Observations ofteleseismic P Wave Coda for Underground Explosions

Transcription

1 PAGEOPH, Vol. 128, Nos. 1/2 (1988) /88/ / Brkh/user Verlag, Basel Observatons oftelesesmc P Wave Coda for Underground Explosons CHRISTOPHER S. LYNNES 1 and THORNE LAY l Abstract--The early P wave coda (5-15 sec after the frst arrval) of underground explosons at the Nevada Test Ste s studed n the tme doman usng 2082 telesesmc short-perod recordngs, wth the ntent of dentfyng near-source contrbutons to the sgnals n the frequency range Hz. Smaller magntude events tend to have relatvely hgh coda levels n the Hz frequency band for both Yucca Flat and Pahute Mesa explosons. Coda complexty n ths low-frequency passband s negatvely correlated wth bural depth for Pahute Mesa events but s only weakly correlated wth depth for Yucca Flat events. Enhanced exctaton of relatvely long-perod scattered waves for smaller, less deeply bured events s requred to explan ths behavor. Coda complexty n the Hz band s postvely correlated wth magntude and depth for Pahute Mesa events, but has no such dependence for Yucca Flat events. Ths may result from systematc varatons between the spectra of drect sgnals and coda arrvals caused by pp nterference for the largest events, all of whch were detonated at Pahute Mesa. Another possble explanaton s a frequency-dependent propagaton effect on the drect sgnals of the larger events, most of whch were located n the center of the mesa overlyng strong lateral velocty gradents n the crust and upper mantle. Event average complexty vares spatally for both test stes, partcularly n the Hz band, provdng evdence for frequency-dependent focussng or scatterng by near-source structure. Both the drect arrvals and the early coda have strong azmuthal ampltude patterns that are produced by defocussng by mantle heterogenety. The drect arrvals have stronger coherent azmuthal patterns than the early coda for Pahute Mesa events, ndcatng more pronounced deep crustal and shallow mantle defocussng for the drect sgnals. However, for Yucca Flat events the drect arrvals have less coherent azmuthal patterns than the coda, suggestng that a hghly varable component of near-source scatterng preferentally affectng the downgong energy s supermposed on a pattern produced by mantle heterogenety that affects the entre sgnal. Ths complcated behavor of the drect arrvals may be the result of trplcatons and caustcs produced by the complex basement structure known to underle the Yucca Flat test ste. The presence of strong azmuthal patterns n the early coda ndcates that source strength estmates based on early coda are subject to bases smlar to those affectng estmates based on drect arrvals. Key words: Scatterng, coda, underground explosons, yeld estmaton, wave propagaton. Introducton Relable determnaton of the energy released by sesmc sources s a fundamental problem n earthquake and exploson sesmology. For relatvely small energy sources, t s necessary to utlze short-perod sesmc waves to determne the source strengths, Department of Geologcal Scences, Unversty of Mchgan, Ann Arbor, MI 48109, U.S.A.

2 232 Chrstopher S. Lynnes and Thorne Lay PAGEOPH, although t s well-known that the scale lengths of heterogenety wthn the earth are such that propagatonal effects are partcularly pronounced for the short-perod waves of nterest. Thus, there has been a concentrated effort to acheve a better theoretcal and emprcal understandng of short-perod wave propagaton processes n the earth. The prmary objectve n the feld of exploson sesmology s to develop a procedure for sesmc yeld estmaton usng short-perod waves that s both precse (.e., consstent) and accurate. The currently preferred methods of yeld estmaton employ sesmc magntude measurements, whch can be dvded nto two classes: those that rely on a drect phase, such as rng b, (based on the frst cycle of the short-perod P waveform); and those that are based on a scattered phase, such as Lg or P coda. The advantage of drect phases s that they can be treated to a large degree as determnstc sgnals, whch are generally expected to be rather smple for underground explosons. Ths allows quanttatve applcaton of correctons for predctable bases arsng from source functon scalng, free-surface nteracton, attenuaton varatons and propagaton effects due to known large-scale three-dmensonal structure (LAY et al., 1984; LAY, 1985; CORMIER, 1987). However, the earth s heterogeneous to such a degree that there appears to be a lmt to the precson that can be obtaned n yeld estmaton usng drect phases (BACHE, 1982). On the other hand, scattered phases must usually be treated stochastcally, but they do have the attrbute that the scattered arrvals sample many raypaths, presumably averagng out to a large extent the varous heterogenetes sampled (AKI, 1982; CORMIER, 1987). Several authors have examned the relatve performance of magntudes based on telesesmc P coda and drect P as relatve sze estmators, concludng that P coda generally yelds more precse sze estmates (BuLLITT and CORMIER,. 1984; BAUMGARDT, 1985; GUPTA et al., 1985; LAY and WELC, 1987). In addton, approxmately half of the energy n telesesmc P coda from shallow sources appears to result from near-source scatterng (DAINTY, 1985). Thus, the P coda may provde nformaton about the source regon, ether through use as a reference for examnng drect sgnal behavor (GuPTA and BLANDFORD, 1987), or by analyss of the coda tself (LAY and WELC, 1987; LAY, 1987a,b). Whle some emprcal progress has been made n usng P coda for both sze estmates and determnng source propertes, the exact orgns of P coda reman somewhat engmatc. It presumably s generated by scatterng from small-scale heterogenetes and may therefore be useful for characterzng crust and mantle velocty varatons. However, the relatve mportance of processes such as mode conversons, forward scatterng or back scatterng s not well establshed. Improvng our knowledge of the processes nvolved n the generaton of P coda s essental to a basc understandng of shortperod wave propagaton n the earth. In ths paper, we wll examne the propertes of the early P coda (defned here as the frst 5-15 sec after the frst arrval of the P wave) from underground explosons at the Nevada Test Ste. The early coda s partcularly nterestng snce there s some

