dxdy
The reciprocal relation is:
( ) ( )
( )
f f x y e d d
i x y
, , =
+
1
4
2
where and are wavenumbers in the x and y directions, respectively, measured in
radians per metre if x and y are in given metres. These are related to spatial
"frequencies" f
x
and f
y
, in cycles per metre.
A grid (in the space domain) is transformed to and from the wavenumber domain
using Fast Fourier Transform (FFT). The equivalent data set in the wavenumber
domain is commonly called Transform. A Transform of a grid is composed of
wavenumbers, which have units of cycles/metre, and have a real and imaginary
component. Just as a grid samples a space domain function at even distance
increments, the Transform samples the Fourier domain function at even increments of
1/(grid size) (cycles/metre) between 0 and the Nyquist wavenumber (1/[2*cell size]).
A given potential field function in the space domain has a single and unique
wavenumber domain function, and vice versa. The addition of two functions
(anomalies) in the space domain is equivalent to the addition of their Transforms.
The energy spectrum is a 2D function of the energy relative to wavenumber and
direction. The radially averaged energy spectrum is a function of wavenumber alone,
and is calculated by averaging the energy for all directions for the same wavenumber.
The Fourier transform of the potential field produced by a prismatic body has a broad
spectrum whose peak location is a function of the depth to the prisms top and bottom
surfaces, and whose magnitude is determined by the prisms density or
magnetization. The peak wavenumber (') can be determined by the following
expression:
Chapter 4: 2D Fast Fourier Transform and MAGMAP Theory 43
( )
t b
t b
h h
h h
=
ln
where:
' is the peak wavenumber in (radians / metre)
h
t
is the depth to the top
h
b
is the depth to the bottom
The spectrum of a bottomless prism peaks at the zero wavenumber according to the
following expression (Bhatacharia, 1966):
( ) f
, =
= +
e
r
hr
2 2
where h is the depth to the top of the prism.
The spectrum for a prism with top and bottom surfaces is:
( ) f , =
e e
h r h r
t b
where h
t
and h
b
are the depths to the top and bottom surfaces, respectively. As the
prism bottom is brought up, the peak moves to higher wavenumbers as illustrated in
the following figure.
0
1
0 1
wavenumber
top = 4
12
20
36
8
no bottom
bottom depth
44 Chapter 4: 2D Fast Fourier Transform and MAGMAP Theory
Considering the spectrum of a fixedsize prism, as the prism depth is increased, the
peak of the spectrum is shifted to lower wavenumbers (the anomaly becomes
broader), and the magnitude of the spectrum is reduced.
0
0
1
1
wavenumber
thickness = 4
top = 4
8
16
An important fact to note in the above figure is that the spectrum of a deep prism does
not exceed the magnitude of the same prism at a lesser depth at any wavenumber
only the peak is shifted to lower wavenumbers. Because of this, there is no way to
separate the effect of deep sources from shallow sources of the same type using
wavenumber filters. This is only possible if the deep sources are of stronger
magnitude, or if the shallow sources have a lesser depth extent.
When considering a grid that is large enough to include many sources, the log
spectrum of this data can be interpreted to determine the statistical depth to the tops
of the sources using the following relationship:
( ) log E r hr = 4
The depth of an ensemble of sources is easily determined by measuring the slope of
the energy (power) spectrum and dividing it by 4. A typical energy spectrum for
magnetic data may exhibit three parts a deep source component, a shallow source
component, and a noise component.
Chapter 4: 2D Fast Fourier Transform and MAGMAP Theory 45
The following figure illustrates the division of an energy spectrum into these three
components.
N
d
s
n
w
E
MAGMAP is commonly used to enhance information of interest in a given 2D data
set, either by removing features considered as noise, or by enhancing the features
of interest. For example, if you are interested in shallow features in a magnetic map,
you might apply a first or second vertical derivative filter to the data in order to
enhance shallow features at the expense of anomalies caused by deeper sources.
MAGMAP takes advantage of the fact that potential field data, by its nature, is very
broadband, so that a single measurement includes the effects due to all the
physical (geological) sources. Resolution of the different sources depends on the
noise level of the measuring system, and the ability to resolve overlapping signals.
The FFT Algorithm
In MAGMAP, Fast Fourier Transform (FFT) is used to convert the space domain
grid data to the Fourier domain. The system applies Fast Fourier Transform (FFT) to
a space domain grid to produce a folded 2D transform as output. As part of the
process, MAGMAP also calculates and saves a radially averaged energy spectrum of
the transform.
The system creates a Fourier domain grid, which is called a Transform. It has the
same name as the input grid, but with the .TRN extension. The transform grid
contains a folded discrete Fourier transform of the input grid.
46 Chapter 4: 2D Fast Fourier Transform and MAGMAP Theory
The size of the transform grid element is 4 bytes; each pair of elements represents the
real and imaginary component of a complex number. The transform is stored in the
same way as the input grid, so that each transform vector (row) represents a vector in
the storage direction of the input grid (X for kx=1, Y for kx=1). The following table
illustrates the logical storage of the transform:
V (n) (1/nv) r i r i r i r i
V (n1) (2/nv) r i r i r i r i
V (n/2+2) (1/2v  1/nv) r i r i r i r i
V (n/2+1) 1/2v (nyq.) r i r i r i r i
V (n/2) 1/2v  1/nv r i r i r i r i
V (3) 2/nv r i r i r i r i
V (2) 1/nv r i r i r i r i
V (1) 0 r i r i r i r i
Vectors 0 1/ne 2/ne 1/2e (nyq.)
Elements E (1,2) E (3,4) E (5,6) E(n+1,n+2)
where:
r, i are real and imaginary components of each transform element
e, v are element and vector separations (cell size)
n is the original grid dimension in cells
The transform element separation (1/ne) and vector separation (1/nv) is 1 / (grid
dimension) cycles/metre. Since both the grid and the grid cell are square, 1/ne = 1/nv.
The Nyquist wavenumber is the largest wavenumber that has been sampled by the
grid, and is defined as one over twice the grid cell size (1/2e and 1/2v, which are also
equal).
