BR112021006744B1 - SEISMIC EXPLORATION METHOD ABOVE A SUBSURFACE REGION - Google Patents

Abstract

SEPARATION OF MULTIPLE SEISMIC SOURCES OF DIFFERENT TYPES BY INVERSION. A method of seismic exploration above a subsurface region containing structural or stratigraphic features conducive to the presence, migration or accumulation of hydrocarbons comprises accessing at least a portion of a mixed seismic source survey, separating the at least two interfering seismic source excitations using inversion separation, producing one or more source clusters based on the separation and using the one or more source clusters to exploit hydrocarbons within said subsurface region. The mixed source seismic survey contains at least two interfering seismic source excitations in it, and the seismic source excitations may be produced by seismic source types having different signatures or frequency characteristics.

Description

CROSS-REFERENCE TO RELATED ORDERS

[0001] This application claims benefit from Provisional Application Serial No. US 62/745,068 filed on October 12, 2018, and entitled “Separation of Multiple Seismic Sources of Different Types by Inversion”, which is incorporated herein by title reference in its entirety for all purposes.

STATEMENT IN RELATION TO RESEARCH OR DEVELOPMENT SPONSORED BY THE FEDERAL GOVERNMENT

[0002] Not applicable.

BACKGROUND

[0003] A seismic survey represents an attempt to image or map the earth's subsurface by sending sound energy below ground and recording the "echoes" that return from the rock layers below. The source of the downward sound energy can come, for example, from seismic blasts or vibrators on land, or air guns in marine environments. During a seismic survey, the energy source is placed at various locations close to the earth's surface above a geological structure of interest. Each time the source is activated, it generates a seismic signal that travels down through the earth, interacts with geological structures on Earth and upon return, is recorded in many locations on the surface. Multiple source/record combinations are then combined to create a nearly continuous subsurface profile that can extend for many kilometers. In a two-dimensional (2-D) seismic survey, registration locations are usually laid out along a single line, whereas in a three-dimensional (3-D) survey, registration locations are distributed across the surface in a grid pattern. In simpler terms, a 2-D seismic line can be thought of as providing a cross-sectional image (vertical slice) of the earth layers as they exist directly below the record locations. A 3-D survey produces a “cube” or volume of data that is, conceptually at least, a 3-D image of the subsurface that lies beneath the survey area. However, in reality, both 2-D and 3-D surveys interrogate some land volume lying below the area covered by the survey.

SUMMARY

[0004] In one embodiment, a method of seismic exploration above a subsurface region containing structural or stratigraphic features conducive to the presence, migration or accumulation of hydrocarbons comprises accessing at least a portion of a mixed seismic source survey, separating the at least two interfering seismic source excitations using inversion separation, producing a plurality of separate source clusters based on the separation, and using the plurality of source clusters to mine for hydrocarbons within said subsurface region. The mixed source seismic survey contains at least two interfering seismic source excitations therein, and the seismic source excitations are produced by at least two seismic sources having different signatures or frequency characteristics.

[0005] In one embodiment, a seismic exploration method comprises initiating a plurality of seismic source excitations above a subsurface region containing structural or stratigraphic features conducive to the presence, migration or accumulation of hydrocarbons, and producing a mixed seismic source survey comprising data obtained from reflections generated by the plurality of seismic source excitations. The plurality of seismic source excitations is produced by at least two seismic sources having different signatures or frequency characteristics, and the mixed source seismic survey contains at least two interfering seismic source excitations therein. The at least two interfering seismic source excitations are separated using inversion separation, and a plurality of source groupings are produced based on the separation of the at least two interfering seismic source excitations.

[0006] In one embodiment, a method of seismic exploration above a subsurface region containing structural or stratigraphic features conducive to the presence, migration or accumulation of hydrocarbons comprises accessing at least a portion of a mixed seismic source survey, separating the at least two interfering seismic source excitations using inversion separation, producing a plurality of separate source clusters based on the separation, and using the plurality of source clusters to mine for hydrocarbons within said subsurface region. The mixed source seismic survey contains at least two interfering seismic source excitations therein, and the seismic source excitations are produced by at least a first seismic source and a second seismic source having different frequency signatures or characteristics. At least one of the first seismic source or the second seismic source is a low frequency source, and at least one of the plurality of separate source arrays contains low frequency data, or data at other frequencies than conventional seismic sources.

[0007] In one embodiment, a method of seismic exploration above a subsurface region containing structural or stratigraphic features conducive to the presence, migration or accumulation of hydrocarbons comprises accessing at least a portion of a mixed seismic source survey and using a plurality of source clusters generated to exploit hydrocarbons within said subsurface region. The mixed source seismic survey contains at least two interfering seismic source excitations therein, and the seismic source excitations are produced by at least two seismic sources having different signatures or frequency characteristics.

[0008] Although multiple embodiments are disclosed, still other embodiments will become apparent to those skilled in the art from the following detailed description. As will be apparent, certain embodiments as disclosed herein are capable of modification in various respects without departing from the spirit and scope of the claims as set forth herein. Accordingly, the detailed description herein below is to be considered as illustrative in nature, not restrictive.

BRIEF DESCRIPTION OF THE DRAWINGS

[0009] The following figures illustrate embodiments of the matter disclosed in this document. The claimed matter can be understood by reference to the following description taken in conjunction with the attached Figures, in which:

[0010] FIG. 1 illustrates a continuous trace for a single receiver spanning multiple source points from a plurality of sources.

[0011] FIG. 2 illustrates aliased data for a first source.

[0012] FIG. 3 illustrates aliased data for a second source.

[0013] FIG. 4 illustrates data from one of the initial iterations for the first source after a smoothness constraint is applied.

[0014] FIG. 5 illustrates data from one of the initial iterations for the second source after a smoothness constraint is applied.

[0015] FIG. 6 illustrates the partially unmerged data for the first source.

[0016] FIG. 7 illustrates the conventional result for the first source for a simulation in which no interference from the second source occurred.

[0017] FIG. 8 illustrates the general background of the present disclosure.

[0018] FIG. 9 illustrates a seismic processing sequence suitable for use with the present disclosure.

[0019] FIG. 10 contains a schematic plan view of a typical merged survey source.

[0020] FIG. 11 schematically illustrates how different recordings can be identified and extracted within the merged source survey.

[0021] FIG. 12 illustrates exemplary operating logic for a representative case where source 1 is impulsive and source 2 is vibratory, with a variable source signature.

[0022] FIG. 13 illustrates an exemplary method for extracting coherent signals from a noisy array.

[0023] FIG. 14 illustrates an exemplary method for replacing “missing” or “tainted” features in a cluster using POCS interpolation.

[0024] Fig 15 shows a diagram of a common receiver array with three interfering seismic sources, each source operating on a uniformly different periodic operating schedule.

DETAILED DESCRIPTION

[0025] Otherwise, unless otherwise specified, any use in any form of the terms “connect”, “engage”, “couple”, “fix” or any other term that describes an interaction between elements is not intended to limit interaction to direct interaction between elements and can also include indirect interaction between described elements. In the following discussion and claims, the terms "including" and "comprising" are used in an open-ended manner and, as such, are to be interpreted to mean "including, but not limited to. . ”. The various features mentioned above, as well as other features and characteristics described in greater detail below, will be readily apparent to those skilled in the art with the aid of the present disclosure upon reading the following detailed description of embodiments, and upon reference to the accompanying drawings.

[0026] A seismic survey is composed of a very large number of individual seismic records or traces. In a typical 2-D survey, there will usually be several tens of thousands of traces, while in a 3-D survey, the number of individual traces can reach multiple millions of traces. Chapter 1, pages 9 to 89, of Seismic Data Processing by Ozdogan Yilmaz, Society of Exploration Geophysicists, 1987, contains general information related to conventional 2-D processing. General background information pertaining to 3-D data acquisition and processing can be found in Chapter 6, pages 384 to 427, of Yilmaz.

[0027] Traditionally, a seismic trace is a digital record of acoustic energy reflecting from inhomogeneities or discontinuities in the subsurface, in which a partial reflection can occur whenever there is a change in the elastic properties of subsurface materials (“primary reflections”). Increasingly, other types of acoustic feedback energy are also being used in geophysics. In addition to primary reflections, acoustic energy, for example, can convert from one wave mode to another, refract along layer boundaries, diffract from scatterers, or be turned back up by velocity inhomogeneities (“waves of division"). Waves can interact with subsurface structures multiple times within the Earth before returning to the surface to be recorded. Depending on the geophysical application, certain waves are considered a “signal”, and certain waves may be considered “noise”.

[0028] Digital samples are usually acquired at intervals of 0.002 second (2 milliseconds or “ms”), although 4 millisecond and 1 millisecond sampling intervals are also common. Each discrete sample in a conventional digital seismic trace is associated with a travel time and, in the case of reflected energy, a round trip time from the source to the reflector and back to the surface again, assuming, of course, that the source and receiver are both located on the surface. Many variations of the conventional source-receiver arrangement are used in practice, for example VSP (vertical seismic profiles) surveys, ocean floor surveys, etc. Furthermore, the surface location of each trace in a seismic survey is carefully tracked and is usually made a part of the trace itself (as part of the trace header information). This allows the seismic information contained within traces to be further correlated with specific surface and subsurface locations, thereby providing a means for posting and contouring seismic data — and attributes extracted therefrom — on a map (e.g., “mapping ”).

[0029] The data in a 3-D survey can be viewed in many different ways. First, horizontal “constant time slices” can be extracted from a stacked or non-stacked seismic volume by collecting all digital samples that occur in the same travel time. This operation results in a horizontal 2-D plane of seismic data. By animating a series of 2-D planes, it is possible for the interpreter to pan the volume, giving the impression that successive layers are being removed, so that the information located below can be observed. Similarly, a vertical plane of seismic data can be taken at an arbitrary azimuth across the volume by collecting and displaying seismic traces lying along a particular line. This operation effectively extracts an individual 2-D seismic line from within the 3-D data volume. It should also be noted that a 3-D dataset can be thought of as being made up of a 5-D dataset that has been reduced in dimensionality by stacking it into a 3-D image. Dimensions are typically time (or “z” depth), “x” (e.g. North-South), “y” (e.g. East-West), source-receiver offset in the x direction, and source-receiver offset in the y direction. While the examples here may focus on the 2-D and 3-D cases, extending the process to four, five, or more dimensions is straightforward.

[0030] Seismic data that has been properly acquired and processed can provide a wealth of information to the explorer, one of the individuals within an oil company whose job it is to locate potential drilling sites. For example, a seismic profile provides the explorer with a broad view of the subsurface structure of rock layers and often reveals important features associated with the trapping and storage of hydrocarbons such as faults, folds, anticlines, nonconformities, and subsurface salt domes and reefs, among others. many others. During computer processing of seismic data, estimates of subsurface rock velocities are routinely generated and near-surface inhomogeneities are detected and displayed. In some cases, seismic data can be used to directly estimate rock porosity, water saturation, and hydrocarbon content. Less obviously, seismic waveform attributes such as phase, peak amplitude, peak-to-trough ratio and a host of other attributes can often be empirically correlated with known hydrocarbon occurrences and such correlation applied to collected seismic data against new ones. exploration targets.

[0031] Obviously, a well-known problem with seismic data is that they are relatively expensive to acquire. In fact, in some cases, the cost of surveying can determine whether or not the profitability of the proposed target is favorable. Therefore, techniques that tend to reduce the cost of each survey are useful. Closely spaced shots from two or more sources can be a strategy to reduce the cost of seismic data acquisition. The idea behind this approach is that a receiver line or field will be deployed and such one or more sources will be activated during a single registration period. Therefore, subsurface reflections from a source excitation can be mixed with those that originated later, ie a “blended source” survey is acquired. This is noted to be in stark contrast to conventional surveying techniques, where subsurface return waves from one source could not overlap return waves from another.

[0032] Merging can happen as a single source is activated again before the Earth return wave current caused by a source activation has had time to deteriorate (called “self-simultaneous” acquisition). It can also happen because multiple sources are activated in the same time window, and at some ranges and offset times, their return wave currents overlap. Both methods can be used at the same time in the same acquisition.

[0033] Seismic acquisition with two or more sources of the same type is possible in the field for both terrestrial and marine acquisition. For example, two air gun boats operating simultaneously can be used as a marine acquisition mode. In some methods, the two boats generally move in a synchronized manner, and generally also fire in a carefully choreographed and coordinated manner ("Distance Separated Simultaneous Source" or "Appearance" methods). In other methods, the sources move independently and operate independently (conventional “ISS” methods). However, the methods in use up to the present time assume that the sources to be separated are of the same type, and after separation, the return signals from all sources are combined into a common data space where the distinction as to which source produced a given bit of seismic energy in the data does not matter for subsequent processing. Return signals are conventionally reflections, but increasingly other types of return signals are also used, e.g. splitting waves, refractions, diffractions, multi-scatter reflections, etc.

