The effect of dependence of observations on hazard validation studies

In countries with an advanced seismic technical culture, where best-practice probabilistic hazard studies are available along with dense seismic networks, there is an increasing interest in validation of hazard maps. This, basically, means trying to quantitatively understand whether probabilities estimated via hazard analysis are consistent with observed frequencies of exceedance of ground motion intensity thresholds. Because the exceedance events of interest are typically rare with respect to the time span covered by data from seismic monitoring networks, a common approach underlying these studies is to pool observations from different sites. The main reason for this is to collect a number of data large enough to convincingly perform a statistical analysis. However, this is often done neglecting the intrinsic stochastic dependence affecting observations at different sites in the same earthquake. On these premises, the presented study demonstrates how this may lead to potentially fallacious conclusions about inadequateness of probabilistic seismic hazard assessment. The study refers, as an example, to an ideal seismic source zone and some recording sites. It is shown, how accounting for the dependence of intensity on magnitude and source to site distance, may change the results of validation from fail to pass. Some considerations with respect to other studies, attempting to validate Italian data via thirty years of seismic observations all across the country, are also made. (authors)