Development of the Inversion Method for Transient Electromagnetic Data

Transient electromagnetic method (TEM) is a geophysical tool to obtain resistivity distribution in the subsurface. Combining with the resistivity property of typical rocks, the TEM method can make inferences of the geological maps underground. The inversion method is the main technique to extract resistivity form the recorded TEM data. There are various inversion methods that have been applied to TEM data, each of which favors different model structures. It is essential to choose the optimal inversion algorithm for a TEM survey in a given geological setting. Thus, this article presents a systematic summary of recent developments of inversion methods for TEM data. We first summarize the basic concept of the TEM inversion theory. Then, the recent developments TEM inverse method are divided into deterministic inversion and stochastic inversion. For the deterministic method, we present the development of constrained inversion and joint inversion. For the stochastic method, we analyze the particle swarm optimization, Bayesian inversion, and TEM pseudo-seismic imaging. Thereafter, we prospect the future research direction of the TEM method.


I. INTRODUCTION
Geophysics is the main method to access information about the earth's interior. Based on physical principles, it can be used to make inferences about the parameters of underground targets by recording data sets on the earth's surface. Because the parameters of strata cannot be directly evaluated from field data, it is necessary to study the propagation of physical fields and inverse theory to reconstruct the details of geological bodies from observed data. The essence of the geophysical inverse problem is to infer the structure of the earth's interior qualitatively or quantitatively based on observed data. As the earth's interior is complex, to make inversion results as close to the real geological structure as possible, we usually use a simple model to approximate the real geological structure and then modify the model parameters based on the fitness of The associate editor coordinating the review of this manuscript and approving it for publication was Geng-Ming Jiang . the forward modeling and the recorded data. The reliability of inversion results largely depends on whether we can find an appropriate model. In the last three decades, with the development of computer science, there have been a number of publications on the inversion of geophysical data by geologists, geophysicists, mathematicians, and engineers [1]- [3].
As an important branch of geophysics, the electromagnetic method infers the presence of underground rocks and ore bodies based on their differences in conductivity and polarizability. The electromagnetic method plays an important role in the exploration of metal, mineral, oil, and gas, and geothermal resources [4]- [7]. As an electromagnetic method, the transient electromagnetic method (TEM) uses a grounded wire or ungrounded loop to transmit a source electromagnetic signal to the ground (Fig. 1). Under the excitation of a source electromagnetic field, a geological body induces eddy currents (a secondary field). After the source is turned off, the eddy currents attenuate gradually rather than disappearing immediately. The secondary magnetic field, which propagates to the earth's surface and is recorded by a receiver in the geophysical electromagnetic system, can be used to determine the electrical distribution of the underground geological body. TEM can be used for high-precision imaging of physical parameters such as resistivity, which has been widely used in mineral and groundwater exploration and other engineering fields [8].
Since the 1990s, researchers have gradually introduced various geophysical inversion methods to invert TEM data. In general, the TEM inversion methods can be divided into two categories. One is the deterministic inversion, such as Occam algorithms [9], Tikhonov regularization inversion, Marquardt inversion, and singular value decomposition. The other is the stochastic inversion, such as simulated annealing algorithms, ant colony algorithms, and Bayesian inversion. The deterministic and stochastic inversion are both effective ways to recover model parameters from the geophysical data sets. However, as the forward modeling of the TEM method is computationally expensive and the deterministic inversion is much more efficient than stochastic inversion, the deterministic method is more favorable for TEM inversion at the early stage of the researches, especially for the multi-dimensional inversion problem.
The accuracy of forward modeling is of great importance for the TEM inversion. As there are several papers summarized the development of multi-dimensional forward and inversion [10], [11]. We will not provide a detailed description of the development of multi-dimensional forward and inversion here. A brief introduction of deterministic inversion of TEM data with a multi-dimensional forward operator is given as follows: Wang and Hohmann [12] applied the conjugate gradient method to the study of TEM inversion and realized a 2D inversion of TEM data. Commer and Newman [13] realized the nonlinear conjugate gradient inversion of 3D TEM data using a grounded wire source. Haber et al. [14] and Haber et al. [15] used the conjugate gradient method to realize a 3D inversion of TEM data from a central loop configuration. Yang and Oldenburg [16] developed an inversion algorithm based on the Gauss-Newton method under a 3D forward modeling of the finite volume method in the time domain and successfully applied it to field data in many mining areas in Canada. Cox et al. [17] used the conjugate gradient method to perform large-scale inversion of 3D TEM data, an approach which has been verified in many mine data and groundwater investigation projects.
The stochastic inversion of TEM data is mainly performed with a 1D forward operator. As the mapping from the recorded decay curve to the resistivity is nonlinear, the stochastic inversion is straightforward for inverting the TEM data. As with the development of computer science in the last few decades, stochastic inversion becomes popular in the geophysical community. Li et al. [18] introduced an algorithm with an artificial neural network into the data processing for TEM inversion and verified its effectiveness. Li et al. [19] combined an adaptive genetic algorithm with the feasible region method to calculate the longitudinal conductance of TEM, which improved the accuracy of the inversion result. Li et al. [20] and Cheng et al. [21] put forward an improved particle swarm optimization (PSO) inversion algorithm to accelerate convergence. Qin et al. [22] combined a genetic algorithm and neural network to shorten the calculation time.
As a variety of models can adequately fit the measured data, constrained inversion using prior information and joint inversion are us to address the ill-posedness of the inverse problem [23]. Wen et al. [24] carried out an Occam inversion of a large-loop TEM. Hu [25] developed a joint inversion scheme for TEM and direct current (DC) sounding data. Auken et al. [26] Bortolozo et al. [29] proposed a lateral constraint inversion method, and Siemon et al. [27] applied lateral constraint inversion to frequency domain aero-electromagnetic data and achieved good results. Viezzoli et al. [28] and Auken et al. [26] realized the inversion of airborne transient electromagnetic data by using the space-constrained inversion method. Bortolozo et al. [29] used a residual function discrete mapping method to perform joint inversion of electric method data and TEM data, and they performed location map analysis to compare the advantages of joint inversion to those of a single method. Lin et al. [30] proposed a lateral constraint inversion method to invert magnetic resonance and TEM data.

