I/O is often a performance bottleneck in global ocean circulation models with fine spatial resolution. In this paper, we present an asynchronous parallel I/O framework and demonstrate its efficacy in the Mass Conservation Ocean Model (MaCOM) as a case study. By largely reducing I/O operations in computing processes and overlapping output in I/O processes with computation in computing processes, this framework significantly improves the performance of the MaCOM. Through both reordering output data for maintaining data continuity and combining file access for reducing file operations, the I/O optimizing algorithms are provided to improve output bandwidth. In the case study of the MaCOM, the cost of output in I/O processes can be overlapped by up to 99% with computation in computing processes as decreasing output frequency. The 1D data output bandwidth with these I/O optimizing algorithms is 3.1 times faster than before optimization at 16 I/O worker processes. Compared to the synchronous parallel I/O framework, the overall performance of MaCOM is improved by 38.8% at 1024 computing processes for a 7-day global ocean forecast with 1 output every 2 h through the asynchronous parallel I/O framework presented in this paper.
Wind field retrieval, a crucial component of weather forecasting, has been significantly enhanced by recent advances in deep learning. However, existing approaches that are primarily focused on wind speed retrieval are limited by their inability to achieve real-time, full-coverage retrievals at large scales. To address this problem, we propose a novel multi-scale fusion retrieval (MFR) method, leveraging geostationary observation satellites. At the mesoscale, MFR incorporates a cloud-to-wind transformer model, which employs local self-attention mechanisms to extract detailed wind field features. At large scales, MFR incorporates a multi-encoder coordinate U-net model, which incorporates multiple encoders and utilises coordinate information to fuse meso- to large-scale features, enabling accurate and regionally complete wind field retrievals, while reducing the computational resources required. The MFR method was validated using Level 1 data from the Himawari-8 satellite, covering a geographic range of 0–60°N and 100–160°E, at a resolution of 0.25°. Wind field retrieval was accomplished within seconds using a single graphics processing unit. The mean absolute error of wind speed obtained by the MFR was 0.97 m/s, surpassing the accuracy of the CFOSAT and HY-2B Level 2B wind field products. The mean absolute error for wind direction achieved by the MFR was 23.31°, outperforming CFOSAT Level 2B products and aligning closely with HY-2B Level 2B products. The MFR represents a pioneering approach for generating initial fields for large-scale grid forecasting models.
Weather radar reflectivity plays a critical role in precipitation estimation and convective storm identification. However, due to terrain limitations and the uneven spatial distribution of radar stations, oceanic regions have long suffered from a lack of radar observations, resulting in extensive monitoring gaps. Geostationary meteorological satellites have wide-area coverage and near-real-time observation capability, offering a viable solution for synthesizing radar reflectivity in these regions. Most previous synthesis studies have adopted fixed time-window data partitioning, which introduces significant noise into visible-light observations under large-scale, low-illumination conditions, thereby degrading synthesis quality. To address this issue, we propose an integrated deep-learning method that combines illumination-based classification and reflectivity synthesis to enhance the accuracy of radar reflectivity synthesis from geostationary meteorological satellites. This approach integrates a classification network with a synthesis network. First, visible-light observations from the Himawari-8 satellite are classified based on illumination conditions to separate valid signals from noise; then, noise-free infrared observations and multimodal fused data are fed into dedicated synthesis networks to generate composite reflectivity products. In experiments, the proposed method outperformed the baseline approach in regions with strong convection (≥35 dBZ), with a 9.5% improvement in the critical success index, a 7.5% increase in the probability of detection, and a 6.1% reduction in the false alarm rate. Additional experiments confirmed the applicability and robustness of the method across various complex scenarios.
High-resolution models are essential for simulating small-scale processes and topographical features, which play a crucial role in understanding meteorological and oceanic events, as well as climatic patterns. High-resolution modeling requires substantial improvement on the parallel scalability of the model to reduce runtime, while massive parallelism is associated with intensive communications. Point-to-point communication is extensively utilized for neighborhood communication in earth models due to its flexibility. The distributed graph topology, first introduced in the MPI version 2.2, provides a scalable and informative communication method. It has demonstrated significant speedups over the point-to-point communication method based on a variety of synthetic and real-world communication graph datasets. But its application in earth models for neighborhood communication is rarely studied. In this study, we implemented neighborhood communication using both the traditional point-to-point communication method and the distributed graph communication topology. We then compared their performance in a case study of the Finite Volume WAve Modeling (FVWAM). Across all tests with 512 to 32,768 processes, the communication time speedup of the distributed graph communication topology ranged from 1.28 to 5.63 compared to the point-to-point communication method. For operational global wave forecasts with 1,024 processes, the runtime of the FVWAM reduced 40.2 % when the point-to-point communication method was replaced by the distributed graph communication topology.
High-resolution models are essential for simulating small-scale processes and topographical features, which play a crucial role in understanding meteorological and oceanic events, as well as climatic patterns. High-resolution modeling requires substantial improvement on the parallel scalability of the model to reduce runtime, while massive parallelism is associated with intensive communications. Point-to-point communication is extensively utilized for neighborhood communication in earth models due to its flexibility. The distributed graph topology, first introduced in the MPI version 2.2, provides a scalable and informative communication method. It has demonstrated significant speedups over the point-to-point communication method based on a variety of synthetic and real-world communication graph datasets. But its application in earth models for neighborhood communication is rarely studied. In this study, we implemented neighborhood communication using both the traditional point-to-point communication method and the distributed graph communication topology. We then compared their performance in a case study of the Finite Volume WAve Modeling (FVWAM). Across all tests with 512 to 32,768 processes, the communication time speedup of the distributed graph communication topology ranged from 1.28 to 5.63 compared to the point-to-point communication method. For operational global wave forecasts with 1,024 processes, the runtime of the FVWAM reduced 40.2 % when the point-to-point communication method was replaced by the distributed graph communication topology.
19 January 2026 in IEEE Transactions on Geoscience and Remote Sensing
Severe convective storms (SCSs) present significant threats due to their rapid onset and destructive potential. Accurate nowcasting is essential to reduce both societal and economic disruption. Traditional deep learning methods often fail to capture fine-scale structures or provide forecasts that are overly smoothed and lack physical realism. Recent generative models, including generative adversarial networks (GANs) and diffusion-based approaches, address some of these limitations but continue to struggle to maintain coherent storm morphology over time. This article introduces a novel two-stage framework for convective storm nowcasting that integrates deterministic forecasting with generative evolution. In the first stage, we propose a multiscale convolutional recurrent network (MS-CRN) that effectively captures both local and global weather patterns to predict short-term storm dynamics. By modeling the complex interactions within small-scale meteorological phenomena while incorporating broader atmospheric trends, this approach substantially improves forecast accuracy. In the second stage, an edge-guided diffusion (EGD) model enhanced with edge-control refines the deterministic outputs, preserving storm boundaries and reducing blurring artifacts. Quantitative evaluations demonstrate that the MS-CRN achieves a mean CSI of 0.3305 and a mean Heidke skill score (HSS) of 0.4358, outperforming strong baselines by 3.9% and 4.3%, respectively. The evolution stage further improves coarse-scale performance, yielding a mean CSI-pool ($4\times 4$ ) of 0.4085, which is 13.9% higher than previous generative methods, particularly under high-reflectivity thresholds. The EGD can be fine-tuned and transferred across forecasting models without retraining the full generative framework, demonstrating its adaptability and practical value for broad integration into nowcasting systems. The source code will be made publicly available (https://github.com/T-mac010203/MS-EdgeCast)