Performance Of Artificial Neural Networks On Different Point Density In Local Geoid Determination

Geoid is a reference surface for physical orthometric heights. Thus precise geoid determination is essential important in geosciences especially in geodesy. For the geoid determination the geometrical method that evaluates GNSS (Global Navigation Satellite Systems) together with levelling data is mostly used in practice. In order to determine geoid surface, many mathematical surfaces and interpolation techniques are applied in this method. Today the rapidly developing artificial intelligence and machine learning technologies by behaving the human brain produce solutions to problems, which have very complex algorithms. In this study, the artificial neural network from artificial intelligence technologies was examined and also its usability was tested in the geoid determination. For this purpose, a study area that covers approximately 2765 km2 was selected and some tests were carried out in this area by using 326 GNSS-levelling points. These points were divided into training and test datasets in order to create various combinations. In this context, some artificial neural network models and polynomial curve surface models were yielded and comparison results were produced. According to numerical results, it was observed that models of artificial neural networks produced better results than the polynomial curve surface models in the homogenous and non-homogeneous point distributions from viewpoint of “Rules of Large Scale Map and Map Data Production”