SeisReconNO: Leveraging a U-Net-Enhanced Fourier neural operator for 3D seismic reconstruction

This paper describes the full cycle of 4D seismic data integration comprised of workflows related to 4D data analysis, quality control of reservoir models and reservoir model updating using both 4D seismic and well production data. These workflows are applied to a deepwater field, where high quality 4D seismic data is available. In the first step, we analyze 4D seismic data and extract multiple attributes to image changes in reservoir properties. Next, we apply different workflows which link 4D seismic data with the reservoir model. Finally, we update the reservoir model automatically by simultaneously honoring the 4D seismic and well production data. We use a novel approach which incorporates 4D seismic amplitude differences without explicitly modeling the full physics in a joint history matching workflow. Introduction Reservoir monitoring using 4D seismic data is becoming an increasingly important tool for reservoir management (Calvert, 2005). Nevertheless, the quantitative integration of both 4D seismic and historical production data into reservoir simulation models is a challenging task, which recently has become an active direction of research. Huang et al. (1997) applied a stochastic optimization method to minimize the mismatch between synthetic and observed seismic data over a reservoir to achieve simultaneous history-matching of 4D seismic and well-by-well production data. Landa (1997) proposed a gradient-based method to integrate both 4D seismic and pressure transient data. Stephen et al. (2006) developed a workflow for multiple-model history matching through simultaneous comparison of spatial information extracted from 4D seismic data as well as individual well-production data. Employing the Neigbourhood Algorithm (NA) as the sampling engine this workflow was applied to the North Sea Schiehallion field. Skjervheim et al. (2007) presented a version of the Ensemble Kalman Filter (EnKF) for continuous model updating capable to match a combination of production and 4D seismic data. They tested the method on a synthetic case and a North Sea field case. Jin et al. (2007, 2008) proposed the combination of the Very Fast Simulated Annealing (VFSA) method with pilot-point parameterization to solve the 4D seismic history-matching inverse problem and applied the workflow to a synthetic case. Castro (2006) proposed a probabilistic approach to perturb a high-resolution 3D geocellular model for integrating data from diverse sources, such as well logs, geological information, 3D/4D seismic, and production data. This workflow was successfully applied on a reservoir of the Oseberg field. Jin et al. (2011) also proposed a flood front based 4D seismic history matching workflow. In this paper, we present a case study of the full cycle of 4D seismic data integration ranging from basic and qualitative 4D attribute analysis to the advanced 4D seismic history matching workflow. 4D seismic attributes analysis This workflow provides an analysis of 4D seismic differences related to changes in reservoir properties. First, timeshifts between baseline and monitor 3D seismic volumes are computed through cross-correlation. Some initial data preparation usually takes place before the cross-correlation of the datasets, including automatic gain control and trace stacking for signal to noise ratio enhancement. Next, the computed timeshift is removed from the monitor survey in order to obtain meaningful 4D difference attributes. At that point, 4D seismic attributes can be extracted from a given gate around time horizons of interest - usually reservoir tops. For reservoirs with large lateral thickness change, top and base reservoir horizons should be used for attribute calculation. Different types of attributes can be extracted, such as the root mean square (RMS), the normalized RMS difference (NRMSD), etc.

Seismic data incorporation in reservoir simulation models history matching (HM) studies has been continuously growing. 4D seismic data, in contrast with well production data, can provide a very good scenario of fluids arrangement along reservoir. In this work we describe how 3D and 4D seismic data gathered in acquisitions performed in Campos Basin was incorporated in Marlim Sul deep water field geological model reconstruction and in assisted HM (AHM). It is taken advantage of both 3D and 4D seismic data in several stages of the study, for instance, in the construction of a new porosity – most influential in impedance – model by using a methodology based on the inversion of synthetic seismic (calculated by petro-elastic model) in porosity through an optimization process that aims to reduce the difference between observed and synthetic impedance, and when defining influential parameters based on fluids displacement registered by seismic signal, by using a technique based on the creation of transmissibility multipliers parameters regions that considers the fluids displacement shown in 4D signal. Another relevant point is the use of information from reservoir and 3D seismic data when weighting the 4D data in the objective function. Combining the above mentioned techniques with the knowledge of the field – supported by the 3D seismic data – which allowed, for instance, identification of faults – where fault transmissibility multipliers were used as parameters in the HM process – a fairly good agreement on the observed well and seismic production data was achieved. HM studies using AHM tools have been shown a much more time-efficient technique when compared to manual HM. The incorporation of 4D seismic data can considerably improve the HM quality by improving the reservoir description, once it increases the ability of describing fluids arrangement and pressure distribution. The techniques successfully applied in the Marlim Sul field HM support these conclusions.

