Specifications

University of Pretoria etd – Combrinck, M (2006)
3.4 Limitations on automation of inversion techniques
Forward modelling as defined in 3.3 cannot be automated as it includes the active
involvement of a geophysicist at every guess of a new model. In fact, inversion is the
process whereby “guessing of models” is taken over by an algorithm or computer.
Although mathematically sound, in practice there are a number of problems which often
inhibits the successful implementation and automation of inversion procedures.
• A priori knowledge of the geology under investigation is needed; i.e. should
inversion be run for a layered earth (with how many layers) or should it be done for
multiple plates (and how many plates)?
• Even if the general structure is known, the initial model should be close to the real
model to obtain mathematical convergence.
• In the case of convergence, it can be difficult to decide whether convergence was
to the desired global minimum or just a local minimum which adds yet another
unknown to the interpretation process.
• Equivalence
• Validity of assumptions; e.g. late time behaviour of models are compared with time
channels exhibiting early or intermediate time behaviour.
• There have to be more data than free parameters which would become a problem
when trying to implement a totally general three-dimensional cube as model (this
would of course be the ultimate solution).
• The most limiting factor however, is the need for a very fast forward modelling
algorithm. TEM algorithms are still very time-consuming for all but the simplest
cases of layered earths and multiple plates.
Practical experience in the interpretation of TEM data with inversion software has shown
that extensive forward modelling was needed before inversions could be run successfully
and that the 3-10% statistical reduction in error did not always mean that a more
geologically plausible model was achieved. In fact, correlation with geological data such as
borehole and structural information proved invaluable to distinguish between mathematical
equivalent models and more often than not mathematical accuracy had to be sacrificed to
obtain geological feasibility in a model. This is especially true in a geologically complex
area. The best of both worlds would naturally be the application of constrained inversion,
21