[0034] Different types of sources have different advantages, and the industry moves towards heterogeneous source procurement to gain the benefits of having different types of sources in a single survey. To save time and expense, the fountains can operate simultaneously. For example, a marine survey might use air guns to cover 2 to 80 Hz, spark gaps to cover 70 to 120 Hz, and a low frequency vibrator to cover 1.5 to 3 Hz, and all these different sources could operate simultaneously and be recorded on the same receivers.

[0035] Certain embodiments may attempt to combine all of these recorded sources into a single high-bandwidth dataset, simulating a single-source survey that has the combined advantages of all different types of sources. Sources cannot be treated as “identical”, however, and sources are usually matched to each other first so that they fit into a combined dataset. In other cases, the types of sources may be sufficiently different, or used for sufficiently different purposes, that a clear distinction must be maintained between them so that the resulting datasets can be processed separately. Certain modalities can also do both at the same time, combining some of the overlapping fonts while leaving others distinct. For example, certain modalities may combine air gun and spark gap data into a common 2 to 120 Hz dataset used for both velocity model building and imaging, but leave the 1.5 to 3 Hz vibrator data low frequency signals to be used only for the initial stages of velocity model building. In areas of complex geology, velocity model building can be a prerequisite for successful imaging; imaging algorithms require a good model of Earth velocity to produce a good result. The same data can also be combined differently depending on the application.

[0036] It also frequently occurs that if two or more seismic acquisitions (which are intended to be separated) are operated close enough to each other in time and space, then sources from one survey are also recorded in the other. This “seismic interference” is conventionally treated as unwanted noise. However, using the methods described in this document, the “interference” can instead be unmerged and removed, or unmerged and separated as a distinct ultra-long displacement dataset, or even treated as an additional welcome and included signal. as another source in the survey.

[0037] In view of the foregoing, one or more embodiments are directed to a system and method for separating multiple seismic sources, in which the seismic sources may have different source signatures, and in which the multiple seismic sources were activated during a record single seismic. In contrast to prior approaches, one or more embodiments of the present systems and methods allow a user to separate sources, in which the seismic waves returned from the separate sources have been acquired superimposed in time. As such, one or more embodiments may obtain two or more separate source data sets that may each have different source information. Using more than one set of sources will allow seismic surveys to be acquired faster if the recorded seismic energy resulting from the different sources can be separated. This method can be especially effective with acquisition of wide azimuth reflection surveys, as the kinematics of the reflectors will be significantly different for different footage, which allows the continuity of reflections to be used in multiple dimensions. Furthermore, the ability to obtain different sets of source data allows different types of information to be obtained from the survey.

[0038] Briefly, the systems and methods of one or more modalities use an inversion-type method to separate information that is captured by a seismic record into datasets from multiple sources. The seismic record is a record of reflections originating from multiple seismic sources (eg Vibroseis®, air guns, low frequency sources, etc.). The seismic record, which can record overlapping reflectors, can be separated into datasets from multiple sources. In some embodiments, sources will be started (eg, triggered) at random times while multiple receivers are registered. Each trigger occurrence can be called a recording. Random or pseudo-random delays between random shots tend to make interference between different sources incoherent while reflections associated with the same source create coherent events. In some embodiments, one or more of the sources may be continuous or semi-continuous, while the other sources start at random times. Separation can be achieved with a numerical inversion process that uses the signatures for each source, the start times of each recording of each source, and the coherence of reflection events between nearby recordings. A signature for a source can correspond to a distinct aspect of the source (eg, the scan performed for each recording, frequency characteristics, etc.). This method has the potential to allow seismic surveys to be acquired faster and less expensively, with better signal quality and bandwidth than has been possible until now.

[0039] Mathematically, in some embodiments, the recorded data can be represented as d=rm, where d is a representation of continuous recorded data containing overlapping signals, m is a representation of data traces that would have been recorded in a seismic experiment conventional one font at a time, where the data for each font activation is in a separate stroke, and r is a merge operator that sums the strokes in m using known font activation times to create the continuous merged data. The systems and methods of one or more modalities describe how to use the merged recorded data d to accurately estimate the desired unmerged data m, which can then be used for further processing, just as if the data had been recorded in separate non-merged surveys. merged. In the case of a continuous sound source, the source activation times can be specified at a convenient regular repetition interval, and r can contain appropriate tapering, so that consecutive “source points” add together to reconstitute the continuous source signal. .

[0040] As represented in a computer system, d will typically be represented as an array with dimensions of sample time and receiver number. In some embodiments, the time axis may be subdivided into fixed-length records. For example, continuous time can be subdivided into one-minute-long dashes by dividing the time axis into two dimensions, “minute” and “time sample within minute”. m will typically be a matrix with dimensions of time, source activation number, source type and receiver number. r will have dimensions of time (possibly subdivided into two dimensions), source type, source number, and receiver number. r will typically be very large but very sparsely populated, and as such r will typically be represented in software applications on a computer by a function call. Typically it will not be stored as a large static array. Instead, its elements can be calculated as needed from a source initiation time table.

[0041] In some embodiments, m may represent “a complete survey”, with a dash for every possible combination of source type, source position, and receiver. Source-receiver combinations that did not actually occur in the survey (and therefore are not represented in the recorded data d) will correspond to zeros in r. In this modality, m can contain features interpolated by the demerge process to fill in acquisition gaps. Interpolation could also be used to replace recorded traces for which the recording and/or receiver were so poor in their performance that the corresponding trace is best treated as "missing", or to replace traces contaminated by a burst of noise. transitional. In other embodiments, m may contain features only for the source-receiver combinations that were actually acquired in the survey. Intermediate cases are also possible, with m representing a “complete survey” for some source types, but only the source-receiver combinations that were actually acquired for other source types. In any case, the problem can be considered to be underdetermined, with more entries in m to solve than there are unique elements in d. Some assumptions may need to be made to troubleshoot this problem. The solution is to note that the recorded data should vary smoothly as a function of source and receiver position for a given source, and use this to constrain the m-estimated unmerged data, also called “model data”. For this to work, the data needs to be acquired in such a way as to ensure that arrivals from different sources are not consistent with each other, typically introducing time lag into source initiation times. The time dither information is contained in the r merge operator. To enforce smoothness, one or more modalities may add an additional constraint m ~ S m, where S is a matrix or operator that describes consistency criteria that are appropriate for each source. The coherence criterion, then, will ensure that the data from the correct sources separate properly at m. The solution would then solve the coupled pair of equations d = r m and m ~ S m. One or more embodiments may also apply a weighting term W and solve the pair of equations W d ~ W r m and m ~ S m, and that the weighting term W is a matrix, or more generally, an operator. This would allow one or more modalities to weight the importance of some receivers over others, balance the amplitudes of different types of sources, or weight different windows of time or frequency differently. Conventionally, these equations could be solved by either a least squares algorithm (e.g. an L2 algorithm) or a fast minimization algorithm (e.g. an L1 algorithm) using one of the many available algorithms that are made available for this purpose. . The disclosed systems and methods provide an alternative method of iterative solving that has been shown to produce superior results. In certain scenarios, there are two or more simultaneous overlapping sources of the same type, and the goal is to create an unmerged dataset equivalent to a traditional survey, where the same set of sources are triggered sequentially and do not overlap. With respect to S, S can be any operator that constrains events in nearby recordings to be similar or coherent. S is not limited to single-direction applications, but can be applied to the full dimensionality of the acquired data (e.g., 2-D, 3-D, 4-D, etc.), or any suitable lower-dimensional subset of the data . Furthermore, it should not be assumed from the previous equation that S is necessarily a linear operator. While in some embodiments S may be a linear operator, in other instances this variable will represent a non-linear operator or a linearized version of it.

[0042] In some embodiments, S represents a fast Fourier transform (“FFT”). For irregularly spaced recordings, S can best be calculated with discrete Fourier transforms or some other method that allows irregular sampling of the recordings. In some embodiments, S may include a 3-dimensional Fourier transform and a clipping operation to remove weak events, which tend to be noise, and pass strong events, which tend to be coherent signals. This allows S to improve the coherence of recorded seismic signals.

[0043] The method can extract time windows from d data that correspond to source activations followed by a listening time interval. This extraction of time windows can be called “d-data aliasing”. If there is no overlapping of data from separate font activations, then aliasing the data would produce m directly. This scenario corresponds to conventional single-source seismic processing. Mathematically, aliasing = rt d, where the aliasing operator rt is the transpose of the aliasing operator r. Where the fonts overlap, then, dashes in the jagged data could be considered to be contaminated with crosstalk from the other font (or fonts). However, due to the timing dithering of source activations, this crosstalk noise will generally be incoherent. The noise, therefore, can be smoothed out by applying S, which removes non-smooth incoherent energy, to create an improved estimate of the msmooth unmerged data. Mathematically, msmooth = S mserriiated = S rt d. The algorithm then calculates the merged seismic data corresponding to this msmooth, which is a rough estimate of the recorded data d.

[0044] The estimated data can be calculated as estimated = r msmooth. If the difference between the actual data “d” and “unestimated” is small enough, then the process has successfully determined a msmooth coherent merged dataset that is consistent with the recorded field data d, and the process is terminated. Otherwise, the method can then compute the data misfits dresiduai = d - estimated, and use the same to improve msmooth. This enhancement is done by performing the following steps. First, the method can update the estimated unmerged dataset: matualized = msmooth + rt dresidual. Then, the method can apply the coherence constraint to that updated estimate, to compute a new msmooth: S matualized ^ msmooth. Finally, the method can then return to the beginning of this paragraph and repeat these steps until the residual data is small enough. Determining what is considered small enough may be selected as criteria for the solution and/or may be selected based on an acceptable reduction or minimization in the residual data, and as a result, may vary from analysis to analysis. In some embodiments, the data residual may be zero, while in other embodiments, the data residual may be considered small enough when the values are less than a noise level in the data.

[0045] In practice, in each iteration of the cycle above, the method can save msmove from the previous iteration (msaivo) and the proposed updated version (mproposto) and then calculate a weighted average of these two to use as the new update: msmove = (1 — à) msaivo + à mproposal. The weighting factor À must be positive but not greater than one above the maximum blending factor, ie the maximum number of overlapping sources present in the data being processed. Every trace in m corresponds to a particular source-receiver pair, so the method can easily implement a weighting term by making À a function of source, receiver, displacement, time, etc. More generally, the weighted average could also be applied in a transformed domain, so À can also be a function of frequency, etc. For example, msmooth = FT-1[ (1 — À) FT(msaivo) + À FT(mproposed) ] , where FT and FT-1 denote Fourier transform and inverse Fourier transform, respectively. Weights could be applied to more than one domain if desired.

[0046] The method may include having source excitations that occur within the same record separated in time by random or pseudorandom time intervals. If recorded recordings are randomly spaced, each recorded recording may be less likely to register coherent energy from other recordings that occurred close in time to the recorded recording, therefore allowing the recordings to be more readily separable. Exciting the recordings at random times, when these recordings are corrected to their individual zero times they will have signals that are coherent from source point to source point, whereas interfering recordings will tend to be incoherent and can be separated by the inversion process taught. in this document. This strengthens the operation of consistency measures in the present separation process.

[0047] Preferably, the acquisition itself will also be designed to accentuate inconsistency between different sources. For example, in conventional positional marine seismic recordings, each source triggers when it reaches the next desired recording position. Desired recording positions are generally evenly spaced. For a vessel moving at a constant speed, recordings will therefore also occur with a regular time increment. As described earlier, for a merged survey, a dither is added to that time to ensure that each source is not consistent with the others. However, one or more modalities can still ensure, even without hesitation, the different sources would still not be coherent by designing the survey so that two vessels never move "in formation", i.e., with one vessel moving with a vector displacement. constant relative to each other. A simple way to ensure this is to assign different speeds to different vessels, so that even if two vessels happen to be moving in the same direction on parallel survey lines, their corresponding recordings would not fall in the same pattern. Different vessel speeds translate to different recording intervals, so recordings are not coherent even in the absence of hesitation.