Substituting petro-elastic model with a new proxy to assimilate time-lapse seismic data considering model errors

In some oilfields with 3D seismic data, the deeper structure cannot be observed due to poor quality deep seismic data. Layer stripping using both seismic and gravity data is a solution for this problem but it cannot get satisfactory results because the horizontal variations in formation density are ignored. We present a variable-density formation separation technique to address this problem. Based on 3D seismic depth data and laterally-variable density derived from 3D seismic velocity data, the upper formation gravity effect is calculated by forward modeling and removed from the Bouguer gravity. The formation-separated gravity anomaly with variable density is obtained, which mainly reflects the deeper geological structure. In block XX of North Africa, the shallow formations seismic data is excellent but the data at the top of basement is poor. The formation-separated gravity anomaly processed under the control of 3D seismic data fits well with the known seismic interpretation and wells. It makes the geological interpretation more reliable.

The fracture direction and its intensity are critical properties related to hydrocarbon characterization and identification. Both these properties have an essential role in identifying the direction of hydrocarbon migration, determining the sweet spot area, and optimizing the drilling design. The velocity variation with azimuth (VVAz) is a well-known method to estimate the fracture direction and its intensity. This method is of widespread interest because it predicts the properties based on seismic data without any practical constraints. Despite this interest, the technique requires rich azimuth 3D seismic data in our case, which is rare. This study aims to apply regularization and interpolation by including the wave front attributes based on the Common Reflection Surface (CRS) method before the VVAz inversion. The motivation of using the CRS method is to enrich the current azimuth of the 3D seismic data and improve the S/N ratio. The synthetic and the real 3D seismic data are evaluated to examine the interpolation scheme of the proposed CRS method’s performance. Based on the evaluation of the 3D seismic data after regularization, the amplitude versus offset (AVO) phenomena, and the VVAz inversion results are relatively consistent (or matched) with the model. A similar result is found for the case of real 3D seismic data. A significant positive correlation between the fracture intensity of FMI and the real seismic data of about 0.9 is obtained. Therefore, CRS can be used as a regularization and interpolation method before the VVAz inversion of the relatively narrow azimuth 3D seismic data.

Some difficulties are frequently associated with the integration of seismic and flow simulation datasets, especially related with the low vertical resolution of the seismic data and the uncertain estimations in the areas between the wells. One challenge is then the integration of both datasets in different scales, in order to take advantage of their characteristics. The present study proposes a redistribution of the reservoir saturation estimated with 4D seismic inversion methods, combining the information provided by the flow simulation in order to improve the quality of the estimations and their resolution. The methodology comprehends a saturation redistribution algorithm that is applied to each reservoir block combining the information of two saturation maps: one derived from the 4D seismic data and the other estimated by the flow simulator. The final saturation estimation follows the vertical distribution given by the simulation data but keep the average behavior observed in the 4D seismic. In order to have a better control of the results, the methodology is applied to a synthetic dataset that includes a reservoir model with different grid resolutions: geomodel, simulation model and seismic model. Two different case studies present the main results of the proposed methodology to improve the saturation predictions. The first case study represents an ideal solution being assumed that the base model (simulation model) is very similar to the reference model (geomodel that represents the true answer), remaining a few differences due to the different scales. In the second case study, the base model is selected between multiple realizations during the uncertainty reduction process. The results shows that is possible to improve the resolution of the saturation variation maps computed from 4D seismic data allowing the identification of new fine scale heterogeneities and providing better estimations of the saturation changes due to production. This methodology can also give additional clues in future history matching procedures through the identification of critical regions. With the continuous calibration of the models during a history matching process the results obtained with the redistribution method tend to improve, approaching the results obtained in the first case study. The proposed redistribution method combines the best characteristics of the seismic and simulation data. It includes the higher sensibility of the seismic data to identify the areal distribution of the main anomalies and inserts the higher sensibility of the simulation data regarding the identification of vertical water flow trends due to gravitational effects. Thus, the procedure introduces, in the maps provided by 4D seismic, new information regarding the injection/production patterns that become more and more reliable as we approach the wells.

