Spherical Gravity Forwarding of Global Discrete Grid Cells by Isoparametric Transformation

Abstract

For regional or even global geophysical problems, the curvature of the geophysical model cannot be approximated as a plane, and its curvature must be considered. Tesseroids can fit the curvature, but their shapes vary from almost rectangular at the equator to almost triangular at the poles, i.e., degradation phenomena. Unlike other spherical discrete grids (e.g., square, triangular, and rhombic grids) that can fit the curvature, the Discrete Global Grid System (DGGS) grid can not only fit the curvature but also effectively avoid degradation phenomena at the poles. In addition, since it has only edge-adjacent grids, DGGS grids have consistent adjacency and excellent angular resolution. Hence, DGGS grids are the best choice for discretizing the sphere into cells with an approximate shape and continuous scale. Compared with the tesseroid, which has no analytical solution but has a well-defined integral limit, the DGGS cell (prisms obtained from DGGS grids) has neither an analytical solution nor a fixed integral limit. Therefore, based on the isoparametric transformation, the non-regular DGGS cell in the system coordinate system is transformed into the regular hexagonal prism in the local coordinate system, and the DGGS-based forwarding algorithm of the gravitational field is realized in the spherical coordinate system. Different coordinate systems have differences in the integral kernels of gravity fields. In the current literature, the forward modeling research of polyhedrons (the DGGS cell, which is a polyhedral cell) is mostly concentrated in the Cartesian coordinate system. Therefore, the reliability of the DGGS-based forwarding algorithm is verified using the tetrahedron-based forwarding algorithm and the tesseroid-based forwarding algorithm with tiny tesseroids. From the numerical results, it can be concluded that if the distance from observations to sources is too small, the corresponding gravity field forwarding results may also have ambiguous values. Therefore, the minimum distance is not recommended for practical applications.

1. Introduction

The curvature of the geophysical model in small/local areas can be represented as a plane. However, its curvature must be considered for regional or even global-scale issues. Because of the ellipsoidal shape of the Earth, the radius of curvature varies at different positions, resulting in changes in both the acceleration of gravity and the intensity and direction of the magnetic field with respect to longitude and latitude [1]. This change is particularly significant on a large-scale area and has important implications for studies such as the calculation of satellite orbits, crustal equilibrium, the inference of the internal structure of the Earth, the origin and evolution of the geomagnetic field, and geomagnetic navigation [2,3]. Therefore, when studying and solving these large-scale geophysical problems, the influence of earth curvature must be fully considered to obtain more accurate and reliable conclusions.

Generally, the mesh (also entitled as physical property unit or cell) of the geophysical model needs to be expressed and processed using spherical coordinates [4,5,6,7]. In the spherical coordinate system, the gravitational potential calculation can be roughly used in two methods: the spherical harmonic function method in the frequency domain and the direct integration method in the spatial domain. For the former, the spherical harmonic function method is more computationally efficient than the latter at low and middle orders (N < 360); however, it is affected by numerical instability at (super) high orders (N > 2700) [8]. Moreover, the spherical harmonic function method cannot properly consider the vertical extension of a given mass distribution [9]. The calculation results are marginally larger in amplitude, which is slightly inconsistent with the actual problem and unsuitable for accurately modeling the global/local gravity field [10].

How to obtain the analytical or numerical solution of gravitational fields has been one of the central problems in the selection of physical property units when performing the forward modeling of gravitational fields based on the direct integration method in the spherical coordinate system [11].

In practical applications, similar to physical property inversion in the Cartesian coordinate system, the spherical shell of the geophysical model is typically split into a set of discrete cells by using radius, latitude, and longitude as parameters. As far as the spherical discrete meshes are concerned, which mainly includes latitude-longitude and polyhedral meshes based on regular polyhedral subdivisions [12], such as point elements [13], line elements [14], surface elements [15], prism cells [16], tesseroids [17], combined cells of prism and tesseroid [18], or polyhedral cells [19].

In terms of fitting the spherical curvature of the geophysical model, the prism cell has a rigorous analytical solution for gravitational fields. However, when dealing with large-scale or global-scale problems, the model’s curvature needs to be corrected by reducing the distance between adjacent prism cells and changing their direction [20]. When the cells are excessively rough, resulting in the geophysical model is discontinuous and a reduction in the gravitational forward modeling calculation accuracy. Therefore, the prism cell cannot fit the model’s spherical curvature well [21]. Compared to other cells, tesseroids can efficiently fit the model’s spherical curvature. However, tesseroids do not have gravitational analytical solutions and must rely on numerical integration methods to obtain corresponding anomaly responses [5,6,22,23]. As latitude increases from the equator to the poles, tesseroids will produce a degradation phenomenon that tesseroids near the poles will degenerate into triangular prisms, generating a large amount of data redundancy, and their calculation error will accumulate toward high latitudes [23,24,25].

As far as polyhedral cells are concerned, Ren et al. [26] used Newton’s integral method to transform the volume integrals of gravity anomalies caused by homogeneous/non-homogeneous polyhedrons into surface integrals over polygonal faces [27]. Furthermore, singularity-free closed solutions are obtained for these surface integrals. The experimental results show that this method has high calculation accuracy. When polyhedral cells are used as an alternative cell/unit in large-scale or global-scale problems [28], significant polyhedral cells/units are required to fit the spherical curvature [21]. Furthermore, at present, gravity and magnetic forward modeling methods based on polyhedral cells generally use the Gauss/divergence theorem [29] to transform the volume integral of a polyhedral into an area/line integral of polygonal surfaces/lines of the polyhedral through various methods [28,30,31], and subsequently obtain the corresponding analytical solution/numerical solution [32]. How to effectively avoid the numerical ambiguity inherent in the divergence theorem [31] while exploiting the efficient computational properties of area/line integrals [33] is a core issue in the gravitational field forward modeling based on the divergence theorem.

In order to avoid the singularities at the poles and non-uniformity of latitude–longitude grids (e.g., as a particular case, tesseroids are latitude–longitude grids), many studies on spherical discrete grids based on regular polyhedrons have been carried out in recent years [12]. Based on the spherical discrete grid method of regular polyhedrons, the Discrete Global Grid System (DGGS) is a spherical fitting method different from the latitude–longitude grids [34]. The DGGS [34,35] are composed of a series of discrete global grids, which are considered to be a kind of earth-fitting grid. DGGS can divide the sphere into multi-level grids with approximate or equal area and shape and a continuous scale. At any resolution, considering the fitting curvature, it can seamlessly cover the global, and the grids with different resolutions form a highly consistent hierarchical structure [36]. Therefore, DGGS grids are the best choice for discretizing the sphere into cells with an approximate shape and continuous scale. To meet different needs, there are many ways to use DGGS to deal with large-scale geospatial data, including Hierarchical Equal Area isoLatitude Pixelisation of a 2-sphere (HEALPix) [37], Open Equal Area Global GRid (OpenEAGGR) [38], Hierarchical Hexagonal Hierarchical Spatial Indexing System (H3) [39], Science Collaboration Environment for New Zealand Grid (SCENZ-Grid) [40], regularized Hierarchical Equal Area isoLatitude Pixelisation of a 2-sphere (rHEALPix) [41], Geographic Grid System (geogrid), and International Society for the Advancement of Spatial Data Handling (ISEA) [42]. When dealing with large-scale heterogeneous geospatial data, DGGS has the characteristics of convenient management, fast storage, and assembly. Currently, it has become an effective tool for efficient exploration and visualization [43].

