A new method recently developed by Hoshiba et al. <1991> was used to separate the effects of scattering Q-1 and intrinsic Q-1 from an analysis of the S wave and its coda in Hawaii, Long Valley, and central California. Unlike the method of Wu <1985>, which involves integration of the entire S wave energy, the new method relies on the integration of the S wave energy for three successive time windows as a function of hypocentral distance. Using the fundamental separability of source, site, and path effects for coda waves, we normalized the energy in each window for many events recorded at many stations to a common site and source. We plotted the geometric spreading-corrected normalized energy as a function of hypocentral distance. The data for all three time windows were then simultaneously fit to Monte Carlo simulations assuming isotropic body wave scattering in a medium of randomly and uniformly distributed scatterers and uniform intrinsic Q-1. In general, for frequencies less than or equal to 6.0 Hz, scattering Q-1, whereas above 6.0 Hz the opposite was true. Model fitting was quite good for frequencies greater than or equal to 6.0 Hz at all distances, despite the model's simplicity. The small range in energy values for any particular time window demonstrates that the site effect can be effectively stripped away using the coda method. Though the model fitting generally worked for 1.5 and 3.0 Hz, the model has difficulty in fitting the whole distance range simultaneously, especially at short distances.

Despite the poor fit at low frequency, the results generally support that in all three regions the scattering Q-1 is strongly frequency dependent, decreasing proportional to frequency or faster, whereas intrinsic Q-1 is considerably less frequency dependent. This suggests that the scale length of heterogeneity responsible for scattering is at least comparable to the wavelength for the lowest frequencies studied, of the order of a few kilometers. The lithosphere studied in all three regions can be characterized as a random medium with velocity fluctuation characterized by exponential or Gaussian autocorrelation functions which predict scattering Q-1 decreasing proportional to frequency or faster. For all frequencies the observed coda Q-1 is intermediate between the total Q-1 and expected coda Q-1 in contrast with theoretical results for an idealized case of uniform distribution of scatterers and homogeneous absorption which predict that coda Q-1 should be close to the intrinsic Q-1. We will discuss possible causes for this discrepancy. ¿ American Geophysical Union 1992