Ground-penetrating Radar Field Data Research Articles

GPR-TransUNet: An improved TransUNet based on self-attention mechanism for ground penetrating radar inversion

Convolutional Neural Networks (CNN) are widely applied to Ground Penetrating Radar (GPR) inversion because they have strong data-driven capabilities and are suitable for the data structure form of GPR. For CNN, the computation increases with the distance that the convolutional block moves from one region to another when it calculates the relationship between two regions. For GPR data, the target reflection exists in the surrounding traces and full time-window of the target, which leads to high degree of remote relationship. In this paper, we propose GPR-TransUNet, a deep-learning based inversion network which use self-attention mechanism. According to the characteristics of GPR data, regression network and GPR-Loss mechanism were used. Both numerical and model experiments were arranged to test the performance of the network, and the result as well as comparative analysis demonstrate the superiority of GPR-TransUNet. Finally, we applied this method to the field GPR data of Guangxi as an attempt.

Enhancing subsurface contamination assessment via ensemble prediction of ground electrical property: A Colorado AMD-impacted wetland case study

Acid mine drainage (AMD) is recognized as a major environmental challenge in the Western United States, particularly in Colorado, leading to extreme subsurface contamination issue. Given Colorado's arid climate and dependence on groundwater, an accurate assessment of AMD-induced contamination is deemed crucial. While in past, machine learning (ML)-based inversion algorithms were used to reconstruct ground electrical properties (GEP) such as relative dielectric permittivity (RDP) from ground penetrating radar (GPR) data for contamination assessment, their inherent non-linear nature can introduce significant uncertainty and non-uniqueness into the reconstructed models. This is a challenge that traditional ML methods are not explicitly designed to address. In this study, a probabilistic hybrid technique has been introduced that combines the DeepLabv3+ architecture-based deep convolutional neural network (DCNN) with an ensemble prediction-based Monte Carlo (MC) dropout method. Different MC dropout rates (1%, 5%, and 10%) were initially evaluated using 1D and 2D synthetic GPR data for accurate and reliable RDP model prediction. The optimal rate was chosen based on minimal prediction uncertainty and the closest alignment of the mean or median model with the true RDP model. Notably, with the optimal MC dropout rate, prediction accuracy of over 95% for the 1D and 2D cases was achieved. Motivated by these results, the hybrid technique was applied to field GPR data collected over an AMD-impacted wetland near Silverton, Colorado. The field results underscored the hybrid technique's ability to predict an accurate subsurface RDP distribution for estimating the spatial extent of AMD-induced contamination. Notably, this technique not only provides a precise assessment of subsurface contamination but also ensures consistent interpretations of subsurface condition by different environmentalists examining the same GPR data. In conclusion, the hybrid technique presents a promising avenue for future environmental studies in regions affected by AMD or other contaminants that alter the natural distribution of GEP.

Adding realistic noise models to synthetic ground‐penetrating radar data

ABSTRACTCost‐effective computing capabilities have paved the road for the use of numerical modelling to develop advanced methods and applications of ground‐penetrating radar (GPR). Realistic synthetic data and the corresponding modelling techniques, respectively, should consider all subsurface and above‐ground aspects that influence GPR wave propagation and the characteristics of recorded signals. Critical aspects that can be realized in modern GPR modelling tools include heterogeneous and frequency‐dependent material properties, complex structures and interface geometries as well as three‐dimensional antenna models, including the interaction between the antenna and the subsurface. However, realistic noise related to the electronic components of a GPR system or ambient electromagnetic noise is often not considered, or simplified by assuming a white Gaussian noise model which is added to the modelled data. We present an approach to include realistic noise scenarios as typically observed in GPR field data into the flow of modelling synthetic GPR data. In our approach, we extract the noise from recorded GPR traces and add it to the modelled GPR data via a convolution‐based process. We illustrate our methodology using a modelling exercise, where we contaminate a synthetic two‐dimensional GPR dataset with frequency‐dependent noise recorded in an urban environment. Comparing our noise‐contaminated synthetic data with field data recorded in a similar environment illustrates that our method allows the generation of synthetic GPR with realistic noise characteristics and further highlights the limitations of assuming pure white Gaussian noise models.