Looking at the above table, you can note that each transform vector (row) represents
a discrete Fourier row in the direction of the input grid vectors. The Fourier elements
within each row start at 0 cycles/metre and extend to the Nyquist wavenumber in 1/ne
increments.
Chapter 4: 2D Fast Fourier Transform and MAGMAP Theory 47
As a result, the transform grid has (n/2 + 1) elements per vector, where n is the
number of elements per vector in the original grid. The transform is folded at the
Nyquist wavenumber in the direction of the grid vectors, so the transform grid has n
vectors.
The Energy Spectrum
In addition to producing the Fourier transform, the Forward FFT option
(FFT2IN.GX) also produces a file containing the radially averaged energy spectrum
(2D Power Spectrum) in a format similar to the following example:
/ 2D RADIALLY AVERAGED POWER SPECTRUM
/
/ WAVENUMBER INTERVAL DWE = 1.428571E01
/ AVERAGE SPECTRAL DENSITY LOG(ETOT) = 1.940511E+01
/
/ CYC/KM #_SAMP LOG_P 3_DEPTH 5_DEPTH
/ 
0.000000E+00 1 6.953915E+00 2.017633E01 *
1.428571E01 8 6.591711E+00 3.892244E01 *
2.857143E01 12 5.556448E+00 5.608587E01 4.363967E01
4.285714E01 16 4.578010E+00 3.591071E01 3.645031E01
5.714285E01 32 4.267114E+00 1.735434E01 2.545823E01
7.142857E01 28 3.954922E+00 2.310963E01 2.454431E01
8.571429E01 40 3.437389E+00 3.316897E01 3.152357E01
1.000000E+00 40 2.764027E+00 3.829211E01 2.908998E01
.
.
.
9.000000E+00 364 6.672572E+00 1.778083E01 5.273030E02
9.142857E+00 440 7.611941E+00 2.138515E01 1.244036E01
9.285714E+00 400 7.440382E+00 1.844887E02 1.012864E01
9.428572E+00 424 7.545702E+00 1.084567E01 1.966618E02
9.571428E+00 420 7.829783E+00 3.100928E02 1.257935E02
9.714286E+00 416 7.434367E+00 3.970939E02 4.879844E02
9.857142E+00 448 7.687212E+00 2.171140E01 *
1.000000E+01 394 8.213891E+00 2.933830E01 *
1.014286E+01 340 1.024793E+01 * *
1.028571E+01 348 1.104301E+01 * *
1.042857E+01 272 1.107746E+01 * *
The radially averaged energy listed in the third column represents the spectral density
(energy) averaged for all grid elements at the wavenumber in the first column. The
second column indicates the number of elements that were used to determine the
average. The energy is normalized by subtracting the log of the average spectral
density.
The 3DEPTH and 5DEPTH columns are ensemble magnetic depth estimates based
on 3 and 5 point averages of the slope of the energy spectrum (Spector and Grant,
1970). The depth to a statistical ensemble of sources is determined by the following
expression:
48 Chapter 4: 2D Fast Fourier Transform and MAGMAP Theory
h = s/ 4
where:
h is depth
s is the slope of the log (energy) spectrum
The above estimates can be used as a rough guide to the depth of magnetic source
populations.
The system enables you to create and view a radially averaged spectrum
automatically. The plot format is shown below.
The above plot illustrates the typical reduction in energy with increasing
wavenumber. The depth estimate is a plot of the 5point depth data from the spectrum
file.
Chapter 5: Preparing Grids for the FFT Processing
Chapter 5: Preparing Grids for the FFT Processing
49
Grid preparation includes the following basic processes:
1. Removing a first order trend from the grid. The removed trend is stored in the
user area of the grid header, and is filtered together with the zero wavenumber.
2. Expanding the grid dimensions by adding dummy areas to the grid edges to
produce a square grid. By default, the grid size is increased by a minimum of
10%, and then the next largest acceptable dimension is selected. The system uses
the FFT algorithm for dimensions up to 2520 x 2520 cells. Beyond that, the
algorithm switches to the power of 2 FFT methods.
3. Replacing all dummy values in the grid with the interpolated values from the
valid parts of the grid.
The following diagram illustrates the effect of the above processes in one dimension.
Because the preprocessed grid must be periodic, it is important to remove a first
order trend before expanding and filling.
Trend Removal (GRIDTRND)
Expand (GRIDXPND) and Fill to Periodic (GRIDFILL)
Original Data
If a significant trend is left in the data, the expansion and filling processes are forced
to introduce a large step function in order to make the data periodic. In the Fourier
domain, this step function predominates, and might cause ringing problems.
Trend Removal Algorithm
The system first removes a first order trend from the data. Because the data is made
periodic when filled, this procedure removes the amount of step function that may be
required to connect the grid at opposite edges.
The trend surface is (by default) calculated using the edge points of the data, so that
no strong anomalies within the grid affect the trend. The trend coefficients (first order
only) are stored in the grid file header and, during filtering, anything applied to the
zero wavenumber of the data is also applied to the trend coefficients.
50 Chapter 5: Preparing Grids for the FFT Processing
Grid Expansion Algorithm
The grid must be expanded in size to have dimensions that are acceptable to the FFT:
2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 24, 28, 30, 36, 40, 42, 48, 56, 60, 70, 72, 80, 84, 90,
112, 120, 126, 140, 144, 168, 180, 210, 240, 252, 280, 315, 336, 360, 420, 504, 560,
630, 720, 840, 1008, 1260, 1280, or 2520 cells. Also, the expansion provides an area
for extending the grid to be smoothly periodic. Grid expansion simply pads the edges
of the grid with dummy values.
The system allows you to specify a minimum percentage for expansion. The default is
10%, after which the next largest acceptable dimension is selected.
In most cases, the default size is acceptable. However, if the wavelength of the
anomalies of interest is more than approximately two times the default padded size,
you must increase the expanded size by increasing the expansion percentage. If the
expansion is too small, any step in the expanded area can adversely affect the
anomalies within the data.
