2005 Bulletin of the Seismological Society of America, Vol. 93, No. 5, pp. 2005–2016, October 2003 Reliability of Envelope Inversion for the High-Frequency Radiation Source Process Using Strong Motion Data: Example of the 1995 Hyogoken Nanbu Earthquake by Jorge Aguirre and Kojiro Irikura Abstract This article presents an application of a procedure to invert the high- frequency radiation process at the source during the 1995 Hyogo-ken Nanbu earth- quake using the envelopes of acceleration waveforms from 16 stations. The inversion uses genetic algorithms that compare observed ground motions with synthetic ones calculated using empirical Green’s functions. Before the inversion, the reliability of the solutions for models with different grid sizes is checked. It is found that the resolution of the high-frequency radiation is strongly dependent on the number of reliable data and, for this case, it is shown that a coarse grid model with 60 parameters provides reliable results. With use of this model, the inversion of the high-frequency radiation distribution for the 1995 Hyogo-ken Nanbu earthquake was performed, and results showed four distinct zones of high-frequency radiation. The first zone is located in subfault 1 near a step-over from subfault 1 to the Nojima fault and the starting point of the rupture. The second zone is located in subfault 2 near the lower limit of the slip inverted from low-frequency data by many authors. The third and fourth zones are located in subfault 3, one of them coinciding with the bifurcation of the fault plane into two fault planes (Sekiguchi et al., 2000). It is concluded that the procedure applied in this article successfully inverts the high-frequency radiation distribution for the 1995 Hyogo-ken Nanbu earthquake. Resolution is restricted to the coarse grid model with 60 parameters because of the limited number of stations, according to the resolution test. The results of the reso- lution test done in this work can not be generalized; however, they show that it is very important to check the resolution before doing this kind of inversion. Introduction The source-rupture process is one of the most complex and unknown factors in the characteristics of an earthquake event. Some knowledge has been obtained through the in- version of data in the low-frequency ranges usually less than 1 Hz (velocity seismograms, accelerograms, Global Posi- tioning System measurements, and interferometer images among others). On the basis of laboratory tests some authors have found friction laws (Dietrich, 1979; Rice and Ruina, 1983) that attempt to explain relations between kinematic and dynamic parameters. The conditions for stress and strength can be measured and controlled in the laboratory, but in actual cases the stress distribution before and after an earthquake and the spatial distribution of strength properties in the fault plane are unknown. The strength of the material for each part of the fault is also unknown. Those properties are impossible to measure directly with the existing tech- nology. The dynamics of the source process has been studied using the results of kinematic inversions combined with crack modeling to obtain the stress drop, strength-excess dis- tributions, rise-time distribution, and dimensionless stress ra- tio S (Fukuyama and Mikumo, 1993). The kinematic models are based on the inversion of waveforms with low-frequency contents (1 Hz). Because the dynamic process is linked to high-frequency motions more than low-frequency motions, it is desirable to obtain the dynamic description of the source process independently of the results estimated from low- frequency motions. Because the high-frequency contents of the acceleration waveforms are highly oscillating and inco- herent, inversion by fitting synthetic waveforms to observed data is not possible. Instead, several authors have used the envelopes of the accelerograms to study the high-frequency radiation from the source (Gusev and Pavlov, 1991; Cocco and Boatwright, 1993; Zeng et al., 1993; Kakehi and Irikura,