人工智能在气象数据集研制中的应用综述  

Review on application of artificial intelligence in meteorological datasets development

作　　者：李波[1] 潘勐 孙越 LI Bo;PAN Meng;SUN Yue(China Yangtze Power Co.,Ltd.,Yichang 443002,China;Hubei Meteorological Information and Technological Support Center,Wuhan 430074,China)

机构地区：[1]中国长江电力股份有限公司,湖北宜昌443002 [2]湖北省气象信息与技术保障中心,湖北武汉430074

出　　处：《人民长江》2025年第1期88-96,共9页Yangtze River

基　　金：中国长江电力股份有限公司科研项目(Z242302005);水利部重大科技计划项目(SKS-2022004)。

摘　　要：长序列、高时空分辨率气象数据集对气象业务和科研具有重要意义,但是应用过程中存在数据质量低、空间分辨率不足等问题。随着人工智能的发展,机器学习以及深度学习算法在气象领域逐步开展应用。针对气象数据集研制中的技术难点问题,梳理了人工智能技术的4个关键应用场景,即观测数据的质量控制、缺测数据的插补和重构、多源数据的融合以及低分辨网格数据的降尺度,对各类机器学习模型在上述场景应用中的优势和不足进行了综述,并采用文献计量方法对研究前沿及发展趋势进行了定量分析。研究表明:相比传统方法,人工智能算法在计算效率、结果准确性、应用灵活性等方面更具竞争力。建议从建立高质量训练数据集、加强多源数据和异构数据的应用以及探索基于气象数据物理机制的建模等多个方面,进一步推动人工智能在气象数据产品研制中的应用。Long-term meteorological datasets with high spatio-temporal resolution are of great significance to meteorological operations and scientific research.However,challenges such as low quality and low spatial resolution emerge during application of existing datasets.With the development of artificial intelligence,machine learning and deep learning algorithms have been applied in the field of meteorology.Addressing the technical challenges in the development of meteorological datasets,we outlined four application scenarios for artificial intelligence technologies:quality control of observations,interpolation and reconstruction of missing data,fusion of multi-source data,and downscaling of low-resolution gridding data.We comprehensively reviewed on the advantages and disadvantages of various machine learning models in these application scenarios and conducted a quantitative analysis on research frontiers and development trends using bibliometric methods.The research showed that compared to traditional methods,artificial intelligence algorithms are more competitive in terms of computational efficiency,accuracy,and application flexibility.It is recommended to further promote the application of artificial intelligence in the development of meteorological data products by establishing high-quality training datasets,enhancing the use of multi-source and heterogeneous data,and exploring modeling based on the physical mechanisms of meteorological data.

关 键 词：气象数据集 人工智能 机器学习 深度学习 质量控制 降尺度 

分 类 号：P409[天文地球—大气科学及气象学]