Grid Filling Algorithm
The system replaces all the dummy values within a grid with the interpolated values.
It starts by interpolating the blank areas of the grid by:
1. Replacing dummies within each grid row so that the grid lines are periodic
2. Replacing dummies within each grid column so that the grid columns are periodic
3. Averaging the results from the row and column filling. The averaging procedure
uses inverse distance weighting to the nearest data within each row and column.
This procedure effectively fills holes in the data and accounts for irregular edges
of the grid.
When filling, you must avoid introducing abrupt step functions in the filled grid. You
can suspect a step problem if you observe ringing in a filtered output grid. Ringing
is a symptom of Gibb's Phenomena, which is observed when you modify the Fourier
Spectrum of a step function. Filters, by definition, must modify Fourier Spectrum;
therefore, filtered maps are susceptible to Gibb's Phenomena if you are not careful
when filling the grid.
Ringing can be identified as a wave pattern that extends away from, or around, a
strong anomaly. The wavelength of the pattern is normally near the size of the strong
feature in the data. Because the system interpolates data beyond the edges of the grid,
it can introduce step functions that cause ringing to spread into a filtered grid.
Minimizing "Ringing" from Grid Filling
MAGMAP provides the following parameters that can be set to minimize ringing
problems when they occur. For a complete summary of the trend removal, grid
Chapter 5: Preparing Grids for the FFT Processing 51
expansion, and filling parameters, see the Setting Trend Removal, Grid Expansion,
and Filling Parameters section at the end of this chapter.
Applying Maximum Entropy Prediction
If required, you can use Maximum Entropy Prediction (MEP) to interpolate the data.
MEP samples the original data near the grid edges to determine its spectral content. It
then predicts a data function that would have the same spectral signature as the
original data. This means that if the original data is smooth, the predicted data is
smooth, and if the original data is noisy, the predicted data is noisy.
As a result, the predicted data will not significantly alter the energy spectrum that
would result from the original data alone. Also, this method allows noisy data on one
edge of a grid to be gradually interpolated into smooth data on the opposite edge of a
grid.
Controlling Edge Effects
Sometimes, the prediction function can produce large ridges that extend away from
the edges of the data. If you suspect ringing caused by an edge effect, look at the pre
processed grid to see if edge filling has produced strong ridges in the filled areas.
This typically happens in the originally rectangular grids for which a large expansion
is required in one direction in order to make the grid square.
If you suspect such an edge problem, you can set the distance at which to roll the data
to zero. Ensure that the roll to zero distance is at least as large as the longest
anomalies of interest along the edges of the data, otherwise the edge anomalies may
be distorted.
Limiting Strong Anomalies Near Grid Edges
Very strong anomalies that are truncated at the edge of a grid can cause problems in
filling because their magnitude is extended into the filled borders of the grid. Step
functions parallel to the grid edges result, and ringing can manifest itself as a streak in
a filtered grid that extends away from the strong anomaly, or appears on the opposite
side of the grid (remember that the grid is considered periodic, so an anomaly on one
edge can effect the opposite edge of the grid).
If required, you can smoothly limit the magnitude of strong anomalies within a
certain distance of the edge of a grid. You can specify the maximum edge magnitude
and distance. All anomalies of greater magnitude than the edge limit will be smoothly
attenuated starting at half the magnitude limit.
In extreme cases, it may be necessary to totally limit edge anomalies by specifying
the maximum magnitude of 0. This is similar to applying the Hamming window to
the grid, which is the more conventional approach used to handle the edges of data in
52 Chapter 5: Preparing Grids for the FFT Processing
the Fourier processing. However, this method produces a pronounced edge ring
around most data grids, which is usually unacceptable.
Limiting Strong Anomalies by Magnitude
Very strong anomalies within the original data area of a grid can also cause ringing
problems in filtered maps. For example, iron formation anomalies in magnetic data
can be many orders of magnitude greater than the surrounding anomalies. The
magnitude of these anomalies is so great that they dominate the Fourier spectrum, and
even small changes to the spectrum result in ringing. Such ringing appears as waves
that surround the large anomaly.
If required, you can smoothly limit any anomalies that exceed a specified magnitude.
Data less than half the limiting magnitude is not changed. Above half the limiting
magnitude, data is smoothly attenuated so that it does not exceed the limit.
If limited anomalies are wide, this can cause the attenuated anomalies to have flat
tops, a fact that you should be aware of when interpreting the resulting filtered maps.
If flattopped anomalies are not wanted, another option is to clip high magnitude
anomalies using Geosoft's Grid Windowing GX (for more information, contact your
Geosoft representative). Clipped areas must be set to dummy values, and the resulting
processed grids will have holes where the anomalies have been clipped.
Setting Trend Removal, Grid Expansion, and Filling
Parameters
As discussed previously in this chapter, Geosoft provides a variety of capabilities for
preparing grids prior to applying the forward FFT. These capabilities are controlled
through the Prepare Grid menu option. When you select this option, the system
displays the FFT2 grid preprocessing dialog. The following table summarizes the
parameters that can be set in this dialog and provides guidelines for setting them.
Trend Removal Parameters
Type of trend surface to
remove
Select order of trend to remove (the default is the first order). Options
are: remove mean, first order, second order, third order.
trend based on
Select or type either 'edge points' or 'all points'. The trend surface to
remove can be calculated either by using all the valid points in the grid,
or by using only the points along the valid edge of the grid. Using the
edge points is often better, especially if there are any largemagnitude
anomalies within the grid.
Chapter 5: Preparing Grids for the FFT Processing 53
Grid Expansion Parameters
% expansion
Type the expansion size (grid is expanded in size by at least this
distance as a percentage of the smallest grid dimension). The expansion
must be about half the size of the broadest features of interest in the
grid. If you are tapering the data to 0, the expansion need not be larger
than the taper distance.
Square or rectangular
expansion
Select or type either 'square' or 'rectangular'. If the grid is small, or if
the wavenumbers of interest approaches the size of the grid, we
recommend square expansion because it minimises side effects that
result from having different wavenumber samples in the X and Y
directions. Rectangular grids can save significant processing time and
disk space when working with large grids.
