SEISMIC-SOURCE AND WAVE-PROPAGATION EFFECTS OF L(G) WAVES IN SCANDINAVIA

A set of 517 recordings of L(g) waves from 151 earthquakes in and arou nd Norway has been used for determination of seismic moment M(0), corn er frequency fo and anelastic attenuation Q(f) The data used have been recorded at source-receiver distances of 20 to 1200 km, with M(L) mag nitudes between 0.8 and 5.0, and the parameters were estimated by inve rtine Fourier spectra from all of the recordings simultaneously. The o bserved spectra were represented;by a source term, a spreading term, a nd an attenuation term, and the inversion was made assuming the geomet rical spreading to be known. Because of the non-linear behaviour of th e spectral shape, the inversion was done iteratively by minimizing the differences between observed and computed spectra. A standard omega(- 2) source model was used in the inversion, supported by near-field obs ervations of small-magnitude earthquakes at the regional NORESS array. A model for geometrical spreading was then established by investigati ng the decay of L(g) waves using synthetic data modelled without anela stic attenuation, for a realistic crustal structure with a Moho depth of 40 km. The results support the standard model of spherical spreadin g at short distances and cylindrical spreading at longer distances, bu t with a transition distance closer to 200 km than to the more commonl y used value of 100 km. The decay rate beyond this distance was found to be close to -1/2 in the frequency domain, equal to the theoretical value for cylindrical spreading, and -3/4 in the time domain, where th e theoretical value for an Airy phase is -5/6. The data used in the in version were amplitude-displacement spectra corrected for instrument r esponse and site effects, reduced (by smoothing) to 64 spectral values and weighted to represent equidistantly spaced points in the log-freq uency domain. Only spectral values with signal-to-noise ratios of at l east four were accepted, with an upper limit set to 10 Hz, and a lower frequency limit determined by the instrument response. The computed M (0) values are quite stable, and exhibit a linear dependency on M(L). The corner frequencies are less well constrained, however, especially for smaller events. In using a model of the type f(0) proportional to M(0)(-delta) we found a delta around 3.4, indicating slightly increasi ng stress drop, with values of less than 10 MPa in all casks (using a Brune model). The resulting Q models of the type Q(f)=qf(eta) yield q values around 440 and eta values around 0.7. This is reasonably close to other anelastic attenuation models found in this area.