Unlike triangles and rhombuses, both of which have edge adjacencies and are therefore inconsistent, the pentagonal/hexagonal cell, which is a kind of DGGS cell (prisms obtained from DGGS grids, which contain spherical pentagons and hexagons) and derived from the ISEA3H grid [44], has only edge neighbors [31,45]. And it is easier to achieve than triangles and rhombuses in the adjacent search of pentagonal/hexagonal grids [12,46]. Compared with square grids (as a particular case, tesseroid are also a kind of square grids), hexagonal grids have excellent angular resolution and also show unified and clear adjacency [34], and their discrete distance measures are closer to Cartesian distances [47].

Essentially, the gravitational forward modeling of DGGS cells in the space domain involves integral calculations. Although tesseroids do not have gravitational analytical solutions, their integration limits are well-defined, so using numerical integration methods can obtain corresponding gravity anomaly responses. However, DGGS cells neither have gravitational analytical solutions nor fixed integration limits.

To solve optimization problems (including integrals) for dealing with complex structures, the finite element method (FEM), as a cornerstone of computational mechanics, has become an indispensable tool of computational in scientific and engineering calculations and numerical simulations [48,49,50,51,52], which plays a crucial role in geophysical research exploring the large-scale atmosphere, oceans, and crust [50]. The target object analyzed by FEM usually has a complex structure [49,53]. Therefore, selecting an appropriate cell is vital to improve the discretization of the target object. Isoparametric elements are widely used in finite element analysis of complex structures because of its excellent adaptability and coordination [54,55].

Presently, to strengthen the use of isoparametric elements in finite element analysis and apply them to solve various practical engineering problems, many scholars have put forward their relevant research insights and achieved specific research results. For example, Jeng and Wexler [56] proposed a new isoparametric finite element numerical algorithm that resolved the variational solution of the boundary integral equations of a general 3D field, which is based on the variational principle and not only approximates the unknown source distributions by the calculation of polynomials but also models surfaces at higher-order, to reduce the errors caused by geometric modeling [56,57].

Lehrenfeld [58] proposed a new geometry-based isoparametric non-fitting FEM for solving problems on non-fitting interfaces, or marginals, and partial differential equations on surfaces. The method is simple to use and relatively new and practical, capable of performing higher-order numerical integration calculations using level set functions over domains as prescribed, with good robustness and high accuracy in dealing with complex geometries [58,59]. Belarbi et al. [60] proposed a new hierarchical isoparametric finite element modeling technique to obtain 2D quadrilateral isoparametric elements that satisfy the “face-core” condition at the interface through a hierarchical method for analyzing the bending problem in sandwich panels. The method can satisfy the continuity of stresses and displacements at the interface, and the model comparison tests show that the algorithm has good accuracy and convergence speed [60,61].

Therefore, drawing a lesson from the unit of analysis process in the FEM [48,62,63], an isoparametric transformation [64,65] is used to transform a non-regular DGGS cell in the system/spherical coordinate system to a regular hexagonal prism in the local/Cartesian coordinate system to determine their integration limits. Then, the gravitational field forward modeling calculation of the DGGS cell is realized under the spherical coordinate system using an isoparametric transformation with the shape functions, integration points, and their integration weights.

2. The Principle of Forward Modeling Tensors and Gravity Gradients

For all observation points, the gravitational effect of each physical property unit can be calculated one by one according to the superposition principle and the corresponding anomalous response can be obtained by cumulative summation. Therefore, the gravitational field forward modeling can be written in matrix form:

$$\mathit{d}=\mathit{Gm}$$

(1)

where $\mathit{d}$ is observation data vector, which can be specifically $g_z$, $g_{xx}$, $g_{xy}$, $g_{xz}$, $g_{yy}$, $g_{yz}$ or $g_{zz}$, etc., with a length of $N_d$; parameter vector $\mathit{m}$ is the densities value of the units; $\mathit{G}$ is the corresponding forward operator (consistent with the inversion kernel matrix) with size of $N_{\mathit{d}}\times N_{\mathit{m}}$; $N_{\mathit{m}}$ and $N_{\mathit{d}}$ are the units’ number and observation points, respectively.

The forward integral kernel of gravitational fields differs in the Cartesian coordinate system and the spherical coordinate system [66]. Taking the tesseroid as an example, as shown in Figure 1, it is assumed that a homogeneous geological body Q with a density $\rho \left(r,\varphi ,\lambda \right)$ exists in the spherical coordinate system. The gravity vector $g_{\alpha }\left(P\right)$ and the gravity gradient tensor $g_{\alpha \beta }\left(P\right)$ at the source external space $P\left(r,\varphi ,\lambda \right)$ of the Q can be expressed as [66], respectively:

$$\left\{\begin{array}{c}{g}_{\alpha }\left(P\right)\\ g_{\alpha \beta }\left(P\right)\end{array}\right\}=\upsilon \rho {\int }_{{\lambda }_1}^{{\lambda }_2}{\int }_{{\varphi }_1}^{{\varphi }_2}{\int }_{r_1}^{r_2}\frac{\kappa }{{\mathit{\ell }}^3}\left\{\begin{array}{c}{\Delta }_{\alpha }\\ 3{\Delta }_{\alpha }{\Delta }_{\beta }{\ell }^{-2}-{\delta }_{\alpha \beta }\end{array}\right\}d{r}^{\prime }d{\varphi }^{\prime }d{\lambda }^{\prime }$$

(2)

where $\alpha ,\beta \in \left\{x,y,z\right\}$, $\upsilon$ = $6.674\times {10}^{-11}$${\mathrm{m}}^3·\mathrm{k}{\mathrm{g}}^{-1}·{\mathrm{s}}^{-2}$ is the gravitational constant, ${\Delta }_x$ = $r^{\prime }\left(\cos \varphi \sin {\varphi }^{\prime }-\sin \varphi \cos {\varphi }^{\prime }\cos \left({\lambda }^{\prime }-\lambda \right)\right)$, ${\Delta }_y$ = $r^{\prime }\cos {\varphi }^{\prime }\sin \left({\lambda }^{\prime }-\lambda \right)$, ${\Delta }_z$ = $r^{\prime }\cos \psi -r$, $\kappa$ = $r^{\prime 2}\cos {\varphi }^{\prime }$, ℓ = $\sqrt{{r^{\prime }}^2+r^2-2r^{\prime }r\cos \psi }$, $\cos \psi$ = $\sin \varphi \sin {\varphi }^{\prime }+\cos \varphi \cos {\varphi }^{\prime }\cos \left({\lambda }^{\prime }-\lambda \right)$.

