Schmucker-Weidelt Lecture Notes, Aarhus, 1975 - MTNet
Schmucker-Weidelt Lecture Notes, Aarhus, 1975 - MTNet
Schmucker-Weidelt Lecture Notes, Aarhus, 1975 - MTNet
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
Electromagnetic Induction<br />
in the Earth<br />
<strong>Lecture</strong>s <strong>Notes</strong> <strong>Aarhus</strong> <strong>1975</strong><br />
Ulrich <strong>Schmucker</strong> and Peter <strong>Weidelt</strong><br />
Laboratoriet for anvendt geofysik<br />
<strong>Aarhus</strong> Universitet
Electromagnetic Induction -- in the Earth<br />
Contents<br />
1. Intrcduction, bas5.c equati.ons<br />
1.1 . General ideas<br />
1.2. Basic equations<br />
2. Electromagnetic induction i r ~ one-diniensional structures<br />
2. ? . The yenerai solution<br />
2.2. Free modes of d-ecay<br />
2-3. .Ca.lculation of the TE-field for a layered structure<br />
2.4. Particular source fields<br />
2,5. Definition of the transfer function C (w ,K)<br />
2.6. Properkies of C(W, K)<br />
3. I.4odel calculations For two-dimensional struc-tures<br />
3.1 . General equations<br />
3.2, Air-half -space and conductor as basic domain<br />
3,3. A~lomalous slab as basic domain<br />
3.4. Anoinalous region as basic domain<br />
4. Model. calculations for three-diwensional structures<br />
4-7. Introduction<br />
4.2. Integral equation metb-06<br />
. 4.3. The surface integral approach to the mndell-iny<br />
p?:obl.em<br />
5. Conversion formu:lae for a t~~7o-d:i1aensi.ol1al TE-field<br />
5.1 . Separation formulae<br />
5.2, Conversion formulae for the f ield components of<br />
a two.-dimensional YE-fri.el-il at the surface of<br />
a one-dimensional st-ructure<br />
6. Approaches to the inverse problem of electromagnetic<br />
iaduction by linearization<br />
6.1. The Backus-Gilbert method<br />
6. 1 . I. Lntroiiuction<br />
..G , I .2. The linear inverse prol~l.em . .<br />
6.1.3. The non~linear inverse problem
6.2. Generalized matrix inversion<br />
6.3. Derivation of the kernels of the 1-inearized inverse<br />
problem of electromagl~etic induction<br />
6.3.1. The. one-dimensional case<br />
6.3.2. Pa.rtial. derivatives in two and three<br />
diniensions<br />
6.4. Quasi-linearization of the one-dimensional inverse<br />
problem of electromagnetic inducti.0~ (Schlfiucker's<br />
Psi-algoritlun)<br />
7, Basic concept of geomagnetic and magnetotell.u~ic<br />
depth sonding<br />
7.1. Gelleral chaxacteristics of the method<br />
7.2. The data and physical properties of internal<br />
matter which arc involved<br />
7.3, Electromagnetic wave propagation and diffusion<br />
through uniform do~nains<br />
Appendix to 7.3.<br />
7.4. Penetration depth of various .types of geomagnetic<br />
variations and the overall distribution of con-<br />
ductivity within the Earth<br />
8. Cats collection and analysis<br />
8.1. Instruments<br />
8.2. Organisation and objectives of field operatLons<br />
8.3. Spectral arlaLysis of ge0magneti.c' induction data<br />
8.4. Data analysis in the time domain, pilot studies<br />
9. Data interpretation as the basLs of selected models<br />
9.1. Layered half-space<br />
9.2. Layered sphere<br />
9 .'3. Non-uniform thin sheet abo-$e layelred substructure<br />
9.4 Non-uniform Layer above layered subs.bructure<br />
10. Geo;?hysical and geological rel.evance ofgeornagnetic in-<br />
duk.kion studies<br />
11. References for general reading
- 1. Introduction. basic equations<br />
-- I.. 1.. General. ideas<br />
The fo1.lowing rnodel j.s the basis ~fall e1ectromagne'ti.c lne-thods<br />
which try to infer the electrical conductivity stl2ucture in the<br />
Earth's subsurface from an anal.ysis of the natural or art.lfici.a~l<br />
electromagnetic surface field: On or above the surface of -the Earth<br />
is situated a tine-dependent electr'omagnetic source. By Fiiraday's<br />
1.a.w the time-varying inagnetic field - 13 induces an elec.trica1 field<br />
- -<br />
this cu~,rent<br />
F, which drives within the conductor a current 1. By kinp5re's law<br />
has a magnetic fi.eld which again is an inducing ageri-t,<br />
and so on. Hence, there is the closed chain<br />
. induction<br />
__I_ i -><br />
.-source<br />
l5<br />
//<br />
E<br />
-. -<br />
*mp.res '011.ct:o.<br />
law<br />
-<br />
The mathematical exprfssion of this closed chain is a second orde-r<br />
~art?.al. differen-tia:l equation resulting after the e1iminati.c.n of<br />
two field quantities from the two first order equations 1 + 3 and<br />
At the surface the ~lectrornagnetic field of the source is dl.sturbcd<br />
by the internal fie1.d~. This di.sirurbance depends on the conduc'ti-
A1~Lernativel.y<br />
p = 30. m, where T in h. (lb 1<br />
Eqs, (la,b) can also be used as a rule of thumb if p varies with<br />
depth and an average resistivity is inserted.<br />
The natural magnetic source fields of the ionospheric and magneto-<br />
spheric currents offer periods approxiniately between 0.1 see and 20<br />
permitting crustal and upper mantle surveys down to a depth of<br />
approximately 800 - 1000 km. On the other side periods between 0.91<br />
and 0.002 sec are most commonly used for Sounding of? -?he first 300 c<br />
with artificial fields.<br />
Electromagnetic methods are applicab1.e for two objectives:<br />
1) ~nvestigation of the change of conductivity with depth, in par.ti-<br />
cular detection and delineation of horizontal interfaces marking<br />
a change of stratigraphy or temperature.<br />
2) Investigation of lateral conductivity variations, in particular<br />
search for local regions wiCh abnormal conductivities (e.g.<br />
meta1lj.c ore deposi-ts, salt domes, sedimeni-dry' basins, zones of<br />
eleva-ted temperature).<br />
.The results of the first investigation are often used. to construct<br />
a normal. conductivi-ty model from which local deviations are measure1<br />
In an interpretation of natural fields the c:-.ange of the electro-<br />
magnetic field quantities both with frequency and with position are<br />
used. Broadly speaking .the dependence on the position provides the<br />
lateral reso1.u-tion and the dependence on frequency gives the reso-<br />
lu-tion with depth.<br />
1.2. Dasic equations -<br />
Let - r be .the position vector and let H, g, 2nd + I he the vectoi-s of<br />
the magnetic field, electrical field and curl-2n.t density, ~?especti-<br />
vely. Using SI u11.i.-ts and a vacuurrl permeability throughout:, the<br />
0<br />
pertinent equations are<br />
cupl - H = - I i- L<br />
11.2)<br />
. -e<br />
CUI?].' - E = --kt H<br />
(1.3)<br />
0-<br />
. .<br />
'- 1 = 0- C<br />
(l.4)<br />
div - !i = 0<br />
(1.5)
All field quanti-ties are functions of position _r and time t. Al.so tl:c<br />
(isotropic) conductivi.-t)~ u is a func.tion of position. -e I 3,s the<br />
current density of the external cilrrent sources being different from<br />
0 only at source points. The displacement current c 0- E is generally<br />
neglected in induction studies. This is justified as follows: within<br />
the conductor -- .the conduc-tion current UC - exceeds the displ.acement<br />
current even in the case of ].owest periods (0.001 sec) and highest<br />
5 2<br />
resis1:ivities (10 Rm) still 10 times. In the vacuum where .the<br />
conduction current is absent the inclusion of the, displacsment<br />
current merely introduces a slight phase shift of the external field<br />
at different points. for in this case the solution of (1.2) and (1.3<br />
(including at the RFIS of (1.2) ) are electromagnetic waves, e. g.<br />
the plane wave e i(k*r - - - u ~ ) where , w is the angular frequency and<br />
.- k the wave vector along the direction of propagation. The phase<br />
difference<br />
4<br />
between P and P2 is<br />
1<br />
,/ case<br />
- .<br />
PI Ax P2<br />
MAX<br />
l,j&t = -- coscl<br />
C<br />
.Z in the very pessimistic<br />
of Ax = 1000 km and T = lse<br />
- H ar~d - .I can he eliminated fPonl (1.. 2 - (1.4). It results<br />
L Induction equation<br />
----- - I<br />
6!t using C1.2) and (1.4) the elec-trical charge density p(1') - is<br />
given by<br />
p = E div - E = - c E - grad I.oga, (1.7)<br />
0 0 -<br />
i.e. there is charge accumulated if an electrical field compunclnt :<br />
parallel to the gradj.en.t. of the conductivi-ky. Physically 'this is<br />
cl.ear,, since the normal. componen-t of .the current density is conti-<br />
nuous whereas the charges accoun-i: for tl-~e corresponding discon-ti-<br />
nuity of the normal electri.cal. field.
The e1ectrica:L effect - of the changes is vel?y important. They modify<br />
by a-ttraction and repulsion .the 'curren-t flow; in particul.ar surface<br />
changes deflec-t the current lines in such a way tllat the norinal<br />
current coriiponent vanishes a-t the surface. Iil con-tras-t the magnetj-c -<br />
effect i.s of the ordel- of that of the disp1acemel-i-t current and<br />
can be neglected.<br />
2. E1ectromagneti.c induction in I-di.mensiona1 structures<br />
2.1. The q,eneral solution<br />
The following vector analytical. identities are used in the sequel<br />
(A<br />
- is a vector, $ a scalar).<br />
div curl - A = 0<br />
(2.1)<br />
curl grad + = 0 (2.2)<br />
div(A+) - = @ div - A -1. & a grad 41 (2.3)<br />
curlCA+) -- = + curlA - Ax - grad @ (2.4)<br />
curl"+ - = grad divA - - AA - (2.5)<br />
In this cha-pter it is assumed -1-hat the el.ec.trica1 conductivity o is<br />
a fun~tion of depth z (positive downwa~ds) on1.y. In this case -the<br />
electromagnetic field inside and outside the conductor can be re-<br />
presented in terix of two scalars and $N(~) denoting the elec-<br />
L -<br />
tric and magr,e"ric. type of solution. These fields are defined as<br />
'% .<br />
follows IS the vector in z-direc-tion):<br />
h A<br />
On using t11.e. ident::.t)i- curlC$z) - = '-z - x gl~adC, Cfrom (2. 4 ) 1, i-t is se@!<br />
.that -the ?$-type solu.!:ion has no nagne-tic z.-component and .the E-type<br />
solution has no electrical .- z-component.
.<br />
The electric field of the E-type solution is tangential to the planes<br />
6 = const. IIence, 2-t is called a TE-mode. (tangei~tial -- e:!ectric). Conversely<br />
the M-type solutj.on is name3 a TM-mode.!!~ required from (1.5)<br />
-the 111agnetic field. of both modes is sol.enoidal (i.e. divl-I - = 0). In<br />
addition the E-field of the TE-mode is also solenoidal.. In this<br />
mode there is no accumulation of charge, all current flow is parallel<br />
to the (x,y)-plane.<br />
ldow we have to derive the differen.ti.al equatior~s for the scala~s<br />
GE and $M. The TE-field (2.8a,b) satisfies already (1.. 3) - (1.5).<br />
It remains to satisfy<br />
~u1~l.H~ = 0 g*.<br />
Using (2.5), (2.2), and (2.4) it results<br />
from which Eollows<br />
(fop if afiax = 0 and af/ay : 0 and fi.0 for x,y -> -, then f=O).<br />
From -this equation fol.lows with the same argurnc1-i-ts<br />
lil urliform domains, (2.9) and (2.10) agree. Using (2.9), (2.10),<br />
(2.5), and (2.4) the TE- and TM-fields in the vacuun oiltside the<br />
sources are<br />
11 = 0, 11: = grad -<br />
--M -PI a L<br />
wL 0.<br />
-E H = grad - az' -.E<br />
E = y z x grad $E<br />
O-<br />
Fields on-tside the conductor<br />
-- II<br />
. ..<br />
(2.12
The vari.ables x,y, and t which do not explici.tly occu~ i.n (2.9)<br />
and (2.10) can be separa-ted from Q E and OM by exponential factors.<br />
Let for OE or OM<br />
- x - Y<br />
A<br />
where K = K x + K and K~ = -<br />
,-.<br />
Then (2.3) and (2.10) mad<br />
K = - K . (2.14a,l<br />
The general solution of (1.2) - (1.5) for a. one-dimensional conduc-<br />
tivity structure is the superposition of a TE- and TM-field. The<br />
total field then reads in components:<br />
--- .--a<br />
a20E<br />
H = -Co$ ) + --x<br />
ay M ax;)~<br />
a ae QE<br />
H = - -(uQ ) +<br />
Y ax M ay a z<br />
H = -(- a<br />
z<br />
ax2 ay2<br />
WE<br />
-<br />
1 a "~l$~) -<br />
Ex-o axaz a~<br />
., 7<br />
1<br />
E = -<br />
y o ayaz<br />
3 - "8,<br />
a I - (-- + a<br />
E~ :)OM<br />
ax2 ay<br />
+ "0 ax<br />
General expression of fie1.d components<br />
- -<br />
At a l>.orizontal di.scon-l-inuity the tangential. compone~~fs of - E and - H<br />
and ille ndriaal. con~ponenl: of I1 arc coiitinuous. Let [ ] deno-te<br />
Z<br />
the j~unp of a p;lrticulaT quanti-ty, Theil frorr~ (2.15~) ]?OF )I z<br />
(2.15a)<br />
(2.15b)<br />
(2.15~<br />
(2.16a)<br />
(2.16b)<br />
(2.16~
[$J satisfies the two-dimensional Lapl-ace equa-tion, is bounded<br />
and \rani-shes at infinity because of the finite extent of the<br />
sources. Hence, from Liouvi.lleLs theorem of function theory<br />
[Q,] = 0, i.e. @ C continuous. Differentiating (2.15a) with respect<br />
to x and (2.15b) with respect to y and adding we infer along the<br />
same lines thxt a$ /az j.s coiltinuous. Conversely differentiating<br />
C<br />
C2i15a) with respect to y and (2.15b) with respect to x and sub-<br />
tracting one obtains oQM continuous. Fina1l.y differentiate (2.16a)<br />
with rbspec-t to x and (2.1.6b) with respect to y and add. It re-<br />
sults that ('licr) a(oQ )/az is con.tinuous 01- 34 /2z is<br />
r! M<br />
continous<br />
if cf tends to a constant value at both sides.<br />
Summarizing: - 1<br />
+EJ a~ M<br />
-- 1<br />
a az<br />
continuous<br />
I Conti.nuity conditions I<br />
This result shows that the TE- and TM-field satisfy disioinl ingab;<br />
k.bLrJ<br />
boundary conditions. Hence, they are comp1etel.y independent and<br />
not coupled.<br />
Frc~n (2.17~) follows<br />
@ (z = 4-0) = 0 (2.18;<br />
M<br />
This boundary coilditioll has a serious drawback for the TM-field:<br />
As a result within the conductor no TM-field can be excited by<br />
external sources.<br />
From (2.18) follows via (2.13) fM(o) = 0. Now<br />
K.iultip:lyi.ng (2.9a) by ~ c f and ~ integrating ) ~ over z from 0 Co z,<br />
integra-tion by parts yj-elds in virtue of fM(o) = 0<br />
From (2.20) and (.2.19). fo1.3.oi..ls for yet?]. Er,equencies
- d<br />
IufM12 > 0 .<br />
dz - -<br />
On the other hand f has to tend to a limit f0r.z -t m. Iience,<br />
M<br />
- f (z)<br />
M<br />
0,<br />
-<br />
O
1<br />
2.2 Free modes of decay-<br />
Whereas external fields can exci-te only the TE-mode within the<br />
conductor, there exist free modes of current decay for both TE-<br />
and TM-fields, Consider the typical potential<br />
where @ and f stand for OE, $M and f,, fM, respectively. First it<br />
L<br />
wlll be shown that the decay constants B (ei.genvalues) must be<br />
positive quantities. Inserting (2.23) in (2.9) and (2.10) we ob-<br />
tain<br />
fECz) =. { K ~ - f3~~0(z)l 'fE(z), (2.24)<br />
where K' = K' -1- K'. The ej-genfunctions have -to satisfy the boundary<br />
X Y<br />
conditions<br />
fE = (km) = 0, f (to) = 0, fr(m) = 0<br />
M M<br />
If there is a perfect conductor at finite depth d then<br />
(2.26)<br />
fE(d) = 0, fi(d) = 0 (2.2Ea)<br />
Multiplying (2.24) and (2.25) by f" aid of" and integrating over z'<br />
E M<br />
from -m to +m and 0 to -, respectively, we obtain on integrating<br />
by parts and using (2.26)<br />
Hence, all decay constants f3 are real and positive. The currcnt<br />
flow for TE-decays is in horizontal planes, whereas the TM-decay<br />
currcnts flow predominantly in vertical planes.<br />
Example :<br />
u = O<br />
0 + 0,-<br />
- z - 0
a) TM-mode<br />
From (2.251, (2.26); (2.26a) foll.ous<br />
f (0) = 0 yields fM[z) = sinyz, whereas f' (d) = 0 requires cosyd=O.<br />
M M<br />
Hence,<br />
The unnormalized eigenfunctions are<br />
Theye exists also an electrical (po-ten.tj.al) field outside the con-<br />
ductor. Find it!<br />
b) TE-mode -<br />
From (2.24), (2.251, (2.26~1) follows<br />
"CZ) = K" fd), Z < 0<br />
fE E<br />
2.. .<br />
f;[z)+Y.'f (z)=O, 0 < z < d, y -f3000 - K2 .<br />
E<br />
with f ( -m) = 0, fE, fc continuous across z = 0, f (dl = 0.<br />
E E E<br />
Then f (d) = 0 leads to the eigenvalue condition<br />
E<br />
casyd -I- (a/y)sinyd = 0.<br />
For I: = 0 the ei.genval.u.es zgree with those of 'i'JI, For 1:c1
The TE-decay systems occur as transients to fit the initial con-<br />
ditions, e.g, whcn sn external fie2.d is switched on.<br />
2.3, Calculation of f,Cz) for a layered structure<br />
Let .the half--space collsist of L uniform layers with conductivi-ties<br />
ul., u2, -", u<br />
the upper. edge of layer m being at h (h = 0). Let<br />
L n~ 1<br />
the index 0 reier to the air half-space.<br />
0 =o<br />
hl=O o Then given in z - < 0 the potential of the source<br />
h~-l<br />
we have to solve (2.9a). i.e.<br />
L- l f"(z) E =.{~"tiw~i a(z))fE(z;.<br />
With the abbreviations<br />
and .the understandin g that 11<br />
I* 1<br />
0<br />
- -, we have<br />
In Eq.. (2.33) h > 0 is -the height of the lo!
with<br />
"<br />
the abbreviation<br />
+ - 1 - ?<br />
=<br />
: - 1<br />
- 2' gm 2 exp{i.a m ( 1 1 l m<br />
, m = L-I , . . . , 1 .<br />
Having computed u:, Eqs. (2.32) and (2.33) yield<br />
-<br />
&=f;/Bo.<br />
Thus the field is specified.<br />
E -.<br />
to calculate B: and 6"<br />
m<br />
separately. Instead only the ratio<br />
~f we are interested in f (2) for z < O only, it is not necessary<br />
y m<br />
- +<br />
-- nm / B;<br />
is requi-red. Because of (2.33) ar?d (2.32) f (z) is given by<br />
E<br />
-KZ 4% Z<br />
f (z) = f-{e + y e ), O
2.4. ~articul'ar source fields<br />
Notation: Xn the sequel only fields with a periodic time function<br />
, --<br />
exp(iwt) are considered. For any field quantity .A(s,t)<br />
,+<br />
we write A(r,w) - with the understanding<br />
that the real or imaginary part of A(g,t) exp(iwl;) is<br />
meant. Furthermore, since only the TE-mode is of interest<br />
we shall drop the subscript "E".<br />
In the last section an algorithm for the cal.culation of the $-poten-<br />
tial in a layered half-space has been given. As input occurs<br />
fe(~,K;w) = fi (Krw) exp(-KZ) , tIie representation of - the source - potential.<br />
in t.he wave-nurnber space. Frolu (2.12a) , i .e. - 13=grad (a@/az)<br />
-<br />
follows that - a$/az is the magnetic scalar potential. Between<br />
f- (K, w) and the source potential be (2,~) exist the following reci:-<br />
0 -<br />
- -- 1<br />
2 SS ie(r,w)e dx dy<br />
--. General field<br />
------<br />
- -<br />
If I$ i.s symmetrical. wi.th respect to a vertical axis, i.e. @ is a<br />
(2.35)<br />
function of r = (x" +y2)1'2 only, e4s.. (2.35) and (2.36) simplify t:o<br />
In the derivation of (2.35a) and (2.3Ga) the cartesian coordinates<br />
were replaced by circular. polar coordinates<br />
x = rcos0, y = rsin0, K = KCOS$, K~ = ~siilq<br />
>:<br />
and use was 1nad.e of the identity
2n -i~cr cos (8-$1 -iKrcU<br />
J e ~B=J e c's8d0-2.fcos (~rcos0)ci0=2n~<br />
0 0 0<br />
11<br />
(icr)<br />
0<br />
- 2.<br />
) 1/2<br />
2 1/2<br />
Froin (2.35a), (2.3Ga) follovis that if 4 is function of r= (x2-I-Y<br />
then fo is a function of K = (I:* f K ) . In certain cases (e.cj.<br />
X Y<br />
example h! below), it is on1.y possible to obtain the potential on<br />
the clrlj-ndrical axis (r = o) in closed form. Then (2.35a) reads<br />
Eq. (2.35b) can be considered as a Laplace-Transform for which the<br />
inversion is<br />
where E is. an appropriate real constant (thus that all singill.arj.-<br />
e<br />
ties of 41 (o,z) are at the left of z = E). Simpler wou.ld be the use<br />
of a table of Laplace transforms.<br />
The spectxal representation of the source is now calculated. for<br />
simple sou-rces :<br />
' &) Vertica.:L magnetic dipo1.e ,-<br />
Locate the dipole at r = (o,o,-h), ki > o and let i.ts moment be<br />
-0 -<br />
*<br />
M - = Mz. - When it is produced by means of a small current loop then<br />
!4 = current x area of the loop. ($5 i.s positive if the di.rect:ion of<br />
the current forms with - 2 a right-handed sys::.em, and negative else.)<br />
From the scalar potential<br />
follows<br />
-<br />
Since 4 shows cylindrical symmetry, v7e obtain front (2.3Sa) on using<br />
the resu1.t<br />
I<br />
Ver kical. magnetic dipole
.<br />
b) CLrcular current loop<br />
Let a and I be the radius and the current of the loop. The potential<br />
has cylindrical syr~.nlctry,but can be expressed by means of simp1.e<br />
functi.ons only on the. axis r =: 0. For t.he n~oment we assume that<br />
the current loop is in the plane z = o. For symnetry reasons there<br />
is only a H z component on the axis.<br />
1 Biot-Savart's law yields<br />
whence<br />
- 1a2 A8<br />
' I a h 6 a<br />
c4 . . .<br />
ILZ = -<br />
4n (z2+a 2) 3/2<br />
I a2<br />
z 2 (z2-ka 2)3i2<br />
- .<br />
l\roli xZ = -av/az = + a2+e/az2. Kence integrating twice on using<br />
Vcm) = 0 = ie(m) we obtain<br />
Now we have to solve (2.35b) , i. e.<br />
Looking up in a table of Laplace transforms:<br />
Sf the plane of thc loop is z = -hI 1 - > 0, then a si1npl.e change<br />
of origin yiel-ds<br />
1 Circular current loop I<br />
I<br />
I
In the 1irnl.t a -+ o, I +- m, M = va2 I fixed,this leads to the result,<br />
(2.37) for the vertical ~nagnetic dipole (5 (x) =x/2 -I-<br />
1<br />
0 (x3) ) .<br />
- - %---<br />
C) IIorPzontal maqnetj-c d:ii>ole<br />
A<br />
IIere M = M s and<br />
- --<br />
-e - . . . ,<br />
' ' -- a$ = - Ms<br />
M: ' '<br />
7.- ,$ €2 =<br />
P<br />
a z<br />
41iR3 4n R (R->z-l-11)<br />
With this result, suppressing al.1 further details the eval.uation of<br />
(2.35) vields<br />
d) Line current<br />
--<br />
Current I at y = 0, z = -h. It results the scalar potential<br />
The corresponding source function is<br />
Line current<br />
- J<br />
The delta function. 6 (K ) occurs since there is no dependence on x.<br />
X<br />
6) ' "P1:arie<br />
--- wave"' ' (elementary\indula-Led - 'field)<br />
Assun:e a wavenumher rc = w and let the scalar magnetic source po-<br />
Y<br />
tential be<br />
leading to a magnetic field<br />
- 1 ~ 1 " lIe=;-TJ - 1 ~ 1 1<br />
H~ = H cos (vy) e I si-n (ply) e<br />
Z 0<br />
Y 0<br />
z<br />
sgn(w).
IIence,<br />
- I<br />
---<br />
"e Q -- sin (wy) e- I w ( z<br />
I{o<br />
and -<br />
2 8 ' ~ ~<br />
f--.(~) =<br />
~ ( K ~ )<br />
0 - j.w 1 ~5~ 1<br />
' { ~ C+w) K<br />
Y<br />
- 6 ( ~ -w)} ,<br />
Y<br />
Elementary undulated field<br />
-- -<br />
tghen the source can be considered as elementary harmonic.field, as<br />
in this case, the Fourier integral representation (2.35) cori>plicates<br />
things only.<br />
In the limi-k w + 0 we obtain a uniform source field in y-d.i.rection<br />
For the induction process a strictly uniform source field is useless<br />
since only a vertical magnetic field component induces. Hcnce w<br />
must be non-zero, no matter how small. If we confine our attention<br />
to a finite part of the infinite horizontal. plane, the dimensions<br />
of that part being smaller than l/lw(, then --.- formally we may put<br />
.w = o and can profit from the particularly simple resulting equations.<br />
Such a source field is called a - quasi-uniform - fie1.d. To the<br />
unrealistic uniform field belongs the source potential<br />
which can no longer be represented in terms of (2.35)
-- 2.5. Definition -- of the transfer function C (10, K)<br />
The TE-potentia.1 is<br />
With this potential we define as the basic transfer function for a<br />
half-space with one--dimensional conductivity structure<br />
C<br />
Note that f depends on K and K separately via, the<br />
X Y<br />
function as an amplitude factor. In the ratio f/fl this factor<br />
v,<br />
drops out and it remains only the dependence on K = since<br />
f is a solution of (2.9a), i.e.<br />
How can C be determined from a. knowledge of the surface electro-<br />
magnetic field, at least theoretically?<br />
From (2.8a,h) results<br />
- -, -I-- i~*r - -<br />
- I I = C U ~ (QZ)=IJ{~K~'+K~~Z~~<br />
~ - ~ - - dx dK<br />
- (0 * Y<br />
Let<br />
A +a, - ..- i ,< * r -- -<br />
- H(K,w) - = -- I /I - ~(x,y~~,w)e dxdy<br />
( 2 ~ -m ) ~<br />
- -<br />
A<br />
t-m , -i~c *r<br />
1<br />
- E(_K,w) = /I g(x,y,O,w)e dxdy (2.46b<br />
( 2 ~ -m ) ~<br />
denote the surface fields in the wavenumnber-frequency space. Then<br />
from (2.45a,b), (2.46arb)<br />
* A<br />
- -<br />
' + K = ~ f' K .i- K~ f z<br />
Y
There are several ways to determine C:<br />
a) From the ratio bctween two orthogonal components - of tj>h<br />
- horizonha1 electric and magnetj.~ fisld -<br />
*<br />
Vectorial multiplication of (2.47a) with z and use of (2.43) yields<br />
* A A<br />
E = -iwu C z x H<br />
.- 0 - -<br />
A<br />
or with any horizontal unit vector - e<br />
A A<br />
- -<br />
e - E<br />
C = .. A*- (2.48)<br />
iwyo'(~xg) *' - K .<br />
A * A A<br />
In particular - e = - x and - e L- yields<br />
b) Prom the ratio of vertical and horizontal - l~agne'iic fi~ld<br />
components<br />
(2.47a) yields immediately<br />
c) Froln the ratiq of internal to external -- part of. a lnagnetj-c<br />
horizontal component<br />
i<br />
In z - < 0 f reads,<br />
(2.49a<br />
- -KZ -,- f+<br />
. , f = f e (2.51<br />
0 0<br />
A<br />
A A<br />
Let Xxe and H be the source part and internal part of Hx. Then<br />
xi<br />
from (2.47a) and (2.51)<br />
Defining<br />
A *<br />
s (K, w) = Hxi (5,'" //IS'xe (5, (6)
we arrive at -<br />
v<br />
The same appl-ies to H Y'<br />
. ~<br />
d) Prom the ratio of vertical. gradients 'at' z = 4-0 --- to the<br />
corresponding field compone~lts at z-o<br />
I'/<br />
Let H~(E,~)<br />
be the vertical gradie!.lt of IIx at z = -1-0 in the wave-<br />
number - frequency domain. Then<br />
On applying (2.44) Eq. i2.47a) yields<br />
A<br />
The same applies to I*,,. For applications of (2.55) the surface<br />
1<br />
value of u must be known. C can also be obta.ined from other field<br />
ratios involving vertical. gradients.<br />
The 111et11ods a) and d) are also appl-icable for a quasi-unifozrr!<br />
source fie,ld; b) ancl c) break down in this case.<br />
The apparent resistivity of magnetotellur~cs is defined as<br />
.- A<br />
According to (2.49a,b) its re:.ation to C is<br />
A
2.6. Properties of C(W~K)<br />
a) Signs, limiting values<br />
According to (2.43) the response function C is defined as<br />
where f satisfies<br />
f" (2) =' I K 1- ~ iwu0u (z) 15 (z)<br />
C has the dimension of a lenqtl~.. Let<br />
-i+<br />
C = g - ih or C = [~[e<br />
Then g - > 0, h - > 0 or 0 - c - < s/2<br />
(2. GO)<br />
Proof: Talce the comp1.e~ conjugate of (2.59) , mu1 tip1.y by f and in-<br />
-<br />
tegrate over z. Integration by parts yields<br />
-f(o) frX(0) = I C [fl<br />
m<br />
0<br />
[2-i-(~2--iw~00)<br />
[fl21dz.<br />
Division by 1 f' (0) 1 leads to the result.<br />
The limiting values of C for w -* 0 and w + CQ are<br />
1<br />
f ; tanh (KH) for W+O<br />
I I for LO+ m -<br />
Ji.wuocr (0)<br />
- In (2.62a) H is the depth of a possib1.e perfect conduc-tor. If<br />
absent, then 1-1 i. 'm, C I jk.<br />
Proof: For w=O the solution of (2,59) vanislling at z=X is<br />
f - sinlz~c(H-z), - whence (2.G2a). - For high frequencj.es f tends t(<br />
.---<br />
'the solutio~l for a uniform ha1.f-space, i.e. f -. exp{fidll,~)~ yield:<br />
(2.62b). This limit is attainecl, if the penetration depth for a<br />
uniform halfspace with u - ~'(oJ,<br />
V<br />
is small co~npared with the scale leng-Lh I/K of the external fie12<br />
and the scale l.engt11 lu(o)/ol (0) 1 of corducti-vity variatiol?.
