APPLICATION OF THE THEORY OF REGIONALIZED VARIABLES TO EARTHQUAKE PARAMETRIC ESTIMATION AND SIMULATION (CALIFORNIA).

Abstract

Historical accounts of earthquakes show a high degree of spatial variability and uncertainty associated with ground motion. For this reason, historical data are not often used as input for earthquake hazard assessment. Regionally, however, earthquake ground motion is related by the concept of attenuation. Seismic hazard assessment techniques usually rely on catalogues of earthquake epicenters together with empirical attenuation relationships to define the seismic hazard for a particular region. Such techniques, however, overlook local variations in ground motion associated with actual earthquakes. A technique for seismic hazard assessment that includes historical data using the theory of regionalized variables and linear estimation techniques best represents ground motion dichotomy. Modified Mercalli intensity observations for the period 1930 through 1971 were treated as regionalized variables to define the seismic hazard for a region of Southern California centered around San Fernando. Despite variations in construction quality and individual sensitivity to ground motion, intensity values associated with seventy percent of the earthquakes that occurred during this period, for which at least five intensity observations were recorded, were accurately treated as regionalized variables. A Gumbel analysis computed using spatially regular data sets developed from these intensity values precisely associated high hazard regions with active faults near San Fernando. Other earthquake ground motion data can also be treated, accurately, as regionalized variables. These data include peak instrument recordings of spectral acceleration, velocity, and displacement. Moreover, response to earthquake ground motion at discrete frequencies, as recorded by response spectra, is also regionalized. These data, therefore, are accurately estimated using kriging. Fundamentally, because earthquake ground motion is shown to be a regionalized variable, all aspects of regionalized variables theory are applicable for these data, including disjunctive kriging, conditional simulation, and co-kriging.