Constructing a SWOT Internal Wave Dataset Using Deep Learning

Abstract. Internal waves (IW) play a crucial role in energy transfer and vertical mixing as key dynamical processes. The Surface Water and Ocean Topography (SWOT) satellite, with its high-resolution sea surface height (SSH) observations, provides new data source for internal wave detecting. This study develops a multi-region automatic internal wave recognition framework named SWOT_IWD and constructs a SWOT internal wave detection dataset (https://doi.org/10.5281/zenodo.17666852, Xi et al. (2025)) covering 13 internal wave-prone regions worldwide from 2023 to 2025. A total of 21,682 SWOT passes are downloaded and processed for internal wave detection across different regions, identifying 2,011 passes containing internal wave signals and detecting a total of 3,264 internal wave signals. The dataset consists of SWOT data and IW labels, and includes visualized internal wave detection result images. The validation results confirm that the average accuracy of the SWOT internal wave dataset is 91.21 %. The study analyzes the spatial distribution, activity frequency, and the relationship between internal waves and topographic coupling across 13 regions included in the dataset. We also conducted a quantitative comparison of three data sources: SWOT, Sentinel-1 C-SAR, and Sentinel-3 OLCI. The results indicated that the detection availability of internal waves using SWOT data reached as high as 29.78 %. The study demonstrates that this dataset can provide high-quality sample data to support internal wave detection based on deep learning. Furthermore, two cases were presented to illustrate the potential of this dataset for internal wave tracking using multi-source remote sensing data.