After many years of studies I have finally finished my degree in geophysics from the University of Copenhagen. It all ended when I handed in my thesis the 15 December and defended it the 5 February. The thesis can be downloaded here: thesis.pdf (11 mb)
UPDATE 12-march-2012: Due to numerous requests I have talked to Bo Holm Jacobsen and Trine Dahl Jensen and they have agreed to provide the forward model and data on this blog post.
The forward model used in thesis was kindly provided by professor Bo Holm Jacobsen (firstname.lastname@example.org) from the geological department of the University in Aarhus, Denmark. The Matlab model can be downloaded here, but all attribution and questions should be give to Bo Holm Jacobsen.
The dataset used in the thesis was kindly provided by Trine Dahl Jensens (email@example.com), senior researcher at the Geological Survey of Denmark and Greenland. The data should only be used as educational and others uses are only allowed with permission by Trine Dahl Jensen.
Receiver function analysis is a seismic method utilizing that fact that seismic P-waves impinging on a subsurface boundary will result in refracted and reflected P and S-waves. The generated waves are only the result of the local structure and by deconvolution the local effect can be isolated in a data set called the receiver function. Reconstructing the local velocity-structure by comparing the observed receiver functions with synthetically generated receiver functions is usually regarded as a highly non-linear inverse problem. In this thesis I will apply four different algorithms to recreate the subsurface velocity structure from a synthetic data set as well as data collected in Greenland. The algorithms are: Uniform search, the Metropolis algorithm, Neighborhood search and the Levenberg-Marquardt algorithm. Overall, the algorithms were successfully in reconstructing the synthetic data. In the Greenland data set the algorithms generated models indicating a low velocity layer which is most likely an artifact of the difference between the wavelet used in the forward algorithm and the wavelet embedded in the data. This problem affected all the algorithms. For the Greenland data only the Metropolis and the Levenberg-Marquardt algorithm found reasonable models. It was also concluded that the Uniform search and the Neighborhood algorithm do not perform well in high dimensional problems.
Thesis defense presentation
As well as the thesis it self the defense is made in LaTeX using the Beamer class. I used the program impressive to present the pdf, see Impressive – presentation using pdf and python | tjansson.dk.
The defense presentation can be downloaded here: thesis-defense.pdf (8.7 mb)