P. Schmoldt, PhD - MTNet - DIAS
P. Schmoldt, PhD - MTNet - DIAS P. Schmoldt, PhD - MTNet - DIAS
6. Using magnetotellurics to gain information about the Earth Fig. 6.5.: Types of minima types, illustrated using of a ball’s behaviour on a curved surface under the effect of gravity; redrawn from Bahr and Simpson [2002]. A ball released at a point ‘A’ will be trapped in the local minimum instead of the global minimum unless a sufficient force is applied to push it over the ridge between the minima. Iterative and stochastic inversion is similar to a ‘trial and error’ process wherein one or more starting models m0 are created and their responses d ′ are calculated using a forward modelling process (Sec. 6.3.2). The misfit of each model, i.e. the difference between model response and measured data, is given in terms of the error function ɛ, viz. ɛ = ||d ′ − d||. (6.29) A common choice of the error function is the L2 norm: ɛ = ||d ′ − d||2 = d ′ − d 2 , (6.30) i.e. the root mean square (RMS) misfit. Model parameters are subsequently adapted during the inversion process in order to minimise ɛ. The crucial part of an inversion approach is to optimise the adaption process in a way that it exhibits a high convergence rate but also adequately samples the model parameter space. The latter is to ensure that a global minimum of the error function is obtained, as opposed to a local one (Fig. 6.5). Non-uniqueness of MT inversion models Except for the idealistic case of 1D subsurface and noise-free data for the complete frequency range, MT inversion is non-unique, i.e. a range of models fit the measured data equally well. The non-uniqueness for the general case of inversion with differential equations was proven by Langer [1933] and for the MT problem by Tikhonov [1965], Bailey [1970], and Weidelt [1972] (Fig. 6.6); see also Parker [1983]; Constable et al. [1987]; Vozoff [1987] for illustration of the non-uniqueness problem in MT inversion. Rough models are superior in terms of data misfit, but often contain resistivity distributions that are not in agreement with physical laws, e.g. conductivities of a few hundreds of Siemens per meter. Therefore, more elaborate criteria are required to evaluate a model, incorporating additional (a priori known) constraints about the characteristics of the model, such as a limited parameter range or the so-called smoothness of the model. The smoothness is usually described in terms of first or second order spatial derivatives of the model param- 120 i
6.3. Deriving subsurface structure using magnetotelluric data Fig. 6.6.: The non-uniqueness problem of magnetotelluric (MT) inversion illustrated using of two very different inversion models (top). Both models reproduce the input data reasonably well (bottom), but describe very different subsurface behaviour. Figure and (dashed) inversion model taken from Weidelt [1972]; MT sounding data from Wiese [1965]; (doted) inversion model used for comparison by Fournier [1968]. eters m, which, in the case of MT application, is calculated using ∇ log(σ) and ∇ 2 log(σ) (or ∇σ and ∇ 2 σ) [e.g. Constable et al., 1987; Smith and Booker, 1988; de Groot-Hedlin and Constable, 1990; Smith and Booker, 1991]. Functionals ψ(m), which consider data misfit as well as model smoothness, have the form ψ(m) = ɛ(m) + τ · ξ(m), (6.31) where ɛ(m) is some measure of the data misfit, ξ(m) is some measure of the model smoothness, and τ is a weighting parameter (often referred to as smoothing parameter). The related inversion processes attempt to find a regularised solution of the inverse problem, i.e. models which yield a minimum of ψ(m). A higher degree of smoothing is generally desirable as it usually yields models with less structure, meaning that the persisting structures are more reliable. However, due to the additional constraints on the model parameters a trade-off between smoothness of the model and data misfit is commonly observed (e.g. 121
- Page 105 and 106: 4.4. Removal of distortion effects
- Page 107 and 108: Parameter Geoelectrical application
