时序形变资料的多核函数滤波方法研究及应用

针对多年时序形变观测资料有效信息提取复杂的问题,对基于多核函数的滤波方法进行研究,得到以下有益结论:(1)当核函数指数为0.5,光滑因子为0.003时,10天及以上核点间隔的滤波模型单位权中误差最小;(2)核点间隔控制滤波信息频谱的高低,间隔越大频谱信息越低,反之则频谱信息越高;(3)因数据缺失部分造成核点减少,当连续减少2个以上时滤波失败,当连续减少2个时数据缺失部分滤波出现失真,当减少1个时滤波效果不受影响;(4)通过对GPS时序资料、定点形变时序资料和非构造形变时序资料的滤波应用,获取不同频谱的信息,验证了本文方法的稳定性和可靠性。

Study and Application of Multi-kernel Function Filtering Method in Time-series Deformation Data Processing

Due to observations of environmental impact carried out by monitoring stations on the China Mainland, we need to deal with the daily data for many years for daily tracking purposes. Existing filtering methods include the wavelet method, least-squares co-location method, Gaussian weighted average method, and Vondark method, etc., but for daily tracking these methods prove to be inconvenient. Aimed at solving the problem of extracting efficient information from sequential deformation observation data over some years, a filtering method, based on multi-kernel function, is studied in this paper. Taking continuous GPS station vertical sequential observation data as an example, we discuss the parameters for the multi-kernel function and their physical meaning. Conclusions are as follows:(1) When the kernel function index is 0.5 and the smoothing factor 0.003, the mean square error of unit weight of the filtering model with a kernel point interval of more than 10 days, is the least. (2) The kernel point interval controls the level of the filtering information frequency spectrum, the larger the interval, the lower the spectrum information; the smaller the interval, the higher the spectrum information. (3) Sometimes kernel points are lost because of missing data. When more than two continuous points are lost, the filtering fails; when two continuous points are lost, part of the filtering waves are distorted because of the missing data; when just one point is lost, the filtering effect is not affected. (4) From the filtering application in the GPS time-series data, the fixed-point deformation time-series data, and the non-tectonic deformation time-series data, information on different spectra are obtained and the stability and reliability of the method verified. This provides a more convenient way to daily process time-series observation data from a large number of stations.