High Performance Computing And Distilling Crustal Deformation Information From GPS Data

With the development of modern science and technology, large scale computing is more and more important to scientific research. On the other hand, with the establishment of"Crustal Movement Observation Network of China", GPS technology has played a more significant role in the domain of crustal monitoring and it has provided high precision, large scale and multi-dimension data for the researchers. GPS data contain abundant information, such as crustal movement information, crustal deformation information and many kinds of noise information, so how to distill the information from GPS data in efficient and fast way is an important question.This paper is based on the idea of solving the above question and it contains 2 parts that the first one aims at high performance computing method and the second aims at the application of these methods in distilling crustal deformation information from GPS data. The fore part is the technique preparation and it includes the discussion of matrix class based on "Message Passing Interface" parallel computing and other relating high performance computing methods. On the other hand, the later part is the main body of this paper, which includes illuminating the relating geodesy knowledge, analyzing the different strain calculating methods, analyzing the GPS time-series and analyzing the crustal deformation field in local region. On the research of high performance computing methods and distilling crustal deformation information from GPS data, some primary results are obtained.1) The research of high performance computing methodBy the discussion of dense matrix in parallel computing, the parallel methods of whole pivot element and column pivot element are given in the matrix inversion operation. When the rank of matrix reaches 2500 and the number of processes reaches 8, the efficiencies of these 2 methods can reach 0.7. Meanwhile, some integrity parallel software, such as PLAPACK, PETSc and TAO, are introduced.Then, the parallel computing methods are used in practice, such as strain field calculating and distilling information from time-series using least square collocation. By using these parallel computing methods, the calculating efficiency increases to a great extent, especially in strain field calculating process with simulating data.2) The analysis of different stain calculating methodThe content of this part contains the analysis of block strain calculating methods and low frequency strain calculating methods. The fore content aims at the formula of Li Yanxing, the formula of Savage and the plane formula of Gauss projection and Lambert projection. By comparing the results of these 3 formulas using simulating data, the conclusion is that the difference between the sphere formula and plane formula increases when the area of points distribution increases and does not relate to the latitude changing of points.On the other hand, the main content of this part includes analyzing the low frequency strain calculating method too. The multi-surface function method and least square collocating method in sphere surface are proposed, and spherical harmonic method is adjusted in the sequence of longitude and latitude when analyzing low frequency strain calculating methods. By comparing of the result obtained by different methods using simulating data, the conclusion is that all the results are similar to the real strain field, so these different methods are validated.3) The analysis of GPS time-seriesFirst, by the research of the least square collocation model and its application in GPS time-series, some different frequency information is distilled. This method has some advantages, which include precise theory foundation, complete functions and can finish data filtering and interpolation at one time. On the other hand, this method can be used as the pretreatment process of spectroscopic analysis and wavelet transformation method. Second, on the analysis of earth rotation data and GPS time series, the transition phenomena can be found in 2004. Then, the centrifugal force distribution due to earth rotation variation is researched in different latitude zone. Meanwhile, the statistics result of north orientation of GPS time-series is given and GPS baseline time-series is discussed.Third, on the research of the method of GPS data separating, the block movement can be classed rigid movement, strain contribution deformation and irregular deformation. After separating information in Huabei region, it is found that the proportions are different. The rigid movement proportion is large and other two are equal small. Considering of earthquake preparation, strain contribution deformation and irregular deformation are relative to it closely and they reflect energy accumulation and fast deformation before earthquake.Last, this part analyzes the stabilization of GPS time-series using the correlation coefficient between the whole time-series and its sub-sample. The result shows that the valid information of GPS time-series is very stable, and the good result can be obtained using even 10% data, so a few data lacking does not affect the analysis result.4) The analysis of regional crustal deformationFist, this part analyses the global plate movement, which using multi-surface function method and Euler model. By the analysis of the movement and deformation of global plate, their characteristics are discussed.Second, the evolution of the velocity and strain field in Chinese mainland is researched, and the result shows that the strain field has adjusted after Kunlun earthquake. The mainly adjustment of area strain field is that it shows tensional characteristic on the west of 91 degree longitude and shows pressure characteristic on the east. On the other hand, the adjustment of maximum shear strain field is that high value zone removes from west to east.Last, on the research of deformation field in local region, the evolution of strain field in Xinjiang and North South seismic belt are discussed and the evolution characteristic are analyzed.