A modeling strategy for ground motion prediction equations based on relative importance analysis

Existing studies typically present the modeling strategies of ground motion models directly, and there remains a lack of methods for determining such strategies in a quantitative manner. In this study, we developed a modeling strategy for ground motion prediction equations based on model relative importance analysis. The method first decomposes the contributions of the different components within the ground motion prediction equation to the model's interpretability and then determines the modeling strategy for the prediction equation based on the contributions in ascending order. Strong motion data from the NGA-West2 project in the western United States were analyzed as a case study, to which the new method was applied to determine the peak ground acceleration prediction equation for the rock sites. The results indicate that the strategy developed herein did not exhibit systematic bias in fitting, thereby demonstrating its validity and rationality. The ground motion prediction equations determined using the modeling strategy also accurately reflected the strong motion characteristics.