) Computation of ---- C for a laxezed half-space --<br />
When interested in the electric field within a layered structure,<br />
we have to compute a set of coefficients Bf according to the rul.es<br />
m<br />
of Sec. 2.3. These coefficients can be used also to express C:<br />
The latter form is also applicable for K'O, whereas a limiting pro-<br />
cess is involved in the former one in this case. If we are only<br />
interested in C then we may proceed as follows: C can be con-<br />
sidered as a continuous function of depth. Then<br />
-+ 1<br />
where % = - exp{?a d 1,d -<br />
2<br />
Hence, (2.65) and (2.66) yield<br />
- hm+l-l~ ~2 = ~ ~ + i ~<br />
m m m m' 1n o 1%<br />
+ tanh(c, d )<br />
I "mCm+l m in -<br />
m a 1 ~ . u ~ C tanh , ~ - (a ~ d ~ )<br />
m<br />
m m<br />
c = --<br />
Starting with C- = l/a baclcvrard recursion using (2.67) leads to<br />
L L<br />
c) A=rox:i.inate . - interpretation of a - one-dimensional - con+uctivity ---<br />
structure<br />
--<br />
If we can assume K = 0, i.e. scale length of external field large<br />
0 .<br />
compared with penetration depth (as generally done in maqneto-<br />
tellurics) then there exists a simple method tc chtain from<br />
C a first approxi.mation of the underl-ying conductivity structure<br />
(<strong>Schmucker</strong>-I
X X<br />
Let C = g - 5-11. Then a first approximation u (z ) of u(z) is ob-<br />
tained by setting<br />
(2.G8)<br />
This cannot be proved rigorous1.y but the' following arguments are<br />
in favour of it:<br />
1) zX = g can be considered as the depth of the "centre of gravity"<br />
.of the in-phase induced current system.<br />
2) It w ill be shorvn below that z" continuously increases when the<br />
frequency decreases. According to a) its maximum value is 3.<br />
3) For a uniform half-space and K=O we have h = T. I-rence,<br />
X<br />
ax is correct in this case. For perfect conductions u +m since<br />
h+O.<br />
This approximate method performs prrticularly well when there is a<br />
monotone increase in con6uctivi.ty. The following .two figures (P- 24)<br />
illustrate capabilities and limitations of the method.<br />
dl Properties of C in the comul'ex frequency plane<br />
For the following considerations it is useEull to coilsider the fre-<br />
quency w as a comples quan-tity. Then in the coniplex frequency<br />
plane outside the -- positive - imasinar~ .> axis C is an analytical func- --<br />
tion of frequency. - For s proof multiply (2.59) by fX and inteqrate<br />
over z. Then the integration by parts yields<br />
Hence, for w not on the positive imaginary axis f' (0,w) cannot<br />
vanish. There are neither isolated poles nor a dense spectrun of<br />
poles (branch cut).<br />
Division of (2.63) by 1 f ' ( 0,~) 1 yields<br />
We can easily deduce Eroni (2.70) that
The approximate interpretation of C using (2.60). In<br />
the left figure .(monotone increase of conductivity)<br />
the zero order approximation interpretes .the data<br />
already complete1.y. When there is a resistive layer<br />
(left hand) the zero order interpretation needs re-<br />
finement.. In the dott.ed line at the left hand, an<br />
approxj-mate phase of C was used. This approximate<br />
phase has been obtained by differentiation of the<br />
double-logarithmic plot of p a (T) (cf. Eq. 2.771.
AS a consequence of (2.71a,b)Cis also free of zeros outside the<br />
positive imaginary axis.<br />
e) Dispersion rel.ntions .<br />
Because of the analytical properties of C, its.rea:L and imaginary<br />
part are not indepc!ndent functions of frequency. Let wo be a point<br />
in the upper w-plane and L be a closed contour consisting of the<br />
real. axis and a large semi-circle Ln the lower w half-plane.<br />
Then ~wO=w+ic<br />
1 C(u')dwr =<br />
-J<br />
n i L w'-wo<br />
since the integrand.is analytical<br />
in L. Due to (2.6213) the large<br />
semicircle does not contribute w ' -plane<br />
and the contour can he confined<br />
to th.e real axis, tIere &ut wl=x and let DJ~=w+~.E (W real, > 0)<br />
tend to the real axis. T.len<br />
where<br />
1 ' . E<br />
lim - = SIX - w)<br />
r+4-0 " ~~+(x-w)<br />
has been used. f denotes the Cauchy principal value of the integral<br />
1.e.t for real frequeilcies<br />
Here g is an even and h an odd function of frequency, i.e.<br />
g(-w) = ~ ( w ) P h(--w) = -h(~).<br />
This is a consequence of (2.70). lience a separation of (2.72) in<br />
its real a.nd imaginary part yiel-ds
I<br />
~ispersi-on relations I<br />
Dispersion relations of this kind occur in many branches of physics.<br />
They are a direct consequence of the causalLty requirement.<br />
Relations corresponding to (2,73a.,b) exist also for modulus and<br />
phase of C. Due to (2.71a) C is also free of zeros in the lower<br />
frequency halfplane. Hence the function<br />
....<br />
log{ C (w) 1<br />
is analytical there and vanishes for lwl + due to (2.62b). Let<br />
-i+ (w)<br />
C(w) = [ ~ ( w ) l e<br />
(2.74)<br />
and assume w > 0. Then the relation corresponding to (2.73b) is<br />
or introducing the apparent resistrivi.ty from (2.57)<br />
where p = l(~(o). There exists a simple approximate version of<br />
0<br />
C2.75). Integration by parts yields<br />
dlog pa(x)<br />
= - w-xdx<br />
J - ' l o g l ~ ~ l - ,<br />
27r 0 dlog x<br />
-1<br />
or since x loglW - '1 is an integrable function pea1:ed at w=x,<br />
w -1- X<br />
in an approsimate evaluation one may draw out of fhe integral the<br />
term dlogpajdlogx! . Hence<br />
>:=w<br />
where the result
has been used. With T = 2 dlogw=-dlogT. The final result is<br />
W'<br />
Since in general double-logarithmic plots of pa(T) are used, a first<br />
approximation of the phase can immediately be obtained from the<br />
slope of a sounding curve. The degree of accuracy can he asserted'<br />
from two examples given at the<br />
left. True phase in broken<br />
. . lines, approximate phase in<br />
nwe,.' -7<br />
full lines.<br />
The simple approximate method<br />
- - - - -. - -<br />
60' ,..-<br />
4). --__ of inversion described in c)<br />
.<br />
w 80.<br />
'. /'<br />
(<strong>Schmucker</strong>-Kuckes relation)<br />
---.-.wc m.3.<br />
II. ..------,' I,. - ".","rn , combined with the approximation<br />
\<br />
-I<br />
' sis&ivity.<br />
Q. I 1 (2.77) provides an extremely<br />
simple tool to derive a zero<br />
'/ ipL'<br />
,m<br />
Ea<br />
70<br />
!D<br />
100 ,000 IOOo3ar< .-- 1 0,<br />
a '? v-<br />
order app?:oximati.on of the true<br />
conductivity from appa.rent re-<br />
f) Inequalities for the freguencyependence of - C<br />
Cauchys formula is
where L is a positively oriented closed contour enclosing only a<br />
domain where C is analytical and the point w. We choose the par-<br />
ticular contour shown at the left. When the radius of the circle<br />
. .<br />
tends to infinity the circle does<br />
not contribute since C (w) = 0(l/&)<br />
for Iwl+-. Hence the contour can be<br />
confined to both sides of the posi-<br />
tive-imaginary axis. On the right<br />
hand side put w'=ih+~, E >' 0. Then<br />
(2.79) yields<br />
m m<br />
= - 1 Im C(iA+E) dh = a q(?,)dA<br />
liin - r x+iw 71<br />
X+i.w '<br />
E++O 0 0<br />
1<br />
where q(h) = - lim - Im C(iA-*E) > 0<br />
1T<br />
E"+O<br />
in virtue of (2.71h) . Summarizing:<br />
The non-negativity of q(X) has the consequence that C must he a<br />
smooth function of frequency. Again let w be a positive frequency<br />
and 1.et<br />
- C=g-'ih.<br />
DeZining<br />
Df: = w -- df - df - d f<br />
dw dlogw dl.0gT<br />
-<br />
- - --<br />
then the fo1lowLn.g constraints apply
(2.82a1b) simply results when (2.80) is split into real and. imagL- ,<br />
nary; it has already been given above (Eq. (2.61 a,b) ) . (2.83) is<br />
proved as follows<br />
The other constraints are proved in a similar way. There are other<br />
constraints involving second and higher derivatives. In terms of<br />
apparent resistivity and phase $ Rqs. (2,83b1a) read:<br />
The slope of a double-logarithmically plotted sounding curve is<br />
- Dpa/pa. As a consequence of (2.85a,b) we have alvrays<br />
The monotone decrease of the real part of C with frequency is a<br />
consequence of<br />
The following figure shows data (full lines) which are inconsistent<br />
a<br />
on the basis of one-dimensional model., since the constraints (2.83a, b) :<br />
are partly-violated. Then the least corrections to the data are<br />
determined that the inequal-ities are satisfied. Since this is only<br />
a necessa-ry coridition, interpretability is not yet granted.
1 2 3 4 CPD 1 2 3 4 CPD<br />
-<br />
g) Dependence of interpretation - on wave-ru~S7er<br />
The fundamental equation is<br />
By the transformations - 1<br />
z = - tanh(~z)<br />
K<br />
it is transformed into<br />
in such a way that<br />
remains unchanged. Hence any C can first be interpreted by a uni-<br />
- -<br />
form external field (K=o) and the result @(z) is then transformed<br />
to the true conductivity by<br />
- -1 1<br />
o(z) = sech4 (ez) .a (- tan11 (KZ) )<br />
K (2.87)<br />
-
For this interpretation the condition<br />
1<br />
C(O) ( K<br />
has to be satisfied on C (w) .<br />
An application of (2.87) is given in the following figure*<br />
When K increases attenuation is interpreted<br />
by geometrical dampping at the expense of<br />
electromagnetic damping due to a perfect<br />
conductor.<br />
.3... M.o.del. .cal.c.ul.at.ions. for. .tw,o-.dimension.al. structures<br />
.-- ~- .- ~ ~~<br />
-- -<br />
3. I.' General. 'equations<br />
We are considering now induction prbblems, where both the conducti-<br />
vity structure and the inducing field are independent of one hori-<br />
zontal coordinate, say x. Compared with Ch,. 2, the class of in-<br />
ducing fields has become more restricted, hut the class of ad-<br />
mitted conductivity structures has been enlarged.<br />
For x-independence, Maxwell's equations<br />
are split into two disjoint sets
The TE-mode has no vertical electric field, ?he TM-mode has no ver-<br />
tical magnetic field. In the treatment of the;+? modes the use of<br />
- -<br />
electromagnetic potentials is not necessary since E x and H x can serve<br />
as pertinent potentials. For conciseness let<br />
- -<br />
H: = Hx I E: = Ex (3.4b,a!<br />
Then E and H satisfy the equations ,<br />
In uniform domains both equations. agree. Eq. (3.5b) resembles the<br />
equation of heat conduction in a non-unifor15 heat conductor.<br />
The continuity of the tangential electric and magnetic field components<br />
at conductivity discontinuities leads to the conditions<br />
-- ..<br />
I -1<br />
E, -- continuous :<br />
an<br />
TE<br />
(3.6a)<br />
H, - 1 - aH<br />
continuous : TM<br />
a -<br />
(3.613)<br />
an<br />
- a is the derivative in direction to the normal of the discontinuity.<br />
an<br />
The E- and H-polarization - shores very different patterns. From (3.lb),<br />
(3.2b) follows that Ex is constant in the air half-space (u=o). Hence<br />
the Tbl-mode admits only a quasi.-uniform inducing magnetic field. In<br />
contrast in the TE-mode any two-dimensional inducing magnetic field<br />
is allowed.<br />
Hence, t17.e source terms to he added on the PJKS of (3. I a, b) are<br />
je(y1z) and Ho 6 (z+h),<br />
-
assuming that the TM magnetic source field j.s due to a uniforrn<br />
sheet current at height z = -h, h > 0. T1lj.s assun~ptj-on, however,<br />
is immaterial for the following.<br />
In the sequel all field quantities are split into a normal and<br />
anoinalous part, denoted by the subscripts "nu and "a" , respectivel-y.<br />
The normal part refers to a one-dimensional conductivity structure.<br />
Let<br />
u(y,zl = un(z) + U,(Y,Z) (3.7)<br />
H(y,z) = Hn(z) + Ha(y,z)<br />
E and H are defined as solutions of the equations<br />
n n<br />
(3.3b)<br />
e<br />
AEn = k2 E + j (3.10a)<br />
n 11<br />
d l d<br />
-(- - 1-1 ) = H z > Or H (0) =<br />
dz ,?2 dz n n ' - n<br />
vanishing for z + m.<br />
In virtue of (3.5a,b), (3.9arb), and (3.103,b) Ea and H satisfy<br />
a<br />
1 d 1. 1 dIln<br />
div (- gradH ) = H a - - - , z > 0<br />
k<br />
a<br />
k * k2 dz -<br />
n<br />
If the anomalous domain is of fin2te extent, E has to vanish uni-<br />
a<br />
forinly at infinity. Under the same condition Ha has to vanish uni-<br />
formly in the lower half-space. At z=o H is zero.<br />
a<br />
If the anomalous domain is of infi.nite extent in horizontal direc-<br />
tion, we can demand only that Ear H +O for z + m.<br />
a<br />
For a numerical solution of (3.11 a) the following three clioices of<br />
a basic dornain are possible (boundaries hatched).<br />
In approach A, (3.11 a) is solved by Finite differences subject to<br />
the boundary condition Ea=O or better subject to an inpedance<br />
boundary collclition (below). In approach B (3.11a) is solved by<br />
finite differences only in the anomal.ous slab. At the ho~izontal<br />
boundaries boundary conditions involicity the normal structure<br />
above and below the slab are applied. I approach C (3.11~~) is re-<br />
duced to an integral equation over the anon~alous domain. These<br />
approaches will now be discussed in details.
i<br />
I Earth<br />
Anomalous<br />
domain<br />
3.2. Air half-space and conductor. as basic domain<br />
(Finite difference method)<br />
For the TE- and TM-mode we have to solve the differential equa-<br />
tions (3.Sa,b), i.e.<br />
with the boundary condltion that the differences E =E-E and<br />
a n<br />
Ha=H-HI> vanish' at infinity. E has to be computed for any given<br />
n<br />
two-dimensional external source field al.ong the lines of Sec.2.3.<br />
The H -field belongs to a uniform external magnetic field.<br />
n<br />
In the finite difference method, the differential operators in<br />
(3.12a,b) are reduced to finite differences. For simplicity a<br />
d<br />
square grid.with grid ~11th h is assumed. Consider the following<br />
configuration of a nodal point 0 and its four neighbours:
Then the evaluation of (3.12a) in uniform subdomains yields<br />
or<br />
E 4-E CE CE - 4Eo = ll2k2E0<br />
1 2 3 4<br />
Within uniform subdomains the same formula qplies to H. Differences<br />
occur if the nodal point 0 is at an interface. Consider as exarrple<br />
a vertical discontinuity:<br />
li<br />
~egio'. I. Region 2<br />
In the absence of region 2 one can write a central difference equa-<br />
tion for E at nodal point O as<br />
where the bracketed superscript indicates the region. In the ab--<br />
sence of region 1 the central difference equation at O is<br />
At the vertical discontinuity we have because of the continuity of<br />
Since also the normal gradient of E is continuous,<br />
The field values E2 (2) and E4 (I) are fictitious and have to be elimi-<br />
nated with the aid of (3.16) and (3.17) fro12 (3.14) and (3.15). The<br />
result is<br />
- where E2 = E2<br />
(1<br />
and E4 Z4<br />
(2)
Hence, the conductivity is to be averaged in the TE-case. Now con-<br />
sider the TM-case:<br />
1<br />
The continui.ty of H and 2 dH(dn yields<br />
I , ,<br />
I~(~)=H(~)=II<br />
H;l 1 =KA2) =<br />
H1 -1 0 0 0' -H3<br />
from where we obtain on eliminating 1-1<br />
.2<br />
(1<br />
and H4<br />
Similar formulae hold for a horizontal discontinuity.<br />
The normal values 05 En and Hn are used as starting as initial<br />
values. At the boundaries we have two choices<br />
a) The boundary values are kept fixed, i.e. E = E<br />
n'<br />
b) Or free boundary values are uked as - ieance - boundary condition,<br />
where n is the direction of the -- outward normal.<br />
--<br />
(3.24) is obtained under the assumption that the anomalous fields<br />
diffuses in form of plane waves outwards, a valid approximation only<br />
if the local penetration depth is small compared with the scale<br />
length of coaductivity changes. It performs poorly at edges and in<br />
isolators. However, better then Ea=O.<br />
Eq. (3.24) yields ascondition at the upper edge of the air layer:<br />
aEa(az = 0 Ci.e, constant horizontal magnetic field). '
The iteration is carried out either alonq rows or colums. Generally<br />
the GauO-Seidel iteration procedure is used with a successive over-<br />
relaxation factor to speed up convergence..<br />
3.3. P.nomalous slab as basic domain<br />
In practice it is not necessary to solve the diffusion equation by<br />
finite differences in t.he total conductor and the air half-space.<br />
Instead it is sufficient to treat the equation only in that slab<br />
which contains the anomalous domain.<br />
Let the anomalous slab be confined to the depth range zl 5 z 5 z2.<br />
Within this domain we have to solve the inho3nogeneous equation<br />
(considering for the moment only the TE-case)<br />
subject to two homogeneous boundary conditiolls at z = z and z2,<br />
1<br />
~Thich involve a for z < z and z > z2 respectively and account for<br />
n 1<br />
the vanishing anomalous field for z -t 2 m. Yihen (3.25) is solved<br />
by finite differences, the discretization invol.ves also the field<br />
values one grid point width above and below the anomalous slab.<br />
The idea is to express these values in terms of a line integral<br />
over Ea at z = z and z2 respectively.<br />
I<br />
Let V and V2 be the half-planes z - < zl and z -- > z2, respectively.<br />
~ e G t r [r). 2, r, be Green's functions which satisfy<br />
" "m<br />
(-0 -<br />
subject to the boundary condition<br />
In V and V2, E is a solution of<br />
1 a<br />
Now Green's formula for two-dimensions states that
The RHS i.s a closed line integral bordering the area over which the<br />
LIIS integral is performed. I11 the RHS differentiation in direction<br />
to the outr,?ard normal is involved.<br />
From (3.28) and (3.26) fol.lows<br />
Identi.fying U with GI V with Ea, Eq. (3.29) yields in virtue of<br />
(3.27)<br />
E -<br />
a Q<br />
around Vm<br />
I<br />
r - a an G ( ~ (&IL)ds.<br />
)<br />
Now G'") and its normal derivative vanish at infinity. Hence, only<br />
the part of the line integral along the axis z = z contrihut-.es.<br />
a a m<br />
Por m=? : - a - - m=2 : - a =--<br />
2n az' an az'<br />
Hence,<br />
Because of (3.26) , Eq. (3.30) depends only on the difference y-yo.<br />
Defining<br />
q . (3.30) reads shorter<br />
For a layered structure in Vm, the kernels R (m)' are easily deter-<br />
mined:
Let us first consider the case m = 2. Assume that there are L uni-<br />
------ (3 ------<br />
I<br />
z =o form layers below z = z2 with conz=zl<br />
ductivities (3 I' (3 2~ * s t a1,r and<br />
upper edges at z = hlr h2, ..., hL<br />
' (hl = z2). In applications, the<br />
z=z,=h, vertical grid width z - z2 will be<br />
'. 4<br />
' so small that zo is in the first<br />
z=z<br />
o uniform layer. Then a solution of<br />
?=h - -2<br />
,r (3.26) having the correct singu-<br />
U<br />
2 z=h3 larity is<br />
.,<br />
IIowever, the boundary condition (3.27) is not yet satisfied. and the<br />
normal conductivity structure has not yet been taken into account.<br />
To achieve this let in<br />
m . o: (z-z ) -a (z-z )<br />
z
4.<br />
Having determined B;, the coefficients 6o and 6L are determined<br />
from the fact that the difference between upward (downward)<br />
travelling waves of (3.33a) ar:d (3.33b) at z = zo must be due to<br />
the primary excitation given by (3.32). Hence,<br />
whence<br />
From (3.30~~)<br />
+ --<br />
Since (3.35) involves only the ratio B1(B1, it can be expressed<br />
in terms of the transfer function C'at z = jz2 (cf. (2.64)):<br />
For a uniform half space 13.35) is simply<br />
(2<br />
x . . .~y-y~.,.z~)<br />
1 - -a1 (2-2 ) (zo-z2)k<br />
= ; 1 e O cos~(y-y~)iix= 1 ~ 1<br />
o<br />
. r: - -0<br />
For kl + 0 (isolator) this yields<br />
[ r-r. [. . . . . . .<br />
(kl l_r_-hll
The case m = 1 can be treated in a quite analogue way. If the<br />
d<br />
anomalous slab extens till the surface, i.e. z = 0, the pertinent<br />
1<br />
kernel is easily derived from (3.36a):<br />
(Because of (3.30a) there has been a change of sign. 1<br />
If there are more normal layers (in addition to the air half-space),<br />
the problem is treated as for m = 2, with the air half-space as<br />
+<br />
last (L-th) layer, we have to calculate 6L and hence B; separately.<br />
We can't use C.<br />
The kernels K (m) are nicely peaked functions. The halfwidth is<br />
approximately 2[zm-zo[, i.e. ttrice the vertical grid width. For an<br />
insulator the tails are comparatively long (-l/y2), for a conductor,<br />
an exponential decrease is inferred from (3.36). In general. two<br />
points to the left and the right of the central point will give a<br />
,<br />
satisfactory approximation:<br />
where<br />
h /2<br />
CO<br />
y: 3h /2<br />
'm)=2 i K(~) (u,zo)du, p1 (m)= JY ~(~)(u,z~)du, pp)= K(~)(U,U~<br />
Po<br />
0 3h /2<br />
J2 Y<br />
-<br />
PWi - Pi# C pi = 1, hy = horizontal griqwidth.<br />
(3.38) expresses in any application of the finite difference formu1<br />
the anoma1.ous part of the electric field outside the anomalous slab<br />
in terms of anomalous .field values at the boundary. At the vertical<br />
boundaries the impedance boundary condition (3.24) is applied.<br />
So far only the TE-case has been considered. The TI$-mode can be<br />
handled similarly, taking only the different boundary condition<br />
into account. The approprate formulas can be worked out as an<br />
exercise.<br />
I
3.4. Anomalous region as basic domain<br />
Inteqral equation method<br />
In the integral equation approach Maxwell's equatiol-Gare transformed<br />
into an integral equation for the electric field over the anomalous<br />
domain. With the usual splitting<br />
o = \ + V a' E = E n + E ~ , k2 = k2 n + k2 a'<br />
we have<br />
or subtracting<br />
AE = k2n + iwv j ,<br />
0<br />
(3.39)<br />
Let Gn be Green's function for the normal conductivity structure, i.e.<br />
AG (r Ir) = ki(g)~~(&l~) - &(L - 5)'<br />
n-o-<br />
G (r [r) can be conceived as the electric, field of a unit line<br />
n -o -<br />
.current placed at % and observed at - r.<br />
Multipl-y (3.41) by Gn, (3.42) by E subtract and intesjrate with<br />
a<br />
respect to - r over the whole space: It results<br />
Greent s theorem (3.29) yields<br />
since Ea and Gn vanish at infinity. Hence, introducing into (3.43)<br />
E instead of Ea,we obtain the integral equation<br />
In (3.44), the inhomogerleous tern En can be computed for a given<br />
normal structure and given external field in a well--known way.<br />
It remains to determine the kernel. G 11 (r -a [r). - It satisfies the reci-<br />
prdcdty relation
i.e. source and receiver are interchangeable. For a proof write<br />
(3.42) for G (r 1 rt ) and a corresponding equation for G (r 1 Y:* ) .<br />
n -a - 1 - -<br />
Multiply these equations crosswise by G (r (r' ) and Gn (rolr' ) , subn-tract<br />
and integrate over the full space. Then the result is ob-<br />
tained on using Green's theorem.<br />
For a sol.ution of (3.44) we have to put a line current at each point<br />
r of the anomalous domain and have to compute the resulting elec-<br />
--O<br />
tric field at each point of this domain replacing its anomaious<br />
structure by the normal conductivity. Assume a rectangular anomalous<br />
don~ain wLth NY cells in. y-direction and iqZ cells in z-direction.<br />
Because of horizontal isotropy the field depends in horizontal d.irec-<br />
tion only on the distance between source axid receiver. Hence, we<br />
need field values only f0r.N.Y horizontal distances. Due to the<br />
layering there is no isotropy in vertical direction. Here we have<br />
to put the line current into the center of each cell. Because of<br />
reciprocity the sorresponding field v.alues have to be calculated<br />
in and below the depth of the source. Hence, for the kernel Gn a<br />
1<br />
total of NY - TNZa(NZ+l) field values has to be calculated.<br />
A second set of kernels is required which transform on using (3.44)<br />
the electrical field in the anomalous domain into the electromagne-<br />
tic surface field for all three components E xr H yr H z . The number of<br />
required kernel data depends on the range where the field is to be<br />
ev~luated.<br />
It remains to calculate G (r ir) for a layered structure. Assume L<br />
n -a -<br />
layers with conductivities oo = 0, al, ...r csL and upper edges<br />
h, ~ 2 ' "'I hL' hL+, - O0.<br />
-<br />
Let the source and observation point be placed in the m-th and p-<br />
layer, respectively, and let in the m-th layer<br />
+ +<br />
6 A-f-(z),<br />
omm<br />
z - < z<br />
0<br />
+ + 6 F3-f (Z),<br />
. Lmm Z > Z - 0
-<br />
6 and 6= can be so adjusted that A: = BL = 1. Since there are no<br />
0<br />
sources in z - < 0 and in z - > zo if z is in the L-th layer,<br />
-<br />
0<br />
= B' = 0. With these starting values the continuity of G and<br />
A. L n<br />
aG /az across interfaces yields the forward and backward recurrence<br />
n<br />
relations<br />
- -<br />
+ + + - 4 -<br />
B-=(l+u /am)qm<br />
m mi- I<br />
Bm,~l+(li-am+l/amlgm Bmtl, m = L-I, ..., 1-1<br />
with<br />
+ +<br />
g-=I /2, g-= (I /2) expf'am (hmtl<br />
0 m<br />
-11 )If m = I, ...f L-I.<br />
In the case )J = L 110 recurrence is required for the B-terms. The<br />
coefficients 60 and 6L are determined from the fact that in (3.46)<br />
the difference in the upward (downward) travelling waves for z -- > zo<br />
and z - < z must be due to the primary excitation given by<br />
0<br />
+ +<br />
where f- = f-(zo). The nominator (including a ) is a Wronskian of<br />
?J u Fr<br />
the differential equation<br />
which is a constant thus ensuring reciprocity. (Proof?)<br />
In applications the anomalous domain is split into rectangul-ar cells,<br />
the electric field is assumed to be constant wlthin each cell. Then<br />
C3.44) reduces to a system of linear equations, which is easily<br />
solved because of its dominant diagonal due to the logarithmic<br />
singularity of the kernels. Either direct elimination or GauB-Seidel<br />
iteration can be applied, the latter being in general quickly<br />
convergent. When E is assumed to be constant withi-n each cell., the<br />
integration over the kernel is easily effected by adding in (3.46)<br />
the factor<br />
4sin (Xhy/2) sinh(a ,h /2) / (XU<br />
* , . 1- z 11 '<br />
_I. r. . . " , _
The integral equation method for the H-polarization case looks<br />
sligh-tly nlore complicated. This is related to the fact that even<br />
for three-dimensional structures Maxiqell's equations l.oo?c simp1.e<br />
when formulated for - E, whereas in the 2-formulation additional gra-<br />
dien'ts of the conductivity arise:<br />
The pertinent equa.tion for H-polarization is (3.5) , i. e.<br />
with the usual splitting<br />
1<br />
div (- gradB) = iwpoH<br />
u<br />
and the additional definitions<br />
the equations for the normal and anomalous part are<br />
and<br />
d l d<br />
-(--- H ) = iwll H , H (o) = H~<br />
dz an dz n o n n ,<br />
1 --<br />
div(pll gradHal = iwpoHa - div(pa gradH)<br />
This equation corresponds to (3.41) for the E-polarization case.<br />
Green's function appropriate to (3.51) is defined as<br />
Physically G can be interpreted as the magnetic field due to an<br />
11<br />
infinite straight line of oscillating magnetic dipoles along the<br />
x-axis.<br />
(3.48)<br />
Multiply (3.51) by Gn and (3.52) by IIa, subtract and integrate over<br />
the whole space. Then
From the generalized Green's theorem<br />
= I ${u - av au<br />
a 11 - v -<br />
an Ids<br />
follows that the LHS of (3.53) vanishes. Applying (3.54) also to<br />
the KHS of (3.53) we obtain, introducing H =<br />
a<br />
N - Hn:<br />
r 7<br />
This is an integral equation for H . It call be cast in a slightly<br />
a<br />
different form, which is particularly suitable for applications<br />
since the high degree of singularity due to a two-fold differentiation<br />
of Gn at - r = r is removed. Because of<br />
-a<br />
div(p a grad~~)=-{iwp~G~-6 (r-r ) 3+pllgradGnegrad (pa/",)<br />
"n - -0<br />
Eq. (3.55) reads alternatively ,<br />
The kernel. of the integral equation consists of two parts. The first<br />
part takes account of the changing concentration of current lines<br />
in anomalous domains. It is a volume effect. The second effect<br />
results from the bending of current lines where conductivity changer<br />
It is essentially a surface effect.<br />
In the case of discontinuous changes in conductivity, which is the<br />
most common assumption, (3.56) needs a slight modification. Assul~c<br />
that the anomalous domain consists of rectangular cells, where the<br />
conductivity is allowed to differ from cell to cell. H is assumed<br />
to be constant in each cell.
