Visualizing eddy structures

Visualization is a crucial technique for helping scientists view and understand complex datasets. To understand the data, a significant objective is to identify scientific structures or features within it and measure them over time. For flow structures in oceanographic observational and simulation data, a key feature is the eddy, which is a swirling structure that extends from the ocean surface into deep water. This thesis focuses on eddy extraction and visualization from both 2D and 3D time-varying observation and simulation datasets. A Hybrid-Eddy-Detection method is proposed, offering high efficiency and comparable accuracy, by combining the physical pattern and geometric structure information of the eddy. A Dual-Streamline-Winding-Angle eddy detection method is introduced to provide a more accurate center and boundary of the eddy, consistent with the natural center of the velocity field. A 2D/3D-Rendering-Network is proposed for a universal 3D feature segmentation method, which significantly reduces the training time by using a 2D network compared to classic 3D deep learning-based segmentation techniques. A Vector-Data-Network is proposed to adapt the deep learning method to vector data in a scientific dataset. With these techniques, this thesis aims to more accurately identify and efficiently visualize complex eddy structures in 3D oceanographic datasets, thereby enabling domain scientists to gain a deeper understanding of ocean dynamics.