Since the tesseroid has no analytical solution, it needs to be processed by numerical integration methods such as the Taylor-series expansion method [23] and the 3D Gauss–Legendre quadrature (GLQ) integration [67,68,69]. Therefore, the GLQ method is used to calculate the gravity vector $g_{\alpha }\left(P\right)$ and the gravity gradient tensor $g_{\alpha \beta }\left(P\right)$. In order to facilitate the calculation, the integral kernel f [6] is introduced and expressed as a general form of triple integrals [70]:

$$\upsilon \rho {\int }_{{\lambda }_s={\lambda }_1}^{{\lambda }_2}{\int }_{{\varphi }_s={\varphi }_1}^{{\varphi }_2}{\int }_{r_s=r_1}^{r_2}f(r_0,{\varphi }_0,{\lambda }_0)d{r}_{s}d{\varphi }_{s}d{\lambda }_s$$

(3)

Using the triple-Gauss quadrature approximation calculation Equation (3) is generally necessary [5]. However, the corresponding integration coefficients must be determined in different integral intervals for various order integrals. It is assumed that the integral interval of Equation (3) along longitude, latitude, and radial direction is $\left[a,b\right]$, and the integration point is $x_k$, and the following form is constructed to obtain the corresponding approximate calculation results:

$${\int }_a^{b}f\left(x\right)dx=\frac{b-a}{2}\sum _{k=1}^n{w}_{k}f\left(x_k\right)$$

(4)

where n is the number of integration points, $w_k$ is the integration weight

$$w_k=\frac{2}{\left(1-x_k^2\right){\left({P_n}^{\prime }\left(x_k\right)\right)}^2}$$

(5)

where ${P_n}^{\prime }\left(x\right)$ is the first derivative of the n-order Legendre function $P_n\left(x\right)$. After scaling the integral interval of Equation (4) to the interval $\left[-1,1\right]$, the corresponding zero point of $P_n\left(x\right)$ is the Gauss point. The root of the higher-order Legendre function $P_n\left(x\right)$ can be obtained using the recursive relation of its first derivative ${P_n}^{\prime }\left(x\right)$.

Using Equation (4), Equation (3) can be re-expressed as a form of the triple numerical integration [70]

$$\upsilon \rho V\sum _{k=1}^{n_{\lambda }}\sum _{j=1}^{n_{\varphi }}\sum _{i=1}^{n_r}{W}_i^r{W}_j^{\varphi }{W}_k^{\lambda }f\left({r^{\prime }}_i,{{\varphi }^{\prime }}_j,{{\lambda }^{\prime }}_k\right)$$

(6)

where $n_{\lambda },n_{\varphi },n_r$ are the number of integration points in longitude, latitude, and radial direction, respectively, and $W_i^r,W_j^{\varphi },W_k^{\lambda }$ are the corresponding integration weights. In practical applications, the number of integration points can control the calculation’s accuracy. The higher the number of integration points, the better the accuracy of the GLQ, but the corresponding computational efficiency will be lower [68,71]. In this article, $n_{\lambda },n_{\varphi },n_r$ is equal to 2, $V=\left(r_2-r_1\right)\left({\varphi }_2-{\varphi }_1\right)\left({\lambda }_1-{\lambda }_2\right)/8$.

3. Forward Modeling of Gravitational Fields Based on DGGS Cells

The tesseroids and triangular or rhombus prisms derived from a discrete spherical shell lack consistency or contain the presence of the degradation phenomenon that spherical discrete grids obtained by subdividing the spherical shell with a regular geometry unit differ in spatial morphology [10,12]. As shown in Figure 2, based on the definition of the DGGS, the first-level grid is constructed on the spherical surface (e.g., other precision grids can also be further constructed on the sphere), including 12 pentagonal and 30 hexagonal DGGS grids. Furthermore, using these grids, the spherical shells within the research area are discretized layer by layer along the radial direction; that is, each layer of the spherical shell is discretized into 12 spherical pentagonal prisms and 30 spherical hexagonal prisms, which are referred to as DGGS cells here. Namely, DGGS cells contain prisms composed of spherical pentagons and spherical hexagons.

Compared with the tesseroid, the DGGS cell has neither a corresponding analytical solution nor a fixed integral limit. Therefore, drawing a lesson from the unit analysis process in the FEM, an isoparametric transformation is used to transform a non-regular DGGS cell from the system coordinate system to a regular hexagonal prism in the local coordinate system.

3.1. The Isoparametric Transformation of DGGS Cells

In finite element analysis, the solution region is frequently partitioned into a finite number of small cells, such as triangles, quadrilaterals, tetrahedrons, rectangles, and so on. Those small cells can be connected to each other through their nodes. In order to obtain the corresponding field values by the FEM, it is necessary to calculate the area/volume integral of each unit in order to establish the system of equations for the corresponding field functions f [72]. Since the forward modeling of gravitational fields in this paper only considers the volume effect, only the volume integral of each small unit $v_e$ is considered. Usually, the volume integral of the field function f can be expressed as

$${\int }_{v_e}fdV={\int }_{v_e}f\left(x,y,z\right)dxdydz$$

(7)

In Figure 1, the field function f cannot obtain the corresponding integral limit for a non-regular hexagonal prism in the system coordinate system. Therefore, it is frequently necessary to introduce a specific geometric transformation to transform the DGGS cell into a regular hexagonal prism to gain a distinct integral limit. If and only if both the field function interpolation and coordinate transformation use the same number of node parameters and the same interpolation function, it is called the isoparametric transformation [65,73]. Isoparametric elements use a mathematical mapping to convert coordinates from one system to another, where the original system is called the system/global coordinate system and the second system is called the natural/local coordinate system.