Grid Filling and Ringing Parameters
grid fill method
Select or type a fill method. When filling the dummy areas, the new
values are determined by extrapolation from the nearest valid parts of
the grid. This extrapolation may be based on inverse distance
weighting or maximum entropy prediction. Maximum entropy is
slower but it creates a filled area more similar in character to the actual
data.
rolloff to 0 at a
distance of (cells)
Specify the number of cells beyond the valid area at which to roll off to
zero. By default, no rolloff is applied. For some grids, the prediction
function can become unreasonable at large distances from the valid
parts of the grid. In these cases, the data can be forced to zero at a
specified distance. This option should only be used on trendremoved
grids.
limit all magnitudes to
be less than
Highmagnitude anomalies can cause problems in filtering systems
such as MAGMAP. With this option, anomalies that exceed half the
specified limit are smoothly attenuated. The attenuation is started at
half the limit, with no values allowed to exceed the limit. This option
must be used only on trendremoved grids.
edge magnitude limit
Highmagnitude anomalies on the edges of the valid area can produce
oscillations in the extrapolated areas. With this option, a limit may be
placed on anomalies along the edges of the grid.
54 Chapter 6: Specifying FFT Filters
Chapter 6: Specifying FFT Filters
All MAGMAP processes are carried out by the application of filters in the Fourier
(wavenumber) domain. Filters are simply multiplied by the transform of the grid on
an elementbyelement basis.
Once a Fourier transform has been created, the application of filters is quite
straightforward. When using the StepByStep method, you select the Define Filters
menu option to either create a new control file or select an existing control file. With
the Interactive filtering method, you select the Interactive Spectrum Filters menu
option to create a control file. When you are ready to proceed, you select the Apply
Filters menu option to apply filters according to instructions defined in your new or
existing MAGMAP control file.
Creating Your Own Filter Control File
The MAGMAP control file used by the Apply Filter option is an ASCII text file that
may be created:
Using a text file editor
By copying, renaming, and editing the mapplot.con file in the C:\Program
Files\Geosoft\Oasis montaj\etc directory. The mapplot.con file is a blank control
file that contains a brief description of the parameters and filters available in
MAGMAP.
Using the MAGMAPStepByStepDefine Filtesr menu option or the
MAGMAPInteractive FilteringInteractive Spectrum Filters menu option.
Following is an example of a MAGMAP control file:
70 /geomagnetic inclination
0 /geomagnetic declination
BPAS 0.0001 0.003 1 /
BTWR 0.0002 4 0 /
TXYZ 0 3 /
CNDN 50 /
CNUP 200 /
COSN 0.001 0.003 2 1 /
The control file must contain 6 or more lines with the following information:
line 1:
A title line for reference only.
line 2:
The nominal height of the magnetic sensor above the ground (normally, the flying
height), or above the level of magnetic sources. This information is used as the default
height for some of the filters, and is not always necessary or relevant.
line 3:
The magnetic inclination (negative in the Southern hemisphere). The inclination is only
used by the reduction to the pole (REDP), reduction to the equator (REDE),
susceptibility (SUSC), and Weiner optimum (OPTM) filters.
Chapter 6: Specifying FFT Filters 55
line 4:
The magnetic declination in degrees of azimuth relative to true North. The inclination is
only used by the reduction to the pole (REDP), reduction to the equator (REDE),
susceptibility (SUSC) and Weiner optimum (OPTM) filters. Note that grid files contain
the direction of the grids Y axis in the grid header as the rotation parameter. Filter takes
this direction into account so that the Line 4 parameter can be the true declination. The
grid header must be correct. Very often, the grid rotation is reported as 0 in the grid
header, in which case the declination specified here must be the declination of magnetic
North relative to the grids Y axis.
line 5:
The nominal total magnetic field strength in nanoTesla (gammas). This value is only
used by the apparent susceptibility map filter (SUSC) in order to derive susceptibility
from magnetization.
line 6+
One or more lines specifying the filters to be applied and their parameters. The following
section documents the available filters.
The forward slash character (/) must terminate a line (with the exception of the title
line), and user comments may follow the slash. After the fifth line, all lines must start
either with a forward slash or a filter name.
Each filter option occupies one or more lines and consists of a fourletter mnemonic
followed by the optional parameter settings. Parameters (if provided) must be
separated by a space. Any number of filters may be applied in a single filtering run.
However, only one output transform is produced. Note that because multiplication is
commutative, the order in which filters are applied is not relevant.
MAGMAP Filters
This section describes the available filters. In each filterspecific subsection, filter
options are listed in alphabetical order. Each description shows the mathematical
expression of the filter followed by a figure if appropriate, then the control file
parameters, and usage notes. The filter expressions use the following basic
expressions:
X wavenumber (complex, radians/ground_unit)
Y wavenumber (complex, radians/ground_unit)
r = +
2 2
Wavenumber (radians/ground_unit)
( =
tan
1
)
Wavenumber direction
N Nyquist wavenumber [1 / (2 * cell size)]
k Wavenumber in cycles/ground_unit (r=2k)
The horizontal axis of the figures represents wavenumbers between 0 and the Nyquist
frequency. All distance references are multiples of the grid cell size. For example,
referring to the filter response drawing for upward continuation (CNUP) filter, if the
56 Chapter 6: Specifying FFT Filters
+
=
n
k
k
k L
0
1
1
= 2
8
16
4
1.0
0.0
0.5
n
0 N
L(k)
k
o
Wavenumber (cycles/ ground_unit)
Parameters:
k
0
The central wavenumber of the filter.
n The degree of the Butterworth filter function. By default, 8.
0/1 A flag (0 or 1). Specifies if a residual (0) high pass or a regional (1)
low pass is required. By default, a regional filter is applied.
Ground_unit The survey ground units used in your grid (e.g., metre, foot, etc.). If
your ground units are in metre, the low and high wavelength cutoff
is in cycles/metre.