3 Vol. 128, 1988 Studes of P Wave Coda for Underground Explosons 233 possblty of treatng t determnstcally as well as statstcally, because the nearsource contrbutons must be generated wthn a restrcted volume around the source. Determnstc treatments that attempt to explan partcular arrvals n the coda are presented by BOUCHEI~ (1973), who analyzed the frst 3 sec of near-source recordngs, and by LAY (1987b) who nverted for sotropc scatterers usng the frst 15 sec of the P wavetran. However, nversons of the latter type are stll very dffcult and poorly posed due to our lack of knowledge of the predomnant scatterng mechansms producng the coda; accordngly, n ths study, we wll examne the statstcal propertes of the early P coda emprcally, to motvate and help desgn future nversons for scatterers. These propertes wll also provde an observatonal bass for evaluatng the sze estmaton capabltes of the early coda. The statstcal propertes of early P coda have already been extensvely studed for explosons at the Pahute Mesa test ste, as well as for the two Novaya Zemlya test stes and the Amchtka test ste by LAY and WELC (1987) and LAY (1987a). One of the nterestng results obtaned n these data-ntensve studes was the detecton of a negatve correlaton of coda "complexty" (power n the coda relatve to the drect arrval) wth both magntude and depth of bural n the Hz frequency band for all stes studed. Smlar correlatons were noted for Novaya Zemlya and Amchtka for the Hz, Hz and Hz bands, whle Pahute Mesa explosons exhbt a postve correlaton of complexty wth magntude/depth n the Hz band. Spatal correlatons of coda complexty were also noted for the Pahute Mesa test ste, wth complexty ncreasng toward the center of the test ste. No correlaton wth tectonc release F-factor was found; thus, t appears that aftershocks do not contrbute sgnfcant hgh frequency energy to the coda, as had been suggested by DOUGLAS (1984) on the bass of analyss of the Pahute Mesa exploson GREELEY. Another nterestng result from the study of Pahute Mesa coda was the detecton of systematc azmuthal patterns n the early coda ampltudes that mrror the patterns n the drect arrvals, albet wth a somewhat smaller range of varaton. In a detaled analyss of the drect sgnals, LYNNES and LAY (1988) demonstrated a correlaton between low m~ ~ and fast travel tmes for Pahute Mesa explosons, whch they attrbuted to strong focussng/defocussng effects n the lthosphere and n the deep mantle. Thus, t appears that the near-source component of the early coda s affected by the same deep mantle heterogenety, rather than havng effectvely averaged out all of the propagatonal effects. Smlar correlatons between coda and drect arrval ampltudes have been observed by BULLITT and CORMIER (1984) and BAUMGARDT (1985). In ths study, we have dgtzed 1063 WWSSN and Canadan Sesmc Network sesmograms from 32 Yucca Flat explosons to complement the 1019 sesmograms from 25 Pahute Mesa tests analyzed by LAY and WELC (1987) and LAY (1987a,b). As n these earler studes, we dentfy the near-source contrbuton to the early P coda by usng a large, well-dstrbuted staton set. The dependence of coda

4 234 Chrstopher S. Lynnes and Thorne Lay PAGEOPH, complexty on varous event parameters such as depth and magntude s determned for the two test stes. We also examne the ndvdual event average complexty values n order to look for coherent spatal varatons wthn the test stes. Such varatons provde prme motvatons for future attempts to dentfy partcular scatterng structures. Fnally, the azmuthal patterns of the drect arrvals and coda for the two nearby test stes wll be compared to determne the relatve mportance of near-source and deep mantle effects. Data Analyss The sesmograms for Yucca Flat events were processed n a nearly dentcal manner to those for Pahute Mesa events, as descrbed n LAY and WELC (1987) and LAY (1987a). At least 15 sec of nose and 15 sec of sgnal were dgtzed and nterpolated at a 0.1 sec sample nterval. The sesmograms were fltered usng three-pole Butterworth bandpass flters wth four dfferent passbfnds: Hz ("broadband"), 0.44).8 Hz (low-frequency), Hz (md-frequency), and Hz (hgh-frequency). The ndvdual nstrument responses were deconvolved from the low-frequency, md-frequency and hgh-frequency data sets before flterng; however, deconvolutons of the "broadband" sesmograms were unstable at the long perod end, so the nstrument response was not removed from ths data set. Snce most of the statons are from the WWSSN, ths results only n a slght ncrease of the varance n the broadband results. In order to characterze the complexty of the coda, the normalzed energy temporal centrod (MCLAUGHLIN and ANDERSON, 1985) of the frst 15 sec of the squared analytc envelope (Fgure 1) was computed. The energy temporal centrod (Cbp) s defned for each bandpass fltered (bp) sesmogram by: Cbp = f015 s2(t)t dt/ 115 sz(t) dr, do where s2(t) s the square of the fltered sgnal envelope. Larger centrod values ndcate a hgher proporton of energy later n the sgnal. A correcton for nose level was appled by subtractng the root-mean-square (RMS) of the nose wndow from the envelope before computng Cbp; ths generally reduced all the event average centrods by comparable amounts. Staton/path correctons were determned by averagng the centrods for all events from a gven test ste recorded at a gven staton. When computng an event average value of Cbp, the staton correctons were appled, and the corrected centrods were azmuthally averaged n 7.5 ~ wndows to reduce bas from uneven azmuthal dstrbuton. The event centrods for each test ste were then demeaned, so that the values reported below are devatons from the average. The RMS of the drect arrval (0-5 sec) and the coda (5-15 sec) were also computed for each sesmogram. The logarthm of the rato RMSS-15/RMS ~ provdes

