Ehsan Jamali Hondori

A random layer-stripping method for seismic reflectivity inversion

Reflection coefficients and arrival times, together with seismic velocities, are significantly important for possible evaluation of reservoir properties in exploration seismology. Reflectivity inversion is one of the robust inverse techniques used to estimate layer properties by minimising misfit error between seismic data and model. On the other hand, the layer-stripping method produces subsurface images via a top-down procedure so that a given layer is modelled after all the upper layers have been inverted. In this paper, we have combined these two methods to develop a new random layer-stripping scheme which first determines the reflectivity series via a random-search algorithm and then estimates P-wave velocities. The first step can be viewed as a variant of sparse spiking deconvolution, and the second step is accomplished by considering empirical relations between density and P-wave velocity. The method has been successfully applied to Marmousi synthetic data to examine dipping reflectors and velocity gradients, and it has been found to be quite reliable for analysing complex structures. A comparison with minimum entropy deconvolution showed that our inversion algorithm gives better results in detecting the amplitudes and arrival times of seismic reflection events.

Spiking Deconvolution; an Inverse Problem Point of View

Summary There are numerous methods which make use of different algorithms to solve the well-known deconvolution problem in seismic data processing. Most of these solutions require restrictive assumptions to seismic wavelet and reflectivity series, in particular wavelet to be known and/or the seismic reflections to be white. Here we use a different approach in the deconvolution that is not sensitive to the phase characteristics of wavelet nor to the whiteness of reflectivity series. We define an inversion problem for deconvolution to avoid from computing the inverse of the seismic wavelet. First, we locate spikes by means of Adaptive Simulated Annealing (ASA) and then compute the amplitudes of them by Least Square method. A comparison between this method and Minimum Entropy Deconvolution (MED) shows that although both methods try to simplify the seismic model, this method yields better results.

Acoustic full waveform inversion for sub-seafloor exploration using vertical cable seismic

Full waveform inversion (FWI) of seismic reflection data has been widely used with different 2D/3D data acquisiton geometries. The method develops a reliable subsurface model by minimizing misfit between observed and simulated waveforms. Vertical Cable Seismic (VCS) is a recent seismic reflection method which uses vertical array of hytdrophones in order to record acoustic energy generated by seismic sources. VCS can acquire high resolution data by deploying hydrophone arrays very near to the seafloor which results in a higher signal to noise ratio. Because of the especial geometry of VCS data it is challenging to develop a velocity model based on the conventional processing techniques. We used a simulation experiment to evaluate the FWI results on seismic reflection data acquired using VCS geometry. Although dominant events in the data were reflections rather than diving waves, which are important for FWI, we could obtain a promising velocity model.

A Reflectivity Guided Elastic Full Waveform Inversion

Full waveform inversion (FWI) produces subsurface images by minimizing the misfit between observed data and calculated model using iterative local optimization algorithms like conjugate gradient method. This approach requires a starting model which should appear in the neighborhood of the global solution of the FWI problem to ensure that the modeled waveforms are less than half a period away from the recorded data. Usually, reflection traveltime tomography is used to create a long-wavelength background velocity model for starting FWI iterations. In this paper we suggest an alternative method to develop the starting model for FWI by using a reflectivity inversion technique. A depth section of migrated data is used to extract the reflection coefficients and impedance section, then the impedance section is converted to velocity model by considering a known density model. The reflectivity inversion can detect subsurface geological structures very well and on the other hand, an approximate known density model is a fair assumption for FWI and does not dramatically affect the long-wavelength model. We applied our method on a part of Marmousi2 model in order to develop P and S wave velocity models via elastic full waveform inversion in the frequency domain.

Full Waveform Inversion of Multi-shot Seismic Data Acquired for Complex Subsurface Structure

We studied full waveform inversion of seismic data acquired in elastic media to develop subsurface images from raw shot gathers. The forward modeling is based on finite difference solution of the elastic wave equation in the frequency domain. Gradient vector is calculated using an adjoint-state technique and pseudo Hessian matrix is used to precondition the gradient vector to update the model parameter via preconditioned conjugate gradient method. Two synthetic examples from crosswell tomography and surface acquisition experiments are presented to examine the ability of the method in proper subsurface imaging.

Efficient Frequency Domain Forward Modeling of Elastic Waveforms

We developed a Matlab package for finite difference frequency domain modeling of elastic waves in heterogeneous media which can be used to efficiently produce synthetic seismic data for full waveform inversion and modeling, or for wave propagation studies. By using 25-point finite difference stencil the number of necessary grid points per shortest shear wavelength reduced to 3.3 with an error smaller than 1%1). Although Matlab is a high level language, which is in general slower than other programing languages, the developed package exploits array processing ability of Matlab to compute the complex impedance matrix without including any loop in the algorithm. This brings a significant increase in computation performance and makes the package useful for developing realistic models. In order to suppress reflections from edges of the computation area Perfectly Matched Layers (PML) technique has been applied. Attenuation characteristics could be modeled easily by introducing complex valued velocities in frequency domain. Some examples show the performance of the package in modeling elastic waves.