· · · Institutional Repository

Seismic interferometry is a technique by which the Green’s function (or impulse response) between two receivers can be acquired from the crosscorrelations of wavefield responses at these receivers. Recent developments of this method has led researchers to exploit active as well as passive seismic wavefields to retrieve surface wave Green’s functions by crosscorrelation. The primary objective of these applications has been to gain near surface resolution from the high frequency content of the active data while gaining greater depth resolution from the low frequency content of the passive data. In these applications however, a Green’s function is retrieved for each data type and therefore a matching filter or a form of joint inversion is required to benefit from the additional bandwidth of both data types. Interferometry by multidimensional deconvolution (MDD) is a relatively new method of Green’s function retrieval that provides several advantages over interferometry by crosscorrelation. This thesis proposes a new method of merging active and passive data during the process of MDD. A primary advantage of this method over the alternatives is that the source signatures are disregarded and only a single Green’s function with the combined characteristics of both the active and passive data is retrieved. Using numerical modelling it is demonstrated that a broadband Green’s function response can be retrieved from combined active and passive data without the need to compensate for the differences in source signatures or variations in amplitude. Merging active and passive data prior to deconvolution may in fact improve the retrieved response due to the additional illumination provided by the supplementary data. In addition to expanding the bandwidth of the retrieved response, this method is shown to be capable of using data from one source type to spatially infill gaps in illumination in another source type when the bandwidth of the two are comparable.

In oil and gas exploration, seismic arrays are deployed by geophycisists to image the subsurface. For passive seismic applications, the data recorded by the sensor array may contain velocity and angle information of the propagating seismic wave. This information can be used to infer the properties of material in different earth layers. In order to find the velocity and arrival angle, beamforming algorithms are applied to estimate the wavenumber-frequency spectrum for the seismic signals. The propagating seismic wave field consists of body waves and surface waves. In some applications, surface waves are interpreted as noise, thus filtering is required to remove the surface waves before or during the implemention of beamforming algorithms. In this thesis, we first introduce a data model. Then several beamforming algorithms based on the data model are discussed, and the performance of the different algorithms is evaluated. Capon beamforming as adopted in seismics has limitations. Robust Capon beamforming which can overcome these limitations is explained in the thesis. For filtering of the surface waves, we propose to first reconstruct the irregularly sampled spatial signal into a uniform array, then design a velocity filter to remove the unwanted low-speed noise (surface waves).

The Nevado del Ruiz volcano is an active and dangerous volcano in the Andean volcanic belt. Measuring seismic activity is one of the most reliable and widely used techniques to monitor and predict renewed volcanic activity. Seismic activity can be caused by several different underlying physical processes. It is of interest to the earth-science observatories monitoring potentially dangerous volcanoes to determine the underlying cause of the registered earthquakes. Typically segmented seismic recordings are classified by hand often based upon their frequency contents. An automated system capable of discriminating reliably between several different seismic recording classes can potentially release the human expert from the labor intensive classification task. An Interesting question concerning the frequency representation of the segmented seismic recordings is: if it is better to use only frequency information in the form of a single spectrum or to use a time frequency representation such as a spectrogram. Furthermore it is of interest to see if the ordering of the spectral frames inside the resulting spectrograms is of importance. In this study a justified spectrogram representation is developed for the segmented recordings from the Nevado del Ruiz volcano. Using this spectrogram representation we also look at five different classification strategies in combination with a large number of different classifiers. Often seismic events such as volcanic tectonic earthquakes, tectonic earthquakes, rockfall etc... are registered by several seismic stations. It is of interest to see if the recordings of multiple stations can be combined to improve classification results. Furthermore it is of interest to see how well the untrained and trained classifier systems generalize to the recordings of other stations.

Since the early days of exploration seismology seismic receiver field arrays have been employed for purposes such as suppressing high wavenumbers present in wavefield, reducing the volume of recorded data and improving its signal-to-noise ratio. These receiver field arrays rely on the receivers being placed on a predetermined geometric layout, a condition which is not always met in the field. Misplacing receivers can have a detrimental effect on the performance of the field array. Fortunately, advances in seismic acquisition now enable a) recording the output of individual receivers and b) knowing with high (but limited) accuracy the actual location of each receiver. It is possible then to form arrays digitally on a computer, a process known as group forming. Group forming can be viewed as a combination of filtering and resampling. We propose two algorithms that take advantage of the positional information available about the receivers and generate a linear space-varying (LSV) filter. The LSV filter is suitable for filtering the nonuniformly sampled data, generating the filtered output on the nominal grid. We examine the relation of our algorithms with other algorithms from the bibliography and investigate their performance on synthetic data.

