Inverse modeling of gravity field data due to finite vertical cylinder using modular neural network and least-squares standard deviation method

In this paper, modular neural network (MNN) inversion has been applied for the parameters approximation of the gravity anomaly causative target. The trained neural network is used for estimating the amplitude coefficient and depths to the top and bottom of a finite vertical cylinder source. The results of the applied neural network method are compared with the results of the least-squares standard deviation method. The inverse modeling has been tested first on synthetic gravity data. The synthetic data are infected with random noise to evaluate the effect of noise on performance of the methods. Both methods show satisfactory results, with and without random noise. The MNN and least squares standard deviation approaches have been applied to two real gravity data due to two salt domes from Iran and USA, where the results comparison shows good agreement with each other. The computed standard errors indicate the generated gravity response of the estimated parameters from MNN has better conformity with the observed gravity anomaly than the generated gravity response from the least squares method. The results of the MNN inversion show the top and bottom depths of the salt dome situated in Iran are about 24.5 m and 63.8 m and for the salt dome situated in USA are about 1451 m and 9263 m, respectively.