Method and device for detecting a subsurface boulder based on acquired full wavefield seismic data

Abstract

A method for detecting a subsurface boulder based on acquired full wavefield seismic data of a region is disclosed. The method is performed by a processor and comprises the steps of: obtaining diffraction data by applying seismic diffraction separation to the full wavefield seismic data, and detecting a boulder by identifying a diffraction pattern representing the boulder from the obtained diffraction data; wherein the full wavefield seismic data comprises 3D ultra-ultra-high-resolution, UUHR, seismic wavefield data. Unlocking insights from Geo-Data, the present invention further relates to improvements in sustainability and environmental developments: together we create a safe and liveable world.

Description

METHOD AND DEVICE FOR DETECTING A SUBSURFACE BOULDER BASED ON ACQUIRED FULL WAVEFIELD SEISMIC DATAFIELD OF THE INVENTION

[0001] The present disclosure generally relates to boulder detection, and more specifically to a method and device for detecting a subsurface boulder based on acquired full wavefield seismic data of a region. Unlocking insights from Geo-Data, the present invention further relates to improvements in sustainability and environmental developments: together we create a safe and liveable world.BACKGROUND OF THE INVENTION

[0002] Detecting subsurface boulders is a critical aspect of geotechnical and engineering projects, such as construction, tunnelling, and resource extraction. These obstacles can significantly affect drilling operations, foundation stability, and excavation processes. Effective identification and characterization of subsurface boulders mitigate risks and ensure safer, more efficient project execution.

[0003] Boulder detection for offshore wind farm installations presents unique challenges compared to traditional boulder detection in other projects, such as tunnelling or onshore construction. In offshore wind farms, detecting boulders is critical for ensuring the stability and proper alignment of monopile or jacket foundations, as even shallow boulders, which tend to be small, can obstruct piledriving operations. In this sense, detecting smaller boulders (e.g., ~50 cm) with high precision is needed for offshore wind farm, whereas traditional projects may focus on larger obstacles or broader subsurface characterization.

[0004] As an example, side-scan sonar focuses on sea-floor boulders and struggles with detecting small boulders because of its resolution limits, especially at greater distances from the sonar device. While larger boulders produce strong acoustic reflections and cast clear acoustic shadows, smaller boulders may blend into background noise or be misinterpreted as seabed roughness. Furthermore, in environments with uneven seabed textures or high levels of clutter, distinguishing small boulders from other features becomes even more challenging.

[0005] Sub-bottom profders target on shallow boulders up to 20 meters below sea floor. It faces difficulties in resolving small boulders due to their limited resolution and signal penetration constraints. Low-frequency waves, which are necessary for deeper penetration, tend to smooth out or bypass small objects, making their acoustic signature less distinct.

[0006] Seismic-based methods provide greater adaptability but encounter difficulties with small-scale targets due to the scale mismatch between seismic wavelengths and the boulders. High-frequencyseismic waves, which could interact with small boulders, attenuate rapidly, limiting their penetration depth. Additionally, the presence of noise, overlapping reflections, and interference from larger geological structures can obscure the subtle diffractions associated with small boulders.

[0007] In view of the above, there is a need for an improved method and device for detecting a subsurface boulder.BRIEF SUMMARY OF THE INVENTION

[0008] In one aspect of the invention there is provided a method for detecting a subsurface boulder based on acquired full wavefield seismic data of a region, the method is performed by a processor and comprises the steps of:

[0009] - obtaining diffraction data by applying seismic diffraction separation to the full wavefield seismic data, and

[0010] - detecting a boulder by identifying a diffraction pattern representing the boulder from the obtained diffraction data,

[0011] wherein the full wavefield seismic data comprises 3D stacked ultra-ultra-high-resolution, UUHR, seismic wavefield data.

[0012] The method of the present disclosure is based on the inventor's insight that subsurface boulders can be detected more accurately by analysing diffraction patterns within high-resolution seismic data. Traditional methods for boulder detection often rely on conventional reflection seismic techniques or lower-resolution data, which can miss small or deeply buried boulders, particularly in complex geological environments. The inventor recognized that diffractions caused by small-scale irregularities, such as boulders, carry unique signatures that can be isolated using seismic diffraction separation techniques, which can be used to detect subsurface obstacles such as boulders.

