深圳市气象局临近预报技术进展

Recent Development of Now casting Techniques in Shenzhen Meteorological Bureau

  • 摘要: 深圳市气象局从2006年开始进行基于雷达的临近预报方法研究。在引进和吸收了国内外对流风暴追踪技术算法的基础上,2008年开发了基于卡尔曼滤波的交叉相关法临近预报方法。但交叉相关法对局地生成和时间空间变化很快回波的预报会出现明显偏差。2013年开发了局部约束光流法临近预报方法,该方法对雷达基数据采用中值滤波进行质量控制,对变化较快的回波预报效果好,但对飑线等西风带系统的临近预报能力稍差。2015年成功研制了粒子滤波融合算法,该方法对局部约束光流法和Harris角点法获取的运动矢量场进行融合获取更接近回波真实运动的运动矢量场进行临近预报,对雷达基数据采用双边滤波进行质量控制,该方法改善了飑线的临近预报。为适应智能临近预报发展形势,2017年开展了基于生成对抗网络(generative adversarial networks,GAN)的人工智能临近预报方法研究,取得一定的成果。持续的临近预报方法研究,极大提升了深圳市气象局临近预报预警水平。

     

    Abstract: According to the requirements of operations, the Meteorological Bureau of Shenzhen Municipality has begun to study the weather nowcasting methods since 2006. In 2006, TITAN (thunderstorm identification, tracking, analysis and nowcasting)algorithm was introduced from the United States. As a supplement to TITAN algorithm, the cross-correlation method has been developed since 2008 and was used in weather service of the 2011 World University Games in Shenzhen. Due to the inherent defects of the cross-correlation method, obvious deviations can be found in the prediction of radar echo that generated locally and quickly changed temporally and spatially. Therefore, optical flow method was developed in 2013 and the method has been improved as particle filtering fusion algorithm to improve weather nowcasting for westerlies system such as squall line. To adapt to the development of intelligent weather nowcasting, the cooperation with Harbin Institute of Technology (Shenzhen) for artificial intelligent nowcasting based on GAN (generative adversarial networks) has been developed since 2017.

     

/

返回文章
返回