II. AN OVERVIEW OF TEM INVERSION THEORY
There are several kinds of TEM configurations with significant differences in their theory and data interpretation techniques. The inverse problem behind the TEM inversion can be expressed as follows: where F is the forward operator, which is based on the mathematical and physical laws of electromagnetic field propagation; d is the observed data and m is the set of model parameters. For a TEM survey, d= (d 1 , d 2 , d 3 , . . . , d M ) is the decay curve of the electromagnetic field after the source is switched off and m=(log(ρ 1 ), log(ρ 2 ), log(ρ 3 ), . . . , log(ρ N )) is the resistivity of each cell for the discretized subsurface. If the model parameters and forward operator are known, the process of calculating the transient electromagnetic response is a forward problem. If, on the contrary, the observation data and forward operators are known, then inferring the structure of the earth's interior or the location of underground mineral deposits is a typical inverse problem (Fig. 2). Because the relationship between the propagation law for an electromagnetic field and the model parameters is nonlinear, the transient electromagnetic inversion problem is nonlinear. Various inversion methods can be used to estimate the model parameters from equation (1). From the perspective of probability, inversion algorithms can be divided into deterministic inversion methods (shown in Fig. 2) and stochastic inversion methods (shown in Fig. 3). Deterministic inversion methods, including the Newton algorithm, steepest descent algorithm, and conjugate gradient method, find a single optimal model along the search direction determined by the sensitivity function. Stochastic inversion methods, including Bayesian algorithms, neural network algorithms, PSO algorithms, simulated annealing algorithms, and genetic algorithms, randomly search the model space and collect several qualified samples to characterize the solution of the inversion problem. The inversion is used to infer the parameters of the underground model by fitting the measured data. However, due to environmental noise and measurement error, it is not correct to reduce the fitting residual to the minimum or even zero.
The numerical simulation of TEM response requires a large amount of calculation. At the early stage, the TEM data are interpreted with approximate imaging techniques [31]. As computing power increases, deterministic inversion, which uses partial derivatives to guide the search of the solution, has gradually become the main method for estimating model parameters from TEM data. Christiansen et al. [32] developed a hybrid inversion scheme to speed up the inversion by using a regular forward operator and an approximated partial derivative. In recent years, the stochastic inversion algorithm has been applied to the TDEM inversion [33]- [36].

