Untitled Document

Calibration and Inversion of Airborne Geophysical Data

R. Brodie and M. Sambridge

 

  Prior to 2006, holistic inversion http://rses.anu.edu.au/cadi/ar/ar05/cadi5.html of frequency-domain airborne electromagnetic data had been successfully applied in situations where there was considerable prior knowledge of the geology and conductivity structure of the survey area to constrain the inversion (Brodie and Sambridge, 2006a). Over the last year we have investigated how well the method works when there is a lack of prior information.   To simulate this scenario we inverted the same datasets without the use of a sophisticated prior reference model, downhole conductivity logs, or watertable depth constraints, all of which had been used in our previous work. We adopted a generalised multi-layer fixed layer thickness inversion model and used a homogenous reference model. To compensate for the lack of hard constraints we imposed vertical smoothness regularisation on the conductivity model to stabilise the inversion.   is possible to extract information from the subsurface along the connecting path between two stations by just using Earth's ambient seismic noise field. Due to the inter-station distance and spectrum characteristics of the noise field, the extracted signal is mainly the Green's function of Rayleigh wave type surface wave for vertical components.   The seismic broadband data was compiled from the temporary and permanent stations across the Australian continent from 1992 to 2006. The data was used to calculate the Green's function between each possible station pairs which resulted in a coverage of the continent as in earthquake tomography studies with over 1000 individual raypaths. Then seismic tomography was set to construct the group velocity image for Australian crust with frequency dependency. The image which was obtained from the seismic tomography, clearly maps the major geological units in the crust. The geologically older parts of the continent in the west have higher group velocities. In contrast to this, the relatively younger Phanerozoic belts are marked with lower group velocities       Figure 1. A comparison of downhole log conductivity measurements and holistic inversion model conductivities.     Although the downhole log data were not actually used in the inversion the correlation between the inversion model conductivities and downhole log data was high, as is demonstrated in Figure 1.

Each point on this figure relates to the average conductivity over 5 m intervals for all the 44 available downhole logs. We have also found that gain and bias calibration parameters were similar to those we had estimated by other methods that had incorporated substantial information.The extracted wavefield which is the Rayleigh wave component of the Green's function, from the correlations of permanent station CTAO with other stations.     Figure 2: The map in A shows the raypath distribution used in the group velocity tomography. Group velocity anomalies for two different frequencies are given in B for 0.2 Hz and C for 0.08 Hz.  

Figure 2 shows the conductivity of layer 12 of the holistic inversion model. Figure 3 shows an example of a section through the inversion models. It can be seen that the top of the conductive zone at around 15-20 m elevation is coincident with the elevation of the known depth of a saline watertable. Our research has led us to concluded that the holistic inversion can be successfully applied in cases where little or no prior information is available (Brodie and Sambridge, 2006b).

Figure 3. A conductivity section through holistic inversion model along the profile labelled A on Figure 2. The top of the watertable surface is shown by the black line at approximately 15-20 m elevation.

Since the multi-layer inversion requires many more inversion parameters we had to parallelise the holistic inversion code. For this we used the MPI paradigm implemented mainly via the PETSc code for distributed sparse matrix algebra. Parallelisation has allowed us to invert whole datasets, rather than subsets, for multi-layer models. We have run inversions with up to 8.07 million data and 3.40 million parameters on 64 nodes of the Terrawulf Cluster.

References:
Brodie, R. and Sambridge, M., 2006, A holistic approach to inversion of frequency-domain airborne EM data: Geophysics, 71, G301-G312.
Brodie, R. and Sambridge, M., 2006, Holistic inversion without prior information: Australian Earth Science Convention 2006 - Melbourne , Australia , ASEG, Extended Abstracts.