04.08.2013 Views

P. Schmoldt, PhD - MTNet - DIAS

P. Schmoldt, PhD - MTNet - DIAS

P. Schmoldt, PhD - MTNet - DIAS

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

RMS−misfit<br />

5<br />

4<br />

3<br />

2<br />

0.1 1 10 100<br />

Roughness(1/τ)<br />

10.2. Inversion for mantle structures<br />

Solid lines:<br />

halfspace<br />

Dashed lines:<br />

layered model<br />

Iteration 1<br />

Iteration 2<br />

Iteration 3<br />

Iteration 4<br />

Iteration 5<br />

Fig. 10.23.: RMS misfit for isotropic 3D inversion models at different iteration steps (colour-coded) using a 100 Ωm halfspace and a<br />

layered arrangement as starting model (solid and dashed lines, respectively); see text for details about the layered starting model. This<br />

initial inversion sequence only involves five iteration steps for each starting model. Results are plotted in terms of RMS misfit versus<br />

roughness 6 for all models at each iteration step. The wsinv3d algorithm [Siripunvaraporn et al., 2005a] yields up to four models at<br />

each iteration step and uses the lowest misfit model as starting model for the following inversion step (cf. Sec. 6.3).<br />

and smoothness (i.e. inverse of model roughness) is observable for a higher number of<br />

iterations, i.e. iteration 5 (models with higher roughness and misfit represent inferior<br />

results). Inversion models of the fifth iteration step exhibit RMS misfits in the range<br />

2.0 – 2.5 (halfspace starting model) and 2.3 – 2.7 (layered starting model); thus, model<br />

responses exhibit an intermediate degree of similarity with the observed data. Despite the<br />

increased misfit of these models (owing to the limited number of iteration steps), results<br />

can be used to examine characteristics of the initial inversion sequence for both starting<br />

models.<br />

For sake of clarity and compactness, illustration is limited here to models from the fifth<br />

iteration, which yield the respective lowest RMS misfit for both starting models (cf. Fig.<br />

10.23). Herein, results are compared using a series of horizontal and vertical slices for<br />

the two models (Fig. 10.24). The most striking difference between models using either a<br />

halfspace or a layered starting model (labelled ‘a’ and ‘b’ in Figure 10.24, respectively)<br />

is the electric resistivity of the region associated with the lithospheric-mantle beneath the<br />

Tajo Basin (at depths between 30 km and 110 km). Inversion models exhibit a strong<br />

correlation in the lithospheric-mantle with their respective starting models, i.e. 100 Ωm<br />

and 1000 Ωm, respectively. Thus indicating that these deeper regions are not strongly<br />

constraint by the data or not well fit by the model at the current inversion step.<br />

Despite the small number of inversion steps, derived isotropic 3D inversion models can<br />

be used to infer features of the Tajo Basin subsurface that are supported by data of the<br />

PICASSO Phase I project. At crustal depths (≤ 30 km), models exhibit some degree of<br />

similarity with each other as well as with the focussed isotropic 2D inversion for crustal<br />

structures (cf. Sec. 10.1.5). However, due to increased cell size and small number of<br />

iteration, some of the features are not as well established as in the focussed inversion. For<br />

example, the extensive near-surface conductor that is modelled in the focussed inversion<br />

263

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!