Strong ocean current influences a marine seismic survey and forces the streamer off-course from the survey line. The sideway drift of the streamer results in that the reflection data are no longer distributed in common midpoint gathers along the survey line but become swath distribution on one side of the ship track. This effect is known as “streamer feathering” which degrades the profile image of the 2D processed seismic data. However, if we have long streamer or closely spaced parallel 2D seismic survey lines, we may turn this deleterious effect into a good opportunity to generate 3D seismic volumes with swath distributed reflection data. We present two case studies in which 2D seismic data were collected offshore eastern Taiwan where the strong Kuroshio Current heavily influenced the ship speed and caused large streamer feathering. The first case is a large-offset 2D seismic profiling data collected using a 6-km long streamer. We processed the swath part of the reflection data in 3D that not only avoids the inappropriate smearing effect in 2D data processing but also generates a 3D seismic volume to help the seismic interpretation. In the second case, we adjusted our 2D survey strategy when realizing that strong Kuroshio Current was causing significant streamer feathering, and collected a set of closely spaced parallel 2D seismic lines. This multi-swath dataset covers a broad area which enables us to generate a 3D seismic volume. Since our datasets are not real 3D seismic data, we have tailored our processing flows to deal with different data configurations and limitations of each dataset. Our results show that not only we have enhanced 2D seismic images of the originally-interested survey lines, but also provide information on 3D geometry of the geological features imaged. The benefits and limitations of utilizing the streamer feathering effect to generate 3D seismic volumes from 2D seismic profile data are reported. Overall, this approach is a considerable way to handle 2D seismic data with large streamer feathering for both avoiding unreliable 2D seismic images and obtaining information on 3D geometry of the geological features imaged.

Traditionally, integration between 4D seismic (4DS) and simulation data has been performed considering the 4DS data deterministically. However, there are uncertainties in the response of seismic. The goal of the methodology presented in this work is to compare the changes of dynamic properties estimated from 4DS and simulation models considering the uncertainties inherent to both data. The relevant reservoir uncertainties can be combined to generate multiple simulation models, which provide maps of dynamic changes, such as pressure change (Δp) and water saturation variation (ΔSw). Available 4DS can also be used to map dynamic changes. Through a stochastic seismic inversion, multiple ΔSw and Δp maps can be obtained from 4DS. After selecting a proper scale (scale transference), we compare the dynamic maps from seismic and simulation data using probabilistic density functions (PDFs), establishing levels of agreement/disagreement between 4DS and simulation data. To validate the methodology we use a synthetic dataset, with moderate complexity and seven uncertainties mapped, such as fault transmissibility, porosity, facies, and permeability. 500 maps of ΔSw and Δp from 4D seismic were generated from prior probabilistic seismic inversion. 500 simulation models previously calibrated using well production data generated the set of 500 maps of ΔSw and Δp from simulation. Applying the methodology, we identify four regions: (1) reservoir locations where both estimates (seismic and simulation) are similar, showing regions properly calibrated, (2) locations where simulation estimates are more precise than 4D seismic, (3) reservoir locations where the data sets indicate divergent estimates, and (4) 4DS estimates are more precise than simulation. This information can be very useful to guide data integration. As an example, we show that region (4) can be used to select the simulation models that reproduce ΔSw or Δp behavior from 4DS, since 4D seismic data is more precise than the simulation estimates in this region. Other useful information from the proposed methodology is that the reservoir zones identified as region (2) can be used as a constraint to reinterpret 4D seismic data, as simulation estimates are more precise. The methodology is a new way to evaluate the information from 4D seismic and simulation data considering uncertainties. The identification of these four regions can be useful in the parametrization phase of the history matching procedure (a complex process), as an additional tool to understand the properties in this procedure. The methodology also indicates possible locations to use reservoir engineering constraints to improve seismic interpretation, in regions where estimates from simulation are more precise than 4D seismic data. Moreover, we can use the methodology to determine critical reservoir locations to be reevaluated, those presenting disagreement between the two data source.