[0048] The above description described how to apply the method to single-type sources, where the different sources are combined into a single volume of unmerged data m. With the advent of new types of seismic sources, such as Wolfspar®, a low frequency vibrating source, there is now a need for systems and methods that can process recorded data using simultaneous acquisition of sources of more than one type. Such systems and methods can process recorded data using simultaneous acquisition from sources of more than one type by implementing the following: (1) generalize the above algorithm for the case of multiple inputs and outputs, (2) modify the algorithm to balance different types of data , so that one font type does not dominate over others in inversion, (3) consider font signatures that can differ drastically in temporal extent, frequency content, etc., and can vary from source point to source point, and/or (4) fonts that may be consistent across different domains.

[0049] The new system and method of one or more embodiments can be illustrated working through a simple example: simultaneous acquisition of a plurality of different types of sources (such as, for example, two air gun sources and a frequency source low). The two air gun sources are of the same type and together form a single unified survey. The low frequency source is of a different type with a different acquisition plan and forms a separate survey. In this example, the method still records a single combined dataset d as before, but now, this method is directed at determining two different m's, one for the air gun survey, and one for the low frequency source survey.

[0050] Mathematically, the method now has d = rA mA + rw mw, where rA is the appropriate blend matrix for the mA air gun model data, and rw is the appropriate blend matrix for the model data mw low frequency source. The objective is to find mA and mw, which produce the recorded data d and also satisfy mA ~ SA mA and mw ~ Sw mw, where SA and Sw are appropriate smoothness constraints that are appropriate for the corresponding source types and acquisition geometries.

[0051] As before, the method aliases the data. FIG. 1 shows a continuous trace for a single receiver spanning multiple source points from both the air guns and the low frequency vibrating source. The arrows point where the low frequency vibratory source can be clearly seen peaking between the higher amplitude airgun arrivals. The method then serrates the data for the air guns, computing mserriihadow = rtA d (yielding the result in FIG. 2), and for the low-frequency source, computing mserriihadow = rtw d (yielding the result in FIG. 3). ). The arrows in FIG. 2 point out particularly severe interference from other sources, with “A” pointing to interference from an air gun source, and “W” pointing to interference from a low frequency vibrating source. The high frequency “noise” visible in Figure 3 is interference from an air gun source. One or more modalities, then, can apply the smoothness constraints that are appropriate to each: msmoothA = SA mserriihadoA = SA rtA d (yielding the result in FIG. 4) and msmoothw = Sw mserriihadow = Sw rtw d (yielding the result in Fig. 5).

[0052] The estimated data can be calculated as estimated = FA msmoothA + rw msmooth. If the difference between the actual data “d” and “unestimated” is small enough, then the method has successfully found coherent merged datasets msuaveA and msuavew that are consistent with the recorded field data d, and the method is finished. Otherwise, the method calculates the data maladjustment dresiduai = d - estimated, and uses the same to improve msuaveA and msuavew. This is done by performing the following steps. First, the method updates the estimated unmerged datasets: matchedA = msmoothA + rtA dresiduai and matchedw = msmoothed + rtw dresiduai . The method then applies the consistency constraints to the updated estimates to compute a new msmoothA and msmooth: SA matualizedA ^ msmoothA and Sw matualizedw ”^ msmooth. Finally, the method can repeat these steps until the data residue is small enough and the method is complete. The method can be applied to more than two types of sources. As an alternative to getting the estimate as estimated = FA msuaveA + rw msuavew, the data estimate can be updated with rA msuaveA and rw msuavew in alternative iterations, which will separate the possible crosstalk between rA msuaveA and rw msuavew. This can also be extended to more font types.

[0053] FIG. 6 shows the resulting unmerged airgun data after several hundred iterations. After enough iterations it closely approximates the exact merged result shown in FIG. 7.

[0054] As in the single data type case, on each iteration of the above cycle, the method saves msuave from the previous iteration and the proposed updated version, and then calculates a weighted average of these two to use the new update: msuaveA = ( 1 — ÀA) msalvoA + ÀA mpropostoA, and mlsoftW = (1 — ÀW) msalvow + ÀW mpropostow. Different weighting factors can be used as needed to balance different source types, and as before, À can also be used to weight by source, source type, receiver, offset, time, etc. For example, if the air gun signals are at much higher amplitudes than the low frequency source signals, the method can start inverting with an ÀA greater than Àw, until the air gun signals converge, then, increase Aw.

[0055] In practice, air guns can operate at amplitudes much greater than the low frequency source (or sources) above about 5 Hz, however, the low frequency source (or sources) can operate at amplitudes much greater than air guns below about 2.5 Hz. Therefore, in some embodiments, weighting would be applied in the frequency domain, with the initial iterations of weighted inversion to emphasize updates of the strongest signals in each frequency band, then a Once the stronger signals stabilize, one or more modalities allow the weaker signals to catch up in later iterations.

[0056] Airguns are generally presumed to have a consistent and repeatable font signature, and much effort has been applied towards this end, however, this may not be true for all font types. The low frequency source, in particular, may not produce an identical signal every time it is activated. However, it continuously records how much water is being displaced and its fountain signature is therefore known in detail for each fountain activation. If the measured font signature is V, then the method can compute a stabilized font signature deconvolution operator using any of the standard techniques. For example, the method can use a traditional formula used in the Frequency domain to calculate a stable approximate inverse: V-1 = V* / (V V* + e2) , with the value of e chosen as needed to stabilize the inversion. If the font signature is varying from font activation to font activation, then the method needs to correct this variation before the method can apply the consistency constraint. Therefore, instead of updated SW W ^ msmoothW, the method would make V SW V 1 updatedW -^ msmoothW • Since the inverse V-1 is an approximation of the true inverse, there will be errors in the result of applying the inversion. However, errors will tend to be random and will be suppressed by the iterative nature of the solution.

[0057] For some choices of S and/or V, smoothing and/or signature matching just for the change in m in each iteration may prove advantageous, where the change can be called mdiff. If the method needs to fix the low frequency source signature but not the airgun source signature, for example, then the update step in the algorithm becomes mdiffA = rtA dresiduai and mdiffW = rtw dresiduai, followed by msuaveA + SA mdiffA "^ msoftA and msmooth + V Sw V-1 mdiffw ^ msoft. Depending on the choice of smoothing operator S and matching or signature operator V, this alternative form of the algorithm can produce a considerably better result. In particular, if the data residue is zero, the algorithm does not perform any further updates, regardless of S and V.

[0058] Although the example above illustrates two different types of fonts, other modalities can be directed to the simultaneous acquisition of more than two types of fonts (such as an “n” number of font types). Mathematically, the method now has d = ri mi + r2 m2 + ... + rn mn, where ri is the appropriate merge matrix for the first mi font type data, and rn is the appropriate merge matrix for the n ° source data type mn. The objective is to find mi, m2, . and mn that produce the recorded data d and also satisfy mi ~ Si mi, m2 ~ S2 m2, and mn ~ Sn mn, where Si, S2, and Sn are appropriate smoothness constraints that are appropriate for the corresponding font types and geometry. acquisition. Some modalities may also choose to treat certain types of noise as another “source”, and resolve it. For example, if the noise is due to surface waves propagating with a slower phase velocity than any seismic arrivals of interest, and the surface waves are properly sampled by the receiver grid (which will occur below a frequency low enough), then certain modalities may encounter a transformed domain, where surface wave data becomes sparse. With such information, certain modalities can then design a noise smoothness constraint that operates in such a domain and exploits such scarcity to accentuate surface wave noise and attenuate the desired seismic signals. If noise is recording generated, one or more modalities may know the recording times. If the noise is continuous and ambient, then certain modalities may use evenly spaced tapered overlapping acquisition time windows for the corresponding “recording times” in the blend/unblend matrices.

[0059] For a controllable source such as a terrestrial or marine vibrator, one or more modalities can deliberately introduce variation from one source activation to another during acquisition, and can also be a way to make different sources inconsistent, allowing them to be separate. One or more embodiments may use a variety of different earth vibrator sweep profiles for this purpose. The existing Wolfspar® low frequency source does not currently allow for this level of flexibility, but can change the signal or phase of your source signatures. A preferred embodiment might use sign/phase shifting instead of time dithering (or to increase time dithering) to break up unwanted coherence between sources.

[0060] One or more embodiments could also use V to not remove the font signature, but instead to match variable font activations to a “pattern” signature. If M is the measured (or estimated) font signatures, and D is the desired font signatures, then one or more modalities could be used, for example V = M D*/ (D D* + ei2) and V-1 = Vo * / (Vo Vo* + £22) , with Vo calculated from M and D exactly like V, but possibly with a different ei, eg e1 = 0. Alternatively, one or more modalities could use V-1 = D M */ (M M* + e22). These formulas would generally be applied in the frequency domain. Applying V-1 converts the actual source signature (M) to the desired source signature (D), which would introduce noise if the desired signature contains frequencies that are not present in the source signature, or present only at a level not sufficiently above the noise. The e2 stabilization parameter limits how much amplification will be applied, and stabilizes the result. V converts the desired (D) font signature back to the desired (M), which will generally be a more stable operation. Typically, one or more modalities would therefore choose to have more stabilization for the inverse operator than the direct operator, i.e., one or more modalities would typically have e2 > ei > 0.

[0061] Note, in the case of marine acquisition, one or more modalities could choose that the desired font signature D does not include the font ghosting effects, but includes the font ghosting in M and, therefore, would use the signature corresponding to deghosting the data. For this to work well, one or more modalities must ensure during acquisition that there is variety in the phantom notches, ie the source depth is variable between recordings. Alternatively, one or more embodiments can ensure that the elements in the source array are not all at the same depth.

[0062] There may be a reason to perform signature mapping even for fonts of a single type. Decades of work have gone into producing consistent and predictable air gun source signatures within its core frequency band, ie around 5 to 40 Hz. Outside this frequency band, less work has been done, and the air guns proved to be much more variable in their source signatures at these extended frequencies. Therefore, it may make sense to treat the port and starboard air gun arrays of a single boat acquisition as “two separate sources”, for example. The two resulting datasets could then be matched against each other to quantify differences in font signatures between what ideally should have been identical fonts. Or, the same air gun array at different times can be thought of as “multiple separate sources”, to capture and correct for unwanted secular variation from a single source.

[0063] One or more embodiments could also separate a source or source type into two or more by frequency. To continue the previous example, one or more modalities can use bandpass filtering to divide the data into two frequency bands, below and above 5 Hz, for example, (with some overlap between them), so that d = down + up. One or more modalities, then, could solve bottom and top as two separate problems. For example, if two airgun arrays are believed to behave identically at higher frequencies but may differ below 5 Hz, then one or more modalities could solve the two different airgun arrays separately for below but below 5 Hz. , as a unique font type for dalto.

[0064] Measuring the variable air gun source signature in the field, for example from hydrophone (or hydrophones) placed near the air gun array, one or more embodiments may also incorporate such knowledge into the process and use the measured signatures M correct the variable airgun source signatures to a desired constant signature D, so that one or more modalities would suffice for a vibrating source with a variable but measured signature.

[0065] While it may be preferable to match different source types with the same source signature before extracting coherent energy, such matching can create problems where the signal-to-noise (S/N) ratio is unsatisfactory for one of the source types . For example, a low frequency source like Wolfspar® can be designed to supplement air guns at low frequencies, and to produce minimal signal at higher frequencies. Therefore, above about 4 Hz, the air guns dominate over the low frequency source, while below 2 Hz, the low frequency source dominates over the air guns. Matching the low frequency source with air guns could create significant noise in the low frequency source signal above 4 Hz, and in the air guns below 2 Hz. The combined problem is unsatisfactorily constrained, as the two source types different frequencies generally overlap significantly over a narrow frequency range of about 2 to 4 Hz. Therefore, one or more modalities can solve the problem for these two different frequency ranges separately. In particular, one embodiment may initially treat the low frequency source above 5 Hz as "noise" and not include it in the higher frequency band inversion. Once the two air gun sources have been merged, then the low frequency source could be introduced, and the AA set to be very small, so that the air gun sources can only make very small updates. For the lower frequency band, the reverse strategy could be employed, solving for the low frequency source first and then adding air guns later. Finally, one modality may end up doing a full bandwidth flip. To generalize the concept to all types of seismic sources, some embodiments may define effective frequency bands that have an acceptable S/N ratio for all sources, and apply very small weighting to the other frequency bands to avoid cross-contamination. between different sources.