III. DETERMINISTIC INVERSION METHOD FOR TEM DATA
One of the methods for solving nonlinear inverse problems is to linearize the nonlinear problem. For the solution of the inverse problem in equation (1), the classical method is to find some operators to minimize the difference between the observed data and the model response: that is, Equation (2) gives the objective function of the leastsquares method. To stabilize the least-squares method, Marquardt [37] proposed the damped least-squares method, which introduces a damping factor to effect a compromise between precision and stability. For an ill-posed inverse problem, regularization is an important and widely used technique. It was first proposed by Tikhonov in the 1960s and is called Tikhonov regularization [3]. Since that time, other regularization methods have been introduced and applied in TEM inversion, including the Landweber iterative method, conjugate gradient method, generalized Newton iterative method, and regularized Gauss-Newton iterative method [1]. The main idea of the Tikhonov regularization method is to solve a well-defined problem by adding a penalty function term into the objective function to obtain a stable and unique solution. In this condition, the objective function based on a simple point is where λ is the regularization parameter, W d is the weighting matrix of the data, and is the first-order roughness matrix of the model. In equation (3), the first term is the residual between the measured data and the modeled response, weighted by the errors of the data. The second term is the regularization term of the designed model. The Tikhonov regularization factor, λ, is used to balance the relationship between the data fitting and model constraint. If λ is low, the stability of the inversion data will decline. If λ is high, the accuracy of the inversion data will decline. The selection of the regularization parameter is important to the effectiveness of Tikhonov regularization. There are two strategies to select regularization parameters: the prior strategy and the posterior strategy. The prior strategy depends only on the properties of operators and has value in theoretical analysis. In the actual calculation, one can use a posteriori strategy in which the selecting and adjusting of regularization parameters are based on the residual in each iteration.
Equation (1) can be approximated about some starting solution using the Taylor expansion where e obs represents the errors of measurement and D is the sensitivity matrix (also known as the Jacobian matrix), which is composed of the partial derivative of the modeled responses with respect to the model parameters, It indicates the influence of the model parameters on the forward response and can be simplified as where d = d − F(m 0 ) and m = m − m 0 . Based on the local approximation in equation (5), the iteration solution of the regularized inverse problem in (3) can be obtained: In this way, the regularized inversion of TEM data can be achieved, combined with model parametrization methods and forward modeling operator. In practice, 1D models play an important role in inverting TEM data. The 1D forward modeling and inversion are based on the assumption that the subsurface is layered (that is, there are no lateral variations in the resistivity distribution). Within a 1D inversion scheme, that assumption is considered valid at each individual measurement location. Under this assumption, the forward modeling of TEM data can be implemented efficiently, allowing various types of regularization methods can be tested and applied.

A. CONSTRAINED INVERSION
Due to the non-uniqueness of the TEM inversion, there could be a number of models that fit the data sets adequately well. It is hard to pick up the optimal model using conventional single point inversion with regularization terms. In these circumstances, the prior information of the model structure is of significant importance for improving the reliability and accuracy of the solution, as proper constraints can be imposed on the model parameters or structures. For example, the quasi-layered model is closer to the real geological situation Auken et al. [38] presented laterally constrained inversion (LCI) to impose the lateral continuity on model parameters. By inverting all the data along the profile in one system, the model with a laterally smooth transition can be obtained. The LCI allows a more complex quasi layered model than the 1D single point inversion. Moreover, compared with 2D or 3D inversion, the LCI inversion substantially reduces calculation time, which is very beneficial for the case with a large number of data.
In the traditional minima functional (3), a lateral constraint term is added to the LCI inversion, which can be expressed as where α, β represent the horizontal and vertical regularization parameters, respectively, and L is the lateral constraint matrix.
The updating equation of the model can be expressed as follows: The LCI method has been further modified to invert the data sets from different geological settings. For example, Viezzoli et al. [28] expanded the LCI into spatially constrained inversion (SCI), in which the constraints are imposed both along the across the profiles. In this way, the information of the parameters from well-resolved areas migrates through the constraints to aid the recovering of the parameters in a poorly constrained area. Vignoli et al. [39] and Vignoli et al. [40] presented a sharply SCI scheme for the scenario when sharp transitions are expected in the model by introducing vertical or horizontal constraints. As is shown in Fig. 4, the boundary of the blocks with distinct resistive contrast is more precisely retrieved using the sharp constraints than conventional SCI inversion. VOLUME 8, 2020

