Contribution of the artificial neural network (ANN) method to the interpolation of the Bouguer gravity anomalies in the region of Lom-Pangar (East-Cameroon)

Abstract

Interpolation methods are frequently used during gravity surveys to improve the coverage of gravity data, particularly in areas where data is scarce and sparse. This is a region which, due to its isolation, has only benefited from a single gravimetric survey campaign, with very little data collected, thus constituting a real obstacle to a geophysical study, even though it has real mining potential (Claude et al. in Adv Remote Sens 10:1, 2021). The ANN (Artificial Neural Network) method is a recent interpolation method applied to gravimetry that imitates the functioning of the biological neural network of the human brain. This paper aims to confirm the effectiveness of the ANN method on the densification of gravity data in the Lom-Pangar region where the distribution of gravity data is particularly weak and irregularly distributed. The Matlab program allowed us to build an ANN architecture with an input layer, a hidden layer, and an output layer. Statistical analysis and regression curves allow us to evaluate the degree of similarity between the Bouguer gravity data obtained in-situ and that calculated by the artificial neural network. In this study, good results were obtained using these statistical parameters: the correlation coefficient (R2 = 0.9811), the root mean square error (RMSE = 0.0804) and the mean bias error (MBE = 0.0003). Even better, these statistical parameters are significantly better compared to those obtained via other classical interpolation methods. The observed Bouguer gravity data and those obtained by the ANN method show relevant similarities, so we obtained a map of Bouguer that showed more pronounced anomalies, with more pronounced shapes and contours; reflecting specific and deeper geological structures compared to those obtained in this region using other methods. The ANN method is therefore appropriate for interpolating gravity data and could be useful in improving the gravity coverage of the Lom-Pangar region and other regions of Cameroon and indeed any region of the world that may experience similar difficulties.

Article highlights

• Designing an artificial neural network adapted to a geophysical study;

• Densify Bouguer gravity anomalies from geographic coordinates;

• Very satisfactory statistical parameters obtained from the neural method, compared to other empirical interpolation methods;

• Real geological features identifiable on the Bouguer anomaly map generated via the neural network method.