Uncertinity of ground motion attenuation relation for predicting earthquake force

Abstract

Several Ground Motion Prediction Model (GMM) were developed in Himalayan region in the past by several researchers. In the present studya comprehensive review the existing GMMs for Himalayan region were carried out. Here an attempt has been madeto identify and collect the seismo-tectonic setting of the region and source of seismic data.Further, applicability of these generated GMMs were inspected thoroughly using available seismic records of Indo Himalayan region. For this purpose, predicted peak ground acceleration (PGA) and Peak Spectral Acceleration (PSA) obtained by using the GMMs for different magnitude and distance are compared with recorded data by computing Root Mean Square Error (RMSE) and Chi-square value. Moreover, Log-likelihood (LLH) scoring method has been applied and proper weights of the GMMs have been proposed.RMSE is employed to quantify the model's goodness of fit, LLH measures the model's ability to capture the distribution of observed data, and chi-square statistics assess the agreement between observed and predicted values. Further, GMMs have been developedto predict the PGAwith the help of different machine learning technique like ANN, ANFIS, FUZZY, GA and RANDOM FOREST. Here also above mentioned statistical analysis were performed in order to analyse their performance. The proposed machine learning-based ground attenuation relation has the potential to significantly improve seismic hazard assessments and engineering design, leading to safer and more resilient structures in earthquake-prone regions.The incorporation of these statistical measures allows us to rigorously assess the appropriateness of our model.