Ground motion and probabilistic hazard

It might be thought that an empirical ground motion prediction model has only to describe the variations in the input data set as accurately as possible in order to be useful, with the proviso that the data set is reasonably extensive and well-selected. If the model is to be used in probabilistic seismic hazard assessment, however, the model will probably be subject to extrapolation beyond the parameter space within which it was constructed, especially for hazard at low annual probabilities. In this case, features of the model, especially its functional form, may turn out to have unexpected and undesirable implications. The end result can be conclusions about the hazard that are clearly not in accordance with commonsense. In this study, two test cases are used to examine the application of some recent ground motion models to probabilistic hazard studies. Problems are found that suggest that, although a ground motion model may be a correct representation of its data set, the effects of the functional form applied can be such that it becomes doubtful whether the model should be used for probabilistic hazard purposes.