球面质心泰森多边形网格海浪预报的向量化检验方法  

Vectorized verification method for Spherical Centroid Voronoi Tessellation grid ocean wave forecasting

作　　者：谢明炎 魏立新[1,2] 高元勇 高志一[1,2] XIE Mingyan;WEI Lixin;GAO Yuanyong;GAO Zhiyi(National Marine Environmental Forecasting Center,Beijing 100081,China;Key Laboratory of Marine Hazards Forecasting,National Marine Environmental Forecasting Center,Ministry of Natural Resources,Beijing 100081,China)

机构地区：[1]国家海洋环境预报中心,北京100081 [2]国家海洋环境预报中心自然资源部海洋灾害预报技术重点实验室,北京100081

出　　处：《海洋预报》2024年第2期34-43,共10页Marine Forecasts

基　　金：国家重点研发计划(2023YFC3107800)。

摘　　要：提出一种适于并行计算设备的向量化检验方法。采用特征半径近似判别法选取搜索域内观测点平均值作为实测值,基于广播、索引等向量化方法及海浪平稳性条件实现模式和观测资料时间和空间的匹配,并将待检区域等分为若干块分别计算以免内存溢出。本算法可在中央处理器(CPU)和图形处理器(GPU)上运行,也适用于其它类型的模式网格。为评估算法实际效果,以有限体积海浪模式产生的球面质心泰森多边形网格预报产品和卫星沿轨道遥感产品分别作为待检和参考数据,结果表明CPU和GPU的计算结果一致;GPU的计算效率较CPU提高30倍;区块大小对效率提升有一定影响,需根据实际算例调试优化。This paper introduces a vectorized inspection method suitable for parallel computing devices.The method employs the feature radius approximate discrimination method to select the average value of observed points within the search domain as the measured value.It relies on vectorization techniques such as broadcasting,indexing,and the smoothness conditions of waves to achieve matching of patterns and observational data in both time and space.Additionally,the algorithm divides the inspection area into several blocks for separate calculations to prevent memory overflow.The algorithm can run on Central Processing Unit(CPU)and Graphics Processing Unit(GPU),and it is applicable to other types of pattern grids.To evaluate the practical effectiveness of the algorithm,Spherical Centroidal Voronoi Tesselations grid forecast products generated by the Finite Volume Wave Model and satellite along-track remote sensing products are used as the test and reference data,respectively.The results show consistency between CPU and GPU computations,with a 30-fold increase in efficiency on the GPU compared to the CPU.The block size has a certain impact on efficiency improvement,requiring tuning and optimization based on actual cases.

关 键 词：精度检验 GPU加速 向量化 高性能检验 

分 类 号：P731.33[天文地球—海洋科学]