Gm. Atkinson et Pg. Somerville, CALIBRATION OF TIME HISTORY SIMULATION METHODS, Bulletin of the Seismological Society of America, 84(2), 1994, pp. 400-414
Ground-motion time histories for use in engineering analyses of struct
ures in eastern North America are often simulated from seismological m
odels, owing to the paucity of real recordings in the magnitude and di
stance ranges of interest. Two simulation methods have been widely use
d in recent years: the stochastic method and the ray-theory method. In
the stochastic method, as implemented in this study, ground motion is
treated as filtered Gaussian noise whose underlying spectrum is deter
mined from an empirical region-specific seismological model of the sou
rce and propagation processes. In the ray-theory method, as implemente
d in this study, the ground motions are simulated by convolving an emp
irical source function with theoretical Green's functions for a specif
ied crustal structure model. This article compares results of the two
simulation methods for four well-recorded ''calibration'' events and a
ssesses the applicability of the methods. The assessment is based on c
omparisons of ground-motion parameters from the simulated data with th
ose of the actual recordings. Ground-motion parameters in the frequenc
y range from 1 to 10 Hz are satisfactorily predicted by both methods.
Averaged over the four events studied, the stochastic method underpred
icts 1-Hz response spectra by 20 to 40% but accurately predicts respon
se spectra for frequencies of greater than 2 Hz; it also accurately pr
edicts peak ground acceleration and velocity. The wave-propagation met
hod underpredicts 1-Hz response spectra by 10 to 40% but accurately pr
edicts response spectra for higher frequencies; it overpredicts peak g
round acceleration and velocity by 10 to 40%. Both methods are impreci
se: the standard error of an estimate is a factor of about 2.2. The bi
as and standard error of an estimate for the wave-propagation method a
re generally slightly lower than for the stochastic method, if the foc
al depth of the event can be specified (i.e., as for a past earthquake
). If the focal depth of the event is not known (i.e., as for a future
earthquake) then the accuracy and precision of the two methods are ab
out the same. The chief advantage of the wave-propagation method is it
s predictive power; since its attenuation function is derived from the
focal depth and crustal structure it does not require knowledge of th
e empirical attenuation function. The chief advantage of the stochasti
c model is its economy and simplicity.