Accelerating Target Tracking in Atmospheric Motion Vector Retrieval Using Openacc

Atmospheric Motion Vector (AMV) provides essential wind field information, playing a key role in typhoon path prediction and intensity analysis. However, the spatial resolution of current mainstream AMV products is relatively low, limiting their ability to meet the high-precision demands of meteorological services. While the China Meteorological Administration's visible channel AMV product achieves a relatively high spatial resolution of 6 km, its long computational time prevents it from being applied in real-time operational scenarios.In this study, we propose a GPU-accelerated high-resolution wind field retrieval algorithm, designed to address the computational bottleneck of the target tracking component within the AMV retrieval process. By decomposing the core calculations into parallel tasks, we leverage OpenACC to efficiently implement parallel computing. Additionally, to overcome the memory limitations of a single GPU, we design a block-based computational strategy, enabling multi-GPU processing for handling larger datasets.Experimental results show that the proposed algorithm achieves significant acceleration, with computational efficiency improved by more than ten times compared to traditional CPU implementations, while maintaining the retrieval accuracy. The algorithm also demonstrates excellent scalability, supporting a wide range of remote sensing data resolutions, from 4000 m down to 500 m. This work presents a feasible technical solution for real-time operational high-resolution AMV retrieval, enhancing the timeliness of typhoon monitoring and numerical weather prediction assimilation.