Comparison of artificial neural network and kriging methods in local geoid modelling

In recent years, GPS (Global Positioning System) is widely used in geodetic measurements by the development of satellite positioning technologies. In engineering applications, vertical coordinate values of GPS measurements must be transformed while the horizontal coordinates values can be easily used. Several suggestions about measure and calculation methods for to evaulate the transformation which will be applied are presented in Large Scale Map and Map Information Production Regulation. To calculation of geoid model by utilizing the GPS measurements in orthometric height known points in a local area and then to provide the solution of the height transformation problem is one of the proposed approach. Geoid height of measurement points need to be known for transformation of ellipsoidal heights derived from GPS measurements to orthometric heights. In this study, 86 geoid height known TUTGA points in Afyonkarahisar and around cities are used. Three appropriate distribution of data are created from these TUTGA points over an area of 360x340 km2. There are different number of reference and test points in each data group. The reference points are taken as a basis in these data groups. The geoid heights of the test points are calculated with Kriging method which is widely used in engineering field and a succesfully interpolation method. And then same test points are calculated with Artificial Neural Network (ANN). Root Mean Square Error (RMSE) values derived from both method are compared and the availability of ANN in determination of local geoid surface model is investigated.