Summary Pre-stack depth migration (PSDM) is one of the most important technologies for accurate subsalt structure imaging. However, in the case of huge salt domes and low signal-to-noise ratios (SNR) of subsalt data, the key to accurate imaging of subsalt structure using PSDM is to establish an accurate velocity model. This paper presents a joint solution of 3D gravity and 3D seismic data. The SNR of seismic data above the several huge salt domes in the area is high, while the SNR below the salt domes is relatively low. Since the target is below the salt domes, the PSDM data can't meet the needs of subsalt structure interpretation. We therefore carried out 3D gravity exploration to solve subsalt structural imaging accuracy problems jointly using 3D gravity data and 3D seismic data. Using the inversion result of 3D gravity formation separation, the cause of clutter in seismic reflections on the interior of salt domes is illustrated and the initial velocity model provided by 3D seismic data is revised in the study area. Through this revised initial velocity model, a new pre-stack depth migrated section with improved subsalt reflection SNR as well as better subsalt structure imaging is achieved. Along with the work flow of joint solution of gravity and seismic data, this paper also gives prerequisites for formation separation technique. Introduction Interval velocity modelling plays a key role in imaging subsalt structure using PSDM technology. Generally, the interval velocity modeling covers two parts: the first is to establish an initial interval velocity model using pre-stack time-migrated seismic data. After PSTM and horizon calibration in the time domain of seismic data, the RMS velocity field as well as the structural model in time domain can be acquired. Once these data are determined, a constrained inversion as well as time-to-depth conversion can be applied to the RMS velocity field to acquire the initial interval velocity model in depth domain. The second part involves using the already established initial interval velocity model in depth domain to apply PSDM to the test line and evaluate the migrated section geophysically and geologically. This is an important step to modify the initial velocity model in depth domain to reasonable solution. The second step is a complex iterative processes, through which the final interval velocity model in depth domain for PSDM processing in the whole work area can be obtained. Obviously, the higher the SNR of seismic data, the finer the layer calibration in the time domain, and the more accurate the established initial interval velocity model. Here, the layer calibration in time domain isn't the chronostratigraphic calibration of interfaces but the interval velocity variation calibration of interfaces, taking corresponding well logging data and underground geological features as constraints for inversion of the RMS velocity field. One of the criterions to judge the reasonability of initial or inverted interval velocity model is the leveled reflection events in the CRP gather. Therefore, we can't judge whether the interval velocity model is reasonable or not if the SNR in the CRP gather is too low. Thus, the SNR of seismic data is an important factor in determining the precision of interval velocity model.

Seismic method is one of geophysical method that is used to determine subsurface condition of the earth. There are two types of seismic data, 2D seismic and 3D seismic. The advantages of using 3D seismic data is it has less uncertainty for imaging the earth subsurface. However, the cost to acquire 3D data is higher than producing 2D data, thus some company needs to invest more money to be able to get a better data quality. One of the ways to have 3D seismic data without having to spend a lot of money is by creating a 3D pseudo seismic data from 2D seismic data. The purpose of this research is to create a 3D pseudo seismic data by using a geostatistical method. The geostatistical method that is used in this research is Kriging interpolation. By the help of Petrel software, kriging interpolation is done to create the 3D seismic. The pseudo 3D seismic results show a good image such as difference in amplitude, also structural geology features of subsurface. In conclusion, the use of kriging interpolation in pseudo 3D seismic still needs more work and development so the results will be good for the next interpretation steps.

Seismic interpretation is an essential factor in the success of field development projects because it allows geologists and engineers to better understand subsurface formations and identify potential resources; however, noisy seismic data, especially in regions with complex geological structures, frequently hinder this process. Such noise emerging from equipment limitations, environmental interference, and inherent geological complexities can obscure essential geological features, making accurate subsurface imaging challenging and reducing the effectiveness of traditional interpretation methods. This paper presents a machine learning (ML) method to enhance the conditioning of 2D and 3D post-stack seismic data by targeting noise and migration artifacts. These issues are often responsible for compromising the quality of seismic data sets. The method this paper presents leverages convolutional neural networks (CNNs) to identify and correct spatial inconsistencies within the data. With minimal preprocessing, requiring only the conversion of data into a 3D NumPy array format, this approach simplifies the workflow while improving data clarity and accuracy. Early statistical analysis of outliers and data normalization provides the ML model with stable inputs, allowing it to iteratively train on raw data and infer the original signal patterns in an unsupervised manner without the need for labeled data sets. The model outputs a conditioned version of the seismic data, significantly reducing noise while preserving critical geological features. The results demonstrate that ML seismic conditioning improves fault imaging, horizon continuity in areas with low signal-to-noise ratios (SNR), and localized smoothing without compromising resolution. Furthermore, the method presented in this paper retains signal fidelity and relative amplitude strengths, ensuring that geological interpretations remain accurate and reliable. By automating the traditional manual task of data conditioning, the proposed solution optimizes interpreter workflows, allowing them to focus on higher-level geological analysis and decision-making. This approach ultimately increases the accuracy of subsurface interpretations and resource exploration, improving the overall quality and efficiency of seismic interpretation projects. In addition, for 2D seismic, traditional methods of denoising rely on expert judgement using tools such as f-x deconvolution. 2D CNNs offer an efficient alternative, especially when training data can be synthesized and leveraged in supervised learning. This method can help the CNN identify and remove noise while preserving pertinent subsurface structures. This paper presents a 2D CNN architecture containing a model trained with synthetic seismic data to improve SNR of field seismic data; hence, interpretation.