The Butterworth filter is excellent for applying straight forward highpass and low
pass filters to data because you can easily control the degree of filter rolloff while
leaving the central wavenumber fixed. If ringing is observed, the degree can be
reduced until the result acceptable. A common but more complicated alternative is the
Cosine filter (COSN).
58 Chapter 6: Specifying FFT Filters
CNDN Downward Continuation
( ) L r e
hr
=
20.0
Wavenumber (radians/ground_unit)
L(r)
0 N
h = 2
h = 4
h = 8
h = 16
1.0
Parameters:
h The distance, in ground units, to continue down relative to the plane of
observation.
r Wavenumber (radians/ground_unit). Note: r = 2k, where k is
cycles/ground_unit
Ground_unit The survey ground units used in your grid (e.g., metre, foot, etc.).
Downward continuation is used to enhance the responses from sources at a depth by
effectively bringing the plane of measurement closer to the sources. Note that it is not
theoretically possible to continue through a potential field source. Since short
wavelength signal can appear to be from shallow sources, it must be removed to
prevent high magnitude and short wavelength noise in the processed data. To do this,
you usually apply some type of lowpass filter, such as the Butterworth or Weiner
Optimum filter. You should use a lowpass filter to remove the short wavelength
noise (as determined by the radially averaged energy spectrum) before applying the
downward continuation filter. The energy spectrum is also a good guide for
determining the depth to which the data can be continued downward.
CNUP Upward Continuation
( ) L r
hr
=
e
1.0
0.0
Wavenumber
(radians/ground_unit)
L(r)
0 N
h = 2
h = 4
h = 8
h = 16
Chapter 6: Specifying FFT Filters 59
Parameters:
h The distance, in ground units, to continue up relative to the plane of
observation.
r Wavenumber (radians/ground_unit). Note: r = 2k, where k is
cycles/ground_unit.
Ground_unit The survey ground units used in your grid (e.g., metre, foot, etc.).
Upward continuation is considered a clean filter because it produces almost no side
effects that may require application of other filters or processes to correct. Because of
this, this filter is often used to remove or minimize the effects of shallow sources and
noise in grids.
Also, upward continued data may be interpreted numerically and with modeling
programs. This is not the case for many other filter processes.
COSN Cosine Rolloff Filter
( )
( )
( )
L k k k
L k
k k
k k
k k k
L k k k
n
= <
=
= >
1
2
0
0
0
1 0
0 1
1
,
cos ,
,
for
for
for
1.0
0.0
0.5
0 N
n=2 1 0.5
k
1
L(k)
Wavenumber cycles/ ground_unit)
k
0
Parameters:
k
0
Low wavenumber starting point of the filter (cutoff wavenumber for
high pass or start of roll off for low pass.
k
1
High wavenumber end point of the filter (start of roll off for high pass
or cutoff wavenumber for low pass.
60 Chapter 6: Specifying FFT Filters
n The degree of the cosine function. The default is a degree of 2 for a
cosine squared roll off.
0/1 0 for residual (highpass) filter; 1 for regional (lowpass) filter. The
default is a lowpass filter.
Ground_unit The survey ground units used in your grid (e.g., metre, foot, etc.).
Because this filter has a smooth shape, and it does not alter the energy spectrum
below the start of roll off (or after the end of roll off in highpass mode), it is
commonly used for simple highpass or lowpass operations. To reduce ringing, the
separation between r
1
and r
0
can be increased.
DCOS Directional Cosine Filter
( ) ( )
(
)
( )
)
/2 +/2
n=2
1
0.5
Chapter 6: Specifying FFT Filters 61
0
N
0
1
N
N
N
0
L
(u,)
u
= 70
n = 2
0
N
0
1
N
N
N
0
L
(u,)
u
n = 1
0
N
0
1
N
N
N
0
L
(u,)
u
n = 0.5
Parameters:
Direction of the filter in degrees (0360 relative to North).
n The degree of the cosine function. By default, a degree of 2 is used to give a
cosine squared function.
0/1 If 1, apply the filter to pass the specified direction; if 0, apply the filter to
reject the specified direction. By default, the direction is rejected.
The directional cosine filter is very good for removing directional features from a
grid. The cosine function makes the filter smooth, so directional ringing effects are
usually not a problem. The rejection (or pass) notch can be narrowed or widened by
setting the degree of the cosine function, so that highly directional features can be
isolated. Decorrugation of poorly levelled magnetic data is a common application for
this filter (see examples).
DENS Apparent Density Calculation
( )
( )
L r
r
tr
=
2 G 1 e

where:
G Gravitational constant.
62 Chapter 6: Specifying FFT Filters
r Wavenumber (radians/ground_unit). Note: r = 2k, where k is
cycles/unit.
Parameters:
t Thickness, in ground units, of the earth model.
d Background density in g/cm
3
, to be added to the density contrast map.
The default is 0, so the density map is produced relative to the average
density.
Ground_unit The survey ground units used in your grid (e.g., metre, foot, etc.).
The data must also be downward continued using the CNDN filter to be close to the
top surface of the source model. Apparent density mapping assumes that an observed
gravity field can be explained by a simple model layer of fixed thickness and varying
density. This is a poor model in most cases.
DPAS Directional Pass/Reject Filter
0
N
0
1
N
N
N
0
L
(u,)
u
= 50
0
= 70
1
Parameters:
0
The low cutoff angle in degrees of azimuth (0360 from North).
1
The high cutoff angle in degrees of azimuth (0360 from North).
0/1 If 1, pass the defined band; if 0, reject the defined band. The default is to pass
the band.
As with the bandpass filter, the directionalpass often suffers from Gibb's
Phenomena (ringing) because the spectrum is cut quite abruptly. We recommend
using the directional cosine filter (DCOS) instead.
Chapter 6: Specifying FFT Filters 63
DRVX Derivative in the X Direction
( ) ( ) L i
n
=
Parameter:
n Order of differentiation.
u The X component of the wavenumber.
i i = 1
The horizontal derivative can be used for creating shaded images, and is required for
some modeling algorithms, such as Euler deconvolution.