B. JOINT INVERSION
The joint inversion of multiple geophysical data sets is an effective way to reduce the ambiguity of the estimated parameters. Broadly speaking, the TEM data are initially joint-inverted with resistivity data or magnetotelluric (MT) to recover a reliable resistivity [41]- [43]. In the subsequent study, the joint inversion of TEM data and magnetic resonance sounding (MRS) data to improve the MRS parameter estimates. However, the structural coupling or the rock physics coupling has not been incorporated into the inversion procedure, which makes them different from the joint inversion in a rigorous way.
The structural coupling strategy is the most commonly used coupling operator in the joint inversion of electromagnetic data. Gallardo and Meju [44] were the first to introduce the cross-gradient function to impose similarity of structures among different parameter models. Since then the cross-gradients have been widely used in joint inversion studies, including the joint inversion of TEM data and seismic data [45]. Here we present a brief introduction of the cross gradients-based joint inversion scheme. The cross gradient functions are defined as follows: τ (x, y, z) = ∇m 1 (x, y, z) × ∇m 2 (x, y, z).
In these equations, m 1 and m 2 represent the different physical parameters in joint inversion and is the gradient operator. ∇m is the gradient at the position (x, y, z). The gradient direction represents the steepest descent direction of the physical parameters at the current point, and the gradient value is the change rate of the model parameter. If the directions of the two gradient vectors are the same (parallel), the crossgradient function of the two vectors is zero; otherwise, it is not zero. In the 3D model, the expansion of the cross-gradient function in three directions is as follows: In the 2D model, the cross-gradient function is zero in two directions.
Joint inversion is based on a regularized inversion algorithm and cross gradient function. The objective function of joint inversion can be written as where W i is the weight matrix of data fitting; λ i is the regularization factor of the model constraint term; d i is the observed data; F i is the forward function; m i is the model parameter; τ (m 1 , m 2 )is the cross gradient function between different model parameters. µ is the weighting term of the cross-gradient function. The joint inversion is realized by minimizing the objective function, which results in a consistent inversion solution.

IV. STOCHASTIC INVERSION OF TEM DATA
As mentioned earlier, in a TEM inversion one needs to choose a compromise between solution speed and solution precision. The above deterministic inversion is widely used in the actual data processing. However, due to the introduction of approximate interpretation and regularization parameters, the result of deterministic inversion depends on the artificial expectation for the solution rather than the estimation of model parameters driven by data. Therefore, it is necessary to develop a data-driven stochastic inversion. In recent years, with the improvement of computing ability, stochastic inversion has been applied to the inversion of TEM data. This method does not need to calculate a sensitivity function to guide the search direction for inversion. Although its inversion efficiency is low, this approach makes it easier to obtain the global optimal solution and uncertainty evaluation, which are difficult to obtain in deterministic inversion.

A. PARTICLE SWARM OPTIMIZATION (PSO)
Particle swarm optimization is a derivative-free global optimization algorithm based on organized behavior of large groups of simple animals, such as flocks of birds, schools of fish, or swarms of locusts. In recent years, with the rapid development of computer science and information technology, the Particle swarm optimization algorithms have been developed fast and applied to solve engineer problems. PSO was proposed by Kennedy and Eberhart [46]. In 1998, Shi and Eberhart [47] proposed a decline inertia factor, which successfully accelerates the convergence in the late stage of iteration. Shi and Eberhart [47] introduced the law of quantum motion in the multidimensional space into the particle swarm optimization algorithm. In 2002, Clerc and Kennedy [48] proposed another concept of construction factor K, which laid a theoretical foundation. Then, Lin and Chen [49] came up with a new formula of construction factor K based on the theory of the best damping ratio, which theoretically guarantees the stability of the algorithm convergence process. In 2017, Patel et al. [50] applied the PSO and gravitational search algorithm (GSA) hybrid optimization to improve the inversion speed of the algorithm. These efforts provide a guarantee for the application of the improved particle swarm optimization algorithm in geophysics. In 2007, to evaluate the applicability of PSO to the inversion of geophysical data, Shaw and Srivastava [51] test PSO for the first time by using DC, induced polarization (IP), and MT data. Particle swarm optimization and its improved algorithms have been widely used in data inversion seismic, magnetic, and electromagnetic methods.
In recent years, the particle swarm optimization (PSO) method has been successfully applied to the electromagnetic method in both underground and above ground environments. Based on the global convergence of the algorithm, the particle swarm optimization inversion of electromagnetic data is developed, which is mainly used for the qualitative interpretation of electromagnetic [21], [52]. In order to make better use of the swarm theory of particle swarm optimization, some scholars proposed to add constraints and carry out joint data inversion, and achieved good results [53], [54]. Particle swarm optimization each particle continually adjusts its speed and trajectory in the search space based on this personal best location and group best location information, moving closer towards the global optimum with each iteration. As seen in nature, this swarm displays a remarkable level of coherence and coordination despite the simplicity of its individual particles. In this case, Suppose that in a j-dimensional search space, M particles form a particle swarm, and the spatial position of the ith particle is X i = X i1 , X i1 , . . . , X ij , i = 1, 2, . . . , M . The PSO equation can be written as: (17) where P best and G best represent personal best location and group best location. k is the weight of the previous velocity; c 1 and c 2 are positive integers, called acceleration coefficients; r 1 and r 2 are two random numbers changing in the range of [0, 1]; and n is the number of iterations. As shown in Fig. 5, it can be seen that the inversion results of the PSO algorithm better reflect the physical characteristics of the theoretical model.
The inversion problem of the geophysical electromagnetic method can be seen as the interference of multiple locally optimal solutions. The deterministic inversion method is easy to fall into the local optimal solution in the search process. In particle swarm optimization, the model is composed of a set of models generated in a certain space, and the model has the characteristics of random distribution. Rich model types and intelligent model-driven form make the algorithm not only have a stronger ability to jump out of local optimal solution, but also can provide more information on the multi-solution of geophysical inversion problem.