5 Vol. 128, 1988 Studes of P Wave Coda for Underground Explosons 235 TAN ATL Nose 10 sec I drect early arrval 1 coda I ~ A centrod 3.2 s Fgure 1 Example of the data analyss procedure. The top trace s the P wavetran wth ts analytc envelope, fltered from Hz for the Yucca Flat event TAN at staton ATE RMS ampltudes are calculated for the frst 5 sec (drect arrval) and the 5-15 sec wndow (early coda). The bottom trace shows the squared envelope, before the nose correcton (sold lne) and after the nose correcton (dashed lne), and the effect of the nose correcton on the energy temporal centrod. another measure of the complexty of the coda, wth hgher values ndcatng relatvely enhanced coda levels. Event averages for log(rmss-15/rms ~ were calculated n the same manner as for the centrods. The log(rmss-15/rms ~ measure was used to compare wth the behavor of the drect arrval (log(rms~ and early coda (log(rmss-zs)); however the centrods wll be preferred n ths study when consderng event average complexty, snce t proved dffcult to apply a stable nose correcton to the log(rmss-15/rms ~ measurements. The correlaton coeffcents, r, of the event average centrods (wthout nose correcton) vs. log(rms 5-15/RMS ~5) are r = 0.94 n the low-frequency band, r n the md-frequency band, and r = 0.94 n the hgh-frequency band for Yucca Flat events. The broadband correlaton coeffcent s only r = 0.72, whch may be due to the fact that the sesmograms were not deconvolved n ths case. In general, however, these two qute dfferent types of coda complexty measures gave very smlar results, as was the case for the Pahute Mesa date (LAY and WEI.C, 1987; LAY, 1987a). Event Average Complexty Correlaton wth mb and depth. The event average centrods n each frequency band for Yucca Flat were correlated wth magntude, bural depth and depth below the water table. However, the only parameter that gave a correlaton coeffcent above 0.5 n any of the frequency bands was magntude. The correlatons of coda

6 236 Chrstopher S. Lynnes and Thorne Lay PAGEOPH, complexty wth magntude and wth bural depth n each frequency band for both Yucca Flat and Pahute Mesa events are lsted n Table 1. Varance-weghted lnear regressons of coda complexty vs. magntude and depth for both test stes are shown n Fgure 2 for the low-frequency and md-frequency bands. Both test stes have lttle magntude dependence n the broadband and hghfrequency band cases. In the md-frequency band, however, the Pahute Mesa events show a postve correlaton of complexty wth magntude (r = 0.62) and depth (r = 0.71). There s no correlaton wth magntude for Yucca Flat, but a smlar trend wth bural depth, although several outlers lead to a poor correlaton coeffcent (r = 0.27). These correlatons may result from varaton of the spectrum of the drect arrval n ths band due to pp nterference. For larger, more deeply bured events, the ncrease n pp delay tme would shft the spectral scallopng from pp nterference toward the mddle of the passband. Ths effect can be predcted by computng synthetc spectra for a large, deep event and a small, shallow event. Two Pahute Mesa explosons were used for these computatons (Fgure 3), because the telesesmc P waveforms for tests at ths ste have been extensvely studed (LAY, 1985). The spectrum of the coda should not have such pronounced scallopng due to pp (GuPTA and BLANDFORD, 1987) and thus the coda ampltudes n ths passband would appear relatvely enrched. In order to demonstrate ths wth actual data, we compared the spectral ampltudes of a 15 s nose wndow to two 15 s wndows n the P wavetran, one ncludng the drect arrval ( s) and one wthout the drect arrval ( s). Ths method of presentaton (Fgure 4) mtgates the possble effects of nose bas between a large and small exploson. COMMODORE s a large, deeply bured Yucca Flat test, whle PIRANHA s a relatvely small, shallow event, wth an average centrod that s 0.7 s less than that of COMMODORE n the md-frequency band. The spectral null related to pp nterference s clearly vsble n the s wndow, whch s domnated by the drect phase, for both events. As n the synthetc case (Fgure 3) the null s shfted nto the md-frequency band for the deeper test. Ths depletes the drect arrval relatve to the coda, n whch the spectral null s much subdued as seen n the pure coda spectra ( s). Nevertheless, the null does exert a small effect on the spectral ampltudes, suggestng that the coda has some component of P-P scatterng. Table 1 Correlaton coeffcents of Cbp wth magntude and depth of bural Test Ste Parameter Hz Hz Hz Hz Yucca Flat m~ b Pahute Mesa m~ b Yucca Flat depth Pahute Mesa depth

