Currently, deep learning is extensively utilized for ship target detection; however, achieving accurate and real-time detection of multi-scale targets remains a significant challenge. Considering the diverse scenes, varied scales, and complex backgrounds of ships in optical remote sensing images, we introduce a network model named YOLO-RSA. The model consists of a backbone feature extraction network, a multi-scale feature pyramid, and a rotated detection head. We conduct thorough tests on the HRSC2016 and DOTA datasets to validate the proposed algorithm. Through ablation experiments, we assess the impact of each improvement component on the model. In comparative experiments, the proposed model surpasses other models in terms of Recall, Precision, and MAP on the HRSC2016 dataset. Finally, in generalization experiments, our proposed ship detection model exhibits excellent detection performance across various scenarios. The method can accurately detect multi-scale ships in the image and provide a basis for marine ship monitoring and port management.

Assessments of wave energy resources form the basis of plans for developing and using wave energy, and provide guidance about where to site wave power stations and how to design wave energy convertors. Various indexes, including the spatial and temporal distributions of wave energy fluxes, the occurrence of the effective significant wave height, monthly and seasonal variability and the frequency of high sea states, were calculated from the recent 37-year European Centre for Medium-Range Weather Forecasts ERA-Interim reanalysis of wave field data. The results showed that areas with abundant and stable exploitable wave energy in the South China Sea were mainly between the Luzon Strait and the southeast of the Indo-China Peninsula, where the annual average wave energy fluxes ranged from 10 to 18 kW/m and the monthly and seasonal variability indexes were less than 2·2 and 1·6, respectively. The wave energy was found to be poor in some coastal areas of the Beibu Gulf, the Gulf of Thailand and the Sulu Sea. In the areas where wave energy was abundant, high sea states occurred frequently, meaning that it was relatively difficult to use the energy. Therefore, the wave energy resources and the frequency of high sea states need to be considered when choosing sites for wave energy farms.

The distribution and spectral properties of colored dissolved organic matter (CDOM) in the northern Bay of Bengal were investigated in June 2016. Based on in situ data collected from 100 CDOM samples at 25 stations, the distribution characteristics of CDOM in the surface layer differed markedly from those at 30 m, 75 m, and 100 m. The CDOM spectral slope ($S_{350-500}$) exhibited a broad range, varying from 0.0026 to 0.0300 nm⁻¹, and showed a significant negative correlation with the absorption coefficient $a_{\mathrm{CDOM}}\left(443\right)$. Analysis of salinity and temperature profiles revealed no obvious correlation between the distribution of $a_{\mathrm{CDOM}}\left(443\right)$ and these physical parameters. A comparative analysis with satellite-derived wind and current data indicated that elevated $a_{\mathrm{CDOM}}\left(443\right)$ values in the northeastern surface waters were primarily associated with the southwest monsoon. In contrast, $a_{\mathrm{CDOM}}\left(443\right)$ values in the lower mixed layer were mainly influenced by the combined effects of geostrophic and eddy currents.

Reliable characterization of dependent wind–wave extremes is essential for offshore wind engineering in monsoon- and tropical-cyclone-affected seas. Using a 2002–2020 COAWST hindcast for an offshore site in the northern South China Sea, we analyze 161,400 common-valid hourly samples of 10-m wind speed (U₁₀), significant wave height (H_s), and peak wave period (Tp) and construct 221 temporally consistent monthly extreme pairs. Validation shows that the hindcast captures temporal co-variability but systematically underestimates wind and wave magnitudes, especially severe tropical-cyclone winds and high sea states; long-return-period estimates are therefore interpreted as methodological sensitivity measures rather than design loads. SJ-DPI and upper-tail sensitivity models based on Kernel–Pareto and GEV marginals are compared to quantify extrapolation uncertainty, while the primary uncertainty interpretation avoids relying on potentially unstable far-tail fits from limited monthly extremes. Under a common maximum pseudo-likelihood framework, no copula family is uniformly optimal, and AND-type exceedance is more sensitive to dependence-family choice than OR-type exceedance. Cool- and warm-season distributions differ significantly, but a 1000-split placebo test shows that part of the apparent long-return-period shift is attributable to the smaller-sample SJ-DPI bandwidths rather than seasonality alone. These results highlight the need to jointly consider concurrent sampling, marginal-tail specification, dependence uncertainty, and seasonal heterogeneity.