Fluvio-deltaic sedimentary systems are of great interest for explorationists because they can form prolific hydrocarbon plays. However, they are also among the most complex and heterogeneous ones encountered in the subsurface. Reservoirs in clinoform systems are difficult to characterize because they show two main types of complexity: complex sedimentology and poor seismic imaging. The former is due to complex internal architecture with many small sedimentary elements often at a sub-seismic scale. Poor seismic imaging occurs because the internal layers of the clinoform often do not differ much in their acoustic properties, in addition, they have thicknesses that are below the vertical resolution of seismic data, and therefore such features do not show up very well on seismic images. Obviously, the most unfavorable situation occurs when both conditions interact. The static model of a fluvio-deltaic (clinoform) reservoir is extremely important because it plays a critical role in the field development planning. There are many ways to build a static model but the most effective way is by integrating seismic and well data through the construction of an acoustic impedance model by inversion of seismic data within a sequence stratigraphic framework. There are several reasons to integrate well-log data into the inverse process in the reservoir characterization workflow, such as the integration of different sources of information in a common earth model, the estimation of the seismic distortion (also known as the wavelet filter), etc. In addition, the low vertical resolution of seismic data is an important motivation to integrate well-log information into the inverse process and thereby complement the relatively dense horizontal coverage of seismic data with high-resolution borehole data (Van Riel and Mesdag, 1988; Van Riel and Pendrel, 2000; Van Riel, 2000; Bosch et al., 2009). Another potential benefit of seismic inversion is the ability to incorporate structural and stratigraphic information of the reservoir in order to differentiate between similar mathematical solutions on the basis of their geological viability. We present two different inversion approaches for poststack, time migrated seismic data and apply them to a clinoform sequence in the North Sea. Both inversion methods are not fully 2D, but more than a series of independently processed 1D inversions. To stress the enforced continuity along the geological structure, we use the name pseudo 2D inversion. The methods use well data as a priori constraints but differ in the way they incorporate structural information. One method uses a discrete layer model from the well that is then propagated laterally along the clinoform layers, which are modeled as sigmoids. The second method uses a constant sampling rate from the well data and employs horizontal and vertical regularization parameters for lateral propagation. Both methods obtain an acoustic impedance image with a high level of detail. The first method has a low level of parameterization embedded in a geological framework and is computationally fast. The second method has a much higher degree of parameterization but is flexible enough to detect deviations in the geological settings of the reservoir, however there is no explicit geological significance and it is computationally much less efficient. Forward seismic modeling of the two inversion results indicates a good match of both methods with the actual seismic data. The methods are especially considered to be useful when seismic data alone do not reveal the actual detailed reservoir architecture, which can be the case either because of their low vertical resolution or exceedingly thin layering.

Seismic Interferometry (SI) with ambient noise is a recently developed method for subsurface imaging that involves cross-correlating ambient noise from passive seismic data to reveal the earth’s reflection response. Based on this technique, TNO aims at developing a methodology for efficient monitoring of CO2 geological sequestration in Carbon Capture and Storage projects. In 2009, as a collaboration with the CO2SINK research project on CO2 sequestration taking place in Ketzin, Germany, TNO installed a permanent seismic array that has now already been recording passive data for 2 years during the CO2-injection. The correlation of the large real datasets requires a prior assessment on the feasibility of SI with ambient noise for time-lapse monitoring of the C02 storage process. Using a 1D layered acoustic velocity model of the Ketzin storage site as well as the layout of the TNO’s permanent array, we perform passive experiments that involve finite-difference modelling of passive wavefields from random noise sources and the cross-correlation of these passive synthetic data to retrieve reflection responses. By repeating the experiment for base and monitor velocity scenarios from CO2 saturation changes in the reservoir, we study the influence of various noise parameters (the number of sources, the sources’ duration, the spatial distribution) on the retrieved base and monitor reflection responses as well as their induced amplitude differences. We show that, for a sufficient number of sources with sufficient durations, the accuracy of the retrieved reflection responses from the CO2 reservoir are close enough to the reference responses from an active surface shot. More generally, the retrieved reflections are improved when more sources lie at or close to stationary points. In addition, when the sources characteristics are identical for both the base and the monitor passive experiments, the amplitude differences between the retrieved base and monitor reflection responses exhibit clear information on reflection changes from the reservoir. However we show that, when the sources’ characteristics, such as their spatial random distribution only, are not close enough for the base and the monitor passive experiments, the induced amplitude differences are difficult to interpret. Processing of the recorded body-wave noise prior to cross-correlation might be then necessary. We conclude that the repeatability of the passive noise is crucial for time-lapse interpretation. Finally, we choose an ideal noise configuration to enhance repeatability of the passive experiments when using different random distributions of the sources. In this special case, plotting picked amplitudes from the retrieved reflection responses for different velocity scenarios illustrates the potential of monitoring the velocity changes from SI with ambient noise.