As shown in Figure 3, for both the system coordinate system and the local coordinate system, the relationship between derivatives and volume elements can be established through isoparametric transformation based on interpolation functions or shape functions N [53,65,74]. The shape function is an essential mathematical tool in finite element analysis and can represent displacements, temperatures, or wave velocities in units [53]. They serve as a bridge to transform continuous problems into equivalent discrete problems. For a typical finite element with $n_e$ node, let the displacements $u\in \left(x,y,z\right)$ at any point within the element be approximated as:

$$u=\sum _{i=1}^{n_e}{N}_i{u}_i$$

(8)

According to the chain rule in calculus, the partial derivative concerning $\xi$ is obtained for the shape function $N_i$ of the i node of the DGGS cell. Then,

$$\frac{\partial N_i}{\partial \xi }=\frac{\partial N_i}{\partial x}\frac{\partial x}{\partial \xi }+\frac{\partial N_i}{\partial y}\frac{\partial y}{\partial \xi }+\frac{\partial N_i}{\partial z}\frac{\partial z}{\partial \xi }$$

(9)

For the other two coordinates $\left(\eta ,\zeta \right)$, a similar expression can also be written. Now, the partial derivatives of $N_i$ to $\xi$, $\eta$ and $\zeta$ are written in the matrix form

$$\left\{\begin{array}{c}\frac{\partial N}{\partial \xi }\\ \frac{\partial N}{\partial \eta }\\ \frac{\partial N}{\partial \zeta }\end{array}\right\}=\left[\begin{array}{ccc}\frac{\partial x}{\partial \xi }& \frac{\partial y}{\partial \xi }& \frac{\partial z}{\partial \xi }\\ \frac{\partial x}{\partial \eta }& \frac{\partial y}{\partial \eta }& \frac{\partial z}{\partial \eta }\\ \frac{\partial x}{\partial \zeta }& \frac{\partial y}{\partial \zeta }& \frac{\partial z}{\partial \zeta }\end{array}\right]\left\{\begin{array}{c}\frac{\partial N}{\partial x}\\ \frac{\partial N}{\partial y}\\ \frac{\partial N}{\partial z}\end{array}\right\}=J\left\{\begin{array}{c}\frac{\partial N}{\partial x}\\ \frac{\partial N}{\partial y}\\ \frac{\partial N}{\partial z}\end{array}\right\}$$

(10)

where J is a $Jacobi$ matrix. In finite element analysis, the physical meaning of $\left|J\right|$ is the volume of the corresponding small unit. Thus, according to Equation (10), there are

$$\begin{array}{cc}J\hfill & =\frac{\partial \left(x,y,z\right)}{\partial \left(\xi ,\eta ,\zeta \right)}=\left[\begin{array}{ccc}\frac{\partial x}{\partial \xi }& \frac{\partial x}{\partial \eta }& \frac{\partial x}{\partial \zeta }\\ \frac{\partial y}{\partial \xi }& \frac{\partial y}{\partial \eta }& \frac{\partial y}{\partial \zeta }\\ \frac{\partial z}{\partial \xi }& \frac{\partial z}{\partial \eta }& \frac{\partial z}{\partial \zeta }\end{array}\right]\hfill \\ \hfill & =\left[\begin{array}{ccc}{\displaystyle \sum _{i=1}^{n_v}}\frac{\partial N_i^{\prime }}{\partial \xi }{x}_i& {\displaystyle \sum _{i=1}^{n_v}}\frac{\partial N_i^{\prime }}{\partial \eta }{x}_i& {\displaystyle \sum _{i=1}^{n_v}}\frac{\partial N_i^{\prime }}{\partial \zeta }{x}_i\\ {\displaystyle \sum _{i=1}^{n_v}}\frac{\partial N_i^{\prime }}{\partial \xi }{y}_i& {\displaystyle \sum _{i=1}^{n_v}}\frac{\partial N_i^{\prime }}{\partial \eta }{y}_i& {\displaystyle \sum _{i=1}^{n_v}}\frac{\partial N_i^{\prime }}{\partial \zeta }{y}_i\\ {\displaystyle \sum _{i=1}^{n_v}}\frac{\partial N_i^{\prime }}{\partial \xi }{z}_i& {\displaystyle \sum _{i=1}^{n_v}}\frac{\partial N_i^{\prime }}{\partial \eta }{z}_i& {\displaystyle \sum _{i=1}^{n_v}}\frac{\partial N_i^{\prime }}{\partial \zeta }{z}_i\end{array}\right]\hfill \\ \hfill & =\left[\begin{array}{cccc}\frac{\partial N_1^{\prime }}{\partial \xi }& \frac{\partial N_2^{\prime }}{\partial \xi }& \cdots & \frac{\partial N_{n_v}^{\prime }}{\partial \xi }\\ \frac{\partial N_1^{\prime }}{\partial \eta }& \frac{\partial N_2^{\prime }}{\partial \eta }& \cdots & \frac{\partial N_{n_v}^{\prime }}{\partial \eta }\\ \frac{\partial N_1^{\prime }}{\partial \zeta }& \frac{\partial N_2^{\prime }}{\partial \zeta }& \cdots & \frac{\partial N_{n_v}^{\prime }}{\partial \zeta }\end{array}\right]\left[\begin{array}{ccc}{x}_1& y_1& z_1\\ x_2& y_2& z_2\\ ⋮& ⋮& ⋮\\ x_{n_v}& y_{n_v}& z_{n_v}\end{array}\right]\hfill \end{array}$$

(11)

Here, $n_v$ is the number of unit nodes. Then Equation (10) can be rewritten as

$$\left\{\begin{array}{c}\frac{\partial N}{\partial x}\\ \frac{\partial N}{\partial y}\\ \frac{\partial N}{\partial z}\end{array}\right\}=J^{-1}\left\{\begin{array}{c}\frac{\partial N}{\partial \xi }\\ \frac{\partial N}{\partial \eta }\\ \frac{\partial N}{\partial \zeta }\end{array}\right\}$$

(12)

where the inverse matrix $J^{-1}=J^{*}/\left|J\right|$, $\left|J\right|$, and $J^{*}$ are the determinant and adjoint matrices of J, respectively.

After the geometric transformation from the system coordinate system to the local coordinate system, the integral of the field function f in the integral domain $v_e$ in the original system coordinate system can have a definite integral limit in the local coordinate system. Assuming that the corresponding integral limit is $\left[a_{\xi },b_{\xi }\right]\times \left[a_{\eta },b_{\eta }\right]\times \left[a_{\zeta },b_{\zeta }\right]$, then the integral of the field function f in the local coordinate system can be expressed as

$${\int }_{a_{\zeta }}^{b_{\zeta }}{\int }_{a_{\eta }}^{b_{\eta }}{\int }_{a_{\xi }}^{b_{\xi }}{f}^{*}\left(\xi ,\eta ,\zeta \right)d\xi d\eta d\zeta$$

(13)

where,

$$f^{*}\left(\xi ,\eta ,\zeta \right)=f\left(x\left(\xi ,\eta ,\zeta \right),y\left(\xi ,\eta ,\zeta \right),z\left(\xi ,\eta ,\zeta \right)\right)\left|J\right|$$

(14)

Using the triple-Gauss quadrature [67,75] in conjuction with Equation (7), Equation (13) can be rewritten as:

$$\begin{array}{c}\begin{array}{c}I={∰}_{v_e}f\left(x,y,z\right)dV={\displaystyle \sum _{i=1}^{n_i}}{\displaystyle \sum _{j=1}^{n_j}}{\displaystyle \sum _{k=1}^{n_k}}{w}_i{w}_j{w}_{k}f\left({\xi }_i,{\eta }_i,{\zeta }_i\right)det\left(J\right)\end{array}\hfill \end{array}$$

(15)

where $n_i,n_j,n_k$ and $w_i,w_j,w_k$ are the number of integration points and the integration weights of the Cartesian coordinate system along the $x,y,z$ axis direction, respectively.