B. BAYESIAN INVERSION
The Bayesian method is based on the Bayes theory for randomly searching model space. With the improvement of computer science, Bayesian inversion has been used to estimate model parameters in geophysical inverse problems in recent years. The Markov chain Monte Carlo (MCMC) method can be used to sample the posterior distribution efficiently. Based on the MCMC sampling algorithm, Geyer and Moller proposed a birth/death MCMC sampling algorithm [55] with a variable number of model parameters. This was improved by Green [56] to the reverse jump (RJ) MCMC algorithm, which laid the theoretical foundation of transdimensional Bayesian inversion. Malinverno [57] applied the trans-dimensional Bayesian method to the DC resistivity method and introduced it into geophysical inversion for the first time. It has since been widely used in the data inversion of geophysical methods, such as seismic, gravity, magnetic, and electromagnetic methods.
In recent years, the Bayesian method has been successfully applied in the electromagnetic method with artificial sources in airborne and marine environments. Based on the efficiency of calculations in the early stage, Bayesian inversion of electromagnetic data has been developed using a 1D forward operator, which is mainly used in quantitative interpretation and reliability evaluation of artificial-source data [58]- [62]. To use the dimensional variation and stinginess of the transdimensional Bayesian algorithm, some scholars have carried out trans-dimensional Bayesian inversion by combining 2D parameterization with a 1D forward operator and have achieved good results [33], [63] Generally in the Bayesian method, the fitting residual and noise model of the recorded data can be used to calculate the model likelihood, while the norm of the model vector (such as smoothness and sparsity) and prior information can be encapsulated in the prior distribution. In this case, the Bayesian equation can be written as where p(m) is the prior distribution that defined the model space containing all the models permitted by the prior information. The prior distribution is independent of the observed data. After obtaining the geophysical data, the model m in the prior distribution is reweighted according to the likelihood function p(d|m), and the posterior probability of the model is obtained. After sampling and weighting, in turn, the posterior distribution composed of the posterior probability of each model is obtained.
In the deterministic inversion method, a unique solution is sought. In the Bayesian method, the model is a random variable characterized by the model ensemble. The form of the solution is the probability distribution of model VOLUME 8, 2020 parameters, which is called a posteriori distribution (as is shown in Fig. 6). This posterior distribution can give the model set under the constraints of fitting residuals and prior information. It cannot only provide the estimation of model parameters that are often provided by deterministic inversion, but it also can determine the uncertainty of parameter estimation under the influence of multiple solutions of the inverse problem, observation error, and noise, so as to provide more information for data interpretation. In addition, because the prior distribution of Bayesian inversion makes it easy to introduce multiple model spaces, it is suitable for joint inversion of multiple parameters, such as parallel sampling of resistivity and charge rate parameters in TEM data [35], joint inversion of artificial-source electromagnetic data and MT data [64].