Then the discontinuity between cell (i, j) and 1 j contribu-tes<br />
to the integral<br />
It has been used that H and p E are continuous across interfaces.<br />
an<br />
The contributioll frcm the i + i j + I interface is<br />
The integral equation is decomposed into a set of linear equations<br />
for the Ha-values at each cell. The electric fLeld is then obtained<br />
by differentiating (3.56) with respect to the coordinates of E ~ .<br />
-<br />
It remains to calculate Gn. At z = 0 G has to sat.i.sfy the boundary<br />
n<br />
condition of Ha, i.e. Ha = 0, : Gn = 0. The boundary condi.tions at<br />
interfaces are G and(l/cr)a~~/az continuous. Assume aqain L uniform<br />
n<br />
layers with conductivities ol, 02, . . . , and upper edges at<br />
OL<br />
hl = 0, h2, ...I hLI h -<br />
~4.1<br />
- -'.<br />
Let the field in the m-th layer be<br />
where
and y can again be so adjusted that C; = DL = 1. = 0 for<br />
Yo L - Gn +<br />
z = 0 requires then C = - I, and Gn + 0 for z + m demands DL = 0<br />
1<br />
Starting with these initial values, the boundary conditions yield<br />
the forward and backward recurrence relations<br />
where<br />
c 1<br />
@ = a /a and gn, = - exp{?a -h 1)<br />
~n in m 2<br />
m h m<br />
p is the source layer. There is no recurrence required. for the C-<br />
term if u = 1 and for the D-terms if y = L.<br />
The free factors follow again frorn the source.represcntation<br />
which must account for the difference for upward and downward<br />
travel-ling waves at z = z . Hence,<br />
0<br />
With the present determination of the field Gn satisfies again<br />
the reciprocity relation G .(r Ir) = G (r 1 r ) . (Proof?)<br />
1 -0 n - -0<br />
-
4. Model calculations. for tl1.r,ee--d~nei~sioi~a1. structures<br />
- -<br />
4.1. Introduction -<br />
In the three-dimensional case the TE- and TM-mode become mixed and<br />
cannot longer be treated separately. Now the differential equation<br />
for a vector field instead of a scalar field is to be solved. In<br />
numerical solutions questions of storage and computer time become<br />
important. Assume as example that in approach A a basic domain with<br />
20 cells in each direction is chosen. In this case, only the storage<br />
of the electric field vector would require 48000 locations. For an<br />
iterative improvement of one field component at least 0.0005 sec<br />
are needed for each cell. This yields 12 sec for a complete itera-<br />
tion, and 20 min for 100 iterations. This appears to be the ].east<br />
time required for this model. Hence methods for a reduction of com-<br />
puter time and.storage are particularly appreciated in this case.<br />
The equation to be solved is<br />
cur1213 - (r) - -t k2 (r) E (r) = - i ~ l(g) i ~ ~ ~<br />
---<br />
where k2.(r) = i~p~c(g).<br />
-<br />
1 (r) is the source current density.<br />
-e -<br />
After the splitting<br />
where E is that solution of<br />
-11<br />
curlZE (r) + k2(r)F, (r) = -impo&<br />
-11 - n--n-<br />
(4.3)<br />
which vanishes at infinity, we obtain for the anomalous field the<br />
two alternative equations<br />
c ~ r l . + ~ k2E ~ = - k2E<br />
-a n-a a<br />
Eq. (4.4a) is the starting point for the volume integral or integral<br />
equation approach, Eq. (4.4b) is the point or-' starting for the sur-<br />
face integral approach.
4.2. In3ral - --- equation method<br />
Let -1 G.(r --a [r), -- i = 1,2,3 be a solution of<br />
cur12c. .-I (r --0 lr)+k2 - n (r)G. - --I (r --0 Ir)<br />
,.<br />
A<br />
- = zi6(g-%), (4.5)<br />
vanishing at infinity. Here, the x. are unit vectors along the<br />
A h -l h A A A<br />
cartesian coordinate axes: x =x, x =<br />
-1 - -2 1, z3 = - 2. 14u'iul&31y (4.5) by<br />
E (r) and (4.4a) by G.<br />
-a - (r I?:) and integrate the difference with re-<br />
-1 -0 -<br />
spect to - r over the whole space. Green's vector theorem<br />
where d-r i.s a volume element, dA a surface element and - 1 the out-<br />
ward normal vector, yields<br />
since and G vanish at infinity. After combining all three compo--<br />
-i<br />
nents and introducing - E instead of E , the vector integral equation<br />
-a<br />
1 -gtg)d~ (4.7)<br />
--<br />
is obtained. Here is Green's tensor being defined as<br />
wj(rJ<br />
% --O<br />
3 3 A h<br />
A<br />
t(r Ir)= C x.G. (r [r)= Z G.. (r .Ir)x: x.<br />
- -1-1 -0 - ir j=1<br />
13 -0 -- -1 -3<br />
i= I<br />
(using dyadic notation). The tensor elements G.. admit a simple<br />
= 7<br />
physical interpretation: Gij (r _o [r) - is the j-th electric field coin-<br />
ponent of an oscillating electric di-pole of unit moment pointlng<br />
in x.-direction, placed in the normal conductivity structure at &;<br />
1 --<br />
the point of observation is - r. Note that the first subscript and<br />
argument refer to the source, the second subscript and argument to<br />
the receiver. Because of the fundamental reci.p~:ocity in electro-<br />
mag~~etl.sm, source and observer parameters axe interchangeable, i.e.<br />
For a proof replace in (4.5) - r by - r', write an analogous equation<br />
for G.(rlrl), multiply cross-wise by G and Gi, integrate the<br />
-3 - - - j -<br />
difference wit11 respect to - r' over the whole space, and obtain (4.9)<br />
on using (4 .G) . Due to (4.9) , the equation (4.7) is alternatj.vely<br />
written
Eq. (4.7) lnvolves integration over the coordinates of the receiver,<br />
(4.10) requires integration over the coordinates of the source.<br />
The kernel G and the inhomogeneous term Ell of the integral equation<br />
(4.7) or (4.10) depend only on the normal. conductivity structure.<br />
To determine the kernel 9 replace first the conductivity within<br />
the anomalous domain by its normal val-ues. Then place at each point<br />
of.the domain successively two n~utually perpendicular horizontal<br />
and one verti-cal dipole and calculate the resulting vector fields<br />
at each point of this domain. At a first glance the work involved<br />
appears to be prohibitive, but it is sharply reduced by the reci-<br />
prociky (4.9) and the isotropy of the normal conductor in horizon-<br />
tal di.rection. Because of (4.9) , from the elements of Green's<br />
tensor<br />
G G G<br />
xx xy xz<br />
the three elements Gxz , G xy' Gyz need not to be calculated when<br />
G G G is computed. From the remaining six elements G has<br />
zy' yxr zy Y Y<br />
the same structure as G xx' only rotated through 90'. The same re-<br />
lation holds between G and GZx. Hence, there are only the four<br />
z Y<br />
independent elements G G G (say). The particular<br />
XX' yx' GZxr zz<br />
symmetry of the Gxxr Gyx, Gzz-conponent in connection with the re-<br />
ciprocity (4.9) then shows that these components need to be eva1ua.-<br />
ted only for points of observation above source points. Consider<br />
for example a vertical dipole at xo=yo=O, zo. Then (4.9) yields<br />
Because of the isotropy of the conductor in horizontal di.recti.on<br />
(4.11 ) is alternatively written<br />
GZZ (O,O,zOI~,y,~) = GZZ (O,O,Z~-X~-Y~Z~).<br />
Now, GZZ has circular symmetry around the z-axis. Hence,
The element GZx is needed for all z and zo. Assume a rectangular<br />
anomal.ous domain, which is decomposed in-to cells with quadratic<br />
horizontal cross-section. It is sufficient to pose the dipoles in<br />
one corner of the ano&aly in the (x,y)-plane. Then assuming NX, NY,<br />
NZ ce1.l.s in x,y,z-direction the total number of required kernels is<br />
There is still a reduction up to a factor 2 possible, if instead<br />
of the corresponding conponents only some auxiliary functions de-<br />
pending only on the horizontal distance from the sourced are cad-<br />
culated .<br />
The corresponding kernels are inost easily computed using a sepalza-<br />
tion into TE- and Tbl-fields as done ,in Section 1.<br />
Let the normal structure consist of L uniform layers with conduc-<br />
tivities 5 , m = 1, ..., L with upper edges at z = h m = I, ..., L<br />
m in<br />
(h = 0).<br />
1<br />
The Green vector GL satisfies in the m-th layer the equation<br />
tie try a solution in the form<br />
where @ corresponds to the TM-potential and Qi to the TE-potential.<br />
i<br />
According to Sec. 2.1, at horizontal interfaces @ and @ satisfy the<br />
continuity conditions<br />
I<br />
a @<br />
041 =I<br />
I<br />
a$<br />
'4, continuous<br />
There is no coupling between $J and @ across boundari-es. Within<br />
uniform layers, but outside sources 4 and $ satisfy identical<br />
differential equations<br />
-*. 7-'<br />
The behaviour of 4 and $ near source p0i.nt.s can be obtained from<br />
. .<br />
the particular forms, which these functions show in a uniform whole-<br />
space :<br />
-kR<br />
.(r .[r) = (k28ij-a2 jaxiay le<br />
. Gij -a -<br />
'j 4nr\k2<br />
,<br />
~=lr-rl - -a<br />
(4.15)
For a dipole - in z-direction this is equal to<br />
a e-kR A e -kR<br />
G = (k" - - grad-) = - curl2 (z --<br />
4a~k'<br />
since<br />
-kR<br />
A (e-kR/@~r~) ) = lc2e / (4aR) -<br />
-z a~ 4Ta2 - 1 I<br />
Comparison of (4.16) and (4.12) shows that for a whole space<br />
The absence of a TE-potential is clear from physical reasons, since<br />
the magnetic field of a vertical. dipole must be confined to hori-<br />
zontal planes (i.e. no vertical. magnetic field, which can only<br />
be produced by a TE-field).<br />
Using Somrnerfeld's integral, C4.17) is written<br />
The f5.el.d of a horizontal electric dipole (in x-direction, say) has<br />
both an electric and magnetic component in z-direction. Hence, a<br />
TM- and TE-potential are needed. 9 +<br />
Since ,- J%i.<br />
a2 a2 1<br />
a2 e<br />
--kl?. m<br />
-a/ z-z ]<br />
G =-(- + -7) 4Jx=--y - -=-- I I A2e O J~ (Xr)dX cos$<br />
Xz ax2 ay 41~k axaz R 4ak20<br />
and<br />
we have<br />
and from<br />
then follows<br />
xsiqn(z-z )<br />
0<br />
J, (Xr)dXcos$ * sign (z-zo) (4.19)<br />
f
With this knowledge of the behaviour of $,, @,, QX in the uniform<br />
whole-space, these functions for a layered medium can be easily<br />
obtained.<br />
Let in the m-th layer hm < z < hmt1<br />
where<br />
where<br />
+ -1-<br />
~ ~ A i f i z - < z0<br />
+ -I-<br />
~ ~ B I ~ f i > z o<br />
z -<br />
+<br />
and f-. = (z-h )Il a2 = h2-tk2. Then starting with<br />
n~<br />
_<br />
in n~ m m<br />
- t - + - + -<br />
+ 1, A =O, B =Or BL=l1 C =I, C=O, D =01 D =I.<br />
ko 0 L 0 0 JJ 1,<br />
The boundary conditions(4.13) lead to the recurrence relations
+<br />
m )}(XI g-<br />
. .<br />
where gi = e~pI+-n~(h~+~-h o = I f2 and y is the layer of the<br />
sou-rce. In the case y = L there is no recurrence'required for the<br />
B and D terms. The free factors are determined in the usual way<br />
simply by ccn~paring upwards and dowriwards travelling \,raves at<br />
z -- z taking the source terms (4.18) - (4.20) into accoun-t.<br />
0 +<br />
Dropping the subscript y on , k2 8; I Bi, C' , D', fi for<br />
Fr U' V P U V<br />
conciseness, we obtain from t4,21) and (4.18)<br />
whence<br />
+<br />
where f- meacs ft (zo) . From (4.22) and (4.20) follows<br />
v<br />
whence<br />
Because of (4.19) +x is discontinuous across z = z o . Hence
After having determined Qzr VJ,; @,, Green's tensor G is obtained<br />
from (4.12). Al.so required is the magnetic field in z - < 0. Only<br />
the TE-part of a horizontal dipole' contributes.<br />
Let P (r lr), i = lr2 be the magnetic field at r due to a dipole<br />
-i -0 - -<br />
in x -direction at r . Then in z < 0<br />
i --o -<br />
0<br />
- 1?$) = curl2 (z*.) " 0 = grad a*i .<br />
0 -3. - 1 az<br />
Now the kernels for the integral equation (4.7) are determined a.nd<br />
it remains to discuss how this equation is sol-ved in practice. The<br />
simplest way certainly is to decompose the anomalous d0lr.ai.n into<br />
a set of N rectangular cells and to assume that the el-ectric field<br />
is constant within each cell. Then there results from (4.7) a set<br />
of 31V linear equations for 3N unknown$. The system has the form<br />
where - x is the vector of unknown field components, & the matrix of<br />
coefficients and 5 the given vector of the normal electric field<br />
values. A direct inversion of this matrix is only feasible for<br />
N - < 50 (say), because of the large amount of storage required. Hence<br />
iterative methods must be used in general.<br />
The simpl.est way would be to start an iterative procedure with<br />
x0 = - -. q. This sequence of approximations, however, converyes only<br />
if the eigenvalue of =z5 A with the largest modulus has a modulus less<br />
than 1, a condition which is certainly not satisfied if cr is large.<br />
a .<br />
Better convergence properties shows the GauB-Seidel iterative scheme,<br />
using during iteration already the updated approximation and a<br />
successive overrelaxation factor:<br />
Let the components of - & be aik. Then (4.28) reads
Then a' new approximation to xi is obtained by<br />
1 1<br />
i- I<br />
3N -<br />
new- old,- -- R<br />
new + c aikxk old - x old<br />
xi -x i 1-a. { C<br />
li k=l<br />
aikxk<br />
k=i<br />
i +gi}<br />
The successive overrelaxation factor R - > 1 has to be chosen suitably.<br />
In many cases R = I (i.e. no overrelaxation) is already a good<br />
choice. If the GauB-Seidel method does not converge then one can<br />
apply a s-pectruin displacement technique: As already mentioned,<br />
simple iteration without updating is only convergent, if the largest<br />
modulus of the eigenvalues is smaller than 1. Eq. (4.28) is equi-<br />
valent to .<br />
- - x = (A - al)x - - + ax<br />
- + 5<br />
(4.29)<br />
The iterative procedure (4.30) will be convexgent, if a can be<br />
chosen in such way that the eigenvalues of - B are of 1aodu1.u~ iess<br />
than 1. If A is an eigenvalue of - A then A - 1 is an eigenvalue<br />
of B. - Consider for example the following situation that - A<br />
has negative real eigenvalues from 0 to Amax, lXrnax 1 > 1. Then a<br />
choice of a = - IXmax[/2 is appropriate. The condition<br />
is then satisfied for all eigenvalues A. The requirements which<br />
m has satisfied can be put in that form that a circle around a<br />
has to include all eigenvalues of A - but has to exclude the point<br />
1. The following figure illustrates the case stated above.<br />
This case applies approximately<br />
to the three-dimensional modelli<br />
problem, where at least the<br />
largest eigenvalues are for a<br />
higher conducting insertion to<br />
a good approximati.on negative<br />
real. For a poor conducting in-<br />
sertion the largest eigenvalues<br />
are essentially positive, hut<br />
smaller than 1.
4.3. The surface inteyral approach to the modelling problem<br />
In the surface integral approach, within the anolr[alous slab the<br />
equation (4.4b) , i. e.<br />
cur-2~ + k 2 ' ~ = - k 2 ~<br />
-a .-a a-n<br />
is solved by finite differences or an equivalent method. The re-<br />
quired field values one grid point width above the upper horizontal<br />
boundary at z = z and below the lower boundary at z = zL are<br />
1<br />
expressed as surface integrals in terms of the tangential component<br />
of ga at z = z and z2, respectively.<br />
.<br />
1<br />
~<br />
Le-t V and V2 be the half-spaces z < z and z > z respectively,<br />
1 1<br />
and let S m = 1.2 be the planes z = z . Let C!mi (r r , r E Vm.<br />
mr in -1 -0 - -0<br />
- r E vm U S be a solution of<br />
m<br />
(i=1,2,3; m=1,2) satisfying for - r E S the boundary condition<br />
m<br />
In Vl and V2, Ea is a solution of .<br />
cur12i3 + k2 E = 0<br />
-a n -a<br />
Multiply (4.34) by G!~), (4.32) by E , integrate the difference with<br />
-1 -a<br />
respect to - r over Vm, and obtain on using (4.61, (4.33) and ga + 0<br />
-<br />
for r -t -:<br />
' r E Vm, or in tensor notation<br />
-0'<br />
,<br />
where<br />
E (r) = (-I)~J curl<br />
-a --o<br />
'm<br />
3 ,-.<br />
x.<br />
(m)<br />
curlG. .<br />
-1 -1<br />
A<br />
(r Irl;{z x E (r)}d~<br />
-0- - -a
Eqs. (4.35aIb) admit a representation of the field values outside<br />
the anomalous layer in terms of the boundary values of the con-<br />
tinuous tangential component of E .<br />
-a<br />
A physical interpretation of Green's vector G subject to (4.33)<br />
-i<br />
is as follows: Reflect the normal conductivity structure for<br />
z < z and z > z2 at the planes z = z and z = z and place a unit<br />
1 1 2<br />
dipole in x -direction at r and an image dipole at<br />
i ,. --o " 'm<br />
r' = r + 2 (zm - z0)zI the moment being, the same for the vertical<br />
-0 -0<br />
dipole and the opposite for the two horizonta~l dipoles. Then the<br />
tangential. component of G!") vanishes at z = z and G!~) is a<br />
-1 111' -1<br />
solution of (4.32) for - r E Vm.<br />
Hence, if Vm is a uniform half-space, G (m) is constructed from the<br />
-i<br />
whole-space formula (4.15) . Eq. (4.3'5) then reads :<br />
m<br />
Eaz = (-1) / J? ( ~ {x-x j )E (g)+ (y-yo)Eo<br />
(5) ld~,<br />
ax CIY<br />
'm<br />
where R = [r - - r 1, k2 = iww cr<br />
--O 0<br />
Eqs. (4.36a-c) contain as important subcase the condition at the<br />
air-earth interface (m = 1, zl = 0, k = 0).<br />
0<br />
Because of the limited range of the kernelstin applications of the<br />
surface integral on1y.a small portion of Smm:ust be cocsj.dered.<br />
For Eax and E the contribution of the region nearest to r is<br />
aY -0<br />
most important. Assuming E and E to be cionstant within's sma1.l<br />
ax aY<br />
disc of radius p centered perpendicularly over r the weight from<br />
-0<br />
(4.36a.b) is<br />
where h = lz - zo[ . There is a contributiorr to E onl-y if Ea,, and<br />
m az<br />
E have a gradient along x and y direction kespectively.<br />
aY
F<br />
At the vertical boundaries the condition E = 0<br />
--a<br />
is relaxed to the<br />
impedance boundary condition<br />
A A<br />
kga.t = n x curl zar - n b E =Of<br />
-a<br />
whiere gat is the tangential component and - & the outward normal.<br />
5. Conversion - formulae for a two-dbensional TE-field<br />
5.1. Separation formu1.ae<br />
In this subsection 5.1. it is assumed that the conductivity does<br />
not change in x-direction and that the inducing magnetic field is<br />
in the (y,z)-plane, i.e. the conditions for the E-polarization<br />
case of Sec. 3.1. Then from a knowledge of the H and Hz component<br />
Y<br />
along a profile from -a to +a at z = o it is possible to separate<br />
the magnetic field into its parts of internal and external origin<br />
without having an additional knowledge of the underlying conducti-<br />
vity structure.<br />
The pertinent differential equation is C3.5a), i.e.<br />
which reducing in z - < 0 to<br />
which has the general solution<br />
where A: describes the external and A: the internal part of the fie1<br />
The magnetic field components at z = 0 are (cE. (3.2.3) and (3.3a)):<br />
Let . .<br />
- - - - . .<br />
H =W +H HZ = H i.I.Izi,<br />
Y ye yi' z e
where the subscripts "en and "i" denote the parts of external and<br />
internal origin at z = o.<br />
Further, let the Fourier transform of any one of the above six<br />
- ~ K Y dy.<br />
quantities be<br />
A<br />
1 +m _ H(K) = - J H(y,o)e<br />
2~ -m<br />
Then (5.4a,b) yields<br />
A A A A<br />
H /xZe =.-i<br />
Ye Y 1<br />
sgn (K) , H . /HZi = +isynCK) (5.7alb)<br />
A product between Fourier transforms in the x-domain transforms to<br />
a convolution integral in the y-domain:<br />
Hence we obtain from (5.7aIb1<br />
- . - -.<br />
H =+KxHze, H . = -I< x<br />
Ye Yl Hzi<br />
-. - - .<br />
= - K x H H =+ICxH<br />
'ze ye' z i. .yi<br />
- €*K<br />
Convergence was forced by a factor e . The resulting convolution<br />
integral exists only in the sense of a Cauchy principzl val-ue, i..e.<br />
(5.8a) for example reads explicitly<br />
+m<br />
. -<br />
H (y) = 1 K (~-~)IX~~(TI)~~<br />
Ye -m
The four equations<br />
A A A A<br />
H = i sgn (K) HZe , H = + isign(~) H<br />
Ye Y i z i<br />
A A A A<br />
can easily be solved for H yer Hze' $ir 'zi*<br />
When transformed into the y-domain it results 4<br />
I Two-dimensional separation formulae I<br />
For practisal purpuses these separation formulae are not very con-<br />
venient, since the kernel decays rather slowly requiring a long<br />
profile to determine the internal and external part at a given '<br />
surface point.<br />
5.2. Conversion - formulae for the field components - of a two- -<br />
dimensional TE-fielci at the surface of a one-dimensional structure<br />
For the separation of the magnetic field components no knowledge<br />
of the two-dimensional conductivity structure is required. However,<br />
the conductivity enters if it is attempted to deduce for instance<br />
r<br />
the total ver-tical component of the magnetic field at the surface<br />
from *he corresponding tangential componenk. If the conductivity<br />
structure is one-dimensional, the conversicn between two component.:<br />
can be effected using a convolution integral, where the kernel is<br />
derived from the one-dimensional structure via the trans5er funs-<br />
'?here S([K[,W) is the ratio between internal and external part of<br />
A h<br />
H i.e. H . /H in the frequency wavenuder domain.<br />
Y' Yl yer<br />
From (5.9a-c) the,yarious conversion formuias for the durface com!31<br />
nents can be derived. The following table gives the definition of
the pertinent kernels and their form for two particularly simple<br />
structures. Note that in the case of a vanishing conductor (i.e.<br />
in the examples h -> m, or k + 0) the kerne1.s K- and M have to<br />
agree with the kernels of (5.8a) and (5.8~1, respectively. Tile<br />
function L occurring in the P-kernel of the uniform half-space<br />
2'<br />
model is the modified Struve function of the second order (cf.<br />
Abramowitz and Stegun, p. 498).
6.<br />
Approaches to the inverse problem of electromaqnetic induction<br />
by linearization<br />
6.1. The Backus-Gilbert method<br />
6.1.1. Introduction<br />
The method of Backus and Gilbert is in the first line a method to<br />
estimate the information contents of a given data set; only in the<br />
second line it is a method to solve a linear inverse problem. The<br />
procedure takes into account that observational errors and incom-<br />
plete data reduce the reliability of a solution of an inverse<br />
problem. It is strictly applicable only to linear inverse problems.<br />
Assume that we are going to investigate an "earth model." - m (r),<br />
where m is a scalar quantity which for simplicity depends only on<br />
one coordinate. For the following examples it w ill be chosen as<br />
the distance from the centre of the earth (to be as close aa<br />
possible to the original approach of Rackus and Gilbert). Then in<br />
a linear inverse problem there exist N linear, functionals - ("rules"),<br />
whLch ascribe to m(r) via data kernels G. (r) numbers 7. (m) in the<br />
3. 1<br />
way<br />
a<br />
gi(m) = j m r)Girdr i = I, ..., N. (6.1)<br />
0<br />
The measured values of gi(m) are the N data yi, i = 1, ..., N. The<br />
"gross earth functionals" g. (m) are linear in m, since it is .<br />
1<br />
assumed that the data kernels G.(r) are independent of m. The in-<br />
1<br />
verse problem consists in choosing m(r) in such a way that the<br />
calculated functionals gi agree with the data y The Backus-Gilber.;<br />
i'<br />
method shows how m(r) is constraint by the given data set. Bcfore<br />
going into details let us give an example. Assume that we are<br />
interested Ln the density distribu{:ion of spherically symmetrical<br />
earth, i.e. m(r) = p (r) , and that our data consist in the ?\ass 1.1<br />
and moment of inertia 0. Then<br />
-
6.1.2. The llnear inverse problem<br />
The data may have two defects:<br />
a) insufficient<br />
b) inaccurate<br />
Certainly in the above problem the two data M and 0 are insufficient<br />
to determine the continuous function p(r). Generally, the properties<br />
a) and b) are inherent to all real data sets when a continuous func-<br />
tion is sought. The lack of data smoothes out details and only<br />
some average quantities are available, the observational error in-<br />
troduces statistical uncertainties in the model. (If we were looking<br />
for a descrete model with fewer parameters than data, then incon-<br />
sistency can arise as a third defect.) Because of the lack of data,<br />
instead of m at r we can obta.in only an averaged quantity ,<br />
0<br />
which is still. subject to statistical incertainties due to errors<br />
in the data. Let<br />
a<br />
< m(ro) > = I A(ro[r)m(r)dr, (6.2)<br />
0<br />
where a is subject to<br />
The latter condition ensures that agrees with m, if m is a con-<br />
stant. A(r ( r) is the window, througlil which the real but unknown<br />
0<br />
function m(r) can be seen. It is the - averaqinq - or resolution Punctio -<br />
The more A at ro resembles a &-function the better is the resolution<br />
at ro. Resolvable are only the projections of m(r1 into the<br />
space of the data kernels Gi. The part of m, which is orthogonal<br />
to the data kernels cannot be resolved. Hence, it is reasonable<br />
to write A as a linear combination of the data kernels<br />
where the coefficients a. (ro) have to be determined in such a way<br />
I<br />
that A is as peaked as possible at ro. The aost obvious choice<br />
would be to minimize<br />
subject to (6.3). For'c~rn~utational ease Backus and Gilbert prefer
to minimize the quantity<br />
a<br />
s = 12 / A2(ro[r) (r-ro)'dr.<br />
0<br />
(6.6)<br />
Since (r-r )2 is small near rot A can be large there. The factor 12<br />
0<br />
is chosen for the fact that if A is a box-car function of width L<br />
1, 0, else<br />
then s = L, i.e. if A is a peaked.function then s in the definition<br />
of (6.6) gives approximately the width of the peak. s(r0) is called<br />
the spread at ro.<br />
We have also taken into account that our measurements y have ob-<br />
i<br />
servational errors Ay i.e.<br />
it<br />
Insertion of (6.4) into (6.2) using (6.1) yields<br />
From (6.7) and (6.8)<br />
and the mean variance is<br />
N<br />
is the covariance matrix of the data errors, which in general is<br />
assumed to be diagonal.<br />
Of course, we would appreciate if the error of m(ro), i.e. E' would<br />
be very small. But also the spread (6.6)<br />
a . N<br />
s = 12 / A' (rolr) (r-ro) ' = C ai (ro) %(rO)s. (rO)<br />
lk<br />
o i., k=l<br />
with a ~ ~<br />
SikCro) = 12 / Gi(r)G -(r) (r-r )'dr<br />
0 k 0<br />
(6.11)
should be small. Hence it required to minimize simultaneously the<br />
quadratic forms<br />
N N<br />
s= C ai% sik 7 c2= C a a E .<br />
i k lk<br />
i,k=l<br />
i , k=l<br />
subject to the condition (6.3), i-e.<br />
There does not ex'ist a set of a. which minimizes s and E' separately.<br />
L<br />
As a compromise only a combination<br />
can be minimized. In (6.15) , c is an arbitrary positive scaling<br />
factor which accounts for the different dimensions of s and E' and<br />
W is a parameter<br />
which weighs the particular importance of s and E ~ For . W == 1 the<br />
spread is minimized without regarding the error of the spatially<br />
averaged quantity . Conversely .for W = 0 the spread s is<br />
large and the error E' is a minimum. Hence, in general there is a<br />
trade-off between resoluti.on - and accurac~, which for a particular<br />
is shown in the following figure.<br />
ro<br />
EZ max<br />
E'<br />
2 -- - -L ---min<br />
I<br />
W=O<br />
I --<br />
s<br />
S<br />
min max<br />
Near s = s . the trade-off curve i.s rather steep. Hence, a small<br />
m m<br />
sacrifice of resolving power will largely reduce the error of the<br />
.average . This uncertainty relation between resolution and<br />
accuracy is the central point 05 the Backus-Gilbert procedure.<br />
, S
It remains to show a way to minimize Q, subject to (6.15). The way,<br />
however, is well-known. One simply intrsduces as (N+1 )-st unknown<br />
a Lagrangian parameter X and minimizes the quantity<br />
a<br />
Q' = Q + h (Cai~li - 1 ) , ui = / Gi (r)dr (6.16)<br />
0<br />
The differentiation of (6.16) with respect to the (N4-1) unknowns<br />
yields the (N+1) linear equations<br />
where Qik = W Sik + (l-W) c Eik.<br />
This system of equations is easily solved. The meaning of X is<br />
revealed by multiplying (6.17a) by ak, adding and using (6.17b)<br />
and (6.15) . It results<br />
In the case W = 1, we have X = -2s. With a knowledge of a (roll<br />
i<br />
is obtained from (6.8) with gi = y. and E' from (6.9).<br />
3-<br />
When is inserted in (6.1) instead of m(r) , it w ill in<br />
general not exactly reproduce the data.<br />
The minimization (6.15)<br />
N<br />
subject to the constraint C aiui = 1 admits for two data (N:=2)<br />
i=1<br />
a simple geometrical interpretation: For constant s and E' these<br />
positive definite quantities are represented by ellipses, the con-<br />
straint is a line in the (al,a2)-plane. For uncorrelated errors,<br />
the principal axes of E~ are the al and a2 axis.<br />
hT1a W varies from 0 to 1 the combinatiomof (al ,a2) on the fat<br />
line are obtained. s and. E' are determined from the ellipses<br />
throughthese points,Sj-nce all s-(~~)-elli~ses are similar, s and^<br />
2
are proportional to the long axes of these ellipses.<br />
6.1.3. The nonlinear inverse problem<br />
The Backus-Gilbert procedure applies only to linear inverse<br />
problems, where according to<br />
a<br />
the gross earth functionals gi have the property that<br />
This means for instance that the data are bui1.t up in an aclditive<br />
way from different parts of the model, i.e., that there is no<br />
coupling between these parts. This certainly does not hold for<br />
electromagnetic inverse problem, where each part of the conductor<br />
is coupled with all other parts.<br />
In nonlinear problems the data kernel Gi(r) will depend on m. Here<br />
it is in general possible to replace (6.18) by<br />
.<br />
a<br />
gi.(mr)-gi~m)=f(m'(r)-m(r))~i(x,m)dr + ~(m'-m)~, (6.18a)<br />
0<br />
where m and m' are two earth models. The data kernel Gi(r,m) is<br />
called the Fr6chet derivative or functional derivative at<br />
model m.
