction in the SYSCWM. Bi-LSTM with the “bi-directional” structure enables information association in temporal dimension. The dynamic coupling vertical mechanism is used to realize the spatial correlation between two adjacent layers of the subsurface ocean, and the difference algorithm is introduced to ensure the accuracy and robustness of the new method. Besides, we construct multifeature datasets to improve data scale and quality and rely on a multistep prediction strategy for multiday prediction. For a more comprehensive evaluation, multiple groups of experiments are set up for comparison, and the RMSE of the new model is reduced to 0.517, and R2 is increased to 0.937, which verifies the good performance of it in both temporal and spatial dimensions.
Published in: IEEE Geoscience and Remote Sensing Letters ( Volume: 21)