Exploration Geophysics - Volume 53, Issue 6, 2022

This paper focuses on low computational efficiency in simulated annealing (SA) inversion of Transient Electromagnetic (TEM) data. Asynchronous multiple Markov chains (MMC) parallel strategy is a very promising SA acceleration method, which can be accelerated almost linearly. However, this method also reduces the accuracy of the solution. To overcome this problem, we added the solution set strategy to the asynchronous MMC parallel simulated annealing (PSA) algorithm for the first time. In this new algorithm, each thread independently searches for direction and exchanges data with the solution set in the shared memory. We used both the synthetic and field data to test the new algorithm. The synthetic data tests showed that the MMC PSA results are better than those of the original MMC PSA. We analyzed the efficiency of the new algorithm. Compared with the sequential VFSA, the maximum speedup of the new algorithm is approximately 10 times. The field data test also showed that the improved MMC PSA algorithm has good practicability. These tests demonstrate that the improved algorithm is effective, showing that its convergence speed is greatly improved without reducing the accuracy.

Distributed acoustic sensing (DAS) is an emerging technology increasingly employed to monitor changes of formation properties, production noise and micro-seismic activity, and as an array of sensors in active seismic surveys. The data recorded with the DAS systems are very rich; some features observed in DAS records are often not well understood, and thus are underutilised. A systematic analysis of the data recorded passively with a DAS system in a 900-m deep well over a period of 12 weeks in the Perth metropolitan area, Western Australia, reveals the presence of several types of ambient energy in the subsurface, such as earthquakes, ocean swell and urban noise. In particular, over 85 days of the experiment, the analysis detected sixteen earthquakes, with epicentres ranging from 126 km to 900 km (for the local events) and from 2300 km to 6400 km (for the remote events). Signals with frequencies below 0.9 Hz are dominated by the oceanic swell. The recorded urban noise includes mine blasting, machinery and traffic. The experiment shows the ability of DAS to detect these events and as such is potentially useful for subsurface characterisation and monitoring.

Seismic data reconstruction is an inverse problem in the geophysical community. Deep learning-based methods directly learn the projection between undersampled and complete data from large training datasets. However, when the feature difference between the test and training datasets increases, the recovery performance is degraded. Fine-tuning the pretrained network on the undersampled test field data has been introduced to adapt the trained network to new data. However, the pseudo-labels of the fine-tuning process need to be obtained by some traditional method in advance. In this paper, we explore a strategy for interpolating regularly sampled aliased seismic data by incorporating deep external and internal learning, i.e. by combing external pretraining with internal unsupervised fine-tuning. The pretrained network is obtained from an external synthetic dataset. Then, the pretrained network is fine-tuned on the internal dataset to obtain the final trained network adapting the test data. The inputs and labels of the internal dataset are generated solely from the currently regularly undersampled test data. As such, the fine-tuning process is totally unsupervised. Finally, the interpolated result is acquired by feeding the test data into the trained network. The proposed algorithm learns the external priors from large external datasets and it learns the internal features from undersampled test data. Thus, the network has an enhanced capacity to adapt to new data. Poststack and prestack field data are provided to verify the performance of the proposed algorithm.

Seismic fault interpretation has great importance in characterising subsurface geology. However, it is in general a manual and time-consuming task, thus, adaptive methods to achieve good fault detection results would be valuable. In this paper, we present a new method using combination of Contourlet transform and fuzzy entropy definition to adaptively detect and extract faults from seismic data. Contourlet is a multiscale and multidirectional filter that has high directionality which decomposes an image to various subscales. Our proposed scheme has 3 phases: first, employing Contourlet to pull out fault information using multidirectional and multiscale property of contourlet (feature extraction), second, using differentiation to boost fault information and calculating the correlation coefficient between the input image and subscales, fault information can be isolated from reflectors adaptively (feature selection) and third, applying a multi-level thresholding approach built on fuzzy partition of the histogram and entropy theory to classify image pixels into fault and non-fault (classification). The adaptive hybrid technique was applied to one synthetic and two real datasets containing fault data, reflectors and random noise. According to the results and their assessments, the proposed scheme has desirably located fault features in the data. We also examined the effect of random noise (Signal to Noise Ratio (SNR) = 2) on our adaptive algorithm which showed the success of our designed technique in the presence of noise.

