地基多通道微波辐射计反演大气温、湿廓线的试验研究  

Research for microwave radiometer remote sensing retrieval of temperature and humidity profiles

作　　者：王小兰[1] 王建凯[2] 李炬[3] 

机构地区：[1]中国气象局气象探测中心,北京100081 [2]中国气象局综合观测司,北京100081 [3]中国气象局北京城市研究所,北京100089

出　　处：《气象水文海洋仪器》2012年第4期1-6,共6页Meteorological,Hydrological and Marine Instruments

基　　金：公益性行业(气象)科研专项(GYHY2010066049)

摘　　要：利用以北京南郊观测场TP/WVP-3000地基微波辐射计的观测亮温和对应的L波段探空温、湿廓线作为3层BP神经网络算法的训练样本的输入和输出数据,在不同天气条件(晴天、云天无雨)和不同季节(夏、秋)情况下组建训练样本集对BP神经网络进行训练,以探空廓线为标准,考察独立测试样本的计算精度,结果发现:(1)研究构建的BP网络的计算精度明显优于地基微波辐射计自带的网络算法(BP-mp);(2)对训练样本进行分类处理后所构建的网络,其反演精度有所下降,因此不必要对样本进行过细的分类;(3)对于大气廓线的反演,按季节划分比按天气条件划分更有意义。可见,优质的训练样本、合理的网络结构和计算方法、足够多的训练样本数量是提高BP神经网络的泛化能力和反演精度的必要条件。In order to analyse the retrieved precision in different weather condition（sunny or cloudy） and seasons（summer or fall）,the brightness temperature observed by ground-based 12-channel microwave radiometer（TP /WVP-3000） and the atmospheric profiles from L-band sounding radar were used as the input and output of BP neural network.Taking the temperature and water vapor density profiles measured by radio sounding as standard value,the retrieved precision of independent samples were counted.The results found that the precision of the constructed BP neural network based on all kinds of classified samples were magnificently better than that（BP-mp） of microwave radiometer.The precision decreased when the training sample database was classified,which means it is unnecessary to classify the training samples.For retrieved precision of atmospheric profiles,the classification according to seasons was more meaningful than to weather conditions.In brief,superior training samples,reasonable network structure and computing method and enough training samples were determinative factors for raising the generalization ability and retrieved precision of BP neural network.

关 键 词：地基微波辐射计 BP神经网络 大气廓线 反演精度 

分 类 号：TB715[一般工业技术—真空技术]