A GPR 2D Teager-Kaiser energy operator based on the multivariate variational mode decomposition

ABSTRACT To delineate the anomalies related to subsurface target bodies and the geophysical events, a novel method based on the effective combination of the multivariate variational mode decomposition (MVMD) and 2D Teager-Kaiser operator (2D-TKEO) is proposed for ground-penetrating radar (GPR) data processing in this letter. The proposed method takes advantage of the multi-channel decomposition scheme of MVMD and the multi-directional localization of 2D-TKEO for tracking energy attributes. The time-frequency spectrum can be generated based on the IMF-slices decomposed by the MVMD, on which the 2D-TKEO is introduced to focus the spectrum energy, then the common frequency slices are extracted to reveal the energy components of different frequency ranges. The forward model and GPR field data are applied here to demonstrate that the time-frequency spectrum and the subsequent common-frequency slices based on the proposed MVMD-2DTKEO approach can depict the subsurface target bodies (e.g., voids and pipes) and subsurface layers with higher resolution and better stratigraphic continuity, compared with that of the S-transform and the VMD-1DTKEO.

Estimating porosity distribution of a heterogeneous alluvial aquifer by GPU-accelerated 3D conditional stochastic inversion of common-offset GPR reflection data

Three-dimensional (3D) high-resolution and accurate characterization of the heterogeneity in an alluvial aquifer is critical to predicting the groundwater flow and contaminant transport. In this study, we extend the previous 2D conditional stochastic inversion method with respect to common-offset ground-penetrating radar (GPR) data to 3D for the estimation of 3D porosity distribution in a heterogeneous alluvial aquifer. First, 3D stochastic realizations of subsurface properties can be obtained using the fast Fourier transform moving average (FFT-MA) method, and the realizations are conditioned to borehole porosity measurements available in the 3D survey area, as well as to the geostatistical parameters derived from the borehole logs and the processed 3D GPR data. Next, the realizations are constantly modified and regenerated via a localized simulated annealing optimization strategy, which ultimately makes their corresponding synthetic data offer an acceptable fit to the GPR data. To accelerate the computational speed of 3D inversion, we design and implement the multi-GPUs multi-node parallelization of the 3D conditional stochastic inversion algorithm. The proposed 3D inversion algorithm is positively verified through the application to field GPR data collected at the Boise Hydrogeophysical Research Site near Boise, Idaho, USA. Our results indicate that the proposed 3D inversion method has the potential to effectively recover the fine-scale porosity distribution of a heterogeneous alluvial aquifer.

Deep-learning-guided high-resolution subsurface reflectivity imaging with application to ground-penetrating radar data

SUMMARY Subsurface reflectivity imaging is one of the most important geophysical characterization methods for revealing subsurface structures. In many cases, accurate subsurface reflectivity imaging is challenging because of, for example, random or coherent noise in the data and sparse source–receiver observation geometry. We develop a deep-learning-guided iterative imaging method to improve subsurface structure imaging. Specifically, we train a supervised neural network to infer a noise-free, high-resolution image from a noisy, low-resolution image and use this estimated image as guidance to regularize least-squares imaging. We develop a systematic method to generate high-quality synthetic training data (data-label pairs) to train the guidance neural network. The trained neural network can provide high-fidelity predictions even for field-data images that are not in the training data. We validate our new imaging method using one synthetic and two field ground-penetrating radar data examples, and find that our method can produce clean, high-resolution subsurface reflectivity images where existing single-pass and least-squares imaging methods fail due to noise and insufficient data coverage.

Efficient source-independent full waveform dual-parameter inversion of common-offset GPR data