[0013] Based on the method of the present disclosure, first diffraction separation is applied to 3D stacked UUHR seismic wavefield data to obtain diffraction data. This process isolates the diffraction energy, which is especially sensitive to small and irregular subsurface features like boulders. Thereafter, boulder(s) are identified by detecting specific diffraction patterns in the obtained diffraction data.

[0014] By identifying these patterns, the method can pinpoint the exact location and characteristics of the boulder, even in cases where it is deeply buried, at for example up to 100 meter below seafloor, or of small size. 3D UUHR seismic wavefield data as used in this method helps to provide detailed view of the subsurface including boulders in the size of for example half a meter

[0015] The UUHR seismic data enhances the resolution, allowing even relatively small boulders (e.g., about 50 cm in size) and relatively deeply buried boulders to be detected, which would be challenging for traditional seismic methods. Moreover, the ability to isolate and identify diffractions from boulders ensures that they are distinguished from other subsurface features, minimizing falsepositives. This is particularly beneficial in applications like offshore wind farm installation, where accurate boulder detection is critical for preventing pile-driving issues and ensuring smooth operations.

[0016] In an example of the present disclosure, the diffraction separation comprises the steps of:

[0017] - deriving reflection data from the full wavefield seismic data;

[0018] - obtaining the diffraction data by subtracting the derived reflection data from the full wavefield seismic data.

[0019] As can be contemplated by those skilled in the art, diffraction separation is achieved by first deriving reflection data from the full seismic wavefield data and then subtracting the reflection data from the full seismic data to isolate the diffraction data. Diffractions, which are often associated with irregular subsurface features, such as boulders, is thereby isolated.

[0020] By separating the diffraction and reflection components of the seismic data, the method enhances the sensitivity to small-scale subsurface features, such as boulders. This approach increases the reliability of boulder detection in complex geological environments, ensuring that features which might otherwise be overlooked in the full wave data comprising the reflection data can be effectively identified.

[0021] In an example of the present disclosure, the step of deriving reflection data comprises the steps of:

[0022] - deriving reflection for each of a plurality of volumes obtained from dividing the full wavefield seismic data by sliding a local window;

[0023] aggregating all reflection in the plurality of volumes to obtain the reflection data.

[0024] By dividing the large volume of seismic data into smaller, localized sections using sliding windows, it becomes possible to focus on specific regions and analyze features in more detail.

[0025] Local processing allows for a more targeted approach to identifying and characterizing subsurface features, such as boulders. Since seismic data can vary significantly across different depths or regions, processing smaller volumes helps to reduce the risk of averaging out important details that might be present in one specific area, shallow boulders that cause relatively weak diffraction may also be detected based on the local refinement.

[0026] In an example of the present disclosure, the step of deriving reflection for each volume comprises the step of:

[0027] - transforming the seismic data volume into frequency domain data using Fourier transform;

[0028] mapping the frequency domain data into a block Hankel matrix;

[0029] - performing local rank reduction on the block Hankel matrix using singular value decomposition, SVD;

[0030] reconstructing a block Hankel matrix of reflection based on the block Hankel matrix with reduced rank; and

[0031] - obtaining the reflection by applying inverse Fourier transform.

[0032] Those skilled in the art understand that rank reduction via SVD essentially filters out the lower-rank components (the reflections) and retains the higher-rank components (the diffractions). By applying local rank reduction, it preserves the lower-rank reflections, which allows the reflections to be subtracted from full wave to obtain the estimated diffraction. This makes it easier to isolate the diffraction patterns that correspond to small, localized subsurface features like boulders.

[0033] In practice, local rank reduction enhances the clarity of diffraction data, making it easier to identify small or irregular features that would otherwise be obscured by the larger and more dominant reflection signals. This is especially important in detecting small objector including boulders having a size of tens of centimetres.

[0034] In an example of the present disclosure, adaptive rank selection is performed in the SVD.

[0035] Instead of using a fixed rank, the method dynamically adjusts the rank based on the local data volume, allowing for more flexible and accurate signal processing. This flexibility ensures that the rank reduction is optimized for each window, leading to better noise reduction and more accurate reflection data, which ultimately aids in boulder detection.

[0036] In an example of the present disclosure, a rank of the block Hankel matrix of the volume is determined based on largest singular value ratio.