[0066] Some embodiments may have a range of source types, each optimized to cover a different frequency band, so that different source types can be combined to simulate a single ultra-wideband source. One more embodiment, then, can design a desired ultra-wideband source signature and partition that signature between the optimal frequency ranges of different source types. So, for example, if D is the desired wideband source signature, and if three source types are spanning the total frequency band, one or more embodiments can divide the source signature into overlapping frequency windows, so that D = Dlow + Dmid + Dal, with the frequency bands chosen to cover the ideal S/N frequency bands of each type of source. The selection of frequency windows can vary from analysis to analysis, as the selection can depend on the frequency content of the seismic sources. For example, a low frequency source might result in a low frequency range being about 0 to 2.5 Hz, an intermediate frequency range being about 2.5 to 5 Hz, and a high frequency range being about from 5 Hz to the Nyquist frequency. Depending on the model and particular seismic sources used, other frequency ranges can be defined and used for the source signature ranges. One or more embodiments could then match each font type to the appropriate low-pass filtered font signature using a V match operator as described earlier. After solving the problem for d = rlow mlow + rmean mmean + raita malta, one or more modalities could then simulate the desired ultra-wideband dataset as dbroadband = V-1low mlow + V-1mean mmedium + V-1high Malta.

[0067] For convenience of implementation, one or more modalities can, for example, use the “desired data” m' = V-1 m as the variable being updated during each iteration, therefore, dbroadband = m'low + m' medium + m'high. This requires adjusting the rest of the formulas to compensate for the change in variables from m to m' in a simple way. Those of ordinary skill in the art will recognize that there are other possible variations on the algorithm that alter the computation but keep the underlying mathematics the same.

[0068] In order to be able to add the different unmerged datasets together into a combined ultra-high bandwidth dataset, all different types of sources need to be represented in the same grid. In practice, different source types would typically have different acquisitions. In particular, lower frequencies can be sampled more coarsely, so many source points can be “missing” for lower frequency sources. One or more modalities can fill in these missing font points during the unmerge process, using the merge operator r to multiply these missing strokes by zero. An appropriate choice of smoothness constraint, then, will have the effect of interpolating the missing source points, so that all frequency bands can be represented on a common grid producing a combined dataset that simulates a single wideband source. .

[0069] Clearly, the correct choice of constraint S smoothness is important to solve the problem. In general, any denoising algorithm that works on a type of clustering that is sufficiently well sampled in the data can be used. One approach is to transform the data into a domain where the desired signal becomes sparse, and the noise non-scarce. All data below some percentage threshold can then be set to zero, and the data can then be transformed back to the original domain. Typically, the data can be subdivided into tapered overlapping subsets, each subset can be smoothed, and the smoothed subsets can then be recombined. This methodology is a noise removal technique. Typically, the transform of choice is a local 2-D, 3-D, 4-D, or even 5-D Fourier transform, which represents the data as a sum of events with linear motion: lines, planes, hyperplanes, etc. . Other transforms such as the Radon transform can also be used. The Radon transform represents the data as a sum of hyperbolas, so if you expect the desired signal to look like hyperbolas, this domain may produce a better result. Other applications use various types of wavelets, and other transform domains have been processed more recently as “curvelets”, “siselets”, etc.

[0070] One or more modalities can use the Fourier transform, and start limiting at a level close to 100%, only keeping the most coherent portion of the data. For example, FIG. 4 shows the inversion at an early stage, with a very strong coherence constraint producing an unnaturally smooth looking result. The high threshold applied in the Fourier domain only allows a narrow bandwidth to pass around the frequencies where the air gun is strongest (around 8 Hz). With each iteration, the threshold is reduced, eventually allowing all or most of the data to pass through in the final iteration (or iterations). A preferred schedule for doing this is threshold = 1 - (iter / niter)2, where a iter is the iteration count and niter is the number of iterations. A threshold of 1 allows only the single largest amplitude sample in the transformed data to pass, 0.5 allows the largest half of the transformed samples to pass, and 0 allows all samples of the transformed data to pass.

[0071] One or more modalities may also use different smoothing thresholds for different font types as another way to stabilize inversion when stronger and weaker fonts are blended. For example, one or more modalities might follow the standard 1 - (iter / niter)2 throttling schedule for the strongest source, but keep the throttling for the weakest source at a high level until the average data residuals have decreased to close to the signal levels of the weakest source. Only then would one or more modalities begin to reduce the threshold by smoothing the weakest source.

[0072] Typically, one or more embodiments may also weight the data before or after transformation, for example, to equalize amplitudes with respect to offset, or to prevent a strong frequency band from dominating (as illustrated in FIG. 4). Where the signal to noise is adequate, the weighting would typically be removed again after throttling. However, where the signal is weak or nonexistent, or the noise is strong, some modalities may have used weighting primarily to remove noise. In such a case, one or more modalities would not remove weighting after throttling, to avoid reintroducing previously removed noise. One or more modalities can also leverage knowledge of how the data was acquired. For example, one or more modalities may first interpolate on missing data using a method such as POCS (convex set projection), leaving the non-missing features unchanged, and then apply a smoothing constraint on a second pass.

[0073] In a case such as a marine nodal acquisition of combined low frequency source and air gun, the air gun recording lines can be spaced approximately (eg 50 meters apart). Ocean floor node array deployment typically “follows” along with airgun engraving lines, however, can considerably outrun airguns. Therefore, there is usually a gap near the midpoint of the survey when all (or nearly all) of the nodes are simultaneously deployed, and the node field is waiting for the airguns to catch up before recovery begins. At approximately this point in the survey, just before the nodes begin to be retrieved rather than deployed, the low frequency source arrives and crosses all of your source lines. This is a way to design the survey to efficiently acquire the ultra-long-shift ultra-low frequency data (which the low-frequency source was designed to acquire) at as many nodes as possible. Since the frequencies are so lower, and low frequency acquisition is often only for velocity model building purposes, not imaging, low frequency source lines can be much thicker than gun source lines of air, allowing a much faster acquisition. Low frequency source recordings may only span two or three weeks, while air gun recordings may span several months. Thus, there may only be a small percentage of air gun recordings that overlap with low frequency source signals.

[0074] This allows some embodiments the freedom to treat air gun traces contaminated by superimposed low frequency source signals as “missing”. In some embodiments, at the start of the inversion, one or more embodiments may treat the contaminated airgun recordings as “missing traces” and interpolate them into the data. As long as there aren't too many of them, this strategy may be viable. Later in the inversion, once the data residues have dropped, some modalities may then switch to other smoothing methods, for example, unmerging the tainted features rather than discarding them.

[0075] Typical vibrating sources produce one dominant frequency at a time, and achieve wider bandwidths by “sweeping” in frequency over time. In contrast, airgun signals are impulsive and broadband. Therefore, at any given time, only a narrow frequency band of the air gun signal is likely to be affected by interference from the vibrator signals. Therefore, one or more modalities can take advantage of this difference by individually flagging affected samples as “missing” in the frequency domain, rather than discarding entire Fourier transform airgun traces. Since smoothing criteria are often applied in the frequency domain, this is a simple extension. It is also commonly necessary for vibrator signals to include “resting” phases, so that the vibrator does not operate continuously. These gaps can be long enough so that, for example, every 3rd or 4th impulsive recording can entirely escape interference from the vibrating source. At low frequencies, these clean recordings can be sufficiently oversampled that they can be used for interpolation into contaminated intervening recordings, which would be treated as “missing” in the early stages of inversion.

[0076] Similarly, low frequency source signals often last many times longer than air gun signals (typically 60 to 300 seconds versus 12 to 15 seconds) and therefore only portions of frequency source sweeps individual drops will be contaminated by strong overlapping airgun signals, and in some cases only for nodes that are close enough to an operational airgun. One or more embodiments may flag all low frequency trace samples that are polluted by air guns above a certain amplitude level, and treat them as "missing" data to be interpolated. The “treat as missing” flag can be specified individually on a sample-by-sample basis in the time domain; it doesn't have to be stroke by stroke. There may be "treat as missing" flags in either the time domain or the frequency domain, or any other domain or domains. The "treat as missing" flag could be a function depending on arguments in multiple domains. In some embodiments, the flag can take on values between 0 and 1, allowing traces or samples to be weighted down, but not completely disregarded, and/or replaced, not with an interpolated value, but with a weighted average between the original value and the interpolated value.

[0077] One or more embodiments can use known air gun firing times to determine which samples are contaminated, but a simpler way to flag contaminated low frequency source data samples is to simply look at the energy in a window at around 8 Hz, the frequency at which airguns are most dominant, and flag all samples where the energy in a window around the sample is above a certain level in that frequency band. A frequency of around 8 Hz is a good value for current airgun arrays, but in general some modalities should use whatever frequency range is most appropriate for distinguishing the interfering source from the source being de-blended. This has the further advantage of capturing air gun shots that were inadvertently not recorded, or air gun recordings that belong to a neighboring survey not captured in the acquisition log files.

[0078] One or more modalities can modify the data acquisition to help ensure that this methodology works. In some embodiments, in any time window when the low frequency source is operating, the method may acquire only odd-numbered airgun source lines. This ensures that the method has good uncontaminated data from the even-numbered source lines. Since low frequency source signals only affect the lowest frequencies, one or more modalities only needed to interpolate “missing” data at the lowest frequencies. Twice the cross-line sample spacing (eg 50 x 2 = 100 meters) is still more than adequate sampling at these frequencies. At higher frequencies, any low frequency source contamination will be small enough that it can be ignored, and certain embodiments may use all airgun source lines. Certain embodiments may also avoid acquiring source lines amenable to division by 3, 4 or other increments, for example when the low frequency source was operating, details of which are possible depending on the overlapping frequency bands of different types of sources and the spacings cross line icon. The important observation is that in typical survey designs low frequencies are extremely overlapping, a fact that certain modalities can use to their advantage.

[0079] The choice of how many dimensions and how to order the data when applying S may depend on the details of the acquisition and, in general, may vary with the type of source. Blast gun sources will generally be densely sampled on a regular grid (50m by 50m, for example), so certain modalities have a wide choice of ways to order the data where they should be smooth. Therefore, for example, for nodal acquisition, certain modalities may apply the SA smoothness constraint to 3-D common receiver arrays. For seismograph cable acquisition, certain modalities can work in either 2-D or 3-D common vector displacement arrays. However, deep ocean nodes are typically not as well sampled (400 to 800 meters by 400 meters, for example). For air gun recording at ocean floor nodes, certain modalities therefore could not use smoothing on a common 3-D recording cluster at conventional air gun frequencies, at least not without some processing to avoid distortion, since the sampling density may be insufficient.

[0080] In contrast, a low-frequency source (e.g., an experimental source like Wolfspar®) can only be acquired in widely spaced thick source lines, and therefore the common receiver low-frequency source clusters recorded in nodes from the ocean floor can be sampled well enough in 2-D, not 3-D common receiver arrays. For each node cluster, the well-sampled geometry axes are source position and line time. The cross-line font position is not sampled well enough to be usable. However, since it operates at such low frequencies, the same receiver node grid that is coarsely sampled at higher frequencies (above 2 Hz) can become well-sampled for the purposes of a low-frequency source, which means that, unlike air guns at conventional frequencies, one or more modalities can apply SW in the common 3-D recording grouping domain. Note that there is no requirement that different sources or types of sources must all have consistency/scarcity constraints applied in the same domain; each must use a consistency and domain constraint that is appropriate for that source.

[0081] This difference in which types of clustering become “well-sampled” (by the Nyquist criterion) over different frequency bands is another reason why it makes sense to divide the problem by frequency band. This can be done simply by splitting the original data into overlapping tapered frequency bands, operating on each separately, and then re-merging the results if desired. Or, the splitting and re-merging can happen in an internal step of the algorithm, for example, during smoothness constraint application. Or, it can be done using frequency-dependent weightings or constraints within a unified general inversion. The acquisition density in relation to the frequency range will also dictate the possible choices of which grid to use for m, and what type of smoothness constraint to apply. For example, being able to invert a complete dataset with some missing recordings interpolated, or just compute unmerged versions of the actual recorded recordings. There may be good reasons for making different choices, not just for different source types, but for different frequency ranges of the same source.

[0082] Airguns are generally reliable, and if there are any problems with source placement, the suboptimal source points can simply be re-recorded. One or more modalities will likely lack the finesse for a new type of source, including a low frequency source such as Wolfspar®. As a result, there will sometimes be places where there are discontinuities or excursions in source lines. This can cause problems in the smoothing step of the algorithm, since the discontinuity will mean that, even if the correct data is not smooth, the smoothing step will try to “repair”.

