Single-station decomposition of seismograms for subevent time histories

We have adapted an iterative least-squares inversion with positivity constraints to the problem of deconvolving the recordings of a small earthquake from the recordings of a large composite earthquake. The inversion uses an F test to minimize the number of non-zero amplitudes or subevents in the solution set, by requiring each added subevent to reduce the variance significantly. We apply this inversion to the analysis of strong motion recordings of a set of moderate-sized (4.0 ≤ M_L ≤ 6.4) aftershocks of the 1976 Friuli earthquake. The earthquake pairs are selected by comparing S trigger times and polarization diagrams. The deconvolved time series are interpreted as the time history of slip events or stress release in the composite earthquake, depending on whether the recordings of the small earthquake are unfiltered or filtered to simulate the waveforms radiated by subevents. Deconvolving with the filtered recordings reduces the variance more rapidly, as a function of the number of subevents, than deconvolving with the unfiltered recordings. The deconvolved time series for three of the four earthquakes analysed indicate that the rupture processes resemble the failure of asperities where the initial subevent has the largest stress drop and the later subevents have longer durations and weaker stress drops.