Agai-n, from a finite erroneous data set we can extract only averaged<br />
estimates with statistical uncertainties, i.e.<br />
As in the linear case A(r Ir) is built up from a linear combination<br />
0<br />
of the data kernels<br />
N<br />
A(rolr) = C a. (rorm)Gi(ro,m).<br />
i= 1 I<br />
Introducing (6.1 9) into (6.20) we obtain<br />
which for nonlinear kernels is different from (6.8) , since i.11 this<br />
nonlinear case gi (m) qi (m) .<br />
In the linear case, two models m and m' which both sa.tisfy the<br />
data lead to the same average model in1 (r ) >. In the nonlinear case,<br />
0<br />
the average models are different; the difference, however, is of<br />
the second order in (m' -m) . (Exercise! )<br />
The Backus-Gilbert procedure in the nonlinear case requires a model<br />
which already nearly fits the data. Then it can give an appraisal<br />
of the inforn~ation contents of a given data set.<br />
6.2. Generalized - matrix inversion<br />
The generalized matrix inversion is an alternative procedure to<br />
the Backus-Gilbert method. It is strictly applicable only to linear<br />
problems, where the model under consideration consists of a set of<br />
discrete unknown parameters. Nonlinear problems are generally<br />
linearized to get in the range of this method. Assume that we :mnt<br />
T<br />
to determine the M component parameter vector p with - p = (plr...,pb5:<br />
and that we have' N functidnalr, (rules) g. i = 1, . . . , N which<br />
1'<br />
assign to any model E a number, which when measured has the average<br />
value y . and variance var (y . ) :<br />
1 1
Suppose that an approximation & to p is known. Then neglecting<br />
terms of order 0 (E - %) ', we have<br />
. Eq. (6.22) constitutes a system of N equations for the M parameter<br />
changes pk - pko. The generalized matrix inversion provides a so-<br />
lution to this system, irrespective of N = M, N < M, or N > M. If<br />
the rank of the system matrix agi/apk is equal to Min (M, N), the<br />
generalized matrix inversion provides in the case M = N (regular<br />
system): the ordinary solrition,M < N (overconstrained system): the<br />
least squares solution,M > N (underdetermined system): the<br />
smallest correction vector p - %.<br />
The generalized inverse exists also in the case when the rank of<br />
agi/apk is smaller than Min(M, N).<br />
After solving the system, the correction is applied to and this<br />
vector in the next step serves as a new approximation to p, thus<br />
starting an iterative scheme.<br />
It is convenient to give all data the same variance ci2 thus<br />
0'<br />
defining as new data and matrix elements<br />
thus weighing in a least squares solution the residuals according<br />
to their accuracy which makes sense. Let<br />
be the parameter correction vector. Then (6.22) reads<br />
corresponds to the data kernels Gi(r) in the Backus-Gilbert<br />
theory). In the generalized matrix inversion first G - is decomposed<br />
into data eigenvectors u and parameter eigenvectors v :<br />
-1 -j<br />
G = U A<br />
T<br />
~<br />
(6.26)<br />
- =--
Here lJ - is a N x P matrix containing the P eigenvectors belonging<br />
to non-zero eigenvalues of the problem<br />
and 1 - is a M x P matrix with the P eigenvectors of the problem<br />
associated with non-zero eigenvalues. P is the rank of 5. - Finally<br />
= A is a P x P diagonal matrix containing just the P nonzero eigenvalues<br />
X . Then the generalized 'inverse of G .- is<br />
j<br />
- H always exists. In the cases mentioned above, - g specializes as<br />
follows<br />
For the proof one has to take into account that because of the<br />
orthogonality and normalization of the eigenvectors one al.ways has<br />
and in addition for<br />
T v v = I (=P-component unit matrix)<br />
=== = (6.30a)<br />
P<br />
For P < Min(M,N) H - cannot be expressed i3 terms of G. -<br />
The generalized inverse provides a solution vector<br />
- -<br />
< x > = H y .<br />
Its relation to the true solution - x is given by<br />
.=_Ax, -- -<br />
where - A is the (M x 21) resolution matrix
- Only for P M, _A - is the M component unit matrix, admitting an<br />
exact determination of - x. (A corresponds approximately to the resolution<br />
function A (ro 1 r) in the Backus--Gilbert theory. But there<br />
are differences: - is symmetrical, A(s[r) not, the norm of A - is<br />
small if the resolution is poor, the norm of A(ro[r) is always 1. )<br />
In generalized matrix inversion exists the same trade-off as in the<br />
Backus-Gilbert method. Cbnsider the covariance matrix of the change:<br />
of.< - x > due to random changes of x:<br />
In particular the variance of is<br />
P . Vk<br />
var (x ) = a2 ZI (3)<br />
Z ,<br />
j=l j<br />
k 0 k = 1,2, .:., M<br />
showing that the variance of xk is largely due to the small eigenvalues<br />
X By discarding small eigenvalues and the corresponding<br />
j'<br />
eigenvectors, the accuracy of x can be increased at the expense<br />
k<br />
of the resolution, since A - will deviate more from a (M x M) unit<br />
matrix if instead of the required M eigenvectors a smaller number<br />
is used. -<br />
The model will in general not.reproduce the data. The repro-<br />
duced data are<br />
= G - - =GWy - - =Ex -<br />
with the information density matrix<br />
T - B=GH=LJQ. - - - -<br />
(6.36)<br />
. . Only in the case N = PI 2 is a unit matrix. In particular, B -<br />
describes the linear dependence of the data in the overconstrained<br />
case. High diagonal values will show that this date contains<br />
specific information on the model which is not contained in other<br />
data. On the other hand a large off-diagonal value shows that this<br />
information is also contained in another data.<br />
Valuable insight in the particular inverse problem can be obtained<br />
by considering the parameter eigenvectors corresponding to high and
small eigenvalues. The parameter vector of a high eigenvalue shows<br />
the parameter combination which can be resolved well, the parameter<br />
eigenvector of a small eigenvalue gives the combination of para-<br />
meters for which only a poor resolution can be obtained.<br />
The generalized inverse is used both to invert a given data set and<br />
to estimate the information contents of this data set when the<br />
final solution is reached.<br />
hring the inversion procedure one has two too1.s to stabilize the<br />
notable unstable process of linearization:<br />
a) application of only a fraction of the con~puted parameter correc--<br />
tion, leading to a trade-off between convergence rate and sta-<br />
bility;<br />
b) decrease of number of eigenvalues taken into account.<br />
In the final estimation of the data contents one might prescribe<br />
for exh parameter a maximum variance. Then one has to determine<br />
from (6.35) the number of eigenvalues leading to a value nearest<br />
to the prescribed one. Finally the row of the resolution matrix<br />
for this particular parameter is calculated.<br />
6.3. 'mri.vation of the kernels for the linearized inverse problem<br />
of electromagnetic induction<br />
6.3.1. The one-dimensional case<br />
Both for the Backus-Gilbert procedure and for the generalized linear<br />
inversion a knowledge of the change of the data due to a small<br />
change in the conductivity structure is required.<br />
In the one-dimensional case the pertinent differential equation and<br />
datum are<br />
, flr(z,w) = I K ~ + iwp0o(z)}f (zrw)<br />
Consider two conductivity profiles o and o with correspondicg<br />
1 2<br />
fields fl and f2. Multiplying the equation for fl with f2 and the<br />
equation for f2 with flr subtracting the resnltinq equations and<br />
-<br />
integrating the difference over z from zero to infinity, we obtain<br />
.I (f;'f2 - f:fl)dz = iwp o 1 (ol-cr2)flf2 dZ<br />
0 0
Now<br />
m<br />
Division by f !, (0) :f; (0) yields<br />
If the difference 60 = a - o is small, f2 in the integral may<br />
2 I<br />
be replaced by fl, since the difference f2 - fl is of the order<br />
of 60 = a2 - o leading to a second order term in 6C = C2<br />
1'<br />
- C1.<br />
Hence to a first order in 6s<br />
m<br />
6C(w) = - iwp l6a(z){ f (z,w) 12 dz<br />
O 0 f' (0,~)<br />
Therefore in the Backus-Gilbert procedure the FrGchet derivative<br />
of C is -iwpb{f(~)/f'(o)}~. In the generalized inversion the deri-<br />
vatives of C with respect to layer conductivities and layer thick-<br />
nesses is required, if a structure with uniform layers is assumed.<br />
Let there be L layers with conductivity am and thickness d in<br />
m<br />
the m-th layer, h < z 2 hm+l (hL+l = -).<br />
m - Then (6.38) yields<br />
since in the last case all layers below the m-th layer are dis-<br />
placed too.'<br />
' '6':3 :2 : Pa,rt?al - derivatives in two- and 'three-dimensions<br />
For two-dimensions only the E-polarization case is considered. The<br />
pertinent equation is (cf . (3.5a) )<br />
AE = iwp a E .<br />
0<br />
Considef again two conductivity structures a and a2.<br />
1
Then<br />
- 77 -<br />
A(E2 - E ) = iwu U (E -E ) + iwc (G -U )E2<br />
1 0 1 2 1 0 2 1<br />
Let GI (r' - [r) - be Greent s function for u i.e.<br />
1'<br />
- and (6.40) by GI (r' I:), integrate<br />
Multiply (6.41) by E2(r1) - El (rr)<br />
the difference with respect to - r.' over the whole (y,z)-plane. Then<br />
Green's theorem (3.29) yields<br />
With the same arguments as in the one-dimensional case we obtain<br />
for small conductivity changes, neglecting second order terms,<br />
i I<br />
The kernel G(rt - [r) - must be determined from the solution of the<br />
integral equation<br />
-<br />
i-<br />
G(rl[r) - - = G ( r r - f ua(rn)Gn(r'[r")~(r"[r)d~"<br />
n- - o - - - - -<br />
Eq. (6.44) results from (6.42) by replacing<br />
This substitution becomes possible, since the 6-function in the<br />
Green's function equations drops out when the difference A(EZ-El)<br />
is formed.<br />
The changes of the magnetic field components are obtained from (6.43<br />
and (6.44) by differentiation with respect to - r. - Methods for the<br />
computation of Gn(rtlr) - - are given in Sec. 3.4.<br />
In the three-dimensional case one obtains in analogy to (6.43) and<br />
(6.44)<br />
where ?is competed by solving the tensor integral equation<br />
(6.44)<br />
Again the magnetic field kernels are obtained by taking the curl<br />
of (6.45) and (6.46) with respect to - r. Methods for calculatincj /.<br />
are given in Sec. 4.2. At the moment, three-dimensional inversion 6<br />
appears to be prohibitive!<br />
-<br />
0
6.4. Quasi-linearization - of the one-dimensional inverse problem<br />
of electromaqnetic induction (Sc1~mucl:er's PSI-alqorithm)<br />
In the one-dimensional problem it is possible to a certain extent<br />
to linearize the inverse problem by introducing new suitable parameters<br />
and data instead of the old ones.<br />
Assume a horizontally layered half-space with L layers having conductivities<br />
u , , . u and thicknesses d<br />
L 1' d2' ..-, dL-l<br />
CdM = m). Let the upper edge of layer m be at z = h and assume<br />
m<br />
that layer conductivLties and thicknesses satisfy the condition<br />
Let k = q. Then the pertinent equation for a quasi-uniform<br />
m<br />
external field is<br />
The transfer function is, as usual,<br />
C(w) = - f,(o)/£;(o) .<br />
Introduce instead of f the new function<br />
m<br />
In the TE-case f and ff are continuous across boundaries, i.e.<br />
As a consequence of (6.48) and<br />
(6.49) + is discontinuous across<br />
boundaries:<br />
'?<br />
z=h<br />
m<br />
The variation of. £ in the m-th layer is<br />
m<br />
. .<br />
-k (z-h )<br />
m<br />
fm(z) = A; e +<br />
+km ( z -hm)<br />
~ m + e<br />
1
whence<br />
2k d -2k d<br />
where a = CA:(A;)~ mm, y=e . m m<br />
I +a<br />
log t-) I -ay<br />
I +a<br />
log (-) I -a<br />
Because of the condition (6.47), y is independent of m.<br />
The quantity la[ is the ratio between the reflected and incident<br />
wave at z = h . Due to the conservation of energy the amplitude<br />
mi 1<br />
of the reflected wave is never larger than the amplitude of the<br />
incident wave. Hence,<br />
i Equality applies to a perfect conductor or insulator. For (a[
Introducing the new transfer function<br />
where u is an arbitrary reference conductivity, we obtain by con-<br />
0<br />
tinued application of (6.53)<br />
.(To "1<br />
= log - + y log - + y a2(h2)<br />
1 O2 .<br />
0 G u<br />
0 1<br />
2<br />
= log - + y log & + y2 log ;;- + y2 Q3(h3)<br />
"1 2 3<br />
L y(w) = (y-I) C ym-I<br />
m<br />
L- 1 0 o<br />
L-<br />
log - - y<br />
I L<br />
. . . . . . . . . . . 0 . . . . . . .<br />
m=l. 0 0<br />
log -,-<br />
In the derivation of (6.55) it has been used that (hL) = 0. Eq.<br />
L<br />
(6.55) is the desired result: it expresses the new data y(w1<br />
linearly in terms of the new model parameter log(o/cro).<br />
, When (6.55) is used for the solution of an inverse problem, one<br />
chooses first an arbitrary oo, and prescribes the constants<br />
c/a m o<br />
d<br />
m<br />
and L. If the occuning conductivity contrasts are not too<br />
sharp, the inversion of (6.55) will already give good estimates<br />
for the layer conductivities. For these estimates the correct<br />
,<br />
response values are calculated, and with the difference between<br />
the data and these values the inversion of (6.55) is repeated,<br />
giving corrections Alog(u /o ) to the previous outcome. At the<br />
m o<br />
end of the iteration procedure, the true thicknesses of the layers<br />
are calculated using (6.47). The errors of u are finally transm<br />
formed into the errors of d . m
.<br />
7. Basic concepts of geomagnetic and magnetotelluric depth sounding<br />
.<br />
. 7.1. . General characteristics of the method<br />
Two types of geophysical surface data can be distinguished to in-<br />
vestigate the distribution of some physical property m(g) of matter<br />
beneath the Earth's surface. The first type is connected with<br />
static or quasi-static phenomena (gravity and magnetic fields), the<br />
second type with time-dependent phenomena (seismic wave propagation)<br />
or with controlled experiments under vartable experimental con-<br />
ditions (DC-geoelectric soundings). Geomagnetic and magnetotellur7ic<br />
soundings utilize the skin-effect of transient electromagnetic<br />
fields. Their penetration into the Earth represents a time-depen-<br />
dent diffusion process, thus the observation of these fields at<br />
the surface produces data of the second type.<br />
The interpretation of static data y = y(R) ...- is non-unique alld an<br />
arbitrary choice can be made among an unlimited number of possible<br />
distributions m(r), - explaining y(R) - equally well. The interpretation<br />
of transient data is with certain constraints unique in the<br />
sense that oniy - one distribution m(r) - can explain the surface data<br />
y = y(R,t), basically because opeadditional variable t (time,<br />
variable parameter of controlled experiment) is involve?,<br />
If the observation of the transient process or the perforrfiance of<br />
the experiment is made at .a single site, the data y '= .~(t) permit<br />
a vertical sounding of the property m = 'm(z), assumed to be a sole<br />
function of depth z beneath that site, If the observations or ex-
'periments are done with profiles or arrays, the data y = y(R,t)<br />
permit a structural sounding of the property m = F(z) + Am(r)<br />
with particular emphasis on lateral variations Am(2).<br />
- .<br />
Here m(z) represents either a global or regional mean distribution.<br />
It may also be the result of vertical soundings at "normal sLtesn<br />
wheqzthe surface data show no indications for lateral variations<br />
of m. Since the dependence of y(R,t) - on m(r) - will be non-linear,<br />
anomalies Ay(R,t)<br />
and<br />
-<br />
-. = y(R,t) - yn(t) will be dependent on Am -<br />
m(z) , i.e. the inierpretation of. second-type -data must proceed on<br />
the basis of a known mean or nornlal distribution F(z) consistent<br />
with data yn("c at a normal site. It should be noted that in the<br />
case of static data of the first type usually no ii~terdependence<br />
between Ay and will exist, i.e, the interpretation of their<br />
anomalies is gependent of global or regional mean distributions<br />
of the relevant property n.<br />
Suppose that for data of the second type the lateral variations<br />
bm(r) - are small in relation to G(z). Then the results of vertical<br />
soundings at many different single sites may be combined to -<br />
approximate a structural distribution m = m + Am. For geomagnetic<br />
induction data the relevant property, namely the electrical. re-<br />
sistivity has usually substantial lateral variations and a one-<br />
dimensional intezpretation as in the case of vertical soundings<br />
will not be adequate. Instead a truly multi-dimensional inter-<br />
pretation of the data is required which is to be based on "normal<br />
data" at selected sites (cf. 8.2). Such normal sites are her@ rather<br />
the exception than the rule.<br />
7.2 The data and physicai properties of internal matter which<br />
are involved<br />
The Earth's magnetic field is subject to small fluctuations g(rtt).<br />
.<br />
They arise from primary sources in the high atmosphere and from<br />
seconda-y sources within the Earth. At the Earth's surface their '<br />
r. amplitude is<br />
-
orders of magnitude smaller' than the clu'asi-stntic main field.<br />
Their depth of ~enetration into the Earth is limited by the "skin-<br />
4Z.L<br />
effect", i.e. by opposing action of electromagnetically induced<br />
eddy currents in conducting matter below the surface . These<br />
currents produce the already mentioned secondary component of H.<br />
They are driven by the eLectric field - E(r,t3 generated by the<br />
changing magnetic flux according to Farad y 's induction law (cf.<br />
7,3), The the-fluctuations in - Fi are referred to "geomagnetic' variations,<br />
those in - E as geo-electric .or "telluric" variations.<br />
The connection between - ti and - E is established by three phy~icai<br />
properties of internal matter: the conductivity a (or its recipro-<br />
cal the resistivity p), the permeability p asid the dielectric con-<br />
stant E. The permeability of ordinary rocks 2s very close to unity<br />
and the approxirnat5,on p = 1 is adequate. The rnagnetic erfect of<br />
displace~~ent durrents which is proportional to E can be neglected<br />
within the earth (cf, Sec. 7.31, leaving the conductivity a as<br />
,the sole prope-i+~':c~nnecting 2 and K. Bcth fzelds tend to zero At<br />
sufficiently great depth, the skin-depth . -<br />
being a characteristic scale length for the depth of penetration<br />
of an oscillatory field exp(iwt) lnto . a . medhm of, resistivity p.<br />
. Geomagnetic induction data are usually presented in terms of trans-<br />
fer functions which connect as -functions of frequency and locations<br />
certain components of - E and - Ii. Examples are the imagneto-telluric<br />
transfer functions:between the tansentidl components 15 C-el I+<br />
and the magnetic transfer functions between the vertical and hori-<br />
zontal conponents of H only. These transfer functions form input<br />
data for - magneto-telluric - and - geomagnetic - depth soundings,<br />
..abbreviated NTS and GDS. . ,<br />
C.<br />
i4otations and units. Rectangular coordinates (x,y,z) are used with<br />
2 positive down. If x is towards local rlagnetic north and y towards<br />
local rnagnetic east, the f ollor~ing<br />
notations of the magnetic and<br />
. .<br />
electric field components are common:<br />
.
The vertical magngtic component HZ is usually simply denoted as Z.<br />
kn<br />
All equations are SI-units, but for H - the tradional geomagnetic<br />
unit ly = GauA may be quoted. A convenient unit for the<br />
telluric field is (mV/km). The magneto-telluric transfer function<br />
between (Ex,E ) and (Hx,H ) will'have in these units the dimension<br />
Y Y<br />
'{mV/km]/y with the conversion<br />
to SI-units. Measuring the period T = 2s/w of time-harmonic field<br />
variations in seconds,<br />
1<br />
p = - km,<br />
2-<br />
if T is measured in hours,<br />
p = 30.2 fi. km,<br />
the unit of p being in either case (am); po = 4s 10'~ volt sec/<br />
. .<br />
Ampin.<br />
7.3 Electromagnetic wave propagation and diffusion through<br />
uniform domains<br />
Consider a volume V for which the physical properties ~,E,u are<br />
constants. Then - E and - H are non-divergent within V and connected<br />
by Maxwell's field equations (. = .a/at)<br />
.... .<br />
rot E = -ppo H, ,<br />
-
Elimination of - E or .,. H yields a second<br />
.<br />
order partial.differentia1<br />
equation .<br />
. V- 2 ~ = pllo(oF - + E E ~<br />
where - F denotes either the electric or the magnetic field vector<br />
(div - F = 0). This equation can be interpreted as a wave equation<br />
or as a diffusion equation, depending on the fastness of the field<br />
variations in relation to the decay time<br />
of free electric charges within V.<br />
. .<br />
Let w be the angular frequency of harmonic field variations., i,e.<br />
- - - -<br />
F = iwF and F = -w~F, Then the basic differential equation is<br />
conveniently written as a wave equation<br />
for WT >> 1 and as a diffusion equation<br />
0<br />
for w~~
the surface. Only near certai-n sulfidic ow-veins the "effective"<br />
decay time T may be in the order of fractions of seconds due to<br />
0<br />
induced polarisation and the propaga-tion terrrt may not be necessarily<br />
small against unity.<br />
The air just above the ground nas'a resistivity in the order of<br />
loi4. Qm, yielding a time constant T of about 10 minutes, Hence,<br />
0<br />
fast fluctuations "propagate"<br />
><br />
while slow variations "diffuse"<br />
.from their ionospheric sources to the Earth's surface,<br />
For geomagnetic soundings only the field at and below the Earth's<br />
surface matters, regardless how the primary field reaches the<br />
surface. It is necessary, however, to make --- one definite assumption -<br />
about the nature of the primary field in the following sense:<br />
Non-di'.'ergent vector fields such as .... E and - H 5n uniform domains<br />
can be decomposed into two parts<br />
-<br />
F = fI +<br />
ZI1<br />
* 6<br />
- -<br />
= rot(rT) + rot ro't(rsS)<br />
h<br />
where - r denotes a unit vector in some specified direction, here<br />
the z-direction. T and S are scalar Cunctions of position, It is<br />
readily seen that the so-called "toroidal" part fI is orthogonal<br />
A<br />
to 2, i.e, in tne here considered case XI is tangential to planes<br />
z = const. The remaining so-called "poloidaln' part zII of F has<br />
three components:<br />
rF"<br />
\ I<br />
VI /<br />
\ \ 1 1 / r<br />
Let now PI and gII be a "toroidal" and a "poloidal" diffusion<br />
vector, both satisfying<br />
V'P - = iwpoo(l +. T~)I',<br />
from which the toroidal and poloidal parts of .-. H are derived by<br />
definition as follows:<br />
-
Observing that<br />
rot pot rot g = - rot(~'p-~rad - divP) - = -iwp o a rotg, the electric<br />
vectors follow then from Maxwell's field equations as<br />
2i<br />
P<br />
E = - rot rot PI; E<br />
-I -I I<br />
= rot PII. '<br />
p2 i(l-ki)<br />
Since the field which is derived from PI has a - tangential ~agnetic<br />
field, it is termed the "TM-mode" of the total field, The field<br />
derived from PII , on the other hand, has a - tangential electric<br />
field and represents the "TE-mode" of the total field.<br />
Suppose that the conductivity is a sole function of depth, o = a(a),<br />
and therefore the diffusion vector vertical ?ownward: - P = (O,O,PZ).<br />
Let Pz be in planes a = const. a harmonic osci.llaCing function<br />
of x and y with the "wave numbers" kx and k Y in x and y-direction:<br />
Then<br />
p(r)<br />
ik . x : po(z) -e - - with - - = (k,,k Y ), ij. (",s?*<br />
rot g = (ik -ikx,O) Pa<br />
Y'<br />
rot rot - P = (ikx/C, iky/C, ]klz) PZ<br />
1 =. ap0/az<br />
where - C and lk1<br />
0<br />
= k:. + k2, It can be shown that C is<br />
Y<br />
a scale length for the depth of penetration of the field into the<br />
uni.form domain. Insert now rot - P and rot rot P into the equations<br />
, . as given above and obtain the following relations between field<br />
components setting wro=O:
Since CI and CII cannot be zero for a finite wave number lkl, TM-<br />
modes will have in the here considered quasi-stationary approxi.ma-<br />
tion zero magnetic fields in nonconducting matter as in air above<br />
the ground. Hence, the existence of TM-modes within the primary<br />
field cannot be ruled out by magnetic observations,sj.nce it' w i l l<br />
be seen above the ground only in.the vertical electric field EZ,<br />
provided that this vertical. electric field i.s not due to local<br />
anomalies of the secondary induced field (s, below), In reality<br />
J<br />
the detection of a primary EZ will be nearly impossible because<br />
changes of the vertical electric field due to changing atmospheric<br />
conditions appear to be orders of magnitude greater than those<br />
connected with ionospheric sources, Therefore it is an ---<br />
assumption<br />
that the primary fields in geomagnetic induction studies are TEfields<br />
.<br />
In that case the secondary induced field above and within a layered<br />
substructure cs = cs (z) is likewise a TE-field. The sum of both<br />
n<br />
fields w i l l be referred to as the "normalt1 field H and E for the<br />
-n -n<br />
considered substructure. Its depth of penetration is characterized<br />
by the response function CII = C which according to the basic re-<br />
n<br />
lations given 'above can be derived from the magneto-telluric "im-<br />
pedance" of the field<br />
-<br />
E E<br />
nx = - 3 = iwPocn<br />
H<br />
ny Hnx<br />
or from the geomagnetic ratios<br />
(MTS=magneto tclluric sounding)<br />
H H<br />
. --.. nz - . v<br />
1kxCn , = ik C (GDSrgeomagnetic depth<br />
H<br />
Hnx nY Yn sounding)<br />
Suppose the internal resistivity is within a limited range also<br />
dependent on - one horizontal coordinate, say x:<br />
u =.un(z)'+ aa(y,z).<br />
Now a local anomaly H~(~,Z) and Ea(y,z) w i l l appear within the<br />
secondary field.
Two special types of such anomalies can be distinguished. If the<br />
electric vector of .the primary fields is linearly polarized in<br />
x-direction, i.e.<br />
and consequently<br />
E = (E , 0, 0)<br />
-n nx<br />
H = (0, H , Hnz),<br />
-n nY<br />
the anomaly has an electric vector likewise only in x-direction:<br />
because the flow of eddy currents will not be changed in direction.<br />
Hence, the anomalous field is a TE-field -<br />
This polarisation of the primary field vector is termed 3ol.arisati<br />
If the electric vector of the primary field is 1inearl.y polarised<br />
in y-direction, the normal field is<br />
and consequently<br />
E = (0, Eny' 0)<br />
-n<br />
H = (H 0, 0)<br />
-n nx'<br />
provided that its depth of penetration is small in comparison to<br />
its reciprocal wave number, yielding H = 0. Only with this con-<br />
the n z<br />
straint is flow of eddy currents to vertical planes<br />
x = const. and the resulting anomalous field w i l l be a TM-fie1.d -<br />
with zero magnetic field above the ground:<br />
E = (0, E<br />
-a ayy Eaz)<br />
H = (Hax, 0, 0).<br />
-a<br />
This pola.risation .of the primary field is termed "H-polarisation". - -.<br />
For three-dimensional structures<br />
O = Ci(x,y,z) = u Cz) + Cia(x,y,z)<br />
n<br />
the anomaly of the induced field will be composed of TE-and TM-<br />
fields which cannot be separated by a special choice of coord:. 7 nates.<br />
There is, however, the following possibility to suppress in model<br />
calculation the TM-mode of the anomalous field:
Suppose the lateral variations u are confined -to a "thin sheet".<br />
a<br />
This sheet may bz imbedded into a layered conductor from which it<br />
must be separated by thin non-conducting layers. Then no currents<br />
can leave or enter the non-uniform sheet and the TM-mode of the<br />
induced field is suppressed. Such models are used to describe the<br />
induction in oceans, assumed to be separated from zones of high<br />
mantle conductivity by a non-conducting crust.<br />
Schemat:ic summary:<br />
Source field Induced field<br />
,,,\~. x,'~ - . . normal anomalous<br />
TE {<br />
TEtTM(genera1)<br />
TE (thin sheet)<br />
xAppGn?3iX to'?, 3 : Recurrence formula for the calculation of the<br />
depth of penetration C for a layered substratum (cf. chapter 2)<br />
7,<br />
r<br />
Definition: C = - 0<br />
aPolaz<br />
a2p<br />
Differentjal equation to be solved: -L? = (iw~~~o + lkI2)po<br />
which satisfies<br />
az2<br />
V - ~ = P iwpo~ .<br />
Continuity condictions:<br />
1. TE-field: - H and E must be continuous which implies<br />
that C-is continuous<br />
-<br />
2. TM-field: H and (Ex,E ,oEZ) are continuous which imp]-ies<br />
v<br />
that oC is continuous.