[0037] This involves analysing the distribution of singular values obtained through SVD and using their relative magnitudes to identify the optimal rank for separating reflection and diffraction energy.

[0038] The largest singular value ratio is particularly effective because it reflects the dominance of energy associated with coherent signals, such as reflections, compared to weaker, less structured signals, such as diffractions. Reflections, being more organized and higher in energy, correspond to the largest singular values in the decomposition having lower ranks. Diffractions, which are less dominant and lower in energy, correspond to smaller singular values having higher ranks. By evaluating the ratio of these values, the method can differentiate between the reflection energy (which tends to dominate the lower ranks) and the diffraction energy (which resides in higher ranks).

[0039] This adaptive rank selection based on singular value ratios ensures that the reflections are accurately modelled and removed while retaining the diffraction components. It allows the method to adapt dynamically to variations in the seismic data, providing an efficient and reliable approach to separate and isolate diffraction signals. This, in turn, enhances the accuracy of identifying subsurface features like boulders, particularly in complex environments where reflection and diffraction signals are closely intertwined.

[0040] In an example of the present disclosure, a size of the local window is statistically determined based on an averaged dimension of collected diffraction curves present in data samples.

[0041] Statistically determining the window size ensures that the window is appropriately sized to capture the relevant features in the seismic data. This leads to improved detection of small boulders, asthe method adapts to the scale of the diffraction curves, ensuring that no important information is overlooked.

[0042] In an example of the present disclosure, the diffraction curves are collected from the data samples using a machine learning method.

[0043] The use of machine learning allows the process of diffraction curve identification, which traditionally requires manual interpretation and can be tedious and time-consuming, to be automated. This speeds up the overall process and reduces the likelihood of human error, allowing for faster and more accurate boulder detection.

[0044] In an example of the present disclosure, the step of identifying a diffraction pattern is performed using an automated pattern recognition algorithm.

[0045] By using an automated pattern recognition algorithm to identify the diffraction pattem(s), the system can quickly and reliably identify diffraction patterns, even in large datasets, making the method more efficient and scalable.

[0046] In an example of the present disclosure, diffraction patterns in horizontal and vertical slices or profiles or sections are analysed.

[0047] Horizontal slices reveal circular features associated with boulders, while vertical slices, which are also referred to as vertical sections or vertical profiles, highlight hyperbolic patterns caused by their depth and wavefront curvature. By examining the data from multiple perspectives, the method enhances the accuracy of identifying boulders, as different views provide complementary insights into the geometry of the diffraction patterns.

[0048] This dual-perspective approach ensures a more reliable detection process by crossvalidating observations across dimensions. It minimizes false positives and provides a comprehensive understanding of boulder locations, which is particularly useful in complex subsurface environments.

[0049] In another example of the present disclosure, diffraction patterns identified in the horizontal and vertical slices are correlated to determine a spatial location of the boulder.

[0050] By ensuring that the identified patterns align across slices, the method eliminates ambiguities and provides high-confidence results. This step is particularly advantageous for UUHR data, where subtle features and noise may complicate interpretation.

[0051] In another example of the present disclosure, the diffraction pattern comprises at least one of a circular and a hyperbolic feature.

[0052] As can be contemplated by those skilled in the art, the diffraction pattern having circular feature is mainly visible in horizontal slices, and hyperbolic feature is mainly visible in vertical slices.

[0053] This distinction is significant because circular patterns help map lateral positions of boulders, while hyperbolic patterns provide depth and positional accuracy. The ability to recognize and utilize both shapes ensures robust identification and localization of boulders, enhancing the overall precision of subsurface mapping.

[0054] In an example of the present disclosure, the method further comprises generating a subsurface boulder map based on the locations of detected boulders.

[0055] The boulder map serves as a practical output for applications such as offshore wind farm pile installation, ensuring precise knowledge of boulder locations. By providing a clear and detailed map, this method significantly reduces the risk of unforeseen obstructions during construction, saving time and costs.

[0056] A second aspect of the present disclosure provides a device for detecting a subsurface boulder based on acquired full wavefield seismic data of a region, the device comprising a processor for performing the method according to the first aspect of the present disclosure.

[0057] A third aspect of the present disclosure provides a computer program product, comprising a computer readable storage medium storing instructions which, when executed on at least one processor, cause the at least one processor to carry out the method according to the first aspect of the present disclosure.

