Zhiyi Fu

Mesoscale eddies are characterized by swirling currents spanning from tens to hundreds of kilometers in diameter three-dimensional attributes holds paramount significance in driving advancements in both oceanographic research and engineering applications. Nonetheless, a notable absence of models capable of adeptly harnessing the scarcity of high-quality annotated marine data, to efficiently discern the three-dimensional morphological attributes of mesoscale eddies, is evident. To address this limitation, this paper constructs an innovative deep-learning-based model termed 3D-EddyNet, tailored for the precise identification and visualization of mesoscale eddies. In contrast to the prevailing 2D models that remain confined to surface-level data, 3D-EddyNet takes full advantage of three-dimensional convolutions to capture the essential characteristics of eddies. It is specifically tailored for recognizing spatial features within mesoscale eddies, including parameters like position, radius, and depth. The combination of dynamic convolutions and residual networks effectively enhances the model’s performance in a synergistic manner. The model employs the PReLU activation function to tackle gradient vanishing issues and improve convergence rates. It also addresses the challenge of foreground–background imbalance through cross-entropy functions. Additionally, to fine-tune the model’s effectiveness during the training phase, techniques such as random dropblock and batch normalization are skillfully incorporated. Furthermore, we created a training dataset using HYCOM data specifically from the South China Sea region. This dataset allowed for a comprehensive analysis of the spatial-temporal distribution and three-dimensional morphology of the eddies, serving as an assessment of the model’s practical effectiveness. The culmination of this analysis reveals an impressive 20% enhancement over 3D-UNet in detection accuracy, coupled with expedited convergence speed. Notably, the results obtained through our detection using empirical data align closely with those obtained by other scholars. The mesoscale eddies within this specific region unveil a discernible northeast-to-southwest distribution pattern, categorized into three principal morphological classifications: bowl-shaped, olive-shaped, and nearly cylindrical, with the bowl-shaped eddies prominently dominating.

Land cover mapping plays a pivotal role in global resource monitoring, sustainable development research, and effective management. However, the complexity of the mapping process, coupled with significant computational and data storage requirements, often leads to delays between data processing and product publication, thereby bringing challenges to creating multi-timesteps large-area products for monitoring dynamic land cover. Therefore, improving the efficiency of each stage in land cover mapping and automating the mapping process is currently an urgent issue to be addressed. This study proposes a high-performance automated large-area land cover mapping framework (HALF). By leveraging Docker and workflow technologies, the HALF effectively tackles model heterogeneity in complex land cover mapping processes, thereby simplifying model deployment and achieving a high degree of decoupling between production models. It optimizes key processes by incorporating high-performance computing techniques. To validate these methods, this study utilized Landsat imagery data and extracted samples using GLC_FCS and FROM_GLC, all of which were acquired at a spatial resolution of 30 m. Several 10° × 10° regions were chosen globally to illustrate the viability of generating large-area land cover using the HALF. In the sample collection phase, the HALF introduced an automated method for generating samples, which overlayed multiple prior products to generate a substantial number of samples, thus saving valuable manpower resources. Additionally, the HALF utilized high-performance computing technology to enhance the efficiency of the sample–image matching phase, thereby achieving a speed that was ten times faster than traditional matching methods. In the mapping stage, the HALF employed adaptive classification models to train the data in each region separately. Moreover, to address the challenge of handling a large number of classification results in a large area, the HALF utilized a parallel mosaicking method for classification results based on the concept of grid division, and the average processing time for a single image was approximately 6.5 s.

Large-scale land cover plays a crucial role in global resource monitoring and management, as well as research on sustainable development. However, the complexity of the mapping process, coupled with significant computational and data storage requirements, often leads to delays between data processing and product publication, creating challenges for dynamic monitoring of large-scale land cover. Therefore, improving the efficiency of each stage in large-scale land cover mapping and automating the mapping process is currently an urgent and critical issue that needs to be addressed. We propose a high-performance automated large-scale land cover mapping framework(HALF) that introduces high-performance computing technology to the field of land cover production. HALF optimizes key processes, such as automated sample point extraction, sample-remote sensing image matching, and large-scale classification result mosaic and update. We selected several 10°×10° regions globally and the research makes several significant contributions:(1)We design HALF for land cover mapping based on docker and CWL-Airflow, which solves the heterogeneity of models between complex processes in land cover mapping and simplifies the model deployment process. By introducing workflow organization, this method achieves a high degree of decoupling between the production models of each stage and the overall process, enhancing the scalability of the framework. (2)HALF propose an automatic sample points method that generates a large number of samples by overlaying and analyzing multiple prior products, thus saving the cost of manual sample selection. Using high-performance computing technology improved the computational efficiency of sample-image matching and feature extraction phase, with 10 times faster than traditional matching methods.(3)HALF propose a high-performance classification result mosaic method based on the idea of grid division. By quickly establishing the spatial relationship between the image and the product and performing parallel computing, the efficiency of the mosaicking in large areas is significantly improved. The average processing time for a single image is around 6.5 seconds.