Source models are determined for the three underground nuclear explosions at the Amchitka test site using seismic observations in the period range 0.5 to 20.0 sec. Empirical yield-scaling relations are inferred from the source models and compared with the predictions of the Haskell, Mueller-Murphy, and finite difference numerical models. Several recent studies of high-frequency, near-field signals and teleseismic short-period P waves for LONGSHOT, MILROW, and CANNIKIN constrain the source functions at periods of 0.5 to 2.0 sec. Teleseismic pS and Rayleigh wave observations are used to constrain the source functions at longer periods. Using a modified Haskell source time function representation given by
the data are best-fit if the corner frequency parameter, K, scales as predicted by the Mueller-Murphy model, and if the amount of overshoot in the reduced displacement potential, which is proportional to B, decreases with increasing yield (depth of burial). The latter behavior is opposite to that predicted by the Mueller-Murphy model and follows from the observation that the long-period level of the explosion potential, ψ∝, increases with yield, W, by ψ∝ ∝ W0.90, or with yield and depth by ψ∝ ∝ W/h1/3. This long-period and overshoot scaling is consistent with that found for some numerical models, and allowing for the depth dependence of the Rayleigh wave excitation, results in the observed MS versus log(W) slope of ∼1. The decrease in overshoot with increasing depth of burial may be the result of the increase in shear strength with increasing overburden pressure. If yield or depth dependence of the source potential overshoot proves to be a general phenomenon, a possibility supported by a preliminary investigation of Pahute Mesa observations, accurate yield estimation will require broadband seismic data. The source function representation adopted is shown to provide an excellent fit to the rise time of very near-in velocity recordings to the rise time with frequencies of 10 Hz and higher.