Evaluation of the applicability of the time‐ and slip‐predictable earthquake recurrence models to Italian seismicity

The applicability of the time‐ and slip‐predictable earthquake recurrence models to Italian seismicity is tested through a more stringent statistical procedure than previously employed. This procedure has the following main features: (1) the original definition, which applied to ‘megafault’ planar geometries is extended to 3‐D source volumes; (2) seismic regions are defined by two independent regionalization criteria based on (a) neotectonics and (b) the spatial clustering of epicenters; (3) analysis is restricted to the complete part of the catalogue, inferring completeness through statistics; (4) the common simple inspection for alignment in the cumulative (or noncumulative) sequence of coseismic slips is translated into an equivalent regression problem, which is studied through the analysis of variance in two stages for the data being compatible with the models; (5) clustered large events within each region are accounted for by ‘attaching’ to each main event all the following occurrences within a given time window; (6) different magnitude thresholds are used to ensure that each event releases strain in a whole region. The procedure is applied to two different sets of data, which, in order to maximize the number of analysable recurrences, both rely or, the moment‐magnitude scaling law to estimate the coseismic slip, and have therefore an intrinsic accuracy similar to that of the sets used in previous studies. The first set of data considers the complete catalogue for Italian seismicity and, capitalizing on the most extended and reliable historical seismic record in the world, contains several recurrences in each region. The second one, based only on instrumental recording, considers the recent activity of a more detailed subset of three low‐seismicity regions. Each region is analysed separately to avoid bias in the regression and since it does not appear feasible to assume that the tectonic setting and the mechanical properties are identical everywhere. This analysis yields that the time‐ and the slip‐predictable models offer a satisfactory fit to reality, respectively, in just two regions and one region out of the 19 globally analysed. The reasons for this negative result, at odds with the positive evidence which has been reported in the literature for the time‐predictable model, are analysed. The potential causes might be several, but applying the same procedure to the flagships of these models, i.e. Bufe, Harsh & Burford (1977), Shimazaki & Nakata (1980) and Papazachos (1989, 1992), yields similar negative results. The origin of this general incapability (with very few exceptions) of the time‐and slip‐predictable models to represent the data appears therefore to be the stringent validation procedure adopted here. Such models appear thus too crude to be a tool of general practical utility, and can be tentatively used for time‐dependent hazard estimates in just a very few regions. This suggests also that seismogenesis obeys constitutive laws more complex than simple linear elastic laws. Copyright © 1995, Wiley Blackwell. All rights reserved