7 Vol. 128, 1988 Studes of P Wave Coda for Underground Explosons 237,E Yucca Flat o.. Pahute Mesa o Yucca Flat 0.6 Pahute Mesa 0 0 I d r-o.06 ~ r= t L -0,6 se( 0.{ 0-0.~ r=0.62 r=-0,94 J r= 0.27 ]~ ro-o.aa L 0,6 /kcbn sec!o r=0.71 r=-o.80 J 4.9 ml~ 5"3b ~50 ' m~ b Depth, m Depth, m Fgure 2 Correlatons of relatve event average centrods (ACbp) wth magntude and bural depth n the lowfrequency and md-frequency bands for Yucca Flat events and Pahute Mesa events. 0 1 BOXCARj~ I I E o -3 t c I -4 ll -" 0 log(frequency) Fgure 3 Synthetc ampltude spectra, calculated usng t* = 0.75 sec and a short-perod WWSSN nstrument for a large, deeply bured event (BOXCAR: yeld = 1300kt, depth= kn) and a smaller, less deeply bured event (HALFBEAK: yeld = 365 kt, depth = km) at Pahute Mesa. The ncrease n pp delay tme for the large event shfts the scallopng effect of pp nterference nto the md-frequency band.

8 238 Chrstopher S. Lynnes and Thorne Lay PAGEOPH, Z Z z o 746 m COMMODORE s ~. / t 10 0 lo-1 1 ] r J Frequency (hz) Z Z + o~ v 102 [ 100 kt 549 m PIRANHA 101 l, * J,,, J, j ' t I t,,, ',, -, J/ s ',"' ' l s I / I I/ 100, t Frequency Fgure 4 Rato of sesmogram spectral ampltude to nose for a large, deep event (COMMODORE) and a small, shallow event (PIRANHA) at Yucca Flat. The dashed lne ndcates the s wndow, whch s domnated by the drect phase, and the s wndow, whch does not nclude the drect phase. (hz) Another possblty s frequency dependent defocussng of the drect arrval for events n the center of the mesa, where most of the larger events were located. Ths second possblty wll be tested later by comparng spatal varatons of complexty and azmuthal patterns n the dfferent frequency bands. In the low-frequency band, both test stes show a negatve correlaton wth magntude (r = for Pahute Mesa and r = for Yucca Flat), wth smaller magntude events havng relatvely hgh coda levels. For Pahute Mesa, the lowfrequency band has a smlar dependence on depth of bural (r = -0.80). For Yucca Flat events, the dependence of complexty on depth s farly weak (r =-0.33), suggestng that the correlaton of complexty wth magntude cannot be ascrbed to depth of bural effects alone, but t s mportant to note that the Pahute Mesa events span a much greater range n bural depth. One possblty for these correlatons s the effect ofpp nterference. However, the parallel spectral curves n the Hz band for both explosons n Fgure 4 ndcate that ths band s relatvely free of such nterference. Furthermore, the effect of nterference should be to decrease the drect arrvals of the larger explosons, as n the md-frequency band, whch would result n a trend opposte of that whch s observed.

9 Vol. 128, 1988 Studes of P Wave Coda for Underground Explosons 239 Another possblty s that the negatve correlaton of complexty wth sze s due to greater nose contamnaton n ths band for smaller explosons. The sgnal/nose rato s ndeed smaller for PIRANHA than for COMMODORE (Fgure 4), but t s stll on average farly hgh wthn ths band for both tests, dppng below 2 only for the coda of PIRANHA. The effect of nose s further reduced by applcaton of the nose correcton to the centrods. Also, although nose wll n general ncrease the centrods, nose that s n phase wth the drect arrval decreases the centrods, further mtgatng the overall effects of nose. These effects are also reduced by the ncluson of low-nose statons, for whch the effects of nose are neglgble. Thus, t seems unlkely that nose contamnaton alone could account for the substantal varaton n centrod wth magntude n the low-frequency band. Spatal varatons ofcornplexty. The spatal varaton of event average complexty n all four passbands for Yucca Flat s shown n Fgure 5. The coherent varatons n hz hz o N 37, L I 1 I - - I I I hz Q2-2.0 hz ~ M +" o +, 37.0 L L "W "W Z~ Cbp, sec Fgure 5 Spatal varaton of event average complexty (ACbp) for the Yucca Flat test ste n the four frequency bands. Damonds ndcate smaller than average coda complexty, and crosses larger than average coda complexty.

10 Chrstopher S. Lynnes and Thorne Lay PAGEOPH, complexty wth locaton are qute strkng. The least coherent varatons occur n the low-frequency band; thus, the complexty n ths band appears to be nfluenced more by magntude than by the source envronment. The most coherent varatons occur n the md-frequency band. Ths result s consstent wth the frequency dependence of spatal varatons for Pahute Mesa, where the md-frequency band had the hghest correlaton wth dstance from the center of the mesa (r = vs. r = 0.24 for the low-frequency band and r = for the hgh-frequency band). The actual spatal patterns are also somewhat dfferent among the dfferent frequency bands: events n the northwest part of the test array have very low complexty n the md-frequency band, but relatvely hgh complexty n the hgh-frequency band; events n the northcentral part, on the other hand, have hgh complexty n the md-frequency band, but average complexty n the hgh-frequency band. The broadband map resembles a subdued verson of the md-frequency band map, due to the predomnance of mdfrequency band energy n the undeconvolved broadband sgnals and the compensatng trends n the dfferent frequency bands. The dfference n patterns among frequency bands suggests a frequency dependence n focussng of the drect arrval or n scatterng contrbutons to the coda. The spatal varatons of coda complexty mply a correlaton wth some property of the source regon. FERGUSON'S (1982) model for Yucca Flat basn, based on hz ~ A Cbp, sec ~ Fgure 6 Depth n meters to Tertary-Paleozoc contact (from FERGUSON, 1982), wth event average complexty for the md-frequency band. l,o