Referring to the article [53,76] and according to the definition of FEM, ensure that the shape function satisfies the following three conditions [77]: (a) only at node i can have a unit value while all other nodes have zero value; (b) disappears on any boundary of element that does not include node i; (c) the determinant of the Jacobian matrix of the resulting shape function concerning the node coordinates is equal to the volume of the regular hexagon prism. Then, the following set of shape functions is given

$$\begin{array}{cc}& \begin{array}{c}{N}_1=-\frac{1}{12}\left(\zeta +1\right)\left(\xi +1\right)\left(4{\eta }^2-3\right)\hfill \\ N_2=\frac{1}{12}\left(\zeta +1\right)\left(2\xi +1\right)\left(2\eta +\sqrt{3}\right)\eta \hfill \\ N_3=-\frac{1}{12}\left(\zeta +1\right)\left(2\xi -1\right)\left(2\eta +\sqrt{3}\right)\eta \hfill \\ N_4=\frac{1}{12}\left(\zeta +1\right)\left(\xi -1\right)\left(4{\eta }^2-3\right)\hfill \\ N_5=-\frac{1}{12}\left(\zeta +1\right)\left(2\xi -1\right)\left(2\eta -\sqrt{3}\right)\eta \hfill \\ N_6=\frac{1}{12}\left(\zeta +1\right)\left(2\xi +1\right)\left(2\eta -\sqrt{3}\right)\eta \hfill \\ N_7=\frac{1}{12}\left(\zeta -1\right)\left(\xi +1\right)\left(4{\eta }^2-3\right)\hfill \\ N_8=-\frac{1}{12}\left(\zeta -1\right)\left(2\xi +1\right)\left(2\eta +\sqrt{3}\right)\eta \hfill \\ N_9=\frac{1}{12}\left(\zeta -1\right)\left(2\xi -1\right)\left(2\eta +\sqrt{3}\right)\eta \hfill \\ N_{10}=-\frac{1}{12}\left(\zeta -1\right)\left(\xi -1\right)\left(4{\eta }^2-3\right)\hfill \\ N_{11}=\frac{1}{12}\left(\zeta -1\right)\left(2\xi -1\right)\left(2\eta -\sqrt{3}\right)\eta \hfill \\ N_{12}=-\frac{1}{12}\left(\zeta -1\right)\left(2\xi +1\right)\left(2\eta -\sqrt{3}\right)\eta \hfill \end{array}\end{array}$$

(16)

3.2. Integral Weights of DGGS Cells

In finite element analysis, integration points are typically placed at specific locations, such as their equipartition points for line/surface elements (see

Table 1. 1D Gauss quadrature scheme for line elements.

Number of Integration Points n	Integration Position ${\mathit{\xi }}_{\mathit{i}}$	Integration Weights ${\mathit{w}}_{\mathit{i}}$	Position of Points
1	${\xi }_1=0$	$w_1=2$
2	${\xi }_1=+1/\sqrt{3}$	$w_1=1$
	${\xi }_2=+1/\sqrt{3}$	$w_2=1$
3	${\xi }_1=-1/\sqrt{3/5}$	$w_1=5/9$
	${\xi }_2=0$	$w_2=8/9$
	${\xi }_3=+1/\sqrt{3/5}$	$w_3=5/9$

and

Table 2. 2D Gauss quadrature scheme for rectangular elements.

Number of Integration Points n	Integration Position ${\mathit{\xi }}_{\mathit{i}}$	Integration Position ${\mathit{\eta }}_{\mathit{i}}$	Integration Weights ${\mathit{w}}_{\mathit{i}}$	Position of Points
1	${\xi }_1=0$	${\eta }_1=0$	$w_1=4$
4	${\xi }_1=-1/\sqrt{3}$	${\eta }_1=-1/\sqrt{3}$	$w_1=1$
	${\xi }_2=+1/\sqrt{3}$	${\eta }_2=-1/\sqrt{3}$	$w_2=1$
	${\xi }_3=-1/\sqrt{3}$	${\eta }_3=+1/\sqrt{3}$	$w_3=1$
	${\xi }_4=+1/\sqrt{3}$	${\eta }_4=+1/\sqrt{3}$	$w_4=1$
9	${\xi }_1=-1/\sqrt{3/5}$	${\eta }_1=-1/\sqrt{3/5}$	$w_1=25/81$
	${\xi }_2=0$	${\eta }_2=-1/\sqrt{3/5}$	$w_2=40/81$
	${\xi }_3=+1/\sqrt{3/5}$	${\eta }_3=-1/\sqrt{3/5}$	$w_3=25/81$
	${\xi }_4=-1/\sqrt{3/5}$	${\eta }_4=0$	$w_4=40/81$
	${\xi }_5=0$	${\eta }_5=0$	$w_5=64/81$
	${\xi }_6=+1/\sqrt{3/5}$	${\eta }_6=0$	$w_6=40/81$
	${\xi }_7=-1/\sqrt{3/5}$	${\eta }_7=+1/\sqrt{3/5}$	$w_7=25/81$
	${\xi }_8=0$	${\eta }_8=+1/\sqrt{3/5}$	$w_8=40/81$
	${\xi }_9=+1/\sqrt{3/5}$	${\eta }_9=+1/\sqrt{3/5}$	$w_9=25/81$

) and their line/surface equipartition points and barycenter (see Figure 4) for volume elements [78].

