Ground-motion prediction by a non-parametric approach

Recently, several new ground-motion prediction equations (GMPEs) have been developed in the U. S. A. (the NGA project) and elsewhere. Unfortunately, the predictions obtained by using different models still differ considerably, although starting from the same database. In this paper, a non-parametric approach, called the Conditional Average Estimator (CAE) method, has been used for ground-motion prediction. The comparison between the CAE results and the predictions obtained by five NGA and one European model suggest that the model predictions depend substantially on the selection of the effective database and on the adopted functional form. Both decisions rely to some extent on judgement, and their influence is especially important at short distances from the source. The differences between the results obtained from the European and NGA databases seem to be of the same or even smaller magnitude than the differences observed between different NGA models, at least at short and moderate distances. Aftershocks in the database generally decrease the median values and increase dispersion. The non-parametric CAE method has proved to be a simple but powerful tool for ground-motion prediction, especially in a research environment. It can be used for quick predictions with different databases and different input parameters within the range of available data. It is easy to add to or remove data from the database, and to check the influence of additional input parameters. With availability of high quality data, the non-parametric approach will become more reliable and more attractive also for practical applications. Copyright (C) 2010 John Wiley & Sons, Ltd.