摘要
海洋中尺度涡对大洋环流、能量平衡及气候变化具有重要调控作用,精准识别和分析涡旋的精细结构对海洋生态保护及灾害预警具有重要意义。然而,传统观测手段存在分辨率不足以及多源数据融合缺失等问题,难以满足亚中尺度精细结构的精准识别。为此,本研究提出了基于合成孔径雷达(SAR)与宽刈幅干涉高度计(SWOT)数据联合深度学习的新方法,首次构建多模态数据深度融合框架,通过SAR的高空间分辨率与SWOT的厘米级高程数据协同,突破单一数据源技术瓶颈,实现涡旋边界、亚中尺度涡丝及能量非对称分布的高精度解析。针对复杂海洋环境,设计物理约束深度学习模型,提出双分支网络架构结合交叉注意力机制,将涡旋极性分类精度提升,较传统阈值分割方法定位误差降低;通过分区块LSTM网络与对抗训练策略,捕捉涡旋强度、轮廓的时空演变。结果表明,该方法能够有效捕捉涡旋的精细结构及其时空变化规律,成功解析涡旋内部的亚中尺度特征,为中尺度涡的研究提供了全新的技术手段。研究为海洋灾害预警、气候模型参数化及生态保护提供了高精度观测基准,推动了海洋动力过程研究从统计描述向机理解析的发展。
Marine mesoscale eddies play a critical role in regulating ocean circulation,energy balance,and climate change.Accurate identification and analysis of their fine-scale structures are essential for marine ecological protection and disaster warning.However,traditional observation methods are limited by resolution constraints and challenges in integrating multi source data,making precise identification of submesoscale fine structures difficult.To address these challenges,this study proposes a novel approach based on joint deep learning using Synthetic Aperture Radar(SAR)and Surface Water Ocean Topography(SWOT)data.For the first time,a multi-modal data fusion framework is developed,leveraging the high spatial resolution of SAR and the centimeter-level elevation accuracy of SWOT.This fusion overcomes the limitations of single data sources,enabling high-precision analysis of eddy boundaries,submesoscale filaments,and asymmetric energy distribution.To tackle complex marine environments,a physics-informed deep learning model is designed,featuring a dual branch network architecture integrated with cross-attention mechanisms.This approach improves eddy polarity classification accuracy and reduces localization errors compared to traditional threshold segmentation methods.By employing a block based LSTM network and adversarial training strategies,the method effectively captures the spatiotemporal evolution of eddy intensity and contours.Experimental results demonstrate that the proposed method effectively identifies fine-scale structures and their spatiotemporal variations,resolving submesoscale features associated with heat and salt transport within eddies. This provides a novel technical approach for mesoscale eddy research and offers a high-precision observational benchmark for marine hazard warning, climate model parameterization, and ecological conservation. The study thus advances ocean dynamic research from statistical description toward mechanistic interpretation.
作者
刘传皓
LIU Chuanhao(Chongben Honors college,Ocean University of China,Qingdao 266100,China)
出处
《海洋通报》
北大核心
2025年第4期574-583,共10页
Marine Science Bulletin
作者简介
刘传皓(2004-),学士,主要从事海洋领域研究,电子邮箱:liuchuanhao@126.com。