[0058] The above mentioned and other features and advantages of the disclosure will be best understood from the following description referring to the attached drawings. In the drawings, like reference numerals denote identical parts or parts performing an identical or comparable function or operation.BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to describe the manner in which the above-recited and other advantages and features of the disclosure can be obtained, a more particular description of the principles briefly described above will be rendered by reference to specific embodiments thereof which are illustrated in the appended drawings. Understanding that these drawings depict only exemplary embodiments of the disclosure and are therefore not to be considered to be limiting of its scope, the principles herein are described and explained with additional specificity and detail through the use of the accompanying drawings in which:

[0060] FIG. 1 schematically illustrates, in a flow chart type of diagram, a method 10 for detecting a subsurface boulder based on acquired full wavefield seismic data of a region according to an embodiment of the present disclosure.

[0061] FIG. 2 schematically illustrates, in a flow chart type of diagram, detailed steps of obtaining a reflection of a specific volume, according to an embodiment of the present disclosure.

[0062] FIGs. 3(a) - 3(c) respectively illustrate exemplary full wavefield seismic data, diffraction data obtained according to the method of the present disclosure and reconstructed reflection data in 3D.

[0063] FIGs. 4(a) - 4(c) respectively illustrate exemplary full wavefield time slice, diffraction time slide and reflection time slice for the 3D data of FIGs. 3(a) to 3(c); and

[0064] FIG. 5(a) - (c) illustrates zoomed-in diagram of the marked region in comparison to the original areas shown in Figures 4(a) to 4(c).DESCRIPTION OF ILLUSTRATIVE EMBODIMENTS

[0065] Embodiments contemplated by the present disclosure will now be described in more detail with reference to the accompanying drawings. The disclosed subject matter should not be construed as limited to only the embodiments set forth herein. Rather, the illustrated embodiments are provided by way of example to convey the scope of the subject matter to those skilled in the art.

[0066] Seismic data input to the method of the present disclosure generally refers to processed seismic data in the form of 3D stacked seismic data.

[0067] Applying diffraction separation to ultra-ultra-high-resolution, UUHR, seismic data introduces several challenges due to the nature of this data and the precision required for accurate analysis. Comparing to conventional exploration seismic data, UUHR data set is unique in that both the temporal and spatial sampling are extremely high, the frequency range is up to 4000 Hz, and the grid bin size is 0.5 meter * 1.0 meter. UUHR mainly focuses on boulders from about 20 meters to 100 meters below sea floor.

[0068] For offshore projects such as windfarm construction, the geological scale is usually within 1 km in space and can be up to 100 meter in depth. While UUHR data provides exceptional details, enabling the detection of small subsurface features like boulders, the data preprocessing steps are very challenging due to the huge influence of sea surface and wave heights.

[0069] In particular, the fine-scale features captured in UUHR data result in more intricate and overlapping reflection and diffraction patterns. Determining an optimal rank for separating reflections and diffractions can become more challenging, as the singular values associated with the two components may not exhibit clear separations.

[0070] Besides, for UUHR data, selecting an appropriate window size for local processing becomes more critical. Windows that are too small may fail to capture sufficient data to represent diffractions adequately. Large windows on the one hand include more complex structures, therefore it can be much more computationally intensive for SVD used in processing the data. Balancing window size to account for the high-resolution nature of UUHR data is a non-trivial task.

[0071] These issues are addressed by the method of the present disclosure, which allows detection of subsurface boulders to be performed in an efficient and reliable way.

[0072] The method of the present disclosure can be executed in a variety of environments, ranging from local computing systems to cloud-based platforms, depending on the scale and requirements of the processing task. For smaller datasets or specific projects, the method can be performed on individualworkstations or high-performance computers equipped with the necessary processing power and software.

[0073] For larger datasets, the method can be executed in cloud-based environments or distributed computing systems. Cloud platforms offer the scalability needed to process vast amounts of seismic data in parallel, enabling the handling of large datasets that span extensive geographic areas. Additionally, cloud platforms allow for efficient resource management and storage, making it easier to manage the computational demands of large-scale data processing tasks.

[0074] Generally, the method can be performed by a computing device having a processor, which is capable of executing the necessary algorithms for boulder detection based on seismic data. This computing device can range from personal workstations to high-performance servers, depending on the scale of the data processing task. Such device is not described here as the present disclosure is not limited to any particular hardware configuration.