In the northern part of the Netherlands the (magnesium containing) salt minerals carnallite and bischofite are extracted using solution mining methods, such as squeeze mining. As a consequence subsidence up to the surface is taking place, potentially causing harm to the environment. Given this fact the local government allows extraction of this salt under the condition that subsidence at the surface does not exceed a pre-defined limit. This makes the importance of salt mining induced subsidence evident. For an accurate prediction the 3-D characteristics of the caverns are important. Currently it is impossible to predict what effect squeeze mining has on the characteristics of the caverns, and thus on the subsidence. This thesis addresses the question whether time-lapse seismic reflection techniques can be used to image and quantify the effects of mining magnesium salt in the north of The Netherlands. The use of seismic time-lapse techniques to indentify produced salt zones has not been investigated before and this study must be considered as a feasibility study, using synthetic seismic data. The questions addressed in this thesis are: • Can the effects of solution salt mining be detected in seismic time-lapse mode? • If so, can these effects be quantified? In our approach we studied the time-lapse effects of different scenarios; representing a vertical and a lateral extension of the mine due to salt production have been evaluated. These scenarios have been transformed into different subsurface models that were an input to an acoustic and elastic finite difference scheme in order to create synthetic data. The geometric and material properties in the scenarios are based on the interpretation of real seismic data. A combination of well data and empirical relations has been used to derive the necessary seismic parameters. The main findings can be summarized as: 1) A seismic reflection of the salt mine is visible in seismic shot records, CMP-gathers and migrated sections. The exact geometry of the mine cannot be distinguished in the data, because of interference effects. 2) To derive time-shifts and amplitude changes caused by geometry and property changes in seismic time-lapse mode, 2-D cross correlation on migrated data was used. This technique allows deriving a horizontal shift as well as a time-shift. The amplitude changes were calculated by comparing the amplitude maximum from a 2-D cross correlation window with the maximum amplitude from a 2-D auto correlation window. The difference is expressed as a percentage. 3) A vertical extension of 5 m causes a potentially detectable time shift of 1.5 ms for the acoustic case and 2.0 ms for the elastic case. The amplitude changes are respectively 5.3% and 7.1%. For a purely lateral extension of 100 m of the caverns no time shift is found for the acoustic and elastic case and the amplitude change is 0.2% and 2.0% respectively. These results show that the amplitude change caused by a vertical extension is significantly higher than the one caused by a lateral extension of the mine. In order to make lateral changes of the salt mine visible one could opt for 1-D cross correlation. The time shifts and amplitudes found are comparable as those found in literature for the oil and gas industry. The final conclusion yields that the effects of solution salt mining can be detected and quantified in seismic time-lapse mode. Some effects in this seismic study are large enough to be seen in real seismic data. It is therefore feasible to use time-lapse seismic to monitor geometric and material changes of an underground solution salt mine.

