Evaluation of a Few Interpolation Techniques of Gravity Values in the Border Region of Brazil and Argentina

Abstract

Least Squares Collocation (LSC) and kriging are the most used techniques to predict gravity values as well as gravity anomalies. The limitations of LSC technique are mainly related in obtaining an adequate co-variance function. Moreover, LSC and kriging predictions depend strongly on known data distribution. Artificial Neural Network (ANN) is a promising tool to be applied in the interpolation problems. Even though, far from the deterministic ones, these techniques are presented as alternatives for interpolating due their good adaptation to several data distribution and easy implementation for fusion of different kinds of data basis. To test the performance of ANN in face of interpolation problems with respect to LSC and kriging, an experiment was developed in a region in the Brazil–Argentina border. Interpolated gravity values were obtained by LSC and kriging and compared with values obtained by ANN considering different data distributions and by using the same test points where gravity values are known. Considering the need of consistency of datum for predicting gravity related values, only a Brazilian data set was used in the present analysis. The smallest number of reference data for training and the low dispersion reveals the ANN as an alternative for LSC and kriging techniques for the usual poor gravity data distribution in South America.