11 Vol. 128, 1988 Studes of P Wave Coda for Underground Explosons 241 borehole data and gravty modelng, s shown n Fgure 6 wth the event average complexty values for the md-frequency band supermposed. The contours ndcate the depth to the Tertary-Paleozoc contact. The events wth the hghest complexty le above the basn axs at the north end, just to the west of the Yucca Fault. The events wth the smplest coda levels are stuated above the east flank of a horst n the basement. Ths suggests that ether the basn shape or the thckness of Tertary tufts controls coda complexty n ths passband. Azmuthal Patterns of Ampltudes and Complexty Slowly varyng azmuthal patterns of m~ b and travel tme resduals have been observed for events at both the Pahute Mesa and Yucca Flat test stes (LYNNES and LAY, 1988). Correlatons between the travel tme and ampltude varatons ndcate that a sgnfcant component s due to focussng and defocussng by velocty heterogenetes n the mantle. MINSTER et al: (1981) and TAYLOR (1983) found localzed hgh velocty regons n the upper mantle to the northeast of Pahute Mesa, whch can produce defocussng that accounts for much of the dfference between the Pahute Mesa and Yucca Flat ampltude patterns (LYNNES and LAY, 1988; CORMIER, 1987). However, the patterns for the two test stes share a large common component, whch LYNNES and LAY (1988) nferred to be due to defocussng and focussng by deep ( > 200 kn) heterogenety. It s thus of nterest to determne the relatve nfluence of such heterogenetes on the drect arrval and the early coda. In order to determne the average azmuthal patterns for the drect arrvals (Iog(RMS~5)), early coda (log(rmss-15)) and complexty (Iog(RMSS-15/RMS~ the staton averages for each ste were calculated usng a least-squares technque that nverts for the best relatve source szes to smultaneously mnmze the varance about the staton means (Larry Ruff, personal communcaton). In order to brng out the slowly varyng component of the azmuthal patterns, curves of the form A 9 sn[2(0-00) ] were regressed through the resultant staton averages as a functon of azmuth (0). The partcular form of the regresson curve s not physcally motvated, but s ntended to parameterze the long-wavelength varatons to facltate comparsons between events and test stes. The F-rato s also calculated as descrbed n LAY et al. (1984) to provde an ndcator of the coherence and sgnfcance of the long-wavelength component. The results for the broadband set are shown for Pahute Mesa and Yucca Flat n Fgure 7. The drect arrvals (log(rms~5)) show azmuthal patterns for both Pahute Mesa and Yucca Flat, wth a somewhat stronger and more coherent pattern for Pahute Mesa. The correlaton n the log(rms ~ staton averages between the test stes s r = 0.83, whle t s r = 0.73 for the log(rms 5-15) values. Thus, a sgnfcant component of the varaton s common to the two test stes. LYNNES and LAY (1988) obtaned a smlar result for mg b and nterpreted the hgher ampltude of varaton for Pahute Mesa as a result of focussng and defocussng by lateral velocty gradents n

12 242 Chrstopher S. Lynnes and Thorne Lay PAGEOPH, Pahute Mesa 0.~ Yucca Flat " o e o 0 9 ~oo o -, ~ -"4 I ~ I 01 0.~ I I~ Ill; I I g I o I I I II ~--F- ~o o ] 0 I L IIt III I I I I -0~ I IIt IIt lltt~ " I I t I I I [ I I I Q8 I I I I I I I I [ I~ o -~ I P I. I I I I 2 1~ I 08! I I I I 1 I! I [ I I 1uu 2fu ' Azmuth, deg Azmuth, deg Fgure 7 Sn 20 regressons for Yucca Flat and Pahute Mesa staton averaged log(rms~ log(rms TM) and Iog(RMSS-lS/RMS ~ n the Hz passband. The sne curves are dashed where not statstcally sgnfcant at the 95% level. the crust and shallow mantle under the mesa. The Iog(RMS ~5) pattern for Pahute Mesa s stronger than the Iog(RMS 5-15) pattern for the same ste, ndcatng that the focussng/defocussng n the Pahute Mesa drect sgnals s less pronounced for the early coda due to the averagng characterstcs of the scattered wavefeld. The log(rm S 5-~5) pattern for Yucca Flat, however, has a stronger long-wavelength component (F-rato = 7.6) wth a slghtly larger range than the pattern for the drect arrvals (F-rato = 3.9) from that ste, although the orentaton of the patterns are qute smlar. Ths behavor s present n many sne curve regressons of log(rms ~ and log(rms 5-15) of the ndvdual events as well. Thus, t seems that both the drect arrvals and the coda are affected, more or less equally, by deep mantle heterogenety, whle the decreased sgnfcance of the drect arrval sne curves appears to be due to ncreased scatter n the assocated staton averages. Ths mght result from very-near-source or near-recever heterogenety. The Yucca Flat events were detonated close to large basement offsets on the Yucca Fault (Fgure 6), whch may cause trplcatons and

