We report here the results of a comprehensive seismic attenuation investigation along the paths connecting Canada's Yellowknife seismic array (YKA) with seven active nuclear explosion testing areas. The data consist of more than 600 explosion-generated teleseismic P wave records. A dual time-frequency averaging technique is used to take advantage of the array recording characteristics without the drawback of the conventional beam-forming, excessive annihilation of high-frequency signal energies. The dual averaging technique, deployed in conjunction with a multiwindow spectral analysis method, yields smooth amplitude spectra whose falloff at high frequencies suffers little from spectral leakage due to the familiar presence of a prominent low-frequency plateau. Measured in terms of t*, the highest attenuation (0.66 s) is found along the path which originates from the Tuamotu test area: somewhat less attenuating are the two paths which depart from the Pahute Mesa (0.59 s) and Yucca Flat (0.50 s) nuclear test areas, both located within the U.S. Nevada Test Site. We find t* for these three paths to be substantially (up to 0.21 s) higher than recently published estimates (e.g., Der et al., 1985). We attribute these disparities largely to differences in spectral leakage control capability between the conventional single window and the improved multiwindow spectral analysis methods.

The least attenuating paths all originate from the Soviet test areas: Novaya Zemlya (NZ), west Kazakhstan, Degelen Mountain (DM), and Shagan River (SR). The last two of these test areas. DM and SR, are both located in east Kazakhstan. The P wave signatures of the Soviet explosions are rich in high-frequency (>4.5 Hz) energies, and the YKA data (0.5--8.0 Hz) support a frequency-dependent t* whose value at high frequencies (>4.5 Hz) is as small as 0.17 s. To gain a grasp of the ramifications of the t* disparity between the multiple-window and the single-window results, we have compared explosion source time functions obtained by the multichannel deconvolution technique of Shumway and Der (1985) in order to assess their sensitivity to the input t* value. In our example involving the deconvolved source functions of five French Tuamotu explosions, we find that a 0.1-s t* difference is large enough to cause clearly discernible signature differences, in terms of the signal frequency content as well as the extractability of a secondary arrival some 0.4 s behind the first P arrival. This secondary arrival is believed to be the depth phase pP, a seismic signature of importance in both yield estimation and earthquake/explosion source discrimination. The absorption band modeling (Minster, 1978a, b) of the French Tuamotu explosion data yields 1.08¿0.05 and 0.079¿0.008 s for t*0 and &tgr;m respectively.

The corresponding parameter estimates derived from the U. S. explosion data are somewhat smaller. The t*0 and &tgr;m estimates are the smallest along the paths which depart from the four Soviet test areas. For the NZ-YKA path the t*0 and &tgr;m estimates are 0.56¿0.08 and 0.061¿0.013 s, respectively. Plagued by a strong trade-off between the two model parameters, these estimates are not tightly constrained, however. ¿American Geophysical Union