[0083] This is similar to a problem that occurred in wave equation migration, when 2-D seismographic cable acquisition was still the norm. Seismograph cable lines may not lie in a straight line, however, they may have curvature due to varying crossover currents during acquisition. Available migration algorithms assumed that the data were truly 2-D, not acquired along a torsion acquisition line. A method called “Azimuth Motion” (AMO) was used to solve this. A simple brute-force method would be applied to flatten the reflectors in the data, using known correct source and receiver positions. For example, data can be flattened using normal motion correction (NMO), calculated using a very simple approximate velocity model. Then, the inverse of such an operation would be applied, however, with the desired source and receiver positions. If the acquisition errors were not too large, forward and inverse operations would approximately cancel each other out. Even if a very approximate correction (since the approximate velocity model is very simple) would still work well enough in practice to allow the data to then be treated as if it were 2-D for the purposes of migration and imaging. Certain modalities may use the same principle for unmerging, for example replacing S with (NMO-1actual_grid NMOregularized_grid) S (NMO-1regularized grid NMOactual grid).

[0084] In practice, certain modalities would probably not apply to forward and inverse NMO operations separately, but would combine them into a single operation (as indicated by grouping the operations in parentheses). Certain modalities can also simply perform smoothing directly on the NMO data, for example, flatten the events, then perform the smoothing, then restore the events to their original locations. Mathematically, this replaces S with (NMO-1actual_grid) S (NMOactual grid). Alternatively, certain modalities can make m represent regularized acquisition, and adjust the formulas to suit including mapping the regularized degree to actual source/receiver positions in the r merge operator.

[0085] Clearly, the method of one or more modalities relies on “coherence-based signal extraction”, the ability to distinguish desired arrivals from those due to interfering sources by their coherence (or lack thereof) in some domain. The data is acquired in a way that makes this separation possible. Hardware-generated random dithering of font initiation times was the preferred conventional method to achieve this described in the literature, however, other methods have also been successfully used in practice. The timing dither can be algorithmic, not truly random, with a fixed repeat pattern of pre-programmed recording timing dithers in the source controllers logic. Several good methods for designing “ideal” hesitation schedules have been described in the literature. In other cases, for example during terrestrial VibroseisTM surveys, hesitation can be achieved by simply allowing the operator of each source to choose when to fire the source. This method relies on unpredictable human operator actions to be “random” enough to avoid unwanted consistency between different sources. In a marine survey, recording in position rather than time can introduce enough randomness. The changing currents and moving waves of the ocean will act to slightly advance and slow the movement of the acquisition vessel, effectively introducing environmentally generated random dithering into source initiation times.

[0086] Depending on the acquisition geometry, the capabilities of the sources, and the choice of smoothing method, other ways of breaking the coherence between interference events may also work well in practice. In one embodiment, the source initiation times are all periodic, however, the periods are carefully chosen so that arrivals generated by one source can be easily separated from arrivals generated by other sources. Fig. 15 shows an example of a serrated common receiver array using initiation times from a first source. Event 1701 represents an arrival generated by the first source. In this domain, event 1701 is well sampled and only has slopes within a certain horizontal range. A second source has a slightly longer period, so 1702 events generated by the second source (but serrated according to the times of the first source) all have much steeper slopes. Although events 1701 and 1702 are well sampled within a coherence analysis window where they overlap (box 1710 showing such a possible coherence window), the events due to the two overlapping sources can easily be separated by their different embeddings. By adding an immersion filtering step to the coherence-based signal extraction, the unwanted interference event can be identified by its non-physically steep slopes and rejected, and this can be enough for the algorithm to then converge to the correct result despite of the perfectly regular periodicity of all sources.

[0087] For other interfering source period choices, events (such as 1703) can become so distorted that, within an analysis window (box 1720), they effectively become a type of noise, and this will result in the same being rejected by coherence-based signal extraction even without any additional immersion filtering step.

[0088] If the interference events have markedly different signatures (such as events 1701 and 1703 in Fig. 15), separation can also be achieved by analyzing the different sources in different domains. In Fig. 15, event 1701 is impulsive, however, events 1703 are blurred in time. After proper filtering to conform signatures, in other parsing window events from the source that generated the 1701 event may become blurred in time, and events from the source that generated the 1703 events can become impulsive, allowing one or the other event is mitigated using an improved scarcity constraint.

[0089] In practice, it may be preferable to combine multiple methods of ensuring the inconsistency of interference events. For example, in one embodiment, acquisition vessels would record in position, not time, ensuring that some natural randomness is added to source initiation times. However, the target speeds of the source vessels would then also be chosen to be different, so that even in the absence of natural time hesitation (for example, if the seas were perfectly flat at some point in the survey) the sources would still they would bump into each other in such a way that, within an analysis window, only one source at a time would appear coherent.

[0090] For fonts of different types that may have very different nominal repetition ranges, the more general design objective is to design the survey so that two different fonts never “hit” each other, that is, no multiple of numbers small integer of the nominal font repetition ranges matches no combination of different fonts. Vessel speeds typically vary within a narrow operating range, so the unhesitated source time interval when recording at position also varies over some range, and this uncertainty often needs to be considered. For example, it is assumed that there is recording of air guns in position, and recording of low frequency source vibration in time. For nominal separation between recording points, the airgun source repeat interval can vary between 11 and 12 seconds, depending on vessel speed. By survey design, the low-frequency (time recording) source should have a nominal unhesitating repetition interval somewhere in the range of 100 to 120 seconds, the precise value being a design parameter that can be chosen. One or more embodiments can then ensure that the low frequency source is not coherent with the air guns by choosing a repetition interval of 109 seconds. This increment is not an integer multiple of the airgun retry interval for any value between 11 and 12 seconds. Mathematically, 9 x 11 to 12 seconds gives a range of 99 to 108 seconds, and 10 x 11 to 12 seconds gives a range of 110 to 120 seconds. Considering a permissible design range of 100 to 120 seconds, one or more modalities may choose a value between 108 to 110 seconds to avoid any possibility of “crashing” with the air guns. If another nearby seismic survey is recording in time instead of position, 109 seconds being prime is also not a multiple of any integer second repeat interval less than 109 seconds. Therefore, by choosing a repetition interval of 109 seconds, one or more modalities can minimize the chance of inadvertent coherence between the low frequency source and any air guns. Those of ordinary skill in the technique will readily see how to apply these principles to other surveying projects.

[0091] All these different options described in the previous paragraphs (both for the external signal separation algorithm and the available coherence-based signal extraction algorithms that are repeatedly used in it) can be combined in a variety of permutations. Clearly many methods are possible, depending on the number and types of sources, their acquisition geometries, the choice of grid to represent the result, the frequency range under consideration, the signatures and spacing of the sources, and/or the desired application. In particular, one or more embodiments may have any number of different types of fonts, and any number of fonts of each type. Fonts can be combined or kept separate in any possible combination. One or more embodiments can combine all sources into a single output, or have one output for each source, or one output for each type of source, or any combination in between, combining some sources into the output but not others, etc.

[0092] Thus, in some embodiments, applying the present inversion process provides reasonably clean receiver clusters that can be used for imaging, pre-stacking analysis such as AVO (Amplitude vs. Displacement) analysis and velocity model building, for example, by Total Waveform Inversion (FWI).

[0093] Acquiring seismic data with recordings where the information recorded from one recording overlaps in time with other recordings has the potential to significantly reduce the time (and cost) required to record a seismic survey. This approach could also allow for more closely spaced recording point intervals (eg during a marine survey) which could potentially provide better seismic imaging that would improve the chances of discovering cost effective quantities of oil and/or gas. Allowing multiple types of sources to be acquired simultaneously allows wider bandwidth data to be acquired as efficiently as conventional air gun data.

[0094] To provide context for the acquisition system, FIG. 8 illustrates the general environment in which the present disclosure would typically be used. A seismic survey is designed by the explorer to cover an area of economic interest. Field acquisition parameters (e.g., recording spacing, line spacing, folding, font type, etc.) are typically selected in conjunction with this step, although it is common to modify the ideal design parameters slightly (or substantially) in the field to accommodate the realities of conducting the survey.

[0095] Seismic data (i.e., seismic traces) are collected in the field 120 over a subsurface target of potential economic importance and are typically then sent to a processing center 150 where the traces will be subjected to various algorithms available for make them more suitable for use in exploration. In some cases, there may be some amount of initial data processing performed while the data is still in the field, and this becomes more common and feasible given the computing power that is available to field crews. As described above, within field 120, there may be two or more overlapping simultaneous sources of the same type and/or there may be two or more overlapping simultaneous sources of different types. As described above, records from different sources may have been combined into a single volume.

[0096] At the processing center, a variety of preparatory processes 130 are typically applied to seismic traces to make them ready for use by the explorer. The processed traces would then be made available for use by the present systems and methods and may be stored, by way of example only, on a storage device such as a hard disk, magnetic tape, solid state drive, magnetic optical disk, DVD or other mass storage media.

[0097] The methods disclosed herein may be implemented in the form of a computer program 140 that has been loaded onto a programmable computer 150 where it is accessible by an interpreter or seismic processor. It is noted that a computer 150 suitable for use with the present disclosure would typically include, in addition to mainframes, servers and workstations, supercomputers, and more generally, a computer or network of computers providing parallel and massively parallel computations, wherein the computational load is spread across two or more processors. As also illustrated in FIG. 8, in some embodiments, some type of digitized zone-of-interest model 160 may be specified by the user and provided as input to the processing computer program. In the case of a 3-D seismic section, the zone of interest model 160 would typically include specifics as to the lateral extent and thickness (which may be variable and can be measured in time, depth, frequency, etc.) of a subsurface target . The exact means by which such zones are created, collected, digitized, stored and read later during program execution are not important to present systems and methods and those skilled in the art will recognize that this can be done in a number of ways.

[0098] A program 140 incorporating the present processes can be transmitted to the computer that must execute the same by means of, for example, a floppy disk, a magnetic disk, a magnetic tape, a magnetic optical disk, an optical disk, a CD -ROM, a DVD disk, a RAM card, flash RAM, a RAM card, a PROM chip, or loaded over a network (eg, a wired or wireless network, etc.). In a typical seismic processing environment, the methods of the present disclosure would be made part of a larger package of software modules that is designed to perform many of the processing steps listed in FIG. 9. After processing by present methods, the resulting traces would then be sorted into clusters, stacked and displayed on a high resolution color computer monitor 170 or in hard copy form as a printed seismic section or map 180. The seismic interpreter then uses the displayed images to assist in identifying subsurface features conducive to the generation, migration or accumulation of hydrocarbons.

[0099] As indicated above, the present disclosure will preferably be made as a part of and incorporated into a conventional seismic processing sequence of the type generally described in FIG. 9. Those of ordinary skill in the art will recognize that the processing steps illustrated in FIG. 9 are only broadly representative of the types of processes that can be applied to such data and the choice and order of processing steps, and the particular available algorithms involved, can vary notably depending on the individual seismic processor, the signal source (dynamite, vibrator , air gun, low frequency generator, etc.), the survey location (land, sea, etc.) of the data, the company processing the data, etc.

[0100] As a first step, and as is generally illustrated in FIG. 9, a 2-D or 3-D seismic survey is conducted through a particular volume of the earth's subsurface (step 210). The data collected in the field consists of unstacked (ie, not summed) seismic traces that contain digital information representative of the volume of land lying below the survey. The methods by which such data are obtained and processed into a form suitable for use by seismic processors and interpreters are well known to those of ordinary skill in the art. It is noted that for purposes of the present disclosure, the seismic survey will be a mixed source survey, in which reflections from a later source activation may interfere with (or coincide in time with) reflections from an earlier one. After the recordings or sources have been separated in accordance with the present disclosure, the unstacked seismic traces resulting from such operation are usable as any other collection of seismic traces will be. Thus, the present systems and methods allow for multiple seismic traces to be obtained from sources that are activated to produce overlapping signals, where the resulting separate seismic data sets are equivalent to or substantially equivalent to seismic data sets obtained using seismic acquisitions. fully separate seismic data and source activations (eg, non-overlapping sources and non-stacked seismic traces).

[0101] The purpose of a seismic survey is to acquire a collection of spatially related seismic traces relative to a subsurface target of some potential economic importance. Data that is suitable for analysis by the methods disclosed herein may consist, for purposes of illustration only, of a 2-D unstacked seismic line, a 2-D unstacked seismic line extracted from a 3-D seismic survey, or, preferably, an unstacked 3-D portion of a 3-D seismic survey. The systems and methods disclosed herein are most effective when applied to a group of stacked seismic traces that have an underlying spatial relationship to a subsurface geological feature. Again for purposes of illustration only, the following discussion will be formulated in terms of traces contained within a 3-D survey (stacked or unstacked as the discussion warrants), although any assembled group of spatially related seismic traces can be used in a similar way. conceivable.

