卷积神经网络和传统算法的雷达面雨量计算效果对比研究  

A comparative study of convolution neural network and traditional algorithm in radar area rainfall calculation

作　　者：李宗飞[1] 陈凯华[1] 赵玉娟[1] Li zongfei;Chen Kaihua;Zhao Yujuan(Tianjin Meteorological Information Center,Tianjin 300000,China)

机构地区：[1]天津市气象信息中心,天津300000

出　　处：《气象研究与应用》2021年第4期89-94,共6页Journal of Meteorological Research and Application

基　　金：天津市气象局科研项目(202126ybxm18);环渤海区域科技协同创新基金项目(QYXM202013)。

摘　　要：利用人工智能方法对雷达反射率进行面雨量计算,使用2D卷积神经网络(Conv2D)和U-Net卷积神经网络的面雨量计算效果与传统的Z-I关系法反演面雨量效果进行对比分析。结果表明,卷积神经网络能够实现面雨量估计,但仍存在平均和聚拢现象,与传统算法各有优缺点。In this paper, for the first time, artificial intelligence was used to calculate the surface rainfall of radar reflectivity. The calculation results of area rainfall using 2D convolution neural network(Conv2D) and Unet convolution network were compared with the traditional Z-I relationship method. The results show that the convolution neural network can estimate the area rainfall, but there are still averaging and convergence phenomena, which has its own advantages and disadvantages compared with the traditional algorithm.

关 键 词：人工智能 神经网络 面雨量 Z-I关系 

分 类 号：P412[天文地球—大气科学及气象学]