The utility of artificial neural networks in geodetic point velocity estimation

Turkish National Fundamental GPS Network (TNFGN) has been established in order to cover the current geodetic needs by General Command of Mapping (GCM) in parallel with the technological developments in satellite positioning systems and the rapidly growing in the use of Global Positioning System (GPS) techniques. Due to the geodetic design of TNFGN with time dimension, the GPS measurements performed in different sessions are required to process with the same reference epoch for consideration of the coordinate displacements of geodetic points caused by the active tectonic structure of Turkey. Furthermore, Large Scale Map and Map Information Production Regulation (LSMMIPR) that came into force in 2005 in parallel with the establishment of TNFGN requires obtaining the coordinates of the densification network points to be created within TNFGN according to the specified reference epoch. These transactions require the velocity vectors (VX, VY, VZ) besides the coordinates of TNFGN points. In the present applications, the velocity field of TNFGN is generated by estimating the velocities of TNFGN points from the velocity vectors of other TNFGN points that are determined by two or more GPS sessions. In the densification networks, the velocities of the constructed points are estimated from the velocities of TNFGN points or from higher order densification points by interpolation methods.In this study, the utility of Artificial Neural Networks (ANN) that have been widely applied in diverse fields of science and engineering by various disciplines for estimation, modelling, classification, prediction, nonlinear regression since the last quarter of the passed century, is investigated for the problem of estimating the geodetic points velocities. The geodetic point velocities are estimated with Back Propagation Artificial Neural Network (BPANN) and Radial Basis Function Neural Network (RBFNN) that have been more widely applied among all other ANN applications by using the velocity information that are determined by GCM as fundamental values. In order to evaluate the performance of BPANN and RBFNN, the velocities are also estimated by Kriging (KRIG) interpolation method that is used by GCM in determining the velocity field of TNFGN and the results are compared in terms of the root mean square error (RMSE). 125 TNFGN points that are located in central and western Anatolian parts of Turkey are selected as the study area and the velocities are estimated on five different geodetic networks that were generated to assess the impact of the point density on the results. In the geodetic networks that the reference points are less than the test points, BPANN gave more accurate results than KRIG. RBFNN gave approximately same accuracy results with KRIG in the geodetic network with the least reference points. When the number of the points that will be estimated are smaller than the number of the points that are estimated, the estimation of geodetic point velocity with the use of BPANN is evaluated to be more effective and accurate than using KRIG.