Chen Wang

Ocean internal waves and eddies are typical mesoscale and submesoscale processes that play critical roles in ocean energy transfer and material transport. Their complex morphology and varying background conditions make detection from Synthetic Aperture Radar (SAR) imagery challenging. In this paper, we propose a novel multi-scale YOLO framework, termed MuLO, which designed to investigate the intrinsic connections among multi-scale ocean dynamic processes and to enable collaborative detection of oceanic eddies and internal waves in SAR imagery. This model integrates a Dynamic Grouping Cascade Module (DGCM) and a Multi-Scale Contextual Aggregation (MSCA) mechanism to enhance weak feature extraction and multi-scale feature fusion. A multi-source spaceborne SAR dataset including X-, C-, and L-band imagery acquired in the South China Sea was constructed, and a multi scale slicing inference strategy was designed for practical analysis. Experiments show that MuLO, with only 2.41M parameters, achieves 90.8% mAP50, outperforming several state-of-the-art models. Multi-source evaluation further demonstrates robust detection across radar bands, with a mean average precision (mAP) of 87.8% in simultaneous eddy and internal wave identification. These results highlight the potential of MuLO for operational monitoring of ocean dynamic processes by spaceborne SAR.

Wind streaks are common and important phenomenon in marine atmospheric boundary layer (MABL), presenting as a weak spatial-quasiperiodic signal in synthetic aperture radar (SAR) imagery. The literature on wind streaks in SAR imagery always focuses on sea surface wind retrieval rather than the relationship between wind streak wavelength and MABL. The objective investigation of the relationship requires automatic and precise wavelength extraction. However, it is a challenging task as the wind streaks are weak signals submerged in intense speckle noise on SAR images, and other phenomena (e.g., waves) can interfere with the extraction at the same time. We proposed an automatic method for extracting wind streak wavelength in SAR imagery. The method comprises a power spectrum density-based extraction algorithm and a subsequent quality control procedure. The algorithm extracts the wind streak wavelength from the 1D power spectrum across the streak direction. This approach differs from conventional methods that rely on the noisy 2D power spectrum of SAR images. The quality control procedure is to eliminate the interference from waves further. We extracted the wind streak wavelengths in 4,445 SAR images using the method. Then, we systematically analyzed the wavelength trends regarding the MABL environment parameters and checked the trends using the Mann-Kendall (MK) significant test. The results show that the wavelength increases as the parameters (sea surface temperature, sea surface air temperature, air humidity, boundary layer height, and vertical derivatives of temperature and humidity) increase. This finding indicates that thermodynamic instability could be significant in the occurrence of wind streaks.

Spaceborne synthetic aperture radar (SAR) provides high-resolution sea surface observations to characterize oceanic and atmospheric variability at submesoscales. Despite significant progress in retrieving wind fields from SAR images, challenges remain particularly for single-antenna measurements where wind speed inversion often requires an external wind direction input. This study introduces and evaluates the use of directional features derived from SAR image sub-look cross-spectra to improve wind retrieval. Specifically, we explore the imaginary part of SAR Mean Cross-Spectra (ΔIMACS) and its differential counterpart, ΔIMACS, which capture directional signatures of range-traveling intermediate wind waves. Using systematically collocated Sentinel-1A wave mode SAR images and model winds, we demonstrate that ΔIMACS exhibits a cosinelike dependence on wind direction at given wind speeds at two incidence angles, while ΔIMACS reveals unique sensitivities that complement radar backscattering and can resolve wind direction ambiguities. A novel cost function combining radar backscattering, ΔIMACS, and ΔIMACS is proposed to retrieve wind vectors directly from SAR data without external wind direction inputs. The proposed scheme is evaluated through comparisons with independent model outputs and buoy measurements, showing improved performance. These findings highlight the potential of ΔIMACS and ΔIMACS to enhance self-sufficient wind inversion, offering a promising pathway for single-polarized SAR wind retrieval.

A dataset of multi‐tagged sea surface roughness synthetic aperture radar (SAR) satellite images was established near Barbados from January to June 2016 to 2019. It is an advancement of the Sentinel‐1 Wave Mode TenGeoP‐SARwv (a labelled SAR imagery dataset of 10 geophysical phenomena from Sentinel‐1 wave mode) dataset that targets SAR marine atmospheric boundary layer (MABL) coherent structures. Twelve tags define roll vortices, convective cells, mixed rolls and convective cells, fronts, rain cells, cold pools and low winds. Examples are provided for each signature. The final dataset is comprised of 2100 Sentinel‐1 wave mode SAR images acquired at 36 incidence angle over an 8° × 8°region centered at 51° W, 15° N. Each image is tagged with one or multiple phenomena by five experts. This strategy extends the TenGeoP‐SARwv by identifying coexisting phenomena within a single SAR image and by the addition of mixed roll/cell states and cold pools. The dataset includes PNG‐formatted SAR image files along with two text files containing the file name, the central latitude/longitude, expert tags for each image, and all dataset metadata. There is a high degree of consensus among expert tags. The dataset complements existing hand‐labelled ocean SAR image datasets and offers the potential for new deep‐learning SAR image classification model developments. Future use is also expected to yield new insights into the tradewind MABL processes such as structure transitions and their relation to the stratification.