Bulletin of the SeismologicalSociety of America, Vol. 87, No. 4, pp. 799-808, August 1997 Evaluation of the Global Applicability of the Regional Time- and Magnitude-Predictable Seismicity Model by Constantinos B. Papazachos and Eleftheria E. Papadimitriou Abstract Two main issues are examined that concern the global applicability of two models that associate earthquake recurrence intervals and the magnitude of the preceding or following event of each interval, namely, the regional time- and mag- nitude-predictable models. Specifically, the statistical significance of the results that are obtained by application of these models to a high-quality data set is tested, and then the effect of the seismic zonation process on these results is examined. A simple application of the time-predictable model to different seismogenic regions using data from a high-quality global data set leads to results that satisfy the 95% confidence limit for different obtained parameters in only 25% of the cases. Using simple Monte Carlo simulation, it is shown that these statistical significance tests often fail simply because of the errors in the magnitude determination and the small magnitude range spanned by the available data. Moreover, these tests examine each seismogenic re- gion separately and ignore the global applicability of such a model. For this reason, a procedure that incorporates the whole global data set is applied. The results and the detailed statistical analysis demonstrate the consistency of the behavior of this recurrence model, as this was established from earlier regional studies. Application of the model to an independent data set shows that the results are robust when different seismic zonation techniques are used. On the contrary, the slip-predictable model is rejected using the same procedure. These results suggest that the time- and magnitude-predictable models can generally be used for practical purposes and haz- ard estimates in active seismogenic regions, provided that an appropriate data sample, as defined in the present article, is available for each region. In~oducfion Recurrence time-distribution models are commonlyused to estimate the hazard due to large earthquakes. The most well studied and widely discussed such models are the time- and slip-predictable models (Shimazaki and Nakata, 1980). Many relevant studies have been performed, dealing mainly with large earthquakes in simple plate boundaries or in single faults, indicating that repeat times are not randomly distrib- uted but follow the time-predictable model (Bufe et al., 1977; Sykes and Quittmeyer, 1981; among others). According to this model, the time between two large earthquakes is pro- portional to the displacement of the preceding large event. Laboratory experiments on stick-slip behavior on pre-exist- ing faults also favor this model (Sykes, 1983). There are, however, some other studies that do not favor such kinds of models (Davis et al., 1989; Kagan and Jackson, 1991). In most parts of the world, the historical record of seis- micity is too short to enable firm conclusions to be made about the repeat time of large earthquakes or possible clus- tering of large events both in space and time. Thus, due to the lack of data for large earthquakes on any individual fault or segment, a generic distribution is estimated by combining data from a number of faults (Jacob, 1984; Nishenko and Buland, 1987), or a more elaborate stochastic model is as- sumed (Kiremidjan and Anagnos, 1984). More recently (Papazachos, 1989, 1992; Papazachos and Papaioannou, 1993), the regional time- and magni- tude-predictable model has been proposed. This model dif- fers from the time-predictable model proposed by Shimazaki and Nakata (1980) in that it holds for seismogenic regions (or sources) that include the main fault, where the largest earthquakes occur, as well as other smaller faults, where smaller earthquakes occur. It is expressed by two relations that give the repeat time, Tt, and the surface-wave magni- tude, Mf, of the next expected mainshock in that source, as a function of the minimum magnitude considered in the data set, M~n; the magnitude of the last mainshock, Mp, which is larger than Mn~n; and the annual moment rate released in that source, m 0. These two relations have the form: 799