[0075] Figure 1 schematically illustrates, in a flow chart type of diagram, a method 10 for detecting a subsurface boulder based on acquired full wavefield seismic data of a region according to an embodiment of the present disclosure. In the present disclosure, the acquired full wavefield seismic data comprises 3D stacked UUHR seismic data.

[0076] The method generally comprises detecting a boulder from diffraction data, while the diffraction data is obtained by performing diffraction separation on the obtained 3D UUHR full wavefield seismic data. The detection is done by identifying a pattern representing the boulder form the obtained diffraction data.

[0077] In seismic data processing, dividing the data into smaller volumes using a sliding local window is a widely adopted approach to handle the complexity and variability of subsurface signals. This method involves segmenting the full seismic dataset into overlapping windows, allowing for localized analysis of specific regions of the data. Each windowed volume represents a subset of the data that can be processed independently, enabling a more detailed and focused examination of subsurface features.

[0078] By isolating smaller volumes, the method helps mitigate the influence of large-scale variations and noise, making it easier to identify localized features such as fractures, voids, or boulders. This is especially beneficial in areas with complex geology, where subsurface features can vary significantly within short distances.

[0079] For UUHR seismic data, choosing an appropriate window size helps to generate more accurate processing result. Statistical methods or adaptive techniques can be used to determine optimal window sizes based on the characteristics of the UUHR data, such as the average dimension of diffraction curves.

[0080] In the present disclosure, only spatial window size is considered as we mainly aim at distinguishing between diffractions and reflections in the spatial dimension.

[0081] According to the present disclosure, a statistical approach is used determine the local window size based on the averaged dimension of diffraction curves observed in the data samples.

[0082] The diffraction curves represent wave propagation patterns caused by irregularities or features in the subsurface, such as boulders or fractures. These curves typically appear as circular or hyperbolic features in seismic data, with the size and shape of the curve being influenced by factors like the size of an object including a boulder and its distance from the seismic source. By examining diffraction curves presented in the stacked seismic data, it is possible to estimate average dimensions, such as a width or extent of the diffraction pattern. This measurement can then inform an appropriate size for the local window that will be used for further processing.

[0083] For instance, in regions where boulders are expected to be located relatively shallow which normally will give rise to relatively small diffractions, with a size of for example from 50cm to 1 meter, the diffraction curves associated with these boulders may have a smaller spatial extent. In such cases, the local window is selected to be small enough to capture the high-frequency details of the diffraction signal but large enough to avoid truncating the feature.

[0084] The statistical approach could involve analysing a set of diffraction curves, calculating their average width, and then using this average value as a guideline for setting the window size. The window size may be selected from a range surrounding the average value. This ensures that the window is optimized for detecting diffraction features while minimizing the inclusion of extraneous data from the surrounding subsurface.

[0085] Exemplary steps for selecting a local window size can comprise: statistically collect size in three dimensions of majority diffraction curves in terms of data samples; then determine averaged dimensions of majority diffraction curves, which is used to determine the local processing window. Data sample as used herein refers to grid points, in the stacked seismic profile.

[0086] It is noted that the overlapping window size in three dimensions is a trade-off parameter (e.g., by trial and error) to balance the computational efficiency and the edge artifacts.

[0087] An exemplary window size may be chosen to cover for example 20 to 50 data samples, which helps to produce more accurate detection result based on UUHR stacked seismic data,

[0088] The statistical method may also take into account the variability of the diffraction curves, such as a standard deviation of their dimensions. This allows for a more flexible window size that can adapt to the variations in diffraction characteristics. For example, in regions where relatively deep boulders are expected, the window size could be increased to accommodate the broader diffraction patterns, while in areas with shallow boulders, the window could be reduced to focus more precisely on the finer details.

[0089] In some instances, machine learning methods could be used to automate this statistical determination process. By training a model to recognize diffraction patterns in the seismic data, it can predict an expected size of the diffraction curves and adjust the window size dynamically based on thedata's characteristics. This adaptive approach allows for the effective handling of a diverse range of subsurface features without manually fine-tuning window parameters for each new data set.

[0090] When an appropriate window size is decided, the method of the present disclosure can be performed based on the flow illustrated in Figure 1.

