We use empirical functions to characterize the shape of band-passed P, P0, and S0 coda envelopes of 93 earthquakes. The data sets consist of earthquake data from the western and northwestern Pacific recorded by arrays of several stations and include both hydrophone and geophone recordings on analog and digital instruments. Each earthquake was band-passed into six octave-wide frequency bands between 0.8 and 50 Hz. The empirical function (a Poisson function, Pn (&ggr;,t)=A &ggr;n+1 tn e-&ggr;t/n!) has parameters A (a measure of energy), &ggr; (falloff rate), and n/&ggr; (time of the maximum). The study showed that these parameters varied in systematic ways. P0/S0 area (A) ratio were found to be considerably higher for hydrophone data in the 20¿--35¿ range than for geophone and hydrophone data in the 8¿--20¿ range. The difference was significantly larger than between vertical and hydrophone data in the same range. In some instances a strong frequency dependence or ''reverse dispersion'' of the time of maximum amplitude (n/&ggr;) was observed, with lower (1--2 Hz) frequencies arriving up to 25 s later than the higher (10--20 Hz) frequencies. P, P0, and S0 average falloff rates &ggr; vary systematically with frequency, where the lowest frequencies most commonly have the minimum falloff rates and the middle frequencies have the maximum falloff rates.

The variation in falloff between earthquakes is large, with the trends only becoming apparent upon averaging many measurements. A strong increase in falloff rates at about 3 Hz may be associated with a decrease in the amount of scatters with scale lengths of about one half the wavelength of a 3-Hz compressional wave, which is about 1.3 km. Falloff rates were seen to decrease with epicentral range, reflecting an increased scattering with length of propagation path. The scattering that produces the coda is thus not concentrated in the source region but distributed along the propagation path. Falloff rates were seen to decrease wtih increasing age of crust. The difference may be due to the different average age of the lithosphere in these two geographical areas. Earthquakes at a depth of 100--300 km have systematically lower P0 falloff values than shallower events. This effect is clearly seen in trench earthquakes and indicates that significant scattering occurs within (or near) the deeper buried part of the subducting plate. The S0/P0 falloff ratio is a factor of 2 lower for the hydrophone data than for the geophone measurements. This difference is most likely due to conversion of shear waves at or below the seafloor. If we assume that the coda decay is controlled by intrinsic attenuation, then measurements of coda decay can be converted into estimates of the quality factor Q. At 10 Hz our falloff rates give QP0=1250, which is a reasonable number for a compressional wave quality factor in the uppermost mantle, and QS0≈QP0/2. The low shear wave Q is in disagreement with inferences made by other observers. However, the fact that the falloff rates are fairly constant in the frequency range 5--30 Hz but vary with propagation distance suggests that attenuation cannot be the only phenomenon determining the decay of the coda. Other factors, such as reverberations in a water column, may play an important part in the coda formation as well and may bias these estimates Q.