Full waveform inversion (FWI) of common-offset ground penetrating radar (GPR) data can make up for the defects of the traditional prospecting technology such as destructive, cumbersome, and low efficiency, which also provides a robust and high-precision tool for quantitative assessment of subsurface prospecting. Nonetheless, conventional GPR-FWI is of limited practicality since the deficiency of accurate source wavelet and tremendous computational cost. We develop an efficient source-independent inversion algorithm based on the convolution method to eliminate the influence of source wavelet and enhance the algorithm efficiency to inverse field data. To suppress the noises induced by the convolution and cross-correlation operations, the time window is introduced in the reference trace which is selected automatically by singular value decomposition (SVD). The stochastic strategy reduces the computational cost by source subsampling. The synthetic model of a complex lens body demonstrates that source-independent FWI with time window has more tolerant to the initial model and noise, and the stochastic strategy improves the efficiency with ensured inversion accuracy. In the field GPR data of a sand trough containing two anomalous bodies, our algorithm has good adaptability and can accurately reconstruct the distribution of the permittivity and conductivity of the subsurface media without accurate estimation of the source wavelet. The results above indicate that the proposed algorithm in this study is valid for common-offset GPR data and improves the inversion efficiency by 1.24 to 3.10 times, which shows great potential to promote qualitative assessment in real-time subsurface exploration.

Drone-mounted ground-penetrating radar surveying: Flight-height considerations for diffraction-based velocity analysis

Recent studies highlight the potential of the drone platform for ground-penetrating radar (GPR) surveying. Most guidance for optimizing drone flight heights is based on maximizing the image quality of target responses, but no study yet considers the impact on diffraction traveltimes. Strong GPR velocity contrasts across the air-ground interface introduce significant refraction effects that distort diffraction hyperbolas and introduce errors into diffraction-based velocity analysis. The severity of these errors is explored with synthetic GPR responses, using ray- and finite-difference approaches, and a field GPR data set acquired over a sequence of diffracting features buried up to 1 m depth. Throughout, GPR antennas with 1000 MHz center frequency are raised from the ground to heights <0.9 m (i.e., 0–3 times the wavelength in air). Velocity estimates are within +10% of modeled values (spanning from 0.07 to 0.13 m/ns) if the antenna height is within 1/2 wavelength in air above the ground surface. Greater heights reduce diffraction curvature, damaging velocity precision, and masking diffractions against a background of subhorizontal reflectivity. Field data highlight further problems of the drone-based platform, with data dominated by reverberations in the air gap and reduced spatial resolution of wavelets at target depth. We suggest that a drone-based platform is unsuitable for diffraction-based velocity analysis, and any future drone surveys are benchmarked against ground-coupled data sets.

Combining ground-penetrating radar sections with different antenna frequencies including time-frequency domain noise suppression filters

Ground-penetrating radar (GPR) is one of the most useful geophysical methods that rapidly and efficiently provide high-resolution subsurface images depending on antenna frequencies and target depth. Recently, to enrich the interpretation of GPR sections, multifrequency sections have been combined at deeper and shallower levels in a profile. In this context, the techniques of combination and image quality improving processes of GPR data also have gained importance. For this, simple summation (SS) and time-shifted balanced summation (TSBS) techniques, frequency-domain local mean notch filter (FDLMNF), frequency-space ( f- x) filter, and automatic gain control (AGC) processes are sequentially applied within a workflow. Although the FDLMNF is used to effectively remove single-frequency spike-like noises without distortion around the notch frequency, the f- x filter is applied to suppress the random noise and improve the lateral traceability of the section, and AGC provides a statistically based gain recovery. In the proposed workflow, after generating synthetic and field GPR data with 100 and 250 MHz frequencies, data were combined using SS and TSBS techniques. Finally, FDLMNF, f- x, and AGC were applied to increase the image quality of this combined section. The result sections were evaluated and compared using the Fourier average amplitude and f- k spectra. According to the spectra, the TSBS technique provides good spectral balancing and offers wider spectral content. At the same time, it was observed that the image quality of the combined GPR sections increased significantly after applying FDLMNF, f- x, and AGC processes. As a result, when the sections resulting from SS and TSBS techniques are examined, the image enhancement processes that protect and strengthen the spectral bandwidth of the data are successful. Thus, the temporal resolution and spatial continuity of GPR sections also will be increased, and a better quality interpretation will be provided.

