Rifting Structure of Central Izu-Ogasawara (Bonin) Arc Crust: Results of Seismic Crustal Imaging - Accretionary Prisms and Convergent Margin Tectonics in the Northwest Pacific Basin - page 79

3 Data and Modeling Procedure

Data quality of the seismic profiles was generally good, and first arrivals could be
traced out to a distance of 70-100 km (Fig. 2 ). To trace refractions and reflections,
we applied a 5-15 Hz band-pass filter and a deconvolution filter when necessary.
In particular, the deconvolution filter helped in picking reflections, because these
were overprinted by reverberations of strong first arrivals. Figure 2 shows examples
of OBS record sections for vertical components on the two lines. Most sections display
refractions through the crust and upper mantle as well as reflections from the Moho.
For modeling, we applied tomographic inversion using first arrivals (GeoCT-II,
Zhang et al. 1998 ) and diffraction-stack migration for reflections (Fujie et al.
2006 ). This inversion has the advantage of allowing structural imaging without
phase identification, although we have to determine if the phases are refractions or
reflections. If reflections are mistakenly picked as refractions, a zone of artificially
low Vp is imposed on the Vp model. In particular, shadow zones without refrac-
tions likely broaden in regions having severe structural variations like those of arc
crust. We paid special attention to pick first arrivals that corresponded to refrac-
tion. In addition, the tomographic inversion easily creates an artificial Vp anomaly
when an inappropriate initial model is compared with the real structure. The OBS
interval of 5 km might be too large to determine shallow structure; therefore, we
applied a Vp model based on MCS velocity analysis to construct the shallow part
of the initial model.
We carried out tomographic inversions with the following steps. We first applied
a tomographic inversion using first arrivals out to 20 km from each OBS. We
applied a second tomographic inversion using results from the first inversion and
first arrivals out to 60 km from each OBS. Then we applied a third inversion using
the second result and all first arrivals. We did this to avoid cases with artificial
anomalies introduced to satisfy time differences between far-offset first arrivals and
synthetics. Using a final model based on the third inversion, we picked and mapped
reflections from the Moho (PmPs) and the top of the subducting plate. Ideally,
regions having sharp Vp gradients should correspond to reflections, which are
imaged by mapping picked reflections. However, this situation is rare because of
the small number of deep refractions and the variation in ray density. Therefore, we
constructed a new initial model using the mapped image of reflections and the final
model of the second inversion. At this step, the initial model has a sharp Vp gradient
matching the distribution of reflection points obtained by the diffraction-stack
migration technique. Then the final inversion was performed again using the new
initial model and all first arrivals, and a new reflection image was also constructed
by the diffraction-stack migration technique.
We applied this procedure to data from both lines. To compare the crustal images
for these lines, it is important to use common specifications for Vp modeling.
Because the images depend on the ray densities, comparing the two images is not
always straightforward. To check the resolution of the obtained image, we used a
checkerboard test (Fig. 3c ) with the following steps. First, we constructed a Vp model
having perturbations of 5% using the initial model from the last inversion, and