Seismic vibrators are used to investigate the structure of the subsurface in the framework of soil and gas exploration, also to locate faults and rupture zones in earthquake investigations. Vibrators introduce seismic waves that propagate through the ground and are received by seismic sensors at some distance from the source. This enables interpretation of the sub-soil structure by inverse analysis. To analyze the behavior of the soil below a seismic vibrator, analytical models were developed in the past in which the behavior of the soil is simplified by linear elasticity. This assumption of linear elastic soil model does not describe real soil behaviors. In this research, a seismic vibrator standing on a homogeneous soil is simulated by means of an axisymmetric finite element model with an advanced soil model, i.e. Hardening Soil model with smallstrain stiffness, called HS-small model in which the non-linear, irreversible soil behaviors as well as small-strain stiffness effects below a seismic vibrator are taken into account. The sign and amplitude of displacements of the ground surface in the near field and the far field are observed for a homogeneous sand and homogeneous clay. A comparison between the linear elastic model and the HS-small model in simulation the soil response is made. The response of the sand and clay under delta-pulse load, minimum-phase-wavelet load, chirp load, and harmonic load with different frequencies (5Hz and 50Hz) and amplitudes (10kN/m2 and 20kN/m2) is simulated and evaluated. The influence of stress-strain dependent soil stiffness and porewater on seismic wave velocities is evaluated by means of seismograms.

Fractures and faults play an important role in controlling the flow and transport properties in a reservoir and that is the main reason for which their characterization is very important in hydrocarbon exploration. In addition to this, the possibility of characterizing fractures can represent a great advantage in other fields, like geothermal exploration, hydrofracturing applications or volcanic risk evaluation. Seismic simulation by finite-difference modeling has been implemented as a tool to characterize fracture networks by testing their seismic signatures. The medium in which fractures are placed is represented by a squared model with the sides measuring 5000m; two receivers arrays, each including 1000 receivers, have been implemented: the first at the top of the model (to record the reflected wavefield), the second at a depth of 4000m (to record the transmitted wavefield). Fractures are randomly positioned in the center of the model, between 1250m and 3750m. In the case of a layered medium, only the middle layer, 1500m thick, contains random fractures. The sensitivity of seismic wave propagation to fractures has been analyzed by testing the influence of different fracture features: length, orientation, density (number of fractures). Even a single fracture affects the incident seismic signal, producing diffracted/scattered and transmitted waves. Moreover, its orientation significantly affects the reflected wavefield in the time domain in terms of amplitude and complexity of the response and in the frequency domain as well, where peak amplitude and peak frequency change depending on the fractures orientation. More than the orientation, the fracture length strongly affects the seismic signal in time and frequency domains. Considering a network of fractures, the imprint of the fracture orientation on the reflected wavefield is significant only in the frequency-wavenumber domain; on the other hand, it is much stronger for the transmitted wavefield in the frequency domain, where the peak amplitude and the peak frequency undergo high variation: in particular, the horizontal fractures produce the strongest frequencies attenuation and the lowest peak frequency. The fracture length variation produces the most interesting signature in the time domain for the reflected wavefield (an increase in the fracture length produces longer coda waves) and in the frequency domain for the transmitted field: in particular, the longest fractures produce the strongest frequencies attenuation and the lowest peak frequency. Significant is the signature of the fracture density, but it is particularly strong on the transmitted wavefield in the frequency domain: in particular, the highest number of fractures produce the same effect of the horizontal and longest fractures (strongest frequencies attenuation and lowest peak frequency). Eventually, the introduction of a fractured layer yields strong change in the incident signal, which is highly attenuated and disrupted. The results of the present work can have implications in all the fields where the detection and the characterization of fractures in the subsurface is vital, such as the geothermal exploration or the hydraulic fracturing applications. Despite of the complexity of a fractured systems, some recurrent trends and characteristic responses have been determined. However, it should kept in mind that fracture features are strictly related (they influence each other) and that different features can yield similar responses: this means that fracture characteristics can not be straightforward inferred from their seismic signatures.

Most of our knowledge of the subsurface comes from the measurement of quantities that are indirectly related to the earth’s structure. Examples are seismic waves, gravity and electromagnetic waves. We consider the use of seismic waves for inference of structural information on an exploration scale. The seismic waves are generated by placing sources at or near the surface and the response is measured with either geophones or hydrophones. From such data, valuable information on the earth’s structure may be inferred up to roughly 10 km depth with a resolution of 50 – 100 m. While such data where interpreted directly in the early days, present-day processing involve detailed mathematical description of wave propagation in terms of density and sound speeds. With the aid of such a mathematical description, the actual seismic experiment may be simulated on a computer. The parameters (density, sound speed) that best describe the subsurface can then be obtained by trying to match the observed data to the simulated data. Such a problem is called an inverse problem. A straightforward way to solve this is to try to fit the data in a least-squares sense. However, due to the specifics of the seismic experiment this only works well for parameters that govern the dynamics of the data. The kinematics are, however, also very important as these are sensitive to large scale velocity changes. To successfully invert the kinematics of the data, the inverse problem needs to be reformulated. In this thesis we investigate such a reformulation. The basic ingredient of this reformulation is the way the mismatch between the data are measured. Instead of subtracting the data – as in the least-squares approach – we infer the kinematic difference (the timeshift, basically) between the simulated and observed data via a weighted norm of the correlation. We discuss the application of this basic principle to both reflection and transmission data. We find that the weighted norm of the correlation is a robust way to measure the shift between complex waveforms and can be successfully applied to cross-well tomography and velocity inversion for horizontally layered media. Preliminary results for non-layered media suggest that the approach is applicable in this case as well.

