ANALYSIS OF PERIOD AND COMPONENT DEPENDENT SPATIAL CHARACTERISTICS OF STRONG GROUND MOTION DISTRIBUTIONS USING MODE DECOMPOSITION AND MACHINE LEARNING

Abstract

 We analyzed the effect of source parameter settings on spatial characteristics of strong ground motion intensity depending on period and orthogonal directionality. The mode decomposition technique was applied to six hundred cases of absolute acceleration response distributions of fault normal and fault parallel components for various periods using strong ground motions simulated for a strike-slip fault. Modal decomposition results were modeled with source parameter settings using a random forest in machine learning. The dominant factor of spatial characteristics was evaluated based on explainable AI which are interpretation methods for machine learning models. Mode 1 represents the attenuation characteristics. Mode 2 and Mode 3 mostly represent the overall intensity level due to perturbated seismic moments, arrangement of asperities and location of hypocenter. The modal characteristics were found to differ by periods and components.