关键词: 高频地波雷达, 海流方向, 卷积神经网络, 长短期记忆神经网络

Abstract:

High-frequency surface wave radars often encounter interference from coastal topography and islands when detecting nearshore areas, leading to significant errors in the synthesized ocean current directions from radar data. Traditional high-frequency surface wave radar inversion algorithms do not account for the impact of these physical factors. To address this issue, leveraging the strengths of convolutional neural networks (CNN) and long short-term memory (LSTM) neural networks, this paper introduces a hybrid CNN-LSTM model. This model incorporates data on sea surface wind, tide, and elevation, allowing for the refinement of ocean current directions obtained from radar measurements. The experimental results show that the CNN-LSTM model, when combined with physical oceanographic factors, can effectively improve the quality of radar-detected ocean current data in areas affected by topography, significantly enhancing the accuracy of the synthesized ocean current direction. After model correction, compared to the original radar data, the directional angle error of the empirical orthogonal function ellipse decreased from 77.10° to 23.06°, the error of the ellipse’s major and minor axes decreased from 0.0896 to 0.0538, and the average flow directional angle error of the ocean currents decreased from 20.82° to 6.21°.

Key words: high-frequency surface wave radar, ocean current direction, convolutional neural network, long short-term memory neural network