There is some challenge in determining the location of the integration points and the integration weights for a regular hexagon or hexagonal prism. As shown in Figure 5a, a cross-section of a regular hexagonal prism is taken as the object of study, and $n_{xy}$ integration points are arranged on the horizontal hexagon in Figure 5b; while the 1D Gauss quadrature scheme (see Table 1) is employed along the vertical direction of the regular hexagonal prism in Figure 5, setting the $n_r$ (the quantity of 1D Gauss integration points) layer of integration points and the corresponding weight coefficients $w_k^r$ for the k-th layer of integration points. By the definition of the triple-Gauss quadrature, Equation (13) is rewritten as

$$I\left(f\right)=\sum _{j=1}^{n_{xy}\times n_r}{w}_{j}f\left({\xi }_j,{\eta }_j,{\zeta }_j\right)det\left(J\right)$$

(17)

where, the triple-Gauss quadrature integration weights $w_j$ = $w_i^{xy}{w}_k^r$, subscripts $j\equiv (k-1)\times n_r+i$, the number of integration points $n_w$ = $n_{xy}\times n_r$, $1\le j\le n_w$, $1\le i\le n_{xy}$, $1\le k\le n_r$, and ${\xi }_j,{\eta }_j,{\zeta }_j$ are the coordinates of the j-th integration point, respectively.

In this paper, especially, $n_r$ = 3 and $n_{xy}$ = 7, the 1D Gauss integration scheme is used along the vertical direction of the regular hexagonal prism (see Table 1) and the customized integration scheme as shown in Figure 5b.

4. Forward Modeling of Gravitational Fields Based on Arbitrary Tetrahedral Cells

How to analyze the accuracy of the gravitational field forward modeling calculation is essential in verifying the correctness of the DGGS-based forwarding algorithm of the gravitational field under the spherical coordinate system proposed in this paper. Since the DGGS cell is a particular case of the polyhedral cell, the polyhedral-based forwarding algorithm can be used to verify the correctness of the DGGS-based forwarding algorithm proposed in this paper [26,28]. However, most existing studies on polyhedral-based forwarding algorithms focus on the Cartesian coordinate system, whereas this paper is involved in the spherical coordinate system. Therefore, there are several points to consider:

• Describing the two spherical surfaces (curved surfaces) above and below the DGGS cell in the spherical coordinate system using the facets of polyhedrons is somewhat challenging;
• A significant number of units and nodes are required if polyhedrons (e.g., tetrahedrons) are directly used to fit the DGGS cell [21,28];
• The gravity field integration kernels under the two coordinate systems are not the same [23,26,66].

Based on the above three factors, considering that the DGGS cell does not have the gravitational field analytical solution, and the polyhedral cell does not have the analytical solution/closed solution of gravitational fields under the spherical coordinate system. Therefore, referring to Grombein et al. [23], the forward integral kernel in the Cartesian coordinate system of the tesseroid is given, and a tetrahedron-based forwarding algorithm is introduced to verify the effectiveness and reliability of the DGGS-based forwarding algorithm proposed in this paper (Figure 6).

Referring to Duczek et al. [79], the shape functions of the tetrahedron can be written as follows:

$$N_1=1-x-y-z,N_2=x,N_3=y,N_4=z$$

(18)

Subsequently, the tetrahedron-based forwarding algorithm of the gravitational field can be calculated using the Gauss integration.

Cao et al. [80] employed the 3D Hammer integration to derive the gravitational field analytical solution for the tetrahedron. The following Equation (2) is utilized to calculate the volume integral of the tetrahedron:

$$\begin{array}{c}\underset{0}{\overset{1}{\int }}\underset{0}{\overset{1-L_1}{\int }}\underset{0}{\overset{1-L_1-L_2}{\int }}F\left(L_1,L_2,L_3,L_4\right)d{L}_3{L}_2{L}_1\hfill \\ =A_{1}F\left(\frac{1}{4},\frac{1}{4},\frac{1}{4},\frac{1}{4}\right)+B_4\left\{F\left(a,b,b,b\right)+F\left(b,a,b,b\right)+F\left(b,b,a,b\right)+F\left(b,b,b,a\right)\right\}\hfill \end{array}$$

(19)

where $L_1,L_2,L_3,L_4$ are volume or natural coordinates, and f and F correspond to the field functions of the system and the natural coordinate system (a kind of local coordinate system), respectively. Parameters $A_1$, $B_4$, a, and b are detailed in

Table 3. Integration weights and position of points of the 3D Hammer integration.

Order	Number of Integration Points	Integration Weights	Position of Points
1	1	$A_1$= 1	1/4, 1/4, 1/4, 1/4
2	4	$B_4$= 1/24	a= 0.585 410 20 b = 0.138 196 60
5	3	a= 1/2, b = 1/6

.

Then, Equation (18) is used as the shape function for tetrahedrons. The 3D first-order Haar integration scheme is utilized to obtain the corresponding gravity vector and its gradient tensor value using Equation (19).

When the integration point is situated within the isoparametric element, an isoparametric transformation is utilized to establish the connection between the integration point in the local coordinate system and the system coordinate system. This relationship ensures the avoidance of the numerical ambiguity problem that tends to occur in traditional volume integration during the implementation of the volume integration process [27,28].

In order to validate the effectiveness and reliability of the DGGS-based forwarding algorithm proposed in this paper, the following three phases are required:

Phase 1. Since the forward integral kernel varies in different coordinate systems, and considering the influence of the number of integration points and KU criteria [22] on the calculation accuracy of tesseroids, it is therefore necessary to compare the computational accuracy of the tetrahedron-based forwarding algorithm and the analytical solution of right rectangular prisms [80], under the Cartesian coordinate system, to verify the correctness of the tetrahedron-based forwarding algorithm.

Phase 2. Considering that the DGGS cell has neither an analytical solution nor a clear integration limit in the spherical coordinate system. Therefore, it is necessary to compare the difference in calculation accuracy between tetrahedrons and tesseroids to verify the correctness of the tetrahedron-based forwarding algorithm Equation (2) in the spherical coordinate system [23].

Phase 3. Based on the algorithm verification of the above two phases in the spherical coordinate system, the difference in calculation accuracy between the tetrahedron-based and DGGS-based forwarding algorithms is compared in order to verify the correctness of the DGGS-based forwarding algorithm proposed in this paper based on isoparametric transformation.

5. Algorithm Verification

5.1. Phase 1: To Verify the Tetrahedron-Based Forwarding Algorithm in the Cartesian Coordinate System

In the first step, to verify the correctness of a tetrahedron-based forwarding algorithm [80]. A cube model with a density of 0.3 g/cm³ is specially constructed with a center point at (0 m, 0 m, 650 m) and a side length of 400 m, as shown in Figure 7. The observation grid ranges from x: −1000 m to 1000 m; y: −1000 m to 1000 m. The spacing of the observation points is 50 m × 50 m, and the points are located on the earth’s surface.

In the second step, the cube is subdivided into tetrahedrons using Delaunay triangulation [81]. It should be noted that if the quality of the tetrahedrons obtained from the Delaunay triangulation discretization is poor, the results of the calculations using Equation (19) may be susceptible to distortion. To ensure calculation accuracy, when using Delaunay triangulation software TetGen (version 1.5) [82], discretizing the cube is necessary to ensure that the odd intensity is less than 1.4.

