Multi-scale Tomography Of The Upper Mantle Beneath The Northeastern Tibetan Plateau

With development for more than 40 years, seismic tomography has become a prevailing method to study the structure of Earth’s interiors using seismic data and one of active domains of seismology. It includes model parameterization, forward calculation, inversion and imaging. Of these, the model parameterization is a mathematical representation of the medium in which seismic waves propagate. It affects many aspects such as the resolution and reliability of results.At present, there exist two problems in seismic tomography:non-uniform distribution of rays and spectral leakage. Usually, the non-uniform distribution of hypocenters and stations is inevitable, which leads to uneven distribution of seismic rays. If this problem cannot be treated properly, spectral leakage will be yielded. Scientists have tried many ways to solve the two problems, like the regularization with weighted damping factors, but these methods do not work well and cannot actually solve these problems. Model parameterization methods are used to solve the problems in the process of parameterization. Another problem in the model parameterization is that this process is not adaptive. For many model parameterizations, the grid size is fixed which means the resolution cannot change actively with the variable ray density, it is the so-called non-adaption. To handle with the spectral leakage and non-adaptation, scientists start to focus on the wavelet transform technique. With the adaptability and multi-scale property, the wavelet analysis can highlight certain properties of signals. Now this technique has been widely used in data compression, de-noising, image recognition, medical imaging and other fields. An adaptive wavelet-based model parameterization still needs to be developed and improved. This thesis introduces a multi-scale tomography by combining the wavelet-based model parameterization, L1-norm regularization and the idea of three-grid method.This thesis first reviews all model parameterization methods used in seismic tomography since the 1970s, and then presents a multi-scale tomography in detail. The previous studies show that:Choosing a proper model parameterization and corresponding forward and inversion methods should be based on the size of the study area, objective, quality of data and computing power, so that it is possible to extract the most accurate and useful information from observational data.It has different resolutions for different data and geophysical methods. The multi-scale parameterization is a perfect choice to realize a joint inversion by several geophysical methods in a same model parameterization.The sparse wavelet model parameterization method can actively adapt to the distribution of data and restrain the spectral leakage. It also can consider the resolution in the spatial and frequency domains at the same time. The sparse multi-scale wavelet model parameterization will be the trend of model parameterization in the future.The State Key Laboratory of Earthquake Dynamics, Institute of Geology, CEA deployed a dense movable array composed of 164 broadband seismic stations in the northeastern Tibetan plateau (called the Gansu array hereafter). We have selected the tele-seismic travel-time data from the big amount of data collected by the seismic array in 3 years and utilize the multi-scale tomography coupled with wavelet-based model parameterization and L1-norm regularization to inverse the 3D P-wave velocity structure of the upper mantle within the depth range of 400km beneath the study area. Comparing the results with the regional geologic and geophysical results, the conclusions of this work are summarized as follows:(1) The upper mantle beneath the northeastern Tibetan plateau exists the low-velocity, while the upper mantle beneath the Yangtze block displays the high-velocity. There is an obvious zoning characteristic in the eastern Tibetan plateau. Low speed character of the P wave is present in the upper mantle of the Songpan block, while high speed in the upper mantle of the Qaidam block.(2) There is a clear block boundary between the northeastern Tibetan plateau and the Yangtze block, which lies between 104°E and 105°E, and shifts eastwards with depth.(3) The Tianshui earthquake in 1654 and Wudu earthquake in 1879 occurred in the collision area between the Yangtze block and Tibet plateau, and the upper mantle beneath the epicenter exhibits a feature of high-low speed transition.The multi-scale tomography based on wavelet used in this study can adapt to non-uniform sampling data, reduce spectral leakage and the uncertainty of the inversion problem effectively. It can also improve the resolution and stability of the seismic tomographic results. Especially for the hyper-dispersion of rays, it is able to acquire more stable and reliable results with high resolutions. The 3D P wave velocity structure of the upper mantle beneath the northeastern Tibetan plateau which is inversed by the multi-scale tomography with wavelet-based model parameterization and L1-norm regularization, is very important to delimit the boundaries between the different blocks in the northeastern Tibetan plateau and solve the geodynamics of the Tibetan plateau.