A neural network-based approach for prediction of PGA and significant duration parameters in the Uttarakhand region of India

The state of Uttarakhand has several prime population centers, and it is considered to be among the most seismically active regions of India. The article presents Artificial Neural Network-based prediction models using Multilayer Perceptron technique for the Himalayan earthquakes specifically for the region of Uttarakhand. Feed Forward Back Propagation Levenberg-Marquardt algorithm-based prediction models are developed for assessing the Peak Ground Acceleration (PGA) and Significant Duration (SD) with the availability of independent parameters such as moment magnitude (M-w), focal depth (F), epicentral distance (E), hypocentral distance (H), and site class (SC) considering either rock or soil site. Two PGA models were developed having high correlation (R) of 0.896 and 0.916 respectively whereas the developed SD model showed correlation value of 0.873. The higher accuracy of the models was ensured by objectivity function (OBJ) values of 0.011 and 0.006 for the two PGA models respectively and 3.6 for the SD model. The developed models are compared with available prediction equations, and it is inferred that the models yield higher accuracy in predicting the earthquake parameters for Uttarakhand state of India. However, it should be noted that the models are suitable for magnitudes (M-w) between 3.0 and 7.0 and for hypocentral distance between 9 and 254 km.