DRVY Derivative in the Y Direction
( ) ( )
n
i L =
Parameter:
n Order of differentiation.
v The Y component of the wavenumber.
i i = 1
The horizontal derivative can be used for creating shaded images, and is required for
some modeling algorithms, such as Euler deconvolution.
DRVZ Derivative in the Z Direction
( ) L r r
n
=
Parameter:
n Order of differentiation.
r Wavenumber (radians/unit). Note: r = 2k, where k is cycles/unit.
The vertical derivative is commonly applied to total magnetic field data to enhance
the shallowest geologic sources in the data. As with other filters that enhance the
highwavenumber components of the spectrum, you must often also apply lowpass
filters to remove highwavenumber noise.
64 Chapter 6: Specifying FFT Filters
GFILT Gravity Earth Filter
( )
r
rz
e
rz
e
G r L
) (
2
2 1
=
When r=0:
( ) ) ( 2
1 2
z z G r L =
Where:
G Gravitational constant
r Wavenumber (radians/ground_unit). Note: r = 2k, where k is cycles/unit.
Parameters:
z
1
Depth to the top of the density layer, in ground units. Must be positive for
layers above calculation level.
z
2
Depth to the bottom of the density layer, in ground units. Must be positive for
layers above calculation level.
Ground_unit The survey ground units used in your grid (e.g., metre, foot, etc.).
GAUS Gaussian Regional/Residual Filter
( )
2
0
2
2
1
k
k
e k L
=
1.0
0.0
0.5
0 N
L(k)
Wavenumber
(cycles/ ground_unit)
K
0
2K
0
3K
0
Parameters:
k
0
Standard deviation of the Gaussian function in cycles/ground_unit
(similar to a cutoff point, except that the function magnitude at this
point is only 0.39).
Chapter 6: Specifying FFT Filters 65
=
), (  if ,
cos cos sin
/
2

Where:
I Geomagnetic inclination
I
a
Inclination for magnitude correction (never less than I). The default is 20
degrees. If I
a

is specified to be less than I, it is set to I.
d Density contrast in g/cm^3.
G Gravitational constant, 6.670E8.
J Magnetization in Gauss.
D Geomagnetic declination.
Parameters:
Direction of wavenumber in degrees of azimuth.
r Wavenumber (radians/ground_unit). Note: r = 2k, where k is
cycles/ground_unit.
Ground_unit The survey ground units used in your grid (e.g., metre, foot, etc.).
Output of PseudoGravity is in mgal if ground_unit is metre, or in 0.3mgal if
groud_unit is foot.
Chapter 6: Specifying FFT Filters 67
HPAS Highpass Filter
( )
( )
L k k k
L k k k
= <
=
0
1
0
0
, for
, for
1.0
0.0
0.5
pass reject
L(k)
k0
Wavenumber (cycles/ground_unit)
Parameters:
k
0
The cutoff wavenumber in cycles/ground_unit. All wavenumbers
below this value are removed.
Ground_unit The survey ground units used in your grid (e.g., metre, foot, etc.).
As with the bandpass filter, the highpass filter is seldom used because the results
usually suffer from Gibb's Phenomena (ringing).
INTG Vertical Integration
L(r) = r
1
Parameters:
r Wavenumber (radians/ground_unit). Note: r = 2k, where k is
cycles/ground_unit.
Ground_unit The survey ground units used in your grid (e.g., metre, foot, etc.).
This filter calculates the vertical integral of the input transform. This is the inverse of
the vertical derivative. The zero wavenumber is set to 0.
68 Chapter 6: Specifying FFT Filters
LPAS Lowpass Filter
L(k) = 1, for k k
0
L(k) = 0, for k > k
0
1.0
0.0
0.5
pass
reject
k0
Wavenumber (cycles/ground_unit)
L(k)
Parameters:
k
0
The cutoff wavenumber in cycles/ground_unit. All wavenumbers
above this value are removed.
Ground_unit The survey ground units used in your grid (e.g., metre, foot, etc.).
As with the bandpass filter, the lowpass filter is seldom used because the results
usually suffer from Gibb's Phenomena (ringing).
OPTM Weiner Optimum Filter
The Weiner optimum filter is intended to remove the effect of white noise from the
magnetic data. White noise is highwavenumber background noise present in the data.
Because magnetic signal is stronger in the direction of the inducing field, the signal
tonoise ratio varies as a function of both inclination and declination. The Weiner
optimum filter takes the variation of signaltonoise ratio into account when applying
the filter.
( )
( )
( )
( )
( )
( )
( )
( )
( ) [ ]
( )
0
4 2 2 4
2
2 2 2
1 0
0 1
0 2
0
0
0
cos 375 . 0 cos sin sin
cos cos sin
,
1 for , 0 ,
for ,
2
cos
,
,
,
for ,
,
,
,
+ +
+ +
=
> =
+
=
<
+
=
k s
s
s
s
s
I I I I
D I I
k
k k k L
k k k
k k
k k
k
k
k L
k k
k
k
k L
Chapter 6: Specifying FFT Filters 69
Where:
I Geomagnetic inclination
D Geomagnetic declination
r
Average radial spectral density
0
Average radial spectral density of noise
k
0
Wavenumber of start of the noise
k
1
Wavenumber at the end of the applied rolloff filter
Parameters:
h The depth at which to interpret an optimum filter from the observed
energy spectrum. By default, the depth is taken as the flying height in
the control file, or the continuation depth if specified by the
Downward Continuation or Apparent Susceptibility filter options (in
ground_units).
k
0
The wavenumber (cycles/ground_unit) at which to start the high
wavenumber roll off. This parameter must be given together with
0
.
By default, this point is the point at which the slope of the observed
energy spectrum rises above the slope defined by the depth (1/4h).
k
1
The wavenumber (cycles/ground_unit) at which to end the high
wavenumber roll off. By default, this point is set to be two times k
0
.
0
Spectral density estimate of noise to be removed by the Weiner filter.
This is in terms of the log of spectral density as reported in the second
column of the energy spectrum. By default, this is calculated as the
average of the spectral density between k
0
and k
1
.
Ground_unit The survey ground units used in your grid (e.g., metre, foot, etc.).
