张文海,李磊. 2019. 人工智能在冰雹识别及临近预报中的初步应用[J]. 气象学报, 77(2):282-291, doi:10.11676/qxxb2019.014
人工智能在冰雹识别及临近预报中的初步应用
A preliminary application of artificial intelligence on the detection and nowcasting of hail weather
投稿时间:2018-04-10  修订日期:2018-07-17
DOI:10.11676/qxxb2019.014
中文关键词:  冰雹识别  临近预报  人工智能  机器学习  贝叶斯分类
英文关键词:Hail detection  Nowcasting  Artificial intelligence  Machine learning  Bayes classifier
基金项目:国家重点研发计划(2016YFC0203600)、国家自然科学基金(41575005)、中国气象局华南区域中心项目(GRMC2015M02)。
作者单位E-mail
张文海 深圳市强风暴科学研究院, 深圳, 518057  
李磊 深圳市气象局, 深圳市国家气候观象台, 深圳, 518040 chonp@163.com 
摘要点击次数: 364
全文下载次数: 484
中文摘要:
      基于广东10部S波段多普勒天气雷达的三维拼图资料,利用机器学习技术开发了一种冰雹识别和临近预报的人工智能算法。算法设计时以雷达回波反射率的垂直和水平扫描数据为基础训练集,将冰雹云的雷达反射率扫描数据作为正样本,将其他雷达反射率扫描数据作为负样本,通过贝叶斯分类法对正、负样本数据集进行机器学习,训练人工智能识别冰雹云内在规律的能力。训练时以广东省2008-2013和2015-2016年的数据作为训练集,使用了2014年广东省12次冰雹过程的数据做检验。对比检验的结果表明,人工智能法比传统的概念模型法击中率高9个百分点。研究结果表明了人工智能对冰雹这类非线性强天气过程具有较强的识别能力。
英文摘要:
      Based on three-dimensional mosaic reflectivity data from ten S-band Doppler radars in Guangdong Province, an artificial intelligence (AI) algorithm for automatic hail detection and nowcasting is developed using the machine learning technology. The training dataset used to develop the algorithm includes vertical and horizontal slice data of the mosaic radar reflectivity, in which the slice data of hail clouds are taken as positive samples, and other data are used as negative samples. The Bayes classifier method is used during the training process of machine learning to establish the capability of AI on recognizing the characteristics of the hail cloud. The data during the period of 2008-2013 and 2015-2016 are taken as training sets, while the observed data during the 12 hail weather processes in 2014 are used to validate the capability of AI. The result of comparative validation is encouraging, and the AI method is 9% higher than the traditional Conceptual Model method on identifying the hit rate. The current study preliminarily shows the strong capability of AI on identifying the nonlinear strong weather processes.
HTML   查看全文   查看/发表评论  下载PDF阅读器
分享按钮