Supplementary Note S1: Resolution, stability and synthetic models

Model resolution matrix (MRM)

We determine the resolution of our velocity models by analysing the model resolution matrix (MRM) provided by the tomographic inversion algorithm. Analysis of the MRM allows us to determine the resolution at each node in the model, and the extent to which any smearing into neighbouring nodes occurs. By inspecting the spread function [Toomey and Foulger, 1989] and the shape of the 70% contour of the resolution kernels at each node, we can determine the spatial distribution and strength of smearing throughout the model. Figure S2 shows these two parameters for our Vp and Vp/Vs models. We find that the all nodes in the forearc are well resolved, particularly those in the onshore forearc. Smearing becomes apparent in the outer-rise and beneath the magmatic arc. The shallow areas of the marine forearc are also poorly resolved since we only use land-based stations in the inversion. The Vp/Vs model is inherently less well-resolved due to the fewer number of S-picks compared to P-picks and the greater error involved with S-picks (Figure S1). The resolution limits (white lines) are based on a spread function of 2 for the P-wave velocity model and a spread function of 4 for the Vp/Vs model.

Bootstrap analysis
We perform 100 inversions using 397 randomly-generated events from a larger catalogue of 750 events to assess whether any features in the observed velocity model are simply artefacts created by a small subset of events. The spatial variation in P-wave velocity standard deviation across the 2D model space is shown in Figure S3. We find that the P-wave velocity determination is most unstable in the outer-rise region; here the final model is strongly dependent on the earthquakes used in the inversion. This is likely due to the incorrect depth estimation of earthquakes in this region. However, we find that the velocity is very well constrained (sigma-Vp < 0.2 km/s) in the region where we observe the P-wave velocity anomaly (130 km from the trench, at 25 km depth). We therefore believe that the high velocity anomaly observed in our final model is not an artefact caused by a small subset of events.

Synthetic recovery tests
We develop synthetic 3D vp and vp/vs models in order to test the resolving capability of our inversion scheme, and to determine whether features observed in the real models are required by the data. Using the same source locations as the real inversion, we traced their rays through the synthetic models, calculating their travel-times. We added Gaussian noise to the travel-times, based on the standard deviation of the automatic picking errors (Figure S1). We use same synthetic model as that used in the tomographic study for south-central Chile [Haberland et al., 2009], so we include features such as the marine forearc, central basin, subducting oceanic crust and the continental mantle. In the middle of our synthetic model, we also include we include a velocity anomaly on the subducting interface to recreate the seamount inferred from our real velocity models. We test two different seamount sizes: a large seamount with 10 km relief, and a small seamount with 5 km relief (Figure S4, S5). We find that no significant artefacts are introduced and the primary features in the forearc are well recovered. The large synthetic seamount is recovered well, although it does appear more rounded (Figure S4). The recovered smaller seamount is recovered well, but its associated anomaly has much smaller topography on the subduction interface. Our observed anomaly (Figure 2) is represented best by the recovered large seamount (Figure S4); therefore, the real data require the presence of a large (5 – 10 km) topographic anomaly on top of the subducting Nazca slab.

References

S1. Toomey, D., & Foulger, G. Tomographic inversion of local earthquake data from the Hengill-   Grensdalur central volcano complex, Iceland, J. Geophys. Res., 94, 17497–17510 (1989). 
S2. Haberland, C., Rietbrock, A., Lange, D., Bataille, K. & Dahm, T. Structure of the seismogenic zone of the southcentral Chilean margin revealed by local earthquake travel-time tomography. J. Geophys. Res., 114, B01317 (2009).