Model :<br />
wo+<br />
Solution for uniform half-space: CI = CII = lk(<br />
(a = oO)<br />
Recurrence formula foY layered half-space'{Cn = C(zn)}:<br />
-<br />
C~~ - C~~~<br />
= Jioy o + [kr2<br />
0 L<br />
The TE-field within the n'th layer at the depth z = zn + E,<br />
Z < Z + E < Z<br />
n+l'<br />
can be calculated from its surface value<br />
n n<br />
according to:<br />
with gtzr,zs) = cosh Bps - sinh Brs/(K,Cr)<br />
h(zr,z ) = cosh @, - sinh B K C<br />
s rs r r<br />
for s.< - n<br />
and gCzn, z~+E) = cosh(Kn~)-sinh(K n E)/K n C n
, ,<br />
The formula for E applies also to E and HZ, the formula for<br />
X Y<br />
Hx also to H .<br />
Y<br />
7.4 Penetration depth of various types of geomagnetic variations<br />
and the overall distribution of conductivity within the Earth<br />
Three types of conduction which vary by orders of magnitude have<br />
to be distinguished:<br />
1. Conduction through rock forming minerals<br />
2. Conduction through fluids in pores and cracks between rockfor-<br />
mine minerals<br />
3. Metallic conduction<br />
li: Since the minerals of crustal and mantle material are expected<br />
to be silicates, the conduction through these minerals will<br />
be that of semi-conductors. Their resistivity is in the order<br />
5<br />
of 10 Om at room temperature but decreases with temperature<br />
T is the absolute temperate, k Boltzmann's constant; a and A<br />
0<br />
are pressure-dependent and composition-dependent constants<br />
within a limited temperature range, for which one specific<br />
mechanism of semi-conduction is predominant. Hence, in plots<br />
of in o versus T-' the o-T dependence should be represented<br />
by joining segments of straight lines. Laboratory experiments<br />
with rocks and minerals at high temperature and varying<br />
pressure have confirmed this piece-wise linear dependence of<br />
lno from T-l. Furthermore, they showed that the composition-<br />
dependence of u is mainly contained in the pre-exponential
parameter u and that the pressure dependence of o and A should<br />
0 0<br />
not be the determining factor for the conductivity down the depth<br />
of several hundred kilometers.<br />
7 (, "C The u-T dependence of Olivin (Mg,<br />
1200 3 b0 60 0 11<br />
Fe ) Si04 has been extensively<br />
T--- 2<br />
investigated by various authors<br />
for temperaturs up to 1400~~ and<br />
pressures up to 30 kbar. This<br />
mineral is thought to be the main<br />
constituent of the upper mantle<br />
with a Mg : Fe ratio of 9 : 1 :<br />
I Si04. The diagram<br />
(Mg0.9 Fe~.l 2<br />
shows the range of u(T)--curves<br />
as obtained for Olivin with 90%<br />
Forsterite CMg2Si041. It is beli.eved that the discrepancies among<br />
-the curves over orders of magni:tude are mainly due to different<br />
degrees of oxydati-on of ~e" to Fe 'I within the olivLn samples<br />
used for the measurement. In fact, the samples may have been oxy-<br />
dized in some cases during the heating experiment as evidenced by<br />
the irreproducibility of the u (TI-curves. However, the r>ar.ge of<br />
possible conductivities for olivin with 10% forsterite is greatly<br />
narrowed in at the high temperature end, where we may expec-t a<br />
resistivity of 10 to 100 Qm for 1400°c, corresponding to a depth<br />
of 100 to 200 km within the mantle.<br />
2.:Electrolytic conduction through salty solutions, filling pores and<br />
cracks of unconsolidated rocks, gives clastic sediments resistivi-<br />
ties from 1 t o 50 Qm. The resistivity of sea water with 3.5 gr<br />
NaCl/liter is 0.25 am. There should be an increase of resistivity<br />
by one or two orders of magnitude at the top of the crystalline<br />
basement beneath sedimentary basins.<br />
Near to the surface the conduction in crystalline rocks is also<br />
electrolytic. Their resistivity may be here as low as 500 Qm, but<br />
it increases rapidly with depth, when the cracks and pores close<br />
under increasing pressure. Below the Mohorovicic-discontinuity<br />
the conduction through grains can be expected to be more effective<br />
than the conduction through pore fluids. If, however, partial
h<br />
melting occurs, the conduction through even a minor molten<br />
fraction of crystalline rocks could Pouer the resistivity by,<br />
say, one order of magnitude.<br />
3.: Metallic conduction is expected to give the material in the oute<br />
core a resistivity of Om. This law value is required by<br />
dynamo theories for the explanation of the main field of the<br />
Earth.<br />
8 6 'f 2 -5 0 + 5 + l'f<br />
.<br />
0 -1<br />
4.0<br />
-&?-xu- 197'<br />
~ ;pa*'?<br />
t<br />
3.2 ES 't<br />
s*nri- ~"r\da~tirr, 1, L- Waf-<br />
I 6. -,b~ loo-l~ok<br />
s~!-,--<br />
4:c ru-,o-r<br />
- ...... > ... .- . -. . -. - - d<br />
3. CI~L~JI.~ KC(;~.L.~~<br />
.<br />
3 7," *\<br />
V"C.L.*<br />
Y.. R;*<br />
The frequency spectrum of geomagnetic variations which can be<br />
utilized for induction studies extends from frequencies of<br />
fractions of a cycle per day to frequencies of 10 to 100 cycles<br />
per second. At t'he low frequency end an overlap with the<br />
spectrum of secular variations occurs which diffuse upwards<br />
from primary sources in the Earth's core.<br />
460<br />
/<br />
j , AMfI:4uL&<br />
/) 0<br />
?<br />
Sc-Qeggti.csp.ectrun1 of geoma~netic variations :<br />
><br />
SccLLl sx<br />
Su-b~t-r<br />
,Wd"t;0*<br />
59 C?,.rcr-KC+ t. )<br />
4 , ,<br />
..L -
(=disturbed)-variations: After magnetic storms the hor,j.zontal<br />
Dst --<br />
-<br />
H-component of the Ear-th's magnetic field shows a worldwide de-<br />
crease. Within a week after stormbeginn H returns asymptotj.cally<br />
to its pre-storm level. This so-called Dst-phase of storms shows<br />
negligible longitudinal dependence, while its latitude dependence<br />
is well described in geomagnetic coordinates by<br />
3 L<br />
is the geomagnetic latitude of the location considered. The<br />
source of the Dst-phase may be visualized as an equatorial - ring<br />
- current (ERC) which encircles the Earth in the equatorial plane<br />
in geomagnetic coordinates:<br />
Tf the interior of the Earth were a perfect insulator and no eddy<br />
currents were induced, the vertical Dst-coniponent would be ZDst -<br />
H, sin @. In reality, only one fourth to one fifth of this value<br />
is observed due to the field of eddy currents which oppose in Z<br />
the primary field of the ERC. Let LDst denote the inductive scale<br />
r -<br />
ijw Q=kso length of DPt, a be the Earth's radius, then for @ : 45'<br />
.,<br />
- -<br />
- 2C~st<br />
'DS t a H~st4<br />
(cf. sec. 8.2)<br />
Hence, with ZDStfHDSt 0.2 the depth of penetration will be 600<br />
km'.
9 (solar quiet) varia-tions: On the da)l-lit side of the Earth thermal<br />
convection in combination with tidal motions generate large-scale<br />
wind-systems in the high atmosphere. 1; the ionosphere these winds<br />
produce electric currents which - for an observer fixed to the<br />
Earth, - move with the sun around the Earth. During equinoxes there<br />
will be two current vortices of equal strength in the northern and<br />
0 0<br />
southern hemisphere, their centers being at roughly 30 N and 30 S.<br />
Due.-to these currents local-time dependent geomagnetic variations<br />
will be observed at a fixed site at the Earth's surface. They are<br />
repeated from day to day in similar form and are called "diurnal"<br />
or Sq-variations:<br />
--<br />
Aiai-n the Z Ampli-tude is much smaller than to be expected from<br />
SQ-<br />
D for an insulating Earth. Tf and D (") deno-te the amplitudes<br />
s Q SQ s ?<br />
of the m1 th subharmonic of the diurnal varl.;i.tions (m=1,2,3 ,I!<br />
(I7 the<br />
corresponding to the periods T = 24, 12, 8, 6 hours) and C s Q<br />
inductive scale length of this subharmonj.~,
with 4 as geographic latitude. From the %Q : FISQ ratio of continental<br />
stations a depth of penetration between 300 and 500 km is inferred.<br />
The penetration depth of SQ in the -ocean basins is still uncertah.<br />
Bays and - Polar Substorms: During the night hours "bay-shaped" de-<br />
flections of the Earth's magnetic fi.eld from the normal level ave<br />
observed from time to time, lasting about one hour. Their amplitude<br />
increases steadily from south to north, reachi-ng its highest value<br />
in the auroral zone. Similar variafions, but much more intense and<br />
rapid, occur during the main phase of magnetic storms, until about<br />
one day after stormbegin.<br />
The source of these so-call-ed "polar substorms" is a shifting and<br />
oscillating current lineament i.n the ionosphere of the aurora zone.<br />
The current wil].be partly cl-osed th~ough field alligned currents<br />
in the magnetosphere, partly by wide-spread ionospheric return<br />
currents in mid latitudes:<br />
S,'L
polar substorlns is j.n mid-latitudes (e.g. Denmarlc, Germany) much<br />
smaller than the ampliiudes of H and D<br />
-<br />
because the vertical field<br />
of induction currents nearly cancels the vertical field of the<br />
polar jet. Under "normal condirtions" the Z:I-l ratio is about 1: '10.<br />
Assu~ning for the mid--latitude substorm field an effective wave<br />
number of 10- 20 000 km, yielding kx 3-6*10-~ as wavenumber, depth<br />
of penetration i.s<br />
'bay<br />
A<br />
I = 150 to 300 km.<br />
bay kx<br />
There are indications tllat the penetration deptll of bays into -the<br />
oceanic substructure is substantialiy smaller. The ocean itself<br />
produces an attenuation of the H-ar~pl.itude of about 75% at the<br />
ocean bottcm, deep basins filled with unconsolidated sediments can<br />
yield a comparable attenua-tion &-6. North German basin).<br />
Pulsations and VLF-emissions: Rapid oscilla-tions of the Earth's fie1<br />
- -<br />
with periods between 5 minu-tes and 1 second are called pu1.sati.011~.<br />
Their ampli-tude increases 1.iIce the ainpli-tude of substorms strongly<br />
when approaching the auroral zone. Their typical midlati.tude<br />
ampli-tude is 1 gamma. The source field structure of bays and pul.-<br />
sstions is similar, the depth of pene-tra-t:ion of pulsations being<br />
largely dependent on the near-surfack conductivity. Lt may rmge<br />
from many kilometers in exposed shield areas to a few hundred<br />
meters and less in sedimentary basins.<br />
The "normal" Z:H ra-tio o:E pulsations is too small to be determS.~-rccl<br />
with any reliabili-ty outside of the auroral zone. However, "an~!naI.oi<br />
Z-pulsations. are frequent and usually of very local. charak-ter. IF<br />
pulsations occur in the form of las-tine harmonic oscilla-:ions , oftel<br />
with a beat-frequency, they are called "pulsation tr>ains1' pt, sin~J.(<br />
pul-satj.on events are ' called "pi" , pul.satj.ons which rt~arlc the be-<br />
ginning of a bay.dis-turbance are called "pc". All three types !la\'c<br />
a clear local time dependence, occuring almost daily:<br />
'F\ 1-1 m- -7,. -
Very rapid oscillations with frequencies between 1 and, say, 100 cps<br />
(=Hz) are called - from the radio enginieers point of view -<br />
"very - - low - frequency emissions: VLF. Their ampl.itudes lie well below<br />
ly. They occur usually intermittendly in "bursts" and are controlled<br />
.in their intensity by the general magnetic activity. In exposed<br />
shield areas they may penetrate downward a few kilometers, but<br />
everywhere else they w i l l be totally attenuated within the very<br />
surface layers.<br />
- Sudden storm -- commencements and sclar flare effects: The begin of a<br />
magnetic storm is usually marked by strong deflections in H and Z<br />
up to 100y within one or two minutes.. This so--called "sudden - - storm<br />
-<br />
commencemen-t" SSC, signals. the inward :notion of the magnetopause<br />
(separating the magnetosphere from the interplanetary space) under<br />
the impact of a suddenly increased solar wind intensity. SSC's<br />
are a world wide, s imultaneous1.y occuring phenornenon. They are,<br />
however, too rarely occuring events (1 per month) for induction<br />
studies.<br />
The momen-tarily increased solar wave radiation, caused by sun spot<br />
e: eruptions, produces a shortlived (5 minutes) in'tensificatrion of the<br />
Sq-system, its magnetic effect is called a "sol.ar flare event". It:<br />
measures a few gammas and like the ssc is a rarely observed varia-<br />
tion type.
Overall - depth - distribution -. Denth of penetation<br />
of resistivity<br />
. .<br />
& ~ n ]<br />
(0) : oceans<br />
(s) : sedimentary basins<br />
(c) : exposed crys-talline<br />
EEJ:<br />
PEJ:<br />
mid-latitude<br />
obs erva-t ions<br />
equatorj.al. e1.ec'tr.c<br />
polar electz~ojet
!<br />
8. Data Collection - and Analysis<br />
8.1 Instruments -<br />
unetic sensors for geomagnetic induction work on land should meet<br />
as many as possi-ble of the following requirements:<br />
(1) Sufficient sensitivity and time resolution<br />
(2) Directional characteristics which allow observation of the<br />
magnetic variation vector in well-defined, preferably ortho-<br />
gonal components.<br />
(3) Stable compensation of the Earth's permanent field, if re-<br />
quired. The sensors should not show "drift" on a time scale<br />
comparable to the slowest variati-ons to be analysed.<br />
(4) Compensation of temperature effects or linear temperature de-<br />
pendence with well defined temperature coefficients for subse-<br />
quent corrections.<br />
C5) Low power consumpti.on to allow field operations on 57et or dry-<br />
cell batteries (less than 100 n;A at 10 to 20 Volts DC con-<br />
tinuous power drain).<br />
C6 1 Minimurn maintenance, all-owing unattended opera-tions over days,<br />
weeks or months, depending on -the period range to be irlvesti-<br />
gated.<br />
(7) Electric outmpu-t signal, adaptab1.e for digital recording<br />
w<br />
No single sensor system can possib1.y meet all requiremen-ts over thc<br />
full frequency range of natural geornagrletic variations and the<br />
- choice of instruments wil.1 depend on the type of variations under<br />
investigation. They record either the total field M Ho + 6H as<br />
?E<br />
sum of -tl;e s-teady field Ho and the variation field 6H, by coml2ensa.-<br />
tion of Ho, the time--derivative fl : 6f'i or a combina-tion of 61.1 nil<br />
x*<br />
& i1. In geomagnetic latitudes the spectral amplitude distr.Lbution of<br />
6H and 611, averagedoverextended perioss of moderate ma~c5tj.c acti-<br />
vity, is the following:<br />
x the variation field 6H only
The next diagram shows the resulting spectral range of adequate<br />
sensitivi.ty and stability for various magnetic sensors, suitable for<br />
geomagnetic studies: Perfornlance on<br />
point (2) to (7)<br />
D<br />
Torsion fibre<br />
B a s s<br />
0 s . 2 3 5 6 7<br />
magnetometer<br />
graph, ~ough<br />
Reitzel variograph<br />
1. )<br />
i<br />
!<br />
I<br />
- I-<br />
i<br />
I<br />
I<br />
-<br />
I<br />
6H + + + t i - -<br />
I I - I<br />
i<br />
I I !<br />
Fluxgate magne- /<br />
tometer (I'oer- i<br />
. ster Sonde) 3.)<br />
6 1-1 i. - - + + J.<br />
Grenet-vario-<br />
meter 1. )<br />
Induction coils<br />
Proton prece-<br />
I<br />
1 !<br />
Kb-vapor Inagne- -------- -. lio+6H - -I- + - - 4-<br />
tometer 3.)<br />
.I. ) Cornpensation of H by mechai-~$.cal torque<br />
0<br />
2.) Bobrov-quartz variometers, Jovilet variometers<br />
3.) Compensation of H with bias fields<br />
0<br />
The horizontal compone~lts of the surface elec-trjc field E which i.s<br />
connected to 6H and 6k can be estimated either from the knowledge<br />
of the depth of penetration C at the period T or from the knowledge<br />
of the resistivity p of the upper layers, assumed to be uniforw.<br />
Measuring H in (y), C in (km), T in (sec) and p i n (Qm) gives<br />
. . .<br />
!<br />
! i !<br />
i 611,6; + - + + - - I<br />
i<br />
i 66 + - + i- i-<br />
: .<br />
ssion magne-to- - -- fi t&H + + + - - .I-<br />
0<br />
meter 3.)<br />
&<br />
r--<br />
Using the spectral amplitudes as given above and C from'previous<br />
sectLon, the following spectral amplitudes of E/H and E are obtained<br />
for mid-la-ti-l-udes during rno:lerate magnetic aeti-vity :
A horizontal electric -- field component in the direction - s is observed<br />
as the fluctuating voltage Es'd between -two electrodes in a distance<br />
d parallel to 2. Due to possible self-polarisation of the electrodes<br />
a quasi-steady voltage of considerable size may be superimposed<br />
upon the fluctuating voltage, truly connected to geomagnetic varia-<br />
-Cions. This self-poten-tial amounts for electrodes in a sa1.t solu-<br />
tion comparable to ocean water to:<br />
> 1 vol-t for unprotected iron rodes<br />
100 mV Cu-CUSO,~ elec-ti-odes<br />
Kaloniel (= biophysical) e1ec:i:vo~<br />
kg-kgC1 (=oceanographic) electro<br />
2<br />
It may vary slowly under changing conditions in the waterbearing soi<br />
Hence, it should be as small as possible in comparison to the voltafi<br />
truly connected to 6H and 6i1. This applies in particular .to periocis<br />
of one hour and more. There are two options to avoid these unwanted<br />
electrode effects: Large electrode spacing (d - > 10 km), hrigh-quali-i:y<br />
electrodes. Preference should be given -1-0 ttte seco~ld option, using<br />
electrode spacings of, say, 100 in. The use G.? a large electlode<br />
spacing does not imply necessarily tha-t a mean electric field,<br />
avera~ed over local. n~ar-surface non-uni.formlties , wi1.l be observed,<br />
since the observed voltage may still be largely de-ter~nined by local<br />
condi-tions near to either one of the electrodes. In any case, the<br />
site of the electrode installati.on should be surveyed with .dipol-<br />
soundings or DC--geoelectrics to ensure layered conditions at -- and<br />
be-tween electrodes.
8.2 Or~anisation and objectives -. of field opcrations<br />
Geomagnetic induction work can be carried out by (i) single--site<br />
soundings, (ii) simultaneous observations along profiles, (iii)<br />
simultaneous observations in arrays covering extended areas. It may<br />
be possible to replace simultaneous by non-simultaneous observations,<br />
provided that the variation field can be "normalized" with respect<br />
to the mean regional variation field. This "normal" field may be<br />
the field at one fixed site with no indications for the presence of<br />
lateral non-unj-formities, or it may be the averaged f'ield obtained<br />
from distant permanent observatories. Usually the normal.isa'tion is<br />
performed in the frequency domain, introducing se-ts of transfer<br />
functions.<br />
Single site vertical sounding:<br />
The resistivity structure within the depth-distance range of pcnetra.<br />
tion is regarded a sole function depth: p = p(z). Its extent is<br />
given by -the modulus of -the induc-tive scale length C(w) at the con-<br />
sidered frequency, whi.ch increases with<br />
- - , --<br />
the increasing period. If a vertical<br />
........... .......... -. .<br />
-. -<br />
J 1 ., . . . ...<br />
.- ..-
For a data reducti.on in the fr3equency domain (cf. Sec. 8.3) the<br />
following set of transfer or response functions w i l l be defined:<br />
For a frequency-space factor Of the source field<br />
the response function C(w) = C n which - characterises the downward<br />
depth of penetration is connected to these transferfunctions accor-<br />
ding to<br />
A Z<br />
n x y<br />
n ikx iwv0<br />
c =- -<br />
The Cagniard apparent resistivity if given hy<br />
the phase of the impedance by<br />
I$(w) = arg(Z ) = arg(C ) + 5 ,<br />
xY n 2<br />
the modified apparent resistivity by<br />
which can he used as an estima-tor of the true resistivi-ty at the<br />
depth<br />
z"(~) = R~IC~I. (cf. Sec. 2.5 and 9.1)<br />
A cornpairison ,of the various equivalent respcrlse func-l-j.ons for a<br />
. , given substructus&ives the fol.lowing displ.ay:<br />
L? *
Tests for the assumption of a layered dis.tri.bution and the source<br />
In addition there are certain constraints with regard to modu.lus,<br />
phase and frequency dependence of the transfer functions. (cf. Sec.<br />
2.6).<br />
The compatability of MTS and GDS vertical sounding results can be<br />
tested by the requirement that<br />
asreadily seen form the second field equation rotE = -imp H .<br />
-n o -n<br />
If the wave-number structure of the source field is not lcnown, a<br />
-<br />
Horizontal Gradient GDS can be formed: Observing that<br />
-<br />
the differentation of E = Cn impo H with respect to y and of<br />
nx nY<br />
= -iou C H wi.th respect to x yields<br />
.Env o n nx<br />
Since the electric surface field may be distorted by lateral non-<br />
uniformi-ties even when their scale length and dep-th is smal.1 in<br />
comparison -to the depth of. penetration, -the E-field observa.tj.oi~s<br />
may be replaced by observing the tangen-tj.al- magnetic field also<br />
at some dep-th d below -the surface, i.e. by conducting a 'vertical -<br />
-----<br />
Gradient GIIS. No knowledge of the wave-number st~uc-ture of the<br />
source field w i l l be required, but"the resi;tivity po between the<br />
' surface and the subsurface points of observations will be reqci-red.<br />
Assume that the lateral gradients of IJnz are small in comparison<br />
to the vertical gradients of H and H in v5.e~ of the condi.tion<br />
n x nY<br />
I -.<br />
I ~ c [
Let q be the tranfer function between Hnx(H at the surfade and<br />
nY<br />
) at the depth z = d:<br />
Hnx(H nY<br />
Then<br />
H (d) = q Hnx(0) where I q 1 < 1.<br />
nx<br />
Single site geo-letic structural sounding: The source field is re-<br />
garded as quasi-uniform (k = 0) and the ver:tical magne-tic component<br />
therefore as anomalous, arising solely from lateral changes of the<br />
resistivity within the depth-distance range of penetration, where<br />
p = Pn(z) + P,(X,Y,Z)\<br />
In this special case the resulting anomalous magnetic vector<br />
H = (Ha,, H , H = H is linearly dependent on the quasi-uniform<br />
-a ay az z<br />
normal magnetic vector H = (Ilnx, H 0) iii the frequency-distance<br />
-n ny'<br />
domain :<br />
denotes a matrix of linear transfer functions as functions of fre-<br />
quency and surface location. This i.mplies that also linear relations<br />
exist between HZ and the total (=observed) ho17izontal variations:<br />
with<br />
These alternative transfer functions A and B can be del-ived now from<br />
observations at a single site. Their graphical display in the form<br />
of Parkinson-Wiese induction arrows indicates the trend of -the sub-<br />
surface resistivity structure which is responsible for the appearance<br />
of anomalous '2-var.i.ations :
The in-phase induction arrow is defined by (in Parkinson's sense of<br />
orientation) by<br />
h<br />
p = - R&(x A + B]<br />
and the ou-t' of-phase arrow by<br />
q + Imagl; A + Bl<br />
h h<br />
where x and y are unit vectors in x- and y-direction. General-ly<br />
speaking, the in-phase arrows point toidards internal concentrations<br />
of induced currents, i.e. to zones of lower than "normal" re-<br />
sistivi.ty at one particular depth.TheYmay point also away from high<br />
resistiv:ity zones around which the induced currents are diverted.<br />
Vertical soundings with station arrays:<br />
The resistivity structure is regarded as layered, p = p(z),but the<br />
inducing source field as non-uniform. Inducing and induced fields<br />
will have ms.tching wave-number spectren with well defined ratios<br />
between spectral components in accordance to the subsurface re-<br />
s i s t ivi-ty structure . h ea-t-k-+h~++t-~o-~~&e-~~idc~~~<br />
e.ve;i:~-~:gh-i-s-k~ol.a-7,-u~.-f m~rnGty--rnaap-~be--~i-~-~P+f+a<br />
Let U and V be field components of the surface field in the frequency<br />
distance domain: U = UCw ,R) , V = VCw ,R). They are deco~nposed into<br />
h h A<br />
the wave-number spectren U(w,k), VCw,k) according to<br />
with<br />
irnh ikR - -<br />
U(w,R) = I I UCw,L,)e dk dk<br />
- '0 Y X
. A<br />
as transfer function between A and B in the wave number domain, which<br />
contains the information about the internal resistivity distribution.<br />
A vertical sounding of this distribution is made by considering<br />
h<br />
R as a function of frequency, k being fixed, or vice versz.<br />
If the array covers the whole globe, the sphericity of the Earth re-<br />
quires the replacement of the krigonometric functions by spherical<br />
harmonics in spherical coordinates Cr,B,A):<br />
a: Garth' s radius ; p:(cos@ : Associated' spherical function. The trans-.<br />
fer functions have then the form<br />
The first and still basic investigations of the Earth's deep conduc-<br />
tivity structure have been carried out by this approach (SCHUSTER,<br />
CHAPMAN and PRICE).<br />
If the array of stations covers only a region of Limited extent, it<br />
may be !impracticable to de'compose the observed field into wave-<br />
spec-tral components or it may be even impossible because only a small.<br />
section of the source field structure has been observed. In that<br />
case vertical array soundings are carried out preferably with response<br />
functi.ons in the frequency-distance domain.<br />
Suppose the source field is quasi-uniform ,i.n one horizon.ta1 direc-- j<br />
tion, say,<br />
U = UCw,y) and V : V(w,y). Ls't<br />
* . %<br />
be the inverse Fourier transform of R. Then, if V is given by the<br />
,. A<br />
roduct of R with U in the (u,k) domain, V will be given by a<br />
corlvolution 05 R with IJ in the (w,y) domain:<br />
+w<br />
with .R U = 1 RCw,y-n) UCq)dq.<br />
-m
If, for instance E = V and H = U, the magnetotelluric relation<br />
A ,. X Y<br />
EX = WM, C II of the k-w domain will transform& into<br />
Y<br />
with N(w,y) as Fourier<br />
-<br />
transform of C(o,k). Observe that reversely<br />
4<br />
-iky<br />
CCw,k) = I N(w,y)e. dy<br />
and therefore<br />
-03<br />
f-<br />
CCw,O) = J N(w,y)dy.<br />
-m<br />
Hence, if H is quasi-uniform within the range of the kernel M,<br />
Y<br />
which is the CAGNIARD-TIKHONOV relaticn con~monly used for single<br />
1<br />
site sounding-s.<br />
In a similar way the magnetic ratio, of vertical to horizontal' va-<br />
riations can be generalized to<br />
with MCw,y) as Fourier transform of ik C(w ,k ) .<br />
Y Y<br />
The response functions N and M have their highes-t valuei. close to<br />
y -<br />
O<br />
and approach zero for distances y which are large i.n compari-<br />
son to the modulus of C(w,k=O): d.;:.rn,-Lc. do-,,o. :.,<br />
----. - ---. .<br />
A.
-- Structural soundi~z with station arrays: - If not only the source<br />
field but also the resistivity structure are laterally non-uniform,<br />
no generally valid linear relations between normal and anomalous<br />
field components exist. In the f~llowing it will be assumed that<br />
the source field is either quasi-uniform or for a given frequency<br />
well represented by a single set of wave numbers. Then the rela-<br />
tions between the various field components can be expressed by<br />
linear transfer functions in the frequency-distance (w,R) domain.<br />
They can be formulated either for the "normal" horizontal field of<br />
the whole array or, if necessary; also for the local horizontal<br />
field only. In either case it is necessary to remove from the ob-<br />
served vertical magnetic component its normal part in accordance<br />
to the normal resistivity dis.tributi.on and the source field struc-<br />
ture. It is assumed that the normal depth of penetration is smal.1<br />
in comparison to the scale length of non-uniformity of the source<br />
~wI,:~I. ;%pi i
. .<br />
These sets of tPansfe$fUnCtions renresek the input ior the model<br />
calculations to explain the anomalous fields H<br />
-a<br />
a resistivity anomaly<br />
or E<br />
-a<br />
in terms of<br />
pn(z) being known.<br />
P,(x,Y,~) = P - Pn(z),<br />
There exist various constraints about acceptable sets of transfer<br />
functions, for examples the functi.ons in either one column of W<br />
must describe a magnetic field of solely internal origin as dis-<br />
cussed below.<br />
The basic complication in the presence of 3-dimensional structures<br />
is due to the fact that the TE and TM mode of the anomalous field<br />
cannot be separated. This separation is possible, however, if the<br />
resistivity structure is 2-dimensional, say<br />
Pa = Pa(yl>z).<br />
If then the normal E-field is linearly<br />
.!<br />
, ,<br />
.. .. .l.<br />
,,' '<br />
:.<br />
;, ;/ ... Y<br />
polarised in xl-direction, the anomaly ,/' ,{ ,> ,,<br />
contains only TE modes with E parallel 1:; %<br />
-a ,,; 71 Y'<br />
A<br />
to x1 and H in planes XI-constant:<br />
-a<br />
E-polarisatio~, If the normal. magnetic field is paral.lsl to x', the<br />
anomalous field is in the TM mode with zero ~nagnetic fields above<br />
the ground. Hence, Ha = 0 for z = 0 and E lies in planes x1-<br />
-a<br />
const. : H-polarisation Ccf. Sect. 7. 3 ). The new sets of transfer<br />
functions in (xl,yl,z) coordinates are given hy<br />
>?here the subscript (\I refers to E-polarisation and -the subscrip-?<br />
(l.) to ~-~olarisation. The resul-ting symmet~y relations for W and<br />
Z in general (x,y,z) coordinates can be derived as follows: Let<br />
be rotation matrices, transforming .- E and -- H fron: (x,y ,z) -to (x' ,yl ,z)<br />
coordinates, c = cosa-and s = sina:<br />
, ,<br />
, !<br />
..,
Since<br />
it follows that<br />
- E 1 = - T E HI= T H -n . ' -a<br />
H ' = T<br />
H<br />
H<br />
n '<br />
E' = Z'HA = Z'TH = TZHn,<br />
- - - n<br />
Z = T-I Z' T.<br />
In a similar way it is readily vkrified that<br />
Hence,<br />
which implies thkt.<br />
W + W =\ih,, and 2. - z =.Z?-Z-~<br />
XX Y Y XY YX<br />
are invariant against rotations and that<br />
"Skew" parameters which characterise the deriation from true 2-<br />
dimensionality are the moduli of the expressions<br />
If a geomagnetic sounding at a single site has been performed near<br />
such a 2-dimensional structure, the relations which connect A and<br />
.B with WZx and W reduce to<br />
ZY<br />
If -the first relation is multi.plied by c and the second relation by<br />
s and then the sum of both is taken, it follows that<br />
This implies that the in-phase and out-of--phase induction a17ro:.?s<br />
will be everywhere perpendicular to the trend of the structure,<br />
which can thus been found.