- Page 109 and 110: 4.4. Removal of distortion effects
- Page 111 and 112: 4.4.5. Caldwell-Bibby-Brown phase t
- Page 113 and 114: 4.4. Removal of distortion effects
- Page 115: Method Applicability Swift angle 2D
- Page 118 and 119: 5. Earth’s properties observable
- Page 120 and 121: 5. Earth’s properties observable
- Page 122 and 123: 5. Earth’s properties observable
- Page 124 and 125: 5. Earth’s properties observable
- Page 126 and 127: 5. Earth’s properties observable
- Page 128 and 129: 5. Earth’s properties observable
- Page 130 and 131: 5. Earth’s properties observable
- Page 132 and 133: 5. Earth’s properties observable
- Page 134 and 135: 5. Earth’s properties observable
- Page 136 and 137: 5. Earth’s properties observable
- Page 138 and 139: 5. Earth’s properties observable
- Page 140 and 141: 5. Earth’s properties observable
- Page 142 and 143: 6. Using magnetotellurics to gain i
- Page 144 and 145: 6. Using magnetotellurics to gain i
- Page 146 and 147: 6. Using magnetotellurics to gain i
- Page 148 and 149: 6. Using magnetotellurics to gain i
- Page 150 and 151: 6. Using magnetotellurics to gain i
- Page 152 and 153: 6. Using magnetotellurics to gain i
- Page 154 and 155: 6. Using magnetotellurics to gain i
- Page 158 and 159: 6. Using magnetotellurics to gain i
- Page 160 and 161: 6. Using magnetotellurics to gain i
- Page 162 and 163: 6. Using magnetotellurics to gain i
- Page 164 and 165: 6. Using magnetotellurics to gain i
- Page 168 and 169: Part II Geology of the study area I
- Page 170 and 171: 7. Geology of the Iberian Peninsula
- Page 172 and 173: 7. Geology of the Iberian Peninsula
- Page 174 and 175: 7. Geology of the Iberian Peninsula
- Page 176 and 177: 7. Geology of the Iberian Peninsula
- Page 178 and 179: 7. Geology of the Iberian Peninsula
- Page 180 and 181: 7. Geology of the Iberian Peninsula
- Page 182 and 183: 7. Geology of the Iberian Peninsula
- Page 184 and 185: 7. Geology of the Iberian Peninsula
- Page 186 and 187: 7. Geology of the Iberian Peninsula
- Page 188 and 189: 7. Geology of the Iberian Peninsula
- Page 190 and 191: 7. Geology of the Iberian Peninsula
- Page 192 and 193: 7. Geology of the Iberian Peninsula
- Page 194 and 195: 7. Geology of the Iberian Peninsula
- Page 196 and 197: 7. Geology of the Iberian Peninsula
- Page 198 and 199: 7. Geology of the Iberian Peninsula
- Page 200 and 201: 7. Geology of the Iberian Peninsula
- Page 202 and 203: 7. Geology of the Iberian Peninsula
- Page 205 and 206: Recovering a synthetic 3D subsurfac
6.3. Deriving subsurface structure using magnetotelluric data<br />
Fig. 6.6.: The non-uniqueness problem of magnetotelluric (MT) inversion illustrated using of two very different inversion models (top).<br />
Both models reproduce the input data reasonably well (bottom), but describe very different subsurface behaviour. Figure and (dashed)<br />
inversion model taken from Weidelt [1972]; MT sounding data from Wiese [1965]; (doted) inversion model used for comparison by<br />
Fournier [1968].<br />
eters m, which, in the case of MT application, is calculated using ∇ log(σ) and ∇ 2 log(σ)<br />
(or ∇σ and ∇ 2 σ) [e.g. Constable et al., 1987; Smith and Booker, 1988; de Groot-Hedlin<br />
and Constable, 1990; Smith and Booker, 1991]. Functionals ψ(m), which consider data<br />
misfit as well as model smoothness, have the form<br />
ψ(m) = ɛ(m) + τ · ξ(m), (6.31)<br />
where ɛ(m) is some measure of the data misfit, ξ(m) is some measure of the model smoothness,<br />
and τ is a weighting parameter (often referred to as smoothing parameter). The related<br />
inversion processes attempt to find a regularised solution of the inverse problem, i.e.<br />
models which yield a minimum of ψ(m). A higher degree of smoothing is generally desirable<br />
as it usually yields models with less structure, meaning that the persisting structures<br />
are more reliable. However, due to the additional constraints on the model parameters a<br />
trade-off between smoothness of the model and data misfit is commonly observed (e.g.<br />
121