[0102] After the seismic data are acquired (step 210), they are typically taken to a processing center or processed in the field in which some initial or preparatory processing steps are applied to them. As illustrated in FIG. 9, a common initial step 215 is designed to edit incoming seismic data in preparation for subsequent processing (eg, demux, gain recovery, wavelet shaping, poor trace removal, etc.). This can be followed by specifying the survey geometry (step 220) and storing a recording/receiver number and surface location as part of each seismic trace header. Once the geometry has been specified, it is customary to perform a velocity analysis and apply a NMO (normal motion) correction to correct each trace in time to account for signal arrival time delays caused by displacement. In some embodiments, the present disclosure may be used in connection with step 215, i.e., in conjunction with or in place of the Vibroseis® wavelet forming/correlation steps, although it certainly may be used elsewhere within this scheme of widespread processing.

[0103] After the initial pre-stacking processing is complete, it is customary to condition the seismic signal onto the unstacked seismic traces before creating stacked (or summed) data volumes (step 230). In FIG. 9, step 230 contains a typical "Signal Processing/Conditioning/Imaging" processing sequence, however, those skilled in the art will recognize that many alternative processes may be used in place of those listed in the Figure. In any case, the primary objective from the explorer's point of view is the production of a stacked seismic volume or, in the case of 2-D data, a stacked seismic line for use in exploration for hydrocarbons within the earth's subsurface.

[0104] As is further suggested in FIG. 9, any digital sample within a stacked seismic volume is uniquely identified by a triplet (X, Y, TIME), with the X and Y coordinates representing some position on the earth's surface, and the time coordinate measuring a recorded arrival time inside the seismic trace (step 240). For purposes of specificity, it will be assumed that the X direction corresponds to the "in-line" direction, and the Y measurement corresponds to the "cross-line" direction, as the terms "in-line" and "cross-line" are generally understood in the art. While time is a preferred and most common vertical axis unit, those skilled in the art will understand that other units which are certainly possible may include, for example, depth or frequency. Additionally, it is a well known fact to those skilled in the art that it is possible to convert seismic traces from one axis unit (eg time) to another (eg depth) using standard mathematical conversion techniques.

[0105] The explorer can perform an initial interpretation 250 of the resulting stacked volume, in which it locates and identifies the reflectors and main faults whenever they occur in the data set. This may be followed by further data enhancement 260 of the stacked or unstacked seismic data and/or attribute generation (step 270) therefrom. In many cases, the explorer will revisit its original interpretation in light of the additional information gained from the attribute generation and data enhancement steps (step 280). As a final step, the explorer will typically use information gleaned from seismic data along with other types of data (magnetic surveys, gravity surveys, LANDSAT data, regional geological studies, well logs, well cores, etc.) to locate resources subsurface structural or stratigraphic features conducive to the generation, accumulation or migration of hydrocarbons (ie, generation outlook 290).

[0106] Returning to the systems and methods disclosed in this document, a method may include separating two or more seismic sources that were activated during a single recording session, in which the similarity between nearby recordings is used to restrict the inversion process and creating a plurality of separate seismic data sets.

[0107] With reference to FIGS. 10 and 11, a merged source survey can be collected by first placing a number of receivers 310 in a 2-D configuration over a target of exploration interest. In some embodiments, there may be only a few or even several thousand 310 receivers in the survey. The receivers 310 can be connected by cables to a central recording unit, they can use wireless transmission to the same, or each receiver can contain some amount of internal data storage in which to record the seismic signals received by it. Those of ordinary skill in the technique will be quite familiar with these types of receiver variations.

[0108] In some embodiments, receivers 310 would be continuously registered for an extended period of time. In some variations, receivers can be registered for a few hours, half a day, a full day, multiple days, etc. The only requirement is that the record must capture at least two source excitations. This is in contrast to common seismic surveying, where receivers are recorded for just a few seconds after a source activates.

[0109] During the period of time that the receivers are being registered, numerous seismic sources 320 will be activated in different locations within the survey area 300. In some embodiments, two or more sources will be used, where the sources may have signatures of different source and/or frequency bands, including any of those described in this document. In the case of a marine survey it is likely that different types of sources will be used, however that is obviously something that is left to the discretion of the survey designer.

[0110] The two or more seismic sources 320 may include any number of discrete or continuous sources. In the marine context, seismic sources 320 may include one or more air guns, spark gaps, vibrators, or the like. In some embodiments, any number, volume, and type of seismic sources can be included in a matrix. As an example, an array can include one or more large volume air guns, one or more medium volume air guns, and/or one or more small volume air guns. Seismic sources can be arranged in various matrices. For example, a collection of seismic sources can be positioned so that they are close together and intended to be activated as part of the same seismic recording. In a marine context, this can be a single source seismic array or multiple source arrays towed by one or more vessels. A sub-array (which may consist of one or more sources) will be understood to be a subset of the sources, with the source (or sources) assigned to each sub-array being intended to be activated simultaneously.

[0111] On earth, the two or more sources 320 may include vibrators, loads, or seismic sound sources. For example, the two or more sources 320 can include earth vibrators (eg, 10 or similar), Mini-Sosie™ lifts, drop weight lifts, etc. One type of seismic survey system uses a vibrator or group of vibrators to provide the source of the acoustic energy (hereinafter called the “Vibroseis source”). A Vibroseis source can generate (“vibrate”) acoustic energy waves at predetermined vibrator points (VPs). PVs can be marked with a stake placed by researchers. Alternatively, global positioning satellite (GPS) equipment can be used to allocate predetermined PVs.

[0112] Other sources may include low frequency sources, such as those used in marine surveys, including Wolfspar® sources. The low-frequency sources could each operate at a single frequency ("monochromatic" low-frequency sources), or cycle between two or more discrete frequencies ("frequency-scaled" low-frequency sources), or sweep over a range of frequencies. narrowband of low frequencies designed to increase the frequency range produced by wideband sources ("narrowband" low frequency sources). The sources could operate to produce waves of constant amplitude, or the amplitude of the waves could vary (taper up and down). In this context, "low frequency" means frequencies less than about 6 to 8 Hz. Some embodiments will be below about 4 Hz, some of which may employ frequencies as low as about 2 Hz, or about 1.5 Hz, or about 0.5 Hz.

[0113] In some embodiments, two or more seismic sources 320 may have different signatures. In Vibroses operations, the vibrator (or vibrators) of the Vibroseis source typically generates a vibration pattern that changes frequency over a predetermined period of time. This vibration pattern is called a vibratory sweep. A typical Vibroseis sweep may be a linear frequency sweep of approximately 5 Hz to 100 Hz and may last on the order of 5 to 30 seconds. Air guns can generate high frequency acoustic waves. Low frequency sources can generate low frequency signals over a longer period of time and, in some embodiments, low frequency sources can operate continuously.

[0114] In some embodiments, source activations will be separated in time by random periods of time. Furthermore, sources can be activated close enough in time that there is some overlapping or blending between recordings. When seismic records are corrected for time zero for each source (i.e., activation time), reflections related to that source will tend to be coherent, however, energy from interfering sources will tend to be incoherent (i.e., reflections will not lines up) as the delay between writes is random. That is, for example, in the case of a survey where each source 320 is a Vibroseis unit, it is contemplated that source activations may be separated by a few seconds in some cases. It is observed that FIG. 10 is not intended to suggest that each source 320 is activated simultaneously, but rather it is intended to indicate that each source is located at a different location within survey area 300. During some surveys, ten or more different sources may be used. used. As an example of an exploration method that would produce data that would be suitable for use with the present disclosure, attention is drawn to WO 2008/025986 (PCT/GB2007/003280) "Seismic Survey Method" which names Howe as its only inventor, whose application is incorporated herein by reference in its entirety. Howe discusses the use of staggered activation of vibrator activations where there is some overlap in the returning subsurface reflections.

[0115] Although the seismic sources 320 may comprise discrete or semi-discrete signatures, one or more of the seismic sources may operate continuously based on a continuous firing pattern or as a continuous emission source. In some embodiments, the method may include continuous shooting, which may also be referred to as "continuous pattern shooting" or "continuous shooting patterns". Specifically, instead of recording one shot pattern and then recording a different shot pattern, there is no time delay and a continuous shot pattern is used which consists of continuous popcorn shots for the case of air gun arrays. . Thus, in some embodiments, there may be gaps of a few to several seconds between two firing patterns, but there need not be. For low frequency sources, the low frequency source can operate continuously over a given period of time to produce the low frequency energy in a continuous manner.

[0116] As described above, different types of fonts may have different characteristics. Different sources having different signatures and/or frequency ranges can allow for differences in source signatures, relevant coherence criteria and acquisition processes (eg 2-D acquisition, 3-D acquisition, etc.) to provide better separation in the acquired data. Exemplary source combinations may include using air guns with vibrators, air guns with spark gaps, air guns with a low frequency source (e.g., a Wolfspar source, etc.), or combinations of three or more. font types.

[0117] FIG. 11 generally suggests what data from a mixed source survey might look like. Each receiver 310 will generate a seismic trace (e.g., trace 405) that could potentially be tens of minutes or several hours (or days, etc.) in length. In this Figure, trace 405 is shown schematically as containing recorded signals from four different source excitations. Although the illustrated signals are shown to be similar, recorded signals may have signatures matching the source. When multiple fonts that have different font signatures are used, the signals can have correspondingly different signatures. Associated with each receiver 310 will be a location on the earth's surface. When the signals that have been recorded from each receiver 310 are properly arranged and displayed, in some embodiments, a 3-D volume will be produced, with each receiver 310 associated with an "X" and a "Y" location, to include locations based on latitude and longitude, etc.

[0118] Preferably, during a merged source survey, the time each source 320 is activated will be observed and recorded, which sources may be located inside or outside the receiver field. In FIG. 11, T1 and T2 represent the known times (as measured from an arbitrary zero time) that two sources were activated, with the parameter “N” generally indicating the length of time (and/or a number of samples ) after source activation during which subsurface reflections from this source can be picked up. In this particular example, and as will be explained in more detail below, the two source activations can be from the same source (eg two Vibroseis sweeps) so that they do not overlap in time. However, with other modalities, source activations can be from different sources. In this arrangement, the interference will most likely come from other seismic sources that were active during the indicated time window or that had subsurface reflections arriving during this same time window. Having said that, the present processes would apply equally if two or more source activations from the same source overlapped in time.

[0119] Generating the plurality of separate seismic data sets can generate useful seismic data for various downstream processing techniques as described herein. In some embodiments, the resulting separate seismic datasets may include different seismic trace data. For example, a first set of seismic data (eg m1) may comprise standard separate seismic trace data useful for traditional processing and imaging. Alternatively, or as a second separate seismic dataset (eg mi, m2, ... mn, etc.), the resulting separate seismic dataset can be used for other forms of processing. It should be noted that the inversion process disclosed herein may allow some elements of individual separate seismic datasets to be interpolated based on the iterative application of the inversion process. This may allow some portion of the resulting separate seismic data sets to be created based on the overlap between the two source excitations, which may produce more complete separate data sets.

[0120] In some embodiments, one or more of the separate seismic data sets may be used for the purpose of full waveform inversion (“FWI”) analysis. FWI is a time- or frequency-based seismic processing technique that provides a more general paradigm for imaging subsurface structures: rather than relying solely on reflected or scattered waves echoing from geological discontinuities on Earth, FWI also uses transmitted/ refracted waves traveling downwards, then turn to become horizontal, and finally turn upwards to emerge as upward seismic waves (possibly at a considerable distance from their origin). The subsurface structures in Earth's advancing, retarding and/or distorting these waveforms transmitted/refracted by their presence, and FWI resolves their location and properties from the characteristic imprints they leave in the data. It is noted that without low-frequency broad shift data containing the transmitted/refracted waves, full waveform inversion often fails and may not resolve subsurface structures (i.e. may produce a useless result). Unfortunately, traditional seismic sources do not provide the low frequency waves that would generally be desired, and more particularly the low frequency data that can be used when full waveform inversion is performed. Therefore, the separation and generation of separate seismic datasets can be used to enable the desired seismic processing. While FWI is a natural application of this method, other uses of the separate data that do not involve imaging or velocity determination are possible.