[0091] At step 11, reflections of a plurality of volumes obtained by dividing the UUHR full wavefield seismic data using a sliding local window are derived. Steps for deriving the reflection will be detailed in the following with reference to another flow chart.

[0092] At step 12, all reflections in the plurality of volumes are aggregated to obtain the reflection data.

[0093] At step 13, the diffraction data is then obtained by subtracting the derived reflection data from the full wavefield seismic data.

[0094] At step 14, boulder(s) are detected by identifying diffraction pattern representing the boulder in the obtained diffraction data.

[0095] Figure 2 schematically illustrates, in a flow chart type of diagram 20, detailed steps of obtaining a reflection of a specific volume, according to an embodiment of the present disclosure.

[0096] At step 21, the seismic data volume obtained by applying a sliding window to the full wavefield seismic data is transformed into the frequency domain using Fourier transform.

[0097] As an example, the seismic data volume can be transformed from the time domain to the frequency domain using fast Fourier transform. Moreover, low and high frequency bounds can be estimated via trial and error to find the optimal processing frequency range.

[0098] At step 22, the frequency domain data is mapped into a block Hankel matrix. Specifically, each frequency slice data is reformatted via Hankelization, which is known to those skilled in the art and will not be elaborated here.

[0099] At step 23, local rank reduction is performed on the block Hankel matrix using singular value decomposition, SVD.

[0100] For the purpose of reflection reconstruction, which is needed for the diffraction separation, an appropriated rank selection is needed, as the rank of the block Hankel matrix is an important factor used to decide whether a singular value corresponds to the reflection energy or the diffraction energy.

[0101] Statistically, rank number can be set from 2 to 6 for all local processing windows. For better performance, the rank number can be selected adaptively by finding the largest jump inside all the singular values for each local processing window.

[0102] Specifically, as an example, SVD is applied on the Hankelization matrix to estimate all the singular values. For the singular values in each window, a singular value ratio of Eq. (1) is considered:

[0104] Where Si represent singular values and R. is the singular value ratio sequence and L is the length of the singular spectrum. The rank of a local windows (data volume) is estimated by selecting the maximum of the singular value ratio sequence R = arg maxtRt.

[0105] The rank selection is performed in an adaptive and automatic way for each window, which allows the reflection to be reconstructed more reliably.

[0106] At step 24, a block Hankel matrix of reflection is reconstructed based on the block Hankel matrix with reduced rank. The reconstructed Hankel matrix is achieved by extracting the low rank components.

[0107] At step 25, the reflection is obtained by applying inverse Fourier transform to the reconstructed Hankel matrix.

[0108] With the reflections of each data volume obtained, the reflection of the whole seismic data set is obtained by aggregating all the reflections. Thereafter, the diffraction data can be derived by subtracting the reflection from the full wavefield data.

[0109] For UUHR seismic data as used in the present disclosure, the UUHR reflections are estimated by extracting the larger principle components based on the above described procedure . UUHR diffractions can be simply estimated by subtracting the UUHR reflections from the original full wavefield data.

[0110] Figures 3(a) - 3(c) respectively illustrate exemplary full wavefield seismic data, diffraction data obtained according to the method of the present disclosure and reconstructed reflection data in 3D.

[0111] Figures 4(a) - 4(c) respectively illustrate exemplary full wavefield time slice, diffraction time slide and reflection time slice for the 3D data of Figures 3(a) to 3(c).

[0112] In the following, the procedure of detecting boulders from the obtained diffraction data will be described.

[0113] Generally speaking, a boulder is detected by identifying a diffraction pattern representing the boulders in the diffraction data.

[0114] Specifically, the diffraction data is analysed for patterns indicative of subsurface boulders, which are commonly characterized by circular or hyperbolic features depending on the geometry of the boulder and the seismic survey parameters.

[0115] The identification procedure involves recognizing shape and / or geometry of the diffraction patterns. According to an exemplary pattern recognition algorithm, the diffraction data is scanned to locate regions exhibiting these geometric features.

[0116] The algorithm can measure the steepness and symmetry of the curves to differentiate boulder-induced diffractions from other anomalies, such as fractures or voids, which may have more irregular or elongated shapes.