Complexities of the near surface and the varying acquisition-surface topography have degrading effects on land seismic images. This is because of trapped Rayleigh waves in the near-surface, and the distortions occurring to the body waves while passing through the complex near-surface. Removal of these two effects depends on the acquisition design and the preprocessing method. The acquisition design proposed in this thesis includes two phases. The first phase involves selecting initial parameters based on wave sampling theory. In the second phase, the subsurface model is incorporated. The main objective of this phase is to evaluate and update the initial design parameters. In this thesis the evaluation is based on the quality of pre-stack imaging results. These results are obtained using the concept of focal beams which allows assessing the effects of an acquisition geometry on the image quality without going into explicit modelling of seismic data. The results obtained in this thesis demonstrate that the complex near-surface causes the seismic wavefronts to be distorted without causing major shadow zones. This is principally different for the situation where the complexities of the underlying macro-model are relatively close to the target. The proposed method of removing the signal distortions is based on wavefield redatuming. This is carried out using one-way focusing operators that are obtained without knowledge of the underlying macro-model. This thesis introduces the estimation of the 3D one-way focusing operators using cross-spreads without a need for trace interpolation. In addition, a method for estimating 2D focusing operators in one-step is introduced and applied on a complex 2D synthetic and field data set. This method is based on global non-parametric inversion of tracked two-way reflection traveltimes in the common source gathers, using Fermat's principle as a constraint.

Full waveform inversion is a tool to obtain high-resolution property models of the subsurface from seismic data. However, the technique is computationally expens- ive and so far no multi-dimensional implementation exists to achieve a resolution that can directly be used for seismic interpretation and characterisation. In this thesis we discuss a method to overcome some of the current limitations of seismic full waveform inversion. The scheme consists of alternating local linear inversions followed by global nonlinear total field estimations. By backpropagating sources and receivers from the acquisition surface into the subsurface, the inversion process can be localised. Inversion of backprogated data to obtain the properties in a local domain only, is a much smaller problem com- pared to direct inversion of surface data, aiming at the properties of the full subsurface at once. As a result we can use the full frequency spectrum, up to 55 Hz and beyond, to obtain property models with a resolution that is compar- able to the one obtained by conventional structural imaging. A high-resolution global subsurface model in terms of properties is then obtained by combining many of these local results. Because all local inversions are fully independent, the scheme is extremely suitable for parallelisation. In the first iteration we assume that the wavefields are propagating in a smooth background model only. Obtaining a high-resolution property model by localised inversion, showing structural features of the subsurface, tells us that wavefield propagation in a smooth background is not a good approximation. This is why in a next step we perform a total wavefield estimation based on the latest combined local inversion results. In this way the full nonlinear relationship between the measured seismic data and the subsurface properties is utilised. We show how nonlinear wavefield effects as multiple scattering, transmission and true travel- times in the inverted medium, can properly be estimated and how they can be used to obtain significantly improved subsurface models. Examples are given where nonlinear full waveform inversion allows us to recover structural informa- tion, particularly very steep dipping geology, that can never be imaged by using a linear data model. It is demonstrated that by using the nonlinear relationship between the measured data and the obtained subsurface properties, the resolution can be increased even further. Full waveform inversion as proposed in this thesis, allows reconstruction of subsurface models with a spatial resolution that goes beyond the equivalent temporal frequencies in the measured seismic data. Consequently, estimation of absolute quantitative properties from a bandlimited seismic signal becomes pos- sible. It is discussed that every scatterer in the subsurface introduces a coda that will have an effect on the imaging of the deeper subsurface. Especially the near- surface can introduce nonlinearity, particularly in terms of multiple scattering, which makes imaging the deeper area of interest, e.g. a reservoir, difficult. In this thesis we show how these internal multiples cannot only be handled, but also how they can contribute to the inversion process by introducing enhanced illumination. The method is first derived in 2-D under the acoustic approximation and assum- ing a constant and known density. Since current implementation of the scheme requires homogeneous background models as input to the inversion, in 2-D the method is demonstrated on synthetic datasets only. Field data application using a 2-D data model needs extension to inhomogeneous background models and to true elastic inversion, which is in principle straightforward, but has not been im- plemented yet. In this thesis we perform these extensions for the 1.5-D case, assuming a ho- rizontally layered subsurface per CMP location. Nonlinear elastic full waveform inversion, aiming at the recovery of three elastic parameters over a reservoir se- quence, is applied to synthetic data as well as to a field dataset from the Middle East. It is shown that nonlinear full waveform inversion is suitable for reservoir characterisation while increased resolution and better results in terms of quantit- ative properties are obtained.

