Velocity structure around the 660-km seismic discontinuity in subduction zones is crucial for understanding the thermal state, chemical composition, and dynamics of the mantle. Because of limited spatial coverage of data and resolution, the complicated structure, especially in sub-slab regions, remains ambiguous.
Topographic variation of the oceanward 660-km discontinuity in the Western Pacific is investigated by stacking S-to-P converted phases using large-array data. Beneath the Philippine Sea and Southern Kuril Islands, the mean depth of this discontinuity is elevated ~10 km and ~6.5 km with respect to the global average depth. Meanwhile, it is normal beneath the Sea of Japan where the subducted slab appears to be near-horizontally flattened above the 660-km discontinuity. Elsewhere, the 660-km discontinuity is ~11.5 km and ~6.4 km shallower beneath the Marianas and Tonga Islands subduction zones, respectively. This topographic variation suggests that the 660-km discontinuity beneath subducting slabs shallows as the dip of the slab increases. Using sources in the western Pacific and stations in China, 3D triplication waveform modeling of structure beneath the Southern Kuril subduction zone suggests that the high-velocity slab (+5% anomaly and ~100km thick) subducts to at least 560 km. No obvious high-velocity anomaly of the slab is observed around the 660 km depth. To match late AB and BC branch arrives at large distances, one of two possible low-velocity anomalies (-3% and ~30km thick) is necessary. The first is located beneath the subducted slab and might be due to a hot sinking anomaly emplaced at shallower depths away from the subduction zone. The second location is on the landward side of the subducted slab and above the 660-km discontinuity. We suggest this low-velocity anomaly could be generated by hot upwelling from lower mantle or small scale convection above the deep slab. Numerical modeling of mantle convection demonstrates that slab entrainment of hot sinking anomalies can explain both the topographic variation of the 660-km discontinuity oceanward of slabs and the possible low-velocity anomaly beneath the subducted slab in the Southern Kuriles.