13 Vol. 128, 1988 Studes of P Wave Coda for Underground Explosons 243 caustcs affectng the drect arrvals and early coda (STEAD and HELMBERGER, 1986). In any case, the homogenzaton of the early coda seems to have lttle effect on subdung the predomnant ampltude varatons nduced by earth structure at both stes. Sne curves were also regressed through the nverse complexty measure log(rms~ 5-15) for Pahute Mesa and Yucca Flat (Fgure 7). As mght be expected, the Pahute Mesa sne curve s sgnfcant (F-rato ), due to the larger range of varatons for log(rms ~ than for (RMSS-ls). The Yucca Flat pattern, on the other hand, does not have any long-wavelength varaton, ndcatng that the drect arrval and the phases makng up the coda are affected smlarly by deep mantle structure. Drect Arrval Early Coda ~ 37,1 q~ o.~ hz 37.0 h -, = _ 37.2,,, I / ~ ~ / hz F~rato o8 hz ":l " 1 " "' *W "W.0 Fgure 8 Map of Yucca Flat showng the F-ratos and azmuths of the negatve lobes of sn 20 regressons for ndvdual events n the low-frequency, md-frequency and hgh-frequency bands. The frst column shows results for log(rms~ the second shows log(rms5-=5). The length of the arrow s proportonal to the F-rato. Arrows for events wth an F-rato lower than 3 are not shown.

14 244 Chrstopher S. Lynnes and Thorne Lay PAGEOPH, In order to further explore the frequency dependence of the focussng and/or scatterng effects, sne curves were regressed through log(rms~ Iog(RMS 5-~ 5) and log(rms~ 5-~5) measures of ndvdual events from both test stes n all four frequency bands. The F-ratos and the drectons of the negatve lobes of the drect arrval and early coda patterns n the low-frequency, md-frequency and hgh-frequency bands for Yucca Flat are shown n Fgure 8. As wth the broadband results, the regressons of log(rms ~ values have systematcally smaller F-ratos than the Iog(RMS 5-~5) measurements; however, the ampltudes and azmuths are qute smlar for a gven frequency band. In addton, there s a systematc clockwse rotaton n the long-wavelength patterns wth frequency from 2~ ~ (azmuth of the negatve lobe) n the low-frequency band to 11~ ~ n the md-frequency band to 17~ ~ n the hgh-frequency band. The F-ratos and azmuths of ndvdual event patterns n the nverse complexty for all four frequency bands are shown n Fgure 9 for both test stes. There were too few sgnfcant sne curves to deduce any frequency dependence n azmuthal patterns of complexty for Yucca Flat, but a spatal dependence does appear n that those few sne curves that are sgnfcant tend to be for events at the southern end of the test l I hz hz ~, I Pahute Mesa ~- Yucca I Flat t J. I ~ 1 J o~ ~ J I ~ [. o,%_. E 37.4 ~ ~ 1 I I l hz hz [ \~ *. * l I _t I _1 _L_~ *W L,11 l,,, *W Fgure 9 Map of the Nevada Test Ste showng the F-ratos and azmuths of the negatve lobes of sn 20 regressons through log(rms~ TM) (nverse complexty) for ndvdual events n the broadband, lowfrequency, md-frequency and hgh-frequency ranges.

15 Vol. 128, 1988 Studes of P Wave Coda for Underground Explosons 245 array. On the other hand, Pahute Mesa, events clearly exhbt a frequency dependence, as there are more events wth coherent patterns n the md-l'requency band than n any other band. Thus, t appears that t s the sgnal n ths passband that produces the sgnfcant azmuthal patterns for Pahute Mesa complexty observed by LAY and WELC (1987) for events n the mddle of the mesa. Furthermore, almost all of the events wth sgnfcant patterns n ths passband are n the central part of the mesa, supportng the hypothess that frequency dependence of focussng/defocussng 9 or scatterng of the drect arrval plays a role n producng these patterns. Because the large events tend to be located n the center of the mesa, frequency-dependent focussng may also help to explan the postve correlaton of complexty wth magntude n the md-frequency band, rather than smply pp varatons. Dscusson One of the most dffcult observatons to explan from ths paper and LAY (1987a) s the dependence of complexty on magntude. All stes studed so far have exhbted enrched long perod (0.44).8 Hz) energy n the coda, relatve to the drect arrval for smaller events. At Pahute Mesa, ths effect could n fact be a dependence on depth of bural. LAY (1987a) suggests that a negatve correlaton of long perod complexty wth depth could ensue from ncreased Raylegh-to-P scatterng for the shallower events. However, the depth dependence s slghtly weaker than the magntude dependence at Pahute Mesa, and s qute a bt weaker than the magntude dependence at Yucca Flat, suggestng that some varable more drectly related to source sze controls long-perod complexty. Such an effect would seem to be ntrnscally nonlnear, and as yet there does not seem to be an adequate explanaton for ths phenomenon. One possblty s that the cavty sze nfluences the exctaton of near-feld Raylegh waves, whch scatter nto the telesesmc wavefeld, or that asymmetry exsts n the spectrum of energy radated horzontally relatve to that radated vertcally. Numercal smulatons are needed to explore these possbltes. The spatal varatons n complexty found for both Pahute Mesa and Yucca Flat strongly suggest that nformaton about the source regon structure s contaned n the telesesmc P wave coda. These varatons are slghtly dependent on the frequency band, beng partcularly coherent n the md-frequency band at both test stes. The spatal dependence of event complexty at Pahute Mesa s dffcult to separate from the sze dependence snce the largest events are located near the center of the mesa. However, the strong azmuthal patterns of complexty and log(rms ~ for events near the center of the mesa suggest that defocussng of the drect arrvals s part of the cause of the enhanced complexty of the coda n the md-frequency band. The spatal dependence n the md-frequency band for Yucca Flat s slghtly easer to nterpret snce event complexty n ths band s not complcated by depth or magntude dependence. Clearly, some near-source factor affects complexty. Possble