Detection and Characterization of Cracks in Highway Pavement with the Amplitude Variation of GPR Diffracted Waves: Insights from Forward Modeling and Field Data

It is important to distinguish between two common defects, fatigue cracks and reflective cracks, and determine their locations (the top and bottom) in the highway pavement because they require individually targeted treatment measures. Ground Penetrating Radar (GPR) has the potential to detect cracks in the highway pavement due to the change of the electromagnetic properties of highway-pavement media, arising from the existences of cracks. By using a theoretical analysis and a numerical simulation, we compare the characteristics of corresponding radargrams, including the amplitude variation of diffracted waves with various models of presetting cracks inside the layered homogeneous media. For those fatigue cracks and reflective cracks extending to the road surface, the amplitude curves of direct ground wave can intuitively indicate the locations of the top of the cracks and qualitatively compare the width of these cracks. Furthermore, we find that the shape and pattern of diffraction hyperbolas of both types of cracks with bottoms at different locations are quite similar, but their amplitudes are significantly different. To be specific, for those cracks with the same width, the amplitude of diffracted waves generated by fatigue cracks is slightly higher than that generated by reflective cracks at the interface between the asphalt surface and the semi-rigid base layer. In contrast, the amplitude of the former is significantly lower than the latter at the interface between the semi-rigid base and the roadbed. We applied these findings to the interpretation of the field GPR data of a highway pavement in China, and successfully identified the locations of the cracks and corresponding types. Our model results and field results clearly show the efficiency of our findings in the detection of cracks for highway-pavement rehabilitation.

Imaging the internal structure of trunks via multiscale phase inversion of ground-penetrating radar data

Ground-penetrating radar (GPR) is a geophysical technique widely used in near-surface noninvasive detecting. It has the ability to obtain a high-resolution internal structure of living tree trunks. Full-waveform inversion (FWI) has been widely used to reconstruct the dielectric constant and conductivity distribution for crosswell applications. However, in some cases, the amplitude information is not reliable due to the antenna coupling, radiation pattern, and other effects. We have developed a multiscale phase inversion (MPI) method, which largely matches the phase information by normalizing the magnitude spectrum; in addition, a natural multiscale approach by integrating the input data with different times is implemented to partly mitigate the local minimal problem. Two synthetic GPR data sets generated from a healthy oak tree trunk and from a decayed trunk are tested by MPI and FWI. A field GPR data set consisting of 30 common-shot GPR data is acquired on a standing white oak tree ( Quercus alba); the MPI and FWI methods are used to reconstruct the dielectric constant distribution of the tree cross section. Results indicate that MPI has more tolerance to the starting model, noise level, and source wavelet. It can provide a more accurate image of the dielectric constant distribution compared to the conventional FWI.

Imaging shallow fault structures by three-dimensional reverse time migration of ground penetration radar data

Ground penetrating radar (GPR) is an effective technology for investigating shallow underground structures using high-frequency electromagnetic waves. Migration is necessary to interpret GPR data because it resolves the true structure from distorted and blurry time-domain profiles. Wave-equation-based reverse time migration (RTM) can provide higher-resolution and more reliable results compared to ray-based migration methods. We have developed a three-dimensional RTM approach based on the finite-difference time-domain algorithm and applied it to field GPR data collected in the northern segment of the Huangzhuang-Gaoliying fault in suburban Beijing. Migrated images clearly show the subsurface structures, with clear indications of underground reflectors in agreement with geological cross-sections proposed in previous studies. Our results demonstrate that GPR is a convenient, non-destructive, and reliable technique that provides a new comprehensive approach to image subsurface conditions, thus allowing effective monitoring of active faults in urban areas.

Numerical simulation of ground‐penetrating radar data for studying the geometry of fault zone

ABSTRACTPalaeoseismology studies the footprints of ancient earthquakes to improve the knowledge about the modern seismicity of the territory. A ground‐penetrating radar (GPR), among other geophysical methods, is used for quick determination of shallow stratigraphy – displaced, oblique layers within the fault zone. GPR data interpretation from diverse and complex reflection patterns of the fault zone heavily depends on the interpreter's experience. The range of different fault zone parameters in which this method can be successfully applied has not yet been investigated. We used a numerical simulation of GPR data to determine how GPR images the elements of faults (fault plane, hanging wall, footwall) in comparison with other reflections. Furthermore, we studied which parameters have the most significant impact on GPR wave patterns. We performed a series of numerical models of a fault, changing its geometry with increasing complexity from elementary models to realistic ones. The resulting synthetic profiles allowed finding specific GPR signatures from the fault plane, the hanging wall and the footwall. We collected field GPR data from two different fault zones and examined them for verification.