Compared with the standard staggered-grid finite-difference (FD) methods, equivalent staggered-grid (ESG) ones can significantly reduce the computational memory for acoustic wave modelling in the variable-density media. To further enhance the simulation efficiency and accuracy, one way is to optimize the FD coefficients, another way is to design new FD stencils. In this paper, we propose a modified ESG (M-ESG) scheme which can significantly accelerate the wavefield simulation process while preserving or even improving the modelling accuracy. We calculate the FD coefficients by approximating the temporal and spatial derivatives simultaneously based on time–space domain (TS-D) dispersion relation of the discrete wave equation. Our M-ESG scheme in the TS-D can maintain basically the same accuracy as the conventional ESG (C-ESG) one when the FD coefficients are derived by the Taylor-series expansion (TE) approach. Note that the TS-D dispersion relation is nonlinear with respect to the FD coefficients of the C-ESG scheme, so it is difficult to obtain the optimized FD coefficients for the discrete wave equation. However, we can minimize the L2-norm error of the dispersion relation based on our M-ESG scheme to implement a linear FD coefficients optimization strategy, which is easy and efficient. Comparisons with TE- and optimization-based C-ESG schemes demonstrate the accuracy, stability, and efficiency superiorities of our TE- and optimization-based M-ESG ones.

The spectral-element method (SEM), which combines the flexibility of the finite element method (FEM) with the accuracy of spectral method, has been successfully applied to simulate seismic wavefields in geological models on different scales. One kind of SEMs that adopts orthogonal Legendre polynomials is widely used in seismology community. In the SEM with orthogonal Legendre polynomials, the Gauss-Lobatto-Legendre (GLL) quadrature rule is employed to calculate the integrals involved in the SEM leading to a diagonal mass matrix. However, the GLL quadrature rule can exactly approximate only integrals with a polynomial degree below 2N-1 (N is the interpolation order in space) and cannot exactly calculate those of polynomials with degree 2N involved in the mass matrix. Therefore, the error of the mass matrix originating from inexact numerical integration may reduce the accuracy of the SEM. To improve the SEM accuracy, we construct a least-squares objective function in terms of numerical and exact integrals to increase the accuracy of the GLL quadrature rule. Then, we utilise the conjugate gradient method to solve the objective function and obtain a set of optimal quadrature weights. The optimal mass matrix can be obtained simultaneously by utilising the GLL quadrature rule with optimal integration weights. The improvement in the numerical accuracy of the SEM with an optimal mass matrix (OSEM) is demonstrated by theoretical analysis and numerical examples.

Granitoid intrusions traditionally form a focus for geothermal heat exploration, as granite is a major host for Heat-Producing Elements (HPE; U, Th & K). Airborne spectral gamma-ray data for the study area highlight variations in HPE abundances in granitic rock units, indicating variation in the Radioactive Heat Production (RHP) values of these rocks. The computed arithmetic means of RHP for granitic rocks range from 0.96 µWm⁻³ for tonalite-quartz diorite to 1.10 µWm⁻³ for granodiorite, followed by a gradual increase to 1.52 µWm⁻³ for monzogranite and 2.51 µWm⁻³ for alkali-feldspar granite. The major control on the distribution of U and Th elements in the granitoid rocks appears to have been primarily of magmatic differentiation and is reflected in the linear correlation between these elements. Besides, subsequent post-magmatic hydrothermal fluids play their important roles in remobilization of profitable secondary U-mineralizations to be trapped and enrichment in the alkali-feldspar granitic rocks.