The Modeling And Analysis Of Seismic Exploration Random Noise On Land

Seismic exploration which is one of the key exploration methods of modern geophysics is used to deduce the geological structure situation under the ground by seismic waves which are exited by explosive or controlled sources propagating in different layers. There are various interruptions during seismic exploration, including the weather, surroundings, human activities, instruments etc. For this reason, the initial seismic data includes much noise which is superposed with the effective signals, and seismic data needs to be implement signal processing digitally before geological interpretation. Since 1960,digital signal processing is introduced into seismic exploration, and some high topics like“three high and one accuracy ”are put forward on seismic data processing. Random noise is one of key factors which reduce the signal noise ratio(SNR) of seismic data, and it appears in some periods of time or some traces record randomly with wide frequency range, random apparent velocity and propagation direction, which brings many difficulties for noise attenuation. There is a lot of random noise suppressing methods for now, and old ones are improved, new ones are proposed continually while it is the first question to know seismic random noise before filteringing it. In spite of randomness in the respects of frequency distribution, propagation direction etc, seismic random noise follows a rule in the respect of statistical characteristics. In recent years, some experts and scholars are trying to select suitable filtering methods on the base of statistical rules of random noise. The analysis of seismic random noise almost is qualitative while less quantitative.For the above reasons, seismic random noise modeling is proposed in this paper, by which random noise can be analyzed qualitatively and quantificationally. Random noise is distinguished by the types of noise sources, the characteristics in seismic records, the spatiotenporal additive effects of different kinds of noise, the near-surface responses and dynamic characteristics of different noise are analyzed. The suitable filtering methods are selected by random noise modeling. The main contributions of this dissertation are summarized as follows.1. The source function model and propagation model of random noise is confirmed. Seismic random noise is classified into natural noise and cultural noise according to the corresponding noise sources. Natural noise which is generated by natural outside force includes the ground surface deformation induced by wind force, microsiesmic driven by tree thunk vibration, and noise induced by branches and leaves rustling in wind. Cultural noise which is generated by human lives includes noise driven by machines running and people walking around the geophones, and noise emitted by cities and towns, traffic, factories in the distance. The source functions are determined on the base of the corresponding theories and experiences. Wave equation is used as the propagation model on random noise. According to Green function, a continuous distribution field source is the sum of point sources. It is assumed that the ground is half-infinite, homogeneous, isotropic, perfectly elastic media, and each kind of sources are distributed on the ground surface around the geophones, and the excitation waves propagate as the ground-roll waves because the forces on the ground surface play a main role for geophones. Random noise is the superposed wave-field excited by all of the point-sources. In the paper, the theoretical model of every kind of noise is built by solving the inhomogeneous wave equations(with each kind of source functions) and superposing all of the wave fields of point sources, according to which, the effects of different factors on noise and the change of noise characteristics in different data collection areas are analyzed theoretically. Based on the theoretical and experimental analysis, it can be seen that the higher the wind speed, the rougher the ground surface, the higher the mountain is, the larger and higher the amplitude and frequency of the noise is. The more the number of trees is, the larger the amplitude of noise is, and the thinner the branches and leaves are, the higher the frequency of noise is.2. The random noise of the real areas is modeled, through which the effectiveness of the modeling methods can be proved. The theoretical models of the random noise in the desert area of Tarim basin and the mountain area of Yunnan are built. It is considered that random noise is a two-dimension wave field, and the characteristics of vibrogram and wave-profile are compared between the simulated noise and the real noise. There are spectrum characteristics, phase locus(only in vibrogram), mean, variance, kurtosis, skewness, frequency distribution, and cumulative distribution function. The random noise in the desert area is composed by noise driven by wind force on the ground surface, near-field cultural noise and far-field cultural noise. The simulated noise record which is the sum of the three kinds of noise is compared with the real noise in the desert. The comparative results both of vibrogram and wave profile show the similar between the simulated noise and the real noise. The random noise of the mountain area in Yunnan is composed by the noise induced by wind force on the ground surface, microseismic generated by tree trunk vibration, noise induced by branches and leaves rustling in wind, near-field cultural noise and far-field cultural noise. The characteristics of vibrogram and wave-profile are compared between the real noise and the simulated noise. It can be seen that the simulated noise is basically the same with the real noise. Unlike the noise in the desert area, the effect speed up and the wind force on the trees are the main factors which induce the high frequency noise in the mountain area. The consistency of the simulated noise and the real noise both in the desert and the mountain area proves the effectiveness of the random noise modeling. Seismic random noise is complex and changeful, and random noise modeling is an approximate result of real noise, it but is the main part.3. The suitable filtering methods are selected by seismic random noise modeling. There are a lot of filtering methods with different advantages, and the filtered results of a same method are different if the noise is different. The corresponding simulated noise can be used on the base of random noise modeling. According to the random noise modeling, the compositions and the characteristics of the random noise in different areas are analyzed, and the appropriate filtering methods can be selected. We can know that the key composition of the random noise in the desert area is the near-field cultural noise, according to which, complex diffusion filtering(CDF) method whose real solution is equal to convolution of Gaussian window and noisy data is selected. CDF can preserve more details than Gaussian filtering. The synthetic records and field data in the desert area are processed by CDF, and the filtered results show the effectiveness of CDF. In the mountain area, the other noise is buried by the noise induced by branches and leaves rustling in wind, the frequency range of the noise is wide, and there are some key high frequency components, according to which, TFPF is selected to suppress the random noise in the mountain area. TFPF is equivalent to a low-pass time-invariant filter, it obtains the filtered signals by estimating the maximums of Wigner-Ville distribution of analytic signal of noisy signals. The filtered results of the synthetic records and field data show that TFPF is effective in the mountain area. The filtered results of CDF used in the mountain area and TFPF used in the desert area are shown, from which, it can be seen that CDF is obviously more suitable for the desert area, and TFPF is more suitable for the moutain area. Median filtering and wavelet transform are tried to be used when we select filtering method, and the filtered results are unsatisfactory. CDF and TFPF are only the examples of suitable methods, and there are other suitable methods. According to seismic random noise modeling, it can be analyzed quantificationally, and it can provide theoretical guidance for selecting suitable filtering methods, and improving signal and noise ratio(SNR).