基于贝叶斯分类器和回波物理特征的C波段雷达非气象回波识别方法和性能分析

Nonmeteorological Echoes Identification Method Based on Bayesian Classifier and Echo Physical Characteristics Using C-Band Radar and Its Performance

天气雷达在观测过程中,通常会受到非气象因子的干扰,产生非气象回波,从而严重影响雷达定量降水估计的精度和短临降水预报的性能。本文使用陕西省西安、延安等7部C波段多普勒天气雷达的体扫观测,构建了基于贝叶斯分类器和回波物理特征的质量控制方法:首先人工提取每部雷达的降水回波、地物回波和晴空回波的反射率因子,并基于提取的不同类型雷达回波,分析了陕西省7部雷达不同类型雷达回波的反射率因子、反射率因子水平纹理、反射率因子沿径向的变化梯度、5 dBZ回波高度、反射率垂直梯度的变化特征,统计得到不同类型雷达回波对应特征量的概率密度分布函数;然后基于统计的概率密度分布函数建立贝叶斯分类器,对雷达回波进行初步识别;在此基础上,结合雷达回波物理特征设计了太阳尖峰识别方法、孤立点去除方法和回波空洞填补方法,进一步识别雷达回波;最后去除非气象回波,得到质量控制后的降水回波数据。利用2019年7~9月陕西省7部雷达的体扫观测数据,系统地分析了雷达质量控制方法的性能,同目前陕西省业务运行的雷达数据质量控制结果进行了对比分析,并使用HSS评分(Heidke skill score)评估了质量控制结果的准确率。结果表明,研发的基于贝叶斯分类器和回波物理特征的雷达质量控制方法能够较好地识别降水回波和非降水回波,识别效果优于业务使用结果,7部雷达数据质量控制结果的HSS评分均在0.75以上。

Nonmeteorological factors usually interfere with weather radars during observation,resulting in nonmeteorological echoes that seriously affect the accuracy of the radar’s quantitative precipitation estimation and the performance of short-term precipitation forecasts.This study uses the scanning observations of C-band Doppler weather radars in Shaanxi Province(Xi’an,Yan’an,etc.)to construct a quality control method based on the Bayesian classifier and physical characteristics of the echo.First,the reflectivity factors of each radar’s precipitation echoes,ground clutter,and clear-air echoes are manually extracted.The reflectivity factor and its horizontal texture,the gradient of the reflectivity factor along the radial direction,the height of five dBZ,and the vertical gradient of the reflectivity of the different types of radar echoes from seven radars in the Shaanxi Province were analyzed based on the different types of radar echoes extracted.Additionally,we analyzed the probability density distribution functions of the corresponding characteristics of different types of radar echoes.Next,a Naïve Bayes classifier is established based on the statistical probability density distribution function to identify the radar echo.Then,combined with the physical characteristics of the echo,the sun spike filter,speckle filter,and hole filling are designed to further identify the echo.Finally,nonmeteorological echoes are removed to obtain precipitation echoes after quality control.The performance of the radar quality control method was systematically analyzed using the scanning observation data of seven radars in Shaanxi Province from July to September 2019.The radar data quality control method results for provincial business operations were compared and analyzed.The accuracy of the quality control results was evaluated using the Heidke skill score(HSS).Results show that the developed radar quality control method based on Bayesian classifiers and echo physical characteristics can better identify precipitation and nonprecipitation echoes,the recognition effect is better than the business results,and the HSS scores of data quality control results for seven radars are all above 0.75.