Estimation of lateral correlation length from deep seismic refection profle based on stochastic model

The fractal nature of crustal seismic velocity heterogeneity makes the interpretation of deep seismic refection data difcult by conventional methods developed in oil and gas exploration. Thus, several statistical approaches have been introduced as promising tools for interpreting the complex refection patterns of deep seismic data. Because stochastic models have been successfully used to describe the heterogeneity of crustal rocks, stochastic parameter estimation has become a potentially powerful tool for recovering information on geometric geological variations. However, there are many factors that infuence parameter estimation, with limited data being a profound one. We present a novel algorithm to estimate the lateral correlation length, an important stochastic parameter, from deep seismic refection data. First, an autoregressive power spectrum-based method was introduced to calculate the autocorrelation function from limited data. Second, the average multi-trace 1D autocorrelation function was used to replace the 2D autocorrelation function to improve the computation efciency, accuracy, and stability. Compared with other algorithms, a velocity model test showed that our method exhibited signifcantly better performance for a small dataset. Then, an appropriately sized sliding window of synthetic seismic data was applied to map the relative variations of lateral stochastic parameters. The results indicated that our method could distinguish the lateral variations in stochastic parameters as well as vertical changes. Finally, the geological meaning of diferent seismic refection patterns was discussed after applying our methods to deep seismic refection feld data. The results demonstrated that lateral correlation can clearly identify Moho discontinuity, crustal refections, and some sedimentary structures.