Research On Application Of Singular Spectrum Analysis In Data Processing And Analysis Of Crustal Deformation

The observatory network of the Crustal Deformation China Seismological Bureau equip with many deformation instruments for researching in earthquake gestation,procession,and the earthquake disaster.It provides a rich observational data of crustal deformation and earth motion for the earthquake research.How to use effective data analysis method to separate out useful information from a large number of terrain data,has become a focus in the study of crustal deformation subject at present.After decades of research,the processing method of crustal deformation observation data is moving time domain analysis to frequency domain analysis,and then extended to the time-frequency analysis,so many useful analysis results were obtained.Singular Spectrum Analysis(SSA)is a method which analysis from the dynamic reconstruction of time series of adaptive analysis.Different from the classical analysis method,it analysis data in time domain,but carrying the characteristics of frequency domain.It has the filtering,denoising,interpolation,detection of periodic signal,and the function of the trend.At present,the SSA method is widely used in time series analysis,and has an application in geodesy,but has not been applied to the tilt and strain data processing.For this reason,this paper analyzes the concrete application of SSA method in time series,tests the applicability and effectiveness of the SSA method based on simulated data.And combining with the data characteristics of tilt and strain observations,we chose different sampling rate and the deformation data from different stations.SSA method preliminary applied to crustal deformation data processing and analysis,we got the following results and conclusions:(1)Proposed the best embedding dimension selection method which combined periodogram method and G-P correlation dimension algorithm,the best reconstruction order selection method which combined turning point method and variance contribution rate,and they validated by the simulated data.At the same time,the basic function of SSA method were also tested with simulated data.The results show that,SSA has the feature of separation the trend for periodic signal,can effectively remove the standard deviation which lower than the noise of the signal amplitude,and can well identify the singular point that exist in signal.(2)SSA method is applied to the tilt and strain data,we select three years of hourly values observed data of Yichang,Yinchuan Yushu station.According to the characteristic form of principal component,We decomposed the original observation data into four characteristic components: tendency item,periodic item,interference and the remaining items;For periodic item,We harmonic analysis and time-frequency analysis the reconstruction series of isolated periodic item,further verified that SSA can effectively isolate the tidal wave of ingredients and improve the precision of observation data;For remaining items,we select observation data of minutely value of Fushun and Wuhai,Yanqing station.After decomposition of SSA,we got refactoring sequences using inflection point method,and proved that the remaining items is background noise produced by the wind disturbance;For interference item,we select tilt and strain data of Dandong,Xuzhou and Jingxian stations.After processing of SSA,the interference component from the data are effectively separated,and the interference of a cycle of about 8 ~ 20 days rainfall was identified;For the abnormal information that exist in interference items,we select a sets of seismic data.After the separation of the interference,we found that the advantage of the interference terms cycle is 3 ~ 21 days,and found abnormal information about 3 months before the earthquake.This may reflect the stress field instability process before the earthquake.(3)SSA is applied to the actual tilt and strain observation data,we chose a slightly different embedding dimension L to the different data of different stations.After a large number of observation data process using SSA,we found that the embedding dimension L = 120 ~ 160 satisfy the trend term paragraphs,interference and periodic separation requirements in the treatment of tilt and strain observation data.