V. TEM PSEUDO-SEISMIC IMAGING
Because seismic exploration research was relatively mature, in the late 1980s, scholars proposed using pseudo-seismic interpretation methods in image processing of electromagnetic exploration data. Some progress has been made with the MT method and the TEM. There are two main kinds of imaging inversion. One is the time-frequency equivalent conversion method. Through an empirical equation, TEM data can be converted into equivalent plane wavefield data, and the reflection coefficient sequence is obtained to imitate the idea of quasi-seismic MT data. The second method combines the wavefield conversion method and pseudo-seismic migration imaging.
The challenge is to find a mathematical method to transform the transient electromagnetic field into a ''wave field,'' which is used to extract characteristics related to propagation in the electromagnetic response, to suppress or remove characteristics related to dispersion and attenuation during electromagnetic wave propagation, and to improve the accuracy in interpreting the TEM with reference to seismic exploration technology. Several researchers have proven that there is a mathematical equivalence between the electromagnetic diffusion equation and the seismic wave equation in the layered-earth medium [65]- [67]. The relationship between the time domain diffusion field Z m (x, t) and the virtual wave field is U (r, τ ) [67].
In the above equation, Z m (t) represents the transient electromagnetic field, U(τ ) represents the virtual wave field, and τ represents the virtual time variable.
The focus of Lee et al. [67] was to transform the wavefield simulation results of the corresponding geoelectric model into a time-domain electromagnetic response [68], [69]. In fact, the study of how to convert known electromagnetic field data into virtual wavefield data and use pseudo-seismic data processing will be beneficial to the subsequent migration procedure and the application of more complex imaging technology [70]- [72]. To improve its computational stability, the transient electromagnetic wave field conversion algorithm has been continuously improved in recent years. Qi et al. [73] introduced the preconditioned regularized conjugate gradient method to complete the transient electromagnetic full time-domain wavefield transformation. Wang and Li [74] used the continuous regularization method to solve the inversion problem of wavefield transformation, constructed a flattening functional, and implemented a numerical solution. Qingquan et al. [75] proposed extracting virtual wavefield information by using a sweep-time transform.

VI. CONCLUSION
The ultimate purpose of TEM exploration is to infer underground geological structures from observed data. Because of the complexity of the earth's internal structure, the reliability of TEM interpretation results depends on whether we can choose appropriate models to approximate the real geological structure. In addition to introducing the mathematical basis of the TEM inverse problem, this article also analyzes simple inversion methods, such as the least-squares method and damped least-squares method, the truncated generalized inverse matrix inversion method, the regularized inversion method, the PSO algorithm, TEM pseudo-seismic imaging, and Bayesian probability inversion.
The results of geophysical inversion are not unique. Reasons for this include the complexity of geological structures, the propagation of geophysical electromagnetic fields, the variety of geological noise, the particularity of the observed environment, the inherent equivalence of the geophysical field, the discreteness and limitations of the observation data, errors contained in observed data and the influence of other field sources. Due to the limitation of single component information and the lack of reliability of inversion results, multi-parameter joint inversion, 3D inversion considering the terrain and 3D inversion considering induced polarization (IP) is important for the study of TEM inversion at present or in the near future.
Deep learning is especially suitable for dealing with uncertain and unstructured information. With the increasing amount of data, data obtained under different geological conditions can be used to train a network, which makes the network more ''informed'' and improves its applicability and the reliability of results. When the collected data is input into the trained network for inversion calculation, the desired geo-electric structure can be obtained. With the progress in life science and computer science, human beings have more self-knowledge. More intelligent and effective algorithms will be developed rapidly for applying to geophysical inversion.
GUOQIANG XUE is currently a Researcher with the Key Laboratory of Mineral Resources, Institute of Geology and Geophysics (subordinate to the Innovation Academy for Earth Science, Chinese Academy of Sciences), and a Professor with the College of Earth and Planetary Sciences, University of Chinese Academy of Sciences, with an emphasis in transient electromagnetic exploration and applications. His research interests include TEM pseudo-seismic interpretation method, TEM tunnel predication studies, large loop TEM exploration technology, TEM response of short-offset excited by grounded electric sources, and analysis of time-varying point charge infinitesimal assumptions in the TEM field. He is an Associate Editor of JEEG and the Journal of Applied Geophysics.
HAI LI (Member, IEEE) received the bachelor's degree from Central South University, in 2011, and the Ph.D. degree from the University of Chinese Academy of Sciences, in 2016. He is currently an Associate Professor with the Key Laboratory of Mineral Resources, Institute of Geology and Geophysics (subordinate to the Innovation Academy for Earth Science, Chinese Academy of Sciences). His research interests include the data processing and imaging of TEM data, the inversion method for CSEM method, the induced polarization effect on TEM data, and the case history of CSEM method. He is a member of SEG and EAGE. XIN WU received the Ph.D. degree in geophysics from the University of Chinese Academy of Sciences, in 2018. He is currently a Postdoctoral Researcher with the Key Laboratory of Mineral Resources, Chinese Academy of Sciences, with an emphasis in theory, technology, and application of the transient electromagnetic method. VOLUME 8, 2020