Fast space-domain evaluation of geodetic surface integrals

Geodetic surface integrals play an important role in the numerical solution of geodetic boundary-value problems. In many cases they can be evaluated using fast methods in the frequency domain (FFT). However, this is not possible in general, because the domain of integration may be non-trivial (as is the surface of the Earth), the kernel function may not be of convolution type, or the data distribution may be heterogeneous. Therefore, fast evaluation strategies are also required in the space domain. They are more difficult to design because only one property is left where a more or less fast evaluation strategy can be built upon: the potential type of the kernel function. Consequently, the idea is not to replace well-established frequency domain techniques, but to supplement them. Our approach to this problem goes in two directions: (1) we use advanced cubature methods where the integration nodes automatically densify in the vicinity of the evaluation points; (2) we use powerful computer hardware, namely MIMD computers with distributed memory. This enables us to evaluate geodetic surface integrals of any practical complexity in reasonable time and accuracy. This is shown in a numerical example. Received: 7 May 1996 / Accepted:17 March 1997     

On the accurate numerical evaluation of geodetic convolution integrals

In the numerical evaluation of geodetic convolution integrals, whether by quadrature or discrete/fast Fourier transform (D/FFT) techniques, the integration kernel is sometimes computed at the centre of the discretised grid cells. For singular kernels—a common case in physical geodesy—this approximation produces significant errors near the computation point, where the kernel changes rapidly across the cell. Rigorously, mean kernels across each whole cell are required. We present one numerical and one analytical method capable of providing estimates of mean kernels for convolution integrals. The numerical method is based on Gauss-Legendre quadrature (GLQ) as efficient integration technique. The analytical approach is based on kernel weighting factors, computed in planar approximation close to the computation point, and used to convert non-planar kernels from point to mean representation. A numerical study exemplifies the benefits of using mean kernels in Stokes’s integral. The method is validated using closed-loop tests based on the EGM2008 global gravity model, revealing that using mean kernels instead of point kernels reduces numerical integration errors by a factor of ~5 (at a grid-resolution of 10 arc min). Analytical mean kernel solutions are then derived for 14 other commonly used geodetic convolution integrals: Hotine, Eötvös, Green-Molodensky, tidal displacement, ocean tide loading, deflection-geoid, Vening-Meinesz, inverse Vening-Meinesz, inverse Stokes, inverse Hotine, terrain correction, primary indirect effect, Molodensky’s G1 term and the Poisson integral. We recommend that mean kernels be used to accurately evaluate geodetic convolution integrals, and the two methods presented here are effective and easy to implement.     

On the inversion of geodetic integrals defined over the sphere using 1-D FFT

An iterative method is presented which performs inversion of integrals defined over the sphere. The method is based on one-dimensional fast Fourier transform (1-D FFT) inversion and is implemented with the projected Landweber technique, which is used to solve constrained least-squares problems reducing the associated 1-D cyclic-convolution error. The results obtained are as precise as the direct matrix inversion approach, but with better computational efficiency. A case study uses the inversion of Hotine’s integral to obtain gravity disturbances from geoid undulations. Numerical convergence is also analyzed and comparisons with respect to the direct matrix inversion method using conjugate gradient (CG) iteration are presented. Like the CG method, the number of iterations needed to get the optimum (i.e., small) error decreases as the measurement noise increases. Nevertheless, for discrete data given over a whole parallel band, the method can be applied directly without implementing the projected Landweber method, since no cyclic convolution error exists.     

大地测量反演问题统一综述

大地测量反演在解决人类面临的资源短缺、自然灾害和生态环境退化问题显示出越来越重要的作用和地位。本文根据R.Parker的理论,首次尝试将近三十年来大地测量反演问题从解的存在性、模型构制、解的非唯一性、解的评价四个方面进行统一,促进了大地测量反演理论和应用的研究。     

The free versus fixed geodetic boundary value problem for different combinations of geodetic observables

Various formulations of the geodetic fixed and free boundary value problem are presented, depending upon the type of boundary data. For the free problem, boundary data of type astronomical latitude, astronomical longitude and a pair of the triplet potential, zero and first-order vertical gradient of gravity are presupposed. For the fixed problem, either the potential or gravity or the vertical gradient of gravity is assumed to be given on the boundary. The potential and its derivatives on the boundary surface are linearized with respect to a reference potential and a reference surface by Taylor expansion. The Eulerian and Lagrangean concepts of a perturbation theory of the nonlinear geodetic boundary value problem are reviewed. Finally the boundary value problems are solved by Hilbert space techniques leading to new generalized Stokes and Hotine functions. Reduced Stokes and Hotine functions are recommended for numerical reasons. For the case of a boundary surface representing the topography a base representation of the solution is achieved by solving an infinite dimensional system of equations. This system of equations is obtained by means of the product-sum-formula for scalar surface spherical harmonics with Wigner 3j-coefficients.     