[0117] To ensure accuracy, the algorithm may incorporate a classification step based on predefined templates or trained models. For instance, machine learning methods, such as support vector machinesor neural networks, can be trained on labelled datasets containing examples of boulder diffractions. These models can then classify detected patterns in the diffraction data, assigning a likelihood score to each identified feature based on how closely it matches the known characteristics of boulder-induced diffractions.

[0118] Once the boulder diffraction patterns are identified, their spatial coordinates and properties are logged. The identified patterns are then mapped to create a visualization of potential boulder locations. In practical scenarios, additional steps such as cross-validation with geological knowledge or nearby seismic features may be used to further refine the results.

[0119] For example, in offshore wind farm planning, where boulder detection is critical for pile installation, the diffraction data might already highlight numerous hyperbolic features. The algorithm would scan these features, analyse their curvature, amplitude, and frequency, and confirm which patterns likely represent boulders. This information would then be used to generate a boulder map, ensuring informed decisions during construction.

[0120] Figures 5(a) - (c) illustrates zoomed-in diagram of the marked region in comparison to the original areas shown in Figures 4(a) to 4(c).

[0121] It is seen from the comparison between Figures 4 and 5 that diffraction separation can help to delineate details under strong reflection energy, which helps with better boulder detection.

[0122] Moreover, refection-only data, as a by-product, can also be used for stratigraphy-based interpretation.

[0123] The invention has been described by reference to certain embodiments discussed above. It will be recognized that these embodiments are susceptible to various modifications and alternative forms well known to those of skill in the art.

[0124] Further modifications in addition to those described above may be made to the structures and techniques described herein without departing from the spirit and scope of the invention. Accordingly, although specific embodiments have been described, these are examples only and are not limiting upon the scope of the invention.

Claims

CLAIMS1. A method for detecting a subsurface boulder based on acquired full wavefield seismic data of a region, the method performed by a processor and comprising the steps of:obtaining diffraction data by applying seismic diffraction separation to the full wavefield seismic data, anddetecting a boulder by identifying a diffraction pattern representing the boulder from the obtained diffraction data;wherein the full wavefield seismic data comprises 3D ultra-ultra-high-resolution, UUHR, stacked seismic wavefield data.

2. The method according to claim 1, wherein the diffraction separation comprises the steps of:deriving reflection data from the full wavefield seismic data, andobtaining the diffraction data by subtracting the derived reflection data from the full wavefield seismic data.

3. The method according to claim 2, wherein the step of deriving reflection data comprises the steps of:deriving reflection for each of a plurality of volumes obtained from dividing the full wavefield seismic data by sliding a local window, andaggregating all reflection in the plurality of volumes to obtain the reflection data.

4. The method according to claim 3, wherein the step of deriving reflection for each volume comprises the step of:transforming the seismic data volume into frequency domain data using Fourier transform;mapping the frequency domain data into a block Hankel matrix;performing local rank reduction on the block Hankel matrix using singular value decomposition, SVD;reconstructing a block Hankel matrix of reflection based on the block Hankel matrix with reduced rank, andobtaining the reflection by applying inverse Fourier transform.

5. The method according to claim 4, wherein adaptive rank selection is performed in the SVD.

6. The method according to claim 5, wherein a rank of the block Hankel matrix of the volume is determined based on largest singular value ratio.

7. The method according to any of the previous claims 3-6, where a size of the local window is statistically determined based on an averaged dimension of collected diffraction curves present in data samples.

8. The method according to claim 7, wherein the diffraction curves are collected from the data samples using a machine learning method.

9. The method according to any of the previous claims, wherein the step of identifying a diffraction pattern is performed using an automated pattern recognition algorithm.

10. The method according to any of the previous claims, wherein diffraction patterns in horizontal and vertical slices are analysed.

11. The method according to claim 10, wherein diffraction patterns identified in the horizontal and vertical slices are correlated to determine a spatial location of the boulder.

12. The method according to any of the previous claims, wherein the diffraction pattern comprises at least one of a circular and a hyperbolic feature.

13. The method according to any of the previous claims, further comprising generating a subsurface boulder map based on locations of detected boulders.

14. A device detecting a subsurface boulder based on acquired full wavefield seismic data of a region, the device comprising a processor for performing the method according to any of the previous claims 1 to 13.

15. A computer program product, comprising instructions which, when executed on at least one processor, cause the at least one processor to carry out the method according to any of the claims 1 to 13.

Images