We present fast and efficient plane-wave migration methods for densely sampled seismic data in both the source and receiver domains. The methods are based on slant stacking over both shot and receiver positions (or offsets) for all the recorded data. If the data-acquisition geometry permits, both inline and crossline source and receiver positions can be incorporated into a multidimensional phase-velocity space, which is regular even for randomly positioned input data. By noting the maximum time dips present in the shot and receiver gathers and constant-offset sections, the number of plane waves required can be estimated, and this generally results in a reduction of the data volume used for migration. The required traveltime computations for depth imaging are independent for each particular plane-wave component. It thus can be used for either the source or the receiver plane waves during extrapolation in phase space, reducing considerably the computational burden. Since only vertical delay times are required, many traveltime techniques can be employed, and the problems with multipathing and first arrivals are either reduced or eliminated. Further, the plane-wave integrals can be pruned to concentrate the image on selected targets. In this way, the computation time can be further reduced, and the technique lends itself naturally to a velocity-modeling scheme where, for example, horizontal and then steeply dipping events are gradually introduced into the velocity analysis. The migration method also lends itself to imaging in anisotropic media because phase space is the natural domain for such an analysis.

Seismic interferometry is a technique for estimating the Green's function that accounts for wave propagation between receivers by correlating the waves recorded at these receivers. We present a derivation of this principle based on the method of stationary phase. Although this derivation is intended to be educational, applicable to simple media only, it provides insight into the physical principle of seismic interferometry. In a homogeneous medium with one horizontal reflector and without a free surface, the correlation of the waves recorded at two receivers correctly gives both the direct wave and the singly reflected waves. When more reflectors are present, a product of the singly reflected waves occurs in the crosscorrelation that leads to spurious multiples when the waves are excited at the surface only. We give a heuristic argument that these spurious multiples disappear when sources below the reflectors are included. We also extend the derivation to a smoothly varying heterogeneous background medium.

The common-focus-point technology (CFP) describes prestack migration by focusing in two steps: emission and detection. The output of the first focusing step represents a CFP gather. This gather defines a shot record that represents the subsurface response resulting from a focused source wavefield. We propose applying the recursive shot-record, depth-migration algorithm to the CFP gathers of a seismic data volume and refer to this process as CFP-gather migration. In the situation of complex geology and/or low signal-to-noise ratio, CFP-based image gathers are easier to interpret for nonalignment than the conventional image gathers. This makes the CFP-based image gathers better suited for velocity analysis. This important property is illustrated by examples on the Marmousi model.

Seismic reflection imaging is a popular method to image, characterize and monitor the Earth's subsurface. In this method, seismic signals are sent into the subsurface and their reflections are collected. Strong heterogeneities in upper sections of the subsurface often pose a problem for imaging deeper sections. To overcome these problems, it has been proposed to place receivers in a horizontal, deviated or vertical borehole and to turn these receivers into virtual sources by seismic interferometry. In this thesis, the correlation-based formalism that undergirds seismic interferometry is replaced by multidimensional deconvolution, yielding several important advantages. It is shown that multidimensional deconvolution improves the radiation pattern of the generated virtual sources and that it removes undesired artifacts. A range of applications is being discussed, including the retrieval of signals from background noise, subsalt imaging and reservoir monitoring.