The free versus fixed geodetic boundary value problem for different combinations of geodetic observables

Various formulations of the geodetic fixed and free boundary value problem are presented, depending upon the type of boundary data. For the free problem, boundary data of type astronomical latitude, astronomical longitude and a pair of the triplet potential, zero and first-order vertical gradient of gravity are presupposed. For the fixed problem, either the potential or gravity or the vertical gradient of gravity is assumed to be given on the boundary. The potential and its derivatives on the boundary surface are linearized with respect to a reference potential and a reference surface by Taylor expansion. The Eulerian and Lagrangean concepts of a perturbation theory of the nonlinear geodetic boundary value problem are reviewed. Finally the boundary value problems are solved by Hilbert space techniques leading to new generalized Stokes and Hotine functions. Reduced Stokes and Hotine functions are recommended for numerical reasons. For the case of a boundary surface representing the topography a base representation of the solution is achieved by solving an infinite dimensional system of equations. This system of equations is obtained by means of the product-sum-formula for scalar surface spherical harmonics with Wigner 3j-coefficients.     

Robust estimation of geodetic datum transformation

The robust estimation of geodetic datum transformation is discussed. The basic principle of robust estimation is introduced. The error influence functions of the robust estimators, together with those of least-squares estimators, are given. Particular attention is given to the robust initial estimates of the transformation parameters, which should have a high breakdown point in order to provide reliable residuals for the following estimation. The median method is applied to solve for robust initial estimates of transformation parameters since it has the highest breakdown point. A smooth weight function is then used to improve the efficiency of the parameter estimates in successive iterative computations. A numerical example is given on a datum transformation between a global positioning system network and the corresponding geodetic network in China. The results show that when the coordinates are contaminated by outliers, the proposed method can still give reasonable results. Received: 25 September 1997 / Accepted: 1 March 1999     

Computation of geodetic direct and indirect problems by computers accumulating increments from geodetic line elements

By choosing sufficiently small elements of the length of the geodetic line, or of the latitude or longitude difference, the other two can be computed at each element and the results can be accumulated to solve the problem with more than twenty significant number accuracy if desired. Ten to twelve number accuracy was computed in the examples of this paper. The geodetic line elements are kept in correct azimuth by Clairaut’s equation for the geodetic line. The computers can do millions of necessary computations very economically in a few seconds. All other published methods solving the direct or indirect problem can be reliably checked against results obtained by this method. The run of geodetic lines around the back side of the Ellipsoid is outlined.     

Strength analysis of geodetic control networks

In strength analysis of horizontal geodetic networks it is appropriate to use pairs of functions which involve the relative position of two points and relative position of three points. Using properly chosen pairs of functions, formulae are given which allow the computation of precision criteria for the orientation and scale of the network as well as its shape. To illustrate the presentation of results, new types of errors ellipses are introduced. Analogies and correlations existing among the adopted functions are introduced by using the concept of orthogonal networks which are defined in the paper.     

On the meaning of geodetic orientation

After deriving models for changes of coordinates and azimuths due to rotations, the investigation considers methods for modeling terrestrial orientation in adjustments of geodetic networks. If a misorientation of a geodetic network exists, this can be due to systematic errors in astronomic longitude or in astronomic azimuth, or in both. A separation of these two effects is not possible in practice. The initial azimuth at the datum origin contributes to the orientation only as much as any other azimuth of the same weight.     

Azimuth-dependent statistics for interpolating geodetic data

Most authors using statistical interpolation techniques on geodetic data have assumed isotropy for the undulation autocorrelation. Tests of actual data,414 deflections of the vertical, indicate this assumption is not valid. The results of interpolation, however, are not very sensitive to the parameters in the covariance function. A special limiting case for which statistical interpolation degenerates into a completely deterministic process is given in the spherical domain. In this case the covariance function has absolutely no effect on the results, so that the covariance of the output of a prediction need not be that assumed for the interpolation. This provides a self-correcting process whereby the information in the data corrects for a poor choice of covariance function. Estimates of the precision of the interpolation, on the other hand, are very sensitive to the covariance function, particularly to the modeling of azimuth dependence. A simple procedure for generalizing isotropic functions to azimuth dependence is given, which provides sufficiently accurate estimates of precision. The advisability of trend removal is illustrated by some numerical examples.