集合背景误差方差中小波阈值去噪方法研究及试验

发布时间：2018-01-08 01:11

本文关键词：集合背景误差方差中小波阈值去噪方法研究及试验　出处：《物理学报》2017年02期 　论文类型：期刊论文

【摘要】：背景误差方差的集合估计值中带有大量采样噪声,在应用之前需进行降噪处理.区别于一般的高斯白噪声,采样噪声具有空间和尺度相关性,部分尺度上的噪声能级远大于平均能级.本文针对背景误差方差中采样噪声的特征,引入小波阈值去噪方法,并根据截断余项的小波系数分布特征发展了一种计算代价很小,能自动修正阈值的算法.一维理想试验结果表明,该方法能滤除大量采样噪声,提高背景误差方差估计值的精度.相对于原来的小波阈值方法,修正阈值后减少了因部分尺度上噪声能级过大导致的残差,去噪后的RMSE减少了13.28%.将该方法应用在实际的集合资料同化系统中,结果表明,小波阈值方法优于谱方法,阈值修正后能在不影响信号的前提下增大小波去噪强度.
[Abstract]:The set estimation of background error variance contains a large amount of sampling noise, which needs to be de-noised before application. Different from general Gao Si white noise, sampling noise has spatial and scale correlation. The noise level on some scales is much larger than the average energy level. In this paper, wavelet threshold denoising method is introduced to deal with the characteristics of sampling noise in the variance of background error. According to the wavelet coefficient distribution of truncated remainder, an algorithm is developed, which can automatically correct the threshold, and the algorithm can filter a large number of sampling noise. Compared with the original wavelet threshold method, the modified threshold reduces the residuals caused by the excessive noise level in some scales. After denoising, the RMSE is reduced by 13.28 steps. The method is applied to the practical data assimilation system. The results show that the wavelet threshold method is better than the spectral method. The wavelet denoising intensity can be increased without affecting the signal after the threshold is corrected.
【作者单位】：国防科技大学海洋科学与工程研究院;国防科技大学计算机学院;
【基金】：国家自然科学基金(批准号:41375113,41475094,41305101,41605070)资助的课题~~
【分类号】：TN713
【正文快照】： 1引言数值天气预报可用一个非线性方程组表示,初始场的精度在很大程度上决定了预报的准确性.为了得到高质量的初始场,变分资料同化需要融合各种观测资料和背景信息[1].其中背景误差协方差矩阵(B)在资料同化中起到了误差结构分布、信息传播、权重分配等作用,因此资料同化的首要

【相似文献】