In magnetotelluric soundings information about the trcnd comes from<br />
the fact that<br />
However, after Z has been rotated into Z' for the thus found angle a,<br />
no distinction is possible which one of the off-diagonal elements of<br />
Z' refers to E- and H-polarisation. If in addition the direction of<br />
the geomagnetic induction arrow is known, such a distinction becomes<br />
possible. Furthermore (z,, [ > [ zLl ,if the structure is better conduc-<br />
ting than its environment and vice versa. Hence, it can be decided<br />
whether the induction arrow points toward a well conducting zone or<br />
away from a poorly conducting zone. .<br />
The dj.sti.nction between the impedances of E- and H-polarisation is<br />
important, if for,a first estimate a 1-dimensional interpretati.on of<br />
.7<br />
Z by a layered. substratum is made. Such an interpretation may give<br />
meaningful results for Zih, but in general no$ for ZL. A test for the<br />
proper choice of the impedance element for a ?-dimensional interpre-<br />
tation comes from the fact that t~ithin a given area Z,, varies less<br />
from place to place than Z1.<br />
A test for self-consistency of the transfer functjons Wh and Wz alonj<br />
a profile y' perpendicular to the strike of a 2-dimensional struc-<br />
ture arises from the purely internal origin of the anoixaly:<br />
WZ = K x Wh and 9Jh = -K x WZ<br />
denotes a convolution of W (or W with the kernel. function l/ny.<br />
h z<br />
-- 8.3 S~ectral Analysis of Geoma~netic 'InductLon Data -<br />
The objective of the data reduction in the frequency domain is the<br />
calculation of transfer functions. They linearly relate functio!?s<br />
.Iv I-<br />
of frequency a field component Z -to one or more other fiel$compo-<br />
nents X, Y , . . . . Let Z(t), X(t) and Y(t) be the observed time Va-<br />
riations of Z, X, -f during a time i.ni-ervall of length T either fpojn
- - .<br />
the same or from different sites. Let Z(w>, X(w), Y(w) be the Fouri-er<br />
transforms of Z(t), XCt) and Y(t). Then a linear relation of the<br />
form<br />
- -<br />
is established in which A and B represent the desired transfer func- -<br />
tions between Z on the one side and X and Y on the other side; 6Z is<br />
the uncorrelated "noise" in 2, assuming X and Y to be noise-free.<br />
As the best fitting transfer - functions will be considered those whicf<br />
produce minLmum noise < 16zI2 > in the statistical average. Here the<br />
average is to be taken either over a number of records -- or within<br />
extended frequency bands of the width which is L times greater<br />
than the ultimate spacing IL'T of individual spectral estimates. The<br />
noise: signal ratio defines the residual e(w) ,%atio of related to<br />
observed signal the . --- coherence R(w):<br />
The coherence in conjunction with the degree of freedom of the<br />
- -<br />
functions A and B.<br />
averaging procedure estab1ishe.s confidence limits for -the transfer<br />
The averaged products of Fourier transforms are denoted as<br />
S<br />
ZZ<br />
., -<br />
= < Z Z * >: power spectrum of Z.<br />
- -<br />
S = < Z Y * >: cross spectrum between Z and Y<br />
ZY<br />
wi-th S = S* .<br />
ZY Y =<br />
In summary, the data reduction involves the following steps<br />
(a) Fourier transformation of tiine records<br />
(b) Calculations of power and cross-spectra<br />
(c) Calculation of transfer functions<br />
(d) )COlculation of confidence limits for the trans-<br />
ferfunctions.<br />
Steps Ca) and (bf can be substi.cuted by the foll.owing alternatives:<br />
(ax) : Calculate auto-correlation functions ) , .. . and cross-<br />
correlation f.urictf.ons R , . . . with T being a time lag,<br />
ZY<br />
with 0 < T < 'rmax.<br />
R (T) = / Z(+-t) Z(t)dt<br />
Z z T<br />
R (T) = / Z(+-t) YCt)dt<br />
ZY T<br />
x<br />
(b ): Take the Fourier transforms of the correlation functions<br />
and obtain as in (b) power- and cross-spectral estimates,<br />
if the averaging is done within frequency bands of the<br />
width m. There is a formal correspondence between the<br />
-- -1.<br />
maximum lag and Af .<br />
The actual performance of the steps (a) to (d) with one or more<br />
setsof records is now described in detail.<br />
-- a. Fourier transformation. - The time series and their spectra are<br />
given, respectively to be found, at discrete values of t and w,<br />
which are equally spaced in time and frequency. Let Z e-tc. be an<br />
n<br />
.instantenuous value of ZCt) for t = tn, n = 0,1,2, . . . N with<br />
At = tntl - t . Outside of the record, extending from t to t<br />
n o N<br />
Z(t) is assumed to be zero.<br />
-<br />
Let Z be the Fourier transform of Z(t) for the frequency f = f .<br />
n1 IT1<br />
Because of the finite length T of the rec6rd the lowest resolvable<br />
frequency and thereby the frequency spacing w i l l be given by<br />
T-I = Af = f and f has to be a multiple of Af. Because Z(t) is<br />
1 m<br />
gj-ven at discrete instances of -time, A t apart, the hi.ghest resol-<br />
-,<br />
vable frequency, called the Ny quist frequenz, w i l l be the reci-<br />
procal of (At-2), i.e.<br />
and<br />
- I<br />
fH - - = MAf<br />
2At
The Fourier integral<br />
- +- -io t T -iw t<br />
n<br />
z(un) = I Z(t)e<br />
dt = J Z(t)e<br />
n<br />
dt<br />
-@2 0<br />
will be evaluated r.umerica1l.y according to th,e trapezoidal formula<br />
of approximation. Setting<br />
and observing tha-t<br />
zo = $[zct 0 ) + 2(tN)-j<br />
"n?n<br />
= 2a fmtn = -- 2 limn<br />
N'<br />
the discrete Fourier transform of Z(t) is<br />
2<br />
L *<br />
A linear trend of the record within the chosen interval may be<br />
wri.tte11 as<br />
with d = Z(tN) - Z(to).<br />
A cor3rection for this trend i.n the frequency domain implies that<br />
the imaginary Fourier transform of Z1(t),<br />
-<br />
is substracted from Z .<br />
m<br />
A second correction arises from the fact that Z(t) does not v?.nish<br />
necessarily outside of the chosen intervall. In that case the<br />
original time function can he multiplied in the rime domain wirh n<br />
weight function W(t) which is zero for t < to and t > tN. The<br />
Fourier transform of the product<br />
W(t) . Z(t)<br />
-<br />
- - - -<br />
is given by7a convolution of Z with the Fourier transform of W:<br />
- 1<br />
W(o) 35 ZCo) - Ff C Wm-& Z; .<br />
A<br />
m<br />
A frequent.1~<br />
used weight function is<br />
' 1 2li T -<br />
WCt) = -11 + cos --(t-t .-<br />
2 T 0<br />
2
which has 'the Fourier transform T for m = 0<br />
"1% 0 71-<br />
t -5<br />
G({l<br />
where <<br />
'm<br />
> is the discs-te Fourie~ transform of W ( t ) * Z(t). This<br />
smoothing procedure of the origi.na1 spectrum is czlled "harming"<br />
after Julius von Hann.<br />
\<br />
\<br />
\<br />
\<br />
\ / -. L- q-4----&<br />
b , I.\/' 3 1<br />
The convolution of W with descrefe values of Z reduces then to<br />
- 1 - -<br />
m T m-m m<br />
-<br />
1-<br />
-(Z. + ZIY m r O<br />
( 2 0<br />
= - C bi A* ZA = m = 1, 2, . . . PI-1<br />
-(Z 4-2 m = M<br />
2 M-1 M<br />
In certain cases it will. be necessary to apply a numerical fi1.ti.r<br />
to the time series to be analysed beforc -. the Fourier transformation.<br />
This filtering process consi.sts i.n a convolution of 2c.t) with a<br />
filter function \d(t) - in the - time do1naj.n and thus corresponds to a<br />
multiplication of Z with W in the frequency domain. But because<br />
the filtering procedure is intended to prepare the time series for<br />
the Fourier transformation it must be carried out in the time domain<br />
If W denotes the filter weight for t = t17, the discrete form of<br />
n<br />
the convolu-ti.on is<br />
Usually even 'filters are employed, W ( t ) = W(-t), to preserve .the<br />
corr-ec-t phase of the Fourier coniponents, The transform of is chose:<br />
in such a way that it acts as a heigh-, lord- or bandpass fj.1-ter for<br />
fr.equency-independent (="whitet') spectra:
The weigh-t . function W is then found by calculating in Four~ieil tl-ansform<br />
of W:<br />
The puTpose of filtering the -time series before making a Fourie?<br />
transformation is<br />
(a! a prewhitening of the spectrum, levelling off peaks and compen-<br />
sating spectral trends<br />
(b) a reduction of unwanted larger spectral values close to zero<br />
frequency,, arising from a background trend in the time series<br />
(c) a suppression of spectral power beyond the Nyquist frequency<br />
to avoid "a.liasingU . Spectral values for the frequencies<br />
*2~-r' f r and . f3M-r>' fM+r' fH-r etc.<br />
are undis-tinguishable (r = 0,l i2, . . . , M-1) :<br />
. Cal'dulation of porter and cross spectra: Let P denote a product<br />
- X - - %<br />
- - m<br />
ZmZIn, ZmYm etc. for a frequency f m = 1,2 . . . .1q. The calcul.a-t'iorl<br />
m'<br />
of power and cross spectral estimates from real. data requi-res that<br />
such produc-ts are averaged with 2a cer-tain degrees of freedom, !C<br />
Suppose -th.a-t the available ~?ecord leng-th just gives the desired re-<br />
- 1<br />
sol.ution Af 2 T for the transfer functions to be determined. In<br />
this case spectral products Pm have .to be derived for a number cf<br />
individual records and then to be avermgcd. Average the sp~:ctral.<br />
products and - not the spectral values! The resulting mean value<br />
is the desired porier or cross spectral es-ti.mate. If L records have<br />
been used, \? = 2L besaucje the sine- and cosine.tr.-,nsfor.~~ls of the<br />
series c011tr~i.b~-te independently to the calculil'tioris of Sm.
- 1<br />
If the record length T allows a frequency resolution Af = T which<br />
is much smalier than a meaningful resolution for the transfer<br />
functions under consideration, then Pm can be averayed over<br />
L = Z/A~ spectral values within (M/L-1) frequency bands of width<br />
-.<br />
Af between fo = 0 and fM = MAf.<br />
Taking the average of Pm withri.n frequency bands implies that EL nu-<br />
merlical filter Q(f) is applied to P(f), yielding the desired power-<br />
cross spec-trum by convolution:<br />
. +M<br />
S = Q x P .- A f L Qm-; P;'<br />
m<br />
-M<br />
(P = 'pi). The filters to be used are even functions of f, bel-<br />
-m<br />
shaped and with zero values outside a range which is smaller then<br />
M Af. Setting Q(f) = 0 for f > f Q'<br />
m<br />
m~<br />
S = Af{ C QA(P m m+m<br />
G=l<br />
m-m<br />
A + P A) + QoPm} .<br />
I I<br />
0.1. 48(<br />
The resultring smoothed spectral estimates Sm have been derivedwith<br />
-<br />
roughly L = &f[Af degrees of freedom. The effective degrees of free-<br />
dom pmay be somewhat smaller, depending on the frequency dependence<br />
of the filter. They are defined as<br />
.y'=<br />
2 varCu)<br />
var (uq)<br />
w11,ere varCu) denotes the variance (=dispersion) of a random va-riable<br />
uCf) and varCuq) the variance of Q x u, hence
Two convenient filters are<br />
3 sinx I,<br />
.the Parzen filter Q(f) = - T = Min!<br />
'626~ m m<br />
Hence., the derivatives of S6z6z with respeci - to ,. the real and imaginary<br />
parts' of the transfer functions A and B have to be zero:<br />
1% nl<br />
aS~z6~ +i---- aS&z6z = = 0 etc.<br />
aRe(An) a1rn(An),
using the notations as - introduced - above. The subscripts m are ommittec!<br />
Elimination of either A or B yields the basic formula& for the de-<br />
termination of transfer functions in geomagnetic and magnetotelluric<br />
The coherence can be derived from<br />
-. *<br />
R' = (AS x z + BSyZ)/SZz<br />
as it is readily seen from<br />
If only a relation between z and X 01: Y. is sought or if X and Y<br />
are linearly independent (S = O), then the above derived relations<br />
XY<br />
reduce to<br />
~2 = a---<br />
S S S<br />
xx yy zz<br />
d. Cal.culation of confi-dence liniits -a,- for the transfer functions<br />
--<br />
- "<br />
In order to establish confidence limits for the transfer functions<br />
A, B of the previous section it is necessary .to fj.nd -the probabili-ty<br />
density func-tion ("pd f") of their devia-tioils fr>om their "true"<br />
" - - -<br />
values Ao, Do. W; assulne that esti.mates of A and B accordi.ng to the<br />
least-square method of the previous sec-tion are wj:thout bias, i. t?.<br />
bl?th~~t sys-tematic errors (E = expected value):<br />
I-lencefor-th, random variables will be wri.tten with capj:l-a1 letters
(e.g. X), their realizations by observation with lower case letters<br />
(e.g. x), and their true values with the subscript "On (e.g. x0).<br />
Let f(X) be the pdf of a variab1.e X. Then the probabi.lity that a<br />
realisation x exceeds a value G is<br />
m<br />
a = J f(X)dX.<br />
G<br />
Hence, there is a "confidence" of<br />
f3 = (1-a) . 100% that x does not<br />
exceed G.<br />
The following pdf's will be needed in this section:<br />
k" C;<br />
(1) The "Gaussian Normal Distribution" of a normalized variable<br />
with the dispersion 1 and zero mean value:<br />
(2) The X2-distribution for v degrees of freedom:<br />
, (3) The Fisher-distribution for N and V2 degrees of freedom:<br />
1<br />
'These distributions are encountered in the fol.lowing problems:<br />
Consider a random variable X which is normally distributed with<br />
Let - 1 n<br />
X = -<br />
n C Xi<br />
i=l<br />
be the, realized .sample mean value of n realizations of X. These<br />
mean val.ues .are also normally distributed with the dispersj.on<br />
accordj-ng to the "central 1.imi-t theorem".
Let<br />
be the realized dispersion within a sample of n observations.<br />
Then the normalized dispersion<br />
w i l l be a random variable with a X2 distribution for U = n-1<br />
degrees of freedom. Let S: and S: be the dispersions within<br />
samples of nl and n2 observations of two different variables<br />
X1 and X Then the ratio of their normalized dispersions<br />
2'<br />
w i l l have a Fisher-probability density distribution E (n-1,n-2).<br />
F<br />
We regard now the Fourier transforms 2, g, i' as random variables<br />
Z, X, Y and denote their real-ization for a single record or a<br />
single frequency component as x, y, z. In a similar way, a and b<br />
shall be the realized transfer functions for a limited number of<br />
records or a limited frequency band G, their true values being<br />
a and bo. We assume now that Z depends linearly only on X and<br />
0<br />
that X is observed error-free:<br />
Then<br />
z = a x +6zo<br />
0 0<br />
6zo is the "realized" not correlated part of z for a single<br />
record. The variable Z shall be normally distributed with the<br />
dispersion<br />
var (Z) = s:,<br />
which w i l l be also the dispersion of 6Z:<br />
By minimizing the uncorrelated power = S for a number<br />
6262<br />
of records or for a number of frequency lines within a frequency<br />
band of the width a realization of the transfer function A is
Observe that the residual, of which -che power has been brought to<br />
a minimum, is<br />
6z=ax - z<br />
0<br />
in contrast to the "true" residual<br />
The random variable<br />
TI I:<br />
11 <br />
has a x2-distribution with n degrees of freedom, where n is the<br />
number of records or the number of lines used to obtain the<br />
average.<br />
If a variable U with a ~~ldistribution of v degrees of freedom is<br />
decomposed into two components U and U2,<br />
1<br />
U1 and U will be likewise x2-distributed and the sum of their<br />
2<br />
degrees of freedom w i l l be v:<br />
This decomposition will be carried out now with U as defined<br />
above: Clearly,<br />
The power of the averaged residual has been mi-nimised by setting<br />
implying that .<br />
u1 -<br />
- .7i<br />
-.<br />
s1<br />
0 . .<br />
has a X2-distril>ution of n - 2 degrecs of fr?eedoni because
must be a distribution of 2 degrees of freedom in v-iew of -the<br />
real and imaginary part of (A - do). The ra-tio<br />
has a Fisher-distribution fF(2, n-2). Using the notations fr'om<br />
above,<br />
and observing that<br />
with R' = [S ["IS S ), the Fisher-distributed ratio becomes<br />
zx zz XX<br />
We have found now the pdf of the deviation of A from its true vzlue<br />
to be in terms of its modulus<br />
Let<br />
G<br />
8 = / fF(2, n-2)dF . .<br />
0<br />
be the probability that F does not exceed the value F = G, observing<br />
that fF is a one-sided pdf. Then there w i l l be the probability f3<br />
that the modulus of A lies within the confidence limits'<br />
in the complex plane.<br />
E<br />
1 = I A I R Jg<br />
The threshold value G for a desired value<br />
of f3 can be obtained in ciosed form because<br />
G<br />
..'<br />
xc [A\<br />
^
Example: n = 12 and @ = 95%:<br />
1<br />
n = 12 and @ = 99%: I/-- G = ~'1003 - 1 = ~43 = 1.22<br />
11- 2 1 ' -<br />
2<br />
5<br />
n = 12 and B = 50%: 4 - - G =<br />
11- 2<br />
42 - 1 = a = 0.39<br />
1<br />
For large n and n; >> lny (y = -1, the following approxiinations are<br />
1-F<br />
valid :<br />
- 2 - 2<br />
G - - lny<br />
n-2 n<br />
This approximation. exemplifies the general propaga-ti-on-of-error law,<br />
namely that the errors are reduced proportional. to the square root<br />
of the number of observations.<br />
Ln the more general case that Z depends on X and Y with a non-zero<br />
coherence between X and Y confidence limits can be obtained in a<br />
similar way: Let<br />
z = a x + boyo + 6zo,<br />
0 0<br />
assuming that now X -- and Y are real-ized error free. Then the Fisher-<br />
distributed ratio, involving the deviation of A and B from a and<br />
0<br />
b ; turns out to be<br />
which has a f (4, n-4) dis.tri.buti.on. The confidence limi-ts for A and<br />
F<br />
R cannot be calculated individuall.y, unless of course S is taken<br />
"jr<br />
to be zero. On the other hand, if we assume that I~-a,l and I R-boI<br />
are equal, Chen
whi.ch allows the determination of cornbined confidence limits for<br />
A and B. The threshold value of F for a given probability fi can<br />
be derived from<br />
1-6 = - 1<br />
(1 t -- 26 - n- 4<br />
) - (- ) with m = -<br />
2 '<br />
2 m<br />
Cl + %G) m+2G<br />
8.4. Data analysis in the time domain, pilot studies<br />
Suppose the time func-tions to be related linearly to each o-ther are<br />
not oscillatory but more like a one-sided asymtotic return to the<br />
undisturbed normal level after a step-wise deflection from it:<br />
It may then be preferable to avoid a Fourier transformation and to<br />
derive i-nductive response functions in the the domain.<br />
If Z depends linearly only on X, i.e.<br />
*<br />
Z(w) = A(w) X(m) + 6ZCw)<br />
in the frequency domain, then Z(t) will be derived from X(t) in<br />
the time_domain by- a convolution of X(t) with the Fourier trans-<br />
form of A,<br />
f"<br />
1 - i.wt<br />
A(t) = - J A(w) e dm.<br />
2s -m<br />
Since ZCt) cannot depend on X(t) at some future time T > 't, the<br />
response function Act) must be zcro for t < 0, yielding<br />
-
-<br />
As a consequence, the real and imaginary part of A(w) wil.1. be rel.ated<br />
as functions of frequency by a dispersion relation:<br />
1<br />
with K(w) = .<br />
The time-domain response Act) is now determined frorn a given record<br />
of Z(t) between t = 0 and t = T by minin>.izing'{6~(t))~ within this<br />
intervall:<br />
. . 1.t is assumed that X(-t) is known also for t < 0. Differentation with<br />
h<br />
respect to ACt=t) gives<br />
Let again Z = ZCt = n.A-t) denote the value of Z at equally spaced<br />
n<br />
instances and let A be the value of ACt') for t' = n'At. Then the<br />
n'<br />
replacement of the integrals by sums yields a<br />
-<br />
system of ].inear<br />
equations for the determination of the A,,, n' 0,1, ... N:<br />
N' N N<br />
c An,EZ x X -1 = c zn xn ;<br />
n-n' n-n +.<br />
n' =o n=o n=o<br />
A A<br />
for n = 0,1,2, ... N.<br />
A<br />
Chosing N < N, the response values can be de+enninc?d by a least-<br />
square fit. 0-therwise a generalized inversior~ of the system of<br />
eyuations can be performed.<br />
Consider the limiting cases that<br />
. .<br />
(1) the frequency response is real ;ind independent of<br />
frequency,<br />
(2) -the frequency response is imaginary and proportional<br />
-1.<br />
to w .<br />
The first case is real-ized by the inductive :scale 1-eng-tl, C above<br />
n<br />
X<br />
a perfect conductor at -the c1cp.th z or by .the i.mpedance for, a thin
sheet of the conductance T above a non-conduc~ing substriit~m:<br />
3:<br />
Cn = z , Zll = 11~. The second case can be realized by the same<br />
models, interchanging Cn and Zn (cf. See. 9-11:<br />
- . -<br />
The Fourier transforms of A1 and A2 are<br />
02<br />
sinwt -<br />
A (t) = b .f x- - sgn(t)nbo<br />
- 2 O-o;,<br />
Hence, in the first case Z depends on X at the same instance of<br />
time only, in the second case on X with equal weight during the<br />
entire past:<br />
These simple relations are very useful to conduct a pilot reduction<br />
of sounding data. Case (1) applies to magnetotelluric soundings in<br />
sedimentary basins, underlain by a crystali5~ne basement, to ver-<br />
tical geomagnetQi.c soundings with very long periods, reachirig the<br />
conductive part of the mantle, and to structural geomagnetic soun-<br />
dings where the perturbed Plow of induction currents is in-phase<br />
with the normal magnetic field.<br />
The baste linear relation for structural geoinagnetic soundings can<br />
be written then si-~nply by products in the ti.rne domain:
This relati-on implies .that .the local geomagnetic d
In an "Untiedt diagram!' the endpoi.nts of the horizol:tal dis-turbance<br />
vector (El H ) are drawn as a curve in (x,y)-coordinates during a<br />
x Y<br />
single event. Lines which connect time instances of equal ver-tical<br />
disturbance M w i l l then be para]-lel to the.intersecting line of<br />
i. z<br />
the prefer~ed plane with the ,(x, y) -plane. The "induction arrows'!<br />
are normal to these lines and their length is given by the ratio<br />
of Haz t o the simultaneous horizontal disturbance vector, projected<br />
onto the direction normal to the connecting lines.<br />
.Parkinson and Wiese diagrams which utilize the H az) I-Ix' H relation:<br />
Y<br />
of nuinerous individual events can be employed also in the case of<br />
2-dimensj.ona1 structures wi-th an anomalous field which is not<br />
exactly in-phase with the normal field. Then more 01' less sinilsoicla'<br />
variations are chosen with readings at well defined times, e.g. at<br />
the time of maximum deviation of Haz from the undisturbed level.
9. ---<br />
Data 5.nterpretatj.on on the basis of selected -- iriodels<br />
9.1 - Layered H a l f - s ~<br />
The interpretation of geomagnetic sounding data by a layered re-<br />
sistivity distribution p = p (z) is appropiate at those single<br />
n<br />
sites or for those arrays which do not show anomalous magnetic<br />
Z-Variati.ons and which have a polarisation-independent magneto-<br />
telluric inwedance:<br />
The transfer functions to be used for the interpretation are<br />
the inductive scale length C (w,O)<br />
n<br />
for zero wavenumber<br />
the impedance<br />
the ratio'of internal to<br />
external parts in the<br />
wavenuinber domain<br />
The transfer function of the Y-algorithm to be uscd for the lineari-<br />
sation of the inverse problem is<br />
where p is arbitrary reference resistivity Ccf. chap. 6.4)<br />
0<br />
Considering these transfer functions a.t one particular frequency ><br />
the parameters of the following models can be derived directly<br />
from tl-tem. These parameters may then be used to resen~ble as func-<br />
tions of frequency certain characteristics of the true resistivity<br />
distribution. All formulas are readily derived from the formulas<br />
for TE-fields at zero wavenumber in chap. 7.3, annex.<br />
(a) Single frequency interpretation of the modulus of transfer-<br />
functions (Cagniard-Tilchonov): the model c0nsi.s-ks. of a uniform<br />
half-space. Its resistivity is
if the period T is inensured in seconds, E in mV/Icm and H in y.<br />
X Y<br />
is the "Cagniardapparent resistivi'ty" of the substrat~uil<br />
pa at<br />
the considered frequency.<br />
(b) Single<br />
1<br />
frequency interpretation of the real part of Cn:<br />
The model consists of a perfect conductor at<br />
W X A<br />
f=m :* layer.<br />
Im(Z<br />
.. .<br />
3:<br />
~ , , ~.<br />
n<br />
, .. ;\<br />
z = R~IC I = --<br />
\ \ \ g ,. = 0 ., , n<br />
\. .,, \\... u"c<br />
is the depth of the "perfect substitute conductor" at the con-<br />
\ .'.., \ . X<br />
the depth z = z beneath a non-conducting -top<br />
sidered frequency, indicating the depth of penetration into<br />
the substratum at that frequency.<br />
(c) Single frequency interpretation of the imaginary part of C :<br />
n<br />
-7- %* The model consists of a thin conductive top -<br />
;' f - '-<br />
layer of the conductance I-~ = 9 cr(z)dz,<br />
\<br />
\. , . .,<br />
0<br />
covering a non-conducting half-space. This ,<br />
\<br />
\ . \ \<br />
apparen-t condoctance is given by<br />
(d) Sing1.e frequency interpretation of amplitude - and phase of resp3nse<br />
functions; 4 = arg{Zn}. 1 0 < @ < ~ / 4 : The model consists of<br />
X<br />
a thin top layer of the conductance T above a uniform substra-<br />
tum OF resistivi-ty p"":<br />
-ER<br />
TX =<br />
p"" = PaS<br />
-Im(C n ) - Real(C n ) , -- Re(Zn) ,- Im(Zn)<br />
--<br />
ullo ! cn I *<br />
1znI2 -<br />
1 + cotZ@<br />
2<br />
2 n/]; < @ < ~ / 2 : the model consists of a non-conducting 'top --<br />
layer of thickness h above a uniform substratum of the re-<br />
X<br />
sistivity p ;<br />
Im(Zn) - Re(Zn)<br />
h = Real(C ) + Im(C,) =<br />
n<br />
u"o<br />
--<br />
p" = 2 wl,;{l:rn(~~)~~ = 2 : Pa 5<br />
, . 1i.tan2@<br />
'
The "modified apparent - - resistivity" p can be used as an estimator<br />
I r.<br />
of the resistivity at the depth zf= Re{Cnl = h 4 ~p with<br />
m<br />
,.<br />
pX = 1 f i k as apparent skin-depth of the substratum. The representation<br />
UlJ0<br />
of the true resistivity distribution p(z) by the modified apparent ,<br />
x<br />
. resistivity distribu-tion p CZ~)'~S usually adquate, if p decreases<br />
with increasing depth. It is-less satisfactory, if p increases with<br />
depth.<br />
(e) Multiple frequency interpretation with a three-layer model:<br />
'Is<br />
-<br />
L<br />
X<br />
This interpretation cornbines the single<br />
frequency in-terpretat3.ons (b) and (c) , i. e.<br />
-TL-.A,.~.-'- I-.-<br />
the model consists of a thin top layer with<br />
2 h dl<br />
the conductance T , a non-conducting inter-<br />
mediate layer of the thickness h above a<br />
perfectly conducting substratum. Then<br />
= wp ~h = is the induction parameter of the model wi-th dl<br />
0<br />
pi<br />
as thickness of the top layer and pl = /l& as its ~%in-depth<br />
W0T<br />
with the requirement that pl > 1, implying that h >>> p the induction is predominan-t1.y<br />
1'<br />
j.n the surface sheet, the response functions<br />
al-e independent of h, thus allowing a unique cleterrr~ination of T.<br />
Ln particular,<br />
Pa<br />
-. -<br />
1<br />
W ~ J ,r2<br />
0<br />
Hence, if y = log pa/p and x = l.og T/To (T = period; po and To<br />
0<br />
are reference values), then<br />
e .<br />
the p vs, period curve in log-log prcsen'cati.on is a straigh-t<br />
a<br />
4 b?i.~:h w as curve parameter<br />
-
0<br />
line ascending under 45 , its intersection point on the x-axis<br />
-<br />
given hy<br />
Po .r2)<br />
= log(21ry -<br />
To<br />
with TI = 1 . lox1 in seconds and po in Qm.<br />
0<br />
If, on the other hand, 11~
Exercise<br />
Geomagne-tic varj.ations. in H (north component), D (east component)<br />
and Z (vertical component) have been recorded at Goettingen from<br />
September 5, 17 h to September G, 2h GMT 1957. During the same time<br />
interval1 records have been obtained also of the telluric field in<br />
its northcomponent EN and its eastcomponent EE. Scale values and<br />
directions of positive change are indicated.<br />
(1) Determine the magneto-telluric impedances EN/D and EE/H for<br />
fast (period 1 10 min) and slow (period 1 1 hour) variations,<br />
evalua-ting peak value readings of pronounced deflections from<br />
the smoothed undisturbed level.<br />
Calculate the apparent resistivity pa and estimate the phase @<br />
0 0 0<br />
of the impedance (0 , f 45 , + 90 ).Interpret pa and @ with<br />
one or two layer models for the two period groups separately.<br />
Then search for a model which could explain both period grLcups<br />
by calculating the response function y and solving a system<br />
o'£ two to four linear equations.<br />
(2) The magnetogram shows pronounced variations in Z which in view<br />
of the low latitude of Goettingen can be interpreted as being<br />
due to a resistivity anoma1.y. Check the correlation of Z with<br />
H. and D by visual inspection. Then choose effects with<br />
distinctly different K : D ratios and determine the coefficient<br />
A and B for the anomalous vertical field Z = AH + BD.<br />
Interpret the result.
