[1] A new P wave tomographic model of the mantle was constructed using more than 10 million travel times. The finite-frequency effect of seismic rays was taken into account by calculating banana-donut kernels at 2 Hz for all first arrival time data, and at 0.1 Hz for broadband differential travel time data. Based on this model, a systematic survey for subducted slab images was developed for the circum-Pacific; including the Kurile, Honshu, Izu-Bonin, Mariana, Java, Tonga-Kermadec, southern and northern South America, and Central America, arcs. This survey revealed a progressive lateral variation of the configuration of slabs along arc(s), which we interpret as an indication for successive stages of slab subduction through the Bullen's transition region with the 660 km discontinuity at the middle. We identified the four distinct stages: I -slab stagnant above the 660 km discontinuity; II -slab penetrating the 660 km discontinuity; III -slab trapped in the uppermost lower mantle (at a depth of 660-1000 km); and IV -slab descending well into the deep lower mantle. The majority of slab images are found to be either at Stage I or III, suggesting that Stages I and III are relatively stable or neutral and II and IV are relatively unstable or transient. There is a remarkable distinction for the deepest hypocentral distribution between slabs at Stage I and slabs at Stages II or III.Citation: Fukao, Y., and M. Obayashi (2013), Subducted slabs stagnant above, penetrating through, and trapped below the 660 km discontinuity,

Abstract. We made a region-by-region examination of subducted slab images along the circum-Pacific for some of the recent global mantle tomographic models, specifically for two high-resolution P velocity models and two long-wavelength S velocity models. We extracted the slab images that are most consistent among different models. We found that subducted slabs tend to be subhorizontally deflected or flattened in the upper and lower mantle transition region, the depth range of which corresponds roughly to the Bullen transition region (400-1000 km). The deflected or flattened slabs reside at different depths, either above or across the 660-km discontinuity as in Chile Andes, Aleutian, Southern Kurile, Japan, and Izu-Bonin; slightly below the discontinuity as in Northern Kurile, Mariana, and Philippine; or well below it as in Peru Andes, Java, and Tonga-

The P wave velocity structure beneath the Western Pacific is found from the International Seismological Center first arrival data. Special attention was paid to the deep structure beneath the Wadati‐Benioff zone. We discretized the whole mantle into blocks with finer blocks in the region of interest to obtain the velocities of all the blocks. This way of discretization minimizes a problem with tomographic studies of regional scale: difficulty in making corrections for the effects outside the region of interest. Our solution is iterative with the alternate step of the relocation of earthquakes, using the whole mantle model of Inoue et al. (1990) as a starting model. A first‐order smoothness constraint was imposed to suppress the possible fluctuation of the solution around the initial model. The essential result depends little on whether the reference spherical model is smooth or discontinuous near 400‐ and 670‐km depths. We examined the resolution by calculating the resolving kernels for selected blocks and by reconstructing the checkerboard test patterns of velocity perturbation and the test structures of subducting lithosphere. The resolution is depth dependent but in general good enough to see the slab configuration beneath the Southern Kurile‐Japan‐Izu‐Bonin arcs and the Java arc. It is relatively poor beneath the Northern Kurile and Mariana arcs. The seismic image of subducting slab beneath the Southern Kurile to Bonin arcs bends to subhorizontal near the leading edge of the Wadati‐Benioff zone and extends continentward over a distance of more than 1000 km. The subhorizontal portion of the slab connects a high‐velocity blob to the bottom that reaches a depth of at least 800 km across the 670‐km discontinuity under the Japan arc. Although the image of the Java slab directly penetrates the 670‐km discontinuity, it then bends to a shallow dip with a considerable spread, reaching a depth of about 1200 km. These results suggest that descending slabs of lithosphere in the Western Pacific tend to be stagnant in the transition zone under a subtle control of the 670‐km discontinuity. Although stagnant slab materials eventually descend into the lower mantle, they no longer maintain their original configuration below the 670‐km discontinuity.

Volcanism that occurs far from plate margins is di cult to explain with the current paradigm of plate tectonics. The Changbaishan volcanic complex, located on the border between China and North Korea, lies approximately 1,300 km away from the Japan Trench subduction zone and is unlikely to result from a mantle plume rising from a thermal boundary layer at the base of the mantle. Here we use seismic images and three-dimensional waveform modelling results obtained from the NECESSArray experiment to identify a slow, continuous seismic anomaly in the mantle beneath Changbaishan. The anomaly extends from just below 660 km depth to the surface beneath Changbaishan and occurs within a gap in the stagnant subducted Pacific Plate. We propose that the anomaly represents hot and buoyant sub-lithospheric mantle that has been entrained beneath the sinking lithosphere of the Pacific Plate and is now escaping through a gap in the subducting slab. We suggest that this subduction-induced upwelling process produces decompression melting that feeds the Changbaishan volcanoes. Subductioninduced upwelling may also explain back-arc volcanism observed at other subduction zones.

[1] We present a new whole mantle P wave tomographic model GAP_P4. We used two data groups; short-period data of more than 10 million picked-up onset times and long-period data of more than 20 thousand differential travel times measured by waveform cross correlation. Finite frequency kernels were calculated at the corresponding frequency bands for both long-and short-period data. With respect to an earlier model GAP_P2, we find important improvements especially in the transition zone and uppermost lower mantle beneath the South China Sea and the southern Philippine Sea owing to broadband ocean bottom seismometers (BBOBSs) deployed in the western Pacific Ocean where station coverage is poor. This new model is different from a model in which the full data set is interpreted with classical ray theory. BBOBS observations should be more useful to sharpen images of subducted slabs than expected from simple raypath coverage arguments.