A new method for ranking and weighting of earthquake ground-motion prediction models

Abstract

Ground-motion prediction equations (GMPEs), which are known as a key component of any seismic hazard analysis, serve as an appropriate tool for estimating the values of ground motion parameters for future earthquake. Toward this goal, candidate ground motion models should be selected in an appropriate way to capture the expected values in the target region. This paper presents a novel, efficient approach for ranking of ground motion prediction equations based on artificial neural network (ANN). The nonlinear nature of ANN is also working as an efficient-robust system for weighting of different GMPEs which could be used in logic tree branch of seismic hazard analysis. An effective type of radial-basis neural network named generalized regression neural networks (GRNN) as a one-pass learning algorithm was chosen in this study. The proposed approach has been tested based on the results achieved using two goodness of fit indicators, Nash-Sutcliffe efficiency coefficient and median LH value which confirms high potential of designed GRNN for ranking of ground motion prediction equations.