The optimum filter is most often used to remove the theoretical effect of all sources
that lie above a specified depth. The filter parameters can be specified or calculated
automatically based on analysis of the energy spectrum.
Although this filter can calculate the parameters of the filter, we recommend that you
confirm that the calculated parameters are reasonable. When the energy spectrum is
not smooth, the filter can choose the wrong point at which to start the noise
calculation. Most often, this point is chosen to be too low, and the resulting maps
appear too smooth.
70 Chapter 6: Specifying FFT Filters
The optimum filter can be quite complex to use and understand. A good alternative is
using the Butterworth filter as a lowpass filter. Determine the wavenumber at which
sources appear too shallow by interpreting the depth estimate in the energyspectrum
plot.
REDE Reduce to the Magnetic Equator
( )
[ ]
[ ] [ ]
I Ia I Ia if
D I I D Ia Ia
D D I i I
L = <
+ +
= ),   ( ,
) ( cos ) ( cos ) ( sin ) ( cos ) ( cos ) ( sin
)) ( cos ( ) cos( ) cos( ) sin(
2 2 2 2 2 2
2 2
where:
I Geomagnetic inclination
D Geomagnetic declination
L() Direction of the wavenumber vector in degrees of azimuth
No parameters.
Reduction to the equator is used in low magnetic latitudes to centre the peaks of
magnetic anomalies over their sources. This can make the data easier to interpret
while not losing any geophysical meaning. Reducing the data to the pole (REDP)
does much the same thing, but at low latitudes. A separate magnitude correction is
usually required to prevent NorthSouth signal in the data from dominating the
results. As a result, reduced to the pole data may present a less honest view of the
data
REDP Reduce to the Magnetic Pole
( )
[ ]
[ ] [ ]
I Ia I Ia if
D I I D Ia Ia
D I i I
L = <
+ +
= ),   ( ,
) ( cos ) ( cos ) ( sin ) ( cos ) ( cos ) ( sin
) cos( ) cos( ) sin(
2 2 2 2 2 2
2
where:
I Geomagnetic inclination
Ia Inclination for magnitude correction (never less than I)
D Geomagnetic declination
Parameters:
I
a
Inclination to be used for the magnitude correction. The default is 20
degrees. If I
a
 is specified to be less than I, it is set to I.
Chapter 6: Specifying FFT Filters 71
Reduction to the pole has a magnitude component (the sin(I) term) and a phase
component (the i cos(I) cos(D) term). When reducing to the pole from equatorial
latitudes, NorthSouth features can be exaggerated due to the strong magnitude
correction (the sin(I) component) that is applied when (D) is /2 (i.e., a magnetic
EastWest wavenumber). By specifying higher latitude for the magnitude correction
alone, this problem can be reduced or eliminated at the expense of undercorrecting
the magnitudes of NorthSouth features.
A magnitude inclination of 90 causes only the phase component to be applied to the
data (no magnitude correction), and a value of 0 (zero) causes phase and magnitude
components to be applied over the entire range.
SUSC Apparent Susceptibility Calculation
The susceptibility filter is, in fact, a compound filter that performs a reduction to the
pole, downward continuation to the source depth, correction for the geometric effect
of a vertical squareended prism, and division by the total magnetic field to yield
susceptibility.
( )
( ) ( ) ( )
( )
( ) ( )]
( )
( ) sin( ar sin )
cos
cos sin
,
cos cos [sin
, 2
1
,
=
+ =
=
=
ar sin
ar
ar
r
D I i I
e r
r r F
r L
a
hr
2
where:
(r) Downward continuation to h
() Reduction to the pole
(r,) Geometric factor of a vertical prism (aa in dimension)
I Geomagnetic inclination
I
a
Pole reduction magnitude inclination
D Geomagnetic declination
F Total geomagnetic field strength
r Wavenumber (radians/ground_unit). Note: r = 2k, where k is
cycles/ground_unit.
72 Chapter 6: Specifying FFT Filters
Wavenumber direction (azimuth)
a Half the cell size
Parameters:
h Depth in ground units, relative to the observation level at which to
calculate the susceptibility. By default, the flying altitude reported in
line 2 of the MAGMAP Control File is used.
I
a
Inclination to which to use the phase component only in the reduction
to the pole. The default is 20. If I
a
 is specified to be less then I, it
is set to I.
Ground_unit The survey ground units used in your grid (e.g., metre, foot, etc.).
The susceptibility filter calculates the apparent magnetic susceptibility of the
magnetic sources using the following assumptions:
The IGRF is removed from the magnetic field .
There is no remnant magnetization
All magnetic responses are caused by a collection of vertical, squareended
prisms of infinite depth extent
The result is in the e.m.u. units
The validity of the results depends on how well the actual observed field conforms to
these assumptions.
TXYZ Conversion between Field Components
TXYZ converts from the total field (T) or vertical field (Z) component of the
magnetic field to any other (X,Y, Z, or T) component.
The following table gives the filter expressions for all possible component field
conversions,
To X To Y To Z To T
From X 1
v / u r / iu P / iu
From Y u / v
1 r / iv P / iv
From Z iu / r
iv / r 1 P / r
From T iu / P
iv / P r / P 1
Chapter 6: Specifying FFT Filters 73
Where:
I Geomagnetic inclination in degrees from the horizon.
D Geomagnetic declination in degrees azimuth.
i
1
u The X wavenumber
v The Y wavenumber
r
The radial wavenumber =
2 2
v u +
Directional cosine of the Total magnetic field of X axis.
Directional cosine of the Total magnetic field of Y axis.
Directional cosine of the Total magnetic field of Z axis.
P
r v u i + + ) (
Theoretically, any one directional component can be used to calculate any other
component. In reality, only the T and Z fields work well as inputs for this conversion
because the formulas for conversion from X or Y contain singularities, i.e., u can go
to zero while v is large, or vice versa, causing the filter coefficient to become infinite.
Filter Examples
This section provides examples of filter application.