16 246 Chrstopher S. Lynnes and Thorne Lay PAGEOPH, canddates are the thckness of the Tertary tufts, whch may gve rse to larger Raylegh waves n the center of the basn (Fgure 6), or the basn shape tself, whch may defocus the drect arrvals to the north for events n the center of the basn. However, t s dffcult to determne what effect these two factors would have wthout sophstcated fnte-dfference modelng (e.g., STEAD and HELMBERGER, 1986). Whle the event average complexty results pont to near-source nformaton n the rato of the coda to the drect arrval, the azmuthal patterns of the early coda for Yucca Flat emphasze the szable contrbuton of a deep mantle effect. The most surprsng result s the hgh sgnfcance of the long wavelength component for the coda relatve to the drect arrval. Ths behavor contrasts wth that for Pahute Mesa, where the drect arrvals seem to be more strongly defocussed to the north by rela- tvely shallow structure than the coda. The ncreased scatter n the Yucca Flat drect sgnals suggests that the drect arrvals from the ndvdual events pass through very shallow structure that has lttle nfluence on the scattered phases makng up the coda (Fgure 10). Thus, the coda comprses arrvals that leave the source regon wth a Pahute Mesa Yucca Flat drect \ arrval Regonal Mantle Heterogenety Fgure 10 Schematc dagram showng the effects of near-source and deep mantle heterogenety at the Nevada Test Ste. Drect arrvals from Pahute Mesa are defocussed by a hgh-velocty structure that does not affect the scattered phase. Drect arrvals from Yucca Flat are complcated by trplcatons and caustcs resultng from the basn structure. All of the arrvals are then affected smlarly by a deep, regonal mantle heterogenety, whch mparts a common azmuthal pattern to drect arrvals and coda from both test stes.

17 Vol. 128, 1988 Studes of P Wave Coda for Underground Explosons 247 smaller ampltude varablty than the drect downgong energy. Then the entre sgnal encounters the deep mantle heterogenety nferred by LYNNES and LAY (1988), whch mparts a common azmuthal pattern. The drect arrvals have the same supermposed varaton, but the sgnfcance of the long wavelength component s reduced by the addtonal scatter from very near-source or near-recever heterogenety. The presence of strong azmuthal patterns n the early P coda for Yucca Flat events rases the queston of whether the early P coda s a more precse estmator of source sze than the drect arrval for Yucca Flat events. To answer ths queston, the event standard devatons were averaged for the three measures mg b, log(rmsws), and log(rms 5 15): the average standard devatons are for mg b, k-.02 for log(rms~ and 0.18 _+.02 for log(rms 5 ~5). Thus, the early P coda actually has somewhat hgher varance than the drect arrval for Yucca Flat events. Whle ths dfference may be partally due to the ncreased effect of nose, most of the varance n the early coda would seem to result from the long wavelength component produced by deep mantle heterogenety. Although the network-averaged coda measurements from Yucca Flat events are less precse than the mb measurements, BAUMGARDT (1985) found that the standard devaton of P coda across NORSAR for Sempaltnsk events was even less than that of the nose across the array. Ths suggests that array-averaged coda measurements are more precse than network-averaged P or P coda. Indeed, azmuthal patterns n the coda such as those from NTS would produce a large degree of scatter n networkaveraged measurements, but very lttle addtonal scatter at a sngle array, leadng to a comparable or hgher precson at the latter. Even more mportant than the precson of the dfferent measures, however, s ther accuracy. Clearly, measurements of the coda from NTS at a sngle staton or array have a hgh probablty of fallng on one of the lobes of the azmuthal pattern from NTS. (NORSAR, for nstance, occurs near the mnmum of the Pahute Mesa and Yucca Flat azmuthal ampltude patterns.) Consequently, sngle arrays may record coda magntudes wth very hgh precson by averagng out the near-recever effects, but cannot average out effects mparted by mantle heterogenety near the source regon, resultng n an absolute bas. Conclusons Data-ntensve analyss of telesesmc P wave sgnals has been used to solate near-source contrbutons to the early coda from NTS events. The 0.44).8 Hz frequency band has a strong negatve correlaton between coda complexty and magntude at both the Pahute Mesa and Yucca Flat test stes, wth smaller events havng hgher coda levels. Effects related to depth of bural alone do not seem suffcent to explan ths correlaton, snce the correlaton of complexty wth depth for Yucca Flat s farly low.

