一种顾及噪声和天气突变的土壤湿度反演方法  

A soil moisture inversion method combining CEEMDAN and elman neural network

作　　者：刘原华 杨成 牛新亮[2] LIU Yuanhua;YANG Cheng;NIU Xinliang(School of Communications and Information Engineering Xi'an University of Posts&Telecommunication,Xi'an 710121,China;Xi'an Branch of China Academy of Space Technology China Academy of Space Technology,Xi'an 710100,China)

机构地区：[1]西安邮电大学通信与信息工程学院,西安710121 [2]中国空间技术研究院西安分院,西安710100

出　　处：《测绘科学》2023年第9期191-201,共11页Science of Surveying and Mapping

基　　金：国家自然科学基金项目(61971348)。

摘　　要：针对传统全球定位系统干涉反射测量GPS-IR土壤湿度反演流程中,信噪比序列因受到噪声干扰,导致信噪比残差序列的特征提取质量不高的问题,使用完全自适应噪声集合经验模态分解CEEMDAN进行信噪比序列的趋势项提取,同时引入Elman神经网络反演土壤湿度。两个测站的试验结果表明:相比使用多项式拟合,使用CEEMDAN方案的反演结果相关系数均值为0.913,均方根误差均值分别为0.033,相关系数均值提升了1.4%,均方根误差均值降低了14.3%;在使用CEEMDAN的基础上,引入Elman神经网络代替BP神经网络的反演结果相关系数均值为0.966,均方根误差均值为0.0185,相关系数均值提升了3.6%,均方根误差均值降低了23.5%。结果证明了CEEMDAN对于一般噪声有一定抑制作用,可提升特征参数的提取质量及Elman神经网络在以雨天为主的环境下,反演效果要优于BP神经网络。In response to the problem of low quality feature extraction of signal-to-noise ratio residual sequences due to noise interference in the traditional GPS-IR(Global Positioning System Interferometric Reflectometry)soil moisture inversion process,complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)is used to extract the trend term of the signal-to-noise ratio sequence,and Elman neural network is introduced to invert soil moisture.The experimental results based on two stations showed that compared to using polynomial fitting,the average correlation coefficient of the inversion results using the CEEMDAN scheme was 0.913,and the average root mean square error was 0.033,respectively.The average correlation coefficient increased by 1.4%,and the average root mean square error decreased by 14.3%;On the basis of using CEEMDAN,the Elman neural network was introduced to replace the BP neural network,and the average correlation coefficient of the inversion results was 0.966,with an average root mean square error of 0.0185.The average correlation coefficient increased by 3.6%,and the average root mean square error decreased by 23.5%.It has been proven that CEEMDAN has a certain inhibitory effect on general noise and can improve the quality of feature parameter extraction.Elman neural network has better inversion performance than BP neural network in rainy environments.

关 键 词：GPS-IR 土壤湿度反演 CEEMDAN ELMAN神经网络 

分 类 号：P237[天文地球—摄影测量与遥感]