. , . ' .,<br />
, . - ~a~neto~ran?;l~<br />
Gottin'gen 1953 September 5-6<br />
.i , $ .
9.2 Layered Sphere -<br />
The sphericity of the real Earth has to be taken into account,<br />
if for the considered frequency the depth of penetration is<br />
- 'comparable to the Earth's radius a divided by N. Here N is the<br />
highest value of the degree n of spherical surface harmonics<br />
which describe the surface field as functions of latitude and<br />
longitude. There is a formal correspondence between the wave<br />
number k and n/a or better Jn(n+l)/a (S. belobr).<br />
Only the long-period variations S and Dst p enetrate deeply enough<br />
9<br />
to make a spherical correction of their plane-earth response<br />
functions necessary. The Dst source field is effectively of the<br />
degree n = 1, while the Sq source field contains spherical har-<br />
monics up to N = 5. In fact, the surface field of the m'th Sq<br />
subharmonic is well described by a spherical harmonic of the<br />
degree n = m + 1 when m ranges from one to four. The ra-ti-os a/N<br />
are therefore roughly 6000 km for Dst and 3000 to 1200 km for Sq.<br />
These values have to be set in relation to the depth of penetration<br />
of Dst from 600 to 1000 km and to the depth of penetration of S 9<br />
from 300 to 500 km.<br />
Hence, in either case there will be the need of a spherical correc-<br />
tion, but it should be noted that it will be bigger for S because<br />
9<br />
the greater spatial. non-uniformity of S outweigh-ts the effect<br />
9<br />
of deeper penetration of Dst.<br />
The theory of electromagnetic induction in conductors of sphecical<br />
symmetry surrounded by non-conducting matter can be summarized<br />
as foll.ows. Let<br />
be the magnetic and electric field vectors of time-harmonic TE-<br />
mode fields in spherical geocentric coordinates; r is the geo-<br />
centric distance, X the angle of longitude and 0 the angle of<br />
' 0<br />
co-latitude (= 90 minus latitude) on a spherical surface. Tlte<br />
diffusion victor from Sec. '7.3 points from external sources<br />
r5adi.all.y inwards toward r = 0:
with<br />
and<br />
rot rotP= -<br />
Pr is expressed on a surface r = const. by a series of spherical<br />
surface harmonics. The diffusion equation V'P - = iwuouv '2 within<br />
the v'the Shel.1 is then solvable by .i separation of variables<br />
(K: = iwpouv):<br />
, .<br />
where the characteristic radial function fm satisfies the ordinary<br />
11<br />
second order differential equation<br />
Its general solution are spherical Besselfunctions jn and n of<br />
n<br />
the first and second kind,<br />
. .- with A~ n and B: as complex-valued constants OF in-tegration. Theym<br />
m<br />
decribe the in--going (B ) and out-going (A ) solu-tion, the latter<br />
n n<br />
being zero below the ul.timate shells, surrounding an uniform<br />
inner core, because nn has a singularity at r = o while jn + 0<br />
for r +- 0.<br />
i<br />
The characteristic scale 1.engi-h of pene-tration is defined as i.<br />
with<br />
m m .<br />
cm = r fn/gn<br />
n<br />
d(r f:)<br />
- -<br />
tz: - dr
The field within the conducting sphere, r - < a, is derived from - P<br />
as described in Sec. 7.3. Observing that the radial component of<br />
rot rot - P reduces by the use of spherical harmonics to<br />
the field components of the spherical harmonic of degree n and<br />
order m are:<br />
and<br />
The impedance of the field at spherical sur'aces, expressed in<br />
terms of c:, is then as in the case of plane donductors given by<br />
For the field outside of the conducting sphere, r - > a, the solution<br />
of the radial function is (in quasi-stationary approximation)<br />
Hence, the surface value of the charlac-terisll-ic scale length for<br />
r = a is found to be<br />
The ratio of internal. -to external. parts of the magnetj.~ surfacc<br />
fleld is by definition
yielding<br />
The ratio of internal to external parts is therefore derived from<br />
cm according to<br />
n<br />
which demonstrates the r$le of n/a, respec-tive1.y (n+l)/a, as equi- c<br />
valent wavenumber of the source fielh in spherical coordinates<br />
(cf. Sec. 9.1).<br />
The spherical version of the input function for the inverse problem,<br />
based on a linearisation according to the $-algorithm, is con-<br />
veniently defined as<br />
with K~ = itopO/pO. It w ill be shown that the spherical ccrrection<br />
0<br />
is of the order (n[~[/a)', if the conductivity is more or less<br />
uniform.<br />
If degree and order of the spherical. harmonic representation of<br />
the source field are independent of frequency (this is true for;D st<br />
but -- not for S ), the inverse problem can be solved by deriving<br />
9<br />
from spherical response functions a prel-iminary plane-earth model<br />
h<br />
This model. is subsequently transformed into a spherical. Earth<br />
model p(r) with WEIDELT's transformation formula:<br />
with
and<br />
An algorithm for the direct problem for spherical conductors can<br />
be formulated as follows. It is designed not to give the transfer<br />
function C: for a given model itself but an auxilary transfer<br />
function<br />
as a direct equivalent to the tranfer function C of plane conm<br />
ductors. To see its connection to C differentiate the characteristic<br />
n'<br />
depth function f: with respect to r, observing that the differen-<br />
tiation of spherical Besselfunctions with respect to their argu-<br />
ment u j.s<br />
and the same for n . Then<br />
n<br />
Hence,<br />
d fm<br />
m = fm + = fm(r/?m - n)<br />
gn n n n<br />
with nC /r as "spherical correction ",<br />
. . n<br />
Continuity of the tangential components of the electric and<br />
A<br />
m<br />
magne.tic field requires that Cn and thereby cm as well are conn<br />
tinuous functions of depth. Let r<br />
L<br />
be the radius of the inner<br />
core of conductivity a surrounded by (L-1) uniform shells, the<br />
L'<br />
-yth shell between r<br />
and rvi-,.<br />
(v = 1,2, ... L).<br />
v
Then<br />
with uL = iKL rL because the in-going solutions are zero within<br />
m<br />
the core. But C for r = r can be expressed also in terms of<br />
n L<br />
thd general solution of the (L-1) 'th shell:<br />
*<br />
m<br />
with ui = iKL-l rL. AS a consequence, C for r = rL-l is now ex-<br />
,. n<br />
pressible in terms of c:(P~) and thus an algorithm can be estah~<br />
~<br />
blished for the calculation of the surface value ~z(r = a). In<br />
1<br />
numerical evaluation it is preferable not to derive the spherical<br />
Besselfunctions for the arguments u and u' -themselves, but to<br />
V v<br />
reformulate the algorithm by expressing jn and nn in terms of<br />
hyperbolic functions (cf. Dover Handbook of Mathematical Function,<br />
FoI?mula 10.2.12):<br />
n<br />
jn(u) = i {gnCz) sinh(z) + g-n-l (z) COS~CZ))<br />
with u = iz and the following recurrence formula for the<br />
gnCn = 0, + 1, + 2 ... ):<br />
- -1<br />
Lf I z 1 >> 1, the approximations z for n = f2, + 4, ... and<br />
-<br />
g n<br />
- -z2 . Cn+l).n/2 for n = + 3, + 5, ... are valid; g-l = 0.<br />
gn<br />
By making use of the addition theorems of hyperbo1.i~ functions<br />
the algorithm can be given a form simular to the recurrence for-<br />
mula of plane conductors (cf. Appendix to Sec.7.3)
with<br />
I -<br />
and- a = gn g-n I<br />
1 6-n-1 En-1<br />
The argument of g is u = i K r , the argument of g' is ul=iK r<br />
n v v v n v v vt.1'<br />
If luvl and [ub[ are large compared to unity, i.e. if the mesa<br />
radius is much~larger than the skin-depth of the v'th shell., the<br />
-1<br />
coefficients a are of the order +(u ul) , the spherical correc-<br />
n -4 :-I or u-l u1-2<br />
tion coefficients E of the order uv uv<br />
I1 V v<br />
Consider for example the case of a uniform sphere of the radius a,<br />
the conductivity a, K = iwu o and u = iKa. Then<br />
0
Assuming the skin-depth to be small in comparison to a,<br />
( I u I >> I), tanh(u) z.1 and with z = u/i = Ka<br />
The resulting approximate spherical transfer function<br />
has its plane-halfspace value except for a spherical correction<br />
of the order Cn/z)'.<br />
9.3. Non-uniform thin sheets above a layered substructure<br />
- ---<br />
The thin-sheet approximation has been introduced by<br />
PRICE to geomagnetic induction problems. It applies to those<br />
variat%on anomalies for which there is reason to believe that<br />
they arise from l-ateral changes of resistivity close to the<br />
surface. For instance, highly resistive basement rocks may be<br />
covered by well conducting sediments of fairly uniform re-<br />
sistivity ps but variable thickness d.<br />
Two conditions have to be satisfied, if this surface cover is<br />
to be treated as a "thin sheet" of variable conductance<br />
.r = d/ps:<br />
(i) Its skin-depth J2ps/wvo at the considered frequency must<br />
exceed the maximum thickness dmax by at least a factor of 2.<br />
(ii) The inductive scald-length c-(u,o) for the matter below the<br />
n<br />
surface cover, assumed to be a layered half-space, must be<br />
large in comparison to dmax.<br />
Under these conditions the electric field within the "-thin sheet1'<br />
may. be. regarded as- constant for 0 .: z < d, reducing the first<br />
field ec~uatlon to its thin-sheet approximation
with<br />
~ = - T E<br />
+<br />
as sheet current density; Htg is the tangential magnetic field<br />
above the sheet at z = 0, H- the tangential magnetic field<br />
;;tg<br />
below the sheet at z = d; z is the unit vector downward.<br />
1 ... , . .<br />
G' 3<br />
for example a sedimentary cover with<br />
I<br />
exceeding 4 km i n thickness. Then condition (i) will be satis-<br />
fied for periods up to 2 minutes and condition (ii) for deep<br />
resi.stivities in excess of, say, 50 Slm.<br />
The response functions for the normal field above the sheet at<br />
z = 0 are readily found from<br />
where T is the constant normal part of T, while C; has to be<br />
n<br />
derived from the given substructure resistivi-ty profile. The<br />
field equations.to be considered for the evaluation of the va- .<br />
. . riation anomaly are<br />
with<br />
rot E = impo EIa,<br />
Z -a
The second field equation implies that the vertical magnetic<br />
field is also to be regarded as a constant between z = 0 and<br />
z = d.<br />
In addition there are boundary conditions for the anolnalous field,<br />
arising from the fact that its primary sourced lie within the<br />
sheet.<br />
Pilot interpretation: In many cases the anomalous field has no<br />
inductive coupling with the substructure because its half-width<br />
is small in comparison to (c,[. In that case no currents are in--<br />
duced by Ha in the subs-tructure and the anomalous magnetic field<br />
is solely due to the anomalous sheet current ja within the sheet.<br />
For reasons of symmetry<br />
Observe that the inductive coupling of -the normal field has not<br />
been neglected, i.e. its internal sources lie within the sheet<br />
as well as within the substructure. Assuming that T~ is known<br />
and that it is possib1.e to identify the normal part of the total<br />
electric field - E = En + Ea either by calculation 01, by observations<br />
outside of the anomaly, the anomalous part of the conductance T., a<br />
t<br />
is readily found from observations of Hatg and<br />
-a<br />
E . In this way,<br />
using the example from above,the variable dep-ti1 of the basement<br />
bel-ow the sedimen.ks,<br />
can be estimated.<br />
If an el-ongated structure is in H-polarisation with respect to<br />
the normal field, the magnetic variation anomaly will disappear<br />
(cf. Sec. 7.3). The pilot investigation of the conductance is done 11~<br />
even inore simply because the sheet current densj.ty must be a<br />
constant in th-e direction normal to the trend:
E~~ T r: j = const. or E T + E a r = j = 0.<br />
n~ a a<br />
This approach has been used by HAACK to obtain a fairly reliable<br />
conductance cross-section through the Rhinegraben.<br />
In the case of E-pol-arisation a different kind of simplification<br />
may be in order: Suppose tHe half-width of the anomaly is<br />
sufficiently small in comparison to (W~~T)-' everwhere. Then no<br />
significant local self-induction clue to HaZ which produces<br />
takes places, i.e. the electric field driving the anomalous<br />
current will be the large-scale induced normal field only:<br />
Assuming E again to be known, the conductance anomaly is now<br />
-n ll<br />
derived from the observation of the tangential magnetic var,iation<br />
anomaly normal to the trend,<br />
In the actual performance with real data all field components<br />
may be expressed in terms of their transfer functions W and Z<br />
and thus be norrnalised with regard to h+<br />
r1tg'<br />
Let T be variable only in one horizontal direction, say, i.n y-<br />
dlrec-tion'which. implies that the anomal.ous fj-eld is also variable<br />
only in that direction, obeying the field equa-tions<br />
CE-polarisation) (H-polarisation)<br />
-<br />
H+ - 1-1- = H - H.k -<br />
ay ay ?ax ax ax - jay<br />
with<br />
and
Here a denotes the conducti.vity at the top of the substratum<br />
0<br />
and ~a~ the anomalous vertical field at z = d which is responsible<br />
for driving currents upwards from the substructure to the sheet<br />
and vice versa.<br />
The boundary conditions, reflecting the "thin-sheet origin" of the<br />
variation anomaly, are<br />
(E-polarisation) (H-polarisation)<br />
for z = 0: H' = -K x H H+ = 0,<br />
a Y az ax<br />
-<br />
for z = d: H- = LII x Haz E = x Eaz.<br />
ay a Y<br />
c<br />
The kernel function of the convolution integrals - are -n 5<br />
H " ' ~ - K * I ~ ' ~ ~ *<br />
1 r "7 ip H<br />
KCy) = -<br />
=-> "' = Q ~ Y ) - : ~ Y ~ . ~ -<br />
"Y<br />
hy?<br />
q*<br />
c(i) U-<br />
+m sin(y k ) -7 LB% HCi; "I<br />
1<br />
LII(w,y) = - I -<br />
dk<br />
+<br />
't H"b x<br />
" oik Y<br />
Ha): !<br />
Y Cn~=(W'!C Y<br />
7<br />
)<br />
I L E,<br />
Y-,.-r-,--* . ./-:<br />
sin(y k )<br />
m'<br />
1<br />
Y<br />
LICw,y) = - I - d k . ' LI" Ea; . 2 - . UCLX = -go.<br />
"-0 ik ~-(w,k)<br />
Y n1 Y {hcifl<<br />
s2x.x:,~.c.r := ----<br />
He-e C-<br />
n I1 is the response function of the substructure for the<br />
anomalous TE-field in the case of E-polarisation and C- the<br />
nI<br />
response function of the substructure for the anomalous TM-field<br />
in the case of H-polarisation.<br />
The kernel K is the separation kernel, introduced in Sec.5.1. It<br />
i<br />
connects K and HaZ in such a manner that the source of the<br />
aY<br />
anomaly is "internal" wh.cn seen from above the sheet. The kernels<br />
L are the Fourier transforms of the response functions, intro-<br />
.duced in Sec.7.3, which connect the tangential and horizontal<br />
field components above a layered half-space. Their application<br />
to the anomalous field at z = d implies that H and E diffuse<br />
-a -a<br />
downward into tlie substructure antidisappear for z -p a. ---<br />
They represent<br />
the.inductive coupling of the anomalous field with the<br />
substructure .'
Field equations and boundary condition together constitute a set<br />
of four linear equations. for an equal number of unlcnown field<br />
components. They are uniquely determined in this way. A solution<br />
toward the anomalous tangential electric field in 8-polarisation,<br />
for instance, gives<br />
A similar solution toward the anomalous current density in H-<br />
polarisation gives<br />
For a given model-distribution these equations are solved numeri-<br />
cally, by setting up a system of linear equations for the unknown<br />
field components at a finite number of grid points along y. The<br />
convolution integrals, involving derivatives of the unknown field<br />
componentswith respect to y, are preferably treated by partial<br />
integration which in effect leads to a convolution of the unknown<br />
field components themselves with the derivatives of the kernels.<br />
It should be noted that the kernels L approach for y -t finite<br />
-1<br />
limiting values, given by' { 2 ~ i ( w ,011 .<br />
Inverse problem for two-dimensional strnc_j;Lms<br />
-.<br />
1.t -is. also possib1.e to consider the anomalous conductance -ra a.s<br />
the unknown quantity to be determined from an observed elongated<br />
variation anomaly, in the actual calculations to be represented by a<br />
set of respective transfer functions. The normal. sheet conductance<br />
T and the resistivity of the substructure, for which the kernels<br />
n<br />
L have to be determined, enter into the calculations as free model<br />
parameters. They can be varied to get the best agreement in ~ ~ ( y )<br />
when uslng more than one frequency of the variation anomaly.
Another point of concern is the reality of the resulting numeri-<br />
cal values of ra. Usually empirical data wi1.l give complex values<br />
and the free parameters should be adjusted al.so to minimize the<br />
imaginary part of the calculated conductance.<br />
In E-polari.sation the elimination of H- gives<br />
a Y<br />
tie can then eliminate either HC and Haz, if -ra is to be determined<br />
aY<br />
from anomalous electric field,or we can derive ra from the anoma-<br />
lous magnetic field by observing tha-t<br />
by integration<br />
. of the second field equation. The norrnal electric field is<br />
derived from the normal magnetic field by setting<br />
when the source field is quasi-uniform and<br />
when the source field is non-uniform; N (w,y) is the Fourier<br />
I1<br />
transform of * '%. (w,k). Cf. Sec. 8.2, "Vertical soundings with<br />
Cln11<br />
station arrays".<br />
In H-polarisation only the anolnalous electric Field is observable<br />
at the surface. For the eli.mination of -the anolnalous magneti.~<br />
field at the lower face of the sheet from the field equations<br />
we use the generalized inlpedance boundary condition for the ano-<br />
malous TM-field at the surface of the substructure (cf. Sec.7.3<br />
und 8.2) :
with N (w,y) being the Fourier transform of C- (w,k 1. Insertion<br />
I n1 Y<br />
above gives<br />
9.4 Non-uniform layers .above and within a layered structure<br />
The source of the surface variation anomaly is assumed to be an<br />
anomalous slab between z = zo and z = z + D in which the re-<br />
0<br />
sistivity changes in vertical &horizontal direction: p=p,+pa.<br />
The region above and below the slab are taken to be layered,<br />
p = pn being here a sole function of depth (cf. Sec.3).<br />
The response functions for the normal variatLon field are defined<br />
for the normal structure as given by p . If the anomaly lies at<br />
n<br />
the transj.tion between two extended normal regions, two normal<br />
sol.utions havk to be formulated. It should be no-Led that in that<br />
case the anomalous field w i l l not disappear bu-t converge outside<br />
of the anomaly toward the difference of the two normal soluti.ons.<br />
Henceforth., the normal structure , the normal response functions J<br />
and the normal fields En and E will be assumed to be known<br />
-n . .<br />
throughou-t the lower conducting half-space. The source-field will<br />
be regarded as quasi-uniform except for source fields in E-polari-
sation in the case of longated anomalies when a non-uniform<br />
source can be permitted (cf. Sec. 9.3, calculation of E from<br />
-nx<br />
H 1).<br />
nY<br />
The principal problem in basing -the interpretation of actual<br />
field data on this type of model consists in a proper choice<br />
of the upper and lower bounds of the anomalous slab. The following<br />
argumepts may be useful. for a sensible choice:<br />
The anomalous region must be reached by the normal variation<br />
field) i.e. z should not be made larger than the depth of pene-<br />
0<br />
tration of the normal field as given by I C (w,0)1 at the highest<br />
n<br />
frequency of an anomalous response. A lower bound for the depth<br />
of the slab is not as readly formulated because the ano~nalous<br />
response does not disappear necessarily when the depth of pene-<br />
tration is much larger than z + D, i.e. when no significant<br />
0<br />
normal induction takes places within the anomalous slab. Only<br />
if the anomaly is elongated and the source field in E-polarisa-<br />
tion, can it be said that the depth zo + 1) must be at least com-<br />
parable to the normal depth of penetration at the lowest fre-<br />
quency of an anomalous response.<br />
Pilot studies: In section 8.2 the general properties of the im-<br />
pedance tensor Z above a non-layered structure have been discussed<br />
and the following rule for elongated anomalies was establi.shec1:<br />
The impedance for E-polarisation does not diverge markedly from<br />
the impedance of a hypothetical. one-dimensional response for the<br />
local resistivity-depth profile, while the impedance for H-pola-<br />
risation will do.so unless the depth of penetration is small in<br />
comparison to the depth of the internal resistivi.-ty anomaly and<br />
the inductive response nearly normal anyway.<br />
Suppose then that an impedance tensor has been obtained at a'lo-<br />
cation y on a profile across a quasi-twodimensional anomaly which<br />
by rotation of coordinates has zero or allnos-t zero diagonal ele-<br />
ments and that a distinction of the offdiagonal elements for E- and<br />
H-polarisation can be made (cf.Sec.8.2). Regarding the E-polari-<br />
sztion response for a first approximation as quasi-normal., a local<br />
inductive scale length
is calculated as function of frequency and location. It is<br />
converted into an apparent CAGKIARD resistivity and phase:<br />
or alternatively into the depth of a perfect substitute con.ductor<br />
and a modified apparent resistivity :<br />
which'can be combined'into a local depth versus apparent resj-sti-<br />
x X<br />
vity profile p (z,, , y).<br />
11<br />
If the magnetic variation anomaly rather than the geoelectric<br />
field has been observed, the anomalous part of El! can be derived<br />
by integration over the anomaly of the vertical magnetic varia-tions,<br />
w11il.e the normal part of E,, is calculated from the normal impedance<br />
outside of the anomaly or derived theoretically for a hypothetical<br />
normal resistivity model:<br />
or in terms of transfer functions with respect to H n l '<br />
using for the transfer function of H the nota-tj.ons of page 113.<br />
az<br />
Magnetotelluric and geomagne-tic depth sounding data along a<br />
profi1.e are in this way readily converted either into CAGNIARD<br />
r~esistivity ,and phase-contours in frequency--distance coordina.i-es ,<br />
'if<br />
into lines of , depth-of-penetration z at a gi.ven frequency,<br />
%<br />
or into modified apparent-resis.tivi-ty- contours in a z -distance<br />
cposs-section Either one of these plots will outline the frequency<br />
range, respectively the depth range, i.11 which the source of the<br />
anomaly can be expected to lie, and provide a rough idea about the<br />
resistivities like1.y to occur within the anona1.ous zone,
Single frequency -- interpretation by perfect conductors - at<br />
--<br />
variable depth<br />
Geomagne-tic variations anomalies show frequc.ntly nearly zero<br />
phase with respect to the normal field, the transfer functions<br />
which connect the cbmponents of H and fin being real func-tions<br />
-a<br />
of frequency and locations. This applies in particular .to two.<br />
types of anomalies. Firstly to those which arise from a non-<br />
uniform surf ace layer, thin enough to allow -the 'ltl~in-sl~ee~t"<br />
approxima-tion of the previous sec-t:ion with predominant .i.nduction<br />
within the sheet (ns>>l), Secondly, it applies to anomalies<br />
above a highly conductive subsurface layer at variab1.e dfpth<br />
benea-th an effectively non-conducting cover.<br />
In the first case En will be in-phase with Hn, in the second<br />
case out-of-phase (cf .Sec.9.1). But it is important to note that<br />
the anomalous variation field H wi1.l be in either casc roughly<br />
a<br />
in-phase -. with Hn.
Both types of anomaly can be explained at a given frequency by<br />
the undulating surface S of a perfect conductor below non-con-<br />
ducting matter, I-ts variable depth below the surface point (x,y)<br />
X x<br />
will be denoted as h (x,y) . Outside of the anomaly h shall be<br />
constant and equal to the real part of Cn at the considered fre-<br />
quenc y .<br />
This kind of interpretation is intended to demonstrate the effect<br />
of lateral changes of in-ternal resistivity on the depth of penetra-<br />
tion as a function of frequency and location, it does not provide,<br />
however, quantitative information about -the resistivities involved<br />
nor does it allow a distinction of the two types of anomalies<br />
mentioned above.<br />
Clearly, the magnetic fie1.d below S must be zero and the magnetic<br />
field vector on S tangential with respect to S because the con-<br />
nor a1<br />
tinui-ty condition for the field conponent toif'?? requires that this<br />
component vanishes just above S.<br />
~irect model problem: For a given shape of S the anoinalous surface<br />
field can be found with the methods of potential field theory,<br />
since Ma will be irrotational and of internal origin above S. If<br />
~ ~<br />
in -particular S has a simple shape independen-t of x, the field<br />
%<br />
' lines for E-polarisation in the (y,z)-plane for z - < h (y) can be<br />
found by conformal mapping as follows:<br />
Let w(y,z) = y(yl,zl) + i.z(yl,zl) be an analytic function whi.ch<br />
maps the line zl=O of rectangular (yl ,zl) coordinates into the line<br />
* z=l> fy) of rectangular (y,z) coordinates. Lines z1 = const, are in-<br />
terpreted as magnetic field lines of a uniform field above a perfect<br />
conduc-tor at constan-t depth, their image in the (y,z)-plane as field<br />
lines of adistorted field above a perfect conductor at the variable
x:<br />
depth h (y), the image of the ultiina.'ce field line zf=O being ta.ngen-<br />
tial to the surface of the conductor as required.<br />
If gn= (Hlly '0) denotes the uniform horizontal field vector at a<br />
point in the original (yl,z') coordinates, the components of the<br />
field vector - H= (H ,Hz) at the image point in the (y,z) coordinates<br />
Y<br />
can be shown to be given by<br />
with<br />
The difference - H - H represents<br />
-n<br />
the anomalous field to be<br />
' Inverse problem: The shape of the surface S can be found inversely<br />
from a given surface anomaly by constructing the internal field<br />
lines of -the field gn+ga. Field lines which have at some distance of<br />
the anpmaly the required normal depth R ~ ( C ) define the surface S,<br />
n<br />
provided of course that the surface thus found does no-t inS:ersec3<<br />
the Earth's surface anywhere.<br />
The ac-tual calculation of internal field lines requires 2 downward<br />
ex-tension of the anomaly through the non-conducting matter above<br />
the perfec-t conductor, using the well developed methods of potential<br />
field continuation towards its sources, In order to obtain sufficien:<br />
stability of thenumericaz process, the anomaly has to be low-pass<br />
filtered prior to the downward con'tinuation with a cut-off at a<br />
reciprocal spa-tial wave number comparable to the maximum depth of<br />
intended downward extrapolation.