Detection of road cavities in urban cities by 3D ground-penetrating radar

Cavities under urban roads have increasingly become a great threat to traffic safety in many cities. As a quick, effective, and high-resolution geophysical method, ground-penetrating radar (GPR) has been widely used to detect and image near-surface objects. However, the interpretation of field GPR data remains challenging. For example, it is hard to distinguish reflections caused by road cavities or other urban utilities with a conventional 2D GPR survey. The superiority of 3D GPR in data interpretation is demonstrated by a laboratory experiment. Two pipes and a glass-made cavity buried in a sandpit show similar hyperbolic reflections in the 2D GPR profiles; thus, they are difficult to discriminate. In contrast, their geometric shapes and dimensions are readily identified in the 3D image reconstructed from the synthetic 3D GPR data set. Based on these ideas, we have developed a car-mounted 3D GPR system with two antenna arrays oriented in different polarization directions and have detected more than 100 cavities in three Chinese cities over the past year. The field data of two such cavities are presented. As a result, the cavity depth, horizontal size, and height can be accurately estimated from the 3D GPR data set. Laboratory and field experimental results indicate that 3D GPR possesses potential in the detection and recognition of road cavities and utilities in the complex urban environment.

Coherent diffraction imaging for enhanced fault and fracture network characterization

Abstract. Faults and fractures represent unique features of the solid Earth and are especially pervasive in the shallow crust. Aside from directly relating to crustal dynamics and the systematic assessment of associated risk, fault and fracture networks enable the efficient migration of fluids and therefore have a direct impact on concrete topics relevant to society, including climate-change-mitigating measures like CO2 sequestration or geothermal exploration and production. Due to their small-scale complexity, fault zones and fracture networks are typically poorly resolved, and their presence can often only be inferred indirectly in seismic and ground-penetrating radar (GPR) subsurface reconstructions. We suggest a largely data-driven framework for the direct imaging of these features by making use of the faint and still often underexplored diffracted portion of the wave field. Finding inspiration in the fields of optics and visual perception, we introduce two different conceptual pathways for coherent diffraction imaging and discuss respective advantages and disadvantages in different contexts of application. At the heart of both of these strategies lies the assessment of data coherence, for which a range of quantitative measures is introduced. To illustrate the versatility and effectiveness of the approach for high-resolution geophysical imaging, several seismic and GPR field data examples are presented, in which the diffracted wave field sheds new light on crustal features like fluvial channels, erosional surfaces, and intricate fault and fracture networks on land and in the marine environment.

On the usage of diffractions in ground-penetrating radar reflection data: Implications for time-lapse gas migration monitoring

Time-lapse ground-penetrating radar (GPR) measurements are used to image/monitor, for example, water infiltration, water table changes, and fluid/gas flow patterns. Although crosshole GPR is often preferred over surface-reflection GPR in such studies, its application is limited by the selected borehole geometry, which may be difficult to define in an optimal way, especially in experiments in which flow pathways are difficult to predict. Surface-reflection GPR data sets are generally faster to collect over relatively large areas and are therefore more efficient for covering the volume when a fast-moving tracer (e.g., gas) may infiltrate a heterogeneous subsurface medium. We have used a diffraction imaging approach on time-lapse surface-reflection GPR data to detect changes in radar wave velocity associated with gas (CO2) injected into a heterogeneous chalk succession. We initially test and evaluate the diffraction imaging approach on synthetic GPR data. Afterward, we apply the methodology to time-lapse GPR field data, and we evaluate the robustness of using information from diffractions in light of the obtained data quality. The synthetic tests indicate that diffractions provide essential information for delimiting the area affected by gas in the heterogeneous chalk section studied. Our field experiment shows that using the diffraction information alone allows for detection of gas-affected zones and, therefore, potential flow characteristics of gas movement. We infer that the CO2 flow patterns in our study most likely are determined by small-scale fractures rather than the porosity/permeability of the rock matrix. Moreover, the approach used may serve as an initial study for future more targeted experiments or for further detail-retrieving full-waveform inversion.