Crosswell seismic data and 3D seismic data are seismic response of different frequency band from the same geologic target. Crosswell seismic data possesses very high resolution in comparing with 3D seismic data for its higher main frequency and broader frequency band. However crosswell seismic data only exist between source well and receiver well, while 3D seismic data has excellent spatial distribution. It is possible to integrate different types of data and get more reliable underground information (Lines et al., 1988; Hu et al.,2007; Heinche et al., 2006). Here we put forward joint frequency expanding (JFE) method of crosswell seismic data and 3D seismic data. We expand 3D seismic data frequency band according to crosswell seismic data, and improve the resolution ability of 3D seismic data. First we use 3D seismic data and crosswell data to build 3D seismic reflect coefficient model, which possesses high resolution, and is consistent with 3D seismic data in structure. Second we use this reflect coefficient model as constraining condition, and joint inversion algorithm is used to get inverted reflect coefficients from 3D seismic data. Finally according to time‐frequency analysis of 3D seismic data, we choose an appropriate high resolution wavelet to convolve with the inverted reflect coefficients, and produces high resolution 3D seismic data. Here we design a processing flow and applied it to Ken 71 data of ShengLi oil field. With the help of crosswell seismic data, 3D seismic data resolution is improved, main frequency is increased from 35Hz to 75Hz, and frequency band is also widened by 30Hz.

There is new and growing enthusiasm for hydrocarbon exploration in the Barents Sea after three recent discoveries: Skrugard, Havis, and Norvarg. We demonstrate how using wide-azimuth 3D controlled-source electromagnetic (CSEM) and 2D seismic data together can improve the identification of prospective areas in the region. An interpretation workflow is presented to show how the challenging resistivity background in the Barents Sea is handled. To do this, we introduce a new inversion attribute called the anomalous vertical resistivity. The inversion attribute is co-visualized with 2D seismic data and used to estimate recoverable reserves. The workflow is illustrated on both synthetic and real data for the Skrugard and Havis discoveries. Both discoveries are identified on CSEM maps and by integrating CSEM data with seismic data. In addition, a new lead is identified. We show that CSEM data carries structural information and that the horizontal resistivity trends can be used with 2D seismic data and well logs to interpret the distribution of good-quality sands. Hydrocarbon reserve calculations based on the 3D CSEM data for the Skrugard and Havis discoveries have P50 values consistent with the publicly available reserve estimates. The reserve estimate for the new lead also shows significant potential.

Seismic reflection studies have been intensively carried out in the Nankai Trough region. However, the role of oceanic crust was not well understood in the plate convergent margin. Recently, Tsuji et al. [2009] identified intraoceanic faults developed as imbricate structures within the subducting Philippine Sea plate off the Kii Peninsula in central Japan manifesting as strong-amplitude reflections observed in an industry-standard 3D seismic reflection data set. Here we use several 2D and 3D seismic reflection data acquired in the whole Nankai Trough region in order to discuss characteristics of intraoceanic faults distributed in the Nankai Trough region. Seismic profiles demonstrate that intraoceanic faults are densely distributed in the Nankai Trough east of the Cape Shionomisaki. Large displacements of a major intraoceanic faults elevate the crust surface, and the offset due to cumulative displacements reaches >1 km. These imbricate intraoceanic faults cut through the oceanic crust as a discontinuous thrust plane. The intraoceanic faults strike nearly parallel to the trend of the trough axis. However the fault traces are bending at the western termination; the fault planes extend upward from side edges of the underlying intraoceanic faults and work as lateral faults. The deformation along the intraoceanic faults may have continued until recently because the shallow sediment as well as the seafloor is deformed due to the fault displacement. Furthermore, the locations of the intraoceanic faults recognized in the seismic data are distributed around the estimated hypocenters of the mainshocks and aftershocks of the 2004 intraplate earthquakes (Mw >7), and their geometry extracted from the 3D seismic data could explain the kind of complex rupture pattern observed during the 2004 events. These observations demonstrate that the intraoceanic faults should be seismogenically active. Furthermore the segmentation of interpolate earthquake off the Cape Shionomisaki is consistent with the ridge originated by the displacement of intraoceanic fault. Because the displacement along the intraoceanic fault is developed with subduction and cuts the plate boundary faults due to their dynamic displacements, there is a possibility that the intraoceanic faults control the interplate earthquake segmentation.

Reconstruction of missing seismic traces based on sparse dictionary learning and the optimization of measurement matrices