We use travel-times from a temporary seismic deployment of 30 broadband seismometers and a national catalog of arrival times to construct a finite frequency teleseismic P-wave tomographic model of the upper mantle beneath astern Indonesia, where subduction of the Indo-Australian plate beneath the Banda Arc transitions to arc-continent collision. The change in tectonics is due to a change from oceanic to continental lithosphere in the lower plate as inferred from geologic mapping and geophysical, geochemical, and geodetic measurements. At this inferred transition, we seismically image the subducted continent-ocean boundary at upper mantle depths that links volcanism on Flores to amagmatic orogenesis on Timor. Our tomographic images reveal a relatively high velocity feature within the upper mantle, which we interpret as the subducted Indo-Australian slab. The slab appears continuous yet deformed as a result of the change in buoyancy due to the composition of the incoming continental lithosphere. Accordingly, there is a difference in dip angle between the oceanic and continental sections of the slab albeit not a gap or discontinuity. We suggest the slab has deformed without tearing to accommodate structural and kinematic changes across the continent-ocean boundary as the two sections of the slab diverge. These results suggest that deformation in tectonic collisions can be localized along a continent-ocean boundary, even at depth. We propose that future slab tearing may develop where we observe slab deformation in our study region and that a similar process may take place in collisions generally.

Our dataset consists of ~20,000 arrival times of teleseismic P and PP waves recorded across Indonesia (Table 1). We manually picked P and PP arrival times from 207 events recorded across the YS network of 31 broadband stations (Miller et al., 2016) between 2014 and 2017 for a total of 9,838 travel time residuals (Figure 2). To this we added 10,117 automatically picked arrivals from 339 events recorded across 59 stations from the AU, IA, and II broadband seismic networks between 2010 and 2018 (Table 1, Figure 2). These data were provided by the Meteorological, Climatological, and Geophysical Agency (BMKG) headquartered in Jakarta, Indonesia. Automatic arrivals were picked using SeisComP3 software using a short-term-average/long-term-average (sta/lta) trigger on broadband data filtered between 0.7-2.0 Hz. We obtained travel-time residuals using predictions from the global one-dimensional velocity model iasp91 (Kennett and Engdahl, 1991). To improve the quality of the manually picked residuals, we employed the cross-correlation optimization scheme of VanDecar and Crosson (1990) in four non-overlapping bandpasses (Table 1). Only measurements with cross-correlation coefficients ≥ 0.8 were used in the inversion. After pruning, all delays were demeaned per event.