In the third step, the gravity vector and its gradient tensor values are obtained using the analytical solution of the right rectangular prism [83] and the tetrahedron-based forwarding algorithm [80]. By comparing with Figure 8, it can be seen that under the Cartesian coordinate system, using the forward integral kernel [80], the tetrahedron-based forward modeling using 3D Hammer integration has high computational accuracy. This demonstrates the tetrahedron-based forwarding algorithm as an instrument for the subsequent analysis of the DGGS-based forwarding algorithm.

Generally, mesh generation is divided into structured and unstructured mesh generation. The so-called structured mesh generation refers to the number of cells adjacent to each internal node after the subdivision is the same, such as in the case of 3D, where the generated cells are generally hexahedrons; unstructured is the opposite of structured, which means that each node can have different adjacent cells. It is easy to make the geometric fitting error [84] larger by using the structured mesh to fit the curved edge/surface, leading to a large error in the numerical solution [80]. Moreover, the nodes of the structured meshes cannot be reasonably distributed in a gradient shape, so the error caused by the geometric discretization of the model cannot be eliminated to the maximum extent. Even if it can be distributed in a certain gradient, it is easy to reduce the shape quality of the structured cells, thus affecting the accuracy of the numerical calculation [85]. Compared with the structured mesh generation method, tetrahedrons have a certain degree of variability and flexibility, i.e., the nodes and units obtained using Tetgen are distributed in a gradient shape in a certain way [86], or some discrete points are inserted (using Tetgen software for multiple time refinement) in the gradient change area [87], which can greatly reduce the geometric fitting error [84].

Cao et al. [80] pointed out that the computational error of the tetrahedron-based forwarding algorithm is proportional to the quality of the tetrahedrons and the geometric fitting error for fitting the model using tetrahedrons [84,88]. Figure 9 shows that the geometric fitting error induced by fitting the cube with tetrahedrons in the Cartesian coordinate system is small, and the residuals between the results of the tetrahedron-based forwarding algorithm Equation (19) and the analytical solutions [83] for the cubic model.

5.2. Phase 2: To Verify the Tetrahedron-Based and Tesseroid-Based Forwarding Algorithms in the Spherical Coordinate System

To facilitate comparative analysis, the second stage uses the four model settings given by Uieda et al. [5], whose names are Experiment 1 (pole), Experiment 2 (equator), Experiment 3 (260 km), and Experiment 4 (30° size). Each of them contains a tesseroid with a thickness of 1 km, and its latitude and longitude range are 89∼90° N/0∼1° E, 0∼1° N/0∼1° E, 89∼90° N/0∼1° E, and 60∼90° N/0∼30° E, respectively. All grids corresponding to the above four models are composed of 51 × 51 evenly distributed calculation points, and the fixed altitudes are 2 km, 2 km, 260 km, and 2 km, respectively.

Subsequently, the Delaunay triangulation technique is used in the spherical coordinate system to discretize the tesseroid into a series of subcells. Due to the difference between the latitude and longitude interval of the tesseroid and the proportional coefficient of the thickness, the latitude and longitude interval is 1°, but the thickness is 1 km. Therefore, the interval of tesseroids can be scaled by referring to the small circle radius or arc length of the latitude in which the tesseroid is located. This is to avoid the distortion of the generated tetrahedron in the spherical coordinate system due to the direct use of latitude and longitude information for Delaunay triangulation, i.e., to ensure that the odd intensity is less than 1.4.

To determine whether a tesseroid must be divided, Uieda et al. [5] introduced an inequality as follows

$$\frac{d}{L_i}\ge D$$

(20)

where, the distance d between the observation point and the geometric center of the tesseroid, and $L_i,i\in \left(r,\varphi ,\lambda \right)$, is the dimensions of the tesseroid (the “side lengths” of Li et al. [89]) along longitude, latitude, and radius, respectively. D denotes the “distance-size ratio”, which is a positive scalar [5].

If the inequality is satisfied for all pairs of observation points and tesseroids, then the tesseroid is undivided. Hence, the distance-size ratio dictates the proximity at which the computation point can be located before necessitating the division of the tesseroid. The value of D indirectly influences the accuracy and computation time of the solution. The d is too short, resulting in ambiguity points in the forward modeling results of gravity fields using the tesseroid, especially at the coincidence of the observation point and the source projection, as shown in Figure 10. Uieda et al. [5] noted that the calculation accuracy of the tesseroid-based forwarding algorithm of the gravitational field is influenced by the location of tesseroids, the height of the observation point, and the distance-size ratio.

To verify the computational accuracy of the tetrahedron-based forwarding algorithm in the spherical coordinate system proposed in this paper, the tesseroid is refined by setting $D=80$, yielding 28,130,947 subcells, then the results of the tetrahedron-based forwarding algorithm, which is obtained using Equation (2), and the tesseroid-based forwarding algorithm as shown in Figure 10 and Figure 11, for Experiment 1 and Experiment 3, respectively.

With a more detailed subdivision of the model, i.e., the corresponding forward modeling, results are smoother as D increases. However, as D increases, more refined tesseroids are involved in the computation, especially when d is small. Therefore, too small d is not common in practical applications. The Tetgen software uses Delaunay triangulation technology to discretize tesseroids into tetrahedrons in the spherical coordinate system. Considering that it is difficult to describe the curved edges/faces of tesseroids by PLC/PSLG [80,90,91], it is also easy to introduce significant geometric errors. The geometric results were refined several times by Tetgen software (see Figure 12).

Figure 13 shows that when the distances d are minimal, even after seven refinements using Tetgen, yielding 423,853 subcells, the residuals between the results of the tetrahedron-based and tesseroid-based forwarding algorithms are still significant. The minimal distances d are also one of the reasons for the large relative error values in Figure 4 and Figure 5 of Uieda et al. [5]. Therefore, small distances d should be avoided for the forwards and inversions of gravitational fields.

Figure 13, Figure 14, Figure 15 and Figure 16 reveal that as the tetrahedrons obtained by Tetgen are further refined, residuals between the results of the tetrahedron-based and tesseroid-based forwarding algorithms gradually approach 0.

By analyzing Figure 10, Figure 11, Figure 12, Figure 13, Figure 14, Figure 15 and Figure 16 simultaneously, when the number of subcells required to refine a cell is small, the tetrahedron-based forwarding results have better accuracy than the tesseroid-based results. These tests indicate that the tetrahedron-based forwarding algorithm proposed in this paper is correct. Therefore, the tetrahedron-based forwarding algorithm can be used as a reference algorithm for analyzing the DGGS-based forwarding algorithm.

