Internal solitary waves (ISWs) play a critical role in ocean mixing, marine biology, and offshore safety. Radar altimeters deliver continuous, all-weather sea surface height anomaly (SSHA) and backscatter observations. However, the along-track nature of both conventional and synthetic aperture radar altimetry (SRAL) limits the intuitive representation of ISW signatures, complicating their automated detection. Here, we propose ISW-DetSRALNet, a deep learning framework designed to detect ISWs along Sentinel-3 SRAL tracks. The core contributions are: (1) the integration of four complementary SRAL parameters to comprehensively capture ISW signatures, such as Sea Surface Height Anomaly, Normalized Radar Cross-Section, Significant Wave Height, and waveform-averaged power; and (2) architecture integrates convolutional layers, bidirectional Long Short-Term Memory (Bi-LSTM) networks, and a frequency – temporal attention mechanism to extract robust ISW features. Trained on 770 manually labelled ISW samples on 333 Sentinel-3 tracks across five regions and validated against optical imagery, SWOT data, and in-situ observations, the model achieves an F1-score of 98.3% and reliably detects weak solitons (~8 cm SSHA) even under high wind conditions ( > 6 m s − 1 ). However, as this high detection accuracy relies on ideal perpendicular geometries, it represents an upper bound relative to general scenarios with diverse intersection angles. Applied to regions such as the Dongsha Atoll waters and the Maluku Sea, the model reveals detailed occurrence characteristics of ISWs. Detection limits arise from ISW-track angle, deep pycnoclines, and wind roughness masking. The approach enables ISW monitoring with altimeters and supports oceanographic and climate modelling efforts.
Xu et al. (Tue,) studied this question.