A field line segment with the horizontal increments dx and dy has<br />
then -the vertical increment<br />
H H<br />
a z a z<br />
az<br />
dz =-dx +-dy =<br />
ax 3~ +H H +H Q2<br />
nx ax nY aY<br />
and the entire field line is the,,constructed by numerical integra:<br />
tion;<br />
An analy-tically solvable inverse problem: Suppose the observed<br />
anomaly' is 2-dimensional and sinusoi.da1,<br />
H (y,z=O) = c sin(k.y) H .<br />
aY nY.<br />
Because of the internal origin of the anomaly<br />
the internal field increasing exponential1.y with depth,<br />
Hence, the slope of field lines at the depth z i s<br />
by integration<br />
z(y) = z - c/k.sin(k-y1.e k.z(y)<br />
0<br />
as the implicit solution with zo as t5e constant of integration.<br />
Here it w i l l be chosen in such a way that the highest point of the<br />
field line is still below the Earth's surface z=0.<br />
We infer from the fi.eLd line equation z(y) that field lines above<br />
sinusoida!. anomalies oscillate non-symmetrically around the depth -1.<br />
z=z with Lj=27i/k as spatial wave length. Their deepest points z=z<br />
0<br />
are at y=L/4,-3L/4,+. ., . ..<br />
y= -L/4,3L/4 ,T.. ...<br />
. .<br />
. . . . , their highest points 2.2- at<br />
u,z<br />
'jar 7, , :f-, , . / ,- - 0 -3<br />
x<br />
- L ....-a' \ 1 f .- ,<br />
/ ,<br />
-r/ ..-<br />
,/' 7 -k> '../%,./'<br />
\.- A--<br />
s u, j., , 0.- , rr .L<br />
1<br />
Z<br />
Y<br />
- i . . '. \<br />
1<br />
Y 1<br />
F: ,.C'C p,<br />
-3
Suppose the spatial wave length of the anomaly is large in'com-<br />
-1.<br />
parison to the deepest point of penetration z . Then the approxi-<br />
mate field line equation<br />
can be used wi-th the requirement tha-t zo is equal to or larger<br />
than c/k. The mean amplitude of the field line oscillations is<br />
readily expressed now in terrns of. the amplitude of the observed<br />
surface anomaly as given by the factor c, namely<br />
If, for example, a moderate anomaly has the ampli.tudd cr0.25 and<br />
a wave length of 628 km, the mean amplitude of the field line'<br />
oscil.lations are given by (zo+'100)/4km, zo? 25 km. Using z =30 km<br />
0<br />
gives a mean ainplitude of 33 km and'the field line oscill.ates<br />
between 4 km and 70 km, This numerical example demonstrates that<br />
even minor ano~nalies require rather large undula-tions in the depth<br />
of a perfect substitute conductor.<br />
'Multi-frequency - - interpretation buocal. ir~duc$ion in isolated<br />
bodies<br />
Most geomagnetic induction anomalies arise from a local rearrange-<br />
ment of large-scale induced currents, In some cases, however, it<br />
may be justified to assume that the source of the anomaly is a<br />
conducting body which is isolated by non-conducting mat-ter from<br />
the large-scale curcent systems. Such a body can be, fop instance,<br />
a crustal lense of high conductivity within the normal1.y highly<br />
resistive crust.<br />
The direc-t - problem: Consider a body of (variable) conductivity<br />
between z=zl and z= 2. ' The sui7roundi.ng matter. at the same depth is<br />
effectively non-conducting at the considered frequency, i,e, large-<br />
scale induced curren-ts which flow in the normal structure above and
- 160 -<br />
below can neither enter nor leave this body.<br />
The inducing field is the normal field HA between zl and z 2 3<br />
regarded as uniform. For simplicity it is assumed that the nor-<br />
mal structure between z=0 and z represents a "thin sheet" of the<br />
1<br />
conductance T ~ . Then the attenuation of H1 with respect to the<br />
n<br />
normal field H at the surface zz0 is given by<br />
n<br />
as thin-sheet induction parameter (Sect.9'3 . . and Appendix to 7.3).<br />
Three aspects of the arising inductiorl probl.em have -to be con-<br />
sidered: (i) the local induction by H' within the body, (ii) the<br />
n<br />
elec-tromagnetic coupling of this body with the normal subst~uc-Lure<br />
below z (iii) the coupling of the body with the surface ccveri;bove<br />
2'<br />
In order to make problem (i) solvable in a straightforrgard manner<br />
uniform bodies of simple shape such as spheres and horizon1:al.<br />
cylinders of infinite length have -to be selected, The anomalous<br />
field due to -the induc-tion in these bodies is of fr,equ.ency.-indepen-<br />
dent geometry. This fact permits si.lnple solutions of The inverse<br />
prohl.em as seen below.<br />
Consider the case of a horizontal cylinder of infi,nite length, the<br />
radius R and -the conductivity u its axis is parallel to x and<br />
z;<br />
intersects the (y, z)-plane at the point (0, zo). The normal horizon-<br />
tal field is. parallel to y and produces hy i.nduction curren-ts which<br />
flow parall.el to x in .the upper halve and anti-parallel to x in "ilie<br />
lower halve of the cylinder. Their magnetic field outside the cy-<br />
linder is -the field of a central 2-dimenstonal di.pole of the monient
m antiparallel to y. It has the components<br />
with r2 = Y2 + (Z~-.Z)~ and cos0 = zs0/r, sin8 = y/r,<br />
The dependence of the dipole moment on the inducing field will<br />
be expressed in terms of a response function, Let mm be the<br />
dip01 moment of a perfectly conducting cylinder which shields<br />
the external fieldwmpletely frorn its interior. Hencc, the in-<br />
duced field at the point (0, z -R) just above the cylinder<br />
0<br />
must be equal to H' i.e.<br />
n'<br />
which implies that it cancels H1 within the cgl.inder as required.<br />
n<br />
The response function of the cylinder with respect to a uniform<br />
inducing field perpendicular to the axis is now defined as the<br />
complex-valued ratio<br />
which can be shown to be a sole function of the dimensionl.ess<br />
induct ion parameter<br />
-.<br />
. ?he modulus of f(il ) will approa.ch zero for nZ >.I, its argument changing from YO for q7 -t- 0 to 0' for<br />
"z<br />
-t -. Hence, at sufficien-tly high frequencies the anomalous<br />
field from the cylinder w i l l be in-phase with H' and not in-<br />
11<br />
crease beyond a certaiii "induc-tive limit", whj.le at sufficiently<br />
low frequencies the anomal.ous field w i l l be out-of-phase and<br />
small in comparison to the field a-i: the i.nductive 1.irnit.<br />
..I
I '. -<br />
L.. . . - I . --.> . ~<br />
4<br />
The transfer function between the anomal.ous and normal field<br />
are now readily expressed in terms of a frequency-dependent<br />
. response function, a frequency-independmt geometric factor,<br />
and a frequency-dependent attenua-tion factor Q' for the normal<br />
field:<br />
wi-th<br />
Q' = (1 4. i nZ)-' ,<br />
An approximate so1.ution of problem (ii) can be obtained by re-<br />
presenting the conductive substructure below the surface sheet<br />
by a perfect conductor at the (frequency-dependent) depth K~(C;)<br />
and by adding to the anomalous dipole field the field of an<br />
image dipo1.e of the moment R ~H;~ at the depth 2sRe(~i)-z~.
Problem (iii) j.nvol.ves mainly the attenuation of the upward<br />
diffusj.ng anomaly ca by ' uniform surface layers. This a.tte.iiuatri.on<br />
is, however, a second order effect and can be neglected, if<br />
at the considered frequency the widespread normal fieid pene-<br />
trates the surface layers to any extent, i.e. if the modulus<br />
of Q' is sufficiently close to unity.<br />
The inverse problem: Assuming the anomalous body to be a uniform<br />
cylinder, the model parameters zo,'R, cr can be uniquely in-<br />
R<br />
ferred from surface observations, providing the transfer func-<br />
tions for H and FIaz are known for at least two locations and<br />
aY<br />
frequencies. For.ming here the ra-tios<br />
the angles 8 and thereby the position- of the cylinder can be<br />
found, The depth of the cylinder axis being kncwn; the dipole<br />
moment RZf(n) can be calculated. Its argument fixes the size<br />
of the induction parameter n and two determinations of nZ at z<br />
different frequencies the radius R and -the conductivi-ty crz. NOW<br />
R and zo being known, the field of the image dipole can be cal--<br />
culated. The inverse procedure is repeated now with the observed<br />
surface anomaly minus the field of the image dipole until con-.<br />
vergence of the model parameters has been reached.<br />
Interpretation with HIOT--SAVART1s law<br />
. . ~<br />
Let - f he the current density vector within a volume dV at the<br />
Q n n h<br />
point - r = (x, y, z). Then according to UZUT-SAVART1s law in SI<br />
units -the magnetic fieled vector at: the point - r = (x, y , z) due<br />
c'<br />
to the line-current element idV is
This law can be used to i-nterpret an observed surface anomaly<br />
of geomagnetic variations in terms of a subsurface distribution<br />
of anomalous induction currents, which in turn are related to<br />
an internal conductivity anomaly a i,e. to local deviations<br />
a'<br />
of a from a normal. layered distribution o (z),<br />
n<br />
It has been pointed out in Sec.8.3 that the anomalous part of<br />
the geomagnetic and geoelectric variation field can be split<br />
into TE and TM modes (= tangential electric and tangential<br />
magnetic modes) and that only -the TE modes produce an observable<br />
magnetic variation anomaly at the Earth's surface. Consequently,<br />
3<br />
only a TE anomalous current density .distribution c~~~ can be<br />
related to the anomalous magnetic surface field Ha, the sub-<br />
script "11" refering to the solutioll I1 of -the diffusion<br />
equation.<br />
Denoting the tangential components of simply with i and<br />
c<br />
-a11 ax<br />
J. the components of the anomalous surface field are given<br />
ay'<br />
A 0<br />
with r2 = (x-x)' + (Y-~ + G2. The inverse problem, namely to<br />
find from a given surface anomaly the internal anomalous current<br />
distribution, has no unique solution. Within certain constrai.nts,<br />
however, it can be made unique, at least in principle. For in-<br />
stance, it is assumed that the anomalous current flow is limited<br />
to ;I certain depth range which implies that the matter between<br />
this depth range and the surface is regarded as non-conducting<br />
at the considered frequency.<br />
The surface anomaly can be extended no:,? dowr~trard t o the top of<br />
.the anomalous depth range with standard metliods which require<br />
A
only adequate smoothing of the observed anomal.ous surface field.<br />
Then a certain well defined depth dep'endence of the anomalous<br />
current density is adopted as discussed below and the anomalous<br />
current distribution can be derived in a straightforward manner.<br />
Suppose an observed anomaly 51, at a given frequcncy has been ex-<br />
plained in this way by a distribu-tion of anomalous internal.<br />
currents. Its connection to the internal conductivity is esta-<br />
blished by the normal and anomalous electric field vector accor-<br />
ding to<br />
Assuming the normal conductivity distribution un(z) to be known,<br />
E as a function of depth 'is readily calculated. There is no<br />
-I1<br />
simple way, however, to derive the anomalous electric field of<br />
the TE mode except by numerical models as discussed below. There<br />
is in particular no justification to regard it as small. i.n com-<br />
parison to the normal electric field and thus to drop the second<br />
term in the above relation. Instead the following argumentati.on<br />
has to be used:<br />
The anomalous electric field in the TE mode can be thought to<br />
contain two distinct components. The first component may be re-<br />
garded as the result of local self-induction due to EaII. I -.<br />
can be neglected a-t suffici.ently low frequencies, when the ;-lzlf-<br />
width of the anomaly is small in comparison to the minimum skin-<br />
depth value within the anomalous zone.<br />
The second component arises from electric charges at boundari-es<br />
and in zones of gradually changing conductivi-by. These charges<br />
produce a quasi-static electric field norroal to bounclarics and<br />
paral7.el -to in-ternal conduc-tivity gradients which ensures the<br />
continuity of the current across boundaries and internal gradient<br />
zones. Hence, this second component of the anomalous electric<br />
field does not disappear, when the frequency becomes small.<br />
It vanishes, however, if anomalous internal currents do not cross<br />
boundaries of gradient zones, i.e. when the normal field is in<br />
E-polarisa-tion with respect to the -tl>end of elongated s'iructures.
Only in this special case will be jus.tificati.on .to neglect the<br />
term oEa at low fr2quenaies and to use the approximation<br />
.to calculate the internal conducti.vj.ty anomaly o from the ano-<br />
a<br />
malous current density and the normal electric field, St w i l l<br />
be advisable to regard here rsa as a sole functi.on of x and y<br />
and thus to postulate the same depth dependence for the anomalous<br />
current and the normal electric field.<br />
Interpretation with numerical models<br />
So far idealized models for a laterally non-uniform subs-tructure<br />
have been considered. They allowed a simplified treatment of<br />
the induction problem and had in common that only few free para-<br />
meters were involved. Once numerical values were attached to<br />
them, the observed surface anomaly of the variation field could<br />
,be converted rather easily into variable model parameters to<br />
characterize the internal change of subsurface conductivity at<br />
some given depth from site to site.<br />
The introduction of more realistic models le'ads to a substantial<br />
increase of the numerical work involved and should be considered<br />
only, if transfer functions for the anomalous surface field are<br />
known with high accuracy. The inverse problem to derive a set<br />
of variable model parameters directly from observations can be<br />
solved by linearisation, i.e. by the introduction of a linear<br />
data kernel, connecting changes of the surface response to<br />
changes in the subsurface conductivity structure (cf. Sec.6).<br />
As before a normal structure o outside of the anomaly is given,<br />
n<br />
the variable model parameters representing only lateral changes<br />
of o with respect to on within the range of the anomaly. Unless
stated otherwise, source fields of lateral homogeneity are<br />
assumed, yielding in conjunction with u the normal fields Hn<br />
n<br />
and Ell as known functions of depth.<br />
Conductivity anomalies aa = o - a are restricted to an anoma-<br />
n<br />
lous slab above or within a laterally uniform structure, exten-<br />
ding in depth from zl to z2, Within This slab two basic types<br />
of conductivity anomalies may be distinguished:<br />
(i) slabs with gradually changing conductivities in horizontal<br />
direction or (ii).slabs which consist of uniform blocs or re-<br />
gions, separated by plane or curved boundaries.<br />
In the first case electric changes are distributed within the<br />
'?<br />
slab, -yielding closed current circuits divl = 0 as required in<br />
quasi-stationary approximation. Consequently, divE - = -(E*grado)/o<br />
will not be zero. In the second case only bloc boundaries carry<br />
free surface changes and the electric field will be non-diqergent<br />
within the blocs. In either case a solution of the diffusion<br />
equation for<br />
-a<br />
H or<br />
-a<br />
E has to be found under the condition that<br />
the anomalous field approaches zero with increasing distance<br />
from the anomalous slab.<br />
Numerical methods for the actual solution of the direct - problem,<br />
when oa is given and Ha or Ea are to be found, have been discussed<br />
in Sec. 2 and 3. Here some additional comments: Models<br />
should be set up in such a way that either the shape of the anomalous<br />
body or the anomalous conductivity is the variable to<br />
be described by a set of model. parameters. After fixi-ng the free<br />
parameters, in particular z and z2, the model parameters for<br />
1<br />
the description of the anomaly are varied un-ti1 agreement is<br />
reached between the observed and calculated transfer functions.<br />
Suppose that models of type (ii) are used, that only vertical<br />
and horizontal boundaries are permitted, and that the anomalous<br />
slab is subdivided by equally spaced horizontal boundaries into<br />
layers. Then the variable model parameters are either the po-<br />
sitions of verti.ca1 boundaries, encl.osing a body of constant
anomalous conductivity oat OP the anomalous conductLvity values<br />
'am<br />
of the blocs m = 1,2,-. .. M, separated by equally spaced<br />
vertical boundaries:<br />
If the search for a best fitting se-t of model parameters is to<br />
be done by an iterative inverse rnethod rather than by trial-and-<br />
error,%inear data kernel matrix G =(g has to be found for<br />
nm<br />
an initial model. In this matrix the element gnm represents the<br />
change yn of the anomalous transfer function at a certain su~face<br />
point and frequency Rn which arises in linear approximation<br />
from a change xm of the model parameter in the m'th bloc of the<br />
starting model. The resulting system of linear equations<br />
- x is solved with the methods of Sec.6. Its solution<br />
Yn<br />
-<br />
m<br />
6nm m<br />
represents the improvement of the initial model, if yn is the<br />
misfit between observed and calculated transfer functions for<br />
the initial model: The process is repeated with the improved<br />
models until the necessary improvements xm are small enough to<br />
justify a linear approximation.<br />
Which complica-tions arise in the general case that not only the<br />
internal conductivity structure but also the inducing source<br />
field are 1aterall.y non-uniform? First of all, in addition to on<br />
as a function of depth the magnetic field H (e) of the external<br />
source must be a known fu11'cti.on of surface location. In the<br />
special case of the equatorial jet field this source field con-<br />
fi.guration w i l l be more or less the same for all day-time varj.a-';iofis<br />
and thus a normalisation.of the observed surface field by the<br />
....
- i<br />
- - i<br />
field at sollie dis-ti~lguishecl surrace point wil.1 be possible, Tor<br />
instance by the horizontal fie1.d at the dip equator,<br />
The conversion formulas of Sec. 5.2 and the methods described<br />
in Sec. 8.2 ("Vertical soundings with station arrays") are now<br />
used to find the inrernal part M (i) of'the normal magnetic field<br />
n<br />
.and the normal electric field En as functions of the surface<br />
loca-ti.on and frequen~y. Finally, the normal surface field is ex-<br />
Ly c oucl.~4wn<br />
tended downwardVwith the spatial Fourier transforms of the down-<br />
ward extension factors as given in the Appendix to Sec. 7.3.<br />
Now the normal field within the anomalous slab is known and the<br />
diffusion equation for the anomalous field can be solved numeri-<br />
cally in the same way as before, when the slab or the anomalous<br />
region themsel.ves are chosen as basic domains for the numerical<br />
calculations. Here a final. schematic summary of the various steps<br />
of the calculations, when the source is the combined field of the<br />
equatorial jet and the low-latitude Sq:<br />
- Interpretation with scale inodel experiments<br />
. . . . . . . . - . . . . - -<br />
Numerical model calculations may be too elaborate, if the induc-<br />
tion by a non-uniform source in a three-din!ensio:lal anomalous<br />
structure has to be considered. An example is the coast effect<br />
of the equatorial jet, when the dip equator crosses the coast-<br />
line under an angle. - No co~nplications arise at those places,<br />
when the dip equator is parallel to xhe coast and the jet field<br />
in E-polarisation (s. above). -<br />
In si-tuation of great complexity the qualitative and even quan-<br />
titative unders-tanding of surface observa-tions may be fur.thcred<br />
by labora-tory scale model experi.ments, sj.niu1.ating the natural<br />
induction process on a reduced scale. Invariance of .the elcc'tro-
the product WUL' constant with L denoting the length scale,<br />
The primary inducing field is produced by an oscilla-ting dipole<br />
source or-by extended current loops, situated as "ionospheric<br />
sources" above an arrangeme& of conductors which represent<br />
the conducting material below the Earth's surface. Alternatively,<br />
the conductors as a whole can be placed into the interior of<br />
coils, say Helmholtz coils, and thus be exposed to a uniform<br />
source field,<br />
.-<br />
\<br />
I<br />
/ I .<br />
1 i:<br />
/'<br />
.
width, however, forces the induced currents to flow in loops<br />
which in strength and phase may be totally controlled by the<br />
edges of the conductor. In order to avoid this unwanted effect<br />
the source field at the level of the conductors should die away<br />
before reaching the edges of - the scale model. For instance, if<br />
a line current source is used, the half-width of the line current<br />
field on the surface of the scale model' should be considerably<br />
smaller then the width of the model..<br />
These complicati.ons do not arise of course, when no attempt is<br />
made to simulate local anomal-ies of a large-scale induced current<br />
system, i.e. when the scale model is placed into the interior<br />
of coils in order to simulate local induction in isolated bodies<br />
(cf. subsection on this topLc in Sec. 9.4).<br />
Here are to men-tion the scale model experiments by GREWET and<br />
LAUNAY who showed how ,a large-scale induction can be simvlated<br />
also by the induction in the interior of coils. Their objective<br />
was to make a scale model of the coast effect at complicated<br />
coastlines. They nptedthat the i.nductive coupling between the<br />
ocean and highly conduc-ting ma-terial at some depth within the<br />
Earth is well represented by a system of image currents at the<br />
level 2 -hx below the ocean. Here h" is again the depth of a<br />
perfect substi-tute conductor for the oceanic substructure at the<br />
considered frequency.<br />
GRENCT and LAUNAY use as model conductors two thin metallic plates<br />
which are connected along two edges by vertical conducting strips.
One of the thin plates represents the ocean and one of its<br />
edges is given the shape .of a certain coastline .to be studied.<br />
The second plate represents the level of the image currents.<br />
The whole arrangement of conductors is placed into a Helmholtz<br />
coil in such a way that the vertical strips are parallel to<br />
. the. magnetic field. Plates and st~-.ips now form a loop normal<br />
to the magnetic flux within the ~elmhoytz coil and thus currents<br />
are induced which flow in the "oceanic" plate parallel to the<br />
"coastline".<br />
In SPITTA's arrangement for th.e study of the coast effect the con-<br />
ductors are placed below a horizontal band-current closed by<br />
a large vertical loop. The oceanic and continental substructure<br />
is represented by a thick metalic plate, the oceans by a thin<br />
metalic sheet which partially covers the plate. The thickness<br />
of the plate is large in comparison to the skin depth and its<br />
width about twice the half-width of the field of the band-<br />
current at the level of the plate. The induced current systems<br />
form closed loops within plate and sheet and can be assumed<br />
to be largely horizontal. By placing one edge of the covering<br />
sheet below the 'center of the band-curren-t the coast effect of<br />
an ionospheric jet can be studied for any angle between coast<br />
and jet.<br />
1,
The ratio of the length-scales model: nature should be in the<br />
6 7<br />
order 1 : 10 or 1 : 10 . In this way current loops of some<br />
1000 km in nature can be reproduced. In SPITTA's' experiment the<br />
ratio of length-scales is 1 : 4.10~. A 4 km thick ocean is re-<br />
presented by an aluminium sheet of 1 rrn thickness, a highly con-<br />
ducting layer in the mantle at 360 km depth'by an aluminum plate<br />
9 cm below the sheet. The width of this plate is 2 m and is<br />
equivalent to 8000 km in nature.<br />
Model conductors niay be chosen from the following materials:<br />
cu - AL - ~b 7 .- 0.5.107 (am)-'<br />
Graphite 3.10~ ' 11<br />
(saturated)<br />
NaCl solution<br />
(concentration of maximum<br />
conductivity )<br />
In SPITTA's experiment the conductivity-ratio model: nature is<br />
7<br />
2.10 : 4. Hence; with a ratio of length scales of 1/4 a<br />
frequency of 1 kHz in the model corresponds to 1/32.10-~ Hz 'lcph<br />
in nature.<br />
DOSSO uses graphite to represent the oceans and highly conduc-<br />
ting material in the deeper mantle, saturated NaC1-solution to<br />
represent the continental surface layers and the poorly conduc-<br />
ting portions of crust and uppermost mantle. Since his model<br />
frequencies are only slightly higher (1 to 60 kHz),.a one orbder<br />
5<br />
of magnitude pester ratio of length-scales (1:lO 1 has to be<br />
used to simulate natural frequencies between 1 cph and 1 cpm.<br />
,<br />
The lists of available model conductors shows that it is diffi-<br />
cult to simulate. conductivity contrasts of 1:10 or 1:100 which<br />
are of particular importance in the natural induction process.<br />
Only salt solutions of variable concentration could provide a<br />
sufficient range in xodel conductivity, but their relatively
low conductivity requires the use of very high frequencies<br />
3<br />
(10 kHz) and extremely large model dimensions (10 m).<br />
In conclusion it should be pointed out that even those scale<br />
model experiments which do not reproduce the natural induction<br />
in a strict quantitative sense may be useful for a descriptive -<br />
interpretation of complicated variation fields. In those cases<br />
only the impedance or the relative changes of the magnetic<br />
field with respect to the field at one distinguished point above<br />
the model will be considered and compared with actual data.<br />
10. Geophysical and geological relevance of geomagnetic<br />
induction studies<br />
In exploration geophysics the magnetotelluric method,<br />
in combination with geomagnetic depth sounding, has been applied<br />
with some success to investigate the conductivity structure of<br />
sedimentary basins. Electromagnetic soundings with artificial<br />
sources as well as DC soundings which truly penetrate through<br />
a sedimentary cover of even moderate thickness are difficult<br />
to conduct on a routine basis. Hence, it seems that electro-<br />
magnetic soundings with natural fields are more efficient than<br />
-any other geoelectric methods in exploring the overall distri-<br />
bution of conductivity in deep basins.<br />
In particular the integrated conductivity T of sediments above<br />
a crystalline basement is well defined by the inductive surface<br />
response to natural EM fields and can be mapped by a survey<br />
with magnetic and geoelectric recording stations. If in addi-<br />
tion some estimate about the mean conductivity of the sediments<br />
can be made from high frequency soundings, the depth of the<br />
crystalline follows directly from T.<br />
If structural details of sedimentary basins are the main interest<br />
of the exploration, a mapping of the electric field only accor-<br />
ding to st'rength and direction for a given polarisation of the<br />
regional horizontal magnetic field will be adequate. The inter-
pretation is handled like a direct current problem in a. thin<br />
conducting plate of variable conductivity. This so-called<br />
"telluric method" represents a very simple lcind of inductive<br />
soundings, but the preferential direction of the superficial<br />
currents thus found usually gives a surprisingly clear im-<br />
pression about the trend of structural elements like grabens,<br />
anticlines etc. The usefulness of this method arises from the<br />
fact that these structural elements can be detected even when<br />
they are buried beneath an undisturbed cover of younger sedi-<br />
ments.<br />
Geomagnetic and magnetotelluric soundings are less useful for<br />
exploration in areas of high surface resistivity, in particular<br />
in crystalline regions. Even pulsations penetrate here too<br />
deeply to yield enough resolution in the shallow depth range of<br />
interest for mining. Audio-frequency soundings with artificial<br />
or even natural sources will be better adapted and are widely<br />
used in mineral exploration.<br />
The probing of deeper parts of crust and mantle with natural<br />
electromagnetic fields will eventually lead to a detailed<br />
knowledge of the internal conductivity distribution down to<br />
about 1000 km depth. Its relation to the downward rise in<br />
temperature is obvious, in fact electromagnetic soundings pro-<br />
vide the only, even though indirect method to derive present-<br />
day temperatures in the upper mantle.
Other derivable properties of mantle material like density and<br />
elastic parameters are.largely determined by the downward in-<br />
crease in pressure with only a second-order dependence on<br />
temperature, which i.s hardly of any use for estimates of mantle<br />
temperatures.<br />
The remarkable rise of conductivity between 600 and 800 krn depth,<br />
however, should not be understood as temperature-related but<br />
as indication for a gradual phase change of the mantle minerals,<br />
possibly with a minor change in chemical composition, this is<br />
with a slight increase of the fe:Mg ratio of the olivinespinell<br />
mineral assembly.<br />
The conductivity beneath the continental upper mantle from 100 km<br />
down to about 600 km is suprisingly uniform and seems to indicate<br />
-that in this -depth range the temperature gradient cannot<br />
be far away from its adiabatic value of roughly 0.5' C/km. The<br />
.mysterious appearance of highly conducting layers in the uppermost<br />
mantle may be connected to magma chambers of partially<br />
molten material and to regional mantle zones of higher than<br />
normal temperatures in general. The expected correlation of high<br />
mantle conductivity, high terrestrial heatflow and magmatic<br />
activity clearly exists in the Rocky Mountains'of North America.<br />
There are also some indication for high conductivities beneath<br />
local thermal areas. The Geysers in California, Owens Valley in<br />
Nevada, possibly Yellowstone and the Hungarian plains are<br />
examples. It should be pointed out, however, that there exist<br />
also regions of.high subcrustal conduc-tivity with absolutely no<br />
correlation to high heatflow or recent magmatic activity. The<br />
most proainent.inland anoma1.y of geomagnetic variations, which<br />
has been found so far, namely the Great Plains or Black Hills<br />
anomaly in North herica,laclcs still any reasonable explanation<br />
,or correlation to other geophysical observations.<br />
0;le . great unsolved problem in geomagnetic inductions<br />
studies is the depth of penetration of slow variations into the<br />
mantle below ocean basins. There are definite reasons to believe
that the upper mantle beneath oceans is hotter than the mantle<br />
beneath continents down to a depth of a few hundred kilometers.<br />
If this is so, a corresphndingly higher conductivity should exist<br />
beneath oceans which could be recognised from a reduced depth<br />
of penetration in comparison to continents. Once a characteristic<br />
conductivity difference between an oceanic and a. continental<br />
substructure has been established, this could be used tb recog-<br />
nize former oceanic mantle material beneath present-day con-<br />
tinents and vice versa.<br />
First houndings with recording instruments at the bottom of the<br />
sea have been carrred out. They confirmed to some extent the<br />
expectation of h'igh conductivities at extremely shallow depth,<br />
but these punctual soundings may not be representative for the<br />
oceans as a whole. Here the development of new experimental<br />
techniques for expedient seafloor operations of magnetic and<br />
geolectric instruments has. to be awaited.<br />
~~sarvatioAs,on. . mid-oceanic islands provide a less expensive<br />
way to study the induction in the oceans which in he case of<br />
substorms and Sq is strongly coupled to the crustal and subthe<br />
ocea s<br />
crustal conductivities beneath . Buf again oceanic islands<br />
are usually volcanic and their substructure may differ from that<br />
of ordinary parts of ocean basins.<br />
The island-effect itself is no obstacle fo$ soundings into the<br />
deep structure. In fact, this effect represents a powerful tool<br />
to investigate the inductive response in the surrounding open<br />
'ocean, since the.theoretica1 distortion of the varia-tion fields<br />
due to the islands can be regarded at sufficiently Low frequency<br />
as a direct current problem for a given pattern of oceanic in-<br />
duction currents at some distance from the island. Setting the<br />
. ..<br />
.density of these currents in relation to the observed magnetic<br />
field on the island gives the impedance of the variation field<br />
, for the surrounding oceari of known integrated conductivity. To a<br />
first approximation induced currents in the ocean do not contribute<br />
to the horizontal magnetic field on the island. Hence, by<br />
knowing their density the horizontal magnetic field on the sea<br />
s<br />
I<br />
i<br />
I:<br />
floor can be calculated from which the inductive response function<br />
for the oceanic substructure follows.<br />
j<br />
!
11. References for general reading<br />
Electroinagnetic induction in the earth and planets. Special<br />
issue of Physics of the Earth and Planet. Interiors, L,<br />
Xo.3, 227-400, 1973.<br />
BERDICHEVSKII, M.N., (Electrical sounding w2th the method of<br />
-magneto-telluric profiling), Nedra Press, Goscow 1968.<br />
FRAZER, M.C.: Geomagnetic depth sounding with arrays of magneto-<br />
meters. Rev. Geophys. Space Phys. -> 12 401-420, 1974.<br />
JANKOGJSKI, J., Techniques and results of magpi~etotelluric and geo-<br />
magnetic deep soundings, Publications OQ? the Institute of<br />
Geophysics, Polish Academy of Sciences, vo1.51i, Warsaw 1972.<br />
KELLER, G.V. and FRISCHKNECHT, F.C.: Electrical methods in Geophy-<br />
sical prospecting. Int. Ser. Monogr. in Electromagnetic Waves,<br />
10, 1-517, 1966.<br />
-<br />
PORATH, H. and DZIEWONSKI, A,: Crustal resistivity anomalies from<br />
geomagnetic depth-sounding studies. Rev. Geophys. Space Phys.<br />
9, 891-915, 1971.<br />
-<br />
RIKITAKE , T. : Electromagnetism and the '~arth 's interior. Develcprnents<br />
in Solid Earth Geophys. - 2, 1-308, 1976.<br />
SCHMUCKER, U.: Anomalies of geomagnetic varizitions in the southwestern<br />
United States. Bull. Scripps Imst .. Oceanogr. -, 13<br />
1-165, 1970. z .<br />
WARD, S.H.: Electromagnetic theory far geophysical application. In:<br />
D.A. Hansen et al. (Eds.), Mining Geophysics, The Sop. of<br />
Expl. Geophysicists, Tulsa, Oklahoma, T.967.<br />
WARD, S.H., PEEPLES, W.J., and RAN, J.: Analysis of geoelectromagne-<br />
tic data. -Methods in Comnii+a+inn;ll --?h\rx