18 248 Chrstopher S. Lynnes and Thorne Lay PAGEOPH, Coherent spatal varatons can also be seen n event-averaged coda complexty for both test stes. These varatons are strongest n the Hz frequency band, and the patterns vary wth frequency for Yucca Flat events. Thus, some frequency dependence of focussng/defocussng or scatterng s ndcated. The Pahute Mesa events n the center of the test array generally have larger coda levels relatve to the drect arrvals. Ths appears to be at least partally due to defocussng of the drect arrvals by hgh veloctes to the northeast under the mesa. The md-frequency event average complexty for Yucca Flat s large n the northern part of the array, above the western axs of the Yucca Flat basn, and small n the northwest, above the east flank of a basement horst. These strong spatal varatons mply a sgnfcant nearsource contrbuton to the coda, whch could be exploted n an nverson for scatterers smlar to that of LAY (1987b) for Pahute Mesa. Strong, slowly varyng azmuthal ampltude patterns n both the drect arrval and coda are apparent for both test stes. Whle the azmuthal pattern s stronger n the drect arrvals than n the early coda for Pahute Mesa events due to crustal and shallow mantle scatterng or defocussng, the reverse s true for Yucca Flat, possbly due to very near-source heterogenety such as basement structure. The common component of the ampltude patterns n the drect arrvals and early coda for the two test stes appears to be due to focussng and defocussng by deep mantle heterogenety. Such heterogenety must be consdered as a possble source of bas when usng coda for sze estmaton. Acknowledgements Jud Sherdan and Martha Ballard dgtzed many of the sesmograms used n ths study. Ths paper benefted much from comments of two anonymous revewers. Ths research was supported by the Sloan Foundaton, a Shell Faculty Career Intaton Grant, and the Defense Advanced Research Projects Agency, and was montored by the Ar Force Geophyscs Laboratory under Contract F K4)010. REFERENCES AKI, K. (1982), Scatterng and attenuaton, Bull. Sesm. Soc. Am. 72, $319-$330. BACHE, T. (1982), Estmatng the yem of underground explosons, Bull. Sesm. Soc. Am. 72, S 131 S168. BAUMGARDT, D. R. (1985), Comparatve analyss of telesesmc P coda and Lg waves from underground nuclear explosons n Eurasa, Bull. Sesm. Soc. Am. 75, BOUCHER, G. (1973), Local sesmc phenomena n the frst three seconds after underground nuclear explosons, J. Geophys. Res. 78, BULLITT, J. T. and CORMIER, V. F. (1984), The relatve performance of m b and alternatve measures of elastc energy n estmatng source sze and exploson yeld, Bull. Sesm. Soc. Am. 74, CORMIER, V. F. (1987), Focusng and defocusng of telesesmc P waves by known 3-d structure beneath Pahute Mesa, Nevada Test Ste, Bull. Sesm. Soc. Am. 77,

19 Vok 128, 1988 Studes of P Wave Coda for Underground Explosons 249 DAINTY, A. M. (1985), Coda observed at NORSAR and NORESS, Fnal Techncal Report, AFGL-TR- 85~)199, Hanscom AFB, MA, 73 p. DOUGLAS, A. (1984), Telesesmc observatons of aftershocks mmedately followng an underground exploson, Geophys. J. Roy. Astr. Soc. 77, FERGUSON, J. F. (1982), Geophyscal nvestgatons of Yucca Flat, Nevada, n Fnal Techncal Report to the AFOSR, Contract no. F C~)010. GUPTA, I. N. and BLANDEORD, R. R. (1987), A study of P waves from NTS explosons--near-source nformaton from telesesme observatons? Bull. Sesm. Soc. Am. 77, GUPTA, I. N., BLANDFORD, R. R., WAGNER, R. A., BURNETTI, J. A., and MCELFRESH, T. W. (1985), Use of P coda for determnaton of yeld of nuclear explosons, Geophys. J. Roy. Astr. Soc. 83, LAY, T. (1985), Estmatng exploson yeld by analytcal waveform comparson, Geophys. J. Roy. Astr. Soc. 82, LAY, T. (1987a), Analyss of near-source contrbutons to early P wave coda for underground explosons: 2. Frequency dependence, Bull. Sesm. Soc. Am. 77, LAY, T. (1987b), Analyss of near-source contrbutons to early P wave coda for underground explosons: 3. Inverson for sotrope scatterers, Bull. Sesm. Soc. Am. 77, LAY, T., BURDICK, L. J., and HELMBERGER, D. V. (1984), Estmatng the yelds of the Amchtka tests by waveform ntercorrelaton, Geophys. J. Roy. Astr. Soc. 78, LAY, T. and WELC, J. (1987), Analyss of near-source contrbutons to early P wave coda for underground explosons: L Waveform complexty, Bull. Sesm. Soc. Am. 77, LYNNES, C. S. and LAY, T. (1988), Analyss of ampltude and travel tme anomales for short-perod P waves from NTS explosons, Geophys. J. 92, McLAUGHLIN, K. F. and ANDERSON, L. M. (1985), P wave dsperson due to scatterng and the effects on attenuaton and magntude determnaton, n The Vela Program (ed. A. U. Kerr) pp. 433~,22. MINSTER, J. B., SAV1NO, J, M., ROOf, W. L., JORDAN, T. H. and MASSO, J. F. (1981), Three-dmensonal velocty structure of the crust and upper mantle beneath the Nevada test ste, Fnal Techncal Report SSS-R-81-53!8, Scence, Systems and Software, La Jolla, Calforna. STEAD, R. J. and HELMBERGER, D. V. (1986), Progress n modelng strong motons of NTS (abstract), LOS, Transactons of the Amercan Geophyscal Unon 67, TAYLOR, S. R. (1983), Three-dmensonal crust and upper mantle structure at the Nevada Test Ste, J. Geophys. Res. 88, Receved March 15, 1987, revsed August 1 I, 1987, accepted August 15, 1987)