Robust retrieval of a seismic point-source time function

A comparison of two waveform-inversion methods designed to retrieve the mec hanism of a seismic source and of its time function is presented using vert ical-component synthetic signals, computed for velocity with a maximum freq uency of 10 Hz. The geometry of the array of recording stations simulates t he northeastern Italy Seismometric Network (OGS Trieste), consisting of 16 stations of which 12 are short period, vertical component only, three are s hort period, three components, and one is broad band. The synthetic seismog rams are inverted using an inconsistent forward modelling technique; that i s, by means of Green's functions (GFs) constructed for a structural model d ifferent from those used to generate the synthetic data. The approach based on 'overparametrization' of the rupture process, by means of independent m oment tensor rate functions (MTRFs), and their subsequent reduction to the source time function (STF) (Method I) is shown to be superior to a traditio nal approach where the rupture process is constrained a priori (Method II). With Method I, the effects of inconsistent structural modelling are partia lly absorbed into the uncorrelated parts of the MTRFs and their reverse sli ps, which allows us to eliminate them by subsequent retention of the STF as their positively constrained correlated part. Method I is shown to be able to yield a reasonable estimate of the STF even in the case when the tradit ional approach fails completely. Inadequacy of the GF, which may occur due to mislocation of the hypocentre, is taken into account by comparing the tw o approaches: the source depth is optimized simultaneously with the determi nation of the mechanism and the source time function. In addition to its ca pacity to handle inaccurate structural models, the overparametrization yiel ding a linear inverse scheme is completely independent from the starting mo del of the mechanism: Method II, using a gradient scheme, can proceed prope rly only if the starting source parameters are sufficiently close to the tr ue ones. The extension of the comparison of the performances between the tw o methods to an M-d 3.0 earthquake near Friuli, in February 1988, recorded by seven stations of the OGS Trieste Network gives results in good agreemen t with the synthetic tests. The orientation of the nodal planes retrieved u sing Method I is in good agreement with the orientation of the source mecha nism retrieved from the polarity of first arrivals, while Method II gives c onsistent results only when starting from the source parameters retrieved u sing Method I.