Identification, tracking, and classification of ocean eddies in along track radar altimetry data using deep learning (EDDY)

使用深度学习 (EDDY) 对沿轨雷达测高数据中的海洋涡流进行识别、跟踪和分类

• 批准号: 444762031
• 负责人: Professor Dr.-Ing. Jürgen Kusche
• 金额: --
• 依托单位: Lehrstuhl für Astronomische, Physikalische und Mathematische Geodäsie
• 依托单位国家: 德国
• 项目类别: Research Grants
• 资助国家: 德国
• 项目状态: 未结题
• 来源: https://gepris.dfg.de/gepris/projekt/444762031?language=en
• 关键词: Identification; tracking; classification; ocean; eddies

All large ocean currents generate eddies, i.e. cyclonically or anticyclonically rotating water masses. While monitoring individual eddies has applications in marine biology and fishery, knowing eddy statistics over larger regions and time periods is required for understanding water mixing and vertical heat transport in the ocean and, thus, a prerequisite for testing ocean models. At mesoscale, eddies are observed in radar altimetry, and methods have been developed to identify, track and classify them in gridded maps of sea surface height derived from multi-mission data sets. However, this procedure has drawbacks since much information is lost in the gridding process. Instead, here we suggest to develop a method that would identify, track, and classify eddies predominantly from along-track altimetry. Additionally, we will work with multiple modalities with complementary views on the phenomenon such as from sea surface temperature maps serving to guide the procedure, which departs from our recently published (preliminary) work. Our method will be based on convolutional neural networks with task-specific network architectures that jointly exploit spatial as well as temporal information in one task. It will be applied to conventional and SAR-altimetry and validated, e.g. with results from the SWOT mission. A comprehensive benchmark with multi-modal remote sensing observations and labeled reference data will be constructed and will be made available to the public.

所有的大洋流都会产生漩涡，即气旋或反气旋旋转的水团。虽然监测单个涡旋在海洋生物学和渔业中有应用，但需要了解更大区域和时间段的涡旋统计数据，以了解海洋中的水混合和垂直热量输送，因此，这是测试海洋模型的先决条件。在中尺度上，在雷达测高中观察到了涡旋，并已制定了在从多任务数据集得出的海面高度网格地图中对其进行识别、跟踪和分类的方法。然而，这个过程有缺点，因为许多信息在网格化过程中丢失。相反，我们建议开发一种方法来识别、跟踪和分类主要来自沿轨道测高的涡流。此外，我们将与多种模式合作，在这一现象上有互补的观点，例如来自用于指导这一程序的海洋表面温度图，这与我们最近发表的(初步)工作不同。我们的方法将基于卷积神经网络，具有任务特定的网络结构，在一个任务中联合利用空间和时间信息。它将应用于常规和合成孔径雷达测高，并通过SWOT任务的结果进行验证。将建立一个综合基准，其中包括多模式遥感观测和标记的参考数据，并将向公众提供。