Susceptibility Map
The following control file (susc.con) applies an optimum depth filter and creates a
magnetic susceptibility map calculated at a depth of 200 metres below the survey
elevation.
susceptibility map
120 /sensor elevation
72 /magnetic inclination
12 /magnetic declination
55000 /total field strength
SUSC 200 /susceptibility map
OPTM /optimum depth filter
The susceptibility map filter includes a downward continuation to the source depth.
Because a downward continuation filter magnitude increases with wavenumber, it
tends to also amplify highwavenumber noise in the data. To prevent this noise from
entering the final map, you often need to remove high wavenumbers that are
considered noise in the data. This is the function of the Weiner Optimum filter
applied by the OPTM option.
74 Chapter 6: Specifying FFT Filters
Second Vertical Derivative
The following control file creates a second vertical derivative. An upward
continuation of 20 metres (equivalent to the cell size) is also applied to reduce the
effect of high wavenumber noise. The Hanning filter applied after processing can also
serve to reduce noise.
second vertical derivative
120 /sensor elevation
72 /magnetic inclination
12 /magnetic declination
55000 /total field strength
DRV2 /second vertical derivative
CNUP 20 /continue up 20 metres.
Decorrugation
Given a grid surveyed with a nominal line spacing of 150 metres, which has a lineto
line levelling problem, the following control file produces a grid that contains the
levelling error only. Following a rule of thumb, the Butterworth highpass filter is set
to four times the line separation in order to only pass frequencies on the order of the
line separation. The directional cosine filter is set to pass wavelengths only in the
direction of the lines (note that NorthSouth line levelling error produces
wavenumbers in the EastWest direction, hence DCOS 90). Because levelling error is
very directional, you can tighten the directional cosine function to an energy value of
0.5.
decorr., 150m line separation, NS lines
120 /sensor elevation
72 /magnetic inclination
12 /magnetic declination
55000 /total field strength
BTWR .00167 8 0 /highpass butterworth
DCOS 90 0.5 1 /directional cosine.
The resulting grid can be subtracted from the original grid to remove the levelling
error.
Some tuning of both the center wavenumber of the Butterworth filter and the energy
of the directional cosine may be required. To remove the more directional signal,
increase the energy of the cosine filter. To remove wider features, decrease the
Butterworth cutoff point. If you see ringing in the data, decrease the order of the
Butterworth filter.
Chapter 7: Applying the Inverse FFT
Chapter 7: Applying the Inverse FFT
75
As described earlier, in the StepByStep method, you use the Define Filters menu
option to select the filters and define their parameters and the Apply Filters menu
option to apply the filters to the transform file. Depending on the processing option
you select ([OK], [FltInv Only], or [Filter Only]) on the FFT2FLT dialog, you can
(i) apply the selected filter, (ii) apply the Inverse FFT (converts the transform file
from the wavenumber domain back to the original space domain) and (iii) apply post
processing (returns the grid to the original dimensions with the trend information
restored).
The three options for controlling the final results you obtain following the Apply
Filters option are detailed below:
Processing Option Selection Result
1
Filtered grid with
postprocessing
[OK] Filtered, spacedomain GRD file with original
grid dimensions and trend information restored
(i.e., postprocessing is added to output file).
2
Filtered grid with no
postprocessing
[FltInv Only]. Filtered, spacedomain GRD file as output file
with gridfilled dimensions and trend information
NOT restored (i.e., no postprocessing is added to
output file).
3
Filter only no
inverse transform
[Filter Only] Filtered transform (*.TRN) file as output file.
When you select the first processing option ([OK]) the transform grid is filtered, the
Inverse FFT is applied (returning the transform file from the wavenumber domain
back to the space domain) and then postprocessing restores the square and periodic
grid to its original dimensions and restores the trend information.
When you select the second processing option ([FltInv Only]) the transform grid is
filtered and the Inverse FFT is applied (returning the transform file from the
wavenumber domain back to the space domain). No postprocessing is applied.
When you select the third processing option ([Filter Only]) the transform grid is
filtered. The Inverse FFT and postprocessing is not applied.
76 References
References
Bhattacharya, B. K., 1966, Continuous spectrum of the total magnetic field anomaly due to a
rectangular prismatic body. Geophysics, Vol. 31, p.97121.
Spector, A. and Grant, F. S., 1970, Statistical models for interpreting aeromagnetic data.
Geophysics, Vol. 35, No.2, p.293302.
Wiener, N., 1949, Extrapolation, Interpolation, and Smoothing of Stationary Time Series.
Cambridge, M.I.T. Press.
Burg, J. P., 1975, Maximum Entropy Special Analysis. Unpublished doctoral dissertation.
Standford University. 168p.
Gupta, V. K., and Grant, F. S., 1985, Mineral exploration aspects of gravity and aeromagnetic
survey in SudburyCobalt area, Ontario. SEG; The Utility of Regional
Gravity and Magnetic Anomaly Maps, W. J. Hinze (Editor) p.392411.
Winograd, S. On Computing the Discrete Fourier Transform, Mathematics of Computation,
Vol.32, N0.141, pp.175199, Jan. 1978.
McClellan, J. H. and Nawab H., Complex GeneralN Winograd Fourier Transform Algorithm
(WFTA), Programs for Digital Signal Processing, IEEE Press, pp.
1.71  1.710, 1979.
Walter R. Roest, Jacob Verhoef, and Mark Pilkington, 1992, Magnetic interpretation using
the 3D analytic signal. Geophysics, Vol. 57, No.1, p.116125.
Ian N. MacLeod, Keith Jones, and TingFan Dai, 1993, 3D Analytic Signal in the
Interpretation of Total Magnetic Field Data at Low magnetic Latitudes.
Exploration Geophysics, Vol.24, p. 679688.
Bruno Verduzco, J. Derek Fairhead, Chris M. Green, and Chris MacKenzie, Feb. 2004, New
insights into magnetic derivatives for structural mapping. The Leading
Edge, p. 116119.
Гораздо больше, чем просто документы.
Откройте для себя все, что может предложить Scribd, включая книги и аудиокниги от крупных издательств.
Отменить можно в любой момент.