GEOPHYSICAL RESEARCH LETTERS, VOL. 13, NO. 6, PAGES 609-612, JUNE 1986 MOMENT-TENSOR SPECTRA OF THE 19 SEPT 85 AND 21 SEPT 85 MICHOACAN, MEXICO, EARTHQUAKES Mark A. Riedesel Institute for Geophysics, University of Texas, Austin, TX 78751 Thomas H. Jordan and Anne F. Sheehan Department of Earth, Atmospheric and Planetary Sciences, MIT, Cambridge, MA 02139 Paul G. Silver Department of Terrestrial Magnetism, Carnegie Institution, Washington, D.C. 20015 Abstract. IDA recordsof the 19 Sept 85 Michoacan earthquake andits large aftershock of 21 Sept85 havebeen used to estimate source-mechanism spectra •l(co) and total-moment spectraMr(co) in 1-mHz bands over the frequency interval 1-11 mHz. Source mechanisms, obtained by the phase-equalization methodof Riedeseland Jordan, show no significant frequency dependence and no significant non-double-couple components. The bestaverage double- couple solution hasa strike, dip andrakeof 289 ø+_ 4ø, 19 ø+_ 15ø and 76ø + 5ø for the main event, and 293ø +_ 3ø, 24ø + 12 ø and73 ø+_ 4ø for the aftershock. Total-moment spectra, obtained by the power-spectral method of Silver andJordan, are parametrized by a total integrated moment M? and a characteristic time'r•. Forthe main event, we obtaih M• ø= (10.7 + 2.0) x 1[32ø N-m and 'r• = 49 + 7 s; •he corresponding parameters for the afters'hock are M•ø = (2.6 _+ 0.6) x 1020 N-m and 'r e =30_ 11 s. Both 6vents are relatively slow; in particular,the aftershock has a larger characteristic time than either the fast-rupturing 29 Nov 78 Oaxaca or the 14 Mar 79 Petalan earthquakes, although the static moments are about the same. Our results support the model of the main shockas a multiple-rupture event with momentreleaseoccuringas long as 100 s after the initial break. Introduction The Michoacan earthquake of 19 Sept 1985 andits large aftershock of 21 Sept 1985 were well recorded on the International Deployment of Accelerometers (IDA) network of long-period gravimeters. We have employed the data from 13 IDA stations to study the low-frequency source properties of these events. The spectrum of a moment-rate tensor M(co) can be factored into a total scalar moment Mr(co ) anda source mechanism •l(co), a second-rank tensor x0ith unit euclidean norm [Silver andJordan, 1982]: M(co) = x/•MT(co) l•l(co ) (1) We have estimated the source-mechanism spectrum •l(co) by the moment-tensor inversion method of Riedesel and Jordan [1985] and the total-moment spectrum Mr(co ) by the technique of Silver andJordan [1982, 1983]. Twelvestations wereemployed for each event, with eleven common to both. The data were edited to eliminate nonlinearities in the initial body waves, the R 1surface-wave packets, and, in some cases, R 2surface waves. All stations recorded R 3 and later wave groups without anyapparent nonlinear distortion. Five hours of record following the first Copyright 1986 by the American Geophysical Union. Paper number 6L6133. 0094-827 6/8 6/006L-6133 $03 . 00 good time pointwereused in the analysis. Transfer functions relating themoment-rate tensor to acceleration were generated frommodel1066A[Gilbert and Dziewonski, 1975] assuming the centroidlatitudes, longitudes and depths reported by Harvard [G. Ekstr6m, personal communication, 1986]: 17.97øN, 102.07øW, 22km for the main event and 17.61øN, 101.48øW, 22 km for theaftershock. For these large events, our analysis is not sensitive to theexact location used, andthe NEIS parameters yield virtually identical results. Source-Mechanism Spectra The data functionals employedby the moment-tensor inversion method of Riedesel andJordan [ 1985]areintegrals of the complex acceleration spectra over narrow frequency bands (0.1 mHz in this study)centered on the fundamental spheroidal modes.The frequency-domain averaging reduces the sensitivity of the estimates to attenuation and splitting [Gilbert, 1973; Jordan, 1978]. The algorithminvolves a nonlinear phase-equalization procedure to compensate for centroidtime shifts,the effectsof unmodeled earth structure, and station timing errors. In this study, estimates M(co,)are obtained over the frequency interval 1-11 mHz by invfirting the integrals in discrete 1-mHz bands centered on the frequencies co_= (n + 1/2) mHz; n = 1, 2.... 10. This •nterval contrans101 fundamental modes, about ten per millihertz, so that for the Michoacan earthquakes, approximately 120 complex numbers are usedto derive the moment-rate tensor in each 1-mHz band. Riedesel [ 1985] has demonstrated thatthisprocedure yields reliable narrow-band estimates of the source-mechanism spectrum to frequencies as high as 11 mHz and that the errors derived for the estimates adequately model the uncertainties induced by noisein the data, including signal-generated noise. Good measures of estimation error are essential in testing for thefrequency dependence of thesource spectrum and the existence of non-double-couple components, but these have generally not beenavailable in the standard procedures for moment-tensor inversion [Dziewonskiet al., 1981; Sipkin, 1982]. Part of the problem is simply the difficulty in manipulating and displaying the autocovariance of a second-rank tensor. Riedeseland Jordan[ 1982] have devised a graph of the source-mechanism tensor -l•l which facilitates thevisualization of itsuncertainties and theiruse in hypothesis testing, and we employ this diagram heLe. The graph is based on the eigenvector expansion l•l = •, •i f•i i•i' Assuming the eigenvectors are orderedsuch that •. _> )• > •, we construct the unit vector •. = Y.)• e•. A plot o• •., •, g9 and f•3 on the unit focal sphere is isbrriorphic to l•l,and'thu'• completely characterizes the source mechanism. On such a plot the three vectors a = (•] - f•,)/•/•, i = (•]- 1/2 $2- 1/2 • )/3'q•, and • = (f• + f• + ?• •/•- represent the values ..3 1 3 . of•. for apure double-couPle, apure compensated hnear vector dipole,and a puredilatation, respectively. The great 609