5.3. Phase 3: To Verify the DGGS-Based Forwarding Algorithm

Phase 3 is to validate the correctness of the DGGS-based forwarding algorithm proposed in this paper. In the first step, the corresponding DGGS grid is generated using Dggrid version V7.0 [92]. For simplicity, the DGGS grid with a resolution of seven and a serial number of 2136 is selected. The top and bottom plates are set to be 40 km and 60 km, respectively, to obtain the corresponding DGGS cell, as shown in Figure 2, in the spherical coordinate system.

In the second step, the DGGS cell is discretized into tetrahedrons using Tetgen software, as shown in Figure 17. In the third step, taking Equation (2) as the gravitational field forward integral kernel for both tetrahedron-based and DGGS-based forwarding algorithms. Then, the result of the DGGS-based forwarding algorithm was calculated by Equation (17) with Equation (16) as the shape function, $n_{xy}=7$ (see Figure 5b) and $n_r=3$ (see Table 1).

Referring to Figure 4 and Figure 5 in the paper of Uieda et al. [5], when the observation altitude is 10 km, the accuracy of the tesseroid-based forwarding algorithm is more than 10% when $D\le 3$. When the observation altitude is 260 km, the accuracy of the tesseroid-based forwarding algorithm is about 0.1% when $D\le 2$. For this reason, the two remaining tests to validate the DGGS-based and tetrahedron-based forwarding algorithms assume that the observation point is located at 260 km and 10 km above the surface for Earth and Luna satellites, respectively, referring to the paper by Uieda et al. [5] for the forward parameter settings.

As shown in Figure 18, Figure 19, Figure 20 and Figure 21 by increasing the observed height and indirectly increasing the distance–size ratio, the residuals between the results of the DGGS-based (see Figure 18) and tetrahedron-based forwarding algorithms are significantly reduced, with a difference of about $10^{-3}$ orders of magnitude.

Referring to Figure 4 and Figure 5 in the paper of Uieda et al. [5], the results in Figure 19 and Figure 21 show that under the same parameter conditions, the computational accuracy of the DGGS-based forwarding algorithm via the isoparametric transformation is equivalent to that of the tesseroid-based forwarding algorithm via the GLQ.

5.4. Additional Test: To Verify the DGGS-Based Forwarding Algorithm Via the Tesseroid-Based Forwarding Algorithm with Tiny Tesseroids

To enhance the credibility of the DGGS-based forwarding algorithm proposed in this paper, by referring to the adaptive tesseroid-based forwarding algorithm proposed in Uieda et al. [5], the 3D space enclosed by the maximum and minimum values of the longitude, latitude, and radial directions of the DGGS cell is divided equally into tiny same-sized tesseroids 8000 = $n_x$ × $n_y$ × $n_z$= 20 × 20 × 20 along the longitude, latitude, and radial directions, respectively. Those tiny tesseroids whose center coordinates are outside the DGGS cell are discarded, while those whose tesseroid center coordinates are inside the DGGS cell are retained. The tesseroid-based forwarding algorithm is used to compute the gravity vector and gravity gradient tensor values of these retained tiny tesseroids. Figure 22 shows the residuals between this experiment and the result of DGGS-based forward modeling (see Figure 18), which are about ∼10⁻² and comparable to the accuracy of Figure 19.

When analyzing the lunar DGGS cell, the residual between the results of the DGGS-based and tesseroid-based forwarding algorithms using tiny tesseroids obtained by setting $n_{xyz}$ ≡ $n_x$ ≡ $n_y$ ≡ $n_z$ = 25, 50, and 100 (see Figure 22), respectively. Results similar to Figure 23 can be obtained, indicating the correctness of the DGGS-based forwarding algorithm. However, the residuals do not decrease with increasing $n_{xyz}$. Compared with Figure 20, it is inferred that this may be caused by the intrinsic influence of the observation source distances on the tesseroid-based forwarding algorithm. Another indication that the tesseroid-based forwarding algorithm cannot satisfy the computational requirements for too-small observation heights.

6. Conclusions and Recommendations

The curvature of the geophysical model cannot be approximated as a plane for regional and even global-scale problems. However, if the spherical curvature is taken into account, the prism cell cannot effectively fit the curvature; otherwise, it could reduce the accuracy of the forward computation. The tesseroid can effectively fit the curvature. However, the tesseroid has no gravitational analytical solution, which must be approximated by numerical methods. The greater the number of integration points, the higher the accuracy, but the lower the calculation efficiency.

Although tesseroids do not have gravitational analytical solutions, their integration limits are well-defined; DGGS cells have neither gravitational analytical solutions nor fixed integration limits. The isoparametric transformation converts the non-regular DGGS cell in the system coordinate system to the regular hexagonal prism in the local coordinate system. Thus, the DGGS-based forwarding algorithm is realized in the spherical coordinate system. Similar to the tesseroid-based algorithm, the integral of the proposed algorithm is implemented in the spherical coordinate system using the spherical gravity integral kernel. Therefore, the DGGS-based forwarding algorithm can also accurately fit the curvature.

Due to the difficulty of directly describing the curved surfaces of DGGS cells using facets of polyhedrons, if polyhedrons (i.e., tetrahedrons) are directly used to fit DGGS spherical shell cells, a large number of cells and nodes are required [21,28]. Furthermore, the gravitational field integral kernels are not the same under different coordinate systems [23,66]. Therefore, referring to the paper of Grombein et al. [23] on the forward integral kernel of the tesseroid in the Cartesian coordinate system given, the tetrahedron-based forwarding algorithm under different coordinate systems is used as a “bridge” to compare and analyze the difference in the calculation accuracy of tetrahedrons, rectangular elements, tesseroids, and DGGS cells. The experimental results show that the DGGS-based forwarding algorithm under the spherical coordinate system proposed in this chapter is correct. As shown in Figure 22 and Figure 23, the same accuracy as the tetrahedron-based forwarding algorithm is achieved when fitting the DGGS cell using tiny tesseroids.

With a more detailed subdivision of the model, i.e., the corresponding forward modeling results are smoother as D increases, which results in more refined tesseroids being involved in the computation, especially when d is small. Therefore, too small d or too large D are not common in practical applications.

By increasing the observation height and indirectly increasing the distance-size ratio D, the residuals between the results of the DGGS-based and the tesseroid-based forwarding algorithms are significantly reduced. Under the same parameter conditions, the computational accuracy of the DGGS-based forwarding algorithm is equivalent to that of the tesseroid-based forwarding algorithm.

The DGGS system is based on icosahedron, there are 12 pentagons for the whole globe. For the research area not greater than ${36}^{\circ }\times {36}^{\circ }$, existing grid data under the software DGGRID (version 7.0) can be directly used by rotating it to cover the measurement area; otherwise, the parameters must be manually adjusted to obtain the corresponding DGGS grid.