Artificial Neural Network: A Powerful Tool for Predicting Gravity Anomaly from Sparse Data

Abstract

Generally, the gravity surveys are developed along roads or other communication ways. This leads to an irregular space distribution and lacking data in large areas, like that containing high mountains, wetlands, lakes and forests. The usual methods for geoid computation from gravity data need a regular grid of gravity anomalies. Numerous methods have been developed for gravity anomalies interpolation at regular distribution. This paper reports on implementation of an interpolation method by using techniques for learning and training of Artificial Neural Networks (ANN) in predicting both free-air and Bouguer gravity anomalies from irregular and sparse data. The method was applied for a region in the Ecuador (5°S - 1°N and 75°W – 81°W) that has strong variations in crustal density and morphology. The free-air gravity anomalies prediction results were compared with the method of Kriging interpolation. The ANN method presented better results in predicting gravity anomalies in the considered region.