A Study Of Complex Near-surface Imaging

With the development of economy,the energy demand is becoming higher.Espe-cially for oil gas mineral resources.Therefore,more and more exploration work should be executed under the condition of complex near-surface structures,and it has become a significant challenge for data processing,geological survey,and risk analysis.The complex situations include discontinuous velocity structures,ragged near-surface to-pography,complex low-velocity layers,complex lithology changes and so on.For ex-ploration geophysics,near-surface long-wavelength statics plays a significant role for handing the cases with complex geological situations.We propose three methods to improve the quality and efficiency of near-surface imaging which is employed for the long-wavelength statics calculation.The methods involve the theories of sparse con-straint,multiscale analysis,and stochastic optimization.We describe them briefly as follows:1.Edge-preserving traveltime tomography with a sparse multiscale imaging con-straintWe assume that the near-surface velocity model is sparse under a known wavelet basis.According to the multiscale representation of the velocity model,we solve a nonlinear subproblem and linear subproblem alternately and obtain the low wavenum-ber velocity structures first,followed by the finer features.The nonlinear subproblem of the objective function is a l2 norm equation with a regularization term which includes an initial model constrained in it.Therefore,it can be solved easily via Gaussian Newton linearization and Conjugate Gradient method.The linear subproblem is a l1 norm equa-tion and we solve it by Fast Iterative Soft-thresholding Algorithm.Such a strategy can avoid solving the nonlinear l1 norm function directly,which is considered to be difficult and instable.We suppress the non-physical smoothness generated by the regularization and keep the true boundary of the subsurface structures by implementing the inversion from low wavenumber to high wavenumber.2.Alternating first-arrival traveltime tomography and multiscale waveform inver-sionTraveltime tomography is the common approach for near-surface imaging.How-ever,it usually gives a solution with low resolution.Also,it cannot invert the velocity models with complex low-velocity layer in it.On the other hand,full waveform inver-sion is proposed for better handling the problems that the traveltime tomography cannot solve.However,the final solution of waveform inversion is significantly affected by the initial model.In data domain,it is well known as the cycle-skipping problem.We develop an alternating inversion strategy by taking the advantages and reducing the dis-advantages of the two methods.The traveltime tomography and waveform inversion is implemented alternately through iterations,the solution obtained from waveform in-version should be the input model and the structural constraint of the traveltime tomog-raphy.Also,the result obtained by traveltime tomography is the input model of wave-form inversion.Furthermore,Waveform inversion is a multiscale approach in which a wavelet transform is applied to better handle the cycle-skipping problem.The new approach keeps both traveltime and waveform fitting well simultaneously and avoids selecting the non-physical parameter between the two approaches like the joint inver-sion method does.3.Highly efficient first-arrival traveltime tomography by stochastic optimizationImproving the efficiency of inversions is significant in industry because the large amount of data is collected for high quality imaging.We propose a highly efficient first-arrival traveltime tomography method by involving a mathematical theory of stochastic optimization.First,we intend to improve the efficiency of traveltime tomography by employing two methods named Sample Average Approximation and Stochastic Ap-proximation.They are two efficient methods of stochastic optimization.We found that Stochastic Approximation is more appropriate for improving the efficiency of traveltime tomography.Then on this basis we propose a fast imaging method base on Stochastic Approximation.The main idea of this method is performing the inversion with a small percentage of data.Additionally,the input data of every iteration is updated to further improve the efficiency of the inversion procedure.Especially for large 3D cases,the new method provides an alomost identical solution as the standard method,and shows an impressive performance on both computational time cost and memory requirement.4,A study on the shingling of the first arrivalsIn land seismic data processing,one usually picks the first arrival to image the near-surface velocity structure with a traveltime inversion method for making long-wavelength statics corrections.However,in many areas,the first arrivals of a shot gather are shingling.For the shingling arrivals,the precondition of first-arrival travel-time tomography fails.In this study,we employ an acoustic full waveform modeling method in a multi-layered half-space to simulate the shingling effects.The numerical tests indicate that when the wave propagates through a thin high-velocity layer which is embedded in a velocity structure with incremental speed,then shingling effects occur.The incursion of the thin high-velocity layer hides the information of the layer right below it,and later-arrival shingling branches with strong energy are refraction from the top of the layer below this hidden layer.For obtaining long-wavelength statics from shingling arrivals,we describe an approach to building an equivalent velocity model of the near-surface by implementing delay-time method while with an unusual picking manner.This approach is applicable because the shingling later-branches are confirmed to be refractions by the forward modeling numerical tests and the delay-time method is an entirely refraction based method.We evaluate the effectiveness of the proposed method in synthetic tests and real case.We also found that one cannot obtain an accurate long-wavelength statics by applying traveltime tomography roughly to shingling cases.