[0121] Having described the different types of sources, the general inversion model and the desirable output seismic data types, the description now turns to the processes useful in implementing the models. FIG. 12 illustrates a logical process flow that would be suitable for use when the sources for a blended survey are one or more air guns and one or more seismic vibrators (for example, a Vibroseis survey). Those of ordinary skill in the art will readily see how to generalize the FIG process flow. 12 for the case of 3 or more font types. The process can be considered as proceeding over 3 nested layers, the outermost of which comprises boxes 1400, 1403, 1499, 1491, 1492, 1493, 1401, 1402, 1480 and 1485.

[0122] Step 1400 is initialization. Three temporary data stores are created, one to hold the residuals, one to hold the unmerged Air Gun data (signal 1), and one to hold the unmerged Vibroseis data (signal 2). The residual is initially just a copy of the recorded seismic data, and the two unmerged signal buffers are initially set to contain all zeros.

[0123] Step 1403 is a check of the possibility that the inversion is done. This can happen because the residual (the difference between the actual recorded data and the predicted recorded data) is small enough, or because a maximum iteration count has been reached. If the condition is satisfied, the merged data (consisting of the two estimated signal traces) is output at step 1499. Otherwise, the new residue is fed to bins 1401 and 1402, an iteration of merged source estimation for air guns and Vibroseis, respectively. Each of these boxes is individually analogous to the method described in document No. US 8,295,124 B2, Abma, in particular the steps described in Figure 8 of Abma at steps 815 to 875. The subject of document No. US 8,295,124 B2 is incorporated herein in its entirety. Each of the 1401 and 1402 issues an up-to-date estimate of the logged data that would have been logged if only the corresponding font type were active (the Air Gun for 1401, and the Vibroseis for 1402), and the data for that font type had been acquired in a traditional unmerged acquisition. These are then negated, the air gun data in box 1480 and the Vibroseis data in box 1485.

[0124] The negated single-source type predictions are then summed in box 1492 with the recorded data (box 1491), producing the updated residual. The newly updated residue is stored in its data staging store (box 1493) and then passed to box 1403. The process repeats until completion. Note that if the prediction were perfect, then the unmerged single source type predictions would add to the actual recorded data, and the output of box 1492, the newly updated residual, would be identically zero. In practice, at the end of the process, the residue will contain incoherent noise, and the de-merging algorithm will therefore have the desirable side-effect of also removing noise from the data. At the same time, in theory, merged acquisition should produce a better result than traditional unmerged acquisition, as a shorter acquisition time means there is less noise for the same amount of signal.

[0125] The intermediate layer comprises the interior of boxes 1401 and 1402. First, the residue is serrated (1410, serrated using air gun strokes, and 1415, serrated using Vibroseis strokes). The buffers containing the current estimated unmerged data (box 1420 for the air gun, and box 1425 for the Vibroseis) are then accessed and each is summed with the corresponding serrated residue, in box 1430 for the air guns and 1435 for Vibroseis. The Vibroseis processing sequence contains the extra step of regularizing the source signature (box 1416), using a signature deconvolution, signature correlation, match filter, etc. A workflow such as the workflow described by Abma in US 8,295,124 B2 would immediately proceed to the data smoothing step to extract the coherent signal (box 1440 for the air gun, and box 1445 for to Vibroseis). When separating different types of sources, it may be necessary to balance their amplitudes. Otherwise, one source type may dominate over the other and the method may not converge. It is also not useful to try to produce a coherent signal that does not exist, for example trying to find a coherent signal above 4 Hz for a low frequency source that produces little to no energy in that frequency band. Attempting to do so will only accentuate the noise, again resulting in the method failing to converge to a usable result. Thus, the modified method includes an additional time/frequency weighting step not found in Abma, box 1432 for the air guns and box 1437 for the Vibroseis. Its function is to balance the relative strengths of different types of sources, and to attenuate the frequencies not produced by the corresponding source. The method then proceeds to coherent signal extraction (steps 1440 for the air gun and 1445 for the Vibroseis) as before. Optionally, more time/frequency weighting can be applied at steps 1442 and 1447 to partially or fully remove weights applied at 1432 and 1437, respectively. If weights were applied to suppress noise, then they are not removed in this step. The result is an update of the estimated unmerged signals, which are added to the previous estimate in steps 1450 and 1455 to produce updated unmerged signal estimates for the Air Guns and Vibroseis, respectively. The updated unmerged signal estimates are stored in their respective data buffers in steps 1460 and 1465. The unmerged traces are then demerged in steps 1470 and 1475 for the Air Guns and Vibroseis, respectively. The Vibroseis flow contains the additional step, in box 1466, of removing the applied font signature in box 1416.

[0126] The innermost layer occurs within boxes 1440 and 1445, coherent signal extraction. Figure 13, box 1500, contains a standard approach as described in Abma for the case of a single source type, and therefore box 1500 can represent the contents of boxes 1440 or 1445. The data is organized into an appropriate grouping where the desired signal must be coherent in box 1510. They are then Fourier transformed in box 1520 such that the desired signal must become "scarce" in that domain. The smallest values in the transformed space are reset to zero at step 1530, to leave the signal untouched while attenuating the noise. The data is then returned to the original domain in step 1540.

[0127] As in Abma, the data will usually be broken into overlapping N-dimensional subcubes, with appropriate tapering in the overlapping zones, and this operation will be performed on each subcube, and the results then merged together again to reconstitute a version with noise removed from the original data.

[0128] An alternative method that can be particularly useful when unmerging different types of data is shown in Figure 14, box 1600. It can be used in place of box 1500, or it can be used first and then the method at 1500 can be used. follow. In this method, the tainted features are marked as “dead” or “missing”, and the standard methods for filling in missing data by interpolation are used to replace the tainted features. The method shown in box 1600 is “POCS interpolation” (Projection For Convex Sets). Contaminated or missing traits will have been identified earlier, and this information is retrieved in box 1650.

[0129] As in box 1500, first the data is organized into appropriate groupings (box 1610). Then, the data is Fourier transformed (box 1620) to become sparse, the smallest values are zeroed or reduced (box 1630), and the data is transformed back to the original domain (box 1640). The difference is that at step 1660 the traces that do not need to be interpolated are replaced with their original values. At step 1670, the result is compared with the previous result, and if it has stopped changing (or the maximum iteration count has been reached) the process exits. Otherwise, the data containing the newly interpolated "missing" traces is sent back for another iteration at box 1620.

[0130] Figures 12 to 14 demonstrate a particular set of preferred embodiments. Those of common skill in the art should readily see how to generalize the methods to other cases, and how to incorporate the many possible alternative options described earlier in the description into the basic workflow.

[0131] It should also be noted that Abma demonstrated two approaches: a constructive approach and a deconstructive approach. The same possibilities apply here. The disclosure above described the generalization of the “constructive approach” to datasets from two or more mixed source types. Those of ordinary skill in the technique should readily be able to see how to similarly generalize the methods described here to use the "deconstructive approach".

[0132] It should be noted that when operations are to be performed on traces of a particular type (e.g., a receiver cluster), it is normally not necessary to bring join seismic traces into memory (e.g., by means of a classification) to in order to apply multi-trace processes to them. Thus, in the above disclosure and the following claims, when it is said that an array (e.g., recording array, receiver array, etc.) is assembled or accessed for further processing, such words are to be interpreted in their broadest sense to cover instances where the features comprising the cluster are processed in place or instantaneously. Therefore, no sorting or other disposition of the data may necessarily be required.

[0133] Furthermore, in some embodiments, the present systems and processes will be adapted for use with a VSP or similar downhole survey. By way of explanation, those of ordinary skill in the art will understand that VSP procurement can be very costly in terms of equipment downtime. Faster recording VSPs with overlapping sources could be used to significantly reduce the costs of such surveys. Therefore, when the phrase “blended seismic survey” is used in this document, such phrase should be interpreted broadly to include both terrestrial and marine 2-D and 3-D surveys, as well as VSPs, crossbore surveys, etc.

[0134] Furthermore, in the previous discussion, the language was expressed in terms of operations performed on conventional seismic data. However, it is understood by those skilled in the art that the disclosure described herein can advantageously be applied in other areas of matter, and used to locate subsurface minerals other than hydrocarbons. By way of example only, the same approach described in this document could potentially be used to process and/or analyze multi-component seismic data, shear wave data, mode converted data, crossbore survey data, VSP, full waveform sonic recordings, controlled source data or other electromagnetic data (CSEM, t-CSEM, etc.) or model-based digital simulations of any of the foregoing. Additionally, the methods claimed below herein can be applied to mathematically transformed versions of these same data traces including, for example: filtered data traces, migrated data traces, frequency domain Fourier transform data traces, frequency domain transformations by discrete orthonormal transforms, instantaneous phase data traces, instantaneous frequency data traces, quadrature traces, analytical traces, etc. In summary, the process disclosed in this document can potentially be applied to a wide variety of geophysical time series types, however, it is preferably applied to a spatially related collection of time series.

[0135] Having described various systems and methods, certain aspects may include, but are not limited to:

[0136] In a first aspect, a seismic exploration method above a subsurface region containing structural or stratigraphic features conducive to the presence, migration or accumulation of hydrocarbons comprises accessing at least a portion of a mixed seismic source survey, in which the A mixed source seismic survey contains at least two interfering seismic source excitations therein, where the seismic source excitations are produced by one or more types of seismic source. Seismic source types may have different signatures or frequency characteristics; separate the at least two interfering seismic source excitations using inversion separation; produce one or more font groupings based on separation; and using the one or more source arrays to exploit hydrocarbons within said subsurface region.

[0137] A second aspect may include the method of the first aspect, wherein separating the at least two interfering seismic source excitations comprises: selecting a first coherence constraint for a first seismic source type from at least two seismic source types; selecting a second coherence constraint for a second seismic source type two at least two seismic source types; and applying the first coherence constraint and the second coherence constraint in the inversion separation to the portion of the merged seismic source survey, wherein the output of the one or more source groupings is based on applying the first coherence constraint and the second constraint of coherence.

[0138] A third aspect may include the method of the second aspect, where the first consistency constraint and the second consistency constraint are different.

[0139] A fourth aspect may include the method of any one of the first to third aspects, wherein separating the at least two interfering seismic source excitations comprises: determining activation times for each interfering seismic source excitation of the at least two excitations from interfering sources; and using the times of activations corresponding to each seismic source excitation to shift the at least two interfering seismic source excitations to a zero time, where the production of the one or more source clusters is based on the displacement of the at least two excitations of Interfering seismic source for time zero.

[0140] A fifth aspect may include the method of any one of the first to fourth aspects, wherein separating the at least two interfering seismic source excitations using inversion separation comprises solving an equation having the form: , for the matrices mi, mi ... mn, to produce the one or more source groupings, where n can be greater than or equal to 1, where mi can be a first source grouping among the one or more source groupings font, mi is another font grouping of the one or more font groupings; d is a matrix representation of the recorded seismic data, ri is a matrix defining excitation times of a first source type from one or more seismic source types, and ri is a matrix defining excitation times of another source type one or more types of seismic source.

[0141] A sixth aspect may include the method of any one of the first to fifth aspects, wherein a plurality of source arrays each have seismic data representative of a different frequency range.

[0142] A seventh aspect may include the method of any one of the first to sixth aspects, further comprising: performing a full waveform inversion analysis on at least one of the source clusters.

[0143] An eighth aspect may include the method of any one of the first to seventh aspects, wherein each of the at least two interfering seismic source excitations is separated in time by a random amount of time.

[0144] A ninth aspect may include the method of any one of the first to eighth aspects, in which at least two types of seismic sources are different and comprise a combination of at least two of: vibrational sources, air gun sources, spark gap sources and low frequency sources.

[0145] A tenth aspect may include the method of any one of the first to ninth aspects, wherein at least one seismic source of the at least two interfering seismic source excitations has altered a signal or phase of its seismic source excitation.

[0146] In an eleventh aspect, a seismic exploration method comprises: initiating a plurality of seismic source excitations above a subsurface region containing structural or stratigraphic features conducive to the presence, migration or accumulation of hydrocarbons, wherein the plurality of seismic source excitation is produced by one or more types of seismic source. Seismic source types may have different signatures or frequency characteristics; and producing a mixed source seismic survey comprising data obtained from reflections generated by the plurality of seismic source excitations, wherein the mixed source seismic survey contains at least two interfering seismic source excitations therein, wherein the at least two interfering seismic source excitations are separated using inversion separation, and wherein one or more source clusters are produced based on the separation of the at least two interfering seismic source excitations.