Finite-difference time domain (FDTD) modeling of ground penetrating radar pulse energy for locating burial sites

Analysis of the finite-difference time domain (FDTD) numerical simulation of ground penetrating radar (GPR) measurement for locating burial sites is described in this paper. Effective, efficient, and reliability interpretation of GPR field data obtained from clandestine sites is very crucial in forensic investigations. The main goal of the study is the prediction of the change in the interaction of the electromagnetic incident on changes in buried bodies with time. In order to achieve this, the research involves the modeling of the GPR electromagnetic pulse energy responses to simulated changes in buried body with time with a view to understand what the results of real field measurement will give. The field measurements were conducted with GPR system manufactured by Mala Geoscience with antennae frequency of 500 MHz, 250 MHz, and 100 MHz. Responses from both synthetic and field radargrams depict the target was intercepted at same time (approximately 25 ns). The results have demonstrated that FDTD modeling is an important tool for enhancing the reliability of GPR data interpretation particularly for forensic study.

Correlation between agrotechnical properties of selected soil types and corresponding GPR response

In this study, we investigated the relationship between ground-penetrating radar (GPR) response and agriculture properties of soil with a view to understanding how the constraint of soil degradation may influence the properties. GPR field data measurements were made at a location with soil types, properties, and disturbances caused by tractor movement. The data were processed, and empirical equations relating soil physical properties and material properties of soil media were considered for the analysis of the field data. The results showed a change in the reflection coefficient and increase in the GPR wave velocity when comparing the records of the initial parts of the GPR records, lasting about 2.5 ns, of the signal response of soil subjected to compaction as a result of 10 tractor passes, with those obtained prior to movement of the tractor. The summation of the absolute value of GPR wave amplitude in the analyzed results clearly shows that the amplitude of the signal corresponding to the compressed ground is twice and even three times smaller than the amplitude recorded before the tractor runs. The results prompted the design of a relatively simple method for tracking changes in soil properties based on the results of GPR measurements, which show that zones subjected to direct tire pressure are easy to delineate and are not limited to the part that is directly under the tire, but extend about 0.5 m. It thus shows that there is a relationship between the penetration resistance induced by a change in porosity and changes in the coefficient of reflectivity and the velocity of radar waves.

Adaptive Ground Clutter Reduction in Ground-Penetrating Radar Data Based on Principal Component Analysis

Singular value decomposition is an effective way to remove ground clutter in ground-penetrating radar (GPR) applications. The main limitation of this method is the selection of principal components to completely reconstruct the ground clutter or the target. To date, no effective criteria or technology have been developed. To solve this problem, a new method is proposed in this paper. The research and analysis presented herein reveal that the root-mean-square height (RMSH) of the first-arrival curve corresponding to the ground clutter has a well-defined positive relationship with the number of singular values associated with the principal components of the ground clutter. The number of singular values of these principal components ( $N$ ) can be precisely determined based on the ground clutter by a linear function, $N = 0.2634D + 1.3086$ , where D represents the RMSH value. In addition, an algorithm called developed histogram equalization was developed to improve the contrast to highlight the targets in denoized GPR data sets. The proposed strategy of extracting the principal components of the ground clutter and highlighting the contrast between the target signal and environmental reflections was successfully applied to the field GPR data, thus demonstrating the practicality and validity of the proposed approach.

A Machine Learning Based Approach for Automatic Rebar Detection and Quantification of Deterioration in Concrete Bridge Deck Ground Penetrating Radar B-scan Images

Ground penetrating radar (GPR) is a non-destructive method (NDT) for subsurface object identification. Interpretation of GPR data is often done manually by an engineer, which is a time-intensive task and requires moderate to significant level of training. The authors proposed a novel machine learning based processing for automatic interpretation and quantification of concrete bridge deck GPR B-scan images. The proposed method is based on combination of image processing, machine learning (ML) data classification, data filtering, and spatial pattern analysis for quantification of deterioration in concrete bridge decks. For the first time, the authors introduced a dataset of 4,000 B-scan images cropped from real bridge deck GPR field data, named DECKGPRH1.0. The proposed method is tested on bridge deck GPR data collected from three bridges with different NBI (National Bridge Inventory) ratings. The results presented indicate that by implementing a ML based classifier and a fine tuned filter, the proposed approach provides a robust solution for automatic quantification GPR field data.