Seismic attenuation model for data gap regions using recorded and simulated ground motions

In this study, seismic attenuation model is derived using recorded and simulated ground motions covering the data gap region of the Indo-Gangetic Plains (IGP), employing artificial neural networks (ANN) methodology. Four independent variables moment magnitude (Mw), focal depth, epicentral distance (Repi), and average shear wave velocity up to 30 m depth (Vs30) are selected to predict peak ground acceleration (PGA) and pseudo-spectral acceleration (PSA) (5% damping) between periods 0.01 to 4 s (twenty-five periods in total), utilizing 926 recordings (PESMOS, CIGN and synthetic database). A feed-forward ANN with Levenberg–Marquardt training algorithm is employed for training the network of input and output dataset. The optimal network architecture obtained in this study consists of 4–9–26 input, hidden and target nodes, respectively. In spite of the absence of presumed functional dependencies in ANN methodology, our model captured a number of sound physical features of earthquake ground motion: magnitude scaling, attenuation with distance and radiation damping. Further, the performance of the model is measured by the standard deviation of the error, σ(ε), and compared with the four widely used conventional GMPEs applicable for IGP region of India. The standard deviations for our model varied between 0.208 and 0.263 which is less than the classical GMPEs at all twenty-six periods of PSA. Finally, the ANN model performance is compared with recorded ground motions at four stations and conventional GMPEs, and the results affirm that this model is competent to predict the response spectrum with good accuracy for IGP region.