Deep earthquake rupture histories determined by global stacking of broadband P waveforms

We conduct a systematic investigation of time histories of moment release in deep earthquakes, whose physical mechanism remains unknown. We have constructed source time functions (STFs) of 111 large deep earthquakes (depth ≥100 km and M w ≥ 6.4) by stacking P waveforms from broadband seismograms recorded by the Global Seismographic Network. Aligning and stacking waveforms cancels noise and allows us to identify the rupture initiation and termination times accurately. Each STF is scaled to a reference moment of 10 19 N m, using the duration‐moment relation τ ∝ M 0 1/3 . In order to investigate depth dependence of the STFs, we bin them into three depth ranges, 100–350 km, 350–550 km, and deeper than 550 km, based on physical properties of the mantle and the distribution of the data. We infer a change in character near 550 km. Earthquakes deeper than about 550 km are shorter, have simpler time functions, and are more uniform in their characteristics. Further, we find longer than expected scaled durations and greater complexity in the 350–550 km depth range. The scaled durations of the 350–550 km earthquakes do not decrease with depth as 1/ V S , instead possessing a slightly longer average scaled duration than those 100–350 km deep. We further categorize the earthquakes by subduction zone and find a similar depth dependence within individual zones. Finally, the events that have achieved highest moment rate in the first 1–2 s after initiation tend to be deeper and/or larger events, the latter suggesting that dynamic rupture propagation affects the final size of the earthquake. The depth dependence of duration and complexity suggests more complicated rupture histories between 350 and 550 km and a possible change in the physical mechanism of rupture at about 550 km.