[0147] A twelfth aspect may include the method of the eleventh aspect, in which the at least two interfering seismic source excitations are separated using inversion separation by solving an equation having the form: , for the matrices mi, mi ... mn, to produce the one or more source groupings, where n can be greater than or equal to 1, where mi can be a first source grouping among the one or more source groupings font, mi may be another font grouping of the one or more font groupings; d is a matrix representation of the recorded seismic data, ri is a matrix defining excitation times of a first source type from one or more seismic source types, and r2 is a matrix defining excitation times of another source type one or more types of seismic source.

[0148] A thirteenth aspect may include the method of the eleventh or twelfth aspect, wherein a plurality of source arrays each have seismic data representative of a different frequency range.

[0149] A fourteenth aspect may include the method of any one of the eleventh to thirteenth aspects, wherein each of the at least two interfering seismic source excitations is separated in time by a random amount of time.

[0150] In a fifteenth aspect, a method of seismic exploration above a subsurface region containing structural or stratigraphic features conducive to the presence, migration or accumulation of hydrocarbons, the method comprising: accessing at least a portion of a seismic source survey mixed source, where the mixed source seismic survey contains at least two interfering seismic source excitations in it, where the seismic source excitations are produced by one or more types of seismic source. The seismic source types may have different frequency signatures or characteristics, wherein at least one of the seismic source types is a low frequency source; separate the at least two interfering seismic source excitations using inversion separation; producing one or more source clusters based on the separation, wherein at least one of the source clusters contains low frequency data or some missing frequencies; and using the one or more source arrays to exploit hydrocarbons within said subsurface region.

[0151] A sixteenth aspect may include the method of the fifteenth aspect, wherein separating the at least two interfering seismic source excitations comprises: selecting a first coherence constraint for a first seismic source type from at least two source types seismic; selecting a second coherence constraint for a second seismic source type two at least two seismic source types; and applying the first coherence constraint and the second coherence constraint in the inversion separation to the portion of the merged seismic source survey, wherein the output of the one or more source groupings is based on applying the first coherence constraint and the second constraint of coherence.

[0152] A seventeenth aspect may include the method of the fifteenth or sixteenth aspect, wherein separating the at least two interfering seismic source excitations using inversion separation comprises solving an equation having the form: for the matrices mi, mi ... mn, to produce the one or more font groupings, where n can be greater than or equal to 1, where mi is a first font grouping of one or more font groupings, mi is another font grouping of the one or more font groupings; d is a matrix representation of the recorded seismic data, ri is a matrix defining excitation times of a first source type from one or more seismic source types, and ri is a matrix defining excitation times of another source type one or more types of seismic source.

[0153] In an eighteenth aspect, a seismic exploration method above a subsurface region containing structural or stratigraphic features conducive to the presence, migration or accumulation of hydrocarbons comprises accessing at least a portion of a mixed seismic source survey, in which the mixed source seismic survey contains at least two interfering seismic source excitations therein, wherein the seismic source excitations are produced by one or more types of seismic source. Seismic source types may have different signatures or frequency characteristics; and using one or more generated source arrays to exploit hydrocarbons within said subsurface region.

[0154] A nineteenth aspect may include the method of the eighteenth aspect, wherein using the plurality of generated source groupings comprises solving an equation having the form: , for the matrices mi, mi ... mn, to produce one or more font groupings, where n can be greater than or equal to 1, where mi is a first font grouping of one or more font groupings, mi is another font grouping of one or more font groupings; d is a matrix representation of the recorded seismic data, ri is a matrix defining excitation times of a first source type from one or more seismic source types, and ri is a matrix defining excitation times of another source type one or more types of seismic source.

[0155] A twentieth aspect may include the method of the eighteenth or nineteenth aspect, wherein at least one seismic source of the at least two interfering seismic source excitations has altered a signal or phase of its seismic source excitation.

[0156] The particular embodiments disclosed above are illustrative only, as the present disclosure can be modified and practiced in different ways, however, evident equivalents to those skilled in the art having the benefit of the teachings in the present document. Furthermore, no limitation is intended to the details of construction or design herein shown, other than as described in the claims below. Therefore, it is evident that the above particular illustrative embodiments can be altered or modified and such variations are considered to be within the scope and spirit of the present disclosure. Alternative modalities that result from the combination, integration and/or omission of modality features (or modalities) are also within the scope of the disclosure. Although compositions and methods are described in the broader terms of "having", "comprising", "containing" or "including" various components or steps, compositions and methods may also "consist essentially of" or "consist of" various components and steps. The use of the term "optionally" in relation to any element of a claim means that the element is required, or alternatively, the element is not required, both alternatives falling within the scope of the claim.

[0157] Numbers and ranges revealed above may vary by some amount. Whenever a numerical range with a lower limit and an upper limit is revealed, any number and any range included within the range are specifically revealed. In particular, every range of values (of the form, "from about a to about b" or, equivalently, "from approximately a to b," or equivalently "from approximately a-b") disclosed herein shall be understood to represent every number and range within the broader range of values. Furthermore, terms in the claims have their pure, ordinary meaning unless otherwise explicitly and clearly defined by the claimant. Furthermore, the indefinite articles "a" or "an", as used in the claims, are defined herein as meaning one or more than one of the element it introduces. If there is any conflict in the use of a word or term in this specification and one or more patents or other documents, definitions that are consistent with this specification shall be adopted.

[0158] Numerous other modifications, equivalents, and alternatives will become apparent to those skilled in the art once the above disclosure is fully appreciated. The following claims are intended to cover such modifications, equivalents and alternatives where applicable. Consequently, the scope of protection is not limited by the above description, but is limited only by the following claims, where such scope includes the subject matter equivalents of the claims.

Claims (20)

1. Method of seismic exploration above a subsurface region containing structural or stratigraphic features conducive to the presence, migration or accumulation of hydrocarbons, the method being characterized by comprising: accessing at least a portion of a mixed seismic source survey, in which the mixed source seismic survey contains at least two interfering seismic source excitations therein, wherein the seismic source excitations are produced by at least two seismic sources having different signatures and different frequency characteristics; separating the at least two interfering seismic source excitations using inversion separation based on the different signatures and different frequency characteristics of the at least two seismic sources; producing a plurality of groupings of separate sources based on the separation; and using the plurality of separate source arrays to exploit hydrocarbons within said subsurface region. 2. Method according to claim 1, characterized in that separating the at least two interfering seismic source excitations comprises: selecting a first coherence constraint for a first seismic source from the at least two seismic sources; selecting a second coherence constraint for a second seismic source from the at least two seismic sources; and applying the first coherence constraint and the second coherence constraint in inversion separation to the portion of the merged seismic source survey, wherein the production of the plurality of source groupings is based on applying the first coherence constraint and the second coherence constraint coherence. 3. Method according to claim 2, characterized in that the first consistency constraint and the second consistency constraint are different. 4. Method, according to claim 1, characterized in that separating the at least two interfering seismic source excitations comprises: determining activation times for each interfering seismic source excitation of the at least two interfering source excitations; and using the times of activations corresponding to each seismic source excitation to shift the at least two interfering seismic source excitations to a zero time, wherein the production of the plurality of source clusters is based on the displacement of the at least two source excitations Interfering seismic data for time zero. 5. Method, according to claim 1, characterized in that separating the at least two interfering seismic source excitations using inversion separation comprises solving an equation having the form: d = ri mi + r m2, for the matrices m1 and m2 for producing the plurality of font arrays, wherein m1 is a first font array among the plurality of font arrays, m2 is a second font array of the plurality of font arrays; d is a matrix representation of the recorded seismic data, ri is a matrix defining excitation times of a first source of the at least two seismic sources, and r is a matrix defining excitation times of a second source of the at least two seismic sources . 6. Method, according to claim 1, characterized by the fact that the plurality of source clusters each have seismic data representative of a different frequency range. 7. Method according to claim 1, characterized in that it further comprises; performing a full waveform inversion analysis on at least one of the plurality of source arrays. 8. Method, according to claim 1, characterized by the fact that each of the at least two interfering seismic source excitations is separated in time by a random period of time. 9. Method, according to claim 1, characterized in that the at least two seismic sources are different and comprise a combination of at least two of: vibrational sources, air gun sources, spark gap sources and frequency sources low. 10. Method according to claim 1, characterized in that at least one seismic source of the at least two interfering seismic source excitations has altered a signal or phase of its seismic source excitation. 11. Method of seismic exploration comprising: initiating a plurality of seismic source excitations above a subsurface region containing structural or stratigraphic features conducive to the presence, migration or accumulation of hydrocarbons, wherein the plurality of seismic source excitations is produced by at least two seismic sources that have different signatures and different frequency characteristics; and producing a mixed source seismic survey comprising data obtained from reflections generated by the plurality of seismic source excitations, wherein the mixed source seismic survey contains at least two interfering seismic source excitations therein, wherein the at least two interfering seismic source excitations are separated using inversion separation based on the different signatures and different frequency characteristics of the at least two seismic sources, and wherein a plurality of source groupings are produced based on the separation of the at least two seismic source excitations interfering seismic source. 12. Method, according to claim 11, characterized in that the at least two interfering seismic source excitations are separated using inversion separation by solving an equation having the form: d = ri mi + r m2, for the matrices m1 and m2 for producing the plurality of font arrays, wherein m1 is a first font array among the plurality of font arrays, m2 is a second font array of the plurality of font arrays; d is a matrix representation of the recorded seismic data, ri is a matrix defining excitation times of a first source of the at least two seismic sources, and r is a matrix defining excitation times of a second source of the at least two seismic sources . 13. Method according to claim 11, characterized in that the plurality of source clusters each have seismic data representative of a different frequency range. 14. Method according to claim 11, characterized in that each of the at least two interfering seismic source excitations is separated in time by a random period of time. 15. Method of seismic exploration above a subsurface region containing structural or stratigraphic features conducive to the presence, migration or accumulation of hydrocarbons, the method being characterized by comprising: accessing at least a portion of a mixed seismic source survey, in which the mixed source seismic survey contains at least two interfering seismic source excitations therein, wherein the seismic source excitations are produced by at least a first seismic source and a second seismic source having different signatures and different frequency characteristics, wherein at least at least one of the first seismic source or the second seismic source is a low frequency source; separating the at least two interfering seismic source excitations using inversion separation based on the different signatures and different frequency characteristics of the at least the first seismic source and the second seismic source; producing a plurality of separate source arrays based on the separation, wherein at least one of the plurality of separate source arrays contains low frequency data or some missing frequencies; and using the plurality of separate source arrays to exploit hydrocarbons within said subsurface region. 16. Method according to claim 15, characterized in that separating the at least two interfering seismic source excitations comprises: selecting a first coherence constraint for a first seismic source from the at least two seismic sources; selecting a second coherence constraint for a second seismic source from the at least two seismic sources; and applying the first coherence constraint and the second coherence constraint in inversion separation to the portion of the merged seismic source survey, wherein the production of the plurality of source groupings is based on applying the first coherence constraint and the second coherence constraint coherence. 17. Method, according to claim 15, characterized in that separating the at least two interfering seismic source excitations using inversion separation comprises solving an equation having the form: d = ri mi + r m2, for the matrices m1 and m2 for producing the plurality of font arrays, wherein m1 is a first font array among the plurality of font arrays, m2 is a second font array of the plurality of font arrays; d is a matrix representation of the recorded seismic data, ri is a matrix defining excitation times of a first source of the at least two seismic sources, and r is a matrix defining excitation times of a second source of the at least two seismic sources . 18. Seismic exploration method above a subsurface region containing structural or stratigraphic features conducive to the presence, migration or accumulation of hydrocarbons, the method being characterized by comprising: accessing at least a portion of a mixed seismic source survey, in which the mixed source seismic survey contains at least two interfering seismic source excitations therein, wherein the seismic source excitations are produced by at least two seismic sources having different signatures and different frequency characteristics; separating the at least two interfering seismic source excitations using inversion separation based on the different signatures and different frequency characteristics of the at least two seismic sources; producing a plurality of groupings of separate sources based on the separation; and using the plurality of generated source clusters to mine for hydrocarbons within said subsurface region. 19. Method, according to claim 18, characterized in that using the plurality of generated source groupings comprises solving an equation having the form: d = ri mi + r m2, for the matrices m1 and m2 to produce the plurality of font arrays, wherein m1 is a first font array among the plurality of font arrays, m2 is a second font array of the plurality of font arrays; d is a matrix representation of the recorded seismic data, ri is a matrix defining excitation times of a first source of the at least two seismic sources, and r is a matrix defining excitation times of a second source of the at least two seismic sources . 20. Method according to claim 18, characterized in that at least one seismic source of the at least two interfering seismic source excitations has altered a signal or phase of its seismic source excitation.