Specifications

University of Pretoria etd – Combrinck, M (2006)

modelling software still approximates complex geological units in terms of simplified

geometries as well as mathematical simplifications sometimes needed to solve a problem

(see 2.3). Since TEM is an active source method (as opposed to magnetics and gravity) the

transmitter current waveform, geometry and position all determine the actual shape and

amplitude of anomalies. Different systems also measure different components of the

magnetic field or its time derivative, again resulting in different anomaly shapes and

amplitudes. Even if all the above factors are accounted for, the TEM method allows only

for modelling of time varying current distributions unlike gravity, for example, where the

measured field is a direct and unique consequence of the subsurface density distribution.

Interpretation of data in sounding format implies that a model is constructed with variation

in depth only, and such a model always takes the form of a number of layer thicknesses

and corresponding conductivities. Data are presented either as emf(t), ∂

B/∂t, B(t) or more

commonly, apparent resistivity as a function of time. As with profiling data, a model is

determined through forward modelling, inversion or curve matching. Due to the one-

dimensional nature of this interpretation, inversion can be applied successfully provided

that the number of layers is known, a good starting model is used and the geology

conforms fairly closely to the one-dimensional assumption made in these algorithms.

3.3 Forward Modelling and Inversion

Forward modelling is a process whereby a geophysicist tries to match field data with

calculated values from a specified model by changing the model parameters until there is a

close correlation between the field and calculated data. Inversion is a mathematical

approach also described as optimization, minimization or solution of a system of non-

linear equations. Most of the various inversion approaches (an exception is Occam’s

inversion) reduce to the guessing of an initial model (initial parameters), forward calculating

the response of this model, determining the difference (error) between the calculated and

measured response and adjusting the initial parameters in a way that would minimize this

difference. A number of forward modelling and inversion algorithms have been published

ranging from a single current filament approximation (Barnett, 1984) through to the most

general case of complete three-dimensional models (Newman et. al., 1986; Newman and

Hohmann, 1988; Xiong, 1992; Wang and Hohmann, 1993).