| Seismic prospecting is the most widely used method for hydrocarbon and mineral exploration. Due to the limitations of the acquisition conditions, seismic records always contain a plentiful of random noise. It brings challenges to identify and extract the effective reflected signals. Besides, the resources which are easy to detect and collect have become exhausted. We need to explore and utilize the resources in deep layers, thin layers and irregular layers. To detect these resources, it requires improving the quality of the seismic records. In other words, the random noise in seismic records should be attenuated more completely. Furthermore, the properties of the land-seismic-prospecting random noise have not been studied scientifically. It has bad effects on the random noise reduction and brings negative impacts on the development of the geophysical prospecting industry. An important step for improving denoising is to scientifically characterize the properties of seismic random noise. The requirements of the industry make the re-definition of the seismic random noise properties become increasing urgent. For these reasons, we start with the investigation of the parametric modeling algorithm of the seismic prospecting random noise. Then, we focus on the stationarity, Gaussianity and linearity of the seismic-prospecting random noise. By applying modern statistical testing methods, the properties of the noise have been intensively investigated. Besides, we also combine the forward modeling experiments with the generative mechanism of the random noise to prove the correctness of our findings in this paper.In this study, we first discuss the parametric modeling problem for the seismic prospecting random noise. On this basis, the statistical testing methods based on surrogates are used to investigate the stationarity, Gaussianity and linearity of the seismic-prospecting random noise. Recently, these testing methods have been controversial discussed in the field of statistical analyzing. The testing methods used in this dissertation have been successfully applied in many other studies, while the effectiveness and reliability of these methods has been proved. In this study, we check the stationarity, Gaussianity and linearity of the noise data, which are collected in the desert, steppe and mountainous areas. Furthermore, we also make comparisons between the statistical results of the random noise acquired in the different environments. From this viewpoint, our findings in this dissertation can represent the general properties of the seismic-prospecting random noise. The main findings of this dissertation are summarized as follows.1. In this study, a parametric modeling algorithm for the land-seismic-prospecting random noise is proposed, which is based on the fractional Brownian motion. By applying the modeling algorithm, we simulate the random noise acquired in different environments. We compare the differences between the properties of the real noise data and its simulated noise records. The results prove the effectiveness of our algorithm. On this basis, the spectral characteristics of the noise are investigated, and the optimum modeling parameters for the noise acquired in different environments are determined.2. We investigate the stationarity of the seismic random noise by applying the statistical testing methods based on surrogates and time-frequency(TF) analysis. From a conventional viewpoint, the nature of the random noise has long been believed to be wide-sense stationary. Our findings suggest that the seismic-prospecting random noise is not always stationary, but locally stationary. Besides, the time length of the noise and the complexity of the acquisition environments also have impacts on the stationarity of the seismic random noise. In general, the stationarity of the random noise deteriorates with the increasing time length of the noise records and the complexity of the acquisition environments. However, the noise could be treated as a stationary series in short time periods. In this study, we also discuss the differences between the stationary and non-stationary noise data in the frequency domain. By analyzing the behavior of the noise data, we determine that the non-stationary noise always has more energy in high-frequency bands. Moreover, the results indicate that the energy distributions of the stationary noise are relatively concentrated. In contrast, the energy distributions of the non-stationary noise are disordered, especially in the high frequencies. On this basis, we deduce that suppress the high-frequency components can improve the stationarity of the random noise, and use some experiments to prove the correctness of our findings.3. The Gaussianity and linearity testing methods based on surrogates and bispectral analysis are used to check the corresponding properties of the short-length land-seismic-prospecting random noise. The seismic random noise is often assumed to be a Gaussian process. Meanwhile, the linearity of the random noise has not been deeply studied yet. The results of this study show that the duration of the random noise does not seriously influence the properties of the short-length noise data. However, the properties of the random noise change obviously with the features of the acquisition environments. Generally, the seismic-prospecting random noise cannot be considered as a Gaussian process. The linearity of the random noise varied with the complexity of the acquisition environments. Specifically, the random noise in a simple environment is superior to that in a complex environment in terms of the linearity. We also compare the differences in the energy distributions between the noise data with different properties. The results indicate that there is little difference between the Gaussian noise and the non-Gaussian noise. However, the differences between the linear noise and the non-linear noise are quite obviously. In general, the energy distributions of the linear noise are more concentrated than that of the non-linear noise. The non-linear noise has more energy in high-frequency bands than the linear noise. We also compare the linearity of the noise before and after low-pass filtering. By the experiment, it is shown that attenuating the high-frequency components can obviously improve the linearity of the random noise.In this study, according to the characteristic of the seismic prospecting technique, a parametric modeling algorithm for the seismic prospecting random noise is proposed. At the same time, our study is the first, to our knowledge, to use modern statistical testing methods to scientifically investigate the stationarity, Gaussianity and linearity of the seismic-prospecting random noise. The characteristics of the seismic random noise have strong relations with the acquisition environments. Due to the variety of the environments, the features of the random noise should not be invariable. From this viewpoint, the traditional understanding about the features of the seismic-prospecting random noise has its own limitations, and too incomprehensive. In this study, we find that the random noise acquired in different acquisition environments has differences in properties. Some features even change with the time length of the noise. On this basis, we use the experiments and theoretical analyzing to demonstrate the accuracy and correctness of our findings. In conclusion, our findings are new re-identifications about the characteristics of the seismic-prospecting random noise. On the other hand, they can also be seen as the necessary developments of the traditional understandings of the seismic random noise. The findings are useful in establishing and developing better models for the land-seismic-prospecting background noise. They also have implications for future noise reduction and signal detection algorithms. From